Commit a9d45396 authored by Joe Thornber's avatar Joe Thornber Committed by Mike Snitzer

dm transaction manager: fix corruption due to non-atomic transaction commit

The persistent-data library used by dm-thin, dm-cache, etc is
transactional.  If anything goes wrong, such as an io error when writing
new metadata or a power failure, then we roll back to the last
transaction.

Atomicity when committing a transaction is achieved by:

a) Never overwriting data from the previous transaction.
b) Writing the superblock last, after all other metadata has hit the
   disk.

This commit and the following commit ("dm: take care to copy the space
map roots before locking the superblock") fix a bug associated with (b).
When committing it was possible for the superblock to still be written
in spite of an io error occurring during the preceeding metadata flush.
With these commits we're careful not to take the write lock out on the
superblock until after the metadata flush has completed.

Change the transaction manager's semantics for dm_tm_commit() to assume
all data has been flushed _before_ the single superblock that is passed
in.

As a prerequisite, split the block manager's block unlocking and
flushing by simplifying dm_bm_flush_and_unlock() to dm_bm_flush().  Now
the unlocking must be done separately.

This issue was discovered by forcing io errors at the crucial time
using dm-flakey.
Signed-off-by: default avatarJoe Thornber <ejt@redhat.com>
Signed-off-by: default avatarMike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
parent 64ab346a
...@@ -530,8 +530,9 @@ static int __begin_transaction_flags(struct dm_cache_metadata *cmd, ...@@ -530,8 +530,9 @@ static int __begin_transaction_flags(struct dm_cache_metadata *cmd,
disk_super = dm_block_data(sblock); disk_super = dm_block_data(sblock);
update_flags(disk_super, mutator); update_flags(disk_super, mutator);
read_superblock_fields(cmd, disk_super); read_superblock_fields(cmd, disk_super);
dm_bm_unlock(sblock);
return dm_bm_flush_and_unlock(cmd->bm, sblock); return dm_bm_flush(cmd->bm);
} }
static int __begin_transaction(struct dm_cache_metadata *cmd) static int __begin_transaction(struct dm_cache_metadata *cmd)
......
...@@ -595,25 +595,14 @@ int dm_bm_unlock(struct dm_block *b) ...@@ -595,25 +595,14 @@ int dm_bm_unlock(struct dm_block *b)
} }
EXPORT_SYMBOL_GPL(dm_bm_unlock); EXPORT_SYMBOL_GPL(dm_bm_unlock);
int dm_bm_flush_and_unlock(struct dm_block_manager *bm, int dm_bm_flush(struct dm_block_manager *bm)
struct dm_block *superblock)
{ {
int r;
if (bm->read_only) if (bm->read_only)
return -EPERM; return -EPERM;
r = dm_bufio_write_dirty_buffers(bm->bufio);
if (unlikely(r)) {
dm_bm_unlock(superblock);
return r;
}
dm_bm_unlock(superblock);
return dm_bufio_write_dirty_buffers(bm->bufio); return dm_bufio_write_dirty_buffers(bm->bufio);
} }
EXPORT_SYMBOL_GPL(dm_bm_flush_and_unlock); EXPORT_SYMBOL_GPL(dm_bm_flush);
void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b) void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b)
{ {
......
...@@ -105,8 +105,7 @@ int dm_bm_unlock(struct dm_block *b); ...@@ -105,8 +105,7 @@ int dm_bm_unlock(struct dm_block *b);
* *
* This method always blocks. * This method always blocks.
*/ */
int dm_bm_flush_and_unlock(struct dm_block_manager *bm, int dm_bm_flush(struct dm_block_manager *bm);
struct dm_block *superblock);
/* /*
* Request data is prefetched into the cache. * Request data is prefetched into the cache.
......
...@@ -154,7 +154,7 @@ int dm_tm_pre_commit(struct dm_transaction_manager *tm) ...@@ -154,7 +154,7 @@ int dm_tm_pre_commit(struct dm_transaction_manager *tm)
if (r < 0) if (r < 0)
return r; return r;
return 0; return dm_bm_flush(tm->bm);
} }
EXPORT_SYMBOL_GPL(dm_tm_pre_commit); EXPORT_SYMBOL_GPL(dm_tm_pre_commit);
...@@ -164,8 +164,9 @@ int dm_tm_commit(struct dm_transaction_manager *tm, struct dm_block *root) ...@@ -164,8 +164,9 @@ int dm_tm_commit(struct dm_transaction_manager *tm, struct dm_block *root)
return -EWOULDBLOCK; return -EWOULDBLOCK;
wipe_shadow_table(tm); wipe_shadow_table(tm);
dm_bm_unlock(root);
return dm_bm_flush_and_unlock(tm->bm, root); return dm_bm_flush(tm->bm);
} }
EXPORT_SYMBOL_GPL(dm_tm_commit); EXPORT_SYMBOL_GPL(dm_tm_commit);
......
...@@ -38,18 +38,17 @@ struct dm_transaction_manager *dm_tm_create_non_blocking_clone(struct dm_transac ...@@ -38,18 +38,17 @@ struct dm_transaction_manager *dm_tm_create_non_blocking_clone(struct dm_transac
/* /*
* We use a 2-phase commit here. * We use a 2-phase commit here.
* *
* i) In the first phase the block manager is told to start flushing, and * i) Make all changes for the transaction *except* for the superblock.
* the changes to the space map are written to disk. You should interrogate * Then call dm_tm_pre_commit() to flush them to disk.
* your particular space map to get detail of its root node etc. to be
* included in your superblock.
* *
* ii) @root will be committed last. You shouldn't use more than the * ii) Lock your superblock. Update. Then call dm_tm_commit() which will
* first 512 bytes of @root if you wish the transaction to survive a power * unlock the superblock and flush it. No other blocks should be updated
* failure. You *must* have a write lock held on @root for both stage (i) * during this period. Care should be taken to never unlock a partially
* and (ii). The commit will drop the write lock. * updated superblock; perform any operations that could fail *before* you
* take the superblock lock.
*/ */
int dm_tm_pre_commit(struct dm_transaction_manager *tm); int dm_tm_pre_commit(struct dm_transaction_manager *tm);
int dm_tm_commit(struct dm_transaction_manager *tm, struct dm_block *root); int dm_tm_commit(struct dm_transaction_manager *tm, struct dm_block *superblock);
/* /*
* These methods are the only way to get hold of a writeable block. * These methods are the only way to get hold of a writeable block.
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment