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_wb_list)
60 list_for_each_entry(inode, &wb->b_io, i_wb_list)
62 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
64 list_for_each_entry(inode, &wb->b_dirty_time, i_wb_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,
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);
314 err = fprop_local_init_percpu(&wb->completions, gfp);
318 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
319 err = percpu_counter_init(&wb->stat[i], 0, gfp);
322 percpu_counter_destroy(&wb->stat[i]);
323 fprop_local_destroy_percpu(&wb->completions);
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);
345 * Drain work list and shutdown the delayed_work. !WB_registered
346 * tells wb_workfn() that @wb is dying and its work_list needs to
347 * be drained no matter what.
349 mod_delayed_work(bdi_wq, &wb->dwork, 0);
350 flush_delayed_work(&wb->dwork);
351 WARN_ON(!list_empty(&wb->work_list));
354 static void wb_exit(struct bdi_writeback *wb)
358 WARN_ON(delayed_work_pending(&wb->dwork));
360 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
361 percpu_counter_destroy(&wb->stat[i]);
363 fprop_local_destroy_percpu(&wb->completions);
366 #ifdef CONFIG_CGROUP_WRITEBACK
368 #include <linux/memcontrol.h>
371 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
372 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU
373 * protected. cgwb_release_wait is used to wait for the completion of cgwb
374 * releases from bdi destruction path.
376 static DEFINE_SPINLOCK(cgwb_lock);
377 static DECLARE_WAIT_QUEUE_HEAD(cgwb_release_wait);
380 * wb_congested_get_create - get or create a wb_congested
381 * @bdi: associated bdi
382 * @blkcg_id: ID of the associated blkcg
383 * @gfp: allocation mask
385 * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one.
386 * The returned wb_congested has its reference count incremented. Returns
389 struct bdi_writeback_congested *
390 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
392 struct bdi_writeback_congested *new_congested = NULL, *congested;
393 struct rb_node **node, *parent;
397 return &bdi->wb_congested;
399 spin_lock_irqsave(&cgwb_lock, flags);
401 node = &bdi->cgwb_congested_tree.rb_node;
404 while (*node != NULL) {
406 congested = container_of(parent, struct bdi_writeback_congested,
408 if (congested->blkcg_id < blkcg_id)
409 node = &parent->rb_left;
410 else if (congested->blkcg_id > blkcg_id)
411 node = &parent->rb_right;
417 /* !found and storage for new one already allocated, insert */
418 congested = new_congested;
419 new_congested = NULL;
420 rb_link_node(&congested->rb_node, parent, node);
421 rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree);
422 atomic_inc(&bdi->usage_cnt);
426 spin_unlock_irqrestore(&cgwb_lock, flags);
428 /* allocate storage for new one and retry */
429 new_congested = kzalloc(sizeof(*new_congested), gfp);
433 atomic_set(&new_congested->refcnt, 0);
434 new_congested->bdi = bdi;
435 new_congested->blkcg_id = blkcg_id;
439 atomic_inc(&congested->refcnt);
440 spin_unlock_irqrestore(&cgwb_lock, flags);
441 kfree(new_congested);
446 * wb_congested_put - put a wb_congested
447 * @congested: wb_congested to put
449 * Put @congested and destroy it if the refcnt reaches zero.
451 void wb_congested_put(struct bdi_writeback_congested *congested)
453 struct backing_dev_info *bdi = congested->bdi;
456 if (congested->blkcg_id == 1)
459 local_irq_save(flags);
460 if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) {
461 local_irq_restore(flags);
465 rb_erase(&congested->rb_node, &congested->bdi->cgwb_congested_tree);
466 spin_unlock_irqrestore(&cgwb_lock, flags);
469 if (atomic_dec_and_test(&bdi->usage_cnt))
470 wake_up_all(&cgwb_release_wait);
473 static void cgwb_release_workfn(struct work_struct *work)
475 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
477 struct backing_dev_info *bdi = wb->bdi;
481 css_put(wb->memcg_css);
482 css_put(wb->blkcg_css);
483 wb_congested_put(wb->congested);
485 fprop_local_destroy_percpu(&wb->memcg_completions);
486 percpu_ref_exit(&wb->refcnt);
490 if (atomic_dec_and_test(&bdi->usage_cnt))
491 wake_up_all(&cgwb_release_wait);
494 static void cgwb_release(struct percpu_ref *refcnt)
496 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
498 schedule_work(&wb->release_work);
501 static void cgwb_kill(struct bdi_writeback *wb)
503 lockdep_assert_held(&cgwb_lock);
505 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
506 list_del(&wb->memcg_node);
507 list_del(&wb->blkcg_node);
508 percpu_ref_kill(&wb->refcnt);
511 static int cgwb_create(struct backing_dev_info *bdi,
512 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
514 struct mem_cgroup *memcg;
515 struct cgroup_subsys_state *blkcg_css;
517 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
518 struct bdi_writeback *wb;
522 memcg = mem_cgroup_from_css(memcg_css);
523 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &blkio_cgrp_subsys);
524 blkcg = css_to_blkcg(blkcg_css);
525 memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
526 blkcg_cgwb_list = &blkcg->cgwb_list;
528 /* look up again under lock and discard on blkcg mismatch */
529 spin_lock_irqsave(&cgwb_lock, flags);
530 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
531 if (wb && wb->blkcg_css != blkcg_css) {
535 spin_unlock_irqrestore(&cgwb_lock, flags);
539 /* need to create a new one */
540 wb = kmalloc(sizeof(*wb), gfp);
544 ret = wb_init(wb, bdi, gfp);
548 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
552 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
556 wb->congested = wb_congested_get_create(bdi, blkcg_css->id, gfp);
557 if (!wb->congested) {
562 wb->memcg_css = memcg_css;
563 wb->blkcg_css = blkcg_css;
564 INIT_WORK(&wb->release_work, cgwb_release_workfn);
565 set_bit(WB_registered, &wb->state);
568 * The root wb determines the registered state of the whole bdi and
569 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
570 * whether they're still online. Don't link @wb if any is dead.
571 * See wb_memcg_offline() and wb_blkcg_offline().
574 spin_lock_irqsave(&cgwb_lock, flags);
575 if (test_bit(WB_registered, &bdi->wb.state) &&
576 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
577 /* we might have raced another instance of this function */
578 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
580 atomic_inc(&bdi->usage_cnt);
581 list_add(&wb->memcg_node, memcg_cgwb_list);
582 list_add(&wb->blkcg_node, blkcg_cgwb_list);
587 spin_unlock_irqrestore(&cgwb_lock, flags);
591 goto err_put_congested;
596 wb_congested_put(wb->congested);
598 fprop_local_destroy_percpu(&wb->memcg_completions);
600 percpu_ref_exit(&wb->refcnt);
611 * wb_get_create - get wb for a given memcg, create if necessary
613 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
614 * @gfp: allocation mask to use
616 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
617 * create one. The returned wb has its refcount incremented.
619 * This function uses css_get() on @memcg_css and thus expects its refcnt
620 * to be positive on invocation. IOW, rcu_read_lock() protection on
621 * @memcg_css isn't enough. try_get it before calling this function.
623 * A wb is keyed by its associated memcg. As blkcg implicitly enables
624 * memcg on the default hierarchy, memcg association is guaranteed to be
625 * more specific (equal or descendant to the associated blkcg) and thus can
626 * identify both the memcg and blkcg associations.
628 * Because the blkcg associated with a memcg may change as blkcg is enabled
629 * and disabled closer to root in the hierarchy, each wb keeps track of
630 * both the memcg and blkcg associated with it and verifies the blkcg on
631 * each lookup. On mismatch, the existing wb is discarded and a new one is
634 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
635 struct cgroup_subsys_state *memcg_css,
638 struct bdi_writeback *wb;
640 might_sleep_if(gfp & __GFP_WAIT);
642 if (!memcg_css->parent)
647 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
649 struct cgroup_subsys_state *blkcg_css;
651 /* see whether the blkcg association has changed */
652 blkcg_css = cgroup_get_e_css(memcg_css->cgroup,
654 if (unlikely(wb->blkcg_css != blkcg_css ||
660 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
665 static void cgwb_bdi_init(struct backing_dev_info *bdi)
667 bdi->wb.memcg_css = mem_cgroup_root_css;
668 bdi->wb.blkcg_css = blkcg_root_css;
669 bdi->wb_congested.blkcg_id = 1;
670 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
671 bdi->cgwb_congested_tree = RB_ROOT;
672 atomic_set(&bdi->usage_cnt, 1);
675 static void cgwb_bdi_destroy(struct backing_dev_info *bdi)
677 struct radix_tree_iter iter;
680 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
682 spin_lock_irq(&cgwb_lock);
683 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
685 spin_unlock_irq(&cgwb_lock);
688 * All cgwb's and their congested states must be shutdown and
689 * released before returning. Drain the usage counter to wait for
690 * all cgwb's and cgwb_congested's ever created on @bdi.
692 atomic_dec(&bdi->usage_cnt);
693 wait_event(cgwb_release_wait, !atomic_read(&bdi->usage_cnt));
697 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
698 * @memcg: memcg being offlined
700 * Also prevents creation of any new wb's associated with @memcg.
702 void wb_memcg_offline(struct mem_cgroup *memcg)
704 LIST_HEAD(to_destroy);
705 struct list_head *memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
706 struct bdi_writeback *wb, *next;
708 spin_lock_irq(&cgwb_lock);
709 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
711 memcg_cgwb_list->next = NULL; /* prevent new wb's */
712 spin_unlock_irq(&cgwb_lock);
716 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
717 * @blkcg: blkcg being offlined
719 * Also prevents creation of any new wb's associated with @blkcg.
721 void wb_blkcg_offline(struct blkcg *blkcg)
723 LIST_HEAD(to_destroy);
724 struct bdi_writeback *wb, *next;
726 spin_lock_irq(&cgwb_lock);
727 list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
729 blkcg->cgwb_list.next = NULL; /* prevent new wb's */
730 spin_unlock_irq(&cgwb_lock);
733 #else /* CONFIG_CGROUP_WRITEBACK */
735 static void cgwb_bdi_init(struct backing_dev_info *bdi) { }
736 static void cgwb_bdi_destroy(struct backing_dev_info *bdi) { }
738 #endif /* CONFIG_CGROUP_WRITEBACK */
740 int bdi_init(struct backing_dev_info *bdi)
747 bdi->max_ratio = 100;
748 bdi->max_prop_frac = FPROP_FRAC_BASE;
749 INIT_LIST_HEAD(&bdi->bdi_list);
750 init_waitqueue_head(&bdi->wb_waitq);
752 err = wb_init(&bdi->wb, bdi, GFP_KERNEL);
756 bdi->wb_congested.state = 0;
757 bdi->wb.congested = &bdi->wb_congested;
762 EXPORT_SYMBOL(bdi_init);
764 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
765 const char *fmt, ...)
770 if (bdi->dev) /* The driver needs to use separate queues per device */
774 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
781 bdi_debug_register(bdi, dev_name(dev));
782 set_bit(WB_registered, &bdi->wb.state);
784 spin_lock_bh(&bdi_lock);
785 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
786 spin_unlock_bh(&bdi_lock);
788 trace_writeback_bdi_register(bdi);
791 EXPORT_SYMBOL(bdi_register);
793 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
795 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
797 EXPORT_SYMBOL(bdi_register_dev);
800 * Remove bdi from bdi_list, and ensure that it is no longer visible
802 static void bdi_remove_from_list(struct backing_dev_info *bdi)
804 spin_lock_bh(&bdi_lock);
805 list_del_rcu(&bdi->bdi_list);
806 spin_unlock_bh(&bdi_lock);
808 synchronize_rcu_expedited();
811 void bdi_destroy(struct backing_dev_info *bdi)
813 /* make sure nobody finds us on the bdi_list anymore */
814 bdi_remove_from_list(bdi);
815 wb_shutdown(&bdi->wb);
816 cgwb_bdi_destroy(bdi);
819 bdi_debug_unregister(bdi);
820 device_unregister(bdi->dev);
826 EXPORT_SYMBOL(bdi_destroy);
829 * For use from filesystems to quickly init and register a bdi associated
830 * with dirty writeback
832 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
837 bdi->capabilities = 0;
842 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
843 atomic_long_inc_return(&bdi_seq));
851 EXPORT_SYMBOL(bdi_setup_and_register);
853 static wait_queue_head_t congestion_wqh[2] = {
854 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
855 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
857 static atomic_t nr_wb_congested[2];
859 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync)
861 wait_queue_head_t *wqh = &congestion_wqh[sync];
864 bit = sync ? WB_sync_congested : WB_async_congested;
865 if (test_and_clear_bit(bit, &congested->state))
866 atomic_dec(&nr_wb_congested[sync]);
867 smp_mb__after_atomic();
868 if (waitqueue_active(wqh))
871 EXPORT_SYMBOL(clear_wb_congested);
873 void set_wb_congested(struct bdi_writeback_congested *congested, int sync)
877 bit = sync ? WB_sync_congested : WB_async_congested;
878 if (!test_and_set_bit(bit, &congested->state))
879 atomic_inc(&nr_wb_congested[sync]);
881 EXPORT_SYMBOL(set_wb_congested);
884 * congestion_wait - wait for a backing_dev to become uncongested
885 * @sync: SYNC or ASYNC IO
886 * @timeout: timeout in jiffies
888 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
889 * write congestion. If no backing_devs are congested then just wait for the
890 * next write to be completed.
892 long congestion_wait(int sync, long timeout)
895 unsigned long start = jiffies;
897 wait_queue_head_t *wqh = &congestion_wqh[sync];
899 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
900 ret = io_schedule_timeout(timeout);
901 finish_wait(wqh, &wait);
903 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
904 jiffies_to_usecs(jiffies - start));
908 EXPORT_SYMBOL(congestion_wait);
911 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
912 * @zone: A zone to check if it is heavily congested
913 * @sync: SYNC or ASYNC IO
914 * @timeout: timeout in jiffies
916 * In the event of a congested backing_dev (any backing_dev) and the given
917 * @zone has experienced recent congestion, this waits for up to @timeout
918 * jiffies for either a BDI to exit congestion of the given @sync queue
919 * or a write to complete.
921 * In the absence of zone congestion, cond_resched() is called to yield
922 * the processor if necessary but otherwise does not sleep.
924 * The return value is 0 if the sleep is for the full timeout. Otherwise,
925 * it is the number of jiffies that were still remaining when the function
926 * returned. return_value == timeout implies the function did not sleep.
928 long wait_iff_congested(struct zone *zone, int sync, long timeout)
931 unsigned long start = jiffies;
933 wait_queue_head_t *wqh = &congestion_wqh[sync];
936 * If there is no congestion, or heavy congestion is not being
937 * encountered in the current zone, yield if necessary instead
938 * of sleeping on the congestion queue
940 if (atomic_read(&nr_wb_congested[sync]) == 0 ||
941 !test_bit(ZONE_CONGESTED, &zone->flags)) {
944 /* In case we scheduled, work out time remaining */
945 ret = timeout - (jiffies - start);
952 /* Sleep until uncongested or a write happens */
953 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
954 ret = io_schedule_timeout(timeout);
955 finish_wait(wqh, &wait);
958 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
959 jiffies_to_usecs(jiffies - start));
963 EXPORT_SYMBOL(wait_iff_congested);
965 int pdflush_proc_obsolete(struct ctl_table *table, int write,
966 void __user *buffer, size_t *lenp, loff_t *ppos)
970 if (*ppos || *lenp < sizeof(kbuf)) {
975 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
977 printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",