1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/blkdev.h>
4 #include <linux/wait.h>
5 #include <linux/rbtree.h>
6 #include <linux/kthread.h>
7 #include <linux/backing-dev.h>
8 #include <linux/blk-cgroup.h>
9 #include <linux/freezer.h>
11 #include <linux/pagemap.h>
13 #include <linux/sched/mm.h>
14 #include <linux/sched.h>
15 #include <linux/module.h>
16 #include <linux/writeback.h>
17 #include <linux/device.h>
18 #include <trace/events/writeback.h>
21 struct backing_dev_info noop_backing_dev_info;
22 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
24 static const char *bdi_unknown_name = "(unknown)";
27 * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU
28 * reader side locking.
30 DEFINE_SPINLOCK(bdi_lock);
31 static u64 bdi_id_cursor;
32 static struct rb_root bdi_tree = RB_ROOT;
35 /* bdi_wq serves all asynchronous writeback tasks */
36 struct workqueue_struct *bdi_wq;
38 #ifdef CONFIG_DEBUG_FS
39 #include <linux/debugfs.h>
40 #include <linux/seq_file.h>
42 static struct dentry *bdi_debug_root;
44 static void bdi_debug_init(void)
46 bdi_debug_root = debugfs_create_dir("bdi", NULL);
49 static int bdi_debug_stats_show(struct seq_file *m, void *v)
51 struct backing_dev_info *bdi = m->private;
52 struct bdi_writeback *wb = &bdi->wb;
53 unsigned long background_thresh;
54 unsigned long dirty_thresh;
55 unsigned long wb_thresh;
56 unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
59 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
60 spin_lock(&wb->list_lock);
61 list_for_each_entry(inode, &wb->b_dirty, i_io_list)
63 list_for_each_entry(inode, &wb->b_io, i_io_list)
65 list_for_each_entry(inode, &wb->b_more_io, i_io_list)
67 list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
68 if (inode->i_state & I_DIRTY_TIME)
70 spin_unlock(&wb->list_lock);
72 global_dirty_limits(&background_thresh, &dirty_thresh);
73 wb_thresh = wb_calc_thresh(wb, dirty_thresh);
76 "BdiWriteback: %10lu kB\n"
77 "BdiReclaimable: %10lu kB\n"
78 "BdiDirtyThresh: %10lu kB\n"
79 "DirtyThresh: %10lu kB\n"
80 "BackgroundThresh: %10lu kB\n"
81 "BdiDirtied: %10lu kB\n"
82 "BdiWritten: %10lu kB\n"
83 "BdiWriteBandwidth: %10lu kBps\n"
87 "b_dirty_time: %10lu\n"
90 (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
91 (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
95 (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
96 (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
97 (unsigned long) K(wb->write_bandwidth),
102 !list_empty(&bdi->bdi_list), bdi->wb.state);
106 DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats);
108 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
110 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
112 debugfs_create_file("stats", 0444, bdi->debug_dir, bdi,
113 &bdi_debug_stats_fops);
116 static void bdi_debug_unregister(struct backing_dev_info *bdi)
118 debugfs_remove_recursive(bdi->debug_dir);
121 static inline void bdi_debug_init(void)
124 static inline void bdi_debug_register(struct backing_dev_info *bdi,
128 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
133 static ssize_t read_ahead_kb_store(struct device *dev,
134 struct device_attribute *attr,
135 const char *buf, size_t count)
137 struct backing_dev_info *bdi = dev_get_drvdata(dev);
138 unsigned long read_ahead_kb;
141 ret = kstrtoul(buf, 10, &read_ahead_kb);
145 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
150 #define BDI_SHOW(name, expr) \
151 static ssize_t name##_show(struct device *dev, \
152 struct device_attribute *attr, char *buf) \
154 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
156 return sysfs_emit(buf, "%lld\n", (long long)expr); \
158 static DEVICE_ATTR_RW(name);
160 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
162 static ssize_t min_ratio_store(struct device *dev,
163 struct device_attribute *attr, const char *buf, size_t count)
165 struct backing_dev_info *bdi = dev_get_drvdata(dev);
169 ret = kstrtouint(buf, 10, &ratio);
173 ret = bdi_set_min_ratio(bdi, ratio);
179 BDI_SHOW(min_ratio, bdi->min_ratio / BDI_RATIO_SCALE)
181 static ssize_t min_ratio_fine_store(struct device *dev,
182 struct device_attribute *attr, const char *buf, size_t count)
184 struct backing_dev_info *bdi = dev_get_drvdata(dev);
188 ret = kstrtouint(buf, 10, &ratio);
192 ret = bdi_set_min_ratio_no_scale(bdi, ratio);
198 BDI_SHOW(min_ratio_fine, bdi->min_ratio)
200 static ssize_t max_ratio_store(struct device *dev,
201 struct device_attribute *attr, const char *buf, size_t count)
203 struct backing_dev_info *bdi = dev_get_drvdata(dev);
207 ret = kstrtouint(buf, 10, &ratio);
211 ret = bdi_set_max_ratio(bdi, ratio);
217 BDI_SHOW(max_ratio, bdi->max_ratio / BDI_RATIO_SCALE)
219 static ssize_t max_ratio_fine_store(struct device *dev,
220 struct device_attribute *attr, const char *buf, size_t count)
222 struct backing_dev_info *bdi = dev_get_drvdata(dev);
226 ret = kstrtouint(buf, 10, &ratio);
230 ret = bdi_set_max_ratio_no_scale(bdi, ratio);
236 BDI_SHOW(max_ratio_fine, bdi->max_ratio)
238 static ssize_t min_bytes_show(struct device *dev,
239 struct device_attribute *attr,
242 struct backing_dev_info *bdi = dev_get_drvdata(dev);
244 return sysfs_emit(buf, "%llu\n", bdi_get_min_bytes(bdi));
247 static ssize_t min_bytes_store(struct device *dev,
248 struct device_attribute *attr, const char *buf, size_t count)
250 struct backing_dev_info *bdi = dev_get_drvdata(dev);
254 ret = kstrtoull(buf, 10, &bytes);
258 ret = bdi_set_min_bytes(bdi, bytes);
264 static DEVICE_ATTR_RW(min_bytes);
266 static ssize_t max_bytes_show(struct device *dev,
267 struct device_attribute *attr,
270 struct backing_dev_info *bdi = dev_get_drvdata(dev);
272 return sysfs_emit(buf, "%llu\n", bdi_get_max_bytes(bdi));
275 static ssize_t max_bytes_store(struct device *dev,
276 struct device_attribute *attr, const char *buf, size_t count)
278 struct backing_dev_info *bdi = dev_get_drvdata(dev);
282 ret = kstrtoull(buf, 10, &bytes);
286 ret = bdi_set_max_bytes(bdi, bytes);
292 static DEVICE_ATTR_RW(max_bytes);
294 static ssize_t stable_pages_required_show(struct device *dev,
295 struct device_attribute *attr,
299 "the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n");
300 return sysfs_emit(buf, "%d\n", 0);
302 static DEVICE_ATTR_RO(stable_pages_required);
304 static ssize_t strict_limit_store(struct device *dev,
305 struct device_attribute *attr, const char *buf, size_t count)
307 struct backing_dev_info *bdi = dev_get_drvdata(dev);
308 unsigned int strict_limit;
311 ret = kstrtouint(buf, 10, &strict_limit);
315 ret = bdi_set_strict_limit(bdi, strict_limit);
322 static ssize_t strict_limit_show(struct device *dev,
323 struct device_attribute *attr, char *buf)
325 struct backing_dev_info *bdi = dev_get_drvdata(dev);
327 return sysfs_emit(buf, "%d\n",
328 !!(bdi->capabilities & BDI_CAP_STRICTLIMIT));
330 static DEVICE_ATTR_RW(strict_limit);
332 static struct attribute *bdi_dev_attrs[] = {
333 &dev_attr_read_ahead_kb.attr,
334 &dev_attr_min_ratio.attr,
335 &dev_attr_min_ratio_fine.attr,
336 &dev_attr_max_ratio.attr,
337 &dev_attr_max_ratio_fine.attr,
338 &dev_attr_min_bytes.attr,
339 &dev_attr_max_bytes.attr,
340 &dev_attr_stable_pages_required.attr,
341 &dev_attr_strict_limit.attr,
344 ATTRIBUTE_GROUPS(bdi_dev);
346 static const struct class bdi_class = {
348 .dev_groups = bdi_dev_groups,
351 static __init int bdi_class_init(void)
355 ret = class_register(&bdi_class);
363 postcore_initcall(bdi_class_init);
365 static int __init default_bdi_init(void)
367 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND |
373 subsys_initcall(default_bdi_init);
375 static void wb_update_bandwidth_workfn(struct work_struct *work)
377 struct bdi_writeback *wb = container_of(to_delayed_work(work),
378 struct bdi_writeback, bw_dwork);
380 wb_update_bandwidth(wb);
384 * Initial write bandwidth: 100 MB/s
386 #define INIT_BW (100 << (20 - PAGE_SHIFT))
388 static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
393 memset(wb, 0, sizeof(*wb));
396 wb->last_old_flush = jiffies;
397 INIT_LIST_HEAD(&wb->b_dirty);
398 INIT_LIST_HEAD(&wb->b_io);
399 INIT_LIST_HEAD(&wb->b_more_io);
400 INIT_LIST_HEAD(&wb->b_dirty_time);
401 spin_lock_init(&wb->list_lock);
403 atomic_set(&wb->writeback_inodes, 0);
404 wb->bw_time_stamp = jiffies;
405 wb->balanced_dirty_ratelimit = INIT_BW;
406 wb->dirty_ratelimit = INIT_BW;
407 wb->write_bandwidth = INIT_BW;
408 wb->avg_write_bandwidth = INIT_BW;
410 spin_lock_init(&wb->work_lock);
411 INIT_LIST_HEAD(&wb->work_list);
412 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
413 INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn);
415 err = fprop_local_init_percpu(&wb->completions, gfp);
419 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
420 err = percpu_counter_init(&wb->stat[i], 0, gfp);
422 goto out_destroy_stat;
429 percpu_counter_destroy(&wb->stat[i]);
430 fprop_local_destroy_percpu(&wb->completions);
434 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
437 * Remove bdi from the global list and shutdown any threads we have running
439 static void wb_shutdown(struct bdi_writeback *wb)
441 /* Make sure nobody queues further work */
442 spin_lock_irq(&wb->work_lock);
443 if (!test_and_clear_bit(WB_registered, &wb->state)) {
444 spin_unlock_irq(&wb->work_lock);
447 spin_unlock_irq(&wb->work_lock);
449 cgwb_remove_from_bdi_list(wb);
451 * Drain work list and shutdown the delayed_work. !WB_registered
452 * tells wb_workfn() that @wb is dying and its work_list needs to
453 * be drained no matter what.
455 mod_delayed_work(bdi_wq, &wb->dwork, 0);
456 flush_delayed_work(&wb->dwork);
457 WARN_ON(!list_empty(&wb->work_list));
458 flush_delayed_work(&wb->bw_dwork);
461 static void wb_exit(struct bdi_writeback *wb)
465 WARN_ON(delayed_work_pending(&wb->dwork));
467 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
468 percpu_counter_destroy(&wb->stat[i]);
470 fprop_local_destroy_percpu(&wb->completions);
473 #ifdef CONFIG_CGROUP_WRITEBACK
475 #include <linux/memcontrol.h>
478 * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and
479 * memcg->cgwb_list. bdi->cgwb_tree is also RCU protected.
481 static DEFINE_SPINLOCK(cgwb_lock);
482 static struct workqueue_struct *cgwb_release_wq;
484 static LIST_HEAD(offline_cgwbs);
485 static void cleanup_offline_cgwbs_workfn(struct work_struct *work);
486 static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn);
488 static void cgwb_free_rcu(struct rcu_head *rcu_head)
490 struct bdi_writeback *wb = container_of(rcu_head,
491 struct bdi_writeback, rcu);
493 percpu_ref_exit(&wb->refcnt);
497 static void cgwb_release_workfn(struct work_struct *work)
499 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
501 struct backing_dev_info *bdi = wb->bdi;
503 mutex_lock(&wb->bdi->cgwb_release_mutex);
506 css_put(wb->memcg_css);
507 css_put(wb->blkcg_css);
508 mutex_unlock(&wb->bdi->cgwb_release_mutex);
510 /* triggers blkg destruction if no online users left */
511 blkcg_unpin_online(wb->blkcg_css);
513 fprop_local_destroy_percpu(&wb->memcg_completions);
515 spin_lock_irq(&cgwb_lock);
516 list_del(&wb->offline_node);
517 spin_unlock_irq(&cgwb_lock);
521 WARN_ON_ONCE(!list_empty(&wb->b_attached));
522 call_rcu(&wb->rcu, cgwb_free_rcu);
525 static void cgwb_release(struct percpu_ref *refcnt)
527 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
529 queue_work(cgwb_release_wq, &wb->release_work);
532 static void cgwb_kill(struct bdi_writeback *wb)
534 lockdep_assert_held(&cgwb_lock);
536 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
537 list_del(&wb->memcg_node);
538 list_del(&wb->blkcg_node);
539 list_add(&wb->offline_node, &offline_cgwbs);
540 percpu_ref_kill(&wb->refcnt);
543 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
545 spin_lock_irq(&cgwb_lock);
546 list_del_rcu(&wb->bdi_node);
547 spin_unlock_irq(&cgwb_lock);
550 static int cgwb_create(struct backing_dev_info *bdi,
551 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
553 struct mem_cgroup *memcg;
554 struct cgroup_subsys_state *blkcg_css;
555 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
556 struct bdi_writeback *wb;
560 memcg = mem_cgroup_from_css(memcg_css);
561 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
562 memcg_cgwb_list = &memcg->cgwb_list;
563 blkcg_cgwb_list = blkcg_get_cgwb_list(blkcg_css);
565 /* look up again under lock and discard on blkcg mismatch */
566 spin_lock_irqsave(&cgwb_lock, flags);
567 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
568 if (wb && wb->blkcg_css != blkcg_css) {
572 spin_unlock_irqrestore(&cgwb_lock, flags);
576 /* need to create a new one */
577 wb = kmalloc(sizeof(*wb), gfp);
583 ret = wb_init(wb, bdi, gfp);
587 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
591 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
595 wb->memcg_css = memcg_css;
596 wb->blkcg_css = blkcg_css;
597 INIT_LIST_HEAD(&wb->b_attached);
598 INIT_WORK(&wb->release_work, cgwb_release_workfn);
599 set_bit(WB_registered, &wb->state);
603 * The root wb determines the registered state of the whole bdi and
604 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
605 * whether they're still online. Don't link @wb if any is dead.
606 * See wb_memcg_offline() and wb_blkcg_offline().
609 spin_lock_irqsave(&cgwb_lock, flags);
610 if (test_bit(WB_registered, &bdi->wb.state) &&
611 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
612 /* we might have raced another instance of this function */
613 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
615 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
616 list_add(&wb->memcg_node, memcg_cgwb_list);
617 list_add(&wb->blkcg_node, blkcg_cgwb_list);
618 blkcg_pin_online(blkcg_css);
623 spin_unlock_irqrestore(&cgwb_lock, flags);
633 fprop_local_destroy_percpu(&wb->memcg_completions);
635 percpu_ref_exit(&wb->refcnt);
646 * wb_get_lookup - get wb for a given memcg
648 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
650 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
651 * refcount incremented.
653 * This function uses css_get() on @memcg_css and thus expects its refcnt
654 * to be positive on invocation. IOW, rcu_read_lock() protection on
655 * @memcg_css isn't enough. try_get it before calling this function.
657 * A wb is keyed by its associated memcg. As blkcg implicitly enables
658 * memcg on the default hierarchy, memcg association is guaranteed to be
659 * more specific (equal or descendant to the associated blkcg) and thus can
660 * identify both the memcg and blkcg associations.
662 * Because the blkcg associated with a memcg may change as blkcg is enabled
663 * and disabled closer to root in the hierarchy, each wb keeps track of
664 * both the memcg and blkcg associated with it and verifies the blkcg on
665 * each lookup. On mismatch, the existing wb is discarded and a new one is
668 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
669 struct cgroup_subsys_state *memcg_css)
671 struct bdi_writeback *wb;
673 if (!memcg_css->parent)
677 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
679 struct cgroup_subsys_state *blkcg_css;
681 /* see whether the blkcg association has changed */
682 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
683 if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
693 * wb_get_create - get wb for a given memcg, create if necessary
695 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
696 * @gfp: allocation mask to use
698 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
699 * create one. See wb_get_lookup() for more details.
701 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
702 struct cgroup_subsys_state *memcg_css,
705 struct bdi_writeback *wb;
710 wb = wb_get_lookup(bdi, memcg_css);
711 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
716 static int cgwb_bdi_init(struct backing_dev_info *bdi)
720 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
721 mutex_init(&bdi->cgwb_release_mutex);
722 init_rwsem(&bdi->wb_switch_rwsem);
724 ret = wb_init(&bdi->wb, bdi, GFP_KERNEL);
726 bdi->wb.memcg_css = &root_mem_cgroup->css;
727 bdi->wb.blkcg_css = blkcg_root_css;
732 static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
734 struct radix_tree_iter iter;
736 struct bdi_writeback *wb;
738 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
740 spin_lock_irq(&cgwb_lock);
741 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
743 spin_unlock_irq(&cgwb_lock);
745 mutex_lock(&bdi->cgwb_release_mutex);
746 spin_lock_irq(&cgwb_lock);
747 while (!list_empty(&bdi->wb_list)) {
748 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
750 spin_unlock_irq(&cgwb_lock);
752 spin_lock_irq(&cgwb_lock);
754 spin_unlock_irq(&cgwb_lock);
755 mutex_unlock(&bdi->cgwb_release_mutex);
759 * cleanup_offline_cgwbs_workfn - try to release dying cgwbs
761 * Try to release dying cgwbs by switching attached inodes to the nearest
762 * living ancestor's writeback. Processed wbs are placed at the end
763 * of the list to guarantee the forward progress.
765 static void cleanup_offline_cgwbs_workfn(struct work_struct *work)
767 struct bdi_writeback *wb;
768 LIST_HEAD(processed);
770 spin_lock_irq(&cgwb_lock);
772 while (!list_empty(&offline_cgwbs)) {
773 wb = list_first_entry(&offline_cgwbs, struct bdi_writeback,
775 list_move(&wb->offline_node, &processed);
778 * If wb is dirty, cleaning up the writeback by switching
779 * attached inodes will result in an effective removal of any
780 * bandwidth restrictions, which isn't the goal. Instead,
781 * it can be postponed until the next time, when all io
782 * will be likely completed. If in the meantime some inodes
783 * will get re-dirtied, they should be eventually switched to
786 if (wb_has_dirty_io(wb))
792 spin_unlock_irq(&cgwb_lock);
793 while (cleanup_offline_cgwb(wb))
795 spin_lock_irq(&cgwb_lock);
800 if (!list_empty(&processed))
801 list_splice_tail(&processed, &offline_cgwbs);
803 spin_unlock_irq(&cgwb_lock);
807 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
808 * @memcg: memcg being offlined
810 * Also prevents creation of any new wb's associated with @memcg.
812 void wb_memcg_offline(struct mem_cgroup *memcg)
814 struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
815 struct bdi_writeback *wb, *next;
817 spin_lock_irq(&cgwb_lock);
818 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
820 memcg_cgwb_list->next = NULL; /* prevent new wb's */
821 spin_unlock_irq(&cgwb_lock);
823 queue_work(system_unbound_wq, &cleanup_offline_cgwbs_work);
827 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
828 * @css: blkcg being offlined
830 * Also prevents creation of any new wb's associated with @blkcg.
832 void wb_blkcg_offline(struct cgroup_subsys_state *css)
834 struct bdi_writeback *wb, *next;
835 struct list_head *list = blkcg_get_cgwb_list(css);
837 spin_lock_irq(&cgwb_lock);
838 list_for_each_entry_safe(wb, next, list, blkcg_node)
840 list->next = NULL; /* prevent new wb's */
841 spin_unlock_irq(&cgwb_lock);
844 static void cgwb_bdi_register(struct backing_dev_info *bdi)
846 spin_lock_irq(&cgwb_lock);
847 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
848 spin_unlock_irq(&cgwb_lock);
851 static int __init cgwb_init(void)
854 * There can be many concurrent release work items overwhelming
855 * system_wq. Put them in a separate wq and limit concurrency.
856 * There's no point in executing many of these in parallel.
858 cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
859 if (!cgwb_release_wq)
864 subsys_initcall(cgwb_init);
866 #else /* CONFIG_CGROUP_WRITEBACK */
868 static int cgwb_bdi_init(struct backing_dev_info *bdi)
870 return wb_init(&bdi->wb, bdi, GFP_KERNEL);
873 static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
875 static void cgwb_bdi_register(struct backing_dev_info *bdi)
877 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
880 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
882 list_del_rcu(&wb->bdi_node);
885 #endif /* CONFIG_CGROUP_WRITEBACK */
887 int bdi_init(struct backing_dev_info *bdi)
891 kref_init(&bdi->refcnt);
893 bdi->max_ratio = 100 * BDI_RATIO_SCALE;
894 bdi->max_prop_frac = FPROP_FRAC_BASE;
895 INIT_LIST_HEAD(&bdi->bdi_list);
896 INIT_LIST_HEAD(&bdi->wb_list);
897 init_waitqueue_head(&bdi->wb_waitq);
898 bdi->last_bdp_sleep = jiffies;
900 return cgwb_bdi_init(bdi);
903 struct backing_dev_info *bdi_alloc(int node_id)
905 struct backing_dev_info *bdi;
907 bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id);
915 bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT;
916 bdi->ra_pages = VM_READAHEAD_PAGES;
917 bdi->io_pages = VM_READAHEAD_PAGES;
918 timer_setup(&bdi->laptop_mode_wb_timer, laptop_mode_timer_fn, 0);
921 EXPORT_SYMBOL(bdi_alloc);
923 static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp)
925 struct rb_node **p = &bdi_tree.rb_node;
926 struct rb_node *parent = NULL;
927 struct backing_dev_info *bdi;
929 lockdep_assert_held(&bdi_lock);
933 bdi = rb_entry(parent, struct backing_dev_info, rb_node);
937 else if (bdi->id < id)
949 * bdi_get_by_id - lookup and get bdi from its id
950 * @id: bdi id to lookup
952 * Find bdi matching @id and get it. Returns NULL if the matching bdi
953 * doesn't exist or is already unregistered.
955 struct backing_dev_info *bdi_get_by_id(u64 id)
957 struct backing_dev_info *bdi = NULL;
960 spin_lock_bh(&bdi_lock);
961 p = bdi_lookup_rb_node(id, NULL);
963 bdi = rb_entry(*p, struct backing_dev_info, rb_node);
966 spin_unlock_bh(&bdi_lock);
971 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
974 struct rb_node *parent, **p;
976 if (bdi->dev) /* The driver needs to use separate queues per device */
979 vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args);
980 dev = device_create(&bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name);
984 cgwb_bdi_register(bdi);
987 bdi_debug_register(bdi, dev_name(dev));
988 set_bit(WB_registered, &bdi->wb.state);
990 spin_lock_bh(&bdi_lock);
992 bdi->id = ++bdi_id_cursor;
994 p = bdi_lookup_rb_node(bdi->id, &parent);
995 rb_link_node(&bdi->rb_node, parent, p);
996 rb_insert_color(&bdi->rb_node, &bdi_tree);
998 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
1000 spin_unlock_bh(&bdi_lock);
1002 trace_writeback_bdi_register(bdi);
1006 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
1011 va_start(args, fmt);
1012 ret = bdi_register_va(bdi, fmt, args);
1016 EXPORT_SYMBOL(bdi_register);
1018 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner)
1020 WARN_ON_ONCE(bdi->owner);
1026 * Remove bdi from bdi_list, and ensure that it is no longer visible
1028 static void bdi_remove_from_list(struct backing_dev_info *bdi)
1030 spin_lock_bh(&bdi_lock);
1031 rb_erase(&bdi->rb_node, &bdi_tree);
1032 list_del_rcu(&bdi->bdi_list);
1033 spin_unlock_bh(&bdi_lock);
1035 synchronize_rcu_expedited();
1038 void bdi_unregister(struct backing_dev_info *bdi)
1040 del_timer_sync(&bdi->laptop_mode_wb_timer);
1042 /* make sure nobody finds us on the bdi_list anymore */
1043 bdi_remove_from_list(bdi);
1044 wb_shutdown(&bdi->wb);
1045 cgwb_bdi_unregister(bdi);
1048 * If this BDI's min ratio has been set, use bdi_set_min_ratio() to
1049 * update the global bdi_min_ratio.
1052 bdi_set_min_ratio(bdi, 0);
1055 bdi_debug_unregister(bdi);
1056 device_unregister(bdi->dev);
1061 put_device(bdi->owner);
1065 EXPORT_SYMBOL(bdi_unregister);
1067 static void release_bdi(struct kref *ref)
1069 struct backing_dev_info *bdi =
1070 container_of(ref, struct backing_dev_info, refcnt);
1072 WARN_ON_ONCE(test_bit(WB_registered, &bdi->wb.state));
1073 WARN_ON_ONCE(bdi->dev);
1078 void bdi_put(struct backing_dev_info *bdi)
1080 kref_put(&bdi->refcnt, release_bdi);
1082 EXPORT_SYMBOL(bdi_put);
1084 struct backing_dev_info *inode_to_bdi(struct inode *inode)
1086 struct super_block *sb;
1089 return &noop_backing_dev_info;
1093 if (sb_is_blkdev_sb(sb))
1094 return I_BDEV(inode)->bd_disk->bdi;
1098 EXPORT_SYMBOL(inode_to_bdi);
1100 const char *bdi_dev_name(struct backing_dev_info *bdi)
1102 if (!bdi || !bdi->dev)
1103 return bdi_unknown_name;
1104 return bdi->dev_name;
1106 EXPORT_SYMBOL_GPL(bdi_dev_name);