drbd: close race when detaching from disk

BUG: unable to handle kernel NULL pointer dereference at 0000000000000058
IP: bd_release+0x21/0x70
Process drbd_w_t7146
Call Trace:
 close_bdev_exclusive
 drbd_free_ldev		[drbd]
 drbd_ldev_destroy	[drbd]
 w_after_state_ch	[drbd]

Race probably went like this:
  state.disk = D_FAILED

... first one to hit zero during D_FAILED:
   put_ldev() /* ----------------> 0 */
     i = atomic_dec_return()
     if (i == 0)
       if (state.disk == D_FAILED)
         schedule_work(go_diskless)
                                /* 1 <------ */ get_ldev_if_state()
   go_diskless()
      do_some_pre_cleanup()                     corresponding put_ldev():
      force_state(D_DISKLESS)   /* 0 <------ */ i = atomic_dec_return()
                                                if (i == 0)
        atomic_inc() /* ---------> 1 */
        state.disk = D_DISKLESS
        schedule_work(after_state_ch)           /* execution pre-empted by IRQ ? */

   after_state_ch()
     put_ldev()
       i = atomic_dec_return()  /* 0 */
       if (i == 0)
         if (state.disk == D_DISKLESS)            if (state.disk == D_DISKLESS)
           drbd_ldev_destroy()                      drbd_ldev_destroy();

Trying to fix this by checking the disk state *before* the
atomic_dec_return(), which implies memory barriers, and by inserting
extra memory barriers around the state assignment in __drbd_set_state().

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>

1 files changed, 7 insertions, 2 deletions

diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index a16f9ae3c98a..a0ffc19ccf0e 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -1946,6 +1946,11 @@ static inline bool is_sync_state(enum drbd_conns connection_state)
 
 static inline void put_ldev(struct drbd_device *device)
 {
+	enum drbd_disk_state ds = device->state.disk;
+	/* We must check the state *before* the atomic_dec becomes visible,
+	 * or we have a theoretical race where someone hitting zero,
+	 * while state still D_FAILED, will then see D_DISKLESS in the
+	 * condition below and calling into destroy, where he must not, yet. */
 	int i = atomic_dec_return(&device->local_cnt);
 
 	/* This may be called from some endio handler,
@@ -1954,10 +1959,10 @@ static inline void put_ldev(struct drbd_device *device)
 	__release(local);
 	D_ASSERT(device, i >= 0);
 	if (i == 0) {
-		if (device->state.disk == D_DISKLESS)
+		if (ds == D_DISKLESS)
 			/* even internal references gone, safe to destroy */
 			drbd_ldev_destroy(device);
-		if (device->state.disk == D_FAILED) {
+		if (ds == D_FAILED) {
 			/* all application IO references gone. */
 			if (!test_and_set_bit(GO_DISKLESS, &device->flags))
 				drbd_queue_work(&first_peer_device(device)->connection->sender_work,