Re: [PATCH] mq_open() honor leading slash

[Jakub Jelinek <jakub@redhat.com>]

On Thu, Apr 15, 2004 at 11:39:51AM -0700, Chris Wright wrote:
> SUSv3 requires the following on mq_open() with name containing leading
> slash:
> 
> 	If name begins with the slash character, then processes
> 	calling mq_open() with the same value of name shall refer to
> 	the same message queue object, as long as that name has not
> 	been removed. If name does not begin with the slash character,
> 	the effect is implementation-defined.  The interpretation of
> 	slash characters other than the leading slash character in name
> 	is implementation-defined.
> 
> The use of lookup_one_len() to force all lookups relative to the
> mqueue_mnt root guarantees absolute rather than relative lookup without
> leading slash, and generates an error on a name that contains any slashes
> at all.  This is inconsitent with the part of the spec that requires
> leading slash to be effectively an absolute path (unless you consider
> -EACCES being "the same message queue object" ;-)
> 
> Patch below simply eats all leading slashes before passing name to
> lookup_one_len() in mq_open() and mq_unlink().

glibc already strips the leading slash in userland.
If you want to do it in the kernel instead, it shouldn't IMHO be silent if it
doesn't see a leading slash or sees more than one. I.e.
	error = -EINVAL;
	if (name[0] != '/')
		goto out_err;
...
	dentry = lookup_one_len(name + 1, mqueue_mnt->mnt_root, strlen(name + 1));

	Jakub