Re: [PATCH] Btrfs: fix tree corruption after multi-thread snapshots and inode cache flush

[Liu Bo <liubo2009@cn.fujitsu.com>]

On 09/29/2011 12:25 PM, Yan, Zheng wrote:
> On 09/29/2011 10:00 AM, Liu Bo wrote:
>> The btrfs snapshotting code requires that once a root has been
>> snapshotted, we don't change it during a commit.
>>
>> But there are two cases to lead to tree corruptions:
>>
>> 1) multi-thread snapshots can commit serveral snapshots in a transaction,
>>    and this may change the src root when processing the following pending
>>    snapshots, which lead to the former snapshots corruptions;
>>
>> 2) the free inode cache was changing the roots when it root the cache,
>>    which lead to corruptions.
>>
> For the case 2, the free inode cache of newly created snapshot is invalid.
> So it's better to avoid modifying snapshotted trees.
> 

For case 2, with flushing dirty inode cache during create_pending_snapshot,
we can avoid modifying snapshotted trees as your advice.

But for case 1, I have no idea how to do the same thing, since we are not
allowed to commit per snapshot, which will make the performance terrible.

thanks,
liubo


>> This fixes things by making sure we force COW the block after we create a
>> snapshot during commiting a transaction, then any changes to the roots
>> will result in COW, and we get all the fs roots and snapshot roots to be
>> consistent.
>>
>> Signed-off-by: Liu Bo <liubo2009@cn.fujitsu.com>
>> Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
>> ---
>>  fs/btrfs/ctree.c       |   17 ++++++++++++++++-
>>  fs/btrfs/ctree.h       |    2 ++
>>  fs/btrfs/transaction.c |    8 ++++++++
>>  3 files changed, 26 insertions(+), 1 deletions(-)
>>
>> diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
>> index 011cab3..49dad7d 100644
>> --- a/fs/btrfs/ctree.c
>> +++ b/fs/btrfs/ctree.c
>> @@ -514,10 +514,25 @@ static inline int should_cow_block(struct btrfs_trans_handle *trans,
>>  				   struct btrfs_root *root,
>>  				   struct extent_buffer *buf)
>>  {
>> +	/* ensure we can see the force_cow */
>> +	smp_rmb();
>> +
>> +	/*
>> +	 * We do not need to cow a block if
>> +	 * 1) this block is not created or changed in this transaction;
>> +	 * 2) this block does not belong to TREE_RELOC tree;
>> +	 * 3) the root is not forced COW.
>> +	 *
>> +	 * What is forced COW:
>> +	 *    when we create snapshot during commiting the transaction,
>> +	 *    after we've finished coping src root, we must COW the shared
>> +	 *    block to ensure the metadata consistency.
>> +	 */
>>  	if (btrfs_header_generation(buf) == trans->transid &&
>>  	    !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
>>  	    !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
>> -	      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
>> +	      btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
>> +	    !root->force_cow)
>>  		return 0;
>>  	return 1;
>>  }
>> diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
>> index 03912c5..bece0df 100644
>> --- a/fs/btrfs/ctree.h
>> +++ b/fs/btrfs/ctree.h
>> @@ -1225,6 +1225,8 @@ struct btrfs_root {
>>  	 * for stat.  It may be used for more later
>>  	 */
>>  	dev_t anon_dev;
>> +
>> +	int force_cow;
>>  };
>>  
>>  struct btrfs_ioctl_defrag_range_args {
>> diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
>> index 7dc36fa..bf6e2b3 100644
>> --- a/fs/btrfs/transaction.c
>> +++ b/fs/btrfs/transaction.c
>> @@ -816,6 +816,10 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
>>  
>>  			btrfs_save_ino_cache(root, trans);
>>  
>> +			/* see comments in should_cow_block() */
>> +			root->force_cow = 0;
>> +			smp_wmb();
>> +
>>  			if (root->commit_root != root->node) {
>>  				mutex_lock(&root->fs_commit_mutex);
>>  				switch_commit_root(root);
>> @@ -976,6 +980,10 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
>>  	btrfs_tree_unlock(old);
>>  	free_extent_buffer(old);
>>  
>> +	/* see comments in should_cow_block() */
>> +	root->force_cow = 1;
>> +	smp_wmb();
>> +
>>  	btrfs_set_root_node(new_root_item, tmp);
>>  	/* record when the snapshot was created in key.offset */
>>  	key.offset = trans->transid;
> 
>