On 06/03/17 10:22, Gavin Shan wrote:
> On Fri, Mar 03, 2017 at 04:59:11PM +1100, Alexey Kardashevskiy wrote:
>> On 03/03/17 15:47, Russell Currey wrote:
>>> eeh_handle_special_event() is called when an EEH event is detected but
>>> can't be narrowed down to a specific PE. This function looks through
>>> every PE to find one in an erroneous state, then calls the regular event
>>> handler eeh_handle_normal_event() once it knows which PE has an error.
>>>
>>> However, if eeh_handle_normal_event() found that the PE cannot possibly
>>> be recovered, it will remove the PE and associated devices. This leads
>>> to a use after free in eeh_handle_special_event() as it attempts to clear
>>> the "recovering" state on the PE after eeh_handle_normal_event() returns.
>>>
>>> Thus, make sure the PE is valid when attempting to clear state in
>>> eeh_handle_special_event().
>>>
>>> Cc: <sta...@vger.kernel.org> #3.10+
>>> Reported-by: Alexey Kardashevskiy <a...@ozlabs.ru>
>>> Signed-off-by: Russell Currey <rus...@russell.cc>
>>> ---
>>> arch/powerpc/kernel/eeh_driver.c | 13 +++++++++++++
>>> 1 file changed, 13 insertions(+)
>>>
>>> diff --git a/arch/powerpc/kernel/eeh_driver.c
>>> b/arch/powerpc/kernel/eeh_driver.c
>>> index b94887165a10..492397298a2a 100644
>>> --- a/arch/powerpc/kernel/eeh_driver.c
>>> +++ b/arch/powerpc/kernel/eeh_driver.c
>>> @@ -983,6 +983,19 @@ static void eeh_handle_special_event(void)
>>> if (rc == EEH_NEXT_ERR_FROZEN_PE ||
>>> rc == EEH_NEXT_ERR_FENCED_PHB) {
>>> eeh_handle_normal_event(pe);
>>> +
>>> + /*
>>> + * eeh_handle_normal_event() can free the PE if it
>>> + * determines that the PE cannot possibly be recovered.
>>> + * Make sure the PE still exists before changing its
>>> + * state.
>>> + */
>>> + if (!pe || (pe->type & EEH_PE_INVALID)
>>> + || (pe->state & EEH_PE_REMOVED)) {
>>
>>
>> The bug is that pe becomes stale after eeh_handle_normal_event() returned
>> and dereferencing it afterwards is broken.
>>
>
> Correct, it won't cause a kernel crash as @pe is deferencing linear mapped
> area whose address is always valid.
Dereferencing pe would not crash but dereferencing any pointer from the
pnv_ioda_pe struct would (as it would random stuff or a poison).
> I think the proper fix would be to use
> eeh_handle_normal_event() to indicate the @pe has been released and don't
> access it any more.
Correct. The problem is that the callstack from my other reply is a bit too
long to make an trivial patch :)
>>
>>
>>> + pr_warn("EEH: not clearing state on bad PE\n");
>
> The message like this isn't meaningful, no need to have it. The messages that
> have prefix "EEH:" is informative messages. We definitely needn't this here.
> However, the message might be not needed in next revision.
>
>>> + continue;
>>> + }
>>> +
>>> eeh_pe_state_clear(pe, EEH_PE_RECOVERING);
>>> } else {
>>> pci_lock_rescan_remove();
>>>
>
> Thanks,
> Gavin
>
--
Alexey
