[LTP] [PATCH] Crash02: Deal with SIGSTOP and SIGSEGV v3

[CAI Q. <ca...@cc...>]

Hi,

Version 3 also fixed compilation failures on IA-64, since there is no
SYS_fork either. It use clone2() instead.

Version 2 fixed compilation failures on IA-64, because there is no
SYS_vfork there. Instead, it uses clone() syscall for vfork(),

clone(child_stack=0, flags=CLONE_VM|CLONE_VFORK|SIGCHLD)

We have seen crash02 test can not finish due to the child process got
SIGSTOP or SIGSEGV when calling random syscalls like this,

# ps aux
...
168      20121  0.0  0.0   3956   336 ?        Ds   15:10   0:00 ./crash02 -e -v 100
168      20122  0.0  0.0   3956   336 ?        T    15:10   0:00 ./crash02 -e -v 100


crash02       X ffff81007fb127a0     0 21272  21271         21273       (L-TLB)
 ffff81004c185d58 0000000000000046 ffff81007b494e08 0000000000000002
 0000000000000002 0000000000000002 ffff81007c5dc860 ffff81007fb127a0
 00000d8ebb800862 000000000000c315 ffff81007c5dca48 0000000300000001
Call Trace:
 [<ffffffff800235a7>] filp_close+0x5c/0x64
 [<ffffffff8001595a>] do_exit+0x913/0x91f
 [<ffffffff80048c18>] cpuset_exit+0x0/0x6c
 [<ffffffff8002ad05>] get_signal_to_deliver+0x42c/0x45a
 [<ffffffff8005a837>] do_notify_resume+0x9c/0x7af
 [<ffffffff80096e2b>] specific_send_sig_info+0x44/0xac
 [<ffffffff8009710e>] force_sig_info+0xae/0xb9
 [<ffffffff80066eed>] do_page_fault+0x81e/0x830
 [<ffffffff8005d6dc>] retint_signal+0x3d/0x79

crash02       X ffff810002376400     0 21273  21271               21272 (L-TLB)
 ffff81004d201f18 0000000000000046 ffff81007b494e08 0000000000000046
 0000000000000046 0000000000000002 ffff810078d3a040 ffffffff802eeae0
 00000d8ebb803430 000000000001a60c ffff810078d3a228 0000000000000000
Call Trace:
 [<ffffffff8001595a>] do_exit+0x913/0x91f
 [<ffffffff80048c18>] cpuset_exit+0x0/0x6c
 [<ffffffff8005d28d>] tracesys+0xd5/0xe0

So we use WUNTRACED option. Also, we found out that if it picks up
vfork(), it will generate SIGSEGV. I am not sure what else can generate
the same, so only blacklist vfork() at the moment.

Signed-off-by: CAI Qian <ca...@cc...>

--- testcases/misc/crash/crash02.c.orig	2009-05-19 19:47:24.000000000 +0800
+++ testcases/misc/crash/crash02.c	2009-05-22 15:57:56.000000000 +0800
@@ -51,6 +51,8 @@
 */
 
 
+#define _GNU_SOURCE
+#include <sys/syscall.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -241,7 +243,7 @@
       my_signal (SIGALRM, monitor_fcn);
       alarm (ntries*MAX_TRY_TIME);
 
-	  pid = wait (&status);
+	  pid = waitpid (-1, &status, WUNTRACED);
 	  if (pid <= 0)
 	  {
 		  perror ("wait");
@@ -298,6 +300,7 @@
 void again_handler (int sig);
 void try_one_crash (int try_num);
 void set_up_signals ();
+int in_blacklist (int sysno);
 
 /* badboy "entry" point */
 
@@ -453,7 +456,10 @@
 {
   long int sysno, arg1, arg2, arg3, arg4, arg5, arg6, arg7;
 
-  sysno = rand()%sysno_max;
+  do {
+    sysno = rand()%sysno_max;
+  } while (in_blacklist(sysno));
+
   arg1 = rand_long();
   arg2 = rand_long();
   arg3 = rand_long();
@@ -472,4 +478,32 @@
   }
 }
 
+/* The following syscalls create new processes which may cause the test
+   unable to finish. */
+int
+in_blacklist (int sysno)
+{
+  int i;
+  const list[] =
+    {
+#ifdef __ia64__
+      SYS_clone2,
+#else
+      /* No SYS_fork(vfork) on IA-64. Instead, it uses,
+         clone(child_stack=0, flags=CLONE_VM|CLONE_VFORK|SIGCHLD)
+         clone2() */
+      SYS_vfork,
+      SYS_fork,
+#endif /* __ia64__ */
+      SYS_clone,
+      -1
+    };
+
+  for (i = 0; list[i] != -1; i++)
+    {
+      if (sysno == list[i])
+        return 1;
+    }
 
+  return 0;
+}