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>
20 struct backing_dev_info noop_backing_dev_info;
21 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
23 static struct class *bdi_class;
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 #define K(x) ((x) << (PAGE_SHIFT - 10))
40 #ifdef CONFIG_DEBUG_FS
41 #include <linux/debugfs.h>
42 #include <linux/seq_file.h>
44 static struct dentry *bdi_debug_root;
46 static void bdi_debug_init(void)
48 bdi_debug_root = debugfs_create_dir("bdi", NULL);
51 static int bdi_debug_stats_show(struct seq_file *m, void *v)
53 struct backing_dev_info *bdi = m->private;
54 struct bdi_writeback *wb = &bdi->wb;
55 unsigned long background_thresh;
56 unsigned long dirty_thresh;
57 unsigned long wb_thresh;
58 unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
61 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
62 spin_lock(&wb->list_lock);
63 list_for_each_entry(inode, &wb->b_dirty, i_io_list)
65 list_for_each_entry(inode, &wb->b_io, i_io_list)
67 list_for_each_entry(inode, &wb->b_more_io, i_io_list)
69 list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
70 if (inode->i_state & I_DIRTY_TIME)
72 spin_unlock(&wb->list_lock);
74 global_dirty_limits(&background_thresh, &dirty_thresh);
75 wb_thresh = wb_calc_thresh(wb, dirty_thresh);
78 "BdiWriteback: %10lu kB\n"
79 "BdiReclaimable: %10lu kB\n"
80 "BdiDirtyThresh: %10lu kB\n"
81 "DirtyThresh: %10lu kB\n"
82 "BackgroundThresh: %10lu kB\n"
83 "BdiDirtied: %10lu kB\n"
84 "BdiWritten: %10lu kB\n"
85 "BdiWriteBandwidth: %10lu kBps\n"
89 "b_dirty_time: %10lu\n"
92 (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
93 (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
97 (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
98 (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
99 (unsigned long) K(wb->write_bandwidth),
104 !list_empty(&bdi->bdi_list), bdi->wb.state);
108 DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats);
110 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
112 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
114 debugfs_create_file("stats", 0444, bdi->debug_dir, bdi,
115 &bdi_debug_stats_fops);
118 static void bdi_debug_unregister(struct backing_dev_info *bdi)
120 debugfs_remove_recursive(bdi->debug_dir);
123 static inline void bdi_debug_init(void)
126 static inline void bdi_debug_register(struct backing_dev_info *bdi,
130 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
135 static ssize_t read_ahead_kb_store(struct device *dev,
136 struct device_attribute *attr,
137 const char *buf, size_t count)
139 struct backing_dev_info *bdi = dev_get_drvdata(dev);
140 unsigned long read_ahead_kb;
143 ret = kstrtoul(buf, 10, &read_ahead_kb);
147 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
152 #define BDI_SHOW(name, expr) \
153 static ssize_t name##_show(struct device *dev, \
154 struct device_attribute *attr, char *buf) \
156 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
158 return sysfs_emit(buf, "%lld\n", (long long)expr); \
160 static DEVICE_ATTR_RW(name);
162 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
164 static ssize_t min_ratio_store(struct device *dev,
165 struct device_attribute *attr, const char *buf, size_t count)
167 struct backing_dev_info *bdi = dev_get_drvdata(dev);
171 ret = kstrtouint(buf, 10, &ratio);
175 ret = bdi_set_min_ratio(bdi, ratio);
181 BDI_SHOW(min_ratio, bdi->min_ratio)
183 static ssize_t max_ratio_store(struct device *dev,
184 struct device_attribute *attr, const char *buf, size_t count)
186 struct backing_dev_info *bdi = dev_get_drvdata(dev);
190 ret = kstrtouint(buf, 10, &ratio);
194 ret = bdi_set_max_ratio(bdi, ratio);
200 BDI_SHOW(max_ratio, bdi->max_ratio)
202 static ssize_t stable_pages_required_show(struct device *dev,
203 struct device_attribute *attr,
207 "the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n");
208 return sysfs_emit(buf, "%d\n", 0);
210 static DEVICE_ATTR_RO(stable_pages_required);
212 static struct attribute *bdi_dev_attrs[] = {
213 &dev_attr_read_ahead_kb.attr,
214 &dev_attr_min_ratio.attr,
215 &dev_attr_max_ratio.attr,
216 &dev_attr_stable_pages_required.attr,
219 ATTRIBUTE_GROUPS(bdi_dev);
221 static __init int bdi_class_init(void)
223 bdi_class = class_create(THIS_MODULE, "bdi");
224 if (IS_ERR(bdi_class))
225 return PTR_ERR(bdi_class);
227 bdi_class->dev_groups = bdi_dev_groups;
232 postcore_initcall(bdi_class_init);
234 static int bdi_init(struct backing_dev_info *bdi);
236 static int __init default_bdi_init(void)
240 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND |
245 err = bdi_init(&noop_backing_dev_info);
249 subsys_initcall(default_bdi_init);
252 * This function is used when the first inode for this wb is marked dirty. It
253 * wakes-up the corresponding bdi thread which should then take care of the
254 * periodic background write-out of dirty inodes. Since the write-out would
255 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
256 * set up a timer which wakes the bdi thread up later.
258 * Note, we wouldn't bother setting up the timer, but this function is on the
259 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
260 * by delaying the wake-up.
262 * We have to be careful not to postpone flush work if it is scheduled for
263 * earlier. Thus we use queue_delayed_work().
265 void wb_wakeup_delayed(struct bdi_writeback *wb)
267 unsigned long timeout;
269 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
270 spin_lock_bh(&wb->work_lock);
271 if (test_bit(WB_registered, &wb->state))
272 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
273 spin_unlock_bh(&wb->work_lock);
276 static void wb_update_bandwidth_workfn(struct work_struct *work)
278 struct bdi_writeback *wb = container_of(to_delayed_work(work),
279 struct bdi_writeback, bw_dwork);
281 wb_update_bandwidth(wb);
285 * Initial write bandwidth: 100 MB/s
287 #define INIT_BW (100 << (20 - PAGE_SHIFT))
289 static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
294 memset(wb, 0, sizeof(*wb));
297 wb->last_old_flush = jiffies;
298 INIT_LIST_HEAD(&wb->b_dirty);
299 INIT_LIST_HEAD(&wb->b_io);
300 INIT_LIST_HEAD(&wb->b_more_io);
301 INIT_LIST_HEAD(&wb->b_dirty_time);
302 spin_lock_init(&wb->list_lock);
304 atomic_set(&wb->writeback_inodes, 0);
305 wb->bw_time_stamp = jiffies;
306 wb->balanced_dirty_ratelimit = INIT_BW;
307 wb->dirty_ratelimit = INIT_BW;
308 wb->write_bandwidth = INIT_BW;
309 wb->avg_write_bandwidth = INIT_BW;
311 spin_lock_init(&wb->work_lock);
312 INIT_LIST_HEAD(&wb->work_list);
313 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
314 INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn);
315 wb->dirty_sleep = jiffies;
317 err = fprop_local_init_percpu(&wb->completions, gfp);
321 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
322 err = percpu_counter_init(&wb->stat[i], 0, gfp);
324 goto out_destroy_stat;
331 percpu_counter_destroy(&wb->stat[i]);
332 fprop_local_destroy_percpu(&wb->completions);
336 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
339 * Remove bdi from the global list and shutdown any threads we have running
341 static void wb_shutdown(struct bdi_writeback *wb)
343 /* Make sure nobody queues further work */
344 spin_lock_bh(&wb->work_lock);
345 if (!test_and_clear_bit(WB_registered, &wb->state)) {
346 spin_unlock_bh(&wb->work_lock);
349 spin_unlock_bh(&wb->work_lock);
351 cgwb_remove_from_bdi_list(wb);
353 * Drain work list and shutdown the delayed_work. !WB_registered
354 * tells wb_workfn() that @wb is dying and its work_list needs to
355 * be drained no matter what.
357 mod_delayed_work(bdi_wq, &wb->dwork, 0);
358 flush_delayed_work(&wb->dwork);
359 WARN_ON(!list_empty(&wb->work_list));
360 flush_delayed_work(&wb->bw_dwork);
363 static void wb_exit(struct bdi_writeback *wb)
367 WARN_ON(delayed_work_pending(&wb->dwork));
369 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
370 percpu_counter_destroy(&wb->stat[i]);
372 fprop_local_destroy_percpu(&wb->completions);
375 #ifdef CONFIG_CGROUP_WRITEBACK
377 #include <linux/memcontrol.h>
380 * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and
381 * memcg->cgwb_list. bdi->cgwb_tree is also RCU protected.
383 static DEFINE_SPINLOCK(cgwb_lock);
384 static struct workqueue_struct *cgwb_release_wq;
386 static LIST_HEAD(offline_cgwbs);
387 static void cleanup_offline_cgwbs_workfn(struct work_struct *work);
388 static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn);
390 static void cgwb_release_workfn(struct work_struct *work)
392 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
394 struct backing_dev_info *bdi = wb->bdi;
396 mutex_lock(&wb->bdi->cgwb_release_mutex);
399 css_put(wb->memcg_css);
400 css_put(wb->blkcg_css);
401 mutex_unlock(&wb->bdi->cgwb_release_mutex);
403 /* triggers blkg destruction if no online users left */
404 blkcg_unpin_online(wb->blkcg_css);
406 fprop_local_destroy_percpu(&wb->memcg_completions);
408 spin_lock_irq(&cgwb_lock);
409 list_del(&wb->offline_node);
410 spin_unlock_irq(&cgwb_lock);
412 percpu_ref_exit(&wb->refcnt);
415 WARN_ON_ONCE(!list_empty(&wb->b_attached));
419 static void cgwb_release(struct percpu_ref *refcnt)
421 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
423 queue_work(cgwb_release_wq, &wb->release_work);
426 static void cgwb_kill(struct bdi_writeback *wb)
428 lockdep_assert_held(&cgwb_lock);
430 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
431 list_del(&wb->memcg_node);
432 list_del(&wb->blkcg_node);
433 list_add(&wb->offline_node, &offline_cgwbs);
434 percpu_ref_kill(&wb->refcnt);
437 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
439 spin_lock_irq(&cgwb_lock);
440 list_del_rcu(&wb->bdi_node);
441 spin_unlock_irq(&cgwb_lock);
444 static int cgwb_create(struct backing_dev_info *bdi,
445 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
447 struct mem_cgroup *memcg;
448 struct cgroup_subsys_state *blkcg_css;
449 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
450 struct bdi_writeback *wb;
454 memcg = mem_cgroup_from_css(memcg_css);
455 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
456 memcg_cgwb_list = &memcg->cgwb_list;
457 blkcg_cgwb_list = blkcg_get_cgwb_list(blkcg_css);
459 /* look up again under lock and discard on blkcg mismatch */
460 spin_lock_irqsave(&cgwb_lock, flags);
461 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
462 if (wb && wb->blkcg_css != blkcg_css) {
466 spin_unlock_irqrestore(&cgwb_lock, flags);
470 /* need to create a new one */
471 wb = kmalloc(sizeof(*wb), gfp);
477 ret = wb_init(wb, bdi, gfp);
481 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
485 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
489 wb->memcg_css = memcg_css;
490 wb->blkcg_css = blkcg_css;
491 INIT_LIST_HEAD(&wb->b_attached);
492 INIT_WORK(&wb->release_work, cgwb_release_workfn);
493 set_bit(WB_registered, &wb->state);
497 * The root wb determines the registered state of the whole bdi and
498 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
499 * whether they're still online. Don't link @wb if any is dead.
500 * See wb_memcg_offline() and wb_blkcg_offline().
503 spin_lock_irqsave(&cgwb_lock, flags);
504 if (test_bit(WB_registered, &bdi->wb.state) &&
505 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
506 /* we might have raced another instance of this function */
507 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
509 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
510 list_add(&wb->memcg_node, memcg_cgwb_list);
511 list_add(&wb->blkcg_node, blkcg_cgwb_list);
512 blkcg_pin_online(blkcg_css);
517 spin_unlock_irqrestore(&cgwb_lock, flags);
527 fprop_local_destroy_percpu(&wb->memcg_completions);
529 percpu_ref_exit(&wb->refcnt);
540 * wb_get_lookup - get wb for a given memcg
542 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
544 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
545 * refcount incremented.
547 * This function uses css_get() on @memcg_css and thus expects its refcnt
548 * to be positive on invocation. IOW, rcu_read_lock() protection on
549 * @memcg_css isn't enough. try_get it before calling this function.
551 * A wb is keyed by its associated memcg. As blkcg implicitly enables
552 * memcg on the default hierarchy, memcg association is guaranteed to be
553 * more specific (equal or descendant to the associated blkcg) and thus can
554 * identify both the memcg and blkcg associations.
556 * Because the blkcg associated with a memcg may change as blkcg is enabled
557 * and disabled closer to root in the hierarchy, each wb keeps track of
558 * both the memcg and blkcg associated with it and verifies the blkcg on
559 * each lookup. On mismatch, the existing wb is discarded and a new one is
562 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
563 struct cgroup_subsys_state *memcg_css)
565 struct bdi_writeback *wb;
567 if (!memcg_css->parent)
571 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
573 struct cgroup_subsys_state *blkcg_css;
575 /* see whether the blkcg association has changed */
576 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
577 if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
587 * wb_get_create - get wb for a given memcg, create if necessary
589 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
590 * @gfp: allocation mask to use
592 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
593 * create one. See wb_get_lookup() for more details.
595 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
596 struct cgroup_subsys_state *memcg_css,
599 struct bdi_writeback *wb;
603 if (!memcg_css->parent)
607 wb = wb_get_lookup(bdi, memcg_css);
608 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
613 static int cgwb_bdi_init(struct backing_dev_info *bdi)
617 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
618 mutex_init(&bdi->cgwb_release_mutex);
619 init_rwsem(&bdi->wb_switch_rwsem);
621 ret = wb_init(&bdi->wb, bdi, GFP_KERNEL);
623 bdi->wb.memcg_css = &root_mem_cgroup->css;
624 bdi->wb.blkcg_css = blkcg_root_css;
629 static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
631 struct radix_tree_iter iter;
633 struct bdi_writeback *wb;
635 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
637 spin_lock_irq(&cgwb_lock);
638 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
640 spin_unlock_irq(&cgwb_lock);
642 mutex_lock(&bdi->cgwb_release_mutex);
643 spin_lock_irq(&cgwb_lock);
644 while (!list_empty(&bdi->wb_list)) {
645 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
647 spin_unlock_irq(&cgwb_lock);
649 spin_lock_irq(&cgwb_lock);
651 spin_unlock_irq(&cgwb_lock);
652 mutex_unlock(&bdi->cgwb_release_mutex);
656 * cleanup_offline_cgwbs_workfn - try to release dying cgwbs
658 * Try to release dying cgwbs by switching attached inodes to the nearest
659 * living ancestor's writeback. Processed wbs are placed at the end
660 * of the list to guarantee the forward progress.
662 static void cleanup_offline_cgwbs_workfn(struct work_struct *work)
664 struct bdi_writeback *wb;
665 LIST_HEAD(processed);
667 spin_lock_irq(&cgwb_lock);
669 while (!list_empty(&offline_cgwbs)) {
670 wb = list_first_entry(&offline_cgwbs, struct bdi_writeback,
672 list_move(&wb->offline_node, &processed);
675 * If wb is dirty, cleaning up the writeback by switching
676 * attached inodes will result in an effective removal of any
677 * bandwidth restrictions, which isn't the goal. Instead,
678 * it can be postponed until the next time, when all io
679 * will be likely completed. If in the meantime some inodes
680 * will get re-dirtied, they should be eventually switched to
683 if (wb_has_dirty_io(wb))
689 spin_unlock_irq(&cgwb_lock);
690 while (cleanup_offline_cgwb(wb))
692 spin_lock_irq(&cgwb_lock);
697 if (!list_empty(&processed))
698 list_splice_tail(&processed, &offline_cgwbs);
700 spin_unlock_irq(&cgwb_lock);
704 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
705 * @memcg: memcg being offlined
707 * Also prevents creation of any new wb's associated with @memcg.
709 void wb_memcg_offline(struct mem_cgroup *memcg)
711 struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
712 struct bdi_writeback *wb, *next;
714 spin_lock_irq(&cgwb_lock);
715 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
717 memcg_cgwb_list->next = NULL; /* prevent new wb's */
718 spin_unlock_irq(&cgwb_lock);
720 queue_work(system_unbound_wq, &cleanup_offline_cgwbs_work);
724 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
725 * @css: blkcg being offlined
727 * Also prevents creation of any new wb's associated with @blkcg.
729 void wb_blkcg_offline(struct cgroup_subsys_state *css)
731 struct bdi_writeback *wb, *next;
732 struct list_head *list = blkcg_get_cgwb_list(css);
734 spin_lock_irq(&cgwb_lock);
735 list_for_each_entry_safe(wb, next, list, blkcg_node)
737 list->next = NULL; /* prevent new wb's */
738 spin_unlock_irq(&cgwb_lock);
741 static void cgwb_bdi_register(struct backing_dev_info *bdi)
743 spin_lock_irq(&cgwb_lock);
744 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
745 spin_unlock_irq(&cgwb_lock);
748 static int __init cgwb_init(void)
751 * There can be many concurrent release work items overwhelming
752 * system_wq. Put them in a separate wq and limit concurrency.
753 * There's no point in executing many of these in parallel.
755 cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
756 if (!cgwb_release_wq)
761 subsys_initcall(cgwb_init);
763 #else /* CONFIG_CGROUP_WRITEBACK */
765 static int cgwb_bdi_init(struct backing_dev_info *bdi)
767 return wb_init(&bdi->wb, bdi, GFP_KERNEL);
770 static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
772 static void cgwb_bdi_register(struct backing_dev_info *bdi)
774 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
777 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
779 list_del_rcu(&wb->bdi_node);
782 #endif /* CONFIG_CGROUP_WRITEBACK */
784 static int bdi_init(struct backing_dev_info *bdi)
790 kref_init(&bdi->refcnt);
792 bdi->max_ratio = 100;
793 bdi->max_prop_frac = FPROP_FRAC_BASE;
794 INIT_LIST_HEAD(&bdi->bdi_list);
795 INIT_LIST_HEAD(&bdi->wb_list);
796 init_waitqueue_head(&bdi->wb_waitq);
798 ret = cgwb_bdi_init(bdi);
803 struct backing_dev_info *bdi_alloc(int node_id)
805 struct backing_dev_info *bdi;
807 bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id);
815 bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT;
816 bdi->ra_pages = VM_READAHEAD_PAGES;
817 bdi->io_pages = VM_READAHEAD_PAGES;
818 timer_setup(&bdi->laptop_mode_wb_timer, laptop_mode_timer_fn, 0);
821 EXPORT_SYMBOL(bdi_alloc);
823 static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp)
825 struct rb_node **p = &bdi_tree.rb_node;
826 struct rb_node *parent = NULL;
827 struct backing_dev_info *bdi;
829 lockdep_assert_held(&bdi_lock);
833 bdi = rb_entry(parent, struct backing_dev_info, rb_node);
837 else if (bdi->id < id)
849 * bdi_get_by_id - lookup and get bdi from its id
850 * @id: bdi id to lookup
852 * Find bdi matching @id and get it. Returns NULL if the matching bdi
853 * doesn't exist or is already unregistered.
855 struct backing_dev_info *bdi_get_by_id(u64 id)
857 struct backing_dev_info *bdi = NULL;
860 spin_lock_bh(&bdi_lock);
861 p = bdi_lookup_rb_node(id, NULL);
863 bdi = rb_entry(*p, struct backing_dev_info, rb_node);
866 spin_unlock_bh(&bdi_lock);
871 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
874 struct rb_node *parent, **p;
876 if (bdi->dev) /* The driver needs to use separate queues per device */
879 vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args);
880 dev = device_create(bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name);
884 cgwb_bdi_register(bdi);
887 bdi_debug_register(bdi, dev_name(dev));
888 set_bit(WB_registered, &bdi->wb.state);
890 spin_lock_bh(&bdi_lock);
892 bdi->id = ++bdi_id_cursor;
894 p = bdi_lookup_rb_node(bdi->id, &parent);
895 rb_link_node(&bdi->rb_node, parent, p);
896 rb_insert_color(&bdi->rb_node, &bdi_tree);
898 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
900 spin_unlock_bh(&bdi_lock);
902 trace_writeback_bdi_register(bdi);
906 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
912 ret = bdi_register_va(bdi, fmt, args);
916 EXPORT_SYMBOL(bdi_register);
918 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner)
920 WARN_ON_ONCE(bdi->owner);
926 * Remove bdi from bdi_list, and ensure that it is no longer visible
928 static void bdi_remove_from_list(struct backing_dev_info *bdi)
930 spin_lock_bh(&bdi_lock);
931 rb_erase(&bdi->rb_node, &bdi_tree);
932 list_del_rcu(&bdi->bdi_list);
933 spin_unlock_bh(&bdi_lock);
935 synchronize_rcu_expedited();
938 void bdi_unregister(struct backing_dev_info *bdi)
940 del_timer_sync(&bdi->laptop_mode_wb_timer);
942 /* make sure nobody finds us on the bdi_list anymore */
943 bdi_remove_from_list(bdi);
944 wb_shutdown(&bdi->wb);
945 cgwb_bdi_unregister(bdi);
948 * If this BDI's min ratio has been set, use bdi_set_min_ratio() to
949 * update the global bdi_min_ratio.
952 bdi_set_min_ratio(bdi, 0);
955 bdi_debug_unregister(bdi);
956 device_unregister(bdi->dev);
961 put_device(bdi->owner);
965 EXPORT_SYMBOL(bdi_unregister);
967 static void release_bdi(struct kref *ref)
969 struct backing_dev_info *bdi =
970 container_of(ref, struct backing_dev_info, refcnt);
972 WARN_ON_ONCE(test_bit(WB_registered, &bdi->wb.state));
973 WARN_ON_ONCE(bdi->dev);
978 void bdi_put(struct backing_dev_info *bdi)
980 kref_put(&bdi->refcnt, release_bdi);
982 EXPORT_SYMBOL(bdi_put);
984 struct backing_dev_info *inode_to_bdi(struct inode *inode)
986 struct super_block *sb;
989 return &noop_backing_dev_info;
993 if (sb_is_blkdev_sb(sb))
994 return I_BDEV(inode)->bd_disk->bdi;
998 EXPORT_SYMBOL(inode_to_bdi);
1000 const char *bdi_dev_name(struct backing_dev_info *bdi)
1002 if (!bdi || !bdi->dev)
1003 return bdi_unknown_name;
1004 return bdi->dev_name;
1006 EXPORT_SYMBOL_GPL(bdi_dev_name);