Linux Audio

Check our new training course

Loading...
v3.15
 
  1
 
  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>
 
 14
 15static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
 16
 17struct 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,
 22};
 23EXPORT_SYMBOL_GPL(default_backing_dev_info);
 24
 25struct backing_dev_info noop_backing_dev_info = {
 26	.name		= "noop",
 27	.capabilities	= BDI_CAP_NO_ACCT_AND_WRITEBACK,
 28};
 29EXPORT_SYMBOL_GPL(noop_backing_dev_info);
 30
 31static struct class *bdi_class;
 32
 33/*
 34 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
 35 * locking.
 36 */
 37DEFINE_SPINLOCK(bdi_lock);
 
 
 38LIST_HEAD(bdi_list);
 39
 40/* bdi_wq serves all asynchronous writeback tasks */
 41struct workqueue_struct *bdi_wq;
 42
 43void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2)
 44{
 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);
 51	}
 52}
 53
 54#ifdef CONFIG_DEBUG_FS
 55#include <linux/debugfs.h>
 56#include <linux/seq_file.h>
 57
 58static struct dentry *bdi_debug_root;
 59
 60static void bdi_debug_init(void)
 61{
 62	bdi_debug_root = debugfs_create_dir("bdi", NULL);
 63}
 64
 65static int bdi_debug_stats_show(struct seq_file *m, void *v)
 66{
 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;
 74
 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);
 84
 85	global_dirty_limits(&background_thresh, &dirty_thresh);
 86	bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
 87
 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
116
117	return 0;
118}
119
120static int bdi_debug_stats_open(struct inode *inode, struct file *file)
121{
122	return single_open(file, bdi_debug_stats_show, inode->i_private);
123}
124
125static 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,
130};
131
132static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
133{
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);
 
137}
138
139static void bdi_debug_unregister(struct backing_dev_info *bdi)
140{
141	debugfs_remove(bdi->debug_stats);
142	debugfs_remove(bdi->debug_dir);
143}
144#else
145static inline void bdi_debug_init(void)
146{
147}
148static inline void bdi_debug_register(struct backing_dev_info *bdi,
149				      const char *name)
150{
151}
152static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
153{
154}
155#endif
156
157static ssize_t read_ahead_kb_store(struct device *dev,
158				  struct device_attribute *attr,
159				  const char *buf, size_t count)
160{
161	struct backing_dev_info *bdi = dev_get_drvdata(dev);
162	unsigned long read_ahead_kb;
163	ssize_t ret;
164
165	ret = kstrtoul(buf, 10, &read_ahead_kb);
166	if (ret < 0)
167		return ret;
168
169	bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
170
171	return count;
172}
173
174#define K(pages) ((pages) << (PAGE_SHIFT - 10))
175
176#define BDI_SHOW(name, expr)						\
177static ssize_t name##_show(struct device *dev,				\
178			   struct device_attribute *attr, char *page)	\
179{									\
180	struct backing_dev_info *bdi = dev_get_drvdata(dev);		\
181									\
182	return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr);	\
183}									\
184static DEVICE_ATTR_RW(name);
185
186BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
187
188static ssize_t min_ratio_store(struct device *dev,
189		struct device_attribute *attr, const char *buf, size_t count)
190{
191	struct backing_dev_info *bdi = dev_get_drvdata(dev);
192	unsigned int ratio;
193	ssize_t ret;
194
195	ret = kstrtouint(buf, 10, &ratio);
196	if (ret < 0)
197		return ret;
198
199	ret = bdi_set_min_ratio(bdi, ratio);
200	if (!ret)
201		ret = count;
202
203	return ret;
204}
205BDI_SHOW(min_ratio, bdi->min_ratio)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
206
207static ssize_t max_ratio_store(struct device *dev,
208		struct device_attribute *attr, const char *buf, size_t count)
209{
210	struct backing_dev_info *bdi = dev_get_drvdata(dev);
211	unsigned int ratio;
212	ssize_t ret;
213
214	ret = kstrtouint(buf, 10, &ratio);
215	if (ret < 0)
216		return ret;
217
218	ret = bdi_set_max_ratio(bdi, ratio);
219	if (!ret)
220		ret = count;
221
222	return ret;
223}
224BDI_SHOW(max_ratio, bdi->max_ratio)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
225
226static ssize_t stable_pages_required_show(struct device *dev,
227					  struct device_attribute *attr,
228					  char *page)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
229{
230	struct backing_dev_info *bdi = dev_get_drvdata(dev);
231
232	return snprintf(page, PAGE_SIZE-1, "%d\n",
233			bdi_cap_stable_pages_required(bdi) ? 1 : 0);
234}
235static DEVICE_ATTR_RO(stable_pages_required);
236
237static struct attribute *bdi_dev_attrs[] = {
238	&dev_attr_read_ahead_kb.attr,
239	&dev_attr_min_ratio.attr,
 
240	&dev_attr_max_ratio.attr,
 
 
 
241	&dev_attr_stable_pages_required.attr,
 
242	NULL,
243};
244ATTRIBUTE_GROUPS(bdi_dev);
245
 
 
 
 
 
246static __init int bdi_class_init(void)
247{
248	bdi_class = class_create(THIS_MODULE, "bdi");
249	if (IS_ERR(bdi_class))
250		return PTR_ERR(bdi_class);
 
 
251
252	bdi_class->dev_groups = bdi_dev_groups;
253	bdi_debug_init();
 
254	return 0;
255}
256postcore_initcall(bdi_class_init);
257
258static int __init default_bdi_init(void)
259{
260	int err;
261
262	bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
263					      WQ_UNBOUND | WQ_SYSFS, 0);
264	if (!bdi_wq)
265		return -ENOMEM;
 
 
 
266
267	err = bdi_init(&default_backing_dev_info);
268	if (!err)
269		bdi_register(&default_backing_dev_info, NULL, "default");
270	err = bdi_init(&noop_backing_dev_info);
271
272	return err;
273}
274subsys_initcall(default_bdi_init);
275
276int bdi_has_dirty_io(struct backing_dev_info *bdi)
 
 
 
 
 
 
277{
278	return wb_has_dirty_io(&bdi->wb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
279}
280
 
 
281/*
282 * This function is used when the first inode for this bdi is marked dirty. It
283 * wakes-up the corresponding bdi thread which should then take care of the
284 * periodic background write-out of dirty inodes. Since the write-out would
285 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
286 * set up a timer which wakes the bdi thread up later.
287 *
288 * Note, we wouldn't bother setting up the timer, but this function is on the
289 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
290 * by delaying the wake-up.
291 *
292 * We have to be careful not to postpone flush work if it is scheduled for
293 * earlier. Thus we use queue_delayed_work().
294 */
295void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
296{
297	unsigned long timeout;
 
 
298
299	timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
300	spin_lock_bh(&bdi->wb_lock);
301	if (test_bit(BDI_registered, &bdi->state))
302		queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
303	spin_unlock_bh(&bdi->wb_lock);
304}
305
 
 
 
 
306/*
307 * Remove bdi from bdi_list, and ensure that it is no longer visible
 
308 */
309static void bdi_remove_from_list(struct backing_dev_info *bdi)
 
 
 
 
 
 
 
310{
311	spin_lock_bh(&bdi_lock);
312	list_del_rcu(&bdi->bdi_list);
313	spin_unlock_bh(&bdi_lock);
314
315	synchronize_rcu_expedited();
 
316}
317
318int bdi_register(struct backing_dev_info *bdi, struct device *parent,
319		const char *fmt, ...)
320{
321	va_list args;
322	struct device *dev;
 
323
324	if (bdi->dev)	/* The driver needs to use separate queues per device */
325		return 0;
326
327	va_start(args, fmt);
328	dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
329	va_end(args);
330	if (IS_ERR(dev))
331		return PTR_ERR(dev);
332
333	bdi->dev = dev;
 
334
335	bdi_debug_register(bdi, dev_name(dev));
336	set_bit(BDI_registered, &bdi->state);
337
338	spin_lock_bh(&bdi_lock);
339	list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
340	spin_unlock_bh(&bdi_lock);
341
342	trace_writeback_bdi_register(bdi);
343	return 0;
 
 
344}
345EXPORT_SYMBOL(bdi_register);
346
347int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
348{
349	return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
 
 
350}
351EXPORT_SYMBOL(bdi_register_dev);
352
353/*
354 * Remove bdi from the global list and shutdown any threads we have running
355 */
356static void bdi_wb_shutdown(struct backing_dev_info *bdi)
357{
358	if (!bdi_cap_writeback_dirty(bdi))
359		return;
360
361	/*
362	 * Make sure nobody finds us on the bdi_list anymore
363	 */
364	bdi_remove_from_list(bdi);
 
 
365
366	/* Make sure nobody queues further work */
367	spin_lock_bh(&bdi->wb_lock);
368	clear_bit(BDI_registered, &bdi->state);
369	spin_unlock_bh(&bdi->wb_lock);
 
 
370
371	/*
372	 * Drain work list and shutdown the delayed_work.  At this point,
373	 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
374	 * is dying and its work_list needs to be drained no matter what.
375	 */
376	mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
377	flush_delayed_work(&bdi->wb.dwork);
378	WARN_ON(!list_empty(&bdi->work_list));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
379
380	/*
381	 * This shouldn't be necessary unless @bdi for some reason has
382	 * unflushed dirty IO after work_list is drained.  Do it anyway
383	 * just in case.
 
384	 */
385	cancel_delayed_work_sync(&bdi->wb.dwork);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
386}
387
388/*
389 * This bdi is going away now, make sure that no super_blocks point to it
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
390 */
391static void bdi_prune_sb(struct backing_dev_info *bdi)
 
392{
393	struct super_block *sb;
 
 
 
394
395	spin_lock(&sb_lock);
396	list_for_each_entry(sb, &super_blocks, s_list) {
397		if (sb->s_bdi == bdi)
398			sb->s_bdi = &default_backing_dev_info;
 
 
 
 
 
 
399	}
400	spin_unlock(&sb_lock);
 
 
401}
402
403void bdi_unregister(struct backing_dev_info *bdi)
 
 
 
 
 
 
 
 
 
 
 
404{
405	struct device *dev = bdi->dev;
406
407	if (dev) {
408		bdi_set_min_ratio(bdi, 0);
409		trace_writeback_bdi_unregister(bdi);
410		bdi_prune_sb(bdi);
411
412		bdi_wb_shutdown(bdi);
413		bdi_debug_unregister(bdi);
 
414
415		spin_lock_bh(&bdi->wb_lock);
416		bdi->dev = NULL;
417		spin_unlock_bh(&bdi->wb_lock);
 
 
 
 
 
 
 
418
419		device_unregister(dev);
 
 
 
420	}
 
421}
422EXPORT_SYMBOL(bdi_unregister);
423
424static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
425{
426	memset(wb, 0, sizeof(*wb));
427
428	wb->bdi = bdi;
429	wb->last_old_flush = jiffies;
430	INIT_LIST_HEAD(&wb->b_dirty);
431	INIT_LIST_HEAD(&wb->b_io);
432	INIT_LIST_HEAD(&wb->b_more_io);
433	spin_lock_init(&wb->list_lock);
434	INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
 
 
 
 
 
 
 
 
 
 
 
 
 
435}
436
437/*
438 * Initial write bandwidth: 100 MB/s
 
 
 
 
439 */
440#define INIT_BW		(100 << (20 - PAGE_SHIFT))
441
442int bdi_init(struct backing_dev_info *bdi)
443{
444	int i, err;
 
445
446	bdi->dev = NULL;
447
448	bdi->min_ratio = 0;
449	bdi->max_ratio = 100;
450	bdi->max_prop_frac = FPROP_FRAC_BASE;
451	spin_lock_init(&bdi->wb_lock);
452	INIT_LIST_HEAD(&bdi->bdi_list);
453	INIT_LIST_HEAD(&bdi->work_list);
454
455	bdi_wb_init(&bdi->wb, bdi);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
456
457	for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
458		err = percpu_counter_init(&bdi->bdi_stat[i], 0);
459		if (err)
460			goto err;
461	}
462
463	bdi->dirty_exceeded = 0;
 
464
465	bdi->bw_time_stamp = jiffies;
466	bdi->written_stamp = 0;
467
468	bdi->balanced_dirty_ratelimit = INIT_BW;
469	bdi->dirty_ratelimit = INIT_BW;
470	bdi->write_bandwidth = INIT_BW;
471	bdi->avg_write_bandwidth = INIT_BW;
 
 
 
 
 
 
472
473	err = fprop_local_init_percpu(&bdi->completions);
 
 
 
 
474
475	if (err) {
476err:
477		while (i--)
478			percpu_counter_destroy(&bdi->bdi_stat[i]);
479	}
480
481	return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
482}
483EXPORT_SYMBOL(bdi_init);
484
485void bdi_destroy(struct backing_dev_info *bdi)
486{
487	int i;
 
 
 
488
 
 
489	/*
490	 * Splice our entries to the default_backing_dev_info, if this
491	 * bdi disappears
 
492	 */
493	if (bdi_has_dirty_io(bdi)) {
494		struct bdi_writeback *dst = &default_backing_dev_info.wb;
 
495
496		bdi_lock_two(&bdi->wb, dst);
497		list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
498		list_splice(&bdi->wb.b_io, &dst->b_io);
499		list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
500		spin_unlock(&bdi->wb.list_lock);
501		spin_unlock(&dst->list_lock);
502	}
503
504	bdi_unregister(bdi);
505
506	/*
507	 * If bdi_unregister() had already been called earlier, the dwork
508	 * could still be pending because bdi_prune_sb() can race with the
509	 * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty().
510	 */
511	cancel_delayed_work_sync(&bdi->wb.dwork);
512
513	for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
514		percpu_counter_destroy(&bdi->bdi_stat[i]);
515
516	fprop_local_destroy_percpu(&bdi->completions);
 
 
517}
518EXPORT_SYMBOL(bdi_destroy);
519
520/*
521 * For use from filesystems to quickly init and register a bdi associated
522 * with dirty writeback
523 */
524int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
525			   unsigned int cap)
526{
527	int err;
 
528
529	bdi->name = name;
530	bdi->capabilities = cap;
531	err = bdi_init(bdi);
532	if (err)
533		return err;
534
535	err = bdi_register(bdi, NULL, "%.28s-%ld", name,
536			   atomic_long_inc_return(&bdi_seq));
537	if (err) {
538		bdi_destroy(bdi);
539		return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
540	}
541
542	return 0;
 
 
543}
544EXPORT_SYMBOL(bdi_setup_and_register);
545
546static wait_queue_head_t congestion_wqh[2] = {
547		__WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
548		__WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
549	};
550static atomic_t nr_bdi_congested[2];
 
 
 
 
 
 
551
552void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
 
 
 
 
 
 
 
 
 
 
 
553{
554	enum bdi_state bit;
555	wait_queue_head_t *wqh = &congestion_wqh[sync];
 
 
 
 
 
 
 
 
 
 
 
556
557	bit = sync ? BDI_sync_congested : BDI_async_congested;
558	if (test_and_clear_bit(bit, &bdi->state))
559		atomic_dec(&nr_bdi_congested[sync]);
560	smp_mb__after_clear_bit();
561	if (waitqueue_active(wqh))
562		wake_up(wqh);
 
 
 
 
 
 
 
 
 
 
 
563}
564EXPORT_SYMBOL(clear_bdi_congested);
565
566void set_bdi_congested(struct backing_dev_info *bdi, int sync)
567{
568	enum bdi_state bit;
 
569
570	bit = sync ? BDI_sync_congested : BDI_async_congested;
571	if (!test_and_set_bit(bit, &bdi->state))
572		atomic_inc(&nr_bdi_congested[sync]);
 
573}
574EXPORT_SYMBOL(set_bdi_congested);
575
576/**
577 * congestion_wait - wait for a backing_dev to become uncongested
578 * @sync: SYNC or ASYNC IO
579 * @timeout: timeout in jiffies
580 *
581 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
582 * write congestion.  If no backing_devs are congested then just wait for the
583 * next write to be completed.
584 */
585long congestion_wait(int sync, long timeout)
586{
587	long ret;
588	unsigned long start = jiffies;
589	DEFINE_WAIT(wait);
590	wait_queue_head_t *wqh = &congestion_wqh[sync];
591
592	prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
593	ret = io_schedule_timeout(timeout);
594	finish_wait(wqh, &wait);
595
596	trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
597					jiffies_to_usecs(jiffies - start));
 
 
 
 
 
 
 
598
599	return ret;
600}
601EXPORT_SYMBOL(congestion_wait);
602
603/**
604 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
605 * @zone: A zone to check if it is heavily congested
606 * @sync: SYNC or ASYNC IO
607 * @timeout: timeout in jiffies
608 *
609 * In the event of a congested backing_dev (any backing_dev) and the given
610 * @zone has experienced recent congestion, this waits for up to @timeout
611 * jiffies for either a BDI to exit congestion of the given @sync queue
612 * or a write to complete.
613 *
614 * In the absence of zone congestion, cond_resched() is called to yield
615 * the processor if necessary but otherwise does not sleep.
616 *
617 * The return value is 0 if the sleep is for the full timeout. Otherwise,
618 * it is the number of jiffies that were still remaining when the function
619 * returned. return_value == timeout implies the function did not sleep.
620 */
621long wait_iff_congested(struct zone *zone, int sync, long timeout)
622{
623	long ret;
624	unsigned long start = jiffies;
625	DEFINE_WAIT(wait);
626	wait_queue_head_t *wqh = &congestion_wqh[sync];
627
628	/*
629	 * If there is no congestion, or heavy congestion is not being
630	 * encountered in the current zone, yield if necessary instead
631	 * of sleeping on the congestion queue
632	 */
633	if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
634			!zone_is_reclaim_congested(zone)) {
635		cond_resched();
636
637		/* In case we scheduled, work out time remaining */
638		ret = timeout - (jiffies - start);
639		if (ret < 0)
640			ret = 0;
641
642		goto out;
 
 
 
643	}
644
645	/* Sleep until uncongested or a write happens */
646	prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
647	ret = io_schedule_timeout(timeout);
648	finish_wait(wqh, &wait);
 
 
649
650out:
651	trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
652					jiffies_to_usecs(jiffies - start));
 
653
654	return ret;
 
 
 
655}
656EXPORT_SYMBOL(wait_iff_congested);
657
658int pdflush_proc_obsolete(struct ctl_table *table, int write,
659			void __user *buffer, size_t *lenp, loff_t *ppos)
660{
661	char kbuf[] = "0\n";
 
 
662
663	if (*ppos || *lenp < sizeof(kbuf)) {
664		*lenp = 0;
665		return 0;
666	}
667
668	if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
669		return -EFAULT;
670	printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
671			table->procname);
672
673	*lenp = 2;
674	*ppos += *lenp;
675	return 2;
 
 
 
 
 
 
 
 
 
676}
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2
   3#include <linux/blkdev.h>
   4#include <linux/wait.h>
   5#include <linux/rbtree.h>
   6#include <linux/kthread.h>
   7#include <linux/backing-dev.h>
   8#include <linux/blk-cgroup.h>
   9#include <linux/freezer.h>
  10#include <linux/fs.h>
  11#include <linux/pagemap.h>
  12#include <linux/mm.h>
  13#include <linux/sched/mm.h>
  14#include <linux/sched.h>
  15#include <linux/module.h>
  16#include <linux/writeback.h>
  17#include <linux/device.h>
  18#include <trace/events/writeback.h>
  19#include "internal.h"
  20
  21struct backing_dev_info noop_backing_dev_info;
 
 
 
 
 
 
 
 
 
 
 
 
 
  22EXPORT_SYMBOL_GPL(noop_backing_dev_info);
  23
  24static const char *bdi_unknown_name = "(unknown)";
  25
  26/*
  27 * bdi_lock protects bdi_tree and updates to bdi_list. bdi_list has RCU
  28 * reader side locking.
  29 */
  30DEFINE_SPINLOCK(bdi_lock);
  31static u64 bdi_id_cursor;
  32static struct rb_root bdi_tree = RB_ROOT;
  33LIST_HEAD(bdi_list);
  34
  35/* bdi_wq serves all asynchronous writeback tasks */
  36struct workqueue_struct *bdi_wq;
  37
 
 
 
 
 
 
 
 
 
 
 
  38#ifdef CONFIG_DEBUG_FS
  39#include <linux/debugfs.h>
  40#include <linux/seq_file.h>
  41
  42static struct dentry *bdi_debug_root;
  43
  44static void bdi_debug_init(void)
  45{
  46	bdi_debug_root = debugfs_create_dir("bdi", NULL);
  47}
  48
  49static int bdi_debug_stats_show(struct seq_file *m, void *v)
  50{
  51	struct backing_dev_info *bdi = m->private;
  52	struct bdi_writeback *wb = &bdi->wb;
  53	unsigned long background_thresh;
  54	unsigned long dirty_thresh;
  55	unsigned long wb_thresh;
  56	unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
  57	struct inode *inode;
  58
  59	nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
  60	spin_lock(&wb->list_lock);
  61	list_for_each_entry(inode, &wb->b_dirty, i_io_list)
  62		nr_dirty++;
  63	list_for_each_entry(inode, &wb->b_io, i_io_list)
  64		nr_io++;
  65	list_for_each_entry(inode, &wb->b_more_io, i_io_list)
  66		nr_more_io++;
  67	list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
  68		if (inode->i_state & I_DIRTY_TIME)
  69			nr_dirty_time++;
  70	spin_unlock(&wb->list_lock);
  71
  72	global_dirty_limits(&background_thresh, &dirty_thresh);
  73	wb_thresh = wb_calc_thresh(wb, dirty_thresh);
  74
 
  75	seq_printf(m,
  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"
  84		   "b_dirty:            %10lu\n"
  85		   "b_io:               %10lu\n"
  86		   "b_more_io:          %10lu\n"
  87		   "b_dirty_time:       %10lu\n"
  88		   "bdi_list:           %10u\n"
  89		   "state:              %10lx\n",
  90		   (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
  91		   (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
  92		   K(wb_thresh),
  93		   K(dirty_thresh),
  94		   K(background_thresh),
  95		   (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
  96		   (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
  97		   (unsigned long) K(wb->write_bandwidth),
  98		   nr_dirty,
  99		   nr_io,
 100		   nr_more_io,
 101		   nr_dirty_time,
 102		   !list_empty(&bdi->bdi_list), bdi->wb.state);
 103
 104	return 0;
 105}
 106DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats);
 
 
 
 
 
 
 
 
 
 
 
 107
 108static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
 109{
 110	bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
 111
 112	debugfs_create_file("stats", 0444, bdi->debug_dir, bdi,
 113			    &bdi_debug_stats_fops);
 114}
 115
 116static void bdi_debug_unregister(struct backing_dev_info *bdi)
 117{
 118	debugfs_remove_recursive(bdi->debug_dir);
 
 119}
 120#else
 121static inline void bdi_debug_init(void)
 122{
 123}
 124static inline void bdi_debug_register(struct backing_dev_info *bdi,
 125				      const char *name)
 126{
 127}
 128static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
 129{
 130}
 131#endif
 132
 133static ssize_t read_ahead_kb_store(struct device *dev,
 134				  struct device_attribute *attr,
 135				  const char *buf, size_t count)
 136{
 137	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 138	unsigned long read_ahead_kb;
 139	ssize_t ret;
 140
 141	ret = kstrtoul(buf, 10, &read_ahead_kb);
 142	if (ret < 0)
 143		return ret;
 144
 145	bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
 146
 147	return count;
 148}
 149
 
 
 150#define BDI_SHOW(name, expr)						\
 151static ssize_t name##_show(struct device *dev,				\
 152			   struct device_attribute *attr, char *buf)	\
 153{									\
 154	struct backing_dev_info *bdi = dev_get_drvdata(dev);		\
 155									\
 156	return sysfs_emit(buf, "%lld\n", (long long)expr);		\
 157}									\
 158static DEVICE_ATTR_RW(name);
 159
 160BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
 161
 162static ssize_t min_ratio_store(struct device *dev,
 163		struct device_attribute *attr, const char *buf, size_t count)
 164{
 165	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 166	unsigned int ratio;
 167	ssize_t ret;
 168
 169	ret = kstrtouint(buf, 10, &ratio);
 170	if (ret < 0)
 171		return ret;
 172
 173	ret = bdi_set_min_ratio(bdi, ratio);
 174	if (!ret)
 175		ret = count;
 176
 177	return ret;
 178}
 179BDI_SHOW(min_ratio, bdi->min_ratio / BDI_RATIO_SCALE)
 180
 181static ssize_t min_ratio_fine_store(struct device *dev,
 182		struct device_attribute *attr, const char *buf, size_t count)
 183{
 184	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 185	unsigned int ratio;
 186	ssize_t ret;
 187
 188	ret = kstrtouint(buf, 10, &ratio);
 189	if (ret < 0)
 190		return ret;
 191
 192	ret = bdi_set_min_ratio_no_scale(bdi, ratio);
 193	if (!ret)
 194		ret = count;
 195
 196	return ret;
 197}
 198BDI_SHOW(min_ratio_fine, bdi->min_ratio)
 199
 200static ssize_t max_ratio_store(struct device *dev,
 201		struct device_attribute *attr, const char *buf, size_t count)
 202{
 203	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 204	unsigned int ratio;
 205	ssize_t ret;
 206
 207	ret = kstrtouint(buf, 10, &ratio);
 208	if (ret < 0)
 209		return ret;
 210
 211	ret = bdi_set_max_ratio(bdi, ratio);
 212	if (!ret)
 213		ret = count;
 214
 215	return ret;
 216}
 217BDI_SHOW(max_ratio, bdi->max_ratio / BDI_RATIO_SCALE)
 218
 219static ssize_t max_ratio_fine_store(struct device *dev,
 220		struct device_attribute *attr, const char *buf, size_t count)
 221{
 222	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 223	unsigned int ratio;
 224	ssize_t ret;
 225
 226	ret = kstrtouint(buf, 10, &ratio);
 227	if (ret < 0)
 228		return ret;
 229
 230	ret = bdi_set_max_ratio_no_scale(bdi, ratio);
 231	if (!ret)
 232		ret = count;
 233
 234	return ret;
 235}
 236BDI_SHOW(max_ratio_fine, bdi->max_ratio)
 237
 238static ssize_t min_bytes_show(struct device *dev,
 239			      struct device_attribute *attr,
 240			      char *buf)
 241{
 242	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 243
 244	return sysfs_emit(buf, "%llu\n", bdi_get_min_bytes(bdi));
 245}
 246
 247static ssize_t min_bytes_store(struct device *dev,
 248		struct device_attribute *attr, const char *buf, size_t count)
 249{
 250	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 251	u64 bytes;
 252	ssize_t ret;
 253
 254	ret = kstrtoull(buf, 10, &bytes);
 255	if (ret < 0)
 256		return ret;
 257
 258	ret = bdi_set_min_bytes(bdi, bytes);
 259	if (!ret)
 260		ret = count;
 261
 262	return ret;
 263}
 264static DEVICE_ATTR_RW(min_bytes);
 265
 266static ssize_t max_bytes_show(struct device *dev,
 267			      struct device_attribute *attr,
 268			      char *buf)
 269{
 270	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 271
 272	return sysfs_emit(buf, "%llu\n", bdi_get_max_bytes(bdi));
 273}
 274
 275static ssize_t max_bytes_store(struct device *dev,
 276		struct device_attribute *attr, const char *buf, size_t count)
 277{
 278	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 279	u64 bytes;
 280	ssize_t ret;
 281
 282	ret = kstrtoull(buf, 10, &bytes);
 283	if (ret < 0)
 284		return ret;
 285
 286	ret = bdi_set_max_bytes(bdi, bytes);
 287	if (!ret)
 288		ret = count;
 289
 290	return ret;
 291}
 292static DEVICE_ATTR_RW(max_bytes);
 293
 294static ssize_t stable_pages_required_show(struct device *dev,
 295					  struct device_attribute *attr,
 296					  char *buf)
 297{
 298	dev_warn_once(dev,
 299		"the stable_pages_required attribute has been removed. Use the stable_writes queue attribute instead.\n");
 300	return sysfs_emit(buf, "%d\n", 0);
 301}
 302static DEVICE_ATTR_RO(stable_pages_required);
 303
 304static ssize_t strict_limit_store(struct device *dev,
 305		struct device_attribute *attr, const char *buf, size_t count)
 306{
 307	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 308	unsigned int strict_limit;
 309	ssize_t ret;
 310
 311	ret = kstrtouint(buf, 10, &strict_limit);
 312	if (ret < 0)
 313		return ret;
 314
 315	ret = bdi_set_strict_limit(bdi, strict_limit);
 316	if (!ret)
 317		ret = count;
 318
 319	return ret;
 320}
 321
 322static ssize_t strict_limit_show(struct device *dev,
 323		struct device_attribute *attr, char *buf)
 324{
 325	struct backing_dev_info *bdi = dev_get_drvdata(dev);
 326
 327	return sysfs_emit(buf, "%d\n",
 328			!!(bdi->capabilities & BDI_CAP_STRICTLIMIT));
 329}
 330static DEVICE_ATTR_RW(strict_limit);
 331
 332static struct attribute *bdi_dev_attrs[] = {
 333	&dev_attr_read_ahead_kb.attr,
 334	&dev_attr_min_ratio.attr,
 335	&dev_attr_min_ratio_fine.attr,
 336	&dev_attr_max_ratio.attr,
 337	&dev_attr_max_ratio_fine.attr,
 338	&dev_attr_min_bytes.attr,
 339	&dev_attr_max_bytes.attr,
 340	&dev_attr_stable_pages_required.attr,
 341	&dev_attr_strict_limit.attr,
 342	NULL,
 343};
 344ATTRIBUTE_GROUPS(bdi_dev);
 345
 346static const struct class bdi_class = {
 347	.name		= "bdi",
 348	.dev_groups	= bdi_dev_groups,
 349};
 350
 351static __init int bdi_class_init(void)
 352{
 353	int ret;
 354
 355	ret = class_register(&bdi_class);
 356	if (ret)
 357		return ret;
 358
 
 359	bdi_debug_init();
 360
 361	return 0;
 362}
 363postcore_initcall(bdi_class_init);
 364
 365static int __init default_bdi_init(void)
 366{
 367	bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_UNBOUND |
 368				 WQ_SYSFS, 0);
 
 
 369	if (!bdi_wq)
 370		return -ENOMEM;
 371	return 0;
 372}
 373subsys_initcall(default_bdi_init);
 374
 375static void wb_update_bandwidth_workfn(struct work_struct *work)
 376{
 377	struct bdi_writeback *wb = container_of(to_delayed_work(work),
 378						struct bdi_writeback, bw_dwork);
 379
 380	wb_update_bandwidth(wb);
 381}
 
 382
 383/*
 384 * Initial write bandwidth: 100 MB/s
 385 */
 386#define INIT_BW		(100 << (20 - PAGE_SHIFT))
 387
 388static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
 389		   gfp_t gfp)
 390{
 391	int i, err;
 392
 393	memset(wb, 0, sizeof(*wb));
 394
 395	wb->bdi = bdi;
 396	wb->last_old_flush = jiffies;
 397	INIT_LIST_HEAD(&wb->b_dirty);
 398	INIT_LIST_HEAD(&wb->b_io);
 399	INIT_LIST_HEAD(&wb->b_more_io);
 400	INIT_LIST_HEAD(&wb->b_dirty_time);
 401	spin_lock_init(&wb->list_lock);
 402
 403	atomic_set(&wb->writeback_inodes, 0);
 404	wb->bw_time_stamp = jiffies;
 405	wb->balanced_dirty_ratelimit = INIT_BW;
 406	wb->dirty_ratelimit = INIT_BW;
 407	wb->write_bandwidth = INIT_BW;
 408	wb->avg_write_bandwidth = INIT_BW;
 409
 410	spin_lock_init(&wb->work_lock);
 411	INIT_LIST_HEAD(&wb->work_list);
 412	INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
 413	INIT_DELAYED_WORK(&wb->bw_dwork, wb_update_bandwidth_workfn);
 414
 415	err = fprop_local_init_percpu(&wb->completions, gfp);
 416	if (err)
 417		return err;
 418
 419	for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
 420		err = percpu_counter_init(&wb->stat[i], 0, gfp);
 421		if (err)
 422			goto out_destroy_stat;
 423	}
 424
 425	return 0;
 426
 427out_destroy_stat:
 428	while (i--)
 429		percpu_counter_destroy(&wb->stat[i]);
 430	fprop_local_destroy_percpu(&wb->completions);
 431	return err;
 432}
 433
 434static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
 435
 436/*
 437 * Remove bdi from the global list and shutdown any threads we have running
 
 
 
 
 
 
 
 
 
 
 
 438 */
 439static void wb_shutdown(struct bdi_writeback *wb)
 440{
 441	/* Make sure nobody queues further work */
 442	spin_lock_irq(&wb->work_lock);
 443	if (!test_and_clear_bit(WB_registered, &wb->state)) {
 444		spin_unlock_irq(&wb->work_lock);
 445		return;
 446	}
 447	spin_unlock_irq(&wb->work_lock);
 448
 449	cgwb_remove_from_bdi_list(wb);
 450	/*
 451	 * Drain work list and shutdown the delayed_work.  !WB_registered
 452	 * tells wb_workfn() that @wb is dying and its work_list needs to
 453	 * be drained no matter what.
 454	 */
 455	mod_delayed_work(bdi_wq, &wb->dwork, 0);
 456	flush_delayed_work(&wb->dwork);
 457	WARN_ON(!list_empty(&wb->work_list));
 458	flush_delayed_work(&wb->bw_dwork);
 459}
 460
 461static void wb_exit(struct bdi_writeback *wb)
 462{
 463	int i;
 464
 465	WARN_ON(delayed_work_pending(&wb->dwork));
 466
 467	for (i = 0; i < NR_WB_STAT_ITEMS; i++)
 468		percpu_counter_destroy(&wb->stat[i]);
 469
 470	fprop_local_destroy_percpu(&wb->completions);
 
 471}
 472
 473#ifdef CONFIG_CGROUP_WRITEBACK
 474
 475#include <linux/memcontrol.h>
 476
 477/*
 478 * cgwb_lock protects bdi->cgwb_tree, blkcg->cgwb_list, offline_cgwbs and
 479 * memcg->cgwb_list.  bdi->cgwb_tree is also RCU protected.
 480 */
 481static DEFINE_SPINLOCK(cgwb_lock);
 482static struct workqueue_struct *cgwb_release_wq;
 483
 484static LIST_HEAD(offline_cgwbs);
 485static void cleanup_offline_cgwbs_workfn(struct work_struct *work);
 486static DECLARE_WORK(cleanup_offline_cgwbs_work, cleanup_offline_cgwbs_workfn);
 487
 488static void cgwb_free_rcu(struct rcu_head *rcu_head)
 489{
 490	struct bdi_writeback *wb = container_of(rcu_head,
 491			struct bdi_writeback, rcu);
 
 492
 493	percpu_ref_exit(&wb->refcnt);
 494	kfree(wb);
 495}
 496
 497static void cgwb_release_workfn(struct work_struct *work)
 
 498{
 499	struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
 500						release_work);
 501	struct backing_dev_info *bdi = wb->bdi;
 502
 503	mutex_lock(&wb->bdi->cgwb_release_mutex);
 504	wb_shutdown(wb);
 505
 506	css_put(wb->memcg_css);
 507	css_put(wb->blkcg_css);
 508	mutex_unlock(&wb->bdi->cgwb_release_mutex);
 
 
 509
 510	/* triggers blkg destruction if no online users left */
 511	blkcg_unpin_online(wb->blkcg_css);
 512
 513	fprop_local_destroy_percpu(&wb->memcg_completions);
 
 514
 515	spin_lock_irq(&cgwb_lock);
 516	list_del(&wb->offline_node);
 517	spin_unlock_irq(&cgwb_lock);
 518
 519	wb_exit(wb);
 520	bdi_put(bdi);
 521	WARN_ON_ONCE(!list_empty(&wb->b_attached));
 522	call_rcu(&wb->rcu, cgwb_free_rcu);
 523}
 
 524
 525static void cgwb_release(struct percpu_ref *refcnt)
 526{
 527	struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
 528						refcnt);
 529	queue_work(cgwb_release_wq, &wb->release_work);
 530}
 
 531
 532static void cgwb_kill(struct bdi_writeback *wb)
 
 
 
 533{
 534	lockdep_assert_held(&cgwb_lock);
 
 535
 536	WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
 537	list_del(&wb->memcg_node);
 538	list_del(&wb->blkcg_node);
 539	list_add(&wb->offline_node, &offline_cgwbs);
 540	percpu_ref_kill(&wb->refcnt);
 541}
 542
 543static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
 544{
 545	spin_lock_irq(&cgwb_lock);
 546	list_del_rcu(&wb->bdi_node);
 547	spin_unlock_irq(&cgwb_lock);
 548}
 549
 550static int cgwb_create(struct backing_dev_info *bdi,
 551		       struct cgroup_subsys_state *memcg_css, gfp_t gfp)
 552{
 553	struct mem_cgroup *memcg;
 554	struct cgroup_subsys_state *blkcg_css;
 555	struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
 556	struct bdi_writeback *wb;
 557	unsigned long flags;
 558	int ret = 0;
 559
 560	memcg = mem_cgroup_from_css(memcg_css);
 561	blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
 562	memcg_cgwb_list = &memcg->cgwb_list;
 563	blkcg_cgwb_list = blkcg_get_cgwb_list(blkcg_css);
 564
 565	/* look up again under lock and discard on blkcg mismatch */
 566	spin_lock_irqsave(&cgwb_lock, flags);
 567	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
 568	if (wb && wb->blkcg_css != blkcg_css) {
 569		cgwb_kill(wb);
 570		wb = NULL;
 571	}
 572	spin_unlock_irqrestore(&cgwb_lock, flags);
 573	if (wb)
 574		goto out_put;
 575
 576	/* need to create a new one */
 577	wb = kmalloc(sizeof(*wb), gfp);
 578	if (!wb) {
 579		ret = -ENOMEM;
 580		goto out_put;
 581	}
 582
 583	ret = wb_init(wb, bdi, gfp);
 584	if (ret)
 585		goto err_free;
 586
 587	ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
 588	if (ret)
 589		goto err_wb_exit;
 590
 591	ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
 592	if (ret)
 593		goto err_ref_exit;
 594
 595	wb->memcg_css = memcg_css;
 596	wb->blkcg_css = blkcg_css;
 597	INIT_LIST_HEAD(&wb->b_attached);
 598	INIT_WORK(&wb->release_work, cgwb_release_workfn);
 599	set_bit(WB_registered, &wb->state);
 600	bdi_get(bdi);
 601
 602	/*
 603	 * The root wb determines the registered state of the whole bdi and
 604	 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
 605	 * whether they're still online.  Don't link @wb if any is dead.
 606	 * See wb_memcg_offline() and wb_blkcg_offline().
 607	 */
 608	ret = -ENODEV;
 609	spin_lock_irqsave(&cgwb_lock, flags);
 610	if (test_bit(WB_registered, &bdi->wb.state) &&
 611	    blkcg_cgwb_list->next && memcg_cgwb_list->next) {
 612		/* we might have raced another instance of this function */
 613		ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
 614		if (!ret) {
 615			list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
 616			list_add(&wb->memcg_node, memcg_cgwb_list);
 617			list_add(&wb->blkcg_node, blkcg_cgwb_list);
 618			blkcg_pin_online(blkcg_css);
 619			css_get(memcg_css);
 620			css_get(blkcg_css);
 621		}
 622	}
 623	spin_unlock_irqrestore(&cgwb_lock, flags);
 624	if (ret) {
 625		if (ret == -EEXIST)
 626			ret = 0;
 627		goto err_fprop_exit;
 628	}
 629	goto out_put;
 630
 631err_fprop_exit:
 632	bdi_put(bdi);
 633	fprop_local_destroy_percpu(&wb->memcg_completions);
 634err_ref_exit:
 635	percpu_ref_exit(&wb->refcnt);
 636err_wb_exit:
 637	wb_exit(wb);
 638err_free:
 639	kfree(wb);
 640out_put:
 641	css_put(blkcg_css);
 642	return ret;
 643}
 644
 645/**
 646 * wb_get_lookup - get wb for a given memcg
 647 * @bdi: target bdi
 648 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
 649 *
 650 * Try to get the wb for @memcg_css on @bdi.  The returned wb has its
 651 * refcount incremented.
 652 *
 653 * This function uses css_get() on @memcg_css and thus expects its refcnt
 654 * to be positive on invocation.  IOW, rcu_read_lock() protection on
 655 * @memcg_css isn't enough.  try_get it before calling this function.
 656 *
 657 * A wb is keyed by its associated memcg.  As blkcg implicitly enables
 658 * memcg on the default hierarchy, memcg association is guaranteed to be
 659 * more specific (equal or descendant to the associated blkcg) and thus can
 660 * identify both the memcg and blkcg associations.
 661 *
 662 * Because the blkcg associated with a memcg may change as blkcg is enabled
 663 * and disabled closer to root in the hierarchy, each wb keeps track of
 664 * both the memcg and blkcg associated with it and verifies the blkcg on
 665 * each lookup.  On mismatch, the existing wb is discarded and a new one is
 666 * created.
 667 */
 668struct bdi_writeback *wb_get_lookup(struct backing_dev_info *bdi,
 669				    struct cgroup_subsys_state *memcg_css)
 670{
 671	struct bdi_writeback *wb;
 672
 673	if (!memcg_css->parent)
 674		return &bdi->wb;
 675
 676	rcu_read_lock();
 677	wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
 678	if (wb) {
 679		struct cgroup_subsys_state *blkcg_css;
 680
 681		/* see whether the blkcg association has changed */
 682		blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
 683		if (unlikely(wb->blkcg_css != blkcg_css || !wb_tryget(wb)))
 684			wb = NULL;
 685		css_put(blkcg_css);
 686	}
 687	rcu_read_unlock();
 688
 689	return wb;
 690}
 691
 692/**
 693 * wb_get_create - get wb for a given memcg, create if necessary
 694 * @bdi: target bdi
 695 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
 696 * @gfp: allocation mask to use
 697 *
 698 * Try to get the wb for @memcg_css on @bdi.  If it doesn't exist, try to
 699 * create one.  See wb_get_lookup() for more details.
 700 */
 701struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
 702				    struct cgroup_subsys_state *memcg_css,
 703				    gfp_t gfp)
 704{
 705	struct bdi_writeback *wb;
 706
 707	might_alloc(gfp);
 
 
 
 708
 709	do {
 710		wb = wb_get_lookup(bdi, memcg_css);
 711	} while (!wb && !cgwb_create(bdi, memcg_css, gfp));
 712
 713	return wb;
 714}
 715
 716static int cgwb_bdi_init(struct backing_dev_info *bdi)
 717{
 718	int ret;
 719
 720	INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
 721	mutex_init(&bdi->cgwb_release_mutex);
 722	init_rwsem(&bdi->wb_switch_rwsem);
 723
 724	ret = wb_init(&bdi->wb, bdi, GFP_KERNEL);
 725	if (!ret) {
 726		bdi->wb.memcg_css = &root_mem_cgroup->css;
 727		bdi->wb.blkcg_css = blkcg_root_css;
 728	}
 729	return ret;
 730}
 
 731
 732static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
 733{
 734	struct radix_tree_iter iter;
 735	void **slot;
 736	struct bdi_writeback *wb;
 737
 738	WARN_ON(test_bit(WB_registered, &bdi->wb.state));
 739
 740	spin_lock_irq(&cgwb_lock);
 741	radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
 742		cgwb_kill(*slot);
 743	spin_unlock_irq(&cgwb_lock);
 744
 745	mutex_lock(&bdi->cgwb_release_mutex);
 746	spin_lock_irq(&cgwb_lock);
 747	while (!list_empty(&bdi->wb_list)) {
 748		wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
 749				      bdi_node);
 750		spin_unlock_irq(&cgwb_lock);
 751		wb_shutdown(wb);
 752		spin_lock_irq(&cgwb_lock);
 753	}
 754	spin_unlock_irq(&cgwb_lock);
 755	mutex_unlock(&bdi->cgwb_release_mutex);
 756}
 757
 758/*
 759 * cleanup_offline_cgwbs_workfn - try to release dying cgwbs
 760 *
 761 * Try to release dying cgwbs by switching attached inodes to the nearest
 762 * living ancestor's writeback. Processed wbs are placed at the end
 763 * of the list to guarantee the forward progress.
 764 */
 765static void cleanup_offline_cgwbs_workfn(struct work_struct *work)
 
 
 766{
 767	struct bdi_writeback *wb;
 768	LIST_HEAD(processed);
 769
 770	spin_lock_irq(&cgwb_lock);
 
 
 
 
 
 
 
 771
 772	while (!list_empty(&offline_cgwbs)) {
 773		wb = list_first_entry(&offline_cgwbs, struct bdi_writeback,
 774				      offline_node);
 775		list_move(&wb->offline_node, &processed);
 776
 777		/*
 778		 * If wb is dirty, cleaning up the writeback by switching
 779		 * attached inodes will result in an effective removal of any
 780		 * bandwidth restrictions, which isn't the goal.  Instead,
 781		 * it can be postponed until the next time, when all io
 782		 * will be likely completed.  If in the meantime some inodes
 783		 * will get re-dirtied, they should be eventually switched to
 784		 * a new cgwb.
 785		 */
 786		if (wb_has_dirty_io(wb))
 787			continue;
 788
 789		if (!wb_tryget(wb))
 790			continue;
 791
 792		spin_unlock_irq(&cgwb_lock);
 793		while (cleanup_offline_cgwb(wb))
 794			cond_resched();
 795		spin_lock_irq(&cgwb_lock);
 796
 797		wb_put(wb);
 
 
 
 798	}
 799
 800	if (!list_empty(&processed))
 801		list_splice_tail(&processed, &offline_cgwbs);
 802
 803	spin_unlock_irq(&cgwb_lock);
 804}
 805
 806/**
 807 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
 808 * @memcg: memcg being offlined
 809 *
 810 * Also prevents creation of any new wb's associated with @memcg.
 811 */
 812void wb_memcg_offline(struct mem_cgroup *memcg)
 813{
 814	struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
 815	struct bdi_writeback *wb, *next;
 816
 817	spin_lock_irq(&cgwb_lock);
 818	list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
 819		cgwb_kill(wb);
 820	memcg_cgwb_list->next = NULL;	/* prevent new wb's */
 821	spin_unlock_irq(&cgwb_lock);
 822
 823	queue_work(system_unbound_wq, &cleanup_offline_cgwbs_work);
 824}
 
 
 
 825
 826/**
 827 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
 828 * @css: blkcg being offlined
 829 *
 830 * Also prevents creation of any new wb's associated with @blkcg.
 831 */
 832void wb_blkcg_offline(struct cgroup_subsys_state *css)
 833{
 834	struct bdi_writeback *wb, *next;
 835	struct list_head *list = blkcg_get_cgwb_list(css);
 836
 837	spin_lock_irq(&cgwb_lock);
 838	list_for_each_entry_safe(wb, next, list, blkcg_node)
 839		cgwb_kill(wb);
 840	list->next = NULL;	/* prevent new wb's */
 841	spin_unlock_irq(&cgwb_lock);
 842}
 
 843
 844static void cgwb_bdi_register(struct backing_dev_info *bdi)
 845{
 846	spin_lock_irq(&cgwb_lock);
 847	list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
 848	spin_unlock_irq(&cgwb_lock);
 849}
 850
 851static int __init cgwb_init(void)
 852{
 853	/*
 854	 * There can be many concurrent release work items overwhelming
 855	 * system_wq.  Put them in a separate wq and limit concurrency.
 856	 * There's no point in executing many of these in parallel.
 857	 */
 858	cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
 859	if (!cgwb_release_wq)
 860		return -ENOMEM;
 861
 862	return 0;
 863}
 864subsys_initcall(cgwb_init);
 
 
 
 
 865
 866#else	/* CONFIG_CGROUP_WRITEBACK */
 867
 868static int cgwb_bdi_init(struct backing_dev_info *bdi)
 869{
 870	return wb_init(&bdi->wb, bdi, GFP_KERNEL);
 871}
 
 
 872
 873static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
 
 874
 875static void cgwb_bdi_register(struct backing_dev_info *bdi)
 876{
 877	list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
 878}
 
 879
 880static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
 
 
 
 
 
 881{
 882	list_del_rcu(&wb->bdi_node);
 883}
 884
 885#endif	/* CONFIG_CGROUP_WRITEBACK */
 
 
 
 
 886
 887int bdi_init(struct backing_dev_info *bdi)
 888{
 889	bdi->dev = NULL;
 890
 891	kref_init(&bdi->refcnt);
 892	bdi->min_ratio = 0;
 893	bdi->max_ratio = 100 * BDI_RATIO_SCALE;
 894	bdi->max_prop_frac = FPROP_FRAC_BASE;
 895	INIT_LIST_HEAD(&bdi->bdi_list);
 896	INIT_LIST_HEAD(&bdi->wb_list);
 897	init_waitqueue_head(&bdi->wb_waitq);
 898	bdi->last_bdp_sleep = jiffies;
 899
 900	return cgwb_bdi_init(bdi);
 901}
 902
 903struct backing_dev_info *bdi_alloc(int node_id)
 904{
 905	struct backing_dev_info *bdi;
 906
 907	bdi = kzalloc_node(sizeof(*bdi), GFP_KERNEL, node_id);
 908	if (!bdi)
 909		return NULL;
 910
 911	if (bdi_init(bdi)) {
 912		kfree(bdi);
 913		return NULL;
 914	}
 915	bdi->capabilities = BDI_CAP_WRITEBACK | BDI_CAP_WRITEBACK_ACCT;
 916	bdi->ra_pages = VM_READAHEAD_PAGES;
 917	bdi->io_pages = VM_READAHEAD_PAGES;
 918	timer_setup(&bdi->laptop_mode_wb_timer, laptop_mode_timer_fn, 0);
 919	return bdi;
 920}
 921EXPORT_SYMBOL(bdi_alloc);
 922
 923static struct rb_node **bdi_lookup_rb_node(u64 id, struct rb_node **parentp)
 924{
 925	struct rb_node **p = &bdi_tree.rb_node;
 926	struct rb_node *parent = NULL;
 927	struct backing_dev_info *bdi;
 928
 929	lockdep_assert_held(&bdi_lock);
 930
 931	while (*p) {
 932		parent = *p;
 933		bdi = rb_entry(parent, struct backing_dev_info, rb_node);
 934
 935		if (bdi->id > id)
 936			p = &(*p)->rb_left;
 937		else if (bdi->id < id)
 938			p = &(*p)->rb_right;
 939		else
 940			break;
 941	}
 942
 943	if (parentp)
 944		*parentp = parent;
 945	return p;
 946}
 
 947
 948/**
 949 * bdi_get_by_id - lookup and get bdi from its id
 950 * @id: bdi id to lookup
 951 *
 952 * Find bdi matching @id and get it.  Returns NULL if the matching bdi
 953 * doesn't exist or is already unregistered.
 954 */
 955struct backing_dev_info *bdi_get_by_id(u64 id)
 956{
 957	struct backing_dev_info *bdi = NULL;
 958	struct rb_node **p;
 959
 960	spin_lock_bh(&bdi_lock);
 961	p = bdi_lookup_rb_node(id, NULL);
 962	if (*p) {
 963		bdi = rb_entry(*p, struct backing_dev_info, rb_node);
 964		bdi_get(bdi);
 965	}
 966	spin_unlock_bh(&bdi_lock);
 967
 968	return bdi;
 969}
 970
 971int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
 972{
 973	struct device *dev;
 974	struct rb_node *parent, **p;
 975
 976	if (bdi->dev)	/* The driver needs to use separate queues per device */
 977		return 0;
 978
 979	vsnprintf(bdi->dev_name, sizeof(bdi->dev_name), fmt, args);
 980	dev = device_create(&bdi_class, NULL, MKDEV(0, 0), bdi, bdi->dev_name);
 981	if (IS_ERR(dev))
 982		return PTR_ERR(dev);
 983
 984	cgwb_bdi_register(bdi);
 985	bdi->dev = dev;
 986
 987	bdi_debug_register(bdi, dev_name(dev));
 988	set_bit(WB_registered, &bdi->wb.state);
 989
 990	spin_lock_bh(&bdi_lock);
 991
 992	bdi->id = ++bdi_id_cursor;
 993
 994	p = bdi_lookup_rb_node(bdi->id, &parent);
 995	rb_link_node(&bdi->rb_node, parent, p);
 996	rb_insert_color(&bdi->rb_node, &bdi_tree);
 997
 998	list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
 999
1000	spin_unlock_bh(&bdi_lock);
1001
1002	trace_writeback_bdi_register(bdi);
1003	return 0;
1004}
 
1005
1006int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
1007{
1008	va_list args;
1009	int ret;
1010
1011	va_start(args, fmt);
1012	ret = bdi_register_va(bdi, fmt, args);
1013	va_end(args);
1014	return ret;
1015}
1016EXPORT_SYMBOL(bdi_register);
1017
1018void bdi_set_owner(struct backing_dev_info *bdi, struct device *owner)
1019{
1020	WARN_ON_ONCE(bdi->owner);
1021	bdi->owner = owner;
1022	get_device(owner);
1023}
 
 
 
 
 
 
 
 
 
 
 
 
 
1024
1025/*
1026 * Remove bdi from bdi_list, and ensure that it is no longer visible
1027 */
1028static void bdi_remove_from_list(struct backing_dev_info *bdi)
1029{
1030	spin_lock_bh(&bdi_lock);
1031	rb_erase(&bdi->rb_node, &bdi_tree);
1032	list_del_rcu(&bdi->bdi_list);
1033	spin_unlock_bh(&bdi_lock);
1034
1035	synchronize_rcu_expedited();
1036}
 
1037
1038void bdi_unregister(struct backing_dev_info *bdi)
1039{
1040	del_timer_sync(&bdi->laptop_mode_wb_timer);
1041
1042	/* make sure nobody finds us on the bdi_list anymore */
1043	bdi_remove_from_list(bdi);
1044	wb_shutdown(&bdi->wb);
1045	cgwb_bdi_unregister(bdi);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1046
1047	/*
1048	 * If this BDI's min ratio has been set, use bdi_set_min_ratio() to
1049	 * update the global bdi_min_ratio.
 
1050	 */
1051	if (bdi->min_ratio)
1052		bdi_set_min_ratio(bdi, 0);
 
 
 
 
 
 
1053
1054	if (bdi->dev) {
1055		bdi_debug_unregister(bdi);
1056		device_unregister(bdi->dev);
1057		bdi->dev = NULL;
1058	}
1059
1060	if (bdi->owner) {
1061		put_device(bdi->owner);
1062		bdi->owner = NULL;
1063	}
1064}
1065EXPORT_SYMBOL(bdi_unregister);
1066
1067static void release_bdi(struct kref *ref)
1068{
1069	struct backing_dev_info *bdi =
1070			container_of(ref, struct backing_dev_info, refcnt);
1071
1072	WARN_ON_ONCE(test_bit(WB_registered, &bdi->wb.state));
1073	WARN_ON_ONCE(bdi->dev);
1074	wb_exit(&bdi->wb);
1075	kfree(bdi);
1076}
 
1077
1078void bdi_put(struct backing_dev_info *bdi)
 
1079{
1080	kref_put(&bdi->refcnt, release_bdi);
1081}
1082EXPORT_SYMBOL(bdi_put);
1083
1084struct backing_dev_info *inode_to_bdi(struct inode *inode)
1085{
1086	struct super_block *sb;
 
1087
1088	if (!inode)
1089		return &noop_backing_dev_info;
1090
1091	sb = inode->i_sb;
1092#ifdef CONFIG_BLOCK
1093	if (sb_is_blkdev_sb(sb))
1094		return I_BDEV(inode)->bd_disk->bdi;
1095#endif
1096	return sb->s_bdi;
1097}
1098EXPORT_SYMBOL(inode_to_bdi);
1099
1100const char *bdi_dev_name(struct backing_dev_info *bdi)
1101{
1102	if (!bdi || !bdi->dev)
1103		return bdi_unknown_name;
1104	return bdi->dev_name;
1105}
1106EXPORT_SYMBOL_GPL(bdi_dev_name);