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/mm.h>
12 #include <linux/sched.h>
13 #include <linux/module.h>
14 #include <linux/writeback.h>
15 #include <linux/device.h>
16 #include <trace/events/writeback.h>
18 struct backing_dev_info noop_backing_dev_info;
19 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
21 static struct class *bdi_class;
22 static const char *bdi_unknown_name = "(unknown)";
25 * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU
26 * reader side locking.
28 DEFINE_SPINLOCK(bdi_lock);
29 static u64 bdi_id_cursor;
30 static struct rb_root bdi_tree = RB_ROOT;
33 /* bdi_wq serves all asynchronous writeback tasks */
34 struct workqueue_struct *bdi_wq;
36 #define K(x) ((x) << (PAGE_SHIFT - 10))
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)
181 static ssize_t max_ratio_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_max_ratio(bdi, ratio);
198 BDI_SHOW(max_ratio, bdi->max_ratio)
200 static ssize_t stable_pages_required_show(struct device *dev,
201 struct device_attribute *attr,
205 "the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n");
206 return sysfs_emit(buf, "%d\n", 0);
208 static DEVICE_ATTR_RO(stable_pages_required);
210 static struct attribute *bdi_dev_attrs[] = {
211 &dev_attr_read_ahead_kb.attr,
212 &dev_attr_min_ratio.attr,
213 &dev_attr_max_ratio.attr,
214 &dev_attr_stable_pages_required.attr,
217 ATTRIBUTE_GROUPS(bdi_dev);
219 static __init int bdi_class_init(void)
221 bdi_class = class_create(THIS_MODULE, "bdi");
222 if (IS_ERR(bdi_class))
223 return PTR_ERR(bdi_class);
225 bdi_class->dev_groups = bdi_dev_groups;
230 postcore_initcall(bdi_class_init);
232 static int bdi_init(struct backing_dev_info *bdi);
234 static int __init default_bdi_init(void)
238 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND |
243 err = bdi_init(&noop_backing_dev_info);
247 subsys_initcall(default_bdi_init);
250 * This function is used when the first inode for this wb is marked dirty. It
251 * wakes-up the corresponding bdi thread which should then take care of the
252 * periodic background write-out of dirty inodes. Since the write-out would
253 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
254 * set up a timer which wakes the bdi thread up later.
256 * Note, we wouldn't bother setting up the timer, but this function is on the
257 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
258 * by delaying the wake-up.
260 * We have to be careful not to postpone flush work if it is scheduled for
261 * earlier. Thus we use queue_delayed_work().
263 void wb_wakeup_delayed(struct bdi_writeback *wb)
265 unsigned long timeout;
267 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
268 spin_lock_bh(&wb->work_lock);
269 if (test_bit(WB_registered, &wb->state))
270 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
271 spin_unlock_bh(&wb->work_lock);
275 * Initial write bandwidth: 100 MB/s
277 #define INIT_BW (100 << (20 - PAGE_SHIFT))
279 static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
284 memset(wb, 0, sizeof(*wb));
289 wb->last_old_flush = jiffies;
290 INIT_LIST_HEAD(&wb->b_dirty);
291 INIT_LIST_HEAD(&wb->b_io);
292 INIT_LIST_HEAD(&wb->b_more_io);
293 INIT_LIST_HEAD(&wb->b_dirty_time);
294 spin_lock_init(&wb->list_lock);
296 wb->bw_time_stamp = jiffies;
297 wb->balanced_dirty_ratelimit = INIT_BW;
298 wb->dirty_ratelimit = INIT_BW;
299 wb->write_bandwidth = INIT_BW;
300 wb->avg_write_bandwidth = INIT_BW;
302 spin_lock_init(&wb->work_lock);
303 INIT_LIST_HEAD(&wb->work_list);
304 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
305 wb->dirty_sleep = jiffies;
307 err = fprop_local_init_percpu(&wb->completions, gfp);
311 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
312 err = percpu_counter_init(&wb->stat[i], 0, gfp);
314 goto out_destroy_stat;
321 percpu_counter_destroy(&wb->stat[i]);
322 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_bh(&wb->work_lock);
338 if (!test_and_clear_bit(WB_registered, &wb->state)) {
339 spin_unlock_bh(&wb->work_lock);
342 spin_unlock_bh(&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));
355 static void wb_exit(struct bdi_writeback *wb)
359 WARN_ON(delayed_work_pending(&wb->dwork));
361 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
362 percpu_counter_destroy(&wb->stat[i]);
364 fprop_local_destroy_percpu(&wb->completions);
365 if (wb != &wb->bdi->wb)
369 #ifdef CONFIG_CGROUP_WRITEBACK
371 #include <linux/memcontrol.h>
374 * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, and memcg->cgwb_list.
375 * bdi->cgwb_tree is also RCU protected.
377 static DEFINE_SPINLOCK(cgwb_lock);
378 static struct workqueue_struct *cgwb_release_wq;
380 static void cgwb_release_workfn(struct work_struct *work)
382 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
384 struct blkcg *blkcg = css_to_blkcg(wb->blkcg_css);
386 mutex_lock(&wb->bdi->cgwb_release_mutex);
389 css_put(wb->memcg_css);
390 css_put(wb->blkcg_css);
391 mutex_unlock(&wb->bdi->cgwb_release_mutex);
393 /* triggers blkg destruction if no online users left */
394 blkcg_unpin_online(blkcg);
396 fprop_local_destroy_percpu(&wb->memcg_completions);
397 percpu_ref_exit(&wb->refcnt);
399 WARN_ON_ONCE(!list_empty(&wb->b_attached));
403 static void cgwb_release(struct percpu_ref *refcnt)
405 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
407 queue_work(cgwb_release_wq, &wb->release_work);
410 static void cgwb_kill(struct bdi_writeback *wb)
412 lockdep_assert_held(&cgwb_lock);
414 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
415 list_del(&wb->memcg_node);
416 list_del(&wb->blkcg_node);
417 percpu_ref_kill(&wb->refcnt);
420 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
422 spin_lock_irq(&cgwb_lock);
423 list_del_rcu(&wb->bdi_node);
424 spin_unlock_irq(&cgwb_lock);
427 static int cgwb_create(struct backing_dev_info *bdi,
428 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
430 struct mem_cgroup *memcg;
431 struct cgroup_subsys_state *blkcg_css;
433 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
434 struct bdi_writeback *wb;
438 memcg = mem_cgroup_from_css(memcg_css);
439 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
440 blkcg = css_to_blkcg(blkcg_css);
441 memcg_cgwb_list = &memcg->cgwb_list;
442 blkcg_cgwb_list = &blkcg->cgwb_list;
444 /* look up again under lock and discard on blkcg mismatch */
445 spin_lock_irqsave(&cgwb_lock, flags);
446 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
447 if (wb && wb->blkcg_css != blkcg_css) {
451 spin_unlock_irqrestore(&cgwb_lock, flags);
455 /* need to create a new one */
456 wb = kmalloc(sizeof(*wb), gfp);
462 ret = wb_init(wb, bdi, gfp);
466 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
470 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
474 wb->memcg_css = memcg_css;
475 wb->blkcg_css = blkcg_css;
476 INIT_LIST_HEAD(&wb->b_attached);
477 INIT_WORK(&wb->release_work, cgwb_release_workfn);
478 set_bit(WB_registered, &wb->state);
481 * The root wb determines the registered state of the whole bdi and
482 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
483 * whether they're still online. Don't link @wb if any is dead.
484 * See wb_memcg_offline() and wb_blkcg_offline().
487 spin_lock_irqsave(&cgwb_lock, flags);
488 if (test_bit(WB_registered, &bdi->wb.state) &&
489 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
490 /* we might have raced another instance of this function */
491 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
493 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
494 list_add(&wb->memcg_node, memcg_cgwb_list);
495 list_add(&wb->blkcg_node, blkcg_cgwb_list);
496 blkcg_pin_online(blkcg);
501 spin_unlock_irqrestore(&cgwb_lock, flags);
510 fprop_local_destroy_percpu(&wb->memcg_completions);
512 percpu_ref_exit(&wb->refcnt);
523 * wb_get_lookup - get wb for a given memcg
525 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
527 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
528 * refcount incremented.
530 * This function uses css_get() on @memcg_css and thus expects its refcnt
531 * to be positive on invocation. IOW, rcu_read_lock() protection on
532 * @memcg_css isn't enough. try_get it before calling this function.
534 * A wb is keyed by its associated memcg. As blkcg implicitly enables
535 * memcg on the default hierarchy, memcg association is guaranteed to be
536 * more specific (equal or descendant to the associated blkcg) and thus can
537 * identify both the memcg and blkcg associations.
539 * Because the blkcg associated with a memcg may change as blkcg is enabled
540 * and disabled closer to root in the hierarchy, each wb keeps track of
541 * both the memcg and blkcg associated with it and verifies the blkcg on
542 * each lookup. On mismatch, the existing wb is discarded and a new one is
545 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
546 struct cgroup_subsys_state *memcg_css)
548 struct bdi_writeback *wb;
550 if (!memcg_css->parent)
554 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
556 struct cgroup_subsys_state *blkcg_css;
558 /* see whether the blkcg association has changed */
559 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
560 if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
570 * wb_get_create - get wb for a given memcg, create if necessary
572 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
573 * @gfp: allocation mask to use
575 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
576 * create one. See wb_get_lookup() for more details.
578 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
579 struct cgroup_subsys_state *memcg_css,
582 struct bdi_writeback *wb;
586 if (!memcg_css->parent)
590 wb = wb_get_lookup(bdi, memcg_css);
591 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
596 static int cgwb_bdi_init(struct backing_dev_info *bdi)
600 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
601 mutex_init(&bdi->cgwb_release_mutex);
602 init_rwsem(&bdi->wb_switch_rwsem);
604 ret = wb_init(&bdi->wb, bdi, GFP_KERNEL);
606 bdi->wb.memcg_css = &root_mem_cgroup->css;
607 bdi->wb.blkcg_css = blkcg_root_css;
612 static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
614 struct radix_tree_iter iter;
616 struct bdi_writeback *wb;
618 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
620 spin_lock_irq(&cgwb_lock);
621 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
623 spin_unlock_irq(&cgwb_lock);
625 mutex_lock(&bdi->cgwb_release_mutex);
626 spin_lock_irq(&cgwb_lock);
627 while (!list_empty(&bdi->wb_list)) {
628 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
630 spin_unlock_irq(&cgwb_lock);
632 spin_lock_irq(&cgwb_lock);
634 spin_unlock_irq(&cgwb_lock);
635 mutex_unlock(&bdi->cgwb_release_mutex);
639 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
640 * @memcg: memcg being offlined
642 * Also prevents creation of any new wb's associated with @memcg.
644 void wb_memcg_offline(struct mem_cgroup *memcg)
646 struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
647 struct bdi_writeback *wb, *next;
649 spin_lock_irq(&cgwb_lock);
650 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
652 memcg_cgwb_list->next = NULL; /* prevent new wb's */
653 spin_unlock_irq(&cgwb_lock);
657 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
658 * @blkcg: blkcg being offlined
660 * Also prevents creation of any new wb's associated with @blkcg.
662 void wb_blkcg_offline(struct blkcg *blkcg)
664 struct bdi_writeback *wb, *next;
666 spin_lock_irq(&cgwb_lock);
667 list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
669 blkcg->cgwb_list.next = NULL; /* prevent new wb's */
670 spin_unlock_irq(&cgwb_lock);
673 static void cgwb_bdi_register(struct backing_dev_info *bdi)
675 spin_lock_irq(&cgwb_lock);
676 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
677 spin_unlock_irq(&cgwb_lock);
680 static int __init cgwb_init(void)
683 * There can be many concurrent release work items overwhelming
684 * system_wq. Put them in a separate wq and limit concurrency.
685 * There's no point in executing many of these in parallel.
687 cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
688 if (!cgwb_release_wq)
693 subsys_initcall(cgwb_init);
695 #else /* CONFIG_CGROUP_WRITEBACK */
697 static int cgwb_bdi_init(struct backing_dev_info *bdi)
699 return wb_init(&bdi->wb, bdi, GFP_KERNEL);
702 static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
704 static void cgwb_bdi_register(struct backing_dev_info *bdi)
706 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
709 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
711 list_del_rcu(&wb->bdi_node);
714 #endif /* CONFIG_CGROUP_WRITEBACK */
716 static int bdi_init(struct backing_dev_info *bdi)
722 kref_init(&bdi->refcnt);
724 bdi->max_ratio = 100;
725 bdi->max_prop_frac = FPROP_FRAC_BASE;
726 INIT_LIST_HEAD(&bdi->bdi_list);
727 INIT_LIST_HEAD(&bdi->wb_list);
728 init_waitqueue_head(&bdi->wb_waitq);
730 ret = cgwb_bdi_init(bdi);
735 struct backing_dev_info *bdi_alloc(int node_id)
737 struct backing_dev_info *bdi;
739 bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id);
747 bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT;
748 bdi->ra_pages = VM_READAHEAD_PAGES;
749 bdi->io_pages = VM_READAHEAD_PAGES;
752 EXPORT_SYMBOL(bdi_alloc);
754 static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp)
756 struct rb_node **p = &bdi_tree.rb_node;
757 struct rb_node *parent = NULL;
758 struct backing_dev_info *bdi;
760 lockdep_assert_held(&bdi_lock);
764 bdi = rb_entry(parent, struct backing_dev_info, rb_node);
768 else if (bdi->id < id)
780 * bdi_get_by_id - lookup and get bdi from its id
781 * @id: bdi id to lookup
783 * Find bdi matching @id and get it. Returns NULL if the matching bdi
784 * doesn't exist or is already unregistered.
786 struct backing_dev_info *bdi_get_by_id(u64 id)
788 struct backing_dev_info *bdi = NULL;
791 spin_lock_bh(&bdi_lock);
792 p = bdi_lookup_rb_node(id, NULL);
794 bdi = rb_entry(*p, struct backing_dev_info, rb_node);
797 spin_unlock_bh(&bdi_lock);
802 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
805 struct rb_node *parent, **p;
807 if (bdi->dev) /* The driver needs to use separate queues per device */
810 vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args);
811 dev = device_create(bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name);
815 cgwb_bdi_register(bdi);
818 bdi_debug_register(bdi, dev_name(dev));
819 set_bit(WB_registered, &bdi->wb.state);
821 spin_lock_bh(&bdi_lock);
823 bdi->id = ++bdi_id_cursor;
825 p = bdi_lookup_rb_node(bdi->id, &parent);
826 rb_link_node(&bdi->rb_node, parent, p);
827 rb_insert_color(&bdi->rb_node, &bdi_tree);
829 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
831 spin_unlock_bh(&bdi_lock);
833 trace_writeback_bdi_register(bdi);
837 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
843 ret = bdi_register_va(bdi, fmt, args);
847 EXPORT_SYMBOL(bdi_register);
849 void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner)
851 WARN_ON_ONCE(bdi->owner);
857 * Remove bdi from bdi_list, and ensure that it is no longer visible
859 static void bdi_remove_from_list(struct backing_dev_info *bdi)
861 spin_lock_bh(&bdi_lock);
862 rb_erase(&bdi->rb_node, &bdi_tree);
863 list_del_rcu(&bdi->bdi_list);
864 spin_unlock_bh(&bdi_lock);
866 synchronize_rcu_expedited();
869 void bdi_unregister(struct backing_dev_info *bdi)
871 /* make sure nobody finds us on the bdi_list anymore */
872 bdi_remove_from_list(bdi);
873 wb_shutdown(&bdi->wb);
874 cgwb_bdi_unregister(bdi);
877 bdi_debug_unregister(bdi);
878 device_unregister(bdi->dev);
883 put_device(bdi->owner);
888 static void release_bdi(struct kref *ref)
890 struct backing_dev_info *bdi =
891 container_of(ref, struct backing_dev_info, refcnt);
893 if (test_bit(WB_registered, &bdi->wb.state))
895 WARN_ON_ONCE(bdi->dev);
900 void bdi_put(struct backing_dev_info *bdi)
902 kref_put(&bdi->refcnt, release_bdi);
904 EXPORT_SYMBOL(bdi_put);
906 const char *bdi_dev_name(struct backing_dev_info *bdi)
908 if (!bdi || !bdi->dev)
909 return bdi_unknown_name;
910 return bdi->dev_name;
912 EXPORT_SYMBOL_GPL(bdi_dev_name);
914 static wait_queue_head_t congestion_wqh[2] = {
915 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
916 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
918 static atomic_t nr_wb_congested[2];
920 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
922 wait_queue_head_t *wqh = &congestion_wqh[sync];
923 enum wb_congested_state bit;
925 bit = sync ? WB_sync_congested : WB_async_congested;
926 if (test_and_clear_bit(bit, &bdi->wb.congested))
927 atomic_dec(&nr_wb_congested[sync]);
928 smp_mb__after_atomic();
929 if (waitqueue_active(wqh))
932 EXPORT_SYMBOL(clear_bdi_congested);
934 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
936 enum wb_congested_state bit;
938 bit = sync ? WB_sync_congested : WB_async_congested;
939 if (!test_and_set_bit(bit, &bdi->wb.congested))
940 atomic_inc(&nr_wb_congested[sync]);
942 EXPORT_SYMBOL(set_bdi_congested);
945 * congestion_wait - wait for a backing_dev to become uncongested
946 * @sync: SYNC or ASYNC IO
947 * @timeout: timeout in jiffies
949 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
950 * write congestion. If no backing_devs are congested then just wait for the
951 * next write to be completed.
953 long congestion_wait(int sync, long timeout)
956 unsigned long start = jiffies;
958 wait_queue_head_t *wqh = &congestion_wqh[sync];
960 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
961 ret = io_schedule_timeout(timeout);
962 finish_wait(wqh, &wait);
964 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
965 jiffies_to_usecs(jiffies - start));
969 EXPORT_SYMBOL(congestion_wait);
972 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
973 * @sync: SYNC or ASYNC IO
974 * @timeout: timeout in jiffies
976 * In the event of a congested backing_dev (any backing_dev) this waits
977 * for up to @timeout jiffies for either a BDI to exit congestion of the
978 * given @sync queue or a write to complete.
980 * The return value is 0 if the sleep is for the full timeout. Otherwise,
981 * it is the number of jiffies that were still remaining when the function
982 * returned. return_value == timeout implies the function did not sleep.
984 long wait_iff_congested(int sync, long timeout)
987 unsigned long start = jiffies;
989 wait_queue_head_t *wqh = &congestion_wqh[sync];
992 * If there is no congestion, yield if necessary instead
993 * of sleeping on the congestion queue
995 if (atomic_read(&nr_wb_congested[sync]) == 0) {
998 /* In case we scheduled, work out time remaining */
999 ret = timeout - (jiffies - start);
1006 /* Sleep until uncongested or a write happens */
1007 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1008 ret = io_schedule_timeout(timeout);
1009 finish_wait(wqh, &wait);
1012 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
1013 jiffies_to_usecs(jiffies - start));
1017 EXPORT_SYMBOL(wait_iff_congested);