TDB2: Goodbye TDB2, Hello NTDB.

This renames everything from tdb2 to ntdb: importantly, we no longer
use the tdb_ namespace, so you can link against both ntdb and tdb if
you want to.

This also enables building of standalone ntdb by the autobuild script.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

lib/ntdb/ABI/ntdb-0.9.sigs

@@ -0,0 +1,39 @@
+ntdb_add_flag: void (struct ntdb_context *, unsigned int)
+ntdb_append: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA, NTDB_DATA)
+ntdb_chainlock: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA)
+ntdb_chainlock_read: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA)
+ntdb_chainunlock: void (struct ntdb_context *, NTDB_DATA)
+ntdb_chainunlock_read: void (struct ntdb_context *, NTDB_DATA)
+ntdb_check_: enum NTDB_ERROR (struct ntdb_context *, enum NTDB_ERROR (*)(NTDB_DATA, NTDB_DATA, void *), void *)
+ntdb_close: int (struct ntdb_context *)
+ntdb_delete: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA)
+ntdb_error: enum NTDB_ERROR (struct ntdb_context *)
+ntdb_errorstr: const char *(enum NTDB_ERROR)
+ntdb_exists: bool (struct ntdb_context *, NTDB_DATA)
+ntdb_fd: int (const struct ntdb_context *)
+ntdb_fetch: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA, NTDB_DATA *)
+ntdb_firstkey: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA *)
+ntdb_foreach_: void (int (*)(struct ntdb_context *, void *), void *)
+ntdb_get_attribute: enum NTDB_ERROR (struct ntdb_context *, union ntdb_attribute *)
+ntdb_get_flags: unsigned int (struct ntdb_context *)
+ntdb_get_seqnum: int64_t (struct ntdb_context *)
+ntdb_lockall: enum NTDB_ERROR (struct ntdb_context *)
+ntdb_lockall_read: enum NTDB_ERROR (struct ntdb_context *)
+ntdb_name: const char *(const struct ntdb_context *)
+ntdb_nextkey: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA *)
+ntdb_open: struct ntdb_context *(const char *, int, int, mode_t, union ntdb_attribute *)
+ntdb_parse_record_: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA, enum NTDB_ERROR (*)(NTDB_DATA, NTDB_DATA, void *), void *)
+ntdb_remove_flag: void (struct ntdb_context *, unsigned int)
+ntdb_repack: enum NTDB_ERROR (struct ntdb_context *)
+ntdb_set_attribute: enum NTDB_ERROR (struct ntdb_context *, const union ntdb_attribute *)
+ntdb_store: enum NTDB_ERROR (struct ntdb_context *, NTDB_DATA, NTDB_DATA, int)
+ntdb_summary: enum NTDB_ERROR (struct ntdb_context *, enum ntdb_summary_flags, char **)
+ntdb_transaction_cancel: void (struct ntdb_context *)
+ntdb_transaction_commit: enum NTDB_ERROR (struct ntdb_context *)
+ntdb_transaction_prepare_commit: enum NTDB_ERROR (struct ntdb_context *)
+ntdb_transaction_start: enum NTDB_ERROR (struct ntdb_context *)
+ntdb_traverse_: int64_t (struct ntdb_context *, int (*)(struct ntdb_context *, NTDB_DATA, NTDB_DATA, void *), void *)
+ntdb_unlockall: void (struct ntdb_context *)
+ntdb_unlockall_read: void (struct ntdb_context *)
+ntdb_unset_attribute: void (struct ntdb_context *, enum ntdb_attribute_type)
+ntdb_wipe_all: enum NTDB_ERROR (struct ntdb_context *)

lib/ntdb/check.c

@@ -0,0 +1,864 @@
+ /*
+   Trivial Database 2: free list/block handling
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "private.h"
+#include <ccan/likely/likely.h>
+#include <ccan/asearch/asearch.h>
+
+/* We keep an ordered array of offsets. */
+static bool append(ntdb_off_t **arr, size_t *num, ntdb_off_t off)
+{
+	ntdb_off_t *new = realloc(*arr, (*num + 1) * sizeof(ntdb_off_t));
+	if (!new)
+		return false;
+	new[(*num)++] = off;
+	*arr = new;
+	return true;
+}
+
+static enum NTDB_ERROR check_header(struct ntdb_context *ntdb, ntdb_off_t *recovery,
+				   uint64_t *features, size_t *num_capabilities)
+{
+	uint64_t hash_test;
+	struct ntdb_header hdr;
+	enum NTDB_ERROR ecode;
+	ntdb_off_t off, next;
+
+	ecode = ntdb_read_convert(ntdb, 0, &hdr, sizeof(hdr));
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+	/* magic food should not be converted, so convert back. */
+	ntdb_convert(ntdb, hdr.magic_food, sizeof(hdr.magic_food));
+
+	hash_test = NTDB_HASH_MAGIC;
+	hash_test = ntdb_hash(ntdb, &hash_test, sizeof(hash_test));
+	if (hdr.hash_test != hash_test) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "check: hash test %llu should be %llu",
+				  (long long)hdr.hash_test,
+				  (long long)hash_test);
+	}
+
+	if (strcmp(hdr.magic_food, NTDB_MAGIC_FOOD) != 0) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "check: bad magic '%.*s'",
+				  (unsigned)sizeof(hdr.magic_food),
+				  hdr.magic_food);
+	}
+
+	/* Features which are used must be a subset of features offered. */
+	if (hdr.features_used & ~hdr.features_offered) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "check: features used (0x%llx) which"
+				  " are not offered (0x%llx)",
+				  (long long)hdr.features_used,
+				  (long long)hdr.features_offered);
+	}
+
+	*features = hdr.features_offered;
+	*recovery = hdr.recovery;
+	if (*recovery) {
+		if (*recovery < sizeof(hdr)
+		    || *recovery > ntdb->file->map_size) {
+			return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+					  "ntdb_check:"
+					  " invalid recovery offset %zu",
+					  (size_t)*recovery);
+		}
+	}
+
+	for (off = hdr.capabilities; off && ecode == NTDB_SUCCESS; off = next) {
+		const struct ntdb_capability *cap;
+		enum NTDB_ERROR e;
+
+		cap = ntdb_access_read(ntdb, off, sizeof(*cap), true);
+		if (NTDB_PTR_IS_ERR(cap)) {
+			return NTDB_PTR_ERR(cap);
+		}
+
+		/* All capabilities are unknown. */
+		e = unknown_capability(ntdb, "ntdb_check", cap->type);
+		next = cap->next;
+		ntdb_access_release(ntdb, cap);
+		if (e)
+			return e;
+		(*num_capabilities)++;
+	}
+
+	/* Don't check reserved: they *can* be used later. */
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR check_hash_tree(struct ntdb_context *ntdb,
+				      ntdb_off_t off, unsigned int group_bits,
+				      uint64_t hprefix,
+				      unsigned hprefix_bits,
+				      ntdb_off_t used[],
+				      size_t num_used,
+				      size_t *num_found,
+				      enum NTDB_ERROR (*check)(NTDB_DATA,
+							      NTDB_DATA, void *),
+				      void *data);
+
+static enum NTDB_ERROR check_hash_chain(struct ntdb_context *ntdb,
+				       ntdb_off_t off,
+				       uint64_t hash,
+				       ntdb_off_t used[],
+				       size_t num_used,
+				       size_t *num_found,
+				       enum NTDB_ERROR (*check)(NTDB_DATA,
+							       NTDB_DATA,
+							       void *),
+				       void *data)
+{
+	struct ntdb_used_record rec;
+	enum NTDB_ERROR ecode;
+
+	ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec));
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (rec_magic(&rec) != NTDB_CHAIN_MAGIC) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: Bad hash chain magic %llu",
+				  (long long)rec_magic(&rec));
+	}
+
+	if (rec_data_length(&rec) != sizeof(struct ntdb_chain)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check:"
+				  " Bad hash chain length %llu vs %zu",
+				  (long long)rec_data_length(&rec),
+				  sizeof(struct ntdb_chain));
+	}
+	if (rec_key_length(&rec) != 0) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: Bad hash chain key length %llu",
+				  (long long)rec_key_length(&rec));
+	}
+	if (rec_hash(&rec) != 0) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: Bad hash chain hash value %llu",
+				  (long long)rec_hash(&rec));
+	}
+
+	off += sizeof(rec);
+	ecode = check_hash_tree(ntdb, off, 0, hash, 64,
+				used, num_used, num_found, check, data);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	off = ntdb_read_off(ntdb, off + offsetof(struct ntdb_chain, next));
+	if (NTDB_OFF_IS_ERR(off)) {
+		return NTDB_OFF_TO_ERR(off);
+	}
+	if (off == 0)
+		return NTDB_SUCCESS;
+	(*num_found)++;
+	return check_hash_chain(ntdb, off, hash, used, num_used, num_found,
+				check, data);
+}
+
+static enum NTDB_ERROR check_hash_record(struct ntdb_context *ntdb,
+					ntdb_off_t off,
+					uint64_t hprefix,
+					unsigned hprefix_bits,
+					ntdb_off_t used[],
+					size_t num_used,
+					size_t *num_found,
+					enum NTDB_ERROR (*check)(NTDB_DATA,
+								NTDB_DATA,
+								void *),
+					void *data)
+{
+	struct ntdb_used_record rec;
+	enum NTDB_ERROR ecode;
+
+	if (hprefix_bits >= 64)
+		return check_hash_chain(ntdb, off, hprefix, used, num_used,
+					num_found, check, data);
+
+	ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec));
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (rec_magic(&rec) != NTDB_HTABLE_MAGIC) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: Bad hash table magic %llu",
+				  (long long)rec_magic(&rec));
+	}
+	if (rec_data_length(&rec)
+	    != sizeof(ntdb_off_t) << NTDB_SUBLEVEL_HASH_BITS) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check:"
+				  " Bad hash table length %llu vs %llu",
+				  (long long)rec_data_length(&rec),
+				  (long long)sizeof(ntdb_off_t)
+				  << NTDB_SUBLEVEL_HASH_BITS);
+	}
+	if (rec_key_length(&rec) != 0) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: Bad hash table key length %llu",
+				  (long long)rec_key_length(&rec));
+	}
+	if (rec_hash(&rec) != 0) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: Bad hash table hash value %llu",
+				  (long long)rec_hash(&rec));
+	}
+
+	off += sizeof(rec);
+	return check_hash_tree(ntdb, off,
+			       NTDB_SUBLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS,
+			       hprefix, hprefix_bits,
+			       used, num_used, num_found, check, data);
+}
+
+static int off_cmp(const ntdb_off_t *a, const ntdb_off_t *b)
+{
+	/* Can overflow an int. */
+	return *a > *b ? 1
+		: *a < *b ? -1
+		: 0;
+}
+
+static uint64_t get_bits(uint64_t h, unsigned num, unsigned *used)
+{
+	*used += num;
+
+	return (h >> (64 - *used)) & ((1U << num) - 1);
+}
+
+static enum NTDB_ERROR check_hash_tree(struct ntdb_context *ntdb,
+				      ntdb_off_t off, unsigned int group_bits,
+				      uint64_t hprefix,
+				      unsigned hprefix_bits,
+				      ntdb_off_t used[],
+				      size_t num_used,
+				      size_t *num_found,
+				      enum NTDB_ERROR (*check)(NTDB_DATA,
+							      NTDB_DATA, void *),
+				      void *data)
+{
+	unsigned int g, b;
+	const ntdb_off_t *hash;
+	struct ntdb_used_record rec;
+	enum NTDB_ERROR ecode;
+
+	hash = ntdb_access_read(ntdb, off,
+			       sizeof(ntdb_off_t)
+			       << (group_bits + NTDB_HASH_GROUP_BITS),
+			       true);
+	if (NTDB_PTR_IS_ERR(hash)) {
+		return NTDB_PTR_ERR(hash);
+	}
+
+	for (g = 0; g < (1 << group_bits); g++) {
+		const ntdb_off_t *group = hash + (g << NTDB_HASH_GROUP_BITS);
+		for (b = 0; b < (1 << NTDB_HASH_GROUP_BITS); b++) {
+			unsigned int bucket, i, used_bits;
+			uint64_t h;
+			ntdb_off_t *p;
+			if (group[b] == 0)
+				continue;
+
+			off = group[b] & NTDB_OFF_MASK;
+			p = asearch(&off, used, num_used, off_cmp);
+			if (!p) {
+				ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						   NTDB_LOG_ERROR,
+						   "ntdb_check: Invalid offset"
+						   " %llu in hash",
+						   (long long)off);
+				goto fail;
+			}
+			/* Mark it invalid. */
+			*p ^= 1;
+			(*num_found)++;
+
+			if (hprefix_bits == 64) {
+				/* Chained entries are unordered. */
+				if (is_subhash(group[b])) {
+					ecode = NTDB_ERR_CORRUPT;
+					ntdb_logerr(ntdb, ecode,
+						   NTDB_LOG_ERROR,
+						   "ntdb_check: Invalid chain"
+						   " entry subhash");
+					goto fail;
+				}
+				h = hash_record(ntdb, off);
+				if (h != hprefix) {
+					ecode = NTDB_ERR_CORRUPT;
+					ntdb_logerr(ntdb, ecode,
+						   NTDB_LOG_ERROR,
+						   "check: bad hash chain"
+						   " placement"
+						   " 0x%llx vs 0x%llx",
+						   (long long)h,
+						   (long long)hprefix);
+					goto fail;
+				}
+				ecode = ntdb_read_convert(ntdb, off, &rec,
+							 sizeof(rec));
+				if (ecode != NTDB_SUCCESS) {
+					goto fail;
+				}
+				goto check;
+			}
+
+			if (is_subhash(group[b])) {
+				uint64_t subprefix;
+				subprefix = (hprefix
+				     << (group_bits + NTDB_HASH_GROUP_BITS))
+					+ g * (1 << NTDB_HASH_GROUP_BITS) + b;
+
+				ecode = check_hash_record(ntdb,
+					       group[b] & NTDB_OFF_MASK,
+					       subprefix,
+					       hprefix_bits
+						       + group_bits
+						       + NTDB_HASH_GROUP_BITS,
+					       used, num_used, num_found,
+					       check, data);
+				if (ecode != NTDB_SUCCESS) {
+					goto fail;
+				}
+				continue;
+			}
+			/* A normal entry */
+
+			/* Does it belong here at all? */
+			h = hash_record(ntdb, off);
+			used_bits = 0;
+			if (get_bits(h, hprefix_bits, &used_bits) != hprefix
+			    && hprefix_bits) {
+				ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						   NTDB_LOG_ERROR,
+						   "check: bad hash placement"
+						   " 0x%llx vs 0x%llx",
+						   (long long)h,
+						   (long long)hprefix);
+				goto fail;
+			}
+
+			/* Does it belong in this group? */
+			if (get_bits(h, group_bits, &used_bits) != g) {
+				ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						   NTDB_LOG_ERROR,
+						   "check: bad group %llu"
+						   " vs %u",
+						   (long long)h, g);
+				goto fail;
+			}
+
+			/* Are bucket bits correct? */
+			bucket = group[b] & NTDB_OFF_HASH_GROUP_MASK;
+			if (get_bits(h, NTDB_HASH_GROUP_BITS, &used_bits)
+			    != bucket) {
+				used_bits -= NTDB_HASH_GROUP_BITS;
+				ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						   NTDB_LOG_ERROR,
+						   "check: bad bucket %u vs %u",
+						   (unsigned)get_bits(h,
+							NTDB_HASH_GROUP_BITS,
+							&used_bits),
+						   bucket);
+				goto fail;
+			}
+
+			/* There must not be any zero entries between
+			 * the bucket it belongs in and this one! */
+			for (i = bucket;
+			     i != b;
+			     i = (i + 1) % (1 << NTDB_HASH_GROUP_BITS)) {
+				if (group[i] == 0) {
+					ecode = NTDB_ERR_CORRUPT;
+					ntdb_logerr(ntdb, ecode,
+						   NTDB_LOG_ERROR,
+						   "check: bad group placement"
+						   " %u vs %u",
+						   b, bucket);
+					goto fail;
+				}
+			}
+
+			ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec));
+			if (ecode != NTDB_SUCCESS) {
+				goto fail;
+			}
+
+			/* Bottom bits must match header. */
+			if ((h & ((1 << 11)-1)) != rec_hash(&rec)) {
+				ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						   NTDB_LOG_ERROR,
+						   "ntdb_check: Bad hash magic"
+						   " at offset %llu"
+						   " (0x%llx vs 0x%llx)",
+						   (long long)off,
+						   (long long)h,
+						   (long long)rec_hash(&rec));
+				goto fail;
+			}
+
+		check:
+			if (check) {
+				NTDB_DATA k, d;
+				const unsigned char *kptr;
+
+				kptr = ntdb_access_read(ntdb,
+						       off + sizeof(rec),
+						       rec_key_length(&rec)
+						       + rec_data_length(&rec),
+						       false);
+				if (NTDB_PTR_IS_ERR(kptr)) {
+					ecode = NTDB_PTR_ERR(kptr);
+					goto fail;
+				}
+
+				k = ntdb_mkdata(kptr, rec_key_length(&rec));
+				d = ntdb_mkdata(kptr + k.dsize,
+					       rec_data_length(&rec));
+				ecode = check(k, d, data);
+				ntdb_access_release(ntdb, kptr);
+				if (ecode != NTDB_SUCCESS) {
+					goto fail;
+				}
+			}
+		}
+	}
+	ntdb_access_release(ntdb, hash);
+	return NTDB_SUCCESS;
+
+fail:
+	ntdb_access_release(ntdb, hash);
+	return ecode;
+}
+
+static enum NTDB_ERROR check_hash(struct ntdb_context *ntdb,
+				 ntdb_off_t used[],
+				 size_t num_used, size_t num_other_used,
+				 enum NTDB_ERROR (*check)(NTDB_DATA, NTDB_DATA, void *),
+				 void *data)
+{
+	/* Free tables and capabilities also show up as used. */
+	size_t num_found = num_other_used;
+	enum NTDB_ERROR ecode;
+
+	ecode = check_hash_tree(ntdb, offsetof(struct ntdb_header, hashtable),
+				NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS,
+				0, 0, used, num_used, &num_found,
+				check, data);
+	if (ecode == NTDB_SUCCESS) {
+		if (num_found != num_used) {
+			ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+					   "ntdb_check: Not all entries"
+					   " are in hash");
+		}
+	}
+	return ecode;
+}
+
+static enum NTDB_ERROR check_free(struct ntdb_context *ntdb,
+				 ntdb_off_t off,
+				 const struct ntdb_free_record *frec,
+				 ntdb_off_t prev, unsigned int ftable,
+				 unsigned int bucket)
+{
+	enum NTDB_ERROR ecode;
+
+	if (frec_magic(frec) != NTDB_FREE_MAGIC) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: offset %llu bad magic 0x%llx",
+				  (long long)off,
+				  (long long)frec->magic_and_prev);
+	}
+	if (frec_ftable(frec) != ftable) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: offset %llu bad freetable %u",
+				  (long long)off, frec_ftable(frec));
+
+	}
+
+	ecode = ntdb->io->oob(ntdb, off,
+			     frec_len(frec)
+			     + sizeof(struct ntdb_used_record),
+			     false);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+	if (size_to_bucket(frec_len(frec)) != bucket) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: offset %llu in wrong bucket"
+				  " (%u vs %u)",
+				  (long long)off,
+				  bucket, size_to_bucket(frec_len(frec)));
+	}
+	if (prev && prev != frec_prev(frec)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: offset %llu bad prev"
+				  " (%llu vs %llu)",
+				  (long long)off,
+				  (long long)prev, (long long)frec_len(frec));
+	}
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR check_free_table(struct ntdb_context *ntdb,
+				       ntdb_off_t ftable_off,
+				       unsigned ftable_num,
+				       ntdb_off_t fr[],
+				       size_t num_free,
+				       size_t *num_found)
+{
+	struct ntdb_freetable ft;
+	ntdb_off_t h;
+	unsigned int i;
+	enum NTDB_ERROR ecode;
+
+	ecode = ntdb_read_convert(ntdb, ftable_off, &ft, sizeof(ft));
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (rec_magic(&ft.hdr) != NTDB_FTABLE_MAGIC
+	    || rec_key_length(&ft.hdr) != 0
+	    || rec_data_length(&ft.hdr) != sizeof(ft) - sizeof(ft.hdr)
+	    || rec_hash(&ft.hdr) != 0) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: Invalid header on free table");
+	}
+
+	for (i = 0; i < NTDB_FREE_BUCKETS; i++) {
+		ntdb_off_t off, prev = 0, *p, first = 0;
+		struct ntdb_free_record f;
+
+		h = bucket_off(ftable_off, i);
+		for (off = ntdb_read_off(ntdb, h); off; off = f.next) {
+			if (NTDB_OFF_IS_ERR(off)) {
+				return NTDB_OFF_TO_ERR(off);
+			}
+			if (!first) {
+				off &= NTDB_OFF_MASK;
+				first = off;
+			}
+			ecode = ntdb_read_convert(ntdb, off, &f, sizeof(f));
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+			ecode = check_free(ntdb, off, &f, prev, ftable_num, i);
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+
+			/* FIXME: Check hash bits */
+			p = asearch(&off, fr, num_free, off_cmp);
+			if (!p) {
+				return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: Invalid offset"
+						  " %llu in free table",
+						  (long long)off);
+			}
+			/* Mark it invalid. */
+			*p ^= 1;
+			(*num_found)++;
+			prev = off;
+		}
+
+		if (first) {
+			/* Now we can check first back pointer. */
+			ecode = ntdb_read_convert(ntdb, first, &f, sizeof(f));
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+			ecode = check_free(ntdb, first, &f, prev, ftable_num, i);
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+		}
+	}
+	return NTDB_SUCCESS;
+}
+
+/* Slow, but should be very rare. */
+ntdb_off_t dead_space(struct ntdb_context *ntdb, ntdb_off_t off)
+{
+	size_t len;
+	enum NTDB_ERROR ecode;
+
+	for (len = 0; off + len < ntdb->file->map_size; len++) {
+		char c;
+		ecode = ntdb->io->tread(ntdb, off, &c, 1);
+		if (ecode != NTDB_SUCCESS) {
+			return NTDB_ERR_TO_OFF(ecode);
+		}
+		if (c != 0 && c != 0x43)
+			break;
+	}
+	return len;
+}
+
+static enum NTDB_ERROR check_linear(struct ntdb_context *ntdb,
+				   ntdb_off_t **used, size_t *num_used,
+				   ntdb_off_t **fr, size_t *num_free,
+				   uint64_t features, ntdb_off_t recovery)
+{
+	ntdb_off_t off;
+	ntdb_len_t len;
+	enum NTDB_ERROR ecode;
+	bool found_recovery = false;
+
+	for (off = sizeof(struct ntdb_header);
+	     off < ntdb->file->map_size;
+	     off += len) {
+		union {
+			struct ntdb_used_record u;
+			struct ntdb_free_record f;
+			struct ntdb_recovery_record r;
+		} rec;
+		/* r is larger: only get that if we need to. */
+		ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec.f));
+		if (ecode != NTDB_SUCCESS) {
+			return ecode;
+		}
+
+		/* If we crash after ftruncate, we can get zeroes or fill. */
+		if (rec.r.magic == NTDB_RECOVERY_INVALID_MAGIC
+		    || rec.r.magic ==  0x4343434343434343ULL) {
+			ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec.r));
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+			if (recovery == off) {
+				found_recovery = true;
+				len = sizeof(rec.r) + rec.r.max_len;
+			} else {
+				len = dead_space(ntdb, off);
+				if (NTDB_OFF_IS_ERR(len)) {
+					return NTDB_OFF_TO_ERR(len);
+				}
+				if (len < sizeof(rec.r)) {
+					return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+							  NTDB_LOG_ERROR,
+							  "ntdb_check: invalid"
+							  " dead space at %zu",
+							  (size_t)off);
+				}
+
+				ntdb_logerr(ntdb, NTDB_SUCCESS, NTDB_LOG_WARNING,
+					   "Dead space at %zu-%zu (of %zu)",
+					   (size_t)off, (size_t)(off + len),
+					   (size_t)ntdb->file->map_size);
+			}
+		} else if (rec.r.magic == NTDB_RECOVERY_MAGIC) {
+			ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec.r));
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+			if (recovery != off) {
+				return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: unexpected"
+						  " recovery record at offset"
+						  " %zu",
+						  (size_t)off);
+			}
+			if (rec.r.len > rec.r.max_len) {
+				return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: invalid recovery"
+						  " length %zu",
+						  (size_t)rec.r.len);
+			}
+			if (rec.r.eof > ntdb->file->map_size) {
+				return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: invalid old EOF"
+						  " %zu", (size_t)rec.r.eof);
+			}
+			found_recovery = true;
+			len = sizeof(rec.r) + rec.r.max_len;
+		} else if (frec_magic(&rec.f) == NTDB_FREE_MAGIC) {
+			len = sizeof(rec.u) + frec_len(&rec.f);
+			if (off + len > ntdb->file->map_size) {
+				return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: free overlength"
+						  " %llu at offset %llu",
+						  (long long)len,
+						  (long long)off);
+			}
+			/* This record should be in free lists. */
+			if (frec_ftable(&rec.f) != NTDB_FTABLE_NONE
+			    && !append(fr, num_free, off)) {
+				return ntdb_logerr(ntdb, NTDB_ERR_OOM,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: tracking %zu'th"
+						  " free record.", *num_free);
+			}
+		} else if (rec_magic(&rec.u) == NTDB_USED_MAGIC
+			   || rec_magic(&rec.u) == NTDB_CHAIN_MAGIC
+			   || rec_magic(&rec.u) == NTDB_HTABLE_MAGIC
+			   || rec_magic(&rec.u) == NTDB_FTABLE_MAGIC
+			   || rec_magic(&rec.u) == NTDB_CAP_MAGIC) {
+			uint64_t klen, dlen, extra;
+
+			/* This record is used! */
+			if (!append(used, num_used, off)) {
+				return ntdb_logerr(ntdb, NTDB_ERR_OOM,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: tracking %zu'th"
+						  " used record.", *num_used);
+			}
+
+			klen = rec_key_length(&rec.u);
+			dlen = rec_data_length(&rec.u);
+			extra = rec_extra_padding(&rec.u);
+
+			len = sizeof(rec.u) + klen + dlen + extra;
+			if (off + len > ntdb->file->map_size) {
+				return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: used overlength"
+						  " %llu at offset %llu",
+						  (long long)len,
+						  (long long)off);
+			}
+
+			if (len < sizeof(rec.f)) {
+				return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						  NTDB_LOG_ERROR,
+						  "ntdb_check: too short record"
+						  " %llu at %llu",
+						  (long long)len,
+						  (long long)off);
+			}
+
+			/* Check that records have correct 0 at end (but may
+			 * not in future). */
+			if (extra && !features
+			    && rec_magic(&rec.u) != NTDB_CAP_MAGIC) {
+				const char *p;
+				char c;
+				p = ntdb_access_read(ntdb, off + sizeof(rec.u)
+						    + klen + dlen, 1, false);
+				if (NTDB_PTR_IS_ERR(p))
+					return NTDB_PTR_ERR(p);
+				c = *p;
+				ntdb_access_release(ntdb, p);
+
+				if (c != '\0') {
+					return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+							  NTDB_LOG_ERROR,
+							  "ntdb_check:"
+							  " non-zero extra"
+							  " at %llu",
+							  (long long)off);
+				}
+			}
+		} else {
+			return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+					  NTDB_LOG_ERROR,
+					  "ntdb_check: Bad magic 0x%llx"
+					  " at offset %zu",
+					  (long long)rec_magic(&rec.u),
+					  (size_t)off);
+		}
+	}
+
+	/* We must have found recovery area if there was one. */
+	if (recovery != 0 && !found_recovery) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "ntdb_check: expected a recovery area at %zu",
+				  (size_t)recovery);
+	}
+
+	return NTDB_SUCCESS;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_check_(struct ntdb_context *ntdb,
+			  enum NTDB_ERROR (*check)(NTDB_DATA, NTDB_DATA, void *),
+			  void *data)
+{
+	ntdb_off_t *fr = NULL, *used = NULL, ft, recovery;
+	size_t num_free = 0, num_used = 0, num_found = 0, num_ftables = 0,
+		num_capabilities = 0;
+	uint64_t features;
+	enum NTDB_ERROR ecode;
+
+	if (ntdb->flags & NTDB_CANT_CHECK) {
+		return ntdb_logerr(ntdb, NTDB_SUCCESS, NTDB_LOG_WARNING,
+				  "ntdb_check: database has unknown capability,"
+				  " cannot check.");
+	}
+
+	ecode = ntdb_allrecord_lock(ntdb, F_RDLCK, NTDB_LOCK_WAIT, false);
+	if (ecode != NTDB_SUCCESS) {
+		return ntdb->last_error = ecode;
+	}
+
+	ecode = ntdb_lock_expand(ntdb, F_RDLCK);
+	if (ecode != NTDB_SUCCESS) {
+		ntdb_allrecord_unlock(ntdb, F_RDLCK);
+		return ntdb->last_error = ecode;
+	}
+
+	ecode = check_header(ntdb, &recovery, &features, &num_capabilities);
+	if (ecode != NTDB_SUCCESS)
+		goto out;
+
+	/* First we do a linear scan, checking all records. */
+	ecode = check_linear(ntdb, &used, &num_used, &fr, &num_free, features,
+			     recovery);
+	if (ecode != NTDB_SUCCESS)
+		goto out;
+
+	for (ft = first_ftable(ntdb); ft; ft = next_ftable(ntdb, ft)) {
+		if (NTDB_OFF_IS_ERR(ft)) {
+			ecode = NTDB_OFF_TO_ERR(ft);
+			goto out;
+		}
+		ecode = check_free_table(ntdb, ft, num_ftables, fr, num_free,
+					 &num_found);
+		if (ecode != NTDB_SUCCESS)
+			goto out;
+		num_ftables++;
+	}
+
+	/* FIXME: Check key uniqueness? */
+	ecode = check_hash(ntdb, used, num_used, num_ftables + num_capabilities,
+			   check, data);
+	if (ecode != NTDB_SUCCESS)
+		goto out;
+
+	if (num_found != num_free) {
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				   "ntdb_check: Not all entries are in"
+				   " free table");
+	}
+
+out:
+	ntdb_allrecord_unlock(ntdb, F_RDLCK);
+	ntdb_unlock_expand(ntdb, F_RDLCK);
+	free(fr);
+	free(used);
+	return ntdb->last_error = ecode;
+}

lib/ntdb/doc/TDB_porting.txt

@@ -0,0 +1,65 @@
+Interface differences between TDB and NTDB.
+
+- ntdb shares 'struct TDB_DATA' with tdb, but TDB defines the TDB_DATA
+  typedef, whereas ntdb defines NTDB_DATA (ie. both are compatible).
+  If you include both ntdb.h and tdb.h, #include tdb.h first,
+  otherwise you'll get a compile error when tdb.h re-defined struct
+  TDB_DATA.
+
+- ntdb functions return NTDB_SUCCESS (ie 0) on success, and a negative
+  error on failure, whereas tdb functions returned 0 on success, and
+  -1 on failure.  tdb then used tdb_error() to determine the error;
+  this is also supported in ntdb to ease backwards compatibility,
+  though the other form is preferred.
+
+- ntdb's ntdb_fetch() returns an error, tdb's returned the data directly
+  (or tdb_null, and you were supposed to check tdb_error() to find out why).
+
+- ntdb's ntdb_nextkey() frees the old key's dptr, in tdb you needed to do
+  this manually.
+
+- tdb's tdb_open/tdb_open_ex took an explicit hash size.  ntdb's hash table
+  resizes as required.
+
+- ntdb uses a linked list of attribute structures to implement logging and
+  alternate hashes.  tdb used tdb_open_ex, which was not extensible.
+
+- ntdb does locking on read-only databases (ie. O_RDONLY passed to ntdb_open).
+  tdb did not: use the NTDB_NOLOCK flag if you want to suppress locking.
+
+- ntdb's log function is simpler than tdb's log function.  The string is
+  already formatted, and it takes an enum ntdb_log_level not a tdb_debug_level,
+  and which has only three values: NTDB_LOG_ERROR, NTDB_LOG_USE_ERROR and
+  NTDB_LOG_WARNING.
+
+- ntdb provides ntdb_deq() for comparing two NTDB_DATA, and ntdb_mkdata() for
+  creating an NTDB_DATA.
+
+- ntdb's ntdb_name() returns a copy of the name even for NTDB_INTERNAL dbs.
+
+- ntdb does not need tdb_reopen() or tdb_reopen_all().  If you call
+  fork() after during certain operations the child should close the
+  tdb, or complete the operations before continuing to use the tdb:
+
+	ntdb_transaction_start(): child must ntdb_transaction_cancel()
+	ntdb_lockall(): child must call ntdb_unlockall()
+	ntdb_lockall_read(): child must call ntdb_unlockall_read()
+	ntdb_chainlock(): child must call ntdb_chainunlock()
+	ntdb_parse() callback: child must return from ntdb_parse()
+
+- ntdb will not open a non-tdb file, even if O_CREAT is specified.
+
+- There is no ntdb_traverse_read.  For operating on TDB files, you can
+  simulate it by ntdb_add_flag(tdb, NTDB_RDONLY); ntdb_traverse();
+  ntdb_remove_flag(tdb, NTDB_RDONLY).  This may be desirable because
+  traverse on TDB files use a write lock on the entire database
+  unless it's read-only.
+
+- Failure inside a transaction (such as a lock function failing) does
+  not implicitly cancel the transaction; you still need to call
+  ntdb_transaction_cancel().
+
+- There is no NTDB_CLEAR_IF_FIRST flag; it has severe scalability and
+  API problems.  If necessary, you can emulate this by using the open
+  hook and placing a 1-byte lock at offset 4.  If your program forks,
+  you will need to place this lock again in the child.

lib/ntdb/free.c

@@ -0,0 +1,976 @@
+ /*
+   Trivial Database 2: free list/block handling
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "private.h"
+#include <ccan/likely/likely.h>
+#include <ccan/ilog/ilog.h>
+#include <time.h>
+#include <assert.h>
+#include <limits.h>
+
+static unsigned fls64(uint64_t val)
+{
+	return ilog64(val);
+}
+
+/* In which bucket would we find a particular record size? (ignoring header) */
+unsigned int size_to_bucket(ntdb_len_t data_len)
+{
+	unsigned int bucket;
+
+	/* We can't have records smaller than this. */
+	assert(data_len >= NTDB_MIN_DATA_LEN);
+
+	/* Ignoring the header... */
+	if (data_len - NTDB_MIN_DATA_LEN <= 64) {
+		/* 0 in bucket 0, 8 in bucket 1... 64 in bucket 8. */
+		bucket = (data_len - NTDB_MIN_DATA_LEN) / 8;
+	} else {
+		/* After that we go power of 2. */
+		bucket = fls64(data_len - NTDB_MIN_DATA_LEN) + 2;
+	}
+
+	if (unlikely(bucket >= NTDB_FREE_BUCKETS))
+		bucket = NTDB_FREE_BUCKETS - 1;
+	return bucket;
+}
+
+ntdb_off_t first_ftable(struct ntdb_context *ntdb)
+{
+	return ntdb_read_off(ntdb, offsetof(struct ntdb_header, free_table));
+}
+
+ntdb_off_t next_ftable(struct ntdb_context *ntdb, ntdb_off_t ftable)
+{
+	return ntdb_read_off(ntdb, ftable + offsetof(struct ntdb_freetable,next));
+}
+
+enum NTDB_ERROR ntdb_ftable_init(struct ntdb_context *ntdb)
+{
+	/* Use reservoir sampling algorithm to select a free list at random. */
+	unsigned int rnd, max = 0, count = 0;
+	ntdb_off_t off;
+
+	ntdb->ftable_off = off = first_ftable(ntdb);
+	ntdb->ftable = 0;
+
+	while (off) {
+		if (NTDB_OFF_IS_ERR(off)) {
+			return NTDB_OFF_TO_ERR(off);
+		}
+
+		rnd = random();
+		if (rnd >= max) {
+			ntdb->ftable_off = off;
+			ntdb->ftable = count;
+			max = rnd;
+		}
+
+		off = next_ftable(ntdb, off);
+		count++;
+	}
+	return NTDB_SUCCESS;
+}
+
+/* Offset of a given bucket. */
+ntdb_off_t bucket_off(ntdb_off_t ftable_off, unsigned bucket)
+{
+	return ftable_off + offsetof(struct ntdb_freetable, buckets)
+		+ bucket * sizeof(ntdb_off_t);
+}
+
+/* Returns free_buckets + 1, or list number to search, or -ve error. */
+static ntdb_off_t find_free_head(struct ntdb_context *ntdb,
+				ntdb_off_t ftable_off,
+				ntdb_off_t bucket)
+{
+	/* Speculatively search for a non-zero bucket. */
+	return ntdb_find_nonzero_off(ntdb, bucket_off(ftable_off, 0),
+				    bucket, NTDB_FREE_BUCKETS);
+}
+
+static void check_list(struct ntdb_context *ntdb, ntdb_off_t b_off)
+{
+#ifdef CCAN_NTDB_DEBUG
+	ntdb_off_t off, prev = 0, first;
+	struct ntdb_free_record r;
+
+	first = off = (ntdb_read_off(ntdb, b_off) & NTDB_OFF_MASK);
+	while (off != 0) {
+		ntdb_read_convert(ntdb, off, &r, sizeof(r));
+		if (frec_magic(&r) != NTDB_FREE_MAGIC)
+			abort();
+		if (prev && frec_prev(&r) != prev)
+			abort();
+		prev = off;
+		off = r.next;
+	}
+
+	if (first) {
+		ntdb_read_convert(ntdb, first, &r, sizeof(r));
+		if (frec_prev(&r) != prev)
+			abort();
+	}
+#endif
+}
+
+/* Remove from free bucket. */
+static enum NTDB_ERROR remove_from_list(struct ntdb_context *ntdb,
+				       ntdb_off_t b_off, ntdb_off_t r_off,
+				       const struct ntdb_free_record *r)
+{
+	ntdb_off_t off, prev_next, head;
+	enum NTDB_ERROR ecode;
+
+	/* Is this only element in list?  Zero out bucket, and we're done. */
+	if (frec_prev(r) == r_off)
+		return ntdb_write_off(ntdb, b_off, 0);
+
+	/* off = &r->prev->next */
+	off = frec_prev(r) + offsetof(struct ntdb_free_record, next);
+
+	/* Get prev->next */
+	prev_next = ntdb_read_off(ntdb, off);
+	if (NTDB_OFF_IS_ERR(prev_next))
+		return NTDB_OFF_TO_ERR(prev_next);
+
+	/* If prev->next == 0, we were head: update bucket to point to next. */
+	if (prev_next == 0) {
+		/* We must preserve upper bits. */
+		head = ntdb_read_off(ntdb, b_off);
+		if (NTDB_OFF_IS_ERR(head))
+			return NTDB_OFF_TO_ERR(head);
+
+		if ((head & NTDB_OFF_MASK) != r_off) {
+			return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+					  "remove_from_list:"
+					  " %llu head %llu on list %llu",
+					  (long long)r_off,
+					  (long long)head,
+					  (long long)b_off);
+		}
+		head = ((head & ~NTDB_OFF_MASK) | r->next);
+		ecode = ntdb_write_off(ntdb, b_off, head);
+		if (ecode != NTDB_SUCCESS)
+			return ecode;
+	} else {
+		/* r->prev->next = r->next */
+		ecode = ntdb_write_off(ntdb, off, r->next);
+		if (ecode != NTDB_SUCCESS)
+			return ecode;
+	}
+
+	/* If we were the tail, off = &head->prev. */
+	if (r->next == 0) {
+		head = ntdb_read_off(ntdb, b_off);
+		if (NTDB_OFF_IS_ERR(head))
+			return NTDB_OFF_TO_ERR(head);
+		head &= NTDB_OFF_MASK;
+		off = head + offsetof(struct ntdb_free_record, magic_and_prev);
+	} else {
+		/* off = &r->next->prev */
+		off = r->next + offsetof(struct ntdb_free_record,
+					 magic_and_prev);
+	}
+
+#ifdef CCAN_NTDB_DEBUG
+	/* *off == r */
+	if ((ntdb_read_off(ntdb, off) & NTDB_OFF_MASK) != r_off) {
+		return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				  "remove_from_list:"
+				  " %llu bad prev in list %llu",
+				  (long long)r_off, (long long)b_off);
+	}
+#endif
+	/* r->next->prev = r->prev */
+	return ntdb_write_off(ntdb, off, r->magic_and_prev);
+}
+
+/* Enqueue in this free bucket: sets coalesce if we've added 128
+ * entries to it. */
+static enum NTDB_ERROR enqueue_in_free(struct ntdb_context *ntdb,
+				      ntdb_off_t b_off,
+				      ntdb_off_t off,
+				      ntdb_len_t len,
+				      bool *coalesce)
+{
+	struct ntdb_free_record new;
+	enum NTDB_ERROR ecode;
+	ntdb_off_t prev, head;
+	uint64_t magic = (NTDB_FREE_MAGIC << (64 - NTDB_OFF_UPPER_STEAL));
+
+	head = ntdb_read_off(ntdb, b_off);
+	if (NTDB_OFF_IS_ERR(head))
+		return NTDB_OFF_TO_ERR(head);
+
+	/* We only need to set ftable_and_len; rest is set in enqueue_in_free */
+	new.ftable_and_len = ((uint64_t)ntdb->ftable
+			      << (64 - NTDB_OFF_UPPER_STEAL))
+		| len;
+
+	/* new->next = head. */
+	new.next = (head & NTDB_OFF_MASK);
+
+	/* First element?  Prev points to ourselves. */
+	if (!new.next) {
+		new.magic_and_prev = (magic | off);
+	} else {
+		/* new->prev = next->prev */
+		prev = ntdb_read_off(ntdb,
+				    new.next + offsetof(struct ntdb_free_record,
+							magic_and_prev));
+		new.magic_and_prev = prev;
+		if (frec_magic(&new) != NTDB_FREE_MAGIC) {
+			return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+					  "enqueue_in_free: %llu bad head"
+					  " prev %llu",
+					  (long long)new.next,
+					  (long long)prev);
+		}
+		/* next->prev = new. */
+		ecode = ntdb_write_off(ntdb, new.next
+				      + offsetof(struct ntdb_free_record,
+						 magic_and_prev),
+				      off | magic);
+		if (ecode != NTDB_SUCCESS) {
+			return ecode;
+		}
+
+#ifdef CCAN_NTDB_DEBUG
+		prev = ntdb_read_off(ntdb, frec_prev(&new)
+				    + offsetof(struct ntdb_free_record, next));
+		if (prev != 0) {
+			return ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+					  "enqueue_in_free:"
+					  " %llu bad tail next ptr %llu",
+					  (long long)frec_prev(&new)
+					  + offsetof(struct ntdb_free_record,
+						     next),
+					  (long long)prev);
+		}
+#endif
+	}
+
+	/* Update enqueue count, but don't set high bit: see NTDB_OFF_IS_ERR */
+	if (*coalesce)
+		head += (1ULL << (64 - NTDB_OFF_UPPER_STEAL));
+	head &= ~(NTDB_OFF_MASK | (1ULL << 63));
+	head |= off;
+
+	ecode = ntdb_write_off(ntdb, b_off, head);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	/* It's time to coalesce if counter wrapped. */
+	if (*coalesce)
+		*coalesce = ((head & ~NTDB_OFF_MASK) == 0);
+
+	return ntdb_write_convert(ntdb, off, &new, sizeof(new));
+}
+
+static ntdb_off_t ftable_offset(struct ntdb_context *ntdb, unsigned int ftable)
+{
+	ntdb_off_t off;
+	unsigned int i;
+
+	if (likely(ntdb->ftable == ftable))
+		return ntdb->ftable_off;
+
+	off = first_ftable(ntdb);
+	for (i = 0; i < ftable; i++) {
+		if (NTDB_OFF_IS_ERR(off)) {
+			break;
+		}
+		off = next_ftable(ntdb, off);
+	}
+	return off;
+}
+
+/* Note: we unlock the current bucket if fail (-ve), or coalesce (+ve) and
+ * need to blatt the *protect record (which is set to an error). */
+static ntdb_len_t coalesce(struct ntdb_context *ntdb,
+			  ntdb_off_t off, ntdb_off_t b_off,
+			  ntdb_len_t data_len,
+			  ntdb_off_t *protect)
+{
+	ntdb_off_t end;
+	struct ntdb_free_record rec;
+	enum NTDB_ERROR ecode;
+
+	ntdb->stats.alloc_coalesce_tried++;
+	end = off + sizeof(struct ntdb_used_record) + data_len;
+
+	while (end < ntdb->file->map_size) {
+		const struct ntdb_free_record *r;
+		ntdb_off_t nb_off;
+		unsigned ftable, bucket;
+
+		r = ntdb_access_read(ntdb, end, sizeof(*r), true);
+		if (NTDB_PTR_IS_ERR(r)) {
+			ecode = NTDB_PTR_ERR(r);
+			goto err;
+		}
+
+		if (frec_magic(r) != NTDB_FREE_MAGIC
+		    || frec_ftable(r) == NTDB_FTABLE_NONE) {
+			ntdb_access_release(ntdb, r);
+			break;
+		}
+
+		ftable = frec_ftable(r);
+		bucket = size_to_bucket(frec_len(r));
+		nb_off = ftable_offset(ntdb, ftable);
+		if (NTDB_OFF_IS_ERR(nb_off)) {
+			ntdb_access_release(ntdb, r);
+			ecode = NTDB_OFF_TO_ERR(nb_off);
+			goto err;
+		}
+		nb_off = bucket_off(nb_off, bucket);
+		ntdb_access_release(ntdb, r);
+
+		/* We may be violating lock order here, so best effort. */
+		if (ntdb_lock_free_bucket(ntdb, nb_off, NTDB_LOCK_NOWAIT)
+		    != NTDB_SUCCESS) {
+			ntdb->stats.alloc_coalesce_lockfail++;
+			break;
+		}
+
+		/* Now we have lock, re-check. */
+		ecode = ntdb_read_convert(ntdb, end, &rec, sizeof(rec));
+		if (ecode != NTDB_SUCCESS) {
+			ntdb_unlock_free_bucket(ntdb, nb_off);
+			goto err;
+		}
+
+		if (unlikely(frec_magic(&rec) != NTDB_FREE_MAGIC)) {
+			ntdb->stats.alloc_coalesce_race++;
+			ntdb_unlock_free_bucket(ntdb, nb_off);
+			break;
+		}
+
+		if (unlikely(frec_ftable(&rec) != ftable)
+		    || unlikely(size_to_bucket(frec_len(&rec)) != bucket)) {
+			ntdb->stats.alloc_coalesce_race++;
+			ntdb_unlock_free_bucket(ntdb, nb_off);
+			break;
+		}
+
+		/* Did we just mess up a record you were hoping to use? */
+		if (end == *protect) {
+			ntdb->stats.alloc_coalesce_iterate_clash++;
+			*protect = NTDB_ERR_TO_OFF(NTDB_ERR_NOEXIST);
+		}
+
+		ecode = remove_from_list(ntdb, nb_off, end, &rec);
+		check_list(ntdb, nb_off);
+		if (ecode != NTDB_SUCCESS) {
+			ntdb_unlock_free_bucket(ntdb, nb_off);
+			goto err;
+		}
+
+		end += sizeof(struct ntdb_used_record) + frec_len(&rec);
+		ntdb_unlock_free_bucket(ntdb, nb_off);
+		ntdb->stats.alloc_coalesce_num_merged++;
+	}
+
+	/* Didn't find any adjacent free? */
+	if (end == off + sizeof(struct ntdb_used_record) + data_len)
+		return 0;
+
+	/* Before we expand, check this isn't one you wanted protected? */
+	if (off == *protect) {
+		*protect = NTDB_ERR_TO_OFF(NTDB_ERR_EXISTS);
+		ntdb->stats.alloc_coalesce_iterate_clash++;
+	}
+
+	/* OK, expand initial record */
+	ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec));
+	if (ecode != NTDB_SUCCESS) {
+		goto err;
+	}
+
+	if (frec_len(&rec) != data_len) {
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+				   "coalesce: expected data len %zu not %zu",
+				   (size_t)data_len, (size_t)frec_len(&rec));
+		goto err;
+	}
+
+	ecode = remove_from_list(ntdb, b_off, off, &rec);
+	check_list(ntdb, b_off);
+	if (ecode != NTDB_SUCCESS) {
+		goto err;
+	}
+
+	/* Try locking violation first.  We don't allow coalesce recursion! */
+	ecode = add_free_record(ntdb, off, end - off, NTDB_LOCK_NOWAIT, false);
+	if (ecode != NTDB_SUCCESS) {
+		/* Need to drop lock.  Can't rely on anything stable. */
+		ntdb->stats.alloc_coalesce_lockfail++;
+		*protect = NTDB_ERR_TO_OFF(NTDB_ERR_CORRUPT);
+
+		/* We have to drop this to avoid deadlocks, so make sure record
+		 * doesn't get coalesced by someone else! */
+		rec.ftable_and_len = (NTDB_FTABLE_NONE
+				      << (64 - NTDB_OFF_UPPER_STEAL))
+			| (end - off - sizeof(struct ntdb_used_record));
+		ecode = ntdb_write_off(ntdb,
+				      off + offsetof(struct ntdb_free_record,
+						     ftable_and_len),
+				      rec.ftable_and_len);
+		if (ecode != NTDB_SUCCESS) {
+			goto err;
+		}
+
+		ntdb_unlock_free_bucket(ntdb, b_off);
+
+		ecode = add_free_record(ntdb, off, end - off, NTDB_LOCK_WAIT,
+					false);
+		if (ecode != NTDB_SUCCESS) {
+			return NTDB_ERR_TO_OFF(ecode);
+		}
+	} else if (NTDB_OFF_IS_ERR(*protect)) {
+		/* For simplicity, we always drop lock if they can't continue */
+		ntdb_unlock_free_bucket(ntdb, b_off);
+	}
+	ntdb->stats.alloc_coalesce_succeeded++;
+
+	/* Return usable length. */
+	return end - off - sizeof(struct ntdb_used_record);
+
+err:
+	/* To unify error paths, we *always* unlock bucket on error. */
+	ntdb_unlock_free_bucket(ntdb, b_off);
+	return NTDB_ERR_TO_OFF(ecode);
+}
+
+/* List is locked: we unlock it. */
+static enum NTDB_ERROR coalesce_list(struct ntdb_context *ntdb,
+				    ntdb_off_t ftable_off,
+				    ntdb_off_t b_off,
+				    unsigned int limit)
+{
+	enum NTDB_ERROR ecode;
+	ntdb_off_t off;
+
+	off = ntdb_read_off(ntdb, b_off);
+	if (NTDB_OFF_IS_ERR(off)) {
+		ecode = NTDB_OFF_TO_ERR(off);
+		goto unlock_err;
+	}
+	/* A little bit of paranoia: counter should be 0. */
+	off &= NTDB_OFF_MASK;
+
+	while (off && limit--) {
+		struct ntdb_free_record rec;
+		ntdb_len_t coal;
+		ntdb_off_t next;
+
+		ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		next = rec.next;
+		coal = coalesce(ntdb, off, b_off, frec_len(&rec), &next);
+		if (NTDB_OFF_IS_ERR(coal)) {
+			/* This has already unlocked on error. */
+			return NTDB_OFF_TO_ERR(coal);
+		}
+		if (NTDB_OFF_IS_ERR(next)) {
+			/* Coalescing had to unlock, so stop. */
+			return NTDB_SUCCESS;
+		}
+		/* Keep going if we're doing well... */
+		limit += size_to_bucket(coal / 16 + NTDB_MIN_DATA_LEN);
+		off = next;
+	}
+
+	/* Now, move those elements to the tail of the list so we get something
+	 * else next time. */
+	if (off) {
+		struct ntdb_free_record oldhrec, newhrec, oldtrec, newtrec;
+		ntdb_off_t oldhoff, oldtoff, newtoff;
+
+		/* The record we were up to is the new head. */
+		ecode = ntdb_read_convert(ntdb, off, &newhrec, sizeof(newhrec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		/* Get the new tail. */
+		newtoff = frec_prev(&newhrec);
+		ecode = ntdb_read_convert(ntdb, newtoff, &newtrec,
+					 sizeof(newtrec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		/* Get the old head. */
+		oldhoff = ntdb_read_off(ntdb, b_off);
+		if (NTDB_OFF_IS_ERR(oldhoff)) {
+			ecode = NTDB_OFF_TO_ERR(oldhoff);
+			goto unlock_err;
+		}
+
+		/* This could happen if they all coalesced away. */
+		if (oldhoff == off)
+			goto out;
+
+		ecode = ntdb_read_convert(ntdb, oldhoff, &oldhrec,
+					 sizeof(oldhrec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		/* Get the old tail. */
+		oldtoff = frec_prev(&oldhrec);
+		ecode = ntdb_read_convert(ntdb, oldtoff, &oldtrec,
+					 sizeof(oldtrec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		/* Old tail's next points to old head. */
+		oldtrec.next = oldhoff;
+
+		/* Old head's prev points to old tail. */
+		oldhrec.magic_and_prev
+			= (NTDB_FREE_MAGIC << (64 - NTDB_OFF_UPPER_STEAL))
+			| oldtoff;
+
+		/* New tail's next is 0. */
+		newtrec.next = 0;
+
+		/* Write out the modified versions. */
+		ecode = ntdb_write_convert(ntdb, oldtoff, &oldtrec,
+					  sizeof(oldtrec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		ecode = ntdb_write_convert(ntdb, oldhoff, &oldhrec,
+					  sizeof(oldhrec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		ecode = ntdb_write_convert(ntdb, newtoff, &newtrec,
+					  sizeof(newtrec));
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+
+		/* And finally link in new head. */
+		ecode = ntdb_write_off(ntdb, b_off, off);
+		if (ecode != NTDB_SUCCESS)
+			goto unlock_err;
+	}
+out:
+	ntdb_unlock_free_bucket(ntdb, b_off);
+	return NTDB_SUCCESS;
+
+unlock_err:
+	ntdb_unlock_free_bucket(ntdb, b_off);
+	return ecode;
+}
+
+/* List must not be locked if coalesce_ok is set. */
+enum NTDB_ERROR add_free_record(struct ntdb_context *ntdb,
+			       ntdb_off_t off, ntdb_len_t len_with_header,
+			       enum ntdb_lock_flags waitflag,
+			       bool coalesce_ok)
+{
+	ntdb_off_t b_off;
+	ntdb_len_t len;
+	enum NTDB_ERROR ecode;
+
+	assert(len_with_header >= sizeof(struct ntdb_free_record));
+
+	len = len_with_header - sizeof(struct ntdb_used_record);
+
+	b_off = bucket_off(ntdb->ftable_off, size_to_bucket(len));
+	ecode = ntdb_lock_free_bucket(ntdb, b_off, waitflag);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	ecode = enqueue_in_free(ntdb, b_off, off, len, &coalesce_ok);
+	check_list(ntdb, b_off);
+
+	/* Coalescing unlocks free list. */
+	if (!ecode && coalesce_ok)
+		ecode = coalesce_list(ntdb, ntdb->ftable_off, b_off, 2);
+	else
+		ntdb_unlock_free_bucket(ntdb, b_off);
+	return ecode;
+}
+
+static size_t adjust_size(size_t keylen, size_t datalen)
+{
+	size_t size = keylen + datalen;
+
+	if (size < NTDB_MIN_DATA_LEN)
+		size = NTDB_MIN_DATA_LEN;
+
+	/* Round to next uint64_t boundary. */
+	return (size + (sizeof(uint64_t) - 1ULL)) & ~(sizeof(uint64_t) - 1ULL);
+}
+
+/* If we have enough left over to be useful, split that off. */
+static size_t record_leftover(size_t keylen, size_t datalen,
+			      bool want_extra, size_t total_len)
+{
+	ssize_t leftover;
+
+	if (want_extra)
+		datalen += datalen / 2;
+	leftover = total_len - adjust_size(keylen, datalen);
+
+	if (leftover < (ssize_t)sizeof(struct ntdb_free_record))
+		return 0;
+
+	return leftover;
+}
+
+/* We need size bytes to put our key and data in. */
+static ntdb_off_t lock_and_alloc(struct ntdb_context *ntdb,
+				ntdb_off_t ftable_off,
+				ntdb_off_t bucket,
+				size_t keylen, size_t datalen,
+				bool want_extra,
+				unsigned magic,
+				unsigned hashlow)
+{
+	ntdb_off_t off, b_off,best_off;
+	struct ntdb_free_record best = { 0 };
+	double multiplier;
+	size_t size = adjust_size(keylen, datalen);
+	enum NTDB_ERROR ecode;
+
+	ntdb->stats.allocs++;
+	b_off = bucket_off(ftable_off, bucket);
+
+	/* FIXME: Try non-blocking wait first, to measure contention. */
+	/* Lock this bucket. */
+	ecode = ntdb_lock_free_bucket(ntdb, b_off, NTDB_LOCK_WAIT);
+	if (ecode != NTDB_SUCCESS) {
+		return NTDB_ERR_TO_OFF(ecode);
+	}
+
+	best.ftable_and_len = -1ULL;
+	best_off = 0;
+
+	/* Get slack if we're after extra. */
+	if (want_extra)
+		multiplier = 1.5;
+	else
+		multiplier = 1.0;
+
+	/* Walk the list to see if any are large enough, getting less fussy
+	 * as we go. */
+	off = ntdb_read_off(ntdb, b_off);
+	if (NTDB_OFF_IS_ERR(off)) {
+		ecode = NTDB_OFF_TO_ERR(off);
+		goto unlock_err;
+	}
+	off &= NTDB_OFF_MASK;
+
+	while (off) {
+		const struct ntdb_free_record *r;
+		ntdb_len_t len;
+		ntdb_off_t next;
+
+		r = ntdb_access_read(ntdb, off, sizeof(*r), true);
+		if (NTDB_PTR_IS_ERR(r)) {
+			ecode = NTDB_PTR_ERR(r);
+			goto unlock_err;
+		}
+
+		if (frec_magic(r) != NTDB_FREE_MAGIC) {
+			ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT, NTDB_LOG_ERROR,
+					   "lock_and_alloc:"
+					   " %llu non-free 0x%llx",
+					   (long long)off,
+					   (long long)r->magic_and_prev);
+			ntdb_access_release(ntdb, r);
+			goto unlock_err;
+		}
+
+		if (frec_len(r) >= size && frec_len(r) < frec_len(&best)) {
+			best_off = off;
+			best = *r;
+		}
+
+		if (frec_len(&best) <= size * multiplier && best_off) {
+			ntdb_access_release(ntdb, r);
+			break;
+		}
+
+		multiplier *= 1.01;
+
+		next = r->next;
+		len = frec_len(r);
+		ntdb_access_release(ntdb, r);
+		off = next;
+	}
+
+	/* If we found anything at all, use it. */
+	if (best_off) {
+		struct ntdb_used_record rec;
+		size_t leftover;
+
+		/* We're happy with this size: take it. */
+		ecode = remove_from_list(ntdb, b_off, best_off, &best);
+		check_list(ntdb, b_off);
+		if (ecode != NTDB_SUCCESS) {
+			goto unlock_err;
+		}
+
+		leftover = record_leftover(keylen, datalen, want_extra,
+					   frec_len(&best));
+
+		assert(keylen + datalen + leftover <= frec_len(&best));
+		/* We need to mark non-free before we drop lock, otherwise
+		 * coalesce() could try to merge it! */
+		ecode = set_header(ntdb, &rec, magic, keylen, datalen,
+				   frec_len(&best) - leftover, hashlow);
+		if (ecode != NTDB_SUCCESS) {
+			goto unlock_err;
+		}
+
+		ecode = ntdb_write_convert(ntdb, best_off, &rec, sizeof(rec));
+		if (ecode != NTDB_SUCCESS) {
+			goto unlock_err;
+		}
+
+		/* For futureproofing, we put a 0 in any unused space. */
+		if (rec_extra_padding(&rec)) {
+			ecode = ntdb->io->twrite(ntdb, best_off + sizeof(rec)
+						+ keylen + datalen, "", 1);
+			if (ecode != NTDB_SUCCESS) {
+				goto unlock_err;
+			}
+		}
+
+		/* Bucket of leftover will be <= current bucket, so nested
+		 * locking is allowed. */
+		if (leftover) {
+			ntdb->stats.alloc_leftover++;
+			ecode = add_free_record(ntdb,
+						best_off + sizeof(rec)
+						+ frec_len(&best) - leftover,
+						leftover, NTDB_LOCK_WAIT, false);
+			if (ecode != NTDB_SUCCESS) {
+				best_off = NTDB_ERR_TO_OFF(ecode);
+			}
+		}
+		ntdb_unlock_free_bucket(ntdb, b_off);
+
+		return best_off;
+	}
+
+	ntdb_unlock_free_bucket(ntdb, b_off);
+	return 0;
+
+unlock_err:
+	ntdb_unlock_free_bucket(ntdb, b_off);
+	return NTDB_ERR_TO_OFF(ecode);
+}
+
+/* Get a free block from current free list, or 0 if none, -ve on error. */
+static ntdb_off_t get_free(struct ntdb_context *ntdb,
+			  size_t keylen, size_t datalen, bool want_extra,
+			  unsigned magic, unsigned hashlow)
+{
+	ntdb_off_t off, ftable_off;
+	ntdb_off_t start_b, b, ftable;
+	bool wrapped = false;
+
+	/* If they are growing, add 50% to get to higher bucket. */
+	if (want_extra)
+		start_b = size_to_bucket(adjust_size(keylen,
+						     datalen + datalen / 2));
+	else
+		start_b = size_to_bucket(adjust_size(keylen, datalen));
+
+	ftable_off = ntdb->ftable_off;
+	ftable = ntdb->ftable;
+	while (!wrapped || ftable_off != ntdb->ftable_off) {
+		/* Start at exact size bucket, and search up... */
+		for (b = find_free_head(ntdb, ftable_off, start_b);
+		     b < NTDB_FREE_BUCKETS;
+		     b = find_free_head(ntdb, ftable_off, b + 1)) {
+			/* Try getting one from list. */
+			off = lock_and_alloc(ntdb, ftable_off,
+					     b, keylen, datalen, want_extra,
+					     magic, hashlow);
+			if (NTDB_OFF_IS_ERR(off))
+				return off;
+			if (off != 0) {
+				if (b == start_b)
+					ntdb->stats.alloc_bucket_exact++;
+				if (b == NTDB_FREE_BUCKETS - 1)
+					ntdb->stats.alloc_bucket_max++;
+				/* Worked?  Stay using this list. */
+				ntdb->ftable_off = ftable_off;
+				ntdb->ftable = ftable;
+				return off;
+			}
+			/* Didn't work.  Try next bucket. */
+		}
+
+		if (NTDB_OFF_IS_ERR(b)) {
+			return b;
+		}
+
+		/* Hmm, try next table. */
+		ftable_off = next_ftable(ntdb, ftable_off);
+		if (NTDB_OFF_IS_ERR(ftable_off)) {
+			return ftable_off;
+		}
+		ftable++;
+
+		if (ftable_off == 0) {
+			wrapped = true;
+			ftable_off = first_ftable(ntdb);
+			if (NTDB_OFF_IS_ERR(ftable_off)) {
+				return ftable_off;
+			}
+			ftable = 0;
+		}
+	}
+
+	return 0;
+}
+
+enum NTDB_ERROR set_header(struct ntdb_context *ntdb,
+			  struct ntdb_used_record *rec,
+			  unsigned magic, uint64_t keylen, uint64_t datalen,
+			  uint64_t actuallen, unsigned hashlow)
+{
+	uint64_t keybits = (fls64(keylen) + 1) / 2;
+
+	/* Use bottom bits of hash, so it's independent of hash table size. */
+	rec->magic_and_meta = (hashlow & ((1 << 11)-1))
+		| ((actuallen - (keylen + datalen)) << 11)
+		| (keybits << 43)
+		| ((uint64_t)magic << 48);
+	rec->key_and_data_len = (keylen | (datalen << (keybits*2)));
+
+	/* Encoding can fail on big values. */
+	if (rec_key_length(rec) != keylen
+	    || rec_data_length(rec) != datalen
+	    || rec_extra_padding(rec) != actuallen - (keylen + datalen)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "Could not encode k=%llu,d=%llu,a=%llu",
+				  (long long)keylen, (long long)datalen,
+				  (long long)actuallen);
+	}
+	return NTDB_SUCCESS;
+}
+
+/* You need 'size', this tells you how much you should expand by. */
+ntdb_off_t ntdb_expand_adjust(ntdb_off_t map_size, ntdb_off_t size)
+{
+	ntdb_off_t new_size, top_size;
+
+	/* limit size in order to avoid using up huge amounts of memory for
+	 * in memory tdbs if an oddball huge record creeps in */
+	if (size > 100 * 1024) {
+		top_size = map_size + size * 2;
+	} else {
+		top_size = map_size + size * 100;
+	}
+
+	/* always make room for at least top_size more records, and at
+	   least 25% more space. if the DB is smaller than 100MiB,
+	   otherwise grow it by 10% only. */
+	if (map_size > 100 * 1024 * 1024) {
+		new_size = map_size * 1.10;
+	} else {
+		new_size = map_size * 1.25;
+	}
+
+	/* Round the database up to a multiple of the page size */
+	if (new_size < top_size)
+		new_size = top_size;
+	return new_size - map_size;
+}
+
+/* Expand the database. */
+static enum NTDB_ERROR ntdb_expand(struct ntdb_context *ntdb, ntdb_len_t size)
+{
+	uint64_t old_size;
+	ntdb_len_t wanted;
+	enum NTDB_ERROR ecode;
+
+	/* Need to hold a hash lock to expand DB: transactions rely on it. */
+	if (!(ntdb->flags & NTDB_NOLOCK)
+	    && !ntdb->file->allrecord_lock.count && !ntdb_has_hash_locks(ntdb)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_expand: must hold lock during expand");
+	}
+
+	/* Only one person can expand file at a time. */
+	ecode = ntdb_lock_expand(ntdb, F_WRLCK);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	/* Someone else may have expanded the file, so retry. */
+	old_size = ntdb->file->map_size;
+	ntdb->io->oob(ntdb, ntdb->file->map_size, 1, true);
+	if (ntdb->file->map_size != old_size) {
+		ntdb_unlock_expand(ntdb, F_WRLCK);
+		return NTDB_SUCCESS;
+	}
+
+	/* Overallocate. */
+	wanted = ntdb_expand_adjust(old_size, size);
+	/* We need room for the record header too. */
+	wanted = adjust_size(0, sizeof(struct ntdb_used_record) + wanted);
+
+	ecode = ntdb->io->expand_file(ntdb, wanted);
+	if (ecode != NTDB_SUCCESS) {
+		ntdb_unlock_expand(ntdb, F_WRLCK);
+		return ecode;
+	}
+
+	/* We need to drop this lock before adding free record. */
+	ntdb_unlock_expand(ntdb, F_WRLCK);
+
+	ntdb->stats.expands++;
+	return add_free_record(ntdb, old_size, wanted, NTDB_LOCK_WAIT, true);
+}
+
+/* This won't fail: it will expand the database if it has to. */
+ntdb_off_t alloc(struct ntdb_context *ntdb, size_t keylen, size_t datalen,
+		uint64_t hash, unsigned magic, bool growing)
+{
+	ntdb_off_t off;
+
+	/* We can't hold pointers during this: we could unmap! */
+	assert(!ntdb->direct_access);
+
+	for (;;) {
+		enum NTDB_ERROR ecode;
+		off = get_free(ntdb, keylen, datalen, growing, magic, hash);
+		if (likely(off != 0))
+			break;
+
+		ecode = ntdb_expand(ntdb, adjust_size(keylen, datalen));
+		if (ecode != NTDB_SUCCESS) {
+			return NTDB_ERR_TO_OFF(ecode);
+		}
+	}
+
+	return off;
+}

lib/ntdb/hash.c

@@ -0,0 +1,894 @@
+ /*
+   Trivial Database 2: hash handling
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "private.h"
+#include <ccan/hash/hash.h>
+#include <assert.h>
+
+/* Default hash function. */
+uint64_t ntdb_jenkins_hash(const void *key, size_t length, uint64_t seed,
+			  void *unused)
+{
+	uint64_t ret;
+	/* hash64_stable assumes lower bits are more important; they are a
+	 * slightly better hash.  We use the upper bits first, so swap them. */
+	ret = hash64_stable((const unsigned char *)key, length, seed);
+	return (ret >> 32) | (ret << 32);
+}
+
+uint64_t ntdb_hash(struct ntdb_context *ntdb, const void *ptr, size_t len)
+{
+	return ntdb->hash_fn(ptr, len, ntdb->hash_seed, ntdb->hash_data);
+}
+
+uint64_t hash_record(struct ntdb_context *ntdb, ntdb_off_t off)
+{
+	const struct ntdb_used_record *r;
+	const void *key;
+	uint64_t klen, hash;
+
+	r = ntdb_access_read(ntdb, off, sizeof(*r), true);
+	if (NTDB_PTR_IS_ERR(r)) {
+		/* FIXME */
+		return 0;
+	}
+
+	klen = rec_key_length(r);
+	ntdb_access_release(ntdb, r);
+
+	key = ntdb_access_read(ntdb, off + sizeof(*r), klen, false);
+	if (NTDB_PTR_IS_ERR(key)) {
+		return 0;
+	}
+
+	hash = ntdb_hash(ntdb, key, klen);
+	ntdb_access_release(ntdb, key);
+	return hash;
+}
+
+/* Get bits from a value. */
+static uint32_t bits_from(uint64_t val, unsigned start, unsigned num)
+{
+	assert(num <= 32);
+	return (val >> start) & ((1U << num) - 1);
+}
+
+/* We take bits from the top: that way we can lock whole sections of the hash
+ * by using lock ranges. */
+static uint32_t use_bits(struct hash_info *h, unsigned num)
+{
+	h->hash_used += num;
+	return bits_from(h->h, 64 - h->hash_used, num);
+}
+
+static ntdb_bool_err key_matches(struct ntdb_context *ntdb,
+				const struct ntdb_used_record *rec,
+				ntdb_off_t off,
+				const NTDB_DATA *key)
+{
+	ntdb_bool_err ret = false;
+	const char *rkey;
+
+	if (rec_key_length(rec) != key->dsize) {
+		ntdb->stats.compare_wrong_keylen++;
+		return ret;
+	}
+
+	rkey = ntdb_access_read(ntdb, off + sizeof(*rec), key->dsize, false);
+	if (NTDB_PTR_IS_ERR(rkey)) {
+		return (ntdb_bool_err)NTDB_PTR_ERR(rkey);
+	}
+	if (memcmp(rkey, key->dptr, key->dsize) == 0)
+		ret = true;
+	else
+		ntdb->stats.compare_wrong_keycmp++;
+	ntdb_access_release(ntdb, rkey);
+	return ret;
+}
+
+/* Does entry match? */
+static ntdb_bool_err match(struct ntdb_context *ntdb,
+			  struct hash_info *h,
+			  const NTDB_DATA *key,
+			  ntdb_off_t val,
+			  struct ntdb_used_record *rec)
+{
+	ntdb_off_t off;
+	enum NTDB_ERROR ecode;
+
+	ntdb->stats.compares++;
+	/* Desired bucket must match. */
+	if (h->home_bucket != (val & NTDB_OFF_HASH_GROUP_MASK)) {
+		ntdb->stats.compare_wrong_bucket++;
+		return false;
+	}
+
+	/* Top bits of offset == next bits of hash. */
+	if (bits_from(val, NTDB_OFF_HASH_EXTRA_BIT, NTDB_OFF_UPPER_STEAL_EXTRA)
+	    != bits_from(h->h, 64 - h->hash_used - NTDB_OFF_UPPER_STEAL_EXTRA,
+		    NTDB_OFF_UPPER_STEAL_EXTRA)) {
+		ntdb->stats.compare_wrong_offsetbits++;
+		return false;
+	}
+
+	off = val & NTDB_OFF_MASK;
+	ecode = ntdb_read_convert(ntdb, off, rec, sizeof(*rec));
+	if (ecode != NTDB_SUCCESS) {
+		return (ntdb_bool_err)ecode;
+	}
+
+	if ((h->h & ((1 << 11)-1)) != rec_hash(rec)) {
+		ntdb->stats.compare_wrong_rechash++;
+		return false;
+	}
+
+	return key_matches(ntdb, rec, off, key);
+}
+
+static ntdb_off_t hbucket_off(ntdb_off_t group_start, unsigned bucket)
+{
+	return group_start
+		+ (bucket % (1 << NTDB_HASH_GROUP_BITS)) * sizeof(ntdb_off_t);
+}
+
+bool is_subhash(ntdb_off_t val)
+{
+	return (val >> NTDB_OFF_UPPER_STEAL_SUBHASH_BIT) & 1;
+}
+
+/* FIXME: Guess the depth, don't over-lock! */
+static ntdb_off_t hlock_range(ntdb_off_t group, ntdb_off_t *size)
+{
+	*size = 1ULL << (64 - (NTDB_TOPLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS));
+	return group << (64 - (NTDB_TOPLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS));
+}
+
+static ntdb_off_t COLD find_in_chain(struct ntdb_context *ntdb,
+				    NTDB_DATA key,
+				    ntdb_off_t chain,
+				    struct hash_info *h,
+				    struct ntdb_used_record *rec,
+				    struct traverse_info *tinfo)
+{
+	ntdb_off_t off, next;
+	enum NTDB_ERROR ecode;
+
+	/* In case nothing is free, we set these to zero. */
+	h->home_bucket = h->found_bucket = 0;
+
+	for (off = chain; off; off = next) {
+		unsigned int i;
+
+		h->group_start = off;
+		ecode = ntdb_read_convert(ntdb, off, h->group, sizeof(h->group));
+		if (ecode != NTDB_SUCCESS) {
+			return NTDB_ERR_TO_OFF(ecode);
+		}
+
+		for (i = 0; i < (1 << NTDB_HASH_GROUP_BITS); i++) {
+			ntdb_off_t recoff;
+			if (!h->group[i]) {
+				/* Remember this empty bucket. */
+				h->home_bucket = h->found_bucket = i;
+				continue;
+			}
+
+			/* We can insert extra bits via add_to_hash
+			 * empty bucket logic. */
+			recoff = h->group[i] & NTDB_OFF_MASK;
+			ecode = ntdb_read_convert(ntdb, recoff, rec,
+						 sizeof(*rec));
+			if (ecode != NTDB_SUCCESS) {
+				return NTDB_ERR_TO_OFF(ecode);
+			}
+
+			ecode = NTDB_OFF_TO_ERR(key_matches(ntdb, rec, recoff,
+							   &key));
+			if (ecode < 0) {
+				return NTDB_ERR_TO_OFF(ecode);
+			}
+			if (ecode == (enum NTDB_ERROR)1) {
+				h->home_bucket = h->found_bucket = i;
+
+				if (tinfo) {
+					tinfo->levels[tinfo->num_levels]
+						.hashtable = off;
+					tinfo->levels[tinfo->num_levels]
+						.total_buckets
+						= 1 << NTDB_HASH_GROUP_BITS;
+					tinfo->levels[tinfo->num_levels].entry
+						= i;
+					tinfo->num_levels++;
+				}
+				return recoff;
+			}
+		}
+		next = ntdb_read_off(ntdb, off
+				    + offsetof(struct ntdb_chain, next));
+		if (NTDB_OFF_IS_ERR(next)) {
+			return next;
+		}
+		if (next)
+			next += sizeof(struct ntdb_used_record);
+	}
+	return 0;
+}
+
+/* This is the core routine which searches the hashtable for an entry.
+ * On error, no locks are held and -ve is returned.
+ * Otherwise, hinfo is filled in (and the optional tinfo).
+ * If not found, the return value is 0.
+ * If found, the return value is the offset, and *rec is the record. */
+ntdb_off_t find_and_lock(struct ntdb_context *ntdb,
+			NTDB_DATA key,
+			int ltype,
+			struct hash_info *h,
+			struct ntdb_used_record *rec,
+			struct traverse_info *tinfo)
+{
+	uint32_t i, group;
+	ntdb_off_t hashtable;
+	enum NTDB_ERROR ecode;
+
+	h->h = ntdb_hash(ntdb, key.dptr, key.dsize);
+	h->hash_used = 0;
+	group = use_bits(h, NTDB_TOPLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS);
+	h->home_bucket = use_bits(h, NTDB_HASH_GROUP_BITS);
+
+	h->hlock_start = hlock_range(group, &h->hlock_range);
+	ecode = ntdb_lock_hashes(ntdb, h->hlock_start, h->hlock_range, ltype,
+				NTDB_LOCK_WAIT);
+	if (ecode != NTDB_SUCCESS) {
+		return NTDB_ERR_TO_OFF(ecode);
+	}
+
+	hashtable = offsetof(struct ntdb_header, hashtable);
+	if (tinfo) {
+		tinfo->toplevel_group = group;
+		tinfo->num_levels = 1;
+		tinfo->levels[0].entry = 0;
+		tinfo->levels[0].hashtable = hashtable
+			+ (group << NTDB_HASH_GROUP_BITS) * sizeof(ntdb_off_t);
+		tinfo->levels[0].total_buckets = 1 << NTDB_HASH_GROUP_BITS;
+	}
+
+	while (h->hash_used <= 64) {
+		/* Read in the hash group. */
+		h->group_start = hashtable
+			+ group * (sizeof(ntdb_off_t) << NTDB_HASH_GROUP_BITS);
+
+		ecode = ntdb_read_convert(ntdb, h->group_start, &h->group,
+					 sizeof(h->group));
+		if (ecode != NTDB_SUCCESS) {
+			goto fail;
+		}
+
+		/* Pointer to another hash table?  Go down... */
+		if (is_subhash(h->group[h->home_bucket])) {
+			hashtable = (h->group[h->home_bucket] & NTDB_OFF_MASK)
+				+ sizeof(struct ntdb_used_record);
+			if (tinfo) {
+				/* When we come back, use *next* bucket */
+				tinfo->levels[tinfo->num_levels-1].entry
+					+= h->home_bucket + 1;
+			}
+			group = use_bits(h, NTDB_SUBLEVEL_HASH_BITS
+					 - NTDB_HASH_GROUP_BITS);
+			h->home_bucket = use_bits(h, NTDB_HASH_GROUP_BITS);
+			if (tinfo) {
+				tinfo->levels[tinfo->num_levels].hashtable
+					= hashtable;
+				tinfo->levels[tinfo->num_levels].total_buckets
+					= 1 << NTDB_SUBLEVEL_HASH_BITS;
+				tinfo->levels[tinfo->num_levels].entry
+					= group << NTDB_HASH_GROUP_BITS;
+				tinfo->num_levels++;
+			}
+			continue;
+		}
+
+		/* It's in this group: search (until 0 or all searched) */
+		for (i = 0, h->found_bucket = h->home_bucket;
+		     i < (1 << NTDB_HASH_GROUP_BITS);
+		     i++, h->found_bucket = ((h->found_bucket+1)
+					     % (1 << NTDB_HASH_GROUP_BITS))) {
+			ntdb_bool_err berr;
+			if (is_subhash(h->group[h->found_bucket]))
+				continue;
+
+			if (!h->group[h->found_bucket])
+				break;
+
+			berr = match(ntdb, h, &key, h->group[h->found_bucket],
+				     rec);
+			if (berr < 0) {
+				ecode = NTDB_OFF_TO_ERR(berr);
+				goto fail;
+			}
+			if (berr) {
+				if (tinfo) {
+					tinfo->levels[tinfo->num_levels-1].entry
+						+= h->found_bucket;
+				}
+				return h->group[h->found_bucket] & NTDB_OFF_MASK;
+			}
+		}
+		/* Didn't find it: h indicates where it would go. */
+		return 0;
+	}
+
+	return find_in_chain(ntdb, key, hashtable, h, rec, tinfo);
+
+fail:
+	ntdb_unlock_hashes(ntdb, h->hlock_start, h->hlock_range, ltype);
+	return NTDB_ERR_TO_OFF(ecode);
+}
+
+/* I wrote a simple test, expanding a hash to 2GB, for the following
+ * cases:
+ * 1) Expanding all the buckets at once,
+ * 2) Expanding the bucket we wanted to place the new entry into.
+ * 3) Expanding the most-populated bucket,
+ *
+ * I measured the worst/average/best density during this process.
+ * 1) 3%/16%/30%
+ * 2) 4%/20%/38%
+ * 3) 6%/22%/41%
+ *
+ * So we figure out the busiest bucket for the moment.
+ */
+static unsigned fullest_bucket(struct ntdb_context *ntdb,
+			       const ntdb_off_t *group,
+			       unsigned new_bucket)
+{
+	unsigned counts[1 << NTDB_HASH_GROUP_BITS] = { 0 };
+	unsigned int i, best_bucket;
+
+	/* Count the new entry. */
+	counts[new_bucket]++;
+	best_bucket = new_bucket;
+
+	for (i = 0; i < (1 << NTDB_HASH_GROUP_BITS); i++) {
+		unsigned this_bucket;
+
+		if (is_subhash(group[i]))
+			continue;
+		this_bucket = group[i] & NTDB_OFF_HASH_GROUP_MASK;
+		if (++counts[this_bucket] > counts[best_bucket])
+			best_bucket = this_bucket;
+	}
+
+	return best_bucket;
+}
+
+static bool put_into_group(ntdb_off_t *group,
+			   unsigned bucket, ntdb_off_t encoded)
+{
+	unsigned int i;
+
+	for (i = 0; i < (1 << NTDB_HASH_GROUP_BITS); i++) {
+		unsigned b = (bucket + i) % (1 << NTDB_HASH_GROUP_BITS);
+
+		if (group[b] == 0) {
+			group[b] = encoded;
+			return true;
+		}
+	}
+	return false;
+}
+
+static void force_into_group(ntdb_off_t *group,
+			     unsigned bucket, ntdb_off_t encoded)
+{
+	if (!put_into_group(group, bucket, encoded))
+		abort();
+}
+
+static ntdb_off_t encode_offset(ntdb_off_t new_off, struct hash_info *h)
+{
+	return h->home_bucket
+		| new_off
+		| ((uint64_t)bits_from(h->h,
+				  64 - h->hash_used - NTDB_OFF_UPPER_STEAL_EXTRA,
+				  NTDB_OFF_UPPER_STEAL_EXTRA)
+		   << NTDB_OFF_HASH_EXTRA_BIT);
+}
+
+/* Simply overwrite the hash entry we found before. */
+enum NTDB_ERROR replace_in_hash(struct ntdb_context *ntdb,
+			       struct hash_info *h,
+			       ntdb_off_t new_off)
+{
+	return ntdb_write_off(ntdb, hbucket_off(h->group_start, h->found_bucket),
+			     encode_offset(new_off, h));
+}
+
+/* We slot in anywhere that's empty in the chain. */
+static enum NTDB_ERROR COLD add_to_chain(struct ntdb_context *ntdb,
+					ntdb_off_t subhash,
+					ntdb_off_t new_off)
+{
+	ntdb_off_t entry;
+	enum NTDB_ERROR ecode;
+
+	entry = ntdb_find_zero_off(ntdb, subhash, 1<<NTDB_HASH_GROUP_BITS);
+	if (NTDB_OFF_IS_ERR(entry)) {
+		return NTDB_OFF_TO_ERR(entry);
+	}
+
+	if (entry == 1 << NTDB_HASH_GROUP_BITS) {
+		ntdb_off_t next;
+
+		next = ntdb_read_off(ntdb, subhash
+				    + offsetof(struct ntdb_chain, next));
+		if (NTDB_OFF_IS_ERR(next)) {
+			return NTDB_OFF_TO_ERR(next);
+		}
+
+		if (!next) {
+			next = alloc(ntdb, 0, sizeof(struct ntdb_chain), 0,
+				     NTDB_CHAIN_MAGIC, false);
+			if (NTDB_OFF_IS_ERR(next))
+				return NTDB_OFF_TO_ERR(next);
+			ecode = zero_out(ntdb,
+					 next+sizeof(struct ntdb_used_record),
+					 sizeof(struct ntdb_chain));
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+			ecode = ntdb_write_off(ntdb, subhash
+					      + offsetof(struct ntdb_chain,
+							 next),
+					      next);
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+		}
+		return add_to_chain(ntdb, next, new_off);
+	}
+
+	return ntdb_write_off(ntdb, subhash + entry * sizeof(ntdb_off_t),
+			     new_off);
+}
+
+/* Add into a newly created subhash. */
+static enum NTDB_ERROR add_to_subhash(struct ntdb_context *ntdb, ntdb_off_t subhash,
+				     unsigned hash_used, ntdb_off_t val)
+{
+	ntdb_off_t off = (val & NTDB_OFF_MASK), *group;
+	struct hash_info h;
+	unsigned int gnum;
+
+	h.hash_used = hash_used;
+
+	if (hash_used + NTDB_SUBLEVEL_HASH_BITS > 64)
+		return add_to_chain(ntdb, subhash, off);
+
+	h.h = hash_record(ntdb, off);
+	gnum = use_bits(&h, NTDB_SUBLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS);
+	h.group_start = subhash
+		+ gnum * (sizeof(ntdb_off_t) << NTDB_HASH_GROUP_BITS);
+	h.home_bucket = use_bits(&h, NTDB_HASH_GROUP_BITS);
+
+	group = ntdb_access_write(ntdb, h.group_start,
+				 sizeof(*group) << NTDB_HASH_GROUP_BITS, true);
+	if (NTDB_PTR_IS_ERR(group)) {
+		return NTDB_PTR_ERR(group);
+	}
+	force_into_group(group, h.home_bucket, encode_offset(off, &h));
+	return ntdb_access_commit(ntdb, group);
+}
+
+static enum NTDB_ERROR expand_group(struct ntdb_context *ntdb, struct hash_info *h)
+{
+	unsigned bucket, num_vals, i, magic;
+	size_t subsize;
+	ntdb_off_t subhash;
+	ntdb_off_t vals[1 << NTDB_HASH_GROUP_BITS];
+	enum NTDB_ERROR ecode;
+
+	/* Attach new empty subhash under fullest bucket. */
+	bucket = fullest_bucket(ntdb, h->group, h->home_bucket);
+
+	if (h->hash_used == 64) {
+		ntdb->stats.alloc_chain++;
+		subsize = sizeof(struct ntdb_chain);
+		magic = NTDB_CHAIN_MAGIC;
+	} else {
+		ntdb->stats.alloc_subhash++;
+		subsize = (sizeof(ntdb_off_t) << NTDB_SUBLEVEL_HASH_BITS);
+		magic = NTDB_HTABLE_MAGIC;
+	}
+
+	subhash = alloc(ntdb, 0, subsize, 0, magic, false);
+	if (NTDB_OFF_IS_ERR(subhash)) {
+		return NTDB_OFF_TO_ERR(subhash);
+	}
+
+	ecode = zero_out(ntdb, subhash + sizeof(struct ntdb_used_record),
+			 subsize);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	/* Remove any which are destined for bucket or are in wrong place. */
+	num_vals = 0;
+	for (i = 0; i < (1 << NTDB_HASH_GROUP_BITS); i++) {
+		unsigned home_bucket = h->group[i] & NTDB_OFF_HASH_GROUP_MASK;
+		if (!h->group[i] || is_subhash(h->group[i]))
+			continue;
+		if (home_bucket == bucket || home_bucket != i) {
+			vals[num_vals++] = h->group[i];
+			h->group[i] = 0;
+		}
+	}
+	/* FIXME: This assert is valid, but we do this during unit test :( */
+	/* assert(num_vals); */
+
+	/* Overwrite expanded bucket with subhash pointer. */
+	h->group[bucket] = subhash | (1ULL << NTDB_OFF_UPPER_STEAL_SUBHASH_BIT);
+
+	/* Point to actual contents of record. */
+	subhash += sizeof(struct ntdb_used_record);
+
+	/* Put values back. */
+	for (i = 0; i < num_vals; i++) {
+		unsigned this_bucket = vals[i] & NTDB_OFF_HASH_GROUP_MASK;
+
+		if (this_bucket == bucket) {
+			ecode = add_to_subhash(ntdb, subhash, h->hash_used,
+					       vals[i]);
+			if (ecode != NTDB_SUCCESS)
+				return ecode;
+		} else {
+			/* There should be room to put this back. */
+			force_into_group(h->group, this_bucket, vals[i]);
+		}
+	}
+	return NTDB_SUCCESS;
+}
+
+enum NTDB_ERROR delete_from_hash(struct ntdb_context *ntdb, struct hash_info *h)
+{
+	unsigned int i, num_movers = 0;
+	ntdb_off_t movers[1 << NTDB_HASH_GROUP_BITS];
+
+	h->group[h->found_bucket] = 0;
+	for (i = 1; i < (1 << NTDB_HASH_GROUP_BITS); i++) {
+		unsigned this_bucket;
+
+		this_bucket = (h->found_bucket+i) % (1 << NTDB_HASH_GROUP_BITS);
+		/* Empty bucket?  We're done. */
+		if (!h->group[this_bucket])
+			break;
+
+		/* Ignore subhashes. */
+		if (is_subhash(h->group[this_bucket]))
+			continue;
+
+		/* If this one is not happy where it is, we'll move it. */
+		if ((h->group[this_bucket] & NTDB_OFF_HASH_GROUP_MASK)
+		    != this_bucket) {
+			movers[num_movers++] = h->group[this_bucket];
+			h->group[this_bucket] = 0;
+		}
+	}
+
+	/* Put back the ones we erased. */
+	for (i = 0; i < num_movers; i++) {
+		force_into_group(h->group, movers[i] & NTDB_OFF_HASH_GROUP_MASK,
+				 movers[i]);
+	}
+
+	/* Now we write back the hash group */
+	return ntdb_write_convert(ntdb, h->group_start,
+				 h->group, sizeof(h->group));
+}
+
+enum NTDB_ERROR add_to_hash(struct ntdb_context *ntdb, struct hash_info *h,
+			   ntdb_off_t new_off)
+{
+	enum NTDB_ERROR ecode;
+
+	/* We hit an empty bucket during search?  That's where it goes. */
+	if (!h->group[h->found_bucket]) {
+		h->group[h->found_bucket] = encode_offset(new_off, h);
+		/* Write back the modified group. */
+		return ntdb_write_convert(ntdb, h->group_start,
+					 h->group, sizeof(h->group));
+	}
+
+	if (h->hash_used > 64)
+		return add_to_chain(ntdb, h->group_start, new_off);
+
+	/* We're full.  Expand. */
+	ecode = expand_group(ntdb, h);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (is_subhash(h->group[h->home_bucket])) {
+		/* We were expanded! */
+		ntdb_off_t hashtable;
+		unsigned int gnum;
+
+		/* Write back the modified group. */
+		ecode = ntdb_write_convert(ntdb, h->group_start, h->group,
+					  sizeof(h->group));
+		if (ecode != NTDB_SUCCESS) {
+			return ecode;
+		}
+
+		/* Move hashinfo down a level. */
+		hashtable = (h->group[h->home_bucket] & NTDB_OFF_MASK)
+			+ sizeof(struct ntdb_used_record);
+		gnum = use_bits(h,NTDB_SUBLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS);
+		h->home_bucket = use_bits(h, NTDB_HASH_GROUP_BITS);
+		h->group_start = hashtable
+			+ gnum * (sizeof(ntdb_off_t) << NTDB_HASH_GROUP_BITS);
+		ecode = ntdb_read_convert(ntdb, h->group_start, &h->group,
+					 sizeof(h->group));
+		if (ecode != NTDB_SUCCESS) {
+			return ecode;
+		}
+	}
+
+	/* Expanding the group must have made room if it didn't choose this
+	 * bucket. */
+	if (put_into_group(h->group, h->home_bucket, encode_offset(new_off,h))){
+		return ntdb_write_convert(ntdb, h->group_start,
+					 h->group, sizeof(h->group));
+	}
+
+	/* This can happen if all hashes in group (and us) dropped into same
+	 * group in subhash. */
+	return add_to_hash(ntdb, h, new_off);
+}
+
+/* Traverse support: returns offset of record, or 0 or -ve error. */
+static ntdb_off_t iterate_hash(struct ntdb_context *ntdb,
+			      struct traverse_info *tinfo)
+{
+	ntdb_off_t off, val, i;
+	struct traverse_level *tlevel;
+
+	tlevel = &tinfo->levels[tinfo->num_levels-1];
+
+again:
+	for (i = ntdb_find_nonzero_off(ntdb, tlevel->hashtable,
+				      tlevel->entry, tlevel->total_buckets);
+	     i != tlevel->total_buckets;
+	     i = ntdb_find_nonzero_off(ntdb, tlevel->hashtable,
+				      i+1, tlevel->total_buckets)) {
+		if (NTDB_OFF_IS_ERR(i)) {
+			return i;
+		}
+
+		val = ntdb_read_off(ntdb, tlevel->hashtable+sizeof(ntdb_off_t)*i);
+		if (NTDB_OFF_IS_ERR(val)) {
+			return val;
+		}
+
+		off = val & NTDB_OFF_MASK;
+
+		/* This makes the delete-all-in-traverse case work
+		 * (and simplifies our logic a little). */
+		if (off == tinfo->prev)
+			continue;
+
+		tlevel->entry = i;
+
+		if (!is_subhash(val)) {
+			/* Found one. */
+			tinfo->prev = off;
+			return off;
+		}
+
+		/* When we come back, we want the next one */
+		tlevel->entry++;
+		tinfo->num_levels++;
+		tlevel++;
+		tlevel->hashtable = off + sizeof(struct ntdb_used_record);
+		tlevel->entry = 0;
+		/* Next level is a chain? */
+		if (unlikely(tinfo->num_levels == NTDB_MAX_LEVELS + 1))
+			tlevel->total_buckets = (1 << NTDB_HASH_GROUP_BITS);
+		else
+			tlevel->total_buckets = (1 << NTDB_SUBLEVEL_HASH_BITS);
+		goto again;
+	}
+
+	/* Nothing there? */
+	if (tinfo->num_levels == 1)
+		return 0;
+
+	/* Handle chained entries. */
+	if (unlikely(tinfo->num_levels == NTDB_MAX_LEVELS + 1)) {
+		tlevel->hashtable = ntdb_read_off(ntdb, tlevel->hashtable
+						 + offsetof(struct ntdb_chain,
+							    next));
+		if (NTDB_OFF_IS_ERR(tlevel->hashtable)) {
+			return tlevel->hashtable;
+		}
+		if (tlevel->hashtable) {
+			tlevel->hashtable += sizeof(struct ntdb_used_record);
+			tlevel->entry = 0;
+			goto again;
+		}
+	}
+
+	/* Go back up and keep searching. */
+	tinfo->num_levels--;
+	tlevel--;
+	goto again;
+}
+
+/* Return success if we find something, NTDB_ERR_NOEXIST if none. */
+enum NTDB_ERROR next_in_hash(struct ntdb_context *ntdb,
+			    struct traverse_info *tinfo,
+			    NTDB_DATA *kbuf, size_t *dlen)
+{
+	const unsigned group_bits = NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS;
+	ntdb_off_t hl_start, hl_range, off;
+	enum NTDB_ERROR ecode;
+
+	while (tinfo->toplevel_group < (1 << group_bits)) {
+		hl_start = (ntdb_off_t)tinfo->toplevel_group
+			<< (64 - group_bits);
+		hl_range = 1ULL << group_bits;
+		ecode = ntdb_lock_hashes(ntdb, hl_start, hl_range, F_RDLCK,
+					NTDB_LOCK_WAIT);
+		if (ecode != NTDB_SUCCESS) {
+			return ecode;
+		}
+
+		off = iterate_hash(ntdb, tinfo);
+		if (off) {
+			struct ntdb_used_record rec;
+
+			if (NTDB_OFF_IS_ERR(off)) {
+				ecode = NTDB_OFF_TO_ERR(off);
+				goto fail;
+			}
+
+			ecode = ntdb_read_convert(ntdb, off, &rec, sizeof(rec));
+			if (ecode != NTDB_SUCCESS) {
+				goto fail;
+			}
+			if (rec_magic(&rec) != NTDB_USED_MAGIC) {
+				ecode = ntdb_logerr(ntdb, NTDB_ERR_CORRUPT,
+						   NTDB_LOG_ERROR,
+						   "next_in_hash:"
+						   " corrupt record at %llu",
+						   (long long)off);
+				goto fail;
+			}
+
+			kbuf->dsize = rec_key_length(&rec);
+
+			/* They want data as well? */
+			if (dlen) {
+				*dlen = rec_data_length(&rec);
+				kbuf->dptr = ntdb_alloc_read(ntdb,
+							    off + sizeof(rec),
+							    kbuf->dsize
+							    + *dlen);
+			} else {
+				kbuf->dptr = ntdb_alloc_read(ntdb,
+							    off + sizeof(rec),
+							    kbuf->dsize);
+			}
+			ntdb_unlock_hashes(ntdb, hl_start, hl_range, F_RDLCK);
+			if (NTDB_PTR_IS_ERR(kbuf->dptr)) {
+				return NTDB_PTR_ERR(kbuf->dptr);
+			}
+			return NTDB_SUCCESS;
+		}
+
+		ntdb_unlock_hashes(ntdb, hl_start, hl_range, F_RDLCK);
+
+		tinfo->toplevel_group++;
+		tinfo->levels[0].hashtable
+			+= (sizeof(ntdb_off_t) << NTDB_HASH_GROUP_BITS);
+		tinfo->levels[0].entry = 0;
+	}
+	return NTDB_ERR_NOEXIST;
+
+fail:
+	ntdb_unlock_hashes(ntdb, hl_start, hl_range, F_RDLCK);
+	return ecode;
+
+}
+
+enum NTDB_ERROR first_in_hash(struct ntdb_context *ntdb,
+			     struct traverse_info *tinfo,
+			     NTDB_DATA *kbuf, size_t *dlen)
+{
+	tinfo->prev = 0;
+	tinfo->toplevel_group = 0;
+	tinfo->num_levels = 1;
+	tinfo->levels[0].hashtable = offsetof(struct ntdb_header, hashtable);
+	tinfo->levels[0].entry = 0;
+	tinfo->levels[0].total_buckets = (1 << NTDB_HASH_GROUP_BITS);
+
+	return next_in_hash(ntdb, tinfo, kbuf, dlen);
+}
+
+/* Even if the entry isn't in this hash bucket, you'd have to lock this
+ * bucket to find it. */
+static enum NTDB_ERROR chainlock(struct ntdb_context *ntdb, const NTDB_DATA *key,
+				int ltype, enum ntdb_lock_flags waitflag,
+				const char *func)
+{
+	enum NTDB_ERROR ecode;
+	uint64_t h = ntdb_hash(ntdb, key->dptr, key->dsize);
+	ntdb_off_t lockstart, locksize;
+	unsigned int group, gbits;
+
+	gbits = NTDB_TOPLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS;
+	group = bits_from(h, 64 - gbits, gbits);
+
+	lockstart = hlock_range(group, &locksize);
+
+	ecode = ntdb_lock_hashes(ntdb, lockstart, locksize, ltype, waitflag);
+	ntdb_trace_1rec(ntdb, func, *key);
+	return ecode;
+}
+
+/* lock/unlock one hash chain. This is meant to be used to reduce
+   contention - it cannot guarantee how many records will be locked */
+_PUBLIC_ enum NTDB_ERROR ntdb_chainlock(struct ntdb_context *ntdb, NTDB_DATA key)
+{
+	return ntdb->last_error = chainlock(ntdb, &key, F_WRLCK, NTDB_LOCK_WAIT,
+					   "ntdb_chainlock");
+}
+
+_PUBLIC_ void ntdb_chainunlock(struct ntdb_context *ntdb, NTDB_DATA key)
+{
+	uint64_t h = ntdb_hash(ntdb, key.dptr, key.dsize);
+	ntdb_off_t lockstart, locksize;
+	unsigned int group, gbits;
+
+	gbits = NTDB_TOPLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS;
+	group = bits_from(h, 64 - gbits, gbits);
+
+	lockstart = hlock_range(group, &locksize);
+
+	ntdb_trace_1rec(ntdb, "ntdb_chainunlock", key);
+	ntdb_unlock_hashes(ntdb, lockstart, locksize, F_WRLCK);
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_chainlock_read(struct ntdb_context *ntdb, NTDB_DATA key)
+{
+	return ntdb->last_error = chainlock(ntdb, &key, F_RDLCK, NTDB_LOCK_WAIT,
+					   "ntdb_chainlock_read");
+}
+
+_PUBLIC_ void ntdb_chainunlock_read(struct ntdb_context *ntdb, NTDB_DATA key)
+{
+	uint64_t h = ntdb_hash(ntdb, key.dptr, key.dsize);
+	ntdb_off_t lockstart, locksize;
+	unsigned int group, gbits;
+
+	gbits = NTDB_TOPLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS;
+	group = bits_from(h, 64 - gbits, gbits);
+
+	lockstart = hlock_range(group, &locksize);
+
+	ntdb_trace_1rec(ntdb, "ntdb_chainunlock_read", key);
+	ntdb_unlock_hashes(ntdb, lockstart, locksize, F_RDLCK);
+}

lib/ntdb/io.c

@@ -0,0 +1,650 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+   Copyright (C) Rusty Russell			   2010
+
+     ** NOTE! The following LGPL license applies to the ntdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "private.h"
+#include <assert.h>
+#include <ccan/likely/likely.h>
+
+void ntdb_munmap(struct ntdb_file *file)
+{
+	if (file->fd == -1)
+		return;
+
+	if (file->map_ptr) {
+		munmap(file->map_ptr, file->map_size);
+		file->map_ptr = NULL;
+	}
+}
+
+enum NTDB_ERROR ntdb_mmap(struct ntdb_context *ntdb)
+{
+	int mmap_flags;
+
+	if (ntdb->flags & NTDB_INTERNAL)
+		return NTDB_SUCCESS;
+
+#ifndef HAVE_INCOHERENT_MMAP
+	if (ntdb->flags & NTDB_NOMMAP)
+		return NTDB_SUCCESS;
+#endif
+
+	if ((ntdb->open_flags & O_ACCMODE) == O_RDONLY)
+		mmap_flags = PROT_READ;
+	else
+		mmap_flags = PROT_READ | PROT_WRITE;
+
+	/* size_t can be smaller than off_t. */
+	if ((size_t)ntdb->file->map_size == ntdb->file->map_size) {
+		ntdb->file->map_ptr = mmap(NULL, ntdb->file->map_size,
+					  mmap_flags,
+					  MAP_SHARED, ntdb->file->fd, 0);
+	} else
+		ntdb->file->map_ptr = MAP_FAILED;
+
+	/*
+	 * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
+	 */
+	if (ntdb->file->map_ptr == MAP_FAILED) {
+		ntdb->file->map_ptr = NULL;
+#ifdef HAVE_INCOHERENT_MMAP
+		/* Incoherent mmap means everyone must mmap! */
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "ntdb_mmap failed for size %lld (%s)",
+				  (long long)ntdb->file->map_size,
+				  strerror(errno));
+#else
+		ntdb_logerr(ntdb, NTDB_SUCCESS, NTDB_LOG_WARNING,
+			   "ntdb_mmap failed for size %lld (%s)",
+			   (long long)ntdb->file->map_size, strerror(errno));
+#endif
+	}
+	return NTDB_SUCCESS;
+}
+
+/* check for an out of bounds access - if it is out of bounds then
+   see if the database has been expanded by someone else and expand
+   if necessary
+   note that "len" is the minimum length needed for the db.
+
+   If probe is true, len being too large isn't a failure.
+*/
+static enum NTDB_ERROR ntdb_oob(struct ntdb_context *ntdb,
+			      ntdb_off_t off, ntdb_len_t len, bool probe)
+{
+	struct stat st;
+	enum NTDB_ERROR ecode;
+
+	/* We can't hold pointers during this: we could unmap! */
+	assert(!ntdb->direct_access
+	       || (ntdb->flags & NTDB_NOLOCK)
+	       || ntdb_has_expansion_lock(ntdb));
+
+	if (len + off < len) {
+		if (probe)
+			return NTDB_SUCCESS;
+
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "ntdb_oob off %llu len %llu wrap\n",
+				  (long long)off, (long long)len);
+	}
+
+	if (len + off <= ntdb->file->map_size)
+		return NTDB_SUCCESS;
+	if (ntdb->flags & NTDB_INTERNAL) {
+		if (probe)
+			return NTDB_SUCCESS;
+
+		ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+			   "ntdb_oob len %lld beyond internal"
+			   " malloc size %lld",
+			   (long long)(off + len),
+			   (long long)ntdb->file->map_size);
+		return NTDB_ERR_IO;
+	}
+
+	ecode = ntdb_lock_expand(ntdb, F_RDLCK);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (fstat(ntdb->file->fd, &st) != 0) {
+		ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+			   "Failed to fstat file: %s", strerror(errno));
+		ntdb_unlock_expand(ntdb, F_RDLCK);
+		return NTDB_ERR_IO;
+	}
+
+	ntdb_unlock_expand(ntdb, F_RDLCK);
+
+	if (st.st_size < off + len) {
+		if (probe)
+			return NTDB_SUCCESS;
+
+		ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+			   "ntdb_oob len %llu beyond eof at %llu",
+			   (long long)(off + len), (long long)st.st_size);
+		return NTDB_ERR_IO;
+	}
+
+	/* Unmap, update size, remap */
+	ntdb_munmap(ntdb->file);
+
+	ntdb->file->map_size = st.st_size;
+	return ntdb_mmap(ntdb);
+}
+
+/* Endian conversion: we only ever deal with 8 byte quantities */
+void *ntdb_convert(const struct ntdb_context *ntdb, void *buf, ntdb_len_t size)
+{
+	assert(size % 8 == 0);
+	if (unlikely((ntdb->flags & NTDB_CONVERT)) && buf) {
+		uint64_t i, *p = (uint64_t *)buf;
+		for (i = 0; i < size / 8; i++)
+			p[i] = bswap_64(p[i]);
+	}
+	return buf;
+}
+
+/* Return first non-zero offset in offset array, or end, or -ve error. */
+/* FIXME: Return the off? */
+uint64_t ntdb_find_nonzero_off(struct ntdb_context *ntdb,
+			      ntdb_off_t base, uint64_t start, uint64_t end)
+{
+	uint64_t i;
+	const uint64_t *val;
+
+	/* Zero vs non-zero is the same unconverted: minor optimization. */
+	val = ntdb_access_read(ntdb, base + start * sizeof(ntdb_off_t),
+			      (end - start) * sizeof(ntdb_off_t), false);
+	if (NTDB_PTR_IS_ERR(val)) {
+		return NTDB_ERR_TO_OFF(NTDB_PTR_ERR(val));
+	}
+
+	for (i = 0; i < (end - start); i++) {
+		if (val[i])
+			break;
+	}
+	ntdb_access_release(ntdb, val);
+	return start + i;
+}
+
+/* Return first zero offset in num offset array, or num, or -ve error. */
+uint64_t ntdb_find_zero_off(struct ntdb_context *ntdb, ntdb_off_t off,
+			   uint64_t num)
+{
+	uint64_t i;
+	const uint64_t *val;
+
+	/* Zero vs non-zero is the same unconverted: minor optimization. */
+	val = ntdb_access_read(ntdb, off, num * sizeof(ntdb_off_t), false);
+	if (NTDB_PTR_IS_ERR(val)) {
+		return NTDB_ERR_TO_OFF(NTDB_PTR_ERR(val));
+	}
+
+	for (i = 0; i < num; i++) {
+		if (!val[i])
+			break;
+	}
+	ntdb_access_release(ntdb, val);
+	return i;
+}
+
+enum NTDB_ERROR zero_out(struct ntdb_context *ntdb, ntdb_off_t off, ntdb_len_t len)
+{
+	char buf[8192] = { 0 };
+	void *p = ntdb->io->direct(ntdb, off, len, true);
+	enum NTDB_ERROR ecode = NTDB_SUCCESS;
+
+	assert(!(ntdb->flags & NTDB_RDONLY));
+	if (NTDB_PTR_IS_ERR(p)) {
+		return NTDB_PTR_ERR(p);
+	}
+	if (p) {
+		memset(p, 0, len);
+		return ecode;
+	}
+	while (len) {
+		unsigned todo = len < sizeof(buf) ? len : sizeof(buf);
+		ecode = ntdb->io->twrite(ntdb, off, buf, todo);
+		if (ecode != NTDB_SUCCESS) {
+			break;
+		}
+		len -= todo;
+		off += todo;
+	}
+	return ecode;
+}
+
+ntdb_off_t ntdb_read_off(struct ntdb_context *ntdb, ntdb_off_t off)
+{
+	ntdb_off_t ret;
+	enum NTDB_ERROR ecode;
+
+	if (likely(!(ntdb->flags & NTDB_CONVERT))) {
+		ntdb_off_t *p = ntdb->io->direct(ntdb, off, sizeof(*p), false);
+		if (NTDB_PTR_IS_ERR(p)) {
+			return NTDB_ERR_TO_OFF(NTDB_PTR_ERR(p));
+		}
+		if (p)
+			return *p;
+	}
+
+	ecode = ntdb_read_convert(ntdb, off, &ret, sizeof(ret));
+	if (ecode != NTDB_SUCCESS) {
+		return NTDB_ERR_TO_OFF(ecode);
+	}
+	return ret;
+}
+
+/* write a lump of data at a specified offset */
+static enum NTDB_ERROR ntdb_write(struct ntdb_context *ntdb, ntdb_off_t off,
+				const void *buf, ntdb_len_t len)
+{
+	enum NTDB_ERROR ecode;
+
+	if (ntdb->flags & NTDB_RDONLY) {
+		return ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_USE_ERROR,
+				  "Write to read-only database");
+	}
+
+	ecode = ntdb->io->oob(ntdb, off, len, false);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (ntdb->file->map_ptr) {
+		memcpy(off + (char *)ntdb->file->map_ptr, buf, len);
+	} else {
+#ifdef HAVE_INCOHERENT_MMAP
+		return NTDB_ERR_IO;
+#else
+		ssize_t ret;
+		ret = pwrite(ntdb->file->fd, buf, len, off);
+		if (ret != len) {
+			/* This shouldn't happen: we avoid sparse files. */
+			if (ret >= 0)
+				errno = ENOSPC;
+
+			return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+					  "ntdb_write: %zi at %zu len=%zu (%s)",
+					  ret, (size_t)off, (size_t)len,
+					  strerror(errno));
+		}
+#endif
+	}
+	return NTDB_SUCCESS;
+}
+
+/* read a lump of data at a specified offset */
+static enum NTDB_ERROR ntdb_read(struct ntdb_context *ntdb, ntdb_off_t off,
+			       void *buf, ntdb_len_t len)
+{
+	enum NTDB_ERROR ecode;
+
+	ecode = ntdb->io->oob(ntdb, off, len, false);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (ntdb->file->map_ptr) {
+		memcpy(buf, off + (char *)ntdb->file->map_ptr, len);
+	} else {
+#ifdef HAVE_INCOHERENT_MMAP
+		return NTDB_ERR_IO;
+#else
+		ssize_t r = pread(ntdb->file->fd, buf, len, off);
+		if (r != len) {
+			return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+					  "ntdb_read failed with %zi at %zu "
+					  "len=%zu (%s) map_size=%zu",
+					  r, (size_t)off, (size_t)len,
+					  strerror(errno),
+					  (size_t)ntdb->file->map_size);
+		}
+#endif
+	}
+	return NTDB_SUCCESS;
+}
+
+enum NTDB_ERROR ntdb_write_convert(struct ntdb_context *ntdb, ntdb_off_t off,
+				 const void *rec, size_t len)
+{
+	enum NTDB_ERROR ecode;
+
+	if (unlikely((ntdb->flags & NTDB_CONVERT))) {
+		void *conv = malloc(len);
+		if (!conv) {
+			return ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+					  "ntdb_write: no memory converting"
+					  " %zu bytes", len);
+		}
+		memcpy(conv, rec, len);
+		ecode = ntdb->io->twrite(ntdb, off,
+					ntdb_convert(ntdb, conv, len), len);
+		free(conv);
+	} else {
+		ecode = ntdb->io->twrite(ntdb, off, rec, len);
+	}
+	return ecode;
+}
+
+enum NTDB_ERROR ntdb_read_convert(struct ntdb_context *ntdb, ntdb_off_t off,
+				void *rec, size_t len)
+{
+	enum NTDB_ERROR ecode = ntdb->io->tread(ntdb, off, rec, len);
+	ntdb_convert(ntdb, rec, len);
+	return ecode;
+}
+
+enum NTDB_ERROR ntdb_write_off(struct ntdb_context *ntdb,
+			     ntdb_off_t off, ntdb_off_t val)
+{
+	if (ntdb->flags & NTDB_RDONLY) {
+		return ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_USE_ERROR,
+				  "Write to read-only database");
+	}
+
+	if (likely(!(ntdb->flags & NTDB_CONVERT))) {
+		ntdb_off_t *p = ntdb->io->direct(ntdb, off, sizeof(*p), true);
+		if (NTDB_PTR_IS_ERR(p)) {
+			return NTDB_PTR_ERR(p);
+		}
+		if (p) {
+			*p = val;
+			return NTDB_SUCCESS;
+		}
+	}
+	return ntdb_write_convert(ntdb, off, &val, sizeof(val));
+}
+
+static void *_ntdb_alloc_read(struct ntdb_context *ntdb, ntdb_off_t offset,
+			     ntdb_len_t len, unsigned int prefix)
+{
+	unsigned char *buf;
+	enum NTDB_ERROR ecode;
+
+	/* some systems don't like zero length malloc */
+	buf = malloc(prefix + len ? prefix + len : 1);
+	if (!buf) {
+		ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_USE_ERROR,
+			   "ntdb_alloc_read malloc failed len=%zu",
+			   (size_t)(prefix + len));
+		return NTDB_ERR_PTR(NTDB_ERR_OOM);
+	} else {
+		ecode = ntdb->io->tread(ntdb, offset, buf+prefix, len);
+		if (unlikely(ecode != NTDB_SUCCESS)) {
+			free(buf);
+			return NTDB_ERR_PTR(ecode);
+		}
+	}
+	return buf;
+}
+
+/* read a lump of data, allocating the space for it */
+void *ntdb_alloc_read(struct ntdb_context *ntdb, ntdb_off_t offset, ntdb_len_t len)
+{
+	return _ntdb_alloc_read(ntdb, offset, len, 0);
+}
+
+static enum NTDB_ERROR fill(struct ntdb_context *ntdb,
+			   const void *buf, size_t size,
+			   ntdb_off_t off, ntdb_len_t len)
+{
+	while (len) {
+		size_t n = len > size ? size : len;
+		ssize_t ret = pwrite(ntdb->file->fd, buf, n, off);
+		if (ret != n) {
+			if (ret >= 0)
+				errno = ENOSPC;
+
+			return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+					  "fill failed:"
+					  " %zi at %zu len=%zu (%s)",
+					  ret, (size_t)off, (size_t)len,
+					  strerror(errno));
+		}
+		len -= n;
+		off += n;
+	}
+	return NTDB_SUCCESS;
+}
+
+/* expand a file.  we prefer to use ftruncate, as that is what posix
+  says to use for mmap expansion */
+static enum NTDB_ERROR ntdb_expand_file(struct ntdb_context *ntdb,
+				      ntdb_len_t addition)
+{
+	char buf[8192];
+	enum NTDB_ERROR ecode;
+
+	if (ntdb->flags & NTDB_RDONLY) {
+		return ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_USE_ERROR,
+				  "Expand on read-only database");
+	}
+
+	if (ntdb->flags & NTDB_INTERNAL) {
+		char *new = realloc(ntdb->file->map_ptr,
+				    ntdb->file->map_size + addition);
+		if (!new) {
+			return ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+					  "No memory to expand database");
+		}
+		ntdb->file->map_ptr = new;
+		ntdb->file->map_size += addition;
+		return NTDB_SUCCESS;
+	} else {
+		/* Unmap before trying to write; old NTDB claimed OpenBSD had
+		 * problem with this otherwise. */
+		ntdb_munmap(ntdb->file);
+
+		/* If this fails, we try to fill anyway. */
+		if (ftruncate(ntdb->file->fd, ntdb->file->map_size + addition))
+			;
+
+		/* now fill the file with something. This ensures that the
+		   file isn't sparse, which would be very bad if we ran out of
+		   disk. This must be done with write, not via mmap */
+		memset(buf, 0x43, sizeof(buf));
+		ecode = fill(ntdb, buf, sizeof(buf), ntdb->file->map_size,
+			     addition);
+		if (ecode != NTDB_SUCCESS)
+			return ecode;
+		ntdb->file->map_size += addition;
+		return ntdb_mmap(ntdb);
+	}
+}
+
+const void *ntdb_access_read(struct ntdb_context *ntdb,
+			    ntdb_off_t off, ntdb_len_t len, bool convert)
+{
+	void *ret = NULL;
+
+	if (likely(!(ntdb->flags & NTDB_CONVERT))) {
+		ret = ntdb->io->direct(ntdb, off, len, false);
+
+		if (NTDB_PTR_IS_ERR(ret)) {
+			return ret;
+		}
+	}
+	if (!ret) {
+		struct ntdb_access_hdr *hdr;
+		hdr = _ntdb_alloc_read(ntdb, off, len, sizeof(*hdr));
+		if (NTDB_PTR_IS_ERR(hdr)) {
+			return hdr;
+		}
+		hdr->next = ntdb->access;
+		ntdb->access = hdr;
+		ret = hdr + 1;
+		if (convert) {
+			ntdb_convert(ntdb, (void *)ret, len);
+		}
+	} else
+		ntdb->direct_access++;
+
+	return ret;
+}
+
+void *ntdb_access_write(struct ntdb_context *ntdb,
+		       ntdb_off_t off, ntdb_len_t len, bool convert)
+{
+	void *ret = NULL;
+
+	if (ntdb->flags & NTDB_RDONLY) {
+		ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_USE_ERROR,
+			   "Write to read-only database");
+		return NTDB_ERR_PTR(NTDB_ERR_RDONLY);
+	}
+
+	if (likely(!(ntdb->flags & NTDB_CONVERT))) {
+		ret = ntdb->io->direct(ntdb, off, len, true);
+
+		if (NTDB_PTR_IS_ERR(ret)) {
+			return ret;
+		}
+	}
+
+	if (!ret) {
+		struct ntdb_access_hdr *hdr;
+		hdr = _ntdb_alloc_read(ntdb, off, len, sizeof(*hdr));
+		if (NTDB_PTR_IS_ERR(hdr)) {
+			return hdr;
+		}
+		hdr->next = ntdb->access;
+		ntdb->access = hdr;
+		hdr->off = off;
+		hdr->len = len;
+		hdr->convert = convert;
+		ret = hdr + 1;
+		if (convert)
+			ntdb_convert(ntdb, (void *)ret, len);
+	} else
+		ntdb->direct_access++;
+
+	return ret;
+}
+
+static struct ntdb_access_hdr **find_hdr(struct ntdb_context *ntdb, const void *p)
+{
+	struct ntdb_access_hdr **hp;
+
+	for (hp = &ntdb->access; *hp; hp = &(*hp)->next) {
+		if (*hp + 1 == p)
+			return hp;
+	}
+	return NULL;
+}
+
+void ntdb_access_release(struct ntdb_context *ntdb, const void *p)
+{
+	struct ntdb_access_hdr *hdr, **hp = find_hdr(ntdb, p);
+
+	if (hp) {
+		hdr = *hp;
+		*hp = hdr->next;
+		free(hdr);
+	} else
+		ntdb->direct_access--;
+}
+
+enum NTDB_ERROR ntdb_access_commit(struct ntdb_context *ntdb, void *p)
+{
+	struct ntdb_access_hdr *hdr, **hp = find_hdr(ntdb, p);
+	enum NTDB_ERROR ecode;
+
+	if (hp) {
+		hdr = *hp;
+		if (hdr->convert)
+			ecode = ntdb_write_convert(ntdb, hdr->off, p, hdr->len);
+		else
+			ecode = ntdb_write(ntdb, hdr->off, p, hdr->len);
+		*hp = hdr->next;
+		free(hdr);
+	} else {
+		ntdb->direct_access--;
+		ecode = NTDB_SUCCESS;
+	}
+
+	return ecode;
+}
+
+static void *ntdb_direct(struct ntdb_context *ntdb, ntdb_off_t off, size_t len,
+			bool write_mode)
+{
+	enum NTDB_ERROR ecode;
+
+	if (unlikely(!ntdb->file->map_ptr))
+		return NULL;
+
+	ecode = ntdb_oob(ntdb, off, len, false);
+	if (unlikely(ecode != NTDB_SUCCESS))
+		return NTDB_ERR_PTR(ecode);
+	return (char *)ntdb->file->map_ptr + off;
+}
+
+void ntdb_inc_seqnum(struct ntdb_context *ntdb)
+{
+	ntdb_off_t seq;
+
+	if (likely(!(ntdb->flags & NTDB_CONVERT))) {
+		int64_t *direct;
+
+		direct = ntdb->io->direct(ntdb,
+					 offsetof(struct ntdb_header, seqnum),
+					 sizeof(*direct), true);
+		if (likely(direct)) {
+			/* Don't let it go negative, even briefly */
+			if (unlikely((*direct) + 1) < 0)
+				*direct = 0;
+			(*direct)++;
+			return;
+		}
+	}
+
+	seq = ntdb_read_off(ntdb, offsetof(struct ntdb_header, seqnum));
+	if (!NTDB_OFF_IS_ERR(seq)) {
+		seq++;
+		if (unlikely((int64_t)seq < 0))
+			seq = 0;
+		ntdb_write_off(ntdb, offsetof(struct ntdb_header, seqnum), seq);
+	}
+}
+
+static const struct ntdb_methods io_methods = {
+	ntdb_read,
+	ntdb_write,
+	ntdb_oob,
+	ntdb_expand_file,
+	ntdb_direct,
+};
+
+/*
+  initialise the default methods table
+*/
+void ntdb_io_init(struct ntdb_context *ntdb)
+{
+	ntdb->io = &io_methods;
+}

lib/ntdb/lock.c

@@ -0,0 +1,883 @@
+ /*
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2005
+   Copyright (C) Paul `Rusty' Russell		   2000
+   Copyright (C) Jeremy Allison			   2000-2003
+
+     ** NOTE! The following LGPL license applies to the ntdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "private.h"
+#include <assert.h>
+#include <ccan/build_assert/build_assert.h>
+
+/* If we were threaded, we could wait for unlock, but we're not, so fail. */
+enum NTDB_ERROR owner_conflict(struct ntdb_context *ntdb, const char *call)
+{
+	return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+			  "%s: lock owned by another ntdb in this process.",
+			  call);
+}
+
+/* If we fork, we no longer really own locks. */
+bool check_lock_pid(struct ntdb_context *ntdb, const char *call, bool log)
+{
+	/* No locks?  No problem! */
+	if (ntdb->file->allrecord_lock.count == 0
+	    && ntdb->file->num_lockrecs == 0) {
+		return true;
+	}
+
+	/* No fork?  No problem! */
+	if (ntdb->file->locker == getpid()) {
+		return true;
+	}
+
+	if (log) {
+		ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+			   "%s: fork() detected after lock acquisition!"
+			   " (%u vs %u)", call, ntdb->file->locker, getpid());
+	}
+	return false;
+}
+
+int ntdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag,
+		   void *unused)
+{
+	struct flock fl;
+	int ret;
+
+	do {
+		fl.l_type = rw;
+		fl.l_whence = SEEK_SET;
+		fl.l_start = off;
+		fl.l_len = len;
+
+		if (waitflag)
+			ret = fcntl(fd, F_SETLKW, &fl);
+		else
+			ret = fcntl(fd, F_SETLK, &fl);
+	} while (ret != 0 && errno == EINTR);
+	return ret;
+}
+
+int ntdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *unused)
+{
+	struct flock fl;
+	int ret;
+
+	do {
+		fl.l_type = F_UNLCK;
+		fl.l_whence = SEEK_SET;
+		fl.l_start = off;
+		fl.l_len = len;
+
+		ret = fcntl(fd, F_SETLKW, &fl);
+	} while (ret != 0 && errno == EINTR);
+	return ret;
+}
+
+static int lock(struct ntdb_context *ntdb,
+		      int rw, off_t off, off_t len, bool waitflag)
+{
+	int ret;
+	if (ntdb->file->allrecord_lock.count == 0
+	    && ntdb->file->num_lockrecs == 0) {
+		ntdb->file->locker = getpid();
+	}
+
+	ntdb->stats.lock_lowlevel++;
+	ret = ntdb->lock_fn(ntdb->file->fd, rw, off, len, waitflag,
+			   ntdb->lock_data);
+	if (!waitflag) {
+		ntdb->stats.lock_nonblock++;
+		if (ret != 0)
+			ntdb->stats.lock_nonblock_fail++;
+	}
+	return ret;
+}
+
+static int unlock(struct ntdb_context *ntdb, int rw, off_t off, off_t len)
+{
+#if 0 /* Check they matched up locks and unlocks correctly. */
+	char line[80];
+	FILE *locks;
+	bool found = false;
+
+	locks = fopen("/proc/locks", "r");
+
+	while (fgets(line, 80, locks)) {
+		char *p;
+		int type, start, l;
+
+		/* eg. 1: FLOCK  ADVISORY  WRITE 2440 08:01:2180826 0 EOF */
+		p = strchr(line, ':') + 1;
+		if (strncmp(p, " POSIX  ADVISORY  ", strlen(" POSIX  ADVISORY  ")))
+			continue;
+		p += strlen(" FLOCK  ADVISORY  ");
+		if (strncmp(p, "READ  ", strlen("READ  ")) == 0)
+			type = F_RDLCK;
+		else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
+			type = F_WRLCK;
+		else
+			abort();
+		p += 6;
+		if (atoi(p) != getpid())
+			continue;
+		p = strchr(strchr(p, ' ') + 1, ' ') + 1;
+		start = atoi(p);
+		p = strchr(p, ' ') + 1;
+		if (strncmp(p, "EOF", 3) == 0)
+			l = 0;
+		else
+			l = atoi(p) - start + 1;
+
+		if (off == start) {
+			if (len != l) {
+				fprintf(stderr, "Len %u should be %u: %s",
+					(int)len, l, line);
+				abort();
+			}
+			if (type != rw) {
+				fprintf(stderr, "Type %s wrong: %s",
+					rw == F_RDLCK ? "READ" : "WRITE", line);
+				abort();
+			}
+			found = true;
+			break;
+		}
+	}
+
+	if (!found) {
+		fprintf(stderr, "Unlock on %u@%u not found!",
+			(int)off, (int)len);
+		abort();
+	}
+
+	fclose(locks);
+#endif
+
+	return ntdb->unlock_fn(ntdb->file->fd, rw, off, len, ntdb->lock_data);
+}
+
+/* a byte range locking function - return 0 on success
+   this functions locks len bytes at the specified offset.
+
+   note that a len of zero means lock to end of file
+*/
+static enum NTDB_ERROR ntdb_brlock(struct ntdb_context *ntdb,
+				 int rw_type, ntdb_off_t offset, ntdb_off_t len,
+				 enum ntdb_lock_flags flags)
+{
+	int ret;
+
+	if (ntdb->flags & NTDB_NOLOCK) {
+		return NTDB_SUCCESS;
+	}
+
+	if (rw_type == F_WRLCK && (ntdb->flags & NTDB_RDONLY)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_USE_ERROR,
+				  "Write lock attempted on read-only database");
+	}
+
+	/* A 32 bit system cannot open a 64-bit file, but it could have
+	 * expanded since then: check here. */
+	if ((size_t)(offset + len) != offset + len) {
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "ntdb_brlock: lock on giant offset %llu",
+				  (long long)(offset + len));
+	}
+
+	ret = lock(ntdb, rw_type, offset, len, flags & NTDB_LOCK_WAIT);
+	if (ret != 0) {
+		/* Generic lock error. errno set by fcntl.
+		 * EAGAIN is an expected return from non-blocking
+		 * locks. */
+		if (!(flags & NTDB_LOCK_PROBE)
+		    && (errno != EAGAIN && errno != EINTR)) {
+			ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				   "ntdb_brlock failed (fd=%d) at"
+				   " offset %zu rw_type=%d flags=%d len=%zu:"
+				   " %s",
+				   ntdb->file->fd, (size_t)offset, rw_type,
+				   flags, (size_t)len, strerror(errno));
+		}
+		return NTDB_ERR_LOCK;
+	}
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR ntdb_brunlock(struct ntdb_context *ntdb,
+				   int rw_type, ntdb_off_t offset, size_t len)
+{
+	if (ntdb->flags & NTDB_NOLOCK) {
+		return NTDB_SUCCESS;
+	}
+
+	if (!check_lock_pid(ntdb, "ntdb_brunlock", true))
+		return NTDB_ERR_LOCK;
+
+	if (unlock(ntdb, rw_type, offset, len) == -1) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_brunlock failed (fd=%d) at offset %zu"
+				  " rw_type=%d len=%zu: %s",
+				  ntdb->file->fd, (size_t)offset, rw_type,
+				  (size_t)len, strerror(errno));
+	}
+	return NTDB_SUCCESS;
+}
+
+/*
+  upgrade a read lock to a write lock. This needs to be handled in a
+  special way as some OSes (such as solaris) have too conservative
+  deadlock detection and claim a deadlock when progress can be
+  made. For those OSes we may loop for a while.
+*/
+enum NTDB_ERROR ntdb_allrecord_upgrade(struct ntdb_context *ntdb, off_t start)
+{
+	int count = 1000;
+
+	if (!check_lock_pid(ntdb, "ntdb_transaction_prepare_commit", true))
+		return NTDB_ERR_LOCK;
+
+	if (ntdb->file->allrecord_lock.count != 1) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_allrecord_upgrade failed:"
+				  " count %u too high",
+				  ntdb->file->allrecord_lock.count);
+	}
+
+	if (ntdb->file->allrecord_lock.off != 1) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_allrecord_upgrade failed:"
+				  " already upgraded?");
+	}
+
+	if (ntdb->file->allrecord_lock.owner != ntdb) {
+		return owner_conflict(ntdb, "ntdb_allrecord_upgrade");
+	}
+
+	while (count--) {
+		struct timeval tv;
+		if (ntdb_brlock(ntdb, F_WRLCK, start, 0,
+			       NTDB_LOCK_WAIT|NTDB_LOCK_PROBE) == NTDB_SUCCESS) {
+			ntdb->file->allrecord_lock.ltype = F_WRLCK;
+			ntdb->file->allrecord_lock.off = 0;
+			return NTDB_SUCCESS;
+		}
+		if (errno != EDEADLK) {
+			break;
+		}
+		/* sleep for as short a time as we can - more portable than usleep() */
+		tv.tv_sec = 0;
+		tv.tv_usec = 1;
+		select(0, NULL, NULL, NULL, &tv);
+	}
+
+	if (errno != EAGAIN && errno != EINTR)
+		ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+			   "ntdb_allrecord_upgrade failed");
+	return NTDB_ERR_LOCK;
+}
+
+static struct ntdb_lock *find_nestlock(struct ntdb_context *ntdb, ntdb_off_t offset,
+				      const struct ntdb_context *owner)
+{
+	unsigned int i;
+
+	for (i=0; i<ntdb->file->num_lockrecs; i++) {
+		if (ntdb->file->lockrecs[i].off == offset) {
+			if (owner && ntdb->file->lockrecs[i].owner != owner)
+				return NULL;
+			return &ntdb->file->lockrecs[i];
+		}
+	}
+	return NULL;
+}
+
+enum NTDB_ERROR ntdb_lock_and_recover(struct ntdb_context *ntdb)
+{
+	enum NTDB_ERROR ecode;
+
+	if (!check_lock_pid(ntdb, "ntdb_transaction_prepare_commit", true))
+		return NTDB_ERR_LOCK;
+
+	ecode = ntdb_allrecord_lock(ntdb, F_WRLCK, NTDB_LOCK_WAIT|NTDB_LOCK_NOCHECK,
+				   false);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	ecode = ntdb_lock_open(ntdb, F_WRLCK, NTDB_LOCK_WAIT|NTDB_LOCK_NOCHECK);
+	if (ecode != NTDB_SUCCESS) {
+		ntdb_allrecord_unlock(ntdb, F_WRLCK);
+		return ecode;
+	}
+	ecode = ntdb_transaction_recover(ntdb);
+	ntdb_unlock_open(ntdb, F_WRLCK);
+	ntdb_allrecord_unlock(ntdb, F_WRLCK);
+
+	return ecode;
+}
+
+/* lock an offset in the database. */
+static enum NTDB_ERROR ntdb_nest_lock(struct ntdb_context *ntdb,
+				    ntdb_off_t offset, int ltype,
+				    enum ntdb_lock_flags flags)
+{
+	struct ntdb_lock *new_lck;
+	enum NTDB_ERROR ecode;
+
+	if (offset > (NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE
+		      + ntdb->file->map_size / 8)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_nest_lock: invalid offset %zu ltype=%d",
+				  (size_t)offset, ltype);
+	}
+
+	if (ntdb->flags & NTDB_NOLOCK)
+		return NTDB_SUCCESS;
+
+	if (!check_lock_pid(ntdb, "ntdb_nest_lock", true)) {
+		return NTDB_ERR_LOCK;
+	}
+
+	ntdb->stats.locks++;
+
+	new_lck = find_nestlock(ntdb, offset, NULL);
+	if (new_lck) {
+		if (new_lck->owner != ntdb) {
+			return owner_conflict(ntdb, "ntdb_nest_lock");
+		}
+
+		if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
+			return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+					  "ntdb_nest_lock:"
+					  " offset %zu has read lock",
+					  (size_t)offset);
+		}
+		/* Just increment the struct, posix locks don't stack. */
+		new_lck->count++;
+		return NTDB_SUCCESS;
+	}
+
+#if 0
+	if (ntdb->file->num_lockrecs
+	    && offset >= NTDB_HASH_LOCK_START
+	    && offset < NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_nest_lock: already have a hash lock?");
+	}
+#endif
+
+	new_lck = (struct ntdb_lock *)realloc(
+		ntdb->file->lockrecs,
+		sizeof(*ntdb->file->lockrecs) * (ntdb->file->num_lockrecs+1));
+	if (new_lck == NULL) {
+		return ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+				  "ntdb_nest_lock:"
+				  " unable to allocate %zu lock struct",
+				  ntdb->file->num_lockrecs + 1);
+	}
+	ntdb->file->lockrecs = new_lck;
+
+	/* Since fcntl locks don't nest, we do a lock for the first one,
+	   and simply bump the count for future ones */
+	ecode = ntdb_brlock(ntdb, ltype, offset, 1, flags);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	/* First time we grab a lock, perhaps someone died in commit? */
+	if (!(flags & NTDB_LOCK_NOCHECK)
+	    && ntdb->file->num_lockrecs == 0) {
+		ntdb_bool_err berr = ntdb_needs_recovery(ntdb);
+		if (berr != false) {
+			ntdb_brunlock(ntdb, ltype, offset, 1);
+
+			if (berr < 0)
+				return NTDB_OFF_TO_ERR(berr);
+			ecode = ntdb_lock_and_recover(ntdb);
+			if (ecode == NTDB_SUCCESS) {
+				ecode = ntdb_brlock(ntdb, ltype, offset, 1,
+						   flags);
+			}
+			if (ecode != NTDB_SUCCESS) {
+				return ecode;
+			}
+		}
+	}
+
+	ntdb->file->lockrecs[ntdb->file->num_lockrecs].owner = ntdb;
+	ntdb->file->lockrecs[ntdb->file->num_lockrecs].off = offset;
+	ntdb->file->lockrecs[ntdb->file->num_lockrecs].count = 1;
+	ntdb->file->lockrecs[ntdb->file->num_lockrecs].ltype = ltype;
+	ntdb->file->num_lockrecs++;
+
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR ntdb_nest_unlock(struct ntdb_context *ntdb,
+				      ntdb_off_t off, int ltype)
+{
+	struct ntdb_lock *lck;
+	enum NTDB_ERROR ecode;
+
+	if (ntdb->flags & NTDB_NOLOCK)
+		return NTDB_SUCCESS;
+
+	lck = find_nestlock(ntdb, off, ntdb);
+	if ((lck == NULL) || (lck->count == 0)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_nest_unlock: no lock for %zu",
+				  (size_t)off);
+	}
+
+	if (lck->count > 1) {
+		lck->count--;
+		return NTDB_SUCCESS;
+	}
+
+	/*
+	 * This lock has count==1 left, so we need to unlock it in the
+	 * kernel. We don't bother with decrementing the in-memory array
+	 * element, we're about to overwrite it with the last array element
+	 * anyway.
+	 */
+	ecode = ntdb_brunlock(ntdb, ltype, off, 1);
+
+	/*
+	 * Shrink the array by overwriting the element just unlocked with the
+	 * last array element.
+	 */
+	*lck = ntdb->file->lockrecs[--ntdb->file->num_lockrecs];
+
+	return ecode;
+}
+
+/*
+  get the transaction lock
+ */
+enum NTDB_ERROR ntdb_transaction_lock(struct ntdb_context *ntdb, int ltype)
+{
+	return ntdb_nest_lock(ntdb, NTDB_TRANSACTION_LOCK, ltype, NTDB_LOCK_WAIT);
+}
+
+/*
+  release the transaction lock
+ */
+void ntdb_transaction_unlock(struct ntdb_context *ntdb, int ltype)
+{
+	ntdb_nest_unlock(ntdb, NTDB_TRANSACTION_LOCK, ltype);
+}
+
+/* We only need to lock individual bytes, but Linux merges consecutive locks
+ * so we lock in contiguous ranges. */
+static enum NTDB_ERROR ntdb_lock_gradual(struct ntdb_context *ntdb,
+				       int ltype, enum ntdb_lock_flags flags,
+				       ntdb_off_t off, ntdb_off_t len)
+{
+	enum NTDB_ERROR ecode;
+	enum ntdb_lock_flags nb_flags = (flags & ~NTDB_LOCK_WAIT);
+
+	if (len <= 1) {
+		/* 0 would mean to end-of-file... */
+		assert(len != 0);
+		/* Single hash.  Just do blocking lock. */
+		return ntdb_brlock(ntdb, ltype, off, len, flags);
+	}
+
+	/* First we try non-blocking. */
+	ecode = ntdb_brlock(ntdb, ltype, off, len, nb_flags);
+	if (ecode != NTDB_ERR_LOCK) {
+		return ecode;
+	}
+
+	/* Try locking first half, then second. */
+	ecode = ntdb_lock_gradual(ntdb, ltype, flags, off, len / 2);
+	if (ecode != NTDB_SUCCESS)
+		return ecode;
+
+	ecode = ntdb_lock_gradual(ntdb, ltype, flags,
+				 off + len / 2, len - len / 2);
+	if (ecode != NTDB_SUCCESS) {
+		ntdb_brunlock(ntdb, ltype, off, len / 2);
+	}
+	return ecode;
+}
+
+/* lock/unlock entire database.  It can only be upgradable if you have some
+ * other way of guaranteeing exclusivity (ie. transaction write lock). */
+enum NTDB_ERROR ntdb_allrecord_lock(struct ntdb_context *ntdb, int ltype,
+				  enum ntdb_lock_flags flags, bool upgradable)
+{
+	enum NTDB_ERROR ecode;
+	ntdb_bool_err berr;
+
+	if (ntdb->flags & NTDB_NOLOCK)
+		return NTDB_SUCCESS;
+
+	if (!check_lock_pid(ntdb, "ntdb_allrecord_lock", true)) {
+		return NTDB_ERR_LOCK;
+	}
+
+	if (ntdb->file->allrecord_lock.count) {
+		if (ntdb->file->allrecord_lock.owner != ntdb) {
+			return owner_conflict(ntdb, "ntdb_allrecord_lock");
+		}
+
+		if (ltype == F_RDLCK
+		    || ntdb->file->allrecord_lock.ltype == F_WRLCK) {
+			ntdb->file->allrecord_lock.count++;
+			return NTDB_SUCCESS;
+		}
+
+		/* a global lock of a different type exists */
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+				  "ntdb_allrecord_lock: already have %s lock",
+				  ntdb->file->allrecord_lock.ltype == F_RDLCK
+				  ? "read" : "write");
+	}
+
+	if (ntdb_has_hash_locks(ntdb)) {
+		/* can't combine global and chain locks */
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+				  "ntdb_allrecord_lock:"
+				  " already have chain lock");
+	}
+
+	if (upgradable && ltype != F_RDLCK) {
+		/* ntdb error: you can't upgrade a write lock! */
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_allrecord_lock:"
+				  " can't upgrade a write lock");
+	}
+
+	ntdb->stats.locks++;
+again:
+	/* Lock hashes, gradually. */
+	ecode = ntdb_lock_gradual(ntdb, ltype, flags, NTDB_HASH_LOCK_START,
+				 NTDB_HASH_LOCK_RANGE);
+	if (ecode != NTDB_SUCCESS)
+		return ecode;
+
+	/* Lock free tables: there to end of file. */
+	ecode = ntdb_brlock(ntdb, ltype,
+			   NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE,
+			   0, flags);
+	if (ecode != NTDB_SUCCESS) {
+		ntdb_brunlock(ntdb, ltype, NTDB_HASH_LOCK_START,
+			     NTDB_HASH_LOCK_RANGE);
+		return ecode;
+	}
+
+	ntdb->file->allrecord_lock.owner = ntdb;
+	ntdb->file->allrecord_lock.count = 1;
+	/* If it's upgradable, it's actually exclusive so we can treat
+	 * it as a write lock. */
+	ntdb->file->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
+	ntdb->file->allrecord_lock.off = upgradable;
+
+	/* Now check for needing recovery. */
+	if (flags & NTDB_LOCK_NOCHECK)
+		return NTDB_SUCCESS;
+
+	berr = ntdb_needs_recovery(ntdb);
+	if (likely(berr == false))
+		return NTDB_SUCCESS;
+
+	ntdb_allrecord_unlock(ntdb, ltype);
+	if (berr < 0)
+		return NTDB_OFF_TO_ERR(berr);
+	ecode = ntdb_lock_and_recover(ntdb);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+	goto again;
+}
+
+enum NTDB_ERROR ntdb_lock_open(struct ntdb_context *ntdb,
+			     int ltype, enum ntdb_lock_flags flags)
+{
+	return ntdb_nest_lock(ntdb, NTDB_OPEN_LOCK, ltype, flags);
+}
+
+void ntdb_unlock_open(struct ntdb_context *ntdb, int ltype)
+{
+	ntdb_nest_unlock(ntdb, NTDB_OPEN_LOCK, ltype);
+}
+
+bool ntdb_has_open_lock(struct ntdb_context *ntdb)
+{
+	return !(ntdb->flags & NTDB_NOLOCK)
+		&& find_nestlock(ntdb, NTDB_OPEN_LOCK, ntdb) != NULL;
+}
+
+enum NTDB_ERROR ntdb_lock_expand(struct ntdb_context *ntdb, int ltype)
+{
+	/* Lock doesn't protect data, so don't check (we recurse if we do!) */
+	return ntdb_nest_lock(ntdb, NTDB_EXPANSION_LOCK, ltype,
+			     NTDB_LOCK_WAIT | NTDB_LOCK_NOCHECK);
+}
+
+void ntdb_unlock_expand(struct ntdb_context *ntdb, int ltype)
+{
+	ntdb_nest_unlock(ntdb, NTDB_EXPANSION_LOCK, ltype);
+}
+
+/* unlock entire db */
+void ntdb_allrecord_unlock(struct ntdb_context *ntdb, int ltype)
+{
+	if (ntdb->flags & NTDB_NOLOCK)
+		return;
+
+	if (ntdb->file->allrecord_lock.count == 0) {
+		ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+			   "ntdb_allrecord_unlock: not locked!");
+		return;
+	}
+
+	if (ntdb->file->allrecord_lock.owner != ntdb) {
+		ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+			   "ntdb_allrecord_unlock: not locked by us!");
+		return;
+	}
+
+	/* Upgradable locks are marked as write locks. */
+	if (ntdb->file->allrecord_lock.ltype != ltype
+	    && (!ntdb->file->allrecord_lock.off || ltype != F_RDLCK)) {
+		ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+			   "ntdb_allrecord_unlock: have %s lock",
+			   ntdb->file->allrecord_lock.ltype == F_RDLCK
+			   ? "read" : "write");
+		return;
+	}
+
+	if (ntdb->file->allrecord_lock.count > 1) {
+		ntdb->file->allrecord_lock.count--;
+		return;
+	}
+
+	ntdb->file->allrecord_lock.count = 0;
+	ntdb->file->allrecord_lock.ltype = 0;
+
+	ntdb_brunlock(ntdb, ltype, NTDB_HASH_LOCK_START, 0);
+}
+
+bool ntdb_has_expansion_lock(struct ntdb_context *ntdb)
+{
+	return find_nestlock(ntdb, NTDB_EXPANSION_LOCK, ntdb) != NULL;
+}
+
+bool ntdb_has_hash_locks(struct ntdb_context *ntdb)
+{
+	unsigned int i;
+
+	for (i=0; i<ntdb->file->num_lockrecs; i++) {
+		if (ntdb->file->lockrecs[i].off >= NTDB_HASH_LOCK_START
+		    && ntdb->file->lockrecs[i].off < (NTDB_HASH_LOCK_START
+						     + NTDB_HASH_LOCK_RANGE))
+			return true;
+	}
+	return false;
+}
+
+static bool ntdb_has_free_lock(struct ntdb_context *ntdb)
+{
+	unsigned int i;
+
+	if (ntdb->flags & NTDB_NOLOCK)
+		return false;
+
+	for (i=0; i<ntdb->file->num_lockrecs; i++) {
+		if (ntdb->file->lockrecs[i].off
+		    > NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE)
+			return true;
+	}
+	return false;
+}
+
+enum NTDB_ERROR ntdb_lock_hashes(struct ntdb_context *ntdb,
+			       ntdb_off_t hash_lock,
+			       ntdb_len_t hash_range,
+			       int ltype, enum ntdb_lock_flags waitflag)
+{
+	/* FIXME: Do this properly, using hlock_range */
+	unsigned l = NTDB_HASH_LOCK_START
+		+ (hash_lock >> (64 - NTDB_HASH_LOCK_RANGE_BITS));
+
+	/* a allrecord lock allows us to avoid per chain locks */
+	if (ntdb->file->allrecord_lock.count) {
+		if (!check_lock_pid(ntdb, "ntdb_lock_hashes", true))
+			return NTDB_ERR_LOCK;
+
+		if (ntdb->file->allrecord_lock.owner != ntdb)
+			return owner_conflict(ntdb, "ntdb_lock_hashes");
+		if (ltype == ntdb->file->allrecord_lock.ltype
+		    || ltype == F_RDLCK) {
+			return NTDB_SUCCESS;
+		}
+
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+				  "ntdb_lock_hashes:"
+				  " already have %s allrecordlock",
+				  ntdb->file->allrecord_lock.ltype == F_RDLCK
+				  ? "read" : "write");
+	}
+
+	if (ntdb_has_free_lock(ntdb)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_lock_hashes: already have free lock");
+	}
+
+	if (ntdb_has_expansion_lock(ntdb)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_lock_hashes:"
+				  " already have expansion lock");
+	}
+
+	return ntdb_nest_lock(ntdb, l, ltype, waitflag);
+}
+
+enum NTDB_ERROR ntdb_unlock_hashes(struct ntdb_context *ntdb,
+				 ntdb_off_t hash_lock,
+				 ntdb_len_t hash_range, int ltype)
+{
+	unsigned l = NTDB_HASH_LOCK_START
+		+ (hash_lock >> (64 - NTDB_HASH_LOCK_RANGE_BITS));
+
+	if (ntdb->flags & NTDB_NOLOCK)
+		return 0;
+
+	/* a allrecord lock allows us to avoid per chain locks */
+	if (ntdb->file->allrecord_lock.count) {
+		if (ntdb->file->allrecord_lock.ltype == F_RDLCK
+		    && ltype == F_WRLCK) {
+			return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+					  "ntdb_unlock_hashes RO allrecord!");
+		}
+		if (ntdb->file->allrecord_lock.owner != ntdb) {
+			return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
+					  "ntdb_unlock_hashes:"
+					  " not locked by us!");
+		}
+		return NTDB_SUCCESS;
+	}
+
+	return ntdb_nest_unlock(ntdb, l, ltype);
+}
+
+/* Hash locks use NTDB_HASH_LOCK_START + the next 30 bits.
+ * Then we begin; bucket offsets are sizeof(ntdb_len_t) apart, so we divide.
+ * The result is that on 32 bit systems we don't use lock values > 2^31 on
+ * files that are less than 4GB.
+ */
+static ntdb_off_t free_lock_off(ntdb_off_t b_off)
+{
+	return NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE
+		+ b_off / sizeof(ntdb_off_t);
+}
+
+enum NTDB_ERROR ntdb_lock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off,
+				    enum ntdb_lock_flags waitflag)
+{
+	assert(b_off >= sizeof(struct ntdb_header));
+
+	if (ntdb->flags & NTDB_NOLOCK)
+		return 0;
+
+	/* a allrecord lock allows us to avoid per chain locks */
+	if (ntdb->file->allrecord_lock.count) {
+		if (!check_lock_pid(ntdb, "ntdb_lock_free_bucket", true))
+			return NTDB_ERR_LOCK;
+
+		if (ntdb->file->allrecord_lock.owner != ntdb) {
+			return owner_conflict(ntdb, "ntdb_lock_free_bucket");
+		}
+
+		if (ntdb->file->allrecord_lock.ltype == F_WRLCK)
+			return 0;
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_lock_free_bucket with"
+				  " read-only allrecordlock!");
+	}
+
+#if 0 /* FIXME */
+	if (ntdb_has_expansion_lock(ntdb)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
+				  "ntdb_lock_free_bucket:"
+				  " already have expansion lock");
+	}
+#endif
+
+	return ntdb_nest_lock(ntdb, free_lock_off(b_off), F_WRLCK, waitflag);
+}
+
+void ntdb_unlock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off)
+{
+	if (ntdb->file->allrecord_lock.count)
+		return;
+
+	ntdb_nest_unlock(ntdb, free_lock_off(b_off), F_WRLCK);
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_lockall(struct ntdb_context *ntdb)
+{
+	return ntdb_allrecord_lock(ntdb, F_WRLCK, NTDB_LOCK_WAIT, false);
+}
+
+_PUBLIC_ void ntdb_unlockall(struct ntdb_context *ntdb)
+{
+	ntdb_allrecord_unlock(ntdb, F_WRLCK);
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_lockall_read(struct ntdb_context *ntdb)
+{
+	return ntdb_allrecord_lock(ntdb, F_RDLCK, NTDB_LOCK_WAIT, false);
+}
+
+_PUBLIC_ void ntdb_unlockall_read(struct ntdb_context *ntdb)
+{
+	ntdb_allrecord_unlock(ntdb, F_RDLCK);
+}
+
+void ntdb_lock_cleanup(struct ntdb_context *ntdb)
+{
+	unsigned int i;
+
+	/* We don't want to warn: they're allowed to close ntdb after fork. */
+	if (!check_lock_pid(ntdb, "ntdb_close", false))
+		return;
+
+	while (ntdb->file->allrecord_lock.count
+	       && ntdb->file->allrecord_lock.owner == ntdb) {
+		ntdb_allrecord_unlock(ntdb, ntdb->file->allrecord_lock.ltype);
+	}
+
+	for (i=0; i<ntdb->file->num_lockrecs; i++) {
+		if (ntdb->file->lockrecs[i].owner == ntdb) {
+			ntdb_nest_unlock(ntdb,
+					ntdb->file->lockrecs[i].off,
+					ntdb->file->lockrecs[i].ltype);
+			i--;
+		}
+	}
+}

lib/ntdb/ntdb.c

@@ -0,0 +1,605 @@
+ /*
+   Trivial Database 2: fetch, store and misc routines.
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "private.h"
+#ifndef HAVE_LIBREPLACE
+#include <ccan/asprintf/asprintf.h>
+#include <stdarg.h>
+#endif
+
+static enum NTDB_ERROR update_rec_hdr(struct ntdb_context *ntdb,
+				     ntdb_off_t off,
+				     ntdb_len_t keylen,
+				     ntdb_len_t datalen,
+				     struct ntdb_used_record *rec,
+				     uint64_t h)
+{
+	uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
+	enum NTDB_ERROR ecode;
+
+	ecode = set_header(ntdb, rec, NTDB_USED_MAGIC, keylen, datalen,
+			   keylen + dataroom, h);
+	if (ecode == NTDB_SUCCESS) {
+		ecode = ntdb_write_convert(ntdb, off, rec, sizeof(*rec));
+	}
+	return ecode;
+}
+
+static enum NTDB_ERROR replace_data(struct ntdb_context *ntdb,
+				   struct hash_info *h,
+				   NTDB_DATA key, NTDB_DATA dbuf,
+				   ntdb_off_t old_off, ntdb_len_t old_room,
+				   bool growing)
+{
+	ntdb_off_t new_off;
+	enum NTDB_ERROR ecode;
+
+	/* Allocate a new record. */
+	new_off = alloc(ntdb, key.dsize, dbuf.dsize, h->h, NTDB_USED_MAGIC,
+			growing);
+	if (NTDB_OFF_IS_ERR(new_off)) {
+		return NTDB_OFF_TO_ERR(new_off);
+	}
+
+	/* We didn't like the existing one: remove it. */
+	if (old_off) {
+		ntdb->stats.frees++;
+		ecode = add_free_record(ntdb, old_off,
+					sizeof(struct ntdb_used_record)
+					+ key.dsize + old_room,
+					NTDB_LOCK_WAIT, true);
+		if (ecode == NTDB_SUCCESS)
+			ecode = replace_in_hash(ntdb, h, new_off);
+	} else {
+		ecode = add_to_hash(ntdb, h, new_off);
+	}
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	new_off += sizeof(struct ntdb_used_record);
+	ecode = ntdb->io->twrite(ntdb, new_off, key.dptr, key.dsize);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	new_off += key.dsize;
+	ecode = ntdb->io->twrite(ntdb, new_off, dbuf.dptr, dbuf.dsize);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	if (ntdb->flags & NTDB_SEQNUM)
+		ntdb_inc_seqnum(ntdb);
+
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR update_data(struct ntdb_context *ntdb,
+				  ntdb_off_t off,
+				  NTDB_DATA dbuf,
+				  ntdb_len_t extra)
+{
+	enum NTDB_ERROR ecode;
+
+	ecode = ntdb->io->twrite(ntdb, off, dbuf.dptr, dbuf.dsize);
+	if (ecode == NTDB_SUCCESS && extra) {
+		/* Put a zero in; future versions may append other data. */
+		ecode = ntdb->io->twrite(ntdb, off + dbuf.dsize, "", 1);
+	}
+	if (ntdb->flags & NTDB_SEQNUM)
+		ntdb_inc_seqnum(ntdb);
+
+	return ecode;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_store(struct ntdb_context *ntdb,
+			 NTDB_DATA key, NTDB_DATA dbuf, int flag)
+{
+	struct hash_info h;
+	ntdb_off_t off;
+	ntdb_len_t old_room = 0;
+	struct ntdb_used_record rec;
+	enum NTDB_ERROR ecode;
+
+	off = find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL);
+	if (NTDB_OFF_IS_ERR(off)) {
+		return ntdb->last_error = NTDB_OFF_TO_ERR(off);
+	}
+
+	/* Now we have lock on this hash bucket. */
+	if (flag == NTDB_INSERT) {
+		if (off) {
+			ecode = NTDB_ERR_EXISTS;
+			goto out;
+		}
+	} else {
+		if (off) {
+			old_room = rec_data_length(&rec)
+				+ rec_extra_padding(&rec);
+			if (old_room >= dbuf.dsize) {
+				/* Can modify in-place.  Easy! */
+				ecode = update_rec_hdr(ntdb, off,
+						       key.dsize, dbuf.dsize,
+						       &rec, h.h);
+				if (ecode != NTDB_SUCCESS) {
+					goto out;
+				}
+				ecode = update_data(ntdb,
+						    off + sizeof(rec)
+						    + key.dsize, dbuf,
+						    old_room - dbuf.dsize);
+				if (ecode != NTDB_SUCCESS) {
+					goto out;
+				}
+				ntdb_unlock_hashes(ntdb, h.hlock_start,
+						  h.hlock_range, F_WRLCK);
+				return ntdb->last_error = NTDB_SUCCESS;
+			}
+		} else {
+			if (flag == NTDB_MODIFY) {
+				/* if the record doesn't exist and we
+				   are in NTDB_MODIFY mode then we should fail
+				   the store */
+				ecode = NTDB_ERR_NOEXIST;
+				goto out;
+			}
+		}
+	}
+
+	/* If we didn't use the old record, this implies we're growing. */
+	ecode = replace_data(ntdb, &h, key, dbuf, off, old_room, off);
+out:
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_WRLCK);
+	return ntdb->last_error = ecode;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_append(struct ntdb_context *ntdb,
+			  NTDB_DATA key, NTDB_DATA dbuf)
+{
+	struct hash_info h;
+	ntdb_off_t off;
+	struct ntdb_used_record rec;
+	ntdb_len_t old_room = 0, old_dlen;
+	unsigned char *newdata;
+	NTDB_DATA new_dbuf;
+	enum NTDB_ERROR ecode;
+
+	off = find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL);
+	if (NTDB_OFF_IS_ERR(off)) {
+		return ntdb->last_error = NTDB_OFF_TO_ERR(off);
+	}
+
+	if (off) {
+		old_dlen = rec_data_length(&rec);
+		old_room = old_dlen + rec_extra_padding(&rec);
+
+		/* Fast path: can append in place. */
+		if (rec_extra_padding(&rec) >= dbuf.dsize) {
+			ecode = update_rec_hdr(ntdb, off, key.dsize,
+					       old_dlen + dbuf.dsize, &rec,
+					       h.h);
+			if (ecode != NTDB_SUCCESS) {
+				goto out;
+			}
+
+			off += sizeof(rec) + key.dsize + old_dlen;
+			ecode = update_data(ntdb, off, dbuf,
+					    rec_extra_padding(&rec));
+			goto out;
+		}
+
+		/* Slow path. */
+		newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
+		if (!newdata) {
+			ecode = ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+					   "ntdb_append:"
+					   " failed to allocate %zu bytes",
+					   (size_t)(key.dsize + old_dlen
+						    + dbuf.dsize));
+			goto out;
+		}
+		ecode = ntdb->io->tread(ntdb, off + sizeof(rec) + key.dsize,
+				       newdata, old_dlen);
+		if (ecode != NTDB_SUCCESS) {
+			goto out_free_newdata;
+		}
+		memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
+		new_dbuf.dptr = newdata;
+		new_dbuf.dsize = old_dlen + dbuf.dsize;
+	} else {
+		newdata = NULL;
+		new_dbuf = dbuf;
+	}
+
+	/* If they're using ntdb_append(), it implies they're growing record. */
+	ecode = replace_data(ntdb, &h, key, new_dbuf, off, old_room, true);
+
+out_free_newdata:
+	free(newdata);
+out:
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_WRLCK);
+	return ntdb->last_error = ecode;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_fetch(struct ntdb_context *ntdb, NTDB_DATA key,
+			 NTDB_DATA *data)
+{
+	ntdb_off_t off;
+	struct ntdb_used_record rec;
+	struct hash_info h;
+	enum NTDB_ERROR ecode;
+
+	off = find_and_lock(ntdb, key, F_RDLCK, &h, &rec, NULL);
+	if (NTDB_OFF_IS_ERR(off)) {
+		return ntdb->last_error = NTDB_OFF_TO_ERR(off);
+	}
+
+	if (!off) {
+		ecode = NTDB_ERR_NOEXIST;
+	} else {
+		data->dsize = rec_data_length(&rec);
+		data->dptr = ntdb_alloc_read(ntdb, off + sizeof(rec) + key.dsize,
+					    data->dsize);
+		if (NTDB_PTR_IS_ERR(data->dptr)) {
+			ecode = NTDB_PTR_ERR(data->dptr);
+		} else
+			ecode = NTDB_SUCCESS;
+	}
+
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_RDLCK);
+	return ntdb->last_error = ecode;
+}
+
+_PUBLIC_ bool ntdb_exists(struct ntdb_context *ntdb, NTDB_DATA key)
+{
+	ntdb_off_t off;
+	struct ntdb_used_record rec;
+	struct hash_info h;
+
+	off = find_and_lock(ntdb, key, F_RDLCK, &h, &rec, NULL);
+	if (NTDB_OFF_IS_ERR(off)) {
+		ntdb->last_error = NTDB_OFF_TO_ERR(off);
+		return false;
+	}
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_RDLCK);
+
+	ntdb->last_error = NTDB_SUCCESS;
+	return off ? true : false;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_delete(struct ntdb_context *ntdb, NTDB_DATA key)
+{
+	ntdb_off_t off;
+	struct ntdb_used_record rec;
+	struct hash_info h;
+	enum NTDB_ERROR ecode;
+
+	off = find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL);
+	if (NTDB_OFF_IS_ERR(off)) {
+		return ntdb->last_error = NTDB_OFF_TO_ERR(off);
+	}
+
+	if (!off) {
+		ecode = NTDB_ERR_NOEXIST;
+		goto unlock;
+	}
+
+	ecode = delete_from_hash(ntdb, &h);
+	if (ecode != NTDB_SUCCESS) {
+		goto unlock;
+	}
+
+	/* Free the deleted entry. */
+	ntdb->stats.frees++;
+	ecode = add_free_record(ntdb, off,
+				sizeof(struct ntdb_used_record)
+				+ rec_key_length(&rec)
+				+ rec_data_length(&rec)
+				+ rec_extra_padding(&rec),
+				NTDB_LOCK_WAIT, true);
+
+	if (ntdb->flags & NTDB_SEQNUM)
+		ntdb_inc_seqnum(ntdb);
+
+unlock:
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_WRLCK);
+	return ntdb->last_error = ecode;
+}
+
+_PUBLIC_ unsigned int ntdb_get_flags(struct ntdb_context *ntdb)
+{
+	return ntdb->flags;
+}
+
+static bool inside_transaction(const struct ntdb_context *ntdb)
+{
+	return ntdb->transaction != NULL;
+}
+
+static bool readonly_changable(struct ntdb_context *ntdb, const char *caller)
+{
+	if (inside_transaction(ntdb)) {
+		ntdb->last_error = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+					     NTDB_LOG_USE_ERROR,
+					     "%s: can't change"
+					     " NTDB_RDONLY inside transaction",
+					     caller);
+		return false;
+	}
+	return true;
+}
+
+_PUBLIC_ void ntdb_add_flag(struct ntdb_context *ntdb, unsigned flag)
+{
+	if (ntdb->flags & NTDB_INTERNAL) {
+		ntdb->last_error = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+					     NTDB_LOG_USE_ERROR,
+					     "ntdb_add_flag: internal db");
+		return;
+	}
+	switch (flag) {
+	case NTDB_NOLOCK:
+		ntdb->flags |= NTDB_NOLOCK;
+		break;
+	case NTDB_NOMMAP:
+		ntdb->flags |= NTDB_NOMMAP;
+#ifndef HAVE_INCOHERENT_MMAP
+		ntdb_munmap(ntdb->file);
+#endif
+		break;
+	case NTDB_NOSYNC:
+		ntdb->flags |= NTDB_NOSYNC;
+		break;
+	case NTDB_SEQNUM:
+		ntdb->flags |= NTDB_SEQNUM;
+		break;
+	case NTDB_ALLOW_NESTING:
+		ntdb->flags |= NTDB_ALLOW_NESTING;
+		break;
+	case NTDB_RDONLY:
+		if (readonly_changable(ntdb, "ntdb_add_flag"))
+			ntdb->flags |= NTDB_RDONLY;
+		break;
+	default:
+		ntdb->last_error = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+					     NTDB_LOG_USE_ERROR,
+					     "ntdb_add_flag: Unknown flag %u",
+					     flag);
+	}
+}
+
+_PUBLIC_ void ntdb_remove_flag(struct ntdb_context *ntdb, unsigned flag)
+{
+	if (ntdb->flags & NTDB_INTERNAL) {
+		ntdb->last_error = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+					     NTDB_LOG_USE_ERROR,
+					     "ntdb_remove_flag: internal db");
+		return;
+	}
+	switch (flag) {
+	case NTDB_NOLOCK:
+		ntdb->flags &= ~NTDB_NOLOCK;
+		break;
+	case NTDB_NOMMAP:
+		ntdb->flags &= ~NTDB_NOMMAP;
+#ifndef HAVE_INCOHERENT_MMAP
+		/* If mmap incoherent, we were mmaping anyway. */
+		ntdb_mmap(ntdb);
+#endif
+		break;
+	case NTDB_NOSYNC:
+		ntdb->flags &= ~NTDB_NOSYNC;
+		break;
+	case NTDB_SEQNUM:
+		ntdb->flags &= ~NTDB_SEQNUM;
+		break;
+	case NTDB_ALLOW_NESTING:
+		ntdb->flags &= ~NTDB_ALLOW_NESTING;
+		break;
+	case NTDB_RDONLY:
+		if ((ntdb->open_flags & O_ACCMODE) == O_RDONLY) {
+			ntdb->last_error = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+						     NTDB_LOG_USE_ERROR,
+						     "ntdb_remove_flag: can't"
+						     " remove NTDB_RDONLY on ntdb"
+						     " opened with O_RDONLY");
+			break;
+		}
+		if (readonly_changable(ntdb, "ntdb_remove_flag"))
+			ntdb->flags &= ~NTDB_RDONLY;
+		break;
+	default:
+		ntdb->last_error = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+					     NTDB_LOG_USE_ERROR,
+					     "ntdb_remove_flag: Unknown flag %u",
+					     flag);
+	}
+}
+
+_PUBLIC_ const char *ntdb_errorstr(enum NTDB_ERROR ecode)
+{
+	/* Gcc warns if you miss a case in the switch, so use that. */
+	switch (NTDB_ERR_TO_OFF(ecode)) {
+	case NTDB_ERR_TO_OFF(NTDB_SUCCESS): return "Success";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_CORRUPT): return "Corrupt database";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_IO): return "IO Error";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_LOCK): return "Locking error";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_OOM): return "Out of memory";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_EXISTS): return "Record exists";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_EINVAL): return "Invalid parameter";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_NOEXIST): return "Record does not exist";
+	case NTDB_ERR_TO_OFF(NTDB_ERR_RDONLY): return "write not permitted";
+	}
+	return "Invalid error code";
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_error(struct ntdb_context *ntdb)
+{
+	return ntdb->last_error;
+}
+
+enum NTDB_ERROR COLD ntdb_logerr(struct ntdb_context *ntdb,
+			       enum NTDB_ERROR ecode,
+			       enum ntdb_log_level level,
+			       const char *fmt, ...)
+{
+	char *message;
+	va_list ap;
+	size_t len;
+	/* ntdb_open paths care about errno, so save it. */
+	int saved_errno = errno;
+
+	if (!ntdb->log_fn)
+		return ecode;
+
+	va_start(ap, fmt);
+	len = vasprintf(&message, fmt, ap);
+	va_end(ap);
+
+	if (len < 0) {
+		ntdb->log_fn(ntdb, NTDB_LOG_ERROR, NTDB_ERR_OOM,
+			    "out of memory formatting message:", ntdb->log_data);
+		ntdb->log_fn(ntdb, level, ecode, fmt, ntdb->log_data);
+	} else {
+		ntdb->log_fn(ntdb, level, ecode, message, ntdb->log_data);
+		free(message);
+	}
+	errno = saved_errno;
+	return ecode;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_parse_record_(struct ntdb_context *ntdb,
+				 NTDB_DATA key,
+				 enum NTDB_ERROR (*parse)(NTDB_DATA k,
+							 NTDB_DATA d,
+							 void *data),
+				 void *data)
+{
+	ntdb_off_t off;
+	struct ntdb_used_record rec;
+	struct hash_info h;
+	enum NTDB_ERROR ecode;
+
+	off = find_and_lock(ntdb, key, F_RDLCK, &h, &rec, NULL);
+	if (NTDB_OFF_IS_ERR(off)) {
+		return ntdb->last_error = NTDB_OFF_TO_ERR(off);
+	}
+
+	if (!off) {
+		ecode = NTDB_ERR_NOEXIST;
+	} else {
+		const void *dptr;
+		dptr = ntdb_access_read(ntdb, off + sizeof(rec) + key.dsize,
+				       rec_data_length(&rec), false);
+		if (NTDB_PTR_IS_ERR(dptr)) {
+			ecode = NTDB_PTR_ERR(dptr);
+		} else {
+			NTDB_DATA d = ntdb_mkdata(dptr, rec_data_length(&rec));
+
+			ecode = parse(key, d, data);
+			ntdb_access_release(ntdb, dptr);
+		}
+	}
+
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_RDLCK);
+	return ntdb->last_error = ecode;
+}
+
+_PUBLIC_ const char *ntdb_name(const struct ntdb_context *ntdb)
+{
+	return ntdb->name;
+}
+
+_PUBLIC_ int64_t ntdb_get_seqnum(struct ntdb_context *ntdb)
+{
+	ntdb_off_t off;
+
+	off = ntdb_read_off(ntdb, offsetof(struct ntdb_header, seqnum));
+	if (NTDB_OFF_IS_ERR(off))
+		ntdb->last_error = NTDB_OFF_TO_ERR(off);
+	else
+		ntdb->last_error = NTDB_SUCCESS;
+	return off;
+}
+
+
+_PUBLIC_ int ntdb_fd(const struct ntdb_context *ntdb)
+{
+	return ntdb->file->fd;
+}
+
+struct traverse_state {
+	enum NTDB_ERROR error;
+	struct ntdb_context *dest_db;
+};
+
+/*
+  traverse function for repacking
+ */
+static int repack_traverse(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA data,
+			   struct traverse_state *state)
+{
+	state->error = ntdb_store(state->dest_db, key, data, NTDB_INSERT);
+	if (state->error != NTDB_SUCCESS) {
+		return -1;
+	}
+	return 0;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_repack(struct ntdb_context *ntdb)
+{
+	struct ntdb_context *tmp_db;
+	struct traverse_state state;
+
+	state.error = ntdb_transaction_start(ntdb);
+	if (state.error != NTDB_SUCCESS) {
+		return state.error;
+	}
+
+	tmp_db = ntdb_open("tmpdb", NTDB_INTERNAL, O_RDWR|O_CREAT, 0, NULL);
+	if (tmp_db == NULL) {
+		state.error = ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+					 __location__
+					 " Failed to create tmp_db");
+		ntdb_transaction_cancel(ntdb);
+		return ntdb->last_error = state.error;
+	}
+
+	state.dest_db = tmp_db;
+	if (ntdb_traverse(ntdb, repack_traverse, &state) < 0) {
+		goto fail;
+	}
+
+	state.error = ntdb_wipe_all(ntdb);
+	if (state.error != NTDB_SUCCESS) {
+		goto fail;
+	}
+
+	state.dest_db = ntdb;
+	if (ntdb_traverse(tmp_db, repack_traverse, &state) < 0) {
+		goto fail;
+	}
+
+	ntdb_close(tmp_db);
+	return ntdb_transaction_commit(ntdb);
+
+fail:
+	ntdb_transaction_cancel(ntdb);
+	ntdb_close(tmp_db);
+	return state.error;
+}

lib/ntdb/ntdb.h

@@ -0,0 +1,901 @@
+#ifndef CCAN_NTDB_H
+#define CCAN_NTDB_H
+
+/*
+   NTDB: trivial database library version 2
+
+   Copyright (C) Andrew Tridgell 1999-2004
+   Copyright (C) Rusty Russell 2010-2012
+
+     ** NOTE! The following LGPL license applies to the ntdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+#ifdef HAVE_LIBREPLACE
+#include <replace.h>
+#else
+#if HAVE_FILE_OFFSET_BITS
+#define _FILE_OFFSET_BITS 64
+#endif
+/* For mode_t */
+#include <sys/types.h>
+/* For O_* flags. */
+#include <sys/stat.h>
+/* For sig_atomic_t. */
+#include <signal.h>
+/* For uint64_t */
+#include <stdint.h>
+/* For bool */
+#include <stdbool.h>
+/* For memcmp */
+#include <string.h>
+#endif
+
+#if HAVE_CCAN
+#include <ccan/compiler/compiler.h>
+#include <ccan/typesafe_cb/typesafe_cb.h>
+#include <ccan/cast/cast.h>
+#else
+#ifndef typesafe_cb_preargs
+/* Failing to have CCAN just mean less typesafe protection, etc. */
+#define typesafe_cb_preargs(rtype, atype, fn, arg, ...)	\
+	((rtype (*)(__VA_ARGS__, atype))(fn))
+#endif
+#ifndef cast_const
+#if defined(__intptr_t_defined) || defined(HAVE_INTPTR_T)
+#define cast_const(type, expr) ((type)((intptr_t)(expr)))
+#else
+#define cast_const(type, expr) ((type *)(expr))
+#endif
+#endif
+#endif /* !HAVE_CCAN */
+
+union ntdb_attribute;
+struct ntdb_context;
+
+/**
+ * struct TDB_DATA - (n)tdb data blob
+ *
+ * To ease compatibility, we use 'struct TDB_DATA' from tdb.h, so if
+ * you want to include both tdb.h and ntdb.h, you need to #include
+ * tdb.h first.
+ */
+#ifndef __TDB_H__
+struct TDB_DATA {
+	unsigned char *dptr;
+	size_t dsize;
+};
+#endif
+
+typedef struct TDB_DATA NTDB_DATA;
+
+/**
+ * ntdb_open - open a database file
+ * @name: the file name (can be NULL if flags contains NTDB_INTERNAL)
+ * @ntdb_flags: options for this database
+ * @open_flags: flags argument for ntdb's open() call.
+ * @mode: mode argument for ntdb's open() call.
+ * @attributes: linked list of extra attributes for this ntdb.
+ *
+ * This call opens (and potentially creates) a database file.
+ * Multiple processes can have the NTDB file open at once.
+ *
+ * On failure it will return NULL, and set errno: it may also call
+ * any log attribute found in @attributes.
+ *
+ * See also:
+ *	union ntdb_attribute
+ */
+struct ntdb_context *ntdb_open(const char *name, int ntdb_flags,
+			       int open_flags, mode_t mode,
+			       union ntdb_attribute *attributes);
+
+
+/* flags for ntdb_open() */
+#define NTDB_DEFAULT 0 /* just a readability place holder */
+#define NTDB_INTERNAL 2 /* don't store on disk */
+#define NTDB_NOLOCK   4 /* don't do any locking */
+#define NTDB_NOMMAP   8 /* don't use mmap */
+#define NTDB_CONVERT 16 /* convert endian */
+#define NTDB_NOSYNC   64 /* don't use synchronous transactions */
+#define NTDB_SEQNUM   128 /* maintain a sequence number */
+#define NTDB_ALLOW_NESTING   256 /* fake nested transactions */
+#define NTDB_RDONLY   512 /* implied by O_RDONLY */
+#define NTDB_CANT_CHECK  2048 /* has a feature which we don't understand */
+
+/**
+ * ntdb_close - close and free a ntdb.
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This always succeeds, in that @ntdb is unusable after this call.  But if
+ * some unexpected error occurred while closing, it will return non-zero
+ * (the only clue as to cause will be via the log attribute).
+ */
+int ntdb_close(struct ntdb_context *ntdb);
+
+/**
+ * enum NTDB_ERROR - error returns for NTDB
+ *
+ * See Also:
+ *	ntdb_errorstr()
+ */
+enum NTDB_ERROR {
+	NTDB_SUCCESS	= 0,	/* No error. */
+	NTDB_ERR_CORRUPT = -1,	/* We read the db, and it was bogus. */
+	NTDB_ERR_IO	= -2,	/* We couldn't read/write the db. */
+	NTDB_ERR_LOCK	= -3,	/* Locking failed. */
+	NTDB_ERR_OOM	= -4,	/* Out of Memory. */
+	NTDB_ERR_EXISTS	= -5,	/* The key already exists. */
+	NTDB_ERR_NOEXIST	= -6,	/* The key does not exist. */
+	NTDB_ERR_EINVAL	= -7,	/* You're using it wrong. */
+	NTDB_ERR_RDONLY	= -8,	/* The database is read-only. */
+	NTDB_ERR_LAST = NTDB_ERR_RDONLY
+};
+
+/**
+ * ntdb_store - store a key/value pair in a ntdb.
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key
+ * @dbuf: the data to associate with the key.
+ * @flag: NTDB_REPLACE, NTDB_INSERT or NTDB_MODIFY.
+ *
+ * This inserts (or overwrites) a key/value pair in the NTDB.  If flag
+ * is NTDB_REPLACE, it doesn't matter whether the key exists or not;
+ * NTDB_INSERT means it must not exist (returns NTDB_ERR_EXISTS otherwise),
+ * and NTDB_MODIFY means it must exist (returns NTDB_ERR_NOEXIST otherwise).
+ *
+ * On success, this returns NTDB_SUCCESS.
+ *
+ * See also:
+ *	ntdb_fetch, ntdb_transaction_start, ntdb_append, ntdb_delete.
+ */
+enum NTDB_ERROR ntdb_store(struct ntdb_context *ntdb,
+			   NTDB_DATA key,
+			   NTDB_DATA dbuf,
+			   int flag);
+
+/* flags to ntdb_store() */
+#define NTDB_REPLACE 1		/* A readability place holder */
+#define NTDB_INSERT 2 		/* Don't overwrite an existing entry */
+#define NTDB_MODIFY 3		/* Don't create an existing entry    */
+
+/**
+ * ntdb_fetch - fetch a value from a ntdb.
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key
+ * @data: pointer to data.
+ *
+ * This looks up a key in the database and sets it in @data.
+ *
+ * If it returns NTDB_SUCCESS, the key was found: it is your
+ * responsibility to call free() on @data->dptr.
+ *
+ * Otherwise, it returns an error (usually, NTDB_ERR_NOEXIST) and @data is
+ * undefined.
+ */
+enum NTDB_ERROR ntdb_fetch(struct ntdb_context *ntdb, NTDB_DATA key,
+			   NTDB_DATA *data);
+
+/**
+ * ntdb_errorstr - map the ntdb error onto a constant readable string
+ * @ecode: the enum NTDB_ERROR to map.
+ *
+ * This is useful for displaying errors to users.
+ */
+const char *ntdb_errorstr(enum NTDB_ERROR ecode);
+
+/**
+ * ntdb_append - append a value to a key/value pair in a ntdb.
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key
+ * @dbuf: the data to append.
+ *
+ * This is equivalent to fetching a record, reallocating .dptr to add the
+ * data, and writing it back, only it's much more efficient.  If the key
+ * doesn't exist, it's equivalent to ntdb_store (with an additional hint that
+ * you expect to expand the record in future).
+ *
+ * See Also:
+ *	ntdb_fetch(), ntdb_store()
+ */
+enum NTDB_ERROR ntdb_append(struct ntdb_context *ntdb,
+			    NTDB_DATA key, NTDB_DATA dbuf);
+
+/**
+ * ntdb_delete - delete a key from a ntdb.
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key to delete.
+ *
+ * Returns NTDB_SUCCESS on success, or an error (usually NTDB_ERR_NOEXIST).
+ *
+ * See Also:
+ *	ntdb_fetch(), ntdb_store()
+ */
+enum NTDB_ERROR ntdb_delete(struct ntdb_context *ntdb, NTDB_DATA key);
+
+/**
+ * ntdb_exists - does a key exist in the database?
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key to search for.
+ *
+ * Returns true if it exists, or false if it doesn't or any other error.
+ */
+bool ntdb_exists(struct ntdb_context *ntdb, NTDB_DATA key);
+
+/**
+ * ntdb_deq - are NTDB_DATA equal?
+ * @a: one NTDB_DATA
+ * @b: another NTDB_DATA
+ */
+static inline bool ntdb_deq(NTDB_DATA a, NTDB_DATA b)
+{
+	return a.dsize == b.dsize && memcmp(a.dptr, b.dptr, a.dsize) == 0;
+}
+
+/**
+ * ntdb_mkdata - make a NTDB_DATA from const data
+ * @p: the constant pointer
+ * @len: the length
+ *
+ * As the dptr member of NTDB_DATA is not constant, you need to
+ * cast it.  This function keeps thost casts in one place, as well as
+ * suppressing the warning some compilers give when casting away a
+ * qualifier (eg. gcc with -Wcast-qual)
+ */
+static inline NTDB_DATA ntdb_mkdata(const void *p, size_t len)
+{
+	NTDB_DATA d;
+	d.dptr = cast_const(void *, p);
+	d.dsize = len;
+	return d;
+}
+
+/**
+ * ntdb_transaction_start - start a transaction
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This begins a series of atomic operations.  Other processes will be able
+ * to read the ntdb, but not alter it (they will block), nor will they see
+ * any changes until ntdb_transaction_commit() is called.
+ *
+ * Note that if the NTDB_ALLOW_NESTING flag is set, a ntdb_transaction_start()
+ * within a transaction will succeed, but it's not a real transaction:
+ * (1) An inner transaction which is committed is not actually committed until
+ *     the outer transaction is; if the outer transaction is cancelled, the
+ *     inner ones are discarded.
+ * (2) ntdb_transaction_cancel() marks the outer transaction as having an error,
+ *     so the final ntdb_transaction_commit() will fail.
+ * (3) the outer transaction will see the results of the inner transaction.
+ *
+ * See Also:
+ *	ntdb_transaction_cancel, ntdb_transaction_commit.
+ */
+enum NTDB_ERROR ntdb_transaction_start(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_transaction_cancel - abandon a transaction
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This aborts a transaction, discarding any changes which were made.
+ * ntdb_close() does this implicitly.
+ */
+void ntdb_transaction_cancel(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_transaction_commit - commit a transaction
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This completes a transaction, writing any changes which were made.
+ *
+ * fsync() is used to commit the transaction (unless NTDB_NOSYNC is set),
+ * making it robust against machine crashes, but very slow compared to
+ * other NTDB operations.
+ *
+ * A failure can only be caused by unexpected errors (eg. I/O or
+ * memory); this is no point looping on transaction failure.
+ *
+ * See Also:
+ *	ntdb_transaction_prepare_commit()
+ */
+enum NTDB_ERROR ntdb_transaction_commit(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_transaction_prepare_commit - prepare to commit a transaction
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This ensures we have the resources to commit a transaction (using
+ * ntdb_transaction_commit): if this succeeds then a transaction will only
+ * fail if the write() or fsync() calls fail.
+ *
+ * If this fails you must still call ntdb_transaction_cancel() to cancel
+ * the transaction.
+ *
+ * See Also:
+ *	ntdb_transaction_commit()
+ */
+enum NTDB_ERROR ntdb_transaction_prepare_commit(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_traverse - traverse a NTDB
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @fn: the function to call for every key/value pair (or NULL)
+ * @p: the pointer to hand to @f
+ *
+ * This walks the NTDB until all they keys have been traversed, or @fn
+ * returns non-zero.  If the traverse function or other processes are
+ * changing data or adding or deleting keys, the traverse may be
+ * unreliable: keys may be skipped or (rarely) visited twice.
+ *
+ * There is one specific exception: the special case of deleting the
+ * current key does not undermine the reliability of the traversal.
+ *
+ * On success, returns the number of keys iterated.  On error returns
+ * a negative enum NTDB_ERROR value.
+ */
+#define ntdb_traverse(ntdb, fn, p)					\
+	ntdb_traverse_(ntdb, typesafe_cb_preargs(int, void *, (fn), (p), \
+						 struct ntdb_context *,	\
+						 NTDB_DATA, NTDB_DATA), (p))
+
+int64_t ntdb_traverse_(struct ntdb_context *ntdb,
+		       int (*fn)(struct ntdb_context *,
+				 NTDB_DATA, NTDB_DATA, void *), void *p);
+
+/**
+ * ntdb_parse_record - operate directly on data in the database.
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key whose record we should hand to @parse
+ * @parse: the function to call for the data
+ * @data: the private pointer to hand to @parse (types must match).
+ *
+ * This avoids a copy for many cases, by handing you a pointer into
+ * the memory-mapped database.  It also locks the record to prevent
+ * other accesses at the same time.
+ *
+ * Do not alter the data handed to parse()!
+ */
+#define ntdb_parse_record(ntdb, key, parse, data)			\
+	ntdb_parse_record_((ntdb), (key),				\
+			   typesafe_cb_preargs(enum NTDB_ERROR, void *,	\
+					       (parse), (data),		\
+					       NTDB_DATA, NTDB_DATA), (data))
+
+enum NTDB_ERROR ntdb_parse_record_(struct ntdb_context *ntdb,
+				   NTDB_DATA key,
+				   enum NTDB_ERROR (*parse)(NTDB_DATA k,
+							    NTDB_DATA d,
+							    void *data),
+				   void *data);
+
+/**
+ * ntdb_get_seqnum - get a database sequence number
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This returns a sequence number: any change to the database from a
+ * ntdb context opened with the NTDB_SEQNUM flag will cause that number
+ * to increment.  Note that the incrementing is unreliable (it is done
+ * without locking), so this is only useful as an optimization.
+ *
+ * For example, you may have a regular database backup routine which
+ * does not operate if the sequence number is unchanged.  In the
+ * unlikely event of a failed increment, it will be backed up next
+ * time any way.
+ *
+ * Returns an enum NTDB_ERROR (ie. negative) on error.
+ */
+int64_t ntdb_get_seqnum(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_firstkey - get the "first" key in a NTDB
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: pointer to key.
+ *
+ * This returns an arbitrary key in the database; with ntdb_nextkey() it allows
+ * open-coded traversal of the database, though it is slightly less efficient
+ * than ntdb_traverse.
+ *
+ * It is your responsibility to free @key->dptr on success.
+ *
+ * Returns NTDB_ERR_NOEXIST if the database is empty.
+ */
+enum NTDB_ERROR ntdb_firstkey(struct ntdb_context *ntdb, NTDB_DATA *key);
+
+/**
+ * ntdb_nextkey - get the "next" key in a NTDB
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: a key returned by ntdb_firstkey() or ntdb_nextkey().
+ *
+ * This returns another key in the database; it will free @key.dptr for
+ * your convenience.
+ *
+ * Returns NTDB_ERR_NOEXIST if there are no more keys.
+ */
+enum NTDB_ERROR ntdb_nextkey(struct ntdb_context *ntdb, NTDB_DATA *key);
+
+/**
+ * ntdb_chainlock - lock a record in the NTDB
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key to lock.
+ *
+ * This prevents any access occurring to a group of keys including @key,
+ * even if @key does not exist.  This allows primitive atomic updates of
+ * records without using transactions.
+ *
+ * You cannot begin a transaction while holding a ntdb_chainlock(), nor can
+ * you do any operations on any other keys in the database.  This also means
+ * that you cannot hold more than one ntdb_chainlock() at a time.
+ *
+ * See Also:
+ *	ntdb_chainunlock()
+ */
+enum NTDB_ERROR ntdb_chainlock(struct ntdb_context *ntdb, NTDB_DATA key);
+
+/**
+ * ntdb_chainunlock - unlock a record in the NTDB
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key to unlock.
+ *
+ * The key must have previously been locked by ntdb_chainlock().
+ */
+void ntdb_chainunlock(struct ntdb_context *ntdb, NTDB_DATA key);
+
+/**
+ * ntdb_chainlock_read - lock a record in the NTDB, for reading
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key to lock.
+ *
+ * This prevents any changes from occurring to a group of keys including @key,
+ * even if @key does not exist.  This allows primitive atomic updates of
+ * records without using transactions.
+ *
+ * You cannot begin a transaction while holding a ntdb_chainlock_read(), nor can
+ * you do any operations on any other keys in the database.  This also means
+ * that you cannot hold more than one ntdb_chainlock()/read() at a time.
+ *
+ * See Also:
+ *	ntdb_chainlock()
+ */
+enum NTDB_ERROR ntdb_chainlock_read(struct ntdb_context *ntdb, NTDB_DATA key);
+
+/**
+ * ntdb_chainunlock_read - unlock a record in the NTDB for reading
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @key: the key to unlock.
+ *
+ * The key must have previously been locked by ntdb_chainlock_read().
+ */
+void ntdb_chainunlock_read(struct ntdb_context *ntdb, NTDB_DATA key);
+
+/**
+ * ntdb_lockall - lock the entire NTDB
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * You cannot hold a ntdb_chainlock while calling this.  It nests, so you
+ * must call ntdb_unlockall as many times as you call ntdb_lockall.
+ */
+enum NTDB_ERROR ntdb_lockall(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_unlockall - unlock the entire NTDB
+ * @ntdb: the ntdb context returned from ntdb_open()
+ */
+void ntdb_unlockall(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_lockall_read - lock the entire NTDB for reading
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This prevents others writing to the database, eg. ntdb_delete, ntdb_store,
+ * ntdb_append, but not ntdb_fetch.
+ *
+ * You cannot hold a ntdb_chainlock while calling this.  It nests, so you
+ * must call ntdb_unlockall_read as many times as you call ntdb_lockall_read.
+ */
+enum NTDB_ERROR ntdb_lockall_read(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_unlockall_read - unlock the entire NTDB for reading
+ * @ntdb: the ntdb context returned from ntdb_open()
+ */
+void ntdb_unlockall_read(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_wipe_all - wipe the database clean
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * Completely erase the database.  This is faster than iterating through
+ * each key and doing ntdb_delete.
+ */
+enum NTDB_ERROR ntdb_wipe_all(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_repack - repack the database
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This repacks the database; if it is suffering from a great deal of
+ * fragmentation this might help.  However, it can take twice the
+ * memory of the existing NTDB.
+ */
+enum NTDB_ERROR ntdb_repack(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_check - check a NTDB for consistency
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @check: function to check each key/data pair (or NULL)
+ * @data: argument for @check, must match type.
+ *
+ * This performs a consistency check of the open database, optionally calling
+ * a check() function on each record so you can do your own data consistency
+ * checks as well.  If check() returns an error, that is returned from
+ * ntdb_check().
+ *
+ * Note that the NTDB uses a feature which we don't understand which
+ * indicates we can't run ntdb_check(), this will log a warning to that
+ * effect and return NTDB_SUCCESS.  You can detect this condition by
+ * looking for NTDB_CANT_CHECK in ntdb_get_flags().
+ *
+ * Returns NTDB_SUCCESS or an error.
+ */
+#define ntdb_check(ntdb, check, data)					\
+	ntdb_check_((ntdb), typesafe_cb_preargs(enum NTDB_ERROR, void *, \
+						(check), (data),	\
+						NTDB_DATA,		\
+						NTDB_DATA),		\
+		    (data))
+
+enum NTDB_ERROR ntdb_check_(struct ntdb_context *ntdb,
+			    enum NTDB_ERROR (*check)(NTDB_DATA k,
+						     NTDB_DATA d,
+						     void *data),
+			    void *data);
+
+/**
+ * ntdb_error - get the last error (not threadsafe)
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * Returns the last error returned by a NTDB function.
+ *
+ * This makes porting from TDB easier, but note that the last error is not
+ * reliable in threaded programs.
+ */
+enum NTDB_ERROR ntdb_error(struct ntdb_context *ntdb);
+
+/**
+ * enum ntdb_summary_flags - flags for ntdb_summary.
+ */
+enum ntdb_summary_flags {
+	NTDB_SUMMARY_HISTOGRAMS = 1 /* Draw graphs in the summary. */
+};
+
+/**
+ * ntdb_summary - return a string describing the NTDB state
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @flags: flags to control the summary output.
+ * @summary: pointer to string to allocate.
+ *
+ * This returns a developer-readable string describing the overall
+ * state of the ntdb, such as the percentage used and sizes of records.
+ * It is designed to provide information about the ntdb at a glance
+ * without displaying any keys or data in the database.
+ *
+ * On success, sets @summary to point to a malloc()'ed nul-terminated
+ * multi-line string.  It is your responsibility to free() it.
+ */
+enum NTDB_ERROR ntdb_summary(struct ntdb_context *ntdb,
+			     enum ntdb_summary_flags flags,
+			     char **summary);
+
+
+/**
+ * ntdb_get_flags - return the flags for a ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This returns the flags on the current ntdb.  Some of these are caused by
+ * the flags argument to ntdb_open(), others (such as NTDB_CONVERT) are
+ * intuited.
+ */
+unsigned int ntdb_get_flags(struct ntdb_context *ntdb);
+
+/**
+ * ntdb_add_flag - set a flag for a ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @flag: one of NTDB_NOLOCK, NTDB_NOMMAP, NTDB_NOSYNC or NTDB_ALLOW_NESTING.
+ *
+ * You can use this to set a flag on the NTDB.  You cannot set these flags
+ * on a NTDB_INTERNAL ntdb.
+ */
+void ntdb_add_flag(struct ntdb_context *ntdb, unsigned flag);
+
+/**
+ * ntdb_remove_flag - unset a flag for a ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @flag: one of NTDB_NOLOCK, NTDB_NOMMAP, NTDB_NOSYNC or NTDB_ALLOW_NESTING.
+ *
+ * You can use this to clear a flag on the NTDB.  You cannot clear flags
+ * on a NTDB_INTERNAL ntdb.
+ */
+void ntdb_remove_flag(struct ntdb_context *ntdb, unsigned flag);
+
+/**
+ * enum ntdb_attribute_type - descriminator for union ntdb_attribute.
+ */
+enum ntdb_attribute_type {
+	NTDB_ATTRIBUTE_LOG = 0,
+	NTDB_ATTRIBUTE_HASH = 1,
+	NTDB_ATTRIBUTE_SEED = 2,
+	NTDB_ATTRIBUTE_STATS = 3,
+	NTDB_ATTRIBUTE_OPENHOOK = 4,
+	NTDB_ATTRIBUTE_FLOCK = 5,
+};
+
+/**
+ * ntdb_get_attribute - get an attribute for an existing ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @attr: the union ntdb_attribute to set.
+ *
+ * This gets an attribute from a NTDB which has previously been set (or
+ * may return the default values).  Set @attr.base.attr to the
+ * attribute type you want get.
+ */
+enum NTDB_ERROR ntdb_get_attribute(struct ntdb_context *ntdb,
+				   union ntdb_attribute *attr);
+
+/**
+ * ntdb_set_attribute - set an attribute for an existing ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @attr: the union ntdb_attribute to set.
+ *
+ * This sets an attribute on a NTDB, overriding any previous attribute
+ * of the same type.  It returns NTDB_ERR_EINVAL if the attribute is
+ * unknown or invalid.
+ *
+ * Note that NTDB_ATTRIBUTE_HASH, NTDB_ATTRIBUTE_SEED, and
+ * NTDB_ATTRIBUTE_OPENHOOK cannot currently be set after ntdb_open.
+ */
+enum NTDB_ERROR ntdb_set_attribute(struct ntdb_context *ntdb,
+				   const union ntdb_attribute *attr);
+
+/**
+ * ntdb_unset_attribute - reset an attribute for an existing ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ * @type: the attribute type to unset.
+ *
+ * This unsets an attribute on a NTDB, returning it to the defaults
+ * (where applicable).
+ *
+ * Note that it only makes sense for NTDB_ATTRIBUTE_LOG and NTDB_ATTRIBUTE_FLOCK
+ * to be unset.
+ */
+void ntdb_unset_attribute(struct ntdb_context *ntdb,
+			  enum ntdb_attribute_type type);
+
+/**
+ * ntdb_name - get the name of a ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This returns a copy of the name string, made at ntdb_open() time.  If that
+ * argument was NULL (possible for a NTDB_INTERNAL db) this will return NULL.
+ *
+ * This is mostly useful for logging.
+ */
+const char *ntdb_name(const struct ntdb_context *ntdb);
+
+/**
+ * ntdb_fd - get the file descriptor of a ntdb
+ * @ntdb: the ntdb context returned from ntdb_open()
+ *
+ * This returns the file descriptor for the underlying database file, or -1
+ * for NTDB_INTERNAL.
+ */
+int ntdb_fd(const struct ntdb_context *ntdb);
+
+/**
+ * ntdb_foreach - iterate through every open NTDB.
+ * @fn: the function to call for every NTDB
+ * @p: the pointer to hand to @fn
+ *
+ * NTDB internally keeps track of all open TDBs; this function allows you to
+ * iterate through them.  If @fn returns non-zero, traversal stops.
+ */
+#define ntdb_foreach(fn, p)						\
+	ntdb_foreach_(typesafe_cb_preargs(int, void *, (fn), (p),	\
+					  struct ntdb_context *), (p))
+
+void ntdb_foreach_(int (*fn)(struct ntdb_context *, void *), void *p);
+
+/**
+ * struct ntdb_attribute_base - common fields for all ntdb attributes.
+ */
+struct ntdb_attribute_base {
+	enum ntdb_attribute_type attr;
+	union ntdb_attribute *next;
+};
+
+/**
+ * enum ntdb_log_level - log levels for ntdb_attribute_log
+ * @NTDB_LOG_ERROR: used to log unrecoverable errors such as I/O errors
+ *		   or internal consistency failures.
+ * @NTDB_LOG_USE_ERROR: used to log usage errors such as invalid parameters
+ *		   or writing to a read-only database.
+ * @NTDB_LOG_WARNING: used for informational messages on issues which
+ *		     are unusual but handled by NTDB internally, such
+ *		     as a failure to mmap or failure to open /dev/urandom.
+ */
+enum ntdb_log_level {
+	NTDB_LOG_ERROR,
+	NTDB_LOG_USE_ERROR,
+	NTDB_LOG_WARNING
+};
+
+/**
+ * struct ntdb_attribute_log - log function attribute
+ *
+ * This attribute provides a hook for you to log errors.
+ */
+struct ntdb_attribute_log {
+	struct ntdb_attribute_base base; /* .attr = NTDB_ATTRIBUTE_LOG */
+	void (*fn)(struct ntdb_context *ntdb,
+		   enum ntdb_log_level level,
+		   enum NTDB_ERROR ecode,
+		   const char *message,
+		   void *data);
+	void *data;
+};
+
+/**
+ * struct ntdb_attribute_hash - hash function attribute
+ *
+ * This attribute allows you to provide an alternative hash function.
+ * This hash function will be handed keys from the database; it will also
+ * be handed the 8-byte NTDB_HASH_MAGIC value for checking the header (the
+ * ntdb_open() will fail if the hash value doesn't match the header).
+ *
+ * Note that if your hash function gives different results on
+ * different machine endians, your ntdb will no longer work across
+ * different architectures!
+ */
+struct ntdb_attribute_hash {
+	struct ntdb_attribute_base base; /* .attr = NTDB_ATTRIBUTE_HASH */
+	uint64_t (*fn)(const void *key, size_t len, uint64_t seed,
+		       void *data);
+	void *data;
+};
+
+/**
+ * struct ntdb_attribute_seed - hash function seed attribute
+ *
+ * The hash function seed is normally taken from /dev/urandom (or equivalent)
+ * but can be set manually here.  This is mainly for testing purposes.
+ */
+struct ntdb_attribute_seed {
+	struct ntdb_attribute_base base; /* .attr = NTDB_ATTRIBUTE_SEED */
+	uint64_t seed;
+};
+
+/**
+ * struct ntdb_attribute_stats - ntdb operational statistics
+ *
+ * This attribute records statistics of various low-level NTDB operations.
+ * This can be used to assist performance evaluation.  This is only
+ * useful for ntdb_get_attribute().
+ *
+ * New fields will be added at the end, hence the "size" argument which
+ * indicates how large your structure is: it must be filled in before
+ * calling ntdb_get_attribute(), which will overwrite it with the size
+ * ntdb knows about.
+ */
+struct ntdb_attribute_stats {
+	struct ntdb_attribute_base base; /* .attr = NTDB_ATTRIBUTE_STATS */
+	size_t size; /* = sizeof(struct ntdb_attribute_stats) */
+	uint64_t allocs;
+	uint64_t   alloc_subhash;
+	uint64_t   alloc_chain;
+	uint64_t   alloc_bucket_exact;
+	uint64_t   alloc_bucket_max;
+	uint64_t   alloc_leftover;
+	uint64_t   alloc_coalesce_tried;
+	uint64_t     alloc_coalesce_iterate_clash;
+	uint64_t     alloc_coalesce_lockfail;
+	uint64_t     alloc_coalesce_race;
+	uint64_t     alloc_coalesce_succeeded;
+	uint64_t       alloc_coalesce_num_merged;
+	uint64_t compares;
+	uint64_t   compare_wrong_bucket;
+	uint64_t   compare_wrong_offsetbits;
+	uint64_t   compare_wrong_keylen;
+	uint64_t   compare_wrong_rechash;
+	uint64_t   compare_wrong_keycmp;
+	uint64_t transactions;
+	uint64_t   transaction_cancel;
+	uint64_t   transaction_nest;
+	uint64_t   transaction_expand_file;
+	uint64_t   transaction_read_direct;
+	uint64_t      transaction_read_direct_fail;
+	uint64_t   transaction_write_direct;
+	uint64_t      transaction_write_direct_fail;
+	uint64_t expands;
+	uint64_t frees;
+	uint64_t locks;
+	uint64_t   lock_lowlevel;
+	uint64_t   lock_nonblock;
+	uint64_t     lock_nonblock_fail;
+};
+
+/**
+ * struct ntdb_attribute_openhook - ntdb special effects hook for open
+ *
+ * This attribute contains a function to call once we have the OPEN_LOCK
+ * for the ntdb, but before we've examined its contents.  If this succeeds,
+ * the ntdb will be populated if it's then zero-length.
+ *
+ * This is a hack to allow support for TDB-style TDB_CLEAR_IF_FIRST
+ * behaviour.
+ */
+struct ntdb_attribute_openhook {
+	struct ntdb_attribute_base base; /* .attr = NTDB_ATTRIBUTE_OPENHOOK */
+	enum NTDB_ERROR (*fn)(int fd, void *data);
+	void *data;
+};
+
+/**
+ * struct ntdb_attribute_flock - ntdb special effects hook for file locking
+ *
+ * This attribute contains function to call to place locks on a file; it can
+ * be used to support non-blocking operations or lock proxying.
+ *
+ * They should return 0 on success, -1 on failure and set errno.
+ *
+ * An error will be logged on error if errno is neither EAGAIN nor EINTR
+ * (normally it would only return EAGAIN if waitflag is false, and
+ * loop internally on EINTR).
+ */
+struct ntdb_attribute_flock {
+	struct ntdb_attribute_base base; /* .attr = NTDB_ATTRIBUTE_FLOCK */
+	int (*lock)(int fd,int rw, off_t off, off_t len, bool waitflag, void *);
+	int (*unlock)(int fd, int rw, off_t off, off_t len, void *);
+	void *data;
+};
+
+/**
+ * union ntdb_attribute - ntdb attributes.
+ *
+ * This represents all the known attributes.
+ *
+ * See also:
+ *	struct ntdb_attribute_log, struct ntdb_attribute_hash,
+ *	struct ntdb_attribute_seed, struct ntdb_attribute_stats,
+ *	struct ntdb_attribute_openhook, struct ntdb_attribute_flock.
+ */
+union ntdb_attribute {
+	struct ntdb_attribute_base base;
+	struct ntdb_attribute_log log;
+	struct ntdb_attribute_hash hash;
+	struct ntdb_attribute_seed seed;
+	struct ntdb_attribute_stats stats;
+	struct ntdb_attribute_openhook openhook;
+	struct ntdb_attribute_flock flock;
+};
+
+#ifdef  __cplusplus
+}
+#endif
+
+#endif /* ntdb.h */

lib/ntdb/ntdb.pc.in

@@ -3,9 +3,9 @@ exec_prefix=@exec_prefix@
 libdir=@libdir@
 includedir=@includedir@
 
-Name: tdb
-Description: A trivial database
+Name: ntdb
+Description: A (not-so) trivial database
 Version: @PACKAGE_VERSION@
-Libs: @LIB_RPATH@ -L${libdir} -ltdb
+Libs: @LIB_RPATH@ -L${libdir} -lntdb
 Cflags: -I${includedir}
 URL: http://tdb.samba.org/

lib/ntdb/open.c

@@ -0,0 +1,768 @@
+ /*
+   Trivial Database 2: opening and closing TDBs
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "private.h"
+#include <ccan/build_assert/build_assert.h>
+#include <assert.h>
+
+/* all tdbs, to detect double-opens (fcntl file don't nest!) */
+static struct ntdb_context *tdbs = NULL;
+
+static struct ntdb_file *find_file(dev_t device, ino_t ino)
+{
+	struct ntdb_context *i;
+
+	for (i = tdbs; i; i = i->next) {
+		if (i->file->device == device && i->file->inode == ino) {
+			i->file->refcnt++;
+			return i->file;
+		}
+	}
+	return NULL;
+}
+
+static bool read_all(int fd, void *buf, size_t len)
+{
+	while (len) {
+		ssize_t ret;
+		ret = read(fd, buf, len);
+		if (ret < 0)
+			return false;
+		if (ret == 0) {
+			/* ETOOSHORT? */
+			errno = EWOULDBLOCK;
+			return false;
+		}
+		buf = (char *)buf + ret;
+		len -= ret;
+	}
+	return true;
+}
+
+static uint64_t random_number(struct ntdb_context *ntdb)
+{
+	int fd;
+	uint64_t ret = 0;
+	struct timeval now;
+
+	fd = open("/dev/urandom", O_RDONLY);
+	if (fd >= 0) {
+		if (read_all(fd, &ret, sizeof(ret))) {
+			close(fd);
+			return ret;
+		}
+		close(fd);
+	}
+	/* FIXME: Untested!  Based on Wikipedia protocol description! */
+	fd = open("/dev/egd-pool", O_RDWR);
+	if (fd >= 0) {
+		/* Command is 1, next byte is size we want to read. */
+		char cmd[2] = { 1, sizeof(uint64_t) };
+		if (write(fd, cmd, sizeof(cmd)) == sizeof(cmd)) {
+			char reply[1 + sizeof(uint64_t)];
+			int r = read(fd, reply, sizeof(reply));
+			if (r > 1) {
+				/* Copy at least some bytes. */
+				memcpy(&ret, reply+1, r - 1);
+				if (reply[0] == sizeof(uint64_t)
+				    && r == sizeof(reply)) {
+					close(fd);
+					return ret;
+				}
+			}
+		}
+		close(fd);
+	}
+
+	/* Fallback: pid and time. */
+	gettimeofday(&now, NULL);
+	ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
+	ntdb_logerr(ntdb, NTDB_SUCCESS, NTDB_LOG_WARNING,
+		   "ntdb_open: random from getpid and time");
+	return ret;
+}
+
+static void ntdb_context_init(struct ntdb_context *ntdb)
+{
+	/* Initialize the NTDB fields here */
+	ntdb_io_init(ntdb);
+	ntdb->direct_access = 0;
+	ntdb->transaction = NULL;
+	ntdb->access = NULL;
+}
+
+struct new_database {
+	struct ntdb_header hdr;
+	struct ntdb_freetable ftable;
+};
+
+/* initialise a new database */
+static enum NTDB_ERROR ntdb_new_database(struct ntdb_context *ntdb,
+				       struct ntdb_attribute_seed *seed,
+				       struct ntdb_header *hdr)
+{
+	/* We make it up in memory, then write it out if not internal */
+	struct new_database newdb;
+	unsigned int magic_len;
+	ssize_t rlen;
+	enum NTDB_ERROR ecode;
+
+	/* Fill in the header */
+	newdb.hdr.version = NTDB_VERSION;
+	if (seed)
+		newdb.hdr.hash_seed = seed->seed;
+	else
+		newdb.hdr.hash_seed = random_number(ntdb);
+	newdb.hdr.hash_test = NTDB_HASH_MAGIC;
+	newdb.hdr.hash_test = ntdb->hash_fn(&newdb.hdr.hash_test,
+					   sizeof(newdb.hdr.hash_test),
+					   newdb.hdr.hash_seed,
+					   ntdb->hash_data);
+	newdb.hdr.recovery = 0;
+	newdb.hdr.features_used = newdb.hdr.features_offered = NTDB_FEATURE_MASK;
+	newdb.hdr.seqnum = 0;
+	newdb.hdr.capabilities = 0;
+	memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
+	/* Initial hashes are empty. */
+	memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
+
+	/* Free is empty. */
+	newdb.hdr.free_table = offsetof(struct new_database, ftable);
+	memset(&newdb.ftable, 0, sizeof(newdb.ftable));
+	ecode = set_header(NULL, &newdb.ftable.hdr, NTDB_FTABLE_MAGIC, 0,
+			   sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
+			   sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
+			   0);
+	if (ecode != NTDB_SUCCESS) {
+		return ecode;
+	}
+
+	/* Magic food */
+	memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
+	strcpy(newdb.hdr.magic_food, NTDB_MAGIC_FOOD);
+
+	/* This creates an endian-converted database, as if read from disk */
+	magic_len = sizeof(newdb.hdr.magic_food);
+	ntdb_convert(ntdb,
+		    (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
+
+	*hdr = newdb.hdr;
+
+	if (ntdb->flags & NTDB_INTERNAL) {
+		ntdb->file->map_size = sizeof(newdb);
+		ntdb->file->map_ptr = malloc(ntdb->file->map_size);
+		if (!ntdb->file->map_ptr) {
+			return ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+					  "ntdb_new_database:"
+					  " failed to allocate");
+		}
+		memcpy(ntdb->file->map_ptr, &newdb, ntdb->file->map_size);
+		return NTDB_SUCCESS;
+	}
+	if (lseek(ntdb->file->fd, 0, SEEK_SET) == -1) {
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "ntdb_new_database:"
+				  " failed to seek: %s", strerror(errno));
+	}
+
+	if (ftruncate(ntdb->file->fd, 0) == -1) {
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "ntdb_new_database:"
+				  " failed to truncate: %s", strerror(errno));
+	}
+
+	rlen = write(ntdb->file->fd, &newdb, sizeof(newdb));
+	if (rlen != sizeof(newdb)) {
+		if (rlen >= 0)
+			errno = ENOSPC;
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "ntdb_new_database: %zi writing header: %s",
+				  rlen, strerror(errno));
+	}
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR ntdb_new_file(struct ntdb_context *ntdb)
+{
+	ntdb->file = malloc(sizeof(*ntdb->file));
+	if (!ntdb->file)
+		return ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+				  "ntdb_open: cannot alloc ntdb_file structure");
+	ntdb->file->num_lockrecs = 0;
+	ntdb->file->lockrecs = NULL;
+	ntdb->file->allrecord_lock.count = 0;
+	ntdb->file->refcnt = 1;
+	ntdb->file->map_ptr = NULL;
+	return NTDB_SUCCESS;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_set_attribute(struct ntdb_context *ntdb,
+				 const union ntdb_attribute *attr)
+{
+	switch (attr->base.attr) {
+	case NTDB_ATTRIBUTE_LOG:
+		ntdb->log_fn = attr->log.fn;
+		ntdb->log_data = attr->log.data;
+		break;
+	case NTDB_ATTRIBUTE_HASH:
+	case NTDB_ATTRIBUTE_SEED:
+	case NTDB_ATTRIBUTE_OPENHOOK:
+		return ntdb->last_error
+			= ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+				     NTDB_LOG_USE_ERROR,
+				     "ntdb_set_attribute:"
+				     " cannot set %s after opening",
+				     attr->base.attr == NTDB_ATTRIBUTE_HASH
+				     ? "NTDB_ATTRIBUTE_HASH"
+				     : attr->base.attr == NTDB_ATTRIBUTE_SEED
+				     ? "NTDB_ATTRIBUTE_SEED"
+				     : "NTDB_ATTRIBUTE_OPENHOOK");
+	case NTDB_ATTRIBUTE_STATS:
+		return ntdb->last_error
+			= ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+				     NTDB_LOG_USE_ERROR,
+				     "ntdb_set_attribute:"
+				     " cannot set NTDB_ATTRIBUTE_STATS");
+	case NTDB_ATTRIBUTE_FLOCK:
+		ntdb->lock_fn = attr->flock.lock;
+		ntdb->unlock_fn = attr->flock.unlock;
+		ntdb->lock_data = attr->flock.data;
+		break;
+	default:
+		return ntdb->last_error
+			= ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+				     NTDB_LOG_USE_ERROR,
+				     "ntdb_set_attribute:"
+				     " unknown attribute type %u",
+				     attr->base.attr);
+	}
+	return NTDB_SUCCESS;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_get_attribute(struct ntdb_context *ntdb,
+				 union ntdb_attribute *attr)
+{
+	switch (attr->base.attr) {
+	case NTDB_ATTRIBUTE_LOG:
+		if (!ntdb->log_fn)
+			return ntdb->last_error = NTDB_ERR_NOEXIST;
+		attr->log.fn = ntdb->log_fn;
+		attr->log.data = ntdb->log_data;
+		break;
+	case NTDB_ATTRIBUTE_HASH:
+		attr->hash.fn = ntdb->hash_fn;
+		attr->hash.data = ntdb->hash_data;
+		break;
+	case NTDB_ATTRIBUTE_SEED:
+		attr->seed.seed = ntdb->hash_seed;
+		break;
+	case NTDB_ATTRIBUTE_OPENHOOK:
+		if (!ntdb->openhook)
+			return ntdb->last_error = NTDB_ERR_NOEXIST;
+		attr->openhook.fn = ntdb->openhook;
+		attr->openhook.data = ntdb->openhook_data;
+		break;
+	case NTDB_ATTRIBUTE_STATS: {
+		size_t size = attr->stats.size;
+		if (size > ntdb->stats.size)
+			size = ntdb->stats.size;
+		memcpy(&attr->stats, &ntdb->stats, size);
+		break;
+	}
+	case NTDB_ATTRIBUTE_FLOCK:
+		attr->flock.lock = ntdb->lock_fn;
+		attr->flock.unlock = ntdb->unlock_fn;
+		attr->flock.data = ntdb->lock_data;
+		break;
+	default:
+		return ntdb->last_error
+			= ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+				     NTDB_LOG_USE_ERROR,
+				     "ntdb_get_attribute:"
+				     " unknown attribute type %u",
+				     attr->base.attr);
+	}
+	attr->base.next = NULL;
+	return NTDB_SUCCESS;
+}
+
+_PUBLIC_ void ntdb_unset_attribute(struct ntdb_context *ntdb,
+			 enum ntdb_attribute_type type)
+{
+	switch (type) {
+	case NTDB_ATTRIBUTE_LOG:
+		ntdb->log_fn = NULL;
+		break;
+	case NTDB_ATTRIBUTE_OPENHOOK:
+		ntdb->openhook = NULL;
+		break;
+	case NTDB_ATTRIBUTE_HASH:
+	case NTDB_ATTRIBUTE_SEED:
+		ntdb_logerr(ntdb, NTDB_ERR_EINVAL, NTDB_LOG_USE_ERROR,
+			   "ntdb_unset_attribute: cannot unset %s after opening",
+			   type == NTDB_ATTRIBUTE_HASH
+			   ? "NTDB_ATTRIBUTE_HASH"
+			   : "NTDB_ATTRIBUTE_SEED");
+		break;
+	case NTDB_ATTRIBUTE_STATS:
+		ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+			   NTDB_LOG_USE_ERROR,
+			   "ntdb_unset_attribute:"
+			   "cannot unset NTDB_ATTRIBUTE_STATS");
+		break;
+	case NTDB_ATTRIBUTE_FLOCK:
+		ntdb->lock_fn = ntdb_fcntl_lock;
+		ntdb->unlock_fn = ntdb_fcntl_unlock;
+		break;
+	default:
+		ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+			   NTDB_LOG_USE_ERROR,
+			   "ntdb_unset_attribute: unknown attribute type %u",
+			   type);
+	}
+}
+
+/* The top three bits of the capability tell us whether it matters. */
+enum NTDB_ERROR unknown_capability(struct ntdb_context *ntdb, const char *caller,
+				  ntdb_off_t type)
+{
+	if (type & NTDB_CAP_NOOPEN) {
+		return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				  "%s: file has unknown capability %llu",
+				  caller, type & NTDB_CAP_NOOPEN);
+	}
+
+	if ((type & NTDB_CAP_NOWRITE) && !(ntdb->flags & NTDB_RDONLY)) {
+		return ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_ERROR,
+				  "%s: file has unknown capability %llu"
+				  " (cannot write to it)",
+				  caller, type & NTDB_CAP_NOOPEN);
+	}
+
+	if (type & NTDB_CAP_NOCHECK) {
+		ntdb->flags |= NTDB_CANT_CHECK;
+	}
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR capabilities_ok(struct ntdb_context *ntdb,
+				      ntdb_off_t capabilities)
+{
+	ntdb_off_t off, next;
+	enum NTDB_ERROR ecode = NTDB_SUCCESS;
+	const struct ntdb_capability *cap;
+
+	/* Check capability list. */
+	for (off = capabilities; off && ecode == NTDB_SUCCESS; off = next) {
+		cap = ntdb_access_read(ntdb, off, sizeof(*cap), true);
+		if (NTDB_PTR_IS_ERR(cap)) {
+			return NTDB_PTR_ERR(cap);
+		}
+
+		switch (cap->type & NTDB_CAP_TYPE_MASK) {
+		/* We don't understand any capabilities (yet). */
+		default:
+			ecode = unknown_capability(ntdb, "ntdb_open", cap->type);
+		}
+		next = cap->next;
+		ntdb_access_release(ntdb, cap);
+	}
+	return ecode;
+}
+
+_PUBLIC_ struct ntdb_context *ntdb_open(const char *name, int ntdb_flags,
+			     int open_flags, mode_t mode,
+			     union ntdb_attribute *attr)
+{
+	struct ntdb_context *ntdb;
+	struct stat st;
+	int saved_errno = 0;
+	uint64_t hash_test;
+	unsigned v;
+	ssize_t rlen;
+	struct ntdb_header hdr;
+	struct ntdb_attribute_seed *seed = NULL;
+	ntdb_bool_err berr;
+	enum NTDB_ERROR ecode;
+	int openlock;
+
+	ntdb = malloc(sizeof(*ntdb) + (name ? strlen(name) + 1 : 0));
+	if (!ntdb) {
+		/* Can't log this */
+		errno = ENOMEM;
+		return NULL;
+	}
+	/* Set name immediately for logging functions. */
+	if (name) {
+		ntdb->name = strcpy((char *)(ntdb + 1), name);
+	} else {
+		ntdb->name = NULL;
+	}
+	ntdb->flags = ntdb_flags;
+	ntdb->log_fn = NULL;
+	ntdb->open_flags = open_flags;
+	ntdb->last_error = NTDB_SUCCESS;
+	ntdb->file = NULL;
+	ntdb->openhook = NULL;
+	ntdb->lock_fn = ntdb_fcntl_lock;
+	ntdb->unlock_fn = ntdb_fcntl_unlock;
+	ntdb->hash_fn = ntdb_jenkins_hash;
+	memset(&ntdb->stats, 0, sizeof(ntdb->stats));
+	ntdb->stats.base.attr = NTDB_ATTRIBUTE_STATS;
+	ntdb->stats.size = sizeof(ntdb->stats);
+
+	while (attr) {
+		switch (attr->base.attr) {
+		case NTDB_ATTRIBUTE_HASH:
+			ntdb->hash_fn = attr->hash.fn;
+			ntdb->hash_data = attr->hash.data;
+			break;
+		case NTDB_ATTRIBUTE_SEED:
+			seed = &attr->seed;
+			break;
+		case NTDB_ATTRIBUTE_OPENHOOK:
+			ntdb->openhook = attr->openhook.fn;
+			ntdb->openhook_data = attr->openhook.data;
+			break;
+		default:
+			/* These are set as normal. */
+			ecode = ntdb_set_attribute(ntdb, attr);
+			if (ecode != NTDB_SUCCESS)
+				goto fail;
+		}
+		attr = attr->base.next;
+	}
+
+	if (ntdb_flags & ~(NTDB_INTERNAL | NTDB_NOLOCK | NTDB_NOMMAP | NTDB_CONVERT
+			  | NTDB_NOSYNC | NTDB_SEQNUM | NTDB_ALLOW_NESTING
+			  | NTDB_RDONLY)) {
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_EINVAL, NTDB_LOG_USE_ERROR,
+				   "ntdb_open: unknown flags %u", ntdb_flags);
+		goto fail;
+	}
+
+	if (seed) {
+		if (!(ntdb_flags & NTDB_INTERNAL) && !(open_flags & O_CREAT)) {
+			ecode = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+					   NTDB_LOG_USE_ERROR,
+					   "ntdb_open:"
+					   " cannot set NTDB_ATTRIBUTE_SEED"
+					   " without O_CREAT.");
+			goto fail;
+		}
+	}
+
+	if ((open_flags & O_ACCMODE) == O_WRONLY) {
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_EINVAL, NTDB_LOG_USE_ERROR,
+				   "ntdb_open: can't open ntdb %s write-only",
+				   name);
+		goto fail;
+	}
+
+	if ((open_flags & O_ACCMODE) == O_RDONLY) {
+		openlock = F_RDLCK;
+		ntdb->flags |= NTDB_RDONLY;
+	} else {
+		if (ntdb_flags & NTDB_RDONLY) {
+			ecode = ntdb_logerr(ntdb, NTDB_ERR_EINVAL,
+					   NTDB_LOG_USE_ERROR,
+					   "ntdb_open: can't use NTDB_RDONLY"
+					   " without O_RDONLY");
+			goto fail;
+		}
+		openlock = F_WRLCK;
+	}
+
+	/* internal databases don't need any of the rest. */
+	if (ntdb->flags & NTDB_INTERNAL) {
+		ntdb->flags |= (NTDB_NOLOCK | NTDB_NOMMAP);
+		ecode = ntdb_new_file(ntdb);
+		if (ecode != NTDB_SUCCESS) {
+			goto fail;
+		}
+		ntdb->file->fd = -1;
+		ecode = ntdb_new_database(ntdb, seed, &hdr);
+		if (ecode == NTDB_SUCCESS) {
+			ntdb_convert(ntdb, &hdr.hash_seed,
+				    sizeof(hdr.hash_seed));
+			ntdb->hash_seed = hdr.hash_seed;
+			ntdb_context_init(ntdb);
+			ntdb_ftable_init(ntdb);
+		}
+		if (ecode != NTDB_SUCCESS) {
+			goto fail;
+		}
+		return ntdb;
+	}
+
+	if (stat(name, &st) != -1)
+		ntdb->file = find_file(st.st_dev, st.st_ino);
+
+	if (!ntdb->file) {
+		int fd;
+
+		if ((fd = open(name, open_flags, mode)) == -1) {
+			/* errno set by open(2) */
+			saved_errno = errno;
+			ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				   "ntdb_open: could not open file %s: %s",
+				   name, strerror(errno));
+			goto fail_errno;
+		}
+
+		/* on exec, don't inherit the fd */
+		v = fcntl(fd, F_GETFD, 0);
+		fcntl(fd, F_SETFD, v | FD_CLOEXEC);
+
+		if (fstat(fd, &st) == -1) {
+			saved_errno = errno;
+			ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				   "ntdb_open: could not stat open %s: %s",
+				   name, strerror(errno));
+			close(fd);
+			goto fail_errno;
+		}
+
+		ecode = ntdb_new_file(ntdb);
+		if (ecode != NTDB_SUCCESS) {
+			close(fd);
+			goto fail;
+		}
+
+		ntdb->file->fd = fd;
+		ntdb->file->device = st.st_dev;
+		ntdb->file->inode = st.st_ino;
+		ntdb->file->map_ptr = NULL;
+		ntdb->file->map_size = 0;
+	}
+
+	/* ensure there is only one process initialising at once */
+	ecode = ntdb_lock_open(ntdb, openlock, NTDB_LOCK_WAIT|NTDB_LOCK_NOCHECK);
+	if (ecode != NTDB_SUCCESS) {
+		saved_errno = errno;
+		goto fail_errno;
+	}
+
+	/* call their open hook if they gave us one. */
+	if (ntdb->openhook) {
+		ecode = ntdb->openhook(ntdb->file->fd, ntdb->openhook_data);
+		if (ecode != NTDB_SUCCESS) {
+			ntdb_logerr(ntdb, ecode, NTDB_LOG_ERROR,
+				   "ntdb_open: open hook failed");
+			goto fail;
+		}
+		open_flags |= O_CREAT;
+	}
+
+	/* If they used O_TRUNC, read will return 0. */
+	rlen = pread(ntdb->file->fd, &hdr, sizeof(hdr), 0);
+	if (rlen == 0 && (open_flags & O_CREAT)) {
+		ecode = ntdb_new_database(ntdb, seed, &hdr);
+		if (ecode != NTDB_SUCCESS) {
+			goto fail;
+		}
+	} else if (rlen < 0) {
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				   "ntdb_open: error %s reading %s",
+				   strerror(errno), name);
+		goto fail;
+	} else if (rlen < sizeof(hdr)
+		   || strcmp(hdr.magic_food, NTDB_MAGIC_FOOD) != 0) {
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				   "ntdb_open: %s is not a ntdb file", name);
+		goto fail;
+	}
+
+	if (hdr.version != NTDB_VERSION) {
+		if (hdr.version == bswap_64(NTDB_VERSION))
+			ntdb->flags |= NTDB_CONVERT;
+		else {
+			/* wrong version */
+			ecode = ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+					   "ntdb_open:"
+					   " %s is unknown version 0x%llx",
+					   name, (long long)hdr.version);
+			goto fail;
+		}
+	} else if (ntdb->flags & NTDB_CONVERT) {
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				   "ntdb_open:"
+				   " %s does not need NTDB_CONVERT",
+				   name);
+		goto fail;
+	}
+
+	ntdb_context_init(ntdb);
+
+	ntdb_convert(ntdb, &hdr, sizeof(hdr));
+	ntdb->hash_seed = hdr.hash_seed;
+	hash_test = NTDB_HASH_MAGIC;
+	hash_test = ntdb_hash(ntdb, &hash_test, sizeof(hash_test));
+	if (hdr.hash_test != hash_test) {
+		/* wrong hash variant */
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+				   "ntdb_open:"
+				   " %s uses a different hash function",
+				   name);
+		goto fail;
+	}
+
+	ecode = capabilities_ok(ntdb, hdr.capabilities);
+	if (ecode != NTDB_SUCCESS) {
+		goto fail;
+	}
+
+	/* Clear any features we don't understand. */
+	if ((open_flags & O_ACCMODE) != O_RDONLY) {
+		hdr.features_used &= NTDB_FEATURE_MASK;
+		ecode = ntdb_write_convert(ntdb, offsetof(struct ntdb_header,
+							features_used),
+					  &hdr.features_used,
+					  sizeof(hdr.features_used));
+		if (ecode != NTDB_SUCCESS)
+			goto fail;
+	}
+
+	ntdb_unlock_open(ntdb, openlock);
+
+	/* This makes sure we have current map_size and mmap. */
+	ecode = ntdb->io->oob(ntdb, ntdb->file->map_size, 1, true);
+	if (unlikely(ecode != NTDB_SUCCESS))
+		goto fail;
+
+	/* Now it's fully formed, recover if necessary. */
+	berr = ntdb_needs_recovery(ntdb);
+	if (unlikely(berr != false)) {
+		if (berr < 0) {
+			ecode = NTDB_OFF_TO_ERR(berr);
+			goto fail;
+		}
+		ecode = ntdb_lock_and_recover(ntdb);
+		if (ecode != NTDB_SUCCESS) {
+			goto fail;
+		}
+	}
+
+	ecode = ntdb_ftable_init(ntdb);
+	if (ecode != NTDB_SUCCESS) {
+		goto fail;
+	}
+
+	ntdb->next = tdbs;
+	tdbs = ntdb;
+	return ntdb;
+
+ fail:
+	/* Map ecode to some logical errno. */
+	switch (NTDB_ERR_TO_OFF(ecode)) {
+	case NTDB_ERR_TO_OFF(NTDB_ERR_CORRUPT):
+	case NTDB_ERR_TO_OFF(NTDB_ERR_IO):
+		saved_errno = EIO;
+		break;
+	case NTDB_ERR_TO_OFF(NTDB_ERR_LOCK):
+		saved_errno = EWOULDBLOCK;
+		break;
+	case NTDB_ERR_TO_OFF(NTDB_ERR_OOM):
+		saved_errno = ENOMEM;
+		break;
+	case NTDB_ERR_TO_OFF(NTDB_ERR_EINVAL):
+		saved_errno = EINVAL;
+		break;
+	default:
+		saved_errno = EINVAL;
+		break;
+	}
+
+fail_errno:
+#ifdef NTDB_TRACE
+	close(ntdb->tracefd);
+#endif
+	if (ntdb->file) {
+		ntdb_lock_cleanup(ntdb);
+		if (--ntdb->file->refcnt == 0) {
+			assert(ntdb->file->num_lockrecs == 0);
+			if (ntdb->file->map_ptr) {
+				if (ntdb->flags & NTDB_INTERNAL) {
+					free(ntdb->file->map_ptr);
+				} else
+					ntdb_munmap(ntdb->file);
+			}
+			if (close(ntdb->file->fd) != 0)
+				ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
+					   "ntdb_open: failed to close ntdb fd"
+					   " on error: %s", strerror(errno));
+			free(ntdb->file->lockrecs);
+			free(ntdb->file);
+		}
+	}
+
+	free(ntdb);
+	errno = saved_errno;
+	return NULL;
+}
+
+_PUBLIC_ int ntdb_close(struct ntdb_context *ntdb)
+{
+	int ret = 0;
+	struct ntdb_context **i;
+
+	ntdb_trace(ntdb, "ntdb_close");
+
+	if (ntdb->transaction) {
+		ntdb_transaction_cancel(ntdb);
+	}
+
+	if (ntdb->file->map_ptr) {
+		if (ntdb->flags & NTDB_INTERNAL)
+			free(ntdb->file->map_ptr);
+		else
+			ntdb_munmap(ntdb->file);
+	}
+	if (ntdb->file) {
+		ntdb_lock_cleanup(ntdb);
+		if (--ntdb->file->refcnt == 0) {
+			ret = close(ntdb->file->fd);
+			free(ntdb->file->lockrecs);
+			free(ntdb->file);
+		}
+	}
+
+	/* Remove from tdbs list */
+	for (i = &tdbs; *i; i = &(*i)->next) {
+		if (*i == ntdb) {
+			*i = ntdb->next;
+			break;
+		}
+	}
+
+#ifdef NTDB_TRACE
+	close(ntdb->tracefd);
+#endif
+	free(ntdb);
+
+	return ret;
+}
+
+_PUBLIC_ void ntdb_foreach_(int (*fn)(struct ntdb_context *, void *), void *p)
+{
+	struct ntdb_context *i;
+
+	for (i = tdbs; i; i = i->next) {
+		if (fn(i, p) != 0)
+			break;
+	}
+}

lib/ntdb/private.h

@@ -0,0 +1,657 @@
+#ifndef NTDB_PRIVATE_H
+#define NTDB_PRIVATE_H
+/*
+  Trivial Database 2: private types and prototypes
+  Copyright (C) Rusty Russell 2010
+
+  This library is free software; you can redistribute it and/or
+  modify it under the terms of the GNU Lesser General Public
+  License as published by the Free Software Foundation; either
+  version 3 of the License, or (at your option) any later version.
+
+  This library is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+  Lesser General Public License for more details.
+
+  You should have received a copy of the GNU Lesser General Public
+  License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "config.h"
+#ifndef HAVE_CCAN
+#error You need ccan to build ntdb!
+#endif
+#include "ntdb.h"
+#include <ccan/compiler/compiler.h>
+#include <ccan/likely/likely.h>
+#include <ccan/endian/endian.h>
+
+#ifdef HAVE_LIBREPLACE
+#include "replace.h"
+#include "system/filesys.h"
+#include "system/time.h"
+#include "system/shmem.h"
+#include "system/select.h"
+#include "system/wait.h"
+#else
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdlib.h>
+#include <stddef.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <stdio.h>
+#include <utime.h>
+#include <unistd.h>
+#endif
+
+#ifndef TEST_IT
+#define TEST_IT(cond)
+#endif
+
+/* #define NTDB_TRACE 1 */
+
+#ifndef __STRING
+#define __STRING(x)    #x
+#endif
+
+#ifndef __STRINGSTRING
+#define __STRINGSTRING(x) __STRING(x)
+#endif
+
+#ifndef __location__
+#define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
+#endif
+
+typedef uint64_t ntdb_len_t;
+typedef uint64_t ntdb_off_t;
+
+#define NTDB_MAGIC_FOOD "NTDB file\n"
+#define NTDB_VERSION ((uint64_t)(0x26011967 + 7))
+#define NTDB_USED_MAGIC ((uint64_t)0x1999)
+#define NTDB_HTABLE_MAGIC ((uint64_t)0x1888)
+#define NTDB_CHAIN_MAGIC ((uint64_t)0x1777)
+#define NTDB_FTABLE_MAGIC ((uint64_t)0x1666)
+#define NTDB_CAP_MAGIC ((uint64_t)0x1555)
+#define NTDB_FREE_MAGIC ((uint64_t)0xFE)
+#define NTDB_HASH_MAGIC (0xA1ABE11A01092008ULL)
+#define NTDB_RECOVERY_MAGIC (0xf53bc0e7ad124589ULL)
+#define NTDB_RECOVERY_INVALID_MAGIC (0x0ULL)
+
+/* Capability bits. */
+#define NTDB_CAP_TYPE_MASK	0x1FFFFFFFFFFFFFFFULL
+#define NTDB_CAP_NOCHECK		0x8000000000000000ULL
+#define NTDB_CAP_NOWRITE		0x4000000000000000ULL
+#define NTDB_CAP_NOOPEN		0x2000000000000000ULL
+
+#define NTDB_OFF_IS_ERR(off) unlikely(off >= (ntdb_off_t)(long)NTDB_ERR_LAST)
+#define NTDB_OFF_TO_ERR(off) ((enum NTDB_ERROR)(long)(off))
+#define NTDB_ERR_TO_OFF(ecode) ((ntdb_off_t)(long)(ecode))
+
+/* Packing errors into pointers and v.v. */
+#define NTDB_PTR_IS_ERR(ptr)						\
+	unlikely((unsigned long)(ptr) >= (unsigned long)NTDB_ERR_LAST)
+#define NTDB_PTR_ERR(p) ((enum NTDB_ERROR)(long)(p))
+#define NTDB_ERR_PTR(err) ((void *)(long)(err))
+
+/* Common case of returning true, false or -ve error. */
+typedef int ntdb_bool_err;
+
+/* Prevent others from opening the file. */
+#define NTDB_OPEN_LOCK 0
+/* Expanding file. */
+#define NTDB_EXPANSION_LOCK 2
+/* Doing a transaction. */
+#define NTDB_TRANSACTION_LOCK 8
+/* Hash chain locks. */
+#define NTDB_HASH_LOCK_START 64
+
+/* Range for hash locks. */
+#define NTDB_HASH_LOCK_RANGE_BITS 30
+#define NTDB_HASH_LOCK_RANGE (1 << NTDB_HASH_LOCK_RANGE_BITS)
+
+/* We have 1024 entries in the top level. */
+#define NTDB_TOPLEVEL_HASH_BITS 10
+/* And 64 entries in each sub-level: thus 64 bits exactly after 9 levels. */
+#define NTDB_SUBLEVEL_HASH_BITS 6
+/* And 8 entries in each group, ie 8 groups per sublevel. */
+#define NTDB_HASH_GROUP_BITS 3
+/* This is currently 10: beyond this we chain. */
+#define NTDB_MAX_LEVELS (1+(64-NTDB_TOPLEVEL_HASH_BITS) / NTDB_SUBLEVEL_HASH_BITS)
+
+/* Extend file by least 100 times larger than needed. */
+#define NTDB_EXTENSION_FACTOR 100
+
+/* We steal bits from the offsets to store hash info. */
+#define NTDB_OFF_HASH_GROUP_MASK ((1ULL << NTDB_HASH_GROUP_BITS) - 1)
+/* We steal this many upper bits, giving a maximum offset of 64 exabytes. */
+#define NTDB_OFF_UPPER_STEAL 8
+#define   NTDB_OFF_UPPER_STEAL_EXTRA 7
+/* The bit number where we store extra hash bits. */
+#define NTDB_OFF_HASH_EXTRA_BIT 57
+#define NTDB_OFF_UPPER_STEAL_SUBHASH_BIT 56
+
+/* Additional features we understand.  Currently: none. */
+#define NTDB_FEATURE_MASK ((uint64_t)0)
+
+/* The bit number where we store the extra hash bits. */
+/* Convenience mask to get actual offset. */
+#define NTDB_OFF_MASK							\
+	(((1ULL << (64 - NTDB_OFF_UPPER_STEAL)) - 1) - NTDB_OFF_HASH_GROUP_MASK)
+
+/* How many buckets in a free list: see size_to_bucket(). */
+#define NTDB_FREE_BUCKETS (64 - NTDB_OFF_UPPER_STEAL)
+
+/* We have to be able to fit a free record here. */
+#define NTDB_MIN_DATA_LEN						\
+	(sizeof(struct ntdb_free_record) - sizeof(struct ntdb_used_record))
+
+/* Indicates this entry is not on an flist (can happen during coalescing) */
+#define NTDB_FTABLE_NONE ((1ULL << NTDB_OFF_UPPER_STEAL) - 1)
+
+struct ntdb_used_record {
+	/* For on-disk compatibility, we avoid bitfields:
+	   magic: 16,        (highest)
+	   key_len_bits: 5,
+	   extra_padding: 32
+	   hash_bits: 11
+	*/
+        uint64_t magic_and_meta;
+	/* The bottom key_len_bits*2 are key length, rest is data length. */
+        uint64_t key_and_data_len;
+};
+
+static inline unsigned rec_key_bits(const struct ntdb_used_record *r)
+{
+	return ((r->magic_and_meta >> 43) & ((1 << 5)-1)) * 2;
+}
+
+static inline uint64_t rec_key_length(const struct ntdb_used_record *r)
+{
+	return r->key_and_data_len & ((1ULL << rec_key_bits(r)) - 1);
+}
+
+static inline uint64_t rec_data_length(const struct ntdb_used_record *r)
+{
+	return r->key_and_data_len >> rec_key_bits(r);
+}
+
+static inline uint64_t rec_extra_padding(const struct ntdb_used_record *r)
+{
+	return (r->magic_and_meta >> 11) & 0xFFFFFFFF;
+}
+
+static inline uint32_t rec_hash(const struct ntdb_used_record *r)
+{
+	return r->magic_and_meta & ((1 << 11) - 1);
+}
+
+static inline uint16_t rec_magic(const struct ntdb_used_record *r)
+{
+	return (r->magic_and_meta >> 48);
+}
+
+struct ntdb_free_record {
+        uint64_t magic_and_prev; /* NTDB_OFF_UPPER_STEAL bits magic, then prev */
+        uint64_t ftable_and_len; /* Len not counting these two fields. */
+	/* This is why the minimum record size is 8 bytes.  */
+	uint64_t next;
+};
+
+static inline uint64_t frec_prev(const struct ntdb_free_record *f)
+{
+	return f->magic_and_prev & ((1ULL << (64 - NTDB_OFF_UPPER_STEAL)) - 1);
+}
+
+static inline uint64_t frec_magic(const struct ntdb_free_record *f)
+{
+	return f->magic_and_prev >> (64 - NTDB_OFF_UPPER_STEAL);
+}
+
+static inline uint64_t frec_len(const struct ntdb_free_record *f)
+{
+	return f->ftable_and_len & ((1ULL << (64 - NTDB_OFF_UPPER_STEAL))-1);
+}
+
+static inline unsigned frec_ftable(const struct ntdb_free_record *f)
+{
+	return f->ftable_and_len >> (64 - NTDB_OFF_UPPER_STEAL);
+}
+
+struct ntdb_recovery_record {
+	uint64_t magic;
+	/* Length of record (add this header to get total length). */
+	uint64_t max_len;
+	/* Length used. */
+	uint64_t len;
+	/* Old length of file before transaction. */
+	uint64_t eof;
+};
+
+/* If we bottom out of the subhashes, we chain. */
+struct ntdb_chain {
+	ntdb_off_t rec[1 << NTDB_HASH_GROUP_BITS];
+	ntdb_off_t next;
+};
+
+/* this is stored at the front of every database */
+struct ntdb_header {
+	char magic_food[64]; /* for /etc/magic */
+	/* FIXME: Make me 32 bit? */
+	uint64_t version; /* version of the code */
+	uint64_t hash_test; /* result of hashing HASH_MAGIC. */
+	uint64_t hash_seed; /* "random" seed written at creation time. */
+	ntdb_off_t free_table; /* (First) free table. */
+	ntdb_off_t recovery; /* Transaction recovery area. */
+
+	uint64_t features_used; /* Features all writers understand */
+	uint64_t features_offered; /* Features offered */
+
+	uint64_t seqnum; /* Sequence number for NTDB_SEQNUM */
+
+	ntdb_off_t capabilities; /* Optional linked list of capabilities. */
+	ntdb_off_t reserved[22];
+
+	/* Top level hash table. */
+	ntdb_off_t hashtable[1ULL << NTDB_TOPLEVEL_HASH_BITS];
+};
+
+struct ntdb_freetable {
+	struct ntdb_used_record hdr;
+	ntdb_off_t next;
+	ntdb_off_t buckets[NTDB_FREE_BUCKETS];
+};
+
+struct ntdb_capability {
+	struct ntdb_used_record hdr;
+	ntdb_off_t type;
+	ntdb_off_t next;
+	/* ... */
+};
+
+/* Information about a particular (locked) hash entry. */
+struct hash_info {
+	/* Full hash value of entry. */
+	uint64_t h;
+	/* Start and length of lock acquired. */
+	ntdb_off_t hlock_start;
+	ntdb_len_t hlock_range;
+	/* Start of hash group. */
+	ntdb_off_t group_start;
+	/* Bucket we belong in. */
+	unsigned int home_bucket;
+	/* Bucket we (or an empty space) were found in. */
+	unsigned int found_bucket;
+	/* How many bits of the hash are already used. */
+	unsigned int hash_used;
+	/* Current working group. */
+	ntdb_off_t group[1 << NTDB_HASH_GROUP_BITS];
+};
+
+struct traverse_info {
+	struct traverse_level {
+		ntdb_off_t hashtable;
+		/* We ignore groups here, and treat it as a big array. */
+		unsigned entry;
+		unsigned int total_buckets;
+	} levels[NTDB_MAX_LEVELS + 1];
+	unsigned int num_levels;
+	unsigned int toplevel_group;
+	/* This makes delete-everything-inside-traverse work as expected. */
+	ntdb_off_t prev;
+};
+
+enum ntdb_lock_flags {
+	/* WAIT == F_SETLKW, NOWAIT == F_SETLK */
+	NTDB_LOCK_NOWAIT = 0,
+	NTDB_LOCK_WAIT = 1,
+	/* If set, don't log an error on failure. */
+	NTDB_LOCK_PROBE = 2,
+	/* If set, don't check for recovery (used by recovery code). */
+	NTDB_LOCK_NOCHECK = 4,
+};
+
+struct ntdb_lock {
+	struct ntdb_context *owner;
+	off_t off;
+	uint32_t count;
+	uint32_t ltype;
+};
+
+/* This is only needed for ntdb_access_commit, but used everywhere to
+ * simplify. */
+struct ntdb_access_hdr {
+	struct ntdb_access_hdr *next;
+	ntdb_off_t off;
+	ntdb_len_t len;
+	bool convert;
+};
+
+struct ntdb_file {
+	/* How many are sharing us? */
+	unsigned int refcnt;
+
+	/* Mmap (if any), or malloc (for NTDB_INTERNAL). */
+	void *map_ptr;
+
+	/* How much space has been mapped (<= current file size) */
+	ntdb_len_t map_size;
+
+	/* The file descriptor (-1 for NTDB_INTERNAL). */
+	int fd;
+
+	/* Lock information */
+	pid_t locker;
+	struct ntdb_lock allrecord_lock;
+	size_t num_lockrecs;
+	struct ntdb_lock *lockrecs;
+
+	/* Identity of this file. */
+	dev_t device;
+	ino_t inode;
+};
+
+struct ntdb_methods {
+	enum NTDB_ERROR (*tread)(struct ntdb_context *, ntdb_off_t, void *,
+				 ntdb_len_t);
+	enum NTDB_ERROR (*twrite)(struct ntdb_context *, ntdb_off_t, const void *,
+				  ntdb_len_t);
+	enum NTDB_ERROR (*oob)(struct ntdb_context *, ntdb_off_t, ntdb_len_t, bool);
+	enum NTDB_ERROR (*expand_file)(struct ntdb_context *, ntdb_len_t);
+	void *(*direct)(struct ntdb_context *, ntdb_off_t, size_t, bool);
+};
+
+/*
+  internal prototypes
+*/
+/* hash.c: */
+uint64_t ntdb_jenkins_hash(const void *key, size_t length, uint64_t seed,
+			   void *unused);
+
+enum NTDB_ERROR first_in_hash(struct ntdb_context *ntdb,
+			      struct traverse_info *tinfo,
+			      NTDB_DATA *kbuf, size_t *dlen);
+
+enum NTDB_ERROR next_in_hash(struct ntdb_context *ntdb,
+			     struct traverse_info *tinfo,
+			     NTDB_DATA *kbuf, size_t *dlen);
+
+/* Hash random memory. */
+uint64_t ntdb_hash(struct ntdb_context *ntdb, const void *ptr, size_t len);
+
+/* Hash on disk. */
+uint64_t hash_record(struct ntdb_context *ntdb, ntdb_off_t off);
+
+/* Find and lock a hash entry (or where it would be). */
+ntdb_off_t find_and_lock(struct ntdb_context *ntdb,
+			 NTDB_DATA key,
+			 int ltype,
+			 struct hash_info *h,
+			 struct ntdb_used_record *rec,
+			 struct traverse_info *tinfo);
+
+enum NTDB_ERROR replace_in_hash(struct ntdb_context *ntdb,
+				struct hash_info *h,
+				ntdb_off_t new_off);
+
+enum NTDB_ERROR add_to_hash(struct ntdb_context *ntdb, struct hash_info *h,
+			    ntdb_off_t new_off);
+
+enum NTDB_ERROR delete_from_hash(struct ntdb_context *ntdb, struct hash_info *h);
+
+/* For ntdb_check */
+bool is_subhash(ntdb_off_t val);
+enum NTDB_ERROR unknown_capability(struct ntdb_context *ntdb, const char *caller,
+				   ntdb_off_t type);
+
+/* free.c: */
+enum NTDB_ERROR ntdb_ftable_init(struct ntdb_context *ntdb);
+
+/* check.c needs these to iterate through free lists. */
+ntdb_off_t first_ftable(struct ntdb_context *ntdb);
+ntdb_off_t next_ftable(struct ntdb_context *ntdb, ntdb_off_t ftable);
+
+/* This returns space or -ve error number. */
+ntdb_off_t alloc(struct ntdb_context *ntdb, size_t keylen, size_t datalen,
+		 uint64_t hash, unsigned magic, bool growing);
+
+/* Put this record in a free list. */
+enum NTDB_ERROR add_free_record(struct ntdb_context *ntdb,
+				ntdb_off_t off, ntdb_len_t len_with_header,
+				enum ntdb_lock_flags waitflag,
+				bool coalesce_ok);
+
+/* Set up header for a used/ftable/htable/chain/capability record. */
+enum NTDB_ERROR set_header(struct ntdb_context *ntdb,
+			   struct ntdb_used_record *rec,
+			   unsigned magic, uint64_t keylen, uint64_t datalen,
+			   uint64_t actuallen, unsigned hashlow);
+
+/* Used by ntdb_check to verify. */
+unsigned int size_to_bucket(ntdb_len_t data_len);
+ntdb_off_t bucket_off(ntdb_off_t ftable_off, unsigned bucket);
+
+/* Used by ntdb_summary */
+ntdb_off_t dead_space(struct ntdb_context *ntdb, ntdb_off_t off);
+
+/* Adjust expansion, used by create_recovery_area */
+ntdb_off_t ntdb_expand_adjust(ntdb_off_t map_size, ntdb_off_t size);
+
+/* io.c: */
+/* Initialize ntdb->methods. */
+void ntdb_io_init(struct ntdb_context *ntdb);
+
+/* Convert endian of the buffer if required. */
+void *ntdb_convert(const struct ntdb_context *ntdb, void *buf, ntdb_len_t size);
+
+/* Unmap and try to map the ntdb. */
+void ntdb_munmap(struct ntdb_file *file);
+enum NTDB_ERROR ntdb_mmap(struct ntdb_context *ntdb);
+
+/* Either alloc a copy, or give direct access.  Release frees or noop. */
+const void *ntdb_access_read(struct ntdb_context *ntdb,
+			     ntdb_off_t off, ntdb_len_t len, bool convert);
+void *ntdb_access_write(struct ntdb_context *ntdb,
+			ntdb_off_t off, ntdb_len_t len, bool convert);
+
+/* Release result of ntdb_access_read/write. */
+void ntdb_access_release(struct ntdb_context *ntdb, const void *p);
+/* Commit result of ntdb_acces_write. */
+enum NTDB_ERROR ntdb_access_commit(struct ntdb_context *ntdb, void *p);
+
+/* Convenience routine to get an offset. */
+ntdb_off_t ntdb_read_off(struct ntdb_context *ntdb, ntdb_off_t off);
+
+/* Write an offset at an offset. */
+enum NTDB_ERROR ntdb_write_off(struct ntdb_context *ntdb, ntdb_off_t off,
+			       ntdb_off_t val);
+
+/* Clear an ondisk area. */
+enum NTDB_ERROR zero_out(struct ntdb_context *ntdb, ntdb_off_t off, ntdb_len_t len);
+
+/* Return a non-zero offset between >= start < end in this array (or end). */
+ntdb_off_t ntdb_find_nonzero_off(struct ntdb_context *ntdb,
+				 ntdb_off_t base,
+				 uint64_t start,
+				 uint64_t end);
+
+/* Return a zero offset in this array, or num. */
+ntdb_off_t ntdb_find_zero_off(struct ntdb_context *ntdb, ntdb_off_t off,
+			      uint64_t num);
+
+/* Allocate and make a copy of some offset. */
+void *ntdb_alloc_read(struct ntdb_context *ntdb, ntdb_off_t offset, ntdb_len_t len);
+
+/* Writes a converted copy of a record. */
+enum NTDB_ERROR ntdb_write_convert(struct ntdb_context *ntdb, ntdb_off_t off,
+				   const void *rec, size_t len);
+
+/* Reads record and converts it */
+enum NTDB_ERROR ntdb_read_convert(struct ntdb_context *ntdb, ntdb_off_t off,
+				  void *rec, size_t len);
+
+/* Bump the seqnum (caller checks for ntdb->flags & NTDB_SEQNUM) */
+void ntdb_inc_seqnum(struct ntdb_context *ntdb);
+
+/* lock.c: */
+/* Print message because another ntdb owns a lock we want. */
+enum NTDB_ERROR owner_conflict(struct ntdb_context *ntdb, const char *call);
+
+/* If we fork, we no longer really own locks. */
+bool check_lock_pid(struct ntdb_context *ntdb, const char *call, bool log);
+
+/* Lock/unlock a range of hashes. */
+enum NTDB_ERROR ntdb_lock_hashes(struct ntdb_context *ntdb,
+				 ntdb_off_t hash_lock, ntdb_len_t hash_range,
+				 int ltype, enum ntdb_lock_flags waitflag);
+enum NTDB_ERROR ntdb_unlock_hashes(struct ntdb_context *ntdb,
+				   ntdb_off_t hash_lock,
+				   ntdb_len_t hash_range, int ltype);
+
+/* For closing the file. */
+void ntdb_lock_cleanup(struct ntdb_context *ntdb);
+
+/* Lock/unlock a particular free bucket. */
+enum NTDB_ERROR ntdb_lock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off,
+				      enum ntdb_lock_flags waitflag);
+void ntdb_unlock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off);
+
+/* Serialize transaction start. */
+enum NTDB_ERROR ntdb_transaction_lock(struct ntdb_context *ntdb, int ltype);
+void ntdb_transaction_unlock(struct ntdb_context *ntdb, int ltype);
+
+/* Do we have any hash locks (ie. via ntdb_chainlock) ? */
+bool ntdb_has_hash_locks(struct ntdb_context *ntdb);
+
+/* Lock entire database. */
+enum NTDB_ERROR ntdb_allrecord_lock(struct ntdb_context *ntdb, int ltype,
+				    enum ntdb_lock_flags flags, bool upgradable);
+void ntdb_allrecord_unlock(struct ntdb_context *ntdb, int ltype);
+enum NTDB_ERROR ntdb_allrecord_upgrade(struct ntdb_context *ntdb, off_t start);
+
+/* Serialize db open. */
+enum NTDB_ERROR ntdb_lock_open(struct ntdb_context *ntdb,
+			       int ltype, enum ntdb_lock_flags flags);
+void ntdb_unlock_open(struct ntdb_context *ntdb, int ltype);
+bool ntdb_has_open_lock(struct ntdb_context *ntdb);
+
+/* Serialize db expand. */
+enum NTDB_ERROR ntdb_lock_expand(struct ntdb_context *ntdb, int ltype);
+void ntdb_unlock_expand(struct ntdb_context *ntdb, int ltype);
+bool ntdb_has_expansion_lock(struct ntdb_context *ntdb);
+
+/* If it needs recovery, grab all the locks and do it. */
+enum NTDB_ERROR ntdb_lock_and_recover(struct ntdb_context *ntdb);
+
+/* Default lock and unlock functions. */
+int ntdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag, void *);
+int ntdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *);
+
+/* transaction.c: */
+enum NTDB_ERROR ntdb_transaction_recover(struct ntdb_context *ntdb);
+ntdb_bool_err ntdb_needs_recovery(struct ntdb_context *ntdb);
+
+struct ntdb_context {
+	/* Single list of all TDBs, to detect multiple opens. */
+	struct ntdb_context *next;
+
+	/* Filename of the database. */
+	const char *name;
+
+	/* Logging function */
+	void (*log_fn)(struct ntdb_context *ntdb,
+		       enum ntdb_log_level level,
+		       enum NTDB_ERROR ecode,
+		       const char *message,
+		       void *data);
+	void *log_data;
+
+	/* Open flags passed to ntdb_open. */
+	int open_flags;
+
+	/* low level (fnctl) lock functions. */
+	int (*lock_fn)(int fd, int rw, off_t off, off_t len, bool w, void *);
+	int (*unlock_fn)(int fd, int rw, off_t off, off_t len, void *);
+	void *lock_data;
+
+	/* the ntdb flags passed to ntdb_open. */
+	uint32_t flags;
+
+	/* Our statistics. */
+	struct ntdb_attribute_stats stats;
+
+	/* The actual file information */
+	struct ntdb_file *file;
+
+	/* Hash function. */
+	uint64_t (*hash_fn)(const void *key, size_t len, uint64_t seed, void *);
+	void *hash_data;
+	uint64_t hash_seed;
+
+	/* Our open hook, if any. */
+	enum NTDB_ERROR (*openhook)(int fd, void *data);
+	void *openhook_data;
+
+	/* Last error we returned. */
+	enum NTDB_ERROR last_error;
+
+	/* Are we accessing directly? (debugging check). */
+	int direct_access;
+
+	/* Set if we are in a transaction. */
+	struct ntdb_transaction *transaction;
+
+	/* What free table are we using? */
+	ntdb_off_t ftable_off;
+	unsigned int ftable;
+
+	/* IO methods: changes for transactions. */
+	const struct ntdb_methods *io;
+
+	/* Direct access information */
+	struct ntdb_access_hdr *access;
+};
+
+/* ntdb.c: */
+enum NTDB_ERROR COLD PRINTF_FMT(4, 5)
+	ntdb_logerr(struct ntdb_context *ntdb,
+		    enum NTDB_ERROR ecode,
+		    enum ntdb_log_level level,
+		    const char *fmt, ...);
+
+#ifdef NTDB_TRACE
+void ntdb_trace(struct ntdb_context *ntdb, const char *op);
+void ntdb_trace_seqnum(struct ntdb_context *ntdb, uint32_t seqnum, const char *op);
+void ntdb_trace_open(struct ntdb_context *ntdb, const char *op,
+		     unsigned hash_size, unsigned ntdb_flags, unsigned open_flags);
+void ntdb_trace_ret(struct ntdb_context *ntdb, const char *op, int ret);
+void ntdb_trace_retrec(struct ntdb_context *ntdb, const char *op, NTDB_DATA ret);
+void ntdb_trace_1rec(struct ntdb_context *ntdb, const char *op,
+		     NTDB_DATA rec);
+void ntdb_trace_1rec_ret(struct ntdb_context *ntdb, const char *op,
+			 NTDB_DATA rec, int ret);
+void ntdb_trace_1rec_retrec(struct ntdb_context *ntdb, const char *op,
+			    NTDB_DATA rec, NTDB_DATA ret);
+void ntdb_trace_2rec_flag_ret(struct ntdb_context *ntdb, const char *op,
+			      NTDB_DATA rec1, NTDB_DATA rec2, unsigned flag,
+			      int ret);
+void ntdb_trace_2rec_retrec(struct ntdb_context *ntdb, const char *op,
+			    NTDB_DATA rec1, NTDB_DATA rec2, NTDB_DATA ret);
+#else
+#define ntdb_trace(ntdb, op)
+#define ntdb_trace_seqnum(ntdb, seqnum, op)
+#define ntdb_trace_open(ntdb, op, hash_size, ntdb_flags, open_flags)
+#define ntdb_trace_ret(ntdb, op, ret)
+#define ntdb_trace_retrec(ntdb, op, ret)
+#define ntdb_trace_1rec(ntdb, op, rec)
+#define ntdb_trace_1rec_ret(ntdb, op, rec, ret)
+#define ntdb_trace_1rec_retrec(ntdb, op, rec, ret)
+#define ntdb_trace_2rec_flag_ret(ntdb, op, rec1, rec2, flag, ret)
+#define ntdb_trace_2rec_retrec(ntdb, op, rec1, rec2, ret)
+#endif /* !NTDB_TRACE */
+
+#endif

lib/ntdb/pyntdb.c

@@ -0,0 +1,591 @@
+/*
+   Unix SMB/CIFS implementation.
+
+   Python interface to ntdb.  Simply modified from tdb version.
+
+   Copyright (C) 2004-2006 Tim Potter <tpot@samba.org>
+   Copyright (C) 2007-2008 Jelmer Vernooij <jelmer@samba.org>
+   Copyright (C) 2011 Rusty Russell <rusty@rustcorp.com.au>
+
+     ** NOTE! The following LGPL license applies to the ntdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include <Python.h>
+#include "replace.h"
+#include "system/filesys.h"
+
+#ifndef Py_RETURN_NONE
+#define Py_RETURN_NONE return Py_INCREF(Py_None), Py_None
+#endif
+
+/* Include ntdb headers */
+#include <ntdb.h>
+
+typedef struct {
+	PyObject_HEAD
+	struct ntdb_context *ctx;
+	bool closed;
+} PyNtdbObject;
+
+staticforward PyTypeObject PyNtdb;
+
+static void PyErr_SetTDBError(enum NTDB_ERROR e)
+{
+	PyErr_SetObject(PyExc_RuntimeError,
+		Py_BuildValue("(i,s)", e, ntdb_errorstr(e)));
+}
+
+static NTDB_DATA PyString_AsNtdb_Data(PyObject *data)
+{
+	NTDB_DATA ret;
+	ret.dptr = (unsigned char *)PyString_AsString(data);
+	ret.dsize = PyString_Size(data);
+	return ret;
+}
+
+static PyObject *PyString_FromNtdb_Data(NTDB_DATA data)
+{
+	PyObject *ret = PyString_FromStringAndSize((const char *)data.dptr,
+						   data.dsize);
+	free(data.dptr);
+	return ret;
+}
+
+#define PyErr_NTDB_ERROR_IS_ERR_RAISE(ret) \
+	if (ret != NTDB_SUCCESS) { \
+		PyErr_SetTDBError(ret); \
+		return NULL; \
+	}
+
+static void stderr_log(struct ntdb_context *ntdb,
+		       enum ntdb_log_level level,
+		       enum NTDB_ERROR ecode,
+		       const char *message,
+		       void *data)
+{
+	fprintf(stderr, "%s:%s:%s\n",
+		ntdb_name(ntdb), ntdb_errorstr(ecode), message);
+}
+
+static PyObject *py_ntdb_open(PyTypeObject *type, PyObject *args, PyObject *kwargs)
+{
+	char *name = NULL;
+	int ntdb_flags = NTDB_DEFAULT, flags = O_RDWR, mode = 0600;
+	struct ntdb_context *ctx;
+	PyNtdbObject *ret;
+	union ntdb_attribute logattr;
+	const char *kwnames[] = { "name", "ntdb_flags", "flags", "mode", NULL };
+
+	if (!PyArg_ParseTupleAndKeywords(args, kwargs, "|siii", cast_const2(char **, kwnames), &name, &ntdb_flags, &flags, &mode))
+		return NULL;
+
+	if (name == NULL) {
+		ntdb_flags |= NTDB_INTERNAL;
+	}
+
+	logattr.log.base.attr = NTDB_ATTRIBUTE_LOG;
+	logattr.log.base.next = NULL;
+	logattr.log.fn = stderr_log;
+	ctx = ntdb_open(name, ntdb_flags, flags, mode, &logattr);
+	if (ctx == NULL) {
+		PyErr_SetFromErrno(PyExc_IOError);
+		return NULL;
+	}
+
+	ret = PyObject_New(PyNtdbObject, &PyNtdb);
+	if (!ret) {
+		ntdb_close(ctx);
+		return NULL;
+	}
+
+	ret->ctx = ctx;
+	ret->closed = false;
+	return (PyObject *)ret;
+}
+
+static PyObject *obj_transaction_cancel(PyNtdbObject *self)
+{
+	ntdb_transaction_cancel(self->ctx);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_transaction_commit(PyNtdbObject *self)
+{
+	enum NTDB_ERROR ret = ntdb_transaction_commit(self->ctx);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_transaction_prepare_commit(PyNtdbObject *self)
+{
+	enum NTDB_ERROR ret = ntdb_transaction_prepare_commit(self->ctx);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_transaction_start(PyNtdbObject *self)
+{
+	enum NTDB_ERROR ret = ntdb_transaction_start(self->ctx);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_lockall(PyNtdbObject *self)
+{
+	enum NTDB_ERROR ret = ntdb_lockall(self->ctx);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_unlockall(PyNtdbObject *self)
+{
+	ntdb_unlockall(self->ctx);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_lockall_read(PyNtdbObject *self)
+{
+	enum NTDB_ERROR ret = ntdb_lockall_read(self->ctx);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_unlockall_read(PyNtdbObject *self)
+{
+	ntdb_unlockall_read(self->ctx);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_close(PyNtdbObject *self)
+{
+	int ret;
+	if (self->closed)
+		Py_RETURN_NONE;
+	ret = ntdb_close(self->ctx);
+	self->closed = true;
+	if (ret != 0) {
+		PyErr_SetTDBError(NTDB_ERR_IO);
+		return NULL;
+	}
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_get(PyNtdbObject *self, PyObject *args)
+{
+	NTDB_DATA key, data;
+	PyObject *py_key;
+	enum NTDB_ERROR ret;
+	if (!PyArg_ParseTuple(args, "O", &py_key))
+		return NULL;
+
+	key = PyString_AsNtdb_Data(py_key);
+	ret = ntdb_fetch(self->ctx, key, &data);
+	if (ret == NTDB_ERR_NOEXIST)
+		Py_RETURN_NONE;
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	return PyString_FromNtdb_Data(data);
+}
+
+static PyObject *obj_append(PyNtdbObject *self, PyObject *args)
+{
+	NTDB_DATA key, data;
+	PyObject *py_key, *py_data;
+	enum NTDB_ERROR ret;
+	if (!PyArg_ParseTuple(args, "OO", &py_key, &py_data))
+		return NULL;
+
+	key = PyString_AsNtdb_Data(py_key);
+	data = PyString_AsNtdb_Data(py_data);
+
+	ret = ntdb_append(self->ctx, key, data);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_firstkey(PyNtdbObject *self)
+{
+	enum NTDB_ERROR ret;
+	NTDB_DATA key;
+
+	ret = ntdb_firstkey(self->ctx, &key);
+	if (ret == NTDB_ERR_NOEXIST)
+		Py_RETURN_NONE;
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+
+	return PyString_FromNtdb_Data(key);
+}
+
+static PyObject *obj_nextkey(PyNtdbObject *self, PyObject *args)
+{
+	NTDB_DATA key;
+	PyObject *py_key;
+	enum NTDB_ERROR ret;
+	if (!PyArg_ParseTuple(args, "O", &py_key))
+		return NULL;
+
+	/* Malloc here, since ntdb_nextkey frees. */
+	key.dsize = PyString_Size(py_key);
+	key.dptr = malloc(key.dsize);
+	memcpy(key.dptr, PyString_AsString(py_key), key.dsize);
+
+	ret = ntdb_nextkey(self->ctx, &key);
+	if (ret == NTDB_ERR_NOEXIST)
+		Py_RETURN_NONE;
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+
+	return PyString_FromNtdb_Data(key);
+}
+
+static PyObject *obj_delete(PyNtdbObject *self, PyObject *args)
+{
+	NTDB_DATA key;
+	PyObject *py_key;
+	enum NTDB_ERROR ret;
+	if (!PyArg_ParseTuple(args, "O", &py_key))
+		return NULL;
+
+	key = PyString_AsNtdb_Data(py_key);
+	ret = ntdb_delete(self->ctx, key);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_has_key(PyNtdbObject *self, PyObject *args)
+{
+	NTDB_DATA key;
+	PyObject *py_key;
+	if (!PyArg_ParseTuple(args, "O", &py_key))
+		return NULL;
+
+	key = PyString_AsNtdb_Data(py_key);
+	if (ntdb_exists(self->ctx, key))
+		return Py_True;
+	if (ntdb_error(self->ctx) != NTDB_ERR_NOEXIST)
+		PyErr_NTDB_ERROR_IS_ERR_RAISE(ntdb_error(self->ctx));
+	return Py_False;
+}
+
+static PyObject *obj_store(PyNtdbObject *self, PyObject *args)
+{
+	NTDB_DATA key, value;
+	enum NTDB_ERROR ret;
+	int flag = NTDB_REPLACE;
+	PyObject *py_key, *py_value;
+
+	if (!PyArg_ParseTuple(args, "OO|i", &py_key, &py_value, &flag))
+		return NULL;
+
+	key = PyString_AsNtdb_Data(py_key);
+	value = PyString_AsNtdb_Data(py_value);
+
+	ret = ntdb_store(self->ctx, key, value, flag);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_add_flag(PyNtdbObject *self, PyObject *args)
+{
+	unsigned flag;
+
+	if (!PyArg_ParseTuple(args, "I", &flag))
+		return NULL;
+
+	ntdb_add_flag(self->ctx, flag);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_remove_flag(PyNtdbObject *self, PyObject *args)
+{
+	unsigned flag;
+
+	if (!PyArg_ParseTuple(args, "I", &flag))
+		return NULL;
+
+	ntdb_remove_flag(self->ctx, flag);
+	Py_RETURN_NONE;
+}
+
+typedef struct {
+	PyObject_HEAD
+	NTDB_DATA current;
+	bool end;
+	PyNtdbObject *iteratee;
+} PyNtdbIteratorObject;
+
+static PyObject *ntdb_iter_next(PyNtdbIteratorObject *self)
+{
+	enum NTDB_ERROR e;
+	PyObject *ret;
+	if (self->end)
+		return NULL;
+	ret = PyString_FromStringAndSize((const char *)self->current.dptr,
+					 self->current.dsize);
+	e = ntdb_nextkey(self->iteratee->ctx, &self->current);
+	if (e == NTDB_ERR_NOEXIST)
+		self->end = true;
+	else
+		PyErr_NTDB_ERROR_IS_ERR_RAISE(e);
+	return ret;
+}
+
+static void ntdb_iter_dealloc(PyNtdbIteratorObject *self)
+{
+	Py_DECREF(self->iteratee);
+	PyObject_Del(self);
+}
+
+PyTypeObject PyNtdbIterator = {
+	.tp_name = "Iterator",
+	.tp_basicsize = sizeof(PyNtdbIteratorObject),
+	.tp_iternext = (iternextfunc)ntdb_iter_next,
+	.tp_dealloc = (destructor)ntdb_iter_dealloc,
+	.tp_flags = Py_TPFLAGS_DEFAULT,
+	.tp_iter = PyObject_SelfIter,
+};
+
+static PyObject *ntdb_object_iter(PyNtdbObject *self)
+{
+	PyNtdbIteratorObject *ret;
+	enum NTDB_ERROR e;
+
+	ret = PyObject_New(PyNtdbIteratorObject, &PyNtdbIterator);
+	if (!ret)
+		return NULL;
+	e = ntdb_firstkey(self->ctx, &ret->current);
+	if (e == NTDB_ERR_NOEXIST) {
+		ret->end = true;
+	} else {
+		PyErr_NTDB_ERROR_IS_ERR_RAISE(e);
+		ret->end = false;
+	}
+	ret->iteratee = self;
+	Py_INCREF(self);
+	return (PyObject *)ret;
+}
+
+static PyObject *obj_clear(PyNtdbObject *self)
+{
+	enum NTDB_ERROR ret = ntdb_wipe_all(self->ctx);
+	PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+	Py_RETURN_NONE;
+}
+
+static PyObject *obj_enable_seqnum(PyNtdbObject *self)
+{
+	ntdb_add_flag(self->ctx, NTDB_SEQNUM);
+	Py_RETURN_NONE;
+}
+
+static PyMethodDef ntdb_object_methods[] = {
+	{ "transaction_cancel", (PyCFunction)obj_transaction_cancel, METH_NOARGS,
+		"S.transaction_cancel() -> None\n"
+		"Cancel the currently active transaction." },
+	{ "transaction_commit", (PyCFunction)obj_transaction_commit, METH_NOARGS,
+		"S.transaction_commit() -> None\n"
+		"Commit the currently active transaction." },
+	{ "transaction_prepare_commit", (PyCFunction)obj_transaction_prepare_commit, METH_NOARGS,
+		"S.transaction_prepare_commit() -> None\n"
+		"Prepare to commit the currently active transaction" },
+	{ "transaction_start", (PyCFunction)obj_transaction_start, METH_NOARGS,
+		"S.transaction_start() -> None\n"
+		"Start a new transaction." },
+	{ "lock_all", (PyCFunction)obj_lockall, METH_NOARGS, NULL },
+	{ "unlock_all", (PyCFunction)obj_unlockall, METH_NOARGS, NULL },
+	{ "read_lock_all", (PyCFunction)obj_lockall_read, METH_NOARGS, NULL },
+	{ "read_unlock_all", (PyCFunction)obj_unlockall_read, METH_NOARGS, NULL },
+	{ "close", (PyCFunction)obj_close, METH_NOARGS, NULL },
+	{ "get", (PyCFunction)obj_get, METH_VARARGS, "S.get(key) -> value\n"
+		"Fetch a value." },
+	{ "append", (PyCFunction)obj_append, METH_VARARGS, "S.append(key, value) -> None\n"
+		"Append data to an existing key." },
+	{ "firstkey", (PyCFunction)obj_firstkey, METH_NOARGS, "S.firstkey() -> data\n"
+		"Return the first key in this database." },
+	{ "nextkey", (PyCFunction)obj_nextkey, METH_NOARGS, "S.nextkey(key) -> data\n"
+		"Return the next key in this database." },
+	{ "delete", (PyCFunction)obj_delete, METH_VARARGS, "S.delete(key) -> None\n"
+		"Delete an entry." },
+	{ "has_key", (PyCFunction)obj_has_key, METH_VARARGS, "S.has_key(key) -> None\n"
+		"Check whether key exists in this database." },
+	{ "store", (PyCFunction)obj_store, METH_VARARGS, "S.store(key, data, flag=REPLACE) -> None"
+		"Store data." },
+	{ "add_flag", (PyCFunction)obj_add_flag, METH_VARARGS, "S.add_flag(flag) -> None" },
+	{ "remove_flag", (PyCFunction)obj_remove_flag, METH_VARARGS, "S.remove_flag(flag) -> None" },
+	{ "iterkeys", (PyCFunction)ntdb_object_iter, METH_NOARGS, "S.iterkeys() -> iterator" },
+	{ "clear", (PyCFunction)obj_clear, METH_NOARGS, "S.clear() -> None\n"
+		"Wipe the entire database." },
+	{ "enable_seqnum", (PyCFunction)obj_enable_seqnum, METH_NOARGS,
+		"S.enable_seqnum() -> None" },
+	{ NULL }
+};
+
+static PyObject *obj_get_flags(PyNtdbObject *self, void *closure)
+{
+	return PyInt_FromLong(ntdb_get_flags(self->ctx));
+}
+
+static PyObject *obj_get_filename(PyNtdbObject *self, void *closure)
+{
+	return PyString_FromString(ntdb_name(self->ctx));
+}
+
+static PyObject *obj_get_seqnum(PyNtdbObject *self, void *closure)
+{
+	return PyInt_FromLong(ntdb_get_seqnum(self->ctx));
+}
+
+
+static PyGetSetDef ntdb_object_getsetters[] = {
+	{ cast_const(char *, "flags"), (getter)obj_get_flags, NULL, NULL },
+	{ cast_const(char *, "filename"), (getter)obj_get_filename, NULL,
+	  cast_const(char *, "The filename of this NTDB file.")},
+	{ cast_const(char *, "seqnum"), (getter)obj_get_seqnum, NULL, NULL },
+	{ NULL }
+};
+
+static PyObject *ntdb_object_repr(PyNtdbObject *self)
+{
+	if (ntdb_get_flags(self->ctx) & NTDB_INTERNAL) {
+		return PyString_FromString("Ntdb(<internal>)");
+	} else {
+		return PyString_FromFormat("Ntdb('%s')", ntdb_name(self->ctx));
+	}
+}
+
+static void ntdb_object_dealloc(PyNtdbObject *self)
+{
+	if (!self->closed)
+		ntdb_close(self->ctx);
+	self->ob_type->tp_free(self);
+}
+
+static PyObject *obj_getitem(PyNtdbObject *self, PyObject *key)
+{
+	NTDB_DATA tkey, val;
+	enum NTDB_ERROR ret;
+
+	if (!PyString_Check(key)) {
+		PyErr_SetString(PyExc_TypeError, "Expected string as key");
+		return NULL;
+	}
+
+	tkey.dptr = (unsigned char *)PyString_AsString(key);
+	tkey.dsize = PyString_Size(key);
+
+	ret = ntdb_fetch(self->ctx, tkey, &val);
+	if (ret == NTDB_ERR_NOEXIST) {
+		PyErr_SetString(PyExc_KeyError, "No such NTDB entry");
+		return NULL;
+	} else {
+		PyErr_NTDB_ERROR_IS_ERR_RAISE(ret);
+		return PyString_FromNtdb_Data(val);
+	}
+}
+
+static int obj_setitem(PyNtdbObject *self, PyObject *key, PyObject *value)
+{
+	NTDB_DATA tkey, tval;
+	enum NTDB_ERROR ret;
+	if (!PyString_Check(key)) {
+		PyErr_SetString(PyExc_TypeError, "Expected string as key");
+		return -1;
+	}
+
+	tkey = PyString_AsNtdb_Data(key);
+
+	if (value == NULL) {
+		ret = ntdb_delete(self->ctx, tkey);
+	} else {
+		if (!PyString_Check(value)) {
+			PyErr_SetString(PyExc_TypeError, "Expected string as value");
+			return -1;
+		}
+
+		tval = PyString_AsNtdb_Data(value);
+
+		ret = ntdb_store(self->ctx, tkey, tval, NTDB_REPLACE);
+	}
+
+	if (ret != NTDB_SUCCESS) {
+		PyErr_SetTDBError(ret);
+		return -1;
+	}
+
+	return ret;
+}
+
+static PyMappingMethods ntdb_object_mapping = {
+	.mp_subscript = (binaryfunc)obj_getitem,
+	.mp_ass_subscript = (objobjargproc)obj_setitem,
+};
+static PyTypeObject PyNtdb = {
+	.tp_name = "ntdb.Ntdb",
+	.tp_basicsize = sizeof(PyNtdbObject),
+	.tp_methods = ntdb_object_methods,
+	.tp_getset = ntdb_object_getsetters,
+	.tp_new = py_ntdb_open,
+	.tp_doc = "A NTDB file",
+	.tp_repr = (reprfunc)ntdb_object_repr,
+	.tp_dealloc = (destructor)ntdb_object_dealloc,
+	.tp_as_mapping = &ntdb_object_mapping,
+	.tp_flags = Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE|Py_TPFLAGS_HAVE_ITER,
+	.tp_iter = (getiterfunc)ntdb_object_iter,
+};
+
+static PyMethodDef ntdb_methods[] = {
+	{ "open", (PyCFunction)py_ntdb_open, METH_VARARGS|METH_KEYWORDS, "open(name, hash_size=0, ntdb_flags=NTDB_DEFAULT, flags=O_RDWR, mode=0600)\n"
+		"Open a NTDB file." },
+	{ NULL }
+};
+
+void inittdb(void);
+void inittdb(void)
+{
+	PyObject *m;
+
+	if (PyType_Ready(&PyNtdb) < 0)
+		return;
+
+	if (PyType_Ready(&PyNtdbIterator) < 0)
+		return;
+
+	m = Py_InitModule3("ntdb", ntdb_methods, "NTDB is a simple key-value database similar to GDBM that supports multiple writers.");
+	if (m == NULL)
+		return;
+
+	PyModule_AddObject(m, "REPLACE", PyInt_FromLong(NTDB_REPLACE));
+	PyModule_AddObject(m, "INSERT", PyInt_FromLong(NTDB_INSERT));
+	PyModule_AddObject(m, "MODIFY", PyInt_FromLong(NTDB_MODIFY));
+
+	PyModule_AddObject(m, "DEFAULT", PyInt_FromLong(NTDB_DEFAULT));
+	PyModule_AddObject(m, "INTERNAL", PyInt_FromLong(NTDB_INTERNAL));
+	PyModule_AddObject(m, "NOLOCK", PyInt_FromLong(NTDB_NOLOCK));
+	PyModule_AddObject(m, "NOMMAP", PyInt_FromLong(NTDB_NOMMAP));
+	PyModule_AddObject(m, "CONVERT", PyInt_FromLong(NTDB_CONVERT));
+	PyModule_AddObject(m, "NOSYNC", PyInt_FromLong(NTDB_NOSYNC));
+	PyModule_AddObject(m, "SEQNUM", PyInt_FromLong(NTDB_SEQNUM));
+	PyModule_AddObject(m, "ALLOW_NESTING", PyInt_FromLong(NTDB_ALLOW_NESTING));
+
+	PyModule_AddObject(m, "__docformat__", PyString_FromString("restructuredText"));
+
+	PyModule_AddObject(m, "__version__", PyString_FromString(PACKAGE_VERSION));
+
+	Py_INCREF(&PyNtdb);
+	PyModule_AddObject(m, "Ntdb", (PyObject *)&PyNtdb);
+
+	Py_INCREF(&PyNtdbIterator);
+}

lib/ntdb/summary.c

@@ -47,25 +47,25 @@
 #define HISTO_WIDTH 70
 #define HISTO_HEIGHT 20
 
-static tdb_off_t count_hash(struct tdb_context *tdb,
-			    tdb_off_t hash_off, unsigned bits)
+static ntdb_off_t count_hash(struct ntdb_context *ntdb,
+			    ntdb_off_t hash_off, unsigned bits)
 {
-	const tdb_off_t *h;
-	tdb_off_t count = 0;
+	const ntdb_off_t *h;
+	ntdb_off_t count = 0;
 	unsigned int i;
 
-	h = tdb_access_read(tdb, hash_off, sizeof(*h) << bits, true);
-	if (TDB_PTR_IS_ERR(h)) {
-		return TDB_ERR_TO_OFF(TDB_PTR_ERR(h));
+	h = ntdb_access_read(ntdb, hash_off, sizeof(*h) << bits, true);
+	if (NTDB_PTR_IS_ERR(h)) {
+		return NTDB_ERR_TO_OFF(NTDB_PTR_ERR(h));
 	}
 	for (i = 0; i < (1 << bits); i++)
 		count += (h[i] != 0);
 
-	tdb_access_release(tdb, h);
+	ntdb_access_release(ntdb, h);
 	return count;
 }
 
-static enum TDB_ERROR summarize(struct tdb_context *tdb,
+static enum NTDB_ERROR summarize(struct ntdb_context *ntdb,
 				struct tally *hashes,
 				struct tally *ftables,
 				struct tally *fr,
@@ -76,39 +76,39 @@ static enum TDB_ERROR summarize(struct tdb_context *tdb,
 				struct tally *chains,
 				size_t *num_caps)
 {
-	tdb_off_t off;
-	tdb_len_t len;
-	tdb_len_t unc = 0;
+	ntdb_off_t off;
+	ntdb_len_t len;
+	ntdb_len_t unc = 0;
 
-	for (off = sizeof(struct tdb_header);
-	     off < tdb->file->map_size;
+	for (off = sizeof(struct ntdb_header);
+	     off < ntdb->file->map_size;
 	     off += len) {
 		const union {
-			struct tdb_used_record u;
-			struct tdb_free_record f;
-			struct tdb_recovery_record r;
+			struct ntdb_used_record u;
+			struct ntdb_free_record f;
+			struct ntdb_recovery_record r;
 		} *p;
 		/* We might not be able to get the whole thing. */
-		p = tdb_access_read(tdb, off, sizeof(p->f), true);
-		if (TDB_PTR_IS_ERR(p)) {
-			return TDB_PTR_ERR(p);
+		p = ntdb_access_read(ntdb, off, sizeof(p->f), true);
+		if (NTDB_PTR_IS_ERR(p)) {
+			return NTDB_PTR_ERR(p);
 		}
-		if (frec_magic(&p->f) != TDB_FREE_MAGIC) {
+		if (frec_magic(&p->f) != NTDB_FREE_MAGIC) {
 			if (unc > 1) {
 				tally_add(uncoal, unc);
 				unc = 0;
 			}
 		}
 
-		if (p->r.magic == TDB_RECOVERY_INVALID_MAGIC
-		    || p->r.magic == TDB_RECOVERY_MAGIC) {
+		if (p->r.magic == NTDB_RECOVERY_INVALID_MAGIC
+		    || p->r.magic == NTDB_RECOVERY_MAGIC) {
 			len = sizeof(p->r) + p->r.max_len;
-		} else if (frec_magic(&p->f) == TDB_FREE_MAGIC) {
+		} else if (frec_magic(&p->f) == NTDB_FREE_MAGIC) {
 			len = frec_len(&p->f);
 			tally_add(fr, len);
 			len += sizeof(p->u);
 			unc++;
-		} else if (rec_magic(&p->u) == TDB_USED_MAGIC) {
+		} else if (rec_magic(&p->u) == NTDB_USED_MAGIC) {
 			len = sizeof(p->u)
 				+ rec_key_length(&p->u)
 				+ rec_data_length(&p->u)
@@ -117,105 +117,105 @@ static enum TDB_ERROR summarize(struct tdb_context *tdb,
 			tally_add(keys, rec_key_length(&p->u));
 			tally_add(data, rec_data_length(&p->u));
 			tally_add(extra, rec_extra_padding(&p->u));
-		} else if (rec_magic(&p->u) == TDB_HTABLE_MAGIC) {
-			tdb_off_t count = count_hash(tdb,
+		} else if (rec_magic(&p->u) == NTDB_HTABLE_MAGIC) {
+			ntdb_off_t count = count_hash(ntdb,
 						     off + sizeof(p->u),
-						     TDB_SUBLEVEL_HASH_BITS);
-			if (TDB_OFF_IS_ERR(count)) {
-				return TDB_OFF_TO_ERR(count);
+						     NTDB_SUBLEVEL_HASH_BITS);
+			if (NTDB_OFF_IS_ERR(count)) {
+				return NTDB_OFF_TO_ERR(count);
 			}
 			tally_add(hashes, count);
 			tally_add(extra, rec_extra_padding(&p->u));
 			len = sizeof(p->u)
 				+ rec_data_length(&p->u)
 				+ rec_extra_padding(&p->u);
-		} else if (rec_magic(&p->u) == TDB_FTABLE_MAGIC) {
+		} else if (rec_magic(&p->u) == NTDB_FTABLE_MAGIC) {
 			len = sizeof(p->u)
 				+ rec_data_length(&p->u)
 				+ rec_extra_padding(&p->u);
 			tally_add(ftables, rec_data_length(&p->u));
 			tally_add(extra, rec_extra_padding(&p->u));
-		} else if (rec_magic(&p->u) == TDB_CHAIN_MAGIC) {
+		} else if (rec_magic(&p->u) == NTDB_CHAIN_MAGIC) {
 			len = sizeof(p->u)
 				+ rec_data_length(&p->u)
 				+ rec_extra_padding(&p->u);
 			tally_add(chains, 1);
 			tally_add(extra, rec_extra_padding(&p->u));
-		} else if (rec_magic(&p->u) == TDB_CAP_MAGIC) {
+		} else if (rec_magic(&p->u) == NTDB_CAP_MAGIC) {
 			len = sizeof(p->u)
 				+ rec_data_length(&p->u)
 				+ rec_extra_padding(&p->u);
 			(*num_caps)++;
 		} else {
-			len = dead_space(tdb, off);
-			if (TDB_OFF_IS_ERR(len)) {
-				return TDB_OFF_TO_ERR(len);
+			len = dead_space(ntdb, off);
+			if (NTDB_OFF_IS_ERR(len)) {
+				return NTDB_OFF_TO_ERR(len);
 			}
 		}
-		tdb_access_release(tdb, p);
+		ntdb_access_release(ntdb, p);
 	}
 	if (unc)
 		tally_add(uncoal, unc);
-	return TDB_SUCCESS;
+	return NTDB_SUCCESS;
 }
 
-static void add_capabilities(struct tdb_context *tdb, char *summary)
+static void add_capabilities(struct ntdb_context *ntdb, char *summary)
 {
-	tdb_off_t off, next;
-	const struct tdb_capability *cap;
+	ntdb_off_t off, next;
+	const struct ntdb_capability *cap;
 	size_t count = 0;
 
 	/* Append to summary. */
 	summary += strlen(summary);
 
-	off = tdb_read_off(tdb, offsetof(struct tdb_header, capabilities));
-	if (TDB_OFF_IS_ERR(off))
+	off = ntdb_read_off(ntdb, offsetof(struct ntdb_header, capabilities));
+	if (NTDB_OFF_IS_ERR(off))
 		return;
 
 	/* Walk capability list. */
 	for (; off; off = next) {
-		cap = tdb_access_read(tdb, off, sizeof(*cap), true);
-		if (TDB_PTR_IS_ERR(cap)) {
+		cap = ntdb_access_read(ntdb, off, sizeof(*cap), true);
+		if (NTDB_PTR_IS_ERR(cap)) {
 			break;
 		}
 		count++;
 		sprintf(summary, CAPABILITY_FORMAT,
-			cap->type & TDB_CAP_TYPE_MASK,
+			cap->type & NTDB_CAP_TYPE_MASK,
 			/* Noopen?  How did we get here? */
-			(cap->type & TDB_CAP_NOOPEN) ? " (unopenable)"
-			: ((cap->type & TDB_CAP_NOWRITE)
-			   && (cap->type & TDB_CAP_NOCHECK)) ? " (uncheckable,read-only)"
-			: (cap->type & TDB_CAP_NOWRITE) ? " (read-only)"
-			: (cap->type & TDB_CAP_NOCHECK) ? " (uncheckable)"
+			(cap->type & NTDB_CAP_NOOPEN) ? " (unopenable)"
+			: ((cap->type & NTDB_CAP_NOWRITE)
+			   && (cap->type & NTDB_CAP_NOCHECK)) ? " (uncheckable,read-only)"
+			: (cap->type & NTDB_CAP_NOWRITE) ? " (read-only)"
+			: (cap->type & NTDB_CAP_NOCHECK) ? " (uncheckable)"
 			: "");
 		summary += strlen(summary);
 		next = cap->next;
-		tdb_access_release(tdb, cap);
+		ntdb_access_release(ntdb, cap);
 	}
 }
 
-_PUBLIC_ enum TDB_ERROR tdb_summary(struct tdb_context *tdb,
-			   enum tdb_summary_flags flags,
+_PUBLIC_ enum NTDB_ERROR ntdb_summary(struct ntdb_context *ntdb,
+			   enum ntdb_summary_flags flags,
 			   char **summary)
 {
-	tdb_len_t len;
+	ntdb_len_t len;
 	size_t num_caps = 0;
 	struct tally *ftables, *hashes, *freet, *keys, *data, *extra, *uncoal,
 		*chains;
 	char *hashesg, *freeg, *keysg, *datag, *extrag, *uncoalg;
-	enum TDB_ERROR ecode;
+	enum NTDB_ERROR ecode;
 
 	hashesg = freeg = keysg = datag = extrag = uncoalg = NULL;
 
-	ecode = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
-	if (ecode != TDB_SUCCESS) {
-		return tdb->last_error = ecode;
+	ecode = ntdb_allrecord_lock(ntdb, F_RDLCK, NTDB_LOCK_WAIT, false);
+	if (ecode != NTDB_SUCCESS) {
+		return ntdb->last_error = ecode;
 	}
 
-	ecode = tdb_lock_expand(tdb, F_RDLCK);
-	if (ecode != TDB_SUCCESS) {
-		tdb_allrecord_unlock(tdb, F_RDLCK);
-		return tdb->last_error = ecode;
+	ecode = ntdb_lock_expand(ntdb, F_RDLCK);
+	if (ecode != NTDB_SUCCESS) {
+		ntdb_allrecord_unlock(ntdb, F_RDLCK);
+		return ntdb->last_error = ecode;
 	}
 
 	/* Start stats off empty. */
@@ -229,19 +229,19 @@ _PUBLIC_ enum TDB_ERROR tdb_summary(struct tdb_context *tdb,
 	chains = tally_new(HISTO_HEIGHT);
 	if (!ftables || !hashes || !freet || !keys || !data || !extra
 	    || !uncoal || !chains) {
-		ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
-				   "tdb_summary: failed to allocate"
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+				   "ntdb_summary: failed to allocate"
 				   " tally structures");
 		goto unlock;
 	}
 
-	ecode = summarize(tdb, hashes, ftables, freet, keys, data, extra,
+	ecode = summarize(ntdb, hashes, ftables, freet, keys, data, extra,
 			  uncoal, chains, &num_caps);
-	if (ecode != TDB_SUCCESS) {
+	if (ecode != NTDB_SUCCESS) {
 		goto unlock;
 	}
 
-	if (flags & TDB_SUMMARY_HISTOGRAMS) {
+	if (flags & NTDB_SUMMARY_HISTOGRAMS) {
 		hashesg = tally_histogram(hashes, HISTO_WIDTH, HISTO_HEIGHT);
 		freeg = tally_histogram(freet, HISTO_WIDTH, HISTO_HEIGHT);
 		keysg = tally_histogram(keys, HISTO_WIDTH, HISTO_HEIGHT);
@@ -263,13 +263,13 @@ _PUBLIC_ enum TDB_ERROR tdb_summary(struct tdb_context *tdb,
 
 	*summary = malloc(len);
 	if (!*summary) {
-		ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
-				   "tdb_summary: failed to allocate string");
+		ecode = ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
+				   "ntdb_summary: failed to allocate string");
 		goto unlock;
 	}
 
 	sprintf(*summary, SUMMARY_FORMAT,
-		(size_t)tdb->file->map_size,
+		(size_t)ntdb->file->map_size,
 		tally_total(keys, NULL) + tally_total(data, NULL),
 		tally_num(keys),
 		tally_min(keys), tally_mean(keys), tally_max(keys),
@@ -284,29 +284,29 @@ _PUBLIC_ enum TDB_ERROR tdb_summary(struct tdb_context *tdb,
 		tally_total(uncoal, NULL),
 		tally_min(uncoal), tally_mean(uncoal), tally_max(uncoal),
 		uncoalg ? uncoalg : "",
-		(unsigned)count_hash(tdb, offsetof(struct tdb_header,
+		(unsigned)count_hash(ntdb, offsetof(struct ntdb_header,
 						   hashtable),
-				     TDB_TOPLEVEL_HASH_BITS),
-		1 << TDB_TOPLEVEL_HASH_BITS,
+				     NTDB_TOPLEVEL_HASH_BITS),
+		1 << NTDB_TOPLEVEL_HASH_BITS,
 		tally_num(chains),
 		tally_num(hashes),
 		tally_min(hashes), tally_mean(hashes), tally_max(hashes),
 		hashesg ? hashesg : "",
-		tally_total(keys, NULL) * 100.0 / tdb->file->map_size,
-		tally_total(data, NULL) * 100.0 / tdb->file->map_size,
-		tally_total(extra, NULL) * 100.0 / tdb->file->map_size,
-		tally_total(freet, NULL) * 100.0 / tdb->file->map_size,
+		tally_total(keys, NULL) * 100.0 / ntdb->file->map_size,
+		tally_total(data, NULL) * 100.0 / ntdb->file->map_size,
+		tally_total(extra, NULL) * 100.0 / ntdb->file->map_size,
+		tally_total(freet, NULL) * 100.0 / ntdb->file->map_size,
 		(tally_num(keys) + tally_num(freet) + tally_num(hashes))
-		* sizeof(struct tdb_used_record) * 100.0 / tdb->file->map_size,
-		tally_num(ftables) * sizeof(struct tdb_freetable)
-		* 100.0 / tdb->file->map_size,
+		* sizeof(struct ntdb_used_record) * 100.0 / ntdb->file->map_size,
+		tally_num(ftables) * sizeof(struct ntdb_freetable)
+		* 100.0 / ntdb->file->map_size,
 		(tally_num(hashes)
-		 * (sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS)
-		 + (sizeof(tdb_off_t) << TDB_TOPLEVEL_HASH_BITS)
-		 + sizeof(struct tdb_chain) * tally_num(chains))
-		* 100.0 / tdb->file->map_size);
+		 * (sizeof(ntdb_off_t) << NTDB_SUBLEVEL_HASH_BITS)
+		 + (sizeof(ntdb_off_t) << NTDB_TOPLEVEL_HASH_BITS)
+		 + sizeof(struct ntdb_chain) * tally_num(chains))
+		* 100.0 / ntdb->file->map_size);
 
-	add_capabilities(tdb, *summary);
+	add_capabilities(ntdb, *summary);
 
 unlock:
 	free(hashesg);
@@ -324,7 +324,7 @@ unlock:
 	free(ftables);
 	free(chains);
 
-	tdb_allrecord_unlock(tdb, F_RDLCK);
-	tdb_unlock_expand(tdb, F_RDLCK);
-	return tdb->last_error = ecode;
+	ntdb_allrecord_unlock(ntdb, F_RDLCK);
+	ntdb_unlock_expand(ntdb, F_RDLCK);
+	return ntdb->last_error = ecode;
 }

lib/ntdb/test/api-12-store.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <ccan/hash/hash.h>
 #include <sys/types.h>
@@ -18,38 +18,38 @@ static uint64_t fixedhash(const void *key, size_t len, uint64_t seed, void *p)
 int main(int argc, char *argv[])
 {
 	unsigned int i, j;
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 	uint64_t seed = 16014841315512641303ULL;
-	union tdb_attribute fixed_hattr
-		= { .hash = { .base = { TDB_ATTRIBUTE_HASH },
+	union ntdb_attribute fixed_hattr
+		= { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
 			      .fn = fixedhash,
 			      .data = &seed } };
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = { (unsigned char *)&j, sizeof(j) };
-	struct tdb_data data = { (unsigned char *)&j, sizeof(j) };
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = { (unsigned char *)&j, sizeof(j) };
+	NTDB_DATA data = { (unsigned char *)&j, sizeof(j) };
 
 	fixed_hattr.base.next = &tap_log_attr;
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * (1 + 500 * 3) + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-12-store.tdb", flags[i],
+		ntdb = ntdb_open("run-12-store.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &fixed_hattr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		/* We seemed to lose some keys.
 		 * Insert and check they're in there! */
 		for (j = 0; j < 500; j++) {
-			struct tdb_data d = { NULL, 0 }; /* Bogus GCC warning */
-			ok1(tdb_store(tdb, key, data, TDB_REPLACE) == 0);
-			ok1(tdb_fetch(tdb, key, &d) == TDB_SUCCESS);
-			ok1(tdb_deq(d, data));
+			NTDB_DATA d = { NULL, 0 }; /* Bogus GCC warning */
+			ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == 0);
+			ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+			ok1(ntdb_deq(d, data));
 			free(d.dptr);
 		}
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/api-13-delete.c

@@ -0,0 +1,205 @@
+#include "private.h" // For NTDB_TOPLEVEL_HASH_BITS
+#include <ccan/hash/hash.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include "ntdb.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+/* We rig the hash so adjacent-numbered records always clash. */
+static uint64_t clash(const void *key, size_t len, uint64_t seed, void *priv)
+{
+	return ((uint64_t)*(const unsigned int *)key)
+		<< (64 - NTDB_TOPLEVEL_HASH_BITS - 1);
+}
+
+/* We use the same seed which we saw a failure on. */
+static uint64_t fixedhash(const void *key, size_t len, uint64_t seed, void *p)
+{
+	return hash64_stable((const unsigned char *)key, len,
+			     *(uint64_t *)p);
+}
+
+static bool store_records(struct ntdb_context *ntdb)
+{
+	int i;
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA d, data = { (unsigned char *)&i, sizeof(i) };
+
+	for (i = 0; i < 1000; i++) {
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
+			return false;
+		ntdb_fetch(ntdb, key, &d);
+		if (!ntdb_deq(d, data))
+			return false;
+		free(d.dptr);
+	}
+	return true;
+}
+
+static void test_val(struct ntdb_context *ntdb, uint64_t val)
+{
+	uint64_t v;
+	NTDB_DATA key = { (unsigned char *)&v, sizeof(v) };
+	NTDB_DATA d, data = { (unsigned char *)&v, sizeof(v) };
+
+	/* Insert an entry, then delete it. */
+	v = val;
+	/* Delete should fail. */
+	ok1(ntdb_delete(ntdb, key) == NTDB_ERR_NOEXIST);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Insert should succeed. */
+	ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Delete should succeed. */
+	ok1(ntdb_delete(ntdb, key) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Re-add it, then add collision. */
+	ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+	v = val + 1;
+	ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Can find both? */
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+	v = val;
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+
+	/* Delete second one. */
+	v = val + 1;
+	ok1(ntdb_delete(ntdb, key) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Re-add */
+	ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Now, try deleting first one. */
+	v = val;
+	ok1(ntdb_delete(ntdb, key) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Can still find second? */
+	v = val + 1;
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+
+	/* Now, this will be ideally placed. */
+	v = val + 2;
+	ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* This will collide with both. */
+	v = val;
+	ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+
+	/* We can still find them all, right? */
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+	v = val + 1;
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+	v = val + 2;
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+
+	/* And if we delete val + 1, that val + 2 should not move! */
+	v = val + 1;
+	ok1(ntdb_delete(ntdb, key) == 0);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	v = val;
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+	v = val + 2;
+	ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+	ok1(d.dsize == data.dsize);
+	free(d.dptr);
+
+	/* Delete those two, so we are empty. */
+	ok1(ntdb_delete(ntdb, key) == 0);
+	v = val;
+	ok1(ntdb_delete(ntdb, key) == 0);
+
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i, j;
+	struct ntdb_context *ntdb;
+	uint64_t seed = 16014841315512641303ULL;
+	union ntdb_attribute clash_hattr
+		= { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
+			      .fn = clash } };
+	union ntdb_attribute fixed_hattr
+		= { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
+			      .fn = fixedhash,
+			      .data = &seed } };
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	/* These two values gave trouble before. */
+	int vals[] = { 755, 837 };
+
+	clash_hattr.base.next = &tap_log_attr;
+	fixed_hattr.base.next = &tap_log_attr;
+
+	plan_tests(sizeof(flags) / sizeof(flags[0])
+		   * (39 * 3 + 5 + sizeof(vals)/sizeof(vals[0])*2) + 1);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("run-13-delete.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &clash_hattr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		/* Check start of hash table. */
+		test_val(ntdb, 0);
+
+		/* Check end of hash table. */
+		test_val(ntdb, -1ULL);
+
+		/* Check mixed bitpattern. */
+		test_val(ntdb, 0x123456789ABCDEF0ULL);
+
+		ok1(!ntdb->file || (ntdb->file->allrecord_lock.count == 0
+				   && ntdb->file->num_lockrecs == 0));
+		ntdb_close(ntdb);
+
+		/* Deleting these entries in the db gave problems. */
+		ntdb = ntdb_open("run-13-delete.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &fixed_hattr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		ok1(store_records(ntdb));
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		for (j = 0; j < sizeof(vals)/sizeof(vals[0]); j++) {
+			NTDB_DATA key;
+
+			key.dptr = (unsigned char *)&vals[j];
+			key.dsize = sizeof(vals[j]);
+			ok1(ntdb_delete(ntdb, key) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		}
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/api-14-exists.c

@@ -0,0 +1,54 @@
+#include "config.h"
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include "logging.h"
+
+static bool test_records(struct ntdb_context *ntdb)
+{
+	int i;
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
+
+	for (i = 0; i < 1000; i++) {
+		if (ntdb_exists(ntdb, key))
+			return false;
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
+			return false;
+		if (!ntdb_exists(ntdb, key))
+			return false;
+	}
+
+	for (i = 0; i < 1000; i++) {
+		if (!ntdb_exists(ntdb, key))
+			return false;
+		if (ntdb_delete(ntdb, key) != 0)
+			return false;
+		if (ntdb_exists(ntdb, key))
+			return false;
+	}
+	return true;
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 2 + 1);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("run-14-exists.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		if (ok1(ntdb))
+			ok1(test_records(ntdb));
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/api-16-wipe_all.c

@@ -0,0 +1,46 @@
+#include "config.h"
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include "logging.h"
+
+static bool add_records(struct ntdb_context *ntdb)
+{
+	int i;
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
+
+	for (i = 0; i < 1000; i++) {
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
+			return false;
+	}
+	return true;
+}
+
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 4 + 1);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("run-16-wipe_all.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		if (ok1(ntdb)) {
+			NTDB_DATA key;
+			ok1(add_records(ntdb));
+			ok1(ntdb_wipe_all(ntdb) == NTDB_SUCCESS);
+			ok1(ntdb_firstkey(ntdb, &key) == NTDB_ERR_NOEXIST);
+			ntdb_close(ntdb);
+		}
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/api-21-parse_record.c

@@ -0,0 +1,67 @@
+#include "config.h"
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include "logging.h"
+
+static enum NTDB_ERROR parse(NTDB_DATA key, NTDB_DATA data, NTDB_DATA *expected)
+{
+	if (!ntdb_deq(data, *expected))
+		return NTDB_ERR_EINVAL;
+	return NTDB_SUCCESS;
+}
+
+static enum NTDB_ERROR parse_err(NTDB_DATA key, NTDB_DATA data, void *unused)
+{
+	return 100;
+}
+
+static bool test_records(struct ntdb_context *ntdb)
+{
+	int i;
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
+
+	for (i = 0; i < 1000; i++) {
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
+			return false;
+	}
+
+	for (i = 0; i < 1000; i++) {
+		if (ntdb_parse_record(ntdb, key, parse, &data) != NTDB_SUCCESS)
+			return false;
+	}
+
+	if (ntdb_parse_record(ntdb, key, parse, &data) != NTDB_ERR_NOEXIST)
+		return false;
+
+	/* Test error return from parse function. */
+	i = 0;
+	if (ntdb_parse_record(ntdb, key, parse_err, NULL) != 100)
+		return false;
+
+	return true;
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 2 + 1);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("api-21-parse_record.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		if (ok1(ntdb))
+			ok1(test_records(ntdb));
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/api-55-transaction.c

@@ -0,0 +1,73 @@
+#include "private.h" // struct ntdb_context
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	unsigned char *buffer;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data;
+
+	buffer = malloc(1000);
+	for (i = 0; i < 1000; i++)
+		buffer[i] = i;
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 20 + 1);
+
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("run-55-transaction.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		ok1(ntdb_transaction_start(ntdb) == 0);
+		data.dptr = buffer;
+		data.dsize = 1000;
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+		ok1(ntdb_fetch(ntdb, key, &data) == NTDB_SUCCESS);
+		ok1(data.dsize == 1000);
+		ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
+		free(data.dptr);
+
+		/* Cancelling a transaction means no store */
+		ntdb_transaction_cancel(ntdb);
+		ok1(ntdb->file->allrecord_lock.count == 0
+		    && ntdb->file->num_lockrecs == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ok1(ntdb_fetch(ntdb, key, &data) == NTDB_ERR_NOEXIST);
+
+		/* Commit the transaction. */
+		ok1(ntdb_transaction_start(ntdb) == 0);
+		data.dptr = buffer;
+		data.dsize = 1000;
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+		ok1(ntdb_fetch(ntdb, key, &data) == NTDB_SUCCESS);
+		ok1(data.dsize == 1000);
+		ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
+		free(data.dptr);
+		ok1(ntdb_transaction_commit(ntdb) == 0);
+		ok1(ntdb->file->allrecord_lock.count == 0
+		    && ntdb->file->num_lockrecs == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ok1(ntdb_fetch(ntdb, key, &data) == NTDB_SUCCESS);
+		ok1(data.dsize == 1000);
+		ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
+		free(data.dptr);
+
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	free(buffer);
+	return exit_status();
+}

lib/ntdb/test/api-80-tdb_fd.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -9,23 +9,23 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 3);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("api-80-tdb_fd.tdb", flags[i],
+		ntdb = ntdb_open("api-80-ntdb_fd.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			continue;
 
-		if (flags[i] & TDB_INTERNAL)
-			ok1(tdb_fd(tdb) == -1);
+		if (flags[i] & NTDB_INTERNAL)
+			ok1(ntdb_fd(ntdb) == -1);
 		else
-			ok1(tdb_fd(tdb) > 2);
-		tdb_close(tdb);
+			ok1(ntdb_fd(ntdb) > 2);
+		ntdb_close(ntdb);
 		ok1(tap_log_messages == 0);
 	}
 	return exit_status();

lib/ntdb/test/api-81-seqnum.c

@@ -0,0 +1,69 @@
+#include "config.h"
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i, seq;
+	struct ntdb_context *ntdb;
+	NTDB_DATA d = { NULL, 0 }; /* Bogus GCC warning */
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 15 + 4 * 13);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("api-81-seqnum.ntdb", flags[i]|NTDB_SEQNUM,
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		if (!ok1(ntdb))
+			continue;
+
+		seq = 0;
+		ok1(ntdb_get_seqnum(ntdb) == seq);
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+		ok1(ntdb_get_seqnum(ntdb) == ++seq);
+		/* Fetch doesn't change seqnum */
+		if (ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS))
+			free(d.dptr);
+		ok1(ntdb_get_seqnum(ntdb) == seq);
+		ok1(ntdb_append(ntdb, key, data) == NTDB_SUCCESS);
+		ok1(ntdb_get_seqnum(ntdb) == ++seq);
+
+		ok1(ntdb_delete(ntdb, key) == NTDB_SUCCESS);
+		ok1(ntdb_get_seqnum(ntdb) == ++seq);
+		/* Empty append works */
+		ok1(ntdb_append(ntdb, key, data) == NTDB_SUCCESS);
+		ok1(ntdb_get_seqnum(ntdb) == ++seq);
+
+		ok1(ntdb_wipe_all(ntdb) == NTDB_SUCCESS);
+		ok1(ntdb_get_seqnum(ntdb) == ++seq);
+
+		if (!(flags[i] & NTDB_INTERNAL)) {
+			ok1(ntdb_transaction_start(ntdb) == NTDB_SUCCESS);
+			ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+			ok1(ntdb_get_seqnum(ntdb) == ++seq);
+			ok1(ntdb_append(ntdb, key, data) == NTDB_SUCCESS);
+			ok1(ntdb_get_seqnum(ntdb) == ++seq);
+			ok1(ntdb_delete(ntdb, key) == NTDB_SUCCESS);
+			ok1(ntdb_get_seqnum(ntdb) == ++seq);
+			ok1(ntdb_transaction_commit(ntdb) == NTDB_SUCCESS);
+			ok1(ntdb_get_seqnum(ntdb) == seq);
+
+			ok1(ntdb_transaction_start(ntdb) == NTDB_SUCCESS);
+			ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+			ok1(ntdb_get_seqnum(ntdb) == seq + 1);
+			ntdb_transaction_cancel(ntdb);
+			ok1(ntdb_get_seqnum(ntdb) == seq);
+		}
+		ntdb_close(ntdb);
+		ok1(tap_log_messages == 0);
+	}
+	return exit_status();
+}

lib/ntdb/test/api-82-lockattr.c

@@ -1,5 +1,5 @@
-#include "private.h" // for tdb_fcntl_unlock
-#include "tdb2.h"
+#include "private.h" // for ntdb_fcntl_unlock
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -35,7 +35,7 @@ static int mylock(int fd, int rw, off_t off, off_t len, bool waitflag,
 }
 
 static int trav_err;
-static int trav(struct tdb_context *tdb, TDB_DATA k, TDB_DATA d, int *terr)
+static int trav(struct ntdb_context *ntdb, NTDB_DATA k, NTDB_DATA d, int *terr)
 {
 	*terr = trav_err;
 	return 0;
@@ -44,193 +44,193 @@ static int trav(struct tdb_context *tdb, TDB_DATA k, TDB_DATA d, int *terr)
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
-	union tdb_attribute lock_attr;
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data = tdb_mkdata("data", 4);
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+	union ntdb_attribute lock_attr;
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
 	int lock_err;
 
-	lock_attr.base.attr = TDB_ATTRIBUTE_FLOCK;
+	lock_attr.base.attr = NTDB_ATTRIBUTE_FLOCK;
 	lock_attr.base.next = &tap_log_attr;
 	lock_attr.flock.lock = mylock;
-	lock_attr.flock.unlock = tdb_fcntl_unlock;
+	lock_attr.flock.unlock = ntdb_fcntl_unlock;
 	lock_attr.flock.data = &lock_err;
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 80);
 
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		struct tdb_data d;
+		NTDB_DATA d;
 
 		/* Nonblocking open; expect no error message. */
 		lock_err = EAGAIN;
-		tdb = tdb_open("run-82-lockattr.tdb", flags[i],
+		ntdb = ntdb_open("run-82-lockattr.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &lock_attr);
 		ok(errno == lock_err, "Errno is %u", errno);
-		ok1(!tdb);
+		ok1(!ntdb);
 		ok1(tap_log_messages == 0);
 
 		lock_err = EINTR;
-		tdb = tdb_open("run-82-lockattr.tdb", flags[i],
+		ntdb = ntdb_open("run-82-lockattr.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &lock_attr);
 		ok(errno == lock_err, "Errno is %u", errno);
-		ok1(!tdb);
+		ok1(!ntdb);
 		ok1(tap_log_messages == 0);
 
 		/* Forced fail open. */
 		lock_err = ENOMEM;
-		tdb = tdb_open("run-82-lockattr.tdb", flags[i],
+		ntdb = ntdb_open("run-82-lockattr.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &lock_attr);
 		ok1(errno == lock_err);
-		ok1(!tdb);
+		ok1(!ntdb);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		lock_err = 0;
-		tdb = tdb_open("run-82-lockattr.tdb", flags[i],
+		ntdb = ntdb_open("run-82-lockattr.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &lock_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			continue;
 		ok1(tap_log_messages == 0);
 
 		/* Nonblocking store. */
 		lock_err = EAGAIN;
-		ok1(tdb_store(tdb, key, data, TDB_REPLACE) == TDB_ERR_LOCK);
+		ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_store(tdb, key, data, TDB_REPLACE) == TDB_ERR_LOCK);
+		ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_store(tdb, key, data, TDB_REPLACE) == TDB_ERR_LOCK);
+		ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		/* Nonblocking fetch. */
 		lock_err = EAGAIN;
-		ok1(!tdb_exists(tdb, key));
+		ok1(!ntdb_exists(ntdb, key));
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(!tdb_exists(tdb, key));
+		ok1(!ntdb_exists(ntdb, key));
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(!tdb_exists(tdb, key));
+		ok1(!ntdb_exists(ntdb, key));
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		lock_err = EAGAIN;
-		ok1(tdb_fetch(tdb, key, &d) == TDB_ERR_LOCK);
+		ok1(ntdb_fetch(ntdb, key, &d) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_fetch(tdb, key, &d) == TDB_ERR_LOCK);
+		ok1(ntdb_fetch(ntdb, key, &d) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_fetch(tdb, key, &d) == TDB_ERR_LOCK);
+		ok1(ntdb_fetch(ntdb, key, &d) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		/* Nonblocking delete. */
 		lock_err = EAGAIN;
-		ok1(tdb_delete(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_delete(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_delete(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_delete(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_delete(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_delete(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		/* Nonblocking locks. */
 		lock_err = EAGAIN;
-		ok1(tdb_chainlock(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_chainlock(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_chainlock(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		lock_err = EAGAIN;
-		ok1(tdb_chainlock_read(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_chainlock_read(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_chainlock_read(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_chainlock_read(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_chainlock_read(tdb, key) == TDB_ERR_LOCK);
+		ok1(ntdb_chainlock_read(ntdb, key) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		lock_err = EAGAIN;
-		ok1(tdb_lockall(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_lockall(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_lockall(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_lockall(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_lockall(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_lockall(ntdb) == NTDB_ERR_LOCK);
 		/* This actually does divide and conquer. */
 		ok1(tap_log_messages > 0);
 		tap_log_messages = 0;
 
 		lock_err = EAGAIN;
-		ok1(tdb_lockall_read(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_lockall_read(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_lockall_read(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_lockall_read(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_lockall_read(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_lockall_read(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages > 0);
 		tap_log_messages = 0;
 
 		/* Nonblocking traverse; go nonblock partway through. */
 		lock_err = 0;
-		ok1(tdb_store(tdb, key, data, TDB_REPLACE) == 0);
+		ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == 0);
 		trav_err = EAGAIN;
-		ok1(tdb_traverse(tdb, trav, &lock_err) == TDB_ERR_LOCK);
+		ok1(ntdb_traverse(ntdb, trav, &lock_err) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		trav_err = EINTR;
 		lock_err = 0;
-		ok1(tdb_traverse(tdb, trav, &lock_err) == TDB_ERR_LOCK);
+		ok1(ntdb_traverse(ntdb, trav, &lock_err) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		trav_err = ENOMEM;
 		lock_err = 0;
-		ok1(tdb_traverse(tdb, trav, &lock_err) == TDB_ERR_LOCK);
+		ok1(ntdb_traverse(ntdb, trav, &lock_err) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		/* Nonblocking transactions. */
 		lock_err = EAGAIN;
-		ok1(tdb_transaction_start(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_transaction_start(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = EINTR;
-		ok1(tdb_transaction_start(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_transaction_start(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 		lock_err = ENOMEM;
-		ok1(tdb_transaction_start(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_transaction_start(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 1);
 		tap_log_messages = 0;
 
 		/* Nonblocking transaction prepare. */
 		lock_err = 0;
-		ok1(tdb_transaction_start(tdb) == 0);
-		ok1(tdb_delete(tdb, key) == 0);
+		ok1(ntdb_transaction_start(ntdb) == 0);
+		ok1(ntdb_delete(ntdb, key) == 0);
 
 		lock_err = EAGAIN;
-		ok1(tdb_transaction_prepare_commit(tdb) == TDB_ERR_LOCK);
+		ok1(ntdb_transaction_prepare_commit(ntdb) == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 
 		lock_err = 0;
-		ok1(tdb_transaction_prepare_commit(tdb) == 0);
-		ok1(tdb_transaction_commit(tdb) == 0);
+		ok1(ntdb_transaction_prepare_commit(ntdb) == 0);
+		ok1(ntdb_transaction_commit(ntdb) == 0);
 
 		/* And the transaction was committed, right? */
-		ok1(!tdb_exists(tdb, key));
-		tdb_close(tdb);
+		ok1(!ntdb_exists(ntdb, key));
+		ntdb_close(ntdb);
 		ok1(tap_log_messages == 0);
 	}
 	return exit_status();

lib/ntdb/test/api-83-openhook.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -11,14 +11,14 @@
 #include "external-agent.h"
 #include "logging.h"
 
-static enum TDB_ERROR clear_if_first(int fd, void *arg)
+static enum NTDB_ERROR clear_if_first(int fd, void *arg)
 {
 /* We hold a lock offset 4 always, so we can tell if anyone is holding it.
- * (This is compatible with tdb1's TDB_CLEAR_IF_FIRST flag).  */
+ * (This is compatible with tdb's TDB_CLEAR_IF_FIRST flag).  */
 	struct flock fl;
 
 	if (arg != clear_if_first)
-		return TDB_ERR_CORRUPT;
+		return NTDB_ERR_CORRUPT;
 
 	fl.l_type = F_WRLCK;
 	fl.l_whence = SEEK_SET;
@@ -29,27 +29,27 @@ static enum TDB_ERROR clear_if_first(int fd, void *arg)
 		/* We must be first ones to open it! */
 		diag("truncating file!");
 		if (ftruncate(fd, 0) != 0) {
-			return TDB_ERR_IO;
+			return NTDB_ERR_IO;
 		}
 	}
 	fl.l_type = F_RDLCK;
 	if (fcntl(fd, F_SETLKW, &fl) != 0) {
-		return TDB_ERR_IO;
+		return NTDB_ERR_IO;
 	}
-	return TDB_SUCCESS;
+	return NTDB_SUCCESS;
 }
 
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 	struct agent *agent;
-	union tdb_attribute cif;
-	struct tdb_data key = tdb_mkdata("key", 3);
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
+	union ntdb_attribute cif;
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
 
-	cif.openhook.base.attr = TDB_ATTRIBUTE_OPENHOOK;
+	cif.openhook.base.attr = NTDB_ATTRIBUTE_OPENHOOK;
 	cif.openhook.base.next = &tap_log_attr;
 	cif.openhook.fn = clear_if_first;
 	cif.openhook.data = clear_if_first;
@@ -58,33 +58,33 @@ int main(int argc, char *argv[])
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 13);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
 		/* Create it */
-		tdb = tdb_open("run-83-openhook.tdb", flags[i],
+		ntdb = ntdb_open("run-83-openhook.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, NULL);
-		ok1(tdb);
-		ok1(tdb_store(tdb, key, key, TDB_REPLACE) == 0);
-		tdb_close(tdb);
+		ok1(ntdb);
+		ok1(ntdb_store(ntdb, key, key, NTDB_REPLACE) == 0);
+		ntdb_close(ntdb);
 
 		/* Now, open with CIF, should clear it. */
-		tdb = tdb_open("run-83-openhook.tdb", flags[i],
+		ntdb = ntdb_open("run-83-openhook.ntdb", flags[i],
 			       O_RDWR, 0, &cif);
-		ok1(tdb);
-		ok1(!tdb_exists(tdb, key));
-		ok1(tdb_store(tdb, key, key, TDB_REPLACE) == 0);
+		ok1(ntdb);
+		ok1(!ntdb_exists(ntdb, key));
+		ok1(ntdb_store(ntdb, key, key, NTDB_REPLACE) == 0);
 
 		/* Agent should not clear it, since it's still open. */
 		ok1(external_agent_operation(agent, OPEN_WITH_HOOK,
-					     "run-83-openhook.tdb") == SUCCESS);
+					     "run-83-openhook.ntdb") == SUCCESS);
 		ok1(external_agent_operation(agent, FETCH, "key") == SUCCESS);
 		ok1(external_agent_operation(agent, CLOSE, "") == SUCCESS);
 
 		/* Still exists for us too. */
-		ok1(tdb_exists(tdb, key));
+		ok1(ntdb_exists(ntdb, key));
 
 		/* Close it, now agent should clear it. */
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 
 		ok1(external_agent_operation(agent, OPEN_WITH_HOOK,
-					     "run-83-openhook.tdb") == SUCCESS);
+					     "run-83-openhook.ntdb") == SUCCESS);
 		ok1(external_agent_operation(agent, FETCH, "key") == FAILED);
 		ok1(external_agent_operation(agent, CLOSE, "") == SUCCESS);
 

lib/ntdb/test/api-91-get-stats.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -11,27 +11,27 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 11);
 
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		union tdb_attribute *attr;
-		struct tdb_data key = tdb_mkdata("key", 3);
+		union ntdb_attribute *attr;
+		NTDB_DATA key = ntdb_mkdata("key", 3);
 
-		tdb = tdb_open("run-91-get-stats.tdb", flags[i],
+		ntdb = ntdb_open("run-91-get-stats.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		ok1(tdb_store(tdb, key, key, TDB_REPLACE) == 0);
+		ok1(ntdb);
+		ok1(ntdb_store(ntdb, key, key, NTDB_REPLACE) == 0);
 
 		/* Use malloc so valgrind will catch overruns. */
 		attr = malloc(sizeof *attr);
-		attr->stats.base.attr = TDB_ATTRIBUTE_STATS;
+		attr->stats.base.attr = NTDB_ATTRIBUTE_STATS;
 		attr->stats.size = sizeof(*attr);
 
-		ok1(tdb_get_attribute(tdb, attr) == 0);
+		ok1(ntdb_get_attribute(ntdb, attr) == 0);
 		ok1(attr->stats.size == sizeof(*attr));
 		ok1(attr->stats.allocs > 0);
 		ok1(attr->stats.expands > 0);
@@ -39,18 +39,18 @@ int main(int argc, char *argv[])
 		free(attr);
 
 		/* Try short one. */
-		attr = malloc(offsetof(struct tdb_attribute_stats, allocs)
+		attr = malloc(offsetof(struct ntdb_attribute_stats, allocs)
 			      + sizeof(attr->stats.allocs));
-		attr->stats.base.attr = TDB_ATTRIBUTE_STATS;
-		attr->stats.size = offsetof(struct tdb_attribute_stats, allocs)
+		attr->stats.base.attr = NTDB_ATTRIBUTE_STATS;
+		attr->stats.size = offsetof(struct ntdb_attribute_stats, allocs)
 			+ sizeof(attr->stats.allocs);
-		ok1(tdb_get_attribute(tdb, attr) == 0);
+		ok1(ntdb_get_attribute(ntdb, attr) == 0);
 		ok1(attr->stats.size == sizeof(*attr));
 		ok1(attr->stats.allocs > 0);
 		free(attr);
 		ok1(tap_log_messages == 0);
 
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 
 	}
 	return exit_status();

lib/ntdb/test/api-92-get-set-readonly.c

@@ -0,0 +1,105 @@
+#include "config.h"
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 48);
+
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		/* RW -> R0 */
+		ntdb = ntdb_open("run-92-get-set-readonly.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		ok1(!(ntdb_get_flags(ntdb) & NTDB_RDONLY));
+
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == NTDB_SUCCESS);
+
+		ntdb_add_flag(ntdb, NTDB_RDONLY);
+		ok1(ntdb_get_flags(ntdb) & NTDB_RDONLY);
+
+		/* Can't store, append, delete. */
+		ok1(ntdb_store(ntdb, key, data, NTDB_MODIFY) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 1);
+		ok1(ntdb_append(ntdb, key, data) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 2);
+		ok1(ntdb_delete(ntdb, key) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 3);
+
+		/* Can't start a transaction, or any write lock. */
+		ok1(ntdb_transaction_start(ntdb) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 4);
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 5);
+		ok1(ntdb_lockall(ntdb) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 6);
+		ok1(ntdb_wipe_all(ntdb) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 7);
+
+		/* Back to RW. */
+		ntdb_remove_flag(ntdb, NTDB_RDONLY);
+		ok1(!(ntdb_get_flags(ntdb) & NTDB_RDONLY));
+
+		ok1(ntdb_store(ntdb, key, data, NTDB_MODIFY) == NTDB_SUCCESS);
+		ok1(ntdb_append(ntdb, key, data) == NTDB_SUCCESS);
+		ok1(ntdb_delete(ntdb, key) == NTDB_SUCCESS);
+
+		ok1(ntdb_transaction_start(ntdb) == NTDB_SUCCESS);
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == NTDB_SUCCESS);
+		ok1(ntdb_transaction_commit(ntdb) == NTDB_SUCCESS);
+
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_SUCCESS);
+		ntdb_chainunlock(ntdb, key);
+		ok1(ntdb_lockall(ntdb) == NTDB_SUCCESS);
+		ntdb_unlockall(ntdb);
+		ok1(ntdb_wipe_all(ntdb) == NTDB_SUCCESS);
+		ok1(tap_log_messages == 7);
+
+		ntdb_close(ntdb);
+
+		/* R0 -> RW */
+		ntdb = ntdb_open("run-92-get-set-readonly.ntdb", flags[i],
+			       O_RDONLY, 0600, &tap_log_attr);
+		ok1(ntdb);
+		ok1(ntdb_get_flags(ntdb) & NTDB_RDONLY);
+
+		/* Can't store, append, delete. */
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 8);
+		ok1(ntdb_append(ntdb, key, data) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 9);
+		ok1(ntdb_delete(ntdb, key) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 10);
+
+		/* Can't start a transaction, or any write lock. */
+		ok1(ntdb_transaction_start(ntdb) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 11);
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 12);
+		ok1(ntdb_lockall(ntdb) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 13);
+		ok1(ntdb_wipe_all(ntdb) == NTDB_ERR_RDONLY);
+		ok1(tap_log_messages == 14);
+
+		/* Can't remove NTDB_RDONLY since we opened with O_RDONLY */
+		ntdb_remove_flag(ntdb, NTDB_RDONLY);
+		ok1(tap_log_messages == 15);
+		ok1(ntdb_get_flags(ntdb) & NTDB_RDONLY);
+		ntdb_close(ntdb);
+
+		ok1(tap_log_messages == 15);
+		tap_log_messages = 0;
+	}
+	return exit_status();
+}

lib/ntdb/test/api-93-repack.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -8,21 +8,21 @@
 
 #define NUM_TESTS 1000
 
-static bool store_all(struct tdb_context *tdb)
+static bool store_all(struct ntdb_context *ntdb)
 {
 	unsigned int i;
-	struct tdb_data key = { (unsigned char *)&i, sizeof(i) };
-	struct tdb_data dbuf = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA dbuf = { (unsigned char *)&i, sizeof(i) };
 
 	for (i = 0; i < NUM_TESTS; i++) {
-		if (tdb_store(tdb, key, dbuf, TDB_INSERT) != TDB_SUCCESS)
+		if (ntdb_store(ntdb, key, dbuf, NTDB_INSERT) != NTDB_SUCCESS)
 			return false;
 	}
 	return true;
 }
 
-static int mark_entry(struct tdb_context *tdb,
-		      TDB_DATA key, TDB_DATA data, bool found[])
+static int mark_entry(struct ntdb_context *ntdb,
+		      NTDB_DATA key, NTDB_DATA data, bool found[])
 {
 	unsigned int num;
 
@@ -51,28 +51,28 @@ int main(int argc, char *argv[])
 {
 	unsigned int i;
 	bool found[NUM_TESTS];
-	struct tdb_context *tdb;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT
 	};
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 6 + 1);
 
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-93-repack.tdb", flags[i],
+		ntdb = ntdb_open("run-93-repack.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			break;
 
-		ok1(store_all(tdb));
+		ok1(store_all(ntdb));
 
-		ok1(tdb_repack(tdb) == TDB_SUCCESS);
+		ok1(ntdb_repack(ntdb) == NTDB_SUCCESS);
 		memset(found, 0, sizeof(found));
-		ok1(tdb_check(tdb, NULL, NULL) == TDB_SUCCESS);
-		ok1(tdb_traverse(tdb, mark_entry, found) == NUM_TESTS);
+		ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
+		ok1(ntdb_traverse(ntdb, mark_entry, found) == NUM_TESTS);
 		ok1(is_all_set(found, NUM_TESTS));
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/api-add-remove-flags.c

@@ -0,0 +1,89 @@
+#include "private.h" // for ntdb_context
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(87);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("run-add-remove-flags.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		ok1(ntdb_get_flags(ntdb) == ntdb->flags);
+		tap_log_messages = 0;
+		ntdb_add_flag(ntdb, NTDB_NOLOCK);
+		if (flags[i] & NTDB_INTERNAL)
+			ok1(tap_log_messages == 1);
+		else {
+			ok1(tap_log_messages == 0);
+			ok1(ntdb_get_flags(ntdb) & NTDB_NOLOCK);
+		}
+
+		tap_log_messages = 0;
+		ntdb_add_flag(ntdb, NTDB_NOMMAP);
+		if (flags[i] & NTDB_INTERNAL)
+			ok1(tap_log_messages == 1);
+		else {
+			ok1(tap_log_messages == 0);
+			ok1(ntdb_get_flags(ntdb) & NTDB_NOMMAP);
+			ok1(ntdb->file->map_ptr == NULL);
+		}
+
+		tap_log_messages = 0;
+		ntdb_add_flag(ntdb, NTDB_NOSYNC);
+		if (flags[i] & NTDB_INTERNAL)
+			ok1(tap_log_messages == 1);
+		else {
+			ok1(tap_log_messages == 0);
+			ok1(ntdb_get_flags(ntdb) & NTDB_NOSYNC);
+		}
+
+		ok1(ntdb_get_flags(ntdb) == ntdb->flags);
+
+		tap_log_messages = 0;
+		ntdb_remove_flag(ntdb, NTDB_NOLOCK);
+		if (flags[i] & NTDB_INTERNAL)
+			ok1(tap_log_messages == 1);
+		else {
+			ok1(tap_log_messages == 0);
+			ok1(!(ntdb_get_flags(ntdb) & NTDB_NOLOCK));
+		}
+
+		tap_log_messages = 0;
+		ntdb_remove_flag(ntdb, NTDB_NOMMAP);
+		if (flags[i] & NTDB_INTERNAL)
+			ok1(tap_log_messages == 1);
+		else {
+			ok1(tap_log_messages == 0);
+			ok1(!(ntdb_get_flags(ntdb) & NTDB_NOMMAP));
+			ok1(ntdb->file->map_ptr != NULL);
+		}
+
+		tap_log_messages = 0;
+		ntdb_remove_flag(ntdb, NTDB_NOSYNC);
+		if (flags[i] & NTDB_INTERNAL)
+			ok1(tap_log_messages == 1);
+		else {
+			ok1(tap_log_messages == 0);
+			ok1(!(ntdb_get_flags(ntdb) & NTDB_NOSYNC));
+		}
+
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/api-check-callback.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -8,77 +8,77 @@
 
 #define NUM_RECORDS 1000
 
-static bool store_records(struct tdb_context *tdb)
+static bool store_records(struct ntdb_context *ntdb)
 {
 	int i;
-	struct tdb_data key = { (unsigned char *)&i, sizeof(i) };
-	struct tdb_data data = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
 
 	for (i = 0; i < NUM_RECORDS; i++)
-		if (tdb_store(tdb, key, data, TDB_REPLACE) != 0)
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
 			return false;
 	return true;
 }
 
-static enum TDB_ERROR check(struct tdb_data key,
-			    struct tdb_data data,
+static enum NTDB_ERROR check(NTDB_DATA key,
+			    NTDB_DATA data,
 			    bool *array)
 {
 	int val;
 
 	if (key.dsize != sizeof(val)) {
 		diag("Wrong key size: %u\n", key.dsize);
-		return TDB_ERR_CORRUPT;
+		return NTDB_ERR_CORRUPT;
 	}
 
 	if (key.dsize != data.dsize
 	    || memcmp(key.dptr, data.dptr, sizeof(val)) != 0) {
 		diag("Key and data differ\n");
-		return TDB_ERR_CORRUPT;
+		return NTDB_ERR_CORRUPT;
 	}
 
 	memcpy(&val, key.dptr, sizeof(val));
 	if (val >= NUM_RECORDS || val < 0) {
 		diag("check value %i\n", val);
-		return TDB_ERR_CORRUPT;
+		return NTDB_ERR_CORRUPT;
 	}
 
 	if (array[val]) {
 		diag("Value %i already seen\n", val);
-		return TDB_ERR_CORRUPT;
+		return NTDB_ERR_CORRUPT;
 	}
 
 	array[val] = true;
-	return TDB_SUCCESS;
+	return NTDB_SUCCESS;
 }
 
 int main(int argc, char *argv[])
 {
 	unsigned int i, j;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 4 + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
 		bool array[NUM_RECORDS];
 
-		tdb = tdb_open("run-check-callback.tdb", flags[i],
+		ntdb = ntdb_open("run-check-callback.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
-		ok1(store_records(tdb));
+		ok1(store_records(ntdb));
 		for (j = 0; j < NUM_RECORDS; j++)
 			array[j] = false;
-		ok1(tdb_check(tdb, check, array) == TDB_SUCCESS);
+		ok1(ntdb_check(ntdb, check, array) == NTDB_SUCCESS);
 		for (j = 0; j < NUM_RECORDS; j++)
 			if (!array[j])
 				break;
 		ok1(j == NUM_RECORDS);
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/api-firstkey-nextkey.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -9,14 +9,14 @@
 
 #define NUM_RECORDS 1000
 
-static bool store_records(struct tdb_context *tdb)
+static bool store_records(struct ntdb_context *ntdb)
 {
 	int i;
-	struct tdb_data key = { (unsigned char *)&i, sizeof(i) };
-	struct tdb_data data = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
 
 	for (i = 0; i < NUM_RECORDS; i++)
-		if (tdb_store(tdb, key, data, TDB_REPLACE) != 0)
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
 			return false;
 	return true;
 }
@@ -26,7 +26,7 @@ struct trav_data {
 	unsigned int calls;
 };
 
-static int trav(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, void *p)
+static int trav(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *p)
 {
 	struct trav_data *td = p;
 	int val;
@@ -36,8 +36,8 @@ static int trav(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, void *p)
 	return 0;
 }
 
-/* Since tdb_nextkey frees dptr, we need to clone it. */
-static TDB_DATA dup_key(TDB_DATA key)
+/* Since ntdb_nextkey frees dptr, we need to clone it. */
+static NTDB_DATA dup_key(NTDB_DATA key)
 {
 	void *p = malloc(key.dsize);
 	memcpy(p, key.dptr, key.dsize);
@@ -50,81 +50,81 @@ int main(int argc, char *argv[])
 	unsigned int i, j;
 	int num;
 	struct trav_data td;
-	TDB_DATA k;
-	struct tdb_context *tdb;
-	union tdb_attribute seed_attr;
-	enum TDB_ERROR ecode;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
+	NTDB_DATA k;
+	struct ntdb_context *ntdb;
+	union ntdb_attribute seed_attr;
+	enum NTDB_ERROR ecode;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
 
-	seed_attr.base.attr = TDB_ATTRIBUTE_SEED;
+	seed_attr.base.attr = NTDB_ATTRIBUTE_SEED;
 	seed_attr.base.next = &tap_log_attr;
 	seed_attr.seed.seed = 6334326220117065685ULL;
 
 	plan_tests(sizeof(flags) / sizeof(flags[0])
 		   * (NUM_RECORDS*6 + (NUM_RECORDS-1)*3 + 22) + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("api-firstkey-nextkey.tdb", flags[i],
+		ntdb = ntdb_open("api-firstkey-nextkey.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600,
 			       &seed_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
-		ok1(tdb_firstkey(tdb, &k) == TDB_ERR_NOEXIST);
+		ok1(ntdb_firstkey(ntdb, &k) == NTDB_ERR_NOEXIST);
 
 		/* One entry... */
 		k.dptr = (unsigned char *)&num;
 		k.dsize = sizeof(num);
 		num = 0;
-		ok1(tdb_store(tdb, k, k, TDB_INSERT) == 0);
-		ok1(tdb_firstkey(tdb, &k) == TDB_SUCCESS);
+		ok1(ntdb_store(ntdb, k, k, NTDB_INSERT) == 0);
+		ok1(ntdb_firstkey(ntdb, &k) == NTDB_SUCCESS);
 		ok1(k.dsize == sizeof(num));
 		ok1(memcmp(k.dptr, &num, sizeof(num)) == 0);
-		ok1(tdb_nextkey(tdb, &k) == TDB_ERR_NOEXIST);
+		ok1(ntdb_nextkey(ntdb, &k) == NTDB_ERR_NOEXIST);
 
 		/* Two entries. */
 		k.dptr = (unsigned char *)&num;
 		k.dsize = sizeof(num);
 		num = 1;
-		ok1(tdb_store(tdb, k, k, TDB_INSERT) == 0);
-		ok1(tdb_firstkey(tdb, &k) == TDB_SUCCESS);
+		ok1(ntdb_store(ntdb, k, k, NTDB_INSERT) == 0);
+		ok1(ntdb_firstkey(ntdb, &k) == NTDB_SUCCESS);
 		ok1(k.dsize == sizeof(num));
 		memcpy(&num, k.dptr, sizeof(num));
 		ok1(num == 0 || num == 1);
-		ok1(tdb_nextkey(tdb, &k) == TDB_SUCCESS);
+		ok1(ntdb_nextkey(ntdb, &k) == NTDB_SUCCESS);
 		ok1(k.dsize == sizeof(j));
 		memcpy(&j, k.dptr, sizeof(j));
 		ok1(j == 0 || j == 1);
 		ok1(j != num);
-		ok1(tdb_nextkey(tdb, &k) == TDB_ERR_NOEXIST);
+		ok1(ntdb_nextkey(ntdb, &k) == NTDB_ERR_NOEXIST);
 
 		/* Clean up. */
 		k.dptr = (unsigned char *)&num;
 		k.dsize = sizeof(num);
 		num = 0;
-		ok1(tdb_delete(tdb, k) == 0);
+		ok1(ntdb_delete(ntdb, k) == 0);
 		num = 1;
-		ok1(tdb_delete(tdb, k) == 0);
+		ok1(ntdb_delete(ntdb, k) == 0);
 
 		/* Now lots of records. */
-		ok1(store_records(tdb));
+		ok1(store_records(ntdb));
 		td.calls = 0;
 
-		num = tdb_traverse(tdb, trav, &td);
+		num = ntdb_traverse(ntdb, trav, &td);
 		ok1(num == NUM_RECORDS);
 		ok1(td.calls == NUM_RECORDS);
 
-		/* Simple loop should match tdb_traverse */
-		for (j = 0, ecode = tdb_firstkey(tdb, &k); j < td.calls; j++) {
+		/* Simple loop should match ntdb_traverse */
+		for (j = 0, ecode = ntdb_firstkey(ntdb, &k); j < td.calls; j++) {
 			int val;
 
-			ok1(ecode == TDB_SUCCESS);
+			ok1(ecode == NTDB_SUCCESS);
 			ok1(k.dsize == sizeof(val));
 			memcpy(&val, k.dptr, k.dsize);
 			ok1(td.records[j] == val);
-			ecode = tdb_nextkey(tdb, &k);
+			ecode = ntdb_nextkey(ntdb, &k);
 		}
 
 		/* But arbitrary orderings should work too. */
@@ -132,26 +132,26 @@ int main(int argc, char *argv[])
 			k.dptr = (unsigned char *)&td.records[j-1];
 			k.dsize = sizeof(td.records[j-1]);
 			k = dup_key(k);
-			ok1(tdb_nextkey(tdb, &k) == TDB_SUCCESS);
+			ok1(ntdb_nextkey(ntdb, &k) == NTDB_SUCCESS);
 			ok1(k.dsize == sizeof(td.records[j]));
 			ok1(memcmp(k.dptr, &td.records[j], k.dsize) == 0);
 			free(k.dptr);
 		}
 
 		/* Even delete should work. */
-		for (j = 0, ecode = tdb_firstkey(tdb, &k);
-		     ecode != TDB_ERR_NOEXIST;
+		for (j = 0, ecode = ntdb_firstkey(ntdb, &k);
+		     ecode != NTDB_ERR_NOEXIST;
 		     j++) {
-			ok1(ecode == TDB_SUCCESS);
+			ok1(ecode == NTDB_SUCCESS);
 			ok1(k.dsize == 4);
-			ok1(tdb_delete(tdb, k) == 0);
-			ecode = tdb_nextkey(tdb, &k);
+			ok1(ntdb_delete(ntdb, k) == 0);
+			ecode = ntdb_nextkey(ntdb, &k);
 		}
 
 		diag("delete using first/nextkey gave %u of %u records",
 		     j, NUM_RECORDS);
 		ok1(j == NUM_RECORDS);
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/api-fork-test.c

@@ -1,17 +1,17 @@
 /* Test forking while holding lock.
  *
  * There are only five ways to do this currently:
- * (1) grab a tdb_chainlock, then fork.
- * (2) grab a tdb_lockall, then fork.
- * (3) grab a tdb_lockall_read, then fork.
+ * (1) grab a ntdb_chainlock, then fork.
+ * (2) grab a ntdb_lockall, then fork.
+ * (3) grab a ntdb_lockall_read, then fork.
  * (4) start a transaction, then fork.
- * (5) fork from inside a tdb_parse() callback.
+ * (5) fork from inside a ntdb_parse() callback.
  *
- * Note that we don't hold a lock across tdb_traverse callbacks, so
+ * Note that we don't hold a lock across ntdb_traverse callbacks, so
  * that doesn't matter.
  */
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -22,40 +22,40 @@
 #include <stdlib.h>
 #include "logging.h"
 
-static enum TDB_ERROR fork_in_parse(TDB_DATA key, TDB_DATA data,
-				    struct tdb_context *tdb)
+static enum NTDB_ERROR fork_in_parse(NTDB_DATA key, NTDB_DATA data,
+				    struct ntdb_context *ntdb)
 {
 	int status;
 
 	if (fork() == 0) {
 		/* We expect this to fail. */
-		if (tdb_store(tdb, key, data, TDB_REPLACE) != TDB_ERR_LOCK)
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != NTDB_ERR_LOCK)
 			exit(1);
 
-		if (tdb_fetch(tdb, key, &data) != TDB_ERR_LOCK)
+		if (ntdb_fetch(ntdb, key, &data) != NTDB_ERR_LOCK)
 			exit(1);
 
 		if (tap_log_messages != 2)
 			exit(2);
 
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 		if (tap_log_messages != 2)
 			exit(3);
 		exit(0);
 	}
 	wait(&status);
 	ok1(WIFEXITED(status) && WEXITSTATUS(status) == 0);
-	return TDB_SUCCESS;
+	return NTDB_SUCCESS;
 }
 
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data = tdb_mkdata("data", 4);
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 14);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
@@ -63,116 +63,116 @@ int main(int argc, char *argv[])
 
 		tap_log_messages = 0;
 
-		tdb = tdb_open("run-fork-test.tdb", flags[i],
+		ntdb = ntdb_open("run-fork-test.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			continue;
 
 		/* Put a record in here. */
-		ok1(tdb_store(tdb, key, data, TDB_REPLACE) == TDB_SUCCESS);
+		ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == NTDB_SUCCESS);
 
-		ok1(tdb_chainlock(tdb, key) == TDB_SUCCESS);
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_SUCCESS);
 		if (fork() == 0) {
 			/* We expect this to fail. */
-			if (tdb_store(tdb, key, data, TDB_REPLACE) != TDB_ERR_LOCK)
+			if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != NTDB_ERR_LOCK)
 				return 1;
 
-			if (tdb_fetch(tdb, key, &data) != TDB_ERR_LOCK)
+			if (ntdb_fetch(ntdb, key, &data) != NTDB_ERR_LOCK)
 				return 1;
 
 			if (tap_log_messages != 2)
 				return 2;
 
-			tdb_chainunlock(tdb, key);
+			ntdb_chainunlock(ntdb, key);
 			if (tap_log_messages != 3)
 				return 3;
-			tdb_close(tdb);
+			ntdb_close(ntdb);
 			if (tap_log_messages != 3)
 				return 4;
 			return 0;
 		}
 		wait(&status);
 		ok1(WIFEXITED(status) && WEXITSTATUS(status) == 0);
-		tdb_chainunlock(tdb, key);
+		ntdb_chainunlock(ntdb, key);
 
-		ok1(tdb_lockall(tdb) == TDB_SUCCESS);
+		ok1(ntdb_lockall(ntdb) == NTDB_SUCCESS);
 		if (fork() == 0) {
 			/* We expect this to fail. */
-			if (tdb_store(tdb, key, data, TDB_REPLACE) != TDB_ERR_LOCK)
+			if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != NTDB_ERR_LOCK)
 				return 1;
 
-			if (tdb_fetch(tdb, key, &data) != TDB_ERR_LOCK)
+			if (ntdb_fetch(ntdb, key, &data) != NTDB_ERR_LOCK)
 				return 1;
 
 			if (tap_log_messages != 2)
 				return 2;
 
-			tdb_unlockall(tdb);
+			ntdb_unlockall(ntdb);
 			if (tap_log_messages != 2)
 				return 3;
-			tdb_close(tdb);
+			ntdb_close(ntdb);
 			if (tap_log_messages != 2)
 				return 4;
 			return 0;
 		}
 		wait(&status);
 		ok1(WIFEXITED(status) && WEXITSTATUS(status) == 0);
-		tdb_unlockall(tdb);
+		ntdb_unlockall(ntdb);
 
-		ok1(tdb_lockall_read(tdb) == TDB_SUCCESS);
+		ok1(ntdb_lockall_read(ntdb) == NTDB_SUCCESS);
 		if (fork() == 0) {
 			/* We expect this to fail. */
 			/* This would always fail anyway... */
-			if (tdb_store(tdb, key, data, TDB_REPLACE) != TDB_ERR_LOCK)
+			if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != NTDB_ERR_LOCK)
 				return 1;
 
-			if (tdb_fetch(tdb, key, &data) != TDB_ERR_LOCK)
+			if (ntdb_fetch(ntdb, key, &data) != NTDB_ERR_LOCK)
 				return 1;
 
 			if (tap_log_messages != 2)
 				return 2;
 
-			tdb_unlockall_read(tdb);
+			ntdb_unlockall_read(ntdb);
 			if (tap_log_messages != 2)
 				return 3;
-			tdb_close(tdb);
+			ntdb_close(ntdb);
 			if (tap_log_messages != 2)
 				return 4;
 			return 0;
 		}
 		wait(&status);
 		ok1(WIFEXITED(status) && WEXITSTATUS(status) == 0);
-		tdb_unlockall_read(tdb);
+		ntdb_unlockall_read(ntdb);
 
-		ok1(tdb_transaction_start(tdb) == TDB_SUCCESS);
+		ok1(ntdb_transaction_start(ntdb) == NTDB_SUCCESS);
 		/* If transactions is empty, noop "commit" succeeds. */
-		ok1(tdb_delete(tdb, key) == TDB_SUCCESS);
+		ok1(ntdb_delete(ntdb, key) == NTDB_SUCCESS);
 		if (fork() == 0) {
 			/* We expect this to fail. */
-			if (tdb_store(tdb, key, data, TDB_REPLACE) != TDB_ERR_LOCK)
+			if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != NTDB_ERR_LOCK)
 				return 1;
 
-			if (tdb_fetch(tdb, key, &data) != TDB_ERR_LOCK)
+			if (ntdb_fetch(ntdb, key, &data) != NTDB_ERR_LOCK)
 				return 1;
 
 			if (tap_log_messages != 2)
 				return 2;
 
-			if (tdb_transaction_commit(tdb) != TDB_ERR_LOCK)
+			if (ntdb_transaction_commit(ntdb) != NTDB_ERR_LOCK)
 				return 3;
 
-			tdb_close(tdb);
+			ntdb_close(ntdb);
 			if (tap_log_messages < 3)
 				return 4;
 			return 0;
 		}
 		wait(&status);
 		ok1(WIFEXITED(status) && WEXITSTATUS(status) == 0);
-		tdb_transaction_cancel(tdb);
+		ntdb_transaction_cancel(ntdb);
 
-		ok1(tdb_parse_record(tdb, key, fork_in_parse, tdb)
-		    == TDB_SUCCESS);
-		tdb_close(tdb);
+		ok1(ntdb_parse_record(ntdb, key, fork_in_parse, ntdb)
+		    == NTDB_SUCCESS);
+		ntdb_close(ntdb);
 		ok1(tap_log_messages == 0);
 	}
 	return exit_status();

lib/ntdb/test/api-locktimeout.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include "system/wait.h"
 #include <sys/types.h>
@@ -94,35 +94,35 @@ static int timeout_lock(int fd, int rw, off_t off, off_t len, bool waitflag,
 	return ret;
 }
 
-static int tdb_chainlock_with_timeout_internal(struct tdb_context *tdb,
-					       TDB_DATA key,
+static int ntdb_chainlock_with_timeout_internal(struct ntdb_context *ntdb,
+					       NTDB_DATA key,
 					       unsigned int timeout,
 					       int rw_type)
 {
-	union tdb_attribute locking;
-	enum TDB_ERROR ecode;
+	union ntdb_attribute locking;
+	enum NTDB_ERROR ecode;
 
 	if (timeout) {
-		locking.base.attr = TDB_ATTRIBUTE_FLOCK;
-		ecode = tdb_get_attribute(tdb, &locking);
-		if (ecode != TDB_SUCCESS)
+		locking.base.attr = NTDB_ATTRIBUTE_FLOCK;
+		ecode = ntdb_get_attribute(ntdb, &locking);
+		if (ecode != NTDB_SUCCESS)
 			return ecode;
 
 		/* Replace locking function with our own. */
 		locking.flock.data = &timeout;
 		locking.flock.lock = timeout_lock;
 
-		ecode = tdb_set_attribute(tdb, &locking);
-		if (ecode != TDB_SUCCESS)
+		ecode = ntdb_set_attribute(ntdb, &locking);
+		if (ecode != NTDB_SUCCESS)
 			return ecode;
 	}
 	if (rw_type == F_RDLCK)
-		ecode = tdb_chainlock_read(tdb, key);
+		ecode = ntdb_chainlock_read(ntdb, key);
 	else
-		ecode = tdb_chainlock(tdb, key);
+		ecode = ntdb_chainlock(ntdb, key);
 
 	if (timeout) {
-		tdb_unset_attribute(tdb, TDB_ATTRIBUTE_FLOCK);
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_FLOCK);
 	}
 	return ecode;
 }
@@ -130,10 +130,10 @@ static int tdb_chainlock_with_timeout_internal(struct tdb_context *tdb,
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	TDB_DATA key = tdb_mkdata("hello", 5);
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
+	struct ntdb_context *ntdb;
+	NTDB_DATA key = ntdb_mkdata("hello", 5);
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
 	struct agent *agent;
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 15);
@@ -141,52 +141,52 @@ int main(int argc, char *argv[])
 	agent = prepare_external_agent();
 
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		enum TDB_ERROR ecode;
-		tdb = tdb_open("run-locktimeout.tdb", flags[i],
+		enum NTDB_ERROR ecode;
+		ntdb = ntdb_open("run-locktimeout.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			break;
 
 		/* Simple cases: should succeed. */
-		ecode = tdb_chainlock_with_timeout_internal(tdb, key, 20,
+		ecode = ntdb_chainlock_with_timeout_internal(ntdb, key, 20,
 							    F_RDLCK);
-		ok1(ecode == TDB_SUCCESS);
+		ok1(ecode == NTDB_SUCCESS);
 		ok1(tap_log_messages == 0);
 
-		tdb_chainunlock_read(tdb, key);
+		ntdb_chainunlock_read(ntdb, key);
 		ok1(tap_log_messages == 0);
 
-		ecode = tdb_chainlock_with_timeout_internal(tdb, key, 20,
+		ecode = ntdb_chainlock_with_timeout_internal(ntdb, key, 20,
 							    F_WRLCK);
-		ok1(ecode == TDB_SUCCESS);
+		ok1(ecode == NTDB_SUCCESS);
 		ok1(tap_log_messages == 0);
 
-		tdb_chainunlock(tdb, key);
+		ntdb_chainunlock(ntdb, key);
 		ok1(tap_log_messages == 0);
 
 		/* OK, get agent to start transaction, then we should time out. */
-		ok1(external_agent_operation(agent, OPEN, "run-locktimeout.tdb")
+		ok1(external_agent_operation(agent, OPEN, "run-locktimeout.ntdb")
 		    == SUCCESS);
 		ok1(external_agent_operation(agent, TRANSACTION_START, "")
 		    == SUCCESS);
-		ecode = tdb_chainlock_with_timeout_internal(tdb, key, 20,
+		ecode = ntdb_chainlock_with_timeout_internal(ntdb, key, 20,
 							    F_WRLCK);
-		ok1(ecode == TDB_ERR_LOCK);
+		ok1(ecode == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 
 		/* Even if we get a different signal, should be fine. */
 		CatchSignal(SIGUSR1, do_nothing);
 		external_agent_operation(agent, SEND_SIGNAL, "");
-		ecode = tdb_chainlock_with_timeout_internal(tdb, key, 20,
+		ecode = ntdb_chainlock_with_timeout_internal(ntdb, key, 20,
 							    F_WRLCK);
-		ok1(ecode == TDB_ERR_LOCK);
+		ok1(ecode == NTDB_ERR_LOCK);
 		ok1(tap_log_messages == 0);
 
 		ok1(external_agent_operation(agent, TRANSACTION_COMMIT, "")
 		    == SUCCESS);
 		ok1(external_agent_operation(agent, CLOSE, "")
 		    == SUCCESS);
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 	free_external_agent(agent);
 	return exit_status();

lib/ntdb/test/api-missing-entries.c

@@ -1,6 +1,6 @@
 /* Another test revealed that we lost an entry.  This reproduces it. */
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include <ccan/hash/hash.h>
 #include "tap-interface.h"
 #include <sys/types.h>
@@ -20,23 +20,23 @@ static uint64_t failhash(const void *key, size_t len, uint64_t seed, void *p)
 int main(int argc, char *argv[])
 {
 	int i;
-	struct tdb_context *tdb;
-	struct tdb_data key = { (unsigned char *)&i, sizeof(i) };
-	struct tdb_data data = { (unsigned char *)&i, sizeof(i) };
-	union tdb_attribute hattr = { .hash = { .base = { TDB_ATTRIBUTE_HASH },
+	struct ntdb_context *ntdb;
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
+	union ntdb_attribute hattr = { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
 						.fn = failhash } };
 
 	hattr.base.next = &tap_log_attr;
 	plan_tests(1 + NUM_RECORDS + 2);
 
-	tdb = tdb_open("run-missing-entries.tdb", TDB_INTERNAL,
+	ntdb = ntdb_open("run-missing-entries.ntdb", NTDB_INTERNAL,
 		       O_RDWR|O_CREAT|O_TRUNC, 0600, &hattr);
-	if (ok1(tdb)) {
+	if (ok1(ntdb)) {
 		for (i = 0; i < NUM_RECORDS; i++) {
-			ok1(tdb_store(tdb, key, data, TDB_REPLACE) == 0);
+			ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == 0);
 		}
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
-		tdb_close(tdb);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/api-open-multiple-times.c

@@ -0,0 +1,83 @@
+#include "config.h"
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb, *ntdb2;
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA d = { NULL, 0 }; /* Bogus GCC warning */
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 28);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("run-open-multiple-times.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		ntdb2 = ntdb_open("run-open-multiple-times.ntdb", flags[i],
+				O_RDWR|O_CREAT, 0600, &tap_log_attr);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ok1(ntdb_check(ntdb2, NULL, NULL) == 0);
+
+		/* Store in one, fetch in the other. */
+		ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == 0);
+		ok1(ntdb_fetch(ntdb2, key, &d) == NTDB_SUCCESS);
+		ok1(ntdb_deq(d, data));
+		free(d.dptr);
+
+		/* Vice versa, with delete. */
+		ok1(ntdb_delete(ntdb2, key) == 0);
+		ok1(ntdb_fetch(ntdb, key, &d) == NTDB_ERR_NOEXIST);
+
+		/* OK, now close first one, check second still good. */
+		ok1(ntdb_close(ntdb) == 0);
+
+		ok1(ntdb_store(ntdb2, key, data, NTDB_REPLACE) == 0);
+		ok1(ntdb_fetch(ntdb2, key, &d) == NTDB_SUCCESS);
+		ok1(ntdb_deq(d, data));
+		free(d.dptr);
+
+		/* Reopen */
+		ntdb = ntdb_open("run-open-multiple-times.ntdb", flags[i],
+			       O_RDWR|O_CREAT, 0600, &tap_log_attr);
+		ok1(ntdb);
+
+		ok1(ntdb_transaction_start(ntdb2) == 0);
+
+		/* Anything in the other one should fail. */
+		ok1(ntdb_fetch(ntdb, key, &d) == NTDB_ERR_LOCK);
+		ok1(tap_log_messages == 1);
+		ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == NTDB_ERR_LOCK);
+		ok1(tap_log_messages == 2);
+		ok1(ntdb_transaction_start(ntdb) == NTDB_ERR_LOCK);
+		ok1(tap_log_messages == 3);
+		ok1(ntdb_chainlock(ntdb, key) == NTDB_ERR_LOCK);
+		ok1(tap_log_messages == 4);
+
+		/* Transaciton should work as normal. */
+		ok1(ntdb_store(ntdb2, key, data, NTDB_REPLACE) == NTDB_SUCCESS);
+
+		/* Now... try closing with locks held. */
+		ok1(ntdb_close(ntdb2) == 0);
+
+		ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+		ok1(ntdb_deq(d, data));
+		free(d.dptr);
+		ok1(ntdb_close(ntdb) == 0);
+		ok1(tap_log_messages == 4);
+		tap_log_messages = 0;
+	}
+
+	return exit_status();
+}

lib/ntdb/test/api-record-expand.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -13,12 +13,12 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data;
 
 	data.dptr = malloc(MAX_SIZE);
 	memset(data.dptr, 0x24, MAX_SIZE);
@@ -26,23 +26,23 @@ int main(int argc, char *argv[])
 	plan_tests(sizeof(flags) / sizeof(flags[0])
 		   * (3 + (1 + (MAX_SIZE/SIZE_STEP)) * 2) + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-record-expand.tdb", flags[i],
+		ntdb = ntdb_open("run-record-expand.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		data.dsize = 0;
-		ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 		for (data.dsize = 0;
 		     data.dsize < MAX_SIZE;
 		     data.dsize += SIZE_STEP) {
 			memset(data.dptr, data.dsize, data.dsize);
-			ok1(tdb_store(tdb, key, data, TDB_MODIFY) == 0);
-			ok1(tdb_check(tdb, NULL, NULL) == 0);
+			ok1(ntdb_store(ntdb, key, data, NTDB_MODIFY) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 		}
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 	ok1(tap_log_messages == 0);
 	free(data.dptr);

lib/ntdb/test/api-simple-delete.c

@@ -0,0 +1,39 @@
+#include "config.h"
+#include "ntdb.h"
+#include "tap-interface.h"
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 7 + 1);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		ntdb = ntdb_open("run-simple-delete.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		if (ntdb) {
+			/* Delete should fail. */
+			ok1(ntdb_delete(ntdb, key) == NTDB_ERR_NOEXIST);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+			/* Insert should succeed. */
+			ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+			/* Delete should now work. */
+			ok1(ntdb_delete(ntdb, key) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+			ntdb_close(ntdb);
+		}
+	}
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/api-summary.c

@@ -1,5 +1,5 @@
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "tap-interface.h"
 #include <sys/types.h>
 #include <sys/stat.h>
@@ -10,38 +10,38 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i, j;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = { (unsigned char *)&j, sizeof(j) };
-	struct tdb_data data = { (unsigned char *)&j, sizeof(j) };
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = { (unsigned char *)&j, sizeof(j) };
+	NTDB_DATA data = { (unsigned char *)&j, sizeof(j) };
 	char *summary;
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * (1 + 2 * 5) + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-summary.tdb", flags[i],
+		ntdb = ntdb_open("run-summary.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		/* Put some stuff in there. */
 		for (j = 0; j < 500; j++) {
 			/* Make sure padding varies to we get some graphs! */
 			data.dsize = j % (sizeof(j) + 1);
-			if (tdb_store(tdb, key, data, TDB_REPLACE) != 0)
-				fail("Storing in tdb");
+			if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
+				fail("Storing in ntdb");
 		}
 
 		for (j = 0;
-		     j <= TDB_SUMMARY_HISTOGRAMS;
-		     j += TDB_SUMMARY_HISTOGRAMS) {
-			ok1(tdb_summary(tdb, j, &summary) == TDB_SUCCESS);
+		     j <= NTDB_SUMMARY_HISTOGRAMS;
+		     j += NTDB_SUMMARY_HISTOGRAMS) {
+			ok1(ntdb_summary(ntdb, j, &summary) == NTDB_SUCCESS);
 			ok1(strstr(summary, "Number of records: 500\n"));
 			ok1(strstr(summary, "Smallest/average/largest keys: 4/4/4\n"));
 			ok1(strstr(summary, "Smallest/average/largest data: 0/2/4\n"));
-			if (j == TDB_SUMMARY_HISTOGRAMS) {
+			if (j == NTDB_SUMMARY_HISTOGRAMS) {
 				ok1(strstr(summary, "|")
 				    && strstr(summary, "*"));
 			} else {
@@ -50,7 +50,7 @@ int main(int argc, char *argv[])
 			}
 			free(summary);
 		}
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/external-agent.c

@@ -14,14 +14,14 @@
 #include <stdio.h>
 #include <stdarg.h>
 
-static struct tdb_context *tdb;
+static struct ntdb_context *ntdb;
 
 void (*external_agent_free)(void *) = free;
 
-static enum TDB_ERROR clear_if_first(int fd, void *arg)
+static enum NTDB_ERROR clear_if_first(int fd, void *arg)
 {
 /* We hold a lock offset 4 always, so we can tell if anyone is holding it.
- * (This is compatible with tdb1's TDB_CLEAR_IF_FIRST flag).  */
+ * (This is compatible with tdb's TDB_CLEAR_IF_FIRST flag).  */
 	struct flock fl;
 
 	fl.l_type = F_WRLCK;
@@ -33,73 +33,73 @@ static enum TDB_ERROR clear_if_first(int fd, void *arg)
 		/* We must be first ones to open it! */
 		diag("agent truncating file!");
 		if (ftruncate(fd, 0) != 0) {
-			return TDB_ERR_IO;
+			return NTDB_ERR_IO;
 		}
 	}
 	fl.l_type = F_RDLCK;
 	if (fcntl(fd, F_SETLKW, &fl) != 0) {
-		return TDB_ERR_IO;
+		return NTDB_ERR_IO;
 	}
-	return TDB_SUCCESS;
+	return NTDB_SUCCESS;
 }
 
 static enum agent_return do_operation(enum operation op, const char *name)
 {
-	TDB_DATA k;
+	NTDB_DATA k;
 	enum agent_return ret;
-	TDB_DATA data;
-	enum TDB_ERROR ecode;
-	union tdb_attribute cif;
+	NTDB_DATA data;
+	enum NTDB_ERROR ecode;
+	union ntdb_attribute cif;
 
-	if (op != OPEN && op != OPEN_WITH_HOOK && !tdb) {
-		diag("external: No tdb open!");
+	if (op != OPEN && op != OPEN_WITH_HOOK && !ntdb) {
+		diag("external: No ntdb open!");
 		return OTHER_FAILURE;
 	}
 
 	diag("external: %s", operation_name(op));
 
-	k = tdb_mkdata(name, strlen(name));
+	k = ntdb_mkdata(name, strlen(name));
 
 	locking_would_block = 0;
 	switch (op) {
 	case OPEN:
-		if (tdb) {
-			diag("Already have tdb %s open", tdb_name(tdb));
+		if (ntdb) {
+			diag("Already have ntdb %s open", ntdb_name(ntdb));
 			return OTHER_FAILURE;
 		}
-		tdb = tdb_open(name, TDB_DEFAULT, O_RDWR, 0, &tap_log_attr);
-		if (!tdb) {
+		ntdb = ntdb_open(name, NTDB_DEFAULT, O_RDWR, 0, &tap_log_attr);
+		if (!ntdb) {
 			if (!locking_would_block)
-				diag("Opening tdb gave %s", strerror(errno));
+				diag("Opening ntdb gave %s", strerror(errno));
 			forget_locking();
 			ret = OTHER_FAILURE;
 		} else
 			ret = SUCCESS;
 		break;
 	case OPEN_WITH_HOOK:
-		if (tdb) {
-			diag("Already have tdb %s open", tdb_name(tdb));
+		if (ntdb) {
+			diag("Already have ntdb %s open", ntdb_name(ntdb));
 			return OTHER_FAILURE;
 		}
-		cif.openhook.base.attr = TDB_ATTRIBUTE_OPENHOOK;
+		cif.openhook.base.attr = NTDB_ATTRIBUTE_OPENHOOK;
 		cif.openhook.base.next = &tap_log_attr;
 		cif.openhook.fn = clear_if_first;
-		tdb = tdb_open(name, TDB_DEFAULT, O_RDWR, 0, &cif);
-		if (!tdb) {
+		ntdb = ntdb_open(name, NTDB_DEFAULT, O_RDWR, 0, &cif);
+		if (!ntdb) {
 			if (!locking_would_block)
-				diag("Opening tdb gave %s", strerror(errno));
+				diag("Opening ntdb gave %s", strerror(errno));
 			forget_locking();
 			ret = OTHER_FAILURE;
 		} else
 			ret = SUCCESS;
 		break;
 	case FETCH:
-		ecode = tdb_fetch(tdb, k, &data);
-		if (ecode == TDB_ERR_NOEXIST) {
+		ecode = ntdb_fetch(ntdb, k, &data);
+		if (ecode == NTDB_ERR_NOEXIST) {
 			ret = FAILED;
 		} else if (ecode < 0) {
 			ret = OTHER_FAILURE;
-		} else if (!tdb_deq(data, k)) {
+		} else if (!ntdb_deq(data, k)) {
 			ret = OTHER_FAILURE;
 			external_agent_free(data.dptr);
 		} else {
@@ -108,23 +108,23 @@ static enum agent_return do_operation(enum operation op, const char *name)
 		}
 		break;
 	case STORE:
-		ret = tdb_store(tdb, k, k, 0) == 0 ? SUCCESS : OTHER_FAILURE;
+		ret = ntdb_store(ntdb, k, k, 0) == 0 ? SUCCESS : OTHER_FAILURE;
 		break;
 	case TRANSACTION_START:
-		ret = tdb_transaction_start(tdb) == 0 ? SUCCESS : OTHER_FAILURE;
+		ret = ntdb_transaction_start(ntdb) == 0 ? SUCCESS : OTHER_FAILURE;
 		break;
 	case TRANSACTION_COMMIT:
-		ret = tdb_transaction_commit(tdb)==0 ? SUCCESS : OTHER_FAILURE;
+		ret = ntdb_transaction_commit(ntdb)==0 ? SUCCESS : OTHER_FAILURE;
 		break;
 	case NEEDS_RECOVERY:
-		ret = external_agent_needs_rec(tdb);
+		ret = external_agent_needs_rec(ntdb);
 		break;
 	case CHECK:
-		ret = tdb_check(tdb, NULL, NULL) == 0 ? SUCCESS : OTHER_FAILURE;
+		ret = ntdb_check(ntdb, NULL, NULL) == 0 ? SUCCESS : OTHER_FAILURE;
 		break;
 	case CLOSE:
-		ret = tdb_close(tdb) == 0 ? SUCCESS : OTHER_FAILURE;
-		tdb = NULL;
+		ret = ntdb_close(ntdb) == 0 ? SUCCESS : OTHER_FAILURE;
+		ntdb = NULL;
 		break;
 	case SEND_SIGNAL:
 		/* We do this async */
@@ -144,7 +144,7 @@ struct agent {
 	int cmdfd, responsefd;
 };
 
-/* Do this before doing any tdb stuff.  Return handle, or NULL. */
+/* Do this before doing any ntdb stuff.  Return handle, or NULL. */
 struct agent *prepare_external_agent(void)
 {
 	int pid, ret;

lib/ntdb/test/external-agent.h

@@ -1,5 +1,5 @@
-#ifndef TDB2_TEST_EXTERNAL_AGENT_H
-#define TDB2_TEST_EXTERNAL_AGENT_H
+#ifndef NTDB_TEST_EXTERNAL_AGENT_H
+#define NTDB_TEST_EXTERNAL_AGENT_H
 
 /* For locking tests, we need a different process to try things at
  * various times. */
@@ -16,7 +16,7 @@ enum operation {
 	CLOSE,
 };
 
-/* Do this before doing any tdb stuff.  Return handle, or -1. */
+/* Do this before doing any ntdb stuff.  Return handle, or -1. */
 struct agent *prepare_external_agent(void);
 
 enum agent_return {
@@ -28,14 +28,14 @@ enum agent_return {
 };
 
 /* Ask the external agent to try to do an operation.
- * name == tdb name for OPEN/OPEN_WITH_CLEAR_IF_FIRST,
+ * name == ntdb name for OPEN/OPEN_WITH_CLEAR_IF_FIRST,
  * record name for FETCH/STORE (store stores name as data too)
  */
 enum agent_return external_agent_operation(struct agent *handle,
 					   enum operation op,
 					   const char *name);
 
-/* Hook into free() on tdb_data in external agent. */
+/* Hook into free() on ntdb_data in external agent. */
 extern void (*external_agent_free)(void *);
 
 /* Mapping enum -> string. */
@@ -45,7 +45,7 @@ const char *operation_name(enum operation op);
 void free_external_agent(struct agent *agent);
 
 /* Internal use: */
-struct tdb_context;
-enum agent_return external_agent_needs_rec(struct tdb_context *tdb);
+struct ntdb_context;
+enum agent_return external_agent_needs_rec(struct ntdb_context *ntdb);
 
-#endif /* TDB2_TEST_EXTERNAL_AGENT_H */
+#endif /* NTDB_TEST_EXTERNAL_AGENT_H */

lib/ntdb/test/failtest_helper.c

@@ -51,8 +51,8 @@ bool exit_check_log(struct tlist_calls *history)
 		if (failmatch(i, URANDOM_READ))
 			continue;
 
-		/* Initial allocation of tdb doesn't log. */
-		if (failmatch(i, INITIAL_TDB_MALLOC))
+		/* Initial allocation of ntdb doesn't log. */
+		if (failmatch(i, INITIAL_NTDB_MALLOC))
 			continue;
 
 		/* We don't block "failures" on non-blocking locks. */
@@ -77,7 +77,7 @@ block_repeat_failures(struct tlist_calls *history)
 	if (failtest_suppress)
 		return FAIL_DONT_FAIL;
 
-	if (failmatch(last, INITIAL_TDB_MALLOC)
+	if (failmatch(last, INITIAL_NTDB_MALLOC)
 	    || failmatch(last, URANDOM_OPEN)
 	    || failmatch(last, URANDOM_READ)) {
 		return FAIL_PROBE;

lib/ntdb/test/failtest_helper.h

@@ -1,10 +1,10 @@
-#ifndef TDB2_TEST_FAILTEST_HELPER_H
-#define TDB2_TEST_FAILTEST_HELPER_H
+#ifndef NTDB_TEST_FAILTEST_HELPER_H
+#define NTDB_TEST_FAILTEST_HELPER_H
 #include <ccan/failtest/failtest.h>
 #include <stdbool.h>
 
 /* FIXME: Check these! */
-#define INITIAL_TDB_MALLOC	"open.c", 403, FAILTEST_MALLOC
+#define INITIAL_NTDB_MALLOC	"open.c", 403, FAILTEST_MALLOC
 #define URANDOM_OPEN		"open.c", 62, FAILTEST_OPEN
 #define URANDOM_READ		"open.c", 42, FAILTEST_READ
 
@@ -16,4 +16,4 @@ enum failtest_result block_repeat_failures(struct tlist_calls *history);
 /* Set this to suppress failure. */
 extern bool failtest_suppress;
 
-#endif /* TDB2_TEST_LOGGING_H */
+#endif /* NTDB_TEST_LOGGING_H */

lib/ntdb/test/helpapi-external-agent.c

@@ -0,0 +1,7 @@
+#include "external-agent.h"
+
+/* This isn't possible with via the ntdb API, but this makes it link. */
+enum agent_return external_agent_needs_rec(struct ntdb_context *ntdb)
+{
+	return FAILED;
+}

lib/ntdb/test/helprun-external-agent.c

@@ -0,0 +1,7 @@
+#include "external-agent.h"
+#include "private.h"
+
+enum agent_return external_agent_needs_rec(struct ntdb_context *ntdb)
+{
+	return ntdb_needs_recovery(ntdb) ? SUCCESS : FAILED;
+}

lib/ntdb/test/helprun-layout.c

@@ -0,0 +1,402 @@
+/* NTDB tools to create various canned database layouts. */
+#include "layout.h"
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#include <ccan/err/err.h>
+#include "logging.h"
+
+struct ntdb_layout *new_ntdb_layout(void)
+{
+	struct ntdb_layout *layout = malloc(sizeof(*layout));
+	layout->num_elems = 0;
+	layout->elem = NULL;
+	return layout;
+}
+
+static void add(struct ntdb_layout *layout, union ntdb_layout_elem elem)
+{
+	layout->elem = realloc(layout->elem,
+			       sizeof(layout->elem[0])
+			       * (layout->num_elems+1));
+	layout->elem[layout->num_elems++] = elem;
+}
+
+void ntdb_layout_add_freetable(struct ntdb_layout *layout)
+{
+	union ntdb_layout_elem elem;
+	elem.base.type = FREETABLE;
+	add(layout, elem);
+}
+
+void ntdb_layout_add_free(struct ntdb_layout *layout, ntdb_len_t len,
+			 unsigned ftable)
+{
+	union ntdb_layout_elem elem;
+	elem.base.type = FREE;
+	elem.free.len = len;
+	elem.free.ftable_num = ftable;
+	add(layout, elem);
+}
+
+void ntdb_layout_add_capability(struct ntdb_layout *layout,
+			       uint64_t type,
+			       bool write_breaks,
+			       bool check_breaks,
+			       bool open_breaks,
+			       ntdb_len_t extra)
+{
+	union ntdb_layout_elem elem;
+	elem.base.type = CAPABILITY;
+	elem.capability.type = type;
+	if (write_breaks)
+		elem.capability.type |= NTDB_CAP_NOWRITE;
+	if (open_breaks)
+		elem.capability.type |= NTDB_CAP_NOOPEN;
+	if (check_breaks)
+		elem.capability.type |= NTDB_CAP_NOCHECK;
+	elem.capability.extra = extra;
+	add(layout, elem);
+}
+
+static NTDB_DATA dup_key(NTDB_DATA key)
+{
+	NTDB_DATA ret;
+	ret.dsize = key.dsize;
+	ret.dptr = malloc(ret.dsize);
+	memcpy(ret.dptr, key.dptr, ret.dsize);
+	return ret;
+}
+
+void ntdb_layout_add_used(struct ntdb_layout *layout,
+			 NTDB_DATA key, NTDB_DATA data,
+			 ntdb_len_t extra)
+{
+	union ntdb_layout_elem elem;
+	elem.base.type = DATA;
+	elem.used.key = dup_key(key);
+	elem.used.data = dup_key(data);
+	elem.used.extra = extra;
+	add(layout, elem);
+}
+
+static ntdb_len_t free_record_len(ntdb_len_t len)
+{
+	return sizeof(struct ntdb_used_record) + len;
+}
+
+static ntdb_len_t data_record_len(struct tle_used *used)
+{
+	ntdb_len_t len;
+	len = sizeof(struct ntdb_used_record)
+		+ used->key.dsize + used->data.dsize + used->extra;
+	assert(len >= sizeof(struct ntdb_free_record));
+	return len;
+}
+
+static ntdb_len_t hashtable_len(struct tle_hashtable *htable)
+{
+	return sizeof(struct ntdb_used_record)
+		+ (sizeof(ntdb_off_t) << NTDB_SUBLEVEL_HASH_BITS)
+		+ htable->extra;
+}
+
+static ntdb_len_t capability_len(struct tle_capability *cap)
+{
+	return sizeof(struct ntdb_capability) + cap->extra;
+}
+
+static ntdb_len_t freetable_len(struct tle_freetable *ftable)
+{
+	return sizeof(struct ntdb_freetable);
+}
+
+static void set_free_record(void *mem, ntdb_len_t len)
+{
+	/* We do all the work in add_to_freetable */
+}
+
+static void add_zero_pad(struct ntdb_used_record *u, size_t len, size_t extra)
+{
+	if (extra)
+		((char *)(u + 1))[len] = '\0';
+}
+
+static void set_data_record(void *mem, struct ntdb_context *ntdb,
+			    struct tle_used *used)
+{
+	struct ntdb_used_record *u = mem;
+
+	set_header(ntdb, u, NTDB_USED_MAGIC, used->key.dsize, used->data.dsize,
+		   used->key.dsize + used->data.dsize + used->extra,
+		   ntdb_hash(ntdb, used->key.dptr, used->key.dsize));
+	memcpy(u + 1, used->key.dptr, used->key.dsize);
+	memcpy((char *)(u + 1) + used->key.dsize,
+	       used->data.dptr, used->data.dsize);
+	add_zero_pad(u, used->key.dsize + used->data.dsize, used->extra);
+}
+
+static void set_hashtable(void *mem, struct ntdb_context *ntdb,
+			  struct tle_hashtable *htable)
+{
+	struct ntdb_used_record *u = mem;
+	ntdb_len_t len = sizeof(ntdb_off_t) << NTDB_SUBLEVEL_HASH_BITS;
+
+	set_header(ntdb, u, NTDB_HTABLE_MAGIC, 0, len, len + htable->extra, 0);
+	memset(u + 1, 0, len);
+	add_zero_pad(u, len, htable->extra);
+}
+
+static void set_capability(void *mem, struct ntdb_context *ntdb,
+			   struct tle_capability *cap, struct ntdb_header *hdr,
+			   ntdb_off_t last_cap)
+{
+	struct ntdb_capability *c = mem;
+	ntdb_len_t len = sizeof(*c) - sizeof(struct ntdb_used_record) + cap->extra;
+
+	c->type = cap->type;
+	c->next = 0;
+	set_header(ntdb, &c->hdr, NTDB_CAP_MAGIC, 0, len, len, 0);
+
+	/* Append to capability list. */
+	if (!last_cap) {
+		hdr->capabilities = cap->base.off;
+	} else {
+		c = (struct ntdb_capability *)((char *)hdr + last_cap);
+		c->next = cap->base.off;
+	}
+}
+
+static void set_freetable(void *mem, struct ntdb_context *ntdb,
+			 struct tle_freetable *freetable, struct ntdb_header *hdr,
+			 ntdb_off_t last_ftable)
+{
+	struct ntdb_freetable *ftable = mem;
+	memset(ftable, 0, sizeof(*ftable));
+	set_header(ntdb, &ftable->hdr, NTDB_FTABLE_MAGIC, 0,
+			sizeof(*ftable) - sizeof(ftable->hdr),
+			sizeof(*ftable) - sizeof(ftable->hdr), 0);
+
+	if (last_ftable) {
+		ftable = (struct ntdb_freetable *)((char *)hdr + last_ftable);
+		ftable->next = freetable->base.off;
+	} else {
+		hdr->free_table = freetable->base.off;
+	}
+}
+
+static void add_to_freetable(struct ntdb_context *ntdb,
+			     ntdb_off_t eoff,
+			     ntdb_off_t elen,
+			     unsigned ftable,
+			     struct tle_freetable *freetable)
+{
+	ntdb->ftable_off = freetable->base.off;
+	ntdb->ftable = ftable;
+	add_free_record(ntdb, eoff, sizeof(struct ntdb_used_record) + elen,
+			NTDB_LOCK_WAIT, false);
+}
+
+static ntdb_off_t hbucket_off(ntdb_off_t group_start, unsigned ingroup)
+{
+	return group_start
+		+ (ingroup % (1 << NTDB_HASH_GROUP_BITS)) * sizeof(ntdb_off_t);
+}
+
+/* Get bits from a value. */
+static uint32_t bits(uint64_t val, unsigned start, unsigned num)
+{
+	assert(num <= 32);
+	return (val >> start) & ((1U << num) - 1);
+}
+
+/* We take bits from the top: that way we can lock whole sections of the hash
+ * by using lock ranges. */
+static uint32_t use_bits(uint64_t h, unsigned num, unsigned *used)
+{
+	*used += num;
+	return bits(h, 64 - *used, num);
+}
+
+static ntdb_off_t encode_offset(ntdb_off_t new_off, unsigned bucket,
+			       uint64_t h)
+{
+	return bucket
+		| new_off
+		| ((uint64_t)bits(h, 64 - NTDB_OFF_UPPER_STEAL_EXTRA,
+				  NTDB_OFF_UPPER_STEAL_EXTRA)
+		   << NTDB_OFF_HASH_EXTRA_BIT);
+}
+
+/* FIXME: Our hash table handling here is primitive: we don't expand! */
+static void add_to_hashtable(struct ntdb_context *ntdb,
+			     ntdb_off_t eoff,
+			     NTDB_DATA key)
+{
+	uint64_t h = ntdb_hash(ntdb, key.dptr, key.dsize);
+	ntdb_off_t b_off, group_start;
+	unsigned i, group, in_group;
+	unsigned used = 0;
+
+	group = use_bits(h, NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS, &used);
+	in_group = use_bits(h, NTDB_HASH_GROUP_BITS, &used);
+
+	group_start = offsetof(struct ntdb_header, hashtable)
+		+ group * (sizeof(ntdb_off_t) << NTDB_HASH_GROUP_BITS);
+
+	for (i = 0; i < (1 << NTDB_HASH_GROUP_BITS); i++) {
+		unsigned bucket = (in_group + i) % (1 << NTDB_HASH_GROUP_BITS);
+
+		b_off = hbucket_off(group_start, bucket);
+		if (ntdb_read_off(ntdb, b_off) == 0) {
+			ntdb_write_off(ntdb, b_off,
+				      encode_offset(eoff, in_group, h));
+			return;
+		}
+	}
+	abort();
+}
+
+static struct tle_freetable *find_ftable(struct ntdb_layout *layout, unsigned num)
+{
+	unsigned i;
+
+	for (i = 0; i < layout->num_elems; i++) {
+		if (layout->elem[i].base.type != FREETABLE)
+			continue;
+		if (num == 0)
+			return &layout->elem[i].ftable;
+		num--;
+	}
+	abort();
+}
+
+/* FIXME: Support NTDB_CONVERT */
+struct ntdb_context *ntdb_layout_get(struct ntdb_layout *layout,
+				   void (*freefn)(void *),
+				   union ntdb_attribute *attr)
+{
+	unsigned int i;
+	ntdb_off_t off, len, last_ftable, last_cap;
+	char *mem;
+	struct ntdb_context *ntdb;
+
+	off = sizeof(struct ntdb_header);
+
+	/* First pass of layout: calc lengths */
+	for (i = 0; i < layout->num_elems; i++) {
+		union ntdb_layout_elem *e = &layout->elem[i];
+		e->base.off = off;
+		switch (e->base.type) {
+		case FREETABLE:
+			len = freetable_len(&e->ftable);
+			break;
+		case FREE:
+			len = free_record_len(e->free.len);
+			break;
+		case DATA:
+			len = data_record_len(&e->used);
+			break;
+		case HASHTABLE:
+			len = hashtable_len(&e->hashtable);
+			break;
+		case CAPABILITY:
+			len = capability_len(&e->capability);
+			break;
+		default:
+			abort();
+		}
+		off += len;
+	}
+
+	mem = malloc(off);
+	/* Fill with some weird pattern. */
+	memset(mem, 0x99, off);
+	/* Now populate our header, cribbing from a real NTDB header. */
+	ntdb = ntdb_open(NULL, NTDB_INTERNAL, O_RDWR, 0, attr);
+	memcpy(mem, ntdb->file->map_ptr, sizeof(struct ntdb_header));
+
+	/* Mug the ntdb we have to make it use this. */
+	freefn(ntdb->file->map_ptr);
+	ntdb->file->map_ptr = mem;
+	ntdb->file->map_size = off;
+
+	last_ftable = 0;
+	last_cap = 0;
+	for (i = 0; i < layout->num_elems; i++) {
+		union ntdb_layout_elem *e = &layout->elem[i];
+		switch (e->base.type) {
+		case FREETABLE:
+			set_freetable(mem + e->base.off, ntdb, &e->ftable,
+				     (struct ntdb_header *)mem, last_ftable);
+			last_ftable = e->base.off;
+			break;
+		case FREE:
+			set_free_record(mem + e->base.off, e->free.len);
+			break;
+		case DATA:
+			set_data_record(mem + e->base.off, ntdb, &e->used);
+			break;
+		case HASHTABLE:
+			set_hashtable(mem + e->base.off, ntdb, &e->hashtable);
+			break;
+		case CAPABILITY:
+			set_capability(mem + e->base.off, ntdb, &e->capability,
+				       (struct ntdb_header *)mem, last_cap);
+			last_cap = e->base.off;
+			break;
+		}
+	}
+	/* Must have a free table! */
+	assert(last_ftable);
+
+	/* Now fill the free and hash tables. */
+	for (i = 0; i < layout->num_elems; i++) {
+		union ntdb_layout_elem *e = &layout->elem[i];
+		switch (e->base.type) {
+		case FREE:
+			add_to_freetable(ntdb, e->base.off, e->free.len,
+					 e->free.ftable_num,
+					 find_ftable(layout, e->free.ftable_num));
+			break;
+		case DATA:
+			add_to_hashtable(ntdb, e->base.off, e->used.key);
+			break;
+		default:
+			break;
+		}
+	}
+
+	ntdb->ftable_off = find_ftable(layout, 0)->base.off;
+	return ntdb;
+}
+
+void ntdb_layout_write(struct ntdb_layout *layout, void (*freefn)(void *),
+		       union ntdb_attribute *attr, const char *filename)
+{
+	struct ntdb_context *ntdb = ntdb_layout_get(layout, freefn, attr);
+	int fd;
+
+	fd = open(filename, O_WRONLY|O_TRUNC|O_CREAT,  0600);
+	if (fd < 0)
+		err(1, "opening %s for writing", filename);
+	if (write(fd, ntdb->file->map_ptr, ntdb->file->map_size)
+	    != ntdb->file->map_size)
+		err(1, "writing %s", filename);
+	close(fd);
+	ntdb_close(ntdb);
+}
+
+void ntdb_layout_free(struct ntdb_layout *layout)
+{
+	unsigned int i;
+
+	for (i = 0; i < layout->num_elems; i++) {
+		if (layout->elem[i].base.type == DATA) {
+			free(layout->elem[i].used.key.dptr);
+			free(layout->elem[i].used.data.dptr);
+		}
+	}
+	free(layout->elem);
+	free(layout);
+}

lib/ntdb/test/layout.h

@@ -1,35 +1,35 @@
-#ifndef TDB2_TEST_LAYOUT_H
-#define TDB2_TEST_LAYOUT_H
+#ifndef NTDB_TEST_LAYOUT_H
+#define NTDB_TEST_LAYOUT_H
 #include "private.h"
 
-struct tdb_layout *new_tdb_layout(void);
-void tdb_layout_add_freetable(struct tdb_layout *layout);
-void tdb_layout_add_free(struct tdb_layout *layout, tdb_len_t len,
+struct ntdb_layout *new_ntdb_layout(void);
+void ntdb_layout_add_freetable(struct ntdb_layout *layout);
+void ntdb_layout_add_free(struct ntdb_layout *layout, ntdb_len_t len,
 			 unsigned ftable);
-void tdb_layout_add_used(struct tdb_layout *layout,
-			 TDB_DATA key, TDB_DATA data,
-			 tdb_len_t extra);
-void tdb_layout_add_capability(struct tdb_layout *layout,
+void ntdb_layout_add_used(struct ntdb_layout *layout,
+			 NTDB_DATA key, NTDB_DATA data,
+			 ntdb_len_t extra);
+void ntdb_layout_add_capability(struct ntdb_layout *layout,
 			       uint64_t type,
 			       bool write_breaks,
 			       bool check_breaks,
 			       bool open_breaks,
-			       tdb_len_t extra);
+			       ntdb_len_t extra);
 
 #if 0 /* FIXME: Allow allocation of subtables */
-void tdb_layout_add_hashtable(struct tdb_layout *layout,
+void ntdb_layout_add_hashtable(struct ntdb_layout *layout,
 			      int htable_parent, /* -1 == toplevel */
 			      unsigned int bucket,
-			      tdb_len_t extra);
+			      ntdb_len_t extra);
 #endif
 /* freefn is needed if we're using failtest_free. */
-struct tdb_context *tdb_layout_get(struct tdb_layout *layout,
+struct ntdb_context *ntdb_layout_get(struct ntdb_layout *layout,
 				   void (*freefn)(void *),
-				   union tdb_attribute *attr);
-void tdb_layout_write(struct tdb_layout *layout, void (*freefn)(void *),
-		       union tdb_attribute *attr, const char *filename);
+				   union ntdb_attribute *attr);
+void ntdb_layout_write(struct ntdb_layout *layout, void (*freefn)(void *),
+		       union ntdb_attribute *attr, const char *filename);
 
-void tdb_layout_free(struct tdb_layout *layout);
+void ntdb_layout_free(struct ntdb_layout *layout);
 
 enum layout_type {
 	FREETABLE, FREE, DATA, HASHTABLE, CAPABILITY
@@ -38,7 +38,7 @@ enum layout_type {
 /* Shared by all union members. */
 struct tle_base {
 	enum layout_type type;
-	tdb_off_t off;
+	ntdb_off_t off;
 };
 
 struct tle_freetable {
@@ -47,31 +47,31 @@ struct tle_freetable {
 
 struct tle_free {
 	struct tle_base base;
-	tdb_len_t len;
+	ntdb_len_t len;
 	unsigned ftable_num;
 };
 
 struct tle_used {
 	struct tle_base base;
-	TDB_DATA key;
-	TDB_DATA data;
-	tdb_len_t extra;
+	NTDB_DATA key;
+	NTDB_DATA data;
+	ntdb_len_t extra;
 };
 
 struct tle_hashtable {
 	struct tle_base base;
 	int parent;
 	unsigned int bucket;
-	tdb_len_t extra;
+	ntdb_len_t extra;
 };
 
 struct tle_capability {
 	struct tle_base base;
 	uint64_t type;
-	tdb_len_t extra;
+	ntdb_len_t extra;
 };
 
-union tdb_layout_elem {
+union ntdb_layout_elem {
 	struct tle_base base;
 	struct tle_freetable ftable;
 	struct tle_free free;
@@ -80,8 +80,8 @@ union tdb_layout_elem {
 	struct tle_capability capability;
 };
 
-struct tdb_layout {
+struct ntdb_layout {
 	unsigned int num_elems;
-	union tdb_layout_elem *elem;
+	union ntdb_layout_elem *elem;
 };
-#endif /* TDB2_TEST_LAYOUT_H */
+#endif /* NTDB_TEST_LAYOUT_H */

lib/ntdb/test/lock-tracking.c

@@ -1,5 +1,5 @@
 /* We save the locks so we can reaquire them. */
-#include "private.h" /* For TDB_HASH_LOCK_START, etc. */
+#include "private.h" /* For NTDB_HASH_LOCK_START, etc. */
 #include <unistd.h>
 #include <fcntl.h>
 #include <stdarg.h>
@@ -88,8 +88,8 @@ int fcntl_with_lockcheck(int fd, int cmd, ... /* arg */ )
 			if (fl_end > i->off && fl_end < i_end)
 				break;
 
-			/* tdb_allrecord_lock does this, handle adjacent: */
-			if (fl->l_start > TDB_HASH_LOCK_START
+			/* ntdb_allrecord_lock does this, handle adjacent: */
+			if (fl->l_start > NTDB_HASH_LOCK_START
 			    && fl->l_start == i_end && fl->l_type == i->type) {
 				if (ret == 0) {
 					i->len = fl->l_len
@@ -102,8 +102,8 @@ int fcntl_with_lockcheck(int fd, int cmd, ... /* arg */ )
 		if (i) {
 			/* Special case: upgrade of allrecord lock. */
 			if (i->type == F_RDLCK && fl->l_type == F_WRLCK
-			    && i->off == TDB_HASH_LOCK_START
-			    && fl->l_start == TDB_HASH_LOCK_START
+			    && i->off == NTDB_HASH_LOCK_START
+			    && fl->l_start == NTDB_HASH_LOCK_START
 			    && i->len == 0
 			    && fl->l_len == 0) {
 				if (ret == 0)

lib/ntdb/test/logging.c

@@ -8,21 +8,21 @@ const char *log_prefix = "";
 char *log_last = NULL;
 bool suppress_logging;
 
-union tdb_attribute tap_log_attr = {
-	.log = { .base = { .attr = TDB_ATTRIBUTE_LOG },
+union ntdb_attribute tap_log_attr = {
+	.log = { .base = { .attr = NTDB_ATTRIBUTE_LOG },
 		 .fn = tap_log_fn }
 };
 
-void tap_log_fn(struct tdb_context *tdb,
-		enum tdb_log_level level,
-		enum TDB_ERROR ecode,
+void tap_log_fn(struct ntdb_context *ntdb,
+		enum ntdb_log_level level,
+		enum NTDB_ERROR ecode,
 		const char *message, void *priv)
 {
 	if (suppress_logging)
 		return;
 
-	diag("tdb log level %u: %s: %s%s",
-	     level, tdb_errorstr(ecode), log_prefix, message);
+	diag("ntdb log level %u: %s: %s%s",
+	     level, ntdb_errorstr(ecode), log_prefix, message);
 	if (log_last)
 		free(log_last);
 	log_last = strdup(message);

lib/ntdb/test/logging.h

@@ -0,0 +1,17 @@
+#ifndef NTDB_TEST_LOGGING_H
+#define NTDB_TEST_LOGGING_H
+#include "ntdb.h"
+#include <stdbool.h>
+#include <string.h>
+
+extern bool suppress_logging;
+extern const char *log_prefix;
+extern unsigned tap_log_messages;
+extern union ntdb_attribute tap_log_attr;
+extern char *log_last;
+
+void tap_log_fn(struct ntdb_context *ntdb,
+		enum ntdb_log_level level,
+		enum NTDB_ERROR ecode,
+		const char *message, void *priv);
+#endif /* NTDB_TEST_LOGGING_H */

lib/ntdb/test/ntdb-source.h

@@ -6,6 +6,6 @@
 #include "lock.c"
 #include "open.c"
 #include "summary.c"
-#include "tdb.c"
+#include "ntdb.c"
 #include "transaction.c"
 #include "traverse.c"

lib/ntdb/test/run-001-encode.c

@@ -1,24 +1,24 @@
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include "logging.h"
 
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_used_record rec;
-	struct tdb_context tdb = { .log_fn = tap_log_fn };
+	struct ntdb_used_record rec;
+	struct ntdb_context ntdb = { .log_fn = tap_log_fn };
 
 	plan_tests(64 + 32 + 48*6 + 1);
 
 	/* We should be able to encode any data value. */
 	for (i = 0; i < 64; i++)
-		ok1(set_header(&tdb, &rec, TDB_USED_MAGIC, 0, 1ULL << i,
+		ok1(set_header(&ntdb, &rec, NTDB_USED_MAGIC, 0, 1ULL << i,
 			       1ULL << i, 0) == 0);
 
 	/* And any key and data with < 64 bits between them. */
 	for (i = 0; i < 32; i++) {
-		tdb_len_t dlen = 1ULL >> (63 - i), klen = 1ULL << i;
-		ok1(set_header(&tdb, &rec, TDB_USED_MAGIC, klen, dlen,
+		ntdb_len_t dlen = 1ULL >> (63 - i), klen = 1ULL << i;
+		ok1(set_header(&ntdb, &rec, NTDB_USED_MAGIC, klen, dlen,
 			       klen + dlen, 0)  == 0);
 	}
 
@@ -28,13 +28,13 @@ int main(int argc, char *argv[])
 		uint64_t klen = 1ULL << (i < 16 ? i : 15);
 		uint64_t dlen = 1ULL << i;
 		uint64_t xlen = 1ULL << (i < 32 ? i : 31);
-		ok1(set_header(&tdb, &rec, TDB_USED_MAGIC, klen, dlen,
+		ok1(set_header(&ntdb, &rec, NTDB_USED_MAGIC, klen, dlen,
 			       klen+dlen+xlen, h) == 0);
 		ok1(rec_key_length(&rec) == klen);
 		ok1(rec_data_length(&rec) == dlen);
 		ok1(rec_extra_padding(&rec) == xlen);
 		ok1((uint64_t)rec_hash(&rec) == h);
-		ok1(rec_magic(&rec) == TDB_USED_MAGIC);
+		ok1(rec_magic(&rec) == NTDB_USED_MAGIC);
 	}
 	ok1(tap_log_messages == 0);
 	return exit_status();

lib/ntdb/test/run-001-fls.c

@@ -1,4 +1,4 @@
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 
 static unsigned int dumb_fls(uint64_t num)

lib/ntdb/test/run-01-new_database.c

@@ -1,5 +1,5 @@
 #include <ccan/failtest/failtest_override.h>
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <ccan/failtest/failtest.h>
 #include "logging.h"
@@ -8,25 +8,25 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
 
 	failtest_init(argc, argv);
 	failtest_hook = block_repeat_failures;
 	failtest_exit_check = exit_check_log;
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 3);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-new_database.tdb", flags[i],
+		ntdb = ntdb_open("run-new_database.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			failtest_exit(exit_status());
 
 		failtest_suppress = true;
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 		failtest_suppress = false;
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 		if (!ok1(tap_log_messages == 0))
 			break;
 	}

lib/ntdb/test/run-02-expand.c

@@ -0,0 +1,62 @@
+#include <ccan/failtest/failtest_override.h>
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include <ccan/failtest/failtest.h>
+#include "logging.h"
+#include "failtest_helper.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	uint64_t val;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 11 + 1);
+
+	failtest_init(argc, argv);
+	failtest_hook = block_repeat_failures;
+	failtest_exit_check = exit_check_log;
+
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		failtest_suppress = true;
+		ntdb = ntdb_open("run-expand.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		if (!ok1(ntdb))
+			break;
+
+		val = ntdb->file->map_size;
+		/* Need some hash lock for expand. */
+		ok1(ntdb_lock_hashes(ntdb, 0, 1, F_WRLCK, NTDB_LOCK_WAIT) == 0);
+		failtest_suppress = false;
+		if (!ok1(ntdb_expand(ntdb, 1) == 0)) {
+			failtest_suppress = true;
+			ntdb_close(ntdb);
+			break;
+		}
+		failtest_suppress = true;
+
+		ok1(ntdb->file->map_size >= val + 1 * NTDB_EXTENSION_FACTOR);
+		ok1(ntdb_unlock_hashes(ntdb, 0, 1, F_WRLCK) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		val = ntdb->file->map_size;
+		ok1(ntdb_lock_hashes(ntdb, 0, 1, F_WRLCK, NTDB_LOCK_WAIT) == 0);
+		failtest_suppress = false;
+		if (!ok1(ntdb_expand(ntdb, 1024) == 0)) {
+			failtest_suppress = true;
+			ntdb_close(ntdb);
+			break;
+		}
+		failtest_suppress = true;
+		ok1(ntdb_unlock_hashes(ntdb, 0, 1, F_WRLCK) == 0);
+		ok1(ntdb->file->map_size >= val + 1024 * NTDB_EXTENSION_FACTOR);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	failtest_exit(exit_status());
+}

lib/ntdb/test/run-03-coalesce.c

@@ -0,0 +1,178 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+#include "layout.h"
+
+static ntdb_len_t free_record_length(struct ntdb_context *ntdb, ntdb_off_t off)
+{
+	struct ntdb_free_record f;
+	enum NTDB_ERROR ecode;
+
+	ecode = ntdb_read_convert(ntdb, off, &f, sizeof(f));
+	if (ecode != NTDB_SUCCESS)
+		return ecode;
+	if (frec_magic(&f) != NTDB_FREE_MAGIC)
+		return NTDB_ERR_CORRUPT;
+	return frec_len(&f);
+}
+
+int main(int argc, char *argv[])
+{
+	ntdb_off_t b_off, test;
+	struct ntdb_context *ntdb;
+	struct ntdb_layout *layout;
+	NTDB_DATA data, key;
+	ntdb_len_t len;
+
+	/* FIXME: Test NTDB_CONVERT */
+	/* FIXME: Test lock order fail. */
+
+	plan_tests(42);
+	data = ntdb_mkdata("world", 5);
+	key = ntdb_mkdata("hello", 5);
+
+	/* No coalescing can be done due to EOF */
+	layout = new_ntdb_layout();
+	ntdb_layout_add_freetable(layout);
+	len = 1024;
+	ntdb_layout_add_free(layout, len, 0);
+	ntdb_layout_write(layout, free, &tap_log_attr, "run-03-coalesce.ntdb");
+	/* NOMMAP is for lockcheck. */
+	ntdb = ntdb_open("run-03-coalesce.ntdb", NTDB_NOMMAP, O_RDWR, 0,
+		       &tap_log_attr);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off) == len);
+
+	/* Figure out which bucket free entry is. */
+	b_off = bucket_off(ntdb->ftable_off, size_to_bucket(len));
+	/* Lock and fail to coalesce. */
+	ok1(ntdb_lock_free_bucket(ntdb, b_off, NTDB_LOCK_WAIT) == 0);
+	test = layout->elem[1].base.off;
+	ok1(coalesce(ntdb, layout->elem[1].base.off, b_off, len, &test)
+	    == 0);
+	ntdb_unlock_free_bucket(ntdb, b_off);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off) == len);
+	ok1(test == layout->elem[1].base.off);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+	ntdb_close(ntdb);
+	ntdb_layout_free(layout);
+
+	/* No coalescing can be done due to used record */
+	layout = new_ntdb_layout();
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_free(layout, 1024, 0);
+	ntdb_layout_add_used(layout, key, data, 6);
+	ntdb_layout_write(layout, free, &tap_log_attr, "run-03-coalesce.ntdb");
+	/* NOMMAP is for lockcheck. */
+	ntdb = ntdb_open("run-03-coalesce.ntdb", NTDB_NOMMAP, O_RDWR, 0,
+		       &tap_log_attr);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off) == 1024);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Figure out which bucket free entry is. */
+	b_off = bucket_off(ntdb->ftable_off, size_to_bucket(1024));
+	/* Lock and fail to coalesce. */
+	ok1(ntdb_lock_free_bucket(ntdb, b_off, NTDB_LOCK_WAIT) == 0);
+	test = layout->elem[1].base.off;
+	ok1(coalesce(ntdb, layout->elem[1].base.off, b_off, 1024, &test)
+	    == 0);
+	ntdb_unlock_free_bucket(ntdb, b_off);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off) == 1024);
+	ok1(test == layout->elem[1].base.off);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+	ntdb_close(ntdb);
+	ntdb_layout_free(layout);
+
+	/* Coalescing can be done due to two free records, then EOF */
+	layout = new_ntdb_layout();
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_free(layout, 1024, 0);
+	ntdb_layout_add_free(layout, 2048, 0);
+	ntdb_layout_write(layout, free, &tap_log_attr, "run-03-coalesce.ntdb");
+	/* NOMMAP is for lockcheck. */
+	ntdb = ntdb_open("run-03-coalesce.ntdb", NTDB_NOMMAP, O_RDWR, 0,
+		       &tap_log_attr);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off) == 1024);
+	ok1(free_record_length(ntdb, layout->elem[2].base.off) == 2048);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Figure out which bucket (first) free entry is. */
+	b_off = bucket_off(ntdb->ftable_off, size_to_bucket(1024));
+	/* Lock and coalesce. */
+	ok1(ntdb_lock_free_bucket(ntdb, b_off, NTDB_LOCK_WAIT) == 0);
+	test = layout->elem[2].base.off;
+	ok1(coalesce(ntdb, layout->elem[1].base.off, b_off, 1024, &test)
+	    == 1024 + sizeof(struct ntdb_used_record) + 2048);
+	/* Should tell us it's erased this one... */
+	ok1(test == NTDB_ERR_NOEXIST);
+	ok1(ntdb->file->allrecord_lock.count == 0 && ntdb->file->num_lockrecs == 0);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off)
+	    == 1024 + sizeof(struct ntdb_used_record) + 2048);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+	ntdb_close(ntdb);
+	ntdb_layout_free(layout);
+
+	/* Coalescing can be done due to two free records, then data */
+	layout = new_ntdb_layout();
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_free(layout, 1024, 0);
+	ntdb_layout_add_free(layout, 512, 0);
+	ntdb_layout_add_used(layout, key, data, 6);
+	ntdb_layout_write(layout, free, &tap_log_attr, "run-03-coalesce.ntdb");
+	/* NOMMAP is for lockcheck. */
+	ntdb = ntdb_open("run-03-coalesce.ntdb", NTDB_NOMMAP, O_RDWR, 0,
+		       &tap_log_attr);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off) == 1024);
+	ok1(free_record_length(ntdb, layout->elem[2].base.off) == 512);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Figure out which bucket free entry is. */
+	b_off = bucket_off(ntdb->ftable_off, size_to_bucket(1024));
+	/* Lock and coalesce. */
+	ok1(ntdb_lock_free_bucket(ntdb, b_off, NTDB_LOCK_WAIT) == 0);
+	test = layout->elem[2].base.off;
+	ok1(coalesce(ntdb, layout->elem[1].base.off, b_off, 1024, &test)
+	    == 1024 + sizeof(struct ntdb_used_record) + 512);
+	ok1(ntdb->file->allrecord_lock.count == 0 && ntdb->file->num_lockrecs == 0);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off)
+	    == 1024 + sizeof(struct ntdb_used_record) + 512);
+	ok1(test == NTDB_ERR_NOEXIST);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+	ntdb_close(ntdb);
+	ntdb_layout_free(layout);
+
+	/* Coalescing can be done due to three free records, then EOF */
+	layout = new_ntdb_layout();
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_free(layout, 1024, 0);
+	ntdb_layout_add_free(layout, 512, 0);
+	ntdb_layout_add_free(layout, 256, 0);
+	ntdb_layout_write(layout, free, &tap_log_attr, "run-03-coalesce.ntdb");
+	/* NOMMAP is for lockcheck. */
+	ntdb = ntdb_open("run-03-coalesce.ntdb", NTDB_NOMMAP, O_RDWR, 0,
+		       &tap_log_attr);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off) == 1024);
+	ok1(free_record_length(ntdb, layout->elem[2].base.off) == 512);
+	ok1(free_record_length(ntdb, layout->elem[3].base.off) == 256);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	/* Figure out which bucket free entry is. */
+	b_off = bucket_off(ntdb->ftable_off, size_to_bucket(1024));
+	/* Lock and coalesce. */
+	ok1(ntdb_lock_free_bucket(ntdb, b_off, NTDB_LOCK_WAIT) == 0);
+	test = layout->elem[2].base.off;
+	ok1(coalesce(ntdb, layout->elem[1].base.off, b_off, 1024, &test)
+	    == 1024 + sizeof(struct ntdb_used_record) + 512
+	    + sizeof(struct ntdb_used_record) + 256);
+	ok1(ntdb->file->allrecord_lock.count == 0
+	    && ntdb->file->num_lockrecs == 0);
+	ok1(free_record_length(ntdb, layout->elem[1].base.off)
+	    == 1024 + sizeof(struct ntdb_used_record) + 512
+	    + sizeof(struct ntdb_used_record) + 256);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+	ntdb_close(ntdb);
+	ntdb_layout_free(layout);
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/run-04-basichash.c

@@ -0,0 +1,260 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+/* We rig the hash so adjacent-numbered records always clash. */
+static uint64_t clash(const void *key, size_t len, uint64_t seed, void *priv)
+{
+	return ((uint64_t)*(const unsigned int *)key)
+		<< (64 - NTDB_TOPLEVEL_HASH_BITS - 1);
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i, j;
+	struct ntdb_context *ntdb;
+	unsigned int v;
+	struct ntdb_used_record rec;
+	NTDB_DATA key = { (unsigned char *)&v, sizeof(v) };
+	NTDB_DATA dbuf = { (unsigned char *)&v, sizeof(v) };
+	union ntdb_attribute hattr = { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
+						.fn = clash } };
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT,
+	};
+
+	hattr.base.next = &tap_log_attr;
+
+	plan_tests(sizeof(flags) / sizeof(flags[0])
+		   * (91 + (2 * ((1 << NTDB_HASH_GROUP_BITS) - 1))) + 1);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		struct hash_info h;
+		ntdb_off_t new_off, off, subhash;
+
+		ntdb = ntdb_open("run-04-basichash.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &hattr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		v = 0;
+		/* Should not find it. */
+		ok1(find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL) == 0);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in group 0, bucket 0. */
+		ok1(h.group_start == offsetof(struct ntdb_header, hashtable));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 0);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS);
+
+		/* Should have lock on bucket 0 */
+		ok1(h.hlock_start == 0);
+		ok1(h.hlock_range ==
+		    1ULL << (64-(NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS)));
+		ok1((ntdb->flags & NTDB_NOLOCK) || ntdb->file->num_lockrecs == 1);
+		ok1((ntdb->flags & NTDB_NOLOCK)
+		    || ntdb->file->lockrecs[0].off == NTDB_HASH_LOCK_START);
+		/* FIXME: Check lock length */
+
+		/* Allocate a new record. */
+		new_off = alloc(ntdb, key.dsize, dbuf.dsize, h.h,
+				NTDB_USED_MAGIC, false);
+		ok1(!NTDB_OFF_IS_ERR(new_off));
+
+		/* We should be able to add it now. */
+		ok1(add_to_hash(ntdb, &h, new_off) == 0);
+
+		/* Make sure we fill it in for later finding. */
+		off = new_off + sizeof(struct ntdb_used_record);
+		ok1(!ntdb->io->twrite(ntdb, off, key.dptr, key.dsize));
+		off += key.dsize;
+		ok1(!ntdb->io->twrite(ntdb, off, dbuf.dptr, dbuf.dsize));
+
+		/* We should be able to unlock that OK. */
+		ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+				      F_WRLCK) == 0);
+
+		/* Database should be consistent. */
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Now, this should give a successful lookup. */
+		ok1(find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL)
+		    == new_off);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in group 0, bucket 0. */
+		ok1(h.group_start == offsetof(struct ntdb_header, hashtable));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 0);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS);
+
+		/* Should have lock on bucket 0 */
+		ok1(h.hlock_start == 0);
+		ok1(h.hlock_range ==
+		    1ULL << (64-(NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS)));
+		ok1((ntdb->flags & NTDB_NOLOCK) || ntdb->file->num_lockrecs == 1);
+		ok1((ntdb->flags & NTDB_NOLOCK)
+		    || ntdb->file->lockrecs[0].off == NTDB_HASH_LOCK_START);
+		/* FIXME: Check lock length */
+
+		ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+				      F_WRLCK) == 0);
+
+		/* Database should be consistent. */
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Test expansion. */
+		v = 1;
+		ok1(find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL) == 0);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in group 0, bucket 1. */
+		ok1(h.group_start == offsetof(struct ntdb_header, hashtable));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 1);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS);
+
+		/* Should have lock on bucket 0 */
+		ok1(h.hlock_start == 0);
+		ok1(h.hlock_range ==
+		    1ULL << (64-(NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS)));
+		ok1((ntdb->flags & NTDB_NOLOCK) || ntdb->file->num_lockrecs == 1);
+		ok1((ntdb->flags & NTDB_NOLOCK)
+		    || ntdb->file->lockrecs[0].off == NTDB_HASH_LOCK_START);
+		/* FIXME: Check lock length */
+
+		/* Make it expand 0'th bucket. */
+		ok1(expand_group(ntdb, &h) == 0);
+		/* First one should be subhash, next should be empty. */
+		ok1(is_subhash(h.group[0]));
+		subhash = (h.group[0] & NTDB_OFF_MASK);
+		for (j = 1; j < (1 << NTDB_HASH_GROUP_BITS); j++)
+			ok1(h.group[j] == 0);
+
+		ok1(ntdb_write_convert(ntdb, h.group_start,
+				      h.group, sizeof(h.group)) == 0);
+		ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+				      F_WRLCK) == 0);
+
+		/* Should be happy with expansion. */
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Should be able to find it. */
+		v = 0;
+		ok1(find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL)
+		    == new_off);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in expanded group 0, bucket 0. */
+		ok1(h.group_start == subhash + sizeof(struct ntdb_used_record));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 0);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS
+		    + NTDB_SUBLEVEL_HASH_BITS);
+
+		/* Should have lock on bucket 0 */
+		ok1(h.hlock_start == 0);
+		ok1(h.hlock_range ==
+		    1ULL << (64-(NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS)));
+		ok1((ntdb->flags & NTDB_NOLOCK) || ntdb->file->num_lockrecs == 1);
+		ok1((ntdb->flags & NTDB_NOLOCK)
+		    || ntdb->file->lockrecs[0].off == NTDB_HASH_LOCK_START);
+		/* FIXME: Check lock length */
+
+		/* Simple delete should work. */
+		ok1(delete_from_hash(ntdb, &h) == 0);
+		ok1(add_free_record(ntdb, new_off,
+				    sizeof(struct ntdb_used_record)
+				    + rec_key_length(&rec)
+				    + rec_data_length(&rec)
+				    + rec_extra_padding(&rec),
+				    NTDB_LOCK_NOWAIT, false) == 0);
+		ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+				      F_WRLCK) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Test second-level expansion: should expand 0th bucket. */
+		v = 0;
+		ok1(find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL) == 0);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in group 0, bucket 0. */
+		ok1(h.group_start == subhash + sizeof(struct ntdb_used_record));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 0);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS+NTDB_SUBLEVEL_HASH_BITS);
+
+		/* Should have lock on bucket 0 */
+		ok1(h.hlock_start == 0);
+		ok1(h.hlock_range ==
+		    1ULL << (64-(NTDB_TOPLEVEL_HASH_BITS-NTDB_HASH_GROUP_BITS)));
+		ok1((ntdb->flags & NTDB_NOLOCK) || ntdb->file->num_lockrecs == 1);
+		ok1((ntdb->flags & NTDB_NOLOCK)
+		    || ntdb->file->lockrecs[0].off == NTDB_HASH_LOCK_START);
+		/* FIXME: Check lock length */
+
+		ok1(expand_group(ntdb, &h) == 0);
+		/* First one should be subhash, next should be empty. */
+		ok1(is_subhash(h.group[0]));
+		subhash = (h.group[0] & NTDB_OFF_MASK);
+		for (j = 1; j < (1 << NTDB_HASH_GROUP_BITS); j++)
+			ok1(h.group[j] == 0);
+		ok1(ntdb_write_convert(ntdb, h.group_start,
+				      h.group, sizeof(h.group)) == 0);
+		ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+				      F_WRLCK) == 0);
+
+		/* Should be happy with expansion. */
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		ok1(find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL) == 0);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in group 0, bucket 0. */
+		ok1(h.group_start == subhash + sizeof(struct ntdb_used_record));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 0);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS
+		    + NTDB_SUBLEVEL_HASH_BITS * 2);
+
+		/* We should be able to add it now. */
+		/* Allocate a new record. */
+		new_off = alloc(ntdb, key.dsize, dbuf.dsize, h.h,
+				NTDB_USED_MAGIC, false);
+		ok1(!NTDB_OFF_IS_ERR(new_off));
+		ok1(add_to_hash(ntdb, &h, new_off) == 0);
+
+		/* Make sure we fill it in for later finding. */
+		off = new_off + sizeof(struct ntdb_used_record);
+		ok1(!ntdb->io->twrite(ntdb, off, key.dptr, key.dsize));
+		off += key.dsize;
+		ok1(!ntdb->io->twrite(ntdb, off, dbuf.dptr, dbuf.dsize));
+
+		/* We should be able to unlock that OK. */
+		ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+				      F_WRLCK) == 0);
+
+		/* Database should be consistent. */
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Should be able to find it. */
+		v = 0;
+		ok1(find_and_lock(ntdb, key, F_WRLCK, &h, &rec, NULL)
+		    == new_off);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in expanded group 0, bucket 0. */
+		ok1(h.group_start == subhash + sizeof(struct ntdb_used_record));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 0);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS
+		    + NTDB_SUBLEVEL_HASH_BITS * 2);
+
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/run-05-readonly-open.c

@@ -1,5 +1,5 @@
 #include <ccan/failtest/failtest_override.h>
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <ccan/failtest/failtest.h>
 #include "logging.h"
@@ -8,55 +8,55 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data = tdb_mkdata("data", 4), d;
-	union tdb_attribute seed_attr;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4), d;
+	union ntdb_attribute seed_attr;
 	unsigned int msgs = 0;
 
 	failtest_init(argc, argv);
 	failtest_hook = block_repeat_failures;
 	failtest_exit_check = exit_check_log;
 
-	seed_attr.base.attr = TDB_ATTRIBUTE_SEED;
+	seed_attr.base.attr = NTDB_ATTRIBUTE_SEED;
 	seed_attr.base.next = &tap_log_attr;
 	seed_attr.seed.seed = 0;
 
 	failtest_suppress = true;
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 11);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-05-readonly-open.tdb", flags[i],
+		ntdb = ntdb_open("run-05-readonly-open.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600,
 			       &seed_attr);
-		ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
-		tdb_close(tdb);
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+		ntdb_close(ntdb);
 
 		failtest_suppress = false;
-		tdb = tdb_open("run-05-readonly-open.tdb", flags[i],
+		ntdb = ntdb_open("run-05-readonly-open.ntdb", flags[i],
 			       O_RDONLY, 0600, &tap_log_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			break;
 		ok1(tap_log_messages == msgs);
 		/* Fetch should succeed, stores should fail. */
-		if (!ok1(tdb_fetch(tdb, key, &d) == 0))
+		if (!ok1(ntdb_fetch(ntdb, key, &d) == 0))
 			goto fail;
-		ok1(tdb_deq(d, data));
+		ok1(ntdb_deq(d, data));
 		free(d.dptr);
-		if (!ok1(tdb_store(tdb, key, data, TDB_MODIFY)
-			 == TDB_ERR_RDONLY))
+		if (!ok1(ntdb_store(ntdb, key, data, NTDB_MODIFY)
+			 == NTDB_ERR_RDONLY))
 			goto fail;
 		ok1(tap_log_messages == ++msgs);
-		if (!ok1(tdb_store(tdb, key, data, TDB_INSERT)
-			 == TDB_ERR_RDONLY))
+		if (!ok1(ntdb_store(ntdb, key, data, NTDB_INSERT)
+			 == NTDB_ERR_RDONLY))
 			goto fail;
 		ok1(tap_log_messages == ++msgs);
 		failtest_suppress = true;
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
-		tdb_close(tdb);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ntdb_close(ntdb);
 		ok1(tap_log_messages == msgs);
-		/* SIGH: failtest bug, it doesn't save the tdb file because
+		/* SIGH: failtest bug, it doesn't save the ntdb file because
 		 * we have it read-only.  If we go around again, it gets
 		 * changed underneath us and things get screwy. */
 		if (failtest_has_failed())
@@ -66,6 +66,6 @@ int main(int argc, char *argv[])
 
 fail:
 	failtest_suppress = true;
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 	failtest_exit(exit_status());
 }

lib/ntdb/test/run-10-simple-store.c

@@ -1,5 +1,5 @@
 #include <ccan/failtest/failtest_override.h>
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <ccan/failtest/failtest.h>
 #include "logging.h"
@@ -8,12 +8,12 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data = tdb_mkdata("data", 4);
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
 
 	failtest_init(argc, argv);
 	failtest_hook = block_repeat_failures;
@@ -22,37 +22,37 @@ int main(int argc, char *argv[])
 	failtest_suppress = true;
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 7 + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-10-simple-store.tdb", flags[i],
+		ntdb = ntdb_open("run-10-simple-store.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			break;
 		/* Modify should fail. */
 		failtest_suppress = false;
-		if (!ok1(tdb_store(tdb, key, data, TDB_MODIFY)
-			 == TDB_ERR_NOEXIST))
+		if (!ok1(ntdb_store(ntdb, key, data, NTDB_MODIFY)
+			 == NTDB_ERR_NOEXIST))
 			goto fail;
 		failtest_suppress = true;
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 		/* Insert should succeed. */
 		failtest_suppress = false;
-		if (!ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0))
+		if (!ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0))
 			goto fail;
 		failtest_suppress = true;
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 		/* Second insert should fail. */
 		failtest_suppress = false;
-		if (!ok1(tdb_store(tdb, key, data, TDB_INSERT)
-			 == TDB_ERR_EXISTS))
+		if (!ok1(ntdb_store(ntdb, key, data, NTDB_INSERT)
+			 == NTDB_ERR_EXISTS))
 			goto fail;
 		failtest_suppress = true;
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
-		tdb_close(tdb);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ntdb_close(ntdb);
 	}
 	ok1(tap_log_messages == 0);
 	failtest_exit(exit_status());
 
 fail:
 	failtest_suppress = true;
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 	failtest_exit(exit_status());
 }

lib/ntdb/test/run-11-simple-fetch.c

@@ -1,5 +1,5 @@
 #include <ccan/failtest/failtest_override.h>
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <ccan/failtest/failtest.h>
 #include "logging.h"
@@ -8,12 +8,12 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data = tdb_mkdata("data", 4);
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
 
 	failtest_init(argc, argv);
 	failtest_hook = block_repeat_failures;
@@ -22,30 +22,30 @@ int main(int argc, char *argv[])
 	failtest_suppress = true;
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 8 + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-11-simple-fetch.tdb", flags[i],
+		ntdb = ntdb_open("run-11-simple-fetch.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (tdb) {
-			struct tdb_data d = { NULL, 0 }; /* Bogus GCC warning */
+		ok1(ntdb);
+		if (ntdb) {
+			NTDB_DATA d = { NULL, 0 }; /* Bogus GCC warning */
 
 			/* fetch should fail. */
 			failtest_suppress = false;
-			if (!ok1(tdb_fetch(tdb, key, &d) == TDB_ERR_NOEXIST))
+			if (!ok1(ntdb_fetch(ntdb, key, &d) == NTDB_ERR_NOEXIST))
 				goto fail;
 			failtest_suppress = true;
-			ok1(tdb_check(tdb, NULL, NULL) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 			/* Insert should succeed. */
-			ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
-			ok1(tdb_check(tdb, NULL, NULL) == 0);
+			ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 			/* Fetch should now work. */
 			failtest_suppress = false;
-			if (!ok1(tdb_fetch(tdb, key, &d) == TDB_SUCCESS))
+			if (!ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS))
 				goto fail;
 			failtest_suppress = true;
-			ok1(tdb_deq(d, data));
+			ok1(ntdb_deq(d, data));
 			free(d.dptr);
-			ok1(tdb_check(tdb, NULL, NULL) == 0);
-			tdb_close(tdb);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+			ntdb_close(ntdb);
 		}
 	}
 	ok1(tap_log_messages == 0);
@@ -53,6 +53,6 @@ int main(int argc, char *argv[])
 
 fail:
 	failtest_suppress = true;
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 	failtest_exit(exit_status());
 }

lib/ntdb/test/run-12-check.c

@@ -1,6 +1,6 @@
 #include "private.h"
 #include <ccan/failtest/failtest_override.h>
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <ccan/failtest/failtest.h>
 #include "logging.h"
@@ -9,12 +9,12 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_INTERNAL,
-			TDB_INTERNAL|TDB_CONVERT,
-			TDB_CONVERT };
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data = tdb_mkdata("data", 4);
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL,
+			NTDB_INTERNAL|NTDB_CONVERT,
+			NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
 
 	failtest_init(argc, argv);
 	failtest_hook = block_repeat_failures;
@@ -23,24 +23,24 @@ int main(int argc, char *argv[])
 	failtest_suppress = true;
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 3 + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-12-check.tdb", flags[i],
+		ntdb = ntdb_open("run-12-check.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
+		ok1(ntdb);
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
 
-		/* This is what we really want to test: tdb_check(). */
+		/* This is what we really want to test: ntdb_check(). */
 		failtest_suppress = false;
-		if (!ok1(tdb_check(tdb, NULL, NULL) == 0))
+		if (!ok1(ntdb_check(ntdb, NULL, NULL) == 0))
 			goto fail;
 		failtest_suppress = true;
 
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 	ok1(tap_log_messages == 0);
 	failtest_exit(exit_status());
 
 fail:
 	failtest_suppress = true;
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 	failtest_exit(exit_status());
 }

lib/ntdb/test/run-15-append.c

@@ -1,4 +1,4 @@
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <ccan/ilog/ilog.h>
 #include "logging.h"
@@ -6,30 +6,30 @@
 #define MAX_SIZE 13100
 #define SIZE_STEP 131
 
-static tdb_off_t tdb_offset(struct tdb_context *tdb, struct tdb_data key)
+static ntdb_off_t ntdb_offset(struct ntdb_context *ntdb, NTDB_DATA key)
 {
-	tdb_off_t off;
-	struct tdb_used_record urec;
+	ntdb_off_t off;
+	struct ntdb_used_record urec;
 	struct hash_info h;
 
-	off = find_and_lock(tdb, key, F_RDLCK, &h, &urec, NULL);
-	if (TDB_OFF_IS_ERR(off))
+	off = find_and_lock(ntdb, key, F_RDLCK, &h, &urec, NULL);
+	if (NTDB_OFF_IS_ERR(off))
 		return 0;
-	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_RDLCK);
 	return off;
 }
 
 int main(int argc, char *argv[])
 {
 	unsigned int i, j, moves;
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 	unsigned char *buffer;
-	tdb_off_t oldoff = 0, newoff;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = tdb_mkdata("key", 3);
-	struct tdb_data data;
+	ntdb_off_t oldoff = 0, newoff;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data;
 
 	buffer = malloc(MAX_SIZE);
 	for (i = 0; i < MAX_SIZE; i++)
@@ -39,89 +39,89 @@ int main(int argc, char *argv[])
 		   * ((3 + MAX_SIZE/SIZE_STEP * 5) * 2 + 7)
 		   + 1);
 
-	/* Using tdb_store. */
+	/* Using ntdb_store. */
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-append.tdb", flags[i],
+		ntdb = ntdb_open("run-append.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		moves = 0;
 		for (j = 0; j < MAX_SIZE; j += SIZE_STEP) {
 			data.dptr = buffer;
 			data.dsize = j;
-			ok1(tdb_store(tdb, key, data, TDB_REPLACE) == 0);
-			ok1(tdb_check(tdb, NULL, NULL) == 0);
-			ok1(tdb_fetch(tdb, key, &data) == TDB_SUCCESS);
+			ok1(ntdb_store(ntdb, key, data, NTDB_REPLACE) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+			ok1(ntdb_fetch(ntdb, key, &data) == NTDB_SUCCESS);
 			ok1(data.dsize == j);
 			ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
 			free(data.dptr);
-			newoff = tdb_offset(tdb, key);
+			newoff = ntdb_offset(ntdb, key);
 			if (newoff != oldoff)
 				moves++;
 			oldoff = newoff;
 		}
-		ok1(!tdb->file || (tdb->file->allrecord_lock.count == 0
-				   && tdb->file->num_lockrecs == 0));
+		ok1(!ntdb->file || (ntdb->file->allrecord_lock.count == 0
+				   && ntdb->file->num_lockrecs == 0));
 		/* We should increase by 50% each time... */
 		ok(moves <= ilog64(j / SIZE_STEP)*2,
 		   "Moved %u times", moves);
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
-	/* Using tdb_append. */
+	/* Using ntdb_append. */
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
 		size_t prev_len = 0;
-		tdb = tdb_open("run-append.tdb", flags[i],
+		ntdb = ntdb_open("run-append.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		moves = 0;
 		for (j = 0; j < MAX_SIZE; j += SIZE_STEP) {
 			data.dptr = buffer + prev_len;
 			data.dsize = j - prev_len;
-			ok1(tdb_append(tdb, key, data) == 0);
-			ok1(tdb_check(tdb, NULL, NULL) == 0);
-			ok1(tdb_fetch(tdb, key, &data) == TDB_SUCCESS);
+			ok1(ntdb_append(ntdb, key, data) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+			ok1(ntdb_fetch(ntdb, key, &data) == NTDB_SUCCESS);
 			ok1(data.dsize == j);
 			ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
 			free(data.dptr);
 			prev_len = data.dsize;
-			newoff = tdb_offset(tdb, key);
+			newoff = ntdb_offset(ntdb, key);
 			if (newoff != oldoff)
 				moves++;
 			oldoff = newoff;
 		}
-		ok1(!tdb->file || (tdb->file->allrecord_lock.count == 0
-				   && tdb->file->num_lockrecs == 0));
+		ok1(!ntdb->file || (ntdb->file->allrecord_lock.count == 0
+				   && ntdb->file->num_lockrecs == 0));
 		/* We should increase by 50% each time... */
 		ok(moves <= ilog64(j / SIZE_STEP)*2,
 		   "Moved %u times", moves);
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-append.tdb", flags[i],
+		ntdb = ntdb_open("run-append.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		/* Huge initial store. */
 		data.dptr = buffer;
 		data.dsize = MAX_SIZE;
-		ok1(tdb_append(tdb, key, data) == 0);
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
-		ok1(tdb_fetch(tdb, key, &data) == TDB_SUCCESS);
+		ok1(ntdb_append(ntdb, key, data) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ok1(ntdb_fetch(ntdb, key, &data) == NTDB_SUCCESS);
 		ok1(data.dsize == MAX_SIZE);
 		ok1(memcmp(data.dptr, buffer, data.dsize) == 0);
 		free(data.dptr);
-		ok1(!tdb->file || (tdb->file->allrecord_lock.count == 0
-				   && tdb->file->num_lockrecs == 0));
-		tdb_close(tdb);
+		ok1(!ntdb->file || (ntdb->file->allrecord_lock.count == 0
+				   && ntdb->file->num_lockrecs == 0));
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/run-20-growhash.c

@@ -0,0 +1,137 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+static uint64_t myhash(const void *key, size_t len, uint64_t seed, void *priv)
+{
+	return *(const uint64_t *)key;
+}
+
+static void add_bits(uint64_t *val, unsigned new, unsigned new_bits,
+		     unsigned *done)
+{
+	*done += new_bits;
+	*val |= ((uint64_t)new << (64 - *done));
+}
+
+static uint64_t make_key(unsigned topgroup, unsigned topbucket,
+			 unsigned subgroup1, unsigned subbucket1,
+			 unsigned subgroup2, unsigned subbucket2)
+{
+	uint64_t key = 0;
+	unsigned done = 0;
+
+	add_bits(&key, topgroup, NTDB_TOPLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS,
+		 &done);
+	add_bits(&key, topbucket, NTDB_HASH_GROUP_BITS, &done);
+	add_bits(&key, subgroup1, NTDB_SUBLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS,
+		 &done);
+	add_bits(&key, subbucket1, NTDB_HASH_GROUP_BITS, &done);
+	add_bits(&key, subgroup2, NTDB_SUBLEVEL_HASH_BITS - NTDB_HASH_GROUP_BITS,
+		 &done);
+	add_bits(&key, subbucket2, NTDB_HASH_GROUP_BITS, &done);
+	return key;
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i, j;
+	struct ntdb_context *ntdb;
+	uint64_t kdata;
+	struct ntdb_used_record rec;
+	NTDB_DATA key = { (unsigned char *)&kdata, sizeof(kdata) };
+	NTDB_DATA dbuf = { (unsigned char *)&kdata, sizeof(kdata) };
+	union ntdb_attribute hattr = { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
+						.fn = myhash } };
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT,
+	};
+
+	hattr.base.next = &tap_log_attr;
+
+	plan_tests(sizeof(flags) / sizeof(flags[0])
+		   * (9 + (20 + 2 * ((1 << NTDB_HASH_GROUP_BITS) - 2))
+		      * (1 << NTDB_HASH_GROUP_BITS)) + 1);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		struct hash_info h;
+
+		ntdb = ntdb_open("run-20-growhash.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &hattr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		/* Fill a group. */
+		for (j = 0; j < (1 << NTDB_HASH_GROUP_BITS); j++) {
+			kdata = make_key(0, j, 0, 0, 0, 0);
+			ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+		}
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Check first still exists. */
+		kdata = make_key(0, 0, 0, 0, 0, 0);
+		ok1(find_and_lock(ntdb, key, F_RDLCK, &h, &rec, NULL) != 0);
+		/* Should have created correct hash. */
+		ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+		/* Should have located space in group 0, bucket 0. */
+		ok1(h.group_start == offsetof(struct ntdb_header, hashtable));
+		ok1(h.home_bucket == 0);
+		ok1(h.found_bucket == 0);
+		ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS);
+		/* Entire group should be full! */
+		for (j = 0; j < (1 << NTDB_HASH_GROUP_BITS); j++)
+			ok1(h.group[j] != 0);
+
+		ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+				      F_RDLCK) == 0);
+
+		/* Now, add one more to each should expand (that) bucket. */
+		for (j = 0; j < (1 << NTDB_HASH_GROUP_BITS); j++) {
+			unsigned int k;
+			kdata = make_key(0, j, 0, 1, 0, 0);
+			ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+			ok1(find_and_lock(ntdb, key, F_RDLCK, &h, &rec, NULL));
+			/* Should have created correct hash. */
+			ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+			/* Should have moved to subhash */
+			ok1(h.group_start >= sizeof(struct ntdb_header));
+			ok1(h.home_bucket == 1);
+			ok1(h.found_bucket == 1);
+			ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS
+			    + NTDB_SUBLEVEL_HASH_BITS);
+			ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+					      F_RDLCK) == 0);
+
+			/* Keep adding, make it expand again. */
+			for (k = 2; k < (1 << NTDB_HASH_GROUP_BITS); k++) {
+				kdata = make_key(0, j, 0, k, 0, 0);
+				ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+				ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+			}
+
+			/* This should tip it over to sub-sub-hash. */
+			kdata = make_key(0, j, 0, 0, 0, 1);
+			ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+			ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+			ok1(find_and_lock(ntdb, key, F_RDLCK, &h, &rec, NULL));
+			/* Should have created correct hash. */
+			ok1(h.h == ntdb_hash(ntdb, key.dptr, key.dsize));
+			/* Should have moved to subhash */
+			ok1(h.group_start >= sizeof(struct ntdb_header));
+			ok1(h.home_bucket == 1);
+			ok1(h.found_bucket == 1);
+			ok1(h.hash_used == NTDB_TOPLEVEL_HASH_BITS
+			    + NTDB_SUBLEVEL_HASH_BITS + NTDB_SUBLEVEL_HASH_BITS);
+			ok1(ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range,
+					      F_RDLCK) == 0);
+		}
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/run-25-hashoverload.c

@@ -0,0 +1,113 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+static uint64_t badhash(const void *key, size_t len, uint64_t seed, void *priv)
+{
+	return 0;
+}
+
+static int trav(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *p)
+{
+	if (p)
+		return ntdb_delete(ntdb, key);
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i, j;
+	struct ntdb_context *ntdb;
+	NTDB_DATA key = { (unsigned char *)&j, sizeof(j) };
+	NTDB_DATA dbuf = { (unsigned char *)&j, sizeof(j) };
+	union ntdb_attribute hattr = { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
+						.fn = badhash } };
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT,
+	};
+
+	hattr.base.next = &tap_log_attr;
+
+	plan_tests(6883);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		NTDB_DATA d = { NULL, 0 }; /* Bogus GCC warning */
+
+		ntdb = ntdb_open("run-25-hashoverload.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &hattr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		/* Fill a group. */
+		for (j = 0; j < (1 << NTDB_HASH_GROUP_BITS); j++) {
+			ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+		}
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Now store one last value: should form chain. */
+		ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Check we can find them all. */
+		for (j = 0; j < (1 << NTDB_HASH_GROUP_BITS) + 1; j++) {
+			ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+			ok1(d.dsize == sizeof(j));
+			ok1(d.dptr != NULL);
+			ok1(d.dptr && memcmp(d.dptr, &j, d.dsize) == 0);
+			free(d.dptr);
+		}
+
+		/* Now add a *lot* more. */
+		for (j = (1 << NTDB_HASH_GROUP_BITS) + 1;
+		     j < (16 << NTDB_HASH_GROUP_BITS);
+		     j++) {
+			ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+			ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+			ok1(d.dsize == sizeof(j));
+			ok1(d.dptr != NULL);
+			ok1(d.dptr && memcmp(d.dptr, &j, d.dsize) == 0);
+			free(d.dptr);
+		}
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Traverse through them. */
+		ok1(ntdb_traverse(ntdb, trav, NULL) == j);
+
+		/* Empty the first chain-worth. */
+		for (j = 0; j < (1 << NTDB_HASH_GROUP_BITS); j++)
+			ok1(ntdb_delete(ntdb, key) == 0);
+
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		for (j = (1 << NTDB_HASH_GROUP_BITS);
+		     j < (16 << NTDB_HASH_GROUP_BITS);
+		     j++) {
+			ok1(ntdb_fetch(ntdb, key, &d) == NTDB_SUCCESS);
+			ok1(d.dsize == sizeof(j));
+			ok1(d.dptr != NULL);
+			ok1(d.dptr && memcmp(d.dptr, &j, d.dsize) == 0);
+			free(d.dptr);
+		}
+
+		/* Traverse through them. */
+		ok1(ntdb_traverse(ntdb, trav, NULL)
+		    == (15 << NTDB_HASH_GROUP_BITS));
+
+		/* Re-add */
+		for (j = 0; j < (1 << NTDB_HASH_GROUP_BITS); j++) {
+			ok1(ntdb_store(ntdb, key, dbuf, NTDB_INSERT) == 0);
+		}
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+		/* Now try deleting as we go. */
+		ok1(ntdb_traverse(ntdb, trav, trav)
+		    == (16 << NTDB_HASH_GROUP_BITS));
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ok1(ntdb_traverse(ntdb, trav, NULL) == 0);
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/run-30-exhaust-before-expand.c

@@ -0,0 +1,71 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+static bool empty_freetable(struct ntdb_context *ntdb)
+{
+	struct ntdb_freetable ftab;
+	unsigned int i;
+
+	/* Now, free table should be completely exhausted in zone 0 */
+	if (ntdb_read_convert(ntdb, ntdb->ftable_off, &ftab, sizeof(ftab)) != 0)
+		abort();
+
+	for (i = 0; i < sizeof(ftab.buckets)/sizeof(ftab.buckets[0]); i++) {
+		if (ftab.buckets[i])
+			return false;
+	}
+	return true;
+}
+
+
+int main(int argc, char *argv[])
+{
+	unsigned int i, j;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 9 + 1);
+
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		NTDB_DATA k;
+		uint64_t size;
+		bool was_empty = false;
+
+		k.dptr = (void *)&j;
+		k.dsize = sizeof(j);
+
+		ntdb = ntdb_open("run-30-exhaust-before-expand.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		ok1(empty_freetable(ntdb));
+		/* Need some hash lock for expand. */
+		ok1(ntdb_lock_hashes(ntdb, 0, 1, F_WRLCK, NTDB_LOCK_WAIT) == 0);
+		/* Create some free space. */
+		ok1(ntdb_expand(ntdb, 1) == 0);
+		ok1(ntdb_unlock_hashes(ntdb, 0, 1, F_WRLCK) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ok1(!empty_freetable(ntdb));
+
+		size = ntdb->file->map_size;
+		/* Insert minimal-length records until we expand. */
+		for (j = 0; ntdb->file->map_size == size; j++) {
+			was_empty = empty_freetable(ntdb);
+			if (ntdb_store(ntdb, k, k, NTDB_INSERT) != 0)
+				err(1, "Failed to store record %i", j);
+		}
+
+		/* Would have been empty before expansion, but no longer. */
+		ok1(was_empty);
+		ok1(!empty_freetable(ntdb));
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/run-35-convert.c

@@ -1,6 +1,6 @@
 #include "private.h"
 #include <ccan/failtest/failtest_override.h>
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <ccan/failtest/failtest.h>
 #include "logging.h"
@@ -9,46 +9,46 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i, messages = 0;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
 
 	failtest_init(argc, argv);
 	failtest_hook = block_repeat_failures;
 	failtest_exit_check = exit_check_log;
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 4);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-35-convert.tdb", flags[i],
+		ntdb = ntdb_open("run-35-convert.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		if (!ok1(tdb))
+		if (!ok1(ntdb))
 			failtest_exit(exit_status());
 
-		tdb_close(tdb);
-		/* If we say TDB_CONVERT, it must be converted */
-		tdb = tdb_open("run-35-convert.tdb",
-			       flags[i]|TDB_CONVERT,
+		ntdb_close(ntdb);
+		/* If we say NTDB_CONVERT, it must be converted */
+		ntdb = ntdb_open("run-35-convert.ntdb",
+			       flags[i]|NTDB_CONVERT,
 			       O_RDWR, 0600, &tap_log_attr);
-		if (flags[i] & TDB_CONVERT) {
-			if (!tdb)
+		if (flags[i] & NTDB_CONVERT) {
+			if (!ntdb)
 				failtest_exit(exit_status());
-			ok1(tdb_get_flags(tdb) & TDB_CONVERT);
-			tdb_close(tdb);
+			ok1(ntdb_get_flags(ntdb) & NTDB_CONVERT);
+			ntdb_close(ntdb);
 		} else {
-			if (!ok1(!tdb && errno == EIO))
+			if (!ok1(!ntdb && errno == EIO))
 				failtest_exit(exit_status());
 			ok1(tap_log_messages == ++messages);
-			if (!ok1(log_last && strstr(log_last, "TDB_CONVERT")))
+			if (!ok1(log_last && strstr(log_last, "NTDB_CONVERT")))
 				failtest_exit(exit_status());
 		}
 
-		/* If don't say TDB_CONVERT, it *may* be converted */
-		tdb = tdb_open("run-35-convert.tdb",
-			       flags[i] & ~TDB_CONVERT,
+		/* If don't say NTDB_CONVERT, it *may* be converted */
+		ntdb = ntdb_open("run-35-convert.ntdb",
+			       flags[i] & ~NTDB_CONVERT,
 			       O_RDWR, 0600, &tap_log_attr);
-		if (!tdb)
+		if (!ntdb)
 			failtest_exit(exit_status());
-		ok1(tdb_get_flags(tdb) == flags[i]);
-		tdb_close(tdb);
+		ok1(ntdb_get_flags(ntdb) == flags[i]);
+		ntdb_close(ntdb);
 	}
 	failtest_exit(exit_status());
 }

lib/ntdb/test/run-50-multiple-freelists.c

@@ -0,0 +1,70 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+#include "layout.h"
+
+int main(int argc, char *argv[])
+{
+	ntdb_off_t off;
+	struct ntdb_context *ntdb;
+	struct ntdb_layout *layout;
+	NTDB_DATA key, data;
+	union ntdb_attribute seed;
+
+	/* This seed value previously tickled a layout.c bug. */
+	seed.base.attr = NTDB_ATTRIBUTE_SEED;
+	seed.seed.seed = 0xb1142bc054d035b4ULL;
+	seed.base.next = &tap_log_attr;
+
+	plan_tests(11);
+	key = ntdb_mkdata("Hello", 5);
+	data = ntdb_mkdata("world", 5);
+
+	/* Create a NTDB with three free tables. */
+	layout = new_ntdb_layout();
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_free(layout, 80, 0);
+	/* Used record prevent coalescing. */
+	ntdb_layout_add_used(layout, key, data, 6);
+	ntdb_layout_add_free(layout, 160, 1);
+	key.dsize--;
+	ntdb_layout_add_used(layout, key, data, 7);
+	ntdb_layout_add_free(layout, 320, 2);
+	key.dsize--;
+	ntdb_layout_add_used(layout, key, data, 8);
+	ntdb_layout_add_free(layout, 40, 0);
+	ntdb = ntdb_layout_get(layout, free, &seed);
+	ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+
+	off = get_free(ntdb, 0, 80 - sizeof(struct ntdb_used_record), 0,
+		       NTDB_USED_MAGIC, 0);
+	ok1(off == layout->elem[3].base.off);
+	ok1(ntdb->ftable_off == layout->elem[0].base.off);
+
+	off = get_free(ntdb, 0, 160 - sizeof(struct ntdb_used_record), 0,
+		       NTDB_USED_MAGIC, 0);
+	ok1(off == layout->elem[5].base.off);
+	ok1(ntdb->ftable_off == layout->elem[1].base.off);
+
+	off = get_free(ntdb, 0, 320 - sizeof(struct ntdb_used_record), 0,
+		       NTDB_USED_MAGIC, 0);
+	ok1(off == layout->elem[7].base.off);
+	ok1(ntdb->ftable_off == layout->elem[2].base.off);
+
+	off = get_free(ntdb, 0, 40 - sizeof(struct ntdb_used_record), 0,
+		       NTDB_USED_MAGIC, 0);
+	ok1(off == layout->elem[9].base.off);
+	ok1(ntdb->ftable_off == layout->elem[0].base.off);
+
+	/* Now we fail. */
+	off = get_free(ntdb, 0, 0, 1, NTDB_USED_MAGIC, 0);
+	ok1(off == 0);
+
+	ntdb_close(ntdb);
+	ntdb_layout_free(layout);
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/run-56-open-during-transaction.c

@@ -11,7 +11,7 @@ static int ftruncate_check(int fd, off_t length);
 #define fcntl fcntl_with_lockcheck
 #define ftruncate ftruncate_check
 
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include <stdlib.h>
 #include <stdbool.h>
@@ -22,7 +22,7 @@ static int ftruncate_check(int fd, off_t length);
 static struct agent *agent;
 static bool opened;
 static int errors = 0;
-#define TEST_DBNAME "run-56-open-during-transaction.tdb"
+#define TEST_DBNAME "run-56-open-during-transaction.ntdb"
 
 #undef write
 #undef pwrite
@@ -80,7 +80,7 @@ static void check_file_intact(int fd)
 	if (ret == SUCCESS) {
 		ret = external_agent_operation(agent, CLOSE, NULL);
 		if (ret != SUCCESS) {
-			diag("Agent failed to close tdb: %s",
+			diag("Agent failed to close ntdb: %s",
 			     agent_return_name(ret));
 			errors++;
 		}
@@ -127,11 +127,11 @@ static int ftruncate_check(int fd, off_t length)
 
 int main(int argc, char *argv[])
 {
-	const int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
+	const int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
 	int i;
-	struct tdb_context *tdb;
-	TDB_DATA key, data;
+	struct ntdb_context *ntdb;
+	NTDB_DATA key, data;
 
 	plan_tests(sizeof(flags)/sizeof(flags[0]) * 5);
 	agent = prepare_external_agent();
@@ -141,24 +141,24 @@ int main(int argc, char *argv[])
 	unlock_callback = after_unlock;
 	for (i = 0; i < sizeof(flags)/sizeof(flags[0]); i++) {
 		diag("Test with %s and %s\n",
-		     (flags[i] & TDB_CONVERT) ? "CONVERT" : "DEFAULT",
-		     (flags[i] & TDB_NOMMAP) ? "no mmap" : "mmap");
+		     (flags[i] & NTDB_CONVERT) ? "CONVERT" : "DEFAULT",
+		     (flags[i] & NTDB_NOMMAP) ? "no mmap" : "mmap");
 		unlink(TEST_DBNAME);
-		tdb = tdb_open(TEST_DBNAME, flags[i],
+		ntdb = ntdb_open(TEST_DBNAME, flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
+		ok1(ntdb);
 
 		opened = true;
-		ok1(tdb_transaction_start(tdb) == 0);
-		key = tdb_mkdata("hi", strlen("hi"));
-		data = tdb_mkdata("world", strlen("world"));
+		ok1(ntdb_transaction_start(ntdb) == 0);
+		key = ntdb_mkdata("hi", strlen("hi"));
+		data = ntdb_mkdata("world", strlen("world"));
 
-		ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
-		ok1(tdb_transaction_commit(tdb) == 0);
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+		ok1(ntdb_transaction_commit(ntdb) == 0);
 		ok(!errors, "We had %u open errors", errors);
 
 		opened = false;
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	return exit_status();

lib/ntdb/test/run-57-die-during-transaction.c

@@ -81,7 +81,7 @@ static void free_all(void)
 #define free free_noleak
 #define realloc realloc_noleak
 
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 
 #undef malloc
 #undef free
@@ -93,6 +93,7 @@ static void free_all(void)
 
 #include <stdbool.h>
 #include <stdarg.h>
+#include <ccan/err/err.h>
 #include <setjmp.h>
 #include "external-agent.h"
 #include "logging.h"
@@ -100,7 +101,7 @@ static void free_all(void)
 static bool in_transaction;
 static int target, current;
 static jmp_buf jmpbuf;
-#define TEST_DBNAME "run-57-die-during-transaction.tdb"
+#define TEST_DBNAME "run-57-die-during-transaction.ntdb"
 #define KEY_STRING "helloworld"
 
 static void maybe_die(int fd)
@@ -153,24 +154,24 @@ static int ftruncate_check(int fd, off_t length)
 
 static bool test_death(enum operation op, struct agent *agent)
 {
-	struct tdb_context *tdb = NULL;
-	TDB_DATA key;
+	struct ntdb_context *ntdb = NULL;
+	NTDB_DATA key;
 	enum agent_return ret;
 	int needed_recovery = 0;
 
 	current = target = 0;
 reset:
 	unlink(TEST_DBNAME);
-	tdb = tdb_open(TEST_DBNAME, TDB_NOMMAP,
+	ntdb = ntdb_open(TEST_DBNAME, NTDB_NOMMAP,
 		       O_CREAT|O_TRUNC|O_RDWR, 0600, &tap_log_attr);
-	if (!tdb) {
-		diag("Failed opening TDB: %s", strerror(errno));
+	if (!ntdb) {
+		diag("Failed opening NTDB: %s", strerror(errno));
 		return false;
 	}
 
 	if (setjmp(jmpbuf) != 0) {
 		/* We're partway through.  Simulate our death. */
-		close(tdb->file->fd);
+		close(ntdb->file->fd);
 		forget_locking();
 		in_transaction = false;
 
@@ -215,7 +216,7 @@ reset:
 		/* Suppress logging as this tries to use closed fd. */
 		suppress_logging = true;
 		suppress_lockcheck = true;
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 		suppress_logging = false;
 		suppress_lockcheck = false;
 		target++;
@@ -225,8 +226,8 @@ reset:
 	}
 
 	/* Put key for agent to fetch. */
-	key = tdb_mkdata(KEY_STRING, strlen(KEY_STRING));
-	if (tdb_store(tdb, key, key, TDB_INSERT) != 0)
+	key = ntdb_mkdata(KEY_STRING, strlen(KEY_STRING));
+	if (ntdb_store(ntdb, key, key, NTDB_INSERT) != 0)
 		return false;
 
 	/* This is the key we insert in transaction. */
@@ -241,20 +242,20 @@ reset:
 		errx(1, "Agent failed find key: %s", agent_return_name(ret));
 
 	in_transaction = true;
-	if (tdb_transaction_start(tdb) != 0)
+	if (ntdb_transaction_start(ntdb) != 0)
 		return false;
 
-	if (tdb_store(tdb, key, key, TDB_INSERT) != 0)
+	if (ntdb_store(ntdb, key, key, NTDB_INSERT) != 0)
 		return false;
 
-	if (tdb_transaction_commit(tdb) != 0)
+	if (ntdb_transaction_commit(ntdb) != 0)
 		return false;
 
 	in_transaction = false;
 
 	/* We made it! */
 	diag("Completed %u runs", current);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 	ret = external_agent_operation(agent, CLOSE, "");
 	if (ret != SUCCESS) {
 		diag("Step %u close failed = %s", current,

lib/ntdb/test/run-64-bit-tdb.c

@@ -0,0 +1,72 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	if (sizeof(off_t) <= 4) {
+		plan_tests(1);
+		pass("No 64 bit off_t");
+		return exit_status();
+	}
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 14);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		off_t old_size;
+		NTDB_DATA k, d;
+		struct hash_info h;
+		struct ntdb_used_record rec;
+		ntdb_off_t off;
+
+		ntdb = ntdb_open("run-64-bit-ntdb.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		old_size = ntdb->file->map_size;
+
+		/* This makes a sparse file */
+		ok1(ftruncate(ntdb->file->fd, 0xFFFFFFF0) == 0);
+		ok1(add_free_record(ntdb, old_size, 0xFFFFFFF0 - old_size,
+				    NTDB_LOCK_WAIT, false) == NTDB_SUCCESS);
+
+		/* Now add a little record past the 4G barrier. */
+		ok1(ntdb_expand_file(ntdb, 100) == NTDB_SUCCESS);
+		ok1(add_free_record(ntdb, 0xFFFFFFF0, 100, NTDB_LOCK_WAIT, false)
+		    == NTDB_SUCCESS);
+
+		ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
+
+		/* Test allocation path. */
+		k = ntdb_mkdata("key", 4);
+		d = ntdb_mkdata("data", 5);
+		ok1(ntdb_store(ntdb, k, d, NTDB_INSERT) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
+
+		/* Make sure it put it at end as we expected. */
+		off = find_and_lock(ntdb, k, F_RDLCK, &h, &rec, NULL);
+		ok1(off >= 0xFFFFFFF0);
+		ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_RDLCK);
+
+		ok1(ntdb_fetch(ntdb, k, &d) == 0);
+		ok1(d.dsize == 5);
+		ok1(strcmp((char *)d.dptr, "data") == 0);
+		free(d.dptr);
+
+		ok1(ntdb_delete(ntdb, k) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
+
+		ntdb_close(ntdb);
+	}
+
+	/* We might get messages about mmap failing, so don't test
+	 * tap_log_messages */
+	return exit_status();
+}

lib/ntdb/test/run-90-get-set-attributes.c

@@ -0,0 +1,159 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+static int mylock(int fd, int rw, off_t off, off_t len, bool waitflag,
+		  void *unused)
+{
+	return 0;
+}
+
+static int myunlock(int fd, int rw, off_t off, off_t len, void *unused)
+{
+	return 0;
+}
+
+static uint64_t hash_fn(const void *key, size_t len, uint64_t seed,
+			void *priv)
+{
+	return 0;
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+	union ntdb_attribute seed_attr;
+	union ntdb_attribute hash_attr;
+	union ntdb_attribute lock_attr;
+
+	seed_attr.base.attr = NTDB_ATTRIBUTE_SEED;
+	seed_attr.base.next = &hash_attr;
+	seed_attr.seed.seed = 100;
+
+	hash_attr.base.attr = NTDB_ATTRIBUTE_HASH;
+	hash_attr.base.next = &lock_attr;
+	hash_attr.hash.fn = hash_fn;
+	hash_attr.hash.data = &hash_attr;
+
+	lock_attr.base.attr = NTDB_ATTRIBUTE_FLOCK;
+	lock_attr.base.next = &tap_log_attr;
+	lock_attr.flock.lock = mylock;
+	lock_attr.flock.unlock = myunlock;
+	lock_attr.flock.data = &lock_attr;
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 50);
+
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		union ntdb_attribute attr;
+
+		/* First open with no attributes. */
+		ntdb = ntdb_open("run-90-get-set-attributes.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, NULL);
+		ok1(ntdb);
+
+		/* Get log on no attributes will fail */
+		attr.base.attr = NTDB_ATTRIBUTE_LOG;
+		ok1(ntdb_get_attribute(ntdb, &attr) == NTDB_ERR_NOEXIST);
+		/* These always work. */
+		attr.base.attr = NTDB_ATTRIBUTE_HASH;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_HASH);
+		ok1(attr.hash.fn == ntdb_jenkins_hash);
+		attr.base.attr = NTDB_ATTRIBUTE_FLOCK;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_FLOCK);
+		ok1(attr.flock.lock == ntdb_fcntl_lock);
+		ok1(attr.flock.unlock == ntdb_fcntl_unlock);
+		attr.base.attr = NTDB_ATTRIBUTE_SEED;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_SEED);
+		/* This is possible, just astronomically unlikely. */
+		ok1(attr.seed.seed != 0);
+
+		/* Unset attributes. */
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_LOG);
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_FLOCK);
+
+		/* Set them. */
+		ok1(ntdb_set_attribute(ntdb, &tap_log_attr) == 0);
+		ok1(ntdb_set_attribute(ntdb, &lock_attr) == 0);
+		/* These should fail. */
+		ok1(ntdb_set_attribute(ntdb, &seed_attr) == NTDB_ERR_EINVAL);
+		ok1(tap_log_messages == 1);
+		ok1(ntdb_set_attribute(ntdb, &hash_attr) == NTDB_ERR_EINVAL);
+		ok1(tap_log_messages == 2);
+		tap_log_messages = 0;
+
+		/* Getting them should work as expected. */
+		attr.base.attr = NTDB_ATTRIBUTE_LOG;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_LOG);
+		ok1(attr.log.fn == tap_log_attr.log.fn);
+		ok1(attr.log.data == tap_log_attr.log.data);
+
+		attr.base.attr = NTDB_ATTRIBUTE_FLOCK;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_FLOCK);
+		ok1(attr.flock.lock == mylock);
+		ok1(attr.flock.unlock == myunlock);
+		ok1(attr.flock.data == &lock_attr);
+
+		/* Unset them again. */
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_FLOCK);
+		ok1(tap_log_messages == 0);
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_LOG);
+		ok1(tap_log_messages == 0);
+
+		ntdb_close(ntdb);
+		ok1(tap_log_messages == 0);
+
+		/* Now open with all attributes. */
+		ntdb = ntdb_open("run-90-get-set-attributes.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600,
+			       &seed_attr);
+
+		ok1(ntdb);
+
+		/* Get will succeed */
+		attr.base.attr = NTDB_ATTRIBUTE_LOG;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_LOG);
+		ok1(attr.log.fn == tap_log_attr.log.fn);
+		ok1(attr.log.data == tap_log_attr.log.data);
+
+		attr.base.attr = NTDB_ATTRIBUTE_HASH;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_HASH);
+		ok1(attr.hash.fn == hash_fn);
+		ok1(attr.hash.data == &hash_attr);
+
+		attr.base.attr = NTDB_ATTRIBUTE_FLOCK;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_FLOCK);
+		ok1(attr.flock.lock == mylock);
+		ok1(attr.flock.unlock == myunlock);
+		ok1(attr.flock.data == &lock_attr);
+
+		attr.base.attr = NTDB_ATTRIBUTE_SEED;
+		ok1(ntdb_get_attribute(ntdb, &attr) == 0);
+		ok1(attr.base.attr == NTDB_ATTRIBUTE_SEED);
+		ok1(attr.seed.seed == seed_attr.seed.seed);
+
+		/* Unset attributes. */
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_HASH);
+		ok1(tap_log_messages == 1);
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_SEED);
+		ok1(tap_log_messages == 2);
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_FLOCK);
+		ntdb_unset_attribute(ntdb, NTDB_ATTRIBUTE_LOG);
+		ok1(tap_log_messages == 2);
+		tap_log_messages = 0;
+
+		ntdb_close(ntdb);
+
+	}
+	return exit_status();
+}

lib/ntdb/test/run-capabilities.c

@@ -1,5 +1,5 @@
 #include <ccan/failtest/failtest_override.h>
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include "logging.h"
 #include "layout.h"
@@ -18,28 +18,28 @@ static size_t len_of(bool breaks_check, bool breaks_write, bool breaks_open)
 	return len;
 }
 
-/* Creates a TDB with various capabilities. */
-static void create_tdb(const char *name,
+/* Creates a NTDB with various capabilities. */
+static void create_ntdb(const char *name,
 		       unsigned int cap,
 		       bool breaks_check,
 		       bool breaks_write,
 		       bool breaks_open, ...)
 {
-	TDB_DATA key, data;
+	NTDB_DATA key, data;
 	va_list ap;
-	struct tdb_layout *layout;
-	struct tdb_context *tdb;
+	struct ntdb_layout *layout;
+	struct ntdb_context *ntdb;
 	int fd;
 
-	key = tdb_mkdata("Hello", 5);
-	data = tdb_mkdata("world", 5);
+	key = ntdb_mkdata("Hello", 5);
+	data = ntdb_mkdata("world", 5);
 
-	/* Create a TDB with some data, and some capabilities */
-	layout = new_tdb_layout();
-	tdb_layout_add_freetable(layout);
-	tdb_layout_add_used(layout, key, data, 6);
-	tdb_layout_add_free(layout, 80, 0);
-	tdb_layout_add_capability(layout, cap,
+	/* Create a NTDB with some data, and some capabilities */
+	layout = new_ntdb_layout();
+	ntdb_layout_add_freetable(layout);
+	ntdb_layout_add_used(layout, key, data, 6);
+	ntdb_layout_add_free(layout, 80, 0);
+	ntdb_layout_add_capability(layout, cap,
 				  breaks_write, breaks_check, breaks_open,
 				  len_of(breaks_check, breaks_write, breaks_open));
 
@@ -50,9 +50,9 @@ static void create_tdb(const char *name,
 		breaks_open = va_arg(ap, int);
 
 		key.dsize--;
-		tdb_layout_add_used(layout, key, data, 11 - key.dsize);
-		tdb_layout_add_free(layout, 80, 0);
-		tdb_layout_add_capability(layout, cap,
+		ntdb_layout_add_used(layout, key, data, 11 - key.dsize);
+		ntdb_layout_add_free(layout, 80, 0);
+		ntdb_layout_add_capability(layout, cap,
 					  breaks_write, breaks_check,
 					  breaks_open,
 					  len_of(breaks_check, breaks_write,
@@ -61,23 +61,23 @@ static void create_tdb(const char *name,
 	va_end(ap);
 
 	/* We open-code this, because we need to use the failtest write. */
-	tdb = tdb_layout_get(layout, failtest_free, &tap_log_attr);
+	ntdb = ntdb_layout_get(layout, failtest_free, &tap_log_attr);
 
 	fd = open(name, O_RDWR|O_TRUNC|O_CREAT, 0600);
 	if (fd < 0)
 		err(1, "opening %s for writing", name);
-	if (write(fd, tdb->file->map_ptr, tdb->file->map_size)
-	    != tdb->file->map_size)
+	if (write(fd, ntdb->file->map_ptr, ntdb->file->map_size)
+	    != ntdb->file->map_size)
 		err(1, "writing %s", name);
 	close(fd);
-	tdb_close(tdb);
-	tdb_layout_free(layout);
+	ntdb_close(ntdb);
+	ntdb_layout_free(layout);
 }
 
 /* Note all the "goto out" early exits: they're to shorten failtest time. */
 int main(int argc, char *argv[])
 {
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 	char *summary;
 
 	failtest_init(argc, argv);
@@ -87,72 +87,72 @@ int main(int argc, char *argv[])
 
 	failtest_suppress = true;
 	/* Capability says you can ignore it? */
-	create_tdb("run-capabilities.tdb", 1, false, false, false, 0);
+	create_ntdb("run-capabilities.ntdb", 1, false, false, false, 0);
 
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDWR, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDWR, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
-	if (!ok1(tdb))
+	if (!ok1(ntdb))
 		goto out;
 	ok1(tap_log_messages == 0);
-	ok1(tdb_check(tdb, NULL, NULL) == TDB_SUCCESS);
+	ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
 	ok1(tap_log_messages == 0);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 
 	/* Two capabilitues say you can ignore them? */
-	create_tdb("run-capabilities.tdb",
+	create_ntdb("run-capabilities.ntdb",
 		   1, false, false, false,
 		   2, false, false, false, 0);
 
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDWR, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDWR, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
-	if (!ok1(tdb))
+	if (!ok1(ntdb))
 		goto out;
 	ok1(tap_log_messages == 0);
-	ok1(tdb_check(tdb, NULL, NULL) == TDB_SUCCESS);
+	ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
 	ok1(tap_log_messages == 0);
-	ok1(tdb_summary(tdb, 0, &summary) == TDB_SUCCESS);
+	ok1(ntdb_summary(ntdb, 0, &summary) == NTDB_SUCCESS);
 	ok1(strstr(summary, "Capability 1\n"));
 	free(summary);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 
 	/* Capability says you can't check. */
-	create_tdb("run-capabilities.tdb",
+	create_ntdb("run-capabilities.ntdb",
 		   1, false, false, false,
 		   2, true, false, false, 0);
 
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDWR, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDWR, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
-	if (!ok1(tdb))
+	if (!ok1(ntdb))
 		goto out;
 	ok1(tap_log_messages == 0);
-	ok1(tdb_get_flags(tdb) & TDB_CANT_CHECK);
-	ok1(tdb_check(tdb, NULL, NULL) == TDB_SUCCESS);
+	ok1(ntdb_get_flags(ntdb) & NTDB_CANT_CHECK);
+	ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
 	/* We expect a warning! */
 	ok1(tap_log_messages == 1);
 	ok1(strstr(log_last, "capabilit"));
-	ok1(tdb_summary(tdb, 0, &summary) == TDB_SUCCESS);
+	ok1(ntdb_summary(ntdb, 0, &summary) == NTDB_SUCCESS);
 	ok1(strstr(summary, "Capability 1\n"));
 	ok1(strstr(summary, "Capability 2 (uncheckable)\n"));
 	free(summary);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 
 	/* Capability says you can't write. */
-	create_tdb("run-capabilities.tdb",
+	create_ntdb("run-capabilities.ntdb",
 		   1, false, false, false,
 		   2, false, true, false, 0);
 
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDWR, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDWR, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
 	/* We expect a message. */
-	ok1(!tdb);
+	ok1(!ntdb);
 	if (!ok1(tap_log_messages == 2))
 		goto out;
 	if (!ok1(strstr(log_last, "unknown")))
@@ -161,48 +161,48 @@ int main(int argc, char *argv[])
 
 	/* We can open it read-only though! */
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDONLY, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDONLY, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
-	if (!ok1(tdb))
+	if (!ok1(ntdb))
 		goto out;
 	ok1(tap_log_messages == 2);
-	ok1(tdb_check(tdb, NULL, NULL) == TDB_SUCCESS);
+	ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
 	ok1(tap_log_messages == 2);
-	ok1(tdb_summary(tdb, 0, &summary) == TDB_SUCCESS);
+	ok1(ntdb_summary(ntdb, 0, &summary) == NTDB_SUCCESS);
 	ok1(strstr(summary, "Capability 1\n"));
 	ok1(strstr(summary, "Capability 2 (read-only)\n"));
 	free(summary);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 
 	/* Capability says you can't open. */
-	create_tdb("run-capabilities.tdb",
+	create_ntdb("run-capabilities.ntdb",
 		   1, false, false, false,
 		   2, false, false, true, 0);
 
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDWR, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDWR, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
 	/* We expect a message. */
-	ok1(!tdb);
+	ok1(!ntdb);
 	if (!ok1(tap_log_messages == 3))
 		goto out;
 	if (!ok1(strstr(log_last, "unknown")))
 		goto out;
 
 	/* Combine capabilities correctly. */
-	create_tdb("run-capabilities.tdb",
+	create_ntdb("run-capabilities.ntdb",
 		   1, false, false, false,
 		   2, true, false, false,
 		   3, false, true, false, 0);
 
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDWR, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDWR, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
 	/* We expect a message. */
-	ok1(!tdb);
+	ok1(!ntdb);
 	if (!ok1(tap_log_messages == 4))
 		goto out;
 	if (!ok1(strstr(log_last, "unknown")))
@@ -211,36 +211,36 @@ int main(int argc, char *argv[])
 
 	/* We can open it read-only though! */
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDONLY, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDONLY, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
-	if (!ok1(tdb))
+	if (!ok1(ntdb))
 		goto out;
 	ok1(tap_log_messages == 4);
-	ok1(tdb_get_flags(tdb) & TDB_CANT_CHECK);
-	ok1(tdb_check(tdb, NULL, NULL) == TDB_SUCCESS);
+	ok1(ntdb_get_flags(ntdb) & NTDB_CANT_CHECK);
+	ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
 	/* We expect a warning! */
 	ok1(tap_log_messages == 5);
 	ok1(strstr(log_last, "unknown"));
-	ok1(tdb_summary(tdb, 0, &summary) == TDB_SUCCESS);
+	ok1(ntdb_summary(ntdb, 0, &summary) == NTDB_SUCCESS);
 	ok1(strstr(summary, "Capability 1\n"));
 	ok1(strstr(summary, "Capability 2 (uncheckable)\n"));
 	ok1(strstr(summary, "Capability 3 (read-only)\n"));
 	free(summary);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 
 	/* Two capability flags in one. */
-	create_tdb("run-capabilities.tdb",
+	create_ntdb("run-capabilities.ntdb",
 		   1, false, false, false,
 		   2, true, true, false,
 		   0);
 
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDWR, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDWR, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
 	/* We expect a message. */
-	ok1(!tdb);
+	ok1(!ntdb);
 	if (!ok1(tap_log_messages == 6))
 		goto out;
 	if (!ok1(strstr(log_last, "unknown")))
@@ -249,22 +249,22 @@ int main(int argc, char *argv[])
 
 	/* We can open it read-only though! */
 	failtest_suppress = false;
-	tdb = tdb_open("run-capabilities.tdb", TDB_DEFAULT, O_RDONLY, 0,
+	ntdb = ntdb_open("run-capabilities.ntdb", NTDB_DEFAULT, O_RDONLY, 0,
 		       &tap_log_attr);
 	failtest_suppress = true;
-	if (!ok1(tdb))
+	if (!ok1(ntdb))
 		goto out;
 	ok1(tap_log_messages == 6);
-	ok1(tdb_get_flags(tdb) & TDB_CANT_CHECK);
-	ok1(tdb_check(tdb, NULL, NULL) == TDB_SUCCESS);
+	ok1(ntdb_get_flags(ntdb) & NTDB_CANT_CHECK);
+	ok1(ntdb_check(ntdb, NULL, NULL) == NTDB_SUCCESS);
 	/* We expect a warning! */
 	ok1(tap_log_messages == 7);
 	ok1(strstr(log_last, "unknown"));
-	ok1(tdb_summary(tdb, 0, &summary) == TDB_SUCCESS);
+	ok1(ntdb_summary(ntdb, 0, &summary) == NTDB_SUCCESS);
 	ok1(strstr(summary, "Capability 1\n"));
 	ok1(strstr(summary, "Capability 2 (uncheckable,read-only)\n"));
 	free(summary);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 
 out:
 	failtest_exit(exit_status());

lib/ntdb/test/run-expand-in-transaction.c

@@ -0,0 +1,36 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = ntdb_mkdata("key", 3);
+	NTDB_DATA data = ntdb_mkdata("data", 4);
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 7 + 1);
+
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		size_t size;
+		ntdb = ntdb_open("run-expand-in-transaction.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+
+		size = ntdb->file->map_size;
+		ok1(ntdb_transaction_start(ntdb) == 0);
+		ok1(ntdb_store(ntdb, key, data, NTDB_INSERT) == 0);
+		ok1(ntdb->file->map_size > size);
+		ok1(ntdb_transaction_commit(ntdb) == 0);
+		ok1(ntdb->file->map_size > size);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ntdb_close(ntdb);
+	}
+
+	ok1(tap_log_messages == 0);
+	return exit_status();
+}

lib/ntdb/test/run-features.c

@@ -1,60 +1,60 @@
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include "logging.h"
 
 int main(int argc, char *argv[])
 {
 	unsigned int i, j;
-	struct tdb_context *tdb;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
-	struct tdb_data key = { (unsigned char *)&j, sizeof(j) };
-	struct tdb_data data = { (unsigned char *)&j, sizeof(j) };
+	struct ntdb_context *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
+	NTDB_DATA key = { (unsigned char *)&j, sizeof(j) };
+	NTDB_DATA data = { (unsigned char *)&j, sizeof(j) };
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 8 + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
 		uint64_t features;
-		tdb = tdb_open("run-features.tdb", flags[i],
+		ntdb = ntdb_open("run-features.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		/* Put some stuff in there. */
 		for (j = 0; j < 100; j++) {
-			if (tdb_store(tdb, key, data, TDB_REPLACE) != 0)
-				fail("Storing in tdb");
+			if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
+				fail("Storing in ntdb");
 		}
 
 		/* Mess with features fields in hdr. */
-		features = (~TDB_FEATURE_MASK ^ 1);
-		ok1(tdb_write_convert(tdb, offsetof(struct tdb_header,
+		features = (~NTDB_FEATURE_MASK ^ 1);
+		ok1(ntdb_write_convert(ntdb, offsetof(struct ntdb_header,
 						    features_used),
 				      &features, sizeof(features)) == 0);
-		ok1(tdb_write_convert(tdb, offsetof(struct tdb_header,
+		ok1(ntdb_write_convert(ntdb, offsetof(struct ntdb_header,
 						    features_offered),
 				      &features, sizeof(features)) == 0);
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 
-		tdb = tdb_open("run-features.tdb", flags[i], O_RDWR, 0,
+		ntdb = ntdb_open("run-features.ntdb", flags[i], O_RDWR, 0,
 			       &tap_log_attr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
 		/* Should not have changed features offered. */
-		ok1(tdb_read_convert(tdb, offsetof(struct tdb_header,
+		ok1(ntdb_read_convert(ntdb, offsetof(struct ntdb_header,
 						   features_offered),
 				     &features, sizeof(features)) == 0);
-		ok1(features == (~TDB_FEATURE_MASK ^ 1));
+		ok1(features == (~NTDB_FEATURE_MASK ^ 1));
 
 		/* Should have cleared unknown bits in features_used. */
-		ok1(tdb_read_convert(tdb, offsetof(struct tdb_header,
+		ok1(ntdb_read_convert(ntdb, offsetof(struct ntdb_header,
 						   features_used),
 				     &features, sizeof(features)) == 0);
-		ok1(features == (1 & TDB_FEATURE_MASK));
+		ok1(features == (1 & NTDB_FEATURE_MASK));
 
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/test/run-lockall.c

@@ -3,25 +3,24 @@
 #include "lock-tracking.h"
 
 #define fcntl fcntl_with_lockcheck
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 
 #include "tap-interface.h"
 #include <stdlib.h>
 #include <stdbool.h>
 #include <stdarg.h>
-#include <ccan/err/err.h>
 #include "external-agent.h"
 #include "logging.h"
 
-#define TEST_DBNAME "run-lockall.tdb"
+#define TEST_DBNAME "run-lockall.ntdb"
 
 #undef fcntl
 
 int main(int argc, char *argv[])
 {
 	struct agent *agent;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
 	int i;
 
 	plan_tests(13 * sizeof(flags)/sizeof(flags[0]) + 1);
@@ -31,38 +30,38 @@ int main(int argc, char *argv[])
 
 	for (i = 0; i < sizeof(flags)/sizeof(flags[0]); i++) {
 		enum agent_return ret;
-		struct tdb_context *tdb;
+		struct ntdb_context *ntdb;
 
-		tdb = tdb_open(TEST_DBNAME, flags[i],
+		ntdb = ntdb_open(TEST_DBNAME, flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		ok1(tdb);
+		ok1(ntdb);
 
 		ret = external_agent_operation(agent, OPEN, TEST_DBNAME);
 		ok1(ret == SUCCESS);
 
-		ok1(tdb_lockall(tdb) == TDB_SUCCESS);
+		ok1(ntdb_lockall(ntdb) == NTDB_SUCCESS);
 		ok1(external_agent_operation(agent, STORE, "key")
 		    == WOULD_HAVE_BLOCKED);
 		ok1(external_agent_operation(agent, FETCH, "key")
 		    == WOULD_HAVE_BLOCKED);
 		/* Test nesting. */
-		ok1(tdb_lockall(tdb) == TDB_SUCCESS);
-		tdb_unlockall(tdb);
-		tdb_unlockall(tdb);
+		ok1(ntdb_lockall(ntdb) == NTDB_SUCCESS);
+		ntdb_unlockall(ntdb);
+		ntdb_unlockall(ntdb);
 
 		ok1(external_agent_operation(agent, STORE, "key") == SUCCESS);
 
-		ok1(tdb_lockall_read(tdb) == TDB_SUCCESS);
+		ok1(ntdb_lockall_read(ntdb) == NTDB_SUCCESS);
 		ok1(external_agent_operation(agent, STORE, "key")
 		    == WOULD_HAVE_BLOCKED);
 		ok1(external_agent_operation(agent, FETCH, "key") == SUCCESS);
-		ok1(tdb_lockall_read(tdb) == TDB_SUCCESS);
-		tdb_unlockall_read(tdb);
-		tdb_unlockall_read(tdb);
+		ok1(ntdb_lockall_read(ntdb) == NTDB_SUCCESS);
+		ntdb_unlockall_read(ntdb);
+		ntdb_unlockall_read(ntdb);
 
 		ok1(external_agent_operation(agent, STORE, "key") == SUCCESS);
 		ok1(external_agent_operation(agent, CLOSE, NULL) == SUCCESS);
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	free_external_agent(agent);

lib/ntdb/test/run-remap-in-read_traverse.c

@@ -1,11 +1,11 @@
-#include "tdb2-source.h"
-/* We had a bug where we marked the tdb read-only for a tdb_traverse_read.
- * If we then expanded the tdb, we would remap read-only, and later SEGV. */
+#include "ntdb-source.h"
+/* We had a bug where we marked the ntdb read-only for a ntdb_traverse_read.
+ * If we then expanded the ntdb, we would remap read-only, and later SEGV. */
 #include "tap-interface.h"
 #include "external-agent.h"
 #include "logging.h"
 
-static bool file_larger(int fd, tdb_len_t size)
+static bool file_larger(int fd, ntdb_len_t size)
 {
 	struct stat st;
 
@@ -13,7 +13,7 @@ static bool file_larger(int fd, tdb_len_t size)
 	return st.st_size != size;
 }
 
-static unsigned add_records_to_grow(struct agent *agent, int fd, tdb_len_t size)
+static unsigned add_records_to_grow(struct agent *agent, int fd, ntdb_len_t size)
 {
 	unsigned int i;
 
@@ -31,27 +31,27 @@ int main(int argc, char *argv[])
 {
 	unsigned int i;
 	struct agent *agent;
-	struct tdb_context *tdb;
-	struct tdb_data d = tdb_mkdata("hello", 5);
-	const char filename[] = "run-remap-in-read_traverse.tdb";
+	struct ntdb_context *ntdb;
+	NTDB_DATA d = ntdb_mkdata("hello", 5);
+	const char filename[] = "run-remap-in-read_traverse.ntdb";
 
 	plan_tests(4);
 
 	agent = prepare_external_agent();
 
-	tdb = tdb_open(filename, TDB_DEFAULT,
+	ntdb = ntdb_open(filename, NTDB_DEFAULT,
 		       O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
 
 	ok1(external_agent_operation(agent, OPEN, filename) == SUCCESS);
-	i = add_records_to_grow(agent, tdb->file->fd, tdb->file->map_size);
+	i = add_records_to_grow(agent, ntdb->file->fd, ntdb->file->map_size);
 
 	/* Do a traverse. */
-	ok1(tdb_traverse(tdb, NULL, NULL) == i);
+	ok1(ntdb_traverse(ntdb, NULL, NULL) == i);
 
 	/* Now store something! */
-	ok1(tdb_store(tdb, d, d, TDB_INSERT) == 0);
+	ok1(ntdb_store(ntdb, d, d, NTDB_INSERT) == 0);
 	ok1(tap_log_messages == 0);
-	tdb_close(tdb);
+	ntdb_close(ntdb);
 	free_external_agent(agent);
 	return exit_status();
 }

lib/ntdb/test/run-seed.c

@@ -0,0 +1,61 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+#include "logging.h"
+
+static int log_count = 0;
+
+/* Normally we get a log when setting random seed. */
+static void my_log_fn(struct ntdb_context *ntdb,
+		      enum ntdb_log_level level,
+		      enum NTDB_ERROR ecode,
+		      const char *message, void *priv)
+{
+	log_count++;
+}
+
+static union ntdb_attribute log_attr = {
+	.log = { .base = { .attr = NTDB_ATTRIBUTE_LOG },
+		 .fn = my_log_fn }
+};
+
+int main(int argc, char *argv[])
+{
+	unsigned int i;
+	struct ntdb_context *ntdb;
+	union ntdb_attribute attr;
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+
+	attr.seed.base.attr = NTDB_ATTRIBUTE_SEED;
+	attr.seed.base.next = &log_attr;
+	attr.seed.seed = 42;
+
+	plan_tests(sizeof(flags) / sizeof(flags[0]) * 4 + 4 * 3);
+	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
+		struct ntdb_header hdr;
+		int fd;
+		ntdb = ntdb_open("run-seed.ntdb", flags[i],
+			       O_RDWR|O_CREAT|O_TRUNC, 0600, &attr);
+		ok1(ntdb);
+		if (!ntdb)
+			continue;
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
+		ok1(ntdb->hash_seed == 42);
+		ok1(log_count == 0);
+		ntdb_close(ntdb);
+
+		if (flags[i] & NTDB_INTERNAL)
+			continue;
+
+		fd = open("run-seed.ntdb", O_RDONLY);
+		ok1(fd >= 0);
+		ok1(read(fd, &hdr, sizeof(hdr)) == sizeof(hdr));
+		if (flags[i] & NTDB_CONVERT)
+			ok1(bswap_64(hdr.hash_seed) == 42);
+		else
+			ok1(hdr.hash_seed == 42);
+		close(fd);
+	}
+	return exit_status();
+}

lib/ntdb/test/run-tdb_errorstr.c

@@ -0,0 +1,52 @@
+#include "ntdb-source.h"
+#include "tap-interface.h"
+
+int main(int argc, char *argv[])
+{
+	enum NTDB_ERROR e;
+	plan_tests(NTDB_ERR_RDONLY*-1 + 2);
+
+	for (e = NTDB_SUCCESS; e >= NTDB_ERR_RDONLY; e--) {
+		switch (e) {
+		case NTDB_SUCCESS:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "Success"));
+			break;
+		case NTDB_ERR_IO:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "IO Error"));
+			break;
+		case NTDB_ERR_LOCK:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "Locking error"));
+			break;
+		case NTDB_ERR_OOM:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "Out of memory"));
+			break;
+		case NTDB_ERR_EXISTS:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "Record exists"));
+			break;
+		case NTDB_ERR_EINVAL:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "Invalid parameter"));
+			break;
+		case NTDB_ERR_NOEXIST:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "Record does not exist"));
+			break;
+		case NTDB_ERR_RDONLY:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "write not permitted"));
+			break;
+		case NTDB_ERR_CORRUPT:
+			ok1(!strcmp(ntdb_errorstr(e),
+				    "Corrupt database"));
+			break;
+		}
+	}
+	ok1(!strcmp(ntdb_errorstr(e), "Invalid error code"));
+
+	return exit_status();
+}

lib/ntdb/test/run-tdb_foreach.c

@@ -1,23 +1,23 @@
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include "logging.h"
 
-static int drop_count(struct tdb_context *tdb, unsigned int *count)
+static int drop_count(struct ntdb_context *ntdb, unsigned int *count)
 {
 	if (--(*count) == 0)
 		return 1;
 	return 0;
 }
 
-static int set_found(struct tdb_context *tdb, bool found[3])
+static int set_found(struct ntdb_context *ntdb, bool found[3])
 {
 	unsigned int idx;
 
-	if (strcmp(tdb_name(tdb), "run-tdb_foreach0.tdb") == 0)
+	if (strcmp(ntdb_name(ntdb), "run-ntdb_foreach0.ntdb") == 0)
 		idx = 0;
-	else if (strcmp(tdb_name(tdb), "run-tdb_foreach1.tdb") == 0)
+	else if (strcmp(ntdb_name(ntdb), "run-ntdb_foreach1.ntdb") == 0)
 		idx = 1;
-	else if (strcmp(tdb_name(tdb), "run-tdb_foreach2.tdb") == 0)
+	else if (strcmp(ntdb_name(ntdb), "run-ntdb_foreach2.ntdb") == 0)
 		idx = 2;
 	else
 		abort();
@@ -32,52 +32,52 @@ int main(int argc, char *argv[])
 {
 	unsigned int i, count;
 	bool found[3];
-	struct tdb_context *tdb0, *tdb1, *tdb2;
-	int flags[] = { TDB_DEFAULT, TDB_NOMMAP,
-			TDB_CONVERT, TDB_NOMMAP|TDB_CONVERT };
+	struct ntdb_context *ntdb0, *ntdb1, *ntdb;
+	int flags[] = { NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_CONVERT, NTDB_NOMMAP|NTDB_CONVERT };
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 8);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb0 = tdb_open("run-tdb_foreach0.tdb", flags[i],
+		ntdb0 = ntdb_open("run-ntdb_foreach0.ntdb", flags[i],
 				O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		tdb1 = tdb_open("run-tdb_foreach1.tdb", flags[i],
+		ntdb1 = ntdb_open("run-ntdb_foreach1.ntdb", flags[i],
 				O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
-		tdb2 = tdb_open("run-tdb_foreach2.tdb", flags[i],
+		ntdb = ntdb_open("run-ntdb_foreach2.ntdb", flags[i],
 				O_RDWR|O_CREAT|O_TRUNC, 0600, &tap_log_attr);
 
 		memset(found, 0, sizeof(found));
-		tdb_foreach(set_found, found);
+		ntdb_foreach(set_found, found);
 		ok1(found[0] && found[1] && found[2]);
 
 		/* Test premature iteration termination */
 		count = 1;
-		tdb_foreach(drop_count, &count);
+		ntdb_foreach(drop_count, &count);
 		ok1(count == 0);
 
-		tdb_close(tdb1);
+		ntdb_close(ntdb1);
 		memset(found, 0, sizeof(found));
-		tdb_foreach(set_found, found);
+		ntdb_foreach(set_found, found);
 		ok1(found[0] && !found[1] && found[2]);
 
-		tdb_close(tdb2);
+		ntdb_close(ntdb);
 		memset(found, 0, sizeof(found));
-		tdb_foreach(set_found, found);
+		ntdb_foreach(set_found, found);
 		ok1(found[0] && !found[1] && !found[2]);
 
-		tdb1 = tdb_open("run-tdb_foreach1.tdb", flags[i],
+		ntdb1 = ntdb_open("run-ntdb_foreach1.ntdb", flags[i],
 				O_RDWR, 0600, &tap_log_attr);
 		memset(found, 0, sizeof(found));
-		tdb_foreach(set_found, found);
+		ntdb_foreach(set_found, found);
 		ok1(found[0] && found[1] && !found[2]);
 
-		tdb_close(tdb0);
+		ntdb_close(ntdb0);
 		memset(found, 0, sizeof(found));
-		tdb_foreach(set_found, found);
+		ntdb_foreach(set_found, found);
 		ok1(!found[0] && found[1] && !found[2]);
 
-		tdb_close(tdb1);
+		ntdb_close(ntdb1);
 		memset(found, 0, sizeof(found));
-		tdb_foreach(set_found, found);
+		ntdb_foreach(set_found, found);
 		ok1(!found[0] && !found[1] && !found[2]);
 		ok1(tap_log_messages == 0);
 	}

lib/ntdb/test/run-traverse.c

@@ -1,4 +1,4 @@
-#include "tdb2-source.h"
+#include "ntdb-source.h"
 #include "tap-interface.h"
 #include "logging.h"
 
@@ -11,14 +11,14 @@ static uint64_t fixedhash(const void *key, size_t len, uint64_t seed, void *p)
 			     *(uint64_t *)p);
 }
 
-static bool store_records(struct tdb_context *tdb)
+static bool store_records(struct ntdb_context *ntdb)
 {
 	int i;
-	struct tdb_data key = { (unsigned char *)&i, sizeof(i) };
-	struct tdb_data data = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA key = { (unsigned char *)&i, sizeof(i) };
+	NTDB_DATA data = { (unsigned char *)&i, sizeof(i) };
 
 	for (i = 0; i < NUM_RECORDS; i++)
-		if (tdb_store(tdb, key, data, TDB_REPLACE) != 0)
+		if (ntdb_store(ntdb, key, data, NTDB_REPLACE) != 0)
 			return false;
 	return true;
 }
@@ -28,10 +28,10 @@ struct trav_data {
 	int low, high;
 	bool mismatch;
 	bool delete;
-	enum TDB_ERROR delete_error;
+	enum NTDB_ERROR delete_error;
 };
 
-static int trav(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
+static int trav(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf,
 		struct trav_data *td)
 {
 	int val;
@@ -49,8 +49,8 @@ static int trav(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 		td->high = val;
 
 	if (td->delete) {
-		td->delete_error = tdb_delete(tdb, key);
-		if (td->delete_error != TDB_SUCCESS) {
+		td->delete_error = ntdb_delete(ntdb, key);
+		if (td->delete_error != NTDB_SUCCESS) {
 			return -1;
 		}
 	}
@@ -64,10 +64,10 @@ struct trav_grow_data {
 	unsigned int calls;
 	unsigned int num_large;
 	bool mismatch;
-	enum TDB_ERROR error;
+	enum NTDB_ERROR error;
 };
 
-static int trav_grow(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
+static int trav_grow(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf,
 		     struct trav_grow_data *tgd)
 {
 	int val;
@@ -87,8 +87,8 @@ static int trav_grow(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 	/* Make a big difference to the database. */
 	dbuf.dptr = buffer;
 	dbuf.dsize = sizeof(buffer);
-	tgd->error = tdb_append(tdb, key, dbuf);
-	if (tgd->error != TDB_SUCCESS) {
+	tgd->error = ntdb_append(ntdb, key, dbuf);
+	if (tgd->error != NTDB_SUCCESS) {
 		return -1;
 	}
 	return 0;
@@ -100,12 +100,12 @@ int main(int argc, char *argv[])
 	int num;
 	struct trav_data td;
 	struct trav_grow_data tgd;
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 	uint64_t seed = 16014841315512641303ULL;
-	int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
-			TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
-			TDB_NOMMAP|TDB_CONVERT };
-	union tdb_attribute hattr = { .hash = { .base = { TDB_ATTRIBUTE_HASH },
+	int flags[] = { NTDB_INTERNAL, NTDB_DEFAULT, NTDB_NOMMAP,
+			NTDB_INTERNAL|NTDB_CONVERT, NTDB_CONVERT,
+			NTDB_NOMMAP|NTDB_CONVERT };
+	union ntdb_attribute hattr = { .hash = { .base = { NTDB_ATTRIBUTE_HASH },
 						.fn = fixedhash,
 						.data = &seed } };
 
@@ -113,16 +113,16 @@ int main(int argc, char *argv[])
 
 	plan_tests(sizeof(flags) / sizeof(flags[0]) * 32 + 1);
 	for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
-		tdb = tdb_open("run-traverse.tdb", flags[i],
+		ntdb = ntdb_open("run-traverse.ntdb", flags[i],
 			       O_RDWR|O_CREAT|O_TRUNC, 0600, &hattr);
-		ok1(tdb);
-		if (!tdb)
+		ok1(ntdb);
+		if (!ntdb)
 			continue;
 
-		ok1(tdb_traverse(tdb, NULL, NULL) == 0);
+		ok1(ntdb_traverse(ntdb, NULL, NULL) == 0);
 
-		ok1(store_records(tdb));
-		num = tdb_traverse(tdb, NULL, NULL);
+		ok1(store_records(ntdb));
+		num = ntdb_traverse(ntdb, NULL, NULL);
 		ok1(num == NUM_RECORDS);
 
 		/* Full traverse. */
@@ -133,7 +133,7 @@ int main(int argc, char *argv[])
 		td.mismatch = false;
 		td.delete = false;
 
-		num = tdb_traverse(tdb, trav, &td);
+		num = ntdb_traverse(ntdb, trav, &td);
 		ok1(num == NUM_RECORDS);
 		ok1(!td.mismatch);
 		ok1(td.calls == NUM_RECORDS);
@@ -148,13 +148,13 @@ int main(int argc, char *argv[])
 		td.mismatch = false;
 		td.delete = false;
 
-		num = tdb_traverse(tdb, trav, &td);
+		num = ntdb_traverse(ntdb, trav, &td);
 		ok1(num == NUM_RECORDS / 2);
 		ok1(!td.mismatch);
 		ok1(td.calls == NUM_RECORDS / 2);
 		ok1(td.low <= NUM_RECORDS / 2);
 		ok1(td.high > NUM_RECORDS / 2);
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 		ok1(tap_log_messages == 0);
 
 		/* Deleting traverse (delete everything). */
@@ -164,38 +164,38 @@ int main(int argc, char *argv[])
 		td.high = INT_MIN;
 		td.mismatch = false;
 		td.delete = true;
-		td.delete_error = TDB_SUCCESS;
-		num = tdb_traverse(tdb, trav, &td);
+		td.delete_error = NTDB_SUCCESS;
+		num = ntdb_traverse(ntdb, trav, &td);
 		ok1(num == NUM_RECORDS);
-		ok1(td.delete_error == TDB_SUCCESS);
+		ok1(td.delete_error == NTDB_SUCCESS);
 		ok1(!td.mismatch);
 		ok1(td.calls == NUM_RECORDS);
 		ok1(td.low == 0);
 		ok1(td.high == NUM_RECORDS - 1);
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 
 		/* Now it's empty! */
-		ok1(tdb_traverse(tdb, NULL, NULL) == 0);
+		ok1(ntdb_traverse(ntdb, NULL, NULL) == 0);
 
 		/* Re-add. */
-		ok1(store_records(tdb));
-		ok1(tdb_traverse(tdb, NULL, NULL) == NUM_RECORDS);
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(store_records(ntdb));
+		ok1(ntdb_traverse(ntdb, NULL, NULL) == NUM_RECORDS);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 
 		/* Grow.  This will cause us to be reshuffled. */
 		tgd.calls = 0;
 		tgd.num_large = 0;
 		tgd.mismatch = false;
-		tgd.error = TDB_SUCCESS;
-		ok1(tdb_traverse(tdb, trav_grow, &tgd) > 1);
+		tgd.error = NTDB_SUCCESS;
+		ok1(ntdb_traverse(ntdb, trav_grow, &tgd) > 1);
 		ok1(tgd.error == 0);
 		ok1(!tgd.mismatch);
-		ok1(tdb_check(tdb, NULL, NULL) == 0);
+		ok1(ntdb_check(ntdb, NULL, NULL) == 0);
 		ok1(tgd.num_large < tgd.calls);
 		diag("growing db: %u calls, %u repeats",
 		     tgd.calls, tgd.num_large);
 
-		tdb_close(tdb);
+		ntdb_close(ntdb);
 	}
 
 	ok1(tap_log_messages == 0);

lib/ntdb/tools/Makefile

@@ -0,0 +1,16 @@
+OBJS:=../../ntdb.o ../../hash.o ../../tally.o
+CFLAGS:=-I../../.. -I.. -Wall -g -O3 #-g -pg
+LDFLAGS:=-L../../..
+
+default: ntdbtorture ntdbtool ntdbdump ntdbrestore mkntdb speed growtdb-bench
+
+ntdbdump: ntdbdump.c $(OBJS)
+ntdbrestore: ntdbrestore.c $(OBJS)
+ntdbtorture: ntdbtorture.c $(OBJS)
+ntdbtool: ntdbtool.c $(OBJS)
+mkntdb: mkntdb.c $(OBJS)
+speed: speed.c $(OBJS)
+growtdb-bench: growtdb-bench.c $(OBJS)
+
+clean:
+	rm -f ntdbtorture ntdbdump ntdbrestore ntdbtool mkntdb speed growtdb-bench

lib/ntdb/tools/growtdb-bench.c

@@ -0,0 +1,114 @@
+#include "ntdb.h"
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include <unistd.h>
+#include <ccan/err/err.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+
+static void logfn(struct ntdb_context *ntdb,
+		  enum ntdb_log_level level,
+		  enum NTDB_ERROR ecode,
+		  const char *message,
+		  void *data)
+{
+	fprintf(stderr, "ntdb:%s:%s:%s\n",
+		ntdb_name(ntdb), ntdb_errorstr(ecode), message);
+}
+
+int main(int argc, char *argv[])
+{
+	unsigned int i, j, users, groups;
+	NTDB_DATA idxkey, idxdata;
+	NTDB_DATA k, d, gk;
+	char cmd[100];
+	struct ntdb_context *ntdb;
+	enum NTDB_ERROR ecode;
+	union ntdb_attribute log;
+
+	if (argc != 3) {
+		printf("Usage: growtdb-bench <users> <groups>\n");
+		exit(1);
+	}
+	users = atoi(argv[1]);
+	groups = atoi(argv[2]);
+
+	sprintf(cmd, "cat /proc/%i/statm", getpid());
+
+	log.base.attr = NTDB_ATTRIBUTE_LOG;
+	log.base.next = NULL;
+	log.log.fn = logfn;
+
+	ntdb = ntdb_open("/tmp/growtdb.ntdb", NTDB_DEFAULT,
+		       O_RDWR|O_CREAT|O_TRUNC, 0600, &log);
+
+	idxkey.dptr = (unsigned char *)"User index";
+	idxkey.dsize = strlen("User index");
+	idxdata.dsize = 51;
+	idxdata.dptr = calloc(idxdata.dsize, 1);
+
+	/* Create users. */
+	k.dsize = 48;
+	k.dptr = calloc(k.dsize, 1);
+	d.dsize = 64;
+	d.dptr = calloc(d.dsize, 1);
+
+	ntdb_transaction_start(ntdb);
+	for (i = 0; i < users; i++) {
+		memcpy(k.dptr, &i, sizeof(i));
+		ecode = ntdb_store(ntdb, k, d, NTDB_INSERT);
+		if (ecode != NTDB_SUCCESS)
+			errx(1, "ntdb insert failed: %s", ntdb_errorstr(ecode));
+
+		/* This simulates a growing index record. */
+		ecode = ntdb_append(ntdb, idxkey, idxdata);
+		if (ecode != NTDB_SUCCESS)
+			errx(1, "ntdb append failed: %s", ntdb_errorstr(ecode));
+	}
+	if ((ecode = ntdb_transaction_commit(ntdb)) != 0)
+		errx(1, "ntdb commit1 failed: %s", ntdb_errorstr(ecode));
+
+	if ((ecode = ntdb_check(ntdb, NULL, NULL)) != 0)
+		errx(1, "ntdb_check failed after initial insert!");
+
+	system(cmd);
+
+	/* Now put them all in groups: add 32 bytes to each record for
+	 * a group. */
+	gk.dsize = 48;
+	gk.dptr = calloc(k.dsize, 1);
+	gk.dptr[gk.dsize-1] = 1;
+
+	d.dsize = 32;
+	for (i = 0; i < groups; i++) {
+		ntdb_transaction_start(ntdb);
+		/* Create the "group". */
+		memcpy(gk.dptr, &i, sizeof(i));
+		ecode = ntdb_store(ntdb, gk, d, NTDB_INSERT);
+		if (ecode != NTDB_SUCCESS)
+			errx(1, "ntdb insert failed: %s", ntdb_errorstr(ecode));
+
+		/* Now populate it. */
+		for (j = 0; j < users; j++) {
+			/* Append to the user. */
+			memcpy(k.dptr, &j, sizeof(j));
+			if ((ecode = ntdb_append(ntdb, k, d)) != 0)
+				errx(1, "ntdb append failed: %s",
+				     ntdb_errorstr(ecode));
+
+			/* Append to the group. */
+			if ((ecode = ntdb_append(ntdb, gk, d)) != 0)
+				errx(1, "ntdb append failed: %s",
+				     ntdb_errorstr(ecode));
+		}
+		if ((ecode = ntdb_transaction_commit(ntdb)) != 0)
+			errx(1, "ntdb commit2 failed: %s", ntdb_errorstr(ecode));
+		if ((ecode = ntdb_check(ntdb, NULL, NULL)) != 0)
+			errx(1, "ntdb_check failed after iteration %i!", i);
+		system(cmd);
+	}
+
+	return 0;
+}

lib/ntdb/tools/mkntdb.c

@@ -1,4 +1,4 @@
-#include "tdb2.h"
+#include "ntdb.h"
 #include <stdlib.h>
 #include <stdio.h>
 #include <fcntl.h>
@@ -7,21 +7,21 @@
 int main(int argc, char *argv[])
 {
 	unsigned int i, num_recs;
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 
 	if (argc != 3 || (num_recs = atoi(argv[2])) == 0)
 		errx(1, "Usage: mktdb <tdbfile> <numrecords>");
 
-	tdb = tdb_open(argv[1], TDB_DEFAULT, O_CREAT|O_TRUNC|O_RDWR, 0600,NULL);
-	if (!tdb)
+	ntdb = ntdb_open(argv[1], NTDB_DEFAULT, O_CREAT|O_TRUNC|O_RDWR, 0600,NULL);
+	if (!ntdb)
 		err(1, "Opening %s", argv[1]);
 
 	for (i = 0; i < num_recs; i++) {
-		TDB_DATA d;
+		NTDB_DATA d;
 
 		d.dptr = (void *)&i;
 		d.dsize = sizeof(i);
-		if (tdb_store(tdb, d, d, TDB_INSERT) != 0)
+		if (ntdb_store(ntdb, d, d, NTDB_INSERT) != 0)
 			err(1, "Failed to store record %i", i);
 	}
 	printf("Done\n");

lib/ntdb/tools/ntdbbackup.c

@@ -1,6 +1,6 @@
 /*
    Unix SMB/CIFS implementation.
-   low level tdb backup and restore utility
+   low level ntdb backup and restore utility
    Copyright (C) Andrew Tridgell              2002
 
    This program is free software; you can redistribute it and/or modify
@@ -19,11 +19,11 @@
 
 /*
 
-  This program is meant for backup/restore of tdb databases. Typical usage would be:
-     tdbbackup *.tdb
+  This program is meant for backup/restore of ntdb databases. Typical usage would be:
+     tdbbackup *.ntdb
   when Samba shuts down cleanly, which will make a backup of all the local databases
   to *.bak files. Then on Samba startup you would use:
-     tdbbackup -v *.tdb
+     tdbbackup -v *.ntdb
   and this will check the databases for corruption and if corruption is detected then
   the backup will be restored.
 
@@ -41,7 +41,7 @@
  */
 
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include "system/filesys.h"
 
 #ifdef HAVE_GETOPT_H
@@ -50,13 +50,13 @@
 
 static int failed;
 
-static void tdb_log(struct tdb_context *tdb,
-		    enum tdb_log_level level,
-		    enum TDB_ERROR ecode,
+static void ntdb_log(struct ntdb_context *ntdb,
+		    enum ntdb_log_level level,
+		    enum NTDB_ERROR ecode,
 		    const char *message,
 		    void *data)
 {
-	fprintf(stderr, "%s:%s\n", tdb_errorstr(ecode), message);
+	fprintf(stderr, "%s:%s\n", ntdb_errorstr(ecode), message);
 }
 
 static char *add_suffix(const char *name, const char *suffix)
@@ -72,15 +72,15 @@ static char *add_suffix(const char *name, const char *suffix)
 	return ret;
 }
 
-static int copy_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
+static int copy_fn(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state)
 {
-	struct tdb_context *tdb_new = (struct tdb_context *)state;
-	enum TDB_ERROR err;
+	struct ntdb_context *ntdb_new = (struct ntdb_context *)state;
+	enum NTDB_ERROR err;
 
-	err = tdb_store(tdb_new, key, dbuf, TDB_INSERT);
+	err = ntdb_store(ntdb_new, key, dbuf, NTDB_INSERT);
 	if (err) {
 		fprintf(stderr,"Failed to insert into %s: %s\n",
-			tdb_name(tdb_new), tdb_errorstr(err));
+			ntdb_name(ntdb_new), ntdb_errorstr(err));
 		failed = 1;
 		return 1;
 	}
@@ -88,75 +88,75 @@ static int copy_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, void *s
 }
 
 
-static int test_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
+static int test_fn(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state)
 {
 	return 0;
 }
 
 /*
-  carefully backup a tdb, validating the contents and
+  carefully backup a ntdb, validating the contents and
   only doing the backup if its OK
   this function is also used for restore
 */
-static int backup_tdb(const char *old_name, const char *new_name)
+static int backup_ntdb(const char *old_name, const char *new_name)
 {
-	struct tdb_context *tdb;
-	struct tdb_context *tdb_new;
+	struct ntdb_context *ntdb;
+	struct ntdb_context *ntdb_new;
 	char *tmp_name;
 	struct stat st;
 	int count1, count2;
-	enum TDB_ERROR err;
-	union tdb_attribute log_attr;
+	enum NTDB_ERROR err;
+	union ntdb_attribute log_attr;
 
 	tmp_name = add_suffix(new_name, ".tmp");
 
-	/* stat the old tdb to find its permissions */
+	/* stat the old ntdb to find its permissions */
 	if (stat(old_name, &st) != 0) {
 		perror(old_name);
 		free(tmp_name);
 		return 1;
 	}
 
-	log_attr.base.attr = TDB_ATTRIBUTE_LOG;
+	log_attr.base.attr = NTDB_ATTRIBUTE_LOG;
 	log_attr.base.next = NULL;
-	log_attr.log.fn = tdb_log;
+	log_attr.log.fn = ntdb_log;
 
-	/* open the old tdb */
-	tdb = tdb_open(old_name, TDB_DEFAULT, O_RDWR, 0, &log_attr);
-	if (!tdb) {
+	/* open the old ntdb */
+	ntdb = ntdb_open(old_name, NTDB_DEFAULT, O_RDWR, 0, &log_attr);
+	if (!ntdb) {
 		printf("Failed to open %s\n", old_name);
 		free(tmp_name);
 		return 1;
 	}
 
 	unlink(tmp_name);
-	tdb_new = tdb_open(tmp_name, TDB_DEFAULT,
+	ntdb_new = ntdb_open(tmp_name, NTDB_DEFAULT,
 			   O_RDWR|O_CREAT|O_EXCL, st.st_mode & 0777,
 			   &log_attr);
-	if (!tdb_new) {
+	if (!ntdb_new) {
 		perror(tmp_name);
 		free(tmp_name);
 		return 1;
 	}
 
-	err = tdb_transaction_start(tdb);
+	err = ntdb_transaction_start(ntdb);
 	if (err) {
-		fprintf(stderr, "Failed to start transaction on old tdb: %s\n",
-			tdb_errorstr(err));
-		tdb_close(tdb);
-		tdb_close(tdb_new);
+		fprintf(stderr, "Failed to start transaction on old ntdb: %s\n",
+			ntdb_errorstr(err));
+		ntdb_close(ntdb);
+		ntdb_close(ntdb_new);
 		unlink(tmp_name);
 		free(tmp_name);
 		return 1;
 	}
 
-	/* lock the backup tdb so that nobody else can change it */
-	err = tdb_lockall(tdb_new);
+	/* lock the backup ntdb so that nobody else can change it */
+	err = ntdb_lockall(ntdb_new);
 	if (err) {
-		fprintf(stderr, "Failed to lock backup tdb: %s\n",
-			tdb_errorstr(err));
-		tdb_close(tdb);
-		tdb_close(tdb_new);
+		fprintf(stderr, "Failed to lock backup ntdb: %s\n",
+			ntdb_errorstr(err));
+		ntdb_close(ntdb);
+		ntdb_close(ntdb_new);
 		unlink(tmp_name);
 		free(tmp_name);
 		return 1;
@@ -165,39 +165,39 @@ static int backup_tdb(const char *old_name, const char *new_name)
 	failed = 0;
 
 	/* traverse and copy */
-	count1 = tdb_traverse(tdb, copy_fn, (void *)tdb_new);
+	count1 = ntdb_traverse(ntdb, copy_fn, (void *)ntdb_new);
 	if (count1 < 0 || failed) {
 		fprintf(stderr,"failed to copy %s\n", old_name);
-		tdb_close(tdb);
-		tdb_close(tdb_new);
+		ntdb_close(ntdb);
+		ntdb_close(ntdb_new);
 		unlink(tmp_name);
 		free(tmp_name);
 		return 1;
 	}
 
-	/* close the old tdb */
-	tdb_close(tdb);
+	/* close the old ntdb */
+	ntdb_close(ntdb);
 
-	/* copy done, unlock the backup tdb */
-	tdb_unlockall(tdb_new);
+	/* copy done, unlock the backup ntdb */
+	ntdb_unlockall(ntdb_new);
 
 #ifdef HAVE_FDATASYNC
-	if (fdatasync(tdb_fd(tdb_new)) != 0) {
+	if (fdatasync(ntdb_fd(ntdb_new)) != 0) {
 #else
-	if (fsync(tdb_fd(tdb_new)) != 0) {
+	if (fsync(ntdb_fd(ntdb_new)) != 0) {
 #endif
 		/* not fatal */
 		fprintf(stderr, "failed to fsync backup file\n");
 	}
 
-	/* close the new tdb and re-open read-only */
-	tdb_close(tdb_new);
+	/* close the new ntdb and re-open read-only */
+	ntdb_close(ntdb_new);
 
 	/* we don't need the hash attr any more */
 	log_attr.base.next = NULL;
 
-	tdb_new = tdb_open(tmp_name, TDB_DEFAULT, O_RDONLY, 0, &log_attr);
-	if (!tdb_new) {
+	ntdb_new = ntdb_open(tmp_name, NTDB_DEFAULT, O_RDONLY, 0, &log_attr);
+	if (!ntdb_new) {
 		fprintf(stderr,"failed to reopen %s\n", tmp_name);
 		unlink(tmp_name);
 		perror(tmp_name);
@@ -205,18 +205,18 @@ static int backup_tdb(const char *old_name, const char *new_name)
 		return 1;
 	}
 
-	/* traverse the new tdb to confirm */
-	count2 = tdb_traverse(tdb_new, test_fn, NULL);
+	/* traverse the new ntdb to confirm */
+	count2 = ntdb_traverse(ntdb_new, test_fn, NULL);
 	if (count2 != count1) {
 		fprintf(stderr,"failed to copy %s\n", old_name);
-		tdb_close(tdb_new);
+		ntdb_close(ntdb_new);
 		unlink(tmp_name);
 		free(tmp_name);
 		return 1;
 	}
 
-	/* close the new tdb and rename it to .bak */
-	tdb_close(tdb_new);
+	/* close the new ntdb and rename it to .bak */
+	ntdb_close(ntdb_new);
 	if (rename(tmp_name, new_name) != 0) {
 		perror(new_name);
 		free(tmp_name);
@@ -229,31 +229,31 @@ static int backup_tdb(const char *old_name, const char *new_name)
 }
 
 /*
-  verify a tdb and if it is corrupt then restore from *.bak
+  verify a ntdb and if it is corrupt then restore from *.bak
 */
-static int verify_tdb(const char *fname, const char *bak_name)
+static int verify_ntdb(const char *fname, const char *bak_name)
 {
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 	int count = -1;
-	union tdb_attribute log_attr;
+	union ntdb_attribute log_attr;
 
-	log_attr.base.attr = TDB_ATTRIBUTE_LOG;
+	log_attr.base.attr = NTDB_ATTRIBUTE_LOG;
 	log_attr.base.next = NULL;
-	log_attr.log.fn = tdb_log;
+	log_attr.log.fn = ntdb_log;
 
-	/* open the tdb */
-	tdb = tdb_open(fname, TDB_DEFAULT, O_RDONLY, 0, &log_attr);
+	/* open the ntdb */
+	ntdb = ntdb_open(fname, NTDB_DEFAULT, O_RDONLY, 0, &log_attr);
 
-	/* traverse the tdb, then close it */
-	if (tdb) {
-		count = tdb_traverse(tdb, test_fn, NULL);
-		tdb_close(tdb);
+	/* traverse the ntdb, then close it */
+	if (ntdb) {
+		count = ntdb_traverse(ntdb, test_fn, NULL);
+		ntdb_close(ntdb);
 	}
 
 	/* count is < 0 means an error */
 	if (count < 0) {
 		printf("restoring %s\n", fname);
-		return backup_tdb(bak_name, fname);
+		return backup_ntdb(bak_name, fname);
 	}
 
 	printf("%s : %d records\n", fname, count);
@@ -278,7 +278,7 @@ static int file_newer(const char *fname1, const char *fname2)
 
 static void usage(void)
 {
-	printf("Usage: tdb2backup [options] <fname...>\n\n");
+	printf("Usage: ntdbbackup [options] <fname...>\n\n");
 	printf("   -h            this help message\n");
 	printf("   -v            verify mode (restore if corrupt)\n");
 	printf("   -s suffix     set the backup suffix\n");
@@ -323,12 +323,12 @@ static void usage(void)
 		bak_name = add_suffix(fname, suffix);
 
 		if (verify) {
-			if (verify_tdb(fname, bak_name) != 0) {
+			if (verify_ntdb(fname, bak_name) != 0) {
 				ret = 1;
 			}
 		} else {
 			if (file_newer(fname, bak_name) &&
-			    backup_tdb(fname, bak_name) != 0) {
+			    backup_ntdb(fname, bak_name) != 0) {
 				ret = 1;
 			}
 		}

lib/ntdb/tools/ntdbdump.c

@@ -1,5 +1,5 @@
 /*
-   simple tdb2 dump util
+   simple ntdb dump util
    Copyright (C) Andrew Tridgell              2001
    Copyright (C) Rusty Russell                2011
 
@@ -17,7 +17,7 @@
    along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #ifdef HAVE_LIBREPLACE
 #include <replace.h>
 #include <system/filesys.h>
@@ -32,7 +32,7 @@
 #include <unistd.h>
 #endif
 
-static void print_data(TDB_DATA d)
+static void print_data(NTDB_DATA d)
 {
 	unsigned char *p = (unsigned char *)d.dptr;
 	int len = d.dsize;
@@ -46,7 +46,7 @@ static void print_data(TDB_DATA d)
 	}
 }
 
-static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
+static int traverse_fn(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state)
 {
 	printf("{\n");
 	printf("key(%d) = \"", (int)key.dsize);
@@ -59,22 +59,22 @@ static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, voi
 	return 0;
 }
 
-static int dump_tdb(const char *fname, const char *keyname)
+static int dump_ntdb(const char *fname, const char *keyname)
 {
-	struct tdb_context *tdb;
-	TDB_DATA key, value;
+	struct ntdb_context *ntdb;
+	NTDB_DATA key, value;
 
-	tdb = tdb_open(fname, 0, O_RDONLY, 0, NULL);
-	if (!tdb) {
+	ntdb = ntdb_open(fname, 0, O_RDONLY, 0, NULL);
+	if (!ntdb) {
 		printf("Failed to open %s\n", fname);
 		return 1;
 	}
 
 	if (!keyname) {
-		tdb_traverse(tdb, traverse_fn, NULL);
+		ntdb_traverse(ntdb, traverse_fn, NULL);
 	} else {
-		key = tdb_mkdata(keyname, strlen(keyname));
-		if (tdb_fetch(tdb, key, &value) != 0) {
+		key = ntdb_mkdata(keyname, strlen(keyname));
+		if (ntdb_fetch(ntdb, key, &value) != 0) {
 			return 1;
 		} else {
 			print_data(value);
@@ -87,7 +87,7 @@ static int dump_tdb(const char *fname, const char *keyname)
 
 static void usage( void)
 {
-	printf( "Usage: tdb2dump [options] <filename>\n\n");
+	printf( "Usage: ntdbdump [options] <filename>\n\n");
 	printf( "   -h          this help message\n");
 	printf( "   -k keyname  dumps value of keyname\n");
 }
@@ -98,7 +98,7 @@ static void usage( void)
 	int c;
 
 	if (argc < 2) {
-		printf("Usage: tdb2dump <fname>\n");
+		printf("Usage: ntdbdump <fname>\n");
 		exit(1);
 	}
 
@@ -118,5 +118,5 @@ static void usage( void)
 
 	fname = argv[optind];
 
-	return dump_tdb(fname, keyname);
+	return dump_ntdb(fname, keyname);
 }

lib/ntdb/tools/ntdbrestore.c

@@ -1,5 +1,5 @@
 /*
-   tdb2restore -- construct a tdb from tdbdump output.
+   ntdbrestore -- construct a ntdb from tdbdump output.
    Copyright (C) Volker Lendecke		2010
    Copyright (C) Simon McVittie			2005
 
@@ -18,7 +18,7 @@
 */
 
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include <assert.h>
 #ifdef HAVE_LIBREPLACE
 #include <replace.h>
@@ -88,7 +88,7 @@ static int read_hex(void) {
 	}
 }
 
-static int read_data(FILE *f, struct tdb_data *d, size_t size) {
+static int read_data(FILE *f, NTDB_DATA *d, size_t size) {
 	int c, low, high;
 	int i;
 
@@ -141,12 +141,12 @@ static int swallow(FILE *f, const char *s, int *eof)
 	return 0;
 }
 
-static bool read_rec(FILE *f, struct tdb_context *tdb, int *eof)
+static bool read_rec(FILE *f, struct ntdb_context *ntdb, int *eof)
 {
 	int length;
-	struct tdb_data key, data;
+	NTDB_DATA key, data;
 	bool ret = false;
-	enum TDB_ERROR e;
+	enum NTDB_ERROR e;
 
 	key.dptr = NULL;
 	data.dptr = NULL;
@@ -175,9 +175,9 @@ static bool read_rec(FILE *f, struct tdb_context *tdb, int *eof)
 	    || (swallow(f, "}\n", NULL) == -1)) {
 		goto fail;
 	}
-	e = tdb_store(tdb, key, data, TDB_INSERT);
-	if (e != TDB_SUCCESS) {
-		fprintf(stderr, "TDB error: %s\n", tdb_errorstr(e));
+	e = ntdb_store(ntdb, key, data, NTDB_INSERT);
+	if (e != NTDB_SUCCESS) {
+		fprintf(stderr, "NTDB error: %s\n", ntdb_errorstr(e));
 		goto fail;
 	}
 
@@ -188,28 +188,28 @@ fail:
 	return ret;
 }
 
-static int restore_tdb(const char *fname)
+static int restore_ntdb(const char *fname)
 {
-	struct tdb_context *tdb;
+	struct ntdb_context *ntdb;
 
-	tdb = tdb_open(fname, 0, O_RDWR|O_CREAT|O_EXCL, 0666, NULL);
-	if (!tdb) {
-		perror("tdb_open");
+	ntdb = ntdb_open(fname, 0, O_RDWR|O_CREAT|O_EXCL, 0666, NULL);
+	if (!ntdb) {
+		perror("ntdb_open");
 		fprintf(stderr, "Failed to open %s\n", fname);
 		return 1;
 	}
 
 	while (1) {
 		int eof = 0;
-		if (!read_rec(stdin, tdb, &eof)) {
+		if (!read_rec(stdin, ntdb, &eof)) {
 			if (eof) {
 				break;
 			}
 			return 1;
 		}
 	}
-	if (tdb_close(tdb)) {
-		fprintf(stderr, "Error closing tdb\n");
+	if (ntdb_close(ntdb)) {
+		fprintf(stderr, "Error closing ntdb\n");
 		return 1;
 	}
 	fprintf(stderr, "EOF\n");
@@ -227,5 +227,5 @@ int main(int argc, char *argv[])
 
 	fname = argv[1];
 
-	return restore_tdb(fname);
+	return restore_ntdb(fname);
 }

lib/ntdb/tools/ntdbtool.c

@@ -21,7 +21,7 @@
 */
 
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #ifdef HAVE_LIBREPLACE
 #include <replace.h>
 #include <system/filesys.h>
@@ -46,13 +46,13 @@ char *arg1, *arg2;
 size_t arg1len, arg2len;
 int bIterate = 0;
 char *line;
-TDB_DATA iterate_kbuf;
+NTDB_DATA iterate_kbuf;
 char cmdline[1024];
 static int disable_mmap;
 
 enum commands {
-	CMD_CREATE_TDB,
-	CMD_OPEN_TDB,
+	CMD_CREATE_NTDB,
+	CMD_OPEN_NTDB,
 	CMD_TRANSACTION_START,
 	CMD_TRANSACTION_COMMIT,
 	CMD_TRANSACTION_CANCEL,
@@ -86,8 +86,8 @@ typedef struct {
 } COMMAND_TABLE;
 
 COMMAND_TABLE cmd_table[] = {
-	{"create",	CMD_CREATE_TDB},
-	{"open",	CMD_OPEN_TDB},
+	{"create",	CMD_CREATE_NTDB},
+	{"open",	CMD_OPEN_NTDB},
 #if 0
 	{"transaction_start",	CMD_TRANSACTION_START},
 	{"transaction_commit",	CMD_TRANSACTION_COMMIT},
@@ -134,23 +134,23 @@ static double _end_timer(void)
 	       (tp2.tv_usec - tp1.tv_usec)*1.0e-6);
 }
 
-static void tdb_log(struct tdb_context *tdb,
-		    enum tdb_log_level level,
-		    enum TDB_ERROR ecode,
+static void ntdb_log(struct ntdb_context *ntdb,
+		    enum ntdb_log_level level,
+		    enum NTDB_ERROR ecode,
 		    const char *message,
 		    void *data)
 {
-	fprintf(stderr, "tdb:%s:%s:%s\n",
-		tdb_name(tdb), tdb_errorstr(ecode), message);
+	fprintf(stderr, "ntdb:%s:%s:%s\n",
+		ntdb_name(ntdb), ntdb_errorstr(ecode), message);
 }
 
-/* a tdb tool for manipulating a tdb database */
+/* a ntdb tool for manipulating a ntdb database */
 
-static struct tdb_context *tdb;
+static struct ntdb_context *ntdb;
 
-static int print_rec(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state);
-static int print_key(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state);
-static int print_hexkey(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state);
+static int print_rec(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state);
+static int print_key(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state);
+static int print_hexkey(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state);
 
 static void print_asc(const char *buf,int len)
 {
@@ -214,7 +214,7 @@ static void help(void)
 "  hexkeys              : dump the database keys as hex values\n"
 "  info                 : print summary info about the database\n"
 "  insert    key  data  : insert a record\n"
-"  move      key  file  : move a record to a destination tdb\n"
+"  move      key  file  : move a record to a destination ntdb\n"
 "  store     key  data  : store a record (replace)\n"
 "  show      key        : show a record by key\n"
 "  delete    key        : delete a record by key\n"
@@ -232,51 +232,51 @@ static void help(void)
 "\n");
 }
 
-static void terror(enum TDB_ERROR err, const char *why)
+static void terror(enum NTDB_ERROR err, const char *why)
 {
-	if (err != TDB_SUCCESS)
-		printf("%s:%s\n", tdb_errorstr(err), why);
+	if (err != NTDB_SUCCESS)
+		printf("%s:%s\n", ntdb_errorstr(err), why);
 	else
 		printf("%s\n", why);
 }
 
-static void create_tdb(const char *tdbname)
+static void create_ntdb(const char *tdbname)
 {
-	union tdb_attribute log_attr;
-	log_attr.base.attr = TDB_ATTRIBUTE_LOG;
+	union ntdb_attribute log_attr;
+	log_attr.base.attr = NTDB_ATTRIBUTE_LOG;
 	log_attr.base.next = NULL;
-	log_attr.log.fn = tdb_log;
+	log_attr.log.fn = ntdb_log;
 
-	if (tdb) tdb_close(tdb);
-	tdb = tdb_open(tdbname, (disable_mmap?TDB_NOMMAP:0),
+	if (ntdb) ntdb_close(ntdb);
+	ntdb = ntdb_open(tdbname, (disable_mmap?NTDB_NOMMAP:0),
 		       O_RDWR | O_CREAT | O_TRUNC, 0600, &log_attr);
-	if (!tdb) {
+	if (!ntdb) {
 		printf("Could not create %s: %s\n", tdbname, strerror(errno));
 	}
 }
 
-static void open_tdb(const char *tdbname)
+static void open_ntdb(const char *tdbname)
 {
-	union tdb_attribute log_attr;
-	log_attr.base.attr = TDB_ATTRIBUTE_LOG;
+	union ntdb_attribute log_attr;
+	log_attr.base.attr = NTDB_ATTRIBUTE_LOG;
 	log_attr.base.next = NULL;
-	log_attr.log.fn = tdb_log;
+	log_attr.log.fn = ntdb_log;
 
-	if (tdb) tdb_close(tdb);
-	tdb = tdb_open(tdbname, disable_mmap?TDB_NOMMAP:0, O_RDWR, 0600,
+	if (ntdb) ntdb_close(ntdb);
+	ntdb = ntdb_open(tdbname, disable_mmap?NTDB_NOMMAP:0, O_RDWR, 0600,
 		       &log_attr);
-	if (!tdb) {
+	if (!ntdb) {
 		printf("Could not open %s: %s\n", tdbname, strerror(errno));
 	}
 }
 
-static void insert_tdb(char *keyname, size_t keylen, char* data, size_t datalen)
+static void insert_ntdb(char *keyname, size_t keylen, char* data, size_t datalen)
 {
-	TDB_DATA key, dbuf;
-	enum TDB_ERROR ecode;
+	NTDB_DATA key, dbuf;
+	enum NTDB_ERROR ecode;
 
 	if ((keyname == NULL) || (keylen == 0)) {
-		terror(TDB_SUCCESS, "need key");
+		terror(NTDB_SUCCESS, "need key");
 		return;
 	}
 
@@ -285,24 +285,24 @@ static void insert_tdb(char *keyname, size_t keylen, char* data, size_t datalen)
 	dbuf.dptr = (unsigned char *)data;
 	dbuf.dsize = datalen;
 
-	ecode = tdb_store(tdb, key, dbuf, TDB_INSERT);
+	ecode = ntdb_store(ntdb, key, dbuf, NTDB_INSERT);
 	if (ecode) {
 		terror(ecode, "insert failed");
 	}
 }
 
-static void store_tdb(char *keyname, size_t keylen, char* data, size_t datalen)
+static void store_ntdb(char *keyname, size_t keylen, char* data, size_t datalen)
 {
-	TDB_DATA key, dbuf;
-	enum TDB_ERROR ecode;
+	NTDB_DATA key, dbuf;
+	enum NTDB_ERROR ecode;
 
 	if ((keyname == NULL) || (keylen == 0)) {
-		terror(TDB_SUCCESS, "need key");
+		terror(NTDB_SUCCESS, "need key");
 		return;
 	}
 
 	if ((data == NULL) || (datalen == 0)) {
-		terror(TDB_SUCCESS, "need data");
+		terror(NTDB_SUCCESS, "need data");
 		return;
 	}
 
@@ -312,52 +312,52 @@ static void store_tdb(char *keyname, size_t keylen, char* data, size_t datalen)
 	dbuf.dsize = datalen;
 
 	printf("Storing key:\n");
-	print_rec(tdb, key, dbuf, NULL);
+	print_rec(ntdb, key, dbuf, NULL);
 
-	ecode = tdb_store(tdb, key, dbuf, TDB_REPLACE);
+	ecode = ntdb_store(ntdb, key, dbuf, NTDB_REPLACE);
 	if (ecode) {
 		terror(ecode, "store failed");
 	}
 }
 
-static void show_tdb(char *keyname, size_t keylen)
+static void show_ntdb(char *keyname, size_t keylen)
 {
-	TDB_DATA key, dbuf;
-	enum TDB_ERROR ecode;
+	NTDB_DATA key, dbuf;
+	enum NTDB_ERROR ecode;
 
 	if ((keyname == NULL) || (keylen == 0)) {
-		terror(TDB_SUCCESS, "need key");
+		terror(NTDB_SUCCESS, "need key");
 		return;
 	}
 
 	key.dptr = (unsigned char *)keyname;
 	key.dsize = keylen;
 
-	ecode = tdb_fetch(tdb, key, &dbuf);
+	ecode = ntdb_fetch(ntdb, key, &dbuf);
 	if (ecode) {
 		terror(ecode, "fetch failed");
 		return;
 	}
 
-	print_rec(tdb, key, dbuf, NULL);
+	print_rec(ntdb, key, dbuf, NULL);
 
 	free( dbuf.dptr );
 }
 
-static void delete_tdb(char *keyname, size_t keylen)
+static void delete_ntdb(char *keyname, size_t keylen)
 {
-	TDB_DATA key;
-	enum TDB_ERROR ecode;
+	NTDB_DATA key;
+	enum NTDB_ERROR ecode;
 
 	if ((keyname == NULL) || (keylen == 0)) {
-		terror(TDB_SUCCESS, "need key");
+		terror(NTDB_SUCCESS, "need key");
 		return;
 	}
 
 	key.dptr = (unsigned char *)keyname;
 	key.dsize = keylen;
 
-	ecode = tdb_delete(tdb, key);
+	ecode = ntdb_delete(ntdb, key);
 	if (ecode) {
 		terror(ecode, "delete failed");
 	}
@@ -365,47 +365,47 @@ static void delete_tdb(char *keyname, size_t keylen)
 
 static void move_rec(char *keyname, size_t keylen, char* tdbname)
 {
-	TDB_DATA key, dbuf;
-	struct tdb_context *dst_tdb;
-	enum TDB_ERROR ecode;
+	NTDB_DATA key, dbuf;
+	struct ntdb_context *dst_ntdb;
+	enum NTDB_ERROR ecode;
 
 	if ((keyname == NULL) || (keylen == 0)) {
-		terror(TDB_SUCCESS, "need key");
+		terror(NTDB_SUCCESS, "need key");
 		return;
 	}
 
 	if ( !tdbname ) {
-		terror(TDB_SUCCESS, "need destination tdb name");
+		terror(NTDB_SUCCESS, "need destination ntdb name");
 		return;
 	}
 
 	key.dptr = (unsigned char *)keyname;
 	key.dsize = keylen;
 
-	ecode = tdb_fetch(tdb, key, &dbuf);
+	ecode = ntdb_fetch(ntdb, key, &dbuf);
 	if (ecode) {
 		terror(ecode, "fetch failed");
 		return;
 	}
 
-	print_rec(tdb, key, dbuf, NULL);
+	print_rec(ntdb, key, dbuf, NULL);
 
-	dst_tdb = tdb_open(tdbname, 0, O_RDWR, 0600, NULL);
-	if ( !dst_tdb ) {
-		terror(TDB_SUCCESS, "unable to open destination tdb");
+	dst_ntdb = ntdb_open(tdbname, 0, O_RDWR, 0600, NULL);
+	if ( !dst_ntdb ) {
+		terror(NTDB_SUCCESS, "unable to open destination ntdb");
 		return;
 	}
 
-	ecode = tdb_store( dst_tdb, key, dbuf, TDB_REPLACE);
+	ecode = ntdb_store( dst_ntdb, key, dbuf, NTDB_REPLACE);
 	if (ecode)
 		terror(ecode, "failed to move record");
 	else
 		printf("record moved\n");
 
-	tdb_close( dst_tdb );
+	ntdb_close( dst_ntdb );
 }
 
-static int print_rec(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
+static int print_rec(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state)
 {
 	printf("\nkey %d bytes\n", (int)key.dsize);
 	print_asc((const char *)key.dptr, key.dsize);
@@ -414,7 +414,7 @@ static int print_rec(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, v
 	return 0;
 }
 
-static int print_key(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
+static int print_key(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state)
 {
 	printf("key %d bytes: ", (int)key.dsize);
 	print_asc((const char *)key.dptr, key.dsize);
@@ -422,7 +422,7 @@ static int print_key(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, v
 	return 0;
 }
 
-static int print_hexkey(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
+static int print_hexkey(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state)
 {
 	printf("key %d bytes\n", (int)key.dsize);
 	print_data((const char *)key.dptr, key.dsize);
@@ -432,18 +432,18 @@ static int print_hexkey(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf
 
 static int total_bytes;
 
-static int traverse_fn(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
+static int traverse_fn(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf, void *state)
 {
 	total_bytes += dbuf.dsize;
 	return 0;
 }
 
-static void info_tdb(void)
+static void info_ntdb(void)
 {
-	enum TDB_ERROR ecode;
+	enum NTDB_ERROR ecode;
 	char *summary;
 
-	ecode = tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &summary);
+	ecode = ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &summary);
 
 	if (ecode) {
 		terror(ecode, "Getting summary");
@@ -453,7 +453,7 @@ static void info_tdb(void)
 	}
 }
 
-static void speed_tdb(const char *tlimit)
+static void speed_ntdb(const char *tlimit)
 {
 	unsigned timelimit = tlimit?atoi(tlimit):0;
 	double t;
@@ -465,11 +465,11 @@ static void speed_tdb(const char *tlimit)
 	_start_timer();
 	do {
 		long int r = random();
-		TDB_DATA key, dbuf;
-		key = tdb_mkdata("store test", strlen("store test"));
+		NTDB_DATA key, dbuf;
+		key = ntdb_mkdata("store test", strlen("store test"));
 		dbuf.dptr = (unsigned char *)&r;
 		dbuf.dsize = sizeof(r);
-		tdb_store(tdb, key, dbuf, TDB_REPLACE);
+		ntdb_store(ntdb, key, dbuf, NTDB_REPLACE);
 		t = _end_timer();
 		ops++;
 	} while (t < timelimit);
@@ -480,11 +480,11 @@ static void speed_tdb(const char *tlimit)
 	_start_timer();
 	do {
 		long int r = random();
-		TDB_DATA key, dbuf;
-		key = tdb_mkdata("store test", strlen("store test"));
+		NTDB_DATA key, dbuf;
+		key = ntdb_mkdata("store test", strlen("store test"));
 		dbuf.dptr = (unsigned char *)&r;
 		dbuf.dsize = sizeof(r);
-		tdb_fetch(tdb, key, &dbuf);
+		ntdb_fetch(ntdb, key, &dbuf);
 		t = _end_timer();
 		ops++;
 	} while (t < timelimit);
@@ -495,13 +495,13 @@ static void speed_tdb(const char *tlimit)
 	_start_timer();
 	do {
 		long int r = random();
-		TDB_DATA key, dbuf;
-		key = tdb_mkdata("transaction test", strlen("transaction test"));
+		NTDB_DATA key, dbuf;
+		key = ntdb_mkdata("transaction test", strlen("transaction test"));
 		dbuf.dptr = (unsigned char *)&r;
 		dbuf.dsize = sizeof(r);
-		tdb_transaction_start(tdb);
-		tdb_store(tdb, key, dbuf, TDB_REPLACE);
-		tdb_transaction_commit(tdb);
+		ntdb_transaction_start(ntdb);
+		ntdb_store(ntdb, key, dbuf, NTDB_REPLACE);
+		ntdb_transaction_commit(ntdb);
 		t = _end_timer();
 		ops++;
 	} while (t < timelimit);
@@ -511,7 +511,7 @@ static void speed_tdb(const char *tlimit)
 	printf("Testing traverse speed for %u seconds\n", timelimit);
 	_start_timer();
 	do {
-		tdb_traverse(tdb, traverse_fn, NULL);
+		ntdb_traverse(ntdb, traverse_fn, NULL);
 		t = _end_timer();
 		ops++;
 	} while (t < timelimit);
@@ -528,7 +528,7 @@ static void toggle_mmap(void)
 	}
 }
 
-static char *tdb_getline(const char *prompt)
+static char *ntdb_getline(const char *prompt)
 {
 	static char thisline[1024];
 	char *p;
@@ -540,45 +540,45 @@ static char *tdb_getline(const char *prompt)
 	return p?thisline:NULL;
 }
 
-static int do_delete_fn(struct tdb_context *the_tdb, TDB_DATA key, TDB_DATA dbuf,
+static int do_delete_fn(struct ntdb_context *the_ntdb, NTDB_DATA key, NTDB_DATA dbuf,
                      void *state)
 {
-    return tdb_delete(the_tdb, key);
+    return ntdb_delete(the_ntdb, key);
 }
 
-static void first_record(struct tdb_context *the_tdb, TDB_DATA *pkey)
+static void first_record(struct ntdb_context *the_ntdb, NTDB_DATA *pkey)
 {
-	TDB_DATA dbuf;
-	enum TDB_ERROR ecode;
-	ecode = tdb_firstkey(the_tdb, pkey);
+	NTDB_DATA dbuf;
+	enum NTDB_ERROR ecode;
+	ecode = ntdb_firstkey(the_ntdb, pkey);
 	if (!ecode)
-		ecode = tdb_fetch(the_tdb, *pkey, &dbuf);
+		ecode = ntdb_fetch(the_ntdb, *pkey, &dbuf);
 	if (ecode) terror(ecode, "fetch failed");
 	else {
-		print_rec(the_tdb, *pkey, dbuf, NULL);
+		print_rec(the_ntdb, *pkey, dbuf, NULL);
 	}
 }
 
-static void next_record(struct tdb_context *the_tdb, TDB_DATA *pkey)
+static void next_record(struct ntdb_context *the_ntdb, NTDB_DATA *pkey)
 {
-	TDB_DATA dbuf;
-	enum TDB_ERROR ecode;
-	ecode = tdb_nextkey(the_tdb, pkey);
+	NTDB_DATA dbuf;
+	enum NTDB_ERROR ecode;
+	ecode = ntdb_nextkey(the_ntdb, pkey);
 
 	if (!ecode)
-		ecode = tdb_fetch(the_tdb, *pkey, &dbuf);
+		ecode = ntdb_fetch(the_ntdb, *pkey, &dbuf);
 	if (ecode)
 		terror(ecode, "fetch failed");
 	else
-		print_rec(the_tdb, *pkey, dbuf, NULL);
+		print_rec(the_ntdb, *pkey, dbuf, NULL);
 }
 
-static void check_db(struct tdb_context *the_tdb)
+static void check_db(struct ntdb_context *the_ntdb)
 {
-	if (!the_tdb) {
+	if (!the_ntdb) {
 		printf("Error: No database opened!\n");
 	} else {
-		if (tdb_check(the_tdb, NULL, NULL) != 0)
+		if (ntdb_check(the_ntdb, NULL, NULL) != 0)
 			printf("Integrity check for the opened database failed.\n");
 		else
 			printf("Database integrity is OK.\n");
@@ -605,54 +605,54 @@ static int do_command(void)
 	}
 
 	switch (mycmd) {
-	case CMD_CREATE_TDB:
+	case CMD_CREATE_NTDB:
 		bIterate = 0;
-		create_tdb(arg1);
+		create_ntdb(arg1);
 		return 0;
-	case CMD_OPEN_TDB:
+	case CMD_OPEN_NTDB:
 		bIterate = 0;
-		open_tdb(arg1);
+		open_ntdb(arg1);
 		return 0;
 	case CMD_SYSTEM:
 		/* Shell command */
 		if (system(arg1) == -1) {
-			terror(TDB_SUCCESS, "system() call failed\n");
+			terror(NTDB_SUCCESS, "system() call failed\n");
 		}
 		return 0;
 	case CMD_QUIT:
 		return 1;
 	default:
 		/* all the rest require a open database */
-		if (!tdb) {
+		if (!ntdb) {
 			bIterate = 0;
-			terror(TDB_SUCCESS, "database not open");
+			terror(NTDB_SUCCESS, "database not open");
 			help();
 			return 0;
 		}
 		switch (mycmd) {
 		case CMD_TRANSACTION_START:
 			bIterate = 0;
-			tdb_transaction_start(tdb);
+			ntdb_transaction_start(ntdb);
 			return 0;
 		case CMD_TRANSACTION_COMMIT:
 			bIterate = 0;
-			tdb_transaction_commit(tdb);
+			ntdb_transaction_commit(ntdb);
 			return 0;
 		case CMD_TRANSACTION_CANCEL:
 			bIterate = 0;
-			tdb_transaction_cancel(tdb);
+			ntdb_transaction_cancel(ntdb);
 			return 0;
 		case CMD_ERASE:
 			bIterate = 0;
-			tdb_traverse(tdb, do_delete_fn, NULL);
+			ntdb_traverse(ntdb, do_delete_fn, NULL);
 			return 0;
 		case CMD_DUMP:
 			bIterate = 0;
-			tdb_traverse(tdb, print_rec, NULL);
+			ntdb_traverse(ntdb, print_rec, NULL);
 			return 0;
 		case CMD_INSERT:
 			bIterate = 0;
-			insert_tdb(arg1, arg1len,arg2,arg2len);
+			insert_ntdb(arg1, arg1len,arg2,arg2len);
 			return 0;
 		case CMD_MOVE:
 			bIterate = 0;
@@ -660,55 +660,55 @@ static int do_command(void)
 			return 0;
 		case CMD_STORE:
 			bIterate = 0;
-			store_tdb(arg1,arg1len,arg2,arg2len);
+			store_ntdb(arg1,arg1len,arg2,arg2len);
 			return 0;
 		case CMD_SHOW:
 			bIterate = 0;
-			show_tdb(arg1, arg1len);
+			show_ntdb(arg1, arg1len);
 			return 0;
 		case CMD_KEYS:
-			tdb_traverse(tdb, print_key, NULL);
+			ntdb_traverse(ntdb, print_key, NULL);
 			return 0;
 		case CMD_HEXKEYS:
-			tdb_traverse(tdb, print_hexkey, NULL);
+			ntdb_traverse(ntdb, print_hexkey, NULL);
 			return 0;
 		case CMD_DELETE:
 			bIterate = 0;
-			delete_tdb(arg1,arg1len);
+			delete_ntdb(arg1,arg1len);
 			return 0;
 #if 0
 		case CMD_LIST_HASH_FREE:
-			tdb_dump_all(tdb);
+			ntdb_dump_all(ntdb);
 			return 0;
 		case CMD_LIST_FREE:
-			tdb_printfreelist(tdb);
+			ntdb_printfreelist(ntdb);
 			return 0;
 #endif
 		case CMD_INFO:
-			info_tdb();
+			info_ntdb();
 			return 0;
 		case CMD_SPEED:
-			speed_tdb(arg1);
+			speed_ntdb(arg1);
 			return 0;
 		case CMD_MMAP:
 			toggle_mmap();
 			return 0;
 		case CMD_FIRST:
 			bIterate = 1;
-			first_record(tdb, &iterate_kbuf);
+			first_record(ntdb, &iterate_kbuf);
 			return 0;
 		case CMD_NEXT:
 			if (bIterate)
-				next_record(tdb, &iterate_kbuf);
+				next_record(ntdb, &iterate_kbuf);
 			return 0;
 		case CMD_CHECK:
-			check_db(tdb);
+			check_db(ntdb);
 			return 0;
 		case CMD_HELP:
 			help();
 			return 0;
-		case CMD_CREATE_TDB:
-		case CMD_OPEN_TDB:
+		case CMD_CREATE_NTDB:
+		case CMD_OPEN_NTDB:
 		case CMD_SYSTEM:
 		case CMD_QUIT:
 			/*
@@ -773,7 +773,7 @@ int main(int argc, char *argv[])
 	case 1:
 	case 2:
 		/* Interactive mode */
-		while ((cmdname = tdb_getline("tdb> "))) {
+		while ((cmdname = ntdb_getline("ntdb> "))) {
 			arg2 = arg1 = NULL;
 			if ((arg1 = strchr((const char *)cmdname,' ')) != NULL) {
 				arg1++;
@@ -804,7 +804,7 @@ int main(int argc, char *argv[])
 		break;
 	}
 
-	if (tdb) tdb_close(tdb);
+	if (ntdb) ntdb_close(ntdb);
 
 	return 0;
 }

lib/ntdb/tools/ntdbtorture.c

@@ -1,9 +1,9 @@
-/* this tests tdb by doing lots of ops from several simultaneous
+/* this tests ntdb by doing lots of ops from several simultaneous
    writers - that stresses the locking code.
 */
 
 #include "config.h"
-#include "tdb2.h"
+#include "ntdb.h"
 #include <ccan/err/err.h>
 #ifdef HAVE_LIBREPLACE
 #include <replace.h>
@@ -35,7 +35,7 @@
 #define KEYLEN 3
 #define DATALEN 100
 
-static struct tdb_context *db;
+static struct ntdb_context *db;
 static int in_transaction;
 static int in_traverse;
 static int error_count;
@@ -44,17 +44,17 @@ static int always_transaction = 0;
 #endif
 static int loopnum;
 static int count_pipe;
-static union tdb_attribute log_attr;
-static union tdb_attribute seed_attr;
+static union ntdb_attribute log_attr;
+static union ntdb_attribute seed_attr;
 
-static void tdb_log(struct tdb_context *tdb,
-		    enum tdb_log_level level,
-		    enum TDB_ERROR ecode,
+static void ntdb_log(struct ntdb_context *ntdb,
+		    enum ntdb_log_level level,
+		    enum NTDB_ERROR ecode,
 		    const char *message,
 		    void *data)
 {
-	printf("tdb:%s:%s:%s\n",
-	       tdb_name(tdb), tdb_errorstr(ecode), message);
+	printf("ntdb:%s:%s:%s\n",
+	       ntdb_name(ntdb), ntdb_errorstr(ecode), message);
 	fflush(stdout);
 #if 0
 	{
@@ -80,10 +80,10 @@ static void segv_handler(int sig, siginfo_t *info, void *p)
 	_exit(11);
 }
 
-static void fatal(struct tdb_context *tdb, const char *why)
+static void fatal(struct ntdb_context *ntdb, const char *why)
 {
 	fprintf(stderr, "%u:%s:%s\n", getpid(), why,
-		tdb ? tdb_errorstr(tdb_error(tdb)) : "(no tdb)");
+		ntdb ? ntdb_errorstr(ntdb_error(ntdb)) : "(no ntdb)");
 	error_count++;
 }
 
@@ -101,12 +101,12 @@ static char *randbuf(int len)
 }
 
 static void addrec_db(void);
-static int modify_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
+static int modify_traverse(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf,
 			   void *state)
 {
 #if CULL_PROB
 	if (random() % CULL_PROB == 0) {
-		tdb_delete(tdb, key);
+		ntdb_delete(ntdb, key);
 	}
 #endif
 
@@ -128,7 +128,7 @@ static void addrec_db(void)
 {
 	int klen, dlen;
 	char *k, *d;
-	TDB_DATA key, data;
+	NTDB_DATA key, data;
 
 	klen = 1 + (rand() % KEYLEN);
 	dlen = 1 + (rand() % DATALEN);
@@ -144,34 +144,34 @@ static void addrec_db(void)
 
 #if REOPEN_PROB
 	if (in_traverse == 0 && in_transaction == 0 && random() % REOPEN_PROB == 0) {
-		tdb_reopen_all(0);
+		ntdb_reopen_all(0);
 		goto next;
 	}
 #endif
 
 #if TRANSACTION_PROB
 	if (in_traverse == 0 && in_transaction == 0 && (always_transaction || random() % TRANSACTION_PROB == 0)) {
-		if (tdb_transaction_start(db) != 0) {
-			fatal(db, "tdb_transaction_start failed");
+		if (ntdb_transaction_start(db) != 0) {
+			fatal(db, "ntdb_transaction_start failed");
 		}
 		in_transaction++;
 		goto next;
 	}
 	if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
 		if (random() % TRANSACTION_PREPARE_PROB == 0) {
-			if (tdb_transaction_prepare_commit(db) != 0) {
-				fatal(db, "tdb_transaction_prepare_commit failed");
+			if (ntdb_transaction_prepare_commit(db) != 0) {
+				fatal(db, "ntdb_transaction_prepare_commit failed");
 			}
 		}
-		if (tdb_transaction_commit(db) != 0) {
-			fatal(db, "tdb_transaction_commit failed");
+		if (ntdb_transaction_commit(db) != 0) {
+			fatal(db, "ntdb_transaction_commit failed");
 		}
 		in_transaction--;
 		goto next;
 	}
 
 	if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
-		tdb_transaction_cancel(db);
+		ntdb_transaction_cancel(db);
 		in_transaction--;
 		goto next;
 	}
@@ -179,15 +179,15 @@ static void addrec_db(void)
 
 #if DELETE_PROB
 	if (random() % DELETE_PROB == 0) {
-		tdb_delete(db, key);
+		ntdb_delete(db, key);
 		goto next;
 	}
 #endif
 
 #if STORE_PROB
 	if (random() % STORE_PROB == 0) {
-		if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
-			fatal(db, "tdb_store failed");
+		if (ntdb_store(db, key, data, NTDB_REPLACE) != 0) {
+			fatal(db, "ntdb_store failed");
 		}
 		goto next;
 	}
@@ -195,8 +195,8 @@ static void addrec_db(void)
 
 #if APPEND_PROB
 	if (random() % APPEND_PROB == 0) {
-		if (tdb_append(db, key, data) != 0) {
-			fatal(db, "tdb_append failed");
+		if (ntdb_append(db, key, data) != 0) {
+			fatal(db, "ntdb_append failed");
 		}
 		goto next;
 	}
@@ -204,16 +204,16 @@ static void addrec_db(void)
 
 #if LOCKSTORE_PROB
 	if (random() % LOCKSTORE_PROB == 0) {
-		tdb_chainlock(db, key);
-		if (tdb_fetch(db, key, &data) != TDB_SUCCESS) {
+		ntdb_chainlock(db, key);
+		if (ntdb_fetch(db, key, &data) != NTDB_SUCCESS) {
 			data.dsize = 0;
 			data.dptr = NULL;
 		}
-		if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
-			fatal(db, "tdb_store failed");
+		if (ntdb_store(db, key, data, NTDB_REPLACE) != 0) {
+			fatal(db, "ntdb_store failed");
 		}
 		if (data.dptr) free(data.dptr);
-		tdb_chainunlock(db, key);
+		ntdb_chainunlock(db, key);
 		goto next;
 	}
 #endif
@@ -222,13 +222,13 @@ static void addrec_db(void)
 	/* FIXME: recursive traverses break transactions? */
 	if (in_traverse == 0 && random() % TRAVERSE_PROB == 0) {
 		in_traverse++;
-		tdb_traverse(db, modify_traverse, NULL);
+		ntdb_traverse(db, modify_traverse, NULL);
 		in_traverse--;
 		goto next;
 	}
 #endif
 
-	if (tdb_fetch(db, key, &data) == TDB_SUCCESS)
+	if (ntdb_fetch(db, key, &data) == NTDB_SUCCESS)
 		free(data.dptr);
 
 next:
@@ -236,16 +236,16 @@ next:
 	free(d);
 }
 
-static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
+static int traverse_fn(struct ntdb_context *ntdb, NTDB_DATA key, NTDB_DATA dbuf,
                        void *state)
 {
-	tdb_delete(tdb, key);
+	ntdb_delete(ntdb, key);
 	return 0;
 }
 
 static void usage(void)
 {
-	printf("Usage: tdb2torture"
+	printf("Usage: ntdbtorture"
 #if TRANSACTION_PROB
 	       " [-t]"
 #endif
@@ -263,13 +263,13 @@ static void send_count_and_suicide(int sig)
 }
 
 static int run_child(const char *filename, int i, int seed, unsigned num_loops,
-		     unsigned start, int tdb_flags)
+		     unsigned start, int ntdb_flags)
 {
 	struct sigaction act = { .sa_sigaction = segv_handler,
 				 .sa_flags = SA_SIGINFO };
 	sigaction(11, &act, NULL);
 
-	db = tdb_open(filename, tdb_flags, O_RDWR | O_CREAT, 0600,
+	db = ntdb_open(filename, ntdb_flags, O_RDWR | O_CREAT, 0600,
 		      &log_attr);
 	if (!db) {
 		fatal(NULL, "db open failed");
@@ -295,29 +295,29 @@ static int run_child(const char *filename, int i, int seed, unsigned num_loops,
 	}
 
 	if (error_count == 0) {
-		tdb_traverse(db, NULL, NULL);
+		ntdb_traverse(db, NULL, NULL);
 #if TRANSACTION_PROB
 		if (always_transaction) {
 			while (in_transaction) {
-				tdb_transaction_cancel(db);
+				ntdb_transaction_cancel(db);
 				in_transaction--;
 			}
-			if (tdb_transaction_start(db) != 0)
-				fatal(db, "tdb_transaction_start failed");
+			if (ntdb_transaction_start(db) != 0)
+				fatal(db, "ntdb_transaction_start failed");
 		}
 #endif
-		tdb_traverse(db, traverse_fn, NULL);
-		tdb_traverse(db, traverse_fn, NULL);
+		ntdb_traverse(db, traverse_fn, NULL);
+		ntdb_traverse(db, traverse_fn, NULL);
 
 #if TRANSACTION_PROB
 		if (always_transaction) {
-			if (tdb_transaction_commit(db) != 0)
-				fatal(db, "tdb_transaction_commit failed");
+			if (ntdb_transaction_commit(db) != 0)
+				fatal(db, "ntdb_transaction_commit failed");
 		}
 #endif
 	}
 
-	tdb_close(db);
+	ntdb_close(db);
 
 	return (error_count < 100 ? error_count : 100);
 }
@@ -350,13 +350,13 @@ int main(int argc, char * const *argv)
 	pid_t *pids;
 	int kill_random = 0;
 	int *done;
-	int tdb_flags = TDB_DEFAULT;
-	char *test_tdb;
+	int ntdb_flags = NTDB_DEFAULT;
+	char *test_ntdb;
 
-	log_attr.base.attr = TDB_ATTRIBUTE_LOG;
+	log_attr.base.attr = NTDB_ATTRIBUTE_LOG;
 	log_attr.base.next = &seed_attr;
-	log_attr.log.fn = tdb_log;
-	seed_attr.base.attr = TDB_ATTRIBUTE_SEED;
+	log_attr.log.fn = ntdb_log;
+	seed_attr.base.attr = NTDB_ATTRIBUTE_SEED;
 	seed_attr.base.next = NULL;
 
 	while ((c = getopt(argc, argv, "n:l:s:thkS")) != -1) {
@@ -371,7 +371,7 @@ int main(int argc, char * const *argv)
 			seed = strtol(optarg, NULL, 0);
 			break;
 		case 'S':
-			tdb_flags = TDB_NOSYNC;
+			ntdb_flags = NTDB_NOSYNC;
 			break;
 		case 't':
 #if TRANSACTION_PROB
@@ -389,9 +389,9 @@ int main(int argc, char * const *argv)
 		}
 	}
 
-	test_tdb = test_path("torture.tdb2");
+	test_ntdb = test_path("torture.ntdb");
 
-	unlink(test_tdb);
+	unlink(test_ntdb);
 
 	if (seed == -1) {
 		seed = (getpid() + time(NULL)) & 0x7FFFFFFF;
@@ -400,8 +400,8 @@ int main(int argc, char * const *argv)
 
 	if (num_procs == 1 && !kill_random) {
 		/* Don't fork for this case, makes debugging easier. */
-		error_count = run_child(test_tdb, 0, seed, num_loops, 0,
-					tdb_flags);
+		error_count = run_child(test_ntdb, 0, seed, num_loops, 0,
+					ntdb_flags);
 		goto done;
 	}
 
@@ -427,8 +427,8 @@ int main(int argc, char * const *argv)
 #endif
 					);
 			}
-			exit(run_child(test_tdb, i, seed, num_loops, 0,
-				       tdb_flags));
+			exit(run_child(test_ntdb, i, seed, num_loops, 0,
+				       ntdb_flags));
 		}
 	}
 
@@ -484,9 +484,9 @@ int main(int argc, char * const *argv)
 				}
 				pids[j] = fork();
 				if (pids[j] == 0)
-					exit(run_child(test_tdb, j, seed,
+					exit(run_child(test_ntdb, j, seed,
 						       num_loops, done[j],
-						       tdb_flags));
+						       ntdb_flags));
 				printf("Restarting child %i for %u-%u\n",
 				       j, done[j], num_loops);
 				continue;
@@ -510,20 +510,20 @@ int main(int argc, char * const *argv)
 
 done:
 	if (error_count == 0) {
-		db = tdb_open(test_tdb, TDB_DEFAULT, O_RDWR | O_CREAT,
+		db = ntdb_open(test_ntdb, NTDB_DEFAULT, O_RDWR | O_CREAT,
 			      0600, &log_attr);
 		if (!db) {
 			fatal(db, "db open failed");
 			exit(1);
 		}
-		if (tdb_check(db, NULL, NULL) != 0) {
+		if (ntdb_check(db, NULL, NULL) != 0) {
 			fatal(db, "db check failed");
 			exit(1);
 		}
-		tdb_close(db);
+		ntdb_close(db);
 		printf("OK\n");
 	}
 
-	free(test_tdb);
+	free(test_ntdb);
 	return error_count;
 }

lib/ntdb/tools/speed.c

@@ -1,4 +1,4 @@
-/* Simple speed test for TDB */
+/* Simple speed test for NTDB */
 #include <ccan/err/err.h>
 #include <time.h>
 #include <sys/types.h>
@@ -10,7 +10,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <stdbool.h>
-#include "tdb2.h"
+#include "ntdb.h"
 
 /* Nanoseconds per operation */
 static size_t normalize(const struct timeval *start,
@@ -30,31 +30,31 @@ static size_t file_size(void)
 {
 	struct stat st;
 
-	if (stat("/tmp/speed.tdb", &st) != 0)
+	if (stat("/tmp/speed.ntdb", &st) != 0)
 		return -1;
 	return st.st_size;
 }
 
-static int count_record(struct tdb_context *tdb,
-			TDB_DATA key, TDB_DATA data, void *p)
+static int count_record(struct ntdb_context *ntdb,
+			NTDB_DATA key, NTDB_DATA data, void *p)
 {
 	int *total = p;
 	*total += *(int *)data.dptr;
 	return 0;
 }
 
-static void dump_and_clear_stats(struct tdb_context **tdb,
+static void dump_and_clear_stats(struct ntdb_context **ntdb,
 				 int flags,
-				 union tdb_attribute *attr)
+				 union ntdb_attribute *attr)
 {
-	union tdb_attribute stats;
-	enum TDB_ERROR ecode;
+	union ntdb_attribute stats;
+	enum NTDB_ERROR ecode;
 
-	stats.base.attr = TDB_ATTRIBUTE_STATS;
+	stats.base.attr = NTDB_ATTRIBUTE_STATS;
 	stats.stats.size = sizeof(stats.stats);
-	ecode = tdb_get_attribute(*tdb, &stats);
-	if (ecode != TDB_SUCCESS)
-		errx(1, "Getting stats: %s", tdb_errorstr(ecode));
+	ecode = ntdb_get_attribute(*ntdb, &stats);
+	if (ecode != NTDB_SUCCESS)
+		errx(1, "Getting stats: %s", ntdb_errorstr(ecode));
 
 	printf("allocs = %llu\n",
 	       (unsigned long long)stats.stats.allocs);
@@ -122,43 +122,43 @@ static void dump_and_clear_stats(struct tdb_context **tdb,
 	       (unsigned long long)stats.stats.lock_nonblock_fail);
 
 	/* Now clear. */
-	tdb_close(*tdb);
-	*tdb = tdb_open("/tmp/speed.tdb", flags, O_RDWR, 0, attr);
+	ntdb_close(*ntdb);
+	*ntdb = ntdb_open("/tmp/speed.ntdb", flags, O_RDWR, 0, attr);
 }
 
-static void tdb_log(struct tdb_context *tdb,
-		    enum tdb_log_level level,
-		    enum TDB_ERROR ecode,
+static void ntdb_log(struct ntdb_context *ntdb,
+		    enum ntdb_log_level level,
+		    enum NTDB_ERROR ecode,
 		    const char *message,
 		    void *data)
 {
-	fprintf(stderr, "tdb:%s:%s:%s\n",
-		tdb_name(tdb), tdb_errorstr(ecode), message);
+	fprintf(stderr, "ntdb:%s:%s:%s\n",
+		ntdb_name(ntdb), ntdb_errorstr(ecode), message);
 }
 
 int main(int argc, char *argv[])
 {
 	unsigned int i, j, num = 1000, stage = 0, stopat = -1;
-	int flags = TDB_DEFAULT;
+	int flags = NTDB_DEFAULT;
 	bool transaction = false, summary = false;
-	TDB_DATA key, data;
-	struct tdb_context *tdb;
+	NTDB_DATA key, data;
+	struct ntdb_context *ntdb;
 	struct timeval start, stop;
-	union tdb_attribute seed, log;
+	union ntdb_attribute seed, log;
 	bool do_stats = false;
-	enum TDB_ERROR ecode;
+	enum NTDB_ERROR ecode;
 
 	/* Try to keep benchmarks even. */
-	seed.base.attr = TDB_ATTRIBUTE_SEED;
+	seed.base.attr = NTDB_ATTRIBUTE_SEED;
 	seed.base.next = NULL;
 	seed.seed.seed = 0;
 
-	log.base.attr = TDB_ATTRIBUTE_LOG;
+	log.base.attr = NTDB_ATTRIBUTE_LOG;
 	log.base.next = &seed;
-	log.log.fn = tdb_log;
+	log.log.fn = ntdb_log;
 
 	if (argv[1] && strcmp(argv[1], "--internal") == 0) {
-		flags = TDB_INTERNAL;
+		flags = NTDB_INTERNAL;
 		argc--;
 		argv++;
 	}
@@ -168,7 +168,7 @@ int main(int argc, char *argv[])
 		argv++;
 	}
 	if (argv[1] && strcmp(argv[1], "--no-sync") == 0) {
-		flags |= TDB_NOSYNC;
+		flags |= NTDB_NOSYNC;
 		argc--;
 		argv++;
 	}
@@ -183,10 +183,10 @@ int main(int argc, char *argv[])
 		argv++;
 	}
 
-	tdb = tdb_open("/tmp/speed.tdb", flags, O_RDWR|O_CREAT|O_TRUNC,
+	ntdb = ntdb_open("/tmp/speed.ntdb", flags, O_RDWR|O_CREAT|O_TRUNC,
 		       0600, &log);
-	if (!tdb)
-		err(1, "Opening /tmp/speed.tdb");
+	if (!ntdb)
+		err(1, "Opening /tmp/speed.ntdb");
 
 	key.dptr = (void *)&i;
 	key.dsize = sizeof(i);
@@ -206,199 +206,199 @@ int main(int argc, char *argv[])
 
 	/* Add 1000 records. */
 	printf("Adding %u records: ", num); fflush(stdout);
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	gettimeofday(&start, NULL);
 	for (i = 0; i < num; i++)
-		if ((ecode = tdb_store(tdb, key, data, TDB_INSERT)) != 0)
-			errx(1, "Inserting key %u in tdb: %s",
-			     i, tdb_errorstr(ecode));
+		if ((ecode = ntdb_store(ntdb, key, data, NTDB_INSERT)) != 0)
+			errx(1, "Inserting key %u in ntdb: %s",
+			     i, ntdb_errorstr(ecode));
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
 
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
 	if (do_stats)
-		dump_and_clear_stats(&tdb, flags, &log);
+		dump_and_clear_stats(&ntdb, flags, &log);
 
 	if (++stage == stopat)
 		exit(0);
 
 	/* Finding 1000 records. */
 	printf("Finding %u records: ", num); fflush(stdout);
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	gettimeofday(&start, NULL);
 	for (i = 0; i < num; i++) {
-		struct tdb_data dbuf;
-		if ((ecode = tdb_fetch(tdb, key, &dbuf)) != TDB_SUCCESS
+		NTDB_DATA dbuf;
+		if ((ecode = ntdb_fetch(ntdb, key, &dbuf)) != NTDB_SUCCESS
 		    || *(int *)dbuf.dptr != i) {
-			errx(1, "Fetching key %u in tdb gave %u",
+			errx(1, "Fetching key %u in ntdb gave %u",
 			     i, ecode ? ecode : *(int *)dbuf.dptr);
 		}
 	}
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
 	if (do_stats)
-		dump_and_clear_stats(&tdb, flags, &log);
+		dump_and_clear_stats(&ntdb, flags, &log);
 	if (++stage == stopat)
 		exit(0);
 
 	/* Missing 1000 records. */
 	printf("Missing %u records: ", num); fflush(stdout);
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	gettimeofday(&start, NULL);
 	for (i = num; i < num*2; i++) {
-		struct tdb_data dbuf;
-		ecode = tdb_fetch(tdb, key, &dbuf);
-		if (ecode != TDB_ERR_NOEXIST)
-			errx(1, "Fetching key %u in tdb gave %s",
-			     i, tdb_errorstr(ecode));
+		NTDB_DATA dbuf;
+		ecode = ntdb_fetch(ntdb, key, &dbuf);
+		if (ecode != NTDB_ERR_NOEXIST)
+			errx(1, "Fetching key %u in ntdb gave %s",
+			     i, ntdb_errorstr(ecode));
 	}
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
 	if (do_stats)
-		dump_and_clear_stats(&tdb, flags, &log);
+		dump_and_clear_stats(&ntdb, flags, &log);
 	if (++stage == stopat)
 		exit(0);
 
 	/* Traverse 1000 records. */
 	printf("Traversing %u records: ", num); fflush(stdout);
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	i = 0;
 	gettimeofday(&start, NULL);
-	if (tdb_traverse(tdb, count_record, &i) != num)
+	if (ntdb_traverse(ntdb, count_record, &i) != num)
 		errx(1, "Traverse returned wrong number of records");
 	if (i != (num - 1) * (num / 2))
 		errx(1, "Traverse tallied to %u", i);
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
 	if (do_stats)
-		dump_and_clear_stats(&tdb, flags, &log);
+		dump_and_clear_stats(&ntdb, flags, &log);
 	if (++stage == stopat)
 		exit(0);
 
 	/* Delete 1000 records (not in order). */
 	printf("Deleting %u records: ", num); fflush(stdout);
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	gettimeofday(&start, NULL);
 	for (j = 0; j < num; j++) {
 		i = (j + 100003) % num;
-		if ((ecode = tdb_delete(tdb, key)) != TDB_SUCCESS)
-			errx(1, "Deleting key %u in tdb: %s",
-			     i, tdb_errorstr(ecode));
+		if ((ecode = ntdb_delete(ntdb, key)) != NTDB_SUCCESS)
+			errx(1, "Deleting key %u in ntdb: %s",
+			     i, ntdb_errorstr(ecode));
 	}
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
 	if (do_stats)
-		dump_and_clear_stats(&tdb, flags, &log);
+		dump_and_clear_stats(&ntdb, flags, &log);
 	if (++stage == stopat)
 		exit(0);
 
 	/* Re-add 1000 records (not in order). */
 	printf("Re-adding %u records: ", num); fflush(stdout);
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	gettimeofday(&start, NULL);
 	for (j = 0; j < num; j++) {
 		i = (j + 100003) % num;
-		if ((ecode = tdb_store(tdb, key, data, TDB_INSERT)) != 0)
-			errx(1, "Inserting key %u in tdb: %s",
-			     i, tdb_errorstr(ecode));
+		if ((ecode = ntdb_store(ntdb, key, data, NTDB_INSERT)) != 0)
+			errx(1, "Inserting key %u in ntdb: %s",
+			     i, ntdb_errorstr(ecode));
 	}
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
 	if (do_stats)
-		dump_and_clear_stats(&tdb, flags, &log);
+		dump_and_clear_stats(&ntdb, flags, &log);
 	if (++stage == stopat)
 		exit(0);
 
 	/* Append 1000 records. */
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	printf("Appending %u records: ", num); fflush(stdout);
 	gettimeofday(&start, NULL);
 	for (i = 0; i < num; i++)
-		if ((ecode = tdb_append(tdb, key, data)) != TDB_SUCCESS)
-			errx(1, "Appending key %u in tdb: %s",
-			     i, tdb_errorstr(ecode));
+		if ((ecode = ntdb_append(ntdb, key, data)) != NTDB_SUCCESS)
+			errx(1, "Appending key %u in ntdb: %s",
+			     i, ntdb_errorstr(ecode));
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
@@ -406,36 +406,36 @@ int main(int argc, char *argv[])
 		exit(0);
 
 	/* Churn 1000 records: not in order! */
-	if (transaction && (ecode = tdb_transaction_start(tdb)))
-		errx(1, "starting transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_start(ntdb)))
+		errx(1, "starting transaction: %s", ntdb_errorstr(ecode));
 	printf("Churning %u records: ", num); fflush(stdout);
 	gettimeofday(&start, NULL);
 	for (j = 0; j < num; j++) {
 		i = (j + 1000019) % num;
-		if ((ecode = tdb_delete(tdb, key)) != TDB_SUCCESS)
-			errx(1, "Deleting key %u in tdb: %s",
-			     i, tdb_errorstr(ecode));
+		if ((ecode = ntdb_delete(ntdb, key)) != NTDB_SUCCESS)
+			errx(1, "Deleting key %u in ntdb: %s",
+			     i, ntdb_errorstr(ecode));
 		i += num;
-		if ((ecode = tdb_store(tdb, key, data, TDB_INSERT)) != 0)
-			errx(1, "Inserting key %u in tdb: %s",
-			     i, tdb_errorstr(ecode));
+		if ((ecode = ntdb_store(ntdb, key, data, NTDB_INSERT)) != 0)
+			errx(1, "Inserting key %u in ntdb: %s",
+			     i, ntdb_errorstr(ecode));
 	}
 	gettimeofday(&stop, NULL);
-	if (transaction && (ecode = tdb_transaction_commit(tdb)))
-		errx(1, "committing transaction: %s", tdb_errorstr(ecode));
+	if (transaction && (ecode = ntdb_transaction_commit(ntdb)))
+		errx(1, "committing transaction: %s", ntdb_errorstr(ecode));
 	printf(" %zu ns (%zu bytes)\n",
 	       normalize(&start, &stop, num), file_size());
 
-	if (tdb_check(tdb, NULL, NULL))
-		errx(1, "tdb_check failed!");
+	if (ntdb_check(ntdb, NULL, NULL))
+		errx(1, "ntdb_check failed!");
 	if (summary) {
 		char *sumstr = NULL;
-		tdb_summary(tdb, TDB_SUMMARY_HISTOGRAMS, &sumstr);
+		ntdb_summary(ntdb, NTDB_SUMMARY_HISTOGRAMS, &sumstr);
 		printf("%s\n", sumstr);
 		free(sumstr);
 	}
 	if (do_stats)
-		dump_and_clear_stats(&tdb, flags, &log);
+		dump_and_clear_stats(&ntdb, flags, &log);
 	if (++stage == stopat)
 		exit(0);
 

lib/ntdb/traverse.c

@@ -0,0 +1,99 @@
+ /*
+   Trivial Database 2: traverse function.
+   Copyright (C) Rusty Russell 2010
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 3 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, see <http://www.gnu.org/licenses/>.
+*/
+#include "private.h"
+#include <ccan/likely/likely.h>
+
+_PUBLIC_ int64_t ntdb_traverse_(struct ntdb_context *ntdb,
+		      int (*fn)(struct ntdb_context *,
+				NTDB_DATA, NTDB_DATA, void *),
+		      void *p)
+{
+	enum NTDB_ERROR ecode;
+	struct traverse_info tinfo;
+	NTDB_DATA k, d;
+	int64_t count = 0;
+
+	k.dptr = NULL;
+	for (ecode = first_in_hash(ntdb, &tinfo, &k, &d.dsize);
+	     ecode == NTDB_SUCCESS;
+	     ecode = next_in_hash(ntdb, &tinfo, &k, &d.dsize)) {
+		d.dptr = k.dptr + k.dsize;
+
+		count++;
+		if (fn && fn(ntdb, k, d, p)) {
+			free(k.dptr);
+			ntdb->last_error = NTDB_SUCCESS;
+			return count;
+		}
+		free(k.dptr);
+	}
+
+	if (ecode != NTDB_ERR_NOEXIST) {
+		return NTDB_ERR_TO_OFF(ntdb->last_error = ecode);
+	}
+	ntdb->last_error = NTDB_SUCCESS;
+	return count;
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_firstkey(struct ntdb_context *ntdb, NTDB_DATA *key)
+{
+	struct traverse_info tinfo;
+
+	return ntdb->last_error = first_in_hash(ntdb, &tinfo, key, NULL);
+}
+
+/* We lock twice, not very efficient.  We could keep last key & tinfo cached. */
+_PUBLIC_ enum NTDB_ERROR ntdb_nextkey(struct ntdb_context *ntdb, NTDB_DATA *key)
+{
+	struct traverse_info tinfo;
+	struct hash_info h;
+	struct ntdb_used_record rec;
+
+	tinfo.prev = find_and_lock(ntdb, *key, F_RDLCK, &h, &rec, &tinfo);
+	free(key->dptr);
+	if (NTDB_OFF_IS_ERR(tinfo.prev)) {
+		return ntdb->last_error = NTDB_OFF_TO_ERR(tinfo.prev);
+	}
+	ntdb_unlock_hashes(ntdb, h.hlock_start, h.hlock_range, F_RDLCK);
+
+	return ntdb->last_error = next_in_hash(ntdb, &tinfo, key, NULL);
+}
+
+static int wipe_one(struct ntdb_context *ntdb,
+		    NTDB_DATA key, NTDB_DATA data, enum NTDB_ERROR *ecode)
+{
+	*ecode = ntdb_delete(ntdb, key);
+	return (*ecode != NTDB_SUCCESS);
+}
+
+_PUBLIC_ enum NTDB_ERROR ntdb_wipe_all(struct ntdb_context *ntdb)
+{
+	enum NTDB_ERROR ecode;
+	int64_t count;
+
+	ecode = ntdb_allrecord_lock(ntdb, F_WRLCK, NTDB_LOCK_WAIT, false);
+	if (ecode != NTDB_SUCCESS)
+		return ntdb->last_error = ecode;
+
+	/* FIXME: Be smarter. */
+	count = ntdb_traverse(ntdb, wipe_one, &ecode);
+	if (count < 0)
+		ecode = NTDB_OFF_TO_ERR(count);
+	ntdb_allrecord_unlock(ntdb, F_WRLCK);
+	return ntdb->last_error = ecode;
+}