Merge tag 'afs-next-20190507' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs

Pull AFS updates from David Howells:
 "A set of fix and development patches for AFS for 5.2.

  Summary:

   - Fix the AFS file locking so that sqlite can run on an AFS mount and
     also so that firefox and gnome can use a homedir that's mounted
     through AFS.

     This required emulation of fine-grained locking when the server
     will only support whole-file locks and no upgrade/downgrade. Four
     modes are provided, settable by mount parameter:

       "flock=local"   - No reference to the server

       "flock=openafs" - Fine-grained locks are local-only, whole-file
                         locks require sufficient server locks

       "flock=strict"  - All locks require sufficient server locks

       "flock=write"   - Always get an exclusive server lock

     If the volume is a read-only or backup volume, then flock=local for
     that volume.

   - Log extra information for a couple of cases where the client mucks
     up somehow: AFS vnode with undefined type and dir check failure -
     in both cases we seem to end up with unfilled data, but the issues
     happen infrequently and are difficult to reproduce at will.

   - Implement silly rename for unlink() and rename().

   - Set i_blocks so that du can get some information about usage.

   - Fix xattr handlers to return the right amount of data and to not
     overflow buffers.

   - Implement getting/setting raw AFS and YFS ACLs as xattrs"

* tag 'afs-next-20190507' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
  afs: Implement YFS ACL setting
  afs: Get YFS ACLs and information through xattrs
  afs: implement acl setting
  afs: Get an AFS3 ACL as an xattr
  afs: Fix getting the afs.fid xattr
  afs: Fix the afs.cell and afs.volume xattr handlers
  afs: Calculate i_blocks based on file size
  afs: Log more information for "kAFS: AFS vnode with undefined type\n"
  afs: Provide mount-time configurable byte-range file locking emulation
  afs: Add more tracepoints
  afs: Implement sillyrename for unlink and rename
  afs: Add directory reload tracepoint
  afs: Handle lock rpc ops failing on a file that got deleted
  afs: Improve dir check failure reports
  afs: Add file locking tracepoints
  afs: Further fix file locking
  afs: Fix AFS file locking to allow fine grained locks
  afs: Calculate lock extend timer from set/extend reply reception
  afs: Split wait from afs_make_call()

1 files changed, 390 insertions, 226 deletions

diff --git a/fs/afs/flock.c b/fs/afs/flock.c
index 6a0174258382..adc88eff7849 100644
--- a/fs/afs/flock.c
+++ b/fs/afs/flock.c
@@ -13,9 +13,11 @@
 
 #define AFS_LOCK_GRANTED	0
 #define AFS_LOCK_PENDING	1
+#define AFS_LOCK_YOUR_TRY	2
 
 struct workqueue_struct *afs_lock_manager;
 
+static void afs_next_locker(struct afs_vnode *vnode, int error);
 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl);
 static void afs_fl_release_private(struct file_lock *fl);
 
@@ -24,6 +26,14 @@ static const struct file_lock_operations afs_lock_ops = {
 	.fl_release_private	= afs_fl_release_private,
 };
 
+static inline void afs_set_lock_state(struct afs_vnode *vnode, enum afs_lock_state state)
+{
+	_debug("STATE %u -> %u", vnode->lock_state, state);
+	vnode->lock_state = state;
+}
+
+static atomic_t afs_file_lock_debug_id;
+
 /*
  * if the callback is broken on this vnode, then the lock may now be available
  */
@@ -31,7 +41,14 @@ void afs_lock_may_be_available(struct afs_vnode *vnode)
 {
 	_enter("{%llx:%llu}", vnode->fid.vid, vnode->fid.vnode);
 
-	queue_delayed_work(afs_lock_manager, &vnode->lock_work, 0);
+	if (vnode->lock_state != AFS_VNODE_LOCK_WAITING_FOR_CB)
+		return;
+
+	spin_lock(&vnode->lock);
+	if (vnode->lock_state == AFS_VNODE_LOCK_WAITING_FOR_CB)
+		afs_next_locker(vnode, 0);
+	trace_afs_flock_ev(vnode, NULL, afs_flock_callback_break, 0);
+	spin_unlock(&vnode->lock);
 }
 
 /*
@@ -40,8 +57,35 @@ void afs_lock_may_be_available(struct afs_vnode *vnode)
  */
 static void afs_schedule_lock_extension(struct afs_vnode *vnode)
 {
-	queue_delayed_work(afs_lock_manager, &vnode->lock_work,
-			   AFS_LOCKWAIT * HZ / 2);
+	ktime_t expires_at, now, duration;
+	u64 duration_j;
+
+	expires_at = ktime_add_ms(vnode->locked_at, AFS_LOCKWAIT * 1000 / 2);
+	now = ktime_get_real();
+	duration = ktime_sub(expires_at, now);
+	if (duration <= 0)
+		duration_j = 0;
+	else
+		duration_j = nsecs_to_jiffies(ktime_to_ns(duration));
+
+	queue_delayed_work(afs_lock_manager, &vnode->lock_work, duration_j);
+}
+
+/*
+ * In the case of successful completion of a lock operation, record the time
+ * the reply appeared and start the lock extension timer.
+ */
+void afs_lock_op_done(struct afs_call *call)
+{
+	struct afs_vnode *vnode = call->reply[0];
+
+	if (call->error == 0) {
+		spin_lock(&vnode->lock);
+		trace_afs_flock_ev(vnode, NULL, afs_flock_timestamp, 0);
+		vnode->locked_at = call->reply_time;
+		afs_schedule_lock_extension(vnode);
+		spin_unlock(&vnode->lock);
+	}
 }
 
 /*
@@ -49,22 +93,90 @@ static void afs_schedule_lock_extension(struct afs_vnode *vnode)
  * first lock in the queue is itself a readlock)
  * - the caller must hold the vnode lock
  */
-static void afs_grant_locks(struct afs_vnode *vnode, struct file_lock *fl)
+static void afs_grant_locks(struct afs_vnode *vnode)
 {
 	struct file_lock *p, *_p;
+	bool exclusive = (vnode->lock_type == AFS_LOCK_WRITE);
 
-	list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
-	if (fl->fl_type == F_RDLCK) {
-		list_for_each_entry_safe(p, _p, &vnode->pending_locks,
-					 fl_u.afs.link) {
-			if (p->fl_type == F_RDLCK) {
-				p->fl_u.afs.state = AFS_LOCK_GRANTED;
-				list_move_tail(&p->fl_u.afs.link,
-					       &vnode->granted_locks);
-				wake_up(&p->fl_wait);
-			}
+	list_for_each_entry_safe(p, _p, &vnode->pending_locks, fl_u.afs.link) {
+		if (!exclusive && p->fl_type == F_WRLCK)
+			continue;
+
+		list_move_tail(&p->fl_u.afs.link, &vnode->granted_locks);
+		p->fl_u.afs.state = AFS_LOCK_GRANTED;
+		trace_afs_flock_op(vnode, p, afs_flock_op_grant);
+		wake_up(&p->fl_wait);
+	}
+}
+
+/*
+ * If an error is specified, reject every pending lock that matches the
+ * authentication and type of the lock we failed to get.  If there are any
+ * remaining lockers, try to wake up one of them to have a go.
+ */
+static void afs_next_locker(struct afs_vnode *vnode, int error)
+{
+	struct file_lock *p, *_p, *next = NULL;
+	struct key *key = vnode->lock_key;
+	unsigned int fl_type = F_RDLCK;
+
+	_enter("");
+
+	if (vnode->lock_type == AFS_LOCK_WRITE)
+		fl_type = F_WRLCK;
+
+	list_for_each_entry_safe(p, _p, &vnode->pending_locks, fl_u.afs.link) {
+		if (error &&
+		    p->fl_type == fl_type &&
+		    afs_file_key(p->fl_file) == key) {
+			list_del_init(&p->fl_u.afs.link);
+			p->fl_u.afs.state = error;
+			wake_up(&p->fl_wait);
 		}
+
+		/* Select the next locker to hand off to. */
+		if (next &&
+		    (next->fl_type == F_WRLCK || p->fl_type == F_RDLCK))
+			continue;
+		next = p;
 	}
+
+	vnode->lock_key = NULL;
+	key_put(key);
+
+	if (next) {
+		afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
+		next->fl_u.afs.state = AFS_LOCK_YOUR_TRY;
+		trace_afs_flock_op(vnode, next, afs_flock_op_wake);
+		wake_up(&next->fl_wait);
+	} else {
+		afs_set_lock_state(vnode, AFS_VNODE_LOCK_NONE);
+		trace_afs_flock_ev(vnode, NULL, afs_flock_no_lockers, 0);
+	}
+
+	_leave("");
+}
+
+/*
+ * Kill off all waiters in the the pending lock queue due to the vnode being
+ * deleted.
+ */
+static void afs_kill_lockers_enoent(struct afs_vnode *vnode)
+{
+	struct file_lock *p;
+
+	afs_set_lock_state(vnode, AFS_VNODE_LOCK_DELETED);
+
+	while (!list_empty(&vnode->pending_locks)) {
+		p = list_entry(vnode->pending_locks.next,
+			       struct file_lock, fl_u.afs.link);
+		list_del_init(&p->fl_u.afs.link);
+		p->fl_u.afs.state = -ENOENT;
+		wake_up(&p->fl_wait);
+	}
+
+	key_put(vnode->lock_key);
+	vnode->lock_key = NULL;
 }
 
 /*
@@ -170,8 +282,6 @@ void afs_lock_work(struct work_struct *work)
 {
 	struct afs_vnode *vnode =
 		container_of(work, struct afs_vnode, lock_work.work);
-	struct file_lock *fl, *next;
-	afs_lock_type_t type;
 	struct key *key;
 	int ret;
 
@@ -183,35 +293,28 @@ again:
 	_debug("wstate %u for %p", vnode->lock_state, vnode);
 	switch (vnode->lock_state) {
 	case AFS_VNODE_LOCK_NEED_UNLOCK:
-		_debug("unlock");
-		vnode->lock_state = AFS_VNODE_LOCK_UNLOCKING;
+		afs_set_lock_state(vnode, AFS_VNODE_LOCK_UNLOCKING);
+		trace_afs_flock_ev(vnode, NULL, afs_flock_work_unlocking, 0);
 		spin_unlock(&vnode->lock);
 
 		/* attempt to release the server lock; if it fails, we just
 		 * wait 5 minutes and it'll expire anyway */
 		ret = afs_release_lock(vnode, vnode->lock_key);
-		if (ret < 0)
+		if (ret < 0 && vnode->lock_state != AFS_VNODE_LOCK_DELETED) {
+			trace_afs_flock_ev(vnode, NULL, afs_flock_release_fail,
+					   ret);
 			printk(KERN_WARNING "AFS:"
 			       " Failed to release lock on {%llx:%llx} error %d\n",
 			       vnode->fid.vid, vnode->fid.vnode, ret);
-
-		spin_lock(&vnode->lock);
-		key_put(vnode->lock_key);
-		vnode->lock_key = NULL;
-		vnode->lock_state = AFS_VNODE_LOCK_NONE;
-
-		if (list_empty(&vnode->pending_locks)) {
-			spin_unlock(&vnode->lock);
-			return;
 		}
 
-		/* The new front of the queue now owns the state variables. */
-		next = list_entry(vnode->pending_locks.next,
-				  struct file_lock, fl_u.afs.link);
-		vnode->lock_key = key_get(afs_file_key(next->fl_file));
-		vnode->lock_type = (next->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE;
-		vnode->lock_state = AFS_VNODE_LOCK_WAITING_FOR_CB;
-		goto again;
+		spin_lock(&vnode->lock);
+		if (ret == -ENOENT)
+			afs_kill_lockers_enoent(vnode);
+		else
+			afs_next_locker(vnode, 0);
+		spin_unlock(&vnode->lock);
+		return;
 
 	/* If we've already got a lock, then it must be time to extend that
 	 * lock as AFS locks time out after 5 minutes.
@@ -222,86 +325,55 @@ again:
 		ASSERT(!list_empty(&vnode->granted_locks));
 
 		key = key_get(vnode->lock_key);
-		vnode->lock_state = AFS_VNODE_LOCK_EXTENDING;
+		afs_set_lock_state(vnode, AFS_VNODE_LOCK_EXTENDING);
+		trace_afs_flock_ev(vnode, NULL, afs_flock_work_extending, 0);
 		spin_unlock(&vnode->lock);
 
 		ret = afs_extend_lock(vnode, key); /* RPC */
 		key_put(key);
 
-		if (ret < 0)
+		if (ret < 0) {
+			trace_afs_flock_ev(vnode, NULL, afs_flock_extend_fail,
+					   ret);
 			pr_warning("AFS: Failed to extend lock on {%llx:%llx} error %d\n",
 				   vnode->fid.vid, vnode->fid.vnode, ret);
+		}
 
 		spin_lock(&vnode->lock);
 
+		if (ret == -ENOENT) {
+			afs_kill_lockers_enoent(vnode);
+			spin_unlock(&vnode->lock);
+			return;
+		}
+
 		if (vnode->lock_state != AFS_VNODE_LOCK_EXTENDING)
 			goto again;
-		vnode->lock_state = AFS_VNODE_LOCK_GRANTED;
+		afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED);
 
-		if (ret == 0)
-			afs_schedule_lock_extension(vnode);
-		else
+		if (ret != 0)
 			queue_delayed_work(afs_lock_manager, &vnode->lock_work,
 					   HZ * 10);
 		spin_unlock(&vnode->lock);
 		_leave(" [ext]");
 		return;
 
-		/* If we don't have a granted lock, then we must've been called
-		 * back by the server, and so if might be possible to get a
-		 * lock we're currently waiting for.
-		 */
+	/* If we're waiting for a callback to indicate lock release, we can't
+	 * actually rely on this, so need to recheck at regular intervals.  The
+	 * problem is that the server might not notify us if the lock just
+	 * expires (say because a client died) rather than being explicitly
+	 * released.
+	 */
 	case AFS_VNODE_LOCK_WAITING_FOR_CB:
-		_debug("get");
-
-		key = key_get(vnode->lock_key);
-		type = vnode->lock_type;
-		vnode->lock_state = AFS_VNODE_LOCK_SETTING;
+		_debug("retry");
+		afs_next_locker(vnode, 0);
 		spin_unlock(&vnode->lock);
+		return;
 
-		ret = afs_set_lock(vnode, key, type); /* RPC */
-		key_put(key);
-
-		spin_lock(&vnode->lock);
-		switch (ret) {
-		case -EWOULDBLOCK:
-			_debug("blocked");
-			break;
-		case 0:
-			_debug("acquired");
-			vnode->lock_state = AFS_VNODE_LOCK_GRANTED;
-			/* Fall through */
-		default:
-			/* Pass the lock or the error onto the first locker in
-			 * the list - if they're looking for this type of lock.
-			 * If they're not, we assume that whoever asked for it
-			 * took a signal.
-			 */
-			if (list_empty(&vnode->pending_locks)) {
-				_debug("withdrawn");
-				vnode->lock_state = AFS_VNODE_LOCK_NEED_UNLOCK;
-				goto again;
-			}
-
-			fl = list_entry(vnode->pending_locks.next,
-					struct file_lock, fl_u.afs.link);
-			type = (fl->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE;
-			if (vnode->lock_type != type) {
-				_debug("changed");
-				vnode->lock_state = AFS_VNODE_LOCK_NEED_UNLOCK;
-				goto again;
-			}
-
-			fl->fl_u.afs.state = ret;
-			if (ret == 0)
-				afs_grant_locks(vnode, fl);
-			else
-				list_del_init(&fl->fl_u.afs.link);
-			wake_up(&fl->fl_wait);
-			spin_unlock(&vnode->lock);
-			_leave(" [granted]");
-			return;
-		}
+	case AFS_VNODE_LOCK_DELETED:
+		afs_kill_lockers_enoent(vnode);
+		spin_unlock(&vnode->lock);
+		return;
 
 		/* Fall through */
 	default:
@@ -320,14 +392,16 @@ again:
  */
 static void afs_defer_unlock(struct afs_vnode *vnode)
 {
-	_enter("");
+	_enter("%u", vnode->lock_state);
 
-	if (vnode->lock_state == AFS_VNODE_LOCK_GRANTED ||
-	    vnode->lock_state == AFS_VNODE_LOCK_EXTENDING) {
+	if (list_empty(&vnode->granted_locks) &&
+	    (vnode->lock_state == AFS_VNODE_LOCK_GRANTED ||
+	     vnode->lock_state == AFS_VNODE_LOCK_EXTENDING)) {
 		cancel_delayed_work(&vnode->lock_work);
 
-		vnode->lock_state = AFS_VNODE_LOCK_NEED_UNLOCK;
-		afs_lock_may_be_available(vnode);
+		afs_set_lock_state(vnode, AFS_VNODE_LOCK_NEED_UNLOCK);
+		trace_afs_flock_ev(vnode, NULL, afs_flock_defer_unlock, 0);
+		queue_delayed_work(afs_lock_manager, &vnode->lock_work, 0);
 	}
 }
 
@@ -336,7 +410,7 @@ static void afs_defer_unlock(struct afs_vnode *vnode)
  * whether we think that we have a locking permit.
  */
 static int afs_do_setlk_check(struct afs_vnode *vnode, struct key *key,
-			      afs_lock_type_t type, bool can_sleep)
+			      enum afs_flock_mode mode, afs_lock_type_t type)
 {
 	afs_access_t access;
 	int ret;
@@ -364,160 +438,177 @@ static int afs_do_setlk_check(struct afs_vnode *vnode, struct key *key,
 	if (type == AFS_LOCK_READ) {
 		if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE | AFS_ACE_LOCK)))
 			return -EACCES;
-		if (vnode->status.lock_count == -1 && !can_sleep)
-			return -EAGAIN; /* Write locked */
 	} else {
 		if (!(access & (AFS_ACE_INSERT | AFS_ACE_WRITE)))
 			return -EACCES;
-		if (vnode->status.lock_count != 0 && !can_sleep)
-			return -EAGAIN; /* Locked */
 	}
 
 	return 0;
 }
 
 /*
- * Remove the front runner from the pending queue.
- * - The caller must hold vnode->lock.
- */
-static void afs_dequeue_lock(struct afs_vnode *vnode, struct file_lock *fl)
-{
-	struct file_lock *next;
-
-	_enter("");
-
-	/* ->lock_type, ->lock_key and ->lock_state only belong to this
-	 * file_lock if we're at the front of the pending queue or if we have
-	 * the lock granted or if the lock_state is NEED_UNLOCK or UNLOCKING.
-	 */
-	if (vnode->granted_locks.next == &fl->fl_u.afs.link &&
-	    vnode->granted_locks.prev == &fl->fl_u.afs.link) {
-		list_del_init(&fl->fl_u.afs.link);
-		afs_defer_unlock(vnode);
-		return;
-	}
-
-	if (!list_empty(&vnode->granted_locks) ||
-	    vnode->pending_locks.next != &fl->fl_u.afs.link) {
-		list_del_init(&fl->fl_u.afs.link);
-		return;
-	}
-
-	list_del_init(&fl->fl_u.afs.link);
-	key_put(vnode->lock_key);
-	vnode->lock_key = NULL;
-	vnode->lock_state = AFS_VNODE_LOCK_NONE;
-
-	if (list_empty(&vnode->pending_locks))
-		return;
-
-	/* The new front of the queue now owns the state variables. */
-	next = list_entry(vnode->pending_locks.next,
-			  struct file_lock, fl_u.afs.link);
-	vnode->lock_key = key_get(afs_file_key(next->fl_file));
-	vnode->lock_type = (next->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE;
-	vnode->lock_state = AFS_VNODE_LOCK_WAITING_FOR_CB;
-	afs_lock_may_be_available(vnode);
-}
-
-/*
  * request a lock on a file on the server
  */
 static int afs_do_setlk(struct file *file, struct file_lock *fl)
 {
 	struct inode *inode = locks_inode(file);
 	struct afs_vnode *vnode = AFS_FS_I(inode);
+	enum afs_flock_mode mode = AFS_FS_S(inode->i_sb)->flock_mode;
 	afs_lock_type_t type;
 	struct key *key = afs_file_key(file);
+	bool partial, no_server_lock = false;
 	int ret;
 
-	_enter("{%llx:%llu},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_type);
+	if (mode == afs_flock_mode_unset)
+		mode = afs_flock_mode_openafs;
 
-	/* only whole-file locks are supported */
-	if (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX)
-		return -EINVAL;
+	_enter("{%llx:%llu},%llu-%llu,%u,%u",
+	       vnode->fid.vid, vnode->fid.vnode,
+	       fl->fl_start, fl->fl_end, fl->fl_type, mode);
 
 	fl->fl_ops = &afs_lock_ops;
 	INIT_LIST_HEAD(&fl->fl_u.afs.link);
 	fl->fl_u.afs.state = AFS_LOCK_PENDING;
 
+	partial = (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX);
 	type = (fl->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE;
+	if (mode == afs_flock_mode_write && partial)
+		type = AFS_LOCK_WRITE;
 
-	ret = afs_do_setlk_check(vnode, key, type, fl->fl_flags & FL_SLEEP);
+	ret = afs_do_setlk_check(vnode, key, mode, type);
 	if (ret < 0)
 		return ret;
 
-	spin_lock(&vnode->lock);
+	trace_afs_flock_op(vnode, fl, afs_flock_op_set_lock);
 
-	/* If we've already got a readlock on the server then we instantly
-	 * grant another readlock, irrespective of whether there are any
-	 * pending writelocks.
+	/* AFS3 protocol only supports full-file locks and doesn't provide any
+	 * method of upgrade/downgrade, so we need to emulate for partial-file
+	 * locks.
+	 *
+	 * The OpenAFS client only gets a server lock for a full-file lock and
+	 * keeps partial-file locks local.  Allow this behaviour to be emulated
+	 * (as the default).
 	 */
-	if (type == AFS_LOCK_READ &&
-	    vnode->lock_state == AFS_VNODE_LOCK_GRANTED &&
-	    vnode->lock_type == AFS_LOCK_READ) {
-		_debug("instant readlock");
-		ASSERT(!list_empty(&vnode->granted_locks));
-		goto share_existing_lock;
+	if (mode == afs_flock_mode_local ||
+	    (partial && mode == afs_flock_mode_openafs)) {
+		no_server_lock = true;
+		goto skip_server_lock;
 	}
 
+	spin_lock(&vnode->lock);
 	list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks);
 
+	ret = -ENOENT;
+	if (vnode->lock_state == AFS_VNODE_LOCK_DELETED)
+		goto error_unlock;
+
+	/* If we've already got a lock on the server then try to move to having
+	 * the VFS grant the requested lock.  Note that this means that other
+	 * clients may get starved out.
+	 */
+	_debug("try %u", vnode->lock_state);
+	if (vnode->lock_state == AFS_VNODE_LOCK_GRANTED) {
+		if (type == AFS_LOCK_READ) {
+			_debug("instant readlock");
+			list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
+			fl->fl_u.afs.state = AFS_LOCK_GRANTED;
+			goto vnode_is_locked_u;
+		}
+
+		if (vnode->lock_type == AFS_LOCK_WRITE) {
+			_debug("instant writelock");
+			list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
+			fl->fl_u.afs.state = AFS_LOCK_GRANTED;
+			goto vnode_is_locked_u;
+		}
+	}
+
+	if (vnode->lock_state == AFS_VNODE_LOCK_NONE &&
+	    !(fl->fl_flags & FL_SLEEP)) {
+		ret = -EAGAIN;
+		if (type == AFS_LOCK_READ) {
+			if (vnode->status.lock_count == -1)
+				goto lock_is_contended; /* Write locked */
+		} else {
+			if (vnode->status.lock_count != 0)
+				goto lock_is_contended; /* Locked */
+		}
+	}
+
 	if (vnode->lock_state != AFS_VNODE_LOCK_NONE)
 		goto need_to_wait;
 
+try_to_lock:
 	/* We don't have a lock on this vnode and we aren't currently waiting
 	 * for one either, so ask the server for a lock.
 	 *
 	 * Note that we need to be careful if we get interrupted by a signal
 	 * after dispatching the request as we may still get the lock, even
 	 * though we don't wait for the reply (it's not too bad a problem - the
-	 * lock will expire in 10 mins anyway).
+	 * lock will expire in 5 mins anyway).
 	 */
-	_debug("not locked");
+	trace_afs_flock_ev(vnode, fl, afs_flock_try_to_lock, 0);
 	vnode->lock_key = key_get(key);
 	vnode->lock_type = type;
-	vnode->lock_state = AFS_VNODE_LOCK_SETTING;
+	afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
 	spin_unlock(&vnode->lock);
 
 	ret = afs_set_lock(vnode, key, type); /* RPC */
 
 	spin_lock(&vnode->lock);
 	switch (ret) {
+	case -EKEYREJECTED:
+	case -EKEYEXPIRED:
+	case -EKEYREVOKED:
+	case -EPERM:
+	case -EACCES:
+		fl->fl_u.afs.state = ret;
+		trace_afs_flock_ev(vnode, fl, afs_flock_fail_perm, ret);
+		list_del_init(&fl->fl_u.afs.link);
+		afs_next_locker(vnode, ret);
+		goto error_unlock;
+
+	case -ENOENT:
+		fl->fl_u.afs.state = ret;
+		trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret);
+		list_del_init(&fl->fl_u.afs.link);
+		afs_kill_lockers_enoent(vnode);
+		goto error_unlock;
+
 	default:
-		goto abort_attempt;
+		fl->fl_u.afs.state = ret;
+		trace_afs_flock_ev(vnode, fl, afs_flock_fail_other, ret);
+		list_del_init(&fl->fl_u.afs.link);
+		afs_next_locker(vnode, 0);
+		goto error_unlock;
 
 	case -EWOULDBLOCK:
 		/* The server doesn't have a lock-waiting queue, so the client
 		 * will have to retry.  The server will break the outstanding
 		 * callbacks on a file when a lock is released.
 		 */
-		_debug("would block");
 		ASSERT(list_empty(&vnode->granted_locks));
 		ASSERTCMP(vnode->pending_locks.next, ==, &fl->fl_u.afs.link);
-		vnode->lock_state = AFS_VNODE_LOCK_WAITING_FOR_CB;
-		goto need_to_wait;
+		goto lock_is_contended;
 
 	case 0:
-		_debug("acquired");
-		break;
+		afs_set_lock_state(vnode, AFS_VNODE_LOCK_GRANTED);
+		trace_afs_flock_ev(vnode, fl, afs_flock_acquired, type);
+		afs_grant_locks(vnode);
+		goto vnode_is_locked_u;
 	}
 
-	/* we've acquired a server lock, but it needs to be renewed after 5
-	 * mins */
-	vnode->lock_state = AFS_VNODE_LOCK_GRANTED;
-	afs_schedule_lock_extension(vnode);
-
-share_existing_lock:
-	/* the lock has been granted as far as we're concerned... */
-	fl->fl_u.afs.state = AFS_LOCK_GRANTED;
-	list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks);
-
-given_lock:
-	/* ... but we do still need to get the VFS's blessing */
+vnode_is_locked_u:
 	spin_unlock(&vnode->lock);
-
-	ret = posix_lock_file(file, fl, NULL);
+vnode_is_locked:
+	/* the lock has been granted by the server... */
+	ASSERTCMP(fl->fl_u.afs.state, ==, AFS_LOCK_GRANTED);
+
+skip_server_lock:
+	/* ... but the VFS still needs to distribute access on this client. */
+	trace_afs_flock_ev(vnode, fl, afs_flock_vfs_locking, 0);
+	ret = locks_lock_file_wait(file, fl);
+	trace_afs_flock_ev(vnode, fl, afs_flock_vfs_lock, ret);
 	if (ret < 0)
 		goto vfs_rejected_lock;
 
@@ -529,38 +620,62 @@ given_lock:
 	_leave(" = 0");
 	return 0;
 
+lock_is_contended:
+	if (!(fl->fl_flags & FL_SLEEP)) {
+		list_del_init(&fl->fl_u.afs.link);
+		afs_next_locker(vnode, 0);
+		ret = -EAGAIN;
+		goto error_unlock;
+	}
+
+	afs_set_lock_state(vnode, AFS_VNODE_LOCK_WAITING_FOR_CB);
+	trace_afs_flock_ev(vnode, fl, afs_flock_would_block, ret);
+	queue_delayed_work(afs_lock_manager, &vnode->lock_work, HZ * 5);
+
 need_to_wait:
 	/* We're going to have to wait.  Either this client doesn't have a lock
 	 * on the server yet and we need to wait for a callback to occur, or
-	 * the client does have a lock on the server, but it belongs to some
-	 * other process(es) and is incompatible with the lock we want.
+	 * the client does have a lock on the server, but it's shared and we
+	 * need an exclusive lock.
 	 */
-	ret = -EAGAIN;
-	if (fl->fl_flags & FL_SLEEP) {
-		spin_unlock(&vnode->lock);
+	spin_unlock(&vnode->lock);
 
-		_debug("sleep");
-		ret = wait_event_interruptible(fl->fl_wait,
-					       fl->fl_u.afs.state != AFS_LOCK_PENDING);
+	trace_afs_flock_ev(vnode, fl, afs_flock_waiting, 0);
+	ret = wait_event_interruptible(fl->fl_wait,
+				       fl->fl_u.afs.state != AFS_LOCK_PENDING);
+	trace_afs_flock_ev(vnode, fl, afs_flock_waited, ret);
 
+	if (fl->fl_u.afs.state >= 0 && fl->fl_u.afs.state != AFS_LOCK_GRANTED) {
 		spin_lock(&vnode->lock);
-	}
 
-	if (fl->fl_u.afs.state == AFS_LOCK_GRANTED)
-		goto given_lock;
-	if (fl->fl_u.afs.state < 0)
-		ret = fl->fl_u.afs.state;
+		switch (fl->fl_u.afs.state) {
+		case AFS_LOCK_YOUR_TRY:
+			fl->fl_u.afs.state = AFS_LOCK_PENDING;
+			goto try_to_lock;
+		case AFS_LOCK_PENDING:
+			if (ret > 0) {
+				/* We need to retry the lock.  We may not be
+				 * notified by the server if it just expired
+				 * rather than being released.
+				 */
+				ASSERTCMP(vnode->lock_state, ==, AFS_VNODE_LOCK_WAITING_FOR_CB);
+				afs_set_lock_state(vnode, AFS_VNODE_LOCK_SETTING);
+				fl->fl_u.afs.state = AFS_LOCK_PENDING;
+				goto try_to_lock;
+			}
+			goto error_unlock;
+		case AFS_LOCK_GRANTED:
+		default:
+			break;
+		}
 
-abort_attempt:
-	/* we aren't going to get the lock, either because we're unwilling to
-	 * wait, or because some signal happened */
-	_debug("abort");
-	afs_dequeue_lock(vnode, fl);
+		spin_unlock(&vnode->lock);
+	}
 
-error_unlock:
-	spin_unlock(&vnode->lock);
-	_leave(" = %d", ret);
-	return ret;
+	if (fl->fl_u.afs.state == AFS_LOCK_GRANTED)
+		goto vnode_is_locked;
+	ret = fl->fl_u.afs.state;
+	goto error;
 
 vfs_rejected_lock:
 	/* The VFS rejected the lock we just obtained, so we have to discard
@@ -568,11 +683,17 @@ vfs_rejected_lock:
 	 * deal with.
 	 */
 	_debug("vfs refused %d", ret);
+	if (no_server_lock)
+		goto error;
 	spin_lock(&vnode->lock);
 	list_del_init(&fl->fl_u.afs.link);
-	if (list_empty(&vnode->granted_locks))
-		afs_defer_unlock(vnode);
-	goto error_unlock;
+	afs_defer_unlock(vnode);
+
+error_unlock:
+	spin_unlock(&vnode->lock);
+error:
+	_leave(" = %d", ret);
+	return ret;
 }
 
 /*
@@ -585,14 +706,12 @@ static int afs_do_unlk(struct file *file, struct file_lock *fl)
 
 	_enter("{%llx:%llu},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_type);
 
+	trace_afs_flock_op(vnode, fl, afs_flock_op_unlock);
+
 	/* Flush all pending writes before doing anything with locks. */
 	vfs_fsync(file, 0);
 
-	/* only whole-file unlocks are supported */
-	if (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX)
-		return -EINVAL;
-
-	ret = posix_lock_file(file, fl, NULL);
+	ret = locks_lock_file_wait(file, fl);
 	_leave(" = %d [%u]", ret, vnode->lock_state);
 	return ret;
 }
@@ -608,6 +727,9 @@ static int afs_do_getlk(struct file *file, struct file_lock *fl)
 
 	_enter("");
 
+	if (vnode->lock_state == AFS_VNODE_LOCK_DELETED)
+		return -ENOENT;
+
 	fl->fl_type = F_UNLCK;
 
 	/* check local lock records first */
@@ -619,12 +741,15 @@ static int afs_do_getlk(struct file *file, struct file_lock *fl)
 			goto error;
 
 		lock_count = READ_ONCE(vnode->status.lock_count);
-		if (lock_count > 0)
-			fl->fl_type = F_RDLCK;
-		else
-			fl->fl_type = F_WRLCK;
-		fl->fl_start = 0;
-		fl->fl_end = OFFSET_MAX;
+		if (lock_count != 0) {
+			if (lock_count > 0)
+				fl->fl_type = F_RDLCK;
+			else
+				fl->fl_type = F_WRLCK;
+			fl->fl_start = 0;
+			fl->fl_end = OFFSET_MAX;
+			fl->fl_pid = 0;
+		}
 	}
 
 	ret = 0;
@@ -639,6 +764,8 @@ error:
 int afs_lock(struct file *file, int cmd, struct file_lock *fl)
 {
 	struct afs_vnode *vnode = AFS_FS_I(locks_inode(file));
+	enum afs_flock_operation op;
+	int ret;
 
 	_enter("{%llx:%llu},%d,{t=%x,fl=%x,r=%Ld:%Ld}",
 	       vnode->fid.vid, vnode->fid.vnode, cmd,
@@ -651,9 +778,23 @@ int afs_lock(struct file *file, int cmd, struct file_lock *fl)
 
 	if (IS_GETLK(cmd))
 		return afs_do_getlk(file, fl);
+
+	fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
+	trace_afs_flock_op(vnode, fl, afs_flock_op_lock);
+
 	if (fl->fl_type == F_UNLCK)
-		return afs_do_unlk(file, fl);
-	return afs_do_setlk(file, fl);
+		ret = afs_do_unlk(file, fl);
+	else
+		ret = afs_do_setlk(file, fl);
+
+	switch (ret) {
+	case 0:		op = afs_flock_op_return_ok; break;
+	case -EAGAIN:	op = afs_flock_op_return_eagain; break;
+	case -EDEADLK:	op = afs_flock_op_return_edeadlk; break;
+	default:	op = afs_flock_op_return_error; break;
+	}
+	trace_afs_flock_op(vnode, fl, op);
+	return ret;
 }
 
 /*
@@ -662,6 +803,8 @@ int afs_lock(struct file *file, int cmd, struct file_lock *fl)
 int afs_flock(struct file *file, int cmd, struct file_lock *fl)
 {
 	struct afs_vnode *vnode = AFS_FS_I(locks_inode(file));
+	enum afs_flock_operation op;
+	int ret;
 
 	_enter("{%llx:%llu},%d,{t=%x,fl=%x}",
 	       vnode->fid.vid, vnode->fid.vnode, cmd,
@@ -677,10 +820,23 @@ int afs_flock(struct file *file, int cmd, struct file_lock *fl)
 	if (!(fl->fl_flags & FL_FLOCK))
 		return -ENOLCK;
 
+	fl->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
+	trace_afs_flock_op(vnode, fl, afs_flock_op_flock);
+
 	/* we're simulating flock() locks using posix locks on the server */
 	if (fl->fl_type == F_UNLCK)
-		return afs_do_unlk(file, fl);
-	return afs_do_setlk(file, fl);
+		ret = afs_do_unlk(file, fl);
+	else
+		ret = afs_do_setlk(file, fl);
+
+	switch (ret) {
+	case 0:		op = afs_flock_op_return_ok; break;
+	case -EAGAIN:	op = afs_flock_op_return_eagain; break;
+	case -EDEADLK:	op = afs_flock_op_return_edeadlk; break;
+	default:	op = afs_flock_op_return_error; break;
+	}
+	trace_afs_flock_op(vnode, fl, op);
+	return ret;
 }
 
 /*
@@ -695,7 +851,10 @@ static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl)
 
 	_enter("");
 
+	new->fl_u.afs.debug_id = atomic_inc_return(&afs_file_lock_debug_id);
+
 	spin_lock(&vnode->lock);
+	trace_afs_flock_op(vnode, new, afs_flock_op_copy_lock);
 	list_add(&new->fl_u.afs.link, &fl->fl_u.afs.link);
 	spin_unlock(&vnode->lock);
 }
@@ -711,7 +870,12 @@ static void afs_fl_release_private(struct file_lock *fl)
 	_enter("");
 
 	spin_lock(&vnode->lock);
-	afs_dequeue_lock(vnode, fl);
+
+	trace_afs_flock_op(vnode, fl, afs_flock_op_release_lock);
+	list_del_init(&fl->fl_u.afs.link);
+	if (list_empty(&vnode->granted_locks))
+		afs_defer_unlock(vnode);
+
 	_debug("state %u for %p", vnode->lock_state, vnode);
 	spin_unlock(&vnode->lock);
 }