Re: [PATCH v2 12/14] fs/configfs: Add a callback to determine attribute visibility

[Joel Becker <jlbec@evilplan.org>]

On Fri, Mar 08, 2024 at 12:35:27PM -0600, Tom Lendacky wrote:
> In order to support dynamic decisions as to whether an attribute should be
> created, add a callback that returns a bool to indicate whether the
> attribute should be display. If no callback is registered, the attribute
> is displayed by default.

I'm curious what the strong value is in this extra callback.  As opposed
to not generating the attribute in the absence of a TPM (why create a
config_item at all?), merely having an empty response from the attribute,
or having `->show()` return -ENODEV or similar.

> 
> Cc: Joel Becker <jlbec@evilplan.org>
> Cc: Christoph Hellwig <hch@lst.de>
> Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
> ---
>  fs/configfs/file.c       |  13 +++++
>  include/linux/configfs.h | 114 +++++++++++++++++++++++++++------------
>  2 files changed, 93 insertions(+), 34 deletions(-)
> 
> diff --git a/fs/configfs/file.c b/fs/configfs/file.c
> index 0ad32150611e..c758bcc11235 100644
> --- a/fs/configfs/file.c
> +++ b/fs/configfs/file.c
> @@ -451,6 +451,12 @@ int configfs_create_file(struct config_item * item, const struct configfs_attrib
>  	umode_t mode = (attr->ca_mode & S_IALLUGO) | S_IFREG;
>  	int error = 0;
>  
> +	if (attr->ca_is_visible) {
> +		mode = attr->ca_is_visible(item, attr);
> +		if (!mode)
> +			return 0;

What value do we get from carrying the mode through here?  The API
proposed is "visible or not", which is a boolean.  Overloading that with
"also set the mode" is confusing, and it also can lead to the divergent
codepath problem you mentioned in your response, where
`->ca_is_visible()` fails to return the mode correctly.  If this was simpl
a boolean hook, the code could read like so:


```
	umode_t mode = (attr->ca_mode & S_IALLUGO) | S_IFREG;
	int error = 0;

	if (attr->ca_is_visible && !attr->ca_is_visible(item, attr))
   		return 0;
```

> diff --git a/include/linux/configfs.h b/include/linux/configfs.h
> index 2606711adb18..18011f78ffde 100644
> --- a/include/linux/configfs.h
> +++ b/include/linux/configfs.h
> @@ -112,39 +112,64 @@ static inline void configfs_add_default_group(struct config_group *new_group,
>  	list_add_tail(&new_group->group_entry, &group->default_groups);
>  }
>  
> +typedef umode_t (*configfs_is_visible_t)(const struct config_item *item,
> +					 const struct configfs_attribute *attr);
> +

We don't use typedefs of op functions anywhere else in configfs or
frankly the entire filesystem API.  Adding one here would just introduce
confusion.

>  struct configfs_attribute {
>  	const char		*ca_name;
>  	struct module 		*ca_owner;
>  	umode_t			ca_mode;
> +	configfs_is_visible_t	ca_is_visible;
>  	ssize_t (*show)(struct config_item *, char *);
>  	ssize_t (*store)(struct config_item *, const char *, size_t);
>  };
>  

Thanks,
Joel


-- 

Life's Little Instruction Book #306

	"Take a nap on Sunday afternoons."

			http://www.jlbec.org/
			jlbec@evilplan.org