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
;
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
;
76 * inode lock is enough here, the bdi->wb_list is protected by
77 * RCU on the reader side
79 nr_wb
= nr_dirty
= nr_io
= nr_more_io
= 0;
80 spin_lock(&inode_lock
);
81 list_for_each_entry(wb
, &bdi
->wb_list
, list
) {
83 list_for_each_entry(inode
, &wb
->b_dirty
, i_list
)
85 list_for_each_entry(inode
, &wb
->b_io
, i_list
)
87 list_for_each_entry(inode
, &wb
->b_more_io
, i_list
)
90 spin_unlock(&inode_lock
);
92 get_dirty_limits(&background_thresh
, &dirty_thresh
, &bdi_thresh
, bdi
);
94 #define K(x) ((x) << (PAGE_SHIFT - 10))
96 "BdiWriteback: %8lu kB\n"
97 "BdiReclaimable: %8lu kB\n"
98 "BdiDirtyThresh: %8lu kB\n"
99 "DirtyThresh: %8lu kB\n"
100 "BackgroundThresh: %8lu kB\n"
101 "WritebackThreads: %8lu\n"
110 (unsigned long) K(bdi_stat(bdi
, BDI_WRITEBACK
)),
111 (unsigned long) K(bdi_stat(bdi
, BDI_RECLAIMABLE
)),
112 K(bdi_thresh
), K(dirty_thresh
),
113 K(background_thresh
), nr_wb
, nr_dirty
, nr_io
, nr_more_io
,
114 !list_empty(&bdi
->bdi_list
), bdi
->state
, bdi
->wb_mask
,
115 !list_empty(&bdi
->wb_list
), bdi
->wb_cnt
);
121 static int bdi_debug_stats_open(struct inode
*inode
, struct file
*file
)
123 return single_open(file
, bdi_debug_stats_show
, inode
->i_private
);
126 static const struct file_operations bdi_debug_stats_fops
= {
127 .open
= bdi_debug_stats_open
,
130 .release
= single_release
,
133 static void bdi_debug_register(struct backing_dev_info
*bdi
, const char *name
)
135 bdi
->debug_dir
= debugfs_create_dir(name
, bdi_debug_root
);
136 bdi
->debug_stats
= debugfs_create_file("stats", 0444, bdi
->debug_dir
,
137 bdi
, &bdi_debug_stats_fops
);
140 static void bdi_debug_unregister(struct backing_dev_info
*bdi
)
142 debugfs_remove(bdi
->debug_stats
);
143 debugfs_remove(bdi
->debug_dir
);
146 static inline void bdi_debug_init(void)
149 static inline void bdi_debug_register(struct backing_dev_info
*bdi
,
153 static inline void bdi_debug_unregister(struct backing_dev_info
*bdi
)
158 static ssize_t
read_ahead_kb_store(struct device
*dev
,
159 struct device_attribute
*attr
,
160 const char *buf
, size_t count
)
162 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
164 unsigned long read_ahead_kb
;
165 ssize_t ret
= -EINVAL
;
167 read_ahead_kb
= simple_strtoul(buf
, &end
, 10);
168 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
169 bdi
->ra_pages
= read_ahead_kb
>> (PAGE_SHIFT
- 10);
175 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
177 #define BDI_SHOW(name, expr) \
178 static ssize_t name##_show(struct device *dev, \
179 struct device_attribute *attr, char *page) \
181 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
183 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
186 BDI_SHOW(read_ahead_kb
, K(bdi
->ra_pages
))
188 static ssize_t
min_ratio_store(struct device
*dev
,
189 struct device_attribute
*attr
, const char *buf
, size_t count
)
191 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
194 ssize_t ret
= -EINVAL
;
196 ratio
= simple_strtoul(buf
, &end
, 10);
197 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
198 ret
= bdi_set_min_ratio(bdi
, ratio
);
204 BDI_SHOW(min_ratio
, bdi
->min_ratio
)
206 static ssize_t
max_ratio_store(struct device
*dev
,
207 struct device_attribute
*attr
, const char *buf
, size_t count
)
209 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
212 ssize_t ret
= -EINVAL
;
214 ratio
= simple_strtoul(buf
, &end
, 10);
215 if (*buf
&& (end
[0] == '\0' || (end
[0] == '\n' && end
[1] == '\0'))) {
216 ret
= bdi_set_max_ratio(bdi
, ratio
);
222 BDI_SHOW(max_ratio
, bdi
->max_ratio
)
224 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
226 static struct device_attribute bdi_dev_attrs
[] = {
227 __ATTR_RW(read_ahead_kb
),
228 __ATTR_RW(min_ratio
),
229 __ATTR_RW(max_ratio
),
233 static __init
int bdi_class_init(void)
235 bdi_class
= class_create(THIS_MODULE
, "bdi");
236 if (IS_ERR(bdi_class
))
237 return PTR_ERR(bdi_class
);
239 bdi_class
->dev_attrs
= bdi_dev_attrs
;
243 postcore_initcall(bdi_class_init
);
245 static int __init
default_bdi_init(void)
249 sync_supers_tsk
= kthread_run(bdi_sync_supers
, NULL
, "sync_supers");
250 BUG_ON(IS_ERR(sync_supers_tsk
));
252 init_timer(&sync_supers_timer
);
253 setup_timer(&sync_supers_timer
, sync_supers_timer_fn
, 0);
254 bdi_arm_supers_timer();
256 err
= bdi_init(&default_backing_dev_info
);
258 bdi_register(&default_backing_dev_info
, NULL
, "default");
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
= {
344 .sync_mode
= WB_SYNC_NONE
,
345 .older_than_this
= NULL
,
350 writeback_inodes_wbc(&wbc
);
354 * kupdated() used to do this. We cannot do it from the bdi_forker_task()
355 * or we risk deadlocking on ->s_umount. The longer term solution would be
356 * to implement sync_supers_bdi() or similar and simply do it from the
357 * bdi writeback tasks individually.
359 static int bdi_sync_supers(void *unused
)
361 set_user_nice(current
, 0);
363 while (!kthread_should_stop()) {
364 set_current_state(TASK_INTERRUPTIBLE
);
368 * Do this periodically, like kupdated() did before.
376 void bdi_arm_supers_timer(void)
380 if (!dirty_writeback_interval
)
383 next
= msecs_to_jiffies(dirty_writeback_interval
* 10) + jiffies
;
384 mod_timer(&sync_supers_timer
, round_jiffies_up(next
));
387 static void sync_supers_timer_fn(unsigned long unused
)
389 wake_up_process(sync_supers_tsk
);
390 bdi_arm_supers_timer();
393 static int bdi_forker_task(void *ptr
)
395 struct bdi_writeback
*me
= ptr
;
397 bdi_task_init(me
->bdi
, me
);
400 struct backing_dev_info
*bdi
, *tmp
;
401 struct bdi_writeback
*wb
;
404 * Temporary measure, we want to make sure we don't see
405 * dirty data on the default backing_dev_info
407 if (wb_has_dirty_io(me
) || !list_empty(&me
->bdi
->work_list
))
408 wb_do_writeback(me
, 0);
410 spin_lock_bh(&bdi_lock
);
413 * Check if any existing bdi's have dirty data without
414 * a thread registered. If so, set that up.
416 list_for_each_entry_safe(bdi
, tmp
, &bdi_list
, bdi_list
) {
419 if (list_empty(&bdi
->work_list
) &&
420 !bdi_has_dirty_io(bdi
))
423 bdi_add_default_flusher_task(bdi
);
426 set_current_state(TASK_INTERRUPTIBLE
);
428 if (list_empty(&bdi_pending_list
)) {
431 spin_unlock_bh(&bdi_lock
);
432 wait
= msecs_to_jiffies(dirty_writeback_interval
* 10);
434 schedule_timeout(wait
);
441 __set_current_state(TASK_RUNNING
);
444 * This is our real job - check for pending entries in
445 * bdi_pending_list, and create the tasks that got added
447 bdi
= list_entry(bdi_pending_list
.next
, struct backing_dev_info
,
449 list_del_init(&bdi
->bdi_list
);
450 spin_unlock_bh(&bdi_lock
);
453 wb
->task
= kthread_run(bdi_start_fn
, wb
, "flush-%s",
456 * If task creation fails, then readd the bdi to
457 * the pending list and force writeout of the bdi
458 * from this forker thread. That will free some memory
459 * and we can try again.
461 if (IS_ERR(wb
->task
)) {
465 * Add this 'bdi' to the back, so we get
466 * a chance to flush other bdi's to free
469 spin_lock_bh(&bdi_lock
);
470 list_add_tail(&bdi
->bdi_list
, &bdi_pending_list
);
471 spin_unlock_bh(&bdi_lock
);
480 static void bdi_add_to_pending(struct rcu_head
*head
)
482 struct backing_dev_info
*bdi
;
484 bdi
= container_of(head
, struct backing_dev_info
, rcu_head
);
485 INIT_LIST_HEAD(&bdi
->bdi_list
);
487 spin_lock(&bdi_lock
);
488 list_add_tail(&bdi
->bdi_list
, &bdi_pending_list
);
489 spin_unlock(&bdi_lock
);
492 * We are now on the pending list, wake up bdi_forker_task()
493 * to finish the job and add us back to the active bdi_list
495 wake_up_process(default_backing_dev_info
.wb
.task
);
499 * Add the default flusher task that gets created for any bdi
500 * that has dirty data pending writeout
502 void static bdi_add_default_flusher_task(struct backing_dev_info
*bdi
)
504 if (!bdi_cap_writeback_dirty(bdi
))
507 if (WARN_ON(!test_bit(BDI_registered
, &bdi
->state
))) {
508 printk(KERN_ERR
"bdi %p/%s is not registered!\n",
514 * Check with the helper whether to proceed adding a task. Will only
515 * abort if we two or more simultanous calls to
516 * bdi_add_default_flusher_task() occured, further additions will block
517 * waiting for previous additions to finish.
519 if (!test_and_set_bit(BDI_pending
, &bdi
->state
)) {
520 list_del_rcu(&bdi
->bdi_list
);
523 * We must wait for the current RCU period to end before
524 * moving to the pending list. So schedule that operation
525 * from an RCU callback.
527 call_rcu(&bdi
->rcu_head
, bdi_add_to_pending
);
532 * Remove bdi from bdi_list, and ensure that it is no longer visible
534 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
536 spin_lock_bh(&bdi_lock
);
537 list_del_rcu(&bdi
->bdi_list
);
538 spin_unlock_bh(&bdi_lock
);
543 int bdi_register(struct backing_dev_info
*bdi
, struct device
*parent
,
544 const char *fmt
, ...)
550 if (bdi
->dev
) /* The driver needs to use separate queues per device */
554 dev
= device_create_vargs(bdi_class
, parent
, MKDEV(0, 0), bdi
, fmt
, args
);
561 spin_lock_bh(&bdi_lock
);
562 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
563 spin_unlock_bh(&bdi_lock
);
568 * Just start the forker thread for our default backing_dev_info,
569 * and add other bdi's to the list. They will get a thread created
570 * on-demand when they need it.
572 if (bdi_cap_flush_forker(bdi
)) {
573 struct bdi_writeback
*wb
= &bdi
->wb
;
575 wb
->task
= kthread_run(bdi_forker_task
, wb
, "bdi-%s",
577 if (IS_ERR(wb
->task
)) {
581 bdi_remove_from_list(bdi
);
586 bdi_debug_register(bdi
, dev_name(dev
));
587 set_bit(BDI_registered
, &bdi
->state
);
591 EXPORT_SYMBOL(bdi_register
);
593 int bdi_register_dev(struct backing_dev_info
*bdi
, dev_t dev
)
595 return bdi_register(bdi
, NULL
, "%u:%u", MAJOR(dev
), MINOR(dev
));
597 EXPORT_SYMBOL(bdi_register_dev
);
600 * Remove bdi from the global list and shutdown any threads we have running
602 static void bdi_wb_shutdown(struct backing_dev_info
*bdi
)
604 struct bdi_writeback
*wb
;
606 if (!bdi_cap_writeback_dirty(bdi
))
610 * If setup is pending, wait for that to complete first
612 wait_on_bit(&bdi
->state
, BDI_pending
, bdi_sched_wait
,
613 TASK_UNINTERRUPTIBLE
);
616 * Make sure nobody finds us on the bdi_list anymore
618 bdi_remove_from_list(bdi
);
621 * Finally, kill the kernel threads. We don't need to be RCU
622 * safe anymore, since the bdi is gone from visibility. Force
623 * unfreeze of the thread before calling kthread_stop(), otherwise
624 * it would never exet if it is currently stuck in the refrigerator.
626 list_for_each_entry(wb
, &bdi
->wb_list
, list
) {
627 thaw_process(wb
->task
);
628 kthread_stop(wb
->task
);
633 * This bdi is going away now, make sure that no super_blocks point to it
635 static void bdi_prune_sb(struct backing_dev_info
*bdi
)
637 struct super_block
*sb
;
640 list_for_each_entry(sb
, &super_blocks
, s_list
) {
641 if (sb
->s_bdi
== bdi
)
644 spin_unlock(&sb_lock
);
647 void bdi_unregister(struct backing_dev_info
*bdi
)
652 if (!bdi_cap_flush_forker(bdi
))
653 bdi_wb_shutdown(bdi
);
654 bdi_debug_unregister(bdi
);
655 device_unregister(bdi
->dev
);
659 EXPORT_SYMBOL(bdi_unregister
);
661 int bdi_init(struct backing_dev_info
*bdi
)
668 bdi
->max_ratio
= 100;
669 bdi
->max_prop_frac
= PROP_FRAC_BASE
;
670 spin_lock_init(&bdi
->wb_lock
);
671 INIT_RCU_HEAD(&bdi
->rcu_head
);
672 INIT_LIST_HEAD(&bdi
->bdi_list
);
673 INIT_LIST_HEAD(&bdi
->wb_list
);
674 INIT_LIST_HEAD(&bdi
->work_list
);
676 bdi_wb_init(&bdi
->wb
, bdi
);
679 * Just one thread support for now, hard code mask and count
684 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++) {
685 err
= percpu_counter_init(&bdi
->bdi_stat
[i
], 0);
690 bdi
->dirty_exceeded
= 0;
691 err
= prop_local_init_percpu(&bdi
->completions
);
696 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
701 EXPORT_SYMBOL(bdi_init
);
703 void bdi_destroy(struct backing_dev_info
*bdi
)
708 * Splice our entries to the default_backing_dev_info, if this
711 if (bdi_has_dirty_io(bdi
)) {
712 struct bdi_writeback
*dst
= &default_backing_dev_info
.wb
;
714 spin_lock(&inode_lock
);
715 list_splice(&bdi
->wb
.b_dirty
, &dst
->b_dirty
);
716 list_splice(&bdi
->wb
.b_io
, &dst
->b_io
);
717 list_splice(&bdi
->wb
.b_more_io
, &dst
->b_more_io
);
718 spin_unlock(&inode_lock
);
723 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++)
724 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
726 prop_local_destroy_percpu(&bdi
->completions
);
728 EXPORT_SYMBOL(bdi_destroy
);
731 * For use from filesystems to quickly init and register a bdi associated
732 * with dirty writeback
734 int bdi_setup_and_register(struct backing_dev_info
*bdi
, char *name
,
741 bdi
->capabilities
= cap
;
746 sprintf(tmp
, "%.28s%s", name
, "-%d");
747 err
= bdi_register(bdi
, NULL
, tmp
, atomic_long_inc_return(&bdi_seq
));
755 EXPORT_SYMBOL(bdi_setup_and_register
);
757 static wait_queue_head_t congestion_wqh
[2] = {
758 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
759 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
762 void clear_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
765 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
767 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
768 clear_bit(bit
, &bdi
->state
);
769 smp_mb__after_clear_bit();
770 if (waitqueue_active(wqh
))
773 EXPORT_SYMBOL(clear_bdi_congested
);
775 void set_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
779 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
780 set_bit(bit
, &bdi
->state
);
782 EXPORT_SYMBOL(set_bdi_congested
);
785 * congestion_wait - wait for a backing_dev to become uncongested
786 * @sync: SYNC or ASYNC IO
787 * @timeout: timeout in jiffies
789 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
790 * write congestion. If no backing_devs are congested then just wait for the
791 * next write to be completed.
793 long congestion_wait(int sync
, long timeout
)
797 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
799 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
800 ret
= io_schedule_timeout(timeout
);
801 finish_wait(wqh
, &wait
);
804 EXPORT_SYMBOL(congestion_wait
);