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 struct backing_dev_info default_backing_dev_info
= {
19 .ra_pages
= VM_MAX_READAHEAD
* 1024 / PAGE_CACHE_SIZE
,
21 .capabilities
= BDI_CAP_MAP_COPY
,
23 EXPORT_SYMBOL_GPL(default_backing_dev_info
);
25 struct backing_dev_info noop_backing_dev_info
= {
27 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
29 EXPORT_SYMBOL_GPL(noop_backing_dev_info
);
31 static struct class *bdi_class
;
34 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
37 DEFINE_SPINLOCK(bdi_lock
);
40 /* bdi_wq serves all asynchronous writeback tasks */
41 struct workqueue_struct
*bdi_wq
;
43 void bdi_lock_two(struct bdi_writeback
*wb1
, struct bdi_writeback
*wb2
)
46 spin_lock(&wb1
->list_lock
);
47 spin_lock_nested(&wb2
->list_lock
, 1);
49 spin_lock(&wb2
->list_lock
);
50 spin_lock_nested(&wb1
->list_lock
, 1);
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
;
75 nr_dirty
= nr_io
= nr_more_io
= 0;
76 spin_lock(&wb
->list_lock
);
77 list_for_each_entry(inode
, &wb
->b_dirty
, i_wb_list
)
79 list_for_each_entry(inode
, &wb
->b_io
, i_wb_list
)
81 list_for_each_entry(inode
, &wb
->b_more_io
, i_wb_list
)
83 spin_unlock(&wb
->list_lock
);
85 global_dirty_limits(&background_thresh
, &dirty_thresh
);
86 bdi_thresh
= bdi_dirty_limit(bdi
, dirty_thresh
);
88 #define K(x) ((x) << (PAGE_SHIFT - 10))
90 "BdiWriteback: %10lu kB\n"
91 "BdiReclaimable: %10lu kB\n"
92 "BdiDirtyThresh: %10lu kB\n"
93 "DirtyThresh: %10lu kB\n"
94 "BackgroundThresh: %10lu kB\n"
95 "BdiDirtied: %10lu kB\n"
96 "BdiWritten: %10lu kB\n"
97 "BdiWriteBandwidth: %10lu kBps\n"
103 (unsigned long) K(bdi_stat(bdi
, BDI_WRITEBACK
)),
104 (unsigned long) K(bdi_stat(bdi
, BDI_RECLAIMABLE
)),
107 K(background_thresh
),
108 (unsigned long) K(bdi_stat(bdi
, BDI_DIRTIED
)),
109 (unsigned long) K(bdi_stat(bdi
, BDI_WRITTEN
)),
110 (unsigned long) K(bdi
->write_bandwidth
),
114 !list_empty(&bdi
->bdi_list
), bdi
->state
);
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
);
162 unsigned long read_ahead_kb
;
165 ret
= kstrtoul(buf
, 10, &read_ahead_kb
);
169 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
);
194 ret
= kstrtouint(buf
, 10, &ratio
);
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
);
213 ret
= kstrtouint(buf
, 10, &ratio
);
217 ret
= bdi_set_max_ratio(bdi
, ratio
);
223 BDI_SHOW(max_ratio
, bdi
->max_ratio
)
225 static ssize_t
stable_pages_required_show(struct device
*dev
,
226 struct device_attribute
*attr
,
229 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
231 return snprintf(page
, PAGE_SIZE
-1, "%d\n",
232 bdi_cap_stable_pages_required(bdi
) ? 1 : 0);
235 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
237 static struct device_attribute bdi_dev_attrs
[] = {
238 __ATTR_RW(read_ahead_kb
),
239 __ATTR_RW(min_ratio
),
240 __ATTR_RW(max_ratio
),
241 __ATTR_RO(stable_pages_required
),
245 static __init
int bdi_class_init(void)
247 bdi_class
= class_create(THIS_MODULE
, "bdi");
248 if (IS_ERR(bdi_class
))
249 return PTR_ERR(bdi_class
);
251 bdi_class
->dev_attrs
= bdi_dev_attrs
;
255 postcore_initcall(bdi_class_init
);
257 static int __init
default_bdi_init(void)
261 bdi_wq
= alloc_workqueue("writeback", WQ_MEM_RECLAIM
| WQ_FREEZABLE
|
262 WQ_UNBOUND
| WQ_SYSFS
, 0);
266 err
= bdi_init(&default_backing_dev_info
);
268 bdi_register(&default_backing_dev_info
, NULL
, "default");
269 err
= bdi_init(&noop_backing_dev_info
);
273 subsys_initcall(default_bdi_init
);
275 int bdi_has_dirty_io(struct backing_dev_info
*bdi
)
277 return wb_has_dirty_io(&bdi
->wb
);
281 * This function is used when the first inode for this bdi is marked dirty. It
282 * wakes-up the corresponding bdi thread which should then take care of the
283 * periodic background write-out of dirty inodes. Since the write-out would
284 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
285 * set up a timer which wakes the bdi thread up later.
287 * Note, we wouldn't bother setting up the timer, but this function is on the
288 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
289 * by delaying the wake-up.
291 void bdi_wakeup_thread_delayed(struct backing_dev_info
*bdi
)
293 unsigned long timeout
;
295 timeout
= msecs_to_jiffies(dirty_writeback_interval
* 10);
296 mod_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, timeout
);
300 * Remove bdi from bdi_list, and ensure that it is no longer visible
302 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
304 spin_lock_bh(&bdi_lock
);
305 list_del_rcu(&bdi
->bdi_list
);
306 spin_unlock_bh(&bdi_lock
);
308 synchronize_rcu_expedited();
310 /* bdi_list is now unused, clear it to mark @bdi dying */
311 INIT_LIST_HEAD(&bdi
->bdi_list
);
314 int bdi_register(struct backing_dev_info
*bdi
, struct device
*parent
,
315 const char *fmt
, ...)
320 if (bdi
->dev
) /* The driver needs to use separate queues per device */
324 dev
= device_create_vargs(bdi_class
, parent
, MKDEV(0, 0), bdi
, fmt
, args
);
331 bdi_debug_register(bdi
, dev_name(dev
));
332 set_bit(BDI_registered
, &bdi
->state
);
334 spin_lock_bh(&bdi_lock
);
335 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
336 spin_unlock_bh(&bdi_lock
);
338 trace_writeback_bdi_register(bdi
);
341 EXPORT_SYMBOL(bdi_register
);
343 int bdi_register_dev(struct backing_dev_info
*bdi
, dev_t dev
)
345 return bdi_register(bdi
, NULL
, "%u:%u", MAJOR(dev
), MINOR(dev
));
347 EXPORT_SYMBOL(bdi_register_dev
);
350 * Remove bdi from the global list and shutdown any threads we have running
352 static void bdi_wb_shutdown(struct backing_dev_info
*bdi
)
354 if (!bdi_cap_writeback_dirty(bdi
))
358 * Make sure nobody finds us on the bdi_list anymore
360 bdi_remove_from_list(bdi
);
363 * Drain work list and shutdown the delayed_work. At this point,
364 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
365 * is dying and its work_list needs to be drained no matter what.
367 mod_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, 0);
368 flush_delayed_work(&bdi
->wb
.dwork
);
369 WARN_ON(!list_empty(&bdi
->work_list
));
372 * This shouldn't be necessary unless @bdi for some reason has
373 * unflushed dirty IO after work_list is drained. Do it anyway
376 cancel_delayed_work_sync(&bdi
->wb
.dwork
);
380 * This bdi is going away now, make sure that no super_blocks point to it
382 static void bdi_prune_sb(struct backing_dev_info
*bdi
)
384 struct super_block
*sb
;
387 list_for_each_entry(sb
, &super_blocks
, s_list
) {
388 if (sb
->s_bdi
== bdi
)
389 sb
->s_bdi
= &default_backing_dev_info
;
391 spin_unlock(&sb_lock
);
394 void bdi_unregister(struct backing_dev_info
*bdi
)
396 struct device
*dev
= bdi
->dev
;
399 bdi_set_min_ratio(bdi
, 0);
400 trace_writeback_bdi_unregister(bdi
);
403 bdi_wb_shutdown(bdi
);
404 bdi_debug_unregister(bdi
);
406 spin_lock_bh(&bdi
->wb_lock
);
408 spin_unlock_bh(&bdi
->wb_lock
);
410 device_unregister(dev
);
413 EXPORT_SYMBOL(bdi_unregister
);
415 static void bdi_wb_init(struct bdi_writeback
*wb
, struct backing_dev_info
*bdi
)
417 memset(wb
, 0, sizeof(*wb
));
420 wb
->last_old_flush
= jiffies
;
421 INIT_LIST_HEAD(&wb
->b_dirty
);
422 INIT_LIST_HEAD(&wb
->b_io
);
423 INIT_LIST_HEAD(&wb
->b_more_io
);
424 spin_lock_init(&wb
->list_lock
);
425 INIT_DELAYED_WORK(&wb
->dwork
, bdi_writeback_workfn
);
429 * Initial write bandwidth: 100 MB/s
431 #define INIT_BW (100 << (20 - PAGE_SHIFT))
433 int bdi_init(struct backing_dev_info
*bdi
)
440 bdi
->max_ratio
= 100;
441 bdi
->max_prop_frac
= FPROP_FRAC_BASE
;
442 spin_lock_init(&bdi
->wb_lock
);
443 INIT_LIST_HEAD(&bdi
->bdi_list
);
444 INIT_LIST_HEAD(&bdi
->work_list
);
446 bdi_wb_init(&bdi
->wb
, bdi
);
448 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++) {
449 err
= percpu_counter_init(&bdi
->bdi_stat
[i
], 0);
454 bdi
->dirty_exceeded
= 0;
456 bdi
->bw_time_stamp
= jiffies
;
457 bdi
->written_stamp
= 0;
459 bdi
->balanced_dirty_ratelimit
= INIT_BW
;
460 bdi
->dirty_ratelimit
= INIT_BW
;
461 bdi
->write_bandwidth
= INIT_BW
;
462 bdi
->avg_write_bandwidth
= INIT_BW
;
464 err
= fprop_local_init_percpu(&bdi
->completions
);
469 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
474 EXPORT_SYMBOL(bdi_init
);
476 void bdi_destroy(struct backing_dev_info
*bdi
)
481 * Splice our entries to the default_backing_dev_info, if this
484 if (bdi_has_dirty_io(bdi
)) {
485 struct bdi_writeback
*dst
= &default_backing_dev_info
.wb
;
487 bdi_lock_two(&bdi
->wb
, dst
);
488 list_splice(&bdi
->wb
.b_dirty
, &dst
->b_dirty
);
489 list_splice(&bdi
->wb
.b_io
, &dst
->b_io
);
490 list_splice(&bdi
->wb
.b_more_io
, &dst
->b_more_io
);
491 spin_unlock(&bdi
->wb
.list_lock
);
492 spin_unlock(&dst
->list_lock
);
498 * If bdi_unregister() had already been called earlier, the dwork
499 * could still be pending because bdi_prune_sb() can race with the
500 * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty().
502 cancel_delayed_work_sync(&bdi
->wb
.dwork
);
504 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++)
505 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
507 fprop_local_destroy_percpu(&bdi
->completions
);
509 EXPORT_SYMBOL(bdi_destroy
);
512 * For use from filesystems to quickly init and register a bdi associated
513 * with dirty writeback
515 int bdi_setup_and_register(struct backing_dev_info
*bdi
, char *name
,
521 bdi
->capabilities
= cap
;
526 err
= bdi_register(bdi
, NULL
, "%.28s-%ld", name
,
527 atomic_long_inc_return(&bdi_seq
));
535 EXPORT_SYMBOL(bdi_setup_and_register
);
537 static wait_queue_head_t congestion_wqh
[2] = {
538 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
539 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
541 static atomic_t nr_bdi_congested
[2];
543 void clear_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
546 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
548 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
549 if (test_and_clear_bit(bit
, &bdi
->state
))
550 atomic_dec(&nr_bdi_congested
[sync
]);
551 smp_mb__after_clear_bit();
552 if (waitqueue_active(wqh
))
555 EXPORT_SYMBOL(clear_bdi_congested
);
557 void set_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
561 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
562 if (!test_and_set_bit(bit
, &bdi
->state
))
563 atomic_inc(&nr_bdi_congested
[sync
]);
565 EXPORT_SYMBOL(set_bdi_congested
);
568 * congestion_wait - wait for a backing_dev to become uncongested
569 * @sync: SYNC or ASYNC IO
570 * @timeout: timeout in jiffies
572 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
573 * write congestion. If no backing_devs are congested then just wait for the
574 * next write to be completed.
576 long congestion_wait(int sync
, long timeout
)
579 unsigned long start
= jiffies
;
581 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
583 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
584 ret
= io_schedule_timeout(timeout
);
585 finish_wait(wqh
, &wait
);
587 trace_writeback_congestion_wait(jiffies_to_usecs(timeout
),
588 jiffies_to_usecs(jiffies
- start
));
592 EXPORT_SYMBOL(congestion_wait
);
595 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
596 * @zone: A zone to check if it is heavily congested
597 * @sync: SYNC or ASYNC IO
598 * @timeout: timeout in jiffies
600 * In the event of a congested backing_dev (any backing_dev) and the given
601 * @zone has experienced recent congestion, this waits for up to @timeout
602 * jiffies for either a BDI to exit congestion of the given @sync queue
603 * or a write to complete.
605 * In the absence of zone congestion, cond_resched() is called to yield
606 * the processor if necessary but otherwise does not sleep.
608 * The return value is 0 if the sleep is for the full timeout. Otherwise,
609 * it is the number of jiffies that were still remaining when the function
610 * returned. return_value == timeout implies the function did not sleep.
612 long wait_iff_congested(struct zone
*zone
, int sync
, long timeout
)
615 unsigned long start
= jiffies
;
617 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
620 * If there is no congestion, or heavy congestion is not being
621 * encountered in the current zone, yield if necessary instead
622 * of sleeping on the congestion queue
624 if (atomic_read(&nr_bdi_congested
[sync
]) == 0 ||
625 !zone_is_reclaim_congested(zone
)) {
628 /* In case we scheduled, work out time remaining */
629 ret
= timeout
- (jiffies
- start
);
636 /* Sleep until uncongested or a write happens */
637 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
638 ret
= io_schedule_timeout(timeout
);
639 finish_wait(wqh
, &wait
);
642 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout
),
643 jiffies_to_usecs(jiffies
- start
));
647 EXPORT_SYMBOL(wait_iff_congested
);
649 int pdflush_proc_obsolete(struct ctl_table
*table
, int write
,
650 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
659 if (copy_to_user(buffer
, kbuf
, sizeof(kbuf
)))
661 printk_once(KERN_WARNING
"%s exported in /proc is scheduled for removal\n",