Bug#867283: Crash in glibc's mktime in low-memory situations

[Paul Eggert]

I proposed a patch here:

https://sourceware.org/bugzilla/show_bug.cgi?id=21716#c1

Please give it a try.

Bug#867283: Crash in glibc's mktime in low-memory situations

[Johannes Schultz]

(Mail was initially only sent to Carlos, sorry for that :))

> If you have the opportunity please file a match bug with upstream at
> sourceware.org/bugzilla. That way upstream is made aware of the issue.

The bug is now being tracked at:
https://sourceware.org/bugzilla/show_bug.cgi?id=21716

Bug#867283: Crash in glibc's mktime in low-memory situations

[Johannes Schultz]

Am 05.07.2017 um 21:56 schrieb Carlos O'Donell:

I agree.

If you have the opportunity please file a match bug with upstream at
sourceware.org/bugzilla. That way upstream is made aware of the issue.


Sure, I'll report the bug there.

Cheers,
Johannes

Bug#867283: Crash in glibc's mktime in low-memory situations

[Carlos O'Donell]

On Wed, Jul 5, 2017 at 1:43 PM, Johannes Schultz  wrote:
>
>> None of the internal assertions in tzfile.c have to do with low
>> memory, they have to do with logical consistency and expected
>> outcomes.
>
> Okay, so let's look at the stack trace again and where it failed.
> The failing line 779 in __tzfile_compute is:
>
> if (__tzname[0] == NULL)
> {
> assert (num_types == 1); // <-- boom
>
> So, where is __tzname[0] being set? Depending on the path taken, it can be
> either of these:
>
> line 627: __tzname[0] = NULL; // initial value
> line 646: __tzname[0] = __tzstring (_names[types[i].idx]);
> line 686: __tzname[0] = __tzstring (zone_names);
> line 756: __tzname[dst] = __tzstring (_names[idx]);
>
> Internally, __tzstring calls malloc. If malloc fails, it returns NULL.
> So it is entirely possible that this assertion will fail because of an
> out-of-memory situation.

Perfect, in which case it *should* be fixed to make this more robust in the
face of low memory.

> No of course this is bad, but so far the integrity has not been compromised.

Agreed.

> It just means that the function really should return with an error now. As
> far as I can see there are currently no facilities for returning an error in
> this particular function, but I guess it really should be able to propagate
> allocation failures to the library functions that call it, so that those can
> return an error to the user (if that is part of their API contract, that
> is).

I agree.

If you have the opportunity please file a match bug with upstream at
sourceware.org/bugzilla. That way upstream is made aware of the issue.

As far as I can see we probably need to fix all cases of __tzstring returning
NULL and do something appropraite. It only ever returns NULL on ENOMEM.

Cheers,
Carlos.

Bug#867283: Crash in glibc's mktime in low-memory situations

[Johannes Schultz]

> None of the internal assertions in tzfile.c have to do with low
> memory, they have to do with logical consistency and expected
> outcomes.

Okay, so let's look at the stack trace again and where it failed.
The failing line 779 in __tzfile_compute is:

if (__tzname[0] == NULL)
{
assert (num_types == 1); // <-- boom

So, where is __tzname[0] being set? Depending on the path taken, it can 
be either of these:


line 627: __tzname[0] = NULL; // initial value
line 646: __tzname[0] = __tzstring (_names[types[i].idx]);
line 686: __tzname[0] = __tzstring (zone_names);
line 756: __tzname[dst] = __tzstring (_names[idx]);

Internally, __tzstring calls malloc. If malloc fails, it returns NULL.
So it is entirely possible that this assertion will fail because of an 
out-of-memory situation.


No of course this is bad, but so far the integrity has not been 
compromised. It just means that the function really should return with 
an error now. As far as I can see there are currently no facilities for 
returning an error in this particular function, but I guess it really 
should be able to propagate allocation failures to the library functions 
that call it, so that those can return an error to the user (if that is 
part of their API contract, that is).


Cheers,
Johannes

Bug#867283: Crash in glibc's mktime in low-memory situations

[Carlos O'Donell]

On Wed, Jul 5, 2017 at 9:13 AM, Johannes Schultz  wrote:
> mktime is supposed to return -1 and, according to cplusplus.com, has a
> no-throw guarantee for C++ code. So even if some internal memory cannot be
> allocated, I expect mktime to return with an error value and not cause a
> SIGABRT.
> I found it difficult to reproduce this outside my afl-instrumented
> environment, but I hope the stack trace helps with verifying whether this is
> a valid bug. If you need a test case to reproduce, let me know.

If an internal assertion in tzfile.c triggers then it means that the
integrity of the library is compromised.

None of the internal assertions in tzfile.c have to do with low
memory, they have to do with logical consistency and expected
outcomes.

I've reviewed tzfile.c, tzset.c, and mktime.c and it's hard for me to
see what low-memory path could trigger the assertion. One possible
hypothesis is that your program is corrupting library data given
certain AFL input.

You will have to review your AFL input in more detail to see why it is
causing a SIGABRT.

In general the library should not SIGABRT for normal conditions like
low-memory. I agree that if you find that is *really* the case, then
we can fix this upstream.

Cheers,
Carlos.

Bug#867283: Crash in glibc's mktime in low-memory situations

[Johannes Schultz]

Package: glibc
Version: 2.19-18+deb8u10

By fuzzing my own software using American Fuzzy Lop and its provided 
libdislocator (an "abusive allocator" library), I found a code path in 
glibc that causes a SIGABRT where it probably shouldn't.


In a low-memory situation, I got the following stack trace:
#0  0x76319067 in __GI_raise (sig=sig@entry=6) at 
../nptl/sysdeps/unix/sysv/linux/raise.c:56

#1  0x7631a448 in __GI_abort () at abort.c:89
#2  0x76312266 in __assert_fail_base (fmt=0x7644af18 
"%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", 
assertion=assertion@entry=0x7644893f "num_types == 1", 
file=file@entry=0x76448936 "tzfile.c", line=line@entry=779, 
function=function@entry=0x7644fc90 <__PRETTY_FUNCTION__.6261> 
"__tzfile_compute") at assert.c:92
#3  0x76312312 in __GI___assert_fail 
(assertion=assertion@entry=0x7644893f "num_types == 1", 
file=file@entry=0x76448936 "tzfile.c", line=line@entry=779, 
function=function@entry=0x7644fc90 <__PRETTY_FUNCTION__.6261> 
"__tzfile_compute") at assert.c:101
#4  0x76391907 in __tzfile_compute (timer=1447111074, 
use_localtime=use_localtime@entry=1, 
leap_correct=leap_correct@entry=0x7fffd848, 
leap_hit=leap_hit@entry=0x7fffd844, tp=tp@entry=0x7fffd960) at 
tzfile.c:779
#5  0x76390429 in __tz_convert (timer=0x7fffd948, 
use_localtime=1, tp=0x7fffd960) at tzset.c:635
#6  0x7638eab0 in ranged_convert (convert=0x7638e940 
<__localtime_r>, t=0x7fffd948, tp=0x7fffd960) at mktime.c:310
#7  0x7638edd5 in __mktime_internal (tp=0x7fffdab0, 
convert=0x7638e940 <__localtime_r>, offset=0x7668bab8 
) at mktime.c:478
#8  0x76c02083 in OpenMPT::mpt::Date::Unix::FromUTC 
(timeUtc=...) at common/mptTime.cpp:115


mktime is supposed to return -1 and, according to cplusplus.com, has a 
no-throw guarantee for C++ code. So even if some internal memory cannot 
be allocated, I expect mktime to return with an error value and not 
cause a SIGABRT.
I found it difficult to reproduce this outside my afl-instrumented 
environment, but I hope the stack trace helps with verifying whether 
this is a valid bug. If you need a test case to reproduce, let me know.


Cheers,
Johannes