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
= {
34 EXPORT_SYMBOL_GPL(noop_backing_dev_info
);
36 static struct class *bdi_class
;
39 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
40 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
43 DEFINE_SPINLOCK(bdi_lock
);
45 LIST_HEAD(bdi_pending_list
);
47 static struct task_struct
*sync_supers_tsk
;
48 static struct timer_list sync_supers_timer
;
50 static int bdi_sync_supers(void *);
51 static void sync_supers_timer_fn(unsigned long);
53 #ifdef CONFIG_DEBUG_FS
54 #include <linux/debugfs.h>
55 #include <linux/seq_file.h>
57 static struct dentry
*bdi_debug_root
;
59 static void bdi_debug_init(void)
61 bdi_debug_root
= debugfs_create_dir("bdi", NULL
);
64 static int bdi_debug_stats_show(struct seq_file
*m
, void *v
)
66 struct backing_dev_info
*bdi
= m
->private;
67 struct bdi_writeback
*wb
= &bdi
->wb
;
68 unsigned long background_thresh
;
69 unsigned long dirty_thresh
;
70 unsigned long bdi_thresh
;
71 unsigned long nr_dirty
, nr_io
, nr_more_io
, nr_wb
;
74 nr_wb
= nr_dirty
= nr_io
= nr_more_io
= 0;
75 spin_lock(&inode_lock
);
76 list_for_each_entry(inode
, &wb
->b_dirty
, i_list
)
78 list_for_each_entry(inode
, &wb
->b_io
, i_list
)
80 list_for_each_entry(inode
, &wb
->b_more_io
, i_list
)
82 spin_unlock(&inode_lock
);
84 get_dirty_limits(&background_thresh
, &dirty_thresh
, &bdi_thresh
, bdi
);
86 #define K(x) ((x) << (PAGE_SHIFT - 10))
88 "BdiWriteback: %8lu kB\n"
89 "BdiReclaimable: %8lu kB\n"
90 "BdiDirtyThresh: %8lu kB\n"
91 "DirtyThresh: %8lu kB\n"
92 "BackgroundThresh: %8lu kB\n"
98 (unsigned long) K(bdi_stat(bdi
, BDI_WRITEBACK
)),
99 (unsigned long) K(bdi_stat(bdi
, BDI_RECLAIMABLE
)),
100 K(bdi_thresh
), K(dirty_thresh
),
101 K(background_thresh
), nr_dirty
, nr_io
, nr_more_io
,
102 !list_empty(&bdi
->bdi_list
), bdi
->state
);
108 static int bdi_debug_stats_open(struct inode
*inode
, struct file
*file
)
110 return single_open(file
, bdi_debug_stats_show
, inode
->i_private
);
113 static const struct file_operations bdi_debug_stats_fops
= {
114 .open
= bdi_debug_stats_open
,
117 .release
= single_release
,
120 static void bdi_debug_register(struct backing_dev_info
*bdi
, const char *name
)
122 bdi
->debug_dir
= debugfs_create_dir(name
, bdi_debug_root
);
123 bdi
->debug_stats
= debugfs_create_file("stats", 0444, bdi
->debug_dir
,
124 bdi
, &bdi_debug_stats_fops
);
127 static void bdi_debug_unregister(struct backing_dev_info
*bdi
)
129 debugfs_remove(bdi
->debug_stats
);
130 debugfs_remove(bdi
->debug_dir
);
133 static inline void bdi_debug_init(void)
136 static inline void bdi_debug_register(struct backing_dev_info
*bdi
,
140 static inline void bdi_debug_unregister(struct backing_dev_info
*bdi
)
145 static ssize_t
read_ahead_kb_store(struct device
*dev
,
146 struct device_attribute
*attr
,
147 const char *buf
, size_t count
)
149 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
151 unsigned long read_ahead_kb
;
152 ssize_t ret
= -EINVAL
;
154 read_ahead_kb
= simple_strtoul(buf
, &end
, 10);
155 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
156 bdi
->ra_pages
= read_ahead_kb
>> (PAGE_SHIFT
- 10);
162 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
164 #define BDI_SHOW(name, expr) \
165 static ssize_t name##_show(struct device *dev, \
166 struct device_attribute *attr, char *page) \
168 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
170 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
173 BDI_SHOW(read_ahead_kb
, K(bdi
->ra_pages
))
175 static ssize_t
min_ratio_store(struct device
*dev
,
176 struct device_attribute
*attr
, const char *buf
, size_t count
)
178 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
181 ssize_t ret
= -EINVAL
;
183 ratio
= simple_strtoul(buf
, &end
, 10);
184 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
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
);
199 ssize_t ret
= -EINVAL
;
201 ratio
= simple_strtoul(buf
, &end
, 10);
202 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
203 ret
= bdi_set_max_ratio(bdi
, ratio
);
209 BDI_SHOW(max_ratio
, bdi
->max_ratio
)
211 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
213 static struct device_attribute bdi_dev_attrs
[] = {
214 __ATTR_RW(read_ahead_kb
),
215 __ATTR_RW(min_ratio
),
216 __ATTR_RW(max_ratio
),
220 static __init
int bdi_class_init(void)
222 bdi_class
= class_create(THIS_MODULE
, "bdi");
223 if (IS_ERR(bdi_class
))
224 return PTR_ERR(bdi_class
);
226 bdi_class
->dev_attrs
= bdi_dev_attrs
;
230 postcore_initcall(bdi_class_init
);
232 static int __init
default_bdi_init(void)
236 sync_supers_tsk
= kthread_run(bdi_sync_supers
, NULL
, "sync_supers");
237 BUG_ON(IS_ERR(sync_supers_tsk
));
239 init_timer(&sync_supers_timer
);
240 setup_timer(&sync_supers_timer
, sync_supers_timer_fn
, 0);
241 bdi_arm_supers_timer();
243 err
= bdi_init(&default_backing_dev_info
);
245 bdi_register(&default_backing_dev_info
, NULL
, "default");
249 subsys_initcall(default_bdi_init
);
251 int bdi_has_dirty_io(struct backing_dev_info
*bdi
)
253 return wb_has_dirty_io(&bdi
->wb
);
256 static void bdi_flush_io(struct backing_dev_info
*bdi
)
258 struct writeback_control wbc
= {
259 .sync_mode
= WB_SYNC_NONE
,
260 .older_than_this
= NULL
,
265 writeback_inodes_wb(&bdi
->wb
, &wbc
);
269 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
270 * or we risk deadlocking on ->s_umount. The longer term solution would be
271 * to implement sync_supers_bdi() or similar and simply do it from the
272 * bdi writeback thread individually.
274 static int bdi_sync_supers(void *unused
)
276 set_user_nice(current
, 0);
278 while (!kthread_should_stop()) {
279 set_current_state(TASK_INTERRUPTIBLE
);
283 * Do this periodically, like kupdated() did before.
291 void bdi_arm_supers_timer(void)
295 if (!dirty_writeback_interval
)
298 next
= msecs_to_jiffies(dirty_writeback_interval
* 10) + jiffies
;
299 mod_timer(&sync_supers_timer
, round_jiffies_up(next
));
302 static void sync_supers_timer_fn(unsigned long unused
)
304 wake_up_process(sync_supers_tsk
);
305 bdi_arm_supers_timer();
308 static void wakeup_timer_fn(unsigned long data
)
310 struct backing_dev_info
*bdi
= (struct backing_dev_info
*)data
;
312 spin_lock_bh(&bdi
->wb_lock
);
314 wake_up_process(bdi
->wb
.task
);
317 * When bdi tasks are inactive for long time, they are killed.
318 * In this case we have to wake-up the forker thread which
319 * should create and run the bdi thread.
321 wake_up_process(default_backing_dev_info
.wb
.task
);
323 spin_unlock_bh(&bdi
->wb_lock
);
327 * This function is used when the first inode for this bdi is marked dirty. It
328 * wakes-up the corresponding bdi thread which should then take care of the
329 * periodic background write-out of dirty inodes. Since the write-out would
330 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
331 * set up a timer which wakes the bdi thread up later.
333 * Note, we wouldn't bother setting up the timer, but this function is on the
334 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
335 * by delaying the wake-up.
337 void bdi_wakeup_thread_delayed(struct backing_dev_info
*bdi
)
339 unsigned long timeout
;
341 timeout
= msecs_to_jiffies(dirty_writeback_interval
* 10);
342 mod_timer(&bdi
->wb
.wakeup_timer
, jiffies
+ timeout
);
346 * Calculate the longest interval (jiffies) bdi threads are allowed to be
349 static unsigned long bdi_longest_inactive(void)
351 unsigned long interval
;
353 interval
= msecs_to_jiffies(dirty_writeback_interval
* 10);
354 return max(5UL * 60 * HZ
, interval
);
357 static int bdi_forker_thread(void *ptr
)
359 struct bdi_writeback
*me
= ptr
;
361 current
->flags
|= PF_FLUSHER
| PF_SWAPWRITE
;
365 * Our parent may run at a different priority, just set us to normal
367 set_user_nice(current
, 0);
370 struct task_struct
*task
= NULL
;
371 struct backing_dev_info
*bdi
;
373 NO_ACTION
, /* Nothing to do */
374 FORK_THREAD
, /* Fork bdi thread */
375 KILL_THREAD
, /* Kill inactive bdi thread */
376 } action
= NO_ACTION
;
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 del_timer(&me
->wakeup_timer
);
384 wb_do_writeback(me
, 0);
387 spin_lock_bh(&bdi_lock
);
388 set_current_state(TASK_INTERRUPTIBLE
);
390 list_for_each_entry(bdi
, &bdi_list
, bdi_list
) {
393 if (!bdi_cap_writeback_dirty(bdi
) ||
394 bdi_cap_flush_forker(bdi
))
397 WARN(!test_bit(BDI_registered
, &bdi
->state
),
398 "bdi %p/%s is not registered!\n", bdi
, bdi
->name
);
400 have_dirty_io
= !list_empty(&bdi
->work_list
) ||
401 wb_has_dirty_io(&bdi
->wb
);
404 * If the bdi has work to do, but the thread does not
407 if (!bdi
->wb
.task
&& have_dirty_io
) {
409 * Set the pending bit - if someone will try to
410 * unregister this bdi - it'll wait on this bit.
412 set_bit(BDI_pending
, &bdi
->state
);
413 action
= FORK_THREAD
;
417 spin_lock_bh(&bdi
->wb_lock
);
419 * If there is no work to do and the bdi thread was
420 * inactive long enough - kill it. The wb_lock is taken
421 * to make sure no-one adds more work to this bdi and
422 * wakes the bdi thread up.
424 if (bdi
->wb
.task
&& !have_dirty_io
&&
425 time_after(jiffies
, bdi
->wb
.last_active
+
426 bdi_longest_inactive())) {
429 spin_unlock(&bdi
->wb_lock
);
430 set_bit(BDI_pending
, &bdi
->state
);
431 action
= KILL_THREAD
;
434 spin_unlock_bh(&bdi
->wb_lock
);
436 spin_unlock_bh(&bdi_lock
);
438 /* Keep working if default bdi still has things to do */
439 if (!list_empty(&me
->bdi
->work_list
))
440 __set_current_state(TASK_RUNNING
);
444 __set_current_state(TASK_RUNNING
);
445 task
= kthread_run(bdi_writeback_thread
, &bdi
->wb
, "flush-%s",
449 * If thread creation fails, force writeout of
450 * the bdi from the thread.
455 * The spinlock makes sure we do not lose
456 * wake-ups when racing with 'bdi_queue_work()'.
458 spin_lock_bh(&bdi
->wb_lock
);
460 spin_unlock_bh(&bdi
->wb_lock
);
465 __set_current_state(TASK_RUNNING
);
470 if (!wb_has_dirty_io(me
) || !dirty_writeback_interval
)
472 * There are no dirty data. The only thing we
473 * should now care about is checking for
474 * inactive bdi threads and killing them. Thus,
475 * let's sleep for longer time, save energy and
476 * be friendly for battery-driven devices.
478 schedule_timeout(bdi_longest_inactive());
480 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval
* 10));
482 /* Back to the main loop */
487 * Clear pending bit and wakeup anybody waiting to tear us down.
489 clear_bit(BDI_pending
, &bdi
->state
);
490 smp_mb__after_clear_bit();
491 wake_up_bit(&bdi
->state
, BDI_pending
);
498 * Remove bdi from bdi_list, and ensure that it is no longer visible
500 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
502 spin_lock_bh(&bdi_lock
);
503 list_del_rcu(&bdi
->bdi_list
);
504 spin_unlock_bh(&bdi_lock
);
509 int bdi_register(struct backing_dev_info
*bdi
, struct device
*parent
,
510 const char *fmt
, ...)
516 if (bdi
->dev
) /* The driver needs to use separate queues per device */
520 dev
= device_create_vargs(bdi_class
, parent
, MKDEV(0, 0), bdi
, fmt
, args
);
527 spin_lock_bh(&bdi_lock
);
528 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
529 spin_unlock_bh(&bdi_lock
);
534 * Just start the forker thread for our default backing_dev_info,
535 * and add other bdi's to the list. They will get a thread created
536 * on-demand when they need it.
538 if (bdi_cap_flush_forker(bdi
)) {
539 struct bdi_writeback
*wb
= &bdi
->wb
;
541 wb
->task
= kthread_run(bdi_forker_thread
, wb
, "bdi-%s",
543 if (IS_ERR(wb
->task
)) {
547 bdi_remove_from_list(bdi
);
552 bdi_debug_register(bdi
, dev_name(dev
));
553 set_bit(BDI_registered
, &bdi
->state
);
554 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])
733 void clear_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
736 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
738 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
739 clear_bit(bit
, &bdi
->state
);
740 smp_mb__after_clear_bit();
741 if (waitqueue_active(wqh
))
744 EXPORT_SYMBOL(clear_bdi_congested
);
746 void set_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
750 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
751 set_bit(bit
, &bdi
->state
);
753 EXPORT_SYMBOL(set_bdi_congested
);
756 * congestion_wait - wait for a backing_dev to become uncongested
757 * @sync: SYNC or ASYNC IO
758 * @timeout: timeout in jiffies
760 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
761 * write congestion. If no backing_devs are congested then just wait for the
762 * next write to be completed.
764 long congestion_wait(int sync
, long timeout
)
768 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
770 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
771 ret
= io_schedule_timeout(timeout
);
772 finish_wait(wqh
, &wait
);
775 EXPORT_SYMBOL(congestion_wait
);