summaryrefslogtreecommitdiff
path: root/fs/btrfs/volumes.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2021-07-13 22:02:07 +0300
committerLinus Torvalds <torvalds@linux-foundation.org>2021-07-13 22:02:07 +0300
commitf02bf8578bd8dd400903291ccebc69665adc911c (patch)
tree9e05913ac5f4f13a2dd1b5db24f2d896d9fbb400 /fs/btrfs/volumes.c
parent7fef2edf7cc753b51f7ccc74993971b0a9c81eca (diff)
parentea32af47f00a046a1f953370514d6d946efe0152 (diff)
downloadlinux-f02bf8578bd8dd400903291ccebc69665adc911c.tar.xz
Merge tag 'for-5.14-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs zoned mode fixes from David Sterba: - fix deadlock when allocating system chunk - fix wrong mutex unlock on an error path - fix extent map splitting for append operation - update and fix message reporting unusable chunk space - don't block when background zone reclaim runs with balance in parallel * tag 'for-5.14-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: btrfs: zoned: fix wrong mutex unlock on failure to allocate log root tree btrfs: don't block if we can't acquire the reclaim lock btrfs: properly split extent_map for REQ_OP_ZONE_APPEND btrfs: rework chunk allocation to avoid exhaustion of the system chunk array btrfs: fix deadlock with concurrent chunk allocations involving system chunks btrfs: zoned: print unusable percentage when reclaiming block groups btrfs: zoned: fix types for u64 division in btrfs_reclaim_bgs_work
Diffstat (limited to 'fs/btrfs/volumes.c')
-rw-r--r--fs/btrfs/volumes.c355
1 files changed, 270 insertions, 85 deletions
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 807502cd6510..1e4d43ffe38b 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -1745,19 +1745,14 @@ again:
extent = btrfs_item_ptr(leaf, path->slots[0],
struct btrfs_dev_extent);
} else {
- btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
goto out;
}
*dev_extent_len = btrfs_dev_extent_length(leaf, extent);
ret = btrfs_del_item(trans, root, path);
- if (ret) {
- btrfs_handle_fs_error(fs_info, ret,
- "Failed to remove dev extent item");
- } else {
+ if (ret == 0)
set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
- }
out:
btrfs_free_path(path);
return ret;
@@ -2942,7 +2937,7 @@ static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
u32 cur;
struct btrfs_key key;
- mutex_lock(&fs_info->chunk_mutex);
+ lockdep_assert_held(&fs_info->chunk_mutex);
array_size = btrfs_super_sys_array_size(super_copy);
ptr = super_copy->sys_chunk_array;
@@ -2972,7 +2967,6 @@ static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
cur += len;
}
}
- mutex_unlock(&fs_info->chunk_mutex);
return ret;
}
@@ -3012,6 +3006,29 @@ struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
return em;
}
+static int remove_chunk_item(struct btrfs_trans_handle *trans,
+ struct map_lookup *map, u64 chunk_offset)
+{
+ int i;
+
+ /*
+ * Removing chunk items and updating the device items in the chunks btree
+ * requires holding the chunk_mutex.
+ * See the comment at btrfs_chunk_alloc() for the details.
+ */
+ lockdep_assert_held(&trans->fs_info->chunk_mutex);
+
+ for (i = 0; i < map->num_stripes; i++) {
+ int ret;
+
+ ret = btrfs_update_device(trans, map->stripes[i].dev);
+ if (ret)
+ return ret;
+ }
+
+ return btrfs_free_chunk(trans, chunk_offset);
+}
+
int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
{
struct btrfs_fs_info *fs_info = trans->fs_info;
@@ -3032,14 +3049,16 @@ int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
return PTR_ERR(em);
}
map = em->map_lookup;
- mutex_lock(&fs_info->chunk_mutex);
- check_system_chunk(trans, map->type);
- mutex_unlock(&fs_info->chunk_mutex);
/*
- * Take the device list mutex to prevent races with the final phase of
- * a device replace operation that replaces the device object associated
- * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
+ * First delete the device extent items from the devices btree.
+ * We take the device_list_mutex to avoid racing with the finishing phase
+ * of a device replace operation. See the comment below before acquiring
+ * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
+ * because that can result in a deadlock when deleting the device extent
+ * items from the devices btree - COWing an extent buffer from the btree
+ * may result in allocating a new metadata chunk, which would attempt to
+ * lock again fs_info->chunk_mutex.
*/
mutex_lock(&fs_devices->device_list_mutex);
for (i = 0; i < map->num_stripes; i++) {
@@ -3061,18 +3080,73 @@ int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
btrfs_clear_space_info_full(fs_info);
mutex_unlock(&fs_info->chunk_mutex);
}
+ }
+ mutex_unlock(&fs_devices->device_list_mutex);
- ret = btrfs_update_device(trans, device);
+ /*
+ * We acquire fs_info->chunk_mutex for 2 reasons:
+ *
+ * 1) Just like with the first phase of the chunk allocation, we must
+ * reserve system space, do all chunk btree updates and deletions, and
+ * update the system chunk array in the superblock while holding this
+ * mutex. This is for similar reasons as explained on the comment at
+ * the top of btrfs_chunk_alloc();
+ *
+ * 2) Prevent races with the final phase of a device replace operation
+ * that replaces the device object associated with the map's stripes,
+ * because the device object's id can change at any time during that
+ * final phase of the device replace operation
+ * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
+ * replaced device and then see it with an ID of
+ * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
+ * the device item, which does not exists on the chunk btree.
+ * The finishing phase of device replace acquires both the
+ * device_list_mutex and the chunk_mutex, in that order, so we are
+ * safe by just acquiring the chunk_mutex.
+ */
+ trans->removing_chunk = true;
+ mutex_lock(&fs_info->chunk_mutex);
+
+ check_system_chunk(trans, map->type);
+
+ ret = remove_chunk_item(trans, map, chunk_offset);
+ /*
+ * Normally we should not get -ENOSPC since we reserved space before
+ * through the call to check_system_chunk().
+ *
+ * Despite our system space_info having enough free space, we may not
+ * be able to allocate extents from its block groups, because all have
+ * an incompatible profile, which will force us to allocate a new system
+ * block group with the right profile, or right after we called
+ * check_system_space() above, a scrub turned the only system block group
+ * with enough free space into RO mode.
+ * This is explained with more detail at do_chunk_alloc().
+ *
+ * So if we get -ENOSPC, allocate a new system chunk and retry once.
+ */
+ if (ret == -ENOSPC) {
+ const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
+ struct btrfs_block_group *sys_bg;
+
+ sys_bg = btrfs_alloc_chunk(trans, sys_flags);
+ if (IS_ERR(sys_bg)) {
+ ret = PTR_ERR(sys_bg);
+ btrfs_abort_transaction(trans, ret);
+ goto out;
+ }
+
+ ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
if (ret) {
- mutex_unlock(&fs_devices->device_list_mutex);
btrfs_abort_transaction(trans, ret);
goto out;
}
- }
- mutex_unlock(&fs_devices->device_list_mutex);
- ret = btrfs_free_chunk(trans, chunk_offset);
- if (ret) {
+ ret = remove_chunk_item(trans, map, chunk_offset);
+ if (ret) {
+ btrfs_abort_transaction(trans, ret);
+ goto out;
+ }
+ } else if (ret) {
btrfs_abort_transaction(trans, ret);
goto out;
}
@@ -3087,6 +3161,15 @@ int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
}
}
+ mutex_unlock(&fs_info->chunk_mutex);
+ trans->removing_chunk = false;
+
+ /*
+ * We are done with chunk btree updates and deletions, so release the
+ * system space we previously reserved (with check_system_chunk()).
+ */
+ btrfs_trans_release_chunk_metadata(trans);
+
ret = btrfs_remove_block_group(trans, chunk_offset, em);
if (ret) {
btrfs_abort_transaction(trans, ret);
@@ -3094,6 +3177,10 @@ int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
}
out:
+ if (trans->removing_chunk) {
+ mutex_unlock(&fs_info->chunk_mutex);
+ trans->removing_chunk = false;
+ }
/* once for us */
free_extent_map(em);
return ret;
@@ -4860,13 +4947,12 @@ static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
u32 array_size;
u8 *ptr;
- mutex_lock(&fs_info->chunk_mutex);
+ lockdep_assert_held(&fs_info->chunk_mutex);
+
array_size = btrfs_super_sys_array_size(super_copy);
if (array_size + item_size + sizeof(disk_key)
- > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
- mutex_unlock(&fs_info->chunk_mutex);
+ > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
return -EFBIG;
- }
ptr = super_copy->sys_chunk_array + array_size;
btrfs_cpu_key_to_disk(&disk_key, key);
@@ -4875,7 +4961,6 @@ static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
memcpy(ptr, chunk, item_size);
item_size += sizeof(disk_key);
btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
- mutex_unlock(&fs_info->chunk_mutex);
return 0;
}
@@ -5225,13 +5310,14 @@ static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
}
}
-static int create_chunk(struct btrfs_trans_handle *trans,
+static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
struct alloc_chunk_ctl *ctl,
struct btrfs_device_info *devices_info)
{
struct btrfs_fs_info *info = trans->fs_info;
struct map_lookup *map = NULL;
struct extent_map_tree *em_tree;
+ struct btrfs_block_group *block_group;
struct extent_map *em;
u64 start = ctl->start;
u64 type = ctl->type;
@@ -5241,7 +5327,7 @@ static int create_chunk(struct btrfs_trans_handle *trans,
map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
if (!map)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
map->num_stripes = ctl->num_stripes;
for (i = 0; i < ctl->ndevs; ++i) {
@@ -5263,7 +5349,7 @@ static int create_chunk(struct btrfs_trans_handle *trans,
em = alloc_extent_map();
if (!em) {
kfree(map);
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
}
set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
em->map_lookup = map;
@@ -5279,12 +5365,12 @@ static int create_chunk(struct btrfs_trans_handle *trans,
if (ret) {
write_unlock(&em_tree->lock);
free_extent_map(em);
- return ret;
+ return ERR_PTR(ret);
}
write_unlock(&em_tree->lock);
- ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
- if (ret)
+ block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
+ if (IS_ERR(block_group))
goto error_del_extent;
for (i = 0; i < map->num_stripes; i++) {
@@ -5304,7 +5390,7 @@ static int create_chunk(struct btrfs_trans_handle *trans,
check_raid56_incompat_flag(info, type);
check_raid1c34_incompat_flag(info, type);
- return 0;
+ return block_group;
error_del_extent:
write_lock(&em_tree->lock);
@@ -5316,34 +5402,36 @@ error_del_extent:
/* One for the tree reference */
free_extent_map(em);
- return ret;
+ return block_group;
}
-int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
+struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
+ u64 type)
{
struct btrfs_fs_info *info = trans->fs_info;
struct btrfs_fs_devices *fs_devices = info->fs_devices;
struct btrfs_device_info *devices_info = NULL;
struct alloc_chunk_ctl ctl;
+ struct btrfs_block_group *block_group;
int ret;
lockdep_assert_held(&info->chunk_mutex);
if (!alloc_profile_is_valid(type, 0)) {
ASSERT(0);
- return -EINVAL;
+ return ERR_PTR(-EINVAL);
}
if (list_empty(&fs_devices->alloc_list)) {
if (btrfs_test_opt(info, ENOSPC_DEBUG))
btrfs_debug(info, "%s: no writable device", __func__);
- return -ENOSPC;
+ return ERR_PTR(-ENOSPC);
}
if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
btrfs_err(info, "invalid chunk type 0x%llx requested", type);
ASSERT(0);
- return -EINVAL;
+ return ERR_PTR(-EINVAL);
}
ctl.start = find_next_chunk(info);
@@ -5353,46 +5441,43 @@ int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
GFP_NOFS);
if (!devices_info)
- return -ENOMEM;
+ return ERR_PTR(-ENOMEM);
ret = gather_device_info(fs_devices, &ctl, devices_info);
- if (ret < 0)
+ if (ret < 0) {
+ block_group = ERR_PTR(ret);
goto out;
+ }
ret = decide_stripe_size(fs_devices, &ctl, devices_info);
- if (ret < 0)
+ if (ret < 0) {
+ block_group = ERR_PTR(ret);
goto out;
+ }
- ret = create_chunk(trans, &ctl, devices_info);
+ block_group = create_chunk(trans, &ctl, devices_info);
out:
kfree(devices_info);
- return ret;
+ return block_group;
}
/*
- * Chunk allocation falls into two parts. The first part does work
- * that makes the new allocated chunk usable, but does not do any operation
- * that modifies the chunk tree. The second part does the work that
- * requires modifying the chunk tree. This division is important for the
- * bootstrap process of adding storage to a seed btrfs.
+ * This function, btrfs_finish_chunk_alloc(), belongs to phase 2.
+ *
+ * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
+ * phases.
*/
int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
u64 chunk_offset, u64 chunk_size)
{
struct btrfs_fs_info *fs_info = trans->fs_info;
- struct btrfs_root *extent_root = fs_info->extent_root;
- struct btrfs_root *chunk_root = fs_info->chunk_root;
- struct btrfs_key key;
struct btrfs_device *device;
- struct btrfs_chunk *chunk;
- struct btrfs_stripe *stripe;
struct extent_map *em;
struct map_lookup *map;
- size_t item_size;
u64 dev_offset;
u64 stripe_size;
- int i = 0;
+ int i;
int ret = 0;
em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
@@ -5400,53 +5485,117 @@ int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
return PTR_ERR(em);
map = em->map_lookup;
- item_size = btrfs_chunk_item_size(map->num_stripes);
stripe_size = em->orig_block_len;
- chunk = kzalloc(item_size, GFP_NOFS);
- if (!chunk) {
- ret = -ENOMEM;
- goto out;
- }
-
/*
* Take the device list mutex to prevent races with the final phase of
* a device replace operation that replaces the device object associated
* with the map's stripes, because the device object's id can change
* at any time during that final phase of the device replace operation
- * (dev-replace.c:btrfs_dev_replace_finishing()).
+ * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
+ * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
+ * resulting in persisting a device extent item with such ID.
*/
mutex_lock(&fs_info->fs_devices->device_list_mutex);
for (i = 0; i < map->num_stripes; i++) {
device = map->stripes[i].dev;
dev_offset = map->stripes[i].physical;
- ret = btrfs_update_device(trans, device);
- if (ret)
- break;
ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
dev_offset, stripe_size);
if (ret)
break;
}
- if (ret) {
- mutex_unlock(&fs_info->fs_devices->device_list_mutex);
+ mutex_unlock(&fs_info->fs_devices->device_list_mutex);
+
+ free_extent_map(em);
+ return ret;
+}
+
+/*
+ * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
+ * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
+ * chunks.
+ *
+ * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
+ * phases.
+ */
+int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
+ struct btrfs_block_group *bg)
+{
+ struct btrfs_fs_info *fs_info = trans->fs_info;
+ struct btrfs_root *extent_root = fs_info->extent_root;
+ struct btrfs_root *chunk_root = fs_info->chunk_root;
+ struct btrfs_key key;
+ struct btrfs_chunk *chunk;
+ struct btrfs_stripe *stripe;
+ struct extent_map *em;
+ struct map_lookup *map;
+ size_t item_size;
+ int i;
+ int ret;
+
+ /*
+ * We take the chunk_mutex for 2 reasons:
+ *
+ * 1) Updates and insertions in the chunk btree must be done while holding
+ * the chunk_mutex, as well as updating the system chunk array in the
+ * superblock. See the comment on top of btrfs_chunk_alloc() for the
+ * details;
+ *
+ * 2) To prevent races with the final phase of a device replace operation
+ * that replaces the device object associated with the map's stripes,
+ * because the device object's id can change at any time during that
+ * final phase of the device replace operation
+ * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
+ * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
+ * which would cause a failure when updating the device item, which does
+ * not exists, or persisting a stripe of the chunk item with such ID.
+ * Here we can't use the device_list_mutex because our caller already
+ * has locked the chunk_mutex, and the final phase of device replace
+ * acquires both mutexes - first the device_list_mutex and then the
+ * chunk_mutex. Using any of those two mutexes protects us from a
+ * concurrent device replace.
+ */
+ lockdep_assert_held(&fs_info->chunk_mutex);
+
+ em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
+ if (IS_ERR(em)) {
+ ret = PTR_ERR(em);
+ btrfs_abort_transaction(trans, ret);
+ return ret;
+ }
+
+ map = em->map_lookup;
+ item_size = btrfs_chunk_item_size(map->num_stripes);
+
+ chunk = kzalloc(item_size, GFP_NOFS);
+ if (!chunk) {
+ ret = -ENOMEM;
+ btrfs_abort_transaction(trans, ret);
goto out;
}
+ for (i = 0; i < map->num_stripes; i++) {
+ struct btrfs_device *device = map->stripes[i].dev;
+
+ ret = btrfs_update_device(trans, device);
+ if (ret)
+ goto out;
+ }
+
stripe = &chunk->stripe;
for (i = 0; i < map->num_stripes; i++) {
- device = map->stripes[i].dev;
- dev_offset = map->stripes[i].physical;
+ struct btrfs_device *device = map->stripes[i].dev;
+ const u64 dev_offset = map->stripes[i].physical;
btrfs_set_stack_stripe_devid(stripe, device->devid);
btrfs_set_stack_stripe_offset(stripe, dev_offset);
memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
stripe++;
}
- mutex_unlock(&fs_info->fs_devices->device_list_mutex);
- btrfs_set_stack_chunk_length(chunk, chunk_size);
+ btrfs_set_stack_chunk_length(chunk, bg->length);
btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
btrfs_set_stack_chunk_type(chunk, map->type);
@@ -5458,15 +5607,18 @@ int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
key.type = BTRFS_CHUNK_ITEM_KEY;
- key.offset = chunk_offset;
+ key.offset = bg->start;
ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
- if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
- /*
- * TODO: Cleanup of inserted chunk root in case of
- * failure.
- */
+ if (ret)
+ goto out;
+
+ bg->chunk_item_inserted = 1;
+
+ if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
+ if (ret)
+ goto out;
}
out:
@@ -5479,16 +5631,41 @@ static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
{
struct btrfs_fs_info *fs_info = trans->fs_info;
u64 alloc_profile;
- int ret;
+ struct btrfs_block_group *meta_bg;
+ struct btrfs_block_group *sys_bg;
+
+ /*
+ * When adding a new device for sprouting, the seed device is read-only
+ * so we must first allocate a metadata and a system chunk. But before
+ * adding the block group items to the extent, device and chunk btrees,
+ * we must first:
+ *
+ * 1) Create both chunks without doing any changes to the btrees, as
+ * otherwise we would get -ENOSPC since the block groups from the
+ * seed device are read-only;
+ *
+ * 2) Add the device item for the new sprout device - finishing the setup
+ * of a new block group requires updating the device item in the chunk
+ * btree, so it must exist when we attempt to do it. The previous step
+ * ensures this does not fail with -ENOSPC.
+ *
+ * After that we can add the block group items to their btrees:
+ * update existing device item in the chunk btree, add a new block group
+ * item to the extent btree, add a new chunk item to the chunk btree and
+ * finally add the new device extent items to the devices btree.
+ */
alloc_profile = btrfs_metadata_alloc_profile(fs_info);
- ret = btrfs_alloc_chunk(trans, alloc_profile);
- if (ret)
- return ret;
+ meta_bg = btrfs_alloc_chunk(trans, alloc_profile);
+ if (IS_ERR(meta_bg))
+ return PTR_ERR(meta_bg);
alloc_profile = btrfs_system_alloc_profile(fs_info);
- ret = btrfs_alloc_chunk(trans, alloc_profile);
- return ret;
+ sys_bg = btrfs_alloc_chunk(trans, alloc_profile);
+ if (IS_ERR(sys_bg))
+ return PTR_ERR(sys_bg);
+
+ return 0;
}
static inline int btrfs_chunk_max_errors(struct map_lookup *map)
@@ -7415,10 +7592,18 @@ int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
total_dev++;
} else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
struct btrfs_chunk *chunk;
+
+ /*
+ * We are only called at mount time, so no need to take
+ * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
+ * we always lock first fs_info->chunk_mutex before
+ * acquiring any locks on the chunk tree. This is a
+ * requirement for chunk allocation, see the comment on
+ * top of btrfs_chunk_alloc() for details.
+ */
+ ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
- mutex_lock(&fs_info->chunk_mutex);
ret = read_one_chunk(&found_key, leaf, chunk);
- mutex_unlock(&fs_info->chunk_mutex);
if (ret)
goto error;
}