Re: [PATCH v5 05/14] VFS/nfsd/cachefiles/ovl: introduce start_removing() and end_removing()

On Thu, 13 Nov 2025, Jeff Layton wrote:
> On Thu, 2025-11-06 at 11:50 +1100, NeilBrown wrote:
> > From: NeilBrown <[email protected]>
> > 
> > start_removing() is similar to start_creating() but will only return a
> > positive dentry with the expectation that it will be removed.  This is
> > used by nfsd, cachefiles, and overlayfs.  They are changed to also use
> > end_removing() to terminate the action begun by start_removing().  This
> > is a simple alias for end_dirop().
> > 
> > Apart from changes to the error paths, as we no longer need to unlock on
> > a lookup error, an effect on callers is that they don't need to test if
> > the found dentry is positive or negative - they can be sure it is
> > positive.
> > 
> > Reviewed-by: Amir Goldstein <[email protected]>
> > Signed-off-by: NeilBrown <[email protected]>
> > ---
> >  fs/cachefiles/namei.c    | 32 ++++++++++++++------------------
> >  fs/namei.c               | 27 +++++++++++++++++++++++++++
> >  fs/nfsd/nfs4recover.c    | 18 +++++-------------
> >  fs/nfsd/vfs.c            | 26 ++++++++++----------------
> >  fs/overlayfs/dir.c       | 15 +++++++--------
> >  fs/overlayfs/overlayfs.h |  8 ++++++++
> >  include/linux/namei.h    | 18 ++++++++++++++++++
> >  7 files changed, 89 insertions(+), 55 deletions(-)
> > 
> > diff --git a/fs/cachefiles/namei.c b/fs/cachefiles/namei.c
> > index 0a136eb434da..c7f0c6ab9b88 100644
> > --- a/fs/cachefiles/namei.c
> > +++ b/fs/cachefiles/namei.c
> > @@ -260,6 +260,7 @@ static int cachefiles_unlink(struct cachefiles_cache *cache,
> >   * - File backed objects are unlinked
> >   * - Directory backed objects are stuffed into the graveyard for userspace to
> >   *   delete
> > + * On entry dir must be locked.  It will be unlocked on exit.
> >   */
> >  int cachefiles_bury_object(struct cachefiles_cache *cache,
> >  			   struct cachefiles_object *object,
> > @@ -274,28 +275,30 @@ int cachefiles_bury_object(struct cachefiles_cache *cache,
> >  
> >  	_enter(",'%pd','%pd'", dir, rep);
> >  
> > +	/* end_removing() will dput() @rep but we need to keep
> > +	 * a ref, so take one now.  This also stops the dentry
> > +	 * being negated when unlinked which we need.
> > +	 */
> > +	dget(rep);
> > +
> >  	if (rep->d_parent != dir) {
> > -		inode_unlock(d_inode(dir));
> > +		end_removing(rep);
> >  		_leave(" = -ESTALE");
> >  		return -ESTALE;
> >  	}
> >  
> >  	/* non-directories can just be unlinked */
> >  	if (!d_is_dir(rep)) {
> > -		dget(rep); /* Stop the dentry being negated if it's only pinned
> > -			    * by a file struct.
> > -			    */
> >  		ret = cachefiles_unlink(cache, object, dir, rep, why);
> > -		dput(rep);
> > +		end_removing(rep);
> >  
> > -		inode_unlock(d_inode(dir));
> >  		_leave(" = %d", ret);
> >  		return ret;
> >  	}
> >  
> >  	/* directories have to be moved to the graveyard */
> >  	_debug("move stale object to graveyard");
> > -	inode_unlock(d_inode(dir));
> > +	end_removing(rep);
> >  
> >  try_again:
> >  	/* first step is to make up a grave dentry in the graveyard */
> > @@ -749,26 +752,20 @@ static struct dentry *cachefiles_lookup_for_cull(struct cachefiles_cache *cache,
> >  	struct dentry *victim;
> >  	int ret = -ENOENT;
> >  
> > -	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
> > +	victim = start_removing(&nop_mnt_idmap, dir, &QSTR(filename));
> >  
> > -	victim = lookup_one(&nop_mnt_idmap, &QSTR(filename), dir);
> >  	if (IS_ERR(victim))
> >  		goto lookup_error;
> > -	if (d_is_negative(victim))
> > -		goto lookup_put;
> >  	if (d_inode(victim)->i_flags & S_KERNEL_FILE)
> >  		goto lookup_busy;
> >  	return victim;
> >  
> >  lookup_busy:
> >  	ret = -EBUSY;
> > -lookup_put:
> > -	inode_unlock(d_inode(dir));
> > -	dput(victim);
> > +	end_removing(victim);
> >  	return ERR_PTR(ret);
> >  
> >  lookup_error:
> > -	inode_unlock(d_inode(dir));
> >  	ret = PTR_ERR(victim);
> >  	if (ret == -ENOENT)
> >  		return ERR_PTR(-ESTALE); /* Probably got retired by the netfs */
> > @@ -816,18 +813,17 @@ int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
> >  
> >  	ret = cachefiles_bury_object(cache, NULL, dir, victim,
> >  				     FSCACHE_OBJECT_WAS_CULLED);
> > +	dput(victim);
> >  	if (ret < 0)
> >  		goto error;
> >  
> >  	fscache_count_culled();
> > -	dput(victim);
> >  	_leave(" = 0");
> >  	return 0;
> >  
> >  error_unlock:
> > -	inode_unlock(d_inode(dir));
> > +	end_removing(victim);
> >  error:
> > -	dput(victim);
> >  	if (ret == -ENOENT)
> >  		return -ESTALE; /* Probably got retired by the netfs */
> >  
> > diff --git a/fs/namei.c b/fs/namei.c
> > index 8873ad0f05b0..38dda29552f6 100644
> > --- a/fs/namei.c
> > +++ b/fs/namei.c
> > @@ -3248,6 +3248,33 @@ struct dentry *start_creating(struct mnt_idmap *idmap, struct dentry *parent,
> >  }
> >  EXPORT_SYMBOL(start_creating);
> >  
> > +/**
> > + * start_removing - prepare to remove a given name with permission checking
> > + * @idmap:  idmap of the mount
> > + * @parent: directory in which to find the name
> > + * @name:   the name to be removed
> > + *
> > + * Locks are taken and a lookup in performed prior to removing
> > + * an object from a directory.  Permission checking (MAY_EXEC) is performed
> > + * against @idmap.
> > + *
> > + * If the name doesn't exist, an error is returned.
> > + *
> > + * end_removing() should be called when removal is complete, or aborted.
> > + *
> > + * Returns: a positive dentry, or an error.
> > + */
> > +struct dentry *start_removing(struct mnt_idmap *idmap, struct dentry *parent,
> > +			      struct qstr *name)
> > +{
> > +	int err = lookup_one_common(idmap, name, parent);
> > +
> > +	if (err)
> > +		return ERR_PTR(err);
> > +	return start_dirop(parent, name, 0);
> > +}
> > +EXPORT_SYMBOL(start_removing);
> > +
> >  #ifdef CONFIG_UNIX98_PTYS
> >  int path_pts(struct path *path)
> >  {
> > diff --git a/fs/nfsd/nfs4recover.c b/fs/nfsd/nfs4recover.c
> > index c247a7c3291c..3eefaa2202e3 100644
> > --- a/fs/nfsd/nfs4recover.c
> > +++ b/fs/nfsd/nfs4recover.c
> > @@ -324,20 +324,12 @@ nfsd4_unlink_clid_dir(char *name, struct nfsd_net *nn)
> >  	dprintk("NFSD: nfsd4_unlink_clid_dir. name %s\n", name);
> >  
> >  	dir = nn->rec_file->f_path.dentry;
> > -	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
> > -	dentry = lookup_one(&nop_mnt_idmap, &QSTR(name), dir);
> > -	if (IS_ERR(dentry)) {
> > -		status = PTR_ERR(dentry);
> > -		goto out_unlock;
> > -	}
> > -	status = -ENOENT;
> > -	if (d_really_is_negative(dentry))
> > -		goto out;
> > +	dentry = start_removing(&nop_mnt_idmap, dir, &QSTR(name));
> > +	if (IS_ERR(dentry))
> > +		return PTR_ERR(dentry);
> > +
> >  	status = vfs_rmdir(&nop_mnt_idmap, d_inode(dir), dentry);
> > -out:
> > -	dput(dentry);
> > -out_unlock:
> > -	inode_unlock(d_inode(dir));
> > +	end_removing(dentry);
> >  	return status;
> >  }
> >  
> > diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> > index 24e501abad0e..6291c371caa7 100644
> > --- a/fs/nfsd/vfs.c
> > +++ b/fs/nfsd/vfs.c
> > @@ -2044,7 +2044,7 @@ nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
> >  {
> >  	struct dentry	*dentry, *rdentry;
> >  	struct inode	*dirp;
> > -	struct inode	*rinode;
> > +	struct inode	*rinode = NULL;
> >  	__be32		err;
> >  	int		host_err;
> >  
> > @@ -2063,24 +2063,21 @@ nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
> >  
> >  	dentry = fhp->fh_dentry;
> >  	dirp = d_inode(dentry);
> > -	inode_lock_nested(dirp, I_MUTEX_PARENT);
> >  
> > -	rdentry = lookup_one(&nop_mnt_idmap, &QSTR_LEN(fname, flen), dentry);
> > +	rdentry = start_removing(&nop_mnt_idmap, dentry, &QSTR_LEN(fname, flen));
> > +
> >  	host_err = PTR_ERR(rdentry);
> >  	if (IS_ERR(rdentry))
> > -		goto out_unlock;
> > +		goto out_drop_write;
> >  
> > -	if (d_really_is_negative(rdentry)) {
> > -		dput(rdentry);
> > -		host_err = -ENOENT;
> > -		goto out_unlock;
> > -	}
> > -	rinode = d_inode(rdentry);
> >  	err = fh_fill_pre_attrs(fhp);
> >  	if (err != nfs_ok)
> >  		goto out_unlock;
> >  
> > +	rinode = d_inode(rdentry);
> > +	/* Prevent truncation until after locks dropped */
> >  	ihold(rinode);
> > +
> >  	if (!type)
> >  		type = d_inode(rdentry)->i_mode & S_IFMT;
> >  
> > @@ -2102,10 +2099,10 @@ nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
> >  	}
> >  	fh_fill_post_attrs(fhp);
> >  
> > -	inode_unlock(dirp);
> > -	if (!host_err)
> > +out_unlock:
> > +	end_removing(rdentry);
> > +	if (!err && !host_err)
> >  		host_err = commit_metadata(fhp);
> > -	dput(rdentry);
> >  	iput(rinode);    /* truncate the inode here */
> >  
> >  out_drop_write:
> > @@ -2123,9 +2120,6 @@ nfsd_unlink(struct svc_rqst *rqstp, struct svc_fh *fhp, int type,
> >  	}
> >  out:
> >  	return err != nfs_ok ? err : nfserrno(host_err);
> > -out_unlock:
> > -	inode_unlock(dirp);
> > -	goto out_drop_write;
> >  }
> >  
> >  /*
> > diff --git a/fs/overlayfs/dir.c b/fs/overlayfs/dir.c
> > index b9160fefbd00..20682afdbd20 100644
> > --- a/fs/overlayfs/dir.c
> > +++ b/fs/overlayfs/dir.c
> > @@ -866,17 +866,17 @@ static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
> >  			goto out;
> >  	}
> >  
> > -	inode_lock_nested(dir, I_MUTEX_PARENT);
> > -	upper = ovl_lookup_upper(ofs, dentry->d_name.name, upperdir,
> > -				 dentry->d_name.len);
> > +	upper = ovl_start_removing_upper(ofs, upperdir,
> > +					 &QSTR_LEN(dentry->d_name.name,
> > +						   dentry->d_name.len));
> >  	err = PTR_ERR(upper);
> >  	if (IS_ERR(upper))
> > -		goto out_unlock;
> > +		goto out_dput;
> >  
> >  	err = -ESTALE;
> >  	if ((opaquedir && upper != opaquedir) ||
> >  	    (!opaquedir && !ovl_matches_upper(dentry, upper)))
> > -		goto out_dput_upper;
> > +		goto out_unlock;
> >  
> >  	if (is_dir)
> >  		err = ovl_do_rmdir(ofs, dir, upper);
> > @@ -892,10 +892,9 @@ static int ovl_remove_upper(struct dentry *dentry, bool is_dir,
> >  	 */
> >  	if (!err)
> >  		d_drop(dentry);
> > -out_dput_upper:
> > -	dput(upper);
> >  out_unlock:
> > -	inode_unlock(dir);
> > +	end_removing(upper);
> > +out_dput:
> >  	dput(opaquedir);
> >  out:
> >  	return err;
> > diff --git a/fs/overlayfs/overlayfs.h b/fs/overlayfs/overlayfs.h
> > index beeba96cfcb2..49ad65f829dc 100644
> > --- a/fs/overlayfs/overlayfs.h
> > +++ b/fs/overlayfs/overlayfs.h
> > @@ -423,6 +423,14 @@ static inline struct dentry *ovl_start_creating_upper(struct ovl_fs *ofs,
> >  			      parent, name);
> >  }
> >  
> > +static inline struct dentry *ovl_start_removing_upper(struct ovl_fs *ofs,
> > +						      struct dentry *parent,
> > +						      struct qstr *name)
> > +{
> > +	return start_removing(ovl_upper_mnt_idmap(ofs),
> > +			      parent, name);
> > +}
> > +
> >  static inline bool ovl_open_flags_need_copy_up(int flags)
> >  {
> >  	if (!flags)
> > diff --git a/include/linux/namei.h b/include/linux/namei.h
> > index 37b72f4a64f0..6d1069f93ebf 100644
> > --- a/include/linux/namei.h
> > +++ b/include/linux/namei.h
> > @@ -91,6 +91,8 @@ struct dentry *lookup_one_positive_killable(struct mnt_idmap *idmap,
> >  
> >  struct dentry *start_creating(struct mnt_idmap *idmap, struct dentry *parent,
> >  			      struct qstr *name);
> > +struct dentry *start_removing(struct mnt_idmap *idmap, struct dentry *parent,
> > +			      struct qstr *name);
> >  
> >  /**
> >   * end_creating - finish action started with start_creating
> > @@ -122,6 +124,22 @@ static inline void end_creating(struct dentry *child, struct dentry *parent)
> >  		end_dirop(child);
> >  }
> >  
> > +/**
> > + * end_removing - finish action started with start_removing
> > + * @child:  dentry returned by start_removing()
> > + * @parent: dentry given to start_removing()
> > + *
> > + * Unlock and release the child.
> > + *
> > + * This is identical to end_dirop().  It can be passed the result of
> > + * start_removing() whether that was successful or not, but it not needed
> > + * if start_removing() failed.
> > + */
> > +static inline void end_removing(struct dentry *child)
> > +{
> > +	end_dirop(child);
> > +}
> > +
> >  extern int follow_down_one(struct path *);
> >  extern int follow_down(struct path *path, unsigned int flags);
> >  extern int follow_up(struct path *);
> 
> This looks fine to me (particularly the knfsd parts), but doesn't ksmbd
> need to be similarly converted?

ksmbd removes names in two places.

ksmbd_vfs_unlink() is changed to use start_removing_dentry() in the next
patch as you noticed when reviewing it.

ksmbd_vfs_remove_file() is called after ksmbd_vfs_kern_path_locked()
which is only used there.  I should possible renamed to
ksmbd_vfs_kern_path_start_removing() and then the "do_lock" flag to
__ksmbd_vfs_kern_path() and ksmbd_vfs_path_lookup() should become
"for_remove" and then the lock/lookup in the "if (do_lock)" branch
(which will become "if (for_remove)") should use start_removing.

I'll insert a patch to do this.

> 
> Reviewed-by: Jeff Layton <[email protected]>

Thanks,
NeilBrown