ceph: keep auth cap when inode has flocks or posix locks

file locks are tracked by inode's auth mds. dropping auth caps
is equivalent to releasing all file locks.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

4 files changed, 60 insertions, 9 deletions

diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index f2550a076edc..6301bf299b7c 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -493,6 +493,7 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
 	ci->i_wb_ref = 0;
 	ci->i_wrbuffer_ref = 0;
 	ci->i_wrbuffer_ref_head = 0;
+	atomic_set(&ci->i_filelock_ref, 0);
 	ci->i_shared_gen = 0;
 	ci->i_rdcache_gen = 0;
 	ci->i_rdcache_revoking = 0;
diff --git a/fs/ceph/locks.c b/fs/ceph/locks.c
index e7cce412f2cf..316d550b9603 100644
--- a/fs/ceph/locks.c
+++ b/fs/ceph/locks.c
@@ -30,19 +30,46 @@ void __init ceph_flock_init(void)
 	get_random_bytes(&lock_secret, sizeof(lock_secret));
 }
 
+static void ceph_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
+{
+	struct inode *inode = file_inode(src->fl_file);
+	atomic_inc(&ceph_inode(inode)->i_filelock_ref);
+}
+
+static void ceph_fl_release_lock(struct file_lock *fl)
+{
+	struct inode *inode = file_inode(fl->fl_file);
+	atomic_dec(&ceph_inode(inode)->i_filelock_ref);
+}
+
+static const struct file_lock_operations ceph_fl_lock_ops = {
+	.fl_copy_lock = ceph_fl_copy_lock,
+	.fl_release_private = ceph_fl_release_lock,
+};
+
 /**
  * Implement fcntl and flock locking functions.
  */
-static int ceph_lock_message(u8 lock_type, u16 operation, struct file *file,
+static int ceph_lock_message(u8 lock_type, u16 operation, struct inode *inode,
 			     int cmd, u8 wait, struct file_lock *fl)
 {
-	struct inode *inode = file_inode(file);
 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
 	struct ceph_mds_request *req;
 	int err;
 	u64 length = 0;
 	u64 owner;
 
+	if (operation == CEPH_MDS_OP_SETFILELOCK) {
+		/*
+		 * increasing i_filelock_ref closes race window between
+		 * handling request reply and adding file_lock struct to
+		 * inode. Otherwise, auth caps may get trimmed in the
+		 * window. Caller function will decrease the counter.
+		 */
+		fl->fl_ops = &ceph_fl_lock_ops;
+		atomic_inc(&ceph_inode(inode)->i_filelock_ref);
+	}
+
 	if (operation != CEPH_MDS_OP_SETFILELOCK || cmd == CEPH_LOCK_UNLOCK)
 		wait = 0;
 
@@ -180,10 +207,11 @@ static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
  */
 int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 {
-	u8 lock_cmd;
+	struct inode *inode = file_inode(file);
 	int err;
-	u8 wait = 0;
 	u16 op = CEPH_MDS_OP_SETFILELOCK;
+	u8 lock_cmd;
+	u8 wait = 0;
 
 	if (!(fl->fl_flags & FL_POSIX))
 		return -ENOLCK;
@@ -199,6 +227,17 @@ int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 	else if (IS_SETLKW(cmd))
 		wait = 1;
 
+	if (op == CEPH_MDS_OP_SETFILELOCK) {
+		/*
+		 * increasing i_filelock_ref closes race window between
+		 * handling request reply and adding file_lock struct to
+		 * inode. Otherwise, i_auth_cap may get trimmed in the
+		 * window. Caller function will decrease the counter.
+		 */
+		fl->fl_ops = &ceph_fl_lock_ops;
+		atomic_inc(&ceph_inode(inode)->i_filelock_ref);
+	}
+
 	if (F_RDLCK == fl->fl_type)
 		lock_cmd = CEPH_LOCK_SHARED;
 	else if (F_WRLCK == fl->fl_type)
@@ -206,7 +245,7 @@ int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 	else
 		lock_cmd = CEPH_LOCK_UNLOCK;
 
-	err = ceph_lock_message(CEPH_LOCK_FCNTL, op, file, lock_cmd, wait, fl);
+	err = ceph_lock_message(CEPH_LOCK_FCNTL, op, inode, lock_cmd, wait, fl);
 	if (!err) {
 		if (op != CEPH_MDS_OP_GETFILELOCK) {
 			dout("mds locked, locking locally");
@@ -215,7 +254,7 @@ int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 				/* undo! This should only happen if
 				 * the kernel detects local
 				 * deadlock. */
-				ceph_lock_message(CEPH_LOCK_FCNTL, op, file,
+				ceph_lock_message(CEPH_LOCK_FCNTL, op, inode,
 						  CEPH_LOCK_UNLOCK, 0, fl);
 				dout("got %d on posix_lock_file, undid lock",
 				     err);
@@ -227,8 +266,9 @@ int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
 
 int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 {
-	u8 lock_cmd;
+	struct inode *inode = file_inode(file);
 	int err;
+	u8 lock_cmd;
 	u8 wait = 0;
 
 	if (!(fl->fl_flags & FL_FLOCK))
@@ -239,6 +279,10 @@ int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 
 	dout("ceph_flock, fl_file: %p", fl->fl_file);
 
+	/* see comment in ceph_lock */
+	fl->fl_ops = &ceph_fl_lock_ops;
+	atomic_inc(&ceph_inode(inode)->i_filelock_ref);
+
 	if (IS_SETLKW(cmd))
 		wait = 1;
 
@@ -250,13 +294,13 @@ int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
 		lock_cmd = CEPH_LOCK_UNLOCK;
 
 	err = ceph_lock_message(CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK,
-				file, lock_cmd, wait, fl);
+				inode, lock_cmd, wait, fl);
 	if (!err) {
 		err = locks_lock_file_wait(file, fl);
 		if (err) {
 			ceph_lock_message(CEPH_LOCK_FLOCK,
 					  CEPH_MDS_OP_SETFILELOCK,
-					  file, CEPH_LOCK_UNLOCK, 0, fl);
+					  inode, CEPH_LOCK_UNLOCK, 0, fl);
 			dout("got %d on locks_lock_file_wait, undid lock", err);
 		}
 	}
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 0687ab3c3267..c8a811db387a 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -1462,6 +1462,11 @@ static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
 			goto out;
 		if ((used | wanted) & CEPH_CAP_ANY_WR)
 			goto out;
+		/* Note: it's possible that i_filelock_ref becomes non-zero
+		 * after dropping auth caps. It doesn't hurt because reply
+		 * of lock mds request will re-add auth caps. */
+		if (atomic_read(&ci->i_filelock_ref) > 0)
+			goto out;
 	}
 	/* The inode has cached pages, but it's no longer used.
 	 * we can safely drop it */
diff --git a/fs/ceph/super.h b/fs/ceph/super.h
index 3e27a28aa44a..100596c49353 100644
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -352,6 +352,7 @@ struct ceph_inode_info {
 	int i_pin_ref;
 	int i_rd_ref, i_rdcache_ref, i_wr_ref, i_wb_ref;
 	int i_wrbuffer_ref, i_wrbuffer_ref_head;
+	atomic_t i_filelock_ref;
 	u32 i_shared_gen;       /* increment each time we get FILE_SHARED */
 	u32 i_rdcache_gen;      /* incremented each time we get FILE_CACHE. */
 	u32 i_rdcache_revoking; /* RDCACHE gen to async invalidate, if any */