[PATCH 1/6] autofs4: allow RCU-walk to walk through autofs4.

From: NeilBrown
Date: Sun Aug 24 2014 - 18:44:58 EST


Any attempt to look up a pathname that passes though an
autofs4 mount is currently forced out of RCU-walk into
REF-walk.

This can significantly hurt performance of many-thread work
loads on many-core systems, especially if the automounted
filesystem supports RCU-walk but doesn't get to benefit from
it.

So if autofs4_d_manage is called with rcu_walk set, only
fail with -ECHILD if it is necessary to wait longer than
a spinlock.

Reviewed-by: Ian Kent <raven@xxxxxxxxxx>
Tested-by: Ian Kent <raven@xxxxxxxxxx>
Signed-off-by: NeilBrown <neilb@xxxxxxx>
---
fs/autofs4/autofs_i.h | 2 +-
fs/autofs4/dev-ioctl.c | 2 +-
fs/autofs4/expire.c | 4 +++-
fs/autofs4/root.c | 44 +++++++++++++++++++++++++++++---------------
4 files changed, 34 insertions(+), 18 deletions(-)

diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
index 9e359fb20c0a..2f1032f12d91 100644
--- a/fs/autofs4/autofs_i.h
+++ b/fs/autofs4/autofs_i.h
@@ -148,7 +148,7 @@ void autofs4_free_ino(struct autofs_info *);

/* Expiration */
int is_autofs4_dentry(struct dentry *);
-int autofs4_expire_wait(struct dentry *dentry);
+int autofs4_expire_wait(struct dentry *dentry, int rcu_walk);
int autofs4_expire_run(struct super_block *, struct vfsmount *,
struct autofs_sb_info *,
struct autofs_packet_expire __user *);
diff --git a/fs/autofs4/dev-ioctl.c b/fs/autofs4/dev-ioctl.c
index 5b570b6efa28..aaf96cb25452 100644
--- a/fs/autofs4/dev-ioctl.c
+++ b/fs/autofs4/dev-ioctl.c
@@ -450,7 +450,7 @@ static int autofs_dev_ioctl_requester(struct file *fp,
ino = autofs4_dentry_ino(path.dentry);
if (ino) {
err = 0;
- autofs4_expire_wait(path.dentry);
+ autofs4_expire_wait(path.dentry, 0);
spin_lock(&sbi->fs_lock);
param->requester.uid = from_kuid_munged(current_user_ns(), ino->uid);
param->requester.gid = from_kgid_munged(current_user_ns(), ino->gid);
diff --git a/fs/autofs4/expire.c b/fs/autofs4/expire.c
index a7be57e39be7..7e2f22ce6954 100644
--- a/fs/autofs4/expire.c
+++ b/fs/autofs4/expire.c
@@ -467,7 +467,7 @@ found:
return expired;
}

-int autofs4_expire_wait(struct dentry *dentry)
+int autofs4_expire_wait(struct dentry *dentry, int rcu_walk)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
@@ -477,6 +477,8 @@ int autofs4_expire_wait(struct dentry *dentry)
spin_lock(&sbi->fs_lock);
if (ino->flags & AUTOFS_INF_EXPIRING) {
spin_unlock(&sbi->fs_lock);
+ if (rcu_walk)
+ return -ECHILD;

DPRINTK("waiting for expire %p name=%.*s",
dentry, dentry->d_name.len, dentry->d_name.name);
diff --git a/fs/autofs4/root.c b/fs/autofs4/root.c
index cdb25ebccc4c..2296c8301b66 100644
--- a/fs/autofs4/root.c
+++ b/fs/autofs4/root.c
@@ -210,7 +210,8 @@ next:
return NULL;
}

-static struct dentry *autofs4_lookup_expiring(struct dentry *dentry)
+static struct dentry *autofs4_lookup_expiring(struct dentry *dentry,
+ bool rcu_walk)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct dentry *parent = dentry->d_parent;
@@ -229,6 +230,11 @@ static struct dentry *autofs4_lookup_expiring(struct dentry *dentry)
struct dentry *expiring;
struct qstr *qstr;

+ if (rcu_walk) {
+ spin_unlock(&sbi->lookup_lock);
+ return ERR_PTR(-ECHILD);
+ }
+
ino = list_entry(p, struct autofs_info, expiring);
expiring = ino->dentry;

@@ -264,13 +270,15 @@ next:
return NULL;
}

-static int autofs4_mount_wait(struct dentry *dentry)
+static int autofs4_mount_wait(struct dentry *dentry, bool rcu_walk)
{
struct autofs_sb_info *sbi = autofs4_sbi(dentry->d_sb);
struct autofs_info *ino = autofs4_dentry_ino(dentry);
int status = 0;

if (ino->flags & AUTOFS_INF_PENDING) {
+ if (rcu_walk)
+ return -ECHILD;
DPRINTK("waiting for mount name=%.*s",
dentry->d_name.len, dentry->d_name.name);
status = autofs4_wait(sbi, dentry, NFY_MOUNT);
@@ -280,20 +288,22 @@ static int autofs4_mount_wait(struct dentry *dentry)
return status;
}

-static int do_expire_wait(struct dentry *dentry)
+static int do_expire_wait(struct dentry *dentry, bool rcu_walk)
{
struct dentry *expiring;

- expiring = autofs4_lookup_expiring(dentry);
+ expiring = autofs4_lookup_expiring(dentry, rcu_walk);
+ if (IS_ERR(expiring))
+ return PTR_ERR(expiring);
if (!expiring)
- return autofs4_expire_wait(dentry);
+ return autofs4_expire_wait(dentry, rcu_walk);
else {
/*
* If we are racing with expire the request might not
* be quite complete, but the directory has been removed
* so it must have been successful, just wait for it.
*/
- autofs4_expire_wait(expiring);
+ autofs4_expire_wait(expiring, 0);
autofs4_del_expiring(expiring);
dput(expiring);
}
@@ -345,7 +355,7 @@ static struct vfsmount *autofs4_d_automount(struct path *path)
* and the directory was removed, so just go ahead and try
* the mount.
*/
- status = do_expire_wait(dentry);
+ status = do_expire_wait(dentry, 0);
if (status && status != -EAGAIN)
return NULL;

@@ -353,7 +363,7 @@ static struct vfsmount *autofs4_d_automount(struct path *path)
spin_lock(&sbi->fs_lock);
if (ino->flags & AUTOFS_INF_PENDING) {
spin_unlock(&sbi->fs_lock);
- status = autofs4_mount_wait(dentry);
+ status = autofs4_mount_wait(dentry, 0);
if (status)
return ERR_PTR(status);
goto done;
@@ -394,7 +404,7 @@ static struct vfsmount *autofs4_d_automount(struct path *path)
}
ino->flags |= AUTOFS_INF_PENDING;
spin_unlock(&sbi->fs_lock);
- status = autofs4_mount_wait(dentry);
+ status = autofs4_mount_wait(dentry, 0);
spin_lock(&sbi->fs_lock);
ino->flags &= ~AUTOFS_INF_PENDING;
if (status) {
@@ -430,21 +440,25 @@ static int autofs4_d_manage(struct dentry *dentry, bool rcu_walk)
return 0;
}

- /* We need to sleep, so we need pathwalk to be in ref-mode */
- if (rcu_walk)
- return -ECHILD;
-
/* Wait for pending expires */
- do_expire_wait(dentry);
+ if (do_expire_wait(dentry, rcu_walk) == -ECHILD)
+ return -ECHILD;

/*
* This dentry may be under construction so wait on mount
* completion.
*/
- status = autofs4_mount_wait(dentry);
+ status = autofs4_mount_wait(dentry, rcu_walk);
if (status)
return status;

+ if (rcu_walk)
+ /* it is always safe to return 0 as the worst that
+ * will happen is we retry in REF-walk mode.
+ * Better than always taking a lock.
+ */
+ return 0;
+
spin_lock(&sbi->fs_lock);
/*
* If the dentry has been selected for expire while we slept


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/