[LTP] [PATCH v1] Fix: dirtyc0w_shmem child process exit with error due to uninitialized lib_pid

[Li Wang]

On Mon, May 12, 2025 at 2:34 PM Petr Vorel <pvorel@suse.cz> wrote:

> > On Sat, May 10, 2025 at 4:54 PM Wei Gao via ltp <ltp@lists.linux.it>
> wrote:
>
> > > The dirtyc0w_shmem test spawns a child process using execlp. This
> child process then calls tst_brk(), which exits early
> > > with a non-zero status because execlp does not inherit the parent's
> lib_pid variable. Consequently, the parent process
> > > incorrectly reports an "Invalid child exit value".
>
> > > This commit addresses this by ensuring the child process has access to
> the necessary lib_pid and main_pid by passing
> > > them through a shared results structure. This prevents the premature
> exit in the child and the subsequent error report
> > > in the parent.
>
> > > Related commit:
> > > commit a1f82704c28d9e027ca899f5ca2841e7fe49de72
> > > lib/tst_test.c: Fix tst_brk() handling
>
> > > Detail failure log:
> > > tst_tmpdir.c:317: TINFO: Using /tmp/LTP_dirSOGVBC as tmpdir (btrfs
> filesystem)
> > > tst_test.c:1938: TINFO: LTP version: 20250507.4a0e3a8fa
> > > tst_test.c:1942: TINFO: Tested kernel: 6.4.0-150700.51-default #1 SMP
> Wed Apr 30 21:35:43 UTC 2025 (6930611) s390x
> > > tst_kconfig.c:88: TINFO: Parsing kernel config '/proc/config.gz'
> > > tst_kconfig.c:678: TINFO: CONFIG_FAULT_INJECTION kernel option
> detected which might slow the execution
> > > tst_test.c:1760: TINFO: Overall timeout per run is 0h 04m 00s
> > > dirtyc0w_shmem.c:54: TINFO: Mounting tmp_dirtyc0w_shmem to
> /tmp/LTP_dirSOGVBC/tmp_dirtyc0w_shmem fstyp=tmpfs flags=0
> > > dirtyc0w_shmem_child.c:160: TCONF: System does not have userfaultfd
> minor fault support for shmem <<<<<<<<<< 1
> > > tst_test.c:481: TBROK: Invalid child (8163) exit value 32 <<<<<<<< 2
> > > dirtyc0w_shmem.c:102: TINFO: Umounting
> /tmp/LTP_dirSOGVBC/tmp_dirtyc0w_shmem
>
> > > tmp_dirtyc0w_shmem.c call execlp to create new process run
> dirtyc0w_shmem_child bin.
>
> > > SAFE_EXECLP("dirtyc0w_shmem_child", "dirtyc0w_shmem_child", NULL)
>
> > > Within dirtyc0w_shmem_child.c trigger
>
> > > tst_brk(TCONF, "System does not have userfaultfd minor fault support
> for shmem")
>
> > > Since execlp does not inherit the parent process's variables lib_pid,
> so it will return TCONF(32) directly.
>
> > > void tst_vbrk_(const char *file, const int lineno, int ttype, const
> char *fmt,
> > >                va_list va)
> > > {
> > > ...
> > >         if (!lib_pid)
> > >                 exit(TTYPE_RESULT(ttype));   <<<<<
> > > ...
> > > }
>
> > > So finally captured by check_child_status report an error.
>
> > > static void check_child_status(pid_t pid, int status)
> > > {
> > > ...
> > >         if (WEXITSTATUS(status))
> > >                 tst_brk(TBROK, "Invalid child (%i) exit value %i",
> pid, WEXITSTATUS(status));  <<<<
> > > }
>
> > > Signed-off-by: Wei Gao <wegao@suse.com>
> > > ---
> > >  lib/tst_test.c | 25 ++++++++++++++-----------
> > >  1 file changed, 14 insertions(+), 11 deletions(-)
>
> > > diff --git a/lib/tst_test.c b/lib/tst_test.c
> > > index 2bb4519dd..b666715b9 100644
> > > --- a/lib/tst_test.c
> > > +++ b/lib/tst_test.c
> > > @@ -59,7 +59,7 @@ static const char *tid;
> > >  static int iterations = 1;
> > >  static float duration = -1;
> > >  static float timeout_mul = -1;
> > > -static pid_t main_pid, lib_pid;
> > > +static pid_t lib_pid;
>
> Wei, you probably noticed tests fail (thanks to Andrea's CI improvements).
> It broke at least tst_needs_cmds02.c I guess due missing main_pid:
>
> $ gdb ./tst_needs_cmds02
> ...
> (gdb) run
> ...
> tst_cmd.c:268: TCONF: Couldn't find 'mkfs.ext45' in $PATH


> Program received signal SIGSEGV, Segmentation fault.
> 0x000055555555bea4 in tst_vbrk_ (file=<optimized out>, lineno=<optimized
> out>, ttype=<optimized out>, fmt=<optimized out>, va=<optimized out>) at
> tst_test.c:397
> 397             if (!results->lib_pid)
> (gdb) bt
> #0  0x000055555555bea4 in tst_vbrk_ (file=<optimized out>,
> lineno=<optimized out>, ttype=<optimized out>, fmt=<optimized out>,
> va=<optimized out>) at tst_test.c:397
> #1  0x000055555555b5be in tst_brk_ (file=file@entry=0x55555557de1c
> "tst_cmd.c", lineno=lineno@entry=268, ttype=ttype@entry=32, fmt=fmt@entry=0x55555557deed
> "Couldn't find '%s' in $PATH")
>     at tst_test.c:460
> #2  0x000055555556bc57 in tst_check_cmd (cmd=0x55555557c643 "mkfs.ext45 >=
> 1.43.0", brk_nosupp=brk_nosupp@entry=1) at tst_cmd.c:268
> #3  0x000055555555e641 in do_setup (argc=1, argv=0x7fffffffcaa8) at
> tst_test.c:1363
> #4  tst_run_tcases (argc=1, argv=0x7fffffffcaa8, self=self@entry=0x55555558a720
> <test>) at tst_test.c:1935
> #5  0x000055555555b044 in main (argc=<optimized out>, argv=<optimized
> out>) at ../../include/tst_test.h:729
>

Oops, I neglected that comment before merging, and it caused CI to fail.
No time to fix it today though (too late). I'll try to look at the details
tomorrow.

But hopefully it'll be fixed by the time I wake up tomorrow :).

Sorry for the careless push.


-- 
Regards,
Li Wang