Re: [PATCH v5 2/3] md/raid10: resize r10bio_pool for reshape

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Critical] Replacing `conf->r10bio_pool` during `raid10_start_reshape()` races with discard I/O completion, leading to mempool Use-After-Free and heap out-of-bounds writes.
--

commit 1d797159800c9f37dc1886dd00e8c5968da7d731
Author: Chen Cheng <[email protected]>
Subject: md/raid10: resize r10bio_pool for reshape

This commit updates raid10_start_reshape() to allocate a new, larger
r10bio_pool when the number of raid disks changes. The old pool is then
destroyed and replaced with the new one.

> diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
> --- a/drivers/md/raid10.c
> +++ b/drivers/md/raid10.c
> @@ -4511,6 +4527,10 @@ static int raid10_start_reshape(struct mddev *mddev)
>  	mddev->raid_disks = conf->geo.raid_disks;
>  	mddev->reshape_position = conf->reshape_progress;
>  	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
> +	if (new_pool) {
> +		mempool_destroy(conf->r10bio_pool);
> +		conf->r10bio_pool = new_pool;
> +	}

[Severity: Critical]
Does this introduce a race condition with discard I/O completion?

In the discard completion path, raid_end_discard_bio() calls bio_endio()
on the master bio before freeing the r10bio:

raid_end_discard_bio() {
        ...
        bio_endio(r10bio->master_bio);
        ...
        free_r10bio(r10bio);
}

Calling bio_endio() invokes md_end_clone_io(), which decrements the
mddev->active_io refcount. If this drops to zero, mddev_suspend() can return
while the thread is still executing raid_end_discard_bio().

If that happens, raid10_start_reshape() can proceed to destroy
conf->r10bio_pool and assign new_pool here, while free_r10bio() is still
about to use it.

Could this lead to free_r10bio() releasing the object into a destroyed
mempool, or freeing a smaller r10_bio object into the newly allocated pool
designed for larger objects (which could cause a heap out-of-bounds write
later when __make_request() clears the devs array)?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2