]> git.openfabrics.org - ~shefty/rdma-dev.git/commitdiff
do_exit(): make sure that we run with get_fs() == USER_DS
authorNelson Elhage <nelhage@ksplice.com>
Thu, 2 Dec 2010 22:31:21 +0000 (14:31 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 2 Dec 2010 22:51:16 +0000 (14:51 -0800)
If a user manages to trigger an oops with fs set to KERNEL_DS, fs is not
otherwise reset before do_exit().  do_exit may later (via mm_release in
fork.c) do a put_user to a user-controlled address, potentially allowing
a user to leverage an oops into a controlled write into kernel memory.

This is only triggerable in the presence of another bug, but this
potentially turns a lot of DoS bugs into privilege escalations, so it's
worth fixing.  I have proof-of-concept code which uses this bug along
with CVE-2010-3849 to write a zero to an arbitrary kernel address, so
I've tested that this is not theoretical.

A more logical place to put this fix might be when we know an oops has
occurred, before we call do_exit(), but that would involve changing
every architecture, in multiple places.

Let's just stick it in do_exit instead.

[akpm@linux-foundation.org: update code comment]
Signed-off-by: Nelson Elhage <nelhage@ksplice.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/exit.c

index 21aa7b3001fb49edce43446dd85898f62ab33d98..676149a4ac5ff497367a484e2b66c01e915ccefd 100644 (file)
@@ -914,6 +914,15 @@ NORET_TYPE void do_exit(long code)
        if (unlikely(!tsk->pid))
                panic("Attempted to kill the idle task!");
 
+       /*
+        * If do_exit is called because this processes oopsed, it's possible
+        * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
+        * continuing. Amongst other possible reasons, this is to prevent
+        * mm_release()->clear_child_tid() from writing to a user-controlled
+        * kernel address.
+        */
+       set_fs(USER_DS);
+
        tracehook_report_exit(&code);
 
        validate_creds_for_do_exit(tsk);