Merge tag 'for-5.7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs updates from David Sterba:
 "A number of core changes that make things work better in general, code
  is simpler and cleaner.

  Core changes:

   - per-inode file extent tree, for in memory tracking of contiguous
     extent ranges to make sure i_size adjustments are accurate

   - tree root structures are protected by reference counts, replacing
     SRCU that did not cover some cases

   - leak detector for tree root structures

   - per-transaction pinned extent tracking

   - buffer heads are replaced by bios for super block access

   - speedup of extent back reference resolution, on an example test
     scenario the runtime of send went down from a hour to minutes

   - factor out locking scheme used for subvolume writer and NOCOW
     exclusion, abstracted as DREW lock, double reader-writer exclusion
     (allow either readers or writers)

   - cleanup and abstract extent allocation policies, preparation for
     zoned device support

   - make reflink/clone_range work on inline extents

   - add more cancellation point for relocation, improves long response
     from 'balance cancel'

   - add page migration callback for data pages

   - switch to guid for uuids, with additional cleanups of the interface

   - make ranged full fsyncs more efficient

   - removal of obsolete ioctl flag BTRFS_SUBVOL_CREATE_ASYNC

   - remove b-tree readahead from delayed refs paths, avoiding seek and
     read unnecessary blocks

  Features:

   - v2 of ioctl to delete subvolumes, allowing to delete by id and more
     future extensions

  Fixes:

   - fix qgroup rescan worker that could block umount

   - fix crash during unmount due to race with delayed inode workers

   - fix dellaloc flushing logic that could create unnecessary chunks
     under heavy load

   - fix missing file extent item for hole after ranged fsync

   - several fixes in relocation error handling

  Other:

   - more documentation of relocation, device replace, space
     reservations

   - many random cleanups"

* tag 'for-5.7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux: (210 commits)
  btrfs: fix missing semaphore unlock in btrfs_sync_file
  btrfs: use nofs allocations for running delayed items
  btrfs: sysfs: Use scnprintf() instead of snprintf()
  btrfs: do not resolve backrefs for roots that are being deleted
  btrfs: track reloc roots based on their commit root bytenr
  btrfs: restart relocate_tree_blocks properly
  btrfs: reloc: reorder reservation before root selection
  btrfs: do not readahead in build_backref_tree
  btrfs: do not use readahead for running delayed refs
  btrfs: Remove async_transid from btrfs_mksubvol/create_subvol/create_snapshot
  btrfs: Remove transid argument from btrfs_ioctl_snap_create_transid
  btrfs: Remove BTRFS_SUBVOL_CREATE_ASYNC support
  btrfs: kill the subvol_srcu
  btrfs: make btrfs_cleanup_fs_roots use the radix tree lock
  btrfs: don't take an extra root ref at allocation time
  btrfs: hold a ref on the root on the dead roots list
  btrfs: make inodes hold a ref on their roots
  btrfs: move the root freeing stuff into btrfs_put_root
  btrfs: move ino_cache_inode dropping out of btrfs_free_fs_root
  btrfs: make the extent buffer leak check per fs info
  ...

1 files changed, 135 insertions, 0 deletions

diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
index 571c4826c428..fb647d8cf527 100644
--- a/fs/btrfs/locking.c
+++ b/fs/btrfs/locking.c
@@ -523,3 +523,138 @@ void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
 		path->locks[i] = 0;
 	}
 }
+
+/*
+ * Loop around taking references on and locking the root node of the tree until
+ * we end up with a lock on the root node.
+ *
+ * Return: root extent buffer with write lock held
+ */
+struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
+{
+	struct extent_buffer *eb;
+
+	while (1) {
+		eb = btrfs_root_node(root);
+		btrfs_tree_lock(eb);
+		if (eb == root->node)
+			break;
+		btrfs_tree_unlock(eb);
+		free_extent_buffer(eb);
+	}
+	return eb;
+}
+
+/*
+ * Loop around taking references on and locking the root node of the tree until
+ * we end up with a lock on the root node.
+ *
+ * Return: root extent buffer with read lock held
+ */
+struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
+{
+	struct extent_buffer *eb;
+
+	while (1) {
+		eb = btrfs_root_node(root);
+		btrfs_tree_read_lock(eb);
+		if (eb == root->node)
+			break;
+		btrfs_tree_read_unlock(eb);
+		free_extent_buffer(eb);
+	}
+	return eb;
+}
+
+/*
+ * DREW locks
+ * ==========
+ *
+ * DREW stands for double-reader-writer-exclusion lock. It's used in situation
+ * where you want to provide A-B exclusion but not AA or BB.
+ *
+ * Currently implementation gives more priority to reader. If a reader and a
+ * writer both race to acquire their respective sides of the lock the writer
+ * would yield its lock as soon as it detects a concurrent reader. Additionally
+ * if there are pending readers no new writers would be allowed to come in and
+ * acquire the lock.
+ */
+
+int btrfs_drew_lock_init(struct btrfs_drew_lock *lock)
+{
+	int ret;
+
+	ret = percpu_counter_init(&lock->writers, 0, GFP_KERNEL);
+	if (ret)
+		return ret;
+
+	atomic_set(&lock->readers, 0);
+	init_waitqueue_head(&lock->pending_readers);
+	init_waitqueue_head(&lock->pending_writers);
+
+	return 0;
+}
+
+void btrfs_drew_lock_destroy(struct btrfs_drew_lock *lock)
+{
+	percpu_counter_destroy(&lock->writers);
+}
+
+/* Return true if acquisition is successful, false otherwise */
+bool btrfs_drew_try_write_lock(struct btrfs_drew_lock *lock)
+{
+	if (atomic_read(&lock->readers))
+		return false;
+
+	percpu_counter_inc(&lock->writers);
+
+	/* Ensure writers count is updated before we check for pending readers */
+	smp_mb();
+	if (atomic_read(&lock->readers)) {
+		btrfs_drew_write_unlock(lock);
+		return false;
+	}
+
+	return true;
+}
+
+void btrfs_drew_write_lock(struct btrfs_drew_lock *lock)
+{
+	while (true) {
+		if (btrfs_drew_try_write_lock(lock))
+			return;
+		wait_event(lock->pending_writers, !atomic_read(&lock->readers));
+	}
+}
+
+void btrfs_drew_write_unlock(struct btrfs_drew_lock *lock)
+{
+	percpu_counter_dec(&lock->writers);
+	cond_wake_up(&lock->pending_readers);
+}
+
+void btrfs_drew_read_lock(struct btrfs_drew_lock *lock)
+{
+	atomic_inc(&lock->readers);
+
+	/*
+	 * Ensure the pending reader count is perceieved BEFORE this reader
+	 * goes to sleep in case of active writers. This guarantees new writers
+	 * won't be allowed and that the current reader will be woken up when
+	 * the last active writer finishes its jobs.
+	 */
+	smp_mb__after_atomic();
+
+	wait_event(lock->pending_readers,
+		   percpu_counter_sum(&lock->writers) == 0);
+}
+
+void btrfs_drew_read_unlock(struct btrfs_drew_lock *lock)
+{
+	/*
+	 * atomic_dec_and_test implies a full barrier, so woken up writers
+	 * are guaranteed to see the decrement
+	 */
+	if (atomic_dec_and_test(&lock->readers))
+		wake_up(&lock->pending_writers);
+}