On Tue, Apr 1, 2025 at 1:14 PM Alexey Dobriyan <adobriyan@xxxxxxxxx> wrote:
>
> > +void proc_make_permanent(struct proc_dir_entry *de)
> > +{
> > + pde_make_permanent(de);
> > +}
> > +EXPORT_SYMBOL(proc_make_permanent);
>
> no, no, no, no
>
> this is wrong!
>
> marking should be done in the context of a module!
>
> the reason it is not exported is because the aren't safeguards against
> module misuse
>
> the flag is supposed to be used in case where
> a) PDE itself is never removed and,
> b) all the code supporting is never removed,
> so that locking can be skipped
>
> this it fine to mark /proc/filesystems because kernel controls it
>
> this is fine to mark /proc/aaa if all module does is to write some
> info to it and deletes it during rmmod
>
> but it is not fine to mark /proc/aaa/bbb if "bbb" is created/deleted
> while module is running,
> locking _must_ be done in this case
Well I'm unhappy to begin with, but did not want to do anything
churn-inducing. The above looks like a minimal solution to me.
The pde_ marking things are in an internal header and I did not want
to move them around.
If anything I'm surprised there is no mechanism to get this done (I
assumed there would be a passable flags argument, but got nothing).
What I need here is that /proc/filesystems thing sorted out, as in this call:
> proc_create_single("filesystems", 0, NULL, filesystems_proc_show);
Would you be ok with adding proc_create_single_permanent() which hides
the logic and is not exported to modules?
--
Mateusz Guzik <mjguzik gmail.com>
