[RFC] [PATCH] selective signal ptracing

[RFC] [PATCH] selective signal ptracing

[John Blackwood]

Selective Ptraced Signal Support - A proposed enhancement

This is a proposal for a ptrace enhancement that adds two new ptrace(2)
commands that let a debugger view and modify the set of signals that
are being ptraced.

By default all signals are ptraced as before. However, a debugger
may now modify the set of per-task ptraced signals, where only the
signals in this ptrace signal mask will be ptraced.

Signals that are not being ptraced will be handled normally by the ptraced
process, and no notification will be passed on to the debugger process.
Only ptraced signals that are received by the ptraced task will cause
that task to stop and notify the debugger.

This new ptrace selective signal support provides a method for improving
debug session performance and for reducing the amount of ptrace-related
interference of a ptraced process, particularly when an an application
is making use of one or more signals that are not currently of interest
to the debug session, but are occuring frequently.

The command interfaces are:

   long int ptrace(enum __ptrace_request request,
                            pid_t target_pid, void *addr, void *data)

   request:
	When PTRACE_SETSIGTRACE is specified, then the set of signals
	in the sigset_t structure (pointed to by the data
	parameter) will become the set of signals that are ptraced for
	the target process.

	When PTRACE_GETSIGTRACE is specified, then the current set
	of ptraced signals will be returned at the sigset_t structure
	location (pointed to by the data parameter).

   pid:	pid of the target/ptraced task

   addr:	not used

   data:	pointer to a sigset_t structure


Tasks that have the PTRACE_O_TRACEFORK, PTRACE_O_TRACECLONE
or PTRACE_O_TRACECLONE option set at the time of a fork, vfork or clone
call that are currently being ptraced will yield child tasks
that inherited their parent task's ptrace signal mask.

When single-stepping and a user's signal handler is executed, the ptraced
task only single-steps/stops in the signal handler if that signal is in
the ptraced signal mask.  Otherwise, the ptraced task will execute the
entire signal handler and then stop at the next instruction in the normal
(non-signal related) execution flow.

The proposed patch below implements ptrace selective signal ptracing
for the i386 and x86_64 architecutres.  Other architectures will ptrace
all signals, in the same way as before.


diff -rup linux-2.6.22-rc4-git5.orig/arch/i386/kernel/ptrace.c 
linux-2.6.22-rc4-git5.new/arch/i386/kernel/ptrace.c
--- linux-2.6.22-rc4-git5.orig/arch/i386/kernel/ptrace.c	2007-06-13 
09:46:37.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/arch/i386/kernel/ptrace.c	2007-06-13 
17:00:38.000000000 -0400
@@ -620,6 +620,44 @@ long arch_ptrace(struct task_struct *chi
  					(struct user_desc __user *) data);
  		break;

+	case PTRACE_GETSIGTRACE: { /* Get the set of ptraced signals. */
+		int i;
+
+		if (!access_ok(VERIFY_WRITE, datap, sizeof(sigset_t))) {
+			ret = -EIO;
+			break;
+		}
+		for (i = 0; i < _NSIG_WORDS; i++) {
+			ret = __put_user(child->ptrace_sigmask.sig[i], datap);
+			if (ret)
+				break;
+			datap++;
+		}
+		break;
+	}
+
+	case PTRACE_SETSIGTRACE: { /* Set ptraced signal mask. */
+		int i;
+		sigset_t new_smask;
+
+		if (!access_ok( VERIFY_READ, datap, sizeof(sigset_t))) {
+			ret = -EIO;
+			break;
+		}
+		for (i = 0; i < _NSIG_WORDS; i++) {
+			ret = __get_user(new_smask.sig[i], datap);
+			if (ret)
+				break;
+			datap++;
+		}
+		if (ret)
+			break;
+
+		/* Update child task's ptrace signal mask. */
+		ptrace_update_traced_signals(child, &new_smask);
+		break;
+	}
+
  	default:
  		ret = ptrace_request(child, request, addr, data);
  		break;
diff -rup linux-2.6.22-rc4-git5.orig/arch/i386/kernel/signal.c 
linux-2.6.22-rc4-git5.new/arch/i386/kernel/signal.c
--- linux-2.6.22-rc4-git5.orig/arch/i386/kernel/signal.c	2007-06-13 
09:46:37.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/arch/i386/kernel/signal.c	2007-06-13 
17:00:38.000000000 -0400
@@ -393,8 +393,12 @@ static int setup_frame(int sig, struct k
  	 * handler too.
  	 */
  	regs->eflags &= ~TF_MASK;
-	if (test_thread_flag(TIF_SINGLESTEP))
-		ptrace_notify(SIGTRAP);
+	if (test_thread_flag(TIF_SINGLESTEP)) {
+		if (sigismember(&current->ptrace_sigmask, sig))
+			ptrace_notify(SIGTRAP);
+		else
+			clear_thread_flag(TIF_SINGLESTEP);
+	}

  #if DEBUG_SIG
  	printk("SIG deliver (%s:%d): sp=%p pc=%p ra=%p\n",
@@ -487,8 +491,12 @@ static int setup_rt_frame(int sig, struc
  	 * handler too.
  	 */
  	regs->eflags &= ~TF_MASK;
-	if (test_thread_flag(TIF_SINGLESTEP))
-		ptrace_notify(SIGTRAP);
+	if (test_thread_flag(TIF_SINGLESTEP)) {
+		if (sigismember(&current->ptrace_sigmask, sig))
+			ptrace_notify(SIGTRAP);
+		else
+			clear_thread_flag(TIF_SINGLESTEP);
+	}

  #if DEBUG_SIG
  	printk("SIG deliver (%s:%d): sp=%p pc=%p ra=%p\n",
@@ -538,7 +546,8 @@ handle_signal(unsigned long sig, siginfo
  	 * that register information in the sigcontext is correct.
  	 */
  	if (unlikely(regs->eflags & TF_MASK)
-	    && likely(current->ptrace & PT_DTRACE)) {
+	    && likely(current->ptrace & PT_DTRACE)
+	    && sigismember(&current->ptrace_sigmask, sig)) {
  		current->ptrace &= ~PT_DTRACE;
  		regs->eflags &= ~TF_MASK;
  	}
diff -rup linux-2.6.22-rc4-git5.orig/arch/x86_64/ia32/ia32_signal.c 
linux-2.6.22-rc4-git5.new/arch/x86_64/ia32/ia32_signal.c
--- linux-2.6.22-rc4-git5.orig/arch/x86_64/ia32/ia32_signal.c	2007-06-13 
09:46:38.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/arch/x86_64/ia32/ia32_signal.c	2007-06-13 
17:00:38.000000000 -0400
@@ -495,8 +495,12 @@ int ia32_setup_frame(int sig, struct k_s

  	set_fs(USER_DS);
  	regs->eflags &= ~TF_MASK;
-	if (test_thread_flag(TIF_SINGLESTEP))
-		ptrace_notify(SIGTRAP);
+	if (test_thread_flag(TIF_SINGLESTEP)) {
+		if (sigismember(&current->ptrace_sigmask, sig))
+			ptrace_notify(SIGTRAP);
+		else
+			clear_thread_flag(TIF_SINGLESTEP);
+	}

  #if DEBUG_SIG
  	printk("SIG deliver (%s:%d): sp=%p pc=%lx ra=%u\n",
@@ -601,8 +605,12 @@ int ia32_setup_rt_frame(int sig, struct

  	set_fs(USER_DS);
  	regs->eflags &= ~TF_MASK;
-	if (test_thread_flag(TIF_SINGLESTEP))
-		ptrace_notify(SIGTRAP);
+	if (test_thread_flag(TIF_SINGLESTEP)) {
+		if (sigismember(&current->ptrace_sigmask, sig))
+			ptrace_notify(SIGTRAP);
+		else
+			clear_thread_flag(TIF_SINGLESTEP);
+	}

  #if DEBUG_SIG
  	printk("SIG deliver (%s:%d): sp=%p pc=%lx ra=%u\n",
diff -rup linux-2.6.22-rc4-git5.orig/arch/x86_64/ia32/ptrace32.c 
linux-2.6.22-rc4-git5.new/arch/x86_64/ia32/ptrace32.c
--- linux-2.6.22-rc4-git5.orig/arch/x86_64/ia32/ptrace32.c	2007-04-25 
23:08:32.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/arch/x86_64/ia32/ptrace32.c	2007-06-13 
17:00:38.000000000 -0400
@@ -247,6 +247,8 @@ asmlinkage long sys32_ptrace(long reques
  	case PTRACE_SETOPTIONS:
  	case PTRACE_SET_THREAD_AREA:
  	case PTRACE_GET_THREAD_AREA:
+	case PTRACE_SETSIGTRACE:
+	case PTRACE_GETSIGTRACE:
  		return sys_ptrace(request, pid, addr, data);

  	default:
diff -rup linux-2.6.22-rc4-git5.orig/arch/x86_64/kernel/ptrace.c 
linux-2.6.22-rc4-git5.new/arch/x86_64/kernel/ptrace.c
--- linux-2.6.22-rc4-git5.orig/arch/x86_64/kernel/ptrace.c	2007-06-13 
09:46:38.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/arch/x86_64/kernel/ptrace.c	2007-06-13 
17:00:38.000000000 -0400
@@ -567,6 +567,48 @@ long arch_ptrace(struct task_struct *chi
  		break;
  	}

+	case PTRACE_GETSIGTRACE: { /* Get the set of ptraced signals. */
+		int i;
+
+		if (!access_ok(
+			VERIFY_WRITE, (unsigned *)data, sizeof(sigset_t)))
+		{
+			ret = -EIO;
+			break;
+		}
+		for (i = 0; i < _NSIG_WORDS; i++) {
+			ret = __put_user(child->ptrace_sigmask.sig[i],
+				(unsigned long *)data);
+			if (ret)
+				break;
+			data += sizeof(unsigned long);
+		}
+		break;
+	}
+
+	case PTRACE_SETSIGTRACE: { /* Set ptraced signal mask. */
+		int i;
+		sigset_t new_smask;
+
+		if (!access_ok(
+			VERIFY_READ, (unsigned *)data, sizeof(sigset_t)))
+		{
+			ret = -EIO;
+			break;
+		}
+		for (i = 0; i < _NSIG_WORDS; i++) {
+			ret = __get_user(new_smask.sig[i], (unsigned long *)data);
+			if (ret)
+				break;
+			data += sizeof(unsigned long);
+		}
+		if (ret)
+			break;
+		/* Update child task's ptrace signal mask. */
+		ptrace_update_traced_signals(child, &new_smask);
+		break;
+	}
+
  	default:
  		ret = ptrace_request(child, request, addr, data);
  		break;
diff -rup linux-2.6.22-rc4-git5.orig/arch/x86_64/kernel/signal.c 
linux-2.6.22-rc4-git5.new/arch/x86_64/kernel/signal.c
--- linux-2.6.22-rc4-git5.orig/arch/x86_64/kernel/signal.c	2007-06-13 
09:46:38.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/arch/x86_64/kernel/signal.c	2007-06-13 
17:00:38.000000000 -0400
@@ -297,8 +297,13 @@ static int setup_rt_frame(int sig, struc
  	set_fs(USER_DS);

  	regs->eflags &= ~TF_MASK;
-	if (test_thread_flag(TIF_SINGLESTEP))
-		ptrace_notify(SIGTRAP);
+	if (test_thread_flag(TIF_SINGLESTEP)) {
+		if (sigismember(&current->ptrace_sigmask, sig))
+			ptrace_notify(SIGTRAP);
+		else
+			clear_thread_flag(TIF_SINGLESTEP);
+	}
+
  #ifdef DEBUG_SIG
  	printk("SIG deliver (%s:%d): sp=%p pc=%lx ra=%p\n",
  		current->comm, current->pid, frame, regs->rip, frame->pretcode);
@@ -355,7 +360,8 @@ handle_signal(unsigned long sig, siginfo
  	 * correct.
  	 */
  	if (unlikely(regs->eflags & TF_MASK)) {
-		if (likely(current->ptrace & PT_DTRACE)) {
+		if (likely(current->ptrace & PT_DTRACE)
+		    && sigismember(&current->ptrace_sigmask, sig)) {
  			current->ptrace &= ~PT_DTRACE;
  			regs->eflags &= ~TF_MASK;
  		}
diff -rup linux-2.6.22-rc4-git5.orig/include/asm-i386/ptrace-abi.h 
linux-2.6.22-rc4-git5.new/include/asm-i386/ptrace-abi.h
--- linux-2.6.22-rc4-git5.orig/include/asm-i386/ptrace-abi.h	2007-04-25 
23:08:32.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/include/asm-i386/ptrace-abi.h	2007-06-13 
17:00:38.000000000 -0400
@@ -36,4 +36,7 @@
  #define PTRACE_SYSEMU		  31
  #define PTRACE_SYSEMU_SINGLESTEP  32

+#define PTRACE_GETSIGTRACE        33
+#define PTRACE_SETSIGTRACE        34
+
  #endif
diff -rup linux-2.6.22-rc4-git5.orig/include/asm-x86_64/ptrace-abi.h 
linux-2.6.22-rc4-git5.new/include/asm-x86_64/ptrace-abi.h
--- linux-2.6.22-rc4-git5.orig/include/asm-x86_64/ptrace-abi.h 
2007-04-25 23:08:32.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/include/asm-x86_64/ptrace-abi.h	2007-06-13 
17:00:38.000000000 -0400
@@ -48,4 +48,7 @@

  #define PTRACE_ARCH_PRCTL	  30	/* arch_prctl for child */

+#define PTRACE_GETSIGTRACE        33
+#define PTRACE_SETSIGTRACE        34
+
  #endif
diff -rup linux-2.6.22-rc4-git5.orig/include/linux/ptrace.h 
linux-2.6.22-rc4-git5.new/include/linux/ptrace.h
--- linux-2.6.22-rc4-git5.orig/include/linux/ptrace.h	2007-04-25 
23:08:32.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/include/linux/ptrace.h	2007-06-13 
17:00:38.000000000 -0400
@@ -92,6 +92,7 @@ extern void ptrace_disable(struct task_s
  extern int ptrace_check_attach(struct task_struct *task, int kill);
  extern int ptrace_request(struct task_struct *child, long request, 
long addr, long data);
  extern void ptrace_notify(int exit_code);
+extern void ptrace_update_traced_signals(struct task_struct *child, 
sigset_t *smask);
  extern void __ptrace_link(struct task_struct *child,
  			  struct task_struct *new_parent);
  extern void __ptrace_unlink(struct task_struct *child);
diff -rup linux-2.6.22-rc4-git5.orig/include/linux/sched.h 
linux-2.6.22-rc4-git5.new/include/linux/sched.h
--- linux-2.6.22-rc4-git5.orig/include/linux/sched.h	2007-06-13 
09:46:52.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/include/linux/sched.h	2007-06-13 
17:00:38.000000000 -0400
@@ -1029,6 +1029,8 @@ struct task_struct {

  	unsigned long ptrace_message;
  	siginfo_t *last_siginfo; /* For ptrace use.  */
+
+	sigset_t ptrace_sigmask; /* Only signals in this mask are ptraced. */
  /*
   * current io wait handle: wait queue entry to use for io waits
   * If this thread is processing aio, this points at the waitqueue
diff -rup linux-2.6.22-rc4-git5.orig/kernel/exit.c 
linux-2.6.22-rc4-git5.new/kernel/exit.c
--- linux-2.6.22-rc4-git5.orig/kernel/exit.c	2007-06-13 
09:46:52.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/kernel/exit.c	2007-06-13 
17:00:38.000000000 -0400
@@ -604,6 +604,18 @@ choose_new_parent(struct task_struct *p,
  	p->real_parent = reaper;
  }

+static inline void
+ptrace_sigmask_exit(struct task_struct *p, int ptraceflags)
+{
+	sigset_t new_sigmask;
+
+	if (!(ptraceflags & PT_PTRACED))
+		return;
+	/* Clear ptrace signal mask. */
+	sigemptyset(&new_sigmask);
+	ptrace_update_traced_signals(p, &new_sigmask);
+}
+
  static void
  reparent_thread(struct task_struct *p, struct task_struct *father, int 
traced)
  {
@@ -621,7 +633,9 @@ reparent_thread(struct task_struct *p, s
  		/* If this child is being traced, then we're the one tracing it
  		 * anyway, so let go of it.
  		 */
+		int ptrace = p->ptrace;
  		p->ptrace = 0;
+		ptrace_sigmask_exit(p, ptrace);
  		remove_parent(p);
  		p->parent = p->real_parent;
  		add_parent(p);
@@ -716,6 +730,7 @@ forget_original_parent(struct task_struc
  		} else {
  			/* reparent ptraced task to its real parent */
  			__ptrace_unlink (p);
+			ptrace_sigmask_exit(p, ptrace);
  			if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
  			    thread_group_empty(p))
  				do_notify_parent(p, p->exit_signal);
diff -rup linux-2.6.22-rc4-git5.orig/kernel/fork.c 
linux-2.6.22-rc4-git5.new/kernel/fork.c
--- linux-2.6.22-rc4-git5.orig/kernel/fork.c	2007-06-13 
09:46:47.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/kernel/fork.c	2007-06-13 
17:00:38.000000000 -0400
@@ -1168,6 +1168,17 @@ static struct task_struct *copy_process(
  	p->exit_state = 0;

  	/*
+	 * If child is being ptraced, then inherit the ptrace signal mask
+	 * from the parent.  When the ptrace signal mask support is not
+	 * used, all signals are ptraced and inherited by the child process.
+	 *
+	 * Clear the ptrace signal mask if this child process is not being
+	 * ptraced.
+	 */
+	if (!(p->ptrace & PT_PTRACED))
+		sigemptyset(&p->ptrace_sigmask);	/* no ptracing */
+
+	/*
  	 * Ok, make it visible to the rest of the system.
  	 * We dont wake it up yet.
  	 */
diff -rup linux-2.6.22-rc4-git5.orig/kernel/ptrace.c 
linux-2.6.22-rc4-git5.new/kernel/ptrace.c
--- linux-2.6.22-rc4-git5.orig/kernel/ptrace.c	2007-06-13 
09:46:47.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/kernel/ptrace.c	2007-06-13 
17:00:38.000000000 -0400
@@ -206,6 +206,9 @@ repeat:
  	if (capable(CAP_SYS_PTRACE))
  		task->ptrace |= PT_PTRACE_CAP;

+	/* All signals are ptraced by default. */
+	sigfillset(&task->ptrace_sigmask);
+
  	__ptrace_link(task, current);

  	force_sig_specific(SIGSTOP, task);
@@ -219,10 +222,17 @@ out:

  static inline void __ptrace_detach(struct task_struct *child, unsigned 
int data)
  {
+	sigset_t new_smask;
+
  	child->exit_code = data;
-	/* .. re-parent .. */
+	/* re-parent */
  	__ptrace_unlink(child);
-	/* .. and wake it up. */
+
+	/* Clear the ptrace signal mask. */
+	sigemptyset(&new_smask);
+	ptrace_update_traced_signals(child, &new_smask);
+
+	/* Wake it up. */
  	if (child->exit_state != EXIT_ZOMBIE)
  		wake_up_process(child);
  }
@@ -412,8 +422,11 @@ int ptrace_traceme(void)
  		/*
  		 * Set the ptrace bit in the process ptrace flags.
  		 */
-		if (!ret)
+		if (!ret) {
  			current->ptrace |= PT_PTRACED;
+			/* All signals are ptraced by default */
+			sigfillset(&current->ptrace_sigmask);
+		}
  	}
  	task_unlock(current);
  	return ret;
diff -rup linux-2.6.22-rc4-git5.orig/kernel/signal.c 
linux-2.6.22-rc4-git5.new/kernel/signal.c
--- linux-2.6.22-rc4-git5.orig/kernel/signal.c	2007-06-13 
09:46:52.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/kernel/signal.c	2007-06-14 
10:28:03.000000000 -0400
@@ -47,7 +47,7 @@ static int sig_ignored(struct task_struc
  	/*
  	 * Tracers always want to know about signals..
  	 */
-	if (t->ptrace & PT_PTRACED)
+	if (t->ptrace & PT_PTRACED && sigismember(&(t)->ptrace_sigmask, sig))
  		return 0;

  	/*
@@ -1610,6 +1610,58 @@ void ptrace_notify(int exit_code)
  	spin_unlock_irq(&current->sighand->siglock);
  }

+/*
+ * void ptrace_update_traced_signals(
+ *			struct task_struct *child, sigset_t *new_smaskp)
+ *
+ *	Update the signal mask of ptraced signals, removing any ignored
+ *	pending signals that are no longer ptraced.  'new_smaskp' points
+ *	to the new ptrace signal mask which will be stored into
+ *	child->ptrace_sigmask.
+ */
+void ptrace_update_traced_signals(struct task_struct *child,
+							sigset_t *new_smaskp)
+{
+	int sig = 0, index = 0;
+	unsigned long int flags;
+	sigset_t prev_smask;
+
+	/* Get signals that were previously, but no longer ptraced. */
+	signandsets(&prev_smask, &child->ptrace_sigmask, new_smaskp);
+
+	child->ptrace_sigmask = *new_smaskp;
+
+	while (!sigisemptyset(&prev_smask)) {
+		while (index < _NSIG_WORDS) {
+			if (prev_smask.sig[index]) {
+                               	sig = ffz(~prev_smask.sig[index]) +
+						(index * _NSIG_BPW) + 1;
+				break;
+			}
+			index++;
+		}
+		/* Remove signal from the previously ptraced signal mask. */
+		sigdelset(&prev_smask, sig);
+
+		spin_lock_irqsave(&child->sighand->siglock, flags);
+
+		if (child->sighand == NULL) {
+			/* This task is exiting, so just return. */
+			spin_unlock_irqrestore(&child->sighand->siglock, flags);
+			return;
+		}
+		if (!sig_ignored(child, sig)) {
+			spin_unlock_irqrestore(&child->sighand->siglock, flags);
+			continue;
+		}
+		/* Remove any queued signals - they should now be ignored. */
+		rm_from_queue(sigmask(sig), &child->pending);
+		rm_from_queue(sigmask(sig), &child->signal->shared_pending);
+
+		spin_unlock_irqrestore(&child->sighand->siglock, flags);
+	}
+}
+
  static void
  finish_stop(int stop_count)
  {
@@ -1750,7 +1802,8 @@ relock:
  		if (!signr)
  			break; /* will return 0 */

-		if ((current->ptrace & PT_PTRACED) && signr != SIGKILL) {
+		if ((current->ptrace & PT_PTRACED) && signr != SIGKILL &&
+				sigismember(&current->ptrace_sigmask, signr)) {
  			ptrace_signal_deliver(regs, cookie);

  			/* Let the debugger run.  */

Re: [RFC] [PATCH] selective signal ptracing

[Roland McGrath]

I am not in favor of any enhancements to the ptrace interface.
It is a terrible interface and just needs to die.


Thanks,
Roland

Re: [RFC] [PATCH] selective signal ptracing

[Alan Cox]

On Fri, 15 Jun 2007 12:42:29 -0700 (PDT)
Roland McGrath <roland@redhat.com> wrote:

> I am not in favor of any enhancements to the ptrace interface.
> It is a terrible interface and just needs to die.

That might make sense if utrace ever looked like it would solve the
questions about platforms like ARM

Re: [RFC] [PATCH] selective signal ptracing

[Roland McGrath]

> That might make sense if utrace ever looked like it would solve the
> questions about platforms like ARM

It certainly will.  The only difficult limitations have been in
communication and understanding.  Please don't perpetuate a generic red
herring without adding any content to the subject.


Thanks,
Roland

Re: [RFC] [PATCH] selective signal ptracing

[Benjamin Herrenschmidt]

On Sat, 2007-06-16 at 00:26 +0100, Alan Cox wrote:
> On Fri, 15 Jun 2007 12:42:29 -0700 (PDT)
> Roland McGrath <roland@redhat.com> wrote:
> 
> > I am not in favor of any enhancements to the ptrace interface.
> > It is a terrible interface and just needs to die.
> 
> That might make sense if utrace ever looked like it would solve the
> questions about platforms like ARM

What are the issues with arch like ARM ? Also, is there a mail thread or
paper somewhere describing utrace, its interfaces and what's needed to
hook it up on a platform ?

/me mostly missed that train

Thanks !

Cheers,
Ben.

Re: [RFC] [PATCH] selective signal ptracing

[Roland McGrath]

> What are the issues with arch like ARM ? 

The interesting class ARM belongs to is machines that don't (or don't
always) have hardware support for single-step.  Maintaining the status quo
of how PTRACE_SINGLESTEP functions on these machines is different in
implementation under utrace than it is for machines that always use
hardware support.  That is the only special complication for ARM, and it is
not really very complicated.  Apparently the way I described the issue in
the past was easily misunderstood.

> Also, is there a mail thread or paper somewhere describing utrace, its
> interfaces and what's needed to hook it up on a platform ?

See http://redhat.com/~roland/utrace/ for the patches and a porting howto.
The patches add a Documentation/ file and kerneldoc stuff for the interfaces.


Thanks,
Roland

Re: [RFC] [PATCH] selective signal ptracing

[Josh Boyer]

On 6/16/07, Roland McGrath <roland@redhat.com> wrote:
> > What are the issues with arch like ARM ?
>
> The interesting class ARM belongs to is machines that don't (or don't
> always) have hardware support for single-step.  Maintaining the status quo
> of how PTRACE_SINGLESTEP functions on these machines is different in
> implementation under utrace than it is for machines that always use
> hardware support.  That is the only special complication for ARM, and it is
> not really very complicated.  Apparently the way I described the issue in
> the past was easily misunderstood.

That isn't just ARM.  There are some embedded PowerPC chips that lack
an easily usable hardware single step.  Just an FYI.

josh

Re: [RFC] [PATCH] selective signal ptracing

[Benjamin Herrenschmidt]

On Mon, 2007-06-18 at 08:15 -0500, Josh Boyer wrote:
> On 6/16/07, Roland McGrath <roland@redhat.com> wrote:
> > > What are the issues with arch like ARM ?
> >
> > The interesting class ARM belongs to is machines that don't (or don't
> > always) have hardware support for single-step.  Maintaining the status quo
> > of how PTRACE_SINGLESTEP functions on these machines is different in
> > implementation under utrace than it is for machines that always use
> > hardware support.  That is the only special complication for ARM, and it is
> > not really very complicated.  Apparently the way I described the issue in
> > the past was easily misunderstood.
> 
> That isn't just ARM.  There are some embedded PowerPC chips that lack
> an easily usable hardware single step.  Just an FYI.

Note that we could use xmon infrastructure for that ... Paul has the
whole shebang done in there, including emulating branches etc...

Ben.

Re: [RFC] [PATCH] selective signal ptracing

[John Blackwood]

 > Subject: Re: [RFC] [PATCH] selective signal ptracing
 > From: Oleg Nesterov <oleg@tv-sign.ru>
 > Date: Sat, 16 Jun 2007 13:24:15 +0400
 > To: John Blackwood <john.blackwood@ccur.com>
 > CC: Alan Cox <alan@lxorguk.ukuu.org.uk>, Roland McGrath 
<roland@redhat.com>, linux-kernel@vger.kernel.org
 >
 > John Blackwood wrote:
 > > >
 > > > By default all signals are ptraced as before. However, a debugger
 > > > may now modify the set of per-task ptraced signals, where only the
 > > > signals in this ptrace signal mask will be ptraced.
 >
 > I must admit, I agree with Roland...
 >
 > > > +void ptrace_update_traced_signals(struct task_struct *child,
 > > > +							sigset_t *new_smaskp)
 > > > +{
 > > > [...snip...]
 > > > +
 > > > +		spin_lock_irqsave(&child->sighand->siglock, flags);
 > > > +
 > > > +		if (child->sighand == NULL) {
 >
 > How so?
 >
 > Oleg.

Hi Oleg.

Right, this above is not a good check for a NULL sighand.

Possibly, this routine could use something like the
rcu_read_lock()/lock_task_sighand() sequences used elsewhere
(shown below).

The whole ptrace_update_traced_signals() where we drain ignored signals
is not dealt with today by ptraced tasks when the debugger exits or
detaches from a target task.  So this whole routine is not entirely
related just to selective signal ptracing.

Thanks for you input.


diff -rup linux-2.6.22-rc4-git5.old/kernel/exit.c 
linux-2.6.22-rc4-git5.new/kernel/exit.c
--- linux-2.6.22-rc4-git5.old/kernel/exit.c	2007-06-18 
14:36:09.000000000 -0400
+++ linux-2.6.22-rc4-git5.new/kernel/exit.c	2007-06-18 
14:31:16.000000000 -0400

+/*
+ * void ptrace_update_traced_signals(
+ *			struct task_struct *child, sigset_t *new_smaskp)
+ *
+ *	Update the signal mask of ptraced signals, removing any ignored
+ *	pending signals that are no longer ptraced.  'new_smaskp' points
+ *	to the new ptrace signal mask which will be stored into
+ *	child->ptrace_sigmask.
+ */
+void ptrace_update_traced_signals(struct task_struct *child,
+							sigset_t *new_smaskp)
+{
+	int sig = 0, index = 0;
+	unsigned long int flags;
+	sigset_t prev_smask;
+
+	/* Get signals that were previously, but no longer ptraced. */
+	signandsets(&prev_smask, &child->ptrace_sigmask, new_smaskp);
+	child->ptrace_sigmask = *new_smaskp;
+	rcu_read_lock();
+	while (!sigisemptyset(&prev_smask)) {
+		while (index < _NSIG_WORDS) {
+			if (prev_smask.sig[index]) {
+                               	sig = ffz(~prev_smask.sig[index]) +
+						(index * _NSIG_BPW) + 1;
+				break;
+			}
+			index++;
+		}
+		/* Remove signal from the previously ptraced signal mask. */
+		sigdelset(&prev_smask, sig);
+
+		if (!lock_task_sighand(child, &flags)) {
+			/* This task is exiting, so just return. */
+			unlock_task_sighand(child, &flags);
+			rcu_read_unlock();
+			return;
+		}
+		if (!sig_ignored(child, sig)) {
+			unlock_task_sighand(child, &flags);
+			continue;
+		}
+		/* Remove any queued signals - they should now be ignored. */
+		rm_from_queue(sigmask(sig), &child->pending);
+		rm_from_queue(sigmask(sig), &child->signal->shared_pending);
+		unlock_task_sighand(child, &flags);
+	}
+	rcu_read_unlock();
+}

Re: [RFC] [PATCH] selective signal ptracing

[Oleg Nesterov]

John Blackwood wrote:
>
> By default all signals are ptraced as before. However, a debugger
> may now modify the set of per-task ptraced signals, where only the
> signals in this ptrace signal mask will be ptraced.

I must admit, I agree with Roland...

> +void ptrace_update_traced_signals(struct task_struct *child,
> +							sigset_t *new_smaskp)
> +{
> [...snip...]
> +
> +		spin_lock_irqsave(&child->sighand->siglock, flags);
> +
> +		if (child->sighand == NULL) {

How so?

Oleg.