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 noop_backing_dev_info
= {
19 .capabilities
= BDI_CAP_NO_ACCT_AND_WRITEBACK
,
22 static struct class *bdi_class
;
25 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
28 DEFINE_SPINLOCK(bdi_lock
);
31 /* bdi_wq serves all asynchronous writeback tasks */
32 struct workqueue_struct
*bdi_wq
;
34 #ifdef CONFIG_DEBUG_FS
35 #include <linux/debugfs.h>
36 #include <linux/seq_file.h>
38 static struct dentry
*bdi_debug_root
;
40 static void bdi_debug_init(void)
42 bdi_debug_root
= debugfs_create_dir("bdi", NULL
);
45 static int bdi_debug_stats_show(struct seq_file
*m
, void *v
)
47 struct backing_dev_info
*bdi
= m
->private;
48 struct bdi_writeback
*wb
= &bdi
->wb
;
49 unsigned long background_thresh
;
50 unsigned long dirty_thresh
;
51 unsigned long bdi_thresh
;
52 unsigned long nr_dirty
, nr_io
, nr_more_io
;
55 nr_dirty
= nr_io
= nr_more_io
= 0;
56 spin_lock(&wb
->list_lock
);
57 list_for_each_entry(inode
, &wb
->b_dirty
, i_wb_list
)
59 list_for_each_entry(inode
, &wb
->b_io
, i_wb_list
)
61 list_for_each_entry(inode
, &wb
->b_more_io
, i_wb_list
)
63 spin_unlock(&wb
->list_lock
);
65 global_dirty_limits(&background_thresh
, &dirty_thresh
);
66 bdi_thresh
= bdi_dirty_limit(bdi
, dirty_thresh
);
68 #define K(x) ((x) << (PAGE_SHIFT - 10))
70 "BdiWriteback: %10lu kB\n"
71 "BdiReclaimable: %10lu kB\n"
72 "BdiDirtyThresh: %10lu kB\n"
73 "DirtyThresh: %10lu kB\n"
74 "BackgroundThresh: %10lu kB\n"
75 "BdiDirtied: %10lu kB\n"
76 "BdiWritten: %10lu kB\n"
77 "BdiWriteBandwidth: %10lu kBps\n"
83 (unsigned long) K(bdi_stat(bdi
, BDI_WRITEBACK
)),
84 (unsigned long) K(bdi_stat(bdi
, BDI_RECLAIMABLE
)),
88 (unsigned long) K(bdi_stat(bdi
, BDI_DIRTIED
)),
89 (unsigned long) K(bdi_stat(bdi
, BDI_WRITTEN
)),
90 (unsigned long) K(bdi
->write_bandwidth
),
94 !list_empty(&bdi
->bdi_list
), bdi
->state
);
100 static int bdi_debug_stats_open(struct inode
*inode
, struct file
*file
)
102 return single_open(file
, bdi_debug_stats_show
, inode
->i_private
);
105 static const struct file_operations bdi_debug_stats_fops
= {
106 .open
= bdi_debug_stats_open
,
109 .release
= single_release
,
112 static void bdi_debug_register(struct backing_dev_info
*bdi
, const char *name
)
114 bdi
->debug_dir
= debugfs_create_dir(name
, bdi_debug_root
);
115 bdi
->debug_stats
= debugfs_create_file("stats", 0444, bdi
->debug_dir
,
116 bdi
, &bdi_debug_stats_fops
);
119 static void bdi_debug_unregister(struct backing_dev_info
*bdi
)
121 debugfs_remove(bdi
->debug_stats
);
122 debugfs_remove(bdi
->debug_dir
);
125 static inline void bdi_debug_init(void)
128 static inline void bdi_debug_register(struct backing_dev_info
*bdi
,
132 static inline void bdi_debug_unregister(struct backing_dev_info
*bdi
)
137 static ssize_t
read_ahead_kb_store(struct device
*dev
,
138 struct device_attribute
*attr
,
139 const char *buf
, size_t count
)
141 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
142 unsigned long read_ahead_kb
;
145 ret
= kstrtoul(buf
, 10, &read_ahead_kb
);
149 bdi
->ra_pages
= read_ahead_kb
>> (PAGE_SHIFT
- 10);
154 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
156 #define BDI_SHOW(name, expr) \
157 static ssize_t name##_show(struct device *dev, \
158 struct device_attribute *attr, char *page) \
160 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
162 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
164 static DEVICE_ATTR_RW(name);
166 BDI_SHOW(read_ahead_kb
, K(bdi
->ra_pages
))
168 static ssize_t
min_ratio_store(struct device
*dev
,
169 struct device_attribute
*attr
, const char *buf
, size_t count
)
171 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
175 ret
= kstrtouint(buf
, 10, &ratio
);
179 ret
= bdi_set_min_ratio(bdi
, ratio
);
185 BDI_SHOW(min_ratio
, bdi
->min_ratio
)
187 static ssize_t
max_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_max_ratio(bdi
, ratio
);
204 BDI_SHOW(max_ratio
, bdi
->max_ratio
)
206 static ssize_t
stable_pages_required_show(struct device
*dev
,
207 struct device_attribute
*attr
,
210 struct backing_dev_info
*bdi
= dev_get_drvdata(dev
);
212 return snprintf(page
, PAGE_SIZE
-1, "%d\n",
213 bdi_cap_stable_pages_required(bdi
) ? 1 : 0);
215 static DEVICE_ATTR_RO(stable_pages_required
);
217 static struct attribute
*bdi_dev_attrs
[] = {
218 &dev_attr_read_ahead_kb
.attr
,
219 &dev_attr_min_ratio
.attr
,
220 &dev_attr_max_ratio
.attr
,
221 &dev_attr_stable_pages_required
.attr
,
224 ATTRIBUTE_GROUPS(bdi_dev
);
226 static __init
int bdi_class_init(void)
228 bdi_class
= class_create(THIS_MODULE
, "bdi");
229 if (IS_ERR(bdi_class
))
230 return PTR_ERR(bdi_class
);
232 bdi_class
->dev_groups
= bdi_dev_groups
;
236 postcore_initcall(bdi_class_init
);
238 static int __init
default_bdi_init(void)
242 bdi_wq
= alloc_workqueue("writeback", WQ_MEM_RECLAIM
| WQ_FREEZABLE
|
243 WQ_UNBOUND
| WQ_SYSFS
, 0);
247 err
= bdi_init(&noop_backing_dev_info
);
251 subsys_initcall(default_bdi_init
);
253 int bdi_has_dirty_io(struct backing_dev_info
*bdi
)
255 return wb_has_dirty_io(&bdi
->wb
);
259 * This function is used when the first inode for this bdi is marked dirty. It
260 * wakes-up the corresponding bdi thread which should then take care of the
261 * periodic background write-out of dirty inodes. Since the write-out would
262 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
263 * set up a timer which wakes the bdi thread up later.
265 * Note, we wouldn't bother setting up the timer, but this function is on the
266 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
267 * by delaying the wake-up.
269 * We have to be careful not to postpone flush work if it is scheduled for
270 * earlier. Thus we use queue_delayed_work().
272 void bdi_wakeup_thread_delayed(struct backing_dev_info
*bdi
)
274 unsigned long timeout
;
276 timeout
= msecs_to_jiffies(dirty_writeback_interval
* 10);
277 spin_lock_bh(&bdi
->wb_lock
);
278 if (test_bit(BDI_registered
, &bdi
->state
))
279 queue_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, timeout
);
280 spin_unlock_bh(&bdi
->wb_lock
);
284 * Remove bdi from bdi_list, and ensure that it is no longer visible
286 static void bdi_remove_from_list(struct backing_dev_info
*bdi
)
288 spin_lock_bh(&bdi_lock
);
289 list_del_rcu(&bdi
->bdi_list
);
290 spin_unlock_bh(&bdi_lock
);
292 synchronize_rcu_expedited();
295 int bdi_register(struct backing_dev_info
*bdi
, struct device
*parent
,
296 const char *fmt
, ...)
301 if (bdi
->dev
) /* The driver needs to use separate queues per device */
305 dev
= device_create_vargs(bdi_class
, parent
, MKDEV(0, 0), bdi
, fmt
, args
);
312 bdi_debug_register(bdi
, dev_name(dev
));
313 set_bit(BDI_registered
, &bdi
->state
);
315 spin_lock_bh(&bdi_lock
);
316 list_add_tail_rcu(&bdi
->bdi_list
, &bdi_list
);
317 spin_unlock_bh(&bdi_lock
);
319 trace_writeback_bdi_register(bdi
);
322 EXPORT_SYMBOL(bdi_register
);
324 int bdi_register_dev(struct backing_dev_info
*bdi
, dev_t dev
)
326 return bdi_register(bdi
, NULL
, "%u:%u", MAJOR(dev
), MINOR(dev
));
328 EXPORT_SYMBOL(bdi_register_dev
);
331 * Remove bdi from the global list and shutdown any threads we have running
333 static void bdi_wb_shutdown(struct backing_dev_info
*bdi
)
335 /* Make sure nobody queues further work */
336 spin_lock_bh(&bdi
->wb_lock
);
337 if (!test_and_clear_bit(BDI_registered
, &bdi
->state
)) {
338 spin_unlock_bh(&bdi
->wb_lock
);
341 spin_unlock_bh(&bdi
->wb_lock
);
344 * Make sure nobody finds us on the bdi_list anymore
346 bdi_remove_from_list(bdi
);
349 * Drain work list and shutdown the delayed_work. At this point,
350 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
351 * is dying and its work_list needs to be drained no matter what.
353 mod_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, 0);
354 flush_delayed_work(&bdi
->wb
.dwork
);
358 * Called when the device behind @bdi has been removed or ejected.
360 * We can't really do much here except for reducing the dirty ratio at
361 * the moment. In the future we should be able to set a flag so that
362 * the filesystem can handle errors at mark_inode_dirty time instead
363 * of only at writeback time.
365 void bdi_unregister(struct backing_dev_info
*bdi
)
367 if (WARN_ON_ONCE(!bdi
->dev
))
370 bdi_set_min_ratio(bdi
, 0);
372 EXPORT_SYMBOL(bdi_unregister
);
374 static void bdi_wb_init(struct bdi_writeback
*wb
, struct backing_dev_info
*bdi
)
376 memset(wb
, 0, sizeof(*wb
));
379 wb
->last_old_flush
= jiffies
;
380 INIT_LIST_HEAD(&wb
->b_dirty
);
381 INIT_LIST_HEAD(&wb
->b_io
);
382 INIT_LIST_HEAD(&wb
->b_more_io
);
383 spin_lock_init(&wb
->list_lock
);
384 INIT_DELAYED_WORK(&wb
->dwork
, bdi_writeback_workfn
);
388 * Initial write bandwidth: 100 MB/s
390 #define INIT_BW (100 << (20 - PAGE_SHIFT))
392 int bdi_init(struct backing_dev_info
*bdi
)
399 bdi
->max_ratio
= 100;
400 bdi
->max_prop_frac
= FPROP_FRAC_BASE
;
401 spin_lock_init(&bdi
->wb_lock
);
402 INIT_LIST_HEAD(&bdi
->bdi_list
);
403 INIT_LIST_HEAD(&bdi
->work_list
);
405 bdi_wb_init(&bdi
->wb
, bdi
);
407 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++) {
408 err
= percpu_counter_init(&bdi
->bdi_stat
[i
], 0, GFP_KERNEL
);
413 bdi
->dirty_exceeded
= 0;
415 bdi
->bw_time_stamp
= jiffies
;
416 bdi
->written_stamp
= 0;
418 bdi
->balanced_dirty_ratelimit
= INIT_BW
;
419 bdi
->dirty_ratelimit
= INIT_BW
;
420 bdi
->write_bandwidth
= INIT_BW
;
421 bdi
->avg_write_bandwidth
= INIT_BW
;
423 err
= fprop_local_init_percpu(&bdi
->completions
, GFP_KERNEL
);
428 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
433 EXPORT_SYMBOL(bdi_init
);
435 void bdi_destroy(struct backing_dev_info
*bdi
)
439 bdi_wb_shutdown(bdi
);
441 WARN_ON(!list_empty(&bdi
->work_list
));
442 WARN_ON(delayed_work_pending(&bdi
->wb
.dwork
));
445 bdi_debug_unregister(bdi
);
446 device_unregister(bdi
->dev
);
450 for (i
= 0; i
< NR_BDI_STAT_ITEMS
; i
++)
451 percpu_counter_destroy(&bdi
->bdi_stat
[i
]);
452 fprop_local_destroy_percpu(&bdi
->completions
);
454 EXPORT_SYMBOL(bdi_destroy
);
457 * For use from filesystems to quickly init and register a bdi associated
458 * with dirty writeback
460 int bdi_setup_and_register(struct backing_dev_info
*bdi
, char *name
)
465 bdi
->capabilities
= 0;
470 err
= bdi_register(bdi
, NULL
, "%.28s-%ld", name
,
471 atomic_long_inc_return(&bdi_seq
));
479 EXPORT_SYMBOL(bdi_setup_and_register
);
481 static wait_queue_head_t congestion_wqh
[2] = {
482 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[0]),
483 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh
[1])
485 static atomic_t nr_bdi_congested
[2];
487 void clear_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
490 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
492 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
493 if (test_and_clear_bit(bit
, &bdi
->state
))
494 atomic_dec(&nr_bdi_congested
[sync
]);
495 smp_mb__after_atomic();
496 if (waitqueue_active(wqh
))
499 EXPORT_SYMBOL(clear_bdi_congested
);
501 void set_bdi_congested(struct backing_dev_info
*bdi
, int sync
)
505 bit
= sync
? BDI_sync_congested
: BDI_async_congested
;
506 if (!test_and_set_bit(bit
, &bdi
->state
))
507 atomic_inc(&nr_bdi_congested
[sync
]);
509 EXPORT_SYMBOL(set_bdi_congested
);
512 * congestion_wait - wait for a backing_dev to become uncongested
513 * @sync: SYNC or ASYNC IO
514 * @timeout: timeout in jiffies
516 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
517 * write congestion. If no backing_devs are congested then just wait for the
518 * next write to be completed.
520 long congestion_wait(int sync
, long timeout
)
523 unsigned long start
= jiffies
;
525 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
527 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
528 ret
= io_schedule_timeout(timeout
);
529 finish_wait(wqh
, &wait
);
531 trace_writeback_congestion_wait(jiffies_to_usecs(timeout
),
532 jiffies_to_usecs(jiffies
- start
));
536 EXPORT_SYMBOL(congestion_wait
);
539 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
540 * @zone: A zone to check if it is heavily congested
541 * @sync: SYNC or ASYNC IO
542 * @timeout: timeout in jiffies
544 * In the event of a congested backing_dev (any backing_dev) and the given
545 * @zone has experienced recent congestion, this waits for up to @timeout
546 * jiffies for either a BDI to exit congestion of the given @sync queue
547 * or a write to complete.
549 * In the absence of zone congestion, cond_resched() is called to yield
550 * the processor if necessary but otherwise does not sleep.
552 * The return value is 0 if the sleep is for the full timeout. Otherwise,
553 * it is the number of jiffies that were still remaining when the function
554 * returned. return_value == timeout implies the function did not sleep.
556 long wait_iff_congested(struct zone
*zone
, int sync
, long timeout
)
559 unsigned long start
= jiffies
;
561 wait_queue_head_t
*wqh
= &congestion_wqh
[sync
];
564 * If there is no congestion, or heavy congestion is not being
565 * encountered in the current zone, yield if necessary instead
566 * of sleeping on the congestion queue
568 if (atomic_read(&nr_bdi_congested
[sync
]) == 0 ||
569 !test_bit(ZONE_CONGESTED
, &zone
->flags
)) {
572 /* In case we scheduled, work out time remaining */
573 ret
= timeout
- (jiffies
- start
);
580 /* Sleep until uncongested or a write happens */
581 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
582 ret
= io_schedule_timeout(timeout
);
583 finish_wait(wqh
, &wait
);
586 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout
),
587 jiffies_to_usecs(jiffies
- start
));
591 EXPORT_SYMBOL(wait_iff_congested
);
593 int pdflush_proc_obsolete(struct ctl_table
*table
, int write
,
594 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
598 if (*ppos
|| *lenp
< sizeof(kbuf
)) {
603 if (copy_to_user(buffer
, kbuf
, sizeof(kbuf
)))
605 printk_once(KERN_WARNING
"%s exported in /proc is scheduled for removal\n",