rbd: avoid a deadlock on header_rwsem when flushing notifies

rbd_unregister_watch() flushes notifies and therefore cannot be called
under header_rwsem because a header update notify takes header_rwsem to
synchronize with "rbd map".  If mapping an image fails after the watch
is established and a header update notify sneaks in, we deadlock when
erroring out from rbd_dev_image_probe().

Move watch registration and unregistration out of the critical section.
The only reason they were put there was to make header_rwsem management
slightly more obvious.

Fixes: 811c66887746 ("rbd: fix rbd map vs notify races")
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Jason Dillaman <dillaman@redhat.com>

diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index 1e0a6b1..ff2377e 100644 (file)
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -4527,6 +4527,10 @@ static void cancel_tasks_sync(struct rbd_device *rbd_dev)
        cancel_work_sync(&rbd_dev->unlock_work);
 }
 
+/*
+ * header_rwsem must not be held to avoid a deadlock with
+ * rbd_dev_refresh() when flushing notifies.
+ */
 static void rbd_unregister_watch(struct rbd_device *rbd_dev)
 {
        cancel_tasks_sync(rbd_dev);
@@ -6907,6 +6911,9 @@ static void rbd_dev_image_release(struct rbd_device *rbd_dev)
  * device.  If this image is the one being mapped (i.e., not a
  * parent), initiate a watch on its header object before using that
  * object to get detailed information about the rbd image.
+ *
+ * On success, returns with header_rwsem held for write if called
+ * with @depth == 0.
  */
 static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
 {
@@ -6936,6 +6943,9 @@ static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
                }
        }
 
+       if (!depth)
+               down_write(&rbd_dev->header_rwsem);
+
        ret = rbd_dev_header_info(rbd_dev);
        if (ret) {
                if (ret == -ENOENT && !need_watch)
@@ -6987,6 +6997,8 @@ static int rbd_dev_image_probe(struct rbd_device *rbd_dev, int depth)
 err_out_probe:
        rbd_dev_unprobe(rbd_dev);
 err_out_watch:
+       if (!depth)
+               up_write(&rbd_dev->header_rwsem);
        if (need_watch)
                rbd_unregister_watch(rbd_dev);
 err_out_format:
@@ -7050,12 +7062,9 @@ static ssize_t do_rbd_add(struct bus_type *bus,
                goto err_out_rbd_dev;
        }
 
-       down_write(&rbd_dev->header_rwsem);
        rc = rbd_dev_image_probe(rbd_dev, 0);
-       if (rc < 0) {
-               up_write(&rbd_dev->header_rwsem);
+       if (rc < 0)
                goto err_out_rbd_dev;
-       }
 
        if (rbd_dev->opts->alloc_size > rbd_dev->layout.object_size) {
                rbd_warn(rbd_dev, "alloc_size adjusted to %u",