child being strace'd does not hang at raise(SIGTSTP)

[Denys Vlasenko]

On 01/18/2012 01:33 PM, Vangelis Koukis wrote:
> Hello,
>
> I've been running into a problem where a child running under strace does
> not stop when it reaches raise(SIGSTOP). This was unexpected at first...
> The attached program [strace-test.c] demoes this behavior,
> and runs differently when being traced.
>
> $ ./strace-test
> Parent: sleeping for a while...
> Parent: SUCCESS. Child killed, sig = 9
>
> $ strace -f ./strace-test 2>/dev/null
> Parent: sleeping for a while...
> Child: FAIL.
> Parent: FAIL. Child exited, status = 1
>
> After quite some experimentation, I came to this, in the manual page:
>
>         A traced process ignores SIGSTOP except on SVR4 platforms.
>
> which I understand is a limitation coming from the way ptrace() works
> under Linux?
>
> So, I tried with SIGTSTP instead. Again, it seems a child process being
> traced will not hang when doing a raise(SIGTSTP) [the signal to use can
> be #defined in strace-test.c]
>
> I've tried with
>
> Linux hostname 2.6.26-2-amd64 #1 SMP Tue Jan 25 05:59:43 UTC 2011 x86_64 GNU/Linux
> and Debian-provided "strace -- version 4.5.17",
>
> as well as
>
> Linux hostname 3.0.4 #1 SMP Tue Aug 30 12:55:46 EEST 2011 i686 GNU/Linux
> with latest strace, compiled from git master, commit
> e73a89d99921f7b9dc3f1350a4eb97c7fdc6032a.
>
> To summarize, is there anything I can do to have straced processes stop
> on reception of a signal? If not, is it a matter of the way ptrace()
> works? Then perhaps the manpage should not mention SIGSTOP only.

This is a known bug in kernel ptrace API, and the fix seems to be ready
and is being pushed into mainline. Concurrent changes for strace are
ready too.


-- 
vda