Re: perf: use after free in perf_remove_from_context

[Peter Zijlstra]

On Wed, May 28, 2014 at 10:31:38PM -0400, Sasha Levin wrote:
> I've just had this:
> 
> 
> [  591.111854] general protection fault: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
> [  591.121057] Dumping ftrace buffer:
> [  591.121057]    (ftrace buffer empty)
> [  591.121057] Modules linked in:
> [  591.121057] CPU: 14 PID: 10272 Comm: trinity-c230 Tainted: G        W     3.15.0-rc7-next-20140528-sasha-00019-gd193e91-dirty #545
> [  591.121057] task: ffff8802df0d3000 ti: ffff8802df0fe000 task.ti: ffff8802df0fe000
> [  591.129130] RIP: perf_event_exit_task (kernel/events/core.c:7492 kernel/events/core.c:7524)
> [  591.129130] RSP: 0000:ffff8802df0ffe28  EFLAGS: 00010246
> [  591.129130] RAX: 0000000000000000 RBX: ffff8802df0d3000 RCX: 0000000000000001
> [  591.129130] RDX: 0000000000000000 RSI: ffffffff97088100 RDI: 0000000000000282
> [  591.129130] RBP: ffff8802df0ffe98 R08: 0000000000000000 R09: 0000000000000000
> [  591.129130] R10: 0000000000000001 R11: 0000000000000000 R12: 6b6b6b6b6b6b6b6b
> [  591.129130] R13: 6b6b6b6b6b6b6b6b R14: ffff8806c9bfd550 R15: 0000000000000001
> [  591.129130] FS:  00007f4b26866700(0000) GS:ffff88036ae00000(0000) knlGS:0000000000000000
> [  591.129130] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
> [  591.129130] CR2: 0000000001a3a9f8 CR3: 000000001702c000 CR4: 00000000000006a0
> [  591.129130] Stack:
> [  591.129130]  ffff8802df0ffe48 ffff8802df09cce0 ffff8802fdf99488 ffff8802fdf99290
> [  591.129130]  ffff8806c9bfd5a8 ffff8806c9bfd670 ffff8802df0ffe88 6b6b6b6b6b6b6b6b
> [  591.129130]  ffff8802df0ffe98 0000000000000001 0000000000000001 ffff8802df0d3000
> [  591.129130] Call Trace:
> [  591.129130] do_exit (kernel/exit.c:766)
> [  591.129130] ? __this_cpu_preempt_check (lib/smp_processor_id.c:63)
> [  591.129130] ? trace_hardirqs_on_caller (kernel/locking/lockdep.c:2557 kernel/locking/lockdep.c:2599)
> [  591.129130] ? trace_hardirqs_on (kernel/locking/lockdep.c:2607)
> [  591.129130] do_group_exit (kernel/exit.c:884)
> [  591.129130] SyS_exit_group (kernel/exit.c:895)
> [  591.129130] tracesys (arch/x86/kernel/entry_64.S:542)
> [ 591.129130] Code: a0 02 00 00 49 89 95 a8 02 00 00 4c 8b 45 a0 4c 89 c7 e8 11 c2 2c 03 48 8b 45 a8 48 89 c7 e8 15 7c ff ff 4c 89 ef e8 bd 7b ff ff <49> 8b 45 00 48 89 45 c8 4c 8b 6d c8 e8 cc 19 f6 ff 85 c0 74 0e
> All code
> ========
>    0:	a0 02 00 00 49 89 95 	movabs 0x2a8958949000002,%al
>    7:	a8 02
>    9:	00 00                	add    %al,(%rax)
>    b:	4c 8b 45 a0          	mov    -0x60(%rbp),%r8
>    f:	4c 89 c7             	mov    %r8,%rdi
>   12:	e8 11 c2 2c 03       	callq  0x32cc228
>   17:	48 8b 45 a8          	mov    -0x58(%rbp),%rax
>   1b:	48 89 c7             	mov    %rax,%rdi
>   1e:	e8 15 7c ff ff       	callq  0xffffffffffff7c38
>   23:	4c 89 ef             	mov    %r13,%rdi
>   26:	e8 bd 7b ff ff       	callq  0xffffffffffff7be8
>   2b:*	49 8b 45 00          	mov    0x0(%r13),%rax		<-- trapping instruction
>   2f:	48 89 45 c8          	mov    %rax,-0x38(%rbp)
>   33:	4c 8b 6d c8          	mov    -0x38(%rbp),%r13
>   37:	e8 cc 19 f6 ff       	callq  0xfffffffffff61a08
>   3c:	85 c0                	test   %eax,%eax
>   3e:	74 0e                	je     0x4e
> 	...
> 
> Code starting with the faulting instruction
> ===========================================
>    0:	49 8b 45 00          	mov    0x0(%r13),%rax
>    4:	48 89 45 c8          	mov    %rax,-0x38(%rbp)
>    8:	4c 8b 6d c8          	mov    -0x38(%rbp),%r13
>    c:	e8 cc 19 f6 ff       	callq  0xfffffffffff619dd
>   11:	85 c0                	test   %eax,%eax
>   13:	74 0e                	je     0x23
> 	...
> [  591.129130] RIP perf_event_exit_task (kernel/events/core.c:7492 kernel/events/core.c:7524)
> [  591.129130]  RSP <ffff8802df0ffe28>
> 
> 
> Seems similar enough.

Shiny, you got the code bits scripted ;-)

Yeah, looks about the same, one of the list elements (I checked, its not
the list head itself, or at least it wasn't when I managed to hit) was
freed while still on the list.

Attachment: pgprjdLbkokgs.pgp
Description: PGP signature