linux-kernel - Re: bio linked list corruption.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 19 Oct 2016 10:43:59 -0700
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Philipp Hahn <pmhahn@...ahn.de>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...nel.org>
Cc:     Chris Mason <clm@...com>, Jens Axboe <axboe@...com>,
        Dave Jones <davej@...emonkey.org.uk>,
        Al Viro <viro@...iv.linux.org.uk>, Josef Bacik <jbacik@...com>,
        David Sterba <dsterba@...e.com>,
        linux-btrfs <linux-btrfs@...r.kernel.org>,
        Linux Kernel <linux-kernel@...r.kernel.org>
Subject: Re: bio linked list corruption.

On Wed, Oct 19, 2016 at 10:09 AM, Philipp Hahn <pmhahn@...ahn.de> wrote:
>
> Nearly a month ago I reported also a "list_add corruption", but with 4.1.6:
> <http://marc.info/?l=linux-kernel&m=147508265316854&w=2>
>
> That server rungs Samba4, which also is a heavy user of xattr.

That one looks very different. In fact, the list that got corrupted
for you has since been changed to a hlist (which is *similar* to our
doubly-linked list, but has a smaller head and does not allow adding
to the end of the list).

Also, the "should be" and "was" values are very close, and switched:

should be ffffffff81ab3ca8, but was ffffffff81ab3cc8
should be ffffffff81ab3cc8, but was ffffffff81ab3ca8

so it actually looks like it was the same data structure. In
particular, it looks like enqueue_timer() ended up racing on adding an
entry to one index in the "base->vectors[]" array, while hitting an
entry that was pointing to another index near-by.

So I don't think it's related. Yours looks like some subtle timer base
race. It smells like a locking problem with timers. I'm not seeing
what it might be, but it *might* have been fixed by doing the
TIMER_MIGRATING bit right in add_timer_on() (commit 22b886dd1018).

Adding some timer people just in case, but I don't think your 4.1
report is related.

                Linus