Hello,
shouldn't ptrace(PTRACE_DETACH, pid, NULL, SIGSTOP) stop the child?

The Valgrind [1] sequence call to stop a child, change the registers and
attach a debugger to the child is as following (or take a look at
[2] function start_debugger(), or the attached test program):

if ((pid = fork()) == 0) {
	ptrace(PTRACE_TRACEME, ...);
	kill(getpid(), SIGSTOP);
	// should never be reached
} else {
	waitpid(pid, ...);
	ptrace(PTRACE_SETREGS, ...)
	ptrace(PTRACE_DETACH, pid, NULL, SIGSTOP)
	// attach gdb to child
}

But the child seems to resume execution after the PTRACE_DETACH call (in
contradiction to what the ptrace man page [3] tells).

Example programm gdb_attach.c attached to the e-mail.

gcc -Wall -g -o g db_attach gdb_attach.c
./gdb_attach 
parent
child
parent: cmdl: gdb gdb_attach 1509
child: should never be reached
<gdb output snipped>
0x400d4411 in nanosleep () from /lib/libc.so.6
(gdb) bt
#0  0x400d4411 in nanosleep () from /lib/libc.so.6
#1  0x400d43a8 in sleep () from /lib/libc.so.6
#2  0x08048704 in start_debugger () at gdb_attach.c:20
#3  0x08048889 in main (argc=1, argv=0xbffff444) at gdb_attach.c:46
#4  0x400499ed in __libc_start_main () from /lib/libc.so.6

One 'workaround' is to send an additional SIGSTOP from the parent:
gcc -Wall -g -DADDITIONAL_SIGSTOP -o gdb_attach gdb_attach.c
 ./gdb_attach 
parent
child
parent: cmdl: gdb gdb_attach 1519
<gdb output>
0x4005aab1 in kill () from /lib/libc.so.6
(gdb) bt
#0  0x4005aab1 in kill () from /lib/libc.so.6
#1  0x080486ec in start_debugger () at gdb_attach.c:17
#2  0x080488c5 in main (argc=1, argv=0xbffff444) at gdb_attach.c:46
#3  0x400499ed in __libc_start_main () from /lib/libc.so.6
(gdb) 

It this a PTRACE_DETACH bug (kernel used 2.4.25)?
If not, is the additional SIGSTOP a valid 'workaround'?
If not, is there a better workaround/solution?

Thanks in advance for your help,
Peter Seiderer


[1] http://valgrind.kde.org/

[2] http://webcvs.kde.org/cgi-bin/cvsweb.cgi/~checkout~/valgrind/coregrind/vg_main.c?rev=1.147&content-type=text/plain

[3] http://techpubs.sgi.com/library/tpl/cgi-bin/getdoc.cgi?coll=linux&db=man&fname=/usr/share/catman/man2/ptrace.2.html
	PTRACE_DETACH: ... Restarts  the stopped child as for PTRACE_CONT ...
	PTRACE_CONT: ... Restarts  the  stopped  child  process.  If data is
	non-zero and not SIGSTOP, it is  interpreted  as  a signal  to be
	delivered to the child ...

-- 
------------------------------------------------------------------------
Peter Seiderer                     E-Mail:  seiderer123@ciselant.de