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>
14 void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
17 EXPORT_SYMBOL(default_unplug_io_fn);
19 struct backing_dev_info default_backing_dev_info = {
21 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
23 .capabilities = BDI_CAP_MAP_COPY,
24 .unplug_io_fn = default_unplug_io_fn,
26 EXPORT_SYMBOL_GPL(default_backing_dev_info);
28 static struct class *bdi_class;
31 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
32 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
35 DEFINE_SPINLOCK(bdi_lock);
37 LIST_HEAD(bdi_pending_list);
39 static struct task_struct *sync_supers_tsk;
40 static struct timer_list sync_supers_timer;
42 static int bdi_sync_supers(void *);
43 static void sync_supers_timer_fn(unsigned long);
44 static void arm_supers_timer(void);
46 static void bdi_add_default_flusher_task(struct backing_dev_info *bdi);
48 #ifdef CONFIG_DEBUG_FS
49 #include <linux/debugfs.h>
50 #include <linux/seq_file.h>
52 static struct dentry *bdi_debug_root;
54 static void bdi_debug_init(void)
56 bdi_debug_root = debugfs_create_dir("bdi", NULL);
59 static int bdi_debug_stats_show(struct seq_file *m, void *v)
61 struct backing_dev_info *bdi = m->private;
62 struct bdi_writeback *wb;
63 unsigned long background_thresh;
64 unsigned long dirty_thresh;
65 unsigned long bdi_thresh;
66 unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
70 * inode lock is enough here, the bdi->wb_list is protected by
71 * RCU on the reader side
73 nr_wb = nr_dirty = nr_io = nr_more_io = 0;
74 spin_lock(&inode_lock);
75 list_for_each_entry(wb, &bdi->wb_list, list) {
77 list_for_each_entry(inode, &wb->b_dirty, i_list)
79 list_for_each_entry(inode, &wb->b_io, i_list)
81 list_for_each_entry(inode, &wb->b_more_io, i_list)
84 spin_unlock(&inode_lock);
86 get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi);
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"
95 "WritebackThreads: %8lu\n"
104 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
105 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
106 K(bdi_thresh), K(dirty_thresh),
107 K(background_thresh), nr_wb, nr_dirty, nr_io, nr_more_io,
108 !list_empty(&bdi->bdi_list), bdi->state, bdi->wb_mask,
109 !list_empty(&bdi->wb_list), bdi->wb_cnt);
115 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
117 return single_open(file, bdi_debug_stats_show, inode->i_private);
120 static const struct file_operations bdi_debug_stats_fops = {
121 .open = bdi_debug_stats_open,
124 .release = single_release,
127 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
129 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
130 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
131 bdi, &bdi_debug_stats_fops);
134 static void bdi_debug_unregister(struct backing_dev_info *bdi)
136 debugfs_remove(bdi->debug_stats);
137 debugfs_remove(bdi->debug_dir);
140 static inline void bdi_debug_init(void)
143 static inline void bdi_debug_register(struct backing_dev_info *bdi,
147 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
152 static ssize_t read_ahead_kb_store(struct device *dev,
153 struct device_attribute *attr,
154 const char *buf, size_t count)
156 struct backing_dev_info *bdi = dev_get_drvdata(dev);
158 unsigned long read_ahead_kb;
159 ssize_t ret = -EINVAL;
161 read_ahead_kb = simple_strtoul(buf, &end, 10);
162 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
163 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
169 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
171 #define BDI_SHOW(name, expr) \
172 static ssize_t name##_show(struct device *dev, \
173 struct device_attribute *attr, char *page) \
175 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
177 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
180 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
182 static ssize_t min_ratio_store(struct device *dev,
183 struct device_attribute *attr, const char *buf, size_t count)
185 struct backing_dev_info *bdi = dev_get_drvdata(dev);
188 ssize_t ret = -EINVAL;
190 ratio = simple_strtoul(buf, &end, 10);
191 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
192 ret = bdi_set_min_ratio(bdi, ratio);
198 BDI_SHOW(min_ratio, bdi->min_ratio)
200 static ssize_t max_ratio_store(struct device *dev,
201 struct device_attribute *attr, const char *buf, size_t count)
203 struct backing_dev_info *bdi = dev_get_drvdata(dev);
206 ssize_t ret = -EINVAL;
208 ratio = simple_strtoul(buf, &end, 10);
209 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
210 ret = bdi_set_max_ratio(bdi, ratio);
216 BDI_SHOW(max_ratio, bdi->max_ratio)
218 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
220 static struct device_attribute bdi_dev_attrs[] = {
221 __ATTR_RW(read_ahead_kb),
222 __ATTR_RW(min_ratio),
223 __ATTR_RW(max_ratio),
227 static __init int bdi_class_init(void)
229 bdi_class = class_create(THIS_MODULE, "bdi");
230 if (IS_ERR(bdi_class))
231 return PTR_ERR(bdi_class);
233 bdi_class->dev_attrs = bdi_dev_attrs;
237 postcore_initcall(bdi_class_init);
239 static int __init default_bdi_init(void)
243 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
244 BUG_ON(IS_ERR(sync_supers_tsk));
246 init_timer(&sync_supers_timer);
247 setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
250 err = bdi_init(&default_backing_dev_info);
252 bdi_register(&default_backing_dev_info, NULL, "default");
256 subsys_initcall(default_bdi_init);
258 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
260 memset(wb, 0, sizeof(*wb));
263 wb->last_old_flush = jiffies;
264 INIT_LIST_HEAD(&wb->b_dirty);
265 INIT_LIST_HEAD(&wb->b_io);
266 INIT_LIST_HEAD(&wb->b_more_io);
269 static void bdi_task_init(struct backing_dev_info *bdi,
270 struct bdi_writeback *wb)
272 struct task_struct *tsk = current;
274 spin_lock(&bdi->wb_lock);
275 list_add_tail_rcu(&wb->list, &bdi->wb_list);
276 spin_unlock(&bdi->wb_lock);
278 tsk->flags |= PF_FLUSHER | PF_SWAPWRITE;
282 * Our parent may run at a different priority, just set us to normal
284 set_user_nice(tsk, 0);
287 static int bdi_start_fn(void *ptr)
289 struct bdi_writeback *wb = ptr;
290 struct backing_dev_info *bdi = wb->bdi;
294 * Add us to the active bdi_list
296 spin_lock_bh(&bdi_lock);
297 list_add_rcu(&bdi->bdi_list, &bdi_list);
298 spin_unlock_bh(&bdi_lock);
300 bdi_task_init(bdi, wb);
303 * Clear pending bit and wakeup anybody waiting to tear us down
305 clear_bit(BDI_pending, &bdi->state);
306 smp_mb__after_clear_bit();
307 wake_up_bit(&bdi->state, BDI_pending);
309 ret = bdi_writeback_task(wb);
312 * Remove us from the list
314 spin_lock(&bdi->wb_lock);
315 list_del_rcu(&wb->list);
316 spin_unlock(&bdi->wb_lock);
319 * Flush any work that raced with us exiting. No new work
320 * will be added, since this bdi isn't discoverable anymore.
322 if (!list_empty(&bdi->work_list))
323 wb_do_writeback(wb, 1);
329 int bdi_has_dirty_io(struct backing_dev_info *bdi)
331 return wb_has_dirty_io(&bdi->wb);
334 static void bdi_flush_io(struct backing_dev_info *bdi)
336 struct writeback_control wbc = {
338 .sync_mode = WB_SYNC_NONE,
339 .older_than_this = NULL,
344 writeback_inodes_wbc(&wbc);
348 * kupdated() used to do this. We cannot do it from the bdi_forker_task()
349 * or we risk deadlocking on ->s_umount. The longer term solution would be
350 * to implement sync_supers_bdi() or similar and simply do it from the
351 * bdi writeback tasks individually.
353 static int bdi_sync_supers(void *unused)
355 set_user_nice(current, 0);
357 while (!kthread_should_stop()) {
358 set_current_state(TASK_INTERRUPTIBLE);
362 * Do this periodically, like kupdated() did before.
370 static void arm_supers_timer(void)
374 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
375 mod_timer(&sync_supers_timer, round_jiffies_up(next));
378 static void sync_supers_timer_fn(unsigned long unused)
380 wake_up_process(sync_supers_tsk);
384 static int bdi_forker_task(void *ptr)
386 struct bdi_writeback *me = ptr;
388 bdi_task_init(me->bdi, me);
391 struct backing_dev_info *bdi, *tmp;
392 struct bdi_writeback *wb;
395 * Temporary measure, we want to make sure we don't see
396 * dirty data on the default backing_dev_info
398 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
399 wb_do_writeback(me, 0);
401 spin_lock_bh(&bdi_lock);
404 * Check if any existing bdi's have dirty data without
405 * a thread registered. If so, set that up.
407 list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) {
410 if (list_empty(&bdi->work_list) &&
411 !bdi_has_dirty_io(bdi))
414 bdi_add_default_flusher_task(bdi);
417 set_current_state(TASK_INTERRUPTIBLE);
419 if (list_empty(&bdi_pending_list)) {
422 spin_unlock_bh(&bdi_lock);
423 wait = msecs_to_jiffies(dirty_writeback_interval * 10);
424 schedule_timeout(wait);
429 __set_current_state(TASK_RUNNING);
432 * This is our real job - check for pending entries in
433 * bdi_pending_list, and create the tasks that got added
435 bdi = list_entry(bdi_pending_list.next, struct backing_dev_info,
437 list_del_init(&bdi->bdi_list);
438 spin_unlock_bh(&bdi_lock);
441 wb->task = kthread_run(bdi_start_fn, wb, "flush-%s",
444 * If task creation fails, then readd the bdi to
445 * the pending list and force writeout of the bdi
446 * from this forker thread. That will free some memory
447 * and we can try again.
449 if (IS_ERR(wb->task)) {
453 * Add this 'bdi' to the back, so we get
454 * a chance to flush other bdi's to free
457 spin_lock_bh(&bdi_lock);
458 list_add_tail(&bdi->bdi_list, &bdi_pending_list);
459 spin_unlock_bh(&bdi_lock);
468 static void bdi_add_to_pending(struct rcu_head *head)
470 struct backing_dev_info *bdi;
472 bdi = container_of(head, struct backing_dev_info, rcu_head);
473 INIT_LIST_HEAD(&bdi->bdi_list);
475 spin_lock(&bdi_lock);
476 list_add_tail(&bdi->bdi_list, &bdi_pending_list);
477 spin_unlock(&bdi_lock);
480 * We are now on the pending list, wake up bdi_forker_task()
481 * to finish the job and add us back to the active bdi_list
483 wake_up_process(default_backing_dev_info.wb.task);
487 * Add the default flusher task that gets created for any bdi
488 * that has dirty data pending writeout
490 void static bdi_add_default_flusher_task(struct backing_dev_info *bdi)
492 if (!bdi_cap_writeback_dirty(bdi))
495 if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) {
496 printk(KERN_ERR "bdi %p/%s is not registered!\n",
502 * Check with the helper whether to proceed adding a task. Will only
503 * abort if we two or more simultanous calls to
504 * bdi_add_default_flusher_task() occured, further additions will block
505 * waiting for previous additions to finish.
507 if (!test_and_set_bit(BDI_pending, &bdi->state)) {
508 list_del_rcu(&bdi->bdi_list);
511 * We must wait for the current RCU period to end before
512 * moving to the pending list. So schedule that operation
513 * from an RCU callback.
515 call_rcu(&bdi->rcu_head, bdi_add_to_pending);
520 * Remove bdi from bdi_list, and ensure that it is no longer visible
522 static void bdi_remove_from_list(struct backing_dev_info *bdi)
524 spin_lock_bh(&bdi_lock);
525 list_del_rcu(&bdi->bdi_list);
526 spin_unlock_bh(&bdi_lock);
531 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
532 const char *fmt, ...)
538 if (bdi->dev) /* The driver needs to use separate queues per device */
542 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
549 spin_lock_bh(&bdi_lock);
550 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
551 spin_unlock_bh(&bdi_lock);
556 * Just start the forker thread for our default backing_dev_info,
557 * and add other bdi's to the list. They will get a thread created
558 * on-demand when they need it.
560 if (bdi_cap_flush_forker(bdi)) {
561 struct bdi_writeback *wb = &bdi->wb;
563 wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s",
565 if (IS_ERR(wb->task)) {
569 bdi_remove_from_list(bdi);
574 bdi_debug_register(bdi, dev_name(dev));
575 set_bit(BDI_registered, &bdi->state);
579 EXPORT_SYMBOL(bdi_register);
581 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
583 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
585 EXPORT_SYMBOL(bdi_register_dev);
588 * Remove bdi from the global list and shutdown any threads we have running
590 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
592 struct bdi_writeback *wb;
594 if (!bdi_cap_writeback_dirty(bdi))
598 * If setup is pending, wait for that to complete first
600 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
601 TASK_UNINTERRUPTIBLE);
604 * Make sure nobody finds us on the bdi_list anymore
606 bdi_remove_from_list(bdi);
609 * Finally, kill the kernel threads. We don't need to be RCU
610 * safe anymore, since the bdi is gone from visibility. Force
611 * unfreeze of the thread before calling kthread_stop(), otherwise
612 * it would never exet if it is currently stuck in the refrigerator.
614 list_for_each_entry(wb, &bdi->wb_list, list) {
615 thaw_process(wb->task);
616 kthread_stop(wb->task);
621 * This bdi is going away now, make sure that no super_blocks point to it
623 static void bdi_prune_sb(struct backing_dev_info *bdi)
625 struct super_block *sb;
628 list_for_each_entry(sb, &super_blocks, s_list) {
629 if (sb->s_bdi == bdi)
632 spin_unlock(&sb_lock);
635 void bdi_unregister(struct backing_dev_info *bdi)
640 if (!bdi_cap_flush_forker(bdi))
641 bdi_wb_shutdown(bdi);
642 bdi_debug_unregister(bdi);
643 device_unregister(bdi->dev);
647 EXPORT_SYMBOL(bdi_unregister);
649 int bdi_init(struct backing_dev_info *bdi)
656 bdi->max_ratio = 100;
657 bdi->max_prop_frac = PROP_FRAC_BASE;
658 spin_lock_init(&bdi->wb_lock);
659 INIT_RCU_HEAD(&bdi->rcu_head);
660 INIT_LIST_HEAD(&bdi->bdi_list);
661 INIT_LIST_HEAD(&bdi->wb_list);
662 INIT_LIST_HEAD(&bdi->work_list);
664 bdi_wb_init(&bdi->wb, bdi);
667 * Just one thread support for now, hard code mask and count
672 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
673 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
678 bdi->dirty_exceeded = 0;
679 err = prop_local_init_percpu(&bdi->completions);
684 percpu_counter_destroy(&bdi->bdi_stat[i]);
689 EXPORT_SYMBOL(bdi_init);
691 void bdi_destroy(struct backing_dev_info *bdi)
696 * Splice our entries to the default_backing_dev_info, if this
699 if (bdi_has_dirty_io(bdi)) {
700 struct bdi_writeback *dst = &default_backing_dev_info.wb;
702 spin_lock(&inode_lock);
703 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
704 list_splice(&bdi->wb.b_io, &dst->b_io);
705 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
706 spin_unlock(&inode_lock);
711 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
712 percpu_counter_destroy(&bdi->bdi_stat[i]);
714 prop_local_destroy_percpu(&bdi->completions);
716 EXPORT_SYMBOL(bdi_destroy);
718 static wait_queue_head_t congestion_wqh[2] = {
719 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
720 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
723 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
726 wait_queue_head_t *wqh = &congestion_wqh[sync];
728 bit = sync ? BDI_sync_congested : BDI_async_congested;
729 clear_bit(bit, &bdi->state);
730 smp_mb__after_clear_bit();
731 if (waitqueue_active(wqh))
734 EXPORT_SYMBOL(clear_bdi_congested);
736 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
740 bit = sync ? BDI_sync_congested : BDI_async_congested;
741 set_bit(bit, &bdi->state);
743 EXPORT_SYMBOL(set_bdi_congested);
746 * congestion_wait - wait for a backing_dev to become uncongested
747 * @sync: SYNC or ASYNC IO
748 * @timeout: timeout in jiffies
750 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
751 * write congestion. If no backing_devs are congested then just wait for the
752 * next write to be completed.
754 long congestion_wait(int sync, long timeout)
758 wait_queue_head_t *wqh = &congestion_wqh[sync];
760 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
761 ret = io_schedule_timeout(timeout);
762 finish_wait(wqh, &wait);
765 EXPORT_SYMBOL(congestion_wait);