1 files changed, 255 insertions, 54 deletions

diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
index 7f9a578a1a20..571c4826c428 100644
--- a/fs/btrfs/locking.c
+++ b/fs/btrfs/locking.c
@@ -13,65 +13,164 @@
 #include "extent_io.h"
 #include "locking.h"
 
+/*
+ * Extent buffer locking
+ * =====================
+ *
+ * The locks use a custom scheme that allows to do more operations than are
+ * available fromt current locking primitives. The building blocks are still
+ * rwlock and wait queues.
+ *
+ * Required semantics:
+ *
+ * - reader/writer exclusion
+ * - writer/writer exclusion
+ * - reader/reader sharing
+ * - spinning lock semantics
+ * - blocking lock semantics
+ * - try-lock semantics for readers and writers
+ * - one level nesting, allowing read lock to be taken by the same thread that
+ *   already has write lock
+ *
+ * The extent buffer locks (also called tree locks) manage access to eb data
+ * related to the storage in the b-tree (keys, items, but not the individual
+ * members of eb).
+ * We want concurrency of many readers and safe updates. The underlying locking
+ * is done by read-write spinlock and the blocking part is implemented using
+ * counters and wait queues.
+ *
+ * spinning semantics - the low-level rwlock is held so all other threads that
+ *                      want to take it are spinning on it.
+ *
+ * blocking semantics - the low-level rwlock is not held but the counter
+ *                      denotes how many times the blocking lock was held;
+ *                      sleeping is possible
+ *
+ * Write lock always allows only one thread to access the data.
+ *
+ *
+ * Debugging
+ * ---------
+ *
+ * There are additional state counters that are asserted in various contexts,
+ * removed from non-debug build to reduce extent_buffer size and for
+ * performance reasons.
+ *
+ *
+ * Lock nesting
+ * ------------
+ *
+ * A write operation on a tree might indirectly start a look up on the same
+ * tree.  This can happen when btrfs_cow_block locks the tree and needs to
+ * lookup free extents.
+ *
+ * btrfs_cow_block
+ *   ..
+ *   alloc_tree_block_no_bg_flush
+ *     btrfs_alloc_tree_block
+ *       btrfs_reserve_extent
+ *         ..
+ *         load_free_space_cache
+ *           ..
+ *           btrfs_lookup_file_extent
+ *             btrfs_search_slot
+ *
+ *
+ * Locking pattern - spinning
+ * --------------------------
+ *
+ * The simple locking scenario, the +--+ denotes the spinning section.
+ *
+ * +- btrfs_tree_lock
+ * | - extent_buffer::rwlock is held
+ * | - no heavy operations should happen, eg. IO, memory allocations, large
+ * |   structure traversals
+ * +- btrfs_tree_unock
+*
+*
+ * Locking pattern - blocking
+ * --------------------------
+ *
+ * The blocking write uses the following scheme.  The +--+ denotes the spinning
+ * section.
+ *
+ * +- btrfs_tree_lock
+ * |
+ * +- btrfs_set_lock_blocking_write
+ *
+ *   - allowed: IO, memory allocations, etc.
+ *
+ * -- btrfs_tree_unlock - note, no explicit unblocking necessary
+ *
+ *
+ * Blocking read is similar.
+ *
+ * +- btrfs_tree_read_lock
+ * |
+ * +- btrfs_set_lock_blocking_read
+ *
+ *  - heavy operations allowed
+ *
+ * +- btrfs_tree_read_unlock_blocking
+ * |
+ * +- btrfs_tree_read_unlock
+ *
+ */
+
 #ifdef CONFIG_BTRFS_DEBUG
-static void btrfs_assert_spinning_writers_get(struct extent_buffer *eb)
+static inline void btrfs_assert_spinning_writers_get(struct extent_buffer *eb)
 {
 	WARN_ON(eb->spinning_writers);
 	eb->spinning_writers++;
 }
 
-static void btrfs_assert_spinning_writers_put(struct extent_buffer *eb)
+static inline void btrfs_assert_spinning_writers_put(struct extent_buffer *eb)
 {
 	WARN_ON(eb->spinning_writers != 1);
 	eb->spinning_writers--;
 }
 
-static void btrfs_assert_no_spinning_writers(struct extent_buffer *eb)
+static inline void btrfs_assert_no_spinning_writers(struct extent_buffer *eb)
 {
 	WARN_ON(eb->spinning_writers);
 }
 
-static void btrfs_assert_spinning_readers_get(struct extent_buffer *eb)
+static inline void btrfs_assert_spinning_readers_get(struct extent_buffer *eb)
 {
 	atomic_inc(&eb->spinning_readers);
 }
 
-static void btrfs_assert_spinning_readers_put(struct extent_buffer *eb)
+static inline void btrfs_assert_spinning_readers_put(struct extent_buffer *eb)
 {
 	WARN_ON(atomic_read(&eb->spinning_readers) == 0);
 	atomic_dec(&eb->spinning_readers);
 }
 
-static void btrfs_assert_tree_read_locks_get(struct extent_buffer *eb)
+static inline void btrfs_assert_tree_read_locks_get(struct extent_buffer *eb)
 {
 	atomic_inc(&eb->read_locks);
 }
 
-static void btrfs_assert_tree_read_locks_put(struct extent_buffer *eb)
+static inline void btrfs_assert_tree_read_locks_put(struct extent_buffer *eb)
 {
 	atomic_dec(&eb->read_locks);
 }
 
-static void btrfs_assert_tree_read_locked(struct extent_buffer *eb)
+static inline void btrfs_assert_tree_read_locked(struct extent_buffer *eb)
 {
 	BUG_ON(!atomic_read(&eb->read_locks));
 }
 
-static void btrfs_assert_tree_write_locks_get(struct extent_buffer *eb)
+static inline void btrfs_assert_tree_write_locks_get(struct extent_buffer *eb)
 {
 	eb->write_locks++;
 }
 
-static void btrfs_assert_tree_write_locks_put(struct extent_buffer *eb)
+static inline void btrfs_assert_tree_write_locks_put(struct extent_buffer *eb)
 {
 	eb->write_locks--;
 }
 
-void btrfs_assert_tree_locked(struct extent_buffer *eb)
-{
-	BUG_ON(!eb->write_locks);
-}
-
 #else
 static void btrfs_assert_spinning_writers_get(struct extent_buffer *eb) { }
 static void btrfs_assert_spinning_writers_put(struct extent_buffer *eb) { }
@@ -81,11 +180,19 @@ static void btrfs_assert_spinning_readers_get(struct extent_buffer *eb) { }
 static void btrfs_assert_tree_read_locked(struct extent_buffer *eb) { }
 static void btrfs_assert_tree_read_locks_get(struct extent_buffer *eb) { }
 static void btrfs_assert_tree_read_locks_put(struct extent_buffer *eb) { }
-void btrfs_assert_tree_locked(struct extent_buffer *eb) { }
 static void btrfs_assert_tree_write_locks_get(struct extent_buffer *eb) { }
 static void btrfs_assert_tree_write_locks_put(struct extent_buffer *eb) { }
 #endif
 
+/*
+ * Mark already held read lock as blocking. Can be nested in write lock by the
+ * same thread.
+ *
+ * Use when there are potentially long operations ahead so other thread waiting
+ * on the lock will not actively spin but sleep instead.
+ *
+ * The rwlock is released and blocking reader counter is increased.
+ */
 void btrfs_set_lock_blocking_read(struct extent_buffer *eb)
 {
 	trace_btrfs_set_lock_blocking_read(eb);
@@ -102,6 +209,14 @@ void btrfs_set_lock_blocking_read(struct extent_buffer *eb)
 	read_unlock(&eb->lock);
 }
 
+/*
+ * Mark already held write lock as blocking.
+ *
+ * Use when there are potentially long operations ahead so other threads
+ * waiting on the lock will not actively spin but sleep instead.
+ *
+ * The rwlock is released and blocking writers is set.
+ */
 void btrfs_set_lock_blocking_write(struct extent_buffer *eb)
 {
 	trace_btrfs_set_lock_blocking_write(eb);
@@ -115,14 +230,19 @@ void btrfs_set_lock_blocking_write(struct extent_buffer *eb)
 	if (eb->blocking_writers == 0) {
 		btrfs_assert_spinning_writers_put(eb);
 		btrfs_assert_tree_locked(eb);
-		eb->blocking_writers++;
+		WRITE_ONCE(eb->blocking_writers, 1);
 		write_unlock(&eb->lock);
 	}
 }
 
 /*
- * take a spinning read lock.  This will wait for any blocking
- * writers
+ * Lock the extent buffer for read. Wait for any writers (spinning or blocking).
+ * Can be nested in write lock by the same thread.
+ *
+ * Use when the locked section does only lightweight actions and busy waiting
+ * would be cheaper than making other threads do the wait/wake loop.
+ *
+ * The rwlock is held upon exit.
  */
 void btrfs_tree_read_lock(struct extent_buffer *eb)
 {
@@ -134,23 +254,24 @@ again:
 	read_lock(&eb->lock);
 	BUG_ON(eb->blocking_writers == 0 &&
 	       current->pid == eb->lock_owner);
-	if (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 = true;
-		read_unlock(&eb->lock);
-		trace_btrfs_tree_read_lock(eb, start_ns);
-		return;
-	}
 	if (eb->blocking_writers) {
+		if (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 = true;
+			read_unlock(&eb->lock);
+			trace_btrfs_tree_read_lock(eb, start_ns);
+			return;
+		}
 		read_unlock(&eb->lock);
 		wait_event(eb->write_lock_wq,
-			   eb->blocking_writers == 0);
+			   READ_ONCE(eb->blocking_writers) == 0);
 		goto again;
 	}
 	btrfs_assert_tree_read_locks_get(eb);
@@ -159,17 +280,19 @@ again:
 }
 
 /*
- * take a spinning read lock.
- * returns 1 if we get the read lock and 0 if we don't
- * this won't wait for blocking writers
+ * Lock extent buffer for read, optimistically expecting that there are no
+ * contending blocking writers. If there are, don't wait.
+ *
+ * Return 1 if the rwlock has been taken, 0 otherwise
  */
 int btrfs_tree_read_lock_atomic(struct extent_buffer *eb)
 {
-	if (eb->blocking_writers)
+	if (READ_ONCE(eb->blocking_writers))
 		return 0;
 
 	read_lock(&eb->lock);
-	if (eb->blocking_writers) {
+	/* Refetch value after lock */
+	if (READ_ONCE(eb->blocking_writers)) {
 		read_unlock(&eb->lock);
 		return 0;
 	}
@@ -180,18 +303,20 @@ int btrfs_tree_read_lock_atomic(struct extent_buffer *eb)
 }
 
 /*
- * returns 1 if we get the read lock and 0 if we don't
- * this won't wait for blocking writers
+ * Try-lock for read. Don't block or wait for contending writers.
+ *
+ * Retrun 1 if the rwlock has been taken, 0 otherwise
  */
 int btrfs_try_tree_read_lock(struct extent_buffer *eb)
 {
-	if (eb->blocking_writers)
+	if (READ_ONCE(eb->blocking_writers))
 		return 0;
 
 	if (!read_trylock(&eb->lock))
 		return 0;
 
-	if (eb->blocking_writers) {
+	/* Refetch value after lock */
+	if (READ_ONCE(eb->blocking_writers)) {
 		read_unlock(&eb->lock);
 		return 0;
 	}
@@ -202,16 +327,19 @@ int btrfs_try_tree_read_lock(struct extent_buffer *eb)
 }
 
 /*
- * returns 1 if we get the read lock and 0 if we don't
- * this won't wait for blocking writers or readers
+ * Try-lock for write. May block until the lock is uncontended, but does not
+ * wait until it is free.
+ *
+ * Retrun 1 if the rwlock has been taken, 0 otherwise
  */
 int btrfs_try_tree_write_lock(struct extent_buffer *eb)
 {
-	if (eb->blocking_writers || atomic_read(&eb->blocking_readers))
+	if (READ_ONCE(eb->blocking_writers) || atomic_read(&eb->blocking_readers))
 		return 0;
 
 	write_lock(&eb->lock);
-	if (eb->blocking_writers || atomic_read(&eb->blocking_readers)) {
+	/* Refetch value after lock */
+	if (READ_ONCE(eb->blocking_writers) || atomic_read(&eb->blocking_readers)) {
 		write_unlock(&eb->lock);
 		return 0;
 	}
@@ -223,7 +351,10 @@ int btrfs_try_tree_write_lock(struct extent_buffer *eb)
 }
 
 /*
- * drop a spinning read lock
+ * Release read lock. Must be used only if the lock is in spinning mode.  If
+ * the read lock is nested, must pair with read lock before the write unlock.
+ *
+ * The rwlock is not held upon exit.
  */
 void btrfs_tree_read_unlock(struct extent_buffer *eb)
 {
@@ -245,7 +376,11 @@ void btrfs_tree_read_unlock(struct extent_buffer *eb)
 }
 
 /*
- * drop a blocking read lock
+ * Release read lock, previously set to blocking by a pairing call to
+ * btrfs_set_lock_blocking_read(). Can be nested in write lock by the same
+ * thread.
+ *
+ * State of rwlock is unchanged, last reader wakes waiting threads.
  */
 void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
 {
@@ -269,8 +404,10 @@ void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
 }
 
 /*
- * take a spinning write lock.  This will wait for both
- * blocking readers or writers
+ * Lock for write. Wait for all blocking and spinning readers and writers. This
+ * starts context where reader lock could be nested by the same thread.
+ *
+ * The rwlock is held for write upon exit.
  */
 void btrfs_tree_lock(struct extent_buffer *eb)
 {
@@ -282,9 +419,11 @@ void btrfs_tree_lock(struct extent_buffer *eb)
 	WARN_ON(eb->lock_owner == current->pid);
 again:
 	wait_event(eb->read_lock_wq, atomic_read(&eb->blocking_readers) == 0);
-	wait_event(eb->write_lock_wq, eb->blocking_writers == 0);
+	wait_event(eb->write_lock_wq, READ_ONCE(eb->blocking_writers) == 0);
 	write_lock(&eb->lock);
-	if (atomic_read(&eb->blocking_readers) || eb->blocking_writers) {
+	/* Refetch value after lock */
+	if (atomic_read(&eb->blocking_readers) ||
+	    READ_ONCE(eb->blocking_writers)) {
 		write_unlock(&eb->lock);
 		goto again;
 	}
@@ -295,10 +434,19 @@ again:
 }
 
 /*
- * drop a spinning or a blocking write lock.
+ * Release the write lock, either blocking or spinning (ie. there's no need
+ * for an explicit blocking unlock, like btrfs_tree_read_unlock_blocking).
+ * This also ends the context for nesting, the read lock must have been
+ * released already.
+ *
+ * Tasks blocked and waiting are woken, rwlock is not held upon exit.
  */
 void btrfs_tree_unlock(struct extent_buffer *eb)
 {
+	/*
+	 * This is read both locked and unlocked but always by the same thread
+	 * that already owns the lock so we don't need to use READ_ONCE
+	 */
 	int blockers = eb->blocking_writers;
 
 	BUG_ON(blockers > 1);
@@ -310,7 +458,8 @@ void btrfs_tree_unlock(struct extent_buffer *eb)
 
 	if (blockers) {
 		btrfs_assert_no_spinning_writers(eb);
-		eb->blocking_writers--;
+		/* Unlocked write */
+		WRITE_ONCE(eb->blocking_writers, 0);
 		/*
 		 * We need to order modifying blocking_writers above with
 		 * actually waking up the sleepers to ensure they see the
@@ -322,3 +471,55 @@ void btrfs_tree_unlock(struct extent_buffer *eb)
 		write_unlock(&eb->lock);
 	}
 }
+
+/*
+ * Set all locked nodes in the path to blocking locks.  This should be done
+ * before scheduling
+ */
+void btrfs_set_path_blocking(struct btrfs_path *p)
+{
+	int i;
+
+	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+		if (!p->nodes[i] || !p->locks[i])
+			continue;
+		/*
+		 * If we currently have a spinning reader or writer lock this
+		 * will bump the count of blocking holders and drop the
+		 * spinlock.
+		 */
+		if (p->locks[i] == BTRFS_READ_LOCK) {
+			btrfs_set_lock_blocking_read(p->nodes[i]);
+			p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
+		} else if (p->locks[i] == BTRFS_WRITE_LOCK) {
+			btrfs_set_lock_blocking_write(p->nodes[i]);
+			p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
+		}
+	}
+}
+
+/*
+ * This releases any locks held in the path starting at level and going all the
+ * way up to the root.
+ *
+ * btrfs_search_slot will keep the lock held on higher nodes in a few corner
+ * cases, such as COW of the block at slot zero in the node.  This ignores
+ * those rules, and it should only be called when there are no more updates to
+ * be done higher up in the tree.
+ */
+void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
+{
+	int i;
+
+	if (path->keep_locks)
+		return;
+
+	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
+		if (!path->nodes[i])
+			continue;
+		if (!path->locks[i])
+			continue;
+		btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
+		path->locks[i] = 0;
+	}
+}