Re: [PATCH v10 8/9] proc: use human-readable values for hidehid

[Alexey Gladkov]

On Thu, Apr 02, 2020 at 11:05:21AM -0500, Eric W. Biederman wrote:
> Alexey Gladkov <gladkov.alexey@xxxxxxxxx> writes:
> 
> > The hidepid parameter values are becoming more and more and it becomes
> > difficult to remember what each new magic number means.
> 
> In principle I like this change.  In practice I think you have just
> broken ABI compatiblity with the new mount ABI.
> 
> In particular the following line seems broken.
> 
> > diff --git a/fs/proc/root.c b/fs/proc/root.c
> > index dbcd96f07c7a..ba782d6e6197 100644
> > --- a/fs/proc/root.c
> > +++ b/fs/proc/root.c
> > @@ -45,7 +45,7 @@ enum proc_param {
> >  
> >  static const struct fs_parameter_spec proc_fs_parameters[] = {
> >  	fsparam_u32("gid",	Opt_gid),
> > -	fsparam_u32("hidepid",	Opt_hidepid),
> > +	fsparam_string("hidepid",	Opt_hidepid),
> >  	fsparam_string("subset",	Opt_subset),
> >  	{}
> >  };
> 
> As I read fs_parser.c fs_param_is_u32 handles string inputs and turns them
> into numbers, and it handles binary numbers.

Yes, you can use: fsconfig(fsfd, FSCONFIG_SET_BINARY, ...); but in this
case the type of parameter will be fs_value_is_blob [1]. This kind of
parameters is handled by fs_param_is_blob(). The fs_param_is_u32 can
handle only parametes with fs_value_is_string type [2].

Am I missing something?

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/fs/fsopen.c#n405
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/fs/fs_parser.c#n215

> However fs_param_is_string
> appears to only handle strings.  It appears to have not capacity to turn
> raw binary numbers into strings.
> 
> So I think we probably need to fix fs_param_is_string to raw binary
> numbers before we can safely make this change to fs/proc/root.c
> 
> David am I reading the fs_parser.c code correctly?  If I am are you ok
> with a change like the above?
> 
> Eric
> 

-- 
Rgrds, legion