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 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
16 void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
19 EXPORT_SYMBOL(default_unplug_io_fn);
21 struct backing_dev_info default_backing_dev_info = {
23 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
25 .capabilities = BDI_CAP_MAP_COPY,
26 .unplug_io_fn = default_unplug_io_fn,
28 EXPORT_SYMBOL_GPL(default_backing_dev_info);
30 struct backing_dev_info noop_backing_dev_info = {
33 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
35 static struct class *bdi_class;
38 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
39 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
42 DEFINE_SPINLOCK(bdi_lock);
44 LIST_HEAD(bdi_pending_list);
46 static struct task_struct *sync_supers_tsk;
47 static struct timer_list sync_supers_timer;
49 static int bdi_sync_supers(void *);
50 static void sync_supers_timer_fn(unsigned long);
52 static void bdi_add_default_flusher_task(struct backing_dev_info *bdi);
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_list)
79 list_for_each_entry(inode, &wb->b_io, i_list)
81 list_for_each_entry(inode, &wb->b_more_io, i_list)
83 spin_unlock(&inode_lock);
85 get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi);
87 #define K(x) ((x) << (PAGE_SHIFT - 10))
89 "BdiWriteback: %8lu kB\n"
90 "BdiReclaimable: %8lu kB\n"
91 "BdiDirtyThresh: %8lu kB\n"
92 "DirtyThresh: %8lu kB\n"
93 "BackgroundThresh: %8lu kB\n"
99 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
100 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
101 K(bdi_thresh), K(dirty_thresh),
102 K(background_thresh), nr_dirty, nr_io, nr_more_io,
103 !list_empty(&bdi->bdi_list), bdi->state);
109 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
111 return single_open(file, bdi_debug_stats_show, inode->i_private);
114 static const struct file_operations bdi_debug_stats_fops = {
115 .open = bdi_debug_stats_open,
118 .release = single_release,
121 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
123 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
124 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
125 bdi, &bdi_debug_stats_fops);
128 static void bdi_debug_unregister(struct backing_dev_info *bdi)
130 debugfs_remove(bdi->debug_stats);
131 debugfs_remove(bdi->debug_dir);
134 static inline void bdi_debug_init(void)
137 static inline void bdi_debug_register(struct backing_dev_info *bdi,
141 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
146 static ssize_t read_ahead_kb_store(struct device *dev,
147 struct device_attribute *attr,
148 const char *buf, size_t count)
150 struct backing_dev_info *bdi = dev_get_drvdata(dev);
152 unsigned long read_ahead_kb;
153 ssize_t ret = -EINVAL;
155 read_ahead_kb = simple_strtoul(buf, &end, 10);
156 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
157 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
163 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
165 #define BDI_SHOW(name, expr) \
166 static ssize_t name##_show(struct device *dev, \
167 struct device_attribute *attr, char *page) \
169 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
171 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
174 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
176 static ssize_t min_ratio_store(struct device *dev,
177 struct device_attribute *attr, const char *buf, size_t count)
179 struct backing_dev_info *bdi = dev_get_drvdata(dev);
182 ssize_t ret = -EINVAL;
184 ratio = simple_strtoul(buf, &end, 10);
185 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
186 ret = bdi_set_min_ratio(bdi, ratio);
192 BDI_SHOW(min_ratio, bdi->min_ratio)
194 static ssize_t max_ratio_store(struct device *dev,
195 struct device_attribute *attr, const char *buf, size_t count)
197 struct backing_dev_info *bdi = dev_get_drvdata(dev);
200 ssize_t ret = -EINVAL;
202 ratio = simple_strtoul(buf, &end, 10);
203 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
204 ret = bdi_set_max_ratio(bdi, ratio);
210 BDI_SHOW(max_ratio, bdi->max_ratio)
212 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
214 static struct device_attribute bdi_dev_attrs[] = {
215 __ATTR_RW(read_ahead_kb),
216 __ATTR_RW(min_ratio),
217 __ATTR_RW(max_ratio),
221 static __init int bdi_class_init(void)
223 bdi_class = class_create(THIS_MODULE, "bdi");
224 if (IS_ERR(bdi_class))
225 return PTR_ERR(bdi_class);
227 bdi_class->dev_attrs = bdi_dev_attrs;
231 postcore_initcall(bdi_class_init);
233 static int __init default_bdi_init(void)
237 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
238 BUG_ON(IS_ERR(sync_supers_tsk));
240 init_timer(&sync_supers_timer);
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");
250 subsys_initcall(default_bdi_init);
252 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
254 memset(wb, 0, sizeof(*wb));
257 wb->last_old_flush = jiffies;
258 INIT_LIST_HEAD(&wb->b_dirty);
259 INIT_LIST_HEAD(&wb->b_io);
260 INIT_LIST_HEAD(&wb->b_more_io);
263 static int bdi_start_fn(void *ptr)
265 struct bdi_writeback *wb = ptr;
266 struct backing_dev_info *bdi = wb->bdi;
270 * Add us to the active bdi_list
272 spin_lock_bh(&bdi_lock);
273 list_add_rcu(&bdi->bdi_list, &bdi_list);
274 spin_unlock_bh(&bdi_lock);
276 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
280 * Our parent may run at a different priority, just set us to normal
282 set_user_nice(current, 0);
285 * Clear pending bit and wakeup anybody waiting to tear us down
287 clear_bit(BDI_pending, &bdi->state);
288 smp_mb__after_clear_bit();
289 wake_up_bit(&bdi->state, BDI_pending);
291 ret = bdi_writeback_task(wb);
296 * Flush any work that raced with us exiting. No new work
297 * will be added, since this bdi isn't discoverable anymore.
299 if (!list_empty(&bdi->work_list))
300 wb_do_writeback(wb, 1);
305 int bdi_has_dirty_io(struct backing_dev_info *bdi)
307 return wb_has_dirty_io(&bdi->wb);
310 static void bdi_flush_io(struct backing_dev_info *bdi)
312 struct writeback_control wbc = {
313 .sync_mode = WB_SYNC_NONE,
314 .older_than_this = NULL,
319 writeback_inodes_wb(&bdi->wb, &wbc);
323 * kupdated() used to do this. We cannot do it from the bdi_forker_task()
324 * or we risk deadlocking on ->s_umount. The longer term solution would be
325 * to implement sync_supers_bdi() or similar and simply do it from the
326 * bdi writeback tasks individually.
328 static int bdi_sync_supers(void *unused)
330 set_user_nice(current, 0);
332 while (!kthread_should_stop()) {
333 set_current_state(TASK_INTERRUPTIBLE);
337 * Do this periodically, like kupdated() did before.
345 void bdi_arm_supers_timer(void)
349 if (!dirty_writeback_interval)
352 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
353 mod_timer(&sync_supers_timer, round_jiffies_up(next));
356 static void sync_supers_timer_fn(unsigned long unused)
358 wake_up_process(sync_supers_tsk);
359 bdi_arm_supers_timer();
362 static int bdi_forker_task(void *ptr)
364 struct bdi_writeback *me = ptr;
366 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
370 * Our parent may run at a different priority, just set us to normal
372 set_user_nice(current, 0);
375 struct backing_dev_info *bdi, *tmp;
376 struct bdi_writeback *wb;
379 * Temporary measure, we want to make sure we don't see
380 * dirty data on the default backing_dev_info
382 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
383 wb_do_writeback(me, 0);
385 spin_lock_bh(&bdi_lock);
388 * Check if any existing bdi's have dirty data without
389 * a thread registered. If so, set that up.
391 list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) {
394 if (list_empty(&bdi->work_list) &&
395 !bdi_has_dirty_io(bdi))
398 bdi_add_default_flusher_task(bdi);
401 set_current_state(TASK_INTERRUPTIBLE);
403 if (list_empty(&bdi_pending_list)) {
406 spin_unlock_bh(&bdi_lock);
407 wait = msecs_to_jiffies(dirty_writeback_interval * 10);
409 schedule_timeout(wait);
416 __set_current_state(TASK_RUNNING);
419 * This is our real job - check for pending entries in
420 * bdi_pending_list, and create the tasks that got added
422 bdi = list_entry(bdi_pending_list.next, struct backing_dev_info,
424 list_del_init(&bdi->bdi_list);
425 spin_unlock_bh(&bdi_lock);
428 wb->task = kthread_run(bdi_start_fn, wb, "flush-%s",
431 * If task creation fails, then readd the bdi to
432 * the pending list and force writeout of the bdi
433 * from this forker thread. That will free some memory
434 * and we can try again.
436 if (IS_ERR(wb->task)) {
440 * Add this 'bdi' to the back, so we get
441 * a chance to flush other bdi's to free
444 spin_lock_bh(&bdi_lock);
445 list_add_tail(&bdi->bdi_list, &bdi_pending_list);
446 spin_unlock_bh(&bdi_lock);
455 static void bdi_add_to_pending(struct rcu_head *head)
457 struct backing_dev_info *bdi;
459 bdi = container_of(head, struct backing_dev_info, rcu_head);
460 INIT_LIST_HEAD(&bdi->bdi_list);
462 spin_lock(&bdi_lock);
463 list_add_tail(&bdi->bdi_list, &bdi_pending_list);
464 spin_unlock(&bdi_lock);
467 * We are now on the pending list, wake up bdi_forker_task()
468 * to finish the job and add us back to the active bdi_list
470 wake_up_process(default_backing_dev_info.wb.task);
474 * Add the default flusher task that gets created for any bdi
475 * that has dirty data pending writeout
477 void static bdi_add_default_flusher_task(struct backing_dev_info *bdi)
479 if (!bdi_cap_writeback_dirty(bdi))
482 if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) {
483 printk(KERN_ERR "bdi %p/%s is not registered!\n",
489 * Check with the helper whether to proceed adding a task. Will only
490 * abort if we two or more simultanous calls to
491 * bdi_add_default_flusher_task() occured, further additions will block
492 * waiting for previous additions to finish.
494 if (!test_and_set_bit(BDI_pending, &bdi->state)) {
495 list_del_rcu(&bdi->bdi_list);
498 * We must wait for the current RCU period to end before
499 * moving to the pending list. So schedule that operation
500 * from an RCU callback.
502 call_rcu(&bdi->rcu_head, bdi_add_to_pending);
507 * Remove bdi from bdi_list, and ensure that it is no longer visible
509 static void bdi_remove_from_list(struct backing_dev_info *bdi)
511 spin_lock_bh(&bdi_lock);
512 list_del_rcu(&bdi->bdi_list);
513 spin_unlock_bh(&bdi_lock);
518 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
519 const char *fmt, ...)
525 if (bdi->dev) /* The driver needs to use separate queues per device */
529 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
536 spin_lock_bh(&bdi_lock);
537 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
538 spin_unlock_bh(&bdi_lock);
543 * Just start the forker thread for our default backing_dev_info,
544 * and add other bdi's to the list. They will get a thread created
545 * on-demand when they need it.
547 if (bdi_cap_flush_forker(bdi)) {
548 struct bdi_writeback *wb = &bdi->wb;
550 wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s",
552 if (IS_ERR(wb->task)) {
556 bdi_remove_from_list(bdi);
561 bdi_debug_register(bdi, dev_name(dev));
562 set_bit(BDI_registered, &bdi->state);
566 EXPORT_SYMBOL(bdi_register);
568 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
570 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
572 EXPORT_SYMBOL(bdi_register_dev);
575 * Remove bdi from the global list and shutdown any threads we have running
577 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
579 if (!bdi_cap_writeback_dirty(bdi))
583 * If setup is pending, wait for that to complete first
585 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
586 TASK_UNINTERRUPTIBLE);
589 * Make sure nobody finds us on the bdi_list anymore
591 bdi_remove_from_list(bdi);
594 * Finally, kill the kernel thread. We don't need to be RCU
595 * safe anymore, since the bdi is gone from visibility. Force
596 * unfreeze of the thread before calling kthread_stop(), otherwise
597 * it would never exet if it is currently stuck in the refrigerator.
600 thaw_process(bdi->wb.task);
601 kthread_stop(bdi->wb.task);
606 * This bdi is going away now, make sure that no super_blocks point to it
608 static void bdi_prune_sb(struct backing_dev_info *bdi)
610 struct super_block *sb;
613 list_for_each_entry(sb, &super_blocks, s_list) {
614 if (sb->s_bdi == bdi)
617 spin_unlock(&sb_lock);
620 void bdi_unregister(struct backing_dev_info *bdi)
625 if (!bdi_cap_flush_forker(bdi))
626 bdi_wb_shutdown(bdi);
627 bdi_debug_unregister(bdi);
628 device_unregister(bdi->dev);
632 EXPORT_SYMBOL(bdi_unregister);
634 int bdi_init(struct backing_dev_info *bdi)
641 bdi->max_ratio = 100;
642 bdi->max_prop_frac = PROP_FRAC_BASE;
643 spin_lock_init(&bdi->wb_lock);
644 INIT_RCU_HEAD(&bdi->rcu_head);
645 INIT_LIST_HEAD(&bdi->bdi_list);
646 INIT_LIST_HEAD(&bdi->work_list);
648 bdi_wb_init(&bdi->wb, bdi);
650 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
651 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
656 bdi->dirty_exceeded = 0;
657 err = prop_local_init_percpu(&bdi->completions);
662 percpu_counter_destroy(&bdi->bdi_stat[i]);
667 EXPORT_SYMBOL(bdi_init);
669 void bdi_destroy(struct backing_dev_info *bdi)
674 * Splice our entries to the default_backing_dev_info, if this
677 if (bdi_has_dirty_io(bdi)) {
678 struct bdi_writeback *dst = &default_backing_dev_info.wb;
680 spin_lock(&inode_lock);
681 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
682 list_splice(&bdi->wb.b_io, &dst->b_io);
683 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
684 spin_unlock(&inode_lock);
689 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
690 percpu_counter_destroy(&bdi->bdi_stat[i]);
692 prop_local_destroy_percpu(&bdi->completions);
694 EXPORT_SYMBOL(bdi_destroy);
697 * For use from filesystems to quickly init and register a bdi associated
698 * with dirty writeback
700 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
707 bdi->capabilities = cap;
712 sprintf(tmp, "%.28s%s", name, "-%d");
713 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
721 EXPORT_SYMBOL(bdi_setup_and_register);
723 static wait_queue_head_t congestion_wqh[2] = {
724 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
725 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
728 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
731 wait_queue_head_t *wqh = &congestion_wqh[sync];
733 bit = sync ? BDI_sync_congested : BDI_async_congested;
734 clear_bit(bit, &bdi->state);
735 smp_mb__after_clear_bit();
736 if (waitqueue_active(wqh))
739 EXPORT_SYMBOL(clear_bdi_congested);
741 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
745 bit = sync ? BDI_sync_congested : BDI_async_congested;
746 set_bit(bit, &bdi->state);
748 EXPORT_SYMBOL(set_bdi_congested);
751 * congestion_wait - wait for a backing_dev to become uncongested
752 * @sync: SYNC or ASYNC IO
753 * @timeout: timeout in jiffies
755 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
756 * write congestion. If no backing_devs are congested then just wait for the
757 * next write to be completed.
759 long congestion_wait(int sync, long timeout)
763 wait_queue_head_t *wqh = &congestion_wqh[sync];
765 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
766 ret = io_schedule_timeout(timeout);
767 finish_wait(wqh, &wait);
770 EXPORT_SYMBOL(congestion_wait);