Re: [PATCH] btrfs: fix error propagation of split bios

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

在 2024/10/10 08:28, Qu Wenruo 写道:
>
>
> 在 2024/10/10 08:08, Qu Wenruo 写道:
[...]
>> And __bi_remaining is only decreased when the cloned or the original bio
>> get its endio function called (bio_endio()).
>>
>> For cloned bios, it's mostly the same chained bio behavior, with extra
>> btrfs write tolerance added.
>
> OK, I see the point. For the cloned ones we can have the following case:
>
> The profile is DUP/RAID1.
>
> For stripe 0, we cloned the original bio, increased
> orig_bio->__bi_remaining.
>
> Then submitted the cloned bio.
>
> But before submitting the original one, cloned one get finished first,
> it call the cloned endio function, which calls back to the endio of the
> original bio.
>
> Then the endio function decrease the __bi_remaining to 0 of the original
> bio, thus it continue to call the endio of the original bio, which freed
> the original bio.

My bad, this is not possible.

The original bio will have 1 as __bi_remaining as the initial value.

So even if the cloned one is finished first, the __bi_remaining will
only stay at 1, not reaching 0, so the original bio will not finish,
thus impossible to free the original bio.

I need to dig further deeper to find out why the NULL pointer
dereference happens.

Thanks,
Qu