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>
30 #include <linux/memcontrol.h>
34 * 4MB minimal write chunk size
36 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
38 struct wb_completion {
43 * Passed into wb_writeback(), essentially a subset of writeback_control
45 struct wb_writeback_work {
47 struct super_block *sb;
48 unsigned long *older_than_this;
49 enum writeback_sync_modes sync_mode;
50 unsigned int tagged_writepages:1;
51 unsigned int for_kupdate:1;
52 unsigned int range_cyclic:1;
53 unsigned int for_background:1;
54 unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */
55 unsigned int auto_free:1; /* free on completion */
56 enum wb_reason reason; /* why was writeback initiated? */
58 struct list_head list; /* pending work list */
59 struct wb_completion *done; /* set if the caller waits */
63 * If one wants to wait for one or more wb_writeback_works, each work's
64 * ->done should be set to a wb_completion defined using the following
65 * macro. Once all work items are issued with wb_queue_work(), the caller
66 * can wait for the completion of all using wb_wait_for_completion(). Work
67 * items which are waited upon aren't freed automatically on completion.
69 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \
70 struct wb_completion cmpl = { \
71 .cnt = ATOMIC_INIT(1), \
76 * If an inode is constantly having its pages dirtied, but then the
77 * updates stop dirtytime_expire_interval seconds in the past, it's
78 * possible for the worst case time between when an inode has its
79 * timestamps updated and when they finally get written out to be two
80 * dirtytime_expire_intervals. We set the default to 12 hours (in
81 * seconds), which means most of the time inodes will have their
82 * timestamps written to disk after 12 hours, but in the worst case a
83 * few inodes might not their timestamps updated for 24 hours.
85 unsigned int dirtytime_expire_interval = 12 * 60 * 60;
87 static inline struct inode *wb_inode(struct list_head *head)
89 return list_entry(head, struct inode, i_io_list);
93 * Include the creation of the trace points after defining the
94 * wb_writeback_work structure and inline functions so that the definition
95 * remains local to this file.
97 #define CREATE_TRACE_POINTS
98 #include <trace/events/writeback.h>
100 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
102 static bool wb_io_lists_populated(struct bdi_writeback *wb)
104 if (wb_has_dirty_io(wb)) {
107 set_bit(WB_has_dirty_io, &wb->state);
108 WARN_ON_ONCE(!wb->avg_write_bandwidth);
109 atomic_long_add(wb->avg_write_bandwidth,
110 &wb->bdi->tot_write_bandwidth);
115 static void wb_io_lists_depopulated(struct bdi_writeback *wb)
117 if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) &&
118 list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) {
119 clear_bit(WB_has_dirty_io, &wb->state);
120 WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth,
121 &wb->bdi->tot_write_bandwidth) < 0);
126 * inode_io_list_move_locked - move an inode onto a bdi_writeback IO list
127 * @inode: inode to be moved
128 * @wb: target bdi_writeback
129 * @head: one of @wb->b_{dirty|io|more_io}
131 * Move @inode->i_io_list to @list of @wb and set %WB_has_dirty_io.
132 * Returns %true if @inode is the first occupant of the !dirty_time IO
133 * lists; otherwise, %false.
135 static bool inode_io_list_move_locked(struct inode *inode,
136 struct bdi_writeback *wb,
137 struct list_head *head)
139 assert_spin_locked(&wb->list_lock);
141 list_move(&inode->i_io_list, head);
143 /* dirty_time doesn't count as dirty_io until expiration */
144 if (head != &wb->b_dirty_time)
145 return wb_io_lists_populated(wb);
147 wb_io_lists_depopulated(wb);
152 * inode_io_list_del_locked - remove an inode from its bdi_writeback IO list
153 * @inode: inode to be removed
154 * @wb: bdi_writeback @inode is being removed from
156 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
157 * clear %WB_has_dirty_io if all are empty afterwards.
159 static void inode_io_list_del_locked(struct inode *inode,
160 struct bdi_writeback *wb)
162 assert_spin_locked(&wb->list_lock);
164 list_del_init(&inode->i_io_list);
165 wb_io_lists_depopulated(wb);
168 static void wb_wakeup(struct bdi_writeback *wb)
170 spin_lock_bh(&wb->work_lock);
171 if (test_bit(WB_registered, &wb->state))
172 mod_delayed_work(bdi_wq, &wb->dwork, 0);
173 spin_unlock_bh(&wb->work_lock);
176 static void wb_queue_work(struct bdi_writeback *wb,
177 struct wb_writeback_work *work)
179 trace_writeback_queue(wb, work);
181 spin_lock_bh(&wb->work_lock);
182 if (!test_bit(WB_registered, &wb->state))
185 atomic_inc(&work->done->cnt);
186 list_add_tail(&work->list, &wb->work_list);
187 mod_delayed_work(bdi_wq, &wb->dwork, 0);
189 spin_unlock_bh(&wb->work_lock);
193 * wb_wait_for_completion - wait for completion of bdi_writeback_works
194 * @bdi: bdi work items were issued to
195 * @done: target wb_completion
197 * Wait for one or more work items issued to @bdi with their ->done field
198 * set to @done, which should have been defined with
199 * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such
200 * work items are completed. Work items which are waited upon aren't freed
201 * automatically on completion.
203 static void wb_wait_for_completion(struct backing_dev_info *bdi,
204 struct wb_completion *done)
206 atomic_dec(&done->cnt); /* put down the initial count */
207 wait_event(bdi->wb_waitq, !atomic_read(&done->cnt));
210 #ifdef CONFIG_CGROUP_WRITEBACK
212 /* parameters for foreign inode detection, see wb_detach_inode() */
213 #define WB_FRN_TIME_SHIFT 13 /* 1s = 2^13, upto 8 secs w/ 16bit */
214 #define WB_FRN_TIME_AVG_SHIFT 3 /* avg = avg * 7/8 + new * 1/8 */
215 #define WB_FRN_TIME_CUT_DIV 2 /* ignore rounds < avg / 2 */
216 #define WB_FRN_TIME_PERIOD (2 * (1 << WB_FRN_TIME_SHIFT)) /* 2s */
218 #define WB_FRN_HIST_SLOTS 16 /* inode->i_wb_frn_history is 16bit */
219 #define WB_FRN_HIST_UNIT (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS)
220 /* each slot's duration is 2s / 16 */
221 #define WB_FRN_HIST_THR_SLOTS (WB_FRN_HIST_SLOTS / 2)
222 /* if foreign slots >= 8, switch */
223 #define WB_FRN_HIST_MAX_SLOTS (WB_FRN_HIST_THR_SLOTS / 2 + 1)
224 /* one round can affect upto 5 slots */
226 void __inode_attach_wb(struct inode *inode, struct page *page)
228 struct backing_dev_info *bdi = inode_to_bdi(inode);
229 struct bdi_writeback *wb = NULL;
231 if (inode_cgwb_enabled(inode)) {
232 struct cgroup_subsys_state *memcg_css;
235 memcg_css = mem_cgroup_css_from_page(page);
236 wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
238 /* must pin memcg_css, see wb_get_create() */
239 memcg_css = task_get_css(current, memory_cgrp_id);
240 wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
249 * There may be multiple instances of this function racing to
250 * update the same inode. Use cmpxchg() to tell the winner.
252 if (unlikely(cmpxchg(&inode->i_wb, NULL, wb)))
257 * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it
258 * @inode: inode of interest with i_lock held
260 * Returns @inode's wb with its list_lock held. @inode->i_lock must be
261 * held on entry and is released on return. The returned wb is guaranteed
262 * to stay @inode's associated wb until its list_lock is released.
264 static struct bdi_writeback *
265 locked_inode_to_wb_and_lock_list(struct inode *inode)
266 __releases(&inode->i_lock)
267 __acquires(&wb->list_lock)
270 struct bdi_writeback *wb = inode_to_wb(inode);
273 * inode_to_wb() association is protected by both
274 * @inode->i_lock and @wb->list_lock but list_lock nests
275 * outside i_lock. Drop i_lock and verify that the
276 * association hasn't changed after acquiring list_lock.
279 spin_unlock(&inode->i_lock);
280 spin_lock(&wb->list_lock);
281 wb_put(wb); /* not gonna deref it anymore */
283 /* i_wb may have changed inbetween, can't use inode_to_wb() */
284 if (likely(wb == inode->i_wb))
285 return wb; /* @inode already has ref */
287 spin_unlock(&wb->list_lock);
289 spin_lock(&inode->i_lock);
294 * inode_to_wb_and_lock_list - determine an inode's wb and lock it
295 * @inode: inode of interest
297 * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held
300 static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode)
301 __acquires(&wb->list_lock)
303 spin_lock(&inode->i_lock);
304 return locked_inode_to_wb_and_lock_list(inode);
307 struct inode_switch_wbs_context {
309 struct bdi_writeback *new_wb;
311 struct rcu_head rcu_head;
312 struct work_struct work;
315 static void inode_switch_wbs_work_fn(struct work_struct *work)
317 struct inode_switch_wbs_context *isw =
318 container_of(work, struct inode_switch_wbs_context, work);
319 struct inode *inode = isw->inode;
320 struct super_block *sb = inode->i_sb;
321 struct address_space *mapping = inode->i_mapping;
322 struct bdi_writeback *old_wb = inode->i_wb;
323 struct bdi_writeback *new_wb = isw->new_wb;
324 struct radix_tree_iter iter;
325 bool switched = false;
329 * By the time control reaches here, RCU grace period has passed
330 * since I_WB_SWITCH assertion and all wb stat update transactions
331 * between unlocked_inode_to_wb_begin/end() are guaranteed to be
332 * synchronizing against mapping->tree_lock.
334 * Grabbing old_wb->list_lock, inode->i_lock and mapping->tree_lock
335 * gives us exclusion against all wb related operations on @inode
336 * including IO list manipulations and stat updates.
338 if (old_wb < new_wb) {
339 spin_lock(&old_wb->list_lock);
340 spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING);
342 spin_lock(&new_wb->list_lock);
343 spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING);
345 spin_lock(&inode->i_lock);
346 spin_lock_irq(&mapping->tree_lock);
349 * Once I_FREEING is visible under i_lock, the eviction path owns
350 * the inode and we shouldn't modify ->i_io_list.
352 if (unlikely(inode->i_state & I_FREEING))
356 * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points
357 * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to
358 * pages actually under underwriteback.
360 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0,
361 PAGECACHE_TAG_DIRTY) {
362 struct page *page = radix_tree_deref_slot_protected(slot,
363 &mapping->tree_lock);
364 if (likely(page) && PageDirty(page)) {
365 __dec_wb_stat(old_wb, WB_RECLAIMABLE);
366 __inc_wb_stat(new_wb, WB_RECLAIMABLE);
370 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0,
371 PAGECACHE_TAG_WRITEBACK) {
372 struct page *page = radix_tree_deref_slot_protected(slot,
373 &mapping->tree_lock);
375 WARN_ON_ONCE(!PageWriteback(page));
376 __dec_wb_stat(old_wb, WB_WRITEBACK);
377 __inc_wb_stat(new_wb, WB_WRITEBACK);
384 * Transfer to @new_wb's IO list if necessary. The specific list
385 * @inode was on is ignored and the inode is put on ->b_dirty which
386 * is always correct including from ->b_dirty_time. The transfer
387 * preserves @inode->dirtied_when ordering.
389 if (!list_empty(&inode->i_io_list)) {
392 inode_io_list_del_locked(inode, old_wb);
393 inode->i_wb = new_wb;
394 list_for_each_entry(pos, &new_wb->b_dirty, i_io_list)
395 if (time_after_eq(inode->dirtied_when,
398 inode_io_list_move_locked(inode, new_wb, pos->i_io_list.prev);
400 inode->i_wb = new_wb;
403 /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */
404 inode->i_wb_frn_winner = 0;
405 inode->i_wb_frn_avg_time = 0;
406 inode->i_wb_frn_history = 0;
410 * Paired with load_acquire in unlocked_inode_to_wb_begin() and
411 * ensures that the new wb is visible if they see !I_WB_SWITCH.
413 smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH);
415 spin_unlock_irq(&mapping->tree_lock);
416 spin_unlock(&inode->i_lock);
417 spin_unlock(&new_wb->list_lock);
418 spin_unlock(&old_wb->list_lock);
427 deactivate_super(sb);
431 static void inode_switch_wbs_rcu_fn(struct rcu_head *rcu_head)
433 struct inode_switch_wbs_context *isw = container_of(rcu_head,
434 struct inode_switch_wbs_context, rcu_head);
436 /* needs to grab bh-unsafe locks, bounce to work item */
437 INIT_WORK(&isw->work, inode_switch_wbs_work_fn);
438 schedule_work(&isw->work);
442 * inode_switch_wbs - change the wb association of an inode
443 * @inode: target inode
444 * @new_wb_id: ID of the new wb
446 * Switch @inode's wb association to the wb identified by @new_wb_id. The
447 * switching is performed asynchronously and may fail silently.
449 static void inode_switch_wbs(struct inode *inode, int new_wb_id)
451 struct backing_dev_info *bdi = inode_to_bdi(inode);
452 struct cgroup_subsys_state *memcg_css;
453 struct inode_switch_wbs_context *isw;
455 /* noop if seems to be already in progress */
456 if (inode->i_state & I_WB_SWITCH)
459 isw = kzalloc(sizeof(*isw), GFP_ATOMIC);
463 /* find and pin the new wb */
465 memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys);
467 isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
472 /* while holding I_WB_SWITCH, no one else can update the association */
473 spin_lock(&inode->i_lock);
475 if (inode->i_state & (I_WB_SWITCH | I_FREEING) ||
476 inode_to_wb(inode) == isw->new_wb)
479 if (!atomic_inc_not_zero(&inode->i_sb->s_active))
482 inode->i_state |= I_WB_SWITCH;
483 spin_unlock(&inode->i_lock);
489 * In addition to synchronizing among switchers, I_WB_SWITCH tells
490 * the RCU protected stat update paths to grab the mapping's
491 * tree_lock so that stat transfer can synchronize against them.
492 * Let's continue after I_WB_SWITCH is guaranteed to be visible.
494 call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn);
498 spin_unlock(&inode->i_lock);
506 * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it
507 * @wbc: writeback_control of interest
508 * @inode: target inode
510 * @inode is locked and about to be written back under the control of @wbc.
511 * Record @inode's writeback context into @wbc and unlock the i_lock. On
512 * writeback completion, wbc_detach_inode() should be called. This is used
513 * to track the cgroup writeback context.
515 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
518 if (!inode_cgwb_enabled(inode)) {
519 spin_unlock(&inode->i_lock);
523 wbc->wb = inode_to_wb(inode);
526 wbc->wb_id = wbc->wb->memcg_css->id;
527 wbc->wb_lcand_id = inode->i_wb_frn_winner;
528 wbc->wb_tcand_id = 0;
530 wbc->wb_lcand_bytes = 0;
531 wbc->wb_tcand_bytes = 0;
534 spin_unlock(&inode->i_lock);
537 * A dying wb indicates that the memcg-blkcg mapping has changed
538 * and a new wb is already serving the memcg. Switch immediately.
540 if (unlikely(wb_dying(wbc->wb)))
541 inode_switch_wbs(inode, wbc->wb_id);
545 * wbc_detach_inode - disassociate wbc from inode and perform foreign detection
546 * @wbc: writeback_control of the just finished writeback
548 * To be called after a writeback attempt of an inode finishes and undoes
549 * wbc_attach_and_unlock_inode(). Can be called under any context.
551 * As concurrent write sharing of an inode is expected to be very rare and
552 * memcg only tracks page ownership on first-use basis severely confining
553 * the usefulness of such sharing, cgroup writeback tracks ownership
554 * per-inode. While the support for concurrent write sharing of an inode
555 * is deemed unnecessary, an inode being written to by different cgroups at
556 * different points in time is a lot more common, and, more importantly,
557 * charging only by first-use can too readily lead to grossly incorrect
558 * behaviors (single foreign page can lead to gigabytes of writeback to be
559 * incorrectly attributed).
561 * To resolve this issue, cgroup writeback detects the majority dirtier of
562 * an inode and transfers the ownership to it. To avoid unnnecessary
563 * oscillation, the detection mechanism keeps track of history and gives
564 * out the switch verdict only if the foreign usage pattern is stable over
565 * a certain amount of time and/or writeback attempts.
567 * On each writeback attempt, @wbc tries to detect the majority writer
568 * using Boyer-Moore majority vote algorithm. In addition to the byte
569 * count from the majority voting, it also counts the bytes written for the
570 * current wb and the last round's winner wb (max of last round's current
571 * wb, the winner from two rounds ago, and the last round's majority
572 * candidate). Keeping track of the historical winner helps the algorithm
573 * to semi-reliably detect the most active writer even when it's not the
576 * Once the winner of the round is determined, whether the winner is
577 * foreign or not and how much IO time the round consumed is recorded in
578 * inode->i_wb_frn_history. If the amount of recorded foreign IO time is
579 * over a certain threshold, the switch verdict is given.
581 void wbc_detach_inode(struct writeback_control *wbc)
583 struct bdi_writeback *wb = wbc->wb;
584 struct inode *inode = wbc->inode;
585 unsigned long avg_time, max_bytes, max_time;
592 history = inode->i_wb_frn_history;
593 avg_time = inode->i_wb_frn_avg_time;
595 /* pick the winner of this round */
596 if (wbc->wb_bytes >= wbc->wb_lcand_bytes &&
597 wbc->wb_bytes >= wbc->wb_tcand_bytes) {
599 max_bytes = wbc->wb_bytes;
600 } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) {
601 max_id = wbc->wb_lcand_id;
602 max_bytes = wbc->wb_lcand_bytes;
604 max_id = wbc->wb_tcand_id;
605 max_bytes = wbc->wb_tcand_bytes;
609 * Calculate the amount of IO time the winner consumed and fold it
610 * into the running average kept per inode. If the consumed IO
611 * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for
612 * deciding whether to switch or not. This is to prevent one-off
613 * small dirtiers from skewing the verdict.
615 max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT,
616 wb->avg_write_bandwidth);
618 avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) -
619 (avg_time >> WB_FRN_TIME_AVG_SHIFT);
621 avg_time = max_time; /* immediate catch up on first run */
623 if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) {
627 * The switch verdict is reached if foreign wb's consume
628 * more than a certain proportion of IO time in a
629 * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot
630 * history mask where each bit represents one sixteenth of
631 * the period. Determine the number of slots to shift into
632 * history from @max_time.
634 slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT),
635 (unsigned long)WB_FRN_HIST_MAX_SLOTS);
637 if (wbc->wb_id != max_id)
638 history |= (1U << slots) - 1;
641 * Switch if the current wb isn't the consistent winner.
642 * If there are multiple closely competing dirtiers, the
643 * inode may switch across them repeatedly over time, which
644 * is okay. The main goal is avoiding keeping an inode on
645 * the wrong wb for an extended period of time.
647 if (hweight32(history) > WB_FRN_HIST_THR_SLOTS)
648 inode_switch_wbs(inode, max_id);
652 * Multiple instances of this function may race to update the
653 * following fields but we don't mind occassional inaccuracies.
655 inode->i_wb_frn_winner = max_id;
656 inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX);
657 inode->i_wb_frn_history = history;
664 * wbc_account_io - account IO issued during writeback
665 * @wbc: writeback_control of the writeback in progress
666 * @page: page being written out
667 * @bytes: number of bytes being written out
669 * @bytes from @page are about to written out during the writeback
670 * controlled by @wbc. Keep the book for foreign inode detection. See
671 * wbc_detach_inode().
673 void wbc_account_io(struct writeback_control *wbc, struct page *page,
679 * pageout() path doesn't attach @wbc to the inode being written
680 * out. This is intentional as we don't want the function to block
681 * behind a slow cgroup. Ultimately, we want pageout() to kick off
682 * regular writeback instead of writing things out itself.
688 id = mem_cgroup_css_from_page(page)->id;
691 if (id == wbc->wb_id) {
692 wbc->wb_bytes += bytes;
696 if (id == wbc->wb_lcand_id)
697 wbc->wb_lcand_bytes += bytes;
699 /* Boyer-Moore majority vote algorithm */
700 if (!wbc->wb_tcand_bytes)
701 wbc->wb_tcand_id = id;
702 if (id == wbc->wb_tcand_id)
703 wbc->wb_tcand_bytes += bytes;
705 wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes);
707 EXPORT_SYMBOL_GPL(wbc_account_io);
710 * inode_congested - test whether an inode is congested
711 * @inode: inode to test for congestion (may be NULL)
712 * @cong_bits: mask of WB_[a]sync_congested bits to test
714 * Tests whether @inode is congested. @cong_bits is the mask of congestion
715 * bits to test and the return value is the mask of set bits.
717 * If cgroup writeback is enabled for @inode, the congestion state is
718 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
719 * associated with @inode is congested; otherwise, the root wb's congestion
722 * @inode is allowed to be NULL as this function is often called on
723 * mapping->host which is NULL for the swapper space.
725 int inode_congested(struct inode *inode, int cong_bits)
728 * Once set, ->i_wb never becomes NULL while the inode is alive.
729 * Start transaction iff ->i_wb is visible.
731 if (inode && inode_to_wb_is_valid(inode)) {
732 struct bdi_writeback *wb;
733 bool locked, congested;
735 wb = unlocked_inode_to_wb_begin(inode, &locked);
736 congested = wb_congested(wb, cong_bits);
737 unlocked_inode_to_wb_end(inode, locked);
741 return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
743 EXPORT_SYMBOL_GPL(inode_congested);
746 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
747 * @wb: target bdi_writeback to split @nr_pages to
748 * @nr_pages: number of pages to write for the whole bdi
750 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
751 * relation to the total write bandwidth of all wb's w/ dirty inodes on
754 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
756 unsigned long this_bw = wb->avg_write_bandwidth;
757 unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth);
759 if (nr_pages == LONG_MAX)
763 * This may be called on clean wb's and proportional distribution
764 * may not make sense, just use the original @nr_pages in those
765 * cases. In general, we wanna err on the side of writing more.
767 if (!tot_bw || this_bw >= tot_bw)
770 return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw);
774 * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
775 * @bdi: target backing_dev_info
776 * @base_work: wb_writeback_work to issue
777 * @skip_if_busy: skip wb's which already have writeback in progress
779 * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
780 * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's
781 * distributed to the busy wbs according to each wb's proportion in the
782 * total active write bandwidth of @bdi.
784 static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
785 struct wb_writeback_work *base_work,
788 struct bdi_writeback *last_wb = NULL;
789 struct bdi_writeback *wb = list_entry(&bdi->wb_list,
790 struct bdi_writeback, bdi_node);
795 list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) {
796 DEFINE_WB_COMPLETION_ONSTACK(fallback_work_done);
797 struct wb_writeback_work fallback_work;
798 struct wb_writeback_work *work;
806 /* SYNC_ALL writes out I_DIRTY_TIME too */
807 if (!wb_has_dirty_io(wb) &&
808 (base_work->sync_mode == WB_SYNC_NONE ||
809 list_empty(&wb->b_dirty_time)))
811 if (skip_if_busy && writeback_in_progress(wb))
814 nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages);
816 work = kmalloc(sizeof(*work), GFP_ATOMIC);
819 work->nr_pages = nr_pages;
821 wb_queue_work(wb, work);
825 /* alloc failed, execute synchronously using on-stack fallback */
826 work = &fallback_work;
828 work->nr_pages = nr_pages;
830 work->done = &fallback_work_done;
832 wb_queue_work(wb, work);
835 * Pin @wb so that it stays on @bdi->wb_list. This allows
836 * continuing iteration from @wb after dropping and
837 * regrabbing rcu read lock.
843 wb_wait_for_completion(bdi, &fallback_work_done);
852 #else /* CONFIG_CGROUP_WRITEBACK */
854 static struct bdi_writeback *
855 locked_inode_to_wb_and_lock_list(struct inode *inode)
856 __releases(&inode->i_lock)
857 __acquires(&wb->list_lock)
859 struct bdi_writeback *wb = inode_to_wb(inode);
861 spin_unlock(&inode->i_lock);
862 spin_lock(&wb->list_lock);
866 static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode)
867 __acquires(&wb->list_lock)
869 struct bdi_writeback *wb = inode_to_wb(inode);
871 spin_lock(&wb->list_lock);
875 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
880 static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
881 struct wb_writeback_work *base_work,
886 if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) {
887 base_work->auto_free = 0;
888 wb_queue_work(&bdi->wb, base_work);
892 #endif /* CONFIG_CGROUP_WRITEBACK */
894 void wb_start_writeback(struct bdi_writeback *wb, long nr_pages,
895 bool range_cyclic, enum wb_reason reason)
897 struct wb_writeback_work *work;
899 if (!wb_has_dirty_io(wb))
903 * This is WB_SYNC_NONE writeback, so if allocation fails just
904 * wakeup the thread for old dirty data writeback
906 work = kzalloc(sizeof(*work), GFP_ATOMIC);
908 trace_writeback_nowork(wb);
913 work->sync_mode = WB_SYNC_NONE;
914 work->nr_pages = nr_pages;
915 work->range_cyclic = range_cyclic;
916 work->reason = reason;
919 wb_queue_work(wb, work);
923 * wb_start_background_writeback - start background writeback
924 * @wb: bdi_writback to write from
927 * This makes sure WB_SYNC_NONE background writeback happens. When
928 * this function returns, it is only guaranteed that for given wb
929 * some IO is happening if we are over background dirty threshold.
930 * Caller need not hold sb s_umount semaphore.
932 void wb_start_background_writeback(struct bdi_writeback *wb)
935 * We just wake up the flusher thread. It will perform background
936 * writeback as soon as there is no other work to do.
938 trace_writeback_wake_background(wb);
943 * Remove the inode from the writeback list it is on.
945 void inode_io_list_del(struct inode *inode)
947 struct bdi_writeback *wb;
949 wb = inode_to_wb_and_lock_list(inode);
950 inode_io_list_del_locked(inode, wb);
951 spin_unlock(&wb->list_lock);
955 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
956 * furthest end of its superblock's dirty-inode list.
958 * Before stamping the inode's ->dirtied_when, we check to see whether it is
959 * already the most-recently-dirtied inode on the b_dirty list. If that is
960 * the case then the inode must have been redirtied while it was being written
961 * out and we don't reset its dirtied_when.
963 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
965 if (!list_empty(&wb->b_dirty)) {
968 tail = wb_inode(wb->b_dirty.next);
969 if (time_before(inode->dirtied_when, tail->dirtied_when))
970 inode->dirtied_when = jiffies;
972 inode_io_list_move_locked(inode, wb, &wb->b_dirty);
976 * requeue inode for re-scanning after bdi->b_io list is exhausted.
978 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
980 inode_io_list_move_locked(inode, wb, &wb->b_more_io);
983 static void inode_sync_complete(struct inode *inode)
985 inode->i_state &= ~I_SYNC;
986 /* If inode is clean an unused, put it into LRU now... */
987 inode_add_lru(inode);
988 /* Waiters must see I_SYNC cleared before being woken up */
990 wake_up_bit(&inode->i_state, __I_SYNC);
993 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
995 bool ret = time_after(inode->dirtied_when, t);
998 * For inodes being constantly redirtied, dirtied_when can get stuck.
999 * It _appears_ to be in the future, but is actually in distant past.
1000 * This test is necessary to prevent such wrapped-around relative times
1001 * from permanently stopping the whole bdi writeback.
1003 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
1008 #define EXPIRE_DIRTY_ATIME 0x0001
1011 * Move expired (dirtied before work->older_than_this) dirty inodes from
1012 * @delaying_queue to @dispatch_queue.
1014 static int move_expired_inodes(struct list_head *delaying_queue,
1015 struct list_head *dispatch_queue,
1017 struct wb_writeback_work *work)
1019 unsigned long *older_than_this = NULL;
1020 unsigned long expire_time;
1022 struct list_head *pos, *node;
1023 struct super_block *sb = NULL;
1024 struct inode *inode;
1028 if ((flags & EXPIRE_DIRTY_ATIME) == 0)
1029 older_than_this = work->older_than_this;
1030 else if (!work->for_sync) {
1031 expire_time = jiffies - (dirtytime_expire_interval * HZ);
1032 older_than_this = &expire_time;
1034 while (!list_empty(delaying_queue)) {
1035 inode = wb_inode(delaying_queue->prev);
1036 if (older_than_this &&
1037 inode_dirtied_after(inode, *older_than_this))
1039 list_move(&inode->i_io_list, &tmp);
1041 if (flags & EXPIRE_DIRTY_ATIME)
1042 set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
1043 if (sb_is_blkdev_sb(inode->i_sb))
1045 if (sb && sb != inode->i_sb)
1050 /* just one sb in list, splice to dispatch_queue and we're done */
1052 list_splice(&tmp, dispatch_queue);
1056 /* Move inodes from one superblock together */
1057 while (!list_empty(&tmp)) {
1058 sb = wb_inode(tmp.prev)->i_sb;
1059 list_for_each_prev_safe(pos, node, &tmp) {
1060 inode = wb_inode(pos);
1061 if (inode->i_sb == sb)
1062 list_move(&inode->i_io_list, dispatch_queue);
1070 * Queue all expired dirty inodes for io, eldest first.
1072 * newly dirtied b_dirty b_io b_more_io
1073 * =============> gf edc BA
1075 * newly dirtied b_dirty b_io b_more_io
1076 * =============> g fBAedc
1078 * +--> dequeue for IO
1080 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
1084 assert_spin_locked(&wb->list_lock);
1085 list_splice_init(&wb->b_more_io, &wb->b_io);
1086 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work);
1087 moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io,
1088 EXPIRE_DIRTY_ATIME, work);
1090 wb_io_lists_populated(wb);
1091 trace_writeback_queue_io(wb, work, moved);
1094 static int write_inode(struct inode *inode, struct writeback_control *wbc)
1098 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
1099 trace_writeback_write_inode_start(inode, wbc);
1100 ret = inode->i_sb->s_op->write_inode(inode, wbc);
1101 trace_writeback_write_inode(inode, wbc);
1108 * Wait for writeback on an inode to complete. Called with i_lock held.
1109 * Caller must make sure inode cannot go away when we drop i_lock.
1111 static void __inode_wait_for_writeback(struct inode *inode)
1112 __releases(inode->i_lock)
1113 __acquires(inode->i_lock)
1115 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
1116 wait_queue_head_t *wqh;
1118 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
1119 while (inode->i_state & I_SYNC) {
1120 spin_unlock(&inode->i_lock);
1121 __wait_on_bit(wqh, &wq, bit_wait,
1122 TASK_UNINTERRUPTIBLE);
1123 spin_lock(&inode->i_lock);
1128 * Wait for writeback on an inode to complete. Caller must have inode pinned.
1130 void inode_wait_for_writeback(struct inode *inode)
1132 spin_lock(&inode->i_lock);
1133 __inode_wait_for_writeback(inode);
1134 spin_unlock(&inode->i_lock);
1138 * Sleep until I_SYNC is cleared. This function must be called with i_lock
1139 * held and drops it. It is aimed for callers not holding any inode reference
1140 * so once i_lock is dropped, inode can go away.
1142 static void inode_sleep_on_writeback(struct inode *inode)
1143 __releases(inode->i_lock)
1146 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
1149 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1150 sleep = inode->i_state & I_SYNC;
1151 spin_unlock(&inode->i_lock);
1154 finish_wait(wqh, &wait);
1158 * Find proper writeback list for the inode depending on its current state and
1159 * possibly also change of its state while we were doing writeback. Here we
1160 * handle things such as livelock prevention or fairness of writeback among
1161 * inodes. This function can be called only by flusher thread - noone else
1162 * processes all inodes in writeback lists and requeueing inodes behind flusher
1163 * thread's back can have unexpected consequences.
1165 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
1166 struct writeback_control *wbc)
1168 if (inode->i_state & I_FREEING)
1172 * Sync livelock prevention. Each inode is tagged and synced in one
1173 * shot. If still dirty, it will be redirty_tail()'ed below. Update
1174 * the dirty time to prevent enqueue and sync it again.
1176 if ((inode->i_state & I_DIRTY) &&
1177 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
1178 inode->dirtied_when = jiffies;
1180 if (wbc->pages_skipped) {
1182 * writeback is not making progress due to locked
1183 * buffers. Skip this inode for now.
1185 redirty_tail(inode, wb);
1189 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
1191 * We didn't write back all the pages. nfs_writepages()
1192 * sometimes bales out without doing anything.
1194 if (wbc->nr_to_write <= 0) {
1195 /* Slice used up. Queue for next turn. */
1196 requeue_io(inode, wb);
1199 * Writeback blocked by something other than
1200 * congestion. Delay the inode for some time to
1201 * avoid spinning on the CPU (100% iowait)
1202 * retrying writeback of the dirty page/inode
1203 * that cannot be performed immediately.
1205 redirty_tail(inode, wb);
1207 } else if (inode->i_state & I_DIRTY) {
1209 * Filesystems can dirty the inode during writeback operations,
1210 * such as delayed allocation during submission or metadata
1211 * updates after data IO completion.
1213 redirty_tail(inode, wb);
1214 } else if (inode->i_state & I_DIRTY_TIME) {
1215 inode->dirtied_when = jiffies;
1216 inode_io_list_move_locked(inode, wb, &wb->b_dirty_time);
1218 /* The inode is clean. Remove from writeback lists. */
1219 inode_io_list_del_locked(inode, wb);
1224 * Write out an inode and its dirty pages. Do not update the writeback list
1225 * linkage. That is left to the caller. The caller is also responsible for
1226 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
1229 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
1231 struct address_space *mapping = inode->i_mapping;
1232 long nr_to_write = wbc->nr_to_write;
1236 WARN_ON(!(inode->i_state & I_SYNC));
1238 trace_writeback_single_inode_start(inode, wbc, nr_to_write);
1240 ret = do_writepages(mapping, wbc);
1243 * Make sure to wait on the data before writing out the metadata.
1244 * This is important for filesystems that modify metadata on data
1245 * I/O completion. We don't do it for sync(2) writeback because it has a
1246 * separate, external IO completion path and ->sync_fs for guaranteeing
1247 * inode metadata is written back correctly.
1249 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
1250 int err = filemap_fdatawait(mapping);
1256 * Some filesystems may redirty the inode during the writeback
1257 * due to delalloc, clear dirty metadata flags right before
1260 spin_lock(&inode->i_lock);
1262 dirty = inode->i_state & I_DIRTY;
1263 if (inode->i_state & I_DIRTY_TIME) {
1264 if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
1265 unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
1266 unlikely(time_after(jiffies,
1267 (inode->dirtied_time_when +
1268 dirtytime_expire_interval * HZ)))) {
1269 dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
1270 trace_writeback_lazytime(inode);
1273 inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
1274 inode->i_state &= ~dirty;
1277 * Paired with smp_mb() in __mark_inode_dirty(). This allows
1278 * __mark_inode_dirty() to test i_state without grabbing i_lock -
1279 * either they see the I_DIRTY bits cleared or we see the dirtied
1282 * I_DIRTY_PAGES is always cleared together above even if @mapping
1283 * still has dirty pages. The flag is reinstated after smp_mb() if
1284 * necessary. This guarantees that either __mark_inode_dirty()
1285 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
1289 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1290 inode->i_state |= I_DIRTY_PAGES;
1292 spin_unlock(&inode->i_lock);
1294 if (dirty & I_DIRTY_TIME)
1295 mark_inode_dirty_sync(inode);
1296 /* Don't write the inode if only I_DIRTY_PAGES was set */
1297 if (dirty & ~I_DIRTY_PAGES) {
1298 int err = write_inode(inode, wbc);
1302 trace_writeback_single_inode(inode, wbc, nr_to_write);
1307 * Write out an inode's dirty pages. Either the caller has an active reference
1308 * on the inode or the inode has I_WILL_FREE set.
1310 * This function is designed to be called for writing back one inode which
1311 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
1312 * and does more profound writeback list handling in writeback_sb_inodes().
1315 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
1316 struct writeback_control *wbc)
1320 spin_lock(&inode->i_lock);
1321 if (!atomic_read(&inode->i_count))
1322 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
1324 WARN_ON(inode->i_state & I_WILL_FREE);
1326 if (inode->i_state & I_SYNC) {
1327 if (wbc->sync_mode != WB_SYNC_ALL)
1330 * It's a data-integrity sync. We must wait. Since callers hold
1331 * inode reference or inode has I_WILL_FREE set, it cannot go
1334 __inode_wait_for_writeback(inode);
1336 WARN_ON(inode->i_state & I_SYNC);
1338 * Skip inode if it is clean and we have no outstanding writeback in
1339 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
1340 * function since flusher thread may be doing for example sync in
1341 * parallel and if we move the inode, it could get skipped. So here we
1342 * make sure inode is on some writeback list and leave it there unless
1343 * we have completely cleaned the inode.
1345 if (!(inode->i_state & I_DIRTY_ALL) &&
1346 (wbc->sync_mode != WB_SYNC_ALL ||
1347 !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
1349 inode->i_state |= I_SYNC;
1350 wbc_attach_and_unlock_inode(wbc, inode);
1352 ret = __writeback_single_inode(inode, wbc);
1354 wbc_detach_inode(wbc);
1355 spin_lock(&wb->list_lock);
1356 spin_lock(&inode->i_lock);
1358 * If inode is clean, remove it from writeback lists. Otherwise don't
1359 * touch it. See comment above for explanation.
1361 if (!(inode->i_state & I_DIRTY_ALL))
1362 inode_io_list_del_locked(inode, wb);
1363 spin_unlock(&wb->list_lock);
1364 inode_sync_complete(inode);
1366 spin_unlock(&inode->i_lock);
1370 static long writeback_chunk_size(struct bdi_writeback *wb,
1371 struct wb_writeback_work *work)
1376 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
1377 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
1378 * here avoids calling into writeback_inodes_wb() more than once.
1380 * The intended call sequence for WB_SYNC_ALL writeback is:
1383 * writeback_sb_inodes() <== called only once
1384 * write_cache_pages() <== called once for each inode
1385 * (quickly) tag currently dirty pages
1386 * (maybe slowly) sync all tagged pages
1388 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
1391 pages = min(wb->avg_write_bandwidth / 2,
1392 global_wb_domain.dirty_limit / DIRTY_SCOPE);
1393 pages = min(pages, work->nr_pages);
1394 pages = round_down(pages + MIN_WRITEBACK_PAGES,
1395 MIN_WRITEBACK_PAGES);
1402 * Write a portion of b_io inodes which belong to @sb.
1404 * Return the number of pages and/or inodes written.
1406 * NOTE! This is called with wb->list_lock held, and will
1407 * unlock and relock that for each inode it ends up doing
1410 static long writeback_sb_inodes(struct super_block *sb,
1411 struct bdi_writeback *wb,
1412 struct wb_writeback_work *work)
1414 struct writeback_control wbc = {
1415 .sync_mode = work->sync_mode,
1416 .tagged_writepages = work->tagged_writepages,
1417 .for_kupdate = work->for_kupdate,
1418 .for_background = work->for_background,
1419 .for_sync = work->for_sync,
1420 .range_cyclic = work->range_cyclic,
1422 .range_end = LLONG_MAX,
1424 unsigned long start_time = jiffies;
1426 long wrote = 0; /* count both pages and inodes */
1428 while (!list_empty(&wb->b_io)) {
1429 struct inode *inode = wb_inode(wb->b_io.prev);
1431 if (inode->i_sb != sb) {
1434 * We only want to write back data for this
1435 * superblock, move all inodes not belonging
1436 * to it back onto the dirty list.
1438 redirty_tail(inode, wb);
1443 * The inode belongs to a different superblock.
1444 * Bounce back to the caller to unpin this and
1445 * pin the next superblock.
1451 * Don't bother with new inodes or inodes being freed, first
1452 * kind does not need periodic writeout yet, and for the latter
1453 * kind writeout is handled by the freer.
1455 spin_lock(&inode->i_lock);
1456 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
1457 spin_unlock(&inode->i_lock);
1458 redirty_tail(inode, wb);
1461 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
1463 * If this inode is locked for writeback and we are not
1464 * doing writeback-for-data-integrity, move it to
1465 * b_more_io so that writeback can proceed with the
1466 * other inodes on s_io.
1468 * We'll have another go at writing back this inode
1469 * when we completed a full scan of b_io.
1471 spin_unlock(&inode->i_lock);
1472 requeue_io(inode, wb);
1473 trace_writeback_sb_inodes_requeue(inode);
1476 spin_unlock(&wb->list_lock);
1479 * We already requeued the inode if it had I_SYNC set and we
1480 * are doing WB_SYNC_NONE writeback. So this catches only the
1483 if (inode->i_state & I_SYNC) {
1484 /* Wait for I_SYNC. This function drops i_lock... */
1485 inode_sleep_on_writeback(inode);
1486 /* Inode may be gone, start again */
1487 spin_lock(&wb->list_lock);
1490 inode->i_state |= I_SYNC;
1491 wbc_attach_and_unlock_inode(&wbc, inode);
1493 write_chunk = writeback_chunk_size(wb, work);
1494 wbc.nr_to_write = write_chunk;
1495 wbc.pages_skipped = 0;
1498 * We use I_SYNC to pin the inode in memory. While it is set
1499 * evict_inode() will wait so the inode cannot be freed.
1501 __writeback_single_inode(inode, &wbc);
1503 wbc_detach_inode(&wbc);
1504 work->nr_pages -= write_chunk - wbc.nr_to_write;
1505 wrote += write_chunk - wbc.nr_to_write;
1507 if (need_resched()) {
1509 * We're trying to balance between building up a nice
1510 * long list of IOs to improve our merge rate, and
1511 * getting those IOs out quickly for anyone throttling
1512 * in balance_dirty_pages(). cond_resched() doesn't
1513 * unplug, so get our IOs out the door before we
1516 blk_flush_plug(current);
1521 spin_lock(&wb->list_lock);
1522 spin_lock(&inode->i_lock);
1523 if (!(inode->i_state & I_DIRTY_ALL))
1525 requeue_inode(inode, wb, &wbc);
1526 inode_sync_complete(inode);
1527 spin_unlock(&inode->i_lock);
1530 * bail out to wb_writeback() often enough to check
1531 * background threshold and other termination conditions.
1534 if (time_is_before_jiffies(start_time + HZ / 10UL))
1536 if (work->nr_pages <= 0)
1543 static long __writeback_inodes_wb(struct bdi_writeback *wb,
1544 struct wb_writeback_work *work)
1546 unsigned long start_time = jiffies;
1549 while (!list_empty(&wb->b_io)) {
1550 struct inode *inode = wb_inode(wb->b_io.prev);
1551 struct super_block *sb = inode->i_sb;
1553 if (!trylock_super(sb)) {
1555 * trylock_super() may fail consistently due to
1556 * s_umount being grabbed by someone else. Don't use
1557 * requeue_io() to avoid busy retrying the inode/sb.
1559 redirty_tail(inode, wb);
1562 wrote += writeback_sb_inodes(sb, wb, work);
1563 up_read(&sb->s_umount);
1565 /* refer to the same tests at the end of writeback_sb_inodes */
1567 if (time_is_before_jiffies(start_time + HZ / 10UL))
1569 if (work->nr_pages <= 0)
1573 /* Leave any unwritten inodes on b_io */
1577 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
1578 enum wb_reason reason)
1580 struct wb_writeback_work work = {
1581 .nr_pages = nr_pages,
1582 .sync_mode = WB_SYNC_NONE,
1586 struct blk_plug plug;
1588 blk_start_plug(&plug);
1589 spin_lock(&wb->list_lock);
1590 if (list_empty(&wb->b_io))
1591 queue_io(wb, &work);
1592 __writeback_inodes_wb(wb, &work);
1593 spin_unlock(&wb->list_lock);
1594 blk_finish_plug(&plug);
1596 return nr_pages - work.nr_pages;
1600 * Explicit flushing or periodic writeback of "old" data.
1602 * Define "old": the first time one of an inode's pages is dirtied, we mark the
1603 * dirtying-time in the inode's address_space. So this periodic writeback code
1604 * just walks the superblock inode list, writing back any inodes which are
1605 * older than a specific point in time.
1607 * Try to run once per dirty_writeback_interval. But if a writeback event
1608 * takes longer than a dirty_writeback_interval interval, then leave a
1611 * older_than_this takes precedence over nr_to_write. So we'll only write back
1612 * all dirty pages if they are all attached to "old" mappings.
1614 static long wb_writeback(struct bdi_writeback *wb,
1615 struct wb_writeback_work *work)
1617 unsigned long wb_start = jiffies;
1618 long nr_pages = work->nr_pages;
1619 unsigned long oldest_jif;
1620 struct inode *inode;
1622 struct blk_plug plug;
1624 oldest_jif = jiffies;
1625 work->older_than_this = &oldest_jif;
1627 blk_start_plug(&plug);
1628 spin_lock(&wb->list_lock);
1631 * Stop writeback when nr_pages has been consumed
1633 if (work->nr_pages <= 0)
1637 * Background writeout and kupdate-style writeback may
1638 * run forever. Stop them if there is other work to do
1639 * so that e.g. sync can proceed. They'll be restarted
1640 * after the other works are all done.
1642 if ((work->for_background || work->for_kupdate) &&
1643 !list_empty(&wb->work_list))
1647 * For background writeout, stop when we are below the
1648 * background dirty threshold
1650 if (work->for_background && !wb_over_bg_thresh(wb))
1654 * Kupdate and background works are special and we want to
1655 * include all inodes that need writing. Livelock avoidance is
1656 * handled by these works yielding to any other work so we are
1659 if (work->for_kupdate) {
1660 oldest_jif = jiffies -
1661 msecs_to_jiffies(dirty_expire_interval * 10);
1662 } else if (work->for_background)
1663 oldest_jif = jiffies;
1665 trace_writeback_start(wb, work);
1666 if (list_empty(&wb->b_io))
1669 progress = writeback_sb_inodes(work->sb, wb, work);
1671 progress = __writeback_inodes_wb(wb, work);
1672 trace_writeback_written(wb, work);
1674 wb_update_bandwidth(wb, wb_start);
1677 * Did we write something? Try for more
1679 * Dirty inodes are moved to b_io for writeback in batches.
1680 * The completion of the current batch does not necessarily
1681 * mean the overall work is done. So we keep looping as long
1682 * as made some progress on cleaning pages or inodes.
1687 * No more inodes for IO, bail
1689 if (list_empty(&wb->b_more_io))
1692 * Nothing written. Wait for some inode to
1693 * become available for writeback. Otherwise
1694 * we'll just busyloop.
1696 if (!list_empty(&wb->b_more_io)) {
1697 trace_writeback_wait(wb, work);
1698 inode = wb_inode(wb->b_more_io.prev);
1699 spin_lock(&inode->i_lock);
1700 spin_unlock(&wb->list_lock);
1701 /* This function drops i_lock... */
1702 inode_sleep_on_writeback(inode);
1703 spin_lock(&wb->list_lock);
1706 spin_unlock(&wb->list_lock);
1707 blk_finish_plug(&plug);
1709 return nr_pages - work->nr_pages;
1713 * Return the next wb_writeback_work struct that hasn't been processed yet.
1715 static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb)
1717 struct wb_writeback_work *work = NULL;
1719 spin_lock_bh(&wb->work_lock);
1720 if (!list_empty(&wb->work_list)) {
1721 work = list_entry(wb->work_list.next,
1722 struct wb_writeback_work, list);
1723 list_del_init(&work->list);
1725 spin_unlock_bh(&wb->work_lock);
1730 * Add in the number of potentially dirty inodes, because each inode
1731 * write can dirty pagecache in the underlying blockdev.
1733 static unsigned long get_nr_dirty_pages(void)
1735 return global_page_state(NR_FILE_DIRTY) +
1736 global_page_state(NR_UNSTABLE_NFS) +
1737 get_nr_dirty_inodes();
1740 static long wb_check_background_flush(struct bdi_writeback *wb)
1742 if (wb_over_bg_thresh(wb)) {
1744 struct wb_writeback_work work = {
1745 .nr_pages = LONG_MAX,
1746 .sync_mode = WB_SYNC_NONE,
1747 .for_background = 1,
1749 .reason = WB_REASON_BACKGROUND,
1752 return wb_writeback(wb, &work);
1758 static long wb_check_old_data_flush(struct bdi_writeback *wb)
1760 unsigned long expired;
1764 * When set to zero, disable periodic writeback
1766 if (!dirty_writeback_interval)
1769 expired = wb->last_old_flush +
1770 msecs_to_jiffies(dirty_writeback_interval * 10);
1771 if (time_before(jiffies, expired))
1774 wb->last_old_flush = jiffies;
1775 nr_pages = get_nr_dirty_pages();
1778 struct wb_writeback_work work = {
1779 .nr_pages = nr_pages,
1780 .sync_mode = WB_SYNC_NONE,
1783 .reason = WB_REASON_PERIODIC,
1786 return wb_writeback(wb, &work);
1793 * Retrieve work items and do the writeback they describe
1795 static long wb_do_writeback(struct bdi_writeback *wb)
1797 struct wb_writeback_work *work;
1800 set_bit(WB_writeback_running, &wb->state);
1801 while ((work = get_next_work_item(wb)) != NULL) {
1802 struct wb_completion *done = work->done;
1804 trace_writeback_exec(wb, work);
1806 wrote += wb_writeback(wb, work);
1808 if (work->auto_free)
1810 if (done && atomic_dec_and_test(&done->cnt))
1811 wake_up_all(&wb->bdi->wb_waitq);
1815 * Check for periodic writeback, kupdated() style
1817 wrote += wb_check_old_data_flush(wb);
1818 wrote += wb_check_background_flush(wb);
1819 clear_bit(WB_writeback_running, &wb->state);
1825 * Handle writeback of dirty data for the device backed by this bdi. Also
1826 * reschedules periodically and does kupdated style flushing.
1828 void wb_workfn(struct work_struct *work)
1830 struct bdi_writeback *wb = container_of(to_delayed_work(work),
1831 struct bdi_writeback, dwork);
1834 set_worker_desc("flush-%s", dev_name(wb->bdi->dev));
1835 current->flags |= PF_SWAPWRITE;
1837 if (likely(!current_is_workqueue_rescuer() ||
1838 !test_bit(WB_registered, &wb->state))) {
1840 * The normal path. Keep writing back @wb until its
1841 * work_list is empty. Note that this path is also taken
1842 * if @wb is shutting down even when we're running off the
1843 * rescuer as work_list needs to be drained.
1846 pages_written = wb_do_writeback(wb);
1847 trace_writeback_pages_written(pages_written);
1848 } while (!list_empty(&wb->work_list));
1851 * bdi_wq can't get enough workers and we're running off
1852 * the emergency worker. Don't hog it. Hopefully, 1024 is
1853 * enough for efficient IO.
1855 pages_written = writeback_inodes_wb(wb, 1024,
1856 WB_REASON_FORKER_THREAD);
1857 trace_writeback_pages_written(pages_written);
1860 if (!list_empty(&wb->work_list))
1861 mod_delayed_work(bdi_wq, &wb->dwork, 0);
1862 else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1863 wb_wakeup_delayed(wb);
1865 current->flags &= ~PF_SWAPWRITE;
1869 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1872 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1874 struct backing_dev_info *bdi;
1877 nr_pages = get_nr_dirty_pages();
1880 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1881 struct bdi_writeback *wb;
1883 if (!bdi_has_dirty_io(bdi))
1886 list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node)
1887 wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages),
1894 * Wake up bdi's periodically to make sure dirtytime inodes gets
1895 * written back periodically. We deliberately do *not* check the
1896 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1897 * kernel to be constantly waking up once there are any dirtytime
1898 * inodes on the system. So instead we define a separate delayed work
1899 * function which gets called much more rarely. (By default, only
1900 * once every 12 hours.)
1902 * If there is any other write activity going on in the file system,
1903 * this function won't be necessary. But if the only thing that has
1904 * happened on the file system is a dirtytime inode caused by an atime
1905 * update, we need this infrastructure below to make sure that inode
1906 * eventually gets pushed out to disk.
1908 static void wakeup_dirtytime_writeback(struct work_struct *w);
1909 static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback);
1911 static void wakeup_dirtytime_writeback(struct work_struct *w)
1913 struct backing_dev_info *bdi;
1916 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1917 struct bdi_writeback *wb;
1919 list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node)
1920 if (!list_empty(&wb->b_dirty_time))
1924 schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
1927 static int __init start_dirtytime_writeback(void)
1929 schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
1932 __initcall(start_dirtytime_writeback);
1934 int dirtytime_interval_handler(struct ctl_table *table, int write,
1935 void __user *buffer, size_t *lenp, loff_t *ppos)
1939 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
1940 if (ret == 0 && write)
1941 mod_delayed_work(system_wq, &dirtytime_work, 0);
1945 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1947 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1948 struct dentry *dentry;
1949 const char *name = "?";
1951 dentry = d_find_alias(inode);
1953 spin_lock(&dentry->d_lock);
1954 name = (const char *) dentry->d_name.name;
1957 "%s(%d): dirtied inode %lu (%s) on %s\n",
1958 current->comm, task_pid_nr(current), inode->i_ino,
1959 name, inode->i_sb->s_id);
1961 spin_unlock(&dentry->d_lock);
1968 * __mark_inode_dirty - internal function
1969 * @inode: inode to mark
1970 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1971 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1972 * mark_inode_dirty_sync.
1974 * Put the inode on the super block's dirty list.
1976 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1977 * dirty list only if it is hashed or if it refers to a blockdev.
1978 * If it was not hashed, it will never be added to the dirty list
1979 * even if it is later hashed, as it will have been marked dirty already.
1981 * In short, make sure you hash any inodes _before_ you start marking
1984 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1985 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1986 * the kernel-internal blockdev inode represents the dirtying time of the
1987 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1988 * page->mapping->host, so the page-dirtying time is recorded in the internal
1991 void __mark_inode_dirty(struct inode *inode, int flags)
1993 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1994 struct super_block *sb = inode->i_sb;
1997 trace_writeback_mark_inode_dirty(inode, flags);
2000 * Don't do this for I_DIRTY_PAGES - that doesn't actually
2001 * dirty the inode itself
2003 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) {
2004 trace_writeback_dirty_inode_start(inode, flags);
2006 if (sb->s_op->dirty_inode)
2007 sb->s_op->dirty_inode(inode, flags);
2009 trace_writeback_dirty_inode(inode, flags);
2011 if (flags & I_DIRTY_INODE)
2012 flags &= ~I_DIRTY_TIME;
2013 dirtytime = flags & I_DIRTY_TIME;
2016 * Paired with smp_mb() in __writeback_single_inode() for the
2017 * following lockless i_state test. See there for details.
2021 if (((inode->i_state & flags) == flags) ||
2022 (dirtytime && (inode->i_state & I_DIRTY_INODE)))
2025 if (unlikely(block_dump))
2026 block_dump___mark_inode_dirty(inode);
2028 spin_lock(&inode->i_lock);
2029 if (dirtytime && (inode->i_state & I_DIRTY_INODE))
2030 goto out_unlock_inode;
2031 if ((inode->i_state & flags) != flags) {
2032 const int was_dirty = inode->i_state & I_DIRTY;
2034 inode_attach_wb(inode, NULL);
2036 if (flags & I_DIRTY_INODE)
2037 inode->i_state &= ~I_DIRTY_TIME;
2038 inode->i_state |= flags;
2041 * If the inode is being synced, just update its dirty state.
2042 * The unlocker will place the inode on the appropriate
2043 * superblock list, based upon its state.
2045 if (inode->i_state & I_SYNC)
2046 goto out_unlock_inode;
2049 * Only add valid (hashed) inodes to the superblock's
2050 * dirty list. Add blockdev inodes as well.
2052 if (!S_ISBLK(inode->i_mode)) {
2053 if (inode_unhashed(inode))
2054 goto out_unlock_inode;
2056 if (inode->i_state & I_FREEING)
2057 goto out_unlock_inode;
2060 * If the inode was already on b_dirty/b_io/b_more_io, don't
2061 * reposition it (that would break b_dirty time-ordering).
2064 struct bdi_writeback *wb;
2065 struct list_head *dirty_list;
2066 bool wakeup_bdi = false;
2068 wb = locked_inode_to_wb_and_lock_list(inode);
2070 WARN(bdi_cap_writeback_dirty(wb->bdi) &&
2071 !test_bit(WB_registered, &wb->state),
2072 "bdi-%s not registered\n", wb->bdi->name);
2074 inode->dirtied_when = jiffies;
2076 inode->dirtied_time_when = jiffies;
2078 if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES))
2079 dirty_list = &wb->b_dirty;
2081 dirty_list = &wb->b_dirty_time;
2083 wakeup_bdi = inode_io_list_move_locked(inode, wb,
2086 spin_unlock(&wb->list_lock);
2087 trace_writeback_dirty_inode_enqueue(inode);
2090 * If this is the first dirty inode for this bdi,
2091 * we have to wake-up the corresponding bdi thread
2092 * to make sure background write-back happens
2095 if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi)
2096 wb_wakeup_delayed(wb);
2101 spin_unlock(&inode->i_lock);
2103 #undef I_DIRTY_INODE
2105 EXPORT_SYMBOL(__mark_inode_dirty);
2108 * The @s_sync_lock is used to serialise concurrent sync operations
2109 * to avoid lock contention problems with concurrent wait_sb_inodes() calls.
2110 * Concurrent callers will block on the s_sync_lock rather than doing contending
2111 * walks. The queueing maintains sync(2) required behaviour as all the IO that
2112 * has been issued up to the time this function is enter is guaranteed to be
2113 * completed by the time we have gained the lock and waited for all IO that is
2114 * in progress regardless of the order callers are granted the lock.
2116 static void wait_sb_inodes(struct super_block *sb)
2118 struct inode *inode, *old_inode = NULL;
2121 * We need to be protected against the filesystem going from
2122 * r/o to r/w or vice versa.
2124 WARN_ON(!rwsem_is_locked(&sb->s_umount));
2126 mutex_lock(&sb->s_sync_lock);
2127 spin_lock(&sb->s_inode_list_lock);
2130 * Data integrity sync. Must wait for all pages under writeback,
2131 * because there may have been pages dirtied before our sync
2132 * call, but which had writeout started before we write it out.
2133 * In which case, the inode may not be on the dirty list, but
2134 * we still have to wait for that writeout.
2136 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
2137 struct address_space *mapping = inode->i_mapping;
2139 spin_lock(&inode->i_lock);
2140 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
2141 (mapping->nrpages == 0)) {
2142 spin_unlock(&inode->i_lock);
2146 spin_unlock(&inode->i_lock);
2147 spin_unlock(&sb->s_inode_list_lock);
2150 * We hold a reference to 'inode' so it couldn't have been
2151 * removed from s_inodes list while we dropped the
2152 * s_inode_list_lock. We cannot iput the inode now as we can
2153 * be holding the last reference and we cannot iput it under
2154 * s_inode_list_lock. So we keep the reference and iput it
2161 * We keep the error status of individual mapping so that
2162 * applications can catch the writeback error using fsync(2).
2163 * See filemap_fdatawait_keep_errors() for details.
2165 filemap_fdatawait_keep_errors(mapping);
2169 spin_lock(&sb->s_inode_list_lock);
2171 spin_unlock(&sb->s_inode_list_lock);
2173 mutex_unlock(&sb->s_sync_lock);
2176 static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr,
2177 enum wb_reason reason, bool skip_if_busy)
2179 DEFINE_WB_COMPLETION_ONSTACK(done);
2180 struct wb_writeback_work work = {
2182 .sync_mode = WB_SYNC_NONE,
2183 .tagged_writepages = 1,
2188 struct backing_dev_info *bdi = sb->s_bdi;
2190 if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info)
2192 WARN_ON(!rwsem_is_locked(&sb->s_umount));
2194 bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy);
2195 wb_wait_for_completion(bdi, &done);
2199 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
2200 * @sb: the superblock
2201 * @nr: the number of pages to write
2202 * @reason: reason why some writeback work initiated
2204 * Start writeback on some inodes on this super_block. No guarantees are made
2205 * on how many (if any) will be written, and this function does not wait
2206 * for IO completion of submitted IO.
2208 void writeback_inodes_sb_nr(struct super_block *sb,
2210 enum wb_reason reason)
2212 __writeback_inodes_sb_nr(sb, nr, reason, false);
2214 EXPORT_SYMBOL(writeback_inodes_sb_nr);
2217 * writeback_inodes_sb - writeback dirty inodes from given super_block
2218 * @sb: the superblock
2219 * @reason: reason why some writeback work was initiated
2221 * Start writeback on some inodes on this super_block. No guarantees are made
2222 * on how many (if any) will be written, and this function does not wait
2223 * for IO completion of submitted IO.
2225 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
2227 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
2229 EXPORT_SYMBOL(writeback_inodes_sb);
2232 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
2233 * @sb: the superblock
2234 * @nr: the number of pages to write
2235 * @reason: the reason of writeback
2237 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
2238 * Returns 1 if writeback was started, 0 if not.
2240 bool try_to_writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr,
2241 enum wb_reason reason)
2243 if (!down_read_trylock(&sb->s_umount))
2246 __writeback_inodes_sb_nr(sb, nr, reason, true);
2247 up_read(&sb->s_umount);
2250 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
2253 * try_to_writeback_inodes_sb - try to start writeback if none underway
2254 * @sb: the superblock
2255 * @reason: reason why some writeback work was initiated
2257 * Implement by try_to_writeback_inodes_sb_nr()
2258 * Returns 1 if writeback was started, 0 if not.
2260 bool try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
2262 return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
2264 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
2267 * sync_inodes_sb - sync sb inode pages
2268 * @sb: the superblock
2270 * This function writes and waits on any dirty inode belonging to this
2273 void sync_inodes_sb(struct super_block *sb)
2275 DEFINE_WB_COMPLETION_ONSTACK(done);
2276 struct wb_writeback_work work = {
2278 .sync_mode = WB_SYNC_ALL,
2279 .nr_pages = LONG_MAX,
2282 .reason = WB_REASON_SYNC,
2285 struct backing_dev_info *bdi = sb->s_bdi;
2288 * Can't skip on !bdi_has_dirty() because we should wait for !dirty
2289 * inodes under writeback and I_DIRTY_TIME inodes ignored by
2290 * bdi_has_dirty() need to be written out too.
2292 if (bdi == &noop_backing_dev_info)
2294 WARN_ON(!rwsem_is_locked(&sb->s_umount));
2296 bdi_split_work_to_wbs(bdi, &work, false);
2297 wb_wait_for_completion(bdi, &done);
2301 EXPORT_SYMBOL(sync_inodes_sb);
2304 * write_inode_now - write an inode to disk
2305 * @inode: inode to write to disk
2306 * @sync: whether the write should be synchronous or not
2308 * This function commits an inode to disk immediately if it is dirty. This is
2309 * primarily needed by knfsd.
2311 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
2313 int write_inode_now(struct inode *inode, int sync)
2315 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
2316 struct writeback_control wbc = {
2317 .nr_to_write = LONG_MAX,
2318 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
2320 .range_end = LLONG_MAX,
2323 if (!mapping_cap_writeback_dirty(inode->i_mapping))
2324 wbc.nr_to_write = 0;
2327 return writeback_single_inode(inode, wb, &wbc);
2329 EXPORT_SYMBOL(write_inode_now);
2332 * sync_inode - write an inode and its pages to disk.
2333 * @inode: the inode to sync
2334 * @wbc: controls the writeback mode
2336 * sync_inode() will write an inode and its pages to disk. It will also
2337 * correctly update the inode on its superblock's dirty inode lists and will
2338 * update inode->i_state.
2340 * The caller must have a ref on the inode.
2342 int sync_inode(struct inode *inode, struct writeback_control *wbc)
2344 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
2346 EXPORT_SYMBOL(sync_inode);
2349 * sync_inode_metadata - write an inode to disk
2350 * @inode: the inode to sync
2351 * @wait: wait for I/O to complete.
2353 * Write an inode to disk and adjust its dirty state after completion.
2355 * Note: only writes the actual inode, no associated data or other metadata.
2357 int sync_inode_metadata(struct inode *inode, int wait)
2359 struct writeback_control wbc = {
2360 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
2361 .nr_to_write = 0, /* metadata-only */
2364 return sync_inode(inode, &wbc);
2366 EXPORT_SYMBOL(sync_inode_metadata);