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 __init default_bdi_init(void)
236 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND |
242 subsys_initcall(default_bdi_init);
245 * This function is used when the first inode for this wb is marked dirty. It
246 * wakes-up the corresponding bdi thread which should then take care of the
247 * periodic background write-out of dirty inodes. Since the write-out would
248 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
249 * set up a timer which wakes the bdi thread up later.
251 * Note, we wouldn't bother setting up the timer, but this function is on the
252 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
253 * by delaying the wake-up.
255 * We have to be careful not to postpone flush work if it is scheduled for
256 * earlier. Thus we use queue_delayed_work().
258 void wb_wakeup_delayed(struct bdi_writeback *wb)
260 unsigned long timeout;
262 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
263 spin_lock_irq(&wb->work_lock);
264 if (test_bit(WB_registered, &wb->state))
265 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
266 spin_unlock_irq(&wb->work_lock);
269 static void wb_update_bandwidth_workfn(struct work_struct *work)
271 struct bdi_writeback *wb = container_of(to_delayed_work(work),
272 struct bdi_writeback, bw_dwork);
274 wb_update_bandwidth(wb);
278 * Initial write bandwidth: 100 MB/s
280 #define INIT_BW (100 << (20 - PAGE_SHIFT))
282 static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
287 memset(wb, 0, sizeof(*wb));
290 wb->last_old_flush = jiffies;
291 INIT_LIST_HEAD(&wb->b_dirty);
292 INIT_LIST_HEAD(&wb->b_io);
293 INIT_LIST_HEAD(&wb->b_more_io);
294 INIT_LIST_HEAD(&wb->b_dirty_time);
295 spin_lock_init(&wb->list_lock);
297 atomic_set(&wb->writeback_inodes, 0);
298 wb->bw_time_stamp = jiffies;
299 wb->balanced_dirty_ratelimit = INIT_BW;
300 wb->dirty_ratelimit = INIT_BW;
301 wb->write_bandwidth = INIT_BW;
302 wb->avg_write_bandwidth = INIT_BW;
304 spin_lock_init(&wb->work_lock);
305 INIT_LIST_HEAD(&wb->work_list);
306 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
307 INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn);
308 wb->dirty_sleep = jiffies;
310 err = fprop_local_init_percpu(&wb->completions, gfp);
314 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
315 err = percpu_counter_init(&wb->stat[i], 0, gfp);
317 goto out_destroy_stat;
324 percpu_counter_destroy(&wb->stat[i]);
325 fprop_local_destroy_percpu(&wb->completions);
329 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
332 * Remove bdi from the global list and shutdown any threads we have running
334 static void wb_shutdown(struct bdi_writeback *wb)
336 /* Make sure nobody queues further work */
337 spin_lock_irq(&wb->work_lock);
338 if (!test_and_clear_bit(WB_registered, &wb->state)) {
339 spin_unlock_irq(&wb->work_lock);
342 spin_unlock_irq(&wb->work_lock);
344 cgwb_remove_from_bdi_list(wb);
346 * Drain work list and shutdown the delayed_work. !WB_registered
347 * tells wb_workfn() that @wb is dying and its work_list needs to
348 * be drained no matter what.
350 mod_delayed_work(bdi_wq, &wb->dwork, 0);
351 flush_delayed_work(&wb->dwork);
352 WARN_ON(!list_empty(&wb->work_list));
353 flush_delayed_work(&wb->bw_dwork);
356 static void wb_exit(struct bdi_writeback *wb)
360 WARN_ON(delayed_work_pending(&wb->dwork));
362 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
363 percpu_counter_destroy(&wb->stat[i]);
365 fprop_local_destroy_percpu(&wb->completions);
368 #ifdef CONFIG_CGROUP_WRITEBACK
370 #include <linux/memcontrol.h>
373 * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and
374 * memcg->cgwb_list. bdi->cgwb_tree is also RCU protected.
376 static DEFINE_SPINLOCK(cgwb_lock);
377 static struct workqueue_struct *cgwb_release_wq;
379 static LIST_HEAD(offline_cgwbs);
380 static void cleanup_offline_cgwbs_workfn(struct work_struct *work);
381 static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn);
383 static void cgwb_free_rcu(struct rcu_head *rcu_head)
385 struct bdi_writeback *wb = container_of(rcu_head,
386 struct bdi_writeback, rcu);
388 percpu_ref_exit(&wb->refcnt);
392 static void cgwb_release_workfn(struct work_struct *work)
394 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
396 struct backing_dev_info *bdi = wb->bdi;
398 mutex_lock(&wb->bdi->cgwb_release_mutex);
401 css_put(wb->memcg_css);
402 css_put(wb->blkcg_css);
403 mutex_unlock(&wb->bdi->cgwb_release_mutex);
405 /* triggers blkg destruction if no online users left */
406 blkcg_unpin_online(wb->blkcg_css);
408 fprop_local_destroy_percpu(&wb->memcg_completions);
410 spin_lock_irq(&cgwb_lock);
411 list_del(&wb->offline_node);
412 spin_unlock_irq(&cgwb_lock);
416 WARN_ON_ONCE(!list_empty(&wb->b_attached));
417 call_rcu(&wb->rcu, cgwb_free_rcu);
420 static void cgwb_release(struct percpu_ref *refcnt)
422 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
424 queue_work(cgwb_release_wq, &wb->release_work);
427 static void cgwb_kill(struct bdi_writeback *wb)
429 lockdep_assert_held(&cgwb_lock);
431 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
432 list_del(&wb->memcg_node);
433 list_del(&wb->blkcg_node);
434 list_add(&wb->offline_node, &offline_cgwbs);
435 percpu_ref_kill(&wb->refcnt);
438 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
440 spin_lock_irq(&cgwb_lock);
441 list_del_rcu(&wb->bdi_node);
442 spin_unlock_irq(&cgwb_lock);
445 static int cgwb_create(struct backing_dev_info *bdi,
446 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
448 struct mem_cgroup *memcg;
449 struct cgroup_subsys_state *blkcg_css;
450 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
451 struct bdi_writeback *wb;
455 memcg = mem_cgroup_from_css(memcg_css);
456 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
457 memcg_cgwb_list = &memcg->cgwb_list;
458 blkcg_cgwb_list = blkcg_get_cgwb_list(blkcg_css);
460 /* look up again under lock and discard on blkcg mismatch */
461 spin_lock_irqsave(&cgwb_lock, flags);
462 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
463 if (wb && wb->blkcg_css != blkcg_css) {
467 spin_unlock_irqrestore(&cgwb_lock, flags);
471 /* need to create a new one */
472 wb = kmalloc(sizeof(*wb), gfp);
478 ret = wb_init(wb, bdi, gfp);
482 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
486 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
490 wb->memcg_css = memcg_css;
491 wb->blkcg_css = blkcg_css;
492 INIT_LIST_HEAD(&wb->b_attached);
493 INIT_WORK(&wb->release_work, cgwb_release_workfn);
494 set_bit(WB_registered, &wb->state);
498 * The root wb determines the registered state of the whole bdi and
499 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
500 * whether they're still online. Don't link @wb if any is dead.
501 * See wb_memcg_offline() and wb_blkcg_offline().
504 spin_lock_irqsave(&cgwb_lock, flags);
505 if (test_bit(WB_registered, &bdi->wb.state) &&
506 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
507 /* we might have raced another instance of this function */
508 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
510 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
511 list_add(&wb->memcg_node, memcg_cgwb_list);
512 list_add(&wb->blkcg_node, blkcg_cgwb_list);
513 blkcg_pin_online(blkcg_css);
518 spin_unlock_irqrestore(&cgwb_lock, flags);
528 fprop_local_destroy_percpu(&wb->memcg_completions);
530 percpu_ref_exit(&wb->refcnt);
541 * wb_get_lookup - get wb for a given memcg
543 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
545 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
546 * refcount incremented.
548 * This function uses css_get() on @memcg_css and thus expects its refcnt
549 * to be positive on invocation. IOW, rcu_read_lock() protection on
550 * @memcg_css isn't enough. try_get it before calling this function.
552 * A wb is keyed by its associated memcg. As blkcg implicitly enables
553 * memcg on the default hierarchy, memcg association is guaranteed to be
554 * more specific (equal or descendant to the associated blkcg) and thus can
555 * identify both the memcg and blkcg associations.
557 * Because the blkcg associated with a memcg may change as blkcg is enabled
558 * and disabled closer to root in the hierarchy, each wb keeps track of
559 * both the memcg and blkcg associated with it and verifies the blkcg on
560 * each lookup. On mismatch, the existing wb is discarded and a new one is
563 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
564 struct cgroup_subsys_state *memcg_css)
566 struct bdi_writeback *wb;
568 if (!memcg_css->parent)
572 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
574 struct cgroup_subsys_state *blkcg_css;
576 /* see whether the blkcg association has changed */
577 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
578 if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
588 * wb_get_create - get wb for a given memcg, create if necessary
590 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
591 * @gfp: allocation mask to use
593 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
594 * create one. See wb_get_lookup() for more details.
596 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
597 struct cgroup_subsys_state *memcg_css,
600 struct bdi_writeback *wb;
604 if (!memcg_css->parent)
608 wb = wb_get_lookup(bdi, memcg_css);
609 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
614 static int cgwb_bdi_init(struct backing_dev_info *bdi)
618 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
619 mutex_init(&bdi->cgwb_release_mutex);
620 init_rwsem(&bdi->wb_switch_rwsem);
622 ret = wb_init(&bdi->wb, bdi, GFP_KERNEL);
624 bdi->wb.memcg_css = &root_mem_cgroup->css;
625 bdi->wb.blkcg_css = blkcg_root_css;
630 static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
632 struct radix_tree_iter iter;
634 struct bdi_writeback *wb;
636 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
638 spin_lock_irq(&cgwb_lock);
639 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
641 spin_unlock_irq(&cgwb_lock);
643 mutex_lock(&bdi->cgwb_release_mutex);
644 spin_lock_irq(&cgwb_lock);
645 while (!list_empty(&bdi->wb_list)) {
646 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
648 spin_unlock_irq(&cgwb_lock);
650 spin_lock_irq(&cgwb_lock);
652 spin_unlock_irq(&cgwb_lock);
653 mutex_unlock(&bdi->cgwb_release_mutex);
657 * cleanup_offline_cgwbs_workfn - try to release dying cgwbs
659 * Try to release dying cgwbs by switching attached inodes to the nearest
660 * living ancestor's writeback. Processed wbs are placed at the end
661 * of the list to guarantee the forward progress.
663 static void cleanup_offline_cgwbs_workfn(struct work_struct *work)
665 struct bdi_writeback *wb;
666 LIST_HEAD(processed);
668 spin_lock_irq(&cgwb_lock);
670 while (!list_empty(&offline_cgwbs)) {
671 wb = list_first_entry(&offline_cgwbs, struct bdi_writeback,
673 list_move(&wb->offline_node, &processed);
676 * If wb is dirty, cleaning up the writeback by switching
677 * attached inodes will result in an effective removal of any
678 * bandwidth restrictions, which isn't the goal. Instead,
679 * it can be postponed until the next time, when all io
680 * will be likely completed. If in the meantime some inodes
681 * will get re-dirtied, they should be eventually switched to
684 if (wb_has_dirty_io(wb))
690 spin_unlock_irq(&cgwb_lock);
691 while (cleanup_offline_cgwb(wb))
693 spin_lock_irq(&cgwb_lock);
698 if (!list_empty(&processed))
699 list_splice_tail(&processed, &offline_cgwbs);
701 spin_unlock_irq(&cgwb_lock);
705 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
706 * @memcg: memcg being offlined
708 * Also prevents creation of any new wb's associated with @memcg.
710 void wb_memcg_offline(struct mem_cgroup *memcg)
712 struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
713 struct bdi_writeback *wb, *next;
715 spin_lock_irq(&cgwb_lock);
716 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
718 memcg_cgwb_list->next = NULL; /* prevent new wb's */
719 spin_unlock_irq(&cgwb_lock);
721 queue_work(system_unbound_wq, &cleanup_offline_cgwbs_work);
725 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
726 * @css: blkcg being offlined
728 * Also prevents creation of any new wb's associated with @blkcg.
730 void wb_blkcg_offline(struct cgroup_subsys_state *css)
732 struct bdi_writeback *wb, *next;
733 struct list_head *list = blkcg_get_cgwb_list(css);
735 spin_lock_irq(&cgwb_lock);
736 list_for_each_entry_safe(wb, next, list, blkcg_node)
738 list->next = NULL; /* prevent new wb's */
739 spin_unlock_irq(&cgwb_lock);
742 static void cgwb_bdi_register(struct backing_dev_info *bdi)
744 spin_lock_irq(&cgwb_lock);
745 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
746 spin_unlock_irq(&cgwb_lock);
749 static int __init cgwb_init(void)
752 * There can be many concurrent release work items overwhelming
753 * system_wq. Put them in a separate wq and limit concurrency.
754 * There's no point in executing many of these in parallel.
756 cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
757 if (!cgwb_release_wq)
762 subsys_initcall(cgwb_init);
764 #else /* CONFIG_CGROUP_WRITEBACK */
766 static int cgwb_bdi_init(struct backing_dev_info *bdi)
768 return wb_init(&bdi->wb, bdi, GFP_KERNEL);
771 static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
773 static void cgwb_bdi_register(struct backing_dev_info *bdi)
775 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
778 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
780 list_del_rcu(&wb->bdi_node);
783 #endif /* CONFIG_CGROUP_WRITEBACK */
785 int bdi_init(struct backing_dev_info *bdi)
789 kref_init(&bdi->refcnt);
791 bdi->max_ratio = 100;
792 bdi->max_prop_frac = FPROP_FRAC_BASE;
793 INIT_LIST_HEAD(&bdi->bdi_list);
794 INIT_LIST_HEAD(&bdi->wb_list);
795 init_waitqueue_head(&bdi->wb_waitq);
797 return cgwb_bdi_init(bdi);
800 struct backing_dev_info *bdi_alloc(int node_id)
802 struct backing_dev_info *bdi;
804 bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id);
812 bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT;
813 bdi->ra_pages = VM_READAHEAD_PAGES;
814 bdi->io_pages = VM_READAHEAD_PAGES;
815 timer_setup(&bdi->laptop_mode_wb_timer, laptop_mode_timer_fn, 0);
818 EXPORT_SYMBOL(bdi_alloc);
820 static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp)
822 struct rb_node **p = &bdi_tree.rb_node;
823 struct rb_node *parent = NULL;
824 struct backing_dev_info *bdi;
826 lockdep_assert_held(&bdi_lock);
830 bdi = rb_entry(parent, struct backing_dev_info, rb_node);
834 else if (bdi->id < id)
846 * bdi_get_by_id - lookup and get bdi from its id
847 * @id: bdi id to lookup
849 * Find bdi matching @id and get it. Returns NULL if the matching bdi
850 * doesn't exist or is already unregistered.
852 struct backing_dev_info *bdi_get_by_id(u64 id)
854 struct backing_dev_info *bdi = NULL;
857 spin_lock_bh(&bdi_lock);
858 p = bdi_lookup_rb_node(id, NULL);
860 bdi = rb_entry(*p, struct backing_dev_info, rb_node);
863 spin_unlock_bh(&bdi_lock);
868 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
871 struct rb_node *parent, **p;
873 if (bdi->dev) /* The driver needs to use separate queues per device */
876 vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args);
877 dev = device_create(bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name);
881 cgwb_bdi_register(bdi);
884 bdi_debug_register(bdi, dev_name(dev));
885 set_bit(WB_registered, &bdi->wb.state);
887 spin_lock_bh(&bdi_lock);
889 bdi->id = ++bdi_id_cursor;
891 p = bdi_lookup_rb_node(bdi->id, &parent);
892 rb_link_node(&bdi->rb_node, parent, p);
893 rb_insert_color(&bdi->rb_node, &bdi_tree);
895 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
897 spin_unlock_bh(&bdi_lock);
899 trace_writeback_bdi_register(bdi);
903 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
909 ret = bdi_register_va(bdi, fmt, args);
913 EXPORT_SYMBOL(bdi_register);
915 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner)
917 WARN_ON_ONCE(bdi->owner);
923 * Remove bdi from bdi_list, and ensure that it is no longer visible
925 static void bdi_remove_from_list(struct backing_dev_info *bdi)
927 spin_lock_bh(&bdi_lock);
928 rb_erase(&bdi->rb_node, &bdi_tree);
929 list_del_rcu(&bdi->bdi_list);
930 spin_unlock_bh(&bdi_lock);
932 synchronize_rcu_expedited();
935 void bdi_unregister(struct backing_dev_info *bdi)
937 del_timer_sync(&bdi->laptop_mode_wb_timer);
939 /* make sure nobody finds us on the bdi_list anymore */
940 bdi_remove_from_list(bdi);
941 wb_shutdown(&bdi->wb);
942 cgwb_bdi_unregister(bdi);
945 * If this BDI's min ratio has been set, use bdi_set_min_ratio() to
946 * update the global bdi_min_ratio.
949 bdi_set_min_ratio(bdi, 0);
952 bdi_debug_unregister(bdi);
953 device_unregister(bdi->dev);
958 put_device(bdi->owner);
962 EXPORT_SYMBOL(bdi_unregister);
964 static void release_bdi(struct kref *ref)
966 struct backing_dev_info *bdi =
967 container_of(ref, struct backing_dev_info, refcnt);
969 WARN_ON_ONCE(test_bit(WB_registered, &bdi->wb.state));
970 WARN_ON_ONCE(bdi->dev);
975 void bdi_put(struct backing_dev_info *bdi)
977 kref_put(&bdi->refcnt, release_bdi);
979 EXPORT_SYMBOL(bdi_put);
981 struct backing_dev_info *inode_to_bdi(struct inode *inode)
983 struct super_block *sb;
986 return &noop_backing_dev_info;
990 if (sb_is_blkdev_sb(sb))
991 return I_BDEV(inode)->bd_disk->bdi;
995 EXPORT_SYMBOL(inode_to_bdi);
997 const char *bdi_dev_name(struct backing_dev_info *bdi)
999 if (!bdi || !bdi->dev)
1000 return bdi_unknown_name;
1001 return bdi->dev_name;
1003 EXPORT_SYMBOL_GPL(bdi_dev_name);