1 // SPDX-License-Identifier: GPL-2.0-only
3 #include <linux/wait.h>
4 #include <linux/rbtree.h>
5 #include <linux/backing-dev.h>
6 #include <linux/kthread.h>
7 #include <linux/freezer.h>
9 #include <linux/pagemap.h>
11 #include <linux/sched.h>
12 #include <linux/module.h>
13 #include <linux/writeback.h>
14 #include <linux/device.h>
15 #include <trace/events/writeback.h>
17 struct backing_dev_info noop_backing_dev_info = {
18 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
20 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
22 static struct class *bdi_class;
23 static const char *bdi_unknown_name = "(unknown)";
26 * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU
27 * reader side locking.
29 DEFINE_SPINLOCK(bdi_lock);
30 static u64 bdi_id_cursor;
31 static struct rb_root bdi_tree = RB_ROOT;
34 /* bdi_wq serves all asynchronous writeback tasks */
35 struct workqueue_struct *bdi_wq;
37 #ifdef CONFIG_DEBUG_FS
38 #include <linux/debugfs.h>
39 #include <linux/seq_file.h>
41 static struct dentry *bdi_debug_root;
43 static void bdi_debug_init(void)
45 bdi_debug_root = debugfs_create_dir("bdi", NULL);
48 static int bdi_debug_stats_show(struct seq_file *m, void *v)
50 struct backing_dev_info *bdi = m->private;
51 struct bdi_writeback *wb = &bdi->wb;
52 unsigned long background_thresh;
53 unsigned long dirty_thresh;
54 unsigned long wb_thresh;
55 unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
58 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
59 spin_lock(&wb->list_lock);
60 list_for_each_entry(inode, &wb->b_dirty, i_io_list)
62 list_for_each_entry(inode, &wb->b_io, i_io_list)
64 list_for_each_entry(inode, &wb->b_more_io, i_io_list)
66 list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
67 if (inode->i_state & I_DIRTY_TIME)
69 spin_unlock(&wb->list_lock);
71 global_dirty_limits(&background_thresh, &dirty_thresh);
72 wb_thresh = wb_calc_thresh(wb, dirty_thresh);
74 #define K(x) ((x) << (PAGE_SHIFT - 10))
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);
107 DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats);
109 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
111 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
113 debugfs_create_file("stats", 0444, bdi->debug_dir, bdi,
114 &bdi_debug_stats_fops);
117 static void bdi_debug_unregister(struct backing_dev_info *bdi)
119 debugfs_remove_recursive(bdi->debug_dir);
122 static inline void bdi_debug_init(void)
125 static inline void bdi_debug_register(struct backing_dev_info *bdi,
129 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
134 static ssize_t read_ahead_kb_store(struct device *dev,
135 struct device_attribute *attr,
136 const char *buf, size_t count)
138 struct backing_dev_info *bdi = dev_get_drvdata(dev);
139 unsigned long read_ahead_kb;
142 ret = kstrtoul(buf, 10, &read_ahead_kb);
146 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
151 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
153 #define BDI_SHOW(name, expr) \
154 static ssize_t name##_show(struct device *dev, \
155 struct device_attribute *attr, char *page) \
157 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
159 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
161 static DEVICE_ATTR_RW(name);
163 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
165 static ssize_t min_ratio_store(struct device *dev,
166 struct device_attribute *attr, const char *buf, size_t count)
168 struct backing_dev_info *bdi = dev_get_drvdata(dev);
172 ret = kstrtouint(buf, 10, &ratio);
176 ret = bdi_set_min_ratio(bdi, ratio);
182 BDI_SHOW(min_ratio, bdi->min_ratio)
184 static ssize_t max_ratio_store(struct device *dev,
185 struct device_attribute *attr, const char *buf, size_t count)
187 struct backing_dev_info *bdi = dev_get_drvdata(dev);
191 ret = kstrtouint(buf, 10, &ratio);
195 ret = bdi_set_max_ratio(bdi, ratio);
201 BDI_SHOW(max_ratio, bdi->max_ratio)
203 static ssize_t stable_pages_required_show(struct device *dev,
204 struct device_attribute *attr,
207 struct backing_dev_info *bdi = dev_get_drvdata(dev);
209 return snprintf(page, PAGE_SIZE-1, "%d\n",
210 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
212 static DEVICE_ATTR_RO(stable_pages_required);
214 static struct attribute *bdi_dev_attrs[] = {
215 &dev_attr_read_ahead_kb.attr,
216 &dev_attr_min_ratio.attr,
217 &dev_attr_max_ratio.attr,
218 &dev_attr_stable_pages_required.attr,
221 ATTRIBUTE_GROUPS(bdi_dev);
223 static __init int bdi_class_init(void)
225 bdi_class = class_create(THIS_MODULE, "bdi");
226 if (IS_ERR(bdi_class))
227 return PTR_ERR(bdi_class);
229 bdi_class->dev_groups = bdi_dev_groups;
234 postcore_initcall(bdi_class_init);
236 static int bdi_init(struct backing_dev_info *bdi);
238 static int __init default_bdi_init(void)
242 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND |
247 err = bdi_init(&noop_backing_dev_info);
251 subsys_initcall(default_bdi_init);
254 * This function is used when the first inode for this wb is marked dirty. It
255 * wakes-up the corresponding bdi thread which should then take care of the
256 * periodic background write-out of dirty inodes. Since the write-out would
257 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
258 * set up a timer which wakes the bdi thread up later.
260 * Note, we wouldn't bother setting up the timer, but this function is on the
261 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
262 * by delaying the wake-up.
264 * We have to be careful not to postpone flush work if it is scheduled for
265 * earlier. Thus we use queue_delayed_work().
267 void wb_wakeup_delayed(struct bdi_writeback *wb)
269 unsigned long timeout;
271 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
272 spin_lock_bh(&wb->work_lock);
273 if (test_bit(WB_registered, &wb->state))
274 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
275 spin_unlock_bh(&wb->work_lock);
279 * Initial write bandwidth: 100 MB/s
281 #define INIT_BW (100 << (20 - PAGE_SHIFT))
283 static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
288 memset(wb, 0, sizeof(*wb));
293 wb->last_old_flush = jiffies;
294 INIT_LIST_HEAD(&wb->b_dirty);
295 INIT_LIST_HEAD(&wb->b_io);
296 INIT_LIST_HEAD(&wb->b_more_io);
297 INIT_LIST_HEAD(&wb->b_dirty_time);
298 spin_lock_init(&wb->list_lock);
300 wb->bw_time_stamp = jiffies;
301 wb->balanced_dirty_ratelimit = INIT_BW;
302 wb->dirty_ratelimit = INIT_BW;
303 wb->write_bandwidth = INIT_BW;
304 wb->avg_write_bandwidth = INIT_BW;
306 spin_lock_init(&wb->work_lock);
307 INIT_LIST_HEAD(&wb->work_list);
308 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
309 wb->dirty_sleep = jiffies;
311 err = fprop_local_init_percpu(&wb->completions, gfp);
315 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
316 err = percpu_counter_init(&wb->stat[i], 0, gfp);
318 goto out_destroy_stat;
325 percpu_counter_destroy(&wb->stat[i]);
326 fprop_local_destroy_percpu(&wb->completions);
333 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
336 * Remove bdi from the global list and shutdown any threads we have running
338 static void wb_shutdown(struct bdi_writeback *wb)
340 /* Make sure nobody queues further work */
341 spin_lock_bh(&wb->work_lock);
342 if (!test_and_clear_bit(WB_registered, &wb->state)) {
343 spin_unlock_bh(&wb->work_lock);
346 spin_unlock_bh(&wb->work_lock);
348 cgwb_remove_from_bdi_list(wb);
350 * Drain work list and shutdown the delayed_work. !WB_registered
351 * tells wb_workfn() that @wb is dying and its work_list needs to
352 * be drained no matter what.
354 mod_delayed_work(bdi_wq, &wb->dwork, 0);
355 flush_delayed_work(&wb->dwork);
356 WARN_ON(!list_empty(&wb->work_list));
359 static void wb_exit(struct bdi_writeback *wb)
363 WARN_ON(delayed_work_pending(&wb->dwork));
365 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
366 percpu_counter_destroy(&wb->stat[i]);
368 fprop_local_destroy_percpu(&wb->completions);
369 if (wb != &wb->bdi->wb)
373 #ifdef CONFIG_CGROUP_WRITEBACK
375 #include <linux/memcontrol.h>
378 * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, and memcg->cgwb_list.
379 * bdi->cgwb_tree is also RCU protected.
381 static DEFINE_SPINLOCK(cgwb_lock);
382 static struct workqueue_struct *cgwb_release_wq;
384 static void cgwb_release_workfn(struct work_struct *work)
386 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
388 struct blkcg *blkcg = css_to_blkcg(wb->blkcg_css);
390 mutex_lock(&wb->bdi->cgwb_release_mutex);
393 css_put(wb->memcg_css);
394 css_put(wb->blkcg_css);
395 mutex_unlock(&wb->bdi->cgwb_release_mutex);
397 /* triggers blkg destruction if no online users left */
398 blkcg_unpin_online(blkcg);
400 fprop_local_destroy_percpu(&wb->memcg_completions);
401 percpu_ref_exit(&wb->refcnt);
406 static void cgwb_release(struct percpu_ref *refcnt)
408 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
410 queue_work(cgwb_release_wq, &wb->release_work);
413 static void cgwb_kill(struct bdi_writeback *wb)
415 lockdep_assert_held(&cgwb_lock);
417 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
418 list_del(&wb->memcg_node);
419 list_del(&wb->blkcg_node);
420 percpu_ref_kill(&wb->refcnt);
423 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
425 spin_lock_irq(&cgwb_lock);
426 list_del_rcu(&wb->bdi_node);
427 spin_unlock_irq(&cgwb_lock);
430 static int cgwb_create(struct backing_dev_info *bdi,
431 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
433 struct mem_cgroup *memcg;
434 struct cgroup_subsys_state *blkcg_css;
436 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
437 struct bdi_writeback *wb;
441 memcg = mem_cgroup_from_css(memcg_css);
442 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
443 blkcg = css_to_blkcg(blkcg_css);
444 memcg_cgwb_list = &memcg->cgwb_list;
445 blkcg_cgwb_list = &blkcg->cgwb_list;
447 /* look up again under lock and discard on blkcg mismatch */
448 spin_lock_irqsave(&cgwb_lock, flags);
449 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
450 if (wb && wb->blkcg_css != blkcg_css) {
454 spin_unlock_irqrestore(&cgwb_lock, flags);
458 /* need to create a new one */
459 wb = kmalloc(sizeof(*wb), gfp);
465 ret = wb_init(wb, bdi, gfp);
469 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
473 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
477 wb->memcg_css = memcg_css;
478 wb->blkcg_css = blkcg_css;
479 INIT_WORK(&wb->release_work, cgwb_release_workfn);
480 set_bit(WB_registered, &wb->state);
483 * The root wb determines the registered state of the whole bdi and
484 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
485 * whether they're still online. Don't link @wb if any is dead.
486 * See wb_memcg_offline() and wb_blkcg_offline().
489 spin_lock_irqsave(&cgwb_lock, flags);
490 if (test_bit(WB_registered, &bdi->wb.state) &&
491 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
492 /* we might have raced another instance of this function */
493 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
495 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
496 list_add(&wb->memcg_node, memcg_cgwb_list);
497 list_add(&wb->blkcg_node, blkcg_cgwb_list);
498 blkcg_pin_online(blkcg);
503 spin_unlock_irqrestore(&cgwb_lock, flags);
512 fprop_local_destroy_percpu(&wb->memcg_completions);
514 percpu_ref_exit(&wb->refcnt);
525 * wb_get_lookup - get wb for a given memcg
527 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
529 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
530 * refcount incremented.
532 * This function uses css_get() on @memcg_css and thus expects its refcnt
533 * to be positive on invocation. IOW, rcu_read_lock() protection on
534 * @memcg_css isn't enough. try_get it before calling this function.
536 * A wb is keyed by its associated memcg. As blkcg implicitly enables
537 * memcg on the default hierarchy, memcg association is guaranteed to be
538 * more specific (equal or descendant to the associated blkcg) and thus can
539 * identify both the memcg and blkcg associations.
541 * Because the blkcg associated with a memcg may change as blkcg is enabled
542 * and disabled closer to root in the hierarchy, each wb keeps track of
543 * both the memcg and blkcg associated with it and verifies the blkcg on
544 * each lookup. On mismatch, the existing wb is discarded and a new one is
547 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
548 struct cgroup_subsys_state *memcg_css)
550 struct bdi_writeback *wb;
552 if (!memcg_css->parent)
556 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
558 struct cgroup_subsys_state *blkcg_css;
560 /* see whether the blkcg association has changed */
561 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
562 if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
572 * wb_get_create - get wb for a given memcg, create if necessary
574 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
575 * @gfp: allocation mask to use
577 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
578 * create one. See wb_get_lookup() for more details.
580 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
581 struct cgroup_subsys_state *memcg_css,
584 struct bdi_writeback *wb;
586 might_sleep_if(gfpflags_allow_blocking(gfp));
588 if (!memcg_css->parent)
592 wb = wb_get_lookup(bdi, memcg_css);
593 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
598 static int cgwb_bdi_init(struct backing_dev_info *bdi)
602 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
603 mutex_init(&bdi->cgwb_release_mutex);
604 init_rwsem(&bdi->wb_switch_rwsem);
606 ret = wb_init(&bdi->wb, bdi, GFP_KERNEL);
608 bdi->wb.memcg_css = &root_mem_cgroup->css;
609 bdi->wb.blkcg_css = blkcg_root_css;
614 static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
616 struct radix_tree_iter iter;
618 struct bdi_writeback *wb;
620 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
622 spin_lock_irq(&cgwb_lock);
623 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
625 spin_unlock_irq(&cgwb_lock);
627 mutex_lock(&bdi->cgwb_release_mutex);
628 spin_lock_irq(&cgwb_lock);
629 while (!list_empty(&bdi->wb_list)) {
630 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
632 spin_unlock_irq(&cgwb_lock);
634 spin_lock_irq(&cgwb_lock);
636 spin_unlock_irq(&cgwb_lock);
637 mutex_unlock(&bdi->cgwb_release_mutex);
641 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
642 * @memcg: memcg being offlined
644 * Also prevents creation of any new wb's associated with @memcg.
646 void wb_memcg_offline(struct mem_cgroup *memcg)
648 struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
649 struct bdi_writeback *wb, *next;
651 spin_lock_irq(&cgwb_lock);
652 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
654 memcg_cgwb_list->next = NULL; /* prevent new wb's */
655 spin_unlock_irq(&cgwb_lock);
659 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
660 * @blkcg: blkcg being offlined
662 * Also prevents creation of any new wb's associated with @blkcg.
664 void wb_blkcg_offline(struct blkcg *blkcg)
666 struct bdi_writeback *wb, *next;
668 spin_lock_irq(&cgwb_lock);
669 list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
671 blkcg->cgwb_list.next = NULL; /* prevent new wb's */
672 spin_unlock_irq(&cgwb_lock);
675 static void cgwb_bdi_register(struct backing_dev_info *bdi)
677 spin_lock_irq(&cgwb_lock);
678 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
679 spin_unlock_irq(&cgwb_lock);
682 static int __init cgwb_init(void)
685 * There can be many concurrent release work items overwhelming
686 * system_wq. Put them in a separate wq and limit concurrency.
687 * There's no point in executing many of these in parallel.
689 cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
690 if (!cgwb_release_wq)
695 subsys_initcall(cgwb_init);
697 #else /* CONFIG_CGROUP_WRITEBACK */
699 static int cgwb_bdi_init(struct backing_dev_info *bdi)
701 return wb_init(&bdi->wb, bdi, GFP_KERNEL);
704 static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
706 static void cgwb_bdi_register(struct backing_dev_info *bdi)
708 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
711 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
713 list_del_rcu(&wb->bdi_node);
716 #endif /* CONFIG_CGROUP_WRITEBACK */
718 static int bdi_init(struct backing_dev_info *bdi)
724 kref_init(&bdi->refcnt);
726 bdi->max_ratio = 100;
727 bdi->max_prop_frac = FPROP_FRAC_BASE;
728 INIT_LIST_HEAD(&bdi->bdi_list);
729 INIT_LIST_HEAD(&bdi->wb_list);
730 init_waitqueue_head(&bdi->wb_waitq);
732 ret = cgwb_bdi_init(bdi);
737 struct backing_dev_info *bdi_alloc(int node_id)
739 struct backing_dev_info *bdi;
741 bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id);
751 EXPORT_SYMBOL(bdi_alloc);
753 static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp)
755 struct rb_node **p = &bdi_tree.rb_node;
756 struct rb_node *parent = NULL;
757 struct backing_dev_info *bdi;
759 lockdep_assert_held(&bdi_lock);
763 bdi = rb_entry(parent, struct backing_dev_info, rb_node);
767 else if (bdi->id < id)
779 * bdi_get_by_id - lookup and get bdi from its id
780 * @id: bdi id to lookup
782 * Find bdi matching @id and get it. Returns NULL if the matching bdi
783 * doesn't exist or is already unregistered.
785 struct backing_dev_info *bdi_get_by_id(u64 id)
787 struct backing_dev_info *bdi = NULL;
790 spin_lock_bh(&bdi_lock);
791 p = bdi_lookup_rb_node(id, NULL);
793 bdi = rb_entry(*p, struct backing_dev_info, rb_node);
796 spin_unlock_bh(&bdi_lock);
801 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
804 struct rb_node *parent, **p;
806 if (bdi->dev) /* The driver needs to use separate queues per device */
809 vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args);
810 dev = device_create(bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name);
814 cgwb_bdi_register(bdi);
817 bdi_debug_register(bdi, dev_name(dev));
818 set_bit(WB_registered, &bdi->wb.state);
820 spin_lock_bh(&bdi_lock);
822 bdi->id = ++bdi_id_cursor;
824 p = bdi_lookup_rb_node(bdi->id, &parent);
825 rb_link_node(&bdi->rb_node, parent, p);
826 rb_insert_color(&bdi->rb_node, &bdi_tree);
828 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
830 spin_unlock_bh(&bdi_lock);
832 trace_writeback_bdi_register(bdi);
836 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
842 ret = bdi_register_va(bdi, fmt, args);
846 EXPORT_SYMBOL(bdi_register);
848 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner)
850 WARN_ON_ONCE(bdi->owner);
856 * Remove bdi from bdi_list, and ensure that it is no longer visible
858 static void bdi_remove_from_list(struct backing_dev_info *bdi)
860 spin_lock_bh(&bdi_lock);
861 rb_erase(&bdi->rb_node, &bdi_tree);
862 list_del_rcu(&bdi->bdi_list);
863 spin_unlock_bh(&bdi_lock);
865 synchronize_rcu_expedited();
868 void bdi_unregister(struct backing_dev_info *bdi)
870 /* make sure nobody finds us on the bdi_list anymore */
871 bdi_remove_from_list(bdi);
872 wb_shutdown(&bdi->wb);
873 cgwb_bdi_unregister(bdi);
876 bdi_debug_unregister(bdi);
877 device_unregister(bdi->dev);
882 put_device(bdi->owner);
887 static void release_bdi(struct kref *ref)
889 struct backing_dev_info *bdi =
890 container_of(ref, struct backing_dev_info, refcnt);
892 if (test_bit(WB_registered, &bdi->wb.state))
894 WARN_ON_ONCE(bdi->dev);
899 void bdi_put(struct backing_dev_info *bdi)
901 kref_put(&bdi->refcnt, release_bdi);
903 EXPORT_SYMBOL(bdi_put);
905 const char *bdi_dev_name(struct backing_dev_info *bdi)
907 if (!bdi || !bdi->dev)
908 return bdi_unknown_name;
909 return bdi->dev_name;
911 EXPORT_SYMBOL_GPL(bdi_dev_name);
913 static wait_queue_head_t congestion_wqh[2] = {
914 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
915 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
917 static atomic_t nr_wb_congested[2];
919 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
921 wait_queue_head_t *wqh = &congestion_wqh[sync];
922 enum wb_congested_state bit;
924 bit = sync ? WB_sync_congested : WB_async_congested;
925 if (test_and_clear_bit(bit, &bdi->wb.congested))
926 atomic_dec(&nr_wb_congested[sync]);
927 smp_mb__after_atomic();
928 if (waitqueue_active(wqh))
931 EXPORT_SYMBOL(clear_bdi_congested);
933 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
935 enum wb_congested_state bit;
937 bit = sync ? WB_sync_congested : WB_async_congested;
938 if (!test_and_set_bit(bit, &bdi->wb.congested))
939 atomic_inc(&nr_wb_congested[sync]);
941 EXPORT_SYMBOL(set_bdi_congested);
944 * congestion_wait - wait for a backing_dev to become uncongested
945 * @sync: SYNC or ASYNC IO
946 * @timeout: timeout in jiffies
948 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
949 * write congestion. If no backing_devs are congested then just wait for the
950 * next write to be completed.
952 long congestion_wait(int sync, long timeout)
955 unsigned long start = jiffies;
957 wait_queue_head_t *wqh = &congestion_wqh[sync];
959 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
960 ret = io_schedule_timeout(timeout);
961 finish_wait(wqh, &wait);
963 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
964 jiffies_to_usecs(jiffies - start));
968 EXPORT_SYMBOL(congestion_wait);
971 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
972 * @sync: SYNC or ASYNC IO
973 * @timeout: timeout in jiffies
975 * In the event of a congested backing_dev (any backing_dev) this waits
976 * for up to @timeout jiffies for either a BDI to exit congestion of the
977 * given @sync queue or a write to complete.
979 * The return value is 0 if the sleep is for the full timeout. Otherwise,
980 * it is the number of jiffies that were still remaining when the function
981 * returned. return_value == timeout implies the function did not sleep.
983 long wait_iff_congested(int sync, long timeout)
986 unsigned long start = jiffies;
988 wait_queue_head_t *wqh = &congestion_wqh[sync];
991 * If there is no congestion, yield if necessary instead
992 * of sleeping on the congestion queue
994 if (atomic_read(&nr_wb_congested[sync]) == 0) {
997 /* In case we scheduled, work out time remaining */
998 ret = timeout - (jiffies - start);
1005 /* Sleep until uncongested or a write happens */
1006 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1007 ret = io_schedule_timeout(timeout);
1008 finish_wait(wqh, &wait);
1011 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
1012 jiffies_to_usecs(jiffies - start));
1016 EXPORT_SYMBOL(wait_iff_congested);