Should not use tst_brk after calling cleanup_ipc()

[fallen]

in the process of exiting, after having printed the Summary, do_cleanup() is called.
Which then calls cleanup_ipc(), which unmaps results. See
https://github.com/linux-test-project/ltp/blob/master/lib/tst_test.c#L924
https://github.com/linux-test-project/ltp/blob/master/lib/tst_test.c#L138

But then, afterward, tst_wallclock_restore() is called which itself can call tst_brk upon some syscall failure:

ltp/lib/tst_wallclock.c

Line 59 in 0410a51

tst_brk(TBROK | TERRNO, "tst_clock_settime() realtime failed");

This is wrong since tst_brk will call do_exit again and will try to de-reference results ( https://github.com/linux-test-project/ltp/blob/master/lib/tst_test.c#L593 ) and this leads to a segfault.

I experienced this segfault while running tests and for some reason the tst_clock_settime call failed.

output example:

Summary:
passed 6
failed 0
skipped 0
warnings 0
tst_wallclock.c:59: BROK: tst_clock_settime() realtime failed: EINVAL

Summary:
Segmentation fault

Another solution would be to make results NULL upon unmapping it.

[metan-ucw]

I guess that we can move the tst_wallclock_restore() before the call to the cleanup_ipc(), that should fix the problem as well.

[fallen]

Yes I agree.

[fallen]

But maybe adding some results = NULL upon unmapping would also be a safe thing to do anyway.

[metan-ucw]

Sure, zeroing results after unmapping sounds good as well.

Will you send a patch or should I do it?

[fallen]

I'll send a patch later. Thanks for your answers.
You prefer the patches to be sent to the mailing list, right?

[metan-ucw]

Yes, mailing list is preferred.

[fallen]

Patch sent, email waiting for moderation.

[fallen]

Patch arrived on the mailinglist: https://lists.linux.it/pipermail/ltp/2019-June/012292.html

[fallen]

re-sent with corrections and missing Signed-off-by line.

[fallen]

Thanks!