Skip to contentSkip to conversation

cleanup ptrace stops and remove notify_parent

PATCH Andi Kleen 2 participants 4 messages Aug 31, 2004 – Sep 4, 2004 1 message not in the archive ptrace support

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Andi Kleen

author
#1 lore raw
Quoted 10 lines · Roland McGrath

This patch is against Linus's current tree.

This adds a new state TASK_TRACED that is used in place of TASK_STOPPED when a thread stops because it is ptraced. Now ptrace operations are only permitted when the target is in TASK_TRACED state, not in TASK_STOPPED. This means that if a process is stopped normally by a job control signal and then you PTRACE_ATTACH to it, you will have to send it a SIGCONT before you can do any ptrace operations on it. (The SIGCONT will be reported to ptrace and then you can discard it instead of passing it through when you call PTRACE_CONT et al.)

Are you sure such a user visible semantic change is a good idea?

I at least have written (not very important, but existing) user space code in the past that assumed it can stop and single step without SIGCONT. I wouldn't be surprised if other debuggers ran into the same issue.

And the Linux debugging world is not gdb only anymore, there are a lot of other users of ptrace around these days ...

I don't think it is very good to change such behaviour in 2.6. Please don't do it.

-Andi

#2 lore raw
Quoted 1 line

Are you sure such a user visible semantic change is a good idea?

Not if it's not too hard to avoid. I think I can make it not come up by converting TASK_STOPPED into TASK_TRACED when you use ptrace. It will just take some synchronization for that case.

Quoted 4 lines

I at least have written (not very important, but existing) user space code in the past that assumed it can stop and single step without SIGCONT. I wouldn't be surprised if other debuggers ran into the same issue.

I assume you mean stop with SIGSTOP, then attach with PTRACE_ATTACH, then step with PTRACE_SINGLESTEP. If you are already attached, then any stop will report to you first in TASK_TRACED and never end up TASK_STOPPED unless you explicitly pass a stop signal through with ptrace.

Thanks,
Roland

#3 lore raw

I don't have a program like yours to test that it was broken with my patch (now in 2.6.9-rc1-mm2). But the following additional patch relative to 2.6.9-rc1-mm2 should do it. I've tested that it doesn't create any new problems. I don't have something right handy that tests the case in question, but you've said that you do.

This makes any ptrace operation that finds the target in TASK_STOPPED state morph it into TASK_TRACED state before doing anything. This necessitates reverting the last_siginfo accesses to check instead of assume last_siginfo is set, since it's no longer impossible to be in TASK_TRACED without being stopped in ptrace_stop (though there are no associated races to worry about).

Thanks,
Roland

kernel/ptrace.c renamed from kernel/ptrace.c.~1~ +14 -5
@@ -81,13 +81,20 @@ int ptrace_check_attach(struct task_stru
 	 * be changed by us so it's not changing right after this.
 	 */
 	read_lock(&tasklist_lock);
-	if ((child->ptrace & PT_PTRACED) && child->parent == current)
+	if ((child->ptrace & PT_PTRACED) && child->parent == current &&
+	    child->signal != NULL) {
 		ret = 0;
+		spin_lock_irq(&child->sighand->siglock);
+		if (child->state == TASK_STOPPED) {
+			child->state = TASK_TRACED;
+		} else if (child->state != TASK_TRACED && !kill) {
+			ret = -ESRCH;
+		}
+		spin_unlock_irq(&child->sighand->siglock);
+	}
 	read_unlock(&tasklist_lock);
 
 	if (!ret && !kill) {
-		if (child->state != TASK_TRACED)
-			return -ESRCH;
 		wait_task_inactive(child);
 	}
 
@@ -298,13 +305,15 @@ static int ptrace_setoptions(struct task
 
 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t __user * data)
 {
-	BUG_ON(child->last_siginfo == NULL);
+	if (child->last_siginfo == NULL)
+		return -EINVAL;
 	return copy_siginfo_to_user(data, child->last_siginfo);
 }
 
 static int ptrace_setsiginfo(struct task_struct *child, siginfo_t __user * data)
 {
-	BUG_ON(child->last_siginfo == NULL);
+	if (child->last_siginfo == NULL)
+		return -EINVAL;
 	if (copy_from_user(child->last_siginfo, data, sizeof (siginfo_t)) != 0)
 		return -EFAULT;
 	return 0;
author
#4 lore raw
Quoted 5 lines · Roland McGrath

I don't have a program like yours to test that it was broken with my patch (now in 2.6.9-rc1-mm2). But the following additional patch relative to 2.6.9-rc1-mm2 should do it. I've tested that it doesn't create any new problems. I don't have something right handy that tests the case in question, but you've said that you do.

I am travelling and can't reach the code right now, but I will test later.

Quoted 7 lines

This makes any ptrace operation that finds the target in TASK_STOPPED state morph it into TASK_TRACED state before doing anything. This necessitates reverting the last_siginfo accesses to check instead of assume last_siginfo is set, since it's no longer impossible to be in TASK_TRACED without being stopped in ptrace_stop (though there are no associated races to worry about).

Looks good, thanks.

-Andi