Re: [PATCH v4 12/15] fs/configfs: Add a callback to determine attribute visibility

[Dan Williams]

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 displayed. If no callback is registered, the attribute
> is displayed by default.
> 
> Cc: Joel Becker <jlbec@xxxxxxxxxxxx>
> Cc: Christoph Hellwig <hch@xxxxxx>
> Co-developed-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> Signed-off-by: Tom Lendacky <thomas.lendacky@xxxxxxx>
> ---
>  fs/configfs/dir.c        | 20 ++++++++++++++++++++
>  include/linux/configfs.h |  3 +++
>  2 files changed, 23 insertions(+)
> 
> diff --git a/fs/configfs/dir.c b/fs/configfs/dir.c
> index 18677cd4e62f..463e66258507 100644
> --- a/fs/configfs/dir.c
> +++ b/fs/configfs/dir.c
> @@ -580,6 +580,7 @@ static void detach_attrs(struct config_item * item)
>  static int populate_attrs(struct config_item *item)
>  {
>  	const struct config_item_type *t = item->ci_type;
> +	struct configfs_group_operations *ops;
>  	struct configfs_attribute *attr;
>  	struct configfs_bin_attribute *bin_attr;
>  	int error = 0;
> @@ -587,14 +588,33 @@ static int populate_attrs(struct config_item *item)
>  
>  	if (!t)
>  		return -EINVAL;
> +
> +	ops = t->ct_group_ops;
> +	if (!ops) {
> +		struct config_group *g = item->ci_group;
> +
> +		/*
> +		 * No item specific group operations, check if the item's group
> +		 * has group operations.
> +		 */
> +		if (g && g->cg_item.ci_type)
> +			ops = g->cg_item.ci_type->ct_group_ops;

Oh, I would not have expected to need to consider any alternate group
ops for attribute visibility beyond t->ct_group_ops. However in my RFC
example I made this mistake:

  static struct configfs_group_operations tsm_report_group_ops = {
         .make_item = tsm_report_make_item,
+        .is_visible = tsm_report_attr_visible,
+        .is_bin_visible = tsm_report_bin_attr_visible,
  };
 
Which in retrospect is the wrong level, and I suspect only reachable if
you do the the above awkward indirection ("ops =
g->cg_item.ci_type->ct_group_ops"). Instead, I was expecting symmetry
with sysfs where the object that carries ->attrs also carries
->is_visible, so something like this:

+ static struct configfs_group_operations tsm_report_attr_group_ops = {
+         .is_visible = tsm_report_attr_visible,
+         .is_bin_visible = tsm_report_bin_attr_visible,
+ };

  const struct config_item_type tsm_report_type = {
          .ct_owner = THIS_MODULE,
          .ct_bin_attrs = tsm_report_bin_attrs,
          .ct_attrs = tsm_report_attrs,
          .ct_item_ops = &tsm_report_item_ops,
+         .ct_group_ops = &tsm_report_attr_group_ops
  };
  EXPORT_SYMBOL_GPL(tsm_report_default_type);

..because is_visible() at the g->cg_item.ci_type->ct_group_ops level
would seem to mean parent directory visibility which is mismatched.

However as I stare at this a bit more it sinks in that configfs "group"
!= sysfs "group". So I am open to the suggestion that ci_item_ops is the
right place to house item attribute visibility callbacks, or even a new
"ci_attr_ops" expressly for this purpose. Either way my expectation is
that config_item_type can get to the visibilty callbacks for its
attributes without needing to traverse any other groups or items.