Why exactly ptrace(PT_TRACE_ME, ...) call after the fork in the debugged app
is not enough for you?

Jaromir

Emmanuel Dreyfus wrote:
> Hi all
> 
> I frequently have the need to stop a process on fork or exec, for debugging
> purposes. It can be done by tricking the process into an infinite loop, or by
> adding somes hacks in the kernel to send a SIGSTOP at the right time.
> 
> Some support in the kernel to do this would be nice. My idea is to use a sysctl
> to set a value (named p_stopchild) stored in struct proc:
> 
> $ sysctl -w proc.$$.stopchild=2                               
> proc.291.stopchild: 0 -> 2
> $ ls
> Here, the shell gets blocked after the fork. It's possible to attach it, or to
> send a SIGCONT. Once the process can continue after the fork, p_stopchild is
> decreased. Because it does not reach zero, we will stop again after the shell
> calls exec to become ls.
> 
> Find below a patch that does the job. Is it good enough to be committed?
> 
> Index: sys/proc.h
> ===================================================================
> RCS file: /cvsroot/syssrc/sys/sys/proc.h,v
> retrieving revision 1.147
> diff -r1.147 proc.h
> 189a190
> >       int             p_stopchild;    /* Stop childs on fork and exec */
> Index: sys/sysctl.h
> ===================================================================
> RCS file: /cvsroot/syssrc/sys/sys/sysctl.h,v
> retrieving revision 1.78
> diff -r1.78 sysctl.h
> 593c593,594
> < #define       PROC_PID_MAXID          3
> ---
> > #define       PROC_PID_STOPCHILD      3
> > #define       PROC_PID_MAXID          4
> 598a600
> >       { "stopchild", CTLTYPE_INT }, \
> Index: kern/kern_sysctl.c
> ===================================================================
> RCS file: /cvsroot/syssrc/sys/kern/kern_sysctl.c,v
> retrieving revision 1.113
> diff -r1.113 kern_sysctl.c
> 746a747,753
> >       if (name[1] == PROC_PID_STOPCHILD) {
> >               if (namelen != 2)
> >                       return EINVAL;
> >               error = sysctl_int(oldp, oldlenp, newp, 
> >                   newlen, &ptmp->p_stopchild);
> >               return error;
> >       }
> Index: kern/kern_exec.c
> ===================================================================
> RCS file: /cvsroot/syssrc/sys/kern/kern_exec.c,v
> retrieving revision 1.160
> diff -r1.160 kern_exec.c
> 740a741,753
> > 
> >       if (p->p_stopchild) {
> >               int s;
> > 
> >               sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
> >               SCHED_LOCK(s);
> >               p->p_stopchild--;
> >               p->p_stat = SSTOP;
> >               mi_switch(p, NULL);
> >               SCHED_ASSERT_UNLOCKED();
> >               splx(s);
> >       }
> > 
> Index: kern/kern_fork.c
> ===================================================================
> RCS file: /cvsroot/syssrc/sys/kern/kern_fork.c,v
> retrieving revision 1.96
> diff -r1.96 kern_fork.c
> 474c474,475
> <        * Make child runnable, set start time, and add to run queue.
> ---
> >        * Make child runnable, set start time, and add to run queue
> >        * except if the parent requested the child to start in SSTOP state.
> 479,480c480,487
> <       p2->p_stat = SRUN;
> <       setrunqueue(p2);
> ---
> >       if (p1->p_stopchild) {
> >               p2->p_stat = SSTOP;
> >               p2->p_stopchild = p1->p_stopchild - 1;
> >               p1->p_stopchild = 0;
> >       } else {
> >               p2->p_stat = SRUN;
> >               setrunqueue(p2);
> >       }
> 
> 
> -- 
> Emmanuel Dreyfus.
> "Le 80x86 n'est pas si complexe - il n'a simplement pas de sens"
> (Mike Johnson, responsable de la conception x86 chez AMD) 
> manu@netbsd.org
> 


-- 
Jaromir Dolecek <jdolecek@NetBSD.org>            http://www.NetBSD.org/
-=- We should be mindful of the potential goal, but as the tantric    -=-
-=- Buddhist masters say, ``You may notice during meditation that you -=-
-=- sometimes levitate or glow.   Do not let this distract you.''     -=-