5 files changed, 56 insertions, 0 deletions

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index aac3d6f4e35b..0378933d163c 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -3564,6 +3564,8 @@ struct reada_control *btrfs_reada_add(struct btrfs_root *root,
 int btrfs_reada_wait(void *handle);
 void btrfs_reada_detach(void *handle);
 int btree_readahead_hook(struct extent_buffer *eb, int err);
+void btrfs_reada_remove_dev(struct btrfs_device *dev);
+void btrfs_reada_undo_remove_dev(struct btrfs_device *dev);
 
 static inline int is_fstree(u64 rootid)
 {
diff --git a/fs/btrfs/dev-replace.c b/fs/btrfs/dev-replace.c
index 4a0243cb9d97..5b9e3f3ace22 100644
--- a/fs/btrfs/dev-replace.c
+++ b/fs/btrfs/dev-replace.c
@@ -688,6 +688,9 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	}
 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 
+	if (!scrub_ret)
+		btrfs_reada_remove_dev(src_device);
+
 	/*
 	 * We have to use this loop approach because at this point src_device
 	 * has to be available for transaction commit to complete, yet new
@@ -696,6 +699,7 @@ static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 	while (1) {
 		trans = btrfs_start_transaction(root, 0);
 		if (IS_ERR(trans)) {
+			btrfs_reada_undo_remove_dev(src_device);
 			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 			return PTR_ERR(trans);
 		}
@@ -746,6 +750,7 @@ error:
 		up_write(&dev_replace->rwsem);
 		mutex_unlock(&fs_info->chunk_mutex);
 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
+		btrfs_reada_undo_remove_dev(src_device);
 		btrfs_rm_dev_replace_blocked(fs_info);
 		if (tgt_device)
 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
diff --git a/fs/btrfs/reada.c b/fs/btrfs/reada.c
index e261c3d0cec7..d9a166eb344e 100644
--- a/fs/btrfs/reada.c
+++ b/fs/btrfs/reada.c
@@ -421,6 +421,9 @@ static struct reada_extent *reada_find_extent(struct btrfs_fs_info *fs_info,
 		if (!dev->bdev)
 			continue;
 
+		if (test_bit(BTRFS_DEV_STATE_NO_READA, &dev->dev_state))
+			continue;
+
 		if (dev_replace_is_ongoing &&
 		    dev == fs_info->dev_replace.tgtdev) {
 			/*
@@ -1022,3 +1025,45 @@ void btrfs_reada_detach(void *handle)
 
 	kref_put(&rc->refcnt, reada_control_release);
 }
+
+/*
+ * Before removing a device (device replace or device remove ioctls), call this
+ * function to wait for all existing readahead requests on the device and to
+ * make sure no one queues more readahead requests for the device.
+ *
+ * Must be called without holding neither the device list mutex nor the device
+ * replace semaphore, otherwise it will deadlock.
+ */
+void btrfs_reada_remove_dev(struct btrfs_device *dev)
+{
+	struct btrfs_fs_info *fs_info = dev->fs_info;
+
+	/* Serialize with readahead extent creation at reada_find_extent(). */
+	spin_lock(&fs_info->reada_lock);
+	set_bit(BTRFS_DEV_STATE_NO_READA, &dev->dev_state);
+	spin_unlock(&fs_info->reada_lock);
+
+	/*
+	 * There might be readahead requests added to the radix trees which
+	 * were not yet added to the readahead work queue. We need to start
+	 * them and wait for their completion, otherwise we can end up with
+	 * use-after-free problems when dropping the last reference on the
+	 * readahead extents and their zones, as they need to access the
+	 * device structure.
+	 */
+	reada_start_machine(fs_info);
+	btrfs_flush_workqueue(fs_info->readahead_workers);
+}
+
+/*
+ * If when removing a device (device replace or device remove ioctls) an error
+ * happens after calling btrfs_reada_remove_dev(), call this to undo what that
+ * function did. This is safe to call even if btrfs_reada_remove_dev() was not
+ * called before.
+ */
+void btrfs_reada_undo_remove_dev(struct btrfs_device *dev)
+{
+	spin_lock(&dev->fs_info->reada_lock);
+	clear_bit(BTRFS_DEV_STATE_NO_READA, &dev->dev_state);
+	spin_unlock(&dev->fs_info->reada_lock);
+}
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 58b9c419a2b6..1991bc5a6f59 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -2099,6 +2099,8 @@ int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
 
 	mutex_unlock(&uuid_mutex);
 	ret = btrfs_shrink_device(device, 0);
+	if (!ret)
+		btrfs_reada_remove_dev(device);
 	mutex_lock(&uuid_mutex);
 	if (ret)
 		goto error_undo;
@@ -2179,6 +2181,7 @@ out:
 	return ret;
 
 error_undo:
+	btrfs_reada_undo_remove_dev(device);
 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
 		mutex_lock(&fs_info->chunk_mutex);
 		list_add(&device->dev_alloc_list,
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index bf27ac07d315..f2177263748e 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -50,6 +50,7 @@ struct btrfs_io_geometry {
 #define BTRFS_DEV_STATE_MISSING		(2)
 #define BTRFS_DEV_STATE_REPLACE_TGT	(3)
 #define BTRFS_DEV_STATE_FLUSH_SENT	(4)
+#define BTRFS_DEV_STATE_NO_READA	(5)
 
 struct btrfs_device {
 	struct list_head dev_list; /* device_list_mutex */