Kernel

Kernel Git Source Tree

Root/fs/fs-writeback.c

Source at commit 9845c1745d3d531a5b9544f5322c62bfb4d4e9bc created 1 year 2 months ago. By Xiangfu, rtc: jz4740 fix hwclock give time out
1	/*
2	* fs/fs-writeback.c
3	*
4	* Copyright (C) 2002, Linus Torvalds.
5	*
6	* Contains all the functions related to writing back and waiting
7	* upon dirty inodes against superblocks, and writing back dirty
8	* pages against inodes. ie: data writeback. Writeout of the
9	* inode itself is not handled here.
10	*
11	* 10Apr2002 Andrew Morton
12	* Split out of fs/inode.c
13	* Additions for address_space-based writeback
14	*/
15
16	#include <linux/kernel.h>
17	#include <linux/module.h>
18	#include <linux/spinlock.h>
19	#include <linux/slab.h>
20	#include <linux/sched.h>
21	#include <linux/fs.h>
22	#include <linux/mm.h>
23	#include <linux/kthread.h>
24	#include <linux/freezer.h>
25	#include <linux/writeback.h>
26	#include <linux/blkdev.h>
27	#include <linux/backing-dev.h>
28	#include <linux/buffer_head.h>
29	#include <linux/tracepoint.h>
30	#include "internal.h"
31
32	/*
33	* Passed into wb_writeback(), essentially a subset of writeback_control
34	*/
35	struct wb_writeback_work {
36	long nr_pages;
37	struct super_block *sb;
38	unsigned long *older_than_this;
39	enum writeback_sync_modes sync_mode;
40	unsigned int tagged_writepages:1;
41	unsigned int for_kupdate:1;
42	unsigned int range_cyclic:1;
43	unsigned int for_background:1;
44	enum wb_reason reason; /* why was writeback initiated? */
45
46	struct list_head list; /* pending work list */
47	struct completion done; / set if the caller waits */
48	};
49
50	/*
51	* Include the creation of the trace points after defining the
52	* wb_writeback_work structure so that the definition remains local to this
53	* file.
54	*/
55	#define CREATE_TRACE_POINTS
56	#include <trace/events/writeback.h>
57
58	/*
59	* We don't actually have pdflush, but this one is exported though /proc...
60	*/
61	int nr_pdflush_threads;
62
63	/**
64	* writeback_in_progress - determine whether there is writeback in progress
65	* @bdi: the device's backing_dev_info structure.
66	*
67	* Determine whether there is writeback waiting to be handled against a
68	* backing device.
69	*/
70	int writeback_in_progress(struct backing_dev_info *bdi)
71	{
72	return test_bit(BDI_writeback_running, &bdi->state);
73	}
74
75	static inline struct backing_dev_info inode_to_bdi(struct inode inode)
76	{
77	struct super_block *sb = inode->i_sb;
78
79	if (strcmp(sb->s_type->name, "bdev") == 0)
80	return inode->i_mapping->backing_dev_info;
81
82	return sb->s_bdi;
83	}
84
85	static inline struct inode wb_inode(struct list_head head)
86	{
87	return list_entry(head, struct inode, i_wb_list);
88	}
89
90	/* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
91	static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
92	{
93	if (bdi->wb.task) {
94	wake_up_process(bdi->wb.task);
95	} else {
96	/*
97	* The bdi thread isn't there, wake up the forker thread which
98	* will create and run it.
99	*/
100	wake_up_process(default_backing_dev_info.wb.task);
101	}
102	}
103
104	static void bdi_queue_work(struct backing_dev_info *bdi,
105	struct wb_writeback_work *work)
106	{
107	trace_writeback_queue(bdi, work);
108
109	spin_lock_bh(&bdi->wb_lock);
110	list_add_tail(&work->list, &bdi->work_list);
111	if (!bdi->wb.task)
112	trace_writeback_nothread(bdi, work);
113	bdi_wakeup_flusher(bdi);
114	spin_unlock_bh(&bdi->wb_lock);
115	}
116
117	static void
118	__bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
119	bool range_cyclic, enum wb_reason reason)
120	{
121	struct wb_writeback_work *work;
122
123	/*
124	* This is WB_SYNC_NONE writeback, so if allocation fails just
125	* wakeup the thread for old dirty data writeback
126	*/
127	work = kzalloc(sizeof(*work), GFP_ATOMIC);
128	if (!work) {
129	if (bdi->wb.task) {
130	trace_writeback_nowork(bdi);
131	wake_up_process(bdi->wb.task);
132	}
133	return;
134	}
135
136	work->sync_mode = WB_SYNC_NONE;
137	work->nr_pages = nr_pages;
138	work->range_cyclic = range_cyclic;
139	work->reason = reason;
140
141	bdi_queue_work(bdi, work);
142	}
143
144	/**
145	* bdi_start_writeback - start writeback
146	* @bdi: the backing device to write from
147	* @nr_pages: the number of pages to write
148	* @reason: reason why some writeback work was initiated
149	*
150	* Description:
151	* This does WB_SYNC_NONE opportunistic writeback. The IO is only
152	* started when this function returns, we make no guarantees on
153	* completion. Caller need not hold sb s_umount semaphore.
154	*
155	*/
156	void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
157	enum wb_reason reason)
158	{
159	__bdi_start_writeback(bdi, nr_pages, true, reason);
160	}
161
162	/**
163	* bdi_start_background_writeback - start background writeback
164	* @bdi: the backing device to write from
165	*
166	* Description:
167	* This makes sure WB_SYNC_NONE background writeback happens. When
168	* this function returns, it is only guaranteed that for given BDI
169	* some IO is happening if we are over background dirty threshold.
170	* Caller need not hold sb s_umount semaphore.
171	*/
172	void bdi_start_background_writeback(struct backing_dev_info *bdi)
173	{
174	/*
175	* We just wake up the flusher thread. It will perform background
176	* writeback as soon as there is no other work to do.
177	*/
178	trace_writeback_wake_background(bdi);
179	spin_lock_bh(&bdi->wb_lock);
180	bdi_wakeup_flusher(bdi);
181	spin_unlock_bh(&bdi->wb_lock);
182	}
183
184	/*
185	* Remove the inode from the writeback list it is on.
186	*/
187	void inode_wb_list_del(struct inode *inode)
188	{
189	struct backing_dev_info *bdi = inode_to_bdi(inode);
190
191	spin_lock(&bdi->wb.list_lock);
192	list_del_init(&inode->i_wb_list);
193	spin_unlock(&bdi->wb.list_lock);
194	}
195
196	/*
197	* Redirty an inode: set its when-it-was dirtied timestamp and move it to the
198	* furthest end of its superblock's dirty-inode list.
199	*
200	* Before stamping the inode's ->dirtied_when, we check to see whether it is
201	* already the most-recently-dirtied inode on the b_dirty list. If that is
202	* the case then the inode must have been redirtied while it was being written
203	* out and we don't reset its dirtied_when.
204	*/
205	static void redirty_tail(struct inode inode, struct bdi_writeback wb)
206	{
207	assert_spin_locked(&wb->list_lock);
208	if (!list_empty(&wb->b_dirty)) {
209	struct inode *tail;
210
211	tail = wb_inode(wb->b_dirty.next);
212	if (time_before(inode->dirtied_when, tail->dirtied_when))
213	inode->dirtied_when = jiffies;
214	}
215	list_move(&inode->i_wb_list, &wb->b_dirty);
216	}
217
218	/*
219	* requeue inode for re-scanning after bdi->b_io list is exhausted.
220	*/
221	static void requeue_io(struct inode inode, struct bdi_writeback wb)
222	{
223	assert_spin_locked(&wb->list_lock);
224	list_move(&inode->i_wb_list, &wb->b_more_io);
225	}
226
227	static void inode_sync_complete(struct inode *inode)
228	{
229	/*
230	* Prevent speculative execution through
231	* spin_unlock(&wb->list_lock);
232	*/
233
234	smp_mb();
235	wake_up_bit(&inode->i_state, __I_SYNC);
236	}
237
238	static bool inode_dirtied_after(struct inode *inode, unsigned long t)
239	{
240	bool ret = time_after(inode->dirtied_when, t);
241	#ifndef CONFIG_64BIT
242	/*
243	* For inodes being constantly redirtied, dirtied_when can get stuck.
244	* It _appears_ to be in the future, but is actually in distant past.
245	* This test is necessary to prevent such wrapped-around relative times
246	* from permanently stopping the whole bdi writeback.
247	*/
248	ret = ret && time_before_eq(inode->dirtied_when, jiffies);
249	#endif
250	return ret;
251	}
252
253	/*
254	* Move expired dirty inodes from @delaying_queue to @dispatch_queue.
255	*/
256	static int move_expired_inodes(struct list_head *delaying_queue,
257	struct list_head *dispatch_queue,
258	struct wb_writeback_work *work)
259	{
260	LIST_HEAD(tmp);
261	struct list_head pos, node;
262	struct super_block *sb = NULL;
263	struct inode *inode;
264	int do_sb_sort = 0;
265	int moved = 0;
266
267	while (!list_empty(delaying_queue)) {
268	inode = wb_inode(delaying_queue->prev);
269	if (work->older_than_this &&
270	inode_dirtied_after(inode, *work->older_than_this))
271	break;
272	if (sb && sb != inode->i_sb)
273	do_sb_sort = 1;
274	sb = inode->i_sb;
275	list_move(&inode->i_wb_list, &tmp);
276	moved++;
277	}
278
279	/* just one sb in list, splice to dispatch_queue and we're done */
280	if (!do_sb_sort) {
281	list_splice(&tmp, dispatch_queue);
282	goto out;
283	}
284
285	/* Move inodes from one superblock together */
286	while (!list_empty(&tmp)) {
287	sb = wb_inode(tmp.prev)->i_sb;
288	list_for_each_prev_safe(pos, node, &tmp) {
289	inode = wb_inode(pos);
290	if (inode->i_sb == sb)
291	list_move(&inode->i_wb_list, dispatch_queue);
292	}
293	}
294	out:
295	return moved;
296	}
297
298	/*
299	* Queue all expired dirty inodes for io, eldest first.
300	* Before
301	* newly dirtied b_dirty b_io b_more_io
302	* =============> gf edc BA
303	* After
304	* newly dirtied b_dirty b_io b_more_io
305	* =============> g fBAedc
306	* \|
307	* +--> dequeue for IO
308	*/
309	static void queue_io(struct bdi_writeback wb, struct wb_writeback_work work)
310	{
311	int moved;
312	assert_spin_locked(&wb->list_lock);
313	list_splice_init(&wb->b_more_io, &wb->b_io);
314	moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
315	trace_writeback_queue_io(wb, work, moved);
316	}
317
318	static int write_inode(struct inode inode, struct writeback_control wbc)
319	{
320	if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
321	return inode->i_sb->s_op->write_inode(inode, wbc);
322	return 0;
323	}
324
325	/*
326	* Wait for writeback on an inode to complete.
327	*/
328	static void inode_wait_for_writeback(struct inode *inode,
329	struct bdi_writeback *wb)
330	{
331	DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
332	wait_queue_head_t *wqh;
333
334	wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
335	while (inode->i_state & I_SYNC) {
336	spin_unlock(&inode->i_lock);
337	spin_unlock(&wb->list_lock);
338	__wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
339	spin_lock(&wb->list_lock);
340	spin_lock(&inode->i_lock);
341	}
342	}
343
344	/*
345	* Write out an inode's dirty pages. Called under wb->list_lock and
346	* inode->i_lock. Either the caller has an active reference on the inode or
347	* the inode has I_WILL_FREE set.
348	*
349	* If `wait' is set, wait on the writeout.
350	*
351	* The whole writeout design is quite complex and fragile. We want to avoid
352	* starvation of particular inodes when others are being redirtied, prevent
353	* livelocks, etc.
354	*/
355	static int
356	writeback_single_inode(struct inode inode, struct bdi_writeback wb,
357	struct writeback_control *wbc)
358	{
359	struct address_space *mapping = inode->i_mapping;
360	long nr_to_write = wbc->nr_to_write;
361	unsigned dirty;
362	int ret;
363
364	assert_spin_locked(&wb->list_lock);
365	assert_spin_locked(&inode->i_lock);
366
367	if (!atomic_read(&inode->i_count))
368	WARN_ON(!(inode->i_state & (I_WILL_FREE\|I_FREEING)));
369	else
370	WARN_ON(inode->i_state & I_WILL_FREE);
371
372	if (inode->i_state & I_SYNC) {
373	/*
374	* If this inode is locked for writeback and we are not doing
375	* writeback-for-data-integrity, move it to b_more_io so that
376	* writeback can proceed with the other inodes on s_io.
377	*
378	* We'll have another go at writing back this inode when we
379	* completed a full scan of b_io.
380	*/
381	if (wbc->sync_mode != WB_SYNC_ALL) {
382	requeue_io(inode, wb);
383	trace_writeback_single_inode_requeue(inode, wbc,
384	nr_to_write);
385	return 0;
386	}
387
388	/*
389	* It's a data-integrity sync. We must wait.
390	*/
391	inode_wait_for_writeback(inode, wb);
392	}
393
394	BUG_ON(inode->i_state & I_SYNC);
395
396	/* Set I_SYNC, reset I_DIRTY_PAGES */
397	inode->i_state \|= I_SYNC;
398	inode->i_state &= ~I_DIRTY_PAGES;
399	spin_unlock(&inode->i_lock);
400	spin_unlock(&wb->list_lock);
401
402	ret = do_writepages(mapping, wbc);
403
404	/*
405	* Make sure to wait on the data before writing out the metadata.
406	* This is important for filesystems that modify metadata on data
407	* I/O completion.
408	*/
409	if (wbc->sync_mode == WB_SYNC_ALL) {
410	int err = filemap_fdatawait(mapping);
411	if (ret == 0)
412	ret = err;
413	}
414
415	/*
416	* Some filesystems may redirty the inode during the writeback
417	* due to delalloc, clear dirty metadata flags right before
418	* write_inode()
419	*/
420	spin_lock(&inode->i_lock);
421	dirty = inode->i_state & I_DIRTY;
422	inode->i_state &= ~(I_DIRTY_SYNC \| I_DIRTY_DATASYNC);
423	spin_unlock(&inode->i_lock);
424	/* Don't write the inode if only I_DIRTY_PAGES was set */
425	if (dirty & (I_DIRTY_SYNC \| I_DIRTY_DATASYNC)) {
426	int err = write_inode(inode, wbc);
427	if (ret == 0)
428	ret = err;
429	}
430
431	spin_lock(&wb->list_lock);
432	spin_lock(&inode->i_lock);
433	inode->i_state &= ~I_SYNC;
434	if (!(inode->i_state & I_FREEING)) {
435	/*
436	* Sync livelock prevention. Each inode is tagged and synced in
437	* one shot. If still dirty, it will be redirty_tail()'ed below.
438	* Update the dirty time to prevent enqueue and sync it again.
439	*/
440	if ((inode->i_state & I_DIRTY) &&
441	(wbc->sync_mode == WB_SYNC_ALL \|\| wbc->tagged_writepages))
442	inode->dirtied_when = jiffies;
443
444	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
445	/*
446	* We didn't write back all the pages. nfs_writepages()
447	* sometimes bales out without doing anything.
448	*/
449	inode->i_state \|= I_DIRTY_PAGES;
450	if (wbc->nr_to_write <= 0) {
451	/*
452	* slice used up: queue for next turn
453	*/
454	requeue_io(inode, wb);
455	} else {
456	/*
457	* Writeback blocked by something other than
458	* congestion. Delay the inode for some time to
459	* avoid spinning on the CPU (100% iowait)
460	* retrying writeback of the dirty page/inode
461	* that cannot be performed immediately.
462	*/
463	redirty_tail(inode, wb);
464	}
465	} else if (inode->i_state & I_DIRTY) {
466	/*
467	* Filesystems can dirty the inode during writeback
468	* operations, such as delayed allocation during
469	* submission or metadata updates after data IO
470	* completion.
471	*/
472	redirty_tail(inode, wb);
473	} else {
474	/*
475	* The inode is clean. At this point we either have
476	* a reference to the inode or it's on it's way out.
477	* No need to add it back to the LRU.
478	*/
479	list_del_init(&inode->i_wb_list);
480	}
481	}
482	inode_sync_complete(inode);
483	trace_writeback_single_inode(inode, wbc, nr_to_write);
484	return ret;
485	}
486
487	static long writeback_chunk_size(struct backing_dev_info *bdi,
488	struct wb_writeback_work *work)
489	{
490	long pages;
491
492	/*
493	* WB_SYNC_ALL mode does livelock avoidance by syncing dirty
494	* inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
495	* here avoids calling into writeback_inodes_wb() more than once.
496	*
497	* The intended call sequence for WB_SYNC_ALL writeback is:
498	*
499	* wb_writeback()
500	* writeback_sb_inodes() <== called only once
501	* write_cache_pages() <== called once for each inode
502	* (quickly) tag currently dirty pages
503	* (maybe slowly) sync all tagged pages
504	*/
505	if (work->sync_mode == WB_SYNC_ALL \|\| work->tagged_writepages)
506	pages = LONG_MAX;
507	else {
508	pages = min(bdi->avg_write_bandwidth / 2,
509	global_dirty_limit / DIRTY_SCOPE);
510	pages = min(pages, work->nr_pages);
511	pages = round_down(pages + MIN_WRITEBACK_PAGES,
512	MIN_WRITEBACK_PAGES);
513	}
514
515	return pages;
516	}
517
518	/*
519	* Write a portion of b_io inodes which belong to @sb.
520	*
521	* If @only_this_sb is true, then find and write all such
522	* inodes. Otherwise write only ones which go sequentially
523	* in reverse order.
524	*
525	* Return the number of pages and/or inodes written.
526	*/
527	static long writeback_sb_inodes(struct super_block *sb,
528	struct bdi_writeback *wb,
529	struct wb_writeback_work *work)
530	{
531	struct writeback_control wbc = {
532	.sync_mode = work->sync_mode,
533	.tagged_writepages = work->tagged_writepages,
534	.for_kupdate = work->for_kupdate,
535	.for_background = work->for_background,
536	.range_cyclic = work->range_cyclic,
537	.range_start = 0,
538	.range_end = LLONG_MAX,
539	};
540	unsigned long start_time = jiffies;
541	long write_chunk;
542	long wrote = 0; /* count both pages and inodes */
543
544	while (!list_empty(&wb->b_io)) {
545	struct inode *inode = wb_inode(wb->b_io.prev);
546
547	if (inode->i_sb != sb) {
548	if (work->sb) {
549	/*
550	* We only want to write back data for this
551	* superblock, move all inodes not belonging
552	* to it back onto the dirty list.
553	*/
554	redirty_tail(inode, wb);
555	continue;
556	}
557
558	/*
559	* The inode belongs to a different superblock.
560	* Bounce back to the caller to unpin this and
561	* pin the next superblock.
562	*/
563	break;
564	}
565
566	/*
567	* Don't bother with new inodes or inodes beeing freed, first
568	* kind does not need peridic writeout yet, and for the latter
569	* kind writeout is handled by the freer.
570	*/
571	spin_lock(&inode->i_lock);
572	if (inode->i_state & (I_NEW \| I_FREEING \| I_WILL_FREE)) {
573	spin_unlock(&inode->i_lock);
574	redirty_tail(inode, wb);
575	continue;
576	}
577	__iget(inode);
578	write_chunk = writeback_chunk_size(wb->bdi, work);
579	wbc.nr_to_write = write_chunk;
580	wbc.pages_skipped = 0;
581
582	writeback_single_inode(inode, wb, &wbc);
583
584	work->nr_pages -= write_chunk - wbc.nr_to_write;
585	wrote += write_chunk - wbc.nr_to_write;
586	if (!(inode->i_state & I_DIRTY))
587	wrote++;
588	if (wbc.pages_skipped) {
589	/*
590	* writeback is not making progress due to locked
591	* buffers. Skip this inode for now.
592	*/
593	redirty_tail(inode, wb);
594	}
595	spin_unlock(&inode->i_lock);
596	spin_unlock(&wb->list_lock);
597	iput(inode);
598	cond_resched();
599	spin_lock(&wb->list_lock);
600	/*
601	* bail out to wb_writeback() often enough to check
602	* background threshold and other termination conditions.
603	*/
604	if (wrote) {
605	if (time_is_before_jiffies(start_time + HZ / 10UL))
606	break;
607	if (work->nr_pages <= 0)
608	break;
609	}
610	}
611	return wrote;
612	}
613
614	static long __writeback_inodes_wb(struct bdi_writeback *wb,
615	struct wb_writeback_work *work)
616	{
617	unsigned long start_time = jiffies;
618	long wrote = 0;
619
620	while (!list_empty(&wb->b_io)) {
621	struct inode *inode = wb_inode(wb->b_io.prev);
622	struct super_block *sb = inode->i_sb;
623
624	if (!grab_super_passive(sb)) {
625	/*
626	* grab_super_passive() may fail consistently due to
627	* s_umount being grabbed by someone else. Don't use
628	* requeue_io() to avoid busy retrying the inode/sb.
629	*/
630	redirty_tail(inode, wb);
631	continue;
632	}
633	wrote += writeback_sb_inodes(sb, wb, work);
634	drop_super(sb);
635
636	/* refer to the same tests at the end of writeback_sb_inodes */
637	if (wrote) {
638	if (time_is_before_jiffies(start_time + HZ / 10UL))
639	break;
640	if (work->nr_pages <= 0)
641	break;
642	}
643	}
644	/* Leave any unwritten inodes on b_io */
645	return wrote;
646	}
647
648	long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
649	enum wb_reason reason)
650	{
651	struct wb_writeback_work work = {
652	.nr_pages = nr_pages,
653	.sync_mode = WB_SYNC_NONE,
654	.range_cyclic = 1,
655	.reason = reason,
656	};
657
658	spin_lock(&wb->list_lock);
659	if (list_empty(&wb->b_io))
660	queue_io(wb, &work);
661	__writeback_inodes_wb(wb, &work);
662	spin_unlock(&wb->list_lock);
663
664	return nr_pages - work.nr_pages;
665	}
666
667	static bool over_bground_thresh(struct backing_dev_info *bdi)
668	{
669	unsigned long background_thresh, dirty_thresh;
670
671	global_dirty_limits(&background_thresh, &dirty_thresh);
672
673	if (global_page_state(NR_FILE_DIRTY) +
674	global_page_state(NR_UNSTABLE_NFS) > background_thresh)
675	return true;
676
677	if (bdi_stat(bdi, BDI_RECLAIMABLE) >
678	bdi_dirty_limit(bdi, background_thresh))
679	return true;
680
681	return false;
682	}
683
684	/*
685	* Called under wb->list_lock. If there are multiple wb per bdi,
686	* only the flusher working on the first wb should do it.
687	*/
688	static void wb_update_bandwidth(struct bdi_writeback *wb,
689	unsigned long start_time)
690	{
691	__bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
692	}
693
694	/*
695	* Explicit flushing or periodic writeback of "old" data.
696	*
697	* Define "old": the first time one of an inode's pages is dirtied, we mark the
698	* dirtying-time in the inode's address_space. So this periodic writeback code
699	* just walks the superblock inode list, writing back any inodes which are
700	* older than a specific point in time.
701	*
702	* Try to run once per dirty_writeback_interval. But if a writeback event
703	* takes longer than a dirty_writeback_interval interval, then leave a
704	* one-second gap.
705	*
706	* older_than_this takes precedence over nr_to_write. So we'll only write back
707	* all dirty pages if they are all attached to "old" mappings.
708	*/
709	static long wb_writeback(struct bdi_writeback *wb,
710	struct wb_writeback_work *work)
711	{
712	unsigned long wb_start = jiffies;
713	long nr_pages = work->nr_pages;
714	unsigned long oldest_jif;
715	struct inode *inode;
716	long progress;
717
718	oldest_jif = jiffies;
719	work->older_than_this = &oldest_jif;
720
721	spin_lock(&wb->list_lock);
722	for (;;) {
723	/*
724	* Stop writeback when nr_pages has been consumed
725	*/
726	if (work->nr_pages <= 0)
727	break;
728
729	/*
730	* Background writeout and kupdate-style writeback may
731	* run forever. Stop them if there is other work to do
732	* so that e.g. sync can proceed. They'll be restarted
733	* after the other works are all done.
734	*/
735	if ((work->for_background \|\| work->for_kupdate) &&
736	!list_empty(&wb->bdi->work_list))
737	break;
738
739	/*
740	* For background writeout, stop when we are below the
741	* background dirty threshold
742	*/
743	if (work->for_background && !over_bground_thresh(wb->bdi))
744	break;
745
746	if (work->for_kupdate) {
747	oldest_jif = jiffies -
748	msecs_to_jiffies(dirty_expire_interval * 10);
749	work->older_than_this = &oldest_jif;
750	}
751
752	trace_writeback_start(wb->bdi, work);
753	if (list_empty(&wb->b_io))
754	queue_io(wb, work);
755	if (work->sb)
756	progress = writeback_sb_inodes(work->sb, wb, work);
757	else
758	progress = __writeback_inodes_wb(wb, work);
759	trace_writeback_written(wb->bdi, work);
760
761	wb_update_bandwidth(wb, wb_start);
762
763	/*
764	* Did we write something? Try for more
765	*
766	* Dirty inodes are moved to b_io for writeback in batches.
767	* The completion of the current batch does not necessarily
768	* mean the overall work is done. So we keep looping as long
769	* as made some progress on cleaning pages or inodes.
770	*/
771	if (progress)
772	continue;
773	/*
774	* No more inodes for IO, bail
775	*/
776	if (list_empty(&wb->b_more_io))
777	break;
778	/*
779	* Nothing written. Wait for some inode to
780	* become available for writeback. Otherwise
781	* we'll just busyloop.
782	*/
783	if (!list_empty(&wb->b_more_io)) {
784	trace_writeback_wait(wb->bdi, work);
785	inode = wb_inode(wb->b_more_io.prev);
786	spin_lock(&inode->i_lock);
787	inode_wait_for_writeback(inode, wb);
788	spin_unlock(&inode->i_lock);
789	}
790	}
791	spin_unlock(&wb->list_lock);
792
793	return nr_pages - work->nr_pages;
794	}
795
796	/*
797	* Return the next wb_writeback_work struct that hasn't been processed yet.
798	*/
799	static struct wb_writeback_work *
800	get_next_work_item(struct backing_dev_info *bdi)
801	{
802	struct wb_writeback_work *work = NULL;
803
804	spin_lock_bh(&bdi->wb_lock);
805	if (!list_empty(&bdi->work_list)) {
806	work = list_entry(bdi->work_list.next,
807	struct wb_writeback_work, list);
808	list_del_init(&work->list);
809	}
810	spin_unlock_bh(&bdi->wb_lock);
811	return work;
812	}
813
814	/*
815	* Add in the number of potentially dirty inodes, because each inode
816	* write can dirty pagecache in the underlying blockdev.
817	*/
818	static unsigned long get_nr_dirty_pages(void)
819	{
820	return global_page_state(NR_FILE_DIRTY) +
821	global_page_state(NR_UNSTABLE_NFS) +
822	get_nr_dirty_inodes();
823	}
824
825	static long wb_check_background_flush(struct bdi_writeback *wb)
826	{
827	if (over_bground_thresh(wb->bdi)) {
828
829	struct wb_writeback_work work = {
830	.nr_pages = LONG_MAX,
831	.sync_mode = WB_SYNC_NONE,
832	.for_background = 1,
833	.range_cyclic = 1,
834	.reason = WB_REASON_BACKGROUND,
835	};
836
837	return wb_writeback(wb, &work);
838	}
839
840	return 0;
841	}
842
843	static long wb_check_old_data_flush(struct bdi_writeback *wb)
844	{
845	unsigned long expired;
846	long nr_pages;
847
848	/*
849	* When set to zero, disable periodic writeback
850	*/
851	if (!dirty_writeback_interval)
852	return 0;
853
854	expired = wb->last_old_flush +
855	msecs_to_jiffies(dirty_writeback_interval * 10);
856	if (time_before(jiffies, expired))
857	return 0;
858
859	wb->last_old_flush = jiffies;
860	nr_pages = get_nr_dirty_pages();
861
862	if (nr_pages) {
863	struct wb_writeback_work work = {
864	.nr_pages = nr_pages,
865	.sync_mode = WB_SYNC_NONE,
866	.for_kupdate = 1,
867	.range_cyclic = 1,
868	.reason = WB_REASON_PERIODIC,
869	};
870
871	return wb_writeback(wb, &work);
872	}
873
874	return 0;
875	}
876
877	/*
878	* Retrieve work items and do the writeback they describe
879	*/
880	long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
881	{
882	struct backing_dev_info *bdi = wb->bdi;
883	struct wb_writeback_work *work;
884	long wrote = 0;
885
886	set_bit(BDI_writeback_running, &wb->bdi->state);
887	while ((work = get_next_work_item(bdi)) != NULL) {
888	/*
889	* Override sync mode, in case we must wait for completion
890	* because this thread is exiting now.
891	*/
892	if (force_wait)
893	work->sync_mode = WB_SYNC_ALL;
894
895	trace_writeback_exec(bdi, work);
896
897	wrote += wb_writeback(wb, work);
898
899	/*
900	* Notify the caller of completion if this is a synchronous
901	* work item, otherwise just free it.
902	*/
903	if (work->done)
904	complete(work->done);
905	else
906	kfree(work);
907	}
908
909	/*
910	* Check for periodic writeback, kupdated() style
911	*/
912	wrote += wb_check_old_data_flush(wb);
913	wrote += wb_check_background_flush(wb);
914	clear_bit(BDI_writeback_running, &wb->bdi->state);
915
916	return wrote;
917	}
918
919	/*
920	* Handle writeback of dirty data for the device backed by this bdi. Also
921	* wakes up periodically and does kupdated style flushing.
922	*/
923	int bdi_writeback_thread(void *data)
924	{
925	struct bdi_writeback *wb = data;
926	struct backing_dev_info *bdi = wb->bdi;
927	long pages_written;
928
929	current->flags \|= PF_SWAPWRITE;
930	set_freezable();
931	wb->last_active = jiffies;
932
933	/*
934	* Our parent may run at a different priority, just set us to normal
935	*/
936	set_user_nice(current, 0);
937
938	trace_writeback_thread_start(bdi);
939
940	while (!kthread_should_stop()) {
941	/*
942	* Remove own delayed wake-up timer, since we are already awake
943	* and we'll take care of the preriodic write-back.
944	*/
945	del_timer(&wb->wakeup_timer);
946
947	pages_written = wb_do_writeback(wb, 0);
948
949	trace_writeback_pages_written(pages_written);
950
951	if (pages_written)
952	wb->last_active = jiffies;
953
954	set_current_state(TASK_INTERRUPTIBLE);
955	if (!list_empty(&bdi->work_list) \|\| kthread_should_stop()) {
956	__set_current_state(TASK_RUNNING);
957	continue;
958	}
959
960	if (wb_has_dirty_io(wb) && dirty_writeback_interval)
961	schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
962	else {
963	/*
964	* We have nothing to do, so can go sleep without any
965	* timeout and save power. When a work is queued or
966	* something is made dirty - we will be woken up.
967	*/
968	schedule();
969	}
970
971	try_to_freeze();
972	}
973
974	/* Flush any work that raced with us exiting */
975	if (!list_empty(&bdi->work_list))
976	wb_do_writeback(wb, 1);
977
978	trace_writeback_thread_stop(bdi);
979	return 0;
980	}
981
982
983	/*
984	* Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
985	* the whole world.
986	*/
987	void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
988	{
989	struct backing_dev_info *bdi;
990
991	if (!nr_pages) {
992	nr_pages = global_page_state(NR_FILE_DIRTY) +
993	global_page_state(NR_UNSTABLE_NFS);
994	}
995
996	rcu_read_lock();
997	list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
998	if (!bdi_has_dirty_io(bdi))
999	continue;
1000	__bdi_start_writeback(bdi, nr_pages, false, reason);
1001	}
1002	rcu_read_unlock();
1003	}
1004
1005	static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1006	{
1007	if (inode->i_ino \|\| strcmp(inode->i_sb->s_id, "bdev")) {
1008	struct dentry *dentry;
1009	const char *name = "?";
1010
1011	dentry = d_find_alias(inode);
1012	if (dentry) {
1013	spin_lock(&dentry->d_lock);
1014	name = (const char *) dentry->d_name.name;
1015	}
1016	printk(KERN_DEBUG
1017	"%s(%d): dirtied inode %lu (%s) on %s\n",
1018	current->comm, task_pid_nr(current), inode->i_ino,
1019	name, inode->i_sb->s_id);
1020	if (dentry) {
1021	spin_unlock(&dentry->d_lock);
1022	dput(dentry);
1023	}
1024	}
1025	}
1026
1027	/**
1028	* __mark_inode_dirty - internal function
1029	* @inode: inode to mark
1030	* @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1031	* Mark an inode as dirty. Callers should use mark_inode_dirty or
1032	* mark_inode_dirty_sync.
1033	*
1034	* Put the inode on the super block's dirty list.
1035	*
1036	* CAREFUL! We mark it dirty unconditionally, but move it onto the
1037	* dirty list only if it is hashed or if it refers to a blockdev.
1038	* If it was not hashed, it will never be added to the dirty list
1039	* even if it is later hashed, as it will have been marked dirty already.
1040	*
1041	* In short, make sure you hash any inodes _before_ you start marking
1042	* them dirty.
1043	*
1044	* Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1045	* the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1046	* the kernel-internal blockdev inode represents the dirtying time of the
1047	* blockdev's pages. This is why for I_DIRTY_PAGES we always use
1048	* page->mapping->host, so the page-dirtying time is recorded in the internal
1049	* blockdev inode.
1050	*/
1051	void __mark_inode_dirty(struct inode *inode, int flags)
1052	{
1053	struct super_block *sb = inode->i_sb;
1054	struct backing_dev_info *bdi = NULL;
1055
1056	/*
1057	* Don't do this for I_DIRTY_PAGES - that doesn't actually
1058	* dirty the inode itself
1059	*/
1060	if (flags & (I_DIRTY_SYNC \| I_DIRTY_DATASYNC)) {
1061	if (sb->s_op->dirty_inode)
1062	sb->s_op->dirty_inode(inode, flags);
1063	}
1064
1065	/*
1066	* make sure that changes are seen by all cpus before we test i_state
1067	* -- mikulas
1068	*/
1069	smp_mb();
1070
1071	/* avoid the locking if we can */
1072	if ((inode->i_state & flags) == flags)
1073	return;
1074
1075	if (unlikely(block_dump))
1076	block_dump___mark_inode_dirty(inode);
1077
1078	spin_lock(&inode->i_lock);
1079	if ((inode->i_state & flags) != flags) {
1080	const int was_dirty = inode->i_state & I_DIRTY;
1081
1082	inode->i_state \|= flags;
1083
1084	/*
1085	* If the inode is being synced, just update its dirty state.
1086	* The unlocker will place the inode on the appropriate
1087	* superblock list, based upon its state.
1088	*/
1089	if (inode->i_state & I_SYNC)
1090	goto out_unlock_inode;
1091
1092	/*
1093	* Only add valid (hashed) inodes to the superblock's
1094	* dirty list. Add blockdev inodes as well.
1095	*/
1096	if (!S_ISBLK(inode->i_mode)) {
1097	if (inode_unhashed(inode))
1098	goto out_unlock_inode;
1099	}
1100	if (inode->i_state & I_FREEING)
1101	goto out_unlock_inode;
1102
1103	/*
1104	* If the inode was already on b_dirty/b_io/b_more_io, don't
1105	* reposition it (that would break b_dirty time-ordering).
1106	*/
1107	if (!was_dirty) {
1108	bool wakeup_bdi = false;
1109	bdi = inode_to_bdi(inode);
1110
1111	if (bdi_cap_writeback_dirty(bdi)) {
1112	WARN(!test_bit(BDI_registered, &bdi->state),
1113	"bdi-%s not registered\n", bdi->name);
1114
1115	/*
1116	* If this is the first dirty inode for this
1117	* bdi, we have to wake-up the corresponding
1118	* bdi thread to make sure background
1119	* write-back happens later.
1120	*/
1121	if (!wb_has_dirty_io(&bdi->wb))
1122	wakeup_bdi = true;
1123	}
1124
1125	spin_unlock(&inode->i_lock);
1126	spin_lock(&bdi->wb.list_lock);
1127	inode->dirtied_when = jiffies;
1128	list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1129	spin_unlock(&bdi->wb.list_lock);
1130
1131	if (wakeup_bdi)
1132	bdi_wakeup_thread_delayed(bdi);
1133	return;
1134	}
1135	}
1136	out_unlock_inode:
1137	spin_unlock(&inode->i_lock);
1138
1139	}
1140	EXPORT_SYMBOL(__mark_inode_dirty);
1141
1142	/*
1143	* Write out a superblock's list of dirty inodes. A wait will be performed
1144	* upon no inodes, all inodes or the final one, depending upon sync_mode.
1145	*
1146	* If older_than_this is non-NULL, then only write out inodes which
1147	* had their first dirtying at a time earlier than *older_than_this.
1148	*
1149	* If `bdi' is non-zero then we're being asked to writeback a specific queue.
1150	* This function assumes that the blockdev superblock's inodes are backed by
1151	* a variety of queues, so all inodes are searched. For other superblocks,
1152	* assume that all inodes are backed by the same queue.
1153	*
1154	* The inodes to be written are parked on bdi->b_io. They are moved back onto
1155	* bdi->b_dirty as they are selected for writing. This way, none can be missed
1156	* on the writer throttling path, and we get decent balancing between many
1157	* throttled threads: we don't want them all piling up on inode_sync_wait.
1158	*/
1159	static void wait_sb_inodes(struct super_block *sb)
1160	{
1161	struct inode inode, old_inode = NULL;
1162
1163	/*
1164	* We need to be protected against the filesystem going from
1165	* r/o to r/w or vice versa.
1166	*/
1167	WARN_ON(!rwsem_is_locked(&sb->s_umount));
1168
1169	spin_lock(&inode_sb_list_lock);
1170
1171	/*
1172	* Data integrity sync. Must wait for all pages under writeback,
1173	* because there may have been pages dirtied before our sync
1174	* call, but which had writeout started before we write it out.
1175	* In which case, the inode may not be on the dirty list, but
1176	* we still have to wait for that writeout.
1177	*/
1178	list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1179	struct address_space *mapping = inode->i_mapping;
1180
1181	spin_lock(&inode->i_lock);
1182	if ((inode->i_state & (I_FREEING\|I_WILL_FREE\|I_NEW)) \|\|
1183	(mapping->nrpages == 0)) {
1184	spin_unlock(&inode->i_lock);
1185	continue;
1186	}
1187	__iget(inode);
1188	spin_unlock(&inode->i_lock);
1189	spin_unlock(&inode_sb_list_lock);
1190
1191	/*
1192	* We hold a reference to 'inode' so it couldn't have been
1193	* removed from s_inodes list while we dropped the
1194	* inode_sb_list_lock. We cannot iput the inode now as we can
1195	* be holding the last reference and we cannot iput it under
1196	* inode_sb_list_lock. So we keep the reference and iput it
1197	* later.
1198	*/
1199	iput(old_inode);
1200	old_inode = inode;
1201
1202	filemap_fdatawait(mapping);
1203
1204	cond_resched();
1205
1206	spin_lock(&inode_sb_list_lock);
1207	}
1208	spin_unlock(&inode_sb_list_lock);
1209	iput(old_inode);
1210	}
1211
1212	/**
1213	* writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1214	* @sb: the superblock
1215	* @nr: the number of pages to write
1216	* @reason: reason why some writeback work initiated
1217	*
1218	* Start writeback on some inodes on this super_block. No guarantees are made
1219	* on how many (if any) will be written, and this function does not wait
1220	* for IO completion of submitted IO.
1221	*/
1222	void writeback_inodes_sb_nr(struct super_block *sb,
1223	unsigned long nr,
1224	enum wb_reason reason)
1225	{
1226	DECLARE_COMPLETION_ONSTACK(done);
1227	struct wb_writeback_work work = {
1228	.sb = sb,
1229	.sync_mode = WB_SYNC_NONE,
1230	.tagged_writepages = 1,
1231	.done = &done,
1232	.nr_pages = nr,
1233	.reason = reason,
1234	};
1235
1236	WARN_ON(!rwsem_is_locked(&sb->s_umount));
1237	bdi_queue_work(sb->s_bdi, &work);
1238	wait_for_completion(&done);
1239	}
1240	EXPORT_SYMBOL(writeback_inodes_sb_nr);
1241
1242	/**
1243	* writeback_inodes_sb - writeback dirty inodes from given super_block
1244	* @sb: the superblock
1245	* @reason: reason why some writeback work was initiated
1246	*
1247	* Start writeback on some inodes on this super_block. No guarantees are made
1248	* on how many (if any) will be written, and this function does not wait
1249	* for IO completion of submitted IO.
1250	*/
1251	void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1252	{
1253	return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1254	}
1255	EXPORT_SYMBOL(writeback_inodes_sb);
1256
1257	/**
1258	* writeback_inodes_sb_if_idle - start writeback if none underway
1259	* @sb: the superblock
1260	* @reason: reason why some writeback work was initiated
1261	*
1262	* Invoke writeback_inodes_sb if no writeback is currently underway.
1263	* Returns 1 if writeback was started, 0 if not.
1264	*/
1265	int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
1266	{
1267	if (!writeback_in_progress(sb->s_bdi)) {
1268	down_read(&sb->s_umount);
1269	writeback_inodes_sb(sb, reason);
1270	up_read(&sb->s_umount);
1271	return 1;
1272	} else
1273	return 0;
1274	}
1275	EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1276
1277	/**
1278	* writeback_inodes_sb_if_idle - start writeback if none underway
1279	* @sb: the superblock
1280	* @nr: the number of pages to write
1281	* @reason: reason why some writeback work was initiated
1282	*
1283	* Invoke writeback_inodes_sb if no writeback is currently underway.
1284	* Returns 1 if writeback was started, 0 if not.
1285	*/
1286	int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1287	unsigned long nr,
1288	enum wb_reason reason)
1289	{
1290	if (!writeback_in_progress(sb->s_bdi)) {
1291	down_read(&sb->s_umount);
1292	writeback_inodes_sb_nr(sb, nr, reason);
1293	up_read(&sb->s_umount);
1294	return 1;
1295	} else
1296	return 0;
1297	}
1298	EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1299
1300	/**
1301	* sync_inodes_sb - sync sb inode pages
1302	* @sb: the superblock
1303	*
1304	* This function writes and waits on any dirty inode belonging to this
1305	* super_block.
1306	*/
1307	void sync_inodes_sb(struct super_block *sb)
1308	{
1309	DECLARE_COMPLETION_ONSTACK(done);
1310	struct wb_writeback_work work = {
1311	.sb = sb,
1312	.sync_mode = WB_SYNC_ALL,
1313	.nr_pages = LONG_MAX,
1314	.range_cyclic = 0,
1315	.done = &done,
1316	.reason = WB_REASON_SYNC,
1317	};
1318
1319	WARN_ON(!rwsem_is_locked(&sb->s_umount));
1320
1321	bdi_queue_work(sb->s_bdi, &work);
1322	wait_for_completion(&done);
1323
1324	wait_sb_inodes(sb);
1325	}
1326	EXPORT_SYMBOL(sync_inodes_sb);
1327
1328	/**
1329	* write_inode_now - write an inode to disk
1330	* @inode: inode to write to disk
1331	* @sync: whether the write should be synchronous or not
1332	*
1333	* This function commits an inode to disk immediately if it is dirty. This is
1334	* primarily needed by knfsd.
1335	*
1336	* The caller must either have a ref on the inode or must have set I_WILL_FREE.
1337	*/
1338	int write_inode_now(struct inode *inode, int sync)
1339	{
1340	struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1341	int ret;
1342	struct writeback_control wbc = {
1343	.nr_to_write = LONG_MAX,
1344	.sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1345	.range_start = 0,
1346	.range_end = LLONG_MAX,
1347	};
1348
1349	if (!mapping_cap_writeback_dirty(inode->i_mapping))
1350	wbc.nr_to_write = 0;
1351
1352	might_sleep();
1353	spin_lock(&wb->list_lock);
1354	spin_lock(&inode->i_lock);
1355	ret = writeback_single_inode(inode, wb, &wbc);
1356	spin_unlock(&inode->i_lock);
1357	spin_unlock(&wb->list_lock);
1358	if (sync)
1359	inode_sync_wait(inode);
1360	return ret;
1361	}
1362	EXPORT_SYMBOL(write_inode_now);
1363
1364	/**
1365	* sync_inode - write an inode and its pages to disk.
1366	* @inode: the inode to sync
1367	* @wbc: controls the writeback mode
1368	*
1369	* sync_inode() will write an inode and its pages to disk. It will also
1370	* correctly update the inode on its superblock's dirty inode lists and will
1371	* update inode->i_state.
1372	*
1373	* The caller must have a ref on the inode.
1374	*/
1375	int sync_inode(struct inode inode, struct writeback_control wbc)
1376	{
1377	struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1378	int ret;
1379
1380	spin_lock(&wb->list_lock);
1381	spin_lock(&inode->i_lock);
1382	ret = writeback_single_inode(inode, wb, wbc);
1383	spin_unlock(&inode->i_lock);
1384	spin_unlock(&wb->list_lock);
1385	return ret;
1386	}
1387	EXPORT_SYMBOL(sync_inode);
1388
1389	/**
1390	* sync_inode_metadata - write an inode to disk
1391	* @inode: the inode to sync
1392	* @wait: wait for I/O to complete.
1393	*
1394	* Write an inode to disk and adjust its dirty state after completion.
1395	*
1396	* Note: only writes the actual inode, no associated data or other metadata.
1397	*/
1398	int sync_inode_metadata(struct inode *inode, int wait)
1399	{
1400	struct writeback_control wbc = {
1401	.sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1402	.nr_to_write = 0, /* metadata-only */
1403	};
1404
1405	return sync_inode(inode, &wbc);
1406	}
1407	EXPORT_SYMBOL(sync_inode_metadata);
1408

Download this file

Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master

Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9

Kernel

Kernel Git Source Tree

Root/fs/fs-writeback.c

interactive