Re: [PATCH v2 1/3] btrfs: avoid double clean up when submit_one_bio() failed

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

On 2022/4/13 21:46, David Sterba wrote:
> On Wed, Apr 13, 2022 at 07:32:41AM +0800, Qu Wenruo wrote:
>> On 2022/4/13 04:41, David Sterba wrote:
>>> On Tue, Apr 12, 2022 at 08:30:13PM +0800, Qu Wenruo wrote:
>>>> The commit 1784b7d502a9 ("btrfs: handle csum lookup errors properly on
>>>> reads") itself is completely fine, it's our existing code not properly
>>>> handling the error from bio submission hook properly.
>>>>
>>>> This patch will make submit_one_bio() to return void so that the callers
>>>> will never be able to do cleanup when bio submission hook fails.
>>>>
>>>> CC: stable@vger.kernel.org # 5.18+
>>>
>>> BTW stable tags are only for released kernels, if it's still in some rc
>>> then Fixes: is appropriate.
>>
>> The problem is I don't have a good idea on which commit to be fixed.
>>
>> Commit 1784b7d502a9 ("btrfs: handle csum lookup errors properly on
>> reads") is completely fine by itself.
>>
>> The problem is there for a long long time, but can only be triggered
>> with IO errors with that newer commit.
>>
>> Should we really use that commit? It looks like a scapegoat to me...
>
> I see, so it does not make sense to put Fixes: if it's not clearly
> caused by the patch, the text description of the problem and references
> to patches that could affect is OK.
>
> Still the stable tag should reflect where the fix applies but 5.18
> hasn't been released so either it's a typo or you know roughly which
> stable kernels should get the fix (5.15, 5.10, etc).

Then I guess we can drop the stable tag.

Before that mentioned commit, btrfs_lookup_bio_csum() will never return
error, thus submit_one_bio() will not really fail (due to IO error).

Although the error path is there for a long long time, we don't have
easy way to trigger the problem.

Thanks,
Qu