2 #include <linux/wait.h>
3 #include <linux/backing-dev.h>
4 #include <linux/kthread.h>
5 #include <linux/freezer.h>
7 #include <linux/pagemap.h>
9 #include <linux/sched.h>
10 #include <linux/module.h>
11 #include <linux/writeback.h>
12 #include <linux/device.h>
13 #include <trace/events/writeback.h>
15 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
17 struct backing_dev_info noop_backing_dev_info = {
19 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
21 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
23 static struct class *bdi_class;
26 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
29 DEFINE_SPINLOCK(bdi_lock);
32 /* bdi_wq serves all asynchronous writeback tasks */
33 struct workqueue_struct *bdi_wq;
35 #ifdef CONFIG_DEBUG_FS
36 #include <linux/debugfs.h>
37 #include <linux/seq_file.h>
39 static struct dentry *bdi_debug_root;
41 static void bdi_debug_init(void)
43 bdi_debug_root = debugfs_create_dir("bdi", NULL);
46 static int bdi_debug_stats_show(struct seq_file *m, void *v)
48 struct backing_dev_info *bdi = m->private;
49 struct bdi_writeback *wb = &bdi->wb;
50 unsigned long background_thresh;
51 unsigned long dirty_thresh;
52 unsigned long wb_thresh;
53 unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
56 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
57 spin_lock(&wb->list_lock);
58 list_for_each_entry(inode, &wb->b_dirty, i_io_list)
60 list_for_each_entry(inode, &wb->b_io, i_io_list)
62 list_for_each_entry(inode, &wb->b_more_io, i_io_list)
64 list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
65 if (inode->i_state & I_DIRTY_TIME)
67 spin_unlock(&wb->list_lock);
69 global_dirty_limits(&background_thresh, &dirty_thresh);
70 wb_thresh = wb_calc_thresh(wb, dirty_thresh);
72 #define K(x) ((x) << (PAGE_SHIFT - 10))
74 "BdiWriteback: %10lu kB\n"
75 "BdiReclaimable: %10lu kB\n"
76 "BdiDirtyThresh: %10lu kB\n"
77 "DirtyThresh: %10lu kB\n"
78 "BackgroundThresh: %10lu kB\n"
79 "BdiDirtied: %10lu kB\n"
80 "BdiWritten: %10lu kB\n"
81 "BdiWriteBandwidth: %10lu kBps\n"
85 "b_dirty_time: %10lu\n"
88 (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
89 (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
93 (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
94 (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
95 (unsigned long) K(wb->write_bandwidth),
100 !list_empty(&bdi->bdi_list), bdi->wb.state);
106 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
108 return single_open(file, bdi_debug_stats_show, inode->i_private);
111 static const struct file_operations bdi_debug_stats_fops = {
112 .open = bdi_debug_stats_open,
115 .release = single_release,
118 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
120 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
121 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
122 bdi, &bdi_debug_stats_fops);
125 static void bdi_debug_unregister(struct backing_dev_info *bdi)
127 debugfs_remove(bdi->debug_stats);
128 debugfs_remove(bdi->debug_dir);
131 static inline void bdi_debug_init(void)
134 static inline void bdi_debug_register(struct backing_dev_info *bdi,
138 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
143 static ssize_t read_ahead_kb_store(struct device *dev,
144 struct device_attribute *attr,
145 const char *buf, size_t count)
147 struct backing_dev_info *bdi = dev_get_drvdata(dev);
148 unsigned long read_ahead_kb;
151 ret = kstrtoul(buf, 10, &read_ahead_kb);
155 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
160 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
162 #define BDI_SHOW(name, expr) \
163 static ssize_t name##_show(struct device *dev, \
164 struct device_attribute *attr, char *page) \
166 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
168 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
170 static DEVICE_ATTR_RW(name);
172 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
174 static ssize_t min_ratio_store(struct device *dev,
175 struct device_attribute *attr, const char *buf, size_t count)
177 struct backing_dev_info *bdi = dev_get_drvdata(dev);
181 ret = kstrtouint(buf, 10, &ratio);
185 ret = bdi_set_min_ratio(bdi, ratio);
191 BDI_SHOW(min_ratio, bdi->min_ratio)
193 static ssize_t max_ratio_store(struct device *dev,
194 struct device_attribute *attr, const char *buf, size_t count)
196 struct backing_dev_info *bdi = dev_get_drvdata(dev);
200 ret = kstrtouint(buf, 10, &ratio);
204 ret = bdi_set_max_ratio(bdi, ratio);
210 BDI_SHOW(max_ratio, bdi->max_ratio)
212 static ssize_t stable_pages_required_show(struct device *dev,
213 struct device_attribute *attr,
216 struct backing_dev_info *bdi = dev_get_drvdata(dev);
218 return snprintf(page, PAGE_SIZE-1, "%d\n",
219 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
221 static DEVICE_ATTR_RO(stable_pages_required);
223 static struct attribute *bdi_dev_attrs[] = {
224 &dev_attr_read_ahead_kb.attr,
225 &dev_attr_min_ratio.attr,
226 &dev_attr_max_ratio.attr,
227 &dev_attr_stable_pages_required.attr,
230 ATTRIBUTE_GROUPS(bdi_dev);
232 static __init int bdi_class_init(void)
234 bdi_class = class_create(THIS_MODULE, "bdi");
235 if (IS_ERR(bdi_class))
236 return PTR_ERR(bdi_class);
238 bdi_class->dev_groups = bdi_dev_groups;
242 postcore_initcall(bdi_class_init);
244 static int __init default_bdi_init(void)
248 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
249 WQ_UNBOUND | WQ_SYSFS, 0);
253 err = bdi_init(&noop_backing_dev_info);
257 subsys_initcall(default_bdi_init);
260 * This function is used when the first inode for this wb is marked dirty. It
261 * wakes-up the corresponding bdi thread which should then take care of the
262 * periodic background write-out of dirty inodes. Since the write-out would
263 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
264 * set up a timer which wakes the bdi thread up later.
266 * Note, we wouldn't bother setting up the timer, but this function is on the
267 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
268 * by delaying the wake-up.
270 * We have to be careful not to postpone flush work if it is scheduled for
271 * earlier. Thus we use queue_delayed_work().
273 void wb_wakeup_delayed(struct bdi_writeback *wb)
275 unsigned long timeout;
277 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
278 spin_lock_bh(&wb->work_lock);
279 if (test_bit(WB_registered, &wb->state))
280 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
281 spin_unlock_bh(&wb->work_lock);
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,
290 int blkcg_id, gfp_t gfp)
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 wb->bw_time_stamp = jiffies;
305 wb->balanced_dirty_ratelimit = INIT_BW;
306 wb->dirty_ratelimit = INIT_BW;
307 wb->write_bandwidth = INIT_BW;
308 wb->avg_write_bandwidth = INIT_BW;
310 spin_lock_init(&wb->work_lock);
311 INIT_LIST_HEAD(&wb->work_list);
312 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
313 wb->dirty_sleep = jiffies;
315 wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp);
319 err = fprop_local_init_percpu(&wb->completions, gfp);
323 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
324 err = percpu_counter_init(&wb->stat[i], 0, gfp);
326 goto out_destroy_stat;
333 percpu_counter_destroy(&wb->stat[i]);
334 fprop_local_destroy_percpu(&wb->completions);
336 wb_congested_put(wb->congested);
341 * Remove bdi from the global list and shutdown any threads we have running
343 static void wb_shutdown(struct bdi_writeback *wb)
345 /* Make sure nobody queues further work */
346 spin_lock_bh(&wb->work_lock);
347 if (!test_and_clear_bit(WB_registered, &wb->state)) {
348 spin_unlock_bh(&wb->work_lock);
351 spin_unlock_bh(&wb->work_lock);
354 * Drain work list and shutdown the delayed_work. !WB_registered
355 * tells wb_workfn() that @wb is dying and its work_list needs to
356 * be drained no matter what.
358 mod_delayed_work(bdi_wq, &wb->dwork, 0);
359 flush_delayed_work(&wb->dwork);
360 WARN_ON(!list_empty(&wb->work_list));
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);
373 wb_congested_put(wb->congested);
376 #ifdef CONFIG_CGROUP_WRITEBACK
378 #include <linux/memcontrol.h>
381 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
382 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU
383 * protected. cgwb_release_wait is used to wait for the completion of cgwb
384 * releases from bdi destruction path.
386 static DEFINE_SPINLOCK(cgwb_lock);
387 static DECLARE_WAIT_QUEUE_HEAD(cgwb_release_wait);
390 * wb_congested_get_create - get or create a wb_congested
391 * @bdi: associated bdi
392 * @blkcg_id: ID of the associated blkcg
393 * @gfp: allocation mask
395 * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one.
396 * The returned wb_congested has its reference count incremented. Returns
399 struct bdi_writeback_congested *
400 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
402 struct bdi_writeback_congested *new_congested = NULL, *congested;
403 struct rb_node **node, *parent;
406 spin_lock_irqsave(&cgwb_lock, flags);
408 node = &bdi->cgwb_congested_tree.rb_node;
411 while (*node != NULL) {
413 congested = container_of(parent, struct bdi_writeback_congested,
415 if (congested->blkcg_id < blkcg_id)
416 node = &parent->rb_left;
417 else if (congested->blkcg_id > blkcg_id)
418 node = &parent->rb_right;
424 /* !found and storage for new one already allocated, insert */
425 congested = new_congested;
426 new_congested = NULL;
427 rb_link_node(&congested->rb_node, parent, node);
428 rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree);
432 spin_unlock_irqrestore(&cgwb_lock, flags);
434 /* allocate storage for new one and retry */
435 new_congested = kzalloc(sizeof(*new_congested), gfp);
439 atomic_set(&new_congested->refcnt, 0);
440 new_congested->bdi = bdi;
441 new_congested->blkcg_id = blkcg_id;
445 atomic_inc(&congested->refcnt);
446 spin_unlock_irqrestore(&cgwb_lock, flags);
447 kfree(new_congested);
452 * wb_congested_put - put a wb_congested
453 * @congested: wb_congested to put
455 * Put @congested and destroy it if the refcnt reaches zero.
457 void wb_congested_put(struct bdi_writeback_congested *congested)
461 local_irq_save(flags);
462 if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) {
463 local_irq_restore(flags);
467 /* bdi might already have been destroyed leaving @congested unlinked */
468 if (congested->bdi) {
469 rb_erase(&congested->rb_node,
470 &congested->bdi->cgwb_congested_tree);
471 congested->bdi = NULL;
474 spin_unlock_irqrestore(&cgwb_lock, flags);
478 static void cgwb_release_workfn(struct work_struct *work)
480 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
482 struct backing_dev_info *bdi = wb->bdi;
484 spin_lock_irq(&cgwb_lock);
485 list_del_rcu(&wb->bdi_node);
486 spin_unlock_irq(&cgwb_lock);
490 css_put(wb->memcg_css);
491 css_put(wb->blkcg_css);
493 fprop_local_destroy_percpu(&wb->memcg_completions);
494 percpu_ref_exit(&wb->refcnt);
498 if (atomic_dec_and_test(&bdi->usage_cnt))
499 wake_up_all(&cgwb_release_wait);
502 static void cgwb_release(struct percpu_ref *refcnt)
504 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
506 schedule_work(&wb->release_work);
509 static void cgwb_kill(struct bdi_writeback *wb)
511 lockdep_assert_held(&cgwb_lock);
513 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
514 list_del(&wb->memcg_node);
515 list_del(&wb->blkcg_node);
516 percpu_ref_kill(&wb->refcnt);
519 static int cgwb_create(struct backing_dev_info *bdi,
520 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
522 struct mem_cgroup *memcg;
523 struct cgroup_subsys_state *blkcg_css;
525 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
526 struct bdi_writeback *wb;
530 memcg = mem_cgroup_from_css(memcg_css);
531 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
532 blkcg = css_to_blkcg(blkcg_css);
533 memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
534 blkcg_cgwb_list = &blkcg->cgwb_list;
536 /* look up again under lock and discard on blkcg mismatch */
537 spin_lock_irqsave(&cgwb_lock, flags);
538 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
539 if (wb && wb->blkcg_css != blkcg_css) {
543 spin_unlock_irqrestore(&cgwb_lock, flags);
547 /* need to create a new one */
548 wb = kmalloc(sizeof(*wb), gfp);
552 ret = wb_init(wb, bdi, blkcg_css->id, gfp);
556 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
560 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
564 wb->memcg_css = memcg_css;
565 wb->blkcg_css = blkcg_css;
566 INIT_WORK(&wb->release_work, cgwb_release_workfn);
567 set_bit(WB_registered, &wb->state);
570 * The root wb determines the registered state of the whole bdi and
571 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
572 * whether they're still online. Don't link @wb if any is dead.
573 * See wb_memcg_offline() and wb_blkcg_offline().
576 spin_lock_irqsave(&cgwb_lock, flags);
577 if (test_bit(WB_registered, &bdi->wb.state) &&
578 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
579 /* we might have raced another instance of this function */
580 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
582 atomic_inc(&bdi->usage_cnt);
583 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
584 list_add(&wb->memcg_node, memcg_cgwb_list);
585 list_add(&wb->blkcg_node, blkcg_cgwb_list);
590 spin_unlock_irqrestore(&cgwb_lock, flags);
599 fprop_local_destroy_percpu(&wb->memcg_completions);
601 percpu_ref_exit(&wb->refcnt);
612 * wb_get_create - get wb for a given memcg, create if necessary
614 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
615 * @gfp: allocation mask to use
617 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
618 * create one. The returned wb has its refcount incremented.
620 * This function uses css_get() on @memcg_css and thus expects its refcnt
621 * to be positive on invocation. IOW, rcu_read_lock() protection on
622 * @memcg_css isn't enough. try_get it before calling this function.
624 * A wb is keyed by its associated memcg. As blkcg implicitly enables
625 * memcg on the default hierarchy, memcg association is guaranteed to be
626 * more specific (equal or descendant to the associated blkcg) and thus can
627 * identify both the memcg and blkcg associations.
629 * Because the blkcg associated with a memcg may change as blkcg is enabled
630 * and disabled closer to root in the hierarchy, each wb keeps track of
631 * both the memcg and blkcg associated with it and verifies the blkcg on
632 * each lookup. On mismatch, the existing wb is discarded and a new one is
635 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
636 struct cgroup_subsys_state *memcg_css,
639 struct bdi_writeback *wb;
641 might_sleep_if(gfpflags_allow_blocking(gfp));
643 if (!memcg_css->parent)
648 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
650 struct cgroup_subsys_state *blkcg_css;
652 /* see whether the blkcg association has changed */
653 blkcg_css = cgroup_get_e_css(memcg_css->cgroup,
655 if (unlikely(wb->blkcg_css != blkcg_css ||
661 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
666 static int cgwb_bdi_init(struct backing_dev_info *bdi)
670 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
671 bdi->cgwb_congested_tree = RB_ROOT;
672 atomic_set(&bdi->usage_cnt, 1);
674 ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
676 bdi->wb.memcg_css = &root_mem_cgroup->css;
677 bdi->wb.blkcg_css = blkcg_root_css;
682 static void cgwb_bdi_destroy(struct backing_dev_info *bdi)
684 struct radix_tree_iter iter;
688 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
690 spin_lock_irq(&cgwb_lock);
692 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
695 while ((rbn = rb_first(&bdi->cgwb_congested_tree))) {
696 struct bdi_writeback_congested *congested =
697 rb_entry(rbn, struct bdi_writeback_congested, rb_node);
699 rb_erase(rbn, &bdi->cgwb_congested_tree);
700 congested->bdi = NULL; /* mark @congested unlinked */
703 spin_unlock_irq(&cgwb_lock);
706 * All cgwb's and their congested states must be shutdown and
707 * released before returning. Drain the usage counter to wait for
708 * all cgwb's and cgwb_congested's ever created on @bdi.
710 atomic_dec(&bdi->usage_cnt);
711 wait_event(cgwb_release_wait, !atomic_read(&bdi->usage_cnt));
715 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
716 * @memcg: memcg being offlined
718 * Also prevents creation of any new wb's associated with @memcg.
720 void wb_memcg_offline(struct mem_cgroup *memcg)
722 LIST_HEAD(to_destroy);
723 struct list_head *memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
724 struct bdi_writeback *wb, *next;
726 spin_lock_irq(&cgwb_lock);
727 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
729 memcg_cgwb_list->next = NULL; /* prevent new wb's */
730 spin_unlock_irq(&cgwb_lock);
734 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
735 * @blkcg: blkcg being offlined
737 * Also prevents creation of any new wb's associated with @blkcg.
739 void wb_blkcg_offline(struct blkcg *blkcg)
741 LIST_HEAD(to_destroy);
742 struct bdi_writeback *wb, *next;
744 spin_lock_irq(&cgwb_lock);
745 list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
747 blkcg->cgwb_list.next = NULL; /* prevent new wb's */
748 spin_unlock_irq(&cgwb_lock);
751 #else /* CONFIG_CGROUP_WRITEBACK */
753 static int cgwb_bdi_init(struct backing_dev_info *bdi)
757 bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL);
758 if (!bdi->wb_congested)
761 err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
763 kfree(bdi->wb_congested);
769 static void cgwb_bdi_destroy(struct backing_dev_info *bdi) { }
771 #endif /* CONFIG_CGROUP_WRITEBACK */
773 int bdi_init(struct backing_dev_info *bdi)
780 bdi->max_ratio = 100;
781 bdi->max_prop_frac = FPROP_FRAC_BASE;
782 INIT_LIST_HEAD(&bdi->bdi_list);
783 INIT_LIST_HEAD(&bdi->wb_list);
784 init_waitqueue_head(&bdi->wb_waitq);
786 ret = cgwb_bdi_init(bdi);
788 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
792 EXPORT_SYMBOL(bdi_init);
794 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
795 const char *fmt, ...)
800 if (bdi->dev) /* The driver needs to use separate queues per device */
804 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
811 bdi_debug_register(bdi, dev_name(dev));
812 set_bit(WB_registered, &bdi->wb.state);
814 spin_lock_bh(&bdi_lock);
815 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
816 spin_unlock_bh(&bdi_lock);
818 trace_writeback_bdi_register(bdi);
821 EXPORT_SYMBOL(bdi_register);
823 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
825 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
827 EXPORT_SYMBOL(bdi_register_dev);
829 int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner)
833 rc = bdi_register(bdi, NULL, "%u:%u", MAJOR(owner->devt),
841 EXPORT_SYMBOL(bdi_register_owner);
844 * Remove bdi from bdi_list, and ensure that it is no longer visible
846 static void bdi_remove_from_list(struct backing_dev_info *bdi)
848 spin_lock_bh(&bdi_lock);
849 list_del_rcu(&bdi->bdi_list);
850 spin_unlock_bh(&bdi_lock);
852 synchronize_rcu_expedited();
855 void bdi_unregister(struct backing_dev_info *bdi)
857 /* make sure nobody finds us on the bdi_list anymore */
858 bdi_remove_from_list(bdi);
859 wb_shutdown(&bdi->wb);
860 cgwb_bdi_destroy(bdi);
863 bdi_debug_unregister(bdi);
864 device_unregister(bdi->dev);
869 put_device(bdi->owner);
874 void bdi_exit(struct backing_dev_info *bdi)
876 WARN_ON_ONCE(bdi->dev);
880 void bdi_destroy(struct backing_dev_info *bdi)
885 EXPORT_SYMBOL(bdi_destroy);
888 * For use from filesystems to quickly init and register a bdi associated
889 * with dirty writeback
891 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
896 bdi->capabilities = 0;
901 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
902 atomic_long_inc_return(&bdi_seq));
910 EXPORT_SYMBOL(bdi_setup_and_register);
912 static wait_queue_head_t congestion_wqh[2] = {
913 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
914 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
916 static atomic_t nr_wb_congested[2];
918 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync)
920 wait_queue_head_t *wqh = &congestion_wqh[sync];
921 enum wb_congested_state bit;
923 bit = sync ? WB_sync_congested : WB_async_congested;
924 if (test_and_clear_bit(bit, &congested->state))
925 atomic_dec(&nr_wb_congested[sync]);
926 smp_mb__after_atomic();
927 if (waitqueue_active(wqh))
930 EXPORT_SYMBOL(clear_wb_congested);
932 void set_wb_congested(struct bdi_writeback_congested *congested, int sync)
934 enum wb_congested_state bit;
936 bit = sync ? WB_sync_congested : WB_async_congested;
937 if (!test_and_set_bit(bit, &congested->state))
938 atomic_inc(&nr_wb_congested[sync]);
940 EXPORT_SYMBOL(set_wb_congested);
943 * congestion_wait - wait for a backing_dev to become uncongested
944 * @sync: SYNC or ASYNC IO
945 * @timeout: timeout in jiffies
947 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
948 * write congestion. If no backing_devs are congested then just wait for the
949 * next write to be completed.
951 long congestion_wait(int sync, long timeout)
954 unsigned long start = jiffies;
956 wait_queue_head_t *wqh = &congestion_wqh[sync];
958 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
959 ret = io_schedule_timeout(timeout);
960 finish_wait(wqh, &wait);
962 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
963 jiffies_to_usecs(jiffies - start));
967 EXPORT_SYMBOL(congestion_wait);
970 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
971 * @pgdat: A pgdat to check if it is heavily congested
972 * @sync: SYNC or ASYNC IO
973 * @timeout: timeout in jiffies
975 * In the event of a congested backing_dev (any backing_dev) and the given
976 * @pgdat has experienced recent congestion, this waits for up to @timeout
977 * jiffies for either a BDI to exit congestion of the given @sync queue
978 * or a write to complete.
980 * In the absence of pgdat congestion, cond_resched() is called to yield
981 * the processor if necessary but otherwise does not sleep.
983 * The return value is 0 if the sleep is for the full timeout. Otherwise,
984 * it is the number of jiffies that were still remaining when the function
985 * returned. return_value == timeout implies the function did not sleep.
987 long wait_iff_congested(struct pglist_data *pgdat, int sync, long timeout)
990 unsigned long start = jiffies;
992 wait_queue_head_t *wqh = &congestion_wqh[sync];
995 * If there is no congestion, or heavy congestion is not being
996 * encountered in the current pgdat, yield if necessary instead
997 * of sleeping on the congestion queue
999 if (atomic_read(&nr_wb_congested[sync]) == 0 ||
1000 !test_bit(PGDAT_CONGESTED, &pgdat->flags)) {
1003 /* In case we scheduled, work out time remaining */
1004 ret = timeout - (jiffies - start);
1011 /* Sleep until uncongested or a write happens */
1012 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1013 ret = io_schedule_timeout(timeout);
1014 finish_wait(wqh, &wait);
1017 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
1018 jiffies_to_usecs(jiffies - start));
1022 EXPORT_SYMBOL(wait_iff_congested);
1024 int pdflush_proc_obsolete(struct ctl_table *table, int write,
1025 void __user *buffer, size_t *lenp, loff_t *ppos)
1027 char kbuf[] = "0\n";
1029 if (*ppos || *lenp < sizeof(kbuf)) {
1034 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
1036 pr_warn_once("%s exported in /proc is scheduled for removal\n",