uprobes/tracing: uprobe_perf_open() forgets to handle the error from uprobe_apply()

PATCH In review Oleg Nesterov 3 participants 9 messages Apr 23, 2014 – Apr 24, 2014 Tracing

lore

Share & tools

lore

Oleg Nesterov

author

2014-04-23 16:58 UTC #0 lore raw

cover uprobes/tracing: uprobe_perf_open() forgets to handle the error from uprobe_apply()

Hello,

Steven, I am going to ask Ingo to pull this fix along with other pending uprobes changes, but please let me know if you want to take this patch.

Oleg.

 kernel/trace/trace_uprobe.c |   46 +++++++++++++++++++++++-------------------
 1 files changed, 25 insertions(+), 21 deletions(-)

Oleg Nesterov

author

2014-04-23 16:58 UTC #1 lore raw

PATCH 1/1 uprobes/tracing: uprobe_perf_open() forgets to handle the error from uprobe_apply()

uprobe_perf_open()->uprobe_apply() can fail, but this error is wrongly ignored. Change uprobe_perf_open() to do uprobe_perf_close() and return the error code in this case.

Change uprobe_perf_close() to propogate the error from uprobe_apply() as well, although it should not fail.

The patch looks more complicated because it moves uprobe_perf_close() up to make it visible to uprobe_perf_open().

Signed-off-by Oleg Nesterov <[email protected]>

---
 kernel/trace/trace_uprobe.c |   46 +++++++++++++++++++++++-------------------
 1 files changed, 25 insertions(+), 21 deletions(-)

kernel/trace/trace_uprobe.c +25 ~~-21~~

index 930e514..9aad3e2 100644
@@ -1003,56 +1003,60 @@ uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 	return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 }
 
-static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
+static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 {
 	bool done;
 
 	write_lock(&tu->filter.rwlock);
 	if (event->hw.tp_target) {
-		/*
-		 * event->parent != NULL means copy_process(), we can avoid
-		 * uprobe_apply(). current->mm must be probed and we can rely
-		 * on dup_mmap() which preserves the already installed bp's.
-		 *
-		 * attr.enable_on_exec means that exec/mmap will install the
-		 * breakpoints we need.
-		 */
+		list_del(&event->hw.tp_list);
 		done = tu->filter.nr_systemwide ||
-			event->parent || event->attr.enable_on_exec ||
+			(event->hw.tp_target->flags & PF_EXITING) ||
 			uprobe_filter_event(tu, event);
-		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
+		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
-		tu->filter.nr_systemwide++;
 	}
 	write_unlock(&tu->filter.rwlock);
 
 	if (!done)
-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 
 	return 0;
 }
 
-static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
+static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 {
 	bool done;
+	int err;
 
 	write_lock(&tu->filter.rwlock);
 	if (event->hw.tp_target) {
-		list_del(&event->hw.tp_list);
+		/*
+		 * event->parent != NULL means copy_process(), we can avoid
+		 * uprobe_apply(). current->mm must be probed and we can rely
+		 * on dup_mmap() which preserves the already installed bp's.
+		 *
+		 * attr.enable_on_exec means that exec/mmap will install the
+		 * breakpoints we need.
+		 */
 		done = tu->filter.nr_systemwide ||
-			(event->hw.tp_target->flags & PF_EXITING) ||
+			event->parent || event->attr.enable_on_exec ||
 			uprobe_filter_event(tu, event);
+		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
-		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
+		tu->filter.nr_systemwide++;
 	}
 	write_unlock(&tu->filter.rwlock);
 
-	if (!done)
-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
-
-	return 0;
+	err = 0;
+	if (!done) {
+		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		if (err)
+			uprobe_perf_close(tu, event);
+	}
+	return err;
 }
 
 static bool uprobe_perf_filter(struct uprobe_consumer *uc,

Signature · 1 line

1.5.5.1

Steven Rostedt

maintainer

2014-04-24 03:17 UTC #2 lore raw

Quoted 20 lines · On Wed, 23 Apr 2014 18:58:30 +0200 Oleg Nesterov <[email protected]> wrote:

uprobe_perf_open()->uprobe_apply() can fail, but this error is wrongly ignored. Change uprobe_perf_open() to do uprobe_perf_close() and return the error code in this case.
Change uprobe_perf_close() to propogate the error from uprobe_apply() as well, although it should not fail.
The patch looks more complicated because it moves uprobe_perf_close() up to make it visible to uprobe_perf_open().
Signed-off-by Oleg Nesterov <[email protected]>
---
 kernel/trace/trace_uprobe.c |   46 +++++++++++++++++++++++-------------------
 1 files changed, 25 insertions(+), 21 deletions(-)
kernel/trace/trace_uprobe.c +0 -0
index 930e514..9aad3e2 100644

kernel/trace/trace_uprobe.c uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event) · line 1006

Quoted 5 lines

kernel/trace/trace_uprobe.c +1 -1

@@ -1003,56 +1003,60 @@ uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 	return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 }
 
-static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
+static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)

Egad, this confused the heck out of me. I didn't notice the swap in functions and was wondering what you were doing. I didn't realize this is what you meant by moving the uprobe_perf_close() up. I was thinking you moved the call up or something, not the function itself physically in the file.

/me tries to continue dazed and confused.

Quoted 63 lines

kernel/trace/trace_uprobe.c +17 ~~-17~~

 {
 	bool done;
 
 	write_lock(&tu->filter.rwlock);
 	if (event->hw.tp_target) {
-		/*
-		 * event->parent != NULL means copy_process(), we can avoid
-		 * uprobe_apply(). current->mm must be probed and we can rely
-		 * on dup_mmap() which preserves the already installed bp's.
-		 *
-		 * attr.enable_on_exec means that exec/mmap will install the
-		 * breakpoints we need.
-		 */
+		list_del(&event->hw.tp_list);
 		done = tu->filter.nr_systemwide ||
-			event->parent || event->attr.enable_on_exec ||
+			(event->hw.tp_target->flags & PF_EXITING) ||
 			uprobe_filter_event(tu, event);
-		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
+		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
-		tu->filter.nr_systemwide++;
 	}
 	write_unlock(&tu->filter.rwlock);
 
 	if (!done)
-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 
 	return 0;
 }
 
-static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
+static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 {
 	bool done;
+	int err;
 
 	write_lock(&tu->filter.rwlock);
 	if (event->hw.tp_target) {
-		list_del(&event->hw.tp_list);
+		/*
+		 * event->parent != NULL means copy_process(), we can avoid
+		 * uprobe_apply(). current->mm must be probed and we can rely
+		 * on dup_mmap() which preserves the already installed bp's.
+		 *
+		 * attr.enable_on_exec means that exec/mmap will install the
+		 * breakpoints we need.
+		 */
 		done = tu->filter.nr_systemwide ||
-			(event->hw.tp_target->flags & PF_EXITING) ||
+			event->parent || event->attr.enable_on_exec ||
 			uprobe_filter_event(tu, event);
+		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
-		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
+		tu->filter.nr_systemwide++;
 	}
 	write_unlock(&tu->filter.rwlock);
 
-	if (!done)

kernel/trace/trace_uprobe.c uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event) · line 1059

Quoted 10 lines

kernel/trace/trace_uprobe.c +7 -3

-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
-
-	return 0;
+	err = 0;
+	if (!done) {
+		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		if (err)
+			uprobe_perf_close(tu, event);
+	}
+	return err;

You can add by Acked-by, but next time, please make this into two patches. One to do the move, the other to do the change.

Thanks!

-- Steve

Quoted 3 lines

kernel/trace/trace_uprobe.c +0 -0

 }
 
 static bool uprobe_perf_filter(struct uprobe_consumer *uc,

Oleg Nesterov

author

2014-04-24 11:55 UTC #3 lore raw

v2 cover uprobes/tracing: uprobe_perf_open()->uprobe_apply() fix

Quoted 6 lines · Steven Rostedt

Egad, this confused the heck out of me. I didn't notice the swap in functions and was wondering what you were doing. I didn't realize this is what you meant by moving the uprobe_perf_close() up. I was thinking you moved the call up or something, not the function itself physically in the file.

...

Quoted 3 lines

You can add by Acked-by, but next time, please make this into two patches. One to do the move, the other to do the change.

OK...

I tried to lessen the number of patches I have ;) But I agree, this simple fix looks too complicated without preparation which only moves the code.

Let me split it then. Result is the same do I preserved the acks, this is what I am going to add to my tree.

Thanks!

Oleg.

Oleg Nesterov

author

2014-04-24 11:55 UTC #4 lore raw

PATCH v2 1/2 uprobes/tracing: Make uprobe_perf_close() visible to uprobe_perf_open()

Preparation. Move uprobe_perf_close() up before uprobe_perf_open() to avoid the forward declaration in the next patch and make it readable.

Signed-off-by Oleg Nesterov <[email protected]>

Acked-by Steven Rostedt <[email protected]>

Acked-by Srikar Dronamraju <[email protected]>

---
 kernel/trace/trace_uprobe.c |   36 ++++++++++++++++++------------------
 1 files changed, 18 insertions(+), 18 deletions(-)

kernel/trace/trace_uprobe.c +18 ~~-18~~

index 930e514..1ed0030 100644
@@ -1003,54 +1003,54 @@ uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
 	return __uprobe_perf_filter(&tu->filter, event->hw.tp_target->mm);
 }
 
-static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
+static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 {
 	bool done;
 
 	write_lock(&tu->filter.rwlock);
 	if (event->hw.tp_target) {
-		/*
-		 * event->parent != NULL means copy_process(), we can avoid
-		 * uprobe_apply(). current->mm must be probed and we can rely
-		 * on dup_mmap() which preserves the already installed bp's.
-		 *
-		 * attr.enable_on_exec means that exec/mmap will install the
-		 * breakpoints we need.
-		 */
+		list_del(&event->hw.tp_list);
 		done = tu->filter.nr_systemwide ||
-			event->parent || event->attr.enable_on_exec ||
+			(event->hw.tp_target->flags & PF_EXITING) ||
 			uprobe_filter_event(tu, event);
-		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
+		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
-		tu->filter.nr_systemwide++;
 	}
 	write_unlock(&tu->filter.rwlock);
 
 	if (!done)
-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 
 	return 0;
 }
 
-static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
+static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 {
 	bool done;
 
 	write_lock(&tu->filter.rwlock);
 	if (event->hw.tp_target) {
-		list_del(&event->hw.tp_list);
+		/*
+		 * event->parent != NULL means copy_process(), we can avoid
+		 * uprobe_apply(). current->mm must be probed and we can rely
+		 * on dup_mmap() which preserves the already installed bp's.
+		 *
+		 * attr.enable_on_exec means that exec/mmap will install the
+		 * breakpoints we need.
+		 */
 		done = tu->filter.nr_systemwide ||
-			(event->hw.tp_target->flags & PF_EXITING) ||
+			event->parent || event->attr.enable_on_exec ||
 			uprobe_filter_event(tu, event);
+		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
-		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
+		tu->filter.nr_systemwide++;
 	}
 	write_unlock(&tu->filter.rwlock);
 
 	if (!done)
-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
+		uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
 
 	return 0;
 }

Signature · 1 line

1.5.5.1

Oleg Nesterov

author

2014-04-24 11:56 UTC #5 lore raw

PATCH v2 2/2 uprobes/tracing: Fix uprobe_perf_open() on uprobe_apply() failure

In reply to Oleg Nesterov

uprobe_perf_open()->uprobe_apply() can fail, but this error is wrongly ignored. Change uprobe_perf_open() to do uprobe_perf_close() and return the error code in this case.

Change uprobe_perf_close() to propogate the error from uprobe_apply() as well, although it should not fail.

Signed-off-by Oleg Nesterov <[email protected]>

Acked-by Steven Rostedt <[email protected]>

Acked-by Srikar Dronamraju <[email protected]>

---
 kernel/trace/trace_uprobe.c |   14 +++++++++-----
 1 files changed, 9 insertions(+), 5 deletions(-)

kernel/trace/trace_uprobe.c +9 -5

index 1ed0030..9aad3e2 100644
@@ -1020,7 +1020,7 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 	write_unlock(&tu->filter.rwlock);
 
 	if (!done)
-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
+		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 
 	return 0;
 }

@@ -1028,6 +1028,7 @@ static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
 static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 {
 	bool done;
+	int err;
 
 	write_lock(&tu->filter.rwlock);
 	if (event->hw.tp_target) {

@@ -1049,10 +1050,13 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 	}
 	write_unlock(&tu->filter.rwlock);
 
-	if (!done)
-		uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
-
-	return 0;
+	err = 0;
+	if (!done) {
+		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		if (err)
+			uprobe_perf_close(tu, event);
+	}
+	return err;
 }
 
 static bool uprobe_perf_filter(struct uprobe_consumer *uc,

Signature · 1 line

1.5.5.1

Steven Rostedt

maintainer

2014-04-24 12:37 UTC #6 lore raw

Re: [PATCH v2 0/2] uprobes/tracing: uprobe_perf_open()->uprobe_apply() fix

In reply to Oleg Nesterov

Quoted 9 lines · On Thu, 24 Apr 2014 13:54:58 +0200 Oleg Nesterov <[email protected]> wrote:

Quoted 6 lines · Steven Rostedt
Egad, this confused the heck out of me. I didn't notice the swap in functions and was wondering what you were doing. I didn't realize this is what you meant by moving the uprobe_perf_close() up. I was thinking you moved the call up or something, not the function itself physically in the file.
...
Quoted 1 line

Quoted 7 lines

Quoted 2 lines
You can add by Acked-by, but next time, please make this into two patches. One to do the move, the other to do the change.
OK...
I tried to lessen the number of patches I have ;) But I agree, this simple fix looks too complicated without preparation which only moves the code.

Yeah, sometimes a simple fix just seems wrong to break into two. But if it helps in code review, it's definitely worth it.

Quoted 3 lines

Let me split it then. Result is the same do I preserved the acks, this is what I am going to add to my tree.

They look good (and much easier to review).

-- Steve

Srikar Dronamraju

Acked-by on v1 1/1

2014-04-24 09:03 UTC #7 lore raw

In reply to Oleg Nesterov

Acked-by Srikar Dronamraju <[email protected]>

(with 2 nits that you can ignore)

kernel/trace/trace_uprobe.c uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event) · line 1019

Quoted 10 lines

kernel/trace/trace_uprobe.c +2 -3

 		done = tu->filter.nr_systemwide ||
-			event->parent || event->attr.enable_on_exec ||
+			(event->hw.tp_target->flags & PF_EXITING) ||
 			uprobe_filter_event(tu, event);
-		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
+		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
-		tu->filter.nr_systemwide++;
 	}

Nit: I think 2 lines can be made into
done = --tu->filter.nr_systemwide;

kernel/trace/trace_uprobe.c uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event) · line 1050

Quoted 10 lines

kernel/trace/trace_uprobe.c +3 -2

 		done = tu->filter.nr_systemwide ||
-			(event->hw.tp_target->flags & PF_EXITING) ||
+			event->parent || event->attr.enable_on_exec ||
 			uprobe_filter_event(tu, event);
+		list_add(&event->hw.tp_list, &tu->filter.perf_events);
 	} else {
-		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
+		tu->filter.nr_systemwide++;
 	}

Nit: Similarly lines can be made into
done = tu->filter.nr_systemwide++;

Signature · 2 lines

Thanks and Regards
Srikar Dronamraju

Oleg Nesterov

author

2014-04-24 12:27 UTC #8 lore raw

Quoted 2 lines · Srikar Dronamraju

Acked-by Srikar Dronamraju <[email protected]>

Thanks, included!

Quoted 7 lines

kernel/trace/trace_uprobe.c uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event) · line 1019
Quoted 4 lines
kernel/trace/trace_uprobe.c +1 -1
+		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
-		tu->filter.nr_systemwide++;
 	}
Nit: I think 2 lines can be made into
done = --tu->filter.nr_systemwide;

...

Quoted 7 lines

kernel/trace/trace_uprobe.c uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event) · line 1050
Quoted 4 lines
kernel/trace/trace_uprobe.c +1 -1
-		tu->filter.nr_systemwide--;
 		done = tu->filter.nr_systemwide;
+		tu->filter.nr_systemwide++;
 	}
Nit: Similarly lines can be made into
done = tu->filter.nr_systemwide++;

Yes, perhaps, but this has nothing to do with this patch, it does not change this code.

And can't resist... you know, initially I wrote this code this way, but then I decided to make it more straightforward to avoid the potential nits from reviewers ;)

Oleg.