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 void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
20 EXPORT_SYMBOL(default_unplug_io_fn);
22 struct backing_dev_info default_backing_dev_info = {
24 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
26 .capabilities = BDI_CAP_MAP_COPY,
27 .unplug_io_fn = default_unplug_io_fn,
29 EXPORT_SYMBOL_GPL(default_backing_dev_info);
31 struct backing_dev_info noop_backing_dev_info = {
33 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
35 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
37 static struct class *bdi_class;
40 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
41 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
44 DEFINE_SPINLOCK(bdi_lock);
46 LIST_HEAD(bdi_pending_list);
48 static struct task_struct *sync_supers_tsk;
49 static struct timer_list sync_supers_timer;
51 static int bdi_sync_supers(void *);
52 static void sync_supers_timer_fn(unsigned long);
54 #ifdef CONFIG_DEBUG_FS
55 #include <linux/debugfs.h>
56 #include <linux/seq_file.h>
58 static struct dentry *bdi_debug_root;
60 static void bdi_debug_init(void)
62 bdi_debug_root = debugfs_create_dir("bdi", NULL);
65 static int bdi_debug_stats_show(struct seq_file *m, void *v)
67 struct backing_dev_info *bdi = m->private;
68 struct bdi_writeback *wb = &bdi->wb;
69 unsigned long background_thresh;
70 unsigned long dirty_thresh;
71 unsigned long bdi_thresh;
72 unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
75 nr_wb = nr_dirty = nr_io = nr_more_io = 0;
76 spin_lock(&inode_lock);
77 list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
79 list_for_each_entry(inode, &wb->b_io, i_wb_list)
81 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
83 spin_unlock(&inode_lock);
85 global_dirty_limits(&background_thresh, &dirty_thresh);
86 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
88 #define K(x) ((x) << (PAGE_SHIFT - 10))
90 "BdiWriteback: %8lu kB\n"
91 "BdiReclaimable: %8lu kB\n"
92 "BdiDirtyThresh: %8lu kB\n"
93 "DirtyThresh: %8lu kB\n"
94 "BackgroundThresh: %8lu kB\n"
100 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
101 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
102 K(bdi_thresh), K(dirty_thresh),
103 K(background_thresh), nr_dirty, nr_io, nr_more_io,
104 !list_empty(&bdi->bdi_list), bdi->state);
110 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
112 return single_open(file, bdi_debug_stats_show, inode->i_private);
115 static const struct file_operations bdi_debug_stats_fops = {
116 .open = bdi_debug_stats_open,
119 .release = single_release,
122 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
124 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
125 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
126 bdi, &bdi_debug_stats_fops);
129 static void bdi_debug_unregister(struct backing_dev_info *bdi)
131 debugfs_remove(bdi->debug_stats);
132 debugfs_remove(bdi->debug_dir);
135 static inline void bdi_debug_init(void)
138 static inline void bdi_debug_register(struct backing_dev_info *bdi,
142 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
147 static ssize_t read_ahead_kb_store(struct device *dev,
148 struct device_attribute *attr,
149 const char *buf, size_t count)
151 struct backing_dev_info *bdi = dev_get_drvdata(dev);
153 unsigned long read_ahead_kb;
154 ssize_t ret = -EINVAL;
156 read_ahead_kb = simple_strtoul(buf, &end, 10);
157 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
158 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
164 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
166 #define BDI_SHOW(name, expr) \
167 static ssize_t name##_show(struct device *dev, \
168 struct device_attribute *attr, char *page) \
170 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
172 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
175 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
177 static ssize_t min_ratio_store(struct device *dev,
178 struct device_attribute *attr, const char *buf, size_t count)
180 struct backing_dev_info *bdi = dev_get_drvdata(dev);
183 ssize_t ret = -EINVAL;
185 ratio = simple_strtoul(buf, &end, 10);
186 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
187 ret = bdi_set_min_ratio(bdi, ratio);
193 BDI_SHOW(min_ratio, bdi->min_ratio)
195 static ssize_t max_ratio_store(struct device *dev,
196 struct device_attribute *attr, const char *buf, size_t count)
198 struct backing_dev_info *bdi = dev_get_drvdata(dev);
201 ssize_t ret = -EINVAL;
203 ratio = simple_strtoul(buf, &end, 10);
204 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
205 ret = bdi_set_max_ratio(bdi, ratio);
211 BDI_SHOW(max_ratio, bdi->max_ratio)
213 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
215 static struct device_attribute bdi_dev_attrs[] = {
216 __ATTR_RW(read_ahead_kb),
217 __ATTR_RW(min_ratio),
218 __ATTR_RW(max_ratio),
222 static __init int bdi_class_init(void)
224 bdi_class = class_create(THIS_MODULE, "bdi");
225 if (IS_ERR(bdi_class))
226 return PTR_ERR(bdi_class);
228 bdi_class->dev_attrs = bdi_dev_attrs;
232 postcore_initcall(bdi_class_init);
234 static int __init default_bdi_init(void)
238 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
239 BUG_ON(IS_ERR(sync_supers_tsk));
241 setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
242 bdi_arm_supers_timer();
244 err = bdi_init(&default_backing_dev_info);
246 bdi_register(&default_backing_dev_info, NULL, "default");
247 err = bdi_init(&noop_backing_dev_info);
251 subsys_initcall(default_bdi_init);
253 int bdi_has_dirty_io(struct backing_dev_info *bdi)
255 return wb_has_dirty_io(&bdi->wb);
258 static void bdi_flush_io(struct backing_dev_info *bdi)
260 struct writeback_control wbc = {
261 .sync_mode = WB_SYNC_NONE,
262 .older_than_this = NULL,
267 writeback_inodes_wb(&bdi->wb, &wbc);
271 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
272 * or we risk deadlocking on ->s_umount. The longer term solution would be
273 * to implement sync_supers_bdi() or similar and simply do it from the
274 * bdi writeback thread individually.
276 static int bdi_sync_supers(void *unused)
278 set_user_nice(current, 0);
280 while (!kthread_should_stop()) {
281 set_current_state(TASK_INTERRUPTIBLE);
285 * Do this periodically, like kupdated() did before.
293 void bdi_arm_supers_timer(void)
297 if (!dirty_writeback_interval)
300 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
301 mod_timer(&sync_supers_timer, round_jiffies_up(next));
304 static void sync_supers_timer_fn(unsigned long unused)
306 wake_up_process(sync_supers_tsk);
307 bdi_arm_supers_timer();
310 static void wakeup_timer_fn(unsigned long data)
312 struct backing_dev_info *bdi = (struct backing_dev_info *)data;
314 spin_lock_bh(&bdi->wb_lock);
316 trace_writeback_wake_thread(bdi);
317 wake_up_process(bdi->wb.task);
320 * When bdi tasks are inactive for long time, they are killed.
321 * In this case we have to wake-up the forker thread which
322 * should create and run the bdi thread.
324 trace_writeback_wake_forker_thread(bdi);
325 wake_up_process(default_backing_dev_info.wb.task);
327 spin_unlock_bh(&bdi->wb_lock);
331 * This function is used when the first inode for this bdi is marked dirty. It
332 * wakes-up the corresponding bdi thread which should then take care of the
333 * periodic background write-out of dirty inodes. Since the write-out would
334 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
335 * set up a timer which wakes the bdi thread up later.
337 * Note, we wouldn't bother setting up the timer, but this function is on the
338 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
339 * by delaying the wake-up.
341 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
343 unsigned long timeout;
345 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
346 mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout);
350 * Calculate the longest interval (jiffies) bdi threads are allowed to be
353 static unsigned long bdi_longest_inactive(void)
355 unsigned long interval;
357 interval = msecs_to_jiffies(dirty_writeback_interval * 10);
358 return max(5UL * 60 * HZ, interval);
361 static int bdi_forker_thread(void *ptr)
363 struct bdi_writeback *me = ptr;
365 current->flags |= PF_SWAPWRITE;
369 * Our parent may run at a different priority, just set us to normal
371 set_user_nice(current, 0);
374 struct task_struct *task = NULL;
375 struct backing_dev_info *bdi;
377 NO_ACTION, /* Nothing to do */
378 FORK_THREAD, /* Fork bdi thread */
379 KILL_THREAD, /* Kill inactive bdi thread */
380 } action = NO_ACTION;
383 * Temporary measure, we want to make sure we don't see
384 * dirty data on the default backing_dev_info
386 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) {
387 del_timer(&me->wakeup_timer);
388 wb_do_writeback(me, 0);
391 spin_lock_bh(&bdi_lock);
392 set_current_state(TASK_INTERRUPTIBLE);
394 list_for_each_entry(bdi, &bdi_list, bdi_list) {
397 if (!bdi_cap_writeback_dirty(bdi) ||
398 bdi_cap_flush_forker(bdi))
401 WARN(!test_bit(BDI_registered, &bdi->state),
402 "bdi %p/%s is not registered!\n", bdi, bdi->name);
404 have_dirty_io = !list_empty(&bdi->work_list) ||
405 wb_has_dirty_io(&bdi->wb);
408 * If the bdi has work to do, but the thread does not
411 if (!bdi->wb.task && have_dirty_io) {
413 * Set the pending bit - if someone will try to
414 * unregister this bdi - it'll wait on this bit.
416 set_bit(BDI_pending, &bdi->state);
417 action = FORK_THREAD;
421 spin_lock(&bdi->wb_lock);
424 * If there is no work to do and the bdi thread was
425 * inactive long enough - kill it. The wb_lock is taken
426 * to make sure no-one adds more work to this bdi and
427 * wakes the bdi thread up.
429 if (bdi->wb.task && !have_dirty_io &&
430 time_after(jiffies, bdi->wb.last_active +
431 bdi_longest_inactive())) {
434 spin_unlock(&bdi->wb_lock);
435 set_bit(BDI_pending, &bdi->state);
436 action = KILL_THREAD;
439 spin_unlock(&bdi->wb_lock);
441 spin_unlock_bh(&bdi_lock);
443 /* Keep working if default bdi still has things to do */
444 if (!list_empty(&me->bdi->work_list))
445 __set_current_state(TASK_RUNNING);
449 __set_current_state(TASK_RUNNING);
450 task = kthread_create(bdi_writeback_thread, &bdi->wb,
451 "flush-%s", dev_name(bdi->dev));
454 * If thread creation fails, force writeout of
455 * the bdi from the thread.
460 * The spinlock makes sure we do not lose
461 * wake-ups when racing with 'bdi_queue_work()'.
462 * And as soon as the bdi thread is visible, we
465 spin_lock_bh(&bdi->wb_lock);
467 spin_unlock_bh(&bdi->wb_lock);
468 wake_up_process(task);
473 __set_current_state(TASK_RUNNING);
478 if (!wb_has_dirty_io(me) || !dirty_writeback_interval)
480 * There are no dirty data. The only thing we
481 * should now care about is checking for
482 * inactive bdi threads and killing them. Thus,
483 * let's sleep for longer time, save energy and
484 * be friendly for battery-driven devices.
486 schedule_timeout(bdi_longest_inactive());
488 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
490 /* Back to the main loop */
495 * Clear pending bit and wakeup anybody waiting to tear us down.
497 clear_bit(BDI_pending, &bdi->state);
498 smp_mb__after_clear_bit();
499 wake_up_bit(&bdi->state, BDI_pending);
506 * Remove bdi from bdi_list, and ensure that it is no longer visible
508 static void bdi_remove_from_list(struct backing_dev_info *bdi)
510 spin_lock_bh(&bdi_lock);
511 list_del_rcu(&bdi->bdi_list);
512 spin_unlock_bh(&bdi_lock);
517 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
518 const char *fmt, ...)
523 if (bdi->dev) /* The driver needs to use separate queues per device */
527 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
535 * Just start the forker thread for our default backing_dev_info,
536 * and add other bdi's to the list. They will get a thread created
537 * on-demand when they need it.
539 if (bdi_cap_flush_forker(bdi)) {
540 struct bdi_writeback *wb = &bdi->wb;
542 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
544 if (IS_ERR(wb->task))
545 return PTR_ERR(wb->task);
548 bdi_debug_register(bdi, dev_name(dev));
549 set_bit(BDI_registered, &bdi->state);
551 spin_lock_bh(&bdi_lock);
552 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
553 spin_unlock_bh(&bdi_lock);
555 trace_writeback_bdi_register(bdi);
558 EXPORT_SYMBOL(bdi_register);
560 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
562 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
564 EXPORT_SYMBOL(bdi_register_dev);
567 * Remove bdi from the global list and shutdown any threads we have running
569 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
571 if (!bdi_cap_writeback_dirty(bdi))
575 * Make sure nobody finds us on the bdi_list anymore
577 bdi_remove_from_list(bdi);
580 * If setup is pending, wait for that to complete first
582 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
583 TASK_UNINTERRUPTIBLE);
586 * Finally, kill the kernel thread. We don't need to be RCU
587 * safe anymore, since the bdi is gone from visibility. Force
588 * unfreeze of the thread before calling kthread_stop(), otherwise
589 * it would never exet if it is currently stuck in the refrigerator.
592 thaw_process(bdi->wb.task);
593 kthread_stop(bdi->wb.task);
598 * This bdi is going away now, make sure that no super_blocks point to it
600 static void bdi_prune_sb(struct backing_dev_info *bdi)
602 struct super_block *sb;
605 list_for_each_entry(sb, &super_blocks, s_list) {
606 if (sb->s_bdi == bdi)
609 spin_unlock(&sb_lock);
612 void bdi_unregister(struct backing_dev_info *bdi)
615 trace_writeback_bdi_unregister(bdi);
617 del_timer_sync(&bdi->wb.wakeup_timer);
619 if (!bdi_cap_flush_forker(bdi))
620 bdi_wb_shutdown(bdi);
621 bdi_debug_unregister(bdi);
622 device_unregister(bdi->dev);
626 EXPORT_SYMBOL(bdi_unregister);
628 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
630 memset(wb, 0, sizeof(*wb));
633 wb->last_old_flush = jiffies;
634 INIT_LIST_HEAD(&wb->b_dirty);
635 INIT_LIST_HEAD(&wb->b_io);
636 INIT_LIST_HEAD(&wb->b_more_io);
637 setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi);
640 int bdi_init(struct backing_dev_info *bdi)
647 bdi->max_ratio = 100;
648 bdi->max_prop_frac = PROP_FRAC_BASE;
649 spin_lock_init(&bdi->wb_lock);
650 INIT_LIST_HEAD(&bdi->bdi_list);
651 INIT_LIST_HEAD(&bdi->work_list);
653 bdi_wb_init(&bdi->wb, bdi);
655 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
656 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
661 bdi->dirty_exceeded = 0;
662 err = prop_local_init_percpu(&bdi->completions);
667 percpu_counter_destroy(&bdi->bdi_stat[i]);
672 EXPORT_SYMBOL(bdi_init);
674 void bdi_destroy(struct backing_dev_info *bdi)
679 * Splice our entries to the default_backing_dev_info, if this
682 if (bdi_has_dirty_io(bdi)) {
683 struct bdi_writeback *dst = &default_backing_dev_info.wb;
685 spin_lock(&inode_lock);
686 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
687 list_splice(&bdi->wb.b_io, &dst->b_io);
688 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
689 spin_unlock(&inode_lock);
694 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
695 percpu_counter_destroy(&bdi->bdi_stat[i]);
697 prop_local_destroy_percpu(&bdi->completions);
699 EXPORT_SYMBOL(bdi_destroy);
702 * For use from filesystems to quickly init and register a bdi associated
703 * with dirty writeback
705 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
712 bdi->capabilities = cap;
717 sprintf(tmp, "%.28s%s", name, "-%d");
718 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
726 EXPORT_SYMBOL(bdi_setup_and_register);
728 static wait_queue_head_t congestion_wqh[2] = {
729 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
730 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
732 static atomic_t nr_bdi_congested[2];
734 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
737 wait_queue_head_t *wqh = &congestion_wqh[sync];
739 bit = sync ? BDI_sync_congested : BDI_async_congested;
740 if (test_and_clear_bit(bit, &bdi->state))
741 atomic_dec(&nr_bdi_congested[sync]);
742 smp_mb__after_clear_bit();
743 if (waitqueue_active(wqh))
746 EXPORT_SYMBOL(clear_bdi_congested);
748 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
752 bit = sync ? BDI_sync_congested : BDI_async_congested;
753 if (!test_and_set_bit(bit, &bdi->state))
754 atomic_inc(&nr_bdi_congested[sync]);
756 EXPORT_SYMBOL(set_bdi_congested);
759 * congestion_wait - wait for a backing_dev to become uncongested
760 * @sync: SYNC or ASYNC IO
761 * @timeout: timeout in jiffies
763 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
764 * write congestion. If no backing_devs are congested then just wait for the
765 * next write to be completed.
767 long congestion_wait(int sync, long timeout)
770 unsigned long start = jiffies;
772 wait_queue_head_t *wqh = &congestion_wqh[sync];
774 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
775 ret = io_schedule_timeout(timeout);
776 finish_wait(wqh, &wait);
778 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
779 jiffies_to_usecs(jiffies - start));
783 EXPORT_SYMBOL(congestion_wait);
786 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
787 * @zone: A zone to check if it is heavily congested
788 * @sync: SYNC or ASYNC IO
789 * @timeout: timeout in jiffies
791 * In the event of a congested backing_dev (any backing_dev) and the given
792 * @zone has experienced recent congestion, this waits for up to @timeout
793 * jiffies for either a BDI to exit congestion of the given @sync queue
794 * or a write to complete.
796 * In the absense of zone congestion, cond_resched() is called to yield
797 * the processor if necessary but otherwise does not sleep.
799 * The return value is 0 if the sleep is for the full timeout. Otherwise,
800 * it is the number of jiffies that were still remaining when the function
801 * returned. return_value == timeout implies the function did not sleep.
803 long wait_iff_congested(struct zone *zone, int sync, long timeout)
806 unsigned long start = jiffies;
808 wait_queue_head_t *wqh = &congestion_wqh[sync];
811 * If there is no congestion, or heavy congestion is not being
812 * encountered in the current zone, yield if necessary instead
813 * of sleeping on the congestion queue
815 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
816 !zone_is_reclaim_congested(zone)) {
819 /* In case we scheduled, work out time remaining */
820 ret = timeout - (jiffies - start);
827 /* Sleep until uncongested or a write happens */
828 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
829 ret = io_schedule_timeout(timeout);
830 finish_wait(wqh, &wait);
833 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
834 jiffies_to_usecs(jiffies - start));
838 EXPORT_SYMBOL(wait_iff_congested);