fork then wait4 bug when being traced?

[Chris Wedgwood]

It appears that when a process is being traced, there is a race or
sorts that can occur which causes wait4 to fail erroneously.
Basically, you can fork and then wait4 on the child yet wait4 returns
ECHILD -- something you would not normally expected.

It seems, I can only replicate this when tracing the process and only
some times (50% or so) as it's appears to be fairly timing critical.
Linux 2.4.0-test3-pre2+some-bits UP.

Here is a trace that shows it happening:

9839  fork()                            = 9840
9839  close(4)                          = 0
9839  fstat(5, {st_mode=S_IFIFO|0600, st_size=0, ...}) = 0
9839  old_mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x40015000
9839  _llseek(5, 0, 0xbfffea24, SEEK_CUR) = -1 ESPIPE (Illegal seek)
9839  write(5, "To: cw\n\nDecoded Message:\n\n", 26) = 26
9839  close(5)                          = 0
9839  wait4(9840, 0xbfffec68, WNOHANG, NULL) = -1 ECHILD (No child processes)

  --cw