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, nr_dirty_time;
55 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 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 list_for_each_entry(inode, &wb->b_dirty_time, i_wb_list)
64 if (inode->i_state & I_DIRTY_TIME)
66 spin_unlock(&wb->list_lock);
68 global_dirty_limits(&background_thresh, &dirty_thresh);
69 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
71 #define K(x) ((x) << (PAGE_SHIFT - 10))
73 "BdiWriteback: %10lu kB\n"
74 "BdiReclaimable: %10lu kB\n"
75 "BdiDirtyThresh: %10lu kB\n"
76 "DirtyThresh: %10lu kB\n"
77 "BackgroundThresh: %10lu kB\n"
78 "BdiDirtied: %10lu kB\n"
79 "BdiWritten: %10lu kB\n"
80 "BdiWriteBandwidth: %10lu kBps\n"
84 "b_dirty_time: %10lu\n"
87 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
88 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
92 (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
93 (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
94 (unsigned long) K(bdi->write_bandwidth),
99 !list_empty(&bdi->bdi_list), bdi->state);
105 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
107 return single_open(file, bdi_debug_stats_show, inode->i_private);
110 static const struct file_operations bdi_debug_stats_fops = {
111 .open = bdi_debug_stats_open,
114 .release = single_release,
117 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
119 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
120 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
121 bdi, &bdi_debug_stats_fops);
124 static void bdi_debug_unregister(struct backing_dev_info *bdi)
126 debugfs_remove(bdi->debug_stats);
127 debugfs_remove(bdi->debug_dir);
130 static inline void bdi_debug_init(void)
133 static inline void bdi_debug_register(struct backing_dev_info *bdi,
137 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
142 static ssize_t read_ahead_kb_store(struct device *dev,
143 struct device_attribute *attr,
144 const char *buf, size_t count)
146 struct backing_dev_info *bdi = dev_get_drvdata(dev);
147 unsigned long read_ahead_kb;
150 ret = kstrtoul(buf, 10, &read_ahead_kb);
154 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
159 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
161 #define BDI_SHOW(name, expr) \
162 static ssize_t name##_show(struct device *dev, \
163 struct device_attribute *attr, char *page) \
165 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
167 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
169 static DEVICE_ATTR_RW(name);
171 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
173 static ssize_t min_ratio_store(struct device *dev,
174 struct device_attribute *attr, const char *buf, size_t count)
176 struct backing_dev_info *bdi = dev_get_drvdata(dev);
180 ret = kstrtouint(buf, 10, &ratio);
184 ret = bdi_set_min_ratio(bdi, ratio);
190 BDI_SHOW(min_ratio, bdi->min_ratio)
192 static ssize_t max_ratio_store(struct device *dev,
193 struct device_attribute *attr, const char *buf, size_t count)
195 struct backing_dev_info *bdi = dev_get_drvdata(dev);
199 ret = kstrtouint(buf, 10, &ratio);
203 ret = bdi_set_max_ratio(bdi, ratio);
209 BDI_SHOW(max_ratio, bdi->max_ratio)
211 static ssize_t stable_pages_required_show(struct device *dev,
212 struct device_attribute *attr,
215 struct backing_dev_info *bdi = dev_get_drvdata(dev);
217 return snprintf(page, PAGE_SIZE-1, "%d\n",
218 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
220 static DEVICE_ATTR_RO(stable_pages_required);
222 static struct attribute *bdi_dev_attrs[] = {
223 &dev_attr_read_ahead_kb.attr,
224 &dev_attr_min_ratio.attr,
225 &dev_attr_max_ratio.attr,
226 &dev_attr_stable_pages_required.attr,
229 ATTRIBUTE_GROUPS(bdi_dev);
231 static __init int bdi_class_init(void)
233 bdi_class = class_create(THIS_MODULE, "bdi");
234 if (IS_ERR(bdi_class))
235 return PTR_ERR(bdi_class);
237 bdi_class->dev_groups = bdi_dev_groups;
241 postcore_initcall(bdi_class_init);
243 static int __init default_bdi_init(void)
247 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
248 WQ_UNBOUND | WQ_SYSFS, 0);
252 err = bdi_init(&noop_backing_dev_info);
256 subsys_initcall(default_bdi_init);
258 int bdi_has_dirty_io(struct backing_dev_info *bdi)
260 return wb_has_dirty_io(&bdi->wb);
264 * This function is used when the first inode for this bdi is marked dirty. It
265 * wakes-up the corresponding bdi thread which should then take care of the
266 * periodic background write-out of dirty inodes. Since the write-out would
267 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
268 * set up a timer which wakes the bdi thread up later.
270 * Note, we wouldn't bother setting up the timer, but this function is on the
271 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
272 * by delaying the wake-up.
274 * We have to be careful not to postpone flush work if it is scheduled for
275 * earlier. Thus we use queue_delayed_work().
277 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
279 unsigned long timeout;
281 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
282 spin_lock_bh(&bdi->wb_lock);
283 if (test_bit(BDI_registered, &bdi->state))
284 queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
285 spin_unlock_bh(&bdi->wb_lock);
289 * Remove bdi from bdi_list, and ensure that it is no longer visible
291 static void bdi_remove_from_list(struct backing_dev_info *bdi)
293 spin_lock_bh(&bdi_lock);
294 list_del_rcu(&bdi->bdi_list);
295 spin_unlock_bh(&bdi_lock);
297 synchronize_rcu_expedited();
300 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
301 const char *fmt, ...)
306 if (bdi->dev) /* The driver needs to use separate queues per device */
310 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
317 bdi_debug_register(bdi, dev_name(dev));
318 set_bit(BDI_registered, &bdi->state);
320 spin_lock_bh(&bdi_lock);
321 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
322 spin_unlock_bh(&bdi_lock);
324 trace_writeback_bdi_register(bdi);
327 EXPORT_SYMBOL(bdi_register);
329 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
331 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
333 EXPORT_SYMBOL(bdi_register_dev);
336 * Remove bdi from the global list and shutdown any threads we have running
338 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
340 /* Make sure nobody queues further work */
341 spin_lock_bh(&bdi->wb_lock);
342 if (!test_and_clear_bit(BDI_registered, &bdi->state)) {
343 spin_unlock_bh(&bdi->wb_lock);
346 spin_unlock_bh(&bdi->wb_lock);
349 * Make sure nobody finds us on the bdi_list anymore
351 bdi_remove_from_list(bdi);
354 * Drain work list and shutdown the delayed_work. At this point,
355 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
356 * is dying and its work_list needs to be drained no matter what.
358 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
359 flush_delayed_work(&bdi->wb.dwork);
363 * Called when the device behind @bdi has been removed or ejected.
365 * We can't really do much here except for reducing the dirty ratio at
366 * the moment. In the future we should be able to set a flag so that
367 * the filesystem can handle errors at mark_inode_dirty time instead
368 * of only at writeback time.
370 void bdi_unregister(struct backing_dev_info *bdi)
372 if (WARN_ON_ONCE(!bdi->dev))
375 bdi_set_min_ratio(bdi, 0);
377 EXPORT_SYMBOL(bdi_unregister);
379 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
381 memset(wb, 0, sizeof(*wb));
384 wb->last_old_flush = jiffies;
385 INIT_LIST_HEAD(&wb->b_dirty);
386 INIT_LIST_HEAD(&wb->b_io);
387 INIT_LIST_HEAD(&wb->b_more_io);
388 INIT_LIST_HEAD(&wb->b_dirty_time);
389 spin_lock_init(&wb->list_lock);
390 INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
394 * Initial write bandwidth: 100 MB/s
396 #define INIT_BW (100 << (20 - PAGE_SHIFT))
398 int bdi_init(struct backing_dev_info *bdi)
405 bdi->max_ratio = 100;
406 bdi->max_prop_frac = FPROP_FRAC_BASE;
407 spin_lock_init(&bdi->wb_lock);
408 INIT_LIST_HEAD(&bdi->bdi_list);
409 INIT_LIST_HEAD(&bdi->work_list);
411 bdi_wb_init(&bdi->wb, bdi);
413 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
414 err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL);
419 bdi->dirty_exceeded = 0;
421 bdi->bw_time_stamp = jiffies;
422 bdi->written_stamp = 0;
424 bdi->balanced_dirty_ratelimit = INIT_BW;
425 bdi->dirty_ratelimit = INIT_BW;
426 bdi->write_bandwidth = INIT_BW;
427 bdi->avg_write_bandwidth = INIT_BW;
429 err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL);
434 percpu_counter_destroy(&bdi->bdi_stat[i]);
439 EXPORT_SYMBOL(bdi_init);
441 void bdi_destroy(struct backing_dev_info *bdi)
445 bdi_wb_shutdown(bdi);
447 WARN_ON(!list_empty(&bdi->work_list));
448 WARN_ON(delayed_work_pending(&bdi->wb.dwork));
451 bdi_debug_unregister(bdi);
452 device_unregister(bdi->dev);
456 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
457 percpu_counter_destroy(&bdi->bdi_stat[i]);
458 fprop_local_destroy_percpu(&bdi->completions);
460 EXPORT_SYMBOL(bdi_destroy);
463 * For use from filesystems to quickly init and register a bdi associated
464 * with dirty writeback
466 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
471 bdi->capabilities = 0;
476 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
477 atomic_long_inc_return(&bdi_seq));
485 EXPORT_SYMBOL(bdi_setup_and_register);
487 static wait_queue_head_t congestion_wqh[2] = {
488 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
489 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
491 static atomic_t nr_bdi_congested[2];
493 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
496 wait_queue_head_t *wqh = &congestion_wqh[sync];
498 bit = sync ? BDI_sync_congested : BDI_async_congested;
499 if (test_and_clear_bit(bit, &bdi->state))
500 atomic_dec(&nr_bdi_congested[sync]);
501 smp_mb__after_atomic();
502 if (waitqueue_active(wqh))
505 EXPORT_SYMBOL(clear_bdi_congested);
507 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
511 bit = sync ? BDI_sync_congested : BDI_async_congested;
512 if (!test_and_set_bit(bit, &bdi->state))
513 atomic_inc(&nr_bdi_congested[sync]);
515 EXPORT_SYMBOL(set_bdi_congested);
518 * congestion_wait - wait for a backing_dev to become uncongested
519 * @sync: SYNC or ASYNC IO
520 * @timeout: timeout in jiffies
522 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
523 * write congestion. If no backing_devs are congested then just wait for the
524 * next write to be completed.
526 long congestion_wait(int sync, long timeout)
529 unsigned long start = jiffies;
531 wait_queue_head_t *wqh = &congestion_wqh[sync];
533 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
534 ret = io_schedule_timeout(timeout);
535 finish_wait(wqh, &wait);
537 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
538 jiffies_to_usecs(jiffies - start));
542 EXPORT_SYMBOL(congestion_wait);
545 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
546 * @zone: A zone to check if it is heavily congested
547 * @sync: SYNC or ASYNC IO
548 * @timeout: timeout in jiffies
550 * In the event of a congested backing_dev (any backing_dev) and the given
551 * @zone has experienced recent congestion, this waits for up to @timeout
552 * jiffies for either a BDI to exit congestion of the given @sync queue
553 * or a write to complete.
555 * In the absence of zone congestion, cond_resched() is called to yield
556 * the processor if necessary but otherwise does not sleep.
558 * The return value is 0 if the sleep is for the full timeout. Otherwise,
559 * it is the number of jiffies that were still remaining when the function
560 * returned. return_value == timeout implies the function did not sleep.
562 long wait_iff_congested(struct zone *zone, int sync, long timeout)
565 unsigned long start = jiffies;
567 wait_queue_head_t *wqh = &congestion_wqh[sync];
570 * If there is no congestion, or heavy congestion is not being
571 * encountered in the current zone, yield if necessary instead
572 * of sleeping on the congestion queue
574 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
575 !test_bit(ZONE_CONGESTED, &zone->flags)) {
578 /* In case we scheduled, work out time remaining */
579 ret = timeout - (jiffies - start);
586 /* Sleep until uncongested or a write happens */
587 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
588 ret = io_schedule_timeout(timeout);
589 finish_wait(wqh, &wait);
592 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
593 jiffies_to_usecs(jiffies - start));
597 EXPORT_SYMBOL(wait_iff_congested);
599 int pdflush_proc_obsolete(struct ctl_table *table, int write,
600 void __user *buffer, size_t *lenp, loff_t *ppos)
604 if (*ppos || *lenp < sizeof(kbuf)) {
609 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
611 printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",