Merge tag 'for-6.0-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs fixes from David Sterba:

 - two fixes for hangs in the umount sequence where threads depend on
   each other and the work must be finished in the right order

 - in zoned mode, wait for flushing all block group metadata IO before
   finishing the zone

* tag 'for-6.0-rc6-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: zoned: wait for extent buffer IOs before finishing a zone
  btrfs: fix hang during unmount when stopping a space reclaim worker
  btrfs: fix hang during unmount when stopping block group reclaim worker

2 files changed, 74 insertions, 8 deletions

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 1af28b066b42..2633137c3e9f 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4475,6 +4475,17 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
 
 	/*
+	 * If we had UNFINISHED_DROPS we could still be processing them, so
+	 * clear that bit and wake up relocation so it can stop.
+	 * We must do this before stopping the block group reclaim task, because
+	 * at btrfs_relocate_block_group() we wait for this bit, and after the
+	 * wait we stop with -EINTR if btrfs_fs_closing() returns non-zero - we
+	 * have just set BTRFS_FS_CLOSING_START, so btrfs_fs_closing() will
+	 * return 1.
+	 */
+	btrfs_wake_unfinished_drop(fs_info);
+
+	/*
 	 * We may have the reclaim task running and relocating a data block group,
 	 * in which case it may create delayed iputs. So stop it before we park
 	 * the cleaner kthread otherwise we can get new delayed iputs after
@@ -4492,12 +4503,6 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	 */
 	kthread_park(fs_info->cleaner_kthread);
 
-	/*
-	 * If we had UNFINISHED_DROPS we could still be processing them, so
-	 * clear that bit and wake up relocation so it can stop.
-	 */
-	btrfs_wake_unfinished_drop(fs_info);
-
 	/* wait for the qgroup rescan worker to stop */
 	btrfs_qgroup_wait_for_completion(fs_info, false);
 
@@ -4520,6 +4525,31 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	/* clear out the rbtree of defraggable inodes */
 	btrfs_cleanup_defrag_inodes(fs_info);
 
+	/*
+	 * After we parked the cleaner kthread, ordered extents may have
+	 * completed and created new delayed iputs. If one of the async reclaim
+	 * tasks is running and in the RUN_DELAYED_IPUTS flush state, then we
+	 * can hang forever trying to stop it, because if a delayed iput is
+	 * added after it ran btrfs_run_delayed_iputs() and before it called
+	 * btrfs_wait_on_delayed_iputs(), it will hang forever since there is
+	 * no one else to run iputs.
+	 *
+	 * So wait for all ongoing ordered extents to complete and then run
+	 * delayed iputs. This works because once we reach this point no one
+	 * can either create new ordered extents nor create delayed iputs
+	 * through some other means.
+	 *
+	 * Also note that btrfs_wait_ordered_roots() is not safe here, because
+	 * it waits for BTRFS_ORDERED_COMPLETE to be set on an ordered extent,
+	 * but the delayed iput for the respective inode is made only when doing
+	 * the final btrfs_put_ordered_extent() (which must happen at
+	 * btrfs_finish_ordered_io() when we are unmounting).
+	 */
+	btrfs_flush_workqueue(fs_info->endio_write_workers);
+	/* Ordered extents for free space inodes. */
+	btrfs_flush_workqueue(fs_info->endio_freespace_worker);
+	btrfs_run_delayed_iputs(fs_info);
+
 	cancel_work_sync(&fs_info->async_reclaim_work);
 	cancel_work_sync(&fs_info->async_data_reclaim_work);
 	cancel_work_sync(&fs_info->preempt_reclaim_work);
diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c
index 62e7007a7e46..73c6929f7be6 100644
--- a/fs/btrfs/zoned.c
+++ b/fs/btrfs/zoned.c
@@ -1918,10 +1918,44 @@ out_unlock:
 	return ret;
 }
 
+static void wait_eb_writebacks(struct btrfs_block_group *block_group)
+{
+	struct btrfs_fs_info *fs_info = block_group->fs_info;
+	const u64 end = block_group->start + block_group->length;
+	struct radix_tree_iter iter;
+	struct extent_buffer *eb;
+	void __rcu **slot;
+
+	rcu_read_lock();
+	radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter,
+				 block_group->start >> fs_info->sectorsize_bits) {
+		eb = radix_tree_deref_slot(slot);
+		if (!eb)
+			continue;
+		if (radix_tree_deref_retry(eb)) {
+			slot = radix_tree_iter_retry(&iter);
+			continue;
+		}
+
+		if (eb->start < block_group->start)
+			continue;
+		if (eb->start >= end)
+			break;
+
+		slot = radix_tree_iter_resume(slot, &iter);
+		rcu_read_unlock();
+		wait_on_extent_buffer_writeback(eb);
+		rcu_read_lock();
+	}
+	rcu_read_unlock();
+}
+
 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
 {
 	struct btrfs_fs_info *fs_info = block_group->fs_info;
 	struct map_lookup *map;
+	const bool is_metadata = (block_group->flags &
+			(BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM));
 	int ret = 0;
 	int i;
 
@@ -1932,8 +1966,7 @@ static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_writ
 	}
 
 	/* Check if we have unwritten allocated space */
-	if ((block_group->flags &
-	     (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)) &&
+	if (is_metadata &&
 	    block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
 		spin_unlock(&block_group->lock);
 		return -EAGAIN;
@@ -1958,6 +1991,9 @@ static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_writ
 		/* No need to wait for NOCOW writers. Zoned mode does not allow that */
 		btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
 					 block_group->length);
+		/* Wait for extent buffers to be written. */
+		if (is_metadata)
+			wait_eb_writebacks(block_group);
 
 		spin_lock(&block_group->lock);