Bug 168775

From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.10) Gecko/20050716 Firefox/1.0.6

Description of problem:
On certain ocasions the wait() and waitpid() calls erroneously generate an ECHILD (No child processes)
error when in fact a child process was successfuly (v)fork()-ed and is running.




Version-Release number of selected component (if applicable):
kernel-smp-2.6.9-11.EL

How reproducible:
Always

Steps to Reproduce:
1. compile the atttached program:
        gcc -W -Wall -g3 -O0 -o test-bin forktest.c -lpthread

2. create a script called 'test-thread' with the following contents:
        <-------------------- CUT HERE ----------------------->
        #/bin/bash
        for i in `seq 1000`; do ./test-bin 1 ; done
        <--------------------- END CUT ----------------------->

3. create a second script called 'test-main' with the following contents:
        <-------------------- CUT HERE ----------------------->
        #!/bin/bash

        for i in `seq 20`; do
            ./test-thread &
        done

        read ABCD

        killall -9 test-thread
        killall -9 test-bin
        <--------------------- END CUT ----------------------->

4. execute the following command:
        ./test-main | grep ">"


Actual Results:  .......................................................................
>>>>>>>>>>>>>>>>>>>>>>>>>> run_level1() waitpid surprise: No child processes
>>>>>>>>>>>>>>>>>>>>>>>>>> kill( [PID_OF_CHILD], 0 ): 0 Success
...................... (repeated) .....................................


Expected Results:  The output of 'test-main' should not contain any lines with '>' characters...

Additional info:

1. the call to kill() is always successful, thus confirming that the process exists and is running

2. the problem seems to only occur under high load. The speed at which processes are spawned also seems to be an important factor: the slower the rate the harder it is to reproduce the problem

The previous kernel version (kernel-smp-2.6.9-5.EL) does not seem to be affected by this issue.