bcachefs: BTREE_ID_logged_ops

Add a new btree for long running logged operations - i.e. for logging
operations that we can't do within a single btree transaction, so that
they can be resumed if we crash.

Keys in the logged operations btree will represent operations in
progress, with the state of the operation stored in the value.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/Makefile

@@ -55,6 +55,7 @@ bcachefs-y		:=	\
 	journal_sb.o		\
 	journal_seq_blacklist.o	\
 	keylist.o		\
+	logged_ops.o		\
 	lru.o			\
 	mean_and_variance.o	\
 	migrate.o		\

fs/bcachefs/bcachefs.h

@@ -454,6 +454,7 @@ enum gc_phase {
 	GC_PHASE_BTREE_bucket_gens,
 	GC_PHASE_BTREE_snapshot_trees,
 	GC_PHASE_BTREE_deleted_inodes,
+	GC_PHASE_BTREE_logged_ops,
 
 	GC_PHASE_PENDING_DELETE,
 };

fs/bcachefs/bcachefs_format.h

@@ -2249,7 +2249,9 @@ enum btree_id_flags {
 	x(snapshot_trees,	15,	0,					\
 	  BIT_ULL(KEY_TYPE_snapshot_tree))					\
 	x(deleted_inodes,	16,	BTREE_ID_SNAPSHOTS,			\
-	  BIT_ULL(KEY_TYPE_set))
+	  BIT_ULL(KEY_TYPE_set))						\
+	x(logged_ops,		17,	0,					\
+	  0)
 
 enum btree_id {
 #define x(name, nr, ...) BTREE_ID_##name = nr,

fs/bcachefs/btree_update.c

@@ -653,6 +653,7 @@ int bch2_btree_insert_nonextent(struct btree_trans *trans,
 	int ret;
 
 	bch2_trans_iter_init(trans, &iter, btree, k->k.p,
+			     BTREE_ITER_CACHED|
 			     BTREE_ITER_NOT_EXTENTS|
 			     BTREE_ITER_INTENT);
 	ret   = bch2_btree_iter_traverse(&iter) ?:
@@ -727,6 +728,23 @@ int bch2_btree_delete_at_buffered(struct btree_trans *trans,
 	return bch2_trans_update_buffered(trans, btree, k);
 }
 
+int bch2_btree_delete(struct btree_trans *trans,
+		      enum btree_id btree, struct bpos pos,
+		      unsigned update_flags)
+{
+	struct btree_iter iter;
+	int ret;
+
+	bch2_trans_iter_init(trans, &iter, btree, pos,
+			     BTREE_ITER_CACHED|
+			     BTREE_ITER_INTENT);
+	ret   = bch2_btree_iter_traverse(&iter) ?:
+		bch2_btree_delete_at(trans, &iter, update_flags);
+	bch2_trans_iter_exit(trans, &iter);
+
+	return ret;
+}
+
 int bch2_btree_delete_range_trans(struct btree_trans *trans, enum btree_id id,
 				  struct bpos start, struct bpos end,
 				  unsigned update_flags,

fs/bcachefs/btree_update.h

@@ -58,6 +58,7 @@ int bch2_btree_delete_extent_at(struct btree_trans *, struct btree_iter *,
 				unsigned, unsigned);
 int bch2_btree_delete_at(struct btree_trans *, struct btree_iter *, unsigned);
 int bch2_btree_delete_at_buffered(struct btree_trans *, enum btree_id, struct bpos);
+int bch2_btree_delete(struct btree_trans *, enum btree_id, struct bpos, unsigned);
 
 int bch2_btree_insert_nonextent(struct btree_trans *, enum btree_id,
 				struct bkey_i *, enum btree_update_flags);

fs/bcachefs/logged_ops.c

@@ -0,0 +1,111 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "bcachefs.h"
+#include "bkey_buf.h"
+#include "btree_update.h"
+#include "error.h"
+#include "logged_ops.h"
+#include "super.h"
+
+struct bch_logged_op_fn {
+	u8		type;
+	int		(*resume)(struct btree_trans *, struct bkey_i *);
+};
+
+static const struct bch_logged_op_fn logged_op_fns[] = {
+#define x(n)		{					\
+	.type		= KEY_TYPE_logged_op_##n,		\
+	.resume		= bch2_resume_logged_op_##n,		\
+},
+	BCH_LOGGED_OPS()
+#undef x
+};
+
+static const struct bch_logged_op_fn *logged_op_fn(enum bch_bkey_type type)
+{
+	for (unsigned i = 0; i < ARRAY_SIZE(logged_op_fns); i++)
+		if (logged_op_fns[i].type == type)
+			return logged_op_fns + i;
+	return NULL;
+}
+
+static int resume_logged_op(struct btree_trans *trans, struct btree_iter *iter,
+			    struct bkey_s_c k)
+{
+	struct bch_fs *c = trans->c;
+	const struct bch_logged_op_fn *fn = logged_op_fn(k.k->type);
+	struct bkey_buf sk;
+	u32 restart_count = trans->restart_count;
+	int ret;
+
+	if (!fn)
+		return 0;
+
+	bch2_bkey_buf_init(&sk);
+	bch2_bkey_buf_reassemble(&sk, c, k);
+
+	ret =   drop_locks_do(trans, (bch2_fs_lazy_rw(c), 0)) ?:
+		fn->resume(trans, sk.k) ?: trans_was_restarted(trans, restart_count);
+
+	bch2_bkey_buf_exit(&sk, c);
+	return ret;
+}
+
+int bch2_resume_logged_ops(struct bch_fs *c)
+{
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	int ret;
+
+	ret = bch2_trans_run(c,
+		for_each_btree_key2(&trans, iter,
+				BTREE_ID_logged_ops, POS_MIN, BTREE_ITER_PREFETCH, k,
+			resume_logged_op(&trans, &iter, k)));
+	if (ret)
+		bch_err_fn(c, ret);
+	return ret;
+}
+
+static int __bch2_logged_op_start(struct btree_trans *trans, struct bkey_i *k)
+{
+	struct btree_iter iter;
+	int ret;
+
+	ret = bch2_bkey_get_empty_slot(trans, &iter, BTREE_ID_logged_ops, POS_MAX);
+	if (ret)
+		return ret;
+
+	k->k.p = iter.pos;
+
+	ret = bch2_trans_update(trans, &iter, k, 0);
+	bch2_trans_iter_exit(trans, &iter);
+	return ret;
+}
+
+int bch2_logged_op_start(struct btree_trans *trans, struct bkey_i *k)
+{
+	return commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
+			 __bch2_logged_op_start(trans, k));
+}
+
+void bch2_logged_op_finish(struct btree_trans *trans, struct bkey_i *k)
+{
+	int ret = commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
+			    bch2_btree_delete(trans, BTREE_ID_logged_ops, k->k.p, 0));
+	/*
+	 * This needs to be a fatal error because we've left an unfinished
+	 * operation in the logged ops btree.
+	 *
+	 * We should only ever see an error here if the filesystem has already
+	 * been shut down, but make sure of that here:
+	 */
+	if (ret) {
+		struct bch_fs *c = trans->c;
+		struct printbuf buf = PRINTBUF;
+
+		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
+		bch2_fs_fatal_error(c, "%s: error deleting logged operation %s: %s",
+				     __func__, buf.buf, bch2_err_str(ret));
+		printbuf_exit(&buf);
+	}
+}

fs/bcachefs/logged_ops.h

@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_LOGGED_OPS_H
+#define _BCACHEFS_LOGGED_OPS_H
+
+#include "bkey.h"
+
+#define BCH_LOGGED_OPS()
+
+static inline int bch2_logged_op_update(struct btree_trans *trans, struct bkey_i *op)
+{
+	return bch2_btree_insert_nonextent(trans, BTREE_ID_logged_ops, op, 0);
+}
+
+int bch2_resume_logged_ops(struct bch_fs *);
+int bch2_logged_op_start(struct btree_trans *, struct bkey_i *);
+void bch2_logged_op_finish(struct btree_trans *, struct bkey_i *);
+
+#endif /* _BCACHEFS_LOGGED_OPS_H */

fs/bcachefs/recovery.c

@@ -20,6 +20,7 @@
 #include "journal_reclaim.h"
 #include "journal_seq_blacklist.h"
 #include "lru.h"
+#include "logged_ops.h"
 #include "move.h"
 #include "quota.h"
 #include "recovery.h"

fs/bcachefs/recovery_types.h

@@ -29,6 +29,7 @@
 	x(check_subvols,		PASS_FSCK)						\
 	x(delete_dead_snapshots,	PASS_FSCK|PASS_UNCLEAN)					\
 	x(fs_upgrade_for_subvolumes,	0)							\
+	x(resume_logged_ops,		PASS_ALWAYS)						\
 	x(check_inodes,			PASS_FSCK)						\
 	x(check_extents,		PASS_FSCK)						\
 	x(check_dirents,		PASS_FSCK)						\

fs/bcachefs/super.c

@@ -791,6 +791,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 	c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
 	if (c->opts.inodes_use_key_cache)
 		c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
+	c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
 
 	c->block_bits		= ilog2(block_sectors(c));
 	c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);