fsnotify: Fix busy inodes during unmount

Detaching of mark connector from fsnotify_put_mark() can race with
unmounting of the filesystem like:

  CPU1				CPU2
fsnotify_put_mark()
  spin_lock(&conn->lock);
  ...
  inode = fsnotify_detach_connector_from_object(conn)
  spin_unlock(&conn->lock);
				generic_shutdown_super()
				  fsnotify_unmount_inodes()
				    sees connector detached for inode
				      -> nothing to do
				  evict_inode()
				    barfs on pending inode reference
  iput(inode);

Resulting in "Busy inodes after unmount" message and possible kernel
oops. Make fsnotify_unmount_inodes() properly wait for outstanding inode
references from detached connectors.

Note that the accounting of outstanding inode references in the
superblock can cause some cacheline contention on the counter. OTOH it
happens only during deletion of the last notification mark from an inode
(or during unlinking of watched inode) and that is not too bad. I have
measured time to create & delete inotify watch 100000 times from 64
processes in parallel (each process having its own inotify group and its
own file on a shared superblock) on a 64 CPU machine. Average and
standard deviation of 15 runs look like:

	Avg		Stddev
Vanilla	9.817400	0.276165
Fixed	9.710467	0.228294

So there's no statistically significant difference.

Fixes: 6b3f05d24d35 ("fsnotify: Detach mark from object list when last reference is dropped")
CC: stable@vger.kernel.org
Signed-off-by: Jan Kara <jack@suse.cz>

diff --git a/fs/notify/fsnotify.c b/fs/notify/fsnotify.c
index 87597550440904646d311f4aa0cb7aa3aefcc8ab..2172ba516c61d536f5f05045e0a47d7dae30cfc1 100644 (file)
--- a/fs/notify/fsnotify.c
+++ b/fs/notify/fsnotify.c
@@ -96,6 +96,9 @@ static void fsnotify_unmount_inodes(struct super_block *sb)
 
        if (iput_inode)
                iput(iput_inode);
+       /* Wait for outstanding inode references from connectors */
+       wait_var_event(&sb->s_fsnotify_inode_refs,
+                      !atomic_long_read(&sb->s_fsnotify_inode_refs));
 }
 
 void fsnotify_sb_delete(struct super_block *sb)

diff --git a/fs/notify/mark.c b/fs/notify/mark.c
index b5172ccb2e60f08fde84e958374a33850a71426a..d2dd16cb5989144d4289cc879a4a34d2af81c216 100644 (file)
--- a/fs/notify/mark.c
+++ b/fs/notify/mark.c
@@ -181,17 +181,20 @@ static void fsnotify_connector_destroy_workfn(struct work_struct *work)
        }
 }
 
-static struct inode *fsnotify_detach_connector_from_object(
-                                       struct fsnotify_mark_connector *conn)
+static void *fsnotify_detach_connector_from_object(
+                                       struct fsnotify_mark_connector *conn,
+                                       unsigned int *type)
 {
        struct inode *inode = NULL;
 
+       *type = conn->type;
        if (conn->type == FSNOTIFY_OBJ_TYPE_DETACHED)
                return NULL;
 
        if (conn->type == FSNOTIFY_OBJ_TYPE_INODE) {
                inode = fsnotify_conn_inode(conn);
                inode->i_fsnotify_mask = 0;
+               atomic_long_inc(&inode->i_sb->s_fsnotify_inode_refs);
        } else if (conn->type == FSNOTIFY_OBJ_TYPE_VFSMOUNT) {
                fsnotify_conn_mount(conn)->mnt_fsnotify_mask = 0;
        } else if (conn->type == FSNOTIFY_OBJ_TYPE_SB) {
@@ -215,10 +218,29 @@ static void fsnotify_final_mark_destroy(struct fsnotify_mark *mark)
        fsnotify_put_group(group);
 }
 
+/* Drop object reference originally held by a connector */
+static void fsnotify_drop_object(unsigned int type, void *objp)
+{
+       struct inode *inode;
+       struct super_block *sb;
+
+       if (!objp)
+               return;
+       /* Currently only inode references are passed to be dropped */
+       if (WARN_ON_ONCE(type != FSNOTIFY_OBJ_TYPE_INODE))
+               return;
+       inode = objp;
+       sb = inode->i_sb;
+       iput(inode);
+       if (atomic_long_dec_and_test(&sb->s_fsnotify_inode_refs))
+               wake_up_var(&sb->s_fsnotify_inode_refs);
+}
+
 void fsnotify_put_mark(struct fsnotify_mark *mark)
 {
        struct fsnotify_mark_connector *conn;
-       struct inode *inode = NULL;
+       void *objp = NULL;
+       unsigned int type = FSNOTIFY_OBJ_TYPE_DETACHED;
        bool free_conn = false;
 
        /* Catch marks that were actually never attached to object */
@@ -238,7 +260,7 @@ void fsnotify_put_mark(struct fsnotify_mark *mark)
        conn = mark->connector;
        hlist_del_init_rcu(&mark->obj_list);
        if (hlist_empty(&conn->list)) {
-               inode = fsnotify_detach_connector_from_object(conn);
+               objp = fsnotify_detach_connector_from_object(conn, &type);
                free_conn = true;
        } else {
                __fsnotify_recalc_mask(conn);
@@ -246,7 +268,7 @@ void fsnotify_put_mark(struct fsnotify_mark *mark)
        mark->connector = NULL;
        spin_unlock(&conn->lock);
 
-       iput(inode);
+       fsnotify_drop_object(type, objp);
 
        if (free_conn) {
                spin_lock(&destroy_lock);
@@ -713,7 +735,8 @@ void fsnotify_destroy_marks(fsnotify_connp_t *connp)
 {
        struct fsnotify_mark_connector *conn;
        struct fsnotify_mark *mark, *old_mark = NULL;
-       struct inode *inode;
+       void *objp;
+       unsigned int type;
 
        conn = fsnotify_grab_connector(connp);
        if (!conn)
@@ -739,11 +762,11 @@ void fsnotify_destroy_marks(fsnotify_connp_t *connp)
         * mark references get dropped. It would lead to strange results such
         * as delaying inode deletion or blocking unmount.
         */
-       inode = fsnotify_detach_connector_from_object(conn);
+       objp = fsnotify_detach_connector_from_object(conn, &type);
        spin_unlock(&conn->lock);
        if (old_mark)
                fsnotify_put_mark(old_mark);
-       iput(inode);
+       fsnotify_drop_object(type, objp);
 }
 
 /*

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 6da94deb957ff7314f92285b4e1dad59deee45e2..00b23b21e78af61425e2ccfac6e402670fb0e174 100644 (file)
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1437,6 +1437,9 @@ struct super_block {
        /* Number of inodes with nlink == 0 but still referenced */
        atomic_long_t s_remove_count;
 
+       /* Pending fsnotify inode refs */
+       atomic_long_t s_fsnotify_inode_refs;
+
        /* Being remounted read-only */
        int s_readonly_remount;