Re: perf: fuzzer KASAN slab-out-of-bounds in snb_uncore_imc_event_del

[Dmitry Vyukov]

On Tue, Nov 15, 2016 at 6:57 AM, Vince Weaver <vincent.weaver@xxxxxxxxx> wrote:
> On Mon, 14 Nov 2016, Vince Weaver wrote:
>
>> Anyway as per the suggestion at Linux Plumbers I enabled KASAN and on my
>> haswell machine it falls over in a few minutes of running the perf_fuzzer.
>>
>> [  205.740194] ==================================================================
>> [  205.748005] BUG: KASAN: slab-out-of-bounds in snb_uncore_imc_event_del+0x6c/0xa0 at addr ffff8800caa43768
>> [  205.758324] Read of size 8 by task perf_fuzzer/6618
>> [  205.763589] CPU: 0 PID: 6618 Comm: perf_fuzzer Not tainted 4.9.0-rc5 #4
>> [  205.770721] Hardware name: LENOVO 10AM000AUS/SHARKBAY, BIOS FBKT72AUS 01/26/2014
>> [  205.778689]  ffff8800c3c479b8 ffffffff816bb796 ffff88011ec00600 ffff8800caa43580
>> [  205.786759]  ffff8800c3c479e0 ffffffff812fb961 ffff8800c3c47a78 ffff8800caa43580
>> [  205.794850]  ffff8800caa43580 ffff8800c3c47a68 ffffffff812fbbd8 ffff8800c3c47a28
>> [  205.802911] Call Trace:
>> [  205.805559]  [<ffffffff816bb796>] dump_stack+0x63/0x8d
>> [  205.811135]  [<ffffffff812fb961>] kasan_object_err+0x21/0x70
>> [  205.817267]  [<ffffffff812fbbd8>] kasan_report_error+0x1d8/0x4c0
>> [  205.823752]  [<ffffffff81133275>] ? __lock_is_held+0x75/0xc0
>> [  205.829868]  [<ffffffff81025b12>] ? snb_uncore_imc_read_counter+0x42/0x50
>> [  205.837198]  [<ffffffff810222e2>] ? uncore_perf_event_update+0xe2/0x160
>> [  205.844337]  [<ffffffff812fc319>] kasan_report+0x39/0x40
>> [  205.850085]  [<ffffffff81025e3c>] ? snb_uncore_imc_event_del+0x6c/0xa0


If you pipe the report through
https://github.com/google/sanitizers/blob/master/address-sanitizer/tools/kasan_symbolize.py
it will give you line numbers and inlined frames.

> The best I can tell this maps to:
>
> static void snb_uncore_imc_event_del(struct perf_event *event, int flags)
> {
>         struct intel_uncore_box *box = uncore_event_to_box(event);
>         int i;
>
>         snb_uncore_imc_event_stop(event, PERF_EF_UPDATE);
>
>         for (i = 0; i < box->n_events; i++) {
>>>>             if (event == box->event_list[i]) {
>                         --box->n_events;
>                         break;
>                 }
>         }
> }
>
> Can this code be right?  Does it actually remove the event?
> The similar code in
>
> static void uncore_pmu_event_del(struct perf_event *event, int flags)
>
> ....
>
>         for (i = 0; i < box->n_events; i++) {
>                 if (event == box->event_list[i]) {
>                         uncore_put_event_constraint(box, event);
>
>                         for (++i; i < box->n_events; i++)
>                                 box->event_list[i - 1] = box->event_list[i];
>
>                         --box->n_events;
>                         break;
>                 }
>         }
>
>
> seems like it is more likely to be correct.
>
> Vince