[PATCH] brd: reduce the brd_devices_mutex scope

[Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>]

As with commit 8b52d8be86d72308 ("loop: reorder loop_exit"),
unregister_blkdev() needs to be called first in order to avoid calling
brd_alloc() from brd_probe() after brd_del_one() from brd_exit(). Then,
we can avoid holding global mutex during add_disk()/del_gendisk() as with
commit 1c500ad706383f1a ("loop: reduce the loop_ctl_mutex scope").

Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
---
I thought that this patch can fix similar circular locking problem at __loop_clr_fd()
(e.g. https://syzkaller.appspot.com/text?tag=CrashReport&x=1026e92e300000 ) in
https://syzkaller.appspot.com/bug?id=7bb10e8b62f83e4d445cdf4c13d69e407e629558 , but
it turned out that breaking major_names_lock => brd_devices_mutex => disk->open_mutex
chain is not sufficient (it might be "block: support delayed holder registration"
that is causing major_names_lock => &disk->open_mutex => &lo->lo_mutex chain).
Thus, submitting as a normal patch.

 drivers/block/brd.c | 44 ++++++++++++++++++++++----------------------
 1 file changed, 22 insertions(+), 22 deletions(-)

diff --git a/drivers/block/brd.c b/drivers/block/brd.c
index 58ec167aa018..fceeff796e07 100644
--- a/drivers/block/brd.c
+++ b/drivers/block/brd.c
@@ -373,10 +373,22 @@ static int brd_alloc(int i)
 	struct gendisk *disk;
 	char buf[DISK_NAME_LEN];
 
+	mutex_lock(&brd_devices_mutex);
+	list_for_each_entry(brd, &brd_devices, brd_list) {
+		if (brd->brd_number != i)
+			continue;
+		mutex_unlock(&brd_devices_mutex);
+		return -EEXIST;
+	}
 	brd = kzalloc(sizeof(*brd), GFP_KERNEL);
-	if (!brd)
+	if (!brd) {
+		mutex_unlock(&brd_devices_mutex);
 		return -ENOMEM;
+	}
 	brd->brd_number		= i;
+	list_add_tail(&brd->brd_list, &brd_devices);
+	mutex_unlock(&brd_devices_mutex);
+
 	spin_lock_init(&brd->brd_lock);
 	INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
 
@@ -411,37 +423,30 @@ static int brd_alloc(int i)
 	blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
 	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
 	add_disk(disk);
-	list_add_tail(&brd->brd_list, &brd_devices);
 
 	return 0;
 
 out_free_dev:
+	mutex_lock(&brd_devices_mutex);
+	list_del(&brd->brd_list);
+	mutex_unlock(&brd_devices_mutex);
 	kfree(brd);
 	return -ENOMEM;
 }
 
 static void brd_probe(dev_t dev)
 {
-	int i = MINOR(dev) / max_part;
-	struct brd_device *brd;
-
-	mutex_lock(&brd_devices_mutex);
-	list_for_each_entry(brd, &brd_devices, brd_list) {
-		if (brd->brd_number == i)
-			goto out_unlock;
-	}
-
-	brd_alloc(i);
-out_unlock:
-	mutex_unlock(&brd_devices_mutex);
+	brd_alloc(MINOR(dev) / max_part);
 }
 
 static void brd_del_one(struct brd_device *brd)
 {
-	list_del(&brd->brd_list);
 	del_gendisk(brd->brd_disk);
 	blk_cleanup_disk(brd->brd_disk);
 	brd_free_pages(brd);
+	mutex_lock(&brd_devices_mutex);
+	list_del(&brd->brd_list);
+	mutex_unlock(&brd_devices_mutex);
 	kfree(brd);
 }
 
@@ -491,25 +496,21 @@ static int __init brd_init(void)
 
 	brd_debugfs_dir = debugfs_create_dir("ramdisk_pages", NULL);
 
-	mutex_lock(&brd_devices_mutex);
 	for (i = 0; i < rd_nr; i++) {
 		err = brd_alloc(i);
 		if (err)
 			goto out_free;
 	}
 
-	mutex_unlock(&brd_devices_mutex);
-
 	pr_info("brd: module loaded\n");
 	return 0;
 
 out_free:
+	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
 	debugfs_remove_recursive(brd_debugfs_dir);
 
 	list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
 		brd_del_one(brd);
-	mutex_unlock(&brd_devices_mutex);
-	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
 
 	pr_info("brd: module NOT loaded !!!\n");
 	return err;
@@ -519,13 +520,12 @@ static void __exit brd_exit(void)
 {
 	struct brd_device *brd, *next;
 
+	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
 	debugfs_remove_recursive(brd_debugfs_dir);
 
 	list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
 		brd_del_one(brd);
 
-	unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
-
 	pr_info("brd: module unloaded\n");
 }
 
-- 
2.30.2