1 files changed, 231 insertions, 37 deletions

diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
index 39bae7761db..5665d214924 100644
--- a/fs/btrfs/locking.c
+++ b/fs/btrfs/locking.c
@@ -16,7 +16,6 @@
  * Boston, MA 021110-1307, USA.
  */
 #include <linux/sched.h>
-#include <linux/gfp.h>
 #include <linux/pagemap.h>
 #include <linux/spinlock.h>
 #include <linux/page-flags.h>
@@ -25,64 +24,259 @@
 #include "extent_io.h"
 #include "locking.h"
 
+static void btrfs_assert_tree_read_locked(struct extent_buffer *eb);
+
 /*
- * locks the per buffer mutex in an extent buffer.  This uses adaptive locks
- * and the spin is not tuned very extensively.  The spinning does make a big
- * difference in almost every workload, but spinning for the right amount of
- * time needs some help.
- *
- * In general, we want to spin as long as the lock holder is doing btree
- * searches, and we should give up if they are in more expensive code.
+ * if we currently have a spinning reader or writer lock
+ * (indicated by the rw flag) this will bump the count
+ * of blocking holders and drop the spinlock.
+ */
+void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw)
+{
+	/*
+	 * no lock is required.  The lock owner may change if
+	 * we have a read lock, but it won't change to or away
+	 * from us.  If we have the write lock, we are the owner
+	 * and it'll never change.
+	 */
+	if (eb->lock_nested && current->pid == eb->lock_owner)
+		return;
+	if (rw == BTRFS_WRITE_LOCK) {
+		if (atomic_read(&eb->blocking_writers) == 0) {
+			WARN_ON(atomic_read(&eb->spinning_writers) != 1);
+			atomic_dec(&eb->spinning_writers);
+			btrfs_assert_tree_locked(eb);
+			atomic_inc(&eb->blocking_writers);
+			write_unlock(&eb->lock);
+		}
+	} else if (rw == BTRFS_READ_LOCK) {
+		btrfs_assert_tree_read_locked(eb);
+		atomic_inc(&eb->blocking_readers);
+		WARN_ON(atomic_read(&eb->spinning_readers) == 0);
+		atomic_dec(&eb->spinning_readers);
+		read_unlock(&eb->lock);
+	}
+	return;
+}
+
+/*
+ * if we currently have a blocking lock, take the spinlock
+ * and drop our blocking count
  */
+void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw)
+{
+	/*
+	 * no lock is required.  The lock owner may change if
+	 * we have a read lock, but it won't change to or away
+	 * from us.  If we have the write lock, we are the owner
+	 * and it'll never change.
+	 */
+	if (eb->lock_nested && current->pid == eb->lock_owner)
+		return;
+
+	if (rw == BTRFS_WRITE_LOCK_BLOCKING) {
+		BUG_ON(atomic_read(&eb->blocking_writers) != 1);
+		write_lock(&eb->lock);
+		WARN_ON(atomic_read(&eb->spinning_writers));
+		atomic_inc(&eb->spinning_writers);
+		if (atomic_dec_and_test(&eb->blocking_writers) &&
+		    waitqueue_active(&eb->write_lock_wq))
+			wake_up(&eb->write_lock_wq);
+	} else if (rw == BTRFS_READ_LOCK_BLOCKING) {
+		BUG_ON(atomic_read(&eb->blocking_readers) == 0);
+		read_lock(&eb->lock);
+		atomic_inc(&eb->spinning_readers);
+		if (atomic_dec_and_test(&eb->blocking_readers) &&
+		    waitqueue_active(&eb->read_lock_wq))
+			wake_up(&eb->read_lock_wq);
+	}
+	return;
+}
+
+/*
+ * take a spinning read lock.  This will wait for any blocking
+ * writers
+ */
+void btrfs_tree_read_lock(struct extent_buffer *eb)
+{
+again:
+	BUG_ON(!atomic_read(&eb->blocking_writers) &&
+	       current->pid == eb->lock_owner);
+
+	read_lock(&eb->lock);
+	if (atomic_read(&eb->blocking_writers) &&
+	    current->pid == eb->lock_owner) {
+		/*
+		 * This extent is already write-locked by our thread. We allow
+		 * an additional read lock to be added because it's for the same
+		 * thread. btrfs_find_all_roots() depends on this as it may be
+		 * called on a partly (write-)locked tree.
+		 */
+		BUG_ON(eb->lock_nested);
+		eb->lock_nested = 1;
+		read_unlock(&eb->lock);
+		return;
+	}
+	if (atomic_read(&eb->blocking_writers)) {
+		read_unlock(&eb->lock);
+		wait_event(eb->write_lock_wq,
+			   atomic_read(&eb->blocking_writers) == 0);
+		goto again;
+	}
+	atomic_inc(&eb->read_locks);
+	atomic_inc(&eb->spinning_readers);
+}
+
+/*
+ * returns 1 if we get the read lock and 0 if we don't
+ * this won't wait for blocking writers
+ */
+int btrfs_try_tree_read_lock(struct extent_buffer *eb)
+{
+	if (atomic_read(&eb->blocking_writers))
+		return 0;
+
+	if (!read_trylock(&eb->lock))
+		return 0;
 
-int btrfs_tree_lock(struct extent_buffer *eb)
+	if (atomic_read(&eb->blocking_writers)) {
+		read_unlock(&eb->lock);
+		return 0;
+	}
+	atomic_inc(&eb->read_locks);
+	atomic_inc(&eb->spinning_readers);
+	return 1;
+}
+
+/*
+ * returns 1 if we get the read lock and 0 if we don't
+ * this won't wait for blocking writers or readers
+ */
+int btrfs_try_tree_write_lock(struct extent_buffer *eb)
 {
-	int i;
+	if (atomic_read(&eb->blocking_writers) ||
+	    atomic_read(&eb->blocking_readers))
+		return 0;
 
-	if (mutex_trylock(&eb->mutex))
+	if (!write_trylock(&eb->lock))
+		return 0;
+
+	if (atomic_read(&eb->blocking_writers) ||
+	    atomic_read(&eb->blocking_readers)) {
+		write_unlock(&eb->lock);
 		return 0;
-	for (i = 0; i < 512; i++) {
-		cpu_relax();
-		if (mutex_trylock(&eb->mutex))
-			return 0;
 	}
-	cpu_relax();
-	mutex_lock_nested(&eb->mutex, BTRFS_MAX_LEVEL - btrfs_header_level(eb));
-	return 0;
+	atomic_inc(&eb->write_locks);
+	atomic_inc(&eb->spinning_writers);
+	eb->lock_owner = current->pid;
+	return 1;
 }
 
-int btrfs_try_tree_lock(struct extent_buffer *eb)
+/*
+ * drop a spinning read lock
+ */
+void btrfs_tree_read_unlock(struct extent_buffer *eb)
 {
-	return mutex_trylock(&eb->mutex);
+	/*
+	 * if we're nested, we have the write lock.  No new locking
+	 * is needed as long as we are the lock owner.
+	 * The write unlock will do a barrier for us, and the lock_nested
+	 * field only matters to the lock owner.
+	 */
+	if (eb->lock_nested && current->pid == eb->lock_owner) {
+		eb->lock_nested = 0;
+		return;
+	}
+	btrfs_assert_tree_read_locked(eb);
+	WARN_ON(atomic_read(&eb->spinning_readers) == 0);
+	atomic_dec(&eb->spinning_readers);
+	atomic_dec(&eb->read_locks);
+	read_unlock(&eb->lock);
 }
 
-int btrfs_tree_unlock(struct extent_buffer *eb)
+/*
+ * drop a blocking read lock
+ */
+void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
 {
-	mutex_unlock(&eb->mutex);
-	return 0;
+	/*
+	 * if we're nested, we have the write lock.  No new locking
+	 * is needed as long as we are the lock owner.
+	 * The write unlock will do a barrier for us, and the lock_nested
+	 * field only matters to the lock owner.
+	 */
+	if (eb->lock_nested && current->pid == eb->lock_owner) {
+		eb->lock_nested = 0;
+		return;
+	}
+	btrfs_assert_tree_read_locked(eb);
+	WARN_ON(atomic_read(&eb->blocking_readers) == 0);
+	if (atomic_dec_and_test(&eb->blocking_readers) &&
+	    waitqueue_active(&eb->read_lock_wq))
+		wake_up(&eb->read_lock_wq);
+	atomic_dec(&eb->read_locks);
 }
 
-int btrfs_tree_locked(struct extent_buffer *eb)
+/*
+ * take a spinning write lock.  This will wait for both
+ * blocking readers or writers
+ */
+void btrfs_tree_lock(struct extent_buffer *eb)
 {
-	return mutex_is_locked(&eb->mutex);
+again:
+	wait_event(eb->read_lock_wq, atomic_read(&eb->blocking_readers) == 0);
+	wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0);
+	write_lock(&eb->lock);
+	if (atomic_read(&eb->blocking_readers)) {
+		write_unlock(&eb->lock);
+		wait_event(eb->read_lock_wq,
+			   atomic_read(&eb->blocking_readers) == 0);
+		goto again;
+	}
+	if (atomic_read(&eb->blocking_writers)) {
+		write_unlock(&eb->lock);
+		wait_event(eb->write_lock_wq,
+			   atomic_read(&eb->blocking_writers) == 0);
+		goto again;
+	}
+	WARN_ON(atomic_read(&eb->spinning_writers));
+	atomic_inc(&eb->spinning_writers);
+	atomic_inc(&eb->write_locks);
+	eb->lock_owner = current->pid;
 }
 
 /*
- * btrfs_search_slot uses this to decide if it should drop its locks
- * before doing something expensive like allocating free blocks for cow.
+ * drop a spinning or a blocking write lock.
  */
-int btrfs_path_lock_waiting(struct btrfs_path *path, int level)
+void btrfs_tree_unlock(struct extent_buffer *eb)
 {
-	int i;
-	struct extent_buffer *eb;
-	for (i = level; i <= level + 1 && i < BTRFS_MAX_LEVEL; i++) {
-		eb = path->nodes[i];
-		if (!eb)
-			break;
+	int blockers = atomic_read(&eb->blocking_writers);
+
+	BUG_ON(blockers > 1);
+
+	btrfs_assert_tree_locked(eb);
+	eb->lock_owner = 0;
+	atomic_dec(&eb->write_locks);
+
+	if (blockers) {
+		WARN_ON(atomic_read(&eb->spinning_writers));
+		atomic_dec(&eb->blocking_writers);
 		smp_mb();
-		if (!list_empty(&eb->mutex.wait_list))
-			return 1;
+		if (waitqueue_active(&eb->write_lock_wq))
+			wake_up(&eb->write_lock_wq);
+	} else {
+		WARN_ON(atomic_read(&eb->spinning_writers) != 1);
+		atomic_dec(&eb->spinning_writers);
+		write_unlock(&eb->lock);
 	}
-	return 0;
 }
 
+void btrfs_assert_tree_locked(struct extent_buffer *eb)
+{
+	BUG_ON(!atomic_read(&eb->write_locks));
+}
+
+static void btrfs_assert_tree_read_locked(struct extent_buffer *eb)
+{
+	BUG_ON(!atomic_read(&eb->read_locks));
+}