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
= {
32 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
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 static void bdi_add_default_flusher_task(struct backing_dev_info
*bdi
);
55 #ifdef CONFIG_DEBUG_FS
56 #include <linux/debugfs.h>
57 #include <linux/seq_file.h>
59 static struct dentry
*bdi_debug_root
;
61 static void bdi_debug_init(void)
63 bdi_debug_root
= debugfs_create_dir("bdi", NULL
);
66 static int bdi_debug_stats_show(struct seq_file
*m
, void *v
)
68 struct backing_dev_info
*bdi
= m
->private;
69 struct bdi_writeback
*wb
;
70 unsigned long background_thresh
;
71 unsigned long dirty_thresh
;
72 unsigned long bdi_thresh
;
73 unsigned long nr_dirty
, nr_io
, nr_more_io
, nr_wb
;
77 * inode lock is enough here, the bdi->wb_list is protected by
78 * RCU on the reader side
80 nr_wb
= nr_dirty
= nr_io
= nr_more_io
= 0;
81 spin_lock(&inode_lock
);
82 list_for_each_entry(wb
, &bdi
->wb_list
, list
) {
84 list_for_each_entry(inode
, &wb
->b_dirty
, i_list
)
86 list_for_each_entry(inode
, &wb
->b_io
, i_list
)
88 list_for_each_entry(inode
, &wb
->b_more_io
, i_list
)
91 spin_unlock(&inode_lock
);
93 get_dirty_limits(&background_thresh
, &dirty_thresh
, &bdi_thresh
, bdi
);
95 #define K(x) ((x) << (PAGE_SHIFT - 10))
97 "BdiWriteback: %8lu kB\n"
98 "BdiReclaimable: %8lu kB\n"
99 "BdiDirtyThresh: %8lu kB\n"
100 "DirtyThresh: %8lu kB\n"
101 "BackgroundThresh: %8lu kB\n"
102 "WritebackThreads: %8lu\n"
109 (unsigned long) K(bdi_stat(bdi
, BDI_WRITEBACK
)),
110 (unsigned long) K(bdi_stat(bdi
, BDI_RECLAIMABLE
)),
111 K(bdi_thresh
), K(dirty_thresh
),
112 K(background_thresh
), nr_wb
, nr_dirty
, nr_io
, nr_more_io
,
113 !list_empty(&bdi
->bdi_list
), bdi
->state
,
114 !list_empty(&bdi
->wb_list
));
120 static int bdi_debug_stats_open(struct inode
*inode
, struct file
*file
)
122 return single_open(file
, bdi_debug_stats_show
, inode
->i_private
);
125 static const struct file_operations bdi_debug_stats_fops
= {
126 .open
= bdi_debug_stats_open
,
129 .release
= single_release
,
132 static void bdi_debug_register(struct backing_dev_info
*bdi
, const char *name
)
134 bdi
->debug_dir
= debugfs_create_dir(name
, bdi_debug_root
);
135 bdi
->debug_stats
= debugfs_create_file("stats", 0444, bdi
->debug_dir
,
136 bdi
, &bdi_debug_stats_fops
);
139 static void bdi_debug_unregister(struct backing_dev_info
*bdi
)
141 debugfs_remove(bdi
->debug_stats
);
142 debugfs_remove(bdi
->debug_dir
);
145 static inline void bdi_debug_init(void)
148 static inline void bdi_debug_register(struct backing_dev_info
*bdi
,
152 static inline void bdi_debug_unregister(struct backing_dev_info
*bdi
)
157 static ssize_t
read_ahead_kb_store(struct device
*dev
,
158 struct device_attribute
*attr
,
159 const char *buf
, size_t count
)
161 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
163 unsigned long read_ahead_kb
;
164 ssize_t ret
= -EINVAL
;
166 read_ahead_kb
= simple_strtoul(buf
, &end
, 10);
167 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
168 bdi
->ra_pages
= read_ahead_kb
>> (PAGE_SHIFT
- 10);
174 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
176 #define BDI_SHOW(name, expr) \
177 static ssize_t name##_show(struct device *dev, \
178 struct device_attribute *attr, char *page) \
180 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
182 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
185 BDI_SHOW(read_ahead_kb
, K(bdi
->ra_pages
))
187 static ssize_t
min_ratio_store(struct device
*dev
,
188 struct device_attribute
*attr
, const char *buf
, size_t count
)
190 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
193 ssize_t ret
= -EINVAL
;
195 ratio
= simple_strtoul(buf
, &end
, 10);
196 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
197 ret
= bdi_set_min_ratio(bdi
, ratio
);
203 BDI_SHOW(min_ratio
, bdi
->min_ratio
)
205 static ssize_t
max_ratio_store(struct device
*dev
,
206 struct device_attribute
*attr
, const char *buf
, size_t count
)
208 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
211 ssize_t ret
= -EINVAL
;
213 ratio
= simple_strtoul(buf
, &end
, 10);
214 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
215 ret
= bdi_set_max_ratio(bdi
, ratio
);
221 BDI_SHOW(max_ratio
, bdi
->max_ratio
)
223 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
225 static struct device_attribute bdi_dev_attrs
[] = {
226 __ATTR_RW(read_ahead_kb
),
227 __ATTR_RW(min_ratio
),
228 __ATTR_RW(max_ratio
),
232 static __init
int bdi_class_init(void)
234 bdi_class
= class_create(THIS_MODULE
, "bdi");
235 if (IS_ERR(bdi_class
))
236 return PTR_ERR(bdi_class
);
238 bdi_class
->dev_attrs
= bdi_dev_attrs
;
242 postcore_initcall(bdi_class_init
);
244 static int __init
default_bdi_init(void)
248 sync_supers_tsk
= kthread_run(bdi_sync_supers
, NULL
, "sync_supers");
249 BUG_ON(IS_ERR(sync_supers_tsk
));
251 init_timer(&sync_supers_timer
);
252 setup_timer(&sync_supers_timer
, sync_supers_timer_fn
, 0);
253 bdi_arm_supers_timer();
255 err
= bdi_init(&default_backing_dev_info
);
257 bdi_register(&default_backing_dev_info
, NULL
, "default");
258 err
= bdi_init(&noop_backing_dev_info
);
262 subsys_initcall(default_bdi_init
);
264 static void bdi_wb_init(struct bdi_writeback
*wb
, struct backing_dev_info
*bdi
)
266 memset(wb
, 0, sizeof(*wb
));
269 wb
->last_old_flush
= jiffies
;
270 INIT_LIST_HEAD(&wb
->b_dirty
);
271 INIT_LIST_HEAD(&wb
->b_io
);
272 INIT_LIST_HEAD(&wb
->b_more_io
);
275 static void bdi_task_init(struct backing_dev_info
*bdi
,
276 struct bdi_writeback
*wb
)
278 struct task_struct
*tsk
= current
;
280 spin_lock(&bdi
->wb_lock
);
281 list_add_tail_rcu(&wb
->list
, &bdi
->wb_list
);
282 spin_unlock(&bdi
->wb_lock
);
284 tsk
->flags
|= PF_FLUSHER
| PF_SWAPWRITE
;
288 * Our parent may run at a different priority, just set us to normal
290 set_user_nice(tsk
, 0);
293 static int bdi_start_fn(void *ptr
)
295 struct bdi_writeback
*wb
= ptr
;
296 struct backing_dev_info
*bdi
= wb
->bdi
;
300 * Add us to the active bdi_list
302 spin_lock_bh(&bdi_lock
);
303 list_add_rcu(&bdi
->bdi_list
, &bdi_list
);
304 spin_unlock_bh(&bdi_lock
);
306 bdi_task_init(bdi
, wb
);
309 * Clear pending bit and wakeup anybody waiting to tear us down
311 clear_bit(BDI_pending
, &bdi
->state
);
312 smp_mb__after_clear_bit();
313 wake_up_bit(&bdi
->state
, BDI_pending
);
315 ret
= bdi_writeback_task(wb
);
318 * Remove us from the list
320 spin_lock(&bdi
->wb_lock
);
321 list_del_rcu(&wb
->list
);
322 spin_unlock(&bdi
->wb_lock
);
325 * Flush any work that raced with us exiting. No new work
326 * will be added, since this bdi isn't discoverable anymore.
328 if (!list_empty(&bdi
->work_list
))
329 wb_do_writeback(wb
, 1);
335 int bdi_has_dirty_io(struct backing_dev_info
*bdi
)
337 return wb_has_dirty_io(&bdi
->wb
);
340 static void bdi_flush_io(struct backing_dev_info
*bdi
)
342 struct writeback_control wbc
= {
343 .sync_mode
= WB_SYNC_NONE
,
344 .older_than_this
= NULL
,
349 writeback_inodes_wb(&bdi
->wb
, &wbc
);
353 * kupdated() used to do this. We cannot do it from the bdi_forker_task()
354 * or we risk deadlocking on ->s_umount. The longer term solution would be
355 * to implement sync_supers_bdi() or similar and simply do it from the
356 * bdi writeback tasks individually.
358 static int bdi_sync_supers(void *unused
)
360 set_user_nice(current
, 0);
362 while (!kthread_should_stop()) {
363 set_current_state(TASK_INTERRUPTIBLE
);
367 * Do this periodically, like kupdated() did before.
375 void bdi_arm_supers_timer(void)
379 if (!dirty_writeback_interval
)
382 next
= msecs_to_jiffies(dirty_writeback_interval
* 10) + jiffies
;
383 mod_timer(&sync_supers_timer
, round_jiffies_up(next
));
386 static void sync_supers_timer_fn(unsigned long unused
)
388 wake_up_process(sync_supers_tsk
);
389 bdi_arm_supers_timer();
392 static int bdi_forker_task(void *ptr
)
394 struct bdi_writeback
*me
= ptr
;
396 bdi_task_init(me
->bdi
, me
);
399 struct backing_dev_info
*bdi
, *tmp
;
400 struct bdi_writeback
*wb
;
403 * Temporary measure, we want to make sure we don't see
404 * dirty data on the default backing_dev_info
406 if (wb_has_dirty_io(me
) || !list_empty(&me
->bdi
->work_list
))
407 wb_do_writeback(me
, 0);
409 spin_lock_bh(&bdi_lock
);
412 * Check if any existing bdi's have dirty data without
413 * a thread registered. If so, set that up.
415 list_for_each_entry_safe(bdi
, tmp
, &bdi_list
, bdi_list
) {
418 if (list_empty(&bdi
->work_list
) &&
419 !bdi_has_dirty_io(bdi
))
422 bdi_add_default_flusher_task(bdi
);
425 set_current_state(TASK_INTERRUPTIBLE
);
427 if (list_empty(&bdi_pending_list
)) {
430 spin_unlock_bh(&bdi_lock
);
431 wait
= msecs_to_jiffies(dirty_writeback_interval
* 10);
433 schedule_timeout(wait
);
440 __set_current_state(TASK_RUNNING
);
443 * This is our real job - check for pending entries in
444 * bdi_pending_list, and create the tasks that got added
446 bdi
= list_entry(bdi_pending_list
.next
, struct backing_dev_info
,
448 list_del_init(&bdi
->bdi_list
);
449 spin_unlock_bh(&bdi_lock
);
452 wb
->task
= kthread_run(bdi_start_fn
, wb
, "flush-%s",
455 * If task creation fails, then readd the bdi to
456 * the pending list and force writeout of the bdi
457 * from this forker thread. That will free some memory
458 * and we can try again.
460 if (IS_ERR(wb
->task
)) {
464 * Add this 'bdi' to the back, so we get
465 * a chance to flush other bdi's to free
468 spin_lock_bh(&bdi_lock
);
469 list_add_tail(&bdi
->bdi_list
, &bdi_pending_list
);
470 spin_unlock_bh(&bdi_lock
);
479 static void bdi_add_to_pending(struct rcu_head
*head
)
481 struct backing_dev_info
*bdi
;
483 bdi
= container_of(head
, struct backing_dev_info
, rcu_head
);
484 INIT_LIST_HEAD(&bdi
->bdi_list
);
486 spin_lock(&bdi_lock
);
487 list_add_tail(&bdi
->bdi_list
, &bdi_pending_list
);
488 spin_unlock(&bdi_lock
);
491 * We are now on the pending list, wake up bdi_forker_task()
492 * to finish the job and add us back to the active bdi_list
494 wake_up_process(default_backing_dev_info
.wb
.task
);
498 * Add the default flusher task that gets created for any bdi
499 * that has dirty data pending writeout
501 void static bdi_add_default_flusher_task(struct backing_dev_info
*bdi
)
503 if (!bdi_cap_writeback_dirty(bdi
))
506 if (WARN_ON(!test_bit(BDI_registered
, &bdi
->state
))) {
507 printk(KERN_ERR
"bdi %p/%s is not registered!\n",
513 * Check with the helper whether to proceed adding a task. Will only
514 * abort if we two or more simultanous calls to
515 * bdi_add_default_flusher_task() occured, further additions will block
516 * waiting for previous additions to finish.
518 if (!test_and_set_bit(BDI_pending
, &bdi
->state
)) {
519 list_del_rcu(&bdi
->bdi_list
);
522 * We must wait for the current RCU period to end before
523 * moving to the pending list. So schedule that operation
524 * from an RCU callback.
526 call_rcu(&bdi
->rcu_head
, bdi_add_to_pending
);
531 * Remove bdi from bdi_list, and ensure that it is no longer visible
533 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
535 spin_lock_bh(&bdi_lock
);
536 list_del_rcu(&bdi
->bdi_list
);
537 spin_unlock_bh(&bdi_lock
);
542 int bdi_register(struct backing_dev_info
*bdi
, struct device
*parent
,
543 const char *fmt
, ...)
549 if (bdi
->dev
) /* The driver needs to use separate queues per device */
553 dev
= device_create_vargs(bdi_class
, parent
, MKDEV(0, 0), bdi
, fmt
, args
);
560 spin_lock_bh(&bdi_lock
);
561 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
562 spin_unlock_bh(&bdi_lock
);
567 * Just start the forker thread for our default backing_dev_info,
568 * and add other bdi's to the list. They will get a thread created
569 * on-demand when they need it.
571 if (bdi_cap_flush_forker(bdi
)) {
572 struct bdi_writeback
*wb
= &bdi
->wb
;
574 wb
->task
= kthread_run(bdi_forker_task
, wb
, "bdi-%s",
576 if (IS_ERR(wb
->task
)) {
580 bdi_remove_from_list(bdi
);
585 bdi_debug_register(bdi
, dev_name(dev
));
586 set_bit(BDI_registered
, &bdi
->state
);
590 EXPORT_SYMBOL(bdi_register
);
592 int bdi_register_dev(struct backing_dev_info
*bdi
, dev_t dev
)
594 return bdi_register(bdi
, NULL
, "%u:%u", MAJOR(dev
), MINOR(dev
));
596 EXPORT_SYMBOL(bdi_register_dev
);
599 * Remove bdi from the global list and shutdown any threads we have running
601 static void bdi_wb_shutdown(struct backing_dev_info
*bdi
)
603 struct bdi_writeback
*wb
;
605 if (!bdi_cap_writeback_dirty(bdi
))
609 * If setup is pending, wait for that to complete first
611 wait_on_bit(&bdi
->state
, BDI_pending
, bdi_sched_wait
,
612 TASK_UNINTERRUPTIBLE
);
615 * Make sure nobody finds us on the bdi_list anymore
617 bdi_remove_from_list(bdi
);
620 * Finally, kill the kernel threads. We don't need to be RCU
621 * safe anymore, since the bdi is gone from visibility. Force
622 * unfreeze of the thread before calling kthread_stop(), otherwise
623 * it would never exet if it is currently stuck in the refrigerator.
625 list_for_each_entry(wb
, &bdi
->wb_list
, list
) {
626 thaw_process(wb
->task
);
627 kthread_stop(wb
->task
);
632 * This bdi is going away now, make sure that no super_blocks point to it
634 static void bdi_prune_sb(struct backing_dev_info
*bdi
)
636 struct super_block
*sb
;
639 list_for_each_entry(sb
, &super_blocks
, s_list
) {
640 if (sb
->s_bdi
== bdi
)
643 spin_unlock(&sb_lock
);
646 void bdi_unregister(struct backing_dev_info
*bdi
)
651 if (!bdi_cap_flush_forker(bdi
))
652 bdi_wb_shutdown(bdi
);
653 bdi_debug_unregister(bdi
);
654 device_unregister(bdi
->dev
);
658 EXPORT_SYMBOL(bdi_unregister
);
660 int bdi_init(struct backing_dev_info
*bdi
)
667 bdi
->max_ratio
= 100;
668 bdi
->max_prop_frac
= PROP_FRAC_BASE
;
669 spin_lock_init(&bdi
->wb_lock
);
670 INIT_RCU_HEAD(&bdi
->rcu_head
);
671 INIT_LIST_HEAD(&bdi
->bdi_list
);
672 INIT_LIST_HEAD(&bdi
->wb_list
);
673 INIT_LIST_HEAD(&bdi
->work_list
);
675 bdi_wb_init(&bdi
->wb
, bdi
);
677 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++) {
678 err
= percpu_counter_init(&bdi
->bdi_stat
[i
], 0);
683 bdi
->dirty_exceeded
= 0;
684 err
= prop_local_init_percpu(&bdi
->completions
);
689 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
694 EXPORT_SYMBOL(bdi_init
);
696 void bdi_destroy(struct backing_dev_info
*bdi
)
701 * Splice our entries to the default_backing_dev_info, if this
704 if (bdi_has_dirty_io(bdi
)) {
705 struct bdi_writeback
*dst
= &default_backing_dev_info
.wb
;
707 spin_lock(&inode_lock
);
708 list_splice(&bdi
->wb
.b_dirty
, &dst
->b_dirty
);
709 list_splice(&bdi
->wb
.b_io
, &dst
->b_io
);
710 list_splice(&bdi
->wb
.b_more_io
, &dst
->b_more_io
);
711 spin_unlock(&inode_lock
);
716 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++)
717 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
719 prop_local_destroy_percpu(&bdi
->completions
);
721 EXPORT_SYMBOL(bdi_destroy
);
724 * For use from filesystems to quickly init and register a bdi associated
725 * with dirty writeback
727 int bdi_setup_and_register(struct backing_dev_info
*bdi
, char *name
,
734 bdi
->capabilities
= cap
;
739 sprintf(tmp
, "%.28s%s", name
, "-%d");
740 err
= bdi_register(bdi
, NULL
, tmp
, atomic_long_inc_return(&bdi_seq
));
748 EXPORT_SYMBOL(bdi_setup_and_register
);
750 static wait_queue_head_t congestion_wqh
[2] = {
751 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
752 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
755 void clear_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
758 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
760 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
761 clear_bit(bit
, &bdi
->state
);
762 smp_mb__after_clear_bit();
763 if (waitqueue_active(wqh
))
766 EXPORT_SYMBOL(clear_bdi_congested
);
768 void set_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
772 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
773 set_bit(bit
, &bdi
->state
);
775 EXPORT_SYMBOL(set_bdi_congested
);
778 * congestion_wait - wait for a backing_dev to become uncongested
779 * @sync: SYNC or ASYNC IO
780 * @timeout: timeout in jiffies
782 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
783 * write congestion. If no backing_devs are congested then just wait for the
784 * next write to be completed.
786 long congestion_wait(int sync
, long timeout
)
790 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
792 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
793 ret
= io_schedule_timeout(timeout
);
794 finish_wait(wqh
, &wait
);
797 EXPORT_SYMBOL(congestion_wait
);