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;
43 unsigned long *older_than_this;
44 enum writeback_sync_modes sync_mode;
45 unsigned int tagged_writepages:1;
46 unsigned int for_kupdate:1;
47 unsigned int range_cyclic:1;
48 unsigned int for_background:1;
49 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
50 unsigned int auto_free:1; /* free on completion */
51 enum wb_reason reason; /* why was writeback initiated? */
53 struct list_head list; /* pending work list */
54 struct completion *done; /* set if the caller waits */
58 * If an inode is constantly having its pages dirtied, but then the
59 * updates stop dirtytime_expire_interval seconds in the past, it's
60 * possible for the worst case time between when an inode has its
61 * timestamps updated and when they finally get written out to be two
62 * dirtytime_expire_intervals. We set the default to 12 hours (in
63 * seconds), which means most of the time inodes will have their
64 * timestamps written to disk after 12 hours, but in the worst case a
65 * few inodes might not their timestamps updated for 24 hours.
67 unsigned int dirtytime_expire_interval = 12 * 60 * 60;
69 static inline struct inode *wb_inode(struct list_head *head)
71 return list_entry(head, struct inode, i_wb_list);
75 * Include the creation of the trace points after defining the
76 * wb_writeback_work structure and inline functions so that the definition
77 * remains local to this file.
79 #define CREATE_TRACE_POINTS
80 #include <trace/events/writeback.h>
82 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
84 static bool wb_io_lists_populated(struct bdi_writeback *wb)
86 if (wb_has_dirty_io(wb)) {
89 set_bit(WB_has_dirty_io, &wb->state);
90 WARN_ON_ONCE(!wb->avg_write_bandwidth);
91 atomic_long_add(wb->avg_write_bandwidth,
92 &wb->bdi->tot_write_bandwidth);
97 static void wb_io_lists_depopulated(struct bdi_writeback *wb)
99 if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) &&
100 list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) {
101 clear_bit(WB_has_dirty_io, &wb->state);
102 WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth,
103 &wb->bdi->tot_write_bandwidth) < 0);
108 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
109 * @inode: inode to be moved
110 * @wb: target bdi_writeback
111 * @head: one of @wb->b_{dirty|io|more_io}
113 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
114 * Returns %true if @inode is the first occupant of the !dirty_time IO
115 * lists; otherwise, %false.
117 static bool inode_wb_list_move_locked(struct inode *inode,
118 struct bdi_writeback *wb,
119 struct list_head *head)
121 assert_spin_locked(&wb->list_lock);
123 list_move(&inode->i_wb_list, head);
125 /* dirty_time doesn't count as dirty_io until expiration */
126 if (head != &wb->b_dirty_time)
127 return wb_io_lists_populated(wb);
129 wb_io_lists_depopulated(wb);
134 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
135 * @inode: inode to be removed
136 * @wb: bdi_writeback @inode is being removed from
138 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
139 * clear %WB_has_dirty_io if all are empty afterwards.
141 static void inode_wb_list_del_locked(struct inode *inode,
142 struct bdi_writeback *wb)
144 assert_spin_locked(&wb->list_lock);
146 list_del_init(&inode->i_wb_list);
147 wb_io_lists_depopulated(wb);
150 static void wb_wakeup(struct bdi_writeback *wb)
152 spin_lock_bh(&wb->work_lock);
153 if (test_bit(WB_registered, &wb->state))
154 mod_delayed_work(bdi_wq, &wb->dwork, 0);
155 spin_unlock_bh(&wb->work_lock);
158 static void wb_queue_work(struct bdi_writeback *wb,
159 struct wb_writeback_work *work)
161 trace_writeback_queue(wb->bdi, work);
163 spin_lock_bh(&wb->work_lock);
164 if (!test_bit(WB_registered, &wb->state)) {
166 complete(work->done);
169 list_add_tail(&work->list, &wb->work_list);
170 mod_delayed_work(bdi_wq, &wb->dwork, 0);
172 spin_unlock_bh(&wb->work_lock);
175 #ifdef CONFIG_CGROUP_WRITEBACK
178 * inode_congested - test whether an inode is congested
179 * @inode: inode to test for congestion
180 * @cong_bits: mask of WB_[a]sync_congested bits to test
182 * Tests whether @inode is congested. @cong_bits is the mask of congestion
183 * bits to test and the return value is the mask of set bits.
185 * If cgroup writeback is enabled for @inode, the congestion state is
186 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
187 * associated with @inode is congested; otherwise, the root wb's congestion
190 int inode_congested(struct inode *inode, int cong_bits)
193 struct bdi_writeback *wb = inode_to_wb(inode);
195 return wb_congested(wb, cong_bits);
198 return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
200 EXPORT_SYMBOL_GPL(inode_congested);
203 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
204 * @wb: target bdi_writeback to split @nr_pages to
205 * @nr_pages: number of pages to write for the whole bdi
207 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
208 * relation to the total write bandwidth of all wb's w/ dirty inodes on
211 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
213 unsigned long this_bw = wb->avg_write_bandwidth;
214 unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth);
216 if (nr_pages == LONG_MAX)
220 * This may be called on clean wb's and proportional distribution
221 * may not make sense, just use the original @nr_pages in those
222 * cases. In general, we wanna err on the side of writing more.
224 if (!tot_bw || this_bw >= tot_bw)
227 return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw);
230 #else /* CONFIG_CGROUP_WRITEBACK */
232 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
237 #endif /* CONFIG_CGROUP_WRITEBACK */
239 void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
240 bool range_cyclic, enum wb_reason reason)
242 struct wb_writeback_work *work;
244 if (!wb_has_dirty_io(wb))
248 * This is WB_SYNC_NONE writeback, so if allocation fails just
249 * wakeup the thread for old dirty data writeback
251 work = kzalloc(sizeof(*work), GFP_ATOMIC);
253 trace_writeback_nowork(wb->bdi);
258 work->sync_mode = WB_SYNC_NONE;
259 work->nr_pages = nr_pages;
260 work->range_cyclic = range_cyclic;
261 work->reason = reason;
264 wb_queue_work(wb, work);
268 * wb_start_background_writeback - start background writeback
269 * @wb: bdi_writback to write from
272 * This makes sure WB_SYNC_NONE background writeback happens. When
273 * this function returns, it is only guaranteed that for given wb
274 * some IO is happening if we are over background dirty threshold.
275 * Caller need not hold sb s_umount semaphore.
277 void wb_start_background_writeback(struct bdi_writeback *wb)
280 * We just wake up the flusher thread. It will perform background
281 * writeback as soon as there is no other work to do.
283 trace_writeback_wake_background(wb->bdi);
288 * Remove the inode from the writeback list it is on.
290 void inode_wb_list_del(struct inode *inode)
292 struct bdi_writeback *wb = inode_to_wb(inode);
294 spin_lock(&wb->list_lock);
295 inode_wb_list_del_locked(inode, wb);
296 spin_unlock(&wb->list_lock);
300 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
301 * furthest end of its superblock's dirty-inode list.
303 * Before stamping the inode's ->dirtied_when, we check to see whether it is
304 * already the most-recently-dirtied inode on the b_dirty list. If that is
305 * the case then the inode must have been redirtied while it was being written
306 * out and we don't reset its dirtied_when.
308 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
310 if (!list_empty(&wb->b_dirty)) {
313 tail = wb_inode(wb->b_dirty.next);
314 if (time_before(inode->dirtied_when, tail->dirtied_when))
315 inode->dirtied_when = jiffies;
317 inode_wb_list_move_locked(inode, wb, &wb->b_dirty);
321 * requeue inode for re-scanning after bdi->b_io list is exhausted.
323 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
325 inode_wb_list_move_locked(inode, wb, &wb->b_more_io);
328 static void inode_sync_complete(struct inode *inode)
330 inode->i_state &= ~I_SYNC;
331 /* If inode is clean an unused, put it into LRU now... */
332 inode_add_lru(inode);
333 /* Waiters must see I_SYNC cleared before being woken up */
335 wake_up_bit(&inode->i_state, __I_SYNC);
338 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
340 bool ret = time_after(inode->dirtied_when, t);
343 * For inodes being constantly redirtied, dirtied_when can get stuck.
344 * It _appears_ to be in the future, but is actually in distant past.
345 * This test is necessary to prevent such wrapped-around relative times
346 * from permanently stopping the whole bdi writeback.
348 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
353 #define EXPIRE_DIRTY_ATIME 0x0001
356 * Move expired (dirtied before work->older_than_this) dirty inodes from
357 * @delaying_queue to @dispatch_queue.
359 static int move_expired_inodes(struct list_head *delaying_queue,
360 struct list_head *dispatch_queue,
362 struct wb_writeback_work *work)
364 unsigned long *older_than_this = NULL;
365 unsigned long expire_time;
367 struct list_head *pos, *node;
368 struct super_block *sb = NULL;
373 if ((flags & EXPIRE_DIRTY_ATIME) == 0)
374 older_than_this = work->older_than_this;
375 else if (!work->for_sync) {
376 expire_time = jiffies - (dirtytime_expire_interval * HZ);
377 older_than_this = &expire_time;
379 while (!list_empty(delaying_queue)) {
380 inode = wb_inode(delaying_queue->prev);
381 if (older_than_this &&
382 inode_dirtied_after(inode, *older_than_this))
384 list_move(&inode->i_wb_list, &tmp);
386 if (flags & EXPIRE_DIRTY_ATIME)
387 set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
388 if (sb_is_blkdev_sb(inode->i_sb))
390 if (sb && sb != inode->i_sb)
395 /* just one sb in list, splice to dispatch_queue and we're done */
397 list_splice(&tmp, dispatch_queue);
401 /* Move inodes from one superblock together */
402 while (!list_empty(&tmp)) {
403 sb = wb_inode(tmp.prev)->i_sb;
404 list_for_each_prev_safe(pos, node, &tmp) {
405 inode = wb_inode(pos);
406 if (inode->i_sb == sb)
407 list_move(&inode->i_wb_list, dispatch_queue);
415 * Queue all expired dirty inodes for io, eldest first.
417 * newly dirtied b_dirty b_io b_more_io
418 * =============> gf edc BA
420 * newly dirtied b_dirty b_io b_more_io
421 * =============> g fBAedc
423 * +--> dequeue for IO
425 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
429 assert_spin_locked(&wb->list_lock);
430 list_splice_init(&wb->b_more_io, &wb->b_io);
431 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work);
432 moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io,
433 EXPIRE_DIRTY_ATIME, work);
435 wb_io_lists_populated(wb);
436 trace_writeback_queue_io(wb, work, moved);
439 static int write_inode(struct inode *inode, struct writeback_control *wbc)
443 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
444 trace_writeback_write_inode_start(inode, wbc);
445 ret = inode->i_sb->s_op->write_inode(inode, wbc);
446 trace_writeback_write_inode(inode, wbc);
453 * Wait for writeback on an inode to complete. Called with i_lock held.
454 * Caller must make sure inode cannot go away when we drop i_lock.
456 static void __inode_wait_for_writeback(struct inode *inode)
457 __releases(inode->i_lock)
458 __acquires(inode->i_lock)
460 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
461 wait_queue_head_t *wqh;
463 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
464 while (inode->i_state & I_SYNC) {
465 spin_unlock(&inode->i_lock);
466 __wait_on_bit(wqh, &wq, bit_wait,
467 TASK_UNINTERRUPTIBLE);
468 spin_lock(&inode->i_lock);
473 * Wait for writeback on an inode to complete. Caller must have inode pinned.
475 void inode_wait_for_writeback(struct inode *inode)
477 spin_lock(&inode->i_lock);
478 __inode_wait_for_writeback(inode);
479 spin_unlock(&inode->i_lock);
483 * Sleep until I_SYNC is cleared. This function must be called with i_lock
484 * held and drops it. It is aimed for callers not holding any inode reference
485 * so once i_lock is dropped, inode can go away.
487 static void inode_sleep_on_writeback(struct inode *inode)
488 __releases(inode->i_lock)
491 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
494 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
495 sleep = inode->i_state & I_SYNC;
496 spin_unlock(&inode->i_lock);
499 finish_wait(wqh, &wait);
503 * Find proper writeback list for the inode depending on its current state and
504 * possibly also change of its state while we were doing writeback. Here we
505 * handle things such as livelock prevention or fairness of writeback among
506 * inodes. This function can be called only by flusher thread - noone else
507 * processes all inodes in writeback lists and requeueing inodes behind flusher
508 * thread's back can have unexpected consequences.
510 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
511 struct writeback_control *wbc)
513 if (inode->i_state & I_FREEING)
517 * Sync livelock prevention. Each inode is tagged and synced in one
518 * shot. If still dirty, it will be redirty_tail()'ed below. Update
519 * the dirty time to prevent enqueue and sync it again.
521 if ((inode->i_state & I_DIRTY) &&
522 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
523 inode->dirtied_when = jiffies;
525 if (wbc->pages_skipped) {
527 * writeback is not making progress due to locked
528 * buffers. Skip this inode for now.
530 redirty_tail(inode, wb);
534 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
536 * We didn't write back all the pages. nfs_writepages()
537 * sometimes bales out without doing anything.
539 if (wbc->nr_to_write <= 0) {
540 /* Slice used up. Queue for next turn. */
541 requeue_io(inode, wb);
544 * Writeback blocked by something other than
545 * congestion. Delay the inode for some time to
546 * avoid spinning on the CPU (100% iowait)
547 * retrying writeback of the dirty page/inode
548 * that cannot be performed immediately.
550 redirty_tail(inode, wb);
552 } else if (inode->i_state & I_DIRTY) {
554 * Filesystems can dirty the inode during writeback operations,
555 * such as delayed allocation during submission or metadata
556 * updates after data IO completion.
558 redirty_tail(inode, wb);
559 } else if (inode->i_state & I_DIRTY_TIME) {
560 inode->dirtied_when = jiffies;
561 inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time);
563 /* The inode is clean. Remove from writeback lists. */
564 inode_wb_list_del_locked(inode, wb);
569 * Write out an inode and its dirty pages. Do not update the writeback list
570 * linkage. That is left to the caller. The caller is also responsible for
571 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
574 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
576 struct address_space *mapping = inode->i_mapping;
577 long nr_to_write = wbc->nr_to_write;
581 WARN_ON(!(inode->i_state & I_SYNC));
583 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
585 ret = do_writepages(mapping, wbc);
588 * Make sure to wait on the data before writing out the metadata.
589 * This is important for filesystems that modify metadata on data
590 * I/O completion. We don't do it for sync(2) writeback because it has a
591 * separate, external IO completion path and ->sync_fs for guaranteeing
592 * inode metadata is written back correctly.
594 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
595 int err = filemap_fdatawait(mapping);
601 * Some filesystems may redirty the inode during the writeback
602 * due to delalloc, clear dirty metadata flags right before
605 spin_lock(&inode->i_lock);
607 dirty = inode->i_state & I_DIRTY;
608 if (inode->i_state & I_DIRTY_TIME) {
609 if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
610 unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
611 unlikely(time_after(jiffies,
612 (inode->dirtied_time_when +
613 dirtytime_expire_interval * HZ)))) {
614 dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
615 trace_writeback_lazytime(inode);
618 inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
619 inode->i_state &= ~dirty;
622 * Paired with smp_mb() in __mark_inode_dirty(). This allows
623 * __mark_inode_dirty() to test i_state without grabbing i_lock -
624 * either they see the I_DIRTY bits cleared or we see the dirtied
627 * I_DIRTY_PAGES is always cleared together above even if @mapping
628 * still has dirty pages. The flag is reinstated after smp_mb() if
629 * necessary. This guarantees that either __mark_inode_dirty()
630 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
634 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
635 inode->i_state |= I_DIRTY_PAGES;
637 spin_unlock(&inode->i_lock);
639 if (dirty & I_DIRTY_TIME)
640 mark_inode_dirty_sync(inode);
641 /* Don't write the inode if only I_DIRTY_PAGES was set */
642 if (dirty & ~I_DIRTY_PAGES) {
643 int err = write_inode(inode, wbc);
647 trace_writeback_single_inode(inode, wbc, nr_to_write);
652 * Write out an inode's dirty pages. Either the caller has an active reference
653 * on the inode or the inode has I_WILL_FREE set.
655 * This function is designed to be called for writing back one inode which
656 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
657 * and does more profound writeback list handling in writeback_sb_inodes().
660 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
661 struct writeback_control *wbc)
665 spin_lock(&inode->i_lock);
666 if (!atomic_read(&inode->i_count))
667 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
669 WARN_ON(inode->i_state & I_WILL_FREE);
671 if (inode->i_state & I_SYNC) {
672 if (wbc->sync_mode != WB_SYNC_ALL)
675 * It's a data-integrity sync. We must wait. Since callers hold
676 * inode reference or inode has I_WILL_FREE set, it cannot go
679 __inode_wait_for_writeback(inode);
681 WARN_ON(inode->i_state & I_SYNC);
683 * Skip inode if it is clean and we have no outstanding writeback in
684 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
685 * function since flusher thread may be doing for example sync in
686 * parallel and if we move the inode, it could get skipped. So here we
687 * make sure inode is on some writeback list and leave it there unless
688 * we have completely cleaned the inode.
690 if (!(inode->i_state & I_DIRTY_ALL) &&
691 (wbc->sync_mode != WB_SYNC_ALL ||
692 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
694 inode->i_state |= I_SYNC;
695 spin_unlock(&inode->i_lock);
697 ret = __writeback_single_inode(inode, wbc);
699 spin_lock(&wb->list_lock);
700 spin_lock(&inode->i_lock);
702 * If inode is clean, remove it from writeback lists. Otherwise don't
703 * touch it. See comment above for explanation.
705 if (!(inode->i_state & I_DIRTY_ALL))
706 inode_wb_list_del_locked(inode, wb);
707 spin_unlock(&wb->list_lock);
708 inode_sync_complete(inode);
710 spin_unlock(&inode->i_lock);
714 static long writeback_chunk_size(struct bdi_writeback *wb,
715 struct wb_writeback_work *work)
720 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
721 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
722 * here avoids calling into writeback_inodes_wb() more than once.
724 * The intended call sequence for WB_SYNC_ALL writeback is:
727 * writeback_sb_inodes() <== called only once
728 * write_cache_pages() <== called once for each inode
729 * (quickly) tag currently dirty pages
730 * (maybe slowly) sync all tagged pages
732 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
735 pages = min(wb->avg_write_bandwidth / 2,
736 global_dirty_limit / DIRTY_SCOPE);
737 pages = min(pages, work->nr_pages);
738 pages = round_down(pages + MIN_WRITEBACK_PAGES,
739 MIN_WRITEBACK_PAGES);
746 * Write a portion of b_io inodes which belong to @sb.
748 * Return the number of pages and/or inodes written.
750 static long writeback_sb_inodes(struct super_block *sb,
751 struct bdi_writeback *wb,
752 struct wb_writeback_work *work)
754 struct writeback_control wbc = {
755 .sync_mode = work->sync_mode,
756 .tagged_writepages = work->tagged_writepages,
757 .for_kupdate = work->for_kupdate,
758 .for_background = work->for_background,
759 .for_sync = work->for_sync,
760 .range_cyclic = work->range_cyclic,
762 .range_end = LLONG_MAX,
764 unsigned long start_time = jiffies;
766 long wrote = 0; /* count both pages and inodes */
768 while (!list_empty(&wb->b_io)) {
769 struct inode *inode = wb_inode(wb->b_io.prev);
771 if (inode->i_sb != sb) {
774 * We only want to write back data for this
775 * superblock, move all inodes not belonging
776 * to it back onto the dirty list.
778 redirty_tail(inode, wb);
783 * The inode belongs to a different superblock.
784 * Bounce back to the caller to unpin this and
785 * pin the next superblock.
791 * Don't bother with new inodes or inodes being freed, first
792 * kind does not need periodic writeout yet, and for the latter
793 * kind writeout is handled by the freer.
795 spin_lock(&inode->i_lock);
796 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
797 spin_unlock(&inode->i_lock);
798 redirty_tail(inode, wb);
801 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
803 * If this inode is locked for writeback and we are not
804 * doing writeback-for-data-integrity, move it to
805 * b_more_io so that writeback can proceed with the
806 * other inodes on s_io.
808 * We'll have another go at writing back this inode
809 * when we completed a full scan of b_io.
811 spin_unlock(&inode->i_lock);
812 requeue_io(inode, wb);
813 trace_writeback_sb_inodes_requeue(inode);
816 spin_unlock(&wb->list_lock);
819 * We already requeued the inode if it had I_SYNC set and we
820 * are doing WB_SYNC_NONE writeback. So this catches only the
823 if (inode->i_state & I_SYNC) {
824 /* Wait for I_SYNC. This function drops i_lock... */
825 inode_sleep_on_writeback(inode);
826 /* Inode may be gone, start again */
827 spin_lock(&wb->list_lock);
830 inode->i_state |= I_SYNC;
831 spin_unlock(&inode->i_lock);
833 write_chunk = writeback_chunk_size(wb, work);
834 wbc.nr_to_write = write_chunk;
835 wbc.pages_skipped = 0;
838 * We use I_SYNC to pin the inode in memory. While it is set
839 * evict_inode() will wait so the inode cannot be freed.
841 __writeback_single_inode(inode, &wbc);
843 work->nr_pages -= write_chunk - wbc.nr_to_write;
844 wrote += write_chunk - wbc.nr_to_write;
845 spin_lock(&wb->list_lock);
846 spin_lock(&inode->i_lock);
847 if (!(inode->i_state & I_DIRTY_ALL))
849 requeue_inode(inode, wb, &wbc);
850 inode_sync_complete(inode);
851 spin_unlock(&inode->i_lock);
852 cond_resched_lock(&wb->list_lock);
854 * bail out to wb_writeback() often enough to check
855 * background threshold and other termination conditions.
858 if (time_is_before_jiffies(start_time + HZ / 10UL))
860 if (work->nr_pages <= 0)
867 static long __writeback_inodes_wb(struct bdi_writeback *wb,
868 struct wb_writeback_work *work)
870 unsigned long start_time = jiffies;
873 while (!list_empty(&wb->b_io)) {
874 struct inode *inode = wb_inode(wb->b_io.prev);
875 struct super_block *sb = inode->i_sb;
877 if (!trylock_super(sb)) {
879 * trylock_super() may fail consistently due to
880 * s_umount being grabbed by someone else. Don't use
881 * requeue_io() to avoid busy retrying the inode/sb.
883 redirty_tail(inode, wb);
886 wrote += writeback_sb_inodes(sb, wb, work);
887 up_read(&sb->s_umount);
889 /* refer to the same tests at the end of writeback_sb_inodes */
891 if (time_is_before_jiffies(start_time + HZ / 10UL))
893 if (work->nr_pages <= 0)
897 /* Leave any unwritten inodes on b_io */
901 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
902 enum wb_reason reason)
904 struct wb_writeback_work work = {
905 .nr_pages = nr_pages,
906 .sync_mode = WB_SYNC_NONE,
911 spin_lock(&wb->list_lock);
912 if (list_empty(&wb->b_io))
914 __writeback_inodes_wb(wb, &work);
915 spin_unlock(&wb->list_lock);
917 return nr_pages - work.nr_pages;
920 static bool over_bground_thresh(struct bdi_writeback *wb)
922 unsigned long background_thresh, dirty_thresh;
924 global_dirty_limits(&background_thresh, &dirty_thresh);
926 if (global_page_state(NR_FILE_DIRTY) +
927 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
930 if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh))
937 * Called under wb->list_lock. If there are multiple wb per bdi,
938 * only the flusher working on the first wb should do it.
940 static void wb_update_bandwidth(struct bdi_writeback *wb,
941 unsigned long start_time)
943 __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time);
947 * Explicit flushing or periodic writeback of "old" data.
949 * Define "old": the first time one of an inode's pages is dirtied, we mark the
950 * dirtying-time in the inode's address_space. So this periodic writeback code
951 * just walks the superblock inode list, writing back any inodes which are
952 * older than a specific point in time.
954 * Try to run once per dirty_writeback_interval. But if a writeback event
955 * takes longer than a dirty_writeback_interval interval, then leave a
958 * older_than_this takes precedence over nr_to_write. So we'll only write back
959 * all dirty pages if they are all attached to "old" mappings.
961 static long wb_writeback(struct bdi_writeback *wb,
962 struct wb_writeback_work *work)
964 unsigned long wb_start = jiffies;
965 long nr_pages = work->nr_pages;
966 unsigned long oldest_jif;
970 oldest_jif = jiffies;
971 work->older_than_this = &oldest_jif;
973 spin_lock(&wb->list_lock);
976 * Stop writeback when nr_pages has been consumed
978 if (work->nr_pages <= 0)
982 * Background writeout and kupdate-style writeback may
983 * run forever. Stop them if there is other work to do
984 * so that e.g. sync can proceed. They'll be restarted
985 * after the other works are all done.
987 if ((work->for_background || work->for_kupdate) &&
988 !list_empty(&wb->work_list))
992 * For background writeout, stop when we are below the
993 * background dirty threshold
995 if (work->for_background && !over_bground_thresh(wb))
999 * Kupdate and background works are special and we want to
1000 * include all inodes that need writing. Livelock avoidance is
1001 * handled by these works yielding to any other work so we are
1004 if (work->for_kupdate) {
1005 oldest_jif = jiffies -
1006 msecs_to_jiffies(dirty_expire_interval * 10);
1007 } else if (work->for_background)
1008 oldest_jif = jiffies;
1010 trace_writeback_start(wb->bdi, work);
1011 if (list_empty(&wb->b_io))
1014 progress = writeback_sb_inodes(work->sb, wb, work);
1016 progress = __writeback_inodes_wb(wb, work);
1017 trace_writeback_written(wb->bdi, work);
1019 wb_update_bandwidth(wb, wb_start);
1022 * Did we write something? Try for more
1024 * Dirty inodes are moved to b_io for writeback in batches.
1025 * The completion of the current batch does not necessarily
1026 * mean the overall work is done. So we keep looping as long
1027 * as made some progress on cleaning pages or inodes.
1032 * No more inodes for IO, bail
1034 if (list_empty(&wb->b_more_io))
1037 * Nothing written. Wait for some inode to
1038 * become available for writeback. Otherwise
1039 * we'll just busyloop.
1041 if (!list_empty(&wb->b_more_io)) {
1042 trace_writeback_wait(wb->bdi, work);
1043 inode = wb_inode(wb->b_more_io.prev);
1044 spin_lock(&inode->i_lock);
1045 spin_unlock(&wb->list_lock);
1046 /* This function drops i_lock... */
1047 inode_sleep_on_writeback(inode);
1048 spin_lock(&wb->list_lock);
1051 spin_unlock(&wb->list_lock);
1053 return nr_pages - work->nr_pages;
1057 * Return the next wb_writeback_work struct that hasn't been processed yet.
1059 static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb)
1061 struct wb_writeback_work *work = NULL;
1063 spin_lock_bh(&wb->work_lock);
1064 if (!list_empty(&wb->work_list)) {
1065 work = list_entry(wb->work_list.next,
1066 struct wb_writeback_work, list);
1067 list_del_init(&work->list);
1069 spin_unlock_bh(&wb->work_lock);
1074 * Add in the number of potentially dirty inodes, because each inode
1075 * write can dirty pagecache in the underlying blockdev.
1077 static unsigned long get_nr_dirty_pages(void)
1079 return global_page_state(NR_FILE_DIRTY) +
1080 global_page_state(NR_UNSTABLE_NFS) +
1081 get_nr_dirty_inodes();
1084 static long wb_check_background_flush(struct bdi_writeback *wb)
1086 if (over_bground_thresh(wb)) {
1088 struct wb_writeback_work work = {
1089 .nr_pages = LONG_MAX,
1090 .sync_mode = WB_SYNC_NONE,
1091 .for_background = 1,
1093 .reason = WB_REASON_BACKGROUND,
1096 return wb_writeback(wb, &work);
1102 static long wb_check_old_data_flush(struct bdi_writeback *wb)
1104 unsigned long expired;
1108 * When set to zero, disable periodic writeback
1110 if (!dirty_writeback_interval)
1113 expired = wb->last_old_flush +
1114 msecs_to_jiffies(dirty_writeback_interval * 10);
1115 if (time_before(jiffies, expired))
1118 wb->last_old_flush = jiffies;
1119 nr_pages = get_nr_dirty_pages();
1122 struct wb_writeback_work work = {
1123 .nr_pages = nr_pages,
1124 .sync_mode = WB_SYNC_NONE,
1127 .reason = WB_REASON_PERIODIC,
1130 return wb_writeback(wb, &work);
1137 * Retrieve work items and do the writeback they describe
1139 static long wb_do_writeback(struct bdi_writeback *wb)
1141 struct wb_writeback_work *work;
1144 set_bit(WB_writeback_running, &wb->state);
1145 while ((work = get_next_work_item(wb)) != NULL) {
1146 struct completion *done = work->done;
1148 trace_writeback_exec(wb->bdi, work);
1150 wrote += wb_writeback(wb, work);
1152 if (work->auto_free)
1159 * Check for periodic writeback, kupdated() style
1161 wrote += wb_check_old_data_flush(wb);
1162 wrote += wb_check_background_flush(wb);
1163 clear_bit(WB_writeback_running, &wb->state);
1169 * Handle writeback of dirty data for the device backed by this bdi. Also
1170 * reschedules periodically and does kupdated style flushing.
1172 void wb_workfn(struct work_struct *work)
1174 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1175 struct bdi_writeback, dwork);
1178 set_worker_desc("flush-%s", dev_name(wb->bdi->dev));
1179 current->flags |= PF_SWAPWRITE;
1181 if (likely(!current_is_workqueue_rescuer() ||
1182 !test_bit(WB_registered, &wb->state))) {
1184 * The normal path. Keep writing back @wb until its
1185 * work_list is empty. Note that this path is also taken
1186 * if @wb is shutting down even when we're running off the
1187 * rescuer as work_list needs to be drained.
1190 pages_written = wb_do_writeback(wb);
1191 trace_writeback_pages_written(pages_written);
1192 } while (!list_empty(&wb->work_list));
1195 * bdi_wq can't get enough workers and we're running off
1196 * the emergency worker. Don't hog it. Hopefully, 1024 is
1197 * enough for efficient IO.
1199 pages_written = writeback_inodes_wb(wb, 1024,
1200 WB_REASON_FORKER_THREAD);
1201 trace_writeback_pages_written(pages_written);
1204 if (!list_empty(&wb->work_list))
1205 mod_delayed_work(bdi_wq, &wb->dwork, 0);
1206 else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1207 wb_wakeup_delayed(wb);
1209 current->flags &= ~PF_SWAPWRITE;
1213 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1216 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1218 struct backing_dev_info *bdi;
1221 nr_pages = get_nr_dirty_pages();
1224 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1225 struct bdi_writeback *wb;
1226 struct wb_iter iter;
1228 if (!bdi_has_dirty_io(bdi))
1231 bdi_for_each_wb(wb, bdi, &iter, 0)
1232 wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages),
1239 * Wake up bdi's periodically to make sure dirtytime inodes gets
1240 * written back periodically. We deliberately do *not* check the
1241 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1242 * kernel to be constantly waking up once there are any dirtytime
1243 * inodes on the system. So instead we define a separate delayed work
1244 * function which gets called much more rarely. (By default, only
1245 * once every 12 hours.)
1247 * If there is any other write activity going on in the file system,
1248 * this function won't be necessary. But if the only thing that has
1249 * happened on the file system is a dirtytime inode caused by an atime
1250 * update, we need this infrastructure below to make sure that inode
1251 * eventually gets pushed out to disk.
1253 static void wakeup_dirtytime_writeback(struct work_struct *w);
1254 static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback);
1256 static void wakeup_dirtytime_writeback(struct work_struct *w)
1258 struct backing_dev_info *bdi;
1261 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1262 struct bdi_writeback *wb;
1263 struct wb_iter iter;
1265 bdi_for_each_wb(wb, bdi, &iter, 0)
1266 if (!list_empty(&bdi->wb.b_dirty_time))
1267 wb_wakeup(&bdi->wb);
1270 schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
1273 static int __init start_dirtytime_writeback(void)
1275 schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
1278 __initcall(start_dirtytime_writeback);
1280 int dirtytime_interval_handler(struct ctl_table *table, int write,
1281 void __user *buffer, size_t *lenp, loff_t *ppos)
1285 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
1286 if (ret == 0 && write)
1287 mod_delayed_work(system_wq, &dirtytime_work, 0);
1291 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1293 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1294 struct dentry *dentry;
1295 const char *name = "?";
1297 dentry = d_find_alias(inode);
1299 spin_lock(&dentry->d_lock);
1300 name = (const char *) dentry->d_name.name;
1303 "%s(%d): dirtied inode %lu (%s) on %s\n",
1304 current->comm, task_pid_nr(current), inode->i_ino,
1305 name, inode->i_sb->s_id);
1307 spin_unlock(&dentry->d_lock);
1314 * __mark_inode_dirty - internal function
1315 * @inode: inode to mark
1316 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1317 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1318 * mark_inode_dirty_sync.
1320 * Put the inode on the super block's dirty list.
1322 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1323 * dirty list only if it is hashed or if it refers to a blockdev.
1324 * If it was not hashed, it will never be added to the dirty list
1325 * even if it is later hashed, as it will have been marked dirty already.
1327 * In short, make sure you hash any inodes _before_ you start marking
1330 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1331 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1332 * the kernel-internal blockdev inode represents the dirtying time of the
1333 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1334 * page->mapping->host, so the page-dirtying time is recorded in the internal
1337 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1338 void __mark_inode_dirty(struct inode *inode, int flags)
1340 struct super_block *sb = inode->i_sb;
1341 struct backing_dev_info *bdi = NULL;
1344 trace_writeback_mark_inode_dirty(inode, flags);
1347 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1348 * dirty the inode itself
1350 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) {
1351 trace_writeback_dirty_inode_start(inode, flags);
1353 if (sb->s_op->dirty_inode)
1354 sb->s_op->dirty_inode(inode, flags);
1356 trace_writeback_dirty_inode(inode, flags);
1358 if (flags & I_DIRTY_INODE)
1359 flags &= ~I_DIRTY_TIME;
1360 dirtytime = flags & I_DIRTY_TIME;
1363 * Paired with smp_mb() in __writeback_single_inode() for the
1364 * following lockless i_state test. See there for details.
1368 if (((inode->i_state & flags) == flags) ||
1369 (dirtytime && (inode->i_state & I_DIRTY_INODE)))
1372 if (unlikely(block_dump))
1373 block_dump___mark_inode_dirty(inode);
1375 spin_lock(&inode->i_lock);
1376 if (dirtytime && (inode->i_state & I_DIRTY_INODE))
1377 goto out_unlock_inode;
1378 if ((inode->i_state & flags) != flags) {
1379 const int was_dirty = inode->i_state & I_DIRTY;
1381 inode_attach_wb(inode, NULL);
1383 if (flags & I_DIRTY_INODE)
1384 inode->i_state &= ~I_DIRTY_TIME;
1385 inode->i_state |= flags;
1388 * If the inode is being synced, just update its dirty state.
1389 * The unlocker will place the inode on the appropriate
1390 * superblock list, based upon its state.
1392 if (inode->i_state & I_SYNC)
1393 goto out_unlock_inode;
1396 * Only add valid (hashed) inodes to the superblock's
1397 * dirty list. Add blockdev inodes as well.
1399 if (!S_ISBLK(inode->i_mode)) {
1400 if (inode_unhashed(inode))
1401 goto out_unlock_inode;
1403 if (inode->i_state & I_FREEING)
1404 goto out_unlock_inode;
1407 * If the inode was already on b_dirty/b_io/b_more_io, don't
1408 * reposition it (that would break b_dirty time-ordering).
1411 struct list_head *dirty_list;
1412 bool wakeup_bdi = false;
1413 bdi = inode_to_bdi(inode);
1415 spin_unlock(&inode->i_lock);
1416 spin_lock(&bdi->wb.list_lock);
1418 WARN(bdi_cap_writeback_dirty(bdi) &&
1419 !test_bit(WB_registered, &bdi->wb.state),
1420 "bdi-%s not registered\n", bdi->name);
1422 inode->dirtied_when = jiffies;
1424 inode->dirtied_time_when = jiffies;
1426 if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES))
1427 dirty_list = &bdi->wb.b_dirty;
1429 dirty_list = &bdi->wb.b_dirty_time;
1431 wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb,
1434 spin_unlock(&bdi->wb.list_lock);
1435 trace_writeback_dirty_inode_enqueue(inode);
1438 * If this is the first dirty inode for this bdi,
1439 * we have to wake-up the corresponding bdi thread
1440 * to make sure background write-back happens
1443 if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi)
1444 wb_wakeup_delayed(&bdi->wb);
1449 spin_unlock(&inode->i_lock);
1452 EXPORT_SYMBOL(__mark_inode_dirty);
1454 static void wait_sb_inodes(struct super_block *sb)
1456 struct inode *inode, *old_inode = NULL;
1459 * We need to be protected against the filesystem going from
1460 * r/o to r/w or vice versa.
1462 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1464 spin_lock(&inode_sb_list_lock);
1467 * Data integrity sync. Must wait for all pages under writeback,
1468 * because there may have been pages dirtied before our sync
1469 * call, but which had writeout started before we write it out.
1470 * In which case, the inode may not be on the dirty list, but
1471 * we still have to wait for that writeout.
1473 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1474 struct address_space *mapping = inode->i_mapping;
1476 spin_lock(&inode->i_lock);
1477 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1478 (mapping->nrpages == 0)) {
1479 spin_unlock(&inode->i_lock);
1483 spin_unlock(&inode->i_lock);
1484 spin_unlock(&inode_sb_list_lock);
1487 * We hold a reference to 'inode' so it couldn't have been
1488 * removed from s_inodes list while we dropped the
1489 * inode_sb_list_lock. We cannot iput the inode now as we can
1490 * be holding the last reference and we cannot iput it under
1491 * inode_sb_list_lock. So we keep the reference and iput it
1497 filemap_fdatawait(mapping);
1501 spin_lock(&inode_sb_list_lock);
1503 spin_unlock(&inode_sb_list_lock);
1508 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1509 * @sb: the superblock
1510 * @nr: the number of pages to write
1511 * @reason: reason why some writeback work initiated
1513 * Start writeback on some inodes on this super_block. No guarantees are made
1514 * on how many (if any) will be written, and this function does not wait
1515 * for IO completion of submitted IO.
1517 void writeback_inodes_sb_nr(struct super_block *sb,
1519 enum wb_reason reason)
1521 DECLARE_COMPLETION_ONSTACK(done);
1522 struct wb_writeback_work work = {
1524 .sync_mode = WB_SYNC_NONE,
1525 .tagged_writepages = 1,
1530 struct backing_dev_info *bdi = sb->s_bdi;
1532 if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info)
1534 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1535 wb_queue_work(&bdi->wb, &work);
1536 wait_for_completion(&done);
1538 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1541 * writeback_inodes_sb - writeback dirty inodes from given super_block
1542 * @sb: the superblock
1543 * @reason: reason why some writeback work was initiated
1545 * Start writeback on some inodes on this super_block. No guarantees are made
1546 * on how many (if any) will be written, and this function does not wait
1547 * for IO completion of submitted IO.
1549 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1551 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1553 EXPORT_SYMBOL(writeback_inodes_sb);
1556 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1557 * @sb: the superblock
1558 * @nr: the number of pages to write
1559 * @reason: the reason of writeback
1561 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1562 * Returns 1 if writeback was started, 0 if not.
1564 int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1566 enum wb_reason reason)
1568 if (writeback_in_progress(&sb->s_bdi->wb))
1571 if (!down_read_trylock(&sb->s_umount))
1574 writeback_inodes_sb_nr(sb, nr, reason);
1575 up_read(&sb->s_umount);
1578 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1581 * try_to_writeback_inodes_sb - try to start writeback if none underway
1582 * @sb: the superblock
1583 * @reason: reason why some writeback work was initiated
1585 * Implement by try_to_writeback_inodes_sb_nr()
1586 * Returns 1 if writeback was started, 0 if not.
1588 int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1590 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1592 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
1595 * sync_inodes_sb - sync sb inode pages
1596 * @sb: the superblock
1598 * This function writes and waits on any dirty inode belonging to this
1601 void sync_inodes_sb(struct super_block *sb)
1603 DECLARE_COMPLETION_ONSTACK(done);
1604 struct wb_writeback_work work = {
1606 .sync_mode = WB_SYNC_ALL,
1607 .nr_pages = LONG_MAX,
1610 .reason = WB_REASON_SYNC,
1613 struct backing_dev_info *bdi = sb->s_bdi;
1615 /* Nothing to do? */
1616 if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info)
1618 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1620 wb_queue_work(&bdi->wb, &work);
1621 wait_for_completion(&done);
1625 EXPORT_SYMBOL(sync_inodes_sb);
1628 * write_inode_now - write an inode to disk
1629 * @inode: inode to write to disk
1630 * @sync: whether the write should be synchronous or not
1632 * This function commits an inode to disk immediately if it is dirty. This is
1633 * primarily needed by knfsd.
1635 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1637 int write_inode_now(struct inode *inode, int sync)
1639 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1640 struct writeback_control wbc = {
1641 .nr_to_write = LONG_MAX,
1642 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1644 .range_end = LLONG_MAX,
1647 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1648 wbc.nr_to_write = 0;
1651 return writeback_single_inode(inode, wb, &wbc);
1653 EXPORT_SYMBOL(write_inode_now);
1656 * sync_inode - write an inode and its pages to disk.
1657 * @inode: the inode to sync
1658 * @wbc: controls the writeback mode
1660 * sync_inode() will write an inode and its pages to disk. It will also
1661 * correctly update the inode on its superblock's dirty inode lists and will
1662 * update inode->i_state.
1664 * The caller must have a ref on the inode.
1666 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1668 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1670 EXPORT_SYMBOL(sync_inode);
1673 * sync_inode_metadata - write an inode to disk
1674 * @inode: the inode to sync
1675 * @wait: wait for I/O to complete.
1677 * Write an inode to disk and adjust its dirty state after completion.
1679 * Note: only writes the actual inode, no associated data or other metadata.
1681 int sync_inode_metadata(struct inode *inode, int wait)
1683 struct writeback_control wbc = {
1684 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1685 .nr_to_write = 0, /* metadata-only */
1688 return sync_inode(inode, &wbc);
1690 EXPORT_SYMBOL(sync_inode_metadata);