full-disclosure - Locked fasync file descriptors can be referenced after free in >= 2.6.28

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>] [day] [month] [year] [list]

Date: Thu, 14 Jan 2010 16:45:02 +0100
From: Tavis Ormandy <taviso@....lonestar.org>
To: full-disclosure@...ts.grok.org.uk
Subject: Locked fasync file descriptors can be referenced
	after free in >= 2.6.28

Locked fasync file descriptors can be referenced after free in >= 2.6.28
-------------------------------------------------------------------------

A problem has been discovered handling locked fasync file descriptors.

Quoting analysis from Linus, "The problem is that the *same* file
descriptor can be on *multiple* fasync lists. It can be on a particular
fasync list just once, but file locking is special, and uses the
'fl->fl_fasync' list to add an arbitrary file to its own fasync list -
regardless of what the underlying device driver or whatever is."

This is an issue because fasync_helper() will clear the FASYNC flag as
it incorrectly assumes a file can only be on one fasync list.

http://lxr.linux.no/#linux+v2.6.30.4/fs/fcntl.c#L566

When the descriptor is finally closed and the file released, the FASYNC
flag will no longer be set and so it will not be removed the fasync list,
thus leaving a dangling reference to the freed structure.

A testcase for this issue which should result in an OOPS caused by a
userland pointer dereference is available here.

http://lock.cmpxchg8b.com/5ebe2294ecd0e0f08eab7690d2a6ee69/create_elf_tables.c

(gdb) bt
#0  __kill_fasync (fa=0xdf294820, sig=29, band=131073) at
/build/buildd/linux-2.6.31/fs/fcntl.c:587
#1  0xc01f4a32 in kill_fasync (fp=0xc089be64, sig=29, band=131073) at
/build/buildd/linux-2.6.31/fs/fcntl.c:614
(gdb) list
592                             return;
593                     }
594                     fown = &fa->fa_file->f_owner;
595                     /* Don't send SIGURG to processes which have not set a
596                        queued signum: SIGURG has its own default signalling
597                        mechanism. */
598                     if (!(sig == SIGURG && fown->signum == 0))
599                             send_sigio(fown, fa->fa_fd, band);
600                     fa = fa->fa_next;
601             }
(gdb) p fa->fa_file->f_owner
$37 = {lock = {raw_lock = {lock = 2621449}}, pid = 0x1c001d, pid_type =
1179403647, uid = 65793, euid = 0, signum = 0}

(this is obviously nonsense, and send_sigio is going to explode)

http://lxr.linux.no/#linux+v2.6.32/fs/fcntl.c#L550

Linus has committed the following patch to address this issue
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=53281b6d3

This bug was discovered by Tavis Ormandy.

Thanks to legendary kernel hackers Linus, Matt Mackall and Al Viro who
did most of the hard work tracking down the root cause.

-- 
-------------------------------------
taviso@....lonestar.org | finger me for my gpg key.
-------------------------------------------------------

_______________________________________________
Full-Disclosure - We believe in it.
Charter: http://lists.grok.org.uk/full-disclosure-charter.html
Hosted and sponsored by Secunia - http://secunia.com/