linux-kernel - Re: [PATCH] fat: Prevent the race of read/write the FAT16 and FAT32 entry

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]

Message-ID: <20250729050413.GF222315@ZenIV>
Date: Tue, 29 Jul 2025 06:04:13 +0100
From: Al Viro <viro@...iv.linux.org.uk>
To: Edward Adam Davis <eadavis@...com>
Cc: hirofumi@...l.parknet.co.jp, linkinjeon@...nel.org,
	linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
	sj1557.seo@...sung.com,
	syzbot+d3c29ed63db6ddf8406e@...kaller.appspotmail.com,
	syzkaller-bugs@...glegroups.com
Subject: Re: [PATCH] fat: Prevent the race of read/write the FAT16 and FAT32
 entry

On Tue, Jul 29, 2025 at 05:53:23AM +0100, Al Viro wrote:

> FAT12 problem is that FAT entries being accessed there are 12-bit, packed in
> pairs into an array of 3-byte values.

PS: they most definitely can cross the cacheline boundaries - cacheline size
is not going to be a multiple of 3 on anything realistic.  Hell, they can
cross *block* boundaries (which is why for FAT12 that code is using two
separate pointers - most of the time they point to adjacent bytes, but
if one byte is in one block and the next one is in another...; that's
what that if in fat12_ent_set_ptr() is doing)...

We could map the entire array contiguously (and it might simplify some of
the logics there), but it's not going to avoid the problem with a single
entry occupying a byte in one cacheline and half of a byte in another.

So nothing like cmpxchg would suffice - we need a spinlock for FAT12 case.