[PATCH] vsock: ratelimit unknown ioctl error message

From: Colin Ian King <redacted>

When exercising the kernel with stress-ng with some ioctl tests the
"Unknown ioctl" error message is spamming the kernel log at a high
rate. Rate limit this message to reduce the noise.

Signed-off-by: Colin Ian King <redacted>
---
 net/vmw_vsock/af_vsock.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
index 9e93bc201cc0..b8feb9223454 100644
--- a/net/vmw_vsock/af_vsock.c
+++ b/net/vmw_vsock/af_vsock.c

@@ -2072,7 +2072,7 @@ static long vsock_dev_do_ioctl(struct file *filp,
 		break;
 
 	default:
-		pr_err("Unknown ioctl %d\n", cmd);
+		pr_err_ratelimited("Unknown ioctl %d\n", cmd);
 		retval = -EINVAL;
 	}
 

-- 
2.27.0

On Fri, Oct 23, 2020 at 01:21:13PM +0100, Colin King wrote:

Make sense, or maybe can we remove the error message returning only the
-EINVAL?

Both cases are fine for me:
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>

		retval = -EINVAL;
	}

-- 
2.27.0

On Fri, 23 Oct 2020 16:09:47 +0200 Stefano Garzarella wrote:

On Fri, Oct 23, 2020 at 01:21:13PM +0100, Colin King wrote:

quoted

From: Colin Ian King <redacted>

When exercising the kernel with stress-ng with some ioctl tests the
"Unknown ioctl" error message is spamming the kernel log at a high
rate. Rate limit this message to reduce the noise.

Signed-off-by: Colin Ian King <redacted>
---
net/vmw_vsock/af_vsock.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
index 9e93bc201cc0..b8feb9223454 100644
--- a/net/vmw_vsock/af_vsock.c
+++ b/net/vmw_vsock/af_vsock.c

@@ -2072,7 +2072,7 @@ static long vsock_dev_do_ioctl(struct file *filp,

		break;

	default:
-		pr_err("Unknown ioctl %d\n", cmd);
+		pr_err_ratelimited("Unknown ioctl %d\n", cmd);  

Make sense, or maybe can we remove the error message returning only the
-EINVAL?

+1

Both cases are fine for me:
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>

From: Stefano Garzarella

Sent: 23 October 2020 15:10

On Fri, Oct 23, 2020 at 01:21:13PM +0100, Colin King wrote:

quoted

From: Colin Ian King <redacted>

When exercising the kernel with stress-ng with some ioctl tests the
"Unknown ioctl" error message is spamming the kernel log at a high
rate. Rate limit this message to reduce the noise.

Signed-off-by: Colin Ian King <redacted>
---
net/vmw_vsock/af_vsock.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
index 9e93bc201cc0..b8feb9223454 100644
--- a/net/vmw_vsock/af_vsock.c
+++ b/net/vmw_vsock/af_vsock.c

@@ -2072,7 +2072,7 @@ static long vsock_dev_do_ioctl(struct file *filp,

		break;

	default:
-		pr_err("Unknown ioctl %d\n", cmd);
+		pr_err_ratelimited("Unknown ioctl %d\n", cmd);

Make sense, or maybe can we remove the error message returning only the
-EINVAL?

Isn't the canonical error for unknown ioctl codes -ENOTTY?

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

On Fri, Oct 23, 2020 at 09:30:59PM +0000, David Laight wrote:

From: Stefano Garzarella

quoted

Sent: 23 October 2020 15:10

On Fri, Oct 23, 2020 at 01:21:13PM +0100, Colin King wrote:

quoted

From: Colin Ian King <redacted>

When exercising the kernel with stress-ng with some ioctl tests the
"Unknown ioctl" error message is spamming the kernel log at a high
rate. Rate limit this message to reduce the noise.

Signed-off-by: Colin Ian King <redacted>
---
net/vmw_vsock/af_vsock.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
index 9e93bc201cc0..b8feb9223454 100644
--- a/net/vmw_vsock/af_vsock.c
+++ b/net/vmw_vsock/af_vsock.c

@@ -2072,7 +2072,7 @@ static long vsock_dev_do_ioctl(struct file *filp,

		break;

	default:
-		pr_err("Unknown ioctl %d\n", cmd);
+		pr_err_ratelimited("Unknown ioctl %d\n", cmd);

Make sense, or maybe can we remove the error message returning only the
-EINVAL?

Isn't the canonical error for unknown ioctl codes -ENOTTY?

Oh, thanks for pointing that out!

I had not paid attention to the error returned, but looking at it I noticed that perhaps the most appropriate would be -ENOIOCTLCMD.
In the ioctl syscall we return -ENOTTY, if the callback returns -ENOIOCTLCMD.

What do you think?

Stefano

From: Stefano Garzarella

Sent: 26 October 2020 08:43

...

quoted

Isn't the canonical error for unknown ioctl codes -ENOTTY?

Oh, thanks for pointing that out!

I had not paid attention to the error returned, but looking at it I
noticed that perhaps the most appropriate would be -ENOIOCTLCMD.
In the ioctl syscall we return -ENOTTY, if the callback returns
-ENOIOCTLCMD.

What do you think?

It is 729 v 443 in favour of ENOTTY (based on grep).

No idea where ENOIOCTLCMD comes from, but ENOTTY probably
goes back to the early 1970s.

The fact that the ioctl wrapper converts the value is a good
hint that userspace expects ENOTTY.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

On Mon, Oct 26, 2020 at 09:13:23AM +0000, David Laight wrote:

From: Stefano Garzarella

quoted

Sent: 26 October 2020 08:43

...

quoted

quoted

Isn't the canonical error for unknown ioctl codes -ENOTTY?

Oh, thanks for pointing that out!

I had not paid attention to the error returned, but looking at it I
noticed that perhaps the most appropriate would be -ENOIOCTLCMD.
In the ioctl syscall we return -ENOTTY, if the callback returns
-ENOIOCTLCMD.

What do you think?

It is 729 v 443 in favour of ENOTTY (based on grep).

Under net/ it is 6 vs 83 in favour of ENOIOCTLCMD.

No idea where ENOIOCTLCMD comes from, but ENOTTY probably
goes back to the early 1970s.

Me too.

The fact that the ioctl wrapper converts the value is a good
hint that userspace expects ENOTTY.

Agree on that, but since we are not interfacing directly with userspace, I think it is better to return the more specific error (ENOIOCTLCMD).

Thanks,
Stefano

From: Stefano Garzarella

Sent: 26 October 2020 09:39

On Mon, Oct 26, 2020 at 09:13:23AM +0000, David Laight wrote:

quoted

From: Stefano Garzarella

quoted

Sent: 26 October 2020 08:43

...

quoted

quoted

Isn't the canonical error for unknown ioctl codes -ENOTTY?

Oh, thanks for pointing that out!

I had not paid attention to the error returned, but looking at it I
noticed that perhaps the most appropriate would be -ENOIOCTLCMD.
In the ioctl syscall we return -ENOTTY, if the callback returns
-ENOIOCTLCMD.

What do you think?

It is 729 v 443 in favour of ENOTTY (based on grep).

Under net/ it is 6 vs 83 in favour of ENOIOCTLCMD.

quoted

No idea where ENOIOCTLCMD comes from, but ENOTTY probably
goes back to the early 1970s.

Me too.

quoted

The fact that the ioctl wrapper converts the value is a good
hint that userspace expects ENOTTY.

Agree on that, but since we are not interfacing directly with userspace,
I think it is better to return the more specific error (ENOIOCTLCMD).

I bet Linux thought it could use a different error code then
found that 'unknown ioctl' was spelt ENOTTY.

Back in the old days error values were probably almost unique.
strerror(EAGIAN) was "No more processes" for a long time!

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

On Mon, Oct 26, 2020 at 09:46:17AM +0000, David Laight wrote:

From: Stefano Garzarella

quoted

Sent: 26 October 2020 09:39

On Mon, Oct 26, 2020 at 09:13:23AM +0000, David Laight wrote:

quoted

From: Stefano Garzarella

quoted

Sent: 26 October 2020 08:43

...

quoted

quoted

Isn't the canonical error for unknown ioctl codes -ENOTTY?

Oh, thanks for pointing that out!

I had not paid attention to the error returned, but looking at it I
noticed that perhaps the most appropriate would be -ENOIOCTLCMD.
In the ioctl syscall we return -ENOTTY, if the callback returns
-ENOIOCTLCMD.

What do you think?

It is 729 v 443 in favour of ENOTTY (based on grep).

Under net/ it is 6 vs 83 in favour of ENOIOCTLCMD.

quoted

No idea where ENOIOCTLCMD comes from, but ENOTTY probably
goes back to the early 1970s.

Me too.

quoted

The fact that the ioctl wrapper converts the value is a good
hint that userspace expects ENOTTY.

Agree on that, but since we are not interfacing directly with userspace,
I think it is better to return the more specific error (ENOIOCTLCMD).

I bet Linux thought it could use a different error code then
found that 'unknown ioctl' was spelt ENOTTY.

It could be :-)

Anyway, as you pointed out, I think we should change the -EINVAL with -ENOTTY or -ENOIOCTLCMD.

@Jakub what do you suggest?

Thanks,
Stefano

On Mon, 26 Oct 2020 11:01:12 +0100 Stefano Garzarella wrote:

On Mon, Oct 26, 2020 at 09:46:17AM +0000, David Laight wrote:

quoted

From: Stefano Garzarella  

quoted

Sent: 26 October 2020 09:39

On Mon, Oct 26, 2020 at 09:13:23AM +0000, David Laight wrote:  

quoted

From: Stefano Garzarella  

quoted

Sent: 26 October 2020 08:43  

...  

quoted

quoted

Isn't the canonical error for unknown ioctl codes -ENOTTY?
 

Oh, thanks for pointing that out!

I had not paid attention to the error returned, but looking at it I
noticed that perhaps the most appropriate would be -ENOIOCTLCMD.
In the ioctl syscall we return -ENOTTY, if the callback returns
-ENOIOCTLCMD.

What do you think?  

It is 729 v 443 in favour of ENOTTY (based on grep).  

Under net/ it is 6 vs 83 in favour of ENOIOCTLCMD.
 

quoted

No idea where ENOIOCTLCMD comes from, but ENOTTY probably
goes back to the early 1970s.  

Me too.
 

quoted

The fact that the ioctl wrapper converts the value is a good
hint that userspace expects ENOTTY.  

Agree on that, but since we are not interfacing directly with userspace,
I think it is better to return the more specific error (ENOIOCTLCMD).  

I bet Linux thought it could use a different error code then
found that 'unknown ioctl' was spelt ENOTTY.  

6 more quoted lines

ENOIOCTLCMD is a kernel-internal high return code (515) which should 
be returned by the driver, but it's then caught inside the core and
translated to ENOTTY which is then returned to user space.

So you're both right, I guess? But the driver should use ENOIOCTLCMD.

On Mon, Oct 26, 2020 at 10:55:48AM -0700, Jakub Kicinski wrote:

On Mon, 26 Oct 2020 11:01:12 +0100 Stefano Garzarella wrote:

quoted

On Mon, Oct 26, 2020 at 09:46:17AM +0000, David Laight wrote:

quoted

From: Stefano Garzarella

quoted

Sent: 26 October 2020 09:39

On Mon, Oct 26, 2020 at 09:13:23AM +0000, David Laight wrote:

quoted

From: Stefano Garzarella

quoted

Sent: 26 October 2020 08:43

...

quoted

quoted

Isn't the canonical error for unknown ioctl codes -ENOTTY?

Oh, thanks for pointing that out!

I had not paid attention to the error returned, but looking at it I
noticed that perhaps the most appropriate would be -ENOIOCTLCMD.
In the ioctl syscall we return -ENOTTY, if the callback returns
-ENOIOCTLCMD.

What do you think?

It is 729 v 443 in favour of ENOTTY (based on grep).

Under net/ it is 6 vs 83 in favour of ENOIOCTLCMD.

quoted

No idea where ENOIOCTLCMD comes from, but ENOTTY probably
goes back to the early 1970s.

Me too.

quoted

The fact that the ioctl wrapper converts the value is a good
hint that userspace expects ENOTTY.

Agree on that, but since we are not interfacing directly with userspace,
I think it is better to return the more specific error (ENOIOCTLCMD).

I bet Linux thought it could use a different error code then
found that 'unknown ioctl' was spelt ENOTTY.

14 more quoted lines

Thanks for clarify!

@Colin, can you send a v2 removing the error message and updating the return value?

Thanks,
Stefano

On Mon, 26 Oct 2020 19:24:57 +0100 Stefano Garzarella wrote:

@Colin, can you send a v2 removing the error message and updating the 
return value?

Not as a single patch, please, these are two different changes.