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 unsigned int single_wait
:1;
57 unsigned int single_done
:1;
58 enum wb_reason reason
; /* why was writeback initiated? */
60 struct list_head list
; /* pending work list */
61 struct wb_completion
*done
; /* set if the caller waits */
65 * If one wants to wait for one or more wb_writeback_works, each work's
66 * ->done should be set to a wb_completion defined using the following
67 * macro. Once all work items are issued with wb_queue_work(), the caller
68 * can wait for the completion of all using wb_wait_for_completion(). Work
69 * items which are waited upon aren't freed automatically on completion.
71 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \
72 struct wb_completion cmpl = { \
73 .cnt = ATOMIC_INIT(1), \
78 * If an inode is constantly having its pages dirtied, but then the
79 * updates stop dirtytime_expire_interval seconds in the past, it's
80 * possible for the worst case time between when an inode has its
81 * timestamps updated and when they finally get written out to be two
82 * dirtytime_expire_intervals. We set the default to 12 hours (in
83 * seconds), which means most of the time inodes will have their
84 * timestamps written to disk after 12 hours, but in the worst case a
85 * few inodes might not their timestamps updated for 24 hours.
87 unsigned int dirtytime_expire_interval
= 12 * 60 * 60;
89 static inline struct inode
*wb_inode(struct list_head
*head
)
91 return list_entry(head
, struct inode
, i_wb_list
);
95 * Include the creation of the trace points after defining the
96 * wb_writeback_work structure and inline functions so that the definition
97 * remains local to this file.
99 #define CREATE_TRACE_POINTS
100 #include <trace/events/writeback.h>
102 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
104 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
106 if (wb_has_dirty_io(wb
)) {
109 set_bit(WB_has_dirty_io
, &wb
->state
);
110 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
111 atomic_long_add(wb
->avg_write_bandwidth
,
112 &wb
->bdi
->tot_write_bandwidth
);
117 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
119 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
120 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
121 clear_bit(WB_has_dirty_io
, &wb
->state
);
122 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
123 &wb
->bdi
->tot_write_bandwidth
) < 0);
128 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
129 * @inode: inode to be moved
130 * @wb: target bdi_writeback
131 * @head: one of @wb->b_{dirty|io|more_io}
133 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
134 * Returns %true if @inode is the first occupant of the !dirty_time IO
135 * lists; otherwise, %false.
137 static bool inode_wb_list_move_locked(struct inode
*inode
,
138 struct bdi_writeback
*wb
,
139 struct list_head
*head
)
141 assert_spin_locked(&wb
->list_lock
);
143 list_move(&inode
->i_wb_list
, head
);
145 /* dirty_time doesn't count as dirty_io until expiration */
146 if (head
!= &wb
->b_dirty_time
)
147 return wb_io_lists_populated(wb
);
149 wb_io_lists_depopulated(wb
);
154 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
155 * @inode: inode to be removed
156 * @wb: bdi_writeback @inode is being removed from
158 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
159 * clear %WB_has_dirty_io if all are empty afterwards.
161 static void inode_wb_list_del_locked(struct inode
*inode
,
162 struct bdi_writeback
*wb
)
164 assert_spin_locked(&wb
->list_lock
);
166 list_del_init(&inode
->i_wb_list
);
167 wb_io_lists_depopulated(wb
);
170 static void wb_wakeup(struct bdi_writeback
*wb
)
172 spin_lock_bh(&wb
->work_lock
);
173 if (test_bit(WB_registered
, &wb
->state
))
174 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
175 spin_unlock_bh(&wb
->work_lock
);
178 static void wb_queue_work(struct bdi_writeback
*wb
,
179 struct wb_writeback_work
*work
)
181 trace_writeback_queue(wb
->bdi
, work
);
183 spin_lock_bh(&wb
->work_lock
);
184 if (!test_bit(WB_registered
, &wb
->state
)) {
185 if (work
->single_wait
)
186 work
->single_done
= 1;
190 atomic_inc(&work
->done
->cnt
);
191 list_add_tail(&work
->list
, &wb
->work_list
);
192 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
194 spin_unlock_bh(&wb
->work_lock
);
198 * wb_wait_for_completion - wait for completion of bdi_writeback_works
199 * @bdi: bdi work items were issued to
200 * @done: target wb_completion
202 * Wait for one or more work items issued to @bdi with their ->done field
203 * set to @done, which should have been defined with
204 * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such
205 * work items are completed. Work items which are waited upon aren't freed
206 * automatically on completion.
208 static void wb_wait_for_completion(struct backing_dev_info
*bdi
,
209 struct wb_completion
*done
)
211 atomic_dec(&done
->cnt
); /* put down the initial count */
212 wait_event(bdi
->wb_waitq
, !atomic_read(&done
->cnt
));
215 #ifdef CONFIG_CGROUP_WRITEBACK
217 /* parameters for foreign inode detection, see wb_detach_inode() */
218 #define WB_FRN_TIME_SHIFT 13 /* 1s = 2^13, upto 8 secs w/ 16bit */
219 #define WB_FRN_TIME_AVG_SHIFT 3 /* avg = avg * 7/8 + new * 1/8 */
220 #define WB_FRN_TIME_CUT_DIV 2 /* ignore rounds < avg / 2 */
221 #define WB_FRN_TIME_PERIOD (2 * (1 << WB_FRN_TIME_SHIFT)) /* 2s */
223 #define WB_FRN_HIST_SLOTS 16 /* inode->i_wb_frn_history is 16bit */
224 #define WB_FRN_HIST_UNIT (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS)
225 /* each slot's duration is 2s / 16 */
226 #define WB_FRN_HIST_THR_SLOTS (WB_FRN_HIST_SLOTS / 2)
227 /* if foreign slots >= 8, switch */
228 #define WB_FRN_HIST_MAX_SLOTS (WB_FRN_HIST_THR_SLOTS / 2 + 1)
229 /* one round can affect upto 5 slots */
231 void __inode_attach_wb(struct inode
*inode
, struct page
*page
)
233 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
234 struct bdi_writeback
*wb
= NULL
;
236 if (inode_cgwb_enabled(inode
)) {
237 struct cgroup_subsys_state
*memcg_css
;
240 memcg_css
= mem_cgroup_css_from_page(page
);
241 wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
243 /* must pin memcg_css, see wb_get_create() */
244 memcg_css
= task_get_css(current
, memory_cgrp_id
);
245 wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
254 * There may be multiple instances of this function racing to
255 * update the same inode. Use cmpxchg() to tell the winner.
257 if (unlikely(cmpxchg(&inode
->i_wb
, NULL
, wb
)))
262 * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it
263 * @inode: inode of interest with i_lock held
265 * Returns @inode's wb with its list_lock held. @inode->i_lock must be
266 * held on entry and is released on return. The returned wb is guaranteed
267 * to stay @inode's associated wb until its list_lock is released.
269 static struct bdi_writeback
*
270 locked_inode_to_wb_and_lock_list(struct inode
*inode
)
271 __releases(&inode
->i_lock
)
272 __acquires(&wb
->list_lock
)
275 struct bdi_writeback
*wb
= inode_to_wb(inode
);
278 * inode_to_wb() association is protected by both
279 * @inode->i_lock and @wb->list_lock but list_lock nests
280 * outside i_lock. Drop i_lock and verify that the
281 * association hasn't changed after acquiring list_lock.
284 spin_unlock(&inode
->i_lock
);
285 spin_lock(&wb
->list_lock
);
286 wb_put(wb
); /* not gonna deref it anymore */
288 /* i_wb may have changed inbetween, can't use inode_to_wb() */
289 if (likely(wb
== inode
->i_wb
))
290 return wb
; /* @inode already has ref */
292 spin_unlock(&wb
->list_lock
);
294 spin_lock(&inode
->i_lock
);
299 * inode_to_wb_and_lock_list - determine an inode's wb and lock it
300 * @inode: inode of interest
302 * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held
305 static struct bdi_writeback
*inode_to_wb_and_lock_list(struct inode
*inode
)
306 __acquires(&wb
->list_lock
)
308 spin_lock(&inode
->i_lock
);
309 return locked_inode_to_wb_and_lock_list(inode
);
312 struct inode_switch_wbs_context
{
314 struct bdi_writeback
*new_wb
;
316 struct rcu_head rcu_head
;
317 struct work_struct work
;
320 static void inode_switch_wbs_work_fn(struct work_struct
*work
)
322 struct inode_switch_wbs_context
*isw
=
323 container_of(work
, struct inode_switch_wbs_context
, work
);
324 struct inode
*inode
= isw
->inode
;
325 struct address_space
*mapping
= inode
->i_mapping
;
326 struct bdi_writeback
*old_wb
= inode
->i_wb
;
327 struct bdi_writeback
*new_wb
= isw
->new_wb
;
328 struct radix_tree_iter iter
;
329 bool switched
= false;
333 * By the time control reaches here, RCU grace period has passed
334 * since I_WB_SWITCH assertion and all wb stat update transactions
335 * between unlocked_inode_to_wb_begin/end() are guaranteed to be
336 * synchronizing against mapping->tree_lock.
338 * Grabbing old_wb->list_lock, inode->i_lock and mapping->tree_lock
339 * gives us exclusion against all wb related operations on @inode
340 * including IO list manipulations and stat updates.
342 if (old_wb
< new_wb
) {
343 spin_lock(&old_wb
->list_lock
);
344 spin_lock_nested(&new_wb
->list_lock
, SINGLE_DEPTH_NESTING
);
346 spin_lock(&new_wb
->list_lock
);
347 spin_lock_nested(&old_wb
->list_lock
, SINGLE_DEPTH_NESTING
);
349 spin_lock(&inode
->i_lock
);
350 spin_lock_irq(&mapping
->tree_lock
);
353 * Once I_FREEING is visible under i_lock, the eviction path owns
354 * the inode and we shouldn't modify ->i_wb_list.
356 if (unlikely(inode
->i_state
& I_FREEING
))
360 * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points
361 * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to
362 * pages actually under underwriteback.
364 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
, 0,
365 PAGECACHE_TAG_DIRTY
) {
366 struct page
*page
= radix_tree_deref_slot_protected(slot
,
367 &mapping
->tree_lock
);
368 if (likely(page
) && PageDirty(page
)) {
369 __dec_wb_stat(old_wb
, WB_RECLAIMABLE
);
370 __inc_wb_stat(new_wb
, WB_RECLAIMABLE
);
374 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
, 0,
375 PAGECACHE_TAG_WRITEBACK
) {
376 struct page
*page
= radix_tree_deref_slot_protected(slot
,
377 &mapping
->tree_lock
);
379 WARN_ON_ONCE(!PageWriteback(page
));
380 __dec_wb_stat(old_wb
, WB_WRITEBACK
);
381 __inc_wb_stat(new_wb
, WB_WRITEBACK
);
388 * Transfer to @new_wb's IO list if necessary. The specific list
389 * @inode was on is ignored and the inode is put on ->b_dirty which
390 * is always correct including from ->b_dirty_time. The transfer
391 * preserves @inode->dirtied_when ordering.
393 if (!list_empty(&inode
->i_wb_list
)) {
396 inode_wb_list_del_locked(inode
, old_wb
);
397 inode
->i_wb
= new_wb
;
398 list_for_each_entry(pos
, &new_wb
->b_dirty
, i_wb_list
)
399 if (time_after_eq(inode
->dirtied_when
,
402 inode_wb_list_move_locked(inode
, new_wb
, pos
->i_wb_list
.prev
);
404 inode
->i_wb
= new_wb
;
407 /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */
408 inode
->i_wb_frn_winner
= 0;
409 inode
->i_wb_frn_avg_time
= 0;
410 inode
->i_wb_frn_history
= 0;
414 * Paired with load_acquire in unlocked_inode_to_wb_begin() and
415 * ensures that the new wb is visible if they see !I_WB_SWITCH.
417 smp_store_release(&inode
->i_state
, inode
->i_state
& ~I_WB_SWITCH
);
419 spin_unlock_irq(&mapping
->tree_lock
);
420 spin_unlock(&inode
->i_lock
);
421 spin_unlock(&new_wb
->list_lock
);
422 spin_unlock(&old_wb
->list_lock
);
434 static void inode_switch_wbs_rcu_fn(struct rcu_head
*rcu_head
)
436 struct inode_switch_wbs_context
*isw
= container_of(rcu_head
,
437 struct inode_switch_wbs_context
, rcu_head
);
439 /* needs to grab bh-unsafe locks, bounce to work item */
440 INIT_WORK(&isw
->work
, inode_switch_wbs_work_fn
);
441 schedule_work(&isw
->work
);
445 * inode_switch_wbs - change the wb association of an inode
446 * @inode: target inode
447 * @new_wb_id: ID of the new wb
449 * Switch @inode's wb association to the wb identified by @new_wb_id. The
450 * switching is performed asynchronously and may fail silently.
452 static void inode_switch_wbs(struct inode
*inode
, int new_wb_id
)
454 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
455 struct cgroup_subsys_state
*memcg_css
;
456 struct inode_switch_wbs_context
*isw
;
458 /* noop if seems to be already in progress */
459 if (inode
->i_state
& I_WB_SWITCH
)
462 isw
= kzalloc(sizeof(*isw
), GFP_ATOMIC
);
466 /* find and pin the new wb */
468 memcg_css
= css_from_id(new_wb_id
, &memory_cgrp_subsys
);
470 isw
->new_wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
475 /* while holding I_WB_SWITCH, no one else can update the association */
476 spin_lock(&inode
->i_lock
);
477 if (inode
->i_state
& (I_WB_SWITCH
| I_FREEING
) ||
478 inode_to_wb(inode
) == isw
->new_wb
) {
479 spin_unlock(&inode
->i_lock
);
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
);
504 * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it
505 * @wbc: writeback_control of interest
506 * @inode: target inode
508 * @inode is locked and about to be written back under the control of @wbc.
509 * Record @inode's writeback context into @wbc and unlock the i_lock. On
510 * writeback completion, wbc_detach_inode() should be called. This is used
511 * to track the cgroup writeback context.
513 void wbc_attach_and_unlock_inode(struct writeback_control
*wbc
,
516 wbc
->wb
= inode_to_wb(inode
);
519 wbc
->wb_id
= wbc
->wb
->memcg_css
->id
;
520 wbc
->wb_lcand_id
= inode
->i_wb_frn_winner
;
521 wbc
->wb_tcand_id
= 0;
523 wbc
->wb_lcand_bytes
= 0;
524 wbc
->wb_tcand_bytes
= 0;
527 spin_unlock(&inode
->i_lock
);
530 * A dying wb indicates that the memcg-blkcg mapping has changed
531 * and a new wb is already serving the memcg. Switch immediately.
533 if (unlikely(wb_dying(wbc
->wb
)))
534 inode_switch_wbs(inode
, wbc
->wb_id
);
538 * wbc_detach_inode - disassociate wbc from inode and perform foreign detection
539 * @wbc: writeback_control of the just finished writeback
541 * To be called after a writeback attempt of an inode finishes and undoes
542 * wbc_attach_and_unlock_inode(). Can be called under any context.
544 * As concurrent write sharing of an inode is expected to be very rare and
545 * memcg only tracks page ownership on first-use basis severely confining
546 * the usefulness of such sharing, cgroup writeback tracks ownership
547 * per-inode. While the support for concurrent write sharing of an inode
548 * is deemed unnecessary, an inode being written to by different cgroups at
549 * different points in time is a lot more common, and, more importantly,
550 * charging only by first-use can too readily lead to grossly incorrect
551 * behaviors (single foreign page can lead to gigabytes of writeback to be
552 * incorrectly attributed).
554 * To resolve this issue, cgroup writeback detects the majority dirtier of
555 * an inode and transfers the ownership to it. To avoid unnnecessary
556 * oscillation, the detection mechanism keeps track of history and gives
557 * out the switch verdict only if the foreign usage pattern is stable over
558 * a certain amount of time and/or writeback attempts.
560 * On each writeback attempt, @wbc tries to detect the majority writer
561 * using Boyer-Moore majority vote algorithm. In addition to the byte
562 * count from the majority voting, it also counts the bytes written for the
563 * current wb and the last round's winner wb (max of last round's current
564 * wb, the winner from two rounds ago, and the last round's majority
565 * candidate). Keeping track of the historical winner helps the algorithm
566 * to semi-reliably detect the most active writer even when it's not the
569 * Once the winner of the round is determined, whether the winner is
570 * foreign or not and how much IO time the round consumed is recorded in
571 * inode->i_wb_frn_history. If the amount of recorded foreign IO time is
572 * over a certain threshold, the switch verdict is given.
574 void wbc_detach_inode(struct writeback_control
*wbc
)
576 struct bdi_writeback
*wb
= wbc
->wb
;
577 struct inode
*inode
= wbc
->inode
;
578 u16 history
= inode
->i_wb_frn_history
;
579 unsigned long avg_time
= inode
->i_wb_frn_avg_time
;
580 unsigned long max_bytes
, max_time
;
583 /* pick the winner of this round */
584 if (wbc
->wb_bytes
>= wbc
->wb_lcand_bytes
&&
585 wbc
->wb_bytes
>= wbc
->wb_tcand_bytes
) {
587 max_bytes
= wbc
->wb_bytes
;
588 } else if (wbc
->wb_lcand_bytes
>= wbc
->wb_tcand_bytes
) {
589 max_id
= wbc
->wb_lcand_id
;
590 max_bytes
= wbc
->wb_lcand_bytes
;
592 max_id
= wbc
->wb_tcand_id
;
593 max_bytes
= wbc
->wb_tcand_bytes
;
597 * Calculate the amount of IO time the winner consumed and fold it
598 * into the running average kept per inode. If the consumed IO
599 * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for
600 * deciding whether to switch or not. This is to prevent one-off
601 * small dirtiers from skewing the verdict.
603 max_time
= DIV_ROUND_UP((max_bytes
>> PAGE_SHIFT
) << WB_FRN_TIME_SHIFT
,
604 wb
->avg_write_bandwidth
);
606 avg_time
+= (max_time
>> WB_FRN_TIME_AVG_SHIFT
) -
607 (avg_time
>> WB_FRN_TIME_AVG_SHIFT
);
609 avg_time
= max_time
; /* immediate catch up on first run */
611 if (max_time
>= avg_time
/ WB_FRN_TIME_CUT_DIV
) {
615 * The switch verdict is reached if foreign wb's consume
616 * more than a certain proportion of IO time in a
617 * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot
618 * history mask where each bit represents one sixteenth of
619 * the period. Determine the number of slots to shift into
620 * history from @max_time.
622 slots
= min(DIV_ROUND_UP(max_time
, WB_FRN_HIST_UNIT
),
623 (unsigned long)WB_FRN_HIST_MAX_SLOTS
);
625 if (wbc
->wb_id
!= max_id
)
626 history
|= (1U << slots
) - 1;
629 * Switch if the current wb isn't the consistent winner.
630 * If there are multiple closely competing dirtiers, the
631 * inode may switch across them repeatedly over time, which
632 * is okay. The main goal is avoiding keeping an inode on
633 * the wrong wb for an extended period of time.
635 if (hweight32(history
) > WB_FRN_HIST_THR_SLOTS
)
636 inode_switch_wbs(inode
, max_id
);
640 * Multiple instances of this function may race to update the
641 * following fields but we don't mind occassional inaccuracies.
643 inode
->i_wb_frn_winner
= max_id
;
644 inode
->i_wb_frn_avg_time
= min(avg_time
, (unsigned long)U16_MAX
);
645 inode
->i_wb_frn_history
= history
;
652 * wbc_account_io - account IO issued during writeback
653 * @wbc: writeback_control of the writeback in progress
654 * @page: page being written out
655 * @bytes: number of bytes being written out
657 * @bytes from @page are about to written out during the writeback
658 * controlled by @wbc. Keep the book for foreign inode detection. See
659 * wbc_detach_inode().
661 void wbc_account_io(struct writeback_control
*wbc
, struct page
*page
,
667 * pageout() path doesn't attach @wbc to the inode being written
668 * out. This is intentional as we don't want the function to block
669 * behind a slow cgroup. Ultimately, we want pageout() to kick off
670 * regular writeback instead of writing things out itself.
676 id
= mem_cgroup_css_from_page(page
)->id
;
679 if (id
== wbc
->wb_id
) {
680 wbc
->wb_bytes
+= bytes
;
684 if (id
== wbc
->wb_lcand_id
)
685 wbc
->wb_lcand_bytes
+= bytes
;
687 /* Boyer-Moore majority vote algorithm */
688 if (!wbc
->wb_tcand_bytes
)
689 wbc
->wb_tcand_id
= id
;
690 if (id
== wbc
->wb_tcand_id
)
691 wbc
->wb_tcand_bytes
+= bytes
;
693 wbc
->wb_tcand_bytes
-= min(bytes
, wbc
->wb_tcand_bytes
);
697 * inode_congested - test whether an inode is congested
698 * @inode: inode to test for congestion
699 * @cong_bits: mask of WB_[a]sync_congested bits to test
701 * Tests whether @inode is congested. @cong_bits is the mask of congestion
702 * bits to test and the return value is the mask of set bits.
704 * If cgroup writeback is enabled for @inode, the congestion state is
705 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
706 * associated with @inode is congested; otherwise, the root wb's congestion
709 int inode_congested(struct inode
*inode
, int cong_bits
)
712 * Once set, ->i_wb never becomes NULL while the inode is alive.
713 * Start transaction iff ->i_wb is visible.
715 if (inode
&& inode_to_wb_is_valid(inode
)) {
716 struct bdi_writeback
*wb
;
717 bool locked
, congested
;
719 wb
= unlocked_inode_to_wb_begin(inode
, &locked
);
720 congested
= wb_congested(wb
, cong_bits
);
721 unlocked_inode_to_wb_end(inode
, locked
);
725 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
727 EXPORT_SYMBOL_GPL(inode_congested
);
730 * wb_wait_for_single_work - wait for completion of a single bdi_writeback_work
731 * @bdi: bdi the work item was issued to
732 * @work: work item to wait for
734 * Wait for the completion of @work which was issued to one of @bdi's
735 * bdi_writeback's. The caller must have set @work->single_wait before
736 * issuing it. This wait operates independently fo
737 * wb_wait_for_completion() and also disables automatic freeing of @work.
739 static void wb_wait_for_single_work(struct backing_dev_info
*bdi
,
740 struct wb_writeback_work
*work
)
742 if (WARN_ON_ONCE(!work
->single_wait
))
745 wait_event(bdi
->wb_waitq
, work
->single_done
);
748 * Paired with smp_wmb() in wb_do_writeback() and ensures that all
749 * modifications to @work prior to assertion of ->single_done is
750 * visible to the caller once this function returns.
756 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
757 * @wb: target bdi_writeback to split @nr_pages to
758 * @nr_pages: number of pages to write for the whole bdi
760 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
761 * relation to the total write bandwidth of all wb's w/ dirty inodes on
764 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
766 unsigned long this_bw
= wb
->avg_write_bandwidth
;
767 unsigned long tot_bw
= atomic_long_read(&wb
->bdi
->tot_write_bandwidth
);
769 if (nr_pages
== LONG_MAX
)
773 * This may be called on clean wb's and proportional distribution
774 * may not make sense, just use the original @nr_pages in those
775 * cases. In general, we wanna err on the side of writing more.
777 if (!tot_bw
|| this_bw
>= tot_bw
)
780 return DIV_ROUND_UP_ULL((u64
)nr_pages
* this_bw
, tot_bw
);
784 * wb_clone_and_queue_work - clone a wb_writeback_work and issue it to a wb
785 * @wb: target bdi_writeback
786 * @base_work: source wb_writeback_work
788 * Try to make a clone of @base_work and issue it to @wb. If cloning
789 * succeeds, %true is returned; otherwise, @base_work is issued directly
790 * and %false is returned. In the latter case, the caller is required to
791 * wait for @base_work's completion using wb_wait_for_single_work().
793 * A clone is auto-freed on completion. @base_work never is.
795 static bool wb_clone_and_queue_work(struct bdi_writeback
*wb
,
796 struct wb_writeback_work
*base_work
)
798 struct wb_writeback_work
*work
;
800 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
804 work
->single_wait
= 0;
808 work
->single_wait
= 1;
810 work
->single_done
= 0;
811 wb_queue_work(wb
, work
);
812 return work
!= base_work
;
816 * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
817 * @bdi: target backing_dev_info
818 * @base_work: wb_writeback_work to issue
819 * @skip_if_busy: skip wb's which already have writeback in progress
821 * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
822 * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's
823 * distributed to the busy wbs according to each wb's proportion in the
824 * total active write bandwidth of @bdi.
826 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
827 struct wb_writeback_work
*base_work
,
830 long nr_pages
= base_work
->nr_pages
;
831 int next_blkcg_id
= 0;
832 struct bdi_writeback
*wb
;
837 if (!bdi_has_dirty_io(bdi
))
841 bdi_for_each_wb(wb
, bdi
, &iter
, next_blkcg_id
) {
842 if (!wb_has_dirty_io(wb
) ||
843 (skip_if_busy
&& writeback_in_progress(wb
)))
846 base_work
->nr_pages
= wb_split_bdi_pages(wb
, nr_pages
);
847 if (!wb_clone_and_queue_work(wb
, base_work
)) {
848 next_blkcg_id
= wb
->blkcg_css
->id
+ 1;
850 wb_wait_for_single_work(bdi
, base_work
);
857 #else /* CONFIG_CGROUP_WRITEBACK */
859 static struct bdi_writeback
*
860 locked_inode_to_wb_and_lock_list(struct inode
*inode
)
861 __releases(&inode
->i_lock
)
862 __acquires(&wb
->list_lock
)
864 struct bdi_writeback
*wb
= inode_to_wb(inode
);
866 spin_unlock(&inode
->i_lock
);
867 spin_lock(&wb
->list_lock
);
871 static struct bdi_writeback
*inode_to_wb_and_lock_list(struct inode
*inode
)
872 __acquires(&wb
->list_lock
)
874 struct bdi_writeback
*wb
= inode_to_wb(inode
);
876 spin_lock(&wb
->list_lock
);
880 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
885 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
886 struct wb_writeback_work
*base_work
,
891 if (bdi_has_dirty_io(bdi
) &&
892 (!skip_if_busy
|| !writeback_in_progress(&bdi
->wb
))) {
893 base_work
->auto_free
= 0;
894 base_work
->single_wait
= 0;
895 base_work
->single_done
= 0;
896 wb_queue_work(&bdi
->wb
, base_work
);
900 #endif /* CONFIG_CGROUP_WRITEBACK */
902 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
903 bool range_cyclic
, enum wb_reason reason
)
905 struct wb_writeback_work
*work
;
907 if (!wb_has_dirty_io(wb
))
911 * This is WB_SYNC_NONE writeback, so if allocation fails just
912 * wakeup the thread for old dirty data writeback
914 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
916 trace_writeback_nowork(wb
->bdi
);
921 work
->sync_mode
= WB_SYNC_NONE
;
922 work
->nr_pages
= nr_pages
;
923 work
->range_cyclic
= range_cyclic
;
924 work
->reason
= reason
;
927 wb_queue_work(wb
, work
);
931 * wb_start_background_writeback - start background writeback
932 * @wb: bdi_writback to write from
935 * This makes sure WB_SYNC_NONE background writeback happens. When
936 * this function returns, it is only guaranteed that for given wb
937 * some IO is happening if we are over background dirty threshold.
938 * Caller need not hold sb s_umount semaphore.
940 void wb_start_background_writeback(struct bdi_writeback
*wb
)
943 * We just wake up the flusher thread. It will perform background
944 * writeback as soon as there is no other work to do.
946 trace_writeback_wake_background(wb
->bdi
);
951 * Remove the inode from the writeback list it is on.
953 void inode_wb_list_del(struct inode
*inode
)
955 struct bdi_writeback
*wb
;
957 wb
= inode_to_wb_and_lock_list(inode
);
958 inode_wb_list_del_locked(inode
, wb
);
959 spin_unlock(&wb
->list_lock
);
963 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
964 * furthest end of its superblock's dirty-inode list.
966 * Before stamping the inode's ->dirtied_when, we check to see whether it is
967 * already the most-recently-dirtied inode on the b_dirty list. If that is
968 * the case then the inode must have been redirtied while it was being written
969 * out and we don't reset its dirtied_when.
971 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
973 if (!list_empty(&wb
->b_dirty
)) {
976 tail
= wb_inode(wb
->b_dirty
.next
);
977 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
978 inode
->dirtied_when
= jiffies
;
980 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
984 * requeue inode for re-scanning after bdi->b_io list is exhausted.
986 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
988 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
991 static void inode_sync_complete(struct inode
*inode
)
993 inode
->i_state
&= ~I_SYNC
;
994 /* If inode is clean an unused, put it into LRU now... */
995 inode_add_lru(inode
);
996 /* Waiters must see I_SYNC cleared before being woken up */
998 wake_up_bit(&inode
->i_state
, __I_SYNC
);
1001 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
1003 bool ret
= time_after(inode
->dirtied_when
, t
);
1004 #ifndef CONFIG_64BIT
1006 * For inodes being constantly redirtied, dirtied_when can get stuck.
1007 * It _appears_ to be in the future, but is actually in distant past.
1008 * This test is necessary to prevent such wrapped-around relative times
1009 * from permanently stopping the whole bdi writeback.
1011 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
1016 #define EXPIRE_DIRTY_ATIME 0x0001
1019 * Move expired (dirtied before work->older_than_this) dirty inodes from
1020 * @delaying_queue to @dispatch_queue.
1022 static int move_expired_inodes(struct list_head
*delaying_queue
,
1023 struct list_head
*dispatch_queue
,
1025 struct wb_writeback_work
*work
)
1027 unsigned long *older_than_this
= NULL
;
1028 unsigned long expire_time
;
1030 struct list_head
*pos
, *node
;
1031 struct super_block
*sb
= NULL
;
1032 struct inode
*inode
;
1036 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
1037 older_than_this
= work
->older_than_this
;
1038 else if (!work
->for_sync
) {
1039 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
1040 older_than_this
= &expire_time
;
1042 while (!list_empty(delaying_queue
)) {
1043 inode
= wb_inode(delaying_queue
->prev
);
1044 if (older_than_this
&&
1045 inode_dirtied_after(inode
, *older_than_this
))
1047 list_move(&inode
->i_wb_list
, &tmp
);
1049 if (flags
& EXPIRE_DIRTY_ATIME
)
1050 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
1051 if (sb_is_blkdev_sb(inode
->i_sb
))
1053 if (sb
&& sb
!= inode
->i_sb
)
1058 /* just one sb in list, splice to dispatch_queue and we're done */
1060 list_splice(&tmp
, dispatch_queue
);
1064 /* Move inodes from one superblock together */
1065 while (!list_empty(&tmp
)) {
1066 sb
= wb_inode(tmp
.prev
)->i_sb
;
1067 list_for_each_prev_safe(pos
, node
, &tmp
) {
1068 inode
= wb_inode(pos
);
1069 if (inode
->i_sb
== sb
)
1070 list_move(&inode
->i_wb_list
, dispatch_queue
);
1078 * Queue all expired dirty inodes for io, eldest first.
1080 * newly dirtied b_dirty b_io b_more_io
1081 * =============> gf edc BA
1083 * newly dirtied b_dirty b_io b_more_io
1084 * =============> g fBAedc
1086 * +--> dequeue for IO
1088 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
1092 assert_spin_locked(&wb
->list_lock
);
1093 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
1094 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
1095 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
1096 EXPIRE_DIRTY_ATIME
, work
);
1098 wb_io_lists_populated(wb
);
1099 trace_writeback_queue_io(wb
, work
, moved
);
1102 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1106 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
1107 trace_writeback_write_inode_start(inode
, wbc
);
1108 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
1109 trace_writeback_write_inode(inode
, wbc
);
1116 * Wait for writeback on an inode to complete. Called with i_lock held.
1117 * Caller must make sure inode cannot go away when we drop i_lock.
1119 static void __inode_wait_for_writeback(struct inode
*inode
)
1120 __releases(inode
->i_lock
)
1121 __acquires(inode
->i_lock
)
1123 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
1124 wait_queue_head_t
*wqh
;
1126 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
1127 while (inode
->i_state
& I_SYNC
) {
1128 spin_unlock(&inode
->i_lock
);
1129 __wait_on_bit(wqh
, &wq
, bit_wait
,
1130 TASK_UNINTERRUPTIBLE
);
1131 spin_lock(&inode
->i_lock
);
1136 * Wait for writeback on an inode to complete. Caller must have inode pinned.
1138 void inode_wait_for_writeback(struct inode
*inode
)
1140 spin_lock(&inode
->i_lock
);
1141 __inode_wait_for_writeback(inode
);
1142 spin_unlock(&inode
->i_lock
);
1146 * Sleep until I_SYNC is cleared. This function must be called with i_lock
1147 * held and drops it. It is aimed for callers not holding any inode reference
1148 * so once i_lock is dropped, inode can go away.
1150 static void inode_sleep_on_writeback(struct inode
*inode
)
1151 __releases(inode
->i_lock
)
1154 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
1157 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
1158 sleep
= inode
->i_state
& I_SYNC
;
1159 spin_unlock(&inode
->i_lock
);
1162 finish_wait(wqh
, &wait
);
1166 * Find proper writeback list for the inode depending on its current state and
1167 * possibly also change of its state while we were doing writeback. Here we
1168 * handle things such as livelock prevention or fairness of writeback among
1169 * inodes. This function can be called only by flusher thread - noone else
1170 * processes all inodes in writeback lists and requeueing inodes behind flusher
1171 * thread's back can have unexpected consequences.
1173 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
1174 struct writeback_control
*wbc
)
1176 if (inode
->i_state
& I_FREEING
)
1180 * Sync livelock prevention. Each inode is tagged and synced in one
1181 * shot. If still dirty, it will be redirty_tail()'ed below. Update
1182 * the dirty time to prevent enqueue and sync it again.
1184 if ((inode
->i_state
& I_DIRTY
) &&
1185 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
1186 inode
->dirtied_when
= jiffies
;
1188 if (wbc
->pages_skipped
) {
1190 * writeback is not making progress due to locked
1191 * buffers. Skip this inode for now.
1193 redirty_tail(inode
, wb
);
1197 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
1199 * We didn't write back all the pages. nfs_writepages()
1200 * sometimes bales out without doing anything.
1202 if (wbc
->nr_to_write
<= 0) {
1203 /* Slice used up. Queue for next turn. */
1204 requeue_io(inode
, wb
);
1207 * Writeback blocked by something other than
1208 * congestion. Delay the inode for some time to
1209 * avoid spinning on the CPU (100% iowait)
1210 * retrying writeback of the dirty page/inode
1211 * that cannot be performed immediately.
1213 redirty_tail(inode
, wb
);
1215 } else if (inode
->i_state
& I_DIRTY
) {
1217 * Filesystems can dirty the inode during writeback operations,
1218 * such as delayed allocation during submission or metadata
1219 * updates after data IO completion.
1221 redirty_tail(inode
, wb
);
1222 } else if (inode
->i_state
& I_DIRTY_TIME
) {
1223 inode
->dirtied_when
= jiffies
;
1224 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
1226 /* The inode is clean. Remove from writeback lists. */
1227 inode_wb_list_del_locked(inode
, wb
);
1232 * Write out an inode and its dirty pages. Do not update the writeback list
1233 * linkage. That is left to the caller. The caller is also responsible for
1234 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
1237 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1239 struct address_space
*mapping
= inode
->i_mapping
;
1240 long nr_to_write
= wbc
->nr_to_write
;
1244 WARN_ON(!(inode
->i_state
& I_SYNC
));
1246 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
1248 ret
= do_writepages(mapping
, wbc
);
1251 * Make sure to wait on the data before writing out the metadata.
1252 * This is important for filesystems that modify metadata on data
1253 * I/O completion. We don't do it for sync(2) writeback because it has a
1254 * separate, external IO completion path and ->sync_fs for guaranteeing
1255 * inode metadata is written back correctly.
1257 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
1258 int err
= filemap_fdatawait(mapping
);
1264 * Some filesystems may redirty the inode during the writeback
1265 * due to delalloc, clear dirty metadata flags right before
1268 spin_lock(&inode
->i_lock
);
1270 dirty
= inode
->i_state
& I_DIRTY
;
1271 if (inode
->i_state
& I_DIRTY_TIME
) {
1272 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
1273 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
1274 unlikely(time_after(jiffies
,
1275 (inode
->dirtied_time_when
+
1276 dirtytime_expire_interval
* HZ
)))) {
1277 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
1278 trace_writeback_lazytime(inode
);
1281 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
1282 inode
->i_state
&= ~dirty
;
1285 * Paired with smp_mb() in __mark_inode_dirty(). This allows
1286 * __mark_inode_dirty() to test i_state without grabbing i_lock -
1287 * either they see the I_DIRTY bits cleared or we see the dirtied
1290 * I_DIRTY_PAGES is always cleared together above even if @mapping
1291 * still has dirty pages. The flag is reinstated after smp_mb() if
1292 * necessary. This guarantees that either __mark_inode_dirty()
1293 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
1297 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
1298 inode
->i_state
|= I_DIRTY_PAGES
;
1300 spin_unlock(&inode
->i_lock
);
1302 if (dirty
& I_DIRTY_TIME
)
1303 mark_inode_dirty_sync(inode
);
1304 /* Don't write the inode if only I_DIRTY_PAGES was set */
1305 if (dirty
& ~I_DIRTY_PAGES
) {
1306 int err
= write_inode(inode
, wbc
);
1310 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
1315 * Write out an inode's dirty pages. Either the caller has an active reference
1316 * on the inode or the inode has I_WILL_FREE set.
1318 * This function is designed to be called for writing back one inode which
1319 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
1320 * and does more profound writeback list handling in writeback_sb_inodes().
1323 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
1324 struct writeback_control
*wbc
)
1328 spin_lock(&inode
->i_lock
);
1329 if (!atomic_read(&inode
->i_count
))
1330 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
1332 WARN_ON(inode
->i_state
& I_WILL_FREE
);
1334 if (inode
->i_state
& I_SYNC
) {
1335 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
1338 * It's a data-integrity sync. We must wait. Since callers hold
1339 * inode reference or inode has I_WILL_FREE set, it cannot go
1342 __inode_wait_for_writeback(inode
);
1344 WARN_ON(inode
->i_state
& I_SYNC
);
1346 * Skip inode if it is clean and we have no outstanding writeback in
1347 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
1348 * function since flusher thread may be doing for example sync in
1349 * parallel and if we move the inode, it could get skipped. So here we
1350 * make sure inode is on some writeback list and leave it there unless
1351 * we have completely cleaned the inode.
1353 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
1354 (wbc
->sync_mode
!= WB_SYNC_ALL
||
1355 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
1357 inode
->i_state
|= I_SYNC
;
1358 wbc_attach_and_unlock_inode(wbc
, inode
);
1360 ret
= __writeback_single_inode(inode
, wbc
);
1362 wbc_detach_inode(wbc
);
1363 spin_lock(&wb
->list_lock
);
1364 spin_lock(&inode
->i_lock
);
1366 * If inode is clean, remove it from writeback lists. Otherwise don't
1367 * touch it. See comment above for explanation.
1369 if (!(inode
->i_state
& I_DIRTY_ALL
))
1370 inode_wb_list_del_locked(inode
, wb
);
1371 spin_unlock(&wb
->list_lock
);
1372 inode_sync_complete(inode
);
1374 spin_unlock(&inode
->i_lock
);
1378 static long writeback_chunk_size(struct bdi_writeback
*wb
,
1379 struct wb_writeback_work
*work
)
1384 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
1385 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
1386 * here avoids calling into writeback_inodes_wb() more than once.
1388 * The intended call sequence for WB_SYNC_ALL writeback is:
1391 * writeback_sb_inodes() <== called only once
1392 * write_cache_pages() <== called once for each inode
1393 * (quickly) tag currently dirty pages
1394 * (maybe slowly) sync all tagged pages
1396 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
1399 pages
= min(wb
->avg_write_bandwidth
/ 2,
1400 global_wb_domain
.dirty_limit
/ DIRTY_SCOPE
);
1401 pages
= min(pages
, work
->nr_pages
);
1402 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
1403 MIN_WRITEBACK_PAGES
);
1410 * Write a portion of b_io inodes which belong to @sb.
1412 * Return the number of pages and/or inodes written.
1414 static long writeback_sb_inodes(struct super_block
*sb
,
1415 struct bdi_writeback
*wb
,
1416 struct wb_writeback_work
*work
)
1418 struct writeback_control wbc
= {
1419 .sync_mode
= work
->sync_mode
,
1420 .tagged_writepages
= work
->tagged_writepages
,
1421 .for_kupdate
= work
->for_kupdate
,
1422 .for_background
= work
->for_background
,
1423 .for_sync
= work
->for_sync
,
1424 .range_cyclic
= work
->range_cyclic
,
1426 .range_end
= LLONG_MAX
,
1428 unsigned long start_time
= jiffies
;
1430 long wrote
= 0; /* count both pages and inodes */
1432 while (!list_empty(&wb
->b_io
)) {
1433 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
1435 if (inode
->i_sb
!= sb
) {
1438 * We only want to write back data for this
1439 * superblock, move all inodes not belonging
1440 * to it back onto the dirty list.
1442 redirty_tail(inode
, wb
);
1447 * The inode belongs to a different superblock.
1448 * Bounce back to the caller to unpin this and
1449 * pin the next superblock.
1455 * Don't bother with new inodes or inodes being freed, first
1456 * kind does not need periodic writeout yet, and for the latter
1457 * kind writeout is handled by the freer.
1459 spin_lock(&inode
->i_lock
);
1460 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
1461 spin_unlock(&inode
->i_lock
);
1462 redirty_tail(inode
, wb
);
1465 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
1467 * If this inode is locked for writeback and we are not
1468 * doing writeback-for-data-integrity, move it to
1469 * b_more_io so that writeback can proceed with the
1470 * other inodes on s_io.
1472 * We'll have another go at writing back this inode
1473 * when we completed a full scan of b_io.
1475 spin_unlock(&inode
->i_lock
);
1476 requeue_io(inode
, wb
);
1477 trace_writeback_sb_inodes_requeue(inode
);
1480 spin_unlock(&wb
->list_lock
);
1483 * We already requeued the inode if it had I_SYNC set and we
1484 * are doing WB_SYNC_NONE writeback. So this catches only the
1487 if (inode
->i_state
& I_SYNC
) {
1488 /* Wait for I_SYNC. This function drops i_lock... */
1489 inode_sleep_on_writeback(inode
);
1490 /* Inode may be gone, start again */
1491 spin_lock(&wb
->list_lock
);
1494 inode
->i_state
|= I_SYNC
;
1495 wbc_attach_and_unlock_inode(&wbc
, inode
);
1497 write_chunk
= writeback_chunk_size(wb
, work
);
1498 wbc
.nr_to_write
= write_chunk
;
1499 wbc
.pages_skipped
= 0;
1502 * We use I_SYNC to pin the inode in memory. While it is set
1503 * evict_inode() will wait so the inode cannot be freed.
1505 __writeback_single_inode(inode
, &wbc
);
1507 wbc_detach_inode(&wbc
);
1508 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
1509 wrote
+= write_chunk
- wbc
.nr_to_write
;
1510 spin_lock(&wb
->list_lock
);
1511 spin_lock(&inode
->i_lock
);
1512 if (!(inode
->i_state
& I_DIRTY_ALL
))
1514 requeue_inode(inode
, wb
, &wbc
);
1515 inode_sync_complete(inode
);
1516 spin_unlock(&inode
->i_lock
);
1517 cond_resched_lock(&wb
->list_lock
);
1519 * bail out to wb_writeback() often enough to check
1520 * background threshold and other termination conditions.
1523 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1525 if (work
->nr_pages
<= 0)
1532 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
1533 struct wb_writeback_work
*work
)
1535 unsigned long start_time
= jiffies
;
1538 while (!list_empty(&wb
->b_io
)) {
1539 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
1540 struct super_block
*sb
= inode
->i_sb
;
1542 if (!trylock_super(sb
)) {
1544 * trylock_super() may fail consistently due to
1545 * s_umount being grabbed by someone else. Don't use
1546 * requeue_io() to avoid busy retrying the inode/sb.
1548 redirty_tail(inode
, wb
);
1551 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
1552 up_read(&sb
->s_umount
);
1554 /* refer to the same tests at the end of writeback_sb_inodes */
1556 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1558 if (work
->nr_pages
<= 0)
1562 /* Leave any unwritten inodes on b_io */
1566 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
1567 enum wb_reason reason
)
1569 struct wb_writeback_work work
= {
1570 .nr_pages
= nr_pages
,
1571 .sync_mode
= WB_SYNC_NONE
,
1576 spin_lock(&wb
->list_lock
);
1577 if (list_empty(&wb
->b_io
))
1578 queue_io(wb
, &work
);
1579 __writeback_inodes_wb(wb
, &work
);
1580 spin_unlock(&wb
->list_lock
);
1582 return nr_pages
- work
.nr_pages
;
1586 * Explicit flushing or periodic writeback of "old" data.
1588 * Define "old": the first time one of an inode's pages is dirtied, we mark the
1589 * dirtying-time in the inode's address_space. So this periodic writeback code
1590 * just walks the superblock inode list, writing back any inodes which are
1591 * older than a specific point in time.
1593 * Try to run once per dirty_writeback_interval. But if a writeback event
1594 * takes longer than a dirty_writeback_interval interval, then leave a
1597 * older_than_this takes precedence over nr_to_write. So we'll only write back
1598 * all dirty pages if they are all attached to "old" mappings.
1600 static long wb_writeback(struct bdi_writeback
*wb
,
1601 struct wb_writeback_work
*work
)
1603 unsigned long wb_start
= jiffies
;
1604 long nr_pages
= work
->nr_pages
;
1605 unsigned long oldest_jif
;
1606 struct inode
*inode
;
1609 oldest_jif
= jiffies
;
1610 work
->older_than_this
= &oldest_jif
;
1612 spin_lock(&wb
->list_lock
);
1615 * Stop writeback when nr_pages has been consumed
1617 if (work
->nr_pages
<= 0)
1621 * Background writeout and kupdate-style writeback may
1622 * run forever. Stop them if there is other work to do
1623 * so that e.g. sync can proceed. They'll be restarted
1624 * after the other works are all done.
1626 if ((work
->for_background
|| work
->for_kupdate
) &&
1627 !list_empty(&wb
->work_list
))
1631 * For background writeout, stop when we are below the
1632 * background dirty threshold
1634 if (work
->for_background
&& !wb_over_bg_thresh(wb
))
1638 * Kupdate and background works are special and we want to
1639 * include all inodes that need writing. Livelock avoidance is
1640 * handled by these works yielding to any other work so we are
1643 if (work
->for_kupdate
) {
1644 oldest_jif
= jiffies
-
1645 msecs_to_jiffies(dirty_expire_interval
* 10);
1646 } else if (work
->for_background
)
1647 oldest_jif
= jiffies
;
1649 trace_writeback_start(wb
->bdi
, work
);
1650 if (list_empty(&wb
->b_io
))
1653 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1655 progress
= __writeback_inodes_wb(wb
, work
);
1656 trace_writeback_written(wb
->bdi
, work
);
1658 wb_update_bandwidth(wb
, wb_start
);
1661 * Did we write something? Try for more
1663 * Dirty inodes are moved to b_io for writeback in batches.
1664 * The completion of the current batch does not necessarily
1665 * mean the overall work is done. So we keep looping as long
1666 * as made some progress on cleaning pages or inodes.
1671 * No more inodes for IO, bail
1673 if (list_empty(&wb
->b_more_io
))
1676 * Nothing written. Wait for some inode to
1677 * become available for writeback. Otherwise
1678 * we'll just busyloop.
1680 if (!list_empty(&wb
->b_more_io
)) {
1681 trace_writeback_wait(wb
->bdi
, work
);
1682 inode
= wb_inode(wb
->b_more_io
.prev
);
1683 spin_lock(&inode
->i_lock
);
1684 spin_unlock(&wb
->list_lock
);
1685 /* This function drops i_lock... */
1686 inode_sleep_on_writeback(inode
);
1687 spin_lock(&wb
->list_lock
);
1690 spin_unlock(&wb
->list_lock
);
1692 return nr_pages
- work
->nr_pages
;
1696 * Return the next wb_writeback_work struct that hasn't been processed yet.
1698 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1700 struct wb_writeback_work
*work
= NULL
;
1702 spin_lock_bh(&wb
->work_lock
);
1703 if (!list_empty(&wb
->work_list
)) {
1704 work
= list_entry(wb
->work_list
.next
,
1705 struct wb_writeback_work
, list
);
1706 list_del_init(&work
->list
);
1708 spin_unlock_bh(&wb
->work_lock
);
1713 * Add in the number of potentially dirty inodes, because each inode
1714 * write can dirty pagecache in the underlying blockdev.
1716 static unsigned long get_nr_dirty_pages(void)
1718 return global_page_state(NR_FILE_DIRTY
) +
1719 global_page_state(NR_UNSTABLE_NFS
) +
1720 get_nr_dirty_inodes();
1723 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1725 if (wb_over_bg_thresh(wb
)) {
1727 struct wb_writeback_work work
= {
1728 .nr_pages
= LONG_MAX
,
1729 .sync_mode
= WB_SYNC_NONE
,
1730 .for_background
= 1,
1732 .reason
= WB_REASON_BACKGROUND
,
1735 return wb_writeback(wb
, &work
);
1741 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1743 unsigned long expired
;
1747 * When set to zero, disable periodic writeback
1749 if (!dirty_writeback_interval
)
1752 expired
= wb
->last_old_flush
+
1753 msecs_to_jiffies(dirty_writeback_interval
* 10);
1754 if (time_before(jiffies
, expired
))
1757 wb
->last_old_flush
= jiffies
;
1758 nr_pages
= get_nr_dirty_pages();
1761 struct wb_writeback_work work
= {
1762 .nr_pages
= nr_pages
,
1763 .sync_mode
= WB_SYNC_NONE
,
1766 .reason
= WB_REASON_PERIODIC
,
1769 return wb_writeback(wb
, &work
);
1776 * Retrieve work items and do the writeback they describe
1778 static long wb_do_writeback(struct bdi_writeback
*wb
)
1780 struct wb_writeback_work
*work
;
1783 set_bit(WB_writeback_running
, &wb
->state
);
1784 while ((work
= get_next_work_item(wb
)) != NULL
) {
1785 struct wb_completion
*done
= work
->done
;
1786 bool need_wake_up
= false;
1788 trace_writeback_exec(wb
->bdi
, work
);
1790 wrote
+= wb_writeback(wb
, work
);
1792 if (work
->single_wait
) {
1793 WARN_ON_ONCE(work
->auto_free
);
1794 /* paired w/ rmb in wb_wait_for_single_work() */
1796 work
->single_done
= 1;
1797 need_wake_up
= true;
1798 } else if (work
->auto_free
) {
1802 if (done
&& atomic_dec_and_test(&done
->cnt
))
1803 need_wake_up
= true;
1806 wake_up_all(&wb
->bdi
->wb_waitq
);
1810 * Check for periodic writeback, kupdated() style
1812 wrote
+= wb_check_old_data_flush(wb
);
1813 wrote
+= wb_check_background_flush(wb
);
1814 clear_bit(WB_writeback_running
, &wb
->state
);
1820 * Handle writeback of dirty data for the device backed by this bdi. Also
1821 * reschedules periodically and does kupdated style flushing.
1823 void wb_workfn(struct work_struct
*work
)
1825 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1826 struct bdi_writeback
, dwork
);
1829 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1830 current
->flags
|= PF_SWAPWRITE
;
1832 if (likely(!current_is_workqueue_rescuer() ||
1833 !test_bit(WB_registered
, &wb
->state
))) {
1835 * The normal path. Keep writing back @wb until its
1836 * work_list is empty. Note that this path is also taken
1837 * if @wb is shutting down even when we're running off the
1838 * rescuer as work_list needs to be drained.
1841 pages_written
= wb_do_writeback(wb
);
1842 trace_writeback_pages_written(pages_written
);
1843 } while (!list_empty(&wb
->work_list
));
1846 * bdi_wq can't get enough workers and we're running off
1847 * the emergency worker. Don't hog it. Hopefully, 1024 is
1848 * enough for efficient IO.
1850 pages_written
= writeback_inodes_wb(wb
, 1024,
1851 WB_REASON_FORKER_THREAD
);
1852 trace_writeback_pages_written(pages_written
);
1855 if (!list_empty(&wb
->work_list
))
1856 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1857 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1858 wb_wakeup_delayed(wb
);
1860 current
->flags
&= ~PF_SWAPWRITE
;
1864 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1867 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1869 struct backing_dev_info
*bdi
;
1872 nr_pages
= get_nr_dirty_pages();
1875 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1876 struct bdi_writeback
*wb
;
1877 struct wb_iter iter
;
1879 if (!bdi_has_dirty_io(bdi
))
1882 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1883 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1890 * Wake up bdi's periodically to make sure dirtytime inodes gets
1891 * written back periodically. We deliberately do *not* check the
1892 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1893 * kernel to be constantly waking up once there are any dirtytime
1894 * inodes on the system. So instead we define a separate delayed work
1895 * function which gets called much more rarely. (By default, only
1896 * once every 12 hours.)
1898 * If there is any other write activity going on in the file system,
1899 * this function won't be necessary. But if the only thing that has
1900 * happened on the file system is a dirtytime inode caused by an atime
1901 * update, we need this infrastructure below to make sure that inode
1902 * eventually gets pushed out to disk.
1904 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1905 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1907 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1909 struct backing_dev_info
*bdi
;
1912 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1913 struct bdi_writeback
*wb
;
1914 struct wb_iter iter
;
1916 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1917 if (!list_empty(&bdi
->wb
.b_dirty_time
))
1918 wb_wakeup(&bdi
->wb
);
1921 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1924 static int __init
start_dirtytime_writeback(void)
1926 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1929 __initcall(start_dirtytime_writeback
);
1931 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1932 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1936 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1937 if (ret
== 0 && write
)
1938 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1942 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1944 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1945 struct dentry
*dentry
;
1946 const char *name
= "?";
1948 dentry
= d_find_alias(inode
);
1950 spin_lock(&dentry
->d_lock
);
1951 name
= (const char *) dentry
->d_name
.name
;
1954 "%s(%d): dirtied inode %lu (%s) on %s\n",
1955 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1956 name
, inode
->i_sb
->s_id
);
1958 spin_unlock(&dentry
->d_lock
);
1965 * __mark_inode_dirty - internal function
1966 * @inode: inode to mark
1967 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1968 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1969 * mark_inode_dirty_sync.
1971 * Put the inode on the super block's dirty list.
1973 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1974 * dirty list only if it is hashed or if it refers to a blockdev.
1975 * If it was not hashed, it will never be added to the dirty list
1976 * even if it is later hashed, as it will have been marked dirty already.
1978 * In short, make sure you hash any inodes _before_ you start marking
1981 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1982 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1983 * the kernel-internal blockdev inode represents the dirtying time of the
1984 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1985 * page->mapping->host, so the page-dirtying time is recorded in the internal
1988 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1989 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1991 struct super_block
*sb
= inode
->i_sb
;
1994 trace_writeback_mark_inode_dirty(inode
, flags
);
1997 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1998 * dirty the inode itself
2000 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
2001 trace_writeback_dirty_inode_start(inode
, flags
);
2003 if (sb
->s_op
->dirty_inode
)
2004 sb
->s_op
->dirty_inode(inode
, flags
);
2006 trace_writeback_dirty_inode(inode
, flags
);
2008 if (flags
& I_DIRTY_INODE
)
2009 flags
&= ~I_DIRTY_TIME
;
2010 dirtytime
= flags
& I_DIRTY_TIME
;
2013 * Paired with smp_mb() in __writeback_single_inode() for the
2014 * following lockless i_state test. See there for details.
2018 if (((inode
->i_state
& flags
) == flags
) ||
2019 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
2022 if (unlikely(block_dump
))
2023 block_dump___mark_inode_dirty(inode
);
2025 spin_lock(&inode
->i_lock
);
2026 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
2027 goto out_unlock_inode
;
2028 if ((inode
->i_state
& flags
) != flags
) {
2029 const int was_dirty
= inode
->i_state
& I_DIRTY
;
2031 inode_attach_wb(inode
, NULL
);
2033 if (flags
& I_DIRTY_INODE
)
2034 inode
->i_state
&= ~I_DIRTY_TIME
;
2035 inode
->i_state
|= flags
;
2038 * If the inode is being synced, just update its dirty state.
2039 * The unlocker will place the inode on the appropriate
2040 * superblock list, based upon its state.
2042 if (inode
->i_state
& I_SYNC
)
2043 goto out_unlock_inode
;
2046 * Only add valid (hashed) inodes to the superblock's
2047 * dirty list. Add blockdev inodes as well.
2049 if (!S_ISBLK(inode
->i_mode
)) {
2050 if (inode_unhashed(inode
))
2051 goto out_unlock_inode
;
2053 if (inode
->i_state
& I_FREEING
)
2054 goto out_unlock_inode
;
2057 * If the inode was already on b_dirty/b_io/b_more_io, don't
2058 * reposition it (that would break b_dirty time-ordering).
2061 struct bdi_writeback
*wb
;
2062 struct list_head
*dirty_list
;
2063 bool wakeup_bdi
= false;
2065 wb
= locked_inode_to_wb_and_lock_list(inode
);
2067 WARN(bdi_cap_writeback_dirty(wb
->bdi
) &&
2068 !test_bit(WB_registered
, &wb
->state
),
2069 "bdi-%s not registered\n", wb
->bdi
->name
);
2071 inode
->dirtied_when
= jiffies
;
2073 inode
->dirtied_time_when
= jiffies
;
2075 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
2076 dirty_list
= &wb
->b_dirty
;
2078 dirty_list
= &wb
->b_dirty_time
;
2080 wakeup_bdi
= inode_wb_list_move_locked(inode
, wb
,
2083 spin_unlock(&wb
->list_lock
);
2084 trace_writeback_dirty_inode_enqueue(inode
);
2087 * If this is the first dirty inode for this bdi,
2088 * we have to wake-up the corresponding bdi thread
2089 * to make sure background write-back happens
2092 if (bdi_cap_writeback_dirty(wb
->bdi
) && wakeup_bdi
)
2093 wb_wakeup_delayed(wb
);
2098 spin_unlock(&inode
->i_lock
);
2101 EXPORT_SYMBOL(__mark_inode_dirty
);
2103 static void wait_sb_inodes(struct super_block
*sb
)
2105 struct inode
*inode
, *old_inode
= NULL
;
2108 * We need to be protected against the filesystem going from
2109 * r/o to r/w or vice versa.
2111 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
2113 spin_lock(&inode_sb_list_lock
);
2116 * Data integrity sync. Must wait for all pages under writeback,
2117 * because there may have been pages dirtied before our sync
2118 * call, but which had writeout started before we write it out.
2119 * In which case, the inode may not be on the dirty list, but
2120 * we still have to wait for that writeout.
2122 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
2123 struct address_space
*mapping
= inode
->i_mapping
;
2125 spin_lock(&inode
->i_lock
);
2126 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
2127 (mapping
->nrpages
== 0)) {
2128 spin_unlock(&inode
->i_lock
);
2132 spin_unlock(&inode
->i_lock
);
2133 spin_unlock(&inode_sb_list_lock
);
2136 * We hold a reference to 'inode' so it couldn't have been
2137 * removed from s_inodes list while we dropped the
2138 * inode_sb_list_lock. We cannot iput the inode now as we can
2139 * be holding the last reference and we cannot iput it under
2140 * inode_sb_list_lock. So we keep the reference and iput it
2146 filemap_fdatawait(mapping
);
2150 spin_lock(&inode_sb_list_lock
);
2152 spin_unlock(&inode_sb_list_lock
);
2156 static void __writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
2157 enum wb_reason reason
, bool skip_if_busy
)
2159 DEFINE_WB_COMPLETION_ONSTACK(done
);
2160 struct wb_writeback_work work
= {
2162 .sync_mode
= WB_SYNC_NONE
,
2163 .tagged_writepages
= 1,
2168 struct backing_dev_info
*bdi
= sb
->s_bdi
;
2170 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
2172 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
2174 bdi_split_work_to_wbs(sb
->s_bdi
, &work
, skip_if_busy
);
2175 wb_wait_for_completion(bdi
, &done
);
2179 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
2180 * @sb: the superblock
2181 * @nr: the number of pages to write
2182 * @reason: reason why some writeback work initiated
2184 * Start writeback on some inodes on this super_block. No guarantees are made
2185 * on how many (if any) will be written, and this function does not wait
2186 * for IO completion of submitted IO.
2188 void writeback_inodes_sb_nr(struct super_block
*sb
,
2190 enum wb_reason reason
)
2192 __writeback_inodes_sb_nr(sb
, nr
, reason
, false);
2194 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
2197 * writeback_inodes_sb - writeback dirty inodes from given super_block
2198 * @sb: the superblock
2199 * @reason: reason why some writeback work was initiated
2201 * Start writeback on some inodes on this super_block. No guarantees are made
2202 * on how many (if any) will be written, and this function does not wait
2203 * for IO completion of submitted IO.
2205 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
2207 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
2209 EXPORT_SYMBOL(writeback_inodes_sb
);
2212 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
2213 * @sb: the superblock
2214 * @nr: the number of pages to write
2215 * @reason: the reason of writeback
2217 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
2218 * Returns 1 if writeback was started, 0 if not.
2220 bool try_to_writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
2221 enum wb_reason reason
)
2223 if (!down_read_trylock(&sb
->s_umount
))
2226 __writeback_inodes_sb_nr(sb
, nr
, reason
, true);
2227 up_read(&sb
->s_umount
);
2230 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
2233 * try_to_writeback_inodes_sb - try to start writeback if none underway
2234 * @sb: the superblock
2235 * @reason: reason why some writeback work was initiated
2237 * Implement by try_to_writeback_inodes_sb_nr()
2238 * Returns 1 if writeback was started, 0 if not.
2240 bool try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
2242 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
2244 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
2247 * sync_inodes_sb - sync sb inode pages
2248 * @sb: the superblock
2250 * This function writes and waits on any dirty inode belonging to this
2253 void sync_inodes_sb(struct super_block
*sb
)
2255 DEFINE_WB_COMPLETION_ONSTACK(done
);
2256 struct wb_writeback_work work
= {
2258 .sync_mode
= WB_SYNC_ALL
,
2259 .nr_pages
= LONG_MAX
,
2262 .reason
= WB_REASON_SYNC
,
2265 struct backing_dev_info
*bdi
= sb
->s_bdi
;
2267 /* Nothing to do? */
2268 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
2270 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
2272 bdi_split_work_to_wbs(bdi
, &work
, false);
2273 wb_wait_for_completion(bdi
, &done
);
2277 EXPORT_SYMBOL(sync_inodes_sb
);
2280 * write_inode_now - write an inode to disk
2281 * @inode: inode to write to disk
2282 * @sync: whether the write should be synchronous or not
2284 * This function commits an inode to disk immediately if it is dirty. This is
2285 * primarily needed by knfsd.
2287 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
2289 int write_inode_now(struct inode
*inode
, int sync
)
2291 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
2292 struct writeback_control wbc
= {
2293 .nr_to_write
= LONG_MAX
,
2294 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
2296 .range_end
= LLONG_MAX
,
2299 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
2300 wbc
.nr_to_write
= 0;
2303 return writeback_single_inode(inode
, wb
, &wbc
);
2305 EXPORT_SYMBOL(write_inode_now
);
2308 * sync_inode - write an inode and its pages to disk.
2309 * @inode: the inode to sync
2310 * @wbc: controls the writeback mode
2312 * sync_inode() will write an inode and its pages to disk. It will also
2313 * correctly update the inode on its superblock's dirty inode lists and will
2314 * update inode->i_state.
2316 * The caller must have a ref on the inode.
2318 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
2320 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
2322 EXPORT_SYMBOL(sync_inode
);
2325 * sync_inode_metadata - write an inode to disk
2326 * @inode: the inode to sync
2327 * @wait: wait for I/O to complete.
2329 * Write an inode to disk and adjust its dirty state after completion.
2331 * Note: only writes the actual inode, no associated data or other metadata.
2333 int sync_inode_metadata(struct inode
*inode
, int wait
)
2335 struct writeback_control wbc
= {
2336 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
2337 .nr_to_write
= 0, /* metadata-only */
2340 return sync_inode(inode
, &wbc
);
2342 EXPORT_SYMBOL(sync_inode_metadata
);