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 static void bdi_wb_init(struct bdi_writeback
*wb
, struct backing_dev_info
*bdi
)
253 memset(wb
, 0, sizeof(*wb
));
256 wb
->last_old_flush
= jiffies
;
257 INIT_LIST_HEAD(&wb
->b_dirty
);
258 INIT_LIST_HEAD(&wb
->b_io
);
259 INIT_LIST_HEAD(&wb
->b_more_io
);
262 int bdi_has_dirty_io(struct backing_dev_info
*bdi
)
264 return wb_has_dirty_io(&bdi
->wb
);
267 static void bdi_flush_io(struct backing_dev_info
*bdi
)
269 struct writeback_control wbc
= {
270 .sync_mode
= WB_SYNC_NONE
,
271 .older_than_this
= NULL
,
276 writeback_inodes_wb(&bdi
->wb
, &wbc
);
280 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
281 * or we risk deadlocking on ->s_umount. The longer term solution would be
282 * to implement sync_supers_bdi() or similar and simply do it from the
283 * bdi writeback thread individually.
285 static int bdi_sync_supers(void *unused
)
287 set_user_nice(current
, 0);
289 while (!kthread_should_stop()) {
290 set_current_state(TASK_INTERRUPTIBLE
);
294 * Do this periodically, like kupdated() did before.
302 void bdi_arm_supers_timer(void)
306 if (!dirty_writeback_interval
)
309 next
= msecs_to_jiffies(dirty_writeback_interval
* 10) + jiffies
;
310 mod_timer(&sync_supers_timer
, round_jiffies_up(next
));
313 static void sync_supers_timer_fn(unsigned long unused
)
315 wake_up_process(sync_supers_tsk
);
316 bdi_arm_supers_timer();
319 static int bdi_forker_thread(void *ptr
)
321 struct bdi_writeback
*me
= ptr
;
323 current
->flags
|= PF_FLUSHER
| PF_SWAPWRITE
;
327 * Our parent may run at a different priority, just set us to normal
329 set_user_nice(current
, 0);
333 struct task_struct
*task
;
334 struct backing_dev_info
*bdi
, *tmp
;
337 * Temporary measure, we want to make sure we don't see
338 * dirty data on the default backing_dev_info
340 if (wb_has_dirty_io(me
) || !list_empty(&me
->bdi
->work_list
))
341 wb_do_writeback(me
, 0);
343 spin_lock_bh(&bdi_lock
);
344 set_current_state(TASK_INTERRUPTIBLE
);
347 * Check if any existing bdi's have dirty data without
348 * a thread registered. If so, set that up.
350 list_for_each_entry_safe(bdi
, tmp
, &bdi_list
, bdi_list
) {
351 if (!bdi_cap_writeback_dirty(bdi
))
355 if (list_empty(&bdi
->work_list
) &&
356 !bdi_has_dirty_io(bdi
))
359 WARN(!test_bit(BDI_registered
, &bdi
->state
),
360 "bdi %p/%s is not registered!\n", bdi
, bdi
->name
);
362 list_del_rcu(&bdi
->bdi_list
);
366 spin_unlock_bh(&bdi_lock
);
368 /* Keep working if default bdi still has things to do */
369 if (!list_empty(&me
->bdi
->work_list
))
370 __set_current_state(TASK_RUNNING
);
375 wait
= msecs_to_jiffies(dirty_writeback_interval
* 10);
377 schedule_timeout(wait
);
384 __set_current_state(TASK_RUNNING
);
387 * Set the pending bit - if someone will try to unregister this
388 * bdi - it'll wait on this bit.
390 set_bit(BDI_pending
, &bdi
->state
);
392 /* Make sure no one uses the picked bdi */
395 task
= kthread_run(bdi_writeback_thread
, &bdi
->wb
, "flush-%s",
399 * If thread creation fails, then readd the bdi back to
400 * the list and force writeout of the bdi from this
401 * forker thread. That will free some memory and we can
402 * try again. Add it to the tail so we get a chance to
403 * flush other bdi's to free memory.
405 spin_lock_bh(&bdi_lock
);
406 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
407 spin_unlock_bh(&bdi_lock
);
418 * Remove bdi from bdi_list, and ensure that it is no longer visible
420 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
422 spin_lock_bh(&bdi_lock
);
423 list_del_rcu(&bdi
->bdi_list
);
424 spin_unlock_bh(&bdi_lock
);
429 int bdi_register(struct backing_dev_info
*bdi
, struct device
*parent
,
430 const char *fmt
, ...)
436 if (bdi
->dev
) /* The driver needs to use separate queues per device */
440 dev
= device_create_vargs(bdi_class
, parent
, MKDEV(0, 0), bdi
, fmt
, args
);
447 spin_lock_bh(&bdi_lock
);
448 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
449 spin_unlock_bh(&bdi_lock
);
454 * Just start the forker thread for our default backing_dev_info,
455 * and add other bdi's to the list. They will get a thread created
456 * on-demand when they need it.
458 if (bdi_cap_flush_forker(bdi
)) {
459 struct bdi_writeback
*wb
= &bdi
->wb
;
461 wb
->task
= kthread_run(bdi_forker_thread
, wb
, "bdi-%s",
463 if (IS_ERR(wb
->task
)) {
467 bdi_remove_from_list(bdi
);
472 bdi_debug_register(bdi
, dev_name(dev
));
473 set_bit(BDI_registered
, &bdi
->state
);
474 trace_writeback_bdi_register(bdi
);
478 EXPORT_SYMBOL(bdi_register
);
480 int bdi_register_dev(struct backing_dev_info
*bdi
, dev_t dev
)
482 return bdi_register(bdi
, NULL
, "%u:%u", MAJOR(dev
), MINOR(dev
));
484 EXPORT_SYMBOL(bdi_register_dev
);
487 * Remove bdi from the global list and shutdown any threads we have running
489 static void bdi_wb_shutdown(struct backing_dev_info
*bdi
)
491 if (!bdi_cap_writeback_dirty(bdi
))
495 * If setup is pending, wait for that to complete first
497 wait_on_bit(&bdi
->state
, BDI_pending
, bdi_sched_wait
,
498 TASK_UNINTERRUPTIBLE
);
501 * Make sure nobody finds us on the bdi_list anymore
503 bdi_remove_from_list(bdi
);
506 * Finally, kill the kernel thread. We don't need to be RCU
507 * safe anymore, since the bdi is gone from visibility. Force
508 * unfreeze of the thread before calling kthread_stop(), otherwise
509 * it would never exet if it is currently stuck in the refrigerator.
512 thaw_process(bdi
->wb
.task
);
513 kthread_stop(bdi
->wb
.task
);
518 * This bdi is going away now, make sure that no super_blocks point to it
520 static void bdi_prune_sb(struct backing_dev_info
*bdi
)
522 struct super_block
*sb
;
525 list_for_each_entry(sb
, &super_blocks
, s_list
) {
526 if (sb
->s_bdi
== bdi
)
529 spin_unlock(&sb_lock
);
532 void bdi_unregister(struct backing_dev_info
*bdi
)
535 trace_writeback_bdi_unregister(bdi
);
538 if (!bdi_cap_flush_forker(bdi
))
539 bdi_wb_shutdown(bdi
);
540 bdi_debug_unregister(bdi
);
541 device_unregister(bdi
->dev
);
545 EXPORT_SYMBOL(bdi_unregister
);
547 int bdi_init(struct backing_dev_info
*bdi
)
554 bdi
->max_ratio
= 100;
555 bdi
->max_prop_frac
= PROP_FRAC_BASE
;
556 spin_lock_init(&bdi
->wb_lock
);
557 INIT_LIST_HEAD(&bdi
->bdi_list
);
558 INIT_LIST_HEAD(&bdi
->work_list
);
560 bdi_wb_init(&bdi
->wb
, bdi
);
562 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++) {
563 err
= percpu_counter_init(&bdi
->bdi_stat
[i
], 0);
568 bdi
->dirty_exceeded
= 0;
569 err
= prop_local_init_percpu(&bdi
->completions
);
574 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
579 EXPORT_SYMBOL(bdi_init
);
581 void bdi_destroy(struct backing_dev_info
*bdi
)
586 * Splice our entries to the default_backing_dev_info, if this
589 if (bdi_has_dirty_io(bdi
)) {
590 struct bdi_writeback
*dst
= &default_backing_dev_info
.wb
;
592 spin_lock(&inode_lock
);
593 list_splice(&bdi
->wb
.b_dirty
, &dst
->b_dirty
);
594 list_splice(&bdi
->wb
.b_io
, &dst
->b_io
);
595 list_splice(&bdi
->wb
.b_more_io
, &dst
->b_more_io
);
596 spin_unlock(&inode_lock
);
601 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++)
602 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
604 prop_local_destroy_percpu(&bdi
->completions
);
606 EXPORT_SYMBOL(bdi_destroy
);
609 * For use from filesystems to quickly init and register a bdi associated
610 * with dirty writeback
612 int bdi_setup_and_register(struct backing_dev_info
*bdi
, char *name
,
619 bdi
->capabilities
= cap
;
624 sprintf(tmp
, "%.28s%s", name
, "-%d");
625 err
= bdi_register(bdi
, NULL
, tmp
, atomic_long_inc_return(&bdi_seq
));
633 EXPORT_SYMBOL(bdi_setup_and_register
);
635 static wait_queue_head_t congestion_wqh
[2] = {
636 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
637 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
640 void clear_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
643 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
645 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
646 clear_bit(bit
, &bdi
->state
);
647 smp_mb__after_clear_bit();
648 if (waitqueue_active(wqh
))
651 EXPORT_SYMBOL(clear_bdi_congested
);
653 void set_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
657 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
658 set_bit(bit
, &bdi
->state
);
660 EXPORT_SYMBOL(set_bdi_congested
);
663 * congestion_wait - wait for a backing_dev to become uncongested
664 * @sync: SYNC or ASYNC IO
665 * @timeout: timeout in jiffies
667 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
668 * write congestion. If no backing_devs are congested then just wait for the
669 * next write to be completed.
671 long congestion_wait(int sync
, long timeout
)
675 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
677 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
678 ret
= io_schedule_timeout(timeout
);
679 finish_wait(wqh
, &wait
);
682 EXPORT_SYMBOL(congestion_wait
);