linux-master/mm/backing-dev.c

63 			     struct bdi_writeback *wb)
67 	spin_lock(&wb->list_lock);
68 	list_for_each_entry(inode, &wb->b_dirty, i_io_list)
70 	list_for_each_entry(inode, &wb->b_io, i_io_list)
72 	list_for_each_entry(inode, &wb->b_more_io, i_io_list)
74 	list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
77 	spin_unlock(&wb->list_lock);
79 	stats->nr_writeback += wb_stat(wb, WB_WRITEBACK);
80 	stats->nr_reclaimable += wb_stat(wb, WB_RECLAIMABLE);
81 	stats->nr_dirtied += wb_stat(wb, WB_DIRTIED);
82 	stats->nr_written += wb_stat(wb, WB_WRITTEN);
83 	stats->wb_thresh += wb_calc_thresh(wb, stats->dirty_thresh);
90 	struct bdi_writeback *wb;
93 	list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) {
94 		if (!wb_tryget(wb))
97 		collect_wb_stats(stats, wb);
98 		wb_put(wb);
106 	collect_wb_stats(stats, &bdi->wb);
152 		   !list_empty(&bdi->bdi_list), bdi->wb.state);
158 static void wb_stats_show(struct seq_file *m, struct bdi_writeback *wb,
176 		   cgroup_ino(wb->memcg_css->cgroup),
185 		   K(wb->avg_write_bandwidth),
190 		   wb->state);
198 	struct bdi_writeback *wb;
203 	list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) {
206 		if (!wb_tryget(wb))
209 		collect_wb_stats(&stats, wb);
212 		 * Calculate thresh of wb in writeback cgroup which is min of
217 		if (mem_cgroup_wb_domain(wb)) {
219 			stats.wb_thresh = min(stats.wb_thresh, cgwb_calc_thresh(wb));
223 		wb_stats_show(m, wb, &stats);
225 		wb_put(wb);
504 	struct bdi_writeback *wb = container_of(to_delayed_work(work),
507 	wb_update_bandwidth(wb);
515 static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
520 	memset(wb, 0, sizeof(*wb));
522 	wb->bdi = bdi;
523 	wb->last_old_flush = jiffies;
524 	INIT_LIST_HEAD(&wb->b_dirty);
525 	INIT_LIST_HEAD(&wb->b_io);
526 	INIT_LIST_HEAD(&wb->b_more_io);
527 	INIT_LIST_HEAD(&wb->b_dirty_time);
528 	spin_lock_init(&wb->list_lock);
530 	atomic_set(&wb->writeback_inodes, 0);
531 	wb->bw_time_stamp = jiffies;
532 	wb->balanced_dirty_ratelimit = INIT_BW;
533 	wb->dirty_ratelimit = INIT_BW;
534 	wb->write_bandwidth = INIT_BW;
535 	wb->avg_write_bandwidth = INIT_BW;
537 	spin_lock_init(&wb->work_lock);
538 	INIT_LIST_HEAD(&wb->work_list);
539 	INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
540 	INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn);
542 	err = fprop_local_init_percpu(&wb->completions, gfp);
546 	err = percpu_counter_init_many(wb->stat, 0, gfp, NR_WB_STAT_ITEMS);
548 		fprop_local_destroy_percpu(&wb->completions);
553 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
558 static void wb_shutdown(struct bdi_writeback *wb)
561 	spin_lock_irq(&wb->work_lock);
562 	if (!test_and_clear_bit(WB_registered, &wb->state)) {
563 		spin_unlock_irq(&wb->work_lock);
566 	spin_unlock_irq(&wb->work_lock);
568 	cgwb_remove_from_bdi_list(wb);
571 	 * tells wb_workfn() that @wb is dying and its work_list needs to
574 	mod_delayed_work(bdi_wq, &wb->dwork, 0);
575 	flush_delayed_work(&wb->dwork);
576 	WARN_ON(!list_empty(&wb->work_list));
577 	flush_delayed_work(&wb->bw_dwork);
580 static void wb_exit(struct bdi_writeback *wb)
582 	WARN_ON(delayed_work_pending(&wb->dwork));
583 	percpu_counter_destroy_many(wb->stat, NR_WB_STAT_ITEMS);
584 	fprop_local_destroy_percpu(&wb->completions);
604 	struct bdi_writeback *wb = container_of(rcu_head,
607 	percpu_ref_exit(&wb->refcnt);
608 	kfree(wb);
613 	struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
615 	struct backing_dev_info *bdi = wb->bdi;
617 	mutex_lock(&wb->bdi->cgwb_release_mutex);
618 	wb_shutdown(wb);
620 	css_put(wb->memcg_css);
621 	css_put(wb->blkcg_css);
622 	mutex_unlock(&wb->bdi->cgwb_release_mutex);
625 	blkcg_unpin_online(wb->blkcg_css);
627 	fprop_local_destroy_percpu(&wb->memcg_completions);
630 	list_del(&wb->offline_node);
633 	wb_exit(wb);
635 	WARN_ON_ONCE(!list_empty(&wb->b_attached));
636 	call_rcu(&wb->rcu, cgwb_free_rcu);
641 	struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
643 	queue_work(cgwb_release_wq, &wb->release_work);
646 static void cgwb_kill(struct bdi_writeback *wb)
650 	WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
651 	list_del(&wb->memcg_node);
652 	list_del(&wb->blkcg_node);
653 	list_add(&wb->offline_node, &offline_cgwbs);
654 	percpu_ref_kill(&wb->refcnt);
657 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
660 	list_del_rcu(&wb->bdi_node);
670 	struct bdi_writeback *wb;
681 	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
682 	if (wb && wb->blkcg_css != blkcg_css) {
683 		cgwb_kill(wb);
684 		wb = NULL;
687 	if (wb)
691 	wb = kmalloc(sizeof(*wb), gfp);
692 	if (!wb) {
697 	ret = wb_init(wb, bdi, gfp);
701 	ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
705 	ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
709 	wb->memcg_css = memcg_css;
710 	wb->blkcg_css = blkcg_css;
711 	INIT_LIST_HEAD(&wb->b_attached);
712 	INIT_WORK(&wb->release_work, cgwb_release_workfn);
713 	set_bit(WB_registered, &wb->state);
717 	 * The root wb determines the registered state of the whole bdi and
719 	 * whether they're still online.  Don't link @wb if any is dead.
724 	if (test_bit(WB_registered, &bdi->wb.state) &&
727 		ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
729 			list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
730 			list_add(&wb->memcg_node, memcg_cgwb_list);
731 			list_add(&wb->blkcg_node, blkcg_cgwb_list);
747 	fprop_local_destroy_percpu(&wb->memcg_completions);
749 	percpu_ref_exit(&wb->refcnt);
751 	wb_exit(wb);
753 	kfree(wb);
760  * wb_get_lookup - get wb for a given memcg
764  * Try to get the wb for @memcg_css on @bdi.  The returned wb has its
771  * A wb is keyed by its associated memcg.  As blkcg implicitly enables
777  * and disabled closer to root in the hierarchy, each wb keeps track of
779  * each lookup.  On mismatch, the existing wb is discarded and a new one is
785 	struct bdi_writeback *wb;
788 		return &bdi->wb;
791 	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
792 	if (wb) {
797 		if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
798 			wb = NULL;
803 	return wb;
807  * wb_get_create - get wb for a given memcg, create if necessary
812  * Try to get the wb for @memcg_css on @bdi.  If it doesn't exist, try to
819 	struct bdi_writeback *wb;
824 		wb = wb_get_lookup(bdi, memcg_css);
825 	} while (!wb && !cgwb_create(bdi, memcg_css, gfp));
827 	return wb;
838 	ret = wb_init(&bdi->wb, bdi, GFP_KERNEL);
840 		bdi->wb.memcg_css = &root_mem_cgroup->css;
841 		bdi->wb.blkcg_css = blkcg_root_css;
850 	struct bdi_writeback *wb;
852 	WARN_ON(test_bit(WB_registered, &bdi->wb.state));
862 		wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
865 		wb_shutdown(wb);
881 	struct bdi_writeback *wb;
887 		wb = list_first_entry(&offline_cgwbs, struct bdi_writeback,
889 		list_move(&wb->offline_node, &processed);
892 		 * If wb is dirty, cleaning up the writeback by switching
900 		if (wb_has_dirty_io(wb))
903 		if (!wb_tryget(wb))
907 		while (cleanup_offline_cgwb(wb))
911 		wb_put(wb);
921  * wb_memcg_offline - kill all wb's associated with a memcg being offlined
924  * Also prevents creation of any new wb's associated with @memcg.
929 	struct bdi_writeback *wb, *next;
932 	list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
933 		cgwb_kill(wb);
934 	memcg_cgwb_list->next = NULL;	/* prevent new wb's */
941  * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
944  * Also prevents creation of any new wb's associated with @blkcg.
948 	struct bdi_writeback *wb, *next;
952 	list_for_each_entry_safe(wb, next, list, blkcg_node)
953 		cgwb_kill(wb);
954 	list->next = NULL;	/* prevent new wb's */
961 	list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
984 	return wb_init(&bdi->wb, bdi, GFP_KERNEL);
991 	list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
994 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
996 	list_del_rcu(&wb->bdi_node);
1102 	set_bit(WB_registered, &bdi->wb.state);
1158 	wb_shutdown(&bdi->wb);
1186 	WARN_ON_ONCE(test_bit(WB_registered, &bdi->wb.state));
1188 	wb_exit(&bdi->wb);