Btrfs: avoid races between super writeout and device list updates

On multi-device filesystems, btrfs writes supers to all of the devices
before considering a sync complete.  There wasn't any additional
locking between super writeout and the device list management code
because device management was done inside a transaction and
super writeout only happened  with no transation writers running.

With the btrfs fsync log and other async transaction updates, this
has been racey for some time.  This adds a mutex to protect
the device list.  The existing volume mutex could not be reused due to
transaction lock ordering requirements.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

3 files changed, 45 insertions, 2 deletions

diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 6c54c210dfd0..b7ddc77fa568 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2111,7 +2111,7 @@ static int write_dev_supers(struct btrfs_device *device,
 
 int write_all_supers(struct btrfs_root *root, int max_mirrors)
 {
-	struct list_head *head = &root->fs_info->fs_devices->devices;
+	struct list_head *head;
 	struct btrfs_device *dev;
 	struct btrfs_super_block *sb;
 	struct btrfs_dev_item *dev_item;
@@ -2126,6 +2126,9 @@ int write_all_supers(struct btrfs_root *root, int max_mirrors)
 
 	sb = &root->fs_info->super_for_commit;
 	dev_item = &sb->dev_item;
+
+	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
+	head = &root->fs_info->fs_devices->devices;
 	list_for_each_entry(dev, head, dev_list) {
 		if (!dev->bdev) {
 			total_errors++;
@@ -2169,6 +2172,7 @@ int write_all_supers(struct btrfs_root *root, int max_mirrors)
 		if (ret)
 			total_errors++;
 	}
+	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 	if (total_errors > max_errors) {
 		printk(KERN_ERR "btrfs: %d errors while writing supers\n",
 		       total_errors);
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 3f4a5932eac9..3ab80e9cd767 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -377,6 +377,7 @@ static noinline int device_list_add(const char *path,
 		memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 		fs_devices->latest_devid = devid;
 		fs_devices->latest_trans = found_transid;
+		mutex_init(&fs_devices->device_list_mutex);
 		device = NULL;
 	} else {
 		device = __find_device(&fs_devices->devices, devid,
@@ -403,7 +404,11 @@ static noinline int device_list_add(const char *path,
 			return -ENOMEM;
 		}
 		INIT_LIST_HEAD(&device->dev_alloc_list);
+
+		mutex_lock(&fs_devices->device_list_mutex);
 		list_add(&device->dev_list, &fs_devices->devices);
+		mutex_unlock(&fs_devices->device_list_mutex);
+
 		device->fs_devices = fs_devices;
 		fs_devices->num_devices++;
 	}
@@ -429,10 +434,12 @@ static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 	INIT_LIST_HEAD(&fs_devices->devices);
 	INIT_LIST_HEAD(&fs_devices->alloc_list);
 	INIT_LIST_HEAD(&fs_devices->list);
+	mutex_init(&fs_devices->device_list_mutex);
 	fs_devices->latest_devid = orig->latest_devid;
 	fs_devices->latest_trans = orig->latest_trans;
 	memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 
+	mutex_lock(&orig->device_list_mutex);
 	list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 		device = kzalloc(sizeof(*device), GFP_NOFS);
 		if (!device)
@@ -454,8 +461,10 @@ static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 		device->fs_devices = fs_devices;
 		fs_devices->num_devices++;
 	}
+	mutex_unlock(&orig->device_list_mutex);
 	return fs_devices;
 error:
+	mutex_unlock(&orig->device_list_mutex);
 	free_fs_devices(fs_devices);
 	return ERR_PTR(-ENOMEM);
 }
@@ -466,6 +475,7 @@ int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
 
 	mutex_lock(&uuid_mutex);
 again:
+	mutex_lock(&fs_devices->device_list_mutex);
 	list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 		if (device->in_fs_metadata)
 			continue;
@@ -485,6 +495,7 @@ again:
 		kfree(device->name);
 		kfree(device);
 	}
+	mutex_unlock(&fs_devices->device_list_mutex);
 
 	if (fs_devices->seed) {
 		fs_devices = fs_devices->seed;
@@ -1135,12 +1146,14 @@ int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 
 		device = NULL;
 		devices = &root->fs_info->fs_devices->devices;
+		mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 		list_for_each_entry(tmp, devices, dev_list) {
 			if (tmp->in_fs_metadata && !tmp->bdev) {
 				device = tmp;
 				break;
 			}
 		}
+		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 		bdev = NULL;
 		bh = NULL;
 		disk_super = NULL;
@@ -1195,7 +1208,16 @@ int btrfs_rm_device(struct btrfs_root *root, char *device_path)
 		goto error_brelse;
 
 	device->in_fs_metadata = 0;
+
+	/*
+	 * the device list mutex makes sure that we don't change
+	 * the device list while someone else is writing out all
+	 * the device supers.
+	 */
+	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 	list_del_init(&device->dev_list);
+	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
+
 	device->fs_devices->num_devices--;
 
 	next_device = list_entry(root->fs_info->fs_devices->devices.next,
@@ -1289,6 +1311,7 @@ static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
 	seed_devices->opened = 1;
 	INIT_LIST_HEAD(&seed_devices->devices);
 	INIT_LIST_HEAD(&seed_devices->alloc_list);
+	mutex_init(&seed_devices->device_list_mutex);
 	list_splice_init(&fs_devices->devices, &seed_devices->devices);
 	list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
 	list_for_each_entry(device, &seed_devices->devices, dev_list) {
@@ -1414,6 +1437,10 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 	mutex_lock(&root->fs_info->volume_mutex);
 
 	devices = &root->fs_info->fs_devices->devices;
+	/*
+	 * we have the volume lock, so we don't need the extra
+	 * device list mutex while reading the list here.
+	 */
 	list_for_each_entry(device, devices, dev_list) {
 		if (device->bdev == bdev) {
 			ret = -EEXIST;
@@ -1468,6 +1495,12 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 	}
 
 	device->fs_devices = root->fs_info->fs_devices;
+
+	/*
+	 * we don't want write_supers to jump in here with our device
+	 * half setup
+	 */
+	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
 	list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
 	list_add(&device->dev_alloc_list,
 		 &root->fs_info->fs_devices->alloc_list);
@@ -1486,6 +1519,7 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
 	total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
 	btrfs_set_super_num_devices(&root->fs_info->super_copy,
 				    total_bytes + 1);
+	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
 
 	if (seeding_dev) {
 		ret = init_first_rw_device(trans, root, device);
diff --git a/fs/btrfs/volumes.h b/fs/btrfs/volumes.h
index 3c1f7310421e..5139a833f721 100644
--- a/fs/btrfs/volumes.h
+++ b/fs/btrfs/volumes.h
@@ -96,7 +96,12 @@ struct btrfs_fs_devices {
 	u64 rw_devices;
 	u64 total_rw_bytes;
 	struct block_device *latest_bdev;
-	/* all of the devices in the FS */
+
+	/* all of the devices in the FS, protected by a mutex
+	 * so we can safely walk it to write out the supers without
+	 * worrying about add/remove by the multi-device code
+	 */
+	struct mutex device_list_mutex;
 	struct list_head devices;
 
 	/* devices not currently being allocated */