lvm2/libdm/libdm-deptree.c

/*
 * Copyright (C) 2005-2014 Red Hat, Inc. All rights reserved.
 *
 * This file is part of the device-mapper userspace tools.
 *
 * This copyrighted material is made available to anyone wishing to use,
 * modify, copy, or redistribute it subject to the terms and conditions
 * of the GNU Lesser General Public License v.2.1.
 *
 * You should have received a copy of the GNU Lesser General Public License
 * along with this program; if not, write to the Free Software Foundation,
 * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#include "dmlib.h"
#include "libdm-targets.h"
#include "libdm-common.h"
#include "kdev_t.h"
#include "dm-ioctl.h"

#include <stdarg.h>
#include <sys/param.h>
#include <sys/utsname.h>

#define MAX_TARGET_PARAMSIZE 500000

#define REPLICATOR_LOCAL_SITE 0

/* Supported segment types */
enum {
	SEG_CACHE,
	SEG_CRYPT,
	SEG_ERROR,
	SEG_LINEAR,
	SEG_MIRRORED,
	SEG_REPLICATOR,
	SEG_REPLICATOR_DEV,
	SEG_SNAPSHOT,
	SEG_SNAPSHOT_ORIGIN,
	SEG_SNAPSHOT_MERGE,
	SEG_STRIPED,
	SEG_ZERO,
	SEG_THIN_POOL,
	SEG_THIN,
	SEG_RAID1,
	SEG_RAID10,
	SEG_RAID4,
	SEG_RAID5_LA,
	SEG_RAID5_RA,
	SEG_RAID5_LS,
	SEG_RAID5_RS,
	SEG_RAID6_ZR,
	SEG_RAID6_NR,
	SEG_RAID6_NC,
};

/* FIXME Add crypt and multipath support */

static const struct {
	unsigned type;
	const char target[16];
} _dm_segtypes[] = {
	{ SEG_CACHE, "cache" },
	{ SEG_CRYPT, "crypt" },
	{ SEG_ERROR, "error" },
	{ SEG_LINEAR, "linear" },
	{ SEG_MIRRORED, "mirror" },
	{ SEG_REPLICATOR, "replicator" },
	{ SEG_REPLICATOR_DEV, "replicator-dev" },
	{ SEG_SNAPSHOT, "snapshot" },
	{ SEG_SNAPSHOT_ORIGIN, "snapshot-origin" },
	{ SEG_SNAPSHOT_MERGE, "snapshot-merge" },
	{ SEG_STRIPED, "striped" },
	{ SEG_ZERO, "zero"},
	{ SEG_THIN_POOL, "thin-pool"},
	{ SEG_THIN, "thin"},
	{ SEG_RAID1, "raid1"},
	{ SEG_RAID10, "raid10"},
	{ SEG_RAID4, "raid4"},
	{ SEG_RAID5_LA, "raid5_la"},
	{ SEG_RAID5_RA, "raid5_ra"},
	{ SEG_RAID5_LS, "raid5_ls"},
	{ SEG_RAID5_RS, "raid5_rs"},
	{ SEG_RAID6_ZR, "raid6_zr"},
	{ SEG_RAID6_NR, "raid6_nr"},
	{ SEG_RAID6_NC, "raid6_nc"},

	/*
	 *WARNING: Since 'raid' target overloads this 1:1 mapping table
	 * for search do not add new enum elements past them!
	 */
	{ SEG_RAID5_LS, "raid5"}, /* same as "raid5_ls" (default for MD also) */
	{ SEG_RAID6_ZR, "raid6"}, /* same as "raid6_zr" */
};

/* Some segment types have a list of areas of other devices attached */
struct seg_area {
	struct dm_list list;

	struct dm_tree_node *dev_node;

	uint64_t offset;

	unsigned rsite_index;		/* Replicator site index */
	struct dm_tree_node *slog;	/* Replicator sync log node */
	uint64_t region_size;		/* Replicator sync log size */
	uint32_t flags;			/* Replicator sync log flags */
};

struct dm_thin_message {
	dm_thin_message_t type;
	union {
		struct {
			uint32_t device_id;
			uint32_t origin_id;
		} m_create_snap;
		struct {
			uint32_t device_id;
		} m_create_thin;
		struct {
			uint32_t device_id;
		} m_delete;
		struct {
			uint64_t current_id;
			uint64_t new_id;
		} m_set_transaction_id;
	} u;
};

struct thin_message {
	struct dm_list list;
	struct dm_thin_message message;
	int expected_errno;
};

/* Replicator-log has a list of sites */
/* FIXME: maybe move to seg_area too? */
struct replicator_site {
	struct dm_list list;

	unsigned rsite_index;
	dm_replicator_mode_t mode;
	uint32_t async_timeout;
	uint32_t fall_behind_ios;
	uint64_t fall_behind_data;
};

/* Per-segment properties */
struct load_segment {
	struct dm_list list;

	unsigned type;

	uint64_t size;

	unsigned area_count;		/* Linear + Striped + Mirrored + Crypt + Replicator */
	struct dm_list areas;		/* Linear + Striped + Mirrored + Crypt + Replicator */

	uint32_t stripe_size;		/* Striped + raid */

	int persistent;			/* Snapshot */
	uint32_t chunk_size;		/* Snapshot + cache */
	struct dm_tree_node *cow;	/* Snapshot */
	struct dm_tree_node *origin;	/* Snapshot + Snapshot origin + Cache */
	struct dm_tree_node *merge;	/* Snapshot */

	struct dm_tree_node *log;	/* Mirror + Replicator */
	uint32_t region_size;		/* Mirror + raid */
	unsigned clustered;		/* Mirror */
	unsigned mirror_area_count;	/* Mirror */
	uint32_t flags;			/* Mirror + raid + Cache */
	char *uuid;			/* Clustered mirror log */

	const char *policy_name;	/* Cache */
	unsigned policy_argc;		/* Cache */
	struct dm_config_node *policy_settings;	/* Cache */

	const char *cipher;		/* Crypt */
	const char *chainmode;		/* Crypt */
	const char *iv;			/* Crypt */
	uint64_t iv_offset;		/* Crypt */
	const char *key;		/* Crypt */

	const char *rlog_type;		/* Replicator */
	struct dm_list rsites;		/* Replicator */
	unsigned rsite_count;		/* Replicator */
	unsigned rdevice_count;		/* Replicator */
	struct dm_tree_node *replicator;/* Replicator-dev */
	uint64_t rdevice_index;		/* Replicator-dev */

	uint64_t rebuilds;		/* raid */
	uint64_t writemostly;		/* raid */
	uint32_t writebehind;		/* raid */
	uint32_t max_recovery_rate;	/* raid kB/sec/disk */
	uint32_t min_recovery_rate;	/* raid kB/sec/disk */

	struct dm_tree_node *metadata;	/* Thin_pool + Cache */
	struct dm_tree_node *pool;	/* Thin_pool, Thin */
	struct dm_tree_node *external;	/* Thin */
	struct dm_list thin_messages;	/* Thin_pool */
	uint64_t transaction_id;	/* Thin_pool */
	uint64_t low_water_mark;	/* Thin_pool */
	uint32_t data_block_size;       /* Thin_pool */
	unsigned skip_block_zeroing;	/* Thin_pool */
	unsigned ignore_discard;	/* Thin_pool target vsn 1.1 */
	unsigned no_discard_passdown;	/* Thin_pool target vsn 1.1 */
	unsigned error_if_no_space;	/* Thin pool target vsn 1.10 */
	uint32_t device_id;		/* Thin */

};

/* Per-device properties */
struct load_properties {
	int read_only;
	uint32_t major;
	uint32_t minor;

	uint32_t read_ahead;
	uint32_t read_ahead_flags;

	unsigned segment_count;
	unsigned size_changed;
	struct dm_list segs;

	const char *new_name;

	/* If immediate_dev_node is set to 1, try to create the dev node
	 * as soon as possible (e.g. in preload stage even during traversal
	 * and processing of dm tree). This will also flush all stacked dev
	 * node operations, synchronizing with udev.
	 */
	unsigned immediate_dev_node;

	/*
	 * If the device size changed from zero and this is set,
	 * don't resume the device immediately, even if the device
	 * has parents.  This works provided the parents do not
	 * validate the device size and is required by pvmove to
	 * avoid starting the mirror resync operation too early.
	 */
	unsigned delay_resume_if_new;

	/* Send messages for this node in preload */
	unsigned send_messages;
};

/* Two of these used to join two nodes with uses and used_by. */
struct dm_tree_link {
	struct dm_list list;
	struct dm_tree_node *node;
};

struct dm_tree_node {
	struct dm_tree *dtree;

	const char *name;
	const char *uuid;
	struct dm_info info;

	struct dm_list uses;       	/* Nodes this node uses */
	struct dm_list used_by;    	/* Nodes that use this node */

	int activation_priority;	/* 0 gets activated first */
	int implicit_deps;		/* 1 device only implicitly referenced */

	uint16_t udev_flags;		/* Udev control flags */

	void *context;			/* External supplied context */

	struct load_properties props;	/* For creation/table (re)load */

	/*
	 * If presuspend of child node is needed
	 * Note: only direct child is allowed
	 */
	struct dm_tree_node *presuspend_node;

	/* Callback */
	dm_node_callback_fn callback;
	void *callback_data;

	/*
	 * TODO:
	 *	Add advanced code which tracks of send ioctls and their
	 *	proper revert operation for more advanced recovery
	 *	Current code serves mostly only to recovery when
	 *	thin pool metadata check fails and command would
	 *	have left active thin data and metadata subvolumes.
	 */
	struct dm_list activated;	/* Head of activated nodes for preload revert */
	struct dm_list activated_list;	/* List of activated nodes for preload revert */
};

struct dm_tree {
	struct dm_pool *mem;
	struct dm_hash_table *devs;
	struct dm_hash_table *uuids;
	struct dm_tree_node root;
	int skip_lockfs;		/* 1 skips lockfs (for non-snapshots) */
	int no_flush;			/* 1 sets noflush (mirrors/multipath) */
	int retry_remove;		/* 1 retries remove if not successful */
	uint32_t cookie;
	const char **optional_uuid_suffixes;	/* uuid suffixes ignored when matching */
};

/*
 * Tree functions.
 */
struct dm_tree *dm_tree_create(void)
{
	struct dm_pool *dmem;
	struct dm_tree *dtree;

	if (!(dmem = dm_pool_create("dtree", 1024)) ||
	    !(dtree = dm_pool_zalloc(dmem, sizeof(*dtree)))) {
		log_error("Failed to allocate dtree.");
		if (dmem)
			dm_pool_destroy(dmem);
		return NULL;
	}

	dtree->root.dtree = dtree;
	dm_list_init(&dtree->root.uses);
	dm_list_init(&dtree->root.used_by);
	dm_list_init(&dtree->root.activated);
	dtree->skip_lockfs = 0;
	dtree->no_flush = 0;
	dtree->mem = dmem;
	dtree->optional_uuid_suffixes = NULL;

	if (!(dtree->devs = dm_hash_create(8))) {
		log_error("dtree hash creation failed");
		dm_pool_destroy(dtree->mem);
		return NULL;
	}

	if (!(dtree->uuids = dm_hash_create(32))) {
		log_error("dtree uuid hash creation failed");
		dm_hash_destroy(dtree->devs);
		dm_pool_destroy(dtree->mem);
		return NULL;
	}

	return dtree;
}

void dm_tree_free(struct dm_tree *dtree)
{
	if (!dtree)
		return;

	dm_hash_destroy(dtree->uuids);
	dm_hash_destroy(dtree->devs);
	dm_pool_destroy(dtree->mem);
}

void dm_tree_set_cookie(struct dm_tree_node *node, uint32_t cookie)
{
	node->dtree->cookie = cookie;
}

uint32_t dm_tree_get_cookie(struct dm_tree_node *node)
{
	return node->dtree->cookie;
}

void dm_tree_skip_lockfs(struct dm_tree_node *dnode)
{
	dnode->dtree->skip_lockfs = 1;
}

void dm_tree_use_no_flush_suspend(struct dm_tree_node *dnode)
{
	dnode->dtree->no_flush = 1;
}

void dm_tree_retry_remove(struct dm_tree_node *dnode)
{
	dnode->dtree->retry_remove = 1;
}

/*
 * Node functions.
 */
static int _nodes_are_linked(const struct dm_tree_node *parent,
			     const struct dm_tree_node *child)
{
	struct dm_tree_link *dlink;

	dm_list_iterate_items(dlink, &parent->uses)
		if (dlink->node == child)
			return 1;

	return 0;
}

static int _link(struct dm_list *list, struct dm_tree_node *node)
{
	struct dm_tree_link *dlink;

	if (!(dlink = dm_pool_alloc(node->dtree->mem, sizeof(*dlink)))) {
		log_error("dtree link allocation failed");
		return 0;
	}

	dlink->node = node;
	dm_list_add(list, &dlink->list);

	return 1;
}

static int _link_nodes(struct dm_tree_node *parent,
		       struct dm_tree_node *child)
{
	if (_nodes_are_linked(parent, child))
		return 1;

	if (!_link(&parent->uses, child))
		return 0;

	if (!_link(&child->used_by, parent))
		return 0;

	return 1;
}

static void _unlink(struct dm_list *list, struct dm_tree_node *node)
{
	struct dm_tree_link *dlink;

	dm_list_iterate_items(dlink, list)
		if (dlink->node == node) {
			dm_list_del(&dlink->list);
			break;
		}
}

static void _unlink_nodes(struct dm_tree_node *parent,
			  struct dm_tree_node *child)
{
	if (!_nodes_are_linked(parent, child))
		return;

	_unlink(&parent->uses, child);
	_unlink(&child->used_by, parent);
}

static int _add_to_toplevel(struct dm_tree_node *node)
{
	return _link_nodes(&node->dtree->root, node);
}

static void _remove_from_toplevel(struct dm_tree_node *node)
{
	_unlink_nodes(&node->dtree->root, node);
}

static int _add_to_bottomlevel(struct dm_tree_node *node)
{
	return _link_nodes(node, &node->dtree->root);
}

static void _remove_from_bottomlevel(struct dm_tree_node *node)
{
	_unlink_nodes(node, &node->dtree->root);
}

static int _link_tree_nodes(struct dm_tree_node *parent, struct dm_tree_node *child)
{
	/* Don't link to root node if child already has a parent */
	if (parent == &parent->dtree->root) {
		if (dm_tree_node_num_children(child, 1))
			return 1;
	} else
		_remove_from_toplevel(child);

	if (child == &child->dtree->root) {
		if (dm_tree_node_num_children(parent, 0))
			return 1;
	} else
		_remove_from_bottomlevel(parent);

	return _link_nodes(parent, child);
}

static struct dm_tree_node *_create_dm_tree_node(struct dm_tree *dtree,
						 const char *name,
						 const char *uuid,
						 struct dm_info *info,
						 void *context,
						 uint16_t udev_flags)
{
	struct dm_tree_node *node;
	dev_t dev;

	if (!(node = dm_pool_zalloc(dtree->mem, sizeof(*node)))) {
		log_error("_create_dm_tree_node alloc failed");
		return NULL;
	}

	node->dtree = dtree;

	node->name = name;
	node->uuid = uuid;
	node->info = *info;
	node->context = context;
	node->udev_flags = udev_flags;

	dm_list_init(&node->uses);
	dm_list_init(&node->used_by);
	dm_list_init(&node->activated);
	dm_list_init(&node->props.segs);

	dev = MKDEV((dev_t)info->major, info->minor);

	if (!dm_hash_insert_binary(dtree->devs, (const char *) &dev,
				sizeof(dev), node)) {
		log_error("dtree node hash insertion failed");
		dm_pool_free(dtree->mem, node);
		return NULL;
	}

	if (uuid && *uuid &&
	    !dm_hash_insert(dtree->uuids, uuid, node)) {
		log_error("dtree uuid hash insertion failed");
		dm_hash_remove_binary(dtree->devs, (const char *) &dev,
				      sizeof(dev));
		dm_pool_free(dtree->mem, node);
		return NULL;
	}

	return node;
}

static struct dm_tree_node *_find_dm_tree_node(struct dm_tree *dtree,
					       uint32_t major, uint32_t minor)
{
	dev_t dev = MKDEV((dev_t)major, minor);

	return dm_hash_lookup_binary(dtree->devs, (const char *) &dev,
				     sizeof(dev));
}

void dm_tree_set_optional_uuid_suffixes(struct dm_tree *dtree, const char **optional_uuid_suffixes)
{
	dtree->optional_uuid_suffixes = optional_uuid_suffixes;
}

static struct dm_tree_node *_find_dm_tree_node_by_uuid(struct dm_tree *dtree,
						       const char *uuid)
{
	struct dm_tree_node *node;
	const char *default_uuid_prefix;
	size_t default_uuid_prefix_len;
	const char *suffix, *suffix_position;
	char uuid_without_suffix[DM_UUID_LEN];
	unsigned i = 0;
	const char **suffix_list = dtree->optional_uuid_suffixes;

	if ((node = dm_hash_lookup(dtree->uuids, uuid))) {
		log_debug("Matched uuid %s in deptree.", uuid);
		return node;
	}

	default_uuid_prefix = dm_uuid_prefix();
	default_uuid_prefix_len = strlen(default_uuid_prefix);

	if (suffix_list && (suffix_position = rindex(uuid, '-'))) {
		while ((suffix = suffix_list[i++])) {
			if (strcmp(suffix_position + 1, suffix))
				continue;

			(void) strncpy(uuid_without_suffix, uuid, sizeof(uuid_without_suffix));
			uuid_without_suffix[suffix_position - uuid] = '\0';

			if ((node = dm_hash_lookup(dtree->uuids, uuid_without_suffix))) {
				log_debug("Matched uuid %s (missing suffix -%s) in deptree.", uuid_without_suffix, suffix);
				return node;
			}

			break;
		};
	}
	
	if (strncmp(uuid, default_uuid_prefix, default_uuid_prefix_len))
		return NULL;

	if ((node = dm_hash_lookup(dtree->uuids, uuid + default_uuid_prefix_len))) {
		log_debug("Matched uuid %s (missing prefix) in deptree.", uuid + default_uuid_prefix_len);
		return node;
	}

	log_debug("Not matched uuid %s in deptree.", uuid + default_uuid_prefix_len);
	return NULL;
}

void dm_tree_node_set_udev_flags(struct dm_tree_node *dnode, uint16_t udev_flags)

{
	struct dm_info *dinfo = &dnode->info;

	if (udev_flags != dnode->udev_flags)
		log_debug_activation("Resetting %s (%" PRIu32 ":%" PRIu32
				     ") udev_flags from 0x%x to 0x%x",
				     dnode->name, dinfo->major, dinfo->minor,
				     dnode->udev_flags, udev_flags);
	dnode->udev_flags = udev_flags;
}

void dm_tree_node_set_read_ahead(struct dm_tree_node *dnode,
				 uint32_t read_ahead,
				 uint32_t read_ahead_flags)
{
	dnode->props.read_ahead = read_ahead;
	dnode->props.read_ahead_flags = read_ahead_flags;
}

void dm_tree_node_set_presuspend_node(struct dm_tree_node *node,
				      struct dm_tree_node *presuspend_node)
{
	node->presuspend_node = presuspend_node;
}

const char *dm_tree_node_get_name(const struct dm_tree_node *node)
{
	return node->info.exists ? node->name : "";
}

const char *dm_tree_node_get_uuid(const struct dm_tree_node *node)
{
	return node->info.exists ? node->uuid : "";
}

const struct dm_info *dm_tree_node_get_info(const struct dm_tree_node *node)
{
	return &node->info;
}

void *dm_tree_node_get_context(const struct dm_tree_node *node)
{
	return node->context;
}

int dm_tree_node_size_changed(const struct dm_tree_node *dnode)
{
	return dnode->props.size_changed;
}

int dm_tree_node_num_children(const struct dm_tree_node *node, uint32_t inverted)
{
	if (inverted) {
		if (_nodes_are_linked(&node->dtree->root, node))
			return 0;
		return dm_list_size(&node->used_by);
	}

	if (_nodes_are_linked(node, &node->dtree->root))
		return 0;

	return dm_list_size(&node->uses);
}

/*
 * Returns 1 if no prefix supplied
 */
static int _uuid_prefix_matches(const char *uuid, const char *uuid_prefix, size_t uuid_prefix_len)
{
	const char *default_uuid_prefix = dm_uuid_prefix();
	size_t default_uuid_prefix_len = strlen(default_uuid_prefix);

	if (!uuid_prefix)
		return 1;

	if (!strncmp(uuid, uuid_prefix, uuid_prefix_len))
		return 1;

	/* Handle transition: active device uuids might be missing the prefix */
	if (uuid_prefix_len <= 4)
		return 0;

	if (!strncmp(uuid, default_uuid_prefix, default_uuid_prefix_len))
		return 0;

	if (strncmp(uuid_prefix, default_uuid_prefix, default_uuid_prefix_len))
		return 0;

	if (!strncmp(uuid, uuid_prefix + default_uuid_prefix_len, uuid_prefix_len - default_uuid_prefix_len))
		return 1;

	return 0;
}

/*
 * Returns 1 if no children.
 */
static int _children_suspended(struct dm_tree_node *node,
			       uint32_t inverted,
			       const char *uuid_prefix,
			       size_t uuid_prefix_len)
{
	struct dm_list *list;
	struct dm_tree_link *dlink;
	const struct dm_info *dinfo;
	const char *uuid;

	if (inverted) {
		if (_nodes_are_linked(&node->dtree->root, node))
			return 1;
		list = &node->used_by;
	} else {
		if (_nodes_are_linked(node, &node->dtree->root))
			return 1;
		list = &node->uses;
	}

	dm_list_iterate_items(dlink, list) {
		if (!(uuid = dm_tree_node_get_uuid(dlink->node))) {
			stack;
			continue;
		}

		/* Ignore if it doesn't belong to this VG */
		if (!_uuid_prefix_matches(uuid, uuid_prefix, uuid_prefix_len))
			continue;

		/* Ignore if parent node wants to presuspend this node */
		if (dlink->node->presuspend_node == node)
			continue;

		if (!(dinfo = dm_tree_node_get_info(dlink->node)))