On Linux MMU architectures this traces vfork children without changing
their behaviour to fork (which is quite a big change even though all
applications are supposed to not care). On Linux no-MMU architectures
it allows child processes to be traced. Before this, all attempts to
trace non-thread children got stuck at vfork. If we're really lucky
and the kernel is sensible, this might let us trace CLONE_NEWPID
grandchildren too.
* util.c [LINUX] (setbpt): When forking, try to enable the
PTRACE_EVENT_{FORK,VFORK,CLONE} events using PTRACE_SETOPTIONS. If
successful, instead of converting fork and vfork to clone equivalents,
and vfork to fork, just leave them as they are. Always return with
tcp->u_arg[arg0_index] set to appropriate clone flags in all cases,
even if clone wasn't actually used.
Signed-off-by: Jamie Lokier <ja...@shareable.org>
---
util.c | 53 +++++++++++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 53 insertions(+), 0 deletions(-)
diff --git a/util.c b/util.c
index 0c4a882..beb397e 100644
--- a/util.c
+++ b/util.c
@@ -1517,6 +1517,59 @@ setbpt(struct tcb *tcp)
return -1;
}
+# ifdef PTRACE_EVENT_FORK
+ /*
+ * If are able to set these options, we don't need a
+ * breakpoint. FIXME: Kernel bugs mean we might need to check
+ * the kernel version number. See Red Hat Bugzilla bug 246330
+ * - "ptrace looses track of a forked child". It says it was
+ * broken for a certain range of versions. See also Red Had
+ * Bugzilla bug 162774 - "strace ignores int3 SIGTRAP" It
+ * includes a patch, and mentions kernels, and a comment in
+ * the patch about when PTRACE_O_TRACESYSGOOD doesn't work
+ * completely, which might be related.
+ */
+ if (ptrace (PTRACE_SETOPTIONS, tcp->pid, (void *) 1,
+ (long) (PTRACE_O_TRACEFORK
+ | PTRACE_O_TRACEVFORK
+ | PTRACE_O_TRACECLONE)) == 0) {
+ /*
+ * Save the clone flags, or fake them, in case no
+ * event is received and internal_fork_child() needs to be
+ * called with these flags.
+ */
+ switch (real_scno) {
+# ifdef SYS_vfork
+ case SYS_vfork:
+ tcp->u_arg[arg0_index] = CLONE_VM|CLONE_VFORK|SIGCHLD;
+ return 0;
+# endif
+# ifdef SYS_fork
+ case SYS_fork:
+ tcp->u_arg[arg0_index] = SIGCHLD;
+ return 0;
+# endif
+
+ default: /* clone, clone2 */
+ /*
+ * No breakpoint, but we still need to save
+ * the clone flags for deciding how to linking
+ * up the child and parent processes.
+ * CHECK: Do we need to call get_arg0? The
+ * old version of the clone/clone2 code down below
+ * just read tcp->u_arg[arg0_index] without
+ * calling get_arg0. Was that a bug?
+ */
+ if (arg_setup (tcp, &state) < 0
+ || get_arg0 (tcp, &state, &tcp->inst[0]) < 0
+ || get_arg1 (tcp, &state, &tcp->inst[1]) < 0)
+ return -1;
+ tcp->u_arg[arg0_index] = tcp->inst[0];
+ return 0;
+ }
+ }
+# endif /* PTRACE_EVENT_FORK */
+
if (arg_setup (tcp, &state) < 0
|| get_arg0 (tcp, &state, &tcp->inst[0]) < 0
|| get_arg1 (tcp, &state, &tcp->inst[1]) < 0)
--
1.7.0.4
_______________________________________________
uClinux-dev mailing list
uClinux-dev@uclinux.org
http://mailman.uclinux.org/mailman/listinfo/uclinux-dev
This message was resent by uclinux-dev@uclinux.org
To unsubscribe see:
http://mailman.uclinux.org/mailman/options/uclinux-dev
