vmpressure: make sure there are no events queued after memcg is offlined
[linux-2.6.git] / mm / backing-dev.c
blobe04454cdb33f4d58809c36b0212aa0ad74c1e22c
2 #include <linux/wait.h>
3 #include <linux/backing-dev.h>
4 #include <linux/kthread.h>
5 #include <linux/freezer.h>
6 #include <linux/fs.h>
7 #include <linux/pagemap.h>
8 #include <linux/mm.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 = {
18 .name = "default",
19 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
20 .state = 0,
21 .capabilities = BDI_CAP_MAP_COPY,
23 EXPORT_SYMBOL_GPL(default_backing_dev_info);
25 struct backing_dev_info noop_backing_dev_info = {
26 .name = "noop",
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
35 * locking.
37 DEFINE_SPINLOCK(bdi_lock);
38 LIST_HEAD(bdi_list);
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)
45 if (wb1 < wb2) {
46 spin_lock(&wb1->list_lock);
47 spin_lock_nested(&wb2->list_lock, 1);
48 } else {
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;
73 struct inode *inode;
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)
78 nr_dirty++;
79 list_for_each_entry(inode, &wb->b_io, i_wb_list)
80 nr_io++;
81 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
82 nr_more_io++;
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))
89 seq_printf(m,
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"
98 "b_dirty: %10lu\n"
99 "b_io: %10lu\n"
100 "b_more_io: %10lu\n"
101 "bdi_list: %10u\n"
102 "state: %10lx\n",
103 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
104 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
105 K(bdi_thresh),
106 K(dirty_thresh),
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),
111 nr_dirty,
112 nr_io,
113 nr_more_io,
114 !list_empty(&bdi->bdi_list), bdi->state);
115 #undef K
117 return 0;
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,
127 .read = seq_read,
128 .llseek = seq_lseek,
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);
144 #else
145 static inline void bdi_debug_init(void)
148 static inline void bdi_debug_register(struct backing_dev_info *bdi,
149 const char *name)
152 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
155 #endif
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;
163 ssize_t ret;
165 ret = kstrtoul(buf, 10, &read_ahead_kb);
166 if (ret < 0)
167 return ret;
169 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
171 return count;
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);
191 unsigned int ratio;
192 ssize_t ret;
194 ret = kstrtouint(buf, 10, &ratio);
195 if (ret < 0)
196 return ret;
198 ret = bdi_set_min_ratio(bdi, ratio);
199 if (!ret)
200 ret = count;
202 return ret;
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);
210 unsigned int ratio;
211 ssize_t ret;
213 ret = kstrtouint(buf, 10, &ratio);
214 if (ret < 0)
215 return ret;
217 ret = bdi_set_max_ratio(bdi, ratio);
218 if (!ret)
219 ret = count;
221 return ret;
223 BDI_SHOW(max_ratio, bdi->max_ratio)
225 static ssize_t stable_pages_required_show(struct device *dev,
226 struct device_attribute *attr,
227 char *page)
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 static struct device_attribute bdi_dev_attrs[] = {
236 __ATTR_RW(read_ahead_kb),
237 __ATTR_RW(min_ratio),
238 __ATTR_RW(max_ratio),
239 __ATTR_RO(stable_pages_required),
240 __ATTR_NULL,
243 static __init int bdi_class_init(void)
245 bdi_class = class_create(THIS_MODULE, "bdi");
246 if (IS_ERR(bdi_class))
247 return PTR_ERR(bdi_class);
249 bdi_class->dev_attrs = bdi_dev_attrs;
250 bdi_debug_init();
251 return 0;
253 postcore_initcall(bdi_class_init);
255 static int __init default_bdi_init(void)
257 int err;
259 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
260 WQ_UNBOUND | WQ_SYSFS, 0);
261 if (!bdi_wq)
262 return -ENOMEM;
264 err = bdi_init(&default_backing_dev_info);
265 if (!err)
266 bdi_register(&default_backing_dev_info, NULL, "default");
267 err = bdi_init(&noop_backing_dev_info);
269 return err;
271 subsys_initcall(default_bdi_init);
273 int bdi_has_dirty_io(struct backing_dev_info *bdi)
275 return wb_has_dirty_io(&bdi->wb);
279 * This function is used when the first inode for this bdi is marked dirty. It
280 * wakes-up the corresponding bdi thread which should then take care of the
281 * periodic background write-out of dirty inodes. Since the write-out would
282 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
283 * set up a timer which wakes the bdi thread up later.
285 * Note, we wouldn't bother setting up the timer, but this function is on the
286 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
287 * by delaying the wake-up.
289 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
291 unsigned long timeout;
293 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
294 mod_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
298 * Remove bdi from bdi_list, and ensure that it is no longer visible
300 static void bdi_remove_from_list(struct backing_dev_info *bdi)
302 spin_lock_bh(&bdi_lock);
303 list_del_rcu(&bdi->bdi_list);
304 spin_unlock_bh(&bdi_lock);
306 synchronize_rcu_expedited();
308 /* bdi_list is now unused, clear it to mark @bdi dying */
309 INIT_LIST_HEAD(&bdi->bdi_list);
312 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
313 const char *fmt, ...)
315 va_list args;
316 struct device *dev;
318 if (bdi->dev) /* The driver needs to use separate queues per device */
319 return 0;
321 va_start(args, fmt);
322 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
323 va_end(args);
324 if (IS_ERR(dev))
325 return PTR_ERR(dev);
327 bdi->dev = dev;
329 bdi_debug_register(bdi, dev_name(dev));
330 set_bit(BDI_registered, &bdi->state);
332 spin_lock_bh(&bdi_lock);
333 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
334 spin_unlock_bh(&bdi_lock);
336 trace_writeback_bdi_register(bdi);
337 return 0;
339 EXPORT_SYMBOL(bdi_register);
341 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
343 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
345 EXPORT_SYMBOL(bdi_register_dev);
348 * Remove bdi from the global list and shutdown any threads we have running
350 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
352 if (!bdi_cap_writeback_dirty(bdi))
353 return;
356 * Make sure nobody finds us on the bdi_list anymore
358 bdi_remove_from_list(bdi);
361 * Drain work list and shutdown the delayed_work. At this point,
362 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
363 * is dying and its work_list needs to be drained no matter what.
365 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
366 flush_delayed_work(&bdi->wb.dwork);
367 WARN_ON(!list_empty(&bdi->work_list));
370 * This shouldn't be necessary unless @bdi for some reason has
371 * unflushed dirty IO after work_list is drained. Do it anyway
372 * just in case.
374 cancel_delayed_work_sync(&bdi->wb.dwork);
378 * This bdi is going away now, make sure that no super_blocks point to it
380 static void bdi_prune_sb(struct backing_dev_info *bdi)
382 struct super_block *sb;
384 spin_lock(&sb_lock);
385 list_for_each_entry(sb, &super_blocks, s_list) {
386 if (sb->s_bdi == bdi)
387 sb->s_bdi = &default_backing_dev_info;
389 spin_unlock(&sb_lock);
392 void bdi_unregister(struct backing_dev_info *bdi)
394 struct device *dev = bdi->dev;
396 if (dev) {
397 bdi_set_min_ratio(bdi, 0);
398 trace_writeback_bdi_unregister(bdi);
399 bdi_prune_sb(bdi);
401 bdi_wb_shutdown(bdi);
402 bdi_debug_unregister(bdi);
404 spin_lock_bh(&bdi->wb_lock);
405 bdi->dev = NULL;
406 spin_unlock_bh(&bdi->wb_lock);
408 device_unregister(dev);
411 EXPORT_SYMBOL(bdi_unregister);
413 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
415 memset(wb, 0, sizeof(*wb));
417 wb->bdi = bdi;
418 wb->last_old_flush = jiffies;
419 INIT_LIST_HEAD(&wb->b_dirty);
420 INIT_LIST_HEAD(&wb->b_io);
421 INIT_LIST_HEAD(&wb->b_more_io);
422 spin_lock_init(&wb->list_lock);
423 INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
427 * Initial write bandwidth: 100 MB/s
429 #define INIT_BW (100 << (20 - PAGE_SHIFT))
431 int bdi_init(struct backing_dev_info *bdi)
433 int i, err;
435 bdi->dev = NULL;
437 bdi->min_ratio = 0;
438 bdi->max_ratio = 100;
439 bdi->max_prop_frac = FPROP_FRAC_BASE;
440 spin_lock_init(&bdi->wb_lock);
441 INIT_LIST_HEAD(&bdi->bdi_list);
442 INIT_LIST_HEAD(&bdi->work_list);
444 bdi_wb_init(&bdi->wb, bdi);
446 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
447 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
448 if (err)
449 goto err;
452 bdi->dirty_exceeded = 0;
454 bdi->bw_time_stamp = jiffies;
455 bdi->written_stamp = 0;
457 bdi->balanced_dirty_ratelimit = INIT_BW;
458 bdi->dirty_ratelimit = INIT_BW;
459 bdi->write_bandwidth = INIT_BW;
460 bdi->avg_write_bandwidth = INIT_BW;
462 err = fprop_local_init_percpu(&bdi->completions);
464 if (err) {
465 err:
466 while (i--)
467 percpu_counter_destroy(&bdi->bdi_stat[i]);
470 return err;
472 EXPORT_SYMBOL(bdi_init);
474 void bdi_destroy(struct backing_dev_info *bdi)
476 int i;
479 * Splice our entries to the default_backing_dev_info, if this
480 * bdi disappears
482 if (bdi_has_dirty_io(bdi)) {
483 struct bdi_writeback *dst = &default_backing_dev_info.wb;
485 bdi_lock_two(&bdi->wb, dst);
486 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
487 list_splice(&bdi->wb.b_io, &dst->b_io);
488 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
489 spin_unlock(&bdi->wb.list_lock);
490 spin_unlock(&dst->list_lock);
493 bdi_unregister(bdi);
496 * If bdi_unregister() had already been called earlier, the dwork
497 * could still be pending because bdi_prune_sb() can race with the
498 * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty().
500 cancel_delayed_work_sync(&bdi->wb.dwork);
502 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
503 percpu_counter_destroy(&bdi->bdi_stat[i]);
505 fprop_local_destroy_percpu(&bdi->completions);
507 EXPORT_SYMBOL(bdi_destroy);
510 * For use from filesystems to quickly init and register a bdi associated
511 * with dirty writeback
513 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
514 unsigned int cap)
516 int err;
518 bdi->name = name;
519 bdi->capabilities = cap;
520 err = bdi_init(bdi);
521 if (err)
522 return err;
524 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
525 atomic_long_inc_return(&bdi_seq));
526 if (err) {
527 bdi_destroy(bdi);
528 return err;
531 return 0;
533 EXPORT_SYMBOL(bdi_setup_and_register);
535 static wait_queue_head_t congestion_wqh[2] = {
536 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
537 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
539 static atomic_t nr_bdi_congested[2];
541 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
543 enum bdi_state bit;
544 wait_queue_head_t *wqh = &congestion_wqh[sync];
546 bit = sync ? BDI_sync_congested : BDI_async_congested;
547 if (test_and_clear_bit(bit, &bdi->state))
548 atomic_dec(&nr_bdi_congested[sync]);
549 smp_mb__after_clear_bit();
550 if (waitqueue_active(wqh))
551 wake_up(wqh);
553 EXPORT_SYMBOL(clear_bdi_congested);
555 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
557 enum bdi_state bit;
559 bit = sync ? BDI_sync_congested : BDI_async_congested;
560 if (!test_and_set_bit(bit, &bdi->state))
561 atomic_inc(&nr_bdi_congested[sync]);
563 EXPORT_SYMBOL(set_bdi_congested);
566 * congestion_wait - wait for a backing_dev to become uncongested
567 * @sync: SYNC or ASYNC IO
568 * @timeout: timeout in jiffies
570 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
571 * write congestion. If no backing_devs are congested then just wait for the
572 * next write to be completed.
574 long congestion_wait(int sync, long timeout)
576 long ret;
577 unsigned long start = jiffies;
578 DEFINE_WAIT(wait);
579 wait_queue_head_t *wqh = &congestion_wqh[sync];
581 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
582 ret = io_schedule_timeout(timeout);
583 finish_wait(wqh, &wait);
585 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
586 jiffies_to_usecs(jiffies - start));
588 return ret;
590 EXPORT_SYMBOL(congestion_wait);
593 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
594 * @zone: A zone to check if it is heavily congested
595 * @sync: SYNC or ASYNC IO
596 * @timeout: timeout in jiffies
598 * In the event of a congested backing_dev (any backing_dev) and the given
599 * @zone has experienced recent congestion, this waits for up to @timeout
600 * jiffies for either a BDI to exit congestion of the given @sync queue
601 * or a write to complete.
603 * In the absence of zone congestion, cond_resched() is called to yield
604 * the processor if necessary but otherwise does not sleep.
606 * The return value is 0 if the sleep is for the full timeout. Otherwise,
607 * it is the number of jiffies that were still remaining when the function
608 * returned. return_value == timeout implies the function did not sleep.
610 long wait_iff_congested(struct zone *zone, int sync, long timeout)
612 long ret;
613 unsigned long start = jiffies;
614 DEFINE_WAIT(wait);
615 wait_queue_head_t *wqh = &congestion_wqh[sync];
618 * If there is no congestion, or heavy congestion is not being
619 * encountered in the current zone, yield if necessary instead
620 * of sleeping on the congestion queue
622 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
623 !zone_is_reclaim_congested(zone)) {
624 cond_resched();
626 /* In case we scheduled, work out time remaining */
627 ret = timeout - (jiffies - start);
628 if (ret < 0)
629 ret = 0;
631 goto out;
634 /* Sleep until uncongested or a write happens */
635 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
636 ret = io_schedule_timeout(timeout);
637 finish_wait(wqh, &wait);
639 out:
640 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
641 jiffies_to_usecs(jiffies - start));
643 return ret;
645 EXPORT_SYMBOL(wait_iff_congested);
647 int pdflush_proc_obsolete(struct ctl_table *table, int write,
648 void __user *buffer, size_t *lenp, loff_t *ppos)
650 char kbuf[] = "0\n";
652 if (*ppos) {
653 *lenp = 0;
654 return 0;
657 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
658 return -EFAULT;
659 printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
660 table->procname);
662 *lenp = 2;
663 *ppos += *lenp;
664 return 2;