bdi: make backing_dev_info->wb.dwork canceling stricter (c0ea1c22) · Commits · 戴 / test

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

+2 −13

Original line number	Diff line number	Diff line
		@@ -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]);

Admin message