diff options
| author |   NeilBrown <neilb@suse.de> | 2014-10-13 15:52:22 -0700 |
|---|---|---|
| committer |   Linus Torvalds <torvalds@linux-foundation.org> | 2014-10-14 02:18:16 +0200 |
| commit | ef16cc5909c907586f4eba5455e183fd319637c8 (patch) | |
| tree | 8c85bca785379b19f215b210a89717e7cd51c4c8 | |
| parent | autofs4: avoid taking fs_lock during rcu-walk (diff) | |
| download | linux-dev-ef16cc5909c907586f4eba5455e183fd319637c8.tar.xz linux-dev-ef16cc5909c907586f4eba5455e183fd319637c8.zip | |
autofs4: d_manage() should return -EISDIR when appropriate in rcu-walk mode.
If rcu-walk mode we don't *have* to return -EISDIR for non-mount-traps
as we will simply drop into REF-walk and handling DCACHE_NEED_AUTOMOUNT
dentrys the slow way. But it is better if we do when possible.
In 'oz_mode', use the same condition as ref-walk: if not a mountpoint,
then it must be -EISDIR.
In regular mode there are most tests needed. Most of them can be
performed without taking any spinlocks. If we find a directory that
isn't obviously empty, and isn't mounted on, we need to call
'simple_empty()' which does take a spinlock. If this turned out to hurt
performance, some other approach could be found to signal when a
directory is known to be empty.
Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Ian Kent <raven@themaw.net>
Tested-by: Ian Kent <raven@themaw.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
| -rw-r--r-- | fs/autofs4/root.c | 26 |
1 files changed, 20 insertions, 6 deletions
diff --git a/fs/autofs4/root.c b/fs/autofs4/root.c index 2296c8301b66..d76d083f2f06 100644 --- a/fs/autofs4/root.c +++ b/fs/autofs4/root.c @@ -433,8 +433,6 @@ static int autofs4_d_manage(struct dentry *dentry, bool rcu_walk) /* The daemon never waits. */ if (autofs4_oz_mode(sbi)) { - if (rcu_walk) - return 0; if (!d_mountpoint(dentry)) return -EISDIR; return 0; @@ -452,12 +450,28 @@ static int autofs4_d_manage(struct dentry *dentry, bool 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. + if (rcu_walk) { + /* We don't need fs_lock in rcu_walk mode, + * just testing 'AUTOFS_INFO_NO_RCU' is enough. + * simple_empty() takes a spinlock, so leave it + * to last. + * We only return -EISDIR when certain this isn't + * a mount-trap. */ + struct inode *inode; + if (ino->flags & (AUTOFS_INF_EXPIRING | AUTOFS_INF_NO_RCU)) + return 0; + if (d_mountpoint(dentry)) + return 0; + inode = ACCESS_ONCE(dentry->d_inode); + if (inode && S_ISLNK(inode->i_mode)) + return -EISDIR; + if (list_empty(&dentry->d_subdirs)) + return 0; + if (!simple_empty(dentry)) + return -EISDIR; return 0; + } spin_lock(&sbi->fs_lock); /* |