bdi: make backing_dev_info->wb.dwork canceling stricter
authorTejun Heo <tj@kernel.org>
Sun, 7 Sep 2014 23:03:59 +0000 (08:03 +0900)
committerJens Axboe <axboe@fb.com>
Mon, 8 Sep 2014 16:00:39 +0000 (10:00 -0600)
Canceling of bdi->wb.dwork is currently a bit mushy.
bdi_wb_shutdown() performs cancel_delayed_work_sync() at the end after
shutting down and flushing the delayed_work and bdi_destroy() tries
yet again after bdi_unregister().

bdi->wb.dwork is queued only after checking BDI_registered while
holding bdi->wb_lock and bdi_wb_shutdown() clears the flag while
holding the same lock and then flushes the delayed_work.  There's no
way the delayed_work can be queued again after that.

Replace the two unnecessary cancel_delayed_work_sync() invocations
with WARNs on pending.  This simplifies and clarifies the code a bit
and will help future changes in further isolating bdi_writeback
handling.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@fb.com>
mm/backing-dev.c

index 4afeefe9e365e6287c6a1f57184d336cb11edfdf..cb7c5e323814015f8dd512da0f606fb35d0fdae0 100644 (file)
@@ -376,13 +376,7 @@ static void bdi_wb_shutdown(struct backing_dev_info *bdi)
        mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
        flush_delayed_work(&bdi->wb.dwork);
        WARN_ON(!list_empty(&bdi->work_list));
-
-       /*
-        * This shouldn't be necessary unless @bdi for some reason has
-        * unflushed dirty IO after work_list is drained.  Do it anyway
-        * just in case.
-        */
-       cancel_delayed_work_sync(&bdi->wb.dwork);
+       WARN_ON(delayed_work_pending(&bdi->wb.dwork));
 }
 
 /*
@@ -497,12 +491,7 @@ void bdi_destroy(struct backing_dev_info *bdi)
 
        bdi_unregister(bdi);
 
-       /*
-        * If bdi_unregister() had already been called earlier, the dwork
-        * could still be pending because bdi_prune_sb() can race with the
-        * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty().
-        */
-       cancel_delayed_work_sync(&bdi->wb.dwork);
+       WARN_ON(delayed_work_pending(&bdi->wb.dwork));
 
        for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
                percpu_counter_destroy(&bdi->bdi_stat[i]);