[PATCH v6 00/15] Create and use APIs to centralise locking for directory ops

[Jeff Layton]

On Fri, 2025-11-14 at 15:23 +0100, Christian Brauner wrote:
> On Fri, Nov 14, 2025 at 01:24:41PM +0100, Christian Brauner wrote:
> > On Thu, Nov 13, 2025 at 11:18:23AM +1100, NeilBrown wrote:
> > > Following is a new version of this series:
> > >  - fixed a bug found by syzbot
> > >  - cleanup suggested by Stephen Smalley
> > >  - added patch for missing updates in smb/server - thanks Jeff Layton
> > 
> > The codeflow right now is very very gnarly in a lot of places which
> > obviously isn't your fault. But start_creating() and end_creating()
> > would very naturally lend themselves to be CLASS() guards.
> >
> > Unrelated: I'm very inclined to slap a patch on top that renames
> > start_creating()/end_creating() and start_dirop()/end_dirop() to
> > vfs_start_creating()/vfs_end_creating() and
> > vfs_start_dirop()/vfs_end_dirop(). After all they are VFS level
> > maintained helpers and I try to be consistent with the naming in the
> > codebase making it very easy to grep.
> 
> @Neil, @Jeff, could you please look at:
> https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git/log/?h=vfs.all
> 
> and specifically at the merge conflict resolution I did for:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git/commit/?h=vfs.all&id=f28c9935f78bffe6fee62f7fb9f6c5af7e30d9b2
> 
> and tell me whether it all looks sane?


I don't see any major issues. I'm kicking off a quick pynfs test run
now with it. One fairly minor nit:

@@ -212,15 +210,13 @@ nfsd4_create_clid_dir(struct nfs4_client *clp)
 		 * In the 4.0 case, we should never get here; but we may
 		 * as well be forgiving and just succeed silently.
 		 */
-		goto out_put;
-	dentry = vfs_mkdir(&nop_mnt_idmap, d_inode(dir), dentry, 0700, NULL);
+		goto out_end;
+	dentry = vfs_mkdir(&nop_mnt_idmap, d_inode(dir), dentry, S_IRWXU, NULL);
 	if (IS_ERR(dentry))
 		status = PTR_ERR(dentry);

I'm not sure if it was Neil's patch or your resolution that changed it,
but the change from 0700 to a symbolic constant is not preferred, IMO.
File modes are one of the few places where I think it's easier to
interpret (octal) numbers rather than symbolic constants.

-- 
Jeff Layton <jlayton at kernel.org>