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 EXPORT_SYMBOL_GPL(default_backing_dev_info);
23 struct backing_dev_info noop_backing_dev_info = {
25 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
28 static struct class *bdi_class;
31 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
34 DEFINE_SPINLOCK(bdi_lock);
37 /* bdi_wq serves all asynchronous writeback tasks */
38 struct workqueue_struct *bdi_wq;
40 #ifdef CONFIG_DEBUG_FS
41 #include <linux/debugfs.h>
42 #include <linux/seq_file.h>
44 static struct dentry *bdi_debug_root;
46 static void bdi_debug_init(void)
48 bdi_debug_root = debugfs_create_dir("bdi", NULL);
51 static int bdi_debug_stats_show(struct seq_file *m, void *v)
53 struct backing_dev_info *bdi = m->private;
54 struct bdi_writeback *wb = &bdi->wb;
55 unsigned long background_thresh;
56 unsigned long dirty_thresh;
57 unsigned long bdi_thresh;
58 unsigned long nr_dirty, nr_io, nr_more_io;
61 nr_dirty = nr_io = nr_more_io = 0;
62 spin_lock(&wb->list_lock);
63 list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
65 list_for_each_entry(inode, &wb->b_io, i_wb_list)
67 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
69 spin_unlock(&wb->list_lock);
71 global_dirty_limits(&background_thresh, &dirty_thresh);
72 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
74 #define K(x) ((x) << (PAGE_SHIFT - 10))
76 "BdiWriteback: %10lu kB\n"
77 "BdiReclaimable: %10lu kB\n"
78 "BdiDirtyThresh: %10lu kB\n"
79 "DirtyThresh: %10lu kB\n"
80 "BackgroundThresh: %10lu kB\n"
81 "BdiDirtied: %10lu kB\n"
82 "BdiWritten: %10lu kB\n"
83 "BdiWriteBandwidth: %10lu kBps\n"
89 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
90 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
94 (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
95 (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
96 (unsigned long) K(bdi->write_bandwidth),
100 !list_empty(&bdi->bdi_list), bdi->state);
106 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
108 return single_open(file, bdi_debug_stats_show, inode->i_private);
111 static const struct file_operations bdi_debug_stats_fops = {
112 .open = bdi_debug_stats_open,
115 .release = single_release,
118 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
120 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
121 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
122 bdi, &bdi_debug_stats_fops);
125 static void bdi_debug_unregister(struct backing_dev_info *bdi)
127 debugfs_remove(bdi->debug_stats);
128 debugfs_remove(bdi->debug_dir);
131 static inline void bdi_debug_init(void)
134 static inline void bdi_debug_register(struct backing_dev_info *bdi,
138 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
143 static ssize_t read_ahead_kb_store(struct device *dev,
144 struct device_attribute *attr,
145 const char *buf, size_t count)
147 struct backing_dev_info *bdi = dev_get_drvdata(dev);
148 unsigned long read_ahead_kb;
151 ret = kstrtoul(buf, 10, &read_ahead_kb);
155 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
160 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
162 #define BDI_SHOW(name, expr) \
163 static ssize_t name##_show(struct device *dev, \
164 struct device_attribute *attr, char *page) \
166 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
168 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
170 static DEVICE_ATTR_RW(name);
172 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
174 static ssize_t min_ratio_store(struct device *dev,
175 struct device_attribute *attr, const char *buf, size_t count)
177 struct backing_dev_info *bdi = dev_get_drvdata(dev);
181 ret = kstrtouint(buf, 10, &ratio);
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);
200 ret = kstrtouint(buf, 10, &ratio);
204 ret = bdi_set_max_ratio(bdi, ratio);
210 BDI_SHOW(max_ratio, bdi->max_ratio)
212 static ssize_t stable_pages_required_show(struct device *dev,
213 struct device_attribute *attr,
216 struct backing_dev_info *bdi = dev_get_drvdata(dev);
218 return snprintf(page, PAGE_SIZE-1, "%d\n",
219 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
221 static DEVICE_ATTR_RO(stable_pages_required);
223 static struct attribute *bdi_dev_attrs[] = {
224 &dev_attr_read_ahead_kb.attr,
225 &dev_attr_min_ratio.attr,
226 &dev_attr_max_ratio.attr,
227 &dev_attr_stable_pages_required.attr,
230 ATTRIBUTE_GROUPS(bdi_dev);
232 static __init int bdi_class_init(void)
234 bdi_class = class_create(THIS_MODULE, "bdi");
235 if (IS_ERR(bdi_class))
236 return PTR_ERR(bdi_class);
238 bdi_class->dev_groups = bdi_dev_groups;
242 postcore_initcall(bdi_class_init);
244 static int __init default_bdi_init(void)
248 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
249 WQ_UNBOUND | WQ_SYSFS, 0);
253 err = bdi_init(&default_backing_dev_info);
255 bdi_register(&default_backing_dev_info, NULL, "default");
256 err = bdi_init(&noop_backing_dev_info);
260 subsys_initcall(default_bdi_init);
262 int bdi_has_dirty_io(struct backing_dev_info *bdi)
264 return wb_has_dirty_io(&bdi->wb);
268 * This function is used when the first inode for this bdi is marked dirty. It
269 * wakes-up the corresponding bdi thread which should then take care of the
270 * periodic background write-out of dirty inodes. Since the write-out would
271 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
272 * set up a timer which wakes the bdi thread up later.
274 * Note, we wouldn't bother setting up the timer, but this function is on the
275 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
276 * by delaying the wake-up.
278 * We have to be careful not to postpone flush work if it is scheduled for
279 * earlier. Thus we use queue_delayed_work().
281 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
283 unsigned long timeout;
285 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
286 spin_lock_bh(&bdi->wb_lock);
287 if (test_bit(BDI_registered, &bdi->state))
288 queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
289 spin_unlock_bh(&bdi->wb_lock);
293 * Remove bdi from bdi_list, and ensure that it is no longer visible
295 static void bdi_remove_from_list(struct backing_dev_info *bdi)
297 spin_lock_bh(&bdi_lock);
298 list_del_rcu(&bdi->bdi_list);
299 spin_unlock_bh(&bdi_lock);
301 synchronize_rcu_expedited();
304 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
305 const char *fmt, ...)
310 if (bdi->dev) /* The driver needs to use separate queues per device */
314 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
321 bdi_debug_register(bdi, dev_name(dev));
322 set_bit(BDI_registered, &bdi->state);
324 spin_lock_bh(&bdi_lock);
325 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
326 spin_unlock_bh(&bdi_lock);
328 trace_writeback_bdi_register(bdi);
331 EXPORT_SYMBOL(bdi_register);
333 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
335 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
337 EXPORT_SYMBOL(bdi_register_dev);
340 * Remove bdi from the global list and shutdown any threads we have running
342 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
344 /* Make sure nobody queues further work */
345 spin_lock_bh(&bdi->wb_lock);
346 if (!test_and_clear_bit(BDI_registered, &bdi->state)) {
347 spin_unlock_bh(&bdi->wb_lock);
350 spin_unlock_bh(&bdi->wb_lock);
353 * Make sure nobody finds us on the bdi_list anymore
355 bdi_remove_from_list(bdi);
358 * Drain work list and shutdown the delayed_work. At this point,
359 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
360 * is dying and its work_list needs to be drained no matter what.
362 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
363 flush_delayed_work(&bdi->wb.dwork);
367 * Called when the device behind @bdi has been removed or ejected.
369 * We can't really do much here except for reducing the dirty ratio at
370 * the moment. In the future we should be able to set a flag so that
371 * the filesystem can handle errors at mark_inode_dirty time instead
372 * of only at writeback time.
374 void bdi_unregister(struct backing_dev_info *bdi)
376 if (WARN_ON_ONCE(!bdi->dev))
379 bdi_set_min_ratio(bdi, 0);
381 EXPORT_SYMBOL(bdi_unregister);
383 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
385 memset(wb, 0, sizeof(*wb));
388 wb->last_old_flush = jiffies;
389 INIT_LIST_HEAD(&wb->b_dirty);
390 INIT_LIST_HEAD(&wb->b_io);
391 INIT_LIST_HEAD(&wb->b_more_io);
392 spin_lock_init(&wb->list_lock);
393 INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
397 * Initial write bandwidth: 100 MB/s
399 #define INIT_BW (100 << (20 - PAGE_SHIFT))
401 int bdi_init(struct backing_dev_info *bdi)
408 bdi->max_ratio = 100;
409 bdi->max_prop_frac = FPROP_FRAC_BASE;
410 spin_lock_init(&bdi->wb_lock);
411 INIT_LIST_HEAD(&bdi->bdi_list);
412 INIT_LIST_HEAD(&bdi->work_list);
414 bdi_wb_init(&bdi->wb, bdi);
416 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
417 err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL);
422 bdi->dirty_exceeded = 0;
424 bdi->bw_time_stamp = jiffies;
425 bdi->written_stamp = 0;
427 bdi->balanced_dirty_ratelimit = INIT_BW;
428 bdi->dirty_ratelimit = INIT_BW;
429 bdi->write_bandwidth = INIT_BW;
430 bdi->avg_write_bandwidth = INIT_BW;
432 err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL);
437 percpu_counter_destroy(&bdi->bdi_stat[i]);
442 EXPORT_SYMBOL(bdi_init);
444 void bdi_destroy(struct backing_dev_info *bdi)
448 bdi_wb_shutdown(bdi);
450 WARN_ON(!list_empty(&bdi->work_list));
451 WARN_ON(delayed_work_pending(&bdi->wb.dwork));
454 bdi_debug_unregister(bdi);
455 device_unregister(bdi->dev);
459 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
460 percpu_counter_destroy(&bdi->bdi_stat[i]);
461 fprop_local_destroy_percpu(&bdi->completions);
463 EXPORT_SYMBOL(bdi_destroy);
466 * For use from filesystems to quickly init and register a bdi associated
467 * with dirty writeback
469 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
474 bdi->capabilities = 0;
479 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
480 atomic_long_inc_return(&bdi_seq));
488 EXPORT_SYMBOL(bdi_setup_and_register);
490 static wait_queue_head_t congestion_wqh[2] = {
491 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
492 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
494 static atomic_t nr_bdi_congested[2];
496 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
499 wait_queue_head_t *wqh = &congestion_wqh[sync];
501 bit = sync ? BDI_sync_congested : BDI_async_congested;
502 if (test_and_clear_bit(bit, &bdi->state))
503 atomic_dec(&nr_bdi_congested[sync]);
504 smp_mb__after_atomic();
505 if (waitqueue_active(wqh))
508 EXPORT_SYMBOL(clear_bdi_congested);
510 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
514 bit = sync ? BDI_sync_congested : BDI_async_congested;
515 if (!test_and_set_bit(bit, &bdi->state))
516 atomic_inc(&nr_bdi_congested[sync]);
518 EXPORT_SYMBOL(set_bdi_congested);
521 * congestion_wait - wait for a backing_dev to become uncongested
522 * @sync: SYNC or ASYNC IO
523 * @timeout: timeout in jiffies
525 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
526 * write congestion. If no backing_devs are congested then just wait for the
527 * next write to be completed.
529 long congestion_wait(int sync, long timeout)
532 unsigned long start = jiffies;
534 wait_queue_head_t *wqh = &congestion_wqh[sync];
536 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
537 ret = io_schedule_timeout(timeout);
538 finish_wait(wqh, &wait);
540 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
541 jiffies_to_usecs(jiffies - start));
545 EXPORT_SYMBOL(congestion_wait);
548 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
549 * @zone: A zone to check if it is heavily congested
550 * @sync: SYNC or ASYNC IO
551 * @timeout: timeout in jiffies
553 * In the event of a congested backing_dev (any backing_dev) and the given
554 * @zone has experienced recent congestion, this waits for up to @timeout
555 * jiffies for either a BDI to exit congestion of the given @sync queue
556 * or a write to complete.
558 * In the absence of zone congestion, cond_resched() is called to yield
559 * the processor if necessary but otherwise does not sleep.
561 * The return value is 0 if the sleep is for the full timeout. Otherwise,
562 * it is the number of jiffies that were still remaining when the function
563 * returned. return_value == timeout implies the function did not sleep.
565 long wait_iff_congested(struct zone *zone, int sync, long timeout)
568 unsigned long start = jiffies;
570 wait_queue_head_t *wqh = &congestion_wqh[sync];
573 * If there is no congestion, or heavy congestion is not being
574 * encountered in the current zone, yield if necessary instead
575 * of sleeping on the congestion queue
577 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
578 !test_bit(ZONE_CONGESTED, &zone->flags)) {
581 /* In case we scheduled, work out time remaining */
582 ret = timeout - (jiffies - start);
589 /* Sleep until uncongested or a write happens */
590 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
591 ret = io_schedule_timeout(timeout);
592 finish_wait(wqh, &wait);
595 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
596 jiffies_to_usecs(jiffies - start));
600 EXPORT_SYMBOL(wait_iff_congested);
602 int pdflush_proc_obsolete(struct ctl_table *table, int write,
603 void __user *buffer, size_t *lenp, loff_t *ppos)
607 if (*ppos || *lenp < sizeof(kbuf)) {
612 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
614 printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",