[RFC,1/1] fs/namespace: defer free_mount from namespace_unlock

Commit Message

Eric Chanudet Jan. 19, 2023, 8:55 p.m. UTC

From: Alexander Larsson <[e-mail]>

Use call_rcu to defer releasing the umount'ed or detached filesystem
when calling namepsace_unlock().

Calling synchronize_rcu_expedited() has a significant cost on RT kernel
that default to rcupdate.rcu_normal_after_boot=1.

For example, on a 6.2-rt1 kernel:
perf stat -r 10 --null --pre 'mount -t tmpfs tmpfs mnt' -- umount mnt
           0.07464 +- 0.00396 seconds time elapsed  ( +-  5.31% )

With this change applied:
perf stat -r 10 --null --pre 'mount -t tmpfs tmpfs mnt' -- umount mnt
        0.00162604 +- 0.00000637 seconds time elapsed  ( +-  0.39% )

Waiting for the grace period before completing the syscall does not seem
mandatory. The struct mount umount'ed are queued up for release in a
separate list and no longer accessible to following syscalls.

Signed-off-by: Alexander Larsson <[e-mail]>
Signed-off-by: Eric Chanudet <[e-mail]>
---
 fs/namespace.c | 42 +++++++++++++++++++++++++++++++++++-------
 1 file changed, 35 insertions(+), 7 deletions(-)

Comments

Al Viro Jan. 19, 2023, 10:03 p.m. UTC | #1

On Thu, Jan 19, 2023 at 03:55:21PM -0500, Eric Chanudet wrote:
> From: Alexander Larsson <[e-mail]>
> 
> Use call_rcu to defer releasing the umount'ed or detached filesystem
> when calling namepsace_unlock().
> 
> Calling synchronize_rcu_expedited() has a significant cost on RT kernel
> that default to rcupdate.rcu_normal_after_boot=1.
> 
> For example, on a 6.2-rt1 kernel:
> perf stat -r 10 --null --pre 'mount -t tmpfs tmpfs mnt' -- umount mnt
>            0.07464 +- 0.00396 seconds time elapsed  ( +-  5.31% )
> 
> With this change applied:
> perf stat -r 10 --null --pre 'mount -t tmpfs tmpfs mnt' -- umount mnt
>         0.00162604 +- 0.00000637 seconds time elapsed  ( +-  0.39% )
> 
> Waiting for the grace period before completing the syscall does not seem
> mandatory. The struct mount umount'ed are queued up for release in a
> separate list and no longer accessible to following syscalls.

You *really* do not want to have umount(2) return without having
the filesystems shut down.

Patch

diff --git a/fs/namespace.c b/fs/namespace.c
index ab467ee58341..11d219a6e83c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -44,6 +44,11 @@  static unsigned int m_hash_shift __read_mostly;
 static unsigned int mp_hash_mask __read_mostly;
 static unsigned int mp_hash_shift __read_mostly;
 
+struct mount_delayed_release {
+	struct rcu_head rcu;
+	struct hlist_head release_list;
+};
+
 static __initdata unsigned long mhash_entries;
 static int __init set_mhash_entries(char *str)
 {
@@ -1582,11 +1587,31 @@  int may_umount(struct vfsmount *mnt)
 
 EXPORT_SYMBOL(may_umount);
 
-static void namespace_unlock(void)
+static void free_mounts(struct hlist_head *mount_list)
 {
-	struct hlist_head head;
 	struct hlist_node *p;
 	struct mount *m;
+
+	hlist_for_each_entry_safe(m, p, mount_list, mnt_umount) {
+		hlist_del(&m->mnt_umount);
+		mntput(&m->mnt);
+	}
+}
+
+static void delayed_mount_release(struct rcu_head *head)
+{
+	struct mount_delayed_release *drelease =
+		container_of(head, struct mount_delayed_release, rcu);
+
+	free_mounts(&drelease->release_list);
+	kfree(drelease);
+}
+
+static void namespace_unlock(void)
+{
+	struct hlist_head head;
+	struct mount_delayed_release *drelease;
+
 	LIST_HEAD(list);
 
 	hlist_move_list(&unmounted, &head);
@@ -1599,12 +1624,15 @@  static void namespace_unlock(void)
 	if (likely(hlist_empty(&head)))
 		return;
 
-	synchronize_rcu_expedited();
-
-	hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
-		hlist_del(&m->mnt_umount);
-		mntput(&m->mnt);
+	drelease = kmalloc(sizeof(*drelease), GFP_KERNEL);
+	if (unlikely(!drelease)) {
+		synchronize_rcu_expedited();
+		free_mounts(&head);
+		return;
 	}
+
+	hlist_move_list(&head, &drelease->release_list);
+	call_rcu(&drelease->rcu, delayed_mount_release);
 }
 
 static inline void namespace_lock(void)