From: Johannes Thumshirn <johannes.thumshirn@wdc.com>

Curretnly we BUG_ON() in btrfs_finish_one_ordered() if we finishing an
ordered-extent that is flagged as NOCOW, but it's checsum list is non-empty.

This is clearly a logic error which we can recover from by aborting the
transaction.

For developer builds which enable CONFIG_BTRFS_ASSERT, also ASSERT() that the
list is empty.

Suggested-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---
 fs/btrfs/inode.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 103ec917ca9d..19ba101dc09c 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3088,7 +3088,10 @@ int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
 
 	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
 		/* Logic error */
-		BUG_ON(!list_empty(&ordered_extent->list));
+		if (list_empty(&ordered_extent->list)) {
+			ASSERT(list_empty(&ordered_extent->list));
+			btrfs_abort_transaction(trans, -EINVAL);
+		}
 
 		btrfs_inode_safe_disk_i_size_write(inode, 0);
 		ret = btrfs_update_inode_fallback(trans, inode);
-- 
2.43.0

在 2024/10/4 18:53, Johannes Thumshirn 写道:
> From: Johannes Thumshirn <johannes.thumshirn@wdc.com>
>
> Curretnly we BUG_ON() in btrfs_finish_one_ordered() if we finishing an
> ordered-extent that is flagged as NOCOW, but it's checsum list is non-empty.
>
> This is clearly a logic error which we can recover from by aborting the
> transaction.
>
> For developer builds which enable CONFIG_BTRFS_ASSERT, also ASSERT() that the
> list is empty.
>
> Suggested-by: Filipe Manana <fdmanana@suse.com>
> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> ---
>   fs/btrfs/inode.c | 5 ++++-
>   1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
> index 103ec917ca9d..19ba101dc09c 100644
> --- a/fs/btrfs/inode.c
> +++ b/fs/btrfs/inode.c
> @@ -3088,7 +3088,10 @@ int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
>
>   	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
>   		/* Logic error */
> -		BUG_ON(!list_empty(&ordered_extent->list));
> +		if (list_empty(&ordered_extent->list)) {
> +			ASSERT(list_empty(&ordered_extent->list));

Will the ASSERT() really get triggered? We just checked the same
list_empty() one line before.

I guess you mean ASSERT(!list_empty()) instead?

Otherwise changing it to ASSERT() and btrfs_abort_transaction() looks
good to me.

Thanks,
Qu

> +			btrfs_abort_transaction(trans, -EINVAL);
> +		}
>
>   		btrfs_inode_safe_disk_i_size_write(inode, 0);
>   		ret = btrfs_update_inode_fallback(trans, inode);

On 04.10.24 11:31, Qu Wenruo wrote:
> 
> 
> 在 2024/10/4 18:53, Johannes Thumshirn 写道:
>> From: Johannes Thumshirn <johannes.thumshirn@wdc.com>
>>
>> Curretnly we BUG_ON() in btrfs_finish_one_ordered() if we finishing an
>> ordered-extent that is flagged as NOCOW, but it's checsum list is non-empty.
>>
>> This is clearly a logic error which we can recover from by aborting the
>> transaction.
>>
>> For developer builds which enable CONFIG_BTRFS_ASSERT, also ASSERT() that the
>> list is empty.
>>
>> Suggested-by: Filipe Manana <fdmanana@suse.com>
>> Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
>> ---
>>    fs/btrfs/inode.c | 5 ++++-
>>    1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
>> index 103ec917ca9d..19ba101dc09c 100644
>> --- a/fs/btrfs/inode.c
>> +++ b/fs/btrfs/inode.c
>> @@ -3088,7 +3088,10 @@ int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
>>
>>    	if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
>>    		/* Logic error */
>> -		BUG_ON(!list_empty(&ordered_extent->list));
>> +		if (list_empty(&ordered_extent->list)) {
>> +			ASSERT(list_empty(&ordered_extent->list));
> 
> Will the ASSERT() really get triggered? We just checked the same
> list_empty() one line before.
> 
> I guess you mean ASSERT(!list_empty()) instead?
> 
> Otherwise changing it to ASSERT() and btrfs_abort_transaction() looks
> good to me.


Of cause you're right! Seems like I need more coffee.