4 * Copyright (C) 2002, Linus Torvalds.
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.
11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/pagemap.h>
24 #include <linux/kthread.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/tracepoint.h>
29 #include <linux/device.h>
33 * 4MB minimal write chunk size
35 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
38 * Passed into wb_writeback(), essentially a subset of writeback_control
40 struct wb_writeback_work {
42 struct super_block *sb;
44 * Write only inodes dirtied before this time. Don't forget to set
45 * older_than_this_is_set when you set this.
47 unsigned long older_than_this;
48 enum writeback_sync_modes sync_mode;
49 unsigned int tagged_writepages:1;
50 unsigned int for_kupdate:1;
51 unsigned int range_cyclic:1;
52 unsigned int for_background:1;
53 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
54 unsigned int older_than_this_is_set:1;
55 enum wb_reason reason; /* why was writeback initiated? */
57 struct list_head list; /* pending work list */
58 struct completion *done; /* set if the caller waits */
62 * writeback_in_progress - determine whether there is writeback in progress
63 * @bdi: the device's backing_dev_info structure.
65 * Determine whether there is writeback waiting to be handled against a
68 int writeback_in_progress(struct backing_dev_info *bdi)
70 return test_bit(BDI_writeback_running, &bdi->state);
72 EXPORT_SYMBOL(writeback_in_progress);
74 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
76 struct super_block *sb = inode->i_sb;
78 if (sb_is_blkdev_sb(sb))
79 return inode->i_mapping->backing_dev_info;
84 static inline struct inode *wb_inode(struct list_head *head)
86 return list_entry(head, struct inode, i_wb_list);
90 * Include the creation of the trace points after defining the
91 * wb_writeback_work structure and inline functions so that the definition
92 * remains local to this file.
94 #define CREATE_TRACE_POINTS
95 #include <trace/events/writeback.h>
97 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
99 static void bdi_queue_work(struct backing_dev_info *bdi,
100 struct wb_writeback_work *work)
102 trace_writeback_queue(bdi, work);
104 spin_lock_bh(&bdi->wb_lock);
105 list_add_tail(&work->list, &bdi->work_list);
106 spin_unlock_bh(&bdi->wb_lock);
108 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
112 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
113 bool range_cyclic, enum wb_reason reason)
115 struct wb_writeback_work *work;
118 * This is WB_SYNC_NONE writeback, so if allocation fails just
119 * wakeup the thread for old dirty data writeback
121 work = kzalloc(sizeof(*work), GFP_ATOMIC);
123 trace_writeback_nowork(bdi);
124 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
128 work->sync_mode = WB_SYNC_NONE;
129 work->nr_pages = nr_pages;
130 work->range_cyclic = range_cyclic;
131 work->reason = reason;
133 bdi_queue_work(bdi, work);
137 * bdi_start_writeback - start writeback
138 * @bdi: the backing device to write from
139 * @nr_pages: the number of pages to write
140 * @reason: reason why some writeback work was initiated
143 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
144 * started when this function returns, we make no guarantees on
145 * completion. Caller need not hold sb s_umount semaphore.
148 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
149 enum wb_reason reason)
151 __bdi_start_writeback(bdi, nr_pages, true, reason);
155 * bdi_start_background_writeback - start background writeback
156 * @bdi: the backing device to write from
159 * This makes sure WB_SYNC_NONE background writeback happens. When
160 * this function returns, it is only guaranteed that for given BDI
161 * some IO is happening if we are over background dirty threshold.
162 * Caller need not hold sb s_umount semaphore.
164 void bdi_start_background_writeback(struct backing_dev_info *bdi)
167 * We just wake up the flusher thread. It will perform background
168 * writeback as soon as there is no other work to do.
170 trace_writeback_wake_background(bdi);
171 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
175 * Remove the inode from the writeback list it is on.
177 void inode_wb_list_del(struct inode *inode)
179 struct backing_dev_info *bdi = inode_to_bdi(inode);
181 spin_lock(&bdi->wb.list_lock);
182 list_del_init(&inode->i_wb_list);
183 spin_unlock(&bdi->wb.list_lock);
187 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
188 * furthest end of its superblock's dirty-inode list.
190 * Before stamping the inode's ->dirtied_when, we check to see whether it is
191 * already the most-recently-dirtied inode on the b_dirty list. If that is
192 * the case then the inode must have been redirtied while it was being written
193 * out and we don't reset its dirtied_when.
195 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
197 assert_spin_locked(&wb->list_lock);
198 if (!list_empty(&wb->b_dirty)) {
201 tail = wb_inode(wb->b_dirty.next);
202 if (time_before(inode->dirtied_when, tail->dirtied_when))
203 inode->dirtied_when = jiffies;
205 list_move(&inode->i_wb_list, &wb->b_dirty);
209 * requeue inode for re-scanning after bdi->b_io list is exhausted.
211 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
213 assert_spin_locked(&wb->list_lock);
214 list_move(&inode->i_wb_list, &wb->b_more_io);
217 static void inode_sync_complete(struct inode *inode)
219 inode->i_state &= ~I_SYNC;
220 /* If inode is clean an unused, put it into LRU now... */
221 inode_add_lru(inode);
222 /* Waiters must see I_SYNC cleared before being woken up */
224 wake_up_bit(&inode->i_state, __I_SYNC);
227 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
229 bool ret = time_after(inode->dirtied_when, t);
232 * For inodes being constantly redirtied, dirtied_when can get stuck.
233 * It _appears_ to be in the future, but is actually in distant past.
234 * This test is necessary to prevent such wrapped-around relative times
235 * from permanently stopping the whole bdi writeback.
237 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
243 * Move expired (dirtied before work->older_than_this) dirty inodes from
244 * @delaying_queue to @dispatch_queue.
246 static int move_expired_inodes(struct list_head *delaying_queue,
247 struct list_head *dispatch_queue,
248 struct wb_writeback_work *work)
251 struct list_head *pos, *node;
252 struct super_block *sb = NULL;
257 WARN_ON_ONCE(!work->older_than_this_is_set);
258 while (!list_empty(delaying_queue)) {
259 inode = wb_inode(delaying_queue->prev);
260 if (inode_dirtied_after(inode, work->older_than_this))
262 list_move(&inode->i_wb_list, &tmp);
264 if (sb_is_blkdev_sb(inode->i_sb))
266 if (sb && sb != inode->i_sb)
271 /* just one sb in list, splice to dispatch_queue and we're done */
273 list_splice(&tmp, dispatch_queue);
277 /* Move inodes from one superblock together */
278 while (!list_empty(&tmp)) {
279 sb = wb_inode(tmp.prev)->i_sb;
280 list_for_each_prev_safe(pos, node, &tmp) {
281 inode = wb_inode(pos);
282 if (inode->i_sb == sb)
283 list_move(&inode->i_wb_list, dispatch_queue);
291 * Queue all expired dirty inodes for io, eldest first.
293 * newly dirtied b_dirty b_io b_more_io
294 * =============> gf edc BA
296 * newly dirtied b_dirty b_io b_more_io
297 * =============> g fBAedc
299 * +--> dequeue for IO
301 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
304 assert_spin_locked(&wb->list_lock);
305 list_splice_init(&wb->b_more_io, &wb->b_io);
306 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
307 trace_writeback_queue_io(wb, work, moved);
310 static int write_inode(struct inode *inode, struct writeback_control *wbc)
314 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
315 trace_writeback_write_inode_start(inode, wbc);
316 ret = inode->i_sb->s_op->write_inode(inode, wbc);
317 trace_writeback_write_inode(inode, wbc);
324 * Wait for writeback on an inode to complete. Called with i_lock held.
325 * Caller must make sure inode cannot go away when we drop i_lock.
327 static void __inode_wait_for_writeback(struct inode *inode)
328 __releases(inode->i_lock)
329 __acquires(inode->i_lock)
331 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
332 wait_queue_head_t *wqh;
334 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
335 while (inode->i_state & I_SYNC) {
336 spin_unlock(&inode->i_lock);
337 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
338 spin_lock(&inode->i_lock);
343 * Wait for writeback on an inode to complete. Caller must have inode pinned.
345 void inode_wait_for_writeback(struct inode *inode)
347 spin_lock(&inode->i_lock);
348 __inode_wait_for_writeback(inode);
349 spin_unlock(&inode->i_lock);
353 * Sleep until I_SYNC is cleared. This function must be called with i_lock
354 * held and drops it. It is aimed for callers not holding any inode reference
355 * so once i_lock is dropped, inode can go away.
357 static void inode_sleep_on_writeback(struct inode *inode)
358 __releases(inode->i_lock)
361 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
364 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
365 sleep = inode->i_state & I_SYNC;
366 spin_unlock(&inode->i_lock);
369 finish_wait(wqh, &wait);
373 * Find proper writeback list for the inode depending on its current state and
374 * possibly also change of its state while we were doing writeback. Here we
375 * handle things such as livelock prevention or fairness of writeback among
376 * inodes. This function can be called only by flusher thread - noone else
377 * processes all inodes in writeback lists and requeueing inodes behind flusher
378 * thread's back can have unexpected consequences.
380 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
381 struct writeback_control *wbc)
383 if (inode->i_state & I_FREEING)
387 * Sync livelock prevention. Each inode is tagged and synced in one
388 * shot. If still dirty, it will be redirty_tail()'ed below. Update
389 * the dirty time to prevent enqueue and sync it again.
391 if ((inode->i_state & I_DIRTY) &&
392 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
393 inode->dirtied_when = jiffies;
395 if (wbc->pages_skipped) {
397 * writeback is not making progress due to locked
398 * buffers. Skip this inode for now.
400 redirty_tail(inode, wb);
404 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
406 * We didn't write back all the pages. nfs_writepages()
407 * sometimes bales out without doing anything.
409 if (wbc->nr_to_write <= 0) {
410 /* Slice used up. Queue for next turn. */
411 requeue_io(inode, wb);
414 * Writeback blocked by something other than
415 * congestion. Delay the inode for some time to
416 * avoid spinning on the CPU (100% iowait)
417 * retrying writeback of the dirty page/inode
418 * that cannot be performed immediately.
420 redirty_tail(inode, wb);
422 } else if (inode->i_state & I_DIRTY) {
424 * Filesystems can dirty the inode during writeback operations,
425 * such as delayed allocation during submission or metadata
426 * updates after data IO completion.
428 redirty_tail(inode, wb);
430 /* The inode is clean. Remove from writeback lists. */
431 list_del_init(&inode->i_wb_list);
436 * Write out an inode and its dirty pages. Do not update the writeback list
437 * linkage. That is left to the caller. The caller is also responsible for
438 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
441 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
443 struct address_space *mapping = inode->i_mapping;
444 long nr_to_write = wbc->nr_to_write;
448 WARN_ON(!(inode->i_state & I_SYNC));
450 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
452 ret = do_writepages(mapping, wbc);
455 * Make sure to wait on the data before writing out the metadata.
456 * This is important for filesystems that modify metadata on data
457 * I/O completion. We don't do it for sync(2) writeback because it has a
458 * separate, external IO completion path and ->sync_fs for guaranteeing
459 * inode metadata is written back correctly.
461 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
462 int err = filemap_fdatawait(mapping);
468 * Some filesystems may redirty the inode during the writeback
469 * due to delalloc, clear dirty metadata flags right before
472 spin_lock(&inode->i_lock);
473 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
474 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
475 inode->i_state &= ~I_DIRTY_PAGES;
476 dirty = inode->i_state & I_DIRTY;
477 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
478 spin_unlock(&inode->i_lock);
479 /* Don't write the inode if only I_DIRTY_PAGES was set */
480 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
481 int err = write_inode(inode, wbc);
485 trace_writeback_single_inode(inode, wbc, nr_to_write);
490 * Write out an inode's dirty pages. Either the caller has an active reference
491 * on the inode or the inode has I_WILL_FREE set.
493 * This function is designed to be called for writing back one inode which
494 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
495 * and does more profound writeback list handling in writeback_sb_inodes().
498 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
499 struct writeback_control *wbc)
503 spin_lock(&inode->i_lock);
504 if (!atomic_read(&inode->i_count))
505 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
507 WARN_ON(inode->i_state & I_WILL_FREE);
509 if (inode->i_state & I_SYNC) {
510 if (wbc->sync_mode != WB_SYNC_ALL)
513 * It's a data-integrity sync. We must wait. Since callers hold
514 * inode reference or inode has I_WILL_FREE set, it cannot go
517 __inode_wait_for_writeback(inode);
519 WARN_ON(inode->i_state & I_SYNC);
521 * Skip inode if it is clean and we have no outstanding writeback in
522 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
523 * function since flusher thread may be doing for example sync in
524 * parallel and if we move the inode, it could get skipped. So here we
525 * make sure inode is on some writeback list and leave it there unless
526 * we have completely cleaned the inode.
528 if (!(inode->i_state & I_DIRTY) &&
529 (wbc->sync_mode != WB_SYNC_ALL ||
530 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
532 inode->i_state |= I_SYNC;
533 spin_unlock(&inode->i_lock);
535 ret = __writeback_single_inode(inode, wbc);
537 spin_lock(&wb->list_lock);
538 spin_lock(&inode->i_lock);
540 * If inode is clean, remove it from writeback lists. Otherwise don't
541 * touch it. See comment above for explanation.
543 if (!(inode->i_state & I_DIRTY))
544 list_del_init(&inode->i_wb_list);
545 spin_unlock(&wb->list_lock);
546 inode_sync_complete(inode);
548 spin_unlock(&inode->i_lock);
552 static long writeback_chunk_size(struct backing_dev_info *bdi,
553 struct wb_writeback_work *work)
558 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
559 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
560 * here avoids calling into writeback_inodes_wb() more than once.
562 * The intended call sequence for WB_SYNC_ALL writeback is:
565 * writeback_sb_inodes() <== called only once
566 * write_cache_pages() <== called once for each inode
567 * (quickly) tag currently dirty pages
568 * (maybe slowly) sync all tagged pages
570 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
573 pages = min(bdi->avg_write_bandwidth / 2,
574 global_dirty_limit / DIRTY_SCOPE);
575 pages = min(pages, work->nr_pages);
576 pages = round_down(pages + MIN_WRITEBACK_PAGES,
577 MIN_WRITEBACK_PAGES);
584 * Write a portion of b_io inodes which belong to @sb.
586 * Return the number of pages and/or inodes written.
588 static long writeback_sb_inodes(struct super_block *sb,
589 struct bdi_writeback *wb,
590 struct wb_writeback_work *work)
592 struct writeback_control wbc = {
593 .sync_mode = work->sync_mode,
594 .tagged_writepages = work->tagged_writepages,
595 .for_kupdate = work->for_kupdate,
596 .for_background = work->for_background,
597 .for_sync = work->for_sync,
598 .range_cyclic = work->range_cyclic,
600 .range_end = LLONG_MAX,
602 unsigned long start_time = jiffies;
604 long wrote = 0; /* count both pages and inodes */
606 while (!list_empty(&wb->b_io)) {
607 struct inode *inode = wb_inode(wb->b_io.prev);
609 if (inode->i_sb != sb) {
612 * We only want to write back data for this
613 * superblock, move all inodes not belonging
614 * to it back onto the dirty list.
616 redirty_tail(inode, wb);
621 * The inode belongs to a different superblock.
622 * Bounce back to the caller to unpin this and
623 * pin the next superblock.
629 * Don't bother with new inodes or inodes being freed, first
630 * kind does not need periodic writeout yet, and for the latter
631 * kind writeout is handled by the freer.
633 spin_lock(&inode->i_lock);
634 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
635 spin_unlock(&inode->i_lock);
636 redirty_tail(inode, wb);
639 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
641 * If this inode is locked for writeback and we are not
642 * doing writeback-for-data-integrity, move it to
643 * b_more_io so that writeback can proceed with the
644 * other inodes on s_io.
646 * We'll have another go at writing back this inode
647 * when we completed a full scan of b_io.
649 spin_unlock(&inode->i_lock);
650 requeue_io(inode, wb);
651 trace_writeback_sb_inodes_requeue(inode);
654 spin_unlock(&wb->list_lock);
657 * We already requeued the inode if it had I_SYNC set and we
658 * are doing WB_SYNC_NONE writeback. So this catches only the
661 if (inode->i_state & I_SYNC) {
662 /* Wait for I_SYNC. This function drops i_lock... */
663 inode_sleep_on_writeback(inode);
664 /* Inode may be gone, start again */
665 spin_lock(&wb->list_lock);
668 inode->i_state |= I_SYNC;
669 spin_unlock(&inode->i_lock);
671 write_chunk = writeback_chunk_size(wb->bdi, work);
672 wbc.nr_to_write = write_chunk;
673 wbc.pages_skipped = 0;
676 * We use I_SYNC to pin the inode in memory. While it is set
677 * evict_inode() will wait so the inode cannot be freed.
679 __writeback_single_inode(inode, &wbc);
681 work->nr_pages -= write_chunk - wbc.nr_to_write;
682 wrote += write_chunk - wbc.nr_to_write;
683 spin_lock(&wb->list_lock);
684 spin_lock(&inode->i_lock);
685 if (!(inode->i_state & I_DIRTY))
687 requeue_inode(inode, wb, &wbc);
688 inode_sync_complete(inode);
689 spin_unlock(&inode->i_lock);
690 cond_resched_lock(&wb->list_lock);
692 * bail out to wb_writeback() often enough to check
693 * background threshold and other termination conditions.
696 if (time_is_before_jiffies(start_time + HZ / 10UL))
698 if (work->nr_pages <= 0)
705 static long __writeback_inodes_wb(struct bdi_writeback *wb,
706 struct wb_writeback_work *work)
708 unsigned long start_time = jiffies;
711 while (!list_empty(&wb->b_io)) {
712 struct inode *inode = wb_inode(wb->b_io.prev);
713 struct super_block *sb = inode->i_sb;
715 if (!grab_super_passive(sb)) {
717 * grab_super_passive() may fail consistently due to
718 * s_umount being grabbed by someone else. Don't use
719 * requeue_io() to avoid busy retrying the inode/sb.
721 redirty_tail(inode, wb);
724 wrote += writeback_sb_inodes(sb, wb, work);
727 /* refer to the same tests at the end of writeback_sb_inodes */
729 if (time_is_before_jiffies(start_time + HZ / 10UL))
731 if (work->nr_pages <= 0)
735 /* Leave any unwritten inodes on b_io */
739 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
740 enum wb_reason reason)
742 struct wb_writeback_work work = {
743 .nr_pages = nr_pages,
744 .sync_mode = WB_SYNC_NONE,
747 .older_than_this = jiffies,
748 .older_than_this_is_set = 1,
751 spin_lock(&wb->list_lock);
752 if (list_empty(&wb->b_io))
754 __writeback_inodes_wb(wb, &work);
755 spin_unlock(&wb->list_lock);
757 return nr_pages - work.nr_pages;
760 static bool over_bground_thresh(struct backing_dev_info *bdi)
762 unsigned long background_thresh, dirty_thresh;
764 global_dirty_limits(&background_thresh, &dirty_thresh);
766 if (global_page_state(NR_FILE_DIRTY) +
767 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
770 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
771 bdi_dirty_limit(bdi, background_thresh))
778 * Called under wb->list_lock. If there are multiple wb per bdi,
779 * only the flusher working on the first wb should do it.
781 static void wb_update_bandwidth(struct bdi_writeback *wb,
782 unsigned long start_time)
784 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
788 * Explicit flushing or periodic writeback of "old" data.
790 * Define "old": the first time one of an inode's pages is dirtied, we mark the
791 * dirtying-time in the inode's address_space. So this periodic writeback code
792 * just walks the superblock inode list, writing back any inodes which are
793 * older than a specific point in time.
795 * Try to run once per dirty_writeback_interval. But if a writeback event
796 * takes longer than a dirty_writeback_interval interval, then leave a
799 * older_than_this takes precedence over nr_to_write. So we'll only write back
800 * all dirty pages if they are all attached to "old" mappings.
802 static long wb_writeback(struct bdi_writeback *wb,
803 struct wb_writeback_work *work)
805 unsigned long wb_start = jiffies;
806 long nr_pages = work->nr_pages;
810 if (!work->older_than_this_is_set) {
811 work->older_than_this = jiffies;
812 work->older_than_this_is_set = 1;
815 spin_lock(&wb->list_lock);
818 * Stop writeback when nr_pages has been consumed
820 if (work->nr_pages <= 0)
824 * Background writeout and kupdate-style writeback may
825 * run forever. Stop them if there is other work to do
826 * so that e.g. sync can proceed. They'll be restarted
827 * after the other works are all done.
829 if ((work->for_background || work->for_kupdate) &&
830 !list_empty(&wb->bdi->work_list))
834 * For background writeout, stop when we are below the
835 * background dirty threshold
837 if (work->for_background && !over_bground_thresh(wb->bdi))
841 * Kupdate and background works are special and we want to
842 * include all inodes that need writing. Livelock avoidance is
843 * handled by these works yielding to any other work so we are
846 if (work->for_kupdate) {
847 work->older_than_this = jiffies -
848 msecs_to_jiffies(dirty_expire_interval * 10);
849 } else if (work->for_background)
850 work->older_than_this = jiffies;
852 trace_writeback_start(wb->bdi, work);
853 if (list_empty(&wb->b_io))
856 progress = writeback_sb_inodes(work->sb, wb, work);
858 progress = __writeback_inodes_wb(wb, work);
859 trace_writeback_written(wb->bdi, work);
861 wb_update_bandwidth(wb, wb_start);
864 * Did we write something? Try for more
866 * Dirty inodes are moved to b_io for writeback in batches.
867 * The completion of the current batch does not necessarily
868 * mean the overall work is done. So we keep looping as long
869 * as made some progress on cleaning pages or inodes.
874 * No more inodes for IO, bail
876 if (list_empty(&wb->b_more_io))
879 * Nothing written. Wait for some inode to
880 * become available for writeback. Otherwise
881 * we'll just busyloop.
883 if (!list_empty(&wb->b_more_io)) {
884 trace_writeback_wait(wb->bdi, work);
885 inode = wb_inode(wb->b_more_io.prev);
886 spin_lock(&inode->i_lock);
887 spin_unlock(&wb->list_lock);
888 /* This function drops i_lock... */
889 inode_sleep_on_writeback(inode);
890 spin_lock(&wb->list_lock);
893 spin_unlock(&wb->list_lock);
895 return nr_pages - work->nr_pages;
899 * Return the next wb_writeback_work struct that hasn't been processed yet.
901 static struct wb_writeback_work *
902 get_next_work_item(struct backing_dev_info *bdi)
904 struct wb_writeback_work *work = NULL;
906 spin_lock_bh(&bdi->wb_lock);
907 if (!list_empty(&bdi->work_list)) {
908 work = list_entry(bdi->work_list.next,
909 struct wb_writeback_work, list);
910 list_del_init(&work->list);
912 spin_unlock_bh(&bdi->wb_lock);
917 * Add in the number of potentially dirty inodes, because each inode
918 * write can dirty pagecache in the underlying blockdev.
920 static unsigned long get_nr_dirty_pages(void)
922 return global_page_state(NR_FILE_DIRTY) +
923 global_page_state(NR_UNSTABLE_NFS) +
924 get_nr_dirty_inodes();
927 static long wb_check_background_flush(struct bdi_writeback *wb)
929 if (over_bground_thresh(wb->bdi)) {
931 struct wb_writeback_work work = {
932 .nr_pages = LONG_MAX,
933 .sync_mode = WB_SYNC_NONE,
936 .reason = WB_REASON_BACKGROUND,
939 return wb_writeback(wb, &work);
945 static long wb_check_old_data_flush(struct bdi_writeback *wb)
947 unsigned long expired;
951 * When set to zero, disable periodic writeback
953 if (!dirty_writeback_interval)
956 expired = wb->last_old_flush +
957 msecs_to_jiffies(dirty_writeback_interval * 10);
958 if (time_before(jiffies, expired))
961 wb->last_old_flush = jiffies;
962 nr_pages = get_nr_dirty_pages();
965 struct wb_writeback_work work = {
966 .nr_pages = nr_pages,
967 .sync_mode = WB_SYNC_NONE,
970 .reason = WB_REASON_PERIODIC,
973 return wb_writeback(wb, &work);
980 * Retrieve work items and do the writeback they describe
982 static long wb_do_writeback(struct bdi_writeback *wb)
984 struct backing_dev_info *bdi = wb->bdi;
985 struct wb_writeback_work *work;
988 set_bit(BDI_writeback_running, &wb->bdi->state);
989 while ((work = get_next_work_item(bdi)) != NULL) {
991 trace_writeback_exec(bdi, work);
993 wrote += wb_writeback(wb, work);
996 * Notify the caller of completion if this is a synchronous
997 * work item, otherwise just free it.
1000 complete(work->done);
1006 * Check for periodic writeback, kupdated() style
1008 wrote += wb_check_old_data_flush(wb);
1009 wrote += wb_check_background_flush(wb);
1010 clear_bit(BDI_writeback_running, &wb->bdi->state);
1016 * Handle writeback of dirty data for the device backed by this bdi. Also
1017 * reschedules periodically and does kupdated style flushing.
1019 void bdi_writeback_workfn(struct work_struct *work)
1021 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1022 struct bdi_writeback, dwork);
1023 struct backing_dev_info *bdi = wb->bdi;
1026 set_worker_desc("flush-%s", dev_name(bdi->dev));
1027 current->flags |= PF_SWAPWRITE;
1029 if (likely(!current_is_workqueue_rescuer() ||
1030 list_empty(&bdi->bdi_list))) {
1032 * The normal path. Keep writing back @bdi until its
1033 * work_list is empty. Note that this path is also taken
1034 * if @bdi is shutting down even when we're running off the
1035 * rescuer as work_list needs to be drained.
1038 pages_written = wb_do_writeback(wb);
1039 trace_writeback_pages_written(pages_written);
1040 } while (!list_empty(&bdi->work_list));
1043 * bdi_wq can't get enough workers and we're running off
1044 * the emergency worker. Don't hog it. Hopefully, 1024 is
1045 * enough for efficient IO.
1047 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1048 WB_REASON_FORKER_THREAD);
1049 trace_writeback_pages_written(pages_written);
1052 if (!list_empty(&bdi->work_list) ||
1053 (wb_has_dirty_io(wb) && dirty_writeback_interval))
1054 queue_delayed_work(bdi_wq, &wb->dwork,
1055 msecs_to_jiffies(dirty_writeback_interval * 10));
1057 current->flags &= ~PF_SWAPWRITE;
1061 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1064 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1066 struct backing_dev_info *bdi;
1069 nr_pages = get_nr_dirty_pages();
1072 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1073 if (!bdi_has_dirty_io(bdi))
1075 __bdi_start_writeback(bdi, nr_pages, false, reason);
1080 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1082 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1083 struct dentry *dentry;
1084 const char *name = "?";
1086 dentry = d_find_alias(inode);
1088 spin_lock(&dentry->d_lock);
1089 name = (const char *) dentry->d_name.name;
1092 "%s(%d): dirtied inode %lu (%s) on %s\n",
1093 current->comm, task_pid_nr(current), inode->i_ino,
1094 name, inode->i_sb->s_id);
1096 spin_unlock(&dentry->d_lock);
1103 * __mark_inode_dirty - internal function
1104 * @inode: inode to mark
1105 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1106 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1107 * mark_inode_dirty_sync.
1109 * Put the inode on the super block's dirty list.
1111 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1112 * dirty list only if it is hashed or if it refers to a blockdev.
1113 * If it was not hashed, it will never be added to the dirty list
1114 * even if it is later hashed, as it will have been marked dirty already.
1116 * In short, make sure you hash any inodes _before_ you start marking
1119 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1120 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1121 * the kernel-internal blockdev inode represents the dirtying time of the
1122 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1123 * page->mapping->host, so the page-dirtying time is recorded in the internal
1126 void __mark_inode_dirty(struct inode *inode, int flags)
1128 struct super_block *sb = inode->i_sb;
1129 struct backing_dev_info *bdi = NULL;
1132 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1133 * dirty the inode itself
1135 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1136 trace_writeback_dirty_inode_start(inode, flags);
1138 if (sb->s_op->dirty_inode)
1139 sb->s_op->dirty_inode(inode, flags);
1141 trace_writeback_dirty_inode(inode, flags);
1145 * make sure that changes are seen by all cpus before we test i_state
1150 /* avoid the locking if we can */
1151 if ((inode->i_state & flags) == flags)
1154 if (unlikely(block_dump))
1155 block_dump___mark_inode_dirty(inode);
1157 spin_lock(&inode->i_lock);
1158 if ((inode->i_state & flags) != flags) {
1159 const int was_dirty = inode->i_state & I_DIRTY;
1161 inode->i_state |= flags;
1164 * If the inode is being synced, just update its dirty state.
1165 * The unlocker will place the inode on the appropriate
1166 * superblock list, based upon its state.
1168 if (inode->i_state & I_SYNC)
1169 goto out_unlock_inode;
1172 * Only add valid (hashed) inodes to the superblock's
1173 * dirty list. Add blockdev inodes as well.
1175 if (!S_ISBLK(inode->i_mode)) {
1176 if (inode_unhashed(inode))
1177 goto out_unlock_inode;
1179 if (inode->i_state & I_FREEING)
1180 goto out_unlock_inode;
1183 * If the inode was already on b_dirty/b_io/b_more_io, don't
1184 * reposition it (that would break b_dirty time-ordering).
1187 bool wakeup_bdi = false;
1188 bdi = inode_to_bdi(inode);
1190 spin_unlock(&inode->i_lock);
1191 spin_lock(&bdi->wb.list_lock);
1192 if (bdi_cap_writeback_dirty(bdi)) {
1193 WARN(!test_bit(BDI_registered, &bdi->state),
1194 "bdi-%s not registered\n", bdi->name);
1197 * If this is the first dirty inode for this
1198 * bdi, we have to wake-up the corresponding
1199 * bdi thread to make sure background
1200 * write-back happens later.
1202 if (!wb_has_dirty_io(&bdi->wb))
1206 inode->dirtied_when = jiffies;
1207 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1208 spin_unlock(&bdi->wb.list_lock);
1211 bdi_wakeup_thread_delayed(bdi);
1216 spin_unlock(&inode->i_lock);
1219 EXPORT_SYMBOL(__mark_inode_dirty);
1221 static void wait_sb_inodes(struct super_block *sb)
1223 struct inode *inode, *old_inode = NULL;
1226 * We need to be protected against the filesystem going from
1227 * r/o to r/w or vice versa.
1229 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1231 spin_lock(&inode_sb_list_lock);
1234 * Data integrity sync. Must wait for all pages under writeback,
1235 * because there may have been pages dirtied before our sync
1236 * call, but which had writeout started before we write it out.
1237 * In which case, the inode may not be on the dirty list, but
1238 * we still have to wait for that writeout.
1240 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1241 struct address_space *mapping = inode->i_mapping;
1243 spin_lock(&inode->i_lock);
1244 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1245 (mapping->nrpages == 0)) {
1246 spin_unlock(&inode->i_lock);
1250 spin_unlock(&inode->i_lock);
1251 spin_unlock(&inode_sb_list_lock);
1254 * We hold a reference to 'inode' so it couldn't have been
1255 * removed from s_inodes list while we dropped the
1256 * inode_sb_list_lock. We cannot iput the inode now as we can
1257 * be holding the last reference and we cannot iput it under
1258 * inode_sb_list_lock. So we keep the reference and iput it
1264 filemap_fdatawait(mapping);
1268 spin_lock(&inode_sb_list_lock);
1270 spin_unlock(&inode_sb_list_lock);
1275 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1276 * @sb: the superblock
1277 * @nr: the number of pages to write
1278 * @reason: reason why some writeback work initiated
1280 * Start writeback on some inodes on this super_block. No guarantees are made
1281 * on how many (if any) will be written, and this function does not wait
1282 * for IO completion of submitted IO.
1284 void writeback_inodes_sb_nr(struct super_block *sb,
1286 enum wb_reason reason)
1288 DECLARE_COMPLETION_ONSTACK(done);
1289 struct wb_writeback_work work = {
1291 .sync_mode = WB_SYNC_NONE,
1292 .tagged_writepages = 1,
1298 if (sb->s_bdi == &noop_backing_dev_info)
1300 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1301 bdi_queue_work(sb->s_bdi, &work);
1302 wait_for_completion(&done);
1304 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1307 * writeback_inodes_sb - writeback dirty inodes from given super_block
1308 * @sb: the superblock
1309 * @reason: reason why some writeback work was initiated
1311 * Start writeback on some inodes on this super_block. No guarantees are made
1312 * on how many (if any) will be written, and this function does not wait
1313 * for IO completion of submitted IO.
1315 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1317 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1319 EXPORT_SYMBOL(writeback_inodes_sb);
1322 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1323 * @sb: the superblock
1324 * @nr: the number of pages to write
1325 * @reason: the reason of writeback
1327 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1328 * Returns 1 if writeback was started, 0 if not.
1330 int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1332 enum wb_reason reason)
1334 if (writeback_in_progress(sb->s_bdi))
1337 if (!down_read_trylock(&sb->s_umount))
1340 writeback_inodes_sb_nr(sb, nr, reason);
1341 up_read(&sb->s_umount);
1344 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1347 * try_to_writeback_inodes_sb - try to start writeback if none underway
1348 * @sb: the superblock
1349 * @reason: reason why some writeback work was initiated
1351 * Implement by try_to_writeback_inodes_sb_nr()
1352 * Returns 1 if writeback was started, 0 if not.
1354 int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1356 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1358 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
1361 * sync_inodes_sb - sync sb inode pages
1362 * @sb: the superblock
1363 * @older_than_this: timestamp
1365 * This function writes and waits on any dirty inode belonging to this
1366 * superblock that has been dirtied before given timestamp.
1368 void sync_inodes_sb(struct super_block *sb, unsigned long older_than_this)
1370 DECLARE_COMPLETION_ONSTACK(done);
1371 struct wb_writeback_work work = {
1373 .sync_mode = WB_SYNC_ALL,
1374 .nr_pages = LONG_MAX,
1375 .older_than_this = older_than_this,
1376 .older_than_this_is_set = 1,
1379 .reason = WB_REASON_SYNC,
1383 /* Nothing to do? */
1384 if (sb->s_bdi == &noop_backing_dev_info)
1386 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1388 bdi_queue_work(sb->s_bdi, &work);
1389 wait_for_completion(&done);
1393 EXPORT_SYMBOL(sync_inodes_sb);
1396 * write_inode_now - write an inode to disk
1397 * @inode: inode to write to disk
1398 * @sync: whether the write should be synchronous or not
1400 * This function commits an inode to disk immediately if it is dirty. This is
1401 * primarily needed by knfsd.
1403 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1405 int write_inode_now(struct inode *inode, int sync)
1407 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1408 struct writeback_control wbc = {
1409 .nr_to_write = LONG_MAX,
1410 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1412 .range_end = LLONG_MAX,
1415 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1416 wbc.nr_to_write = 0;
1419 return writeback_single_inode(inode, wb, &wbc);
1421 EXPORT_SYMBOL(write_inode_now);
1424 * sync_inode - write an inode and its pages to disk.
1425 * @inode: the inode to sync
1426 * @wbc: controls the writeback mode
1428 * sync_inode() will write an inode and its pages to disk. It will also
1429 * correctly update the inode on its superblock's dirty inode lists and will
1430 * update inode->i_state.
1432 * The caller must have a ref on the inode.
1434 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1436 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1438 EXPORT_SYMBOL(sync_inode);
1441 * sync_inode_metadata - write an inode to disk
1442 * @inode: the inode to sync
1443 * @wait: wait for I/O to complete.
1445 * Write an inode to disk and adjust its dirty state after completion.
1447 * Note: only writes the actual inode, no associated data or other metadata.
1449 int sync_inode_metadata(struct inode *inode, int wait)
1451 struct writeback_control wbc = {
1452 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1453 .nr_to_write = 0, /* metadata-only */
1456 return sync_inode(inode, &wbc);
1458 EXPORT_SYMBOL(sync_inode_metadata);