Re: [PATCH -next v2 1/3] driver core: add error handling for devtmpfs_create_node()

Re: [PATCH -next v2 1/3] driver core: add error handling for devtmpfs_create_node()

[Nathan Chancellor]

Hi Longlong,

On Fri, Feb 10, 2023 at 09:54:42AM +0000, Longlong Xia wrote:
> In some cases, devtmpfs_create_node() can return error value.
> So, make use of it.
> 
> Signed-off-by: Longlong Xia <xialonglong1@huawei.com>
> ---
>  drivers/base/core.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/base/core.c b/drivers/base/core.c
> index 7dab705f2937..aaa3088e5456 100644
> --- a/drivers/base/core.c
> +++ b/drivers/base/core.c
> @@ -3405,7 +3405,9 @@ int device_add(struct device *dev)
>  		if (error)
>  			goto SysEntryError;
>  
> -		devtmpfs_create_node(dev);
> +		error = devtmpfs_create_node(dev);
> +		if (error)
> +			goto DevtmpfsError;
>  	}
>  
>  	/* Notify clients of device addition.  This call must come
> @@ -3461,6 +3463,8 @@ int device_add(struct device *dev)
>  done:
>  	put_device(dev);
>  	return error;
> + DevtmpfsError:
> +	device_remove_sys_dev_entry(dev);
>   SysEntryError:
>  	if (MAJOR(dev->devt))
>  		device_remove_file(dev, &dev_attr_dev);
> -- 
> 2.25.1
> 

After this change in -next as commit 31b4b6730fd4 ("driver core: add
error handling for devtmpfs_create_node()"), my test machines failed to
boot after the rootfs could not be mounted. I added some logging to see
which device was failing, which triggers a few times with the exact same
message:

  device: 'btrfs-control': devtmpfs_create_node() failed, err = -17

with -17 being -EEXIST. I am not sure why this device is getting
registered more than once, it appears to occur during module insertion
though, as I am able to get to systemd starting within the initrd.

Should this particular return value be downgraded to a warning so that
the device still loads or should the driver be fixed? I have cc'd the
btrfs mailing list, in case they have any input.

Cheers,
Nathan

Re: [PATCH -next v2 1/3] driver core: add error handling for devtmpfs_create_node()

[Nathan Chancellor]

On Mon, Feb 13, 2023 at 05:14:27PM -0700, Nathan Chancellor wrote:
> Hi Longlong,
> 
> On Fri, Feb 10, 2023 at 09:54:42AM +0000, Longlong Xia wrote:
> > In some cases, devtmpfs_create_node() can return error value.
> > So, make use of it.
> > 
> > Signed-off-by: Longlong Xia <xialonglong1@huawei.com>
> > ---
> >  drivers/base/core.c | 6 +++++-
> >  1 file changed, 5 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/base/core.c b/drivers/base/core.c
> > index 7dab705f2937..aaa3088e5456 100644
> > --- a/drivers/base/core.c
> > +++ b/drivers/base/core.c
> > @@ -3405,7 +3405,9 @@ int device_add(struct device *dev)
> >  		if (error)
> >  			goto SysEntryError;
> >  
> > -		devtmpfs_create_node(dev);
> > +		error = devtmpfs_create_node(dev);
> > +		if (error)
> > +			goto DevtmpfsError;
> >  	}
> >  
> >  	/* Notify clients of device addition.  This call must come
> > @@ -3461,6 +3463,8 @@ int device_add(struct device *dev)
> >  done:
> >  	put_device(dev);
> >  	return error;
> > + DevtmpfsError:
> > +	device_remove_sys_dev_entry(dev);
> >   SysEntryError:
> >  	if (MAJOR(dev->devt))
> >  		device_remove_file(dev, &dev_attr_dev);
> > -- 
> > 2.25.1
> > 
> 
> After this change in -next as commit 31b4b6730fd4 ("driver core: add
> error handling for devtmpfs_create_node()"), my test machines failed to
> boot after the rootfs could not be mounted. I added some logging to see
> which device was failing, which triggers a few times with the exact same
> message:
> 
>   device: 'btrfs-control': devtmpfs_create_node() failed, err = -17
> 
> with -17 being -EEXIST. I am not sure why this device is getting
> registered more than once, it appears to occur during module insertion
> though, as I am able to get to systemd starting within the initrd.
> 
> Should this particular return value be downgraded to a warning so that
> the device still loads or should the driver be fixed? I have cc'd the
> btrfs mailing list, in case they have any input.

Apologies, I see now that the second patch in this series is logging for
this, sorry for not using or noticing that sooner.

I am now looking at some of my other test machines and I see messages
similar to the one above for other devices:

[   16.268685] sound timer: failed to create snd/timer, ret = -17
[   16.274666] ALSA: unable to register timer device (-17)
[   16.500253] sound timer: failed to create snd/timer, ret = -17
[   16.506124] ALSA: unable to register timer device (-17)

[   18.273067] misc rfkill: failed to create rfkill, ret = -17
[   18.464836] misc rfkill: failed to create rfkill, ret = -17
[   19.366149] misc tun: failed to create net/tun, ret = -17
[   19.371613] tun: Can't register misc device 200
[   19.552362] misc tun: failed to create net/tun, ret = -17
[   19.557839] tun: Can't register misc device 200
[   23.124993] misc tun: failed to create net/tun, ret = -17
[   23.130460] tun: Can't register misc device 200
[   23.299267] misc tun: failed to create net/tun, ret = -17
[   23.299319] tun: Can't register misc device 200
[   27.050363] misc tun: failed to create net/tun, ret = -17
[   27.055824] tun: Can't register misc device 200
[   27.181496] misc tun: failed to create net/tun, ret = -17

So it seems like this is not a btrfs specific problem?

Cheers,
Nathan

Re: [PATCH -next v2 1/3] driver core: add error handling for devtmpfs_create_node()

[Greg KH]

On Mon, Feb 13, 2023 at 05:33:06PM -0700, Nathan Chancellor wrote:
> On Mon, Feb 13, 2023 at 05:14:27PM -0700, Nathan Chancellor wrote:
> > Hi Longlong,
> > 
> > On Fri, Feb 10, 2023 at 09:54:42AM +0000, Longlong Xia wrote:
> > > In some cases, devtmpfs_create_node() can return error value.
> > > So, make use of it.
> > > 
> > > Signed-off-by: Longlong Xia <xialonglong1@huawei.com>
> > > ---
> > >  drivers/base/core.c | 6 +++++-
> > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/base/core.c b/drivers/base/core.c
> > > index 7dab705f2937..aaa3088e5456 100644
> > > --- a/drivers/base/core.c
> > > +++ b/drivers/base/core.c
> > > @@ -3405,7 +3405,9 @@ int device_add(struct device *dev)
> > >  		if (error)
> > >  			goto SysEntryError;
> > >  
> > > -		devtmpfs_create_node(dev);
> > > +		error = devtmpfs_create_node(dev);
> > > +		if (error)
> > > +			goto DevtmpfsError;
> > >  	}
> > >  
> > >  	/* Notify clients of device addition.  This call must come
> > > @@ -3461,6 +3463,8 @@ int device_add(struct device *dev)
> > >  done:
> > >  	put_device(dev);
> > >  	return error;
> > > + DevtmpfsError:
> > > +	device_remove_sys_dev_entry(dev);
> > >   SysEntryError:
> > >  	if (MAJOR(dev->devt))
> > >  		device_remove_file(dev, &dev_attr_dev);
> > > -- 
> > > 2.25.1
> > > 
> > 
> > After this change in -next as commit 31b4b6730fd4 ("driver core: add
> > error handling for devtmpfs_create_node()"), my test machines failed to
> > boot after the rootfs could not be mounted. I added some logging to see
> > which device was failing, which triggers a few times with the exact same
> > message:
> > 
> >   device: 'btrfs-control': devtmpfs_create_node() failed, err = -17
> > 
> > with -17 being -EEXIST. I am not sure why this device is getting
> > registered more than once, it appears to occur during module insertion
> > though, as I am able to get to systemd starting within the initrd.
> > 
> > Should this particular return value be downgraded to a warning so that
> > the device still loads or should the driver be fixed? I have cc'd the
> > btrfs mailing list, in case they have any input.
> 
> Apologies, I see now that the second patch in this series is logging for
> this, sorry for not using or noticing that sooner.
> 
> I am now looking at some of my other test machines and I see messages
> similar to the one above for other devices:
> 
> [   16.268685] sound timer: failed to create snd/timer, ret = -17
> [   16.274666] ALSA: unable to register timer device (-17)
> [   16.500253] sound timer: failed to create snd/timer, ret = -17
> [   16.506124] ALSA: unable to register timer device (-17)
> 
> [   18.273067] misc rfkill: failed to create rfkill, ret = -17
> [   18.464836] misc rfkill: failed to create rfkill, ret = -17
> [   19.366149] misc tun: failed to create net/tun, ret = -17
> [   19.371613] tun: Can't register misc device 200
> [   19.552362] misc tun: failed to create net/tun, ret = -17
> [   19.557839] tun: Can't register misc device 200
> [   23.124993] misc tun: failed to create net/tun, ret = -17
> [   23.130460] tun: Can't register misc device 200
> [   23.299267] misc tun: failed to create net/tun, ret = -17
> [   23.299319] tun: Can't register misc device 200
> [   27.050363] misc tun: failed to create net/tun, ret = -17
> [   27.055824] tun: Can't register misc device 200
> [   27.181496] misc tun: failed to create net/tun, ret = -17
> 
> So it seems like this is not a btrfs specific problem?

Ok, let me go revert the series for now.  Longlong, please feel free to
resubmit after testing this a bit better.

thanks,

greg k-h