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 = {
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 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,
284 int blkcg_id, gfp_t gfp)
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 wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp);
312 if (!wb->congested) {
317 err = fprop_local_init_percpu(&wb->completions, gfp);
321 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
322 err = percpu_counter_init(&wb->stat[i], 0, gfp);
324 goto out_destroy_stat;
331 percpu_counter_destroy(&wb->stat[i]);
332 fprop_local_destroy_percpu(&wb->completions);
334 wb_congested_put(wb->congested);
341 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
344 * Remove bdi from the global list and shutdown any threads we have running
346 static void wb_shutdown(struct bdi_writeback *wb)
348 /* Make sure nobody queues further work */
349 spin_lock_bh(&wb->work_lock);
350 if (!test_and_clear_bit(WB_registered, &wb->state)) {
351 spin_unlock_bh(&wb->work_lock);
354 spin_unlock_bh(&wb->work_lock);
356 cgwb_remove_from_bdi_list(wb);
358 * Drain work list and shutdown the delayed_work. !WB_registered
359 * tells wb_workfn() that @wb is dying and its work_list needs to
360 * be drained no matter what.
362 mod_delayed_work(bdi_wq, &wb->dwork, 0);
363 flush_delayed_work(&wb->dwork);
364 WARN_ON(!list_empty(&wb->work_list));
367 static void wb_exit(struct bdi_writeback *wb)
371 WARN_ON(delayed_work_pending(&wb->dwork));
373 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
374 percpu_counter_destroy(&wb->stat[i]);
376 fprop_local_destroy_percpu(&wb->completions);
377 wb_congested_put(wb->congested);
378 if (wb != &wb->bdi->wb)
382 #ifdef CONFIG_CGROUP_WRITEBACK
384 #include <linux/memcontrol.h>
387 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
388 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU
391 static DEFINE_SPINLOCK(cgwb_lock);
392 static struct workqueue_struct *cgwb_release_wq;
395 * wb_congested_get_create - get or create a wb_congested
396 * @bdi: associated bdi
397 * @blkcg_id: ID of the associated blkcg
398 * @gfp: allocation mask
400 * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one.
401 * The returned wb_congested has its reference count incremented. Returns
404 struct bdi_writeback_congested *
405 wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
407 struct bdi_writeback_congested *new_congested = NULL, *congested;
408 struct rb_node **node, *parent;
411 spin_lock_irqsave(&cgwb_lock, flags);
413 node = &bdi->cgwb_congested_tree.rb_node;
416 while (*node != NULL) {
418 congested = rb_entry(parent, struct bdi_writeback_congested,
420 if (congested->blkcg_id < blkcg_id)
421 node = &parent->rb_left;
422 else if (congested->blkcg_id > blkcg_id)
423 node = &parent->rb_right;
429 /* !found and storage for new one already allocated, insert */
430 congested = new_congested;
431 rb_link_node(&congested->rb_node, parent, node);
432 rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree);
433 spin_unlock_irqrestore(&cgwb_lock, flags);
437 spin_unlock_irqrestore(&cgwb_lock, flags);
439 /* allocate storage for new one and retry */
440 new_congested = kzalloc(sizeof(*new_congested), gfp);
444 refcount_set(&new_congested->refcnt, 1);
445 new_congested->__bdi = bdi;
446 new_congested->blkcg_id = blkcg_id;
450 refcount_inc(&congested->refcnt);
451 spin_unlock_irqrestore(&cgwb_lock, flags);
452 kfree(new_congested);
457 * wb_congested_put - put a wb_congested
458 * @congested: wb_congested to put
460 * Put @congested and destroy it if the refcnt reaches zero.
462 void wb_congested_put(struct bdi_writeback_congested *congested)
466 if (!refcount_dec_and_lock_irqsave(&congested->refcnt, &cgwb_lock, &flags))
469 /* bdi might already have been destroyed leaving @congested unlinked */
470 if (congested->__bdi) {
471 rb_erase(&congested->rb_node,
472 &congested->__bdi->cgwb_congested_tree);
473 congested->__bdi = NULL;
476 spin_unlock_irqrestore(&cgwb_lock, flags);
480 static void cgwb_release_workfn(struct work_struct *work)
482 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
484 struct blkcg *blkcg = css_to_blkcg(wb->blkcg_css);
486 mutex_lock(&wb->bdi->cgwb_release_mutex);
489 css_put(wb->memcg_css);
490 css_put(wb->blkcg_css);
491 mutex_unlock(&wb->bdi->cgwb_release_mutex);
493 /* triggers blkg destruction if cgwb_refcnt becomes zero */
494 blkcg_cgwb_put(blkcg);
496 fprop_local_destroy_percpu(&wb->memcg_completions);
497 percpu_ref_exit(&wb->refcnt);
502 static void cgwb_release(struct percpu_ref *refcnt)
504 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
506 queue_work(cgwb_release_wq, &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 void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
521 spin_lock_irq(&cgwb_lock);
522 list_del_rcu(&wb->bdi_node);
523 spin_unlock_irq(&cgwb_lock);
526 static int cgwb_create(struct backing_dev_info *bdi,
527 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
529 struct mem_cgroup *memcg;
530 struct cgroup_subsys_state *blkcg_css;
532 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
533 struct bdi_writeback *wb;
537 memcg = mem_cgroup_from_css(memcg_css);
538 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
539 blkcg = css_to_blkcg(blkcg_css);
540 memcg_cgwb_list = &memcg->cgwb_list;
541 blkcg_cgwb_list = &blkcg->cgwb_list;
543 /* look up again under lock and discard on blkcg mismatch */
544 spin_lock_irqsave(&cgwb_lock, flags);
545 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
546 if (wb && wb->blkcg_css != blkcg_css) {
550 spin_unlock_irqrestore(&cgwb_lock, flags);
554 /* need to create a new one */
555 wb = kmalloc(sizeof(*wb), gfp);
561 ret = wb_init(wb, bdi, blkcg_css->id, gfp);
565 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
569 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
573 wb->memcg_css = memcg_css;
574 wb->blkcg_css = blkcg_css;
575 INIT_WORK(&wb->release_work, cgwb_release_workfn);
576 set_bit(WB_registered, &wb->state);
579 * The root wb determines the registered state of the whole bdi and
580 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
581 * whether they're still online. Don't link @wb if any is dead.
582 * See wb_memcg_offline() and wb_blkcg_offline().
585 spin_lock_irqsave(&cgwb_lock, flags);
586 if (test_bit(WB_registered, &bdi->wb.state) &&
587 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
588 /* we might have raced another instance of this function */
589 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
591 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
592 list_add(&wb->memcg_node, memcg_cgwb_list);
593 list_add(&wb->blkcg_node, blkcg_cgwb_list);
594 blkcg_cgwb_get(blkcg);
599 spin_unlock_irqrestore(&cgwb_lock, flags);
608 fprop_local_destroy_percpu(&wb->memcg_completions);
610 percpu_ref_exit(&wb->refcnt);
621 * wb_get_lookup - get wb for a given memcg
623 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
625 * Try to get the wb for @memcg_css on @bdi. The returned wb has its
626 * refcount incremented.
628 * This function uses css_get() on @memcg_css and thus expects its refcnt
629 * to be positive on invocation. IOW, rcu_read_lock() protection on
630 * @memcg_css isn't enough. try_get it before calling this function.
632 * A wb is keyed by its associated memcg. As blkcg implicitly enables
633 * memcg on the default hierarchy, memcg association is guaranteed to be
634 * more specific (equal or descendant to the associated blkcg) and thus can
635 * identify both the memcg and blkcg associations.
637 * Because the blkcg associated with a memcg may change as blkcg is enabled
638 * and disabled closer to root in the hierarchy, each wb keeps track of
639 * both the memcg and blkcg associated with it and verifies the blkcg on
640 * each lookup. On mismatch, the existing wb is discarded and a new one is
643 struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
644 struct cgroup_subsys_state *memcg_css)
646 struct bdi_writeback *wb;
648 if (!memcg_css->parent)
652 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
654 struct cgroup_subsys_state *blkcg_css;
656 /* see whether the blkcg association has changed */
657 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
658 if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
668 * wb_get_create - get wb for a given memcg, create if necessary
670 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
671 * @gfp: allocation mask to use
673 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
674 * create one. See wb_get_lookup() for more details.
676 struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
677 struct cgroup_subsys_state *memcg_css,
680 struct bdi_writeback *wb;
682 might_sleep_if(gfpflags_allow_blocking(gfp));
684 if (!memcg_css->parent)
688 wb = wb_get_lookup(bdi, memcg_css);
689 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
694 static int cgwb_bdi_init(struct backing_dev_info *bdi)
698 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
699 bdi->cgwb_congested_tree = RB_ROOT;
700 mutex_init(&bdi->cgwb_release_mutex);
701 init_rwsem(&bdi->wb_switch_rwsem);
703 ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
705 bdi->wb.memcg_css = &root_mem_cgroup->css;
706 bdi->wb.blkcg_css = blkcg_root_css;
711 static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
713 struct radix_tree_iter iter;
715 struct bdi_writeback *wb;
717 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
719 spin_lock_irq(&cgwb_lock);
720 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
722 spin_unlock_irq(&cgwb_lock);
724 mutex_lock(&bdi->cgwb_release_mutex);
725 spin_lock_irq(&cgwb_lock);
726 while (!list_empty(&bdi->wb_list)) {
727 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
729 spin_unlock_irq(&cgwb_lock);
731 spin_lock_irq(&cgwb_lock);
733 spin_unlock_irq(&cgwb_lock);
734 mutex_unlock(&bdi->cgwb_release_mutex);
738 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
739 * @memcg: memcg being offlined
741 * Also prevents creation of any new wb's associated with @memcg.
743 void wb_memcg_offline(struct mem_cgroup *memcg)
745 struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
746 struct bdi_writeback *wb, *next;
748 spin_lock_irq(&cgwb_lock);
749 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
751 memcg_cgwb_list->next = NULL; /* prevent new wb's */
752 spin_unlock_irq(&cgwb_lock);
756 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
757 * @blkcg: blkcg being offlined
759 * Also prevents creation of any new wb's associated with @blkcg.
761 void wb_blkcg_offline(struct blkcg *blkcg)
763 struct bdi_writeback *wb, *next;
765 spin_lock_irq(&cgwb_lock);
766 list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
768 blkcg->cgwb_list.next = NULL; /* prevent new wb's */
769 spin_unlock_irq(&cgwb_lock);
772 static void cgwb_bdi_exit(struct backing_dev_info *bdi)
776 spin_lock_irq(&cgwb_lock);
777 while ((rbn = rb_first(&bdi->cgwb_congested_tree))) {
778 struct bdi_writeback_congested *congested =
779 rb_entry(rbn, struct bdi_writeback_congested, rb_node);
781 rb_erase(rbn, &bdi->cgwb_congested_tree);
782 congested->__bdi = NULL; /* mark @congested unlinked */
784 spin_unlock_irq(&cgwb_lock);
787 static void cgwb_bdi_register(struct backing_dev_info *bdi)
789 spin_lock_irq(&cgwb_lock);
790 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
791 spin_unlock_irq(&cgwb_lock);
794 static int __init cgwb_init(void)
797 * There can be many concurrent release work items overwhelming
798 * system_wq. Put them in a separate wq and limit concurrency.
799 * There's no point in executing many of these in parallel.
801 cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
802 if (!cgwb_release_wq)
807 subsys_initcall(cgwb_init);
809 #else /* CONFIG_CGROUP_WRITEBACK */
811 static int cgwb_bdi_init(struct backing_dev_info *bdi)
815 bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL);
816 if (!bdi->wb_congested)
819 refcount_set(&bdi->wb_congested->refcnt, 1);
821 err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
823 wb_congested_put(bdi->wb_congested);
829 static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
831 static void cgwb_bdi_exit(struct backing_dev_info *bdi)
833 wb_congested_put(bdi->wb_congested);
836 static void cgwb_bdi_register(struct backing_dev_info *bdi)
838 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
841 static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
843 list_del_rcu(&wb->bdi_node);
846 #endif /* CONFIG_CGROUP_WRITEBACK */
848 static int bdi_init(struct backing_dev_info *bdi)
854 kref_init(&bdi->refcnt);
856 bdi->max_ratio = 100;
857 bdi->max_prop_frac = FPROP_FRAC_BASE;
858 INIT_LIST_HEAD(&bdi->bdi_list);
859 INIT_LIST_HEAD(&bdi->wb_list);
860 init_waitqueue_head(&bdi->wb_waitq);
862 ret = cgwb_bdi_init(bdi);
867 struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id)
869 struct backing_dev_info *bdi;
871 bdi = kmalloc_node(sizeof(struct backing_dev_info),
872 gfp_mask | __GFP_ZERO, node_id);
882 EXPORT_SYMBOL(bdi_alloc_node);
884 static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp)
886 struct rb_node **p = &bdi_tree.rb_node;
887 struct rb_node *parent = NULL;
888 struct backing_dev_info *bdi;
890 lockdep_assert_held(&bdi_lock);
894 bdi = rb_entry(parent, struct backing_dev_info, rb_node);
898 else if (bdi->id < id)
910 * bdi_get_by_id - lookup and get bdi from its id
911 * @id: bdi id to lookup
913 * Find bdi matching @id and get it. Returns NULL if the matching bdi
914 * doesn't exist or is already unregistered.
916 struct backing_dev_info *bdi_get_by_id(u64 id)
918 struct backing_dev_info *bdi = NULL;
921 spin_lock_bh(&bdi_lock);
922 p = bdi_lookup_rb_node(id, NULL);
924 bdi = rb_entry(*p, struct backing_dev_info, rb_node);
927 spin_unlock_bh(&bdi_lock);
932 int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
935 struct rb_node *parent, **p;
937 if (bdi->dev) /* The driver needs to use separate queues per device */
940 dev = device_create_vargs(bdi_class, NULL, MKDEV(0, 0), bdi, fmt, args);
944 cgwb_bdi_register(bdi);
947 bdi_debug_register(bdi, dev_name(dev));
948 set_bit(WB_registered, &bdi->wb.state);
950 spin_lock_bh(&bdi_lock);
952 bdi->id = ++bdi_id_cursor;
954 p = bdi_lookup_rb_node(bdi->id, &parent);
955 rb_link_node(&bdi->rb_node, parent, p);
956 rb_insert_color(&bdi->rb_node, &bdi_tree);
958 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
960 spin_unlock_bh(&bdi_lock);
962 trace_writeback_bdi_register(bdi);
965 EXPORT_SYMBOL(bdi_register_va);
967 int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
973 ret = bdi_register_va(bdi, fmt, args);
977 EXPORT_SYMBOL(bdi_register);
979 int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner)
983 rc = bdi_register(bdi, "%u:%u", MAJOR(owner->devt), MINOR(owner->devt));
986 /* Leaking owner reference... */
992 EXPORT_SYMBOL(bdi_register_owner);
995 * Remove bdi from bdi_list, and ensure that it is no longer visible
997 static void bdi_remove_from_list(struct backing_dev_info *bdi)
999 spin_lock_bh(&bdi_lock);
1000 rb_erase(&bdi->rb_node, &bdi_tree);
1001 list_del_rcu(&bdi->bdi_list);
1002 spin_unlock_bh(&bdi_lock);
1004 synchronize_rcu_expedited();
1007 void bdi_unregister(struct backing_dev_info *bdi)
1009 /* make sure nobody finds us on the bdi_list anymore */
1010 bdi_remove_from_list(bdi);
1011 wb_shutdown(&bdi->wb);
1012 cgwb_bdi_unregister(bdi);
1015 bdi_debug_unregister(bdi);
1016 device_unregister(bdi->dev);
1021 put_device(bdi->owner);
1026 static void release_bdi(struct kref *ref)
1028 struct backing_dev_info *bdi =
1029 container_of(ref, struct backing_dev_info, refcnt);
1031 if (test_bit(WB_registered, &bdi->wb.state))
1032 bdi_unregister(bdi);
1033 WARN_ON_ONCE(bdi->dev);
1039 void bdi_put(struct backing_dev_info *bdi)
1041 kref_put(&bdi->refcnt, release_bdi);
1043 EXPORT_SYMBOL(bdi_put);
1045 static wait_queue_head_t congestion_wqh[2] = {
1046 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
1047 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
1049 static atomic_t nr_wb_congested[2];
1051 void clear_wb_congested(struct bdi_writeback_congested *congested, int sync)
1053 wait_queue_head_t *wqh = &congestion_wqh[sync];
1054 enum wb_congested_state bit;
1056 bit = sync ? WB_sync_congested : WB_async_congested;
1057 if (test_and_clear_bit(bit, &congested->state))
1058 atomic_dec(&nr_wb_congested[sync]);
1059 smp_mb__after_atomic();
1060 if (waitqueue_active(wqh))
1063 EXPORT_SYMBOL(clear_wb_congested);
1065 void set_wb_congested(struct bdi_writeback_congested *congested, int sync)
1067 enum wb_congested_state bit;
1069 bit = sync ? WB_sync_congested : WB_async_congested;
1070 if (!test_and_set_bit(bit, &congested->state))
1071 atomic_inc(&nr_wb_congested[sync]);
1073 EXPORT_SYMBOL(set_wb_congested);
1076 * congestion_wait - wait for a backing_dev to become uncongested
1077 * @sync: SYNC or ASYNC IO
1078 * @timeout: timeout in jiffies
1080 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
1081 * write congestion. If no backing_devs are congested then just wait for the
1082 * next write to be completed.
1084 long congestion_wait(int sync, long timeout)
1087 unsigned long start = jiffies;
1089 wait_queue_head_t *wqh = &congestion_wqh[sync];
1091 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1092 ret = io_schedule_timeout(timeout);
1093 finish_wait(wqh, &wait);
1095 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
1096 jiffies_to_usecs(jiffies - start));
1100 EXPORT_SYMBOL(congestion_wait);
1103 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
1104 * @sync: SYNC or ASYNC IO
1105 * @timeout: timeout in jiffies
1107 * In the event of a congested backing_dev (any backing_dev) this waits
1108 * for up to @timeout jiffies for either a BDI to exit congestion of the
1109 * given @sync queue or a write to complete.
1111 * The return value is 0 if the sleep is for the full timeout. Otherwise,
1112 * it is the number of jiffies that were still remaining when the function
1113 * returned. return_value == timeout implies the function did not sleep.
1115 long wait_iff_congested(int sync, long timeout)
1118 unsigned long start = jiffies;
1120 wait_queue_head_t *wqh = &congestion_wqh[sync];
1123 * If there is no congestion, yield if necessary instead
1124 * of sleeping on the congestion queue
1126 if (atomic_read(&nr_wb_congested[sync]) == 0) {
1129 /* In case we scheduled, work out time remaining */
1130 ret = timeout - (jiffies - start);
1137 /* Sleep until uncongested or a write happens */
1138 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1139 ret = io_schedule_timeout(timeout);
1140 finish_wait(wqh, &wait);
1143 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
1144 jiffies_to_usecs(jiffies - start));
1148 EXPORT_SYMBOL(wait_iff_congested);