ptrace weirdness with 9.0-CURRENT

[Kostik Belousov]

On Wed, Feb 09, 2011 at 01:49:52AM +0200, Kostik Belousov wrote:
> On Wed, Feb 09, 2011 at 12:42:15AM +0200, Ali Polatel wrote:
> > Hello everyone,
> > 
> > I'm the developer of pinktrace - http://dev.exherbo.org/~alip/pinktrace/
> > - a simple ptrace() wrapper library for FreeBSD and Linux. I have set up
> > a FreeBSD-9.0-CURRENT VM today to test various new features recently
> > added to ptrace(). This is about a behaviour difference between
> > 8.1-RELEASE and 9.0-CURRENT which I've noticed through a unit test of
> > pinktrace. I don't want to bother you with the internals of this library
> > so I'll briefly explain the problem.
> > 
> > I've inserted the testcase I've used below. The aim is to trace a
> > open(NULL, 0) call which should fail with EFAULT. Running this on two
> > different VMs I get:
> > 
> > % uname -a
> > FreeBSD  9.0-CURRENT FreeBSD 9.0-CURRENT #0: Wed Feb  9 05:02:31 EET 2011     root@:/usr/obj/usr/src/sys/GENERIC  amd64
> > % sudo cat /root/world.txt
> > --------------------------------------------------------------
> > >>> World build completed on Wed Feb  9 00:23:30 EET 2011
> > --------------------------------------------------------------
> > % gcc -Wall ptrace-amd64-fbsd-return.c
> > % ./a.out
> > retval:0 error:0
> > 
> > $ uname -a
> > FreeBSD  8.1-RELEASE FreeBSD 8.1-RELEASE #0: Mon Jul 19 02:36:49 UTC 2010     root at mason.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC  amd64
> > $ gcc -Wall ptrace-amd64-fbsd-return.c
> > $ ./a.out
> > retval:14 error:1
> > $ 
> > 
> > Important note: I couldn't notice a problem with truss tracing a
> > open(NULL, 0) call so I think this is a problem with my testcase.
> > I'll be happy if you can shed some light on what I'm doing wrong here:
> There is no issue with ptrace(2). Your test fails because, apparently,
> rtld in HEAD calls setjmp(3) when resolving symbols, and setjmp(3)
> calls sigprocmask(2). The end result is that you get SCX event for
> sigprocmask, and not for your open(2).
> 
> The issue with sigprocmask call from setjmp shall be fixed, but this
> is not an issue with ptrace(2).

The following should fix the problem.

diff --git a/libexec/rtld-elf/rtld.c b/libexec/rtld-elf/rtld.c
index 50ab393..948cf49 100644
--- a/libexec/rtld-elf/rtld.c
+++ b/libexec/rtld-elf/rtld.c
@@ -560,7 +560,7 @@ _rtld_bind(Obj_Entry *obj, Elf_Size reloff)
     RtldLockState lockstate;
 
     rlock_acquire(rtld_bind_lock, &lockstate);
-    if (setjmp(lockstate.env) != 0)
+    if (sigsetjmp(lockstate.env, 0) != 0)
 	    lock_upgrade(rtld_bind_lock, &lockstate);
     if (obj->pltrel)
 	rel = (const Elf_Rel *) ((caddr_t) obj->pltrel + reloff);
@@ -2142,7 +2142,7 @@ dlopen(const char *name, int mode)
     ld_tracing = (mode & RTLD_TRACE) == 0 ? NULL : "1";
     if (ld_tracing != NULL) {
 	rlock_acquire(rtld_bind_lock, &lockstate);
-	if (setjmp(lockstate.env) != 0)
+	if (sigsetjmp(lockstate.env, 0) != 0)
 	    lock_upgrade(rtld_bind_lock, &lockstate);
 	environ = (char **)*get_program_var_addr("environ", &lockstate);
 	lock_release(rtld_bind_lock, &lockstate);
@@ -2264,7 +2264,7 @@ do_dlsym(void *handle, const char *name, void *retaddr, const Ver_Entry *ve,
     req.lockstate = &lockstate;
 
     rlock_acquire(rtld_bind_lock, &lockstate);
-    if (setjmp(lockstate.env) != 0)
+    if (sigsetjmp(lockstate.env, 0) != 0)
 	    lock_upgrade(rtld_bind_lock, &lockstate);
     if (handle == NULL || handle == RTLD_NEXT ||
 	handle == RTLD_DEFAULT || handle == RTLD_SELF) {
diff --git a/libexec/rtld-elf/rtld.h b/libexec/rtld-elf/rtld.h
index 8941d29..bb365a7 100644
--- a/libexec/rtld-elf/rtld.h
+++ b/libexec/rtld-elf/rtld.h
@@ -276,7 +276,7 @@ typedef struct Struct_DoneList {
 
 struct Struct_RtldLockState {
 	int lockstate;
-	jmp_buf env;
+	sigjmp_buf env;
 };
 
 /*
diff --git a/libexec/rtld-elf/rtld_lock.c b/libexec/rtld-elf/rtld_lock.c
index e76a4da..024e1e2 100644
--- a/libexec/rtld-elf/rtld_lock.c
+++ b/libexec/rtld-elf/rtld_lock.c
@@ -259,7 +259,7 @@ lock_restart_for_upgrade(RtldLockState *lockstate)
 	case RTLD_LOCK_WLOCKED:
 		break;
 	case RTLD_LOCK_RLOCKED:
-		longjmp(lockstate->env, 1);
+		siglongjmp(lockstate->env, 1);
 		break;
 	default:
 		assert(0);
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-hackers/attachments/20110208/d39f48c9/attachment.pgp