[5.15.y] Request to backport c567c86b90d4 (md/raid10 fstrim kernel crash)

[5.15.y] Request to backport c567c86b90d4 (md/raid10 fstrim kernel crash)

[EOS - Manlio Morini]

Hello,

I'd like to request the backport of the following upstream commit to the 
Linux 5.15.y stable series:

---

c567c86b90d4715081adfe5eb812141a5b6b4883
md: move initialization and destruction of 'io_acct_set' to md.c

---

The regression was introduced in 5.15.181 by commit 4a05f7ae3371 
("md/raid10: fix missing discard IO accounting"), which depends on the 
above commit.

Without this prerequisite, running fstrim on an MD RAID10 array can 
cause a NULL pointer dereference and crash the kernel.

I have reproduced this crash on my Slackware workstation running Linux 
5.15.223.

The issue was independently investigated and fixed by Canonical in 
Ubuntu's 5.15 kernel:

* https://bugs.launchpad.net/bugs/2117395

*https://lists.ubuntu.com/archives/kernel-team/2025-July/161560.html

The prerequisite was also backported to the Linux 6.1.y stable series:

* https://lists.openwall.net/linux-kernel/2025/05/06/1629

My original report:

* 
https://blog.eosdev.it/2025/07/attenzione-fstrim-su-raid-10-crash-del.html

Since this is a reproducible kernel crash caused by a missing dependency 
of an already backported commit, could the necessary fix also be 
considered for the 5.15.y stable series?

Thank you for your work on maintaining the stable kernels.

Best regards,
Manlio Morini

Re: [5.15.y] Request to backport c567c86b90d4 (md/raid10 fstrim kernel crash)

[Sasha Levin]

> I'd like to request the backport of the following upstream commit to the
> Linux 5.15.y stable series:
>
> c567c86b90d4715081adfe5eb812141a5b6b4883
> md: move initialization and destruction of 'io_acct_set' to md.c

Queued for 5.15, thanks. The io_acct_set exit went into md_stop() and
md_free() rather than __md_stop(), since that is where 5.15 tears down the
other biosets.

-- 
Thanks,
Sasha