Skip to contentSkip to conversation

trace: Return ENOTCONN instead of EBADF

PATCH In review Peter Enderborg 2 participants 4 messages Oct 12, 2020 Tracing

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 trace: Return ENOTCONN instead of EBADF

When there is no clients listening on event the trace return EBADF. The file is not a bad file descriptor and to get the userspace able to do a proper error handling it need a different error code that separate a bad file descriptor from a empty listening.

---
 kernel/trace/trace.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
kernel/trace/trace.c +4 -4
index d3e5de717df2..6e592bf736df 100644
@@ -6651,8 +6651,8 @@ tracing_mark_write(struct file *filp, const char __user *ubuf,
 	event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
 					    irq_flags, preempt_count());
 	if (unlikely(!event))
-		/* Ring buffer disabled, return as if not open for write */
-		return -EBADF;
+		/* Ring buffer disabled, return as if not connected */
+		return -ENOTCONN;
 
 	entry = ring_buffer_event_data(event);
 	entry->ip = _THIS_IP_;
@@ -6731,8 +6731,8 @@ tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
 	event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
 					    irq_flags, preempt_count());
 	if (!event)
-		/* Ring buffer disabled, return as if not open for write */
-		return -EBADF;
+		/* Ring buffer disabled, return not connected */
+		return -ENOTCONN;
 
 	entry = ring_buffer_event_data(event);
 
Signature · 1 line
2.17.1
maintainer
#1 lore raw
Quoted 4 lines · On Mon, 12 Oct 2020 10:26:42 +0200 Peter Enderborg <[email protected]>…

When there is no clients listening on event the trace return EBADF. The file is not a bad file descriptor and to get the userspace able to do a proper error handling it need a different error code that separate a bad file descriptor from a empty listening.

I have no problem with this patch, but your description is incorrect. And before making this change, I want to make sure that what you think is happening is actually happening.

This has nothing to do with "clients listening". This happens when the ring buffer is disabled for some reason. The most likely case of this happening is if someone sets /sys/kernel/tracing/tracing_on to zero.

If this is still something you want applied, please update the change log to a more accurate scenario.

Thanks,

-- Steve

Quoted 32 lines
---
 kernel/trace/trace.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
kernel/trace/trace.c +4 -4
index d3e5de717df2..6e592bf736df 100644
@@ -6651,8 +6651,8 @@ tracing_mark_write(struct file *filp, const char __user *ubuf,
 	event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
 					    irq_flags, preempt_count());
 	if (unlikely(!event))
-		/* Ring buffer disabled, return as if not open for write */
-		return -EBADF;
+		/* Ring buffer disabled, return as if not connected */
+		return -ENOTCONN;
 
 	entry = ring_buffer_event_data(event);
 	entry->ip = _THIS_IP_;
@@ -6731,8 +6731,8 @@ tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
 	event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
 					    irq_flags, preempt_count());
 	if (!event)
-		/* Ring buffer disabled, return as if not open for write */
-		return -EBADF;
+		/* Ring buffer disabled, return not connected */
+		return -ENOTCONN;
 
 	entry = ring_buffer_event_data(event);
 
#2 lore raw
Quoted 7 lines · Steven Rostedt
Quoted 4 lines · On Mon, 12 Oct 2020 10:26:42 +0200 Peter Enderborg <[email protected]>…

When there is no clients listening on event the trace return EBADF. The file is not a bad file descriptor and to get the userspace able to do a proper error handling it need a different error code that separate a bad file descriptor from a empty listening.

Quoted 7 lines

I have no problem with this patch, but your description is incorrect. And before making this change, I want to make sure that what you think is happening is actually happening.

This has nothing to do with "clients listening". This happens when the ring buffer is disabled for some reason. The most likely case of this happening is if someone sets /sys/kernel/tracing/tracing_on to zero.

I see that as no one is listening. You start to listen by setting this tracing on some instance, but that is for trace_pipe. Is it the same flag for raw access and all ways you can invoke a trace?

Would

"When there is no instances listening on events the trace return
EBADF, it is when the tracing_on is off globally. The file is not a bad file descriptor and to get the userspace able to do a proper error handling it need a different error code that separate a bad file descriptor from a empty listening."

be a ok?

Quoted 39 lines

If this is still something you want applied, please update the change log to a more accurate scenario.

Thanks,

-- Steve

Quoted 31 lines
---
 kernel/trace/trace.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
kernel/trace/trace.c +4 -4
index d3e5de717df2..6e592bf736df 100644
@@ -6651,8 +6651,8 @@ tracing_mark_write(struct file *filp, const char __user *ubuf,
 	event = __trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
 					    irq_flags, preempt_count());
 	if (unlikely(!event))
-		/* Ring buffer disabled, return as if not open for write */
-		return -EBADF;
+		/* Ring buffer disabled, return as if not connected */
+		return -ENOTCONN;
 
 	entry = ring_buffer_event_data(event);
 	entry->ip = _THIS_IP_;
@@ -6731,8 +6731,8 @@ tracing_mark_raw_write(struct file *filp, const char __user *ubuf,
 	event = __trace_buffer_lock_reserve(buffer, TRACE_RAW_DATA, size,
 					    irq_flags, preempt_count());
 	if (!event)
-		/* Ring buffer disabled, return as if not open for write */
-		return -EBADF;
+		/* Ring buffer disabled, return not connected */
+		return -ENOTCONN;
 
 	entry = ring_buffer_event_data(event);
 
maintainer
#3 lore raw
Quoted 12 lines · Peter
Quoted 11 lines · Steven Rostedt
Quoted 4 lines · On Mon, 12 Oct 2020 10:26:42 +0200 Peter Enderborg <[email protected]>…

When there is no clients listening on event the trace return EBADF. The file is not a bad file descriptor and to get the userspace able to do a proper error handling it need a different error code that separate a bad file descriptor from a empty listening.

I have no problem with this patch, but your description is incorrect. And before making this change, I want to make sure that what you think is happening is actually happening.

Quoted 7 lines
Quoted 3 lines

This has nothing to do with "clients listening". This happens when the ring buffer is disabled for some reason. The most likely case of this happening is if someone sets /sys/kernel/tracing/tracing_on to zero.

I see that as no one is listening. You start to listen by setting this tracing on some instance, but that is for trace_pipe. Is it the same flag for raw access and all ways you can invoke a trace?

I don't know what you mean by "setting this tracing on some instance".

When you boot up, who is listening? tracing_on is enabled. But there's no listener.

Quoted 4 lines

Would

"When there is no instances listening on events the trace return

What instance are you talking about? What do you think needs to be listening. And no, trace_pipe plays zero role in this.

Quoted 6 lines

EBADF, it is when the tracing_on is off globally. The file is not a bad file descriptor and to get the userspace able to do a proper error handling it need a different error code that separate a bad file descriptor from a empty listening."

be a ok?

No. Because I don't understand any of the above!

There's no a producer / consumer here. It's a read / write enabled or disabled. If the ring buffer is disabled from write, and you try to write to the ring buffer, you get an error.

Basically, it's the same analogy as trying to write to a read-only file. Perhaps the proper error is EPERM.

-- Steve