Re: [PATCH] Introduce {sysfs,device}_create_file_mode

[Daniel Mack]

On Wed, May 12, 2010 at 10:15:46PM +0400, Anton Vorontsov wrote:
> We need to create attributes with different modes across devices.
> We can do this by modifying attr.mode between device_create_file
> invocations, but that is racy in case of globally defined attrs.
> 
> Luckily, there's sysfs_add_file_mode() function that seems to do
> exactly what we want, and if we use it, we don't need any locks
> to avoid races. Though, it isn't exposed via device-drivers core
> API.
> 
> Now that we need it, introduce sysfs_create_file_mode()
> and device_create_file_mode() functions. With these, creating
> attributes with different modes is a joy.

Ah, nice. Thanks for caring! If this is accepted, I'll respin my patches
and make them use it.

Thanks,
Daniel


> 
> Signed-off-by: Anton Vorontsov <cbouatmailru@xxxxxxxxx>
> ---
> 
> On Wed, May 12, 2010 at 12:28:25AM +0200, Daniel Mack wrote:
> > On Tue, May 11, 2010 at 10:23:47PM +0400, Anton Vorontsov wrote:
> > > Yes, power_supply_attrs is a global array, and you shouldn't change
> > > it between power_supply_register() calls.
> > > 
> > > If you don't see why it's a bad idea in general, think about it
> > > other way, a race:
> > > 
> > > ...someone registers psy0 with attr X marked as read-only...
> > > ...code flow stops before device_create_file(psy0, global->mode)..
> > > [preempt]
> > > ...someone registers psy1 with attr X marked as writable...
> > > ...you set up global->mode to 0644...
> > > [preempt again]
> > > ...we end up calling device_create_file(psy0, 0644)...
> > 
> > Ah, I see. But the struct passed in is just used as template, right?
> 
> In general, you can't assume that it's just a template,
> otherwise 'const ptrdiff_t off = attr - power_supply_attrs;'
> wouldn't work.
> 
> I.e. 'attr' points to exact attribute you specified during
> registration. Weather its fields are used as a template or
> not, we don't know (well, we know how the sysfs core uses them
> *today*, but we don't know how sysfs will use them tomorrow).
> 
> > So
> > for the particular case you outlined, a simple lock should do the trick,
> > right?
> 
> Yes, for the particular case.
> 
> Still, I don't think that playing with attr->mode is a good
> idea.
> 
> If we don't want to use attr->mode for mode specifier (and I
> think we don't), we must implement a clean API for such a case.
> 
> Luckily, there's sysfs_add_file_mode, so if anyone will want
> to change attr->mode behaviour, one will have to deal with
> that function, and so we'll be safe.
> 
> And we must use that function, not just silently play
> attr->mode games during registration. That also avoids
> any need for locking.
> 
> 
> Greg, does the patch look OK? If so, I'd like it to go
> via battery-2.6.git tree, along with patches that need
> that one.
> 
> You can see the original thread here:
> http://lkml.org/lkml/2010/5/11/71
> 
> See power_supply_create_attrs(), there we'll use
> device_create_file_mode() instead of messing with
> power_supply_attrs[psy->properties[j]].attr.mode.
> 
>  drivers/base/core.c    |   21 +++++++++++++++++++++
>  fs/sysfs/file.c        |   16 ++++++++++++++++
>  include/linux/device.h |    3 +++
>  include/linux/sysfs.h  |   10 ++++++++++
>  4 files changed, 50 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/base/core.c b/drivers/base/core.c
> index b56a0ba..1bc6134 100644
> --- a/drivers/base/core.c
> +++ b/drivers/base/core.c
> @@ -440,6 +440,27 @@ struct kset *devices_kset;
>   * device_create_file - create sysfs attribute file for device.
>   * @dev: device.
>   * @attr: device attribute descriptor.
> + * @mode: file mode.
> + *
> + * This function is similar to device_create_file(), except that it
> + * doesn't use attr->mode for mode specifier.
> + */
> +int device_create_file_mode(struct device *dev,
> +			    const struct device_attribute *attr,
> +			    mode_t mode)
> +{
> +	int error = 0;
> +
> +	if (dev)
> +		error = sysfs_create_file_mode(&dev->kobj, &attr->attr, mode);
> +	return error;
> +}
> +EXPORT_SYMBOL_GPL(device_create_file_mode);
> +
> +/**
> + * device_create_file - create sysfs attribute file for device.
> + * @dev: device.
> + * @attr: device attribute descriptor.
>   */
>  int device_create_file(struct device *dev,
>  		       const struct device_attribute *attr)
> diff --git a/fs/sysfs/file.c b/fs/sysfs/file.c
> index e222b25..7bebb60 100644
> --- a/fs/sysfs/file.c
> +++ b/fs/sysfs/file.c
> @@ -528,6 +528,22 @@ int sysfs_add_file(struct sysfs_dirent *dir_sd, const struct attribute *attr,
>  	return sysfs_add_file_mode(dir_sd, attr, type, attr->mode);
>  }
>  
> +/**
> + *	sysfs_create_file_mode - create an attribute file for an object.
> + *	@kobj:	object we're creating for.
> + *	@attr:	attribute descriptor.
> + *	@mode:	file mode.
> + *
> + *	This function is similar to sysfs_create_file(), except that it
> + *	doesn't use attr->mode for mode specifier.
> + */
> +int sysfs_create_file_mode(struct kobject *kobj, const struct attribute *attr,
> +			   mode_t mode)
> +{
> +	BUG_ON(!kobj || !kobj->sd || !attr);
> +
> +	return sysfs_add_file_mode(kobj->sd, attr, SYSFS_KOBJ_ATTR, mode);
> +}
>  
>  /**
>   *	sysfs_create_file - create an attribute file for an object.
> diff --git a/include/linux/device.h b/include/linux/device.h
> index 1821928..8277364 100644
> --- a/include/linux/device.h
> +++ b/include/linux/device.h
> @@ -337,6 +337,9 @@ struct device_attribute {
>  #define DEVICE_ATTR(_name, _mode, _show, _store) \
>  struct device_attribute dev_attr_##_name = __ATTR(_name, _mode, _show, _store)
>  
> +extern int __must_check device_create_file_mode(struct device *device,
> +					const struct device_attribute *entry,
> +					mode_t mode);
>  extern int __must_check device_create_file(struct device *device,
>  					const struct device_attribute *entry);
>  extern void device_remove_file(struct device *dev,
> diff --git a/include/linux/sysfs.h b/include/linux/sysfs.h
> index f0496b3..222774f 100644
> --- a/include/linux/sysfs.h
> +++ b/include/linux/sysfs.h
> @@ -130,6 +130,9 @@ int __must_check sysfs_rename_dir(struct kobject *kobj, const char *new_name);
>  int __must_check sysfs_move_dir(struct kobject *kobj,
>  				struct kobject *new_parent_kobj);
>  
> +int __must_check sysfs_create_file_mode(struct kobject *kobj,
> +					const struct attribute *attr,
> +					mode_t mode);
>  int __must_check sysfs_create_file(struct kobject *kobj,
>  				   const struct attribute *attr);
>  int __must_check sysfs_create_files(struct kobject *kobj,
> @@ -202,6 +205,13 @@ static inline int sysfs_move_dir(struct kobject *kobj,
>  	return 0;
>  }
>  
> +static inline int sysfs_create_file_mode(struct kobject *kobj,
> +					 const struct attribute *attr,
> +					 mode_t mode)
> +{
> +	return 0;
> +}
> +
>  static inline int sysfs_create_file(struct kobject *kobj,
>  				    const struct attribute *attr)
>  {
> -- 
> 1.7.0.5
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/