2003-01-15 16:26:12

by Daniel Barlow

Subject: 2.4.x, can't ptrace a task created with clone() ?

Tested with 4.2.20 and 2.4.21-pre3, after PTRACE_ATTACHing to a
process I created with clone(), wait() returns ECHILD . strace
of the parent shows

clone(child_stack=0x80599a8, flags=CLONE_VM) = 1292
rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
rt_sigaction(SIGCHLD, NULL, {SIG_DFL}, 8) = 0
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
nanosleep({5, 0}, {5, 0}) = 0
ptrace(PTRACE_ATTACH, 1292, 0, 0) = 0
--- SIGCHLD (Child exited) ---
wait4(-1, 0xbffffbf0, 0, NULL) = -1 ECHILD (No child processes)
write(2, "waitpid: No child processes\n", 28) = 28

If I replace the clone in my test program with a fork, everything
works as I expected. Is this a bug, or are my expectations at fault?
if the latter, how _do_ I attach to one of my cloned children?

Here's the test program I've been using:

#include <sched.h>
#include <sys/ptrace.h>
#include <linux/user.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/wait.h>


void child(void *d)
{
int foo;
fprintf(stderr,"child stack is somewhere around 0x%x\n",&foo);
while(1){
fprintf(stderr, "still here\n");
sleep(3);
}
}

main()
{
pid_t kid;
int waitr;
struct user_regs_struct regs;

kid=clone(child,SIGSTKSZ*8+malloc(SIGSTKSZ*8), CLONE_VM ,0);
/* kid=fork(); */
if(kid<0) perror("clone");
if(kid==0) child(0);
sleep(5);
if(kid>0) {
if(ptrace(PTRACE_ATTACH,kid,0,0)) perror("PTRACE_ATTACH");
if(wait(&waitr)==-1) {
perror("waitpid");
exit(1);
}
if(ptrace(PTRACE_GETREGS,kid,0,&regs)) perror("PTRACE_GETREGS");
fprintf(stderr,"child stack pointer is 0x%x\n",regs.esp);
if(ptrace(PTRACE_DETACH,kid,0,0)) perror("PTRACE_DETACH");
kill(kid,15);
}
}

libc 2.3.1, in case that makes a difference. Let me know if there's
anything else you wanted to ask


-dan

--

http://www.cliki.net/ - Link farm for free CL-on-Unix resources

2003-01-15 20:19:22

by Daniel Jacobowitz

Subject: Re: 2.4.x, can't ptrace a task created with clone() ?

On Wed, Jan 15, 2003 at 04:34:58PM +0000, Daniel Barlow wrote:
>
> Tested with 4.2.20 and 2.4.21-pre3, after PTRACE_ATTACHing to a
> process I created with clone(), wait() returns ECHILD . strace
> of the parent shows
>
> clone(child_stack=0x80599a8, flags=CLONE_VM) = 1292
> rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
> rt_sigaction(SIGCHLD, NULL, {SIG_DFL}, 8) = 0
> rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
> nanosleep({5, 0}, {5, 0}) = 0
> ptrace(PTRACE_ATTACH, 1292, 0, 0) = 0
> --- SIGCHLD (Child exited) ---
> wait4(-1, 0xbffffbf0, 0, NULL) = -1 ECHILD (No child processes)
> write(2, "waitpid: No child processes\n", 28) = 28
>
> If I replace the clone in my test program with a fork, everything
> works as I expected. Is this a bug, or are my expectations at fault?
> if the latter, how _do_ I attach to one of my cloned children?

You did attach. But if you want to wait for a cloned child, you need
to use __WCLONE or __WALL.

--
Daniel Jacobowitz
MontaVista Software Debian GNU/Linux Developer