The recent change from using d_hash_and_lookup() to using
try_lookup_noperm() inadvertently introduce a d_revalidate() call when
the lookup was successful.  Steven French reports that this resulted in
worse than halving of performance in some cases.

Prior to the offending patch the only caller of try_lookup_noperm() was
autofs which does not need the d_revalidate().  So it is safe to remove
the d_revalidate() call providing we stop using try_lookup_noperm() to
implement lookup_noperm().

The "try_" in the name is strongly suggestive that the caller isn't
expecting much effort, so it seems reasonable to avoid the effort of
d_revalidate().

Fixes: 06c567403ae5 ("Use try_lookup_noperm() instead of d_hash_and_lookup() outside of VFS")
Reported-by: Steve French <smfrench@gmail.com>
Link: https://lore.kernel.org/all/CAH2r5mu5SfBrdc2CFHwzft8=n9koPMk+Jzwpy-oUMx-wCRCesQ@mail.gmail.com/
Signed-off-by: NeilBrown <neil@brown.name>
---
 fs/namei.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/fs/namei.c b/fs/namei.c
index 4bb889fc980b..f761cafaeaad 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -2917,7 +2917,8 @@ static int lookup_one_common(struct mnt_idmap *idmap,
  * @base:	base directory to lookup from
  *
  * Look up a dentry by name in the dcache, returning NULL if it does not
- * currently exist.  The function does not try to create a dentry.
+ * currently exist.  The function does not try to create a dentry and if one
+ * is found it doesn't try to revalidate it.
  *
  * Note that this routine is purely a helper for filesystem usage and should
  * not be called by generic code.  It does no permission checking.
@@ -2933,7 +2934,7 @@ struct dentry *try_lookup_noperm(struct qstr *name, struct dentry *base)
 	if (err)
 		return ERR_PTR(err);
 
-	return lookup_dcache(name, base, 0);
+	return d_lookup(base, name);
 }
 EXPORT_SYMBOL(try_lookup_noperm);
 
@@ -3057,14 +3058,22 @@ EXPORT_SYMBOL(lookup_one_positive_unlocked);
  * Note that this routine is purely a helper for filesystem usage and should
  * not be called by generic code. It does no permission checking.
  *
- * Unlike lookup_noperm, it should be called without the parent
+ * Unlike lookup_noperm(), it should be called without the parent
  * i_rwsem held, and will take the i_rwsem itself if necessary.
+ *
+ * Unlike try_lookup_noperm() it *does* revalidate the dentry if it already
+ * existed.
  */
 struct dentry *lookup_noperm_unlocked(struct qstr *name, struct dentry *base)
 {
 	struct dentry *ret;
+	int err;
 
-	ret = try_lookup_noperm(name, base);
+	err = lookup_noperm_common(name, base);
+	if (err)
+		return ERR_PTR(err);
+
+	ret = lookup_dcache(name, base, 0);
 	if (!ret)
 		ret = lookup_slow(name, base, 0);
 	return ret;
-- 
2.49.0

On Tue, 10 Jun 2025 11:04:04 +1000, NeilBrown wrote:
> The recent change from using d_hash_and_lookup() to using
> try_lookup_noperm() inadvertently introduce a d_revalidate() call when
> the lookup was successful.  Steven French reports that this resulted in
> worse than halving of performance in some cases.
> 
> Prior to the offending patch the only caller of try_lookup_noperm() was
> autofs which does not need the d_revalidate().  So it is safe to remove
> the d_revalidate() call providing we stop using try_lookup_noperm() to
> implement lookup_noperm().
> 
> [...]

Applied to the vfs.fixes branch of the vfs/vfs.git tree.
Patches in the vfs.fixes branch should appear in linux-next soon.

Please report any outstanding bugs that were missed during review in a
new review to the original patch series allowing us to drop it.

It's encouraged to provide Acked-bys and Reviewed-bys even though the
patch has now been applied. If possible patch trailers will be updated.

Note that commit hashes shown below are subject to change due to rebase,
trailer updates or similar. If in doubt, please check the listed branch.

tree:   https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git
branch: vfs.fixes

[1/1] VFS: change try_lookup_noperm() to skip revalidation
      https://git.kernel.org/vfs/vfs/c/ad5a0351064c

Tested-by: Steve French <stfrench@microsoft.com>

I verified that it fixed the performance regression in generic/676
(see e.g. a full test run this morning with the patch on 6.16-rc1
http://smb311-linux-testing.southcentralus.cloudapp.azure.com/#/builders/3/builds/497),
the test took 10:48 vs. 23 to 30 minutes without the patch.

I also saw similar performance yesterday with 6.16-rc1 with reverting
the patch ("Use try_lookup_noperm() instead of d_hash_and_lookup()
outside of VFS"
) For that run test generic/676 took 9:32.

On Mon, Jun 9, 2025 at 8:04 PM NeilBrown <neil@brown.name> wrote:
>
>
> The recent change from using d_hash_and_lookup() to using
> try_lookup_noperm() inadvertently introduce a d_revalidate() call when
> the lookup was successful.  Steven French reports that this resulted in
> worse than halving of performance in some cases.
>
> Prior to the offending patch the only caller of try_lookup_noperm() was
> autofs which does not need the d_revalidate().  So it is safe to remove
> the d_revalidate() call providing we stop using try_lookup_noperm() to
> implement lookup_noperm().
>
> The "try_" in the name is strongly suggestive that the caller isn't
> expecting much effort, so it seems reasonable to avoid the effort of
> d_revalidate().
>
> Fixes: 06c567403ae5 ("Use try_lookup_noperm() instead of d_hash_and_lookup() outside of VFS")
> Reported-by: Steve French <smfrench@gmail.com>
> Link: https://lore.kernel.org/all/CAH2r5mu5SfBrdc2CFHwzft8=n9koPMk+Jzwpy-oUMx-wCRCesQ@mail.gmail.com/
> Signed-off-by: NeilBrown <neil@brown.name>
> ---
>  fs/namei.c | 17 +++++++++++++----
>  1 file changed, 13 insertions(+), 4 deletions(-)
>
> diff --git a/fs/namei.c b/fs/namei.c
> index 4bb889fc980b..f761cafaeaad 100644
> --- a/fs/namei.c
> +++ b/fs/namei.c
> @@ -2917,7 +2917,8 @@ static int lookup_one_common(struct mnt_idmap *idmap,
>   * @base:      base directory to lookup from
>   *
>   * Look up a dentry by name in the dcache, returning NULL if it does not
> - * currently exist.  The function does not try to create a dentry.
> + * currently exist.  The function does not try to create a dentry and if one
> + * is found it doesn't try to revalidate it.
>   *
>   * Note that this routine is purely a helper for filesystem usage and should
>   * not be called by generic code.  It does no permission checking.
> @@ -2933,7 +2934,7 @@ struct dentry *try_lookup_noperm(struct qstr *name, struct dentry *base)
>         if (err)
>                 return ERR_PTR(err);
>
> -       return lookup_dcache(name, base, 0);
> +       return d_lookup(base, name);
>  }
>  EXPORT_SYMBOL(try_lookup_noperm);
>
> @@ -3057,14 +3058,22 @@ EXPORT_SYMBOL(lookup_one_positive_unlocked);
>   * Note that this routine is purely a helper for filesystem usage and should
>   * not be called by generic code. It does no permission checking.
>   *
> - * Unlike lookup_noperm, it should be called without the parent
> + * Unlike lookup_noperm(), it should be called without the parent
>   * i_rwsem held, and will take the i_rwsem itself if necessary.
> + *
> + * Unlike try_lookup_noperm() it *does* revalidate the dentry if it already
> + * existed.
>   */
>  struct dentry *lookup_noperm_unlocked(struct qstr *name, struct dentry *base)
>  {
>         struct dentry *ret;
> +       int err;
>
> -       ret = try_lookup_noperm(name, base);
> +       err = lookup_noperm_common(name, base);
> +       if (err)
> +               return ERR_PTR(err);
> +
> +       ret = lookup_dcache(name, base, 0);
>         if (!ret)
>                 ret = lookup_slow(name, base, 0);
>         return ret;
> --
> 2.49.0
>


-- 
Thanks,

Steve