Bug 209106

LTC Owner is: skannery.com
LTC Originator is: yuweiy.com


Problem description:
Ltrace hangs when trace child process with -f option

Hardware Environment
    Machine type (p650, x235, SF2, etc.):SF4+
    Cpu type (Power4, Power5, IA-64, etc.):Power5+

Please provide access information for the machine if it is available.
ltc3.cn.ibm.com(root/.pasroot)

Is this reproducible?
    If so, how long does it (did it) take to reproduce it?
    Describe the steps:

[root@ltc3 home]# cat trace-fork.c
/* Ltrace Test : trace-fork.c
   Objectives  : Verify that ltrace can trace to child process after
   fork called.

   This file was written by Yao Qi <qiyao.com>.  */

#include <stdio.h>
#include <sys/types.h>
#include <unistd.h>
void
child ()
{
  printf("Fork Child\n");
  sleep(1);
}

int
main ()
{
  pid_t pid;
  pid = fork ();

  if (pid == -1)
    printf("fork failed!\n");
  else if (pid == 0)
    child();
  else
    {
      printf("My child pid is %d\n",pid);
      wait();
    }
  return 0;
}
[root@ltc3 home]# gcc -o trace-fork trace-fork.c
[root@ltc3 home]# ltrace -f ./trace-fork
unexpected breakpoint at 0xffd63e0
_start(1, 0xfe56fa44, 0xfe56fa4c, 0xfe56fac0, 0xffce9a0 <unfinished ...>
__libc_start_main(1, 0xfe56fa44, 0xfe56fa4c, 0xfe56fac0, 0xffce9a0 <unfinished ...>
fork( <unfinished ...>
fork( <unfinished ...>
[pid 12879] --- SIGSTOP (Stopped (signal)) ---
[pid 12878] <... fork resumed> )                                               
                                          = 12879
[pid 12878] printf("My child pid is %d\n", 12879 <unfinished ...>
[pid 12878] fork(My child pid is 12879
)                                                                              
                         = 22
[pid 12878] wait(0x16              -----------> Hang

 
Additional information:
[root@ltc3 home]# rpm -q ltrace
ltrace-0.5-1.0.45svn.5

Analysis: Applied manually the patch given thru bug 24646 for tracing fork. Hang
problem is getting addressed with the patch. But the library calls of the forked
child is not getting traced. This is because the breakpoints for the library
symbols of the child process is not getting inserted properly. Analysing further..

Wei,
  I have prepared a patch for this problem. While patch for tracing fork
provided in bug 24646 addressed the hang issue, I had to make few more
modifications in addition to that patch to address the issue on tracing the
child process.

This patch is prepared over ltrace-0.5-1.0.45svn.5 level. Please verify this
patch and let me know the results.

Still there are some issues like "puts" getting printed instead of "printf",
incorrect data getting printed as parameters for "puts", unexpected breakpoints,
etc. These issues need to be looked at separately. 
Thanks, Supriya

<Test patch-1 part of Consolidated patch>
Testpatch for addressing hang while tracing child


Supriya,
I had tested the patch from above , there is no hang for ltrace while tracing
child, but it cant trace a process by its PID any more.

[root@blade9 ~]# /usr/src/redhat/BUILD/ltrace-0.5/ltrace -f ./trace-fork
unexpected breakpoint at 0xffd63e0
__libc_start_main(1, 0xf8f6faa4, 0xf8f6faac, 0xf8f6fb0c, 0xffce9a0 <unfinished ...>
fork( <unfinished ...>
fork()                                                                         
                 = 13262
[pid 13261] printf("My child pid is %d\n", 13262 <unfinished ...>
[pid 13261] fork(My child pid is 13262
)                                                                               = 22
[pid 13261] wait(0x16 <unfinished ...>
[pid 13262] puts("Fork Child"Fork Child
)                                                                   = 11
[pid 13262] sleep(1 <unfinished ...>
[pid 13262]
puts("\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377"...)
= 0
[pid 13262] +++ exited (status 0) +++
--- SIGCHLD (Child exited) ---
<... wait resumed> )                                                           
                 = -1
+++ exited (status 0) +++


[root@blade9 ~]# ./attach-process &
[1] 13270
[root@blade9 ~]# /usr/src/redhat/BUILD/ltrace-0.5/ltrace -p 13270
[root@blade9 ~]#
[1]+  Done                    ./attach-process
[root@blade9 ~]#


Wei,
  Can you pls apply this patch (trace_pid_patch) in addition to the previous one
and verify whether tracing a pid is working.
Thanks, Supriya

<Test patch-2 part of Consolidated patch>
Test patch for tracing pid


Tested ltrace with the patch-1 and patch-2, and it works fine for tracing
process by pid, and the same result for tracing child as comment earlier.

[root@blade9 ~]# ./attach-process &
[1] 15622
[root@blade9 ~]# /usr/src/redhat/BUILD/ltrace-0.5/ltrace -p 15622
sleep(3)                                                                       
                 = 0
+++ exited (status 0) +++
[1]+  Done                    ./attach-process
[root@blade9 ~]#