Re: [PATCH v3 1/1] dcache: Translating dentry into pathname without taking rename_lock

[Al Viro <viro@ZenIV.linux.org.uk>]

On Sat, Sep 07, 2013 at 04:01:10AM +0100, Al Viro wrote:
> 	* plain seqretry loop (d_lookup(), is_subdir(), autofs4_getpath(),
> ceph_misc_build_path(), [cifs] build_path_from_dentry(), nfs_path(),
      _mds_, actually - sorry.
> [audit] handle_path())
> 	* try seqretry once, then switch to write_seqlock() (the things
> that got unified into d_walk())
> 	* try seqretry three times, then switch to write_seqlock() (d_path()
> and friends)
> 	* several pure write_seqlock() users (d_move(), d_set_mounted(),
> d_materialize_unique())

BTW, autofs4_getpath() looks really odd:
static int autofs4_getpath(struct autofs_sb_info *sbi,
                           struct dentry *dentry, char **name)
and *name is never modified in it.  Why not simply pass it by value?
Moreover, I'm not sure I understand what do we need sbi->fs_lock in
there.  Other than that, it's very close to dentry_path() (well, that
and different calling conventions).  Ian?

ceph_mds_build_path() is similar, but it does kmalloc to store the result
and grabs ->d_lock for ->d_name protection.  This one, BTW, is much more
likely to get stalls - it ends up doing kmalloc on each attempt (after
having calculated the current length).  Bugger if I understand what's wrong
with simply grabbing a page and doing that once - before everything else...

build_path_from_dentry() - same story, might very well have been the source
of ceph_mds_build_path().

nfs_path() - not far from open-coded dentry_path() with some postprocessing,
uses ->d_lock for ->d_name protection.