Return code reports success when malloc test fails

[Rod Smith]

I ran stress-ng 0.05.12 on a 32-bit computer and got the following output:

rodsmith@tunesmith:~$ stress-ng --aggressive --verify --timeout 60 --malloc 0
stress-ng: info: [5500] dispatching hogs: 2 malloc
*** Error in `stress-ng': free(): invalid next size (fast): 0x07af3fe8 ***
stress-ng: info: [5500] successful run completed in 0.17s
rodsmith@tunesmith:~$ echo $?
0

Clearly, the test failed -- it both reported an error and completed in a (reported) 0.17 seconds rather than the requested 60 seconds. The return code, though, was 0, indicating success.

The system on which I encountered this problem reports errors in this test >50% of the time, so I can easily test any code changes.

[Colin Ian King (colin-king)]

==30870== Syscall param mincore(vec) points to unaddressable byte(s)
==30870== at 0x41EB9E9: mincore (syscall-template.S:81)
==30870== by 0x808CE10: mincore_touch_pages (mincore.c:61)
==30870== by 0x806C457: stress_malloc (stress-malloc.c:218)
==30870== by 0x80FC404: stress_run (in /home/king/repos/stress-ng/stress-ng)
==30870== by 0x804F279: main (stress-ng.c:3115)
==30870== Address 0x45fd29b is 0 bytes after a block of size 11 alloc'd
==30870== at 0x402E0D8: calloc (in /usr/lib/valgrind/vgpreload_memcheck-x86-linux.so)
==30870== by 0x808CDF9: mincore_touch_pages (mincore.c:57)
==30870== by 0x806C457: stress_malloc (stress-malloc.c:218)
==30870== by 0x80FC404: stress_run (in /home/king/repos/stress-ng/stress-ng)
==30870== by 0x804F279: main (stress-ng.c:3115)
==30870==

[Jeff Lane  (bladernr)]

I believe that's by design for now... I think I asked Colin the same question (about hte numa test) when I first started looking at it. IIRC, he offered to change that behaviour, but at the time I didn't worry about it too much, thinking we could just interpret the output and generate the right exit code from that (similar to what fwts_test does to ensure it only passes a failing exit code on high or critical fwts errors.

[Colin Ian King (colin-king)]

I believe it's a rounding error in

addr[i] = calloc(len / n, len * n);

...
...

(void)mincore_touch_pages(addr[i], len);

[Colin Ian King (colin-king)]

The bug is that the address being passed to mincore() should be page aligned.

[Colin Ian King (colin-king)]

Hi Rod, do you mind testing stress-ng on your 32 bit platform with the latest fixes from my git repo:

git clone git://kernel.ubuntu.com/cking/stress-ng
cd stress-ng
make
./stress-ng --aggressive --verify --timeout 60 --malloc 0

I think the fix should do the trick.

[Colin Ian King (colin-king)]

Fixes:

http://kernel.ubuntu.com/git/cking/stress-ng.git/commit/?id=9ca71e9681a1769b029a51f24863ebe0bf38249d
http://kernel.ubuntu.com/git/cking/stress-ng.git/commit/?id=12d734ed3e58da65170853fe0a0c8528573ae042

[Rod Smith (rodsmith)]

I've run it about a dozen times with no errors.

For clarification, should the return code be non-0 if an error like this occurs? The man page seems to say it should:

EXIT STATUS
         Status Description
           0 Success.
           1 Error; incorrect user options or a fatal resource issue
                    (for example, out of memory).
           2 One or more stressors failed.

[Colin Ian King (colin-king)]

It should be a 1, for some reason a SIGSEGV is not being picked up by the parent. I'll see if I can fix that too.

[Colin Ian King (colin-king)]

I've tweaked the code now to pick up crashes from the child stressors and report them as errors, for example, I deliberately made the io stressor segfault - and this is what we get:

~/repos/stress-ng$ ./stress-ng --io 1 -v
stress-ng: debug: [13494] 4 processors online, 4 processors configured
stress-ng: info: [13494] defaulting to a 86400 second run per stressor
stress-ng: info: [13494] dispatching hogs: 1 iosync
stress-ng: info: [13494] cache allocate: default cache size: 3072K
stress-ng: debug: [13494] starting stressors
stress-ng: debug: [13494] 1 stressor spawned
stress-ng: debug: [13495] stress-ng-iosync: started [13495] (instance 0)
stress-ng: debug: [13494] process 13495 (stress-ng-iosync) terminated on signal: 11 (Segmentation fault)
stress-ng: debug: [13494] process [13495] terminated
stress-ng: info: [13494] unsuccessful run completed in 0.00s
echo $?
2

(2 because the stressor failed).

[Colin Ian King (colin-king)]

http://kernel.ubuntu.com/git/cking/stress-ng.git/commit/?id=738458008e0516703b825705ea6a1e4ee4e8cbeb