~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/fs-writeback.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * fs/fs-writeback.c
  3  *
  4  * Copyright (C) 2002, Linus Torvalds.
  5  *
  6  * Contains all the functions related to writing back and waiting
  7  * upon dirty inodes against superblocks, and writing back dirty
  8  * pages against inodes.  ie: data writeback.  Writeout of the
  9  * inode itself is not handled here.
 10  *
 11  * 10Apr2002    Andrew Morton
 12  *              Split out of fs/inode.c
 13  *              Additions for address_space-based writeback
 14  */
 15 
 16 #include <linux/kernel.h>
 17 #include <linux/export.h>
 18 #include <linux/spinlock.h>
 19 #include <linux/slab.h>
 20 #include <linux/sched.h>
 21 #include <linux/fs.h>
 22 #include <linux/mm.h>
 23 #include <linux/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>
 31 #include "internal.h"
 32 
 33 /*
 34  * 4MB minimal write chunk size
 35  */
 36 #define MIN_WRITEBACK_PAGES     (4096UL >> (PAGE_SHIFT - 10))
 37 
 38 struct wb_completion {
 39         atomic_t                cnt;
 40 };
 41 
 42 /*
 43  * Passed into wb_writeback(), essentially a subset of writeback_control
 44  */
 45 struct wb_writeback_work {
 46         long nr_pages;
 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? */
 57 
 58         struct list_head list;          /* pending work list */
 59         struct wb_completion *done;     /* set if the caller waits */
 60 };
 61 
 62 /*
 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.
 68  */
 69 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl)                              \
 70         struct wb_completion cmpl = {                                   \
 71                 .cnt            = ATOMIC_INIT(1),                       \
 72         }
 73 
 74 
 75 /*
 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.
 84  */
 85 unsigned int dirtytime_expire_interval = 12 * 60 * 60;
 86 
 87 static inline struct inode *wb_inode(struct list_head *head)
 88 {
 89         return list_entry(head, struct inode, i_io_list);
 90 }
 91 
 92 /*
 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.
 96  */
 97 #define CREATE_TRACE_POINTS
 98 #include <trace/events/writeback.h>
 99 
100 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
101 
102 static bool wb_io_lists_populated(struct bdi_writeback *wb)
103 {
104         if (wb_has_dirty_io(wb)) {
105                 return false;
106         } else {
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);
111                 return true;
112         }
113 }
114 
115 static void wb_io_lists_depopulated(struct bdi_writeback *wb)
116 {
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);
122         }
123 }
124 
125 /**
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|dirty_time}
130  *
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.
134  */
135 static bool inode_io_list_move_locked(struct inode *inode,
136                                       struct bdi_writeback *wb,
137                                       struct list_head *head)
138 {
139         assert_spin_locked(&wb->list_lock);
140 
141         list_move(&inode->i_io_list, head);
142 
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);
146 
147         wb_io_lists_depopulated(wb);
148         return false;
149 }
150 
151 /**
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
155  *
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.
158  */
159 static void inode_io_list_del_locked(struct inode *inode,
160                                      struct bdi_writeback *wb)
161 {
162         assert_spin_locked(&wb->list_lock);
163 
164         list_del_init(&inode->i_io_list);
165         wb_io_lists_depopulated(wb);
166 }
167 
168 static void wb_wakeup(struct bdi_writeback *wb)
169 {
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);
174 }
175 
176 static void finish_writeback_work(struct bdi_writeback *wb,
177                                   struct wb_writeback_work *work)
178 {
179         struct wb_completion *done = work->done;
180 
181         if (work->auto_free)
182                 kfree(work);
183         if (done && atomic_dec_and_test(&done->cnt))
184                 wake_up_all(&wb->bdi->wb_waitq);
185 }
186 
187 static void wb_queue_work(struct bdi_writeback *wb,
188                           struct wb_writeback_work *work)
189 {
190         trace_writeback_queue(wb, work);
191 
192         if (work->done)
193                 atomic_inc(&work->done->cnt);
194 
195         spin_lock_bh(&wb->work_lock);
196 
197         if (test_bit(WB_registered, &wb->state)) {
198                 list_add_tail(&work->list, &wb->work_list);
199                 mod_delayed_work(bdi_wq, &wb->dwork, 0);
200         } else
201                 finish_writeback_work(wb, work);
202 
203         spin_unlock_bh(&wb->work_lock);
204 }
205 
206 /**
207  * wb_wait_for_completion - wait for completion of bdi_writeback_works
208  * @bdi: bdi work items were issued to
209  * @done: target wb_completion
210  *
211  * Wait for one or more work items issued to @bdi with their ->done field
212  * set to @done, which should have been defined with
213  * DEFINE_WB_COMPLETION_ONSTACK().  This function returns after all such
214  * work items are completed.  Work items which are waited upon aren't freed
215  * automatically on completion.
216  */
217 static void wb_wait_for_completion(struct backing_dev_info *bdi,
218                                    struct wb_completion *done)
219 {
220         atomic_dec(&done->cnt);         /* put down the initial count */
221         wait_event(bdi->wb_waitq, !atomic_read(&done->cnt));
222 }
223 
224 #ifdef CONFIG_CGROUP_WRITEBACK
225 
226 /* parameters for foreign inode detection, see wb_detach_inode() */
227 #define WB_FRN_TIME_SHIFT       13      /* 1s = 2^13, upto 8 secs w/ 16bit */
228 #define WB_FRN_TIME_AVG_SHIFT   3       /* avg = avg * 7/8 + new * 1/8 */
229 #define WB_FRN_TIME_CUT_DIV     2       /* ignore rounds < avg / 2 */
230 #define WB_FRN_TIME_PERIOD      (2 * (1 << WB_FRN_TIME_SHIFT))  /* 2s */
231 
232 #define WB_FRN_HIST_SLOTS       16      /* inode->i_wb_frn_history is 16bit */
233 #define WB_FRN_HIST_UNIT        (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS)
234                                         /* each slot's duration is 2s / 16 */
235 #define WB_FRN_HIST_THR_SLOTS   (WB_FRN_HIST_SLOTS / 2)
236                                         /* if foreign slots >= 8, switch */
237 #define WB_FRN_HIST_MAX_SLOTS   (WB_FRN_HIST_THR_SLOTS / 2 + 1)
238                                         /* one round can affect upto 5 slots */
239 
240 static atomic_t isw_nr_in_flight = ATOMIC_INIT(0);
241 static struct workqueue_struct *isw_wq;
242 
243 void __inode_attach_wb(struct inode *inode, struct page *page)
244 {
245         struct backing_dev_info *bdi = inode_to_bdi(inode);
246         struct bdi_writeback *wb = NULL;
247 
248         if (inode_cgwb_enabled(inode)) {
249                 struct cgroup_subsys_state *memcg_css;
250 
251                 if (page) {
252                         memcg_css = mem_cgroup_css_from_page(page);
253                         wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
254                 } else {
255                         /* must pin memcg_css, see wb_get_create() */
256                         memcg_css = task_get_css(current, memory_cgrp_id);
257                         wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
258                         css_put(memcg_css);
259                 }
260         }
261 
262         if (!wb)
263                 wb = &bdi->wb;
264 
265         /*
266          * There may be multiple instances of this function racing to
267          * update the same inode.  Use cmpxchg() to tell the winner.
268          */
269         if (unlikely(cmpxchg(&inode->i_wb, NULL, wb)))
270                 wb_put(wb);
271 }
272 
273 /**
274  * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it
275  * @inode: inode of interest with i_lock held
276  *
277  * Returns @inode's wb with its list_lock held.  @inode->i_lock must be
278  * held on entry and is released on return.  The returned wb is guaranteed
279  * to stay @inode's associated wb until its list_lock is released.
280  */
281 static struct bdi_writeback *
282 locked_inode_to_wb_and_lock_list(struct inode *inode)
283         __releases(&inode->i_lock)
284         __acquires(&wb->list_lock)
285 {
286         while (true) {
287                 struct bdi_writeback *wb = inode_to_wb(inode);
288 
289                 /*
290                  * inode_to_wb() association is protected by both
291                  * @inode->i_lock and @wb->list_lock but list_lock nests
292                  * outside i_lock.  Drop i_lock and verify that the
293                  * association hasn't changed after acquiring list_lock.
294                  */
295                 wb_get(wb);
296                 spin_unlock(&inode->i_lock);
297                 spin_lock(&wb->list_lock);
298 
299                 /* i_wb may have changed inbetween, can't use inode_to_wb() */
300                 if (likely(wb == inode->i_wb)) {
301                         wb_put(wb);     /* @inode already has ref */
302                         return wb;
303                 }
304 
305                 spin_unlock(&wb->list_lock);
306                 wb_put(wb);
307                 cpu_relax();
308                 spin_lock(&inode->i_lock);
309         }
310 }
311 
312 /**
313  * inode_to_wb_and_lock_list - determine an inode's wb and lock it
314  * @inode: inode of interest
315  *
316  * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held
317  * on entry.
318  */
319 static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode)
320         __acquires(&wb->list_lock)
321 {
322         spin_lock(&inode->i_lock);
323         return locked_inode_to_wb_and_lock_list(inode);
324 }
325 
326 struct inode_switch_wbs_context {
327         struct inode            *inode;
328         struct bdi_writeback    *new_wb;
329 
330         struct rcu_head         rcu_head;
331         struct work_struct      work;
332 };
333 
334 static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi)
335 {
336         down_write(&bdi->wb_switch_rwsem);
337 }
338 
339 static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi)
340 {
341         up_write(&bdi->wb_switch_rwsem);
342 }
343 
344 static void inode_switch_wbs_work_fn(struct work_struct *work)
345 {
346         struct inode_switch_wbs_context *isw =
347                 container_of(work, struct inode_switch_wbs_context, work);
348         struct inode *inode = isw->inode;
349         struct backing_dev_info *bdi = inode_to_bdi(inode);
350         struct address_space *mapping = inode->i_mapping;
351         struct bdi_writeback *old_wb = inode->i_wb;
352         struct bdi_writeback *new_wb = isw->new_wb;
353         XA_STATE(xas, &mapping->i_pages, 0);
354         struct page *page;
355         bool switched = false;
356 
357         /*
358          * If @inode switches cgwb membership while sync_inodes_sb() is
359          * being issued, sync_inodes_sb() might miss it.  Synchronize.
360          */
361         down_read(&bdi->wb_switch_rwsem);
362 
363         /*
364          * By the time control reaches here, RCU grace period has passed
365          * since I_WB_SWITCH assertion and all wb stat update transactions
366          * between unlocked_inode_to_wb_begin/end() are guaranteed to be
367          * synchronizing against the i_pages lock.
368          *
369          * Grabbing old_wb->list_lock, inode->i_lock and the i_pages lock
370          * gives us exclusion against all wb related operations on @inode
371          * including IO list manipulations and stat updates.
372          */
373         if (old_wb < new_wb) {
374                 spin_lock(&old_wb->list_lock);
375                 spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING);
376         } else {
377                 spin_lock(&new_wb->list_lock);
378                 spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING);
379         }
380         spin_lock(&inode->i_lock);
381         xa_lock_irq(&mapping->i_pages);
382 
383         /*
384          * Once I_FREEING is visible under i_lock, the eviction path owns
385          * the inode and we shouldn't modify ->i_io_list.
386          */
387         if (unlikely(inode->i_state & I_FREEING))
388                 goto skip_switch;
389 
390         /*
391          * Count and transfer stats.  Note that PAGECACHE_TAG_DIRTY points
392          * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to
393          * pages actually under writeback.
394          */
395         xas_for_each_marked(&xas, page, ULONG_MAX, PAGECACHE_TAG_DIRTY) {
396                 if (PageDirty(page)) {
397                         dec_wb_stat(old_wb, WB_RECLAIMABLE);
398                         inc_wb_stat(new_wb, WB_RECLAIMABLE);
399                 }
400         }
401 
402         xas_set(&xas, 0);
403         xas_for_each_marked(&xas, page, ULONG_MAX, PAGECACHE_TAG_WRITEBACK) {
404                 WARN_ON_ONCE(!PageWriteback(page));
405                 dec_wb_stat(old_wb, WB_WRITEBACK);
406                 inc_wb_stat(new_wb, WB_WRITEBACK);
407         }
408 
409         wb_get(new_wb);
410 
411         /*
412          * Transfer to @new_wb's IO list if necessary.  The specific list
413          * @inode was on is ignored and the inode is put on ->b_dirty which
414          * is always correct including from ->b_dirty_time.  The transfer
415          * preserves @inode->dirtied_when ordering.
416          */
417         if (!list_empty(&inode->i_io_list)) {
418                 struct inode *pos;
419 
420                 inode_io_list_del_locked(inode, old_wb);
421                 inode->i_wb = new_wb;
422                 list_for_each_entry(pos, &new_wb->b_dirty, i_io_list)
423                         if (time_after_eq(inode->dirtied_when,
424                                           pos->dirtied_when))
425                                 break;
426                 inode_io_list_move_locked(inode, new_wb, pos->i_io_list.prev);
427         } else {
428                 inode->i_wb = new_wb;
429         }
430 
431         /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */
432         inode->i_wb_frn_winner = 0;
433         inode->i_wb_frn_avg_time = 0;
434         inode->i_wb_frn_history = 0;
435         switched = true;
436 skip_switch:
437         /*
438          * Paired with load_acquire in unlocked_inode_to_wb_begin() and
439          * ensures that the new wb is visible if they see !I_WB_SWITCH.
440          */
441         smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH);
442 
443         xa_unlock_irq(&mapping->i_pages);
444         spin_unlock(&inode->i_lock);
445         spin_unlock(&new_wb->list_lock);
446         spin_unlock(&old_wb->list_lock);
447 
448         up_read(&bdi->wb_switch_rwsem);
449 
450         if (switched) {
451                 wb_wakeup(new_wb);
452                 wb_put(old_wb);
453         }
454         wb_put(new_wb);
455 
456         iput(inode);
457         kfree(isw);
458 
459         atomic_dec(&isw_nr_in_flight);
460 }
461 
462 static void inode_switch_wbs_rcu_fn(struct rcu_head *rcu_head)
463 {
464         struct inode_switch_wbs_context *isw = container_of(rcu_head,
465                                 struct inode_switch_wbs_context, rcu_head);
466 
467         /* needs to grab bh-unsafe locks, bounce to work item */
468         INIT_WORK(&isw->work, inode_switch_wbs_work_fn);
469         queue_work(isw_wq, &isw->work);
470 }
471 
472 /**
473  * inode_switch_wbs - change the wb association of an inode
474  * @inode: target inode
475  * @new_wb_id: ID of the new wb
476  *
477  * Switch @inode's wb association to the wb identified by @new_wb_id.  The
478  * switching is performed asynchronously and may fail silently.
479  */
480 static void inode_switch_wbs(struct inode *inode, int new_wb_id)
481 {
482         struct backing_dev_info *bdi = inode_to_bdi(inode);
483         struct cgroup_subsys_state *memcg_css;
484         struct inode_switch_wbs_context *isw;
485 
486         /* noop if seems to be already in progress */
487         if (inode->i_state & I_WB_SWITCH)
488                 return;
489 
490         /*
491          * Avoid starting new switches while sync_inodes_sb() is in
492          * progress.  Otherwise, if the down_write protected issue path
493          * blocks heavily, we might end up starting a large number of
494          * switches which will block on the rwsem.
495          */
496         if (!down_read_trylock(&bdi->wb_switch_rwsem))
497                 return;
498 
499         isw = kzalloc(sizeof(*isw), GFP_ATOMIC);
500         if (!isw)
501                 goto out_unlock;
502 
503         /* find and pin the new wb */
504         rcu_read_lock();
505         memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys);
506         if (memcg_css)
507                 isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
508         rcu_read_unlock();
509         if (!isw->new_wb)
510                 goto out_free;
511 
512         /* while holding I_WB_SWITCH, no one else can update the association */
513         spin_lock(&inode->i_lock);
514         if (!(inode->i_sb->s_flags & SB_ACTIVE) ||
515             inode->i_state & (I_WB_SWITCH | I_FREEING) ||
516             inode_to_wb(inode) == isw->new_wb) {
517                 spin_unlock(&inode->i_lock);
518                 goto out_free;
519         }
520         inode->i_state |= I_WB_SWITCH;
521         __iget(inode);
522         spin_unlock(&inode->i_lock);
523 
524         isw->inode = inode;
525 
526         /*
527          * In addition to synchronizing among switchers, I_WB_SWITCH tells
528          * the RCU protected stat update paths to grab the i_page
529          * lock so that stat transfer can synchronize against them.
530          * Let's continue after I_WB_SWITCH is guaranteed to be visible.
531          */
532         call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn);
533 
534         atomic_inc(&isw_nr_in_flight);
535 
536         goto out_unlock;
537 
538 out_free:
539         if (isw->new_wb)
540                 wb_put(isw->new_wb);
541         kfree(isw);
542 out_unlock:
543         up_read(&bdi->wb_switch_rwsem);
544 }
545 
546 /**
547  * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it
548  * @wbc: writeback_control of interest
549  * @inode: target inode
550  *
551  * @inode is locked and about to be written back under the control of @wbc.
552  * Record @inode's writeback context into @wbc and unlock the i_lock.  On
553  * writeback completion, wbc_detach_inode() should be called.  This is used
554  * to track the cgroup writeback context.
555  */
556 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
557                                  struct inode *inode)
558 {
559         if (!inode_cgwb_enabled(inode)) {
560                 spin_unlock(&inode->i_lock);
561                 return;
562         }
563 
564         wbc->wb = inode_to_wb(inode);
565         wbc->inode = inode;
566 
567         wbc->wb_id = wbc->wb->memcg_css->id;
568         wbc->wb_lcand_id = inode->i_wb_frn_winner;
569         wbc->wb_tcand_id = 0;
570         wbc->wb_bytes = 0;
571         wbc->wb_lcand_bytes = 0;
572         wbc->wb_tcand_bytes = 0;
573 
574         wb_get(wbc->wb);
575         spin_unlock(&inode->i_lock);
576 
577         /*
578          * A dying wb indicates that the memcg-blkcg mapping has changed
579          * and a new wb is already serving the memcg.  Switch immediately.
580          */
581         if (unlikely(wb_dying(wbc->wb)))
582                 inode_switch_wbs(inode, wbc->wb_id);
583 }
584 
585 /**
586  * wbc_detach_inode - disassociate wbc from inode and perform foreign detection
587  * @wbc: writeback_control of the just finished writeback
588  *
589  * To be called after a writeback attempt of an inode finishes and undoes
590  * wbc_attach_and_unlock_inode().  Can be called under any context.
591  *
592  * As concurrent write sharing of an inode is expected to be very rare and
593  * memcg only tracks page ownership on first-use basis severely confining
594  * the usefulness of such sharing, cgroup writeback tracks ownership
595  * per-inode.  While the support for concurrent write sharing of an inode
596  * is deemed unnecessary, an inode being written to by different cgroups at
597  * different points in time is a lot more common, and, more importantly,
598  * charging only by first-use can too readily lead to grossly incorrect
599  * behaviors (single foreign page can lead to gigabytes of writeback to be
600  * incorrectly attributed).
601  *
602  * To resolve this issue, cgroup writeback detects the majority dirtier of
603  * an inode and transfers the ownership to it.  To avoid unnnecessary
604  * oscillation, the detection mechanism keeps track of history and gives
605  * out the switch verdict only if the foreign usage pattern is stable over
606  * a certain amount of time and/or writeback attempts.
607  *
608  * On each writeback attempt, @wbc tries to detect the majority writer
609  * using Boyer-Moore majority vote algorithm.  In addition to the byte
610  * count from the majority voting, it also counts the bytes written for the
611  * current wb and the last round's winner wb (max of last round's current
612  * wb, the winner from two rounds ago, and the last round's majority
613  * candidate).  Keeping track of the historical winner helps the algorithm
614  * to semi-reliably detect the most active writer even when it's not the
615  * absolute majority.
616  *
617  * Once the winner of the round is determined, whether the winner is
618  * foreign or not and how much IO time the round consumed is recorded in
619  * inode->i_wb_frn_history.  If the amount of recorded foreign IO time is
620  * over a certain threshold, the switch verdict is given.
621  */
622 void wbc_detach_inode(struct writeback_control *wbc)
623 {
624         struct bdi_writeback *wb = wbc->wb;
625         struct inode *inode = wbc->inode;
626         unsigned long avg_time, max_bytes, max_time;
627         u16 history;
628         int max_id;
629 
630         if (!wb)
631                 return;
632 
633         history = inode->i_wb_frn_history;
634         avg_time = inode->i_wb_frn_avg_time;
635 
636         /* pick the winner of this round */
637         if (wbc->wb_bytes >= wbc->wb_lcand_bytes &&
638             wbc->wb_bytes >= wbc->wb_tcand_bytes) {
639                 max_id = wbc->wb_id;
640                 max_bytes = wbc->wb_bytes;
641         } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) {
642                 max_id = wbc->wb_lcand_id;
643                 max_bytes = wbc->wb_lcand_bytes;
644         } else {
645                 max_id = wbc->wb_tcand_id;
646                 max_bytes = wbc->wb_tcand_bytes;
647         }
648 
649         /*
650          * Calculate the amount of IO time the winner consumed and fold it
651          * into the running average kept per inode.  If the consumed IO
652          * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for
653          * deciding whether to switch or not.  This is to prevent one-off
654          * small dirtiers from skewing the verdict.
655          */
656         max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT,
657                                 wb->avg_write_bandwidth);
658         if (avg_time)
659                 avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) -
660                             (avg_time >> WB_FRN_TIME_AVG_SHIFT);
661         else
662                 avg_time = max_time;    /* immediate catch up on first run */
663 
664         if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) {
665                 int slots;
666 
667                 /*
668                  * The switch verdict is reached if foreign wb's consume
669                  * more than a certain proportion of IO time in a
670                  * WB_FRN_TIME_PERIOD.  This is loosely tracked by 16 slot
671                  * history mask where each bit represents one sixteenth of
672                  * the period.  Determine the number of slots to shift into
673                  * history from @max_time.
674                  */
675                 slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT),
676                             (unsigned long)WB_FRN_HIST_MAX_SLOTS);
677                 history <<= slots;
678                 if (wbc->wb_id != max_id)
679                         history |= (1U << slots) - 1;
680 
681                 /*
682                  * Switch if the current wb isn't the consistent winner.
683                  * If there are multiple closely competing dirtiers, the
684                  * inode may switch across them repeatedly over time, which
685                  * is okay.  The main goal is avoiding keeping an inode on
686                  * the wrong wb for an extended period of time.
687                  */
688                 if (hweight32(history) > WB_FRN_HIST_THR_SLOTS)
689                         inode_switch_wbs(inode, max_id);
690         }
691 
692         /*
693          * Multiple instances of this function may race to update the
694          * following fields but we don't mind occassional inaccuracies.
695          */
696         inode->i_wb_frn_winner = max_id;
697         inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX);
698         inode->i_wb_frn_history = history;
699 
700         wb_put(wbc->wb);
701         wbc->wb = NULL;
702 }
703 
704 /**
705  * wbc_account_io - account IO issued during writeback
706  * @wbc: writeback_control of the writeback in progress
707  * @page: page being written out
708  * @bytes: number of bytes being written out
709  *
710  * @bytes from @page are about to written out during the writeback
711  * controlled by @wbc.  Keep the book for foreign inode detection.  See
712  * wbc_detach_inode().
713  */
714 void wbc_account_io(struct writeback_control *wbc, struct page *page,
715                     size_t bytes)
716 {
717         int id;
718 
719         /*
720          * pageout() path doesn't attach @wbc to the inode being written
721          * out.  This is intentional as we don't want the function to block
722          * behind a slow cgroup.  Ultimately, we want pageout() to kick off
723          * regular writeback instead of writing things out itself.
724          */
725         if (!wbc->wb)
726                 return;
727 
728         id = mem_cgroup_css_from_page(page)->id;
729 
730         if (id == wbc->wb_id) {
731                 wbc->wb_bytes += bytes;
732                 return;
733         }
734 
735         if (id == wbc->wb_lcand_id)
736                 wbc->wb_lcand_bytes += bytes;
737 
738         /* Boyer-Moore majority vote algorithm */
739         if (!wbc->wb_tcand_bytes)
740                 wbc->wb_tcand_id = id;
741         if (id == wbc->wb_tcand_id)
742                 wbc->wb_tcand_bytes += bytes;
743         else
744                 wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes);
745 }
746 EXPORT_SYMBOL_GPL(wbc_account_io);
747 
748 /**
749  * inode_congested - test whether an inode is congested
750  * @inode: inode to test for congestion (may be NULL)
751  * @cong_bits: mask of WB_[a]sync_congested bits to test
752  *
753  * Tests whether @inode is congested.  @cong_bits is the mask of congestion
754  * bits to test and the return value is the mask of set bits.
755  *
756  * If cgroup writeback is enabled for @inode, the congestion state is
757  * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
758  * associated with @inode is congested; otherwise, the root wb's congestion
759  * state is used.
760  *
761  * @inode is allowed to be NULL as this function is often called on
762  * mapping->host which is NULL for the swapper space.
763  */
764 int inode_congested(struct inode *inode, int cong_bits)
765 {
766         /*
767          * Once set, ->i_wb never becomes NULL while the inode is alive.
768          * Start transaction iff ->i_wb is visible.
769          */
770         if (inode && inode_to_wb_is_valid(inode)) {
771                 struct bdi_writeback *wb;
772                 struct wb_lock_cookie lock_cookie = {};
773                 bool congested;
774 
775                 wb = unlocked_inode_to_wb_begin(inode, &lock_cookie);
776                 congested = wb_congested(wb, cong_bits);
777                 unlocked_inode_to_wb_end(inode, &lock_cookie);
778                 return congested;
779         }
780 
781         return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
782 }
783 EXPORT_SYMBOL_GPL(inode_congested);
784 
785 /**
786  * wb_split_bdi_pages - split nr_pages to write according to bandwidth
787  * @wb: target bdi_writeback to split @nr_pages to
788  * @nr_pages: number of pages to write for the whole bdi
789  *
790  * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
791  * relation to the total write bandwidth of all wb's w/ dirty inodes on
792  * @wb->bdi.
793  */
794 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
795 {
796         unsigned long this_bw = wb->avg_write_bandwidth;
797         unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth);
798 
799         if (nr_pages == LONG_MAX)
800                 return LONG_MAX;
801 
802         /*
803          * This may be called on clean wb's and proportional distribution
804          * may not make sense, just use the original @nr_pages in those
805          * cases.  In general, we wanna err on the side of writing more.
806          */
807         if (!tot_bw || this_bw >= tot_bw)
808                 return nr_pages;
809         else
810                 return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw);
811 }
812 
813 /**
814  * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
815  * @bdi: target backing_dev_info
816  * @base_work: wb_writeback_work to issue
817  * @skip_if_busy: skip wb's which already have writeback in progress
818  *
819  * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
820  * have dirty inodes.  If @base_work->nr_page isn't %LONG_MAX, it's
821  * distributed to the busy wbs according to each wb's proportion in the
822  * total active write bandwidth of @bdi.
823  */
824 static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
825                                   struct wb_writeback_work *base_work,
826                                   bool skip_if_busy)
827 {
828         struct bdi_writeback *last_wb = NULL;
829         struct bdi_writeback *wb = list_entry(&bdi->wb_list,
830                                               struct bdi_writeback, bdi_node);
831 
832         might_sleep();
833 restart:
834         rcu_read_lock();
835         list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) {
836                 DEFINE_WB_COMPLETION_ONSTACK(fallback_work_done);
837                 struct wb_writeback_work fallback_work;
838                 struct wb_writeback_work *work;
839                 long nr_pages;
840 
841                 if (last_wb) {
842                         wb_put(last_wb);
843                         last_wb = NULL;
844                 }
845 
846                 /* SYNC_ALL writes out I_DIRTY_TIME too */
847                 if (!wb_has_dirty_io(wb) &&
848                     (base_work->sync_mode == WB_SYNC_NONE ||
849                      list_empty(&wb->b_dirty_time)))
850                         continue;
851                 if (skip_if_busy && writeback_in_progress(wb))
852                         continue;
853 
854                 nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages);
855 
856                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
857                 if (work) {
858                         *work = *base_work;
859                         work->nr_pages = nr_pages;
860                         work->auto_free = 1;
861                         wb_queue_work(wb, work);
862                         continue;
863                 }
864 
865                 /* alloc failed, execute synchronously using on-stack fallback */
866                 work = &fallback_work;
867                 *work = *base_work;
868                 work->nr_pages = nr_pages;
869                 work->auto_free = 0;
870                 work->done = &fallback_work_done;
871 
872                 wb_queue_work(wb, work);
873 
874                 /*
875                  * Pin @wb so that it stays on @bdi->wb_list.  This allows
876                  * continuing iteration from @wb after dropping and
877                  * regrabbing rcu read lock.
878                  */
879                 wb_get(wb);
880                 last_wb = wb;
881 
882                 rcu_read_unlock();
883                 wb_wait_for_completion(bdi, &fallback_work_done);
884                 goto restart;
885         }
886         rcu_read_unlock();
887 
888         if (last_wb)
889                 wb_put(last_wb);
890 }
891 
892 /**
893  * cgroup_writeback_umount - flush inode wb switches for umount
894  *
895  * This function is called when a super_block is about to be destroyed and
896  * flushes in-flight inode wb switches.  An inode wb switch goes through
897  * RCU and then workqueue, so the two need to be flushed in order to ensure
898  * that all previously scheduled switches are finished.  As wb switches are
899  * rare occurrences and synchronize_rcu() can take a while, perform
900  * flushing iff wb switches are in flight.
901  */
902 void cgroup_writeback_umount(void)
903 {
904         if (atomic_read(&isw_nr_in_flight)) {
905                 /*
906                  * Use rcu_barrier() to wait for all pending callbacks to
907                  * ensure that all in-flight wb switches are in the workqueue.
908                  */
909                 rcu_barrier();
910                 flush_workqueue(isw_wq);
911         }
912 }
913 
914 static int __init cgroup_writeback_init(void)
915 {
916         isw_wq = alloc_workqueue("inode_switch_wbs", 0, 0);
917         if (!isw_wq)
918                 return -ENOMEM;
919         return 0;
920 }
921 fs_initcall(cgroup_writeback_init);
922 
923 #else   /* CONFIG_CGROUP_WRITEBACK */
924 
925 static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) { }
926 static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) { }
927 
928 static struct bdi_writeback *
929 locked_inode_to_wb_and_lock_list(struct inode *inode)
930         __releases(&inode->i_lock)
931         __acquires(&wb->list_lock)
932 {
933         struct bdi_writeback *wb = inode_to_wb(inode);
934 
935         spin_unlock(&inode->i_lock);
936         spin_lock(&wb->list_lock);
937         return wb;
938 }
939 
940 static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode)
941         __acquires(&wb->list_lock)
942 {
943         struct bdi_writeback *wb = inode_to_wb(inode);
944 
945         spin_lock(&wb->list_lock);
946         return wb;
947 }
948 
949 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
950 {
951         return nr_pages;
952 }
953 
954 static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
955                                   struct wb_writeback_work *base_work,
956                                   bool skip_if_busy)
957 {
958         might_sleep();
959 
960         if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) {
961                 base_work->auto_free = 0;
962                 wb_queue_work(&bdi->wb, base_work);
963         }
964 }
965 
966 #endif  /* CONFIG_CGROUP_WRITEBACK */
967 
968 /*
969  * Add in the number of potentially dirty inodes, because each inode
970  * write can dirty pagecache in the underlying blockdev.
971  */
972 static unsigned long get_nr_dirty_pages(void)
973 {
974         return global_node_page_state(NR_FILE_DIRTY) +
975                 global_node_page_state(NR_UNSTABLE_NFS) +
976                 get_nr_dirty_inodes();
977 }
978 
979 static void wb_start_writeback(struct bdi_writeback *wb, enum wb_reason reason)
980 {
981         if (!wb_has_dirty_io(wb))
982                 return;
983 
984         /*
985          * All callers of this function want to start writeback of all
986          * dirty pages. Places like vmscan can call this at a very
987          * high frequency, causing pointless allocations of tons of
988          * work items and keeping the flusher threads busy retrieving
989          * that work. Ensure that we only allow one of them pending and
990          * inflight at the time.
991          */
992         if (test_bit(WB_start_all, &wb->state) ||
993             test_and_set_bit(WB_start_all, &wb->state))
994                 return;
995 
996         wb->start_all_reason = reason;
997         wb_wakeup(wb);
998 }
999 
1000 /**
1001  * wb_start_background_writeback - start background writeback
1002  * @wb: bdi_writback to write from
1003  *
1004  * Description:
1005  *   This makes sure WB_SYNC_NONE background writeback happens. When
1006  *   this function returns, it is only guaranteed that for given wb
1007  *   some IO is happening if we are over background dirty threshold.
1008  *   Caller need not hold sb s_umount semaphore.
1009  */
1010 void wb_start_background_writeback(struct bdi_writeback *wb)
1011 {
1012         /*
1013          * We just wake up the flusher thread. It will perform background
1014          * writeback as soon as there is no other work to do.
1015          */
1016         trace_writeback_wake_background(wb);
1017         wb_wakeup(wb);
1018 }
1019 
1020 /*
1021  * Remove the inode from the writeback list it is on.
1022  */
1023 void inode_io_list_del(struct inode *inode)
1024 {
1025         struct bdi_writeback *wb;
1026 
1027         wb = inode_to_wb_and_lock_list(inode);
1028         inode_io_list_del_locked(inode, wb);
1029         spin_unlock(&wb->list_lock);
1030 }
1031 
1032 /*
1033  * mark an inode as under writeback on the sb
1034  */
1035 void sb_mark_inode_writeback(struct inode *inode)
1036 {
1037         struct super_block *sb = inode->i_sb;
1038         unsigned long flags;
1039 
1040         if (list_empty(&inode->i_wb_list)) {
1041                 spin_lock_irqsave(&sb->s_inode_wblist_lock, flags);
1042                 if (list_empty(&inode->i_wb_list)) {
1043                         list_add_tail(&inode->i_wb_list, &sb->s_inodes_wb);
1044                         trace_sb_mark_inode_writeback(inode);
1045                 }
1046                 spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags);
1047         }
1048 }
1049 
1050 /*
1051  * clear an inode as under writeback on the sb
1052  */
1053 void sb_clear_inode_writeback(struct inode *inode)
1054 {
1055         struct super_block *sb = inode->i_sb;
1056         unsigned long flags;
1057 
1058         if (!list_empty(&inode->i_wb_list)) {
1059                 spin_lock_irqsave(&sb->s_inode_wblist_lock, flags);
1060                 if (!list_empty(&inode->i_wb_list)) {
1061                         list_del_init(&inode->i_wb_list);
1062                         trace_sb_clear_inode_writeback(inode);
1063                 }
1064                 spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags);
1065         }
1066 }
1067 
1068 /*
1069  * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
1070  * furthest end of its superblock's dirty-inode list.
1071  *
1072  * Before stamping the inode's ->dirtied_when, we check to see whether it is
1073  * already the most-recently-dirtied inode on the b_dirty list.  If that is
1074  * the case then the inode must have been redirtied while it was being written
1075  * out and we don't reset its dirtied_when.
1076  */
1077 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
1078 {
1079         if (!list_empty(&wb->b_dirty)) {
1080                 struct inode *tail;
1081 
1082                 tail = wb_inode(wb->b_dirty.next);
1083                 if (time_before(inode->dirtied_when, tail->dirtied_when))
1084                         inode->dirtied_when = jiffies;
1085         }
1086         inode_io_list_move_locked(inode, wb, &wb->b_dirty);
1087 }
1088 
1089 /*
1090  * requeue inode for re-scanning after bdi->b_io list is exhausted.
1091  */
1092 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
1093 {
1094         inode_io_list_move_locked(inode, wb, &wb->b_more_io);
1095 }
1096 
1097 static void inode_sync_complete(struct inode *inode)
1098 {
1099         inode->i_state &= ~I_SYNC;
1100         /* If inode is clean an unused, put it into LRU now... */
1101         inode_add_lru(inode);
1102         /* Waiters must see I_SYNC cleared before being woken up */
1103         smp_mb();
1104         wake_up_bit(&inode->i_state, __I_SYNC);
1105 }
1106 
1107 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
1108 {
1109         bool ret = time_after(inode->dirtied_when, t);
1110 #ifndef CONFIG_64BIT
1111         /*
1112          * For inodes being constantly redirtied, dirtied_when can get stuck.
1113          * It _appears_ to be in the future, but is actually in distant past.
1114          * This test is necessary to prevent such wrapped-around relative times
1115          * from permanently stopping the whole bdi writeback.
1116          */
1117         ret = ret && time_before_eq(inode->dirtied_when, jiffies);
1118 #endif
1119         return ret;
1120 }
1121 
1122 #define EXPIRE_DIRTY_ATIME 0x0001
1123 
1124 /*
1125  * Move expired (dirtied before work->older_than_this) dirty inodes from
1126  * @delaying_queue to @dispatch_queue.
1127  */
1128 static int move_expired_inodes(struct list_head *delaying_queue,
1129                                struct list_head *dispatch_queue,
1130                                int flags,
1131                                struct wb_writeback_work *work)
1132 {
1133         unsigned long *older_than_this = NULL;
1134         unsigned long expire_time;
1135         LIST_HEAD(tmp);
1136         struct list_head *pos, *node;
1137         struct super_block *sb = NULL;
1138         struct inode *inode;
1139         int do_sb_sort = 0;
1140         int moved = 0;
1141 
1142         if ((flags & EXPIRE_DIRTY_ATIME) == 0)
1143                 older_than_this = work->older_than_this;
1144         else if (!work->for_sync) {
1145                 expire_time = jiffies - (dirtytime_expire_interval * HZ);
1146                 older_than_this = &expire_time;
1147         }
1148         while (!list_empty(delaying_queue)) {
1149                 inode = wb_inode(delaying_queue->prev);
1150                 if (older_than_this &&
1151                     inode_dirtied_after(inode, *older_than_this))
1152                         break;
1153                 list_move(&inode->i_io_list, &tmp);
1154                 moved++;
1155                 if (flags & EXPIRE_DIRTY_ATIME)
1156                         set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
1157                 if (sb_is_blkdev_sb(inode->i_sb))
1158                         continue;
1159                 if (sb && sb != inode->i_sb)
1160                         do_sb_sort = 1;
1161                 sb = inode->i_sb;
1162         }
1163 
1164         /* just one sb in list, splice to dispatch_queue and we're done */
1165         if (!do_sb_sort) {
1166                 list_splice(&tmp, dispatch_queue);
1167                 goto out;
1168         }
1169 
1170         /* Move inodes from one superblock together */
1171         while (!list_empty(&tmp)) {
1172                 sb = wb_inode(tmp.prev)->i_sb;
1173                 list_for_each_prev_safe(pos, node, &tmp) {
1174                         inode = wb_inode(pos);
1175                         if (inode->i_sb == sb)
1176                                 list_move(&inode->i_io_list, dispatch_queue);
1177                 }
1178         }
1179 out:
1180         return moved;
1181 }
1182 
1183 /*
1184  * Queue all expired dirty inodes for io, eldest first.
1185  * Before
1186  *         newly dirtied     b_dirty    b_io    b_more_io
1187  *         =============>    gf         edc     BA
1188  * After
1189  *         newly dirtied     b_dirty    b_io    b_more_io
1190  *         =============>    g          fBAedc
1191  *                                           |
1192  *                                           +--> dequeue for IO
1193  */
1194 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
1195 {
1196         int moved;
1197 
1198         assert_spin_locked(&wb->list_lock);
1199         list_splice_init(&wb->b_more_io, &wb->b_io);
1200         moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work);
1201         moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io,
1202                                      EXPIRE_DIRTY_ATIME, work);
1203         if (moved)
1204                 wb_io_lists_populated(wb);
1205         trace_writeback_queue_io(wb, work, moved);
1206 }
1207 
1208 static int write_inode(struct inode *inode, struct writeback_control *wbc)
1209 {
1210         int ret;
1211 
1212         if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
1213                 trace_writeback_write_inode_start(inode, wbc);
1214                 ret = inode->i_sb->s_op->write_inode(inode, wbc);
1215                 trace_writeback_write_inode(inode, wbc);
1216                 return ret;
1217         }
1218         return 0;
1219 }
1220 
1221 /*
1222  * Wait for writeback on an inode to complete. Called with i_lock held.
1223  * Caller must make sure inode cannot go away when we drop i_lock.
1224  */
1225 static void __inode_wait_for_writeback(struct inode *inode)
1226         __releases(inode->i_lock)
1227         __acquires(inode->i_lock)
1228 {
1229         DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
1230         wait_queue_head_t *wqh;
1231 
1232         wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
1233         while (inode->i_state & I_SYNC) {
1234                 spin_unlock(&inode->i_lock);
1235                 __wait_on_bit(wqh, &wq, bit_wait,
1236                               TASK_UNINTERRUPTIBLE);
1237                 spin_lock(&inode->i_lock);
1238         }
1239 }
1240 
1241 /*
1242  * Wait for writeback on an inode to complete. Caller must have inode pinned.
1243  */
1244 void inode_wait_for_writeback(struct inode *inode)
1245 {
1246         spin_lock(&inode->i_lock);
1247         __inode_wait_for_writeback(inode);
1248         spin_unlock(&inode->i_lock);
1249 }
1250 
1251 /*
1252  * Sleep until I_SYNC is cleared. This function must be called with i_lock
1253  * held and drops it. It is aimed for callers not holding any inode reference
1254  * so once i_lock is dropped, inode can go away.
1255  */
1256 static void inode_sleep_on_writeback(struct inode *inode)
1257         __releases(inode->i_lock)
1258 {
1259         DEFINE_WAIT(wait);
1260         wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
1261         int sleep;
1262 
1263         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1264         sleep = inode->i_state & I_SYNC;
1265         spin_unlock(&inode->i_lock);
1266         if (sleep)
1267                 schedule();
1268         finish_wait(wqh, &wait);
1269 }
1270 
1271 /*
1272  * Find proper writeback list for the inode depending on its current state and
1273  * possibly also change of its state while we were doing writeback.  Here we
1274  * handle things such as livelock prevention or fairness of writeback among
1275  * inodes. This function can be called only by flusher thread - noone else
1276  * processes all inodes in writeback lists and requeueing inodes behind flusher
1277  * thread's back can have unexpected consequences.
1278  */
1279 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
1280                           struct writeback_control *wbc)
1281 {
1282         if (inode->i_state & I_FREEING)
1283                 return;
1284 
1285         /*
1286          * Sync livelock prevention. Each inode is tagged and synced in one
1287          * shot. If still dirty, it will be redirty_tail()'ed below.  Update
1288          * the dirty time to prevent enqueue and sync it again.
1289          */
1290         if ((inode->i_state & I_DIRTY) &&
1291             (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
1292                 inode->dirtied_when = jiffies;
1293 
1294         if (wbc->pages_skipped) {
1295                 /*
1296                  * writeback is not making progress due to locked
1297                  * buffers. Skip this inode for now.
1298                  */
1299                 redirty_tail(inode, wb);
1300                 return;
1301         }
1302 
1303         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
1304                 /*
1305                  * We didn't write back all the pages.  nfs_writepages()
1306                  * sometimes bales out without doing anything.
1307                  */
1308                 if (wbc->nr_to_write <= 0) {
1309                         /* Slice used up. Queue for next turn. */
1310                         requeue_io(inode, wb);
1311                 } else {
1312                         /*
1313                          * Writeback blocked by something other than
1314                          * congestion. Delay the inode for some time to
1315                          * avoid spinning on the CPU (100% iowait)
1316                          * retrying writeback of the dirty page/inode
1317                          * that cannot be performed immediately.
1318                          */
1319                         redirty_tail(inode, wb);
1320                 }
1321         } else if (inode->i_state & I_DIRTY) {
1322                 /*
1323                  * Filesystems can dirty the inode during writeback operations,
1324                  * such as delayed allocation during submission or metadata
1325                  * updates after data IO completion.
1326                  */
1327                 redirty_tail(inode, wb);
1328         } else if (inode->i_state & I_DIRTY_TIME) {
1329                 inode->dirtied_when = jiffies;
1330                 inode_io_list_move_locked(inode, wb, &wb->b_dirty_time);
1331         } else {
1332                 /* The inode is clean. Remove from writeback lists. */
1333                 inode_io_list_del_locked(inode, wb);
1334         }
1335 }
1336 
1337 /*
1338  * Write out an inode and its dirty pages. Do not update the writeback list
1339  * linkage. That is left to the caller. The caller is also responsible for
1340  * setting I_SYNC flag and calling inode_sync_complete() to clear it.
1341  */
1342 static int
1343 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
1344 {
1345         struct address_space *mapping = inode->i_mapping;
1346         long nr_to_write = wbc->nr_to_write;
1347         unsigned dirty;
1348         int ret;
1349 
1350         WARN_ON(!(inode->i_state & I_SYNC));
1351 
1352         trace_writeback_single_inode_start(inode, wbc, nr_to_write);
1353 
1354         ret = do_writepages(mapping, wbc);
1355 
1356         /*
1357          * Make sure to wait on the data before writing out the metadata.
1358          * This is important for filesystems that modify metadata on data
1359          * I/O completion. We don't do it for sync(2) writeback because it has a
1360          * separate, external IO completion path and ->sync_fs for guaranteeing
1361          * inode metadata is written back correctly.
1362          */
1363         if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
1364                 int err = filemap_fdatawait(mapping);
1365                 if (ret == 0)
1366                         ret = err;
1367         }
1368 
1369         /*
1370          * Some filesystems may redirty the inode during the writeback
1371          * due to delalloc, clear dirty metadata flags right before
1372          * write_inode()
1373          */
1374         spin_lock(&inode->i_lock);
1375 
1376         dirty = inode->i_state & I_DIRTY;
1377         if (inode->i_state & I_DIRTY_TIME) {
1378                 if ((dirty & I_DIRTY_INODE) ||
1379                     wbc->sync_mode == WB_SYNC_ALL ||
1380                     unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
1381                     unlikely(time_after(jiffies,
1382                                         (inode->dirtied_time_when +
1383                                          dirtytime_expire_interval * HZ)))) {
1384                         dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
1385                         trace_writeback_lazytime(inode);
1386                 }
1387         } else
1388                 inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
1389         inode->i_state &= ~dirty;
1390 
1391         /*
1392          * Paired with smp_mb() in __mark_inode_dirty().  This allows
1393          * __mark_inode_dirty() to test i_state without grabbing i_lock -
1394          * either they see the I_DIRTY bits cleared or we see the dirtied
1395          * inode.
1396          *
1397          * I_DIRTY_PAGES is always cleared together above even if @mapping
1398          * still has dirty pages.  The flag is reinstated after smp_mb() if
1399          * necessary.  This guarantees that either __mark_inode_dirty()
1400          * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
1401          */
1402         smp_mb();
1403 
1404         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1405                 inode->i_state |= I_DIRTY_PAGES;
1406 
1407         spin_unlock(&inode->i_lock);
1408 
1409         if (dirty & I_DIRTY_TIME)
1410                 mark_inode_dirty_sync(inode);
1411         /* Don't write the inode if only I_DIRTY_PAGES was set */
1412         if (dirty & ~I_DIRTY_PAGES) {
1413                 int err = write_inode(inode, wbc);
1414                 if (ret == 0)
1415                         ret = err;
1416         }
1417         trace_writeback_single_inode(inode, wbc, nr_to_write);
1418         return ret;
1419 }
1420 
1421 /*
1422  * Write out an inode's dirty pages. Either the caller has an active reference
1423  * on the inode or the inode has I_WILL_FREE set.
1424  *
1425  * This function is designed to be called for writing back one inode which
1426  * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
1427  * and does more profound writeback list handling in writeback_sb_inodes().
1428  */
1429 static int writeback_single_inode(struct inode *inode,
1430                                   struct writeback_control *wbc)
1431 {
1432         struct bdi_writeback *wb;
1433         int ret = 0;
1434 
1435         spin_lock(&inode->i_lock);
1436         if (!atomic_read(&inode->i_count))
1437                 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
1438         else
1439                 WARN_ON(inode->i_state & I_WILL_FREE);
1440 
1441         if (inode->i_state & I_SYNC) {
1442                 if (wbc->sync_mode != WB_SYNC_ALL)
1443                         goto out;
1444                 /*
1445                  * It's a data-integrity sync. We must wait. Since callers hold
1446                  * inode reference or inode has I_WILL_FREE set, it cannot go
1447                  * away under us.
1448                  */
1449                 __inode_wait_for_writeback(inode);
1450         }
1451         WARN_ON(inode->i_state & I_SYNC);
1452         /*
1453          * Skip inode if it is clean and we have no outstanding writeback in
1454          * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
1455          * function since flusher thread may be doing for example sync in
1456          * parallel and if we move the inode, it could get skipped. So here we
1457          * make sure inode is on some writeback list and leave it there unless
1458          * we have completely cleaned the inode.
1459          */
1460         if (!(inode->i_state & I_DIRTY_ALL) &&
1461             (wbc->sync_mode != WB_SYNC_ALL ||
1462              !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
1463                 goto out;
1464         inode->i_state |= I_SYNC;
1465         wbc_attach_and_unlock_inode(wbc, inode);
1466 
1467         ret = __writeback_single_inode(inode, wbc);
1468 
1469         wbc_detach_inode(wbc);
1470 
1471         wb = inode_to_wb_and_lock_list(inode);
1472         spin_lock(&inode->i_lock);
1473         /*
1474          * If inode is clean, remove it from writeback lists. Otherwise don't
1475          * touch it. See comment above for explanation.
1476          */
1477         if (!(inode->i_state & I_DIRTY_ALL))
1478                 inode_io_list_del_locked(inode, wb);
1479         spin_unlock(&wb->list_lock);
1480         inode_sync_complete(inode);
1481 out:
1482         spin_unlock(&inode->i_lock);
1483         return ret;
1484 }
1485 
1486 static long writeback_chunk_size(struct bdi_writeback *wb,
1487                                  struct wb_writeback_work *work)
1488 {
1489         long pages;
1490 
1491         /*
1492          * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
1493          * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
1494          * here avoids calling into writeback_inodes_wb() more than once.
1495          *
1496          * The intended call sequence for WB_SYNC_ALL writeback is:
1497          *
1498          *      wb_writeback()
1499          *          writeback_sb_inodes()       <== called only once
1500          *              write_cache_pages()     <== called once for each inode
1501          *                   (quickly) tag currently dirty pages
1502          *                   (maybe slowly) sync all tagged pages
1503          */
1504         if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
1505                 pages = LONG_MAX;
1506         else {
1507                 pages = min(wb->avg_write_bandwidth / 2,
1508                             global_wb_domain.dirty_limit / DIRTY_SCOPE);
1509                 pages = min(pages, work->nr_pages);
1510                 pages = round_down(pages + MIN_WRITEBACK_PAGES,
1511                                    MIN_WRITEBACK_PAGES);
1512         }
1513 
1514         return pages;
1515 }
1516 
1517 /*
1518  * Write a portion of b_io inodes which belong to @sb.
1519  *
1520  * Return the number of pages and/or inodes written.
1521  *
1522  * NOTE! This is called with wb->list_lock held, and will
1523  * unlock and relock that for each inode it ends up doing
1524  * IO for.
1525  */
1526 static long writeback_sb_inodes(struct super_block *sb,
1527                                 struct bdi_writeback *wb,
1528                                 struct wb_writeback_work *work)
1529 {
1530         struct writeback_control wbc = {
1531                 .sync_mode              = work->sync_mode,
1532                 .tagged_writepages      = work->tagged_writepages,
1533                 .for_kupdate            = work->for_kupdate,
1534                 .for_background         = work->for_background,
1535                 .for_sync               = work->for_sync,
1536                 .range_cyclic           = work->range_cyclic,
1537                 .range_start            = 0,
1538                 .range_end              = LLONG_MAX,
1539         };
1540         unsigned long start_time = jiffies;
1541         long write_chunk;
1542         long wrote = 0;  /* count both pages and inodes */
1543 
1544         while (!list_empty(&wb->b_io)) {
1545                 struct inode *inode = wb_inode(wb->b_io.prev);
1546                 struct bdi_writeback *tmp_wb;
1547 
1548                 if (inode->i_sb != sb) {
1549                         if (work->sb) {
1550                                 /*
1551                                  * We only want to write back data for this
1552                                  * superblock, move all inodes not belonging
1553                                  * to it back onto the dirty list.
1554                                  */
1555                                 redirty_tail(inode, wb);
1556                                 continue;
1557                         }
1558 
1559                         /*
1560                          * The inode belongs to a different superblock.
1561                          * Bounce back to the caller to unpin this and
1562                          * pin the next superblock.
1563                          */
1564                         break;
1565                 }
1566 
1567                 /*
1568                  * Don't bother with new inodes or inodes being freed, first
1569                  * kind does not need periodic writeout yet, and for the latter
1570                  * kind writeout is handled by the freer.
1571                  */
1572                 spin_lock(&inode->i_lock);
1573                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
1574                         spin_unlock(&inode->i_lock);
1575                         redirty_tail(inode, wb);
1576                         continue;
1577                 }
1578                 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
1579                         /*
1580                          * If this inode is locked for writeback and we are not
1581                          * doing writeback-for-data-integrity, move it to
1582                          * b_more_io so that writeback can proceed with the
1583                          * other inodes on s_io.
1584                          *
1585                          * We'll have another go at writing back this inode
1586                          * when we completed a full scan of b_io.
1587                          */
1588                         spin_unlock(&inode->i_lock);
1589                         requeue_io(inode, wb);
1590                         trace_writeback_sb_inodes_requeue(inode);
1591                         continue;
1592                 }
1593                 spin_unlock(&wb->list_lock);
1594 
1595                 /*
1596                  * We already requeued the inode if it had I_SYNC set and we
1597                  * are doing WB_SYNC_NONE writeback. So this catches only the
1598                  * WB_SYNC_ALL case.
1599                  */
1600                 if (inode->i_state & I_SYNC) {
1601                         /* Wait for I_SYNC. This function drops i_lock... */
1602                         inode_sleep_on_writeback(inode);
1603                         /* Inode may be gone, start again */
1604                         spin_lock(&wb->list_lock);
1605                         continue;
1606                 }
1607                 inode->i_state |= I_SYNC;
1608                 wbc_attach_and_unlock_inode(&wbc, inode);
1609 
1610                 write_chunk = writeback_chunk_size(wb, work);
1611                 wbc.nr_to_write = write_chunk;
1612                 wbc.pages_skipped = 0;
1613 
1614                 /*
1615                  * We use I_SYNC to pin the inode in memory. While it is set
1616                  * evict_inode() will wait so the inode cannot be freed.
1617                  */
1618                 __writeback_single_inode(inode, &wbc);
1619 
1620                 wbc_detach_inode(&wbc);
1621                 work->nr_pages -= write_chunk - wbc.nr_to_write;
1622                 wrote += write_chunk - wbc.nr_to_write;
1623 
1624                 if (need_resched()) {
1625                         /*
1626                          * We're trying to balance between building up a nice
1627                          * long list of IOs to improve our merge rate, and
1628                          * getting those IOs out quickly for anyone throttling
1629                          * in balance_dirty_pages().  cond_resched() doesn't
1630                          * unplug, so get our IOs out the door before we
1631                          * give up the CPU.
1632                          */
1633                         blk_flush_plug(current);
1634                         cond_resched();
1635                 }
1636 
1637                 /*
1638                  * Requeue @inode if still dirty.  Be careful as @inode may
1639                  * have been switched to another wb in the meantime.
1640                  */
1641                 tmp_wb = inode_to_wb_and_lock_list(inode);
1642                 spin_lock(&inode->i_lock);
1643                 if (!(inode->i_state & I_DIRTY_ALL))
1644                         wrote++;
1645                 requeue_inode(inode, tmp_wb, &wbc);
1646                 inode_sync_complete(inode);
1647                 spin_unlock(&inode->i_lock);
1648 
1649                 if (unlikely(tmp_wb != wb)) {
1650                         spin_unlock(&tmp_wb->list_lock);
1651                         spin_lock(&wb->list_lock);
1652                 }
1653 
1654                 /*
1655                  * bail out to wb_writeback() often enough to check
1656                  * background threshold and other termination conditions.
1657                  */
1658                 if (wrote) {
1659                         if (time_is_before_jiffies(start_time + HZ / 10UL))
1660                                 break;
1661                         if (work->nr_pages <= 0)
1662                                 break;
1663                 }
1664         }
1665         return wrote;
1666 }
1667 
1668 static long __writeback_inodes_wb(struct bdi_writeback *wb,
1669                                   struct wb_writeback_work *work)
1670 {
1671         unsigned long start_time = jiffies;
1672         long wrote = 0;
1673 
1674         while (!list_empty(&wb->b_io)) {
1675                 struct inode *inode = wb_inode(wb->b_io.prev);
1676                 struct super_block *sb = inode->i_sb;
1677 
1678                 if (!trylock_super(sb)) {
1679                         /*
1680                          * trylock_super() may fail consistently due to
1681                          * s_umount being grabbed by someone else. Don't use
1682                          * requeue_io() to avoid busy retrying the inode/sb.
1683                          */
1684                         redirty_tail(inode, wb);
1685                         continue;
1686                 }
1687                 wrote += writeback_sb_inodes(sb, wb, work);
1688                 up_read(&sb->s_umount);
1689 
1690                 /* refer to the same tests at the end of writeback_sb_inodes */
1691                 if (wrote) {
1692                         if (time_is_before_jiffies(start_time + HZ / 10UL))
1693                                 break;
1694                         if (work->nr_pages <= 0)
1695                                 break;
1696                 }
1697         }
1698         /* Leave any unwritten inodes on b_io */
1699         return wrote;
1700 }
1701 
1702 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
1703                                 enum wb_reason reason)
1704 {
1705         struct wb_writeback_work work = {
1706                 .nr_pages       = nr_pages,
1707                 .sync_mode      = WB_SYNC_NONE,
1708                 .range_cyclic   = 1,
1709                 .reason         = reason,
1710         };
1711         struct blk_plug plug;
1712 
1713         blk_start_plug(&plug);
1714         spin_lock(&wb->list_lock);
1715         if (list_empty(&wb->b_io))
1716                 queue_io(wb, &work);
1717         __writeback_inodes_wb(wb, &work);
1718         spin_unlock(&wb->list_lock);
1719         blk_finish_plug(&plug);
1720 
1721         return nr_pages - work.nr_pages;
1722 }
1723 
1724 /*
1725  * Explicit flushing or periodic writeback of "old" data.
1726  *
1727  * Define "old": the first time one of an inode's pages is dirtied, we mark the
1728  * dirtying-time in the inode's address_space.  So this periodic writeback code
1729  * just walks the superblock inode list, writing back any inodes which are
1730  * older than a specific point in time.
1731  *
1732  * Try to run once per dirty_writeback_interval.  But if a writeback event
1733  * takes longer than a dirty_writeback_interval interval, then leave a
1734  * one-second gap.
1735  *
1736  * older_than_this takes precedence over nr_to_write.  So we'll only write back
1737  * all dirty pages if they are all attached to "old" mappings.
1738  */
1739 static long wb_writeback(struct bdi_writeback *wb,
1740                          struct wb_writeback_work *work)
1741 {
1742         unsigned long wb_start = jiffies;
1743         long nr_pages = work->nr_pages;
1744         unsigned long oldest_jif;
1745         struct inode *inode;
1746         long progress;
1747         struct blk_plug plug;
1748 
1749         oldest_jif = jiffies;
1750         work->older_than_this = &oldest_jif;
1751 
1752         blk_start_plug(&plug);
1753         spin_lock(&wb->list_lock);
1754         for (;;) {
1755                 /*
1756                  * Stop writeback when nr_pages has been consumed
1757                  */
1758                 if (work->nr_pages <= 0)
1759                         break;
1760 
1761                 /*
1762                  * Background writeout and kupdate-style writeback may
1763                  * run forever. Stop them if there is other work to do
1764                  * so that e.g. sync can proceed. They'll be restarted
1765                  * after the other works are all done.
1766                  */
1767                 if ((work->for_background || work->for_kupdate) &&
1768                     !list_empty(&wb->work_list))
1769                         break;
1770 
1771                 /*
1772                  * For background writeout, stop when we are below the
1773                  * background dirty threshold
1774                  */
1775                 if (work->for_background && !wb_over_bg_thresh(wb))
1776                         break;
1777 
1778                 /*
1779                  * Kupdate and background works are special and we want to
1780                  * include all inodes that need writing. Livelock avoidance is
1781                  * handled by these works yielding to any other work so we are
1782                  * safe.
1783                  */
1784                 if (work->for_kupdate) {
1785                         oldest_jif = jiffies -
1786                                 msecs_to_jiffies(dirty_expire_interval * 10);
1787                 } else if (work->for_background)
1788                         oldest_jif = jiffies;
1789 
1790                 trace_writeback_start(wb, work);
1791                 if (list_empty(&wb->b_io))
1792                         queue_io(wb, work);
1793                 if (work->sb)
1794                         progress = writeback_sb_inodes(work->sb, wb, work);
1795                 else
1796                         progress = __writeback_inodes_wb(wb, work);
1797                 trace_writeback_written(wb, work);
1798 
1799                 wb_update_bandwidth(wb, wb_start);
1800 
1801                 /*
1802                  * Did we write something? Try for more
1803                  *
1804                  * Dirty inodes are moved to b_io for writeback in batches.
1805                  * The completion of the current batch does not necessarily
1806                  * mean the overall work is done. So we keep looping as long
1807                  * as made some progress on cleaning pages or inodes.
1808                  */
1809                 if (progress)
1810                         continue;
1811                 /*
1812                  * No more inodes for IO, bail
1813                  */
1814                 if (list_empty(&wb->b_more_io))
1815                         break;
1816                 /*
1817                  * Nothing written. Wait for some inode to
1818                  * become available for writeback. Otherwise
1819                  * we'll just busyloop.
1820                  */
1821                 trace_writeback_wait(wb, work);
1822                 inode = wb_inode(wb->b_more_io.prev);
1823                 spin_lock(&inode->i_lock);
1824                 spin_unlock(&wb->list_lock);
1825                 /* This function drops i_lock... */
1826                 inode_sleep_on_writeback(inode);
1827                 spin_lock(&wb->list_lock);
1828         }
1829         spin_unlock(&wb->list_lock);
1830         blk_finish_plug(&plug);
1831 
1832         return nr_pages - work->nr_pages;
1833 }
1834 
1835 /*
1836  * Return the next wb_writeback_work struct that hasn't been processed yet.
1837  */
1838 static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb)
1839 {
1840         struct wb_writeback_work *work = NULL;
1841 
1842         spin_lock_bh(&wb->work_lock);
1843         if (!list_empty(&wb->work_list)) {
1844                 work = list_entry(wb->work_list.next,
1845                                   struct wb_writeback_work, list);
1846                 list_del_init(&work->list);
1847         }
1848         spin_unlock_bh(&wb->work_lock);
1849         return work;
1850 }
1851 
1852 static long wb_check_background_flush(struct bdi_writeback *wb)
1853 {
1854         if (wb_over_bg_thresh(wb)) {
1855 
1856                 struct wb_writeback_work work = {
1857                         .nr_pages       = LONG_MAX,
1858                         .sync_mode      = WB_SYNC_NONE,
1859                         .for_background = 1,
1860                         .range_cyclic   = 1,
1861                         .reason         = WB_REASON_BACKGROUND,
1862                 };
1863 
1864                 return wb_writeback(wb, &work);
1865         }
1866 
1867         return 0;
1868 }
1869 
1870 static long wb_check_old_data_flush(struct bdi_writeback *wb)
1871 {
1872         unsigned long expired;
1873         long nr_pages;
1874 
1875         /*
1876          * When set to zero, disable periodic writeback
1877          */
1878         if (!dirty_writeback_interval)
1879                 return 0;
1880 
1881         expired = wb->last_old_flush +
1882                         msecs_to_jiffies(dirty_writeback_interval * 10);
1883         if (time_before(jiffies, expired))
1884                 return 0;
1885 
1886         wb->last_old_flush = jiffies;
1887         nr_pages = get_nr_dirty_pages();
1888 
1889         if (nr_pages) {
1890                 struct wb_writeback_work work = {
1891                         .nr_pages       = nr_pages,
1892                         .sync_mode      = WB_SYNC_NONE,
1893                         .for_kupdate    = 1,
1894                         .range_cyclic   = 1,
1895                         .reason         = WB_REASON_PERIODIC,
1896                 };
1897 
1898                 return wb_writeback(wb, &work);
1899         }
1900 
1901         return 0;
1902 }
1903 
1904 static long wb_check_start_all(struct bdi_writeback *wb)
1905 {
1906         long nr_pages;
1907 
1908         if (!test_bit(WB_start_all, &wb->state))
1909                 return 0;
1910 
1911         nr_pages = get_nr_dirty_pages();
1912         if (nr_pages) {
1913                 struct wb_writeback_work work = {
1914                         .nr_pages       = wb_split_bdi_pages(wb, nr_pages),
1915                         .sync_mode      = WB_SYNC_NONE,
1916                         .range_cyclic   = 1,
1917                         .reason         = wb->start_all_reason,
1918                 };
1919 
1920                 nr_pages = wb_writeback(wb, &work);
1921         }
1922 
1923         clear_bit(WB_start_all, &wb->state);
1924         return nr_pages;
1925 }
1926 
1927 
1928 /*
1929  * Retrieve work items and do the writeback they describe
1930  */
1931 static long wb_do_writeback(struct bdi_writeback *wb)
1932 {
1933         struct wb_writeback_work *work;
1934         long wrote = 0;
1935 
1936         set_bit(WB_writeback_running, &wb->state);
1937         while ((work = get_next_work_item(wb)) != NULL) {
1938                 trace_writeback_exec(wb, work);
1939                 wrote += wb_writeback(wb, work);
1940                 finish_writeback_work(wb, work);
1941         }
1942 
1943         /*
1944          * Check for a flush-everything request
1945          */
1946         wrote += wb_check_start_all(wb);
1947 
1948         /*
1949          * Check for periodic writeback, kupdated() style
1950          */
1951         wrote += wb_check_old_data_flush(wb);
1952         wrote += wb_check_background_flush(wb);
1953         clear_bit(WB_writeback_running, &wb->state);
1954 
1955         return wrote;
1956 }
1957 
1958 /*
1959  * Handle writeback of dirty data for the device backed by this bdi. Also
1960  * reschedules periodically and does kupdated style flushing.
1961  */
1962 void wb_workfn(struct work_struct *work)
1963 {
1964         struct bdi_writeback *wb = container_of(to_delayed_work(work),
1965                                                 struct bdi_writeback, dwork);
1966         long pages_written;
1967 
1968         set_worker_desc("flush-%s", dev_name(wb->bdi->dev));
1969         current->flags |= PF_SWAPWRITE;
1970 
1971         if (likely(!current_is_workqueue_rescuer() ||
1972                    !test_bit(WB_registered, &wb->state))) {
1973                 /*
1974                  * The normal path.  Keep writing back @wb until its
1975                  * work_list is empty.  Note that this path is also taken
1976                  * if @wb is shutting down even when we're running off the
1977                  * rescuer as work_list needs to be drained.
1978                  */
1979                 do {
1980                         pages_written = wb_do_writeback(wb);
1981                         trace_writeback_pages_written(pages_written);
1982                 } while (!list_empty(&wb->work_list));
1983         } else {
1984                 /*
1985                  * bdi_wq can't get enough workers and we're running off
1986                  * the emergency worker.  Don't hog it.  Hopefully, 1024 is
1987                  * enough for efficient IO.
1988                  */
1989                 pages_written = writeback_inodes_wb(wb, 1024,
1990                                                     WB_REASON_FORKER_THREAD);
1991                 trace_writeback_pages_written(pages_written);
1992         }
1993 
1994         if (!list_empty(&wb->work_list))
1995                 wb_wakeup(wb);
1996         else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1997                 wb_wakeup_delayed(wb);
1998 
1999         current->flags &= ~PF_SWAPWRITE;
2000 }
2001 
2002 /*
2003  * Start writeback of `nr_pages' pages on this bdi. If `nr_pages' is zero,
2004  * write back the whole world.
2005  */
2006 static void __wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
2007                                          enum wb_reason reason)
2008 {
2009         struct bdi_writeback *wb;
2010 
2011         if (!bdi_has_dirty_io(bdi))
2012                 return;
2013 
2014         list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node)
2015                 wb_start_writeback(wb, reason);
2016 }
2017 
2018 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
2019                                 enum wb_reason reason)
2020 {
2021         rcu_read_lock();
2022         __wakeup_flusher_threads_bdi(bdi, reason);
2023         rcu_read_unlock();
2024 }
2025 
2026 /*
2027  * Wakeup the flusher threads to start writeback of all currently dirty pages
2028  */
2029 void wakeup_flusher_threads(enum wb_reason reason)
2030 {
2031         struct backing_dev_info *bdi;
2032 
2033         /*
2034          * If we are expecting writeback progress we must submit plugged IO.
2035          */
2036         if (blk_needs_flush_plug(current))
2037                 blk_schedule_flush_plug(current);
2038 
2039         rcu_read_lock();
2040         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list)
2041                 __wakeup_flusher_threads_bdi(bdi, reason);
2042         rcu_read_unlock();
2043 }
2044 
2045 /*
2046  * Wake up bdi's periodically to make sure dirtytime inodes gets
2047  * written back periodically.  We deliberately do *not* check the
2048  * b_dirtytime list in wb_has_dirty_io(), since this would cause the
2049  * kernel to be constantly waking up once there are any dirtytime
2050  * inodes on the system.  So instead we define a separate delayed work
2051  * function which gets called much more rarely.  (By default, only
2052  * once every 12 hours.)
2053  *
2054  * If there is any other write activity going on in the file system,
2055  * this function won't be necessary.  But if the only thing that has
2056  * happened on the file system is a dirtytime inode caused by an atime
2057  * update, we need this infrastructure below to make sure that inode
2058  * eventually gets pushed out to disk.
2059  */
2060 static void wakeup_dirtytime_writeback(struct work_struct *w);
2061 static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback);
2062 
2063 static void wakeup_dirtytime_writeback(struct work_struct *w)
2064 {
2065         struct backing_dev_info *bdi;
2066 
2067         rcu_read_lock();
2068         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
2069                 struct bdi_writeback *wb;
2070 
2071                 list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node)
2072                         if (!list_empty(&wb->b_dirty_time))
2073                                 wb_wakeup(wb);
2074         }
2075         rcu_read_unlock();
2076         schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
2077 }
2078 
2079 static int __init start_dirtytime_writeback(void)
2080 {
2081         schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
2082         return 0;
2083 }
2084 __initcall(start_dirtytime_writeback);
2085 
2086 int dirtytime_interval_handler(struct ctl_table *table, int write,
2087                                void __user *buffer, size_t *lenp, loff_t *ppos)
2088 {
2089         int ret;
2090 
2091         ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2092         if (ret == 0 && write)
2093                 mod_delayed_work(system_wq, &dirtytime_work, 0);
2094         return ret;
2095 }
2096 
2097 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
2098 {
2099         if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
2100                 struct dentry *dentry;
2101                 const char *name = "?";
2102 
2103                 dentry = d_find_alias(inode);
2104                 if (dentry) {
2105                         spin_lock(&dentry->d_lock);
2106                         name = (const char *) dentry->d_name.name;
2107                 }
2108                 printk(KERN_DEBUG
2109                        "%s(%d): dirtied inode %lu (%s) on %s\n",
2110                        current->comm, task_pid_nr(current), inode->i_ino,
2111                        name, inode->i_sb->s_id);
2112                 if (dentry) {
2113                         spin_unlock(&dentry->d_lock);
2114                         dput(dentry);
2115                 }
2116         }
2117 }
2118 
2119 /**
2120  * __mark_inode_dirty - internal function
2121  *
2122  * @inode: inode to mark
2123  * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
2124  *
2125  * Mark an inode as dirty. Callers should use mark_inode_dirty or
2126  * mark_inode_dirty_sync.
2127  *
2128  * Put the inode on the super block's dirty list.
2129  *
2130  * CAREFUL! We mark it dirty unconditionally, but move it onto the
2131  * dirty list only if it is hashed or if it refers to a blockdev.
2132  * If it was not hashed, it will never be added to the dirty list
2133  * even if it is later hashed, as it will have been marked dirty already.
2134  *
2135  * In short, make sure you hash any inodes _before_ you start marking
2136  * them dirty.
2137  *
2138  * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
2139  * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
2140  * the kernel-internal blockdev inode represents the dirtying time of the
2141  * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
2142  * page->mapping->host, so the page-dirtying time is recorded in the internal
2143  * blockdev inode.
2144  */
2145 void __mark_inode_dirty(struct inode *inode, int flags)
2146 {
2147         struct super_block *sb = inode->i_sb;
2148         int dirtytime;
2149 
2150         trace_writeback_mark_inode_dirty(inode, flags);
2151 
2152         /*
2153          * Don't do this for I_DIRTY_PAGES - that doesn't actually
2154          * dirty the inode itself
2155          */
2156         if (flags & (I_DIRTY_INODE | I_DIRTY_TIME)) {
2157                 trace_writeback_dirty_inode_start(inode, flags);
2158 
2159                 if (sb->s_op->dirty_inode)
2160                         sb->s_op->dirty_inode(inode, flags);
2161 
2162                 trace_writeback_dirty_inode(inode, flags);
2163         }
2164         if (flags & I_DIRTY_INODE)
2165                 flags &= ~I_DIRTY_TIME;
2166         dirtytime = flags & I_DIRTY_TIME;
2167 
2168         /*
2169          * Paired with smp_mb() in __writeback_single_inode() for the
2170          * following lockless i_state test.  See there for details.
2171          */
2172         smp_mb();
2173 
2174         if (((inode->i_state & flags) == flags) ||
2175             (dirtytime && (inode->i_state & I_DIRTY_INODE)))
2176                 return;
2177 
2178         if (unlikely(block_dump))
2179                 block_dump___mark_inode_dirty(inode);
2180 
2181         spin_lock(&inode->i_lock);
2182         if (dirtytime && (inode->i_state & I_DIRTY_INODE))
2183                 goto out_unlock_inode;
2184         if ((inode->i_state & flags) != flags) {
2185                 const int was_dirty = inode->i_state & I_DIRTY;
2186 
2187                 inode_attach_wb(inode, NULL);
2188 
2189                 if (flags & I_DIRTY_INODE)
2190                         inode->i_state &= ~I_DIRTY_TIME;
2191                 inode->i_state |= flags;
2192 
2193                 /*
2194                  * If the inode is being synced, just update its dirty state.
2195                  * The unlocker will place the inode on the appropriate
2196                  * superblock list, based upon its state.
2197                  */
2198                 if (inode->i_state & I_SYNC)
2199                         goto out_unlock_inode;
2200 
2201                 /*
2202                  * Only add valid (hashed) inodes to the superblock's
2203                  * dirty list.  Add blockdev inodes as well.
2204                  */
2205                 if (!S_ISBLK(inode->i_mode)) {
2206                         if (inode_unhashed(inode))
2207                                 goto out_unlock_inode;
2208                 }
2209                 if (inode->i_state & I_FREEING)
2210                         goto out_unlock_inode;
2211 
2212                 /*
2213                  * If the inode was already on b_dirty/b_io/b_more_io, don't
2214                  * reposition it (that would break b_dirty time-ordering).
2215                  */
2216                 if (!was_dirty) {
2217                         struct bdi_writeback *wb;
2218                         struct list_head *dirty_list;
2219                         bool wakeup_bdi = false;
2220 
2221                         wb = locked_inode_to_wb_and_lock_list(inode);
2222 
2223                         WARN(bdi_cap_writeback_dirty(wb->bdi) &&
2224                              !test_bit(WB_registered, &wb->state),
2225                              "bdi-%s not registered\n", wb->bdi->name);
2226 
2227                         inode->dirtied_when = jiffies;
2228                         if (dirtytime)
2229                                 inode->dirtied_time_when = jiffies;
2230 
2231                         if (inode->i_state & I_DIRTY)
2232                                 dirty_list = &wb->b_dirty;
2233                         else
2234                                 dirty_list = &wb->b_dirty_time;
2235 
2236                         wakeup_bdi = inode_io_list_move_locked(inode, wb,
2237                                                                dirty_list);
2238 
2239                         spin_unlock(&wb->list_lock);
2240                         trace_writeback_dirty_inode_enqueue(inode);
2241 
2242                         /*
2243                          * If this is the first dirty inode for this bdi,
2244                          * we have to wake-up the corresponding bdi thread
2245                          * to make sure background write-back happens
2246                          * later.
2247                          */
2248                         if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi)
2249                                 wb_wakeup_delayed(wb);
2250                         return;
2251                 }
2252         }
2253 out_unlock_inode:
2254         spin_unlock(&inode->i_lock);
2255 }
2256 EXPORT_SYMBOL(__mark_inode_dirty);
2257 
2258 /*
2259  * The @s_sync_lock is used to serialise concurrent sync operations
2260  * to avoid lock contention problems with concurrent wait_sb_inodes() calls.
2261  * Concurrent callers will block on the s_sync_lock rather than doing contending
2262  * walks. The queueing maintains sync(2) required behaviour as all the IO that
2263  * has been issued up to the time this function is enter is guaranteed to be
2264  * completed by the time we have gained the lock and waited for all IO that is
2265  * in progress regardless of the order callers are granted the lock.
2266  */
2267 static void wait_sb_inodes(struct super_block *sb)
2268 {
2269         LIST_HEAD(sync_list);
2270 
2271         /*
2272          * We need to be protected against the filesystem going from
2273          * r/o to r/w or vice versa.
2274          */
2275         WARN_ON(!rwsem_is_locked(&sb->s_umount));
2276 
2277         mutex_lock(&sb->s_sync_lock);
2278 
2279         /*
2280          * Splice the writeback list onto a temporary list to avoid waiting on
2281          * inodes that have started writeback after this point.
2282          *
2283          * Use rcu_read_lock() to keep the inodes around until we have a
2284          * reference. s_inode_wblist_lock protects sb->s_inodes_wb as well as
2285          * the local list because inodes can be dropped from either by writeback
2286          * completion.
2287          */
2288         rcu_read_lock();
2289         spin_lock_irq(&sb->s_inode_wblist_lock);
2290         list_splice_init(&sb->s_inodes_wb, &sync_list);
2291 
2292         /*
2293          * Data integrity sync. Must wait for all pages under writeback, because
2294          * there may have been pages dirtied before our sync call, but which had
2295          * writeout started before we write it out.  In which case, the inode
2296          * may not be on the dirty list, but we still have to wait for that
2297          * writeout.
2298          */
2299         while (!list_empty(&sync_list)) {
2300                 struct inode *inode = list_first_entry(&sync_list, struct inode,
2301                                                        i_wb_list);
2302                 struct address_space *mapping = inode->i_mapping;
2303 
2304                 /*
2305                  * Move each inode back to the wb list before we drop the lock
2306                  * to preserve consistency between i_wb_list and the mapping
2307                  * writeback tag. Writeback completion is responsible to remove
2308                  * the inode from either list once the writeback tag is cleared.
2309                  */
2310                 list_move_tail(&inode->i_wb_list, &sb->s_inodes_wb);
2311 
2312                 /*
2313                  * The mapping can appear untagged while still on-list since we
2314                  * do not have the mapping lock. Skip it here, wb completion
2315                  * will remove it.
2316                  */
2317                 if (!mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
2318                         continue;
2319 
2320                 spin_unlock_irq(&sb->s_inode_wblist_lock);
2321 
2322                 spin_lock(&inode->i_lock);
2323                 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
2324                         spin_unlock(&inode->i_lock);
2325 
2326                         spin_lock_irq(&sb->s_inode_wblist_lock);
2327                         continue;
2328                 }
2329                 __iget(inode);
2330                 spin_unlock(&inode->i_lock);
2331                 rcu_read_unlock();
2332 
2333                 /*
2334                  * We keep the error status of individual mapping so that
2335                  * applications can catch the writeback error using fsync(2).
2336                  * See filemap_fdatawait_keep_errors() for details.
2337                  */
2338                 filemap_fdatawait_keep_errors(mapping);
2339 
2340                 cond_resched();
2341 
2342                 iput(inode);
2343 
2344                 rcu_read_lock();
2345                 spin_lock_irq(&sb->s_inode_wblist_lock);
2346         }
2347         spin_unlock_irq(&sb->s_inode_wblist_lock);
2348         rcu_read_unlock();
2349         mutex_unlock(&sb->s_sync_lock);
2350 }
2351 
2352 static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr,
2353                                      enum wb_reason reason, bool skip_if_busy)
2354 {
2355         DEFINE_WB_COMPLETION_ONSTACK(done);
2356         struct wb_writeback_work work = {
2357                 .sb                     = sb,
2358                 .sync_mode              = WB_SYNC_NONE,
2359                 .tagged_writepages      = 1,
2360                 .done                   = &done,
2361                 .nr_pages               = nr,
2362                 .reason                 = reason,
2363         };
2364         struct backing_dev_info *bdi = sb->s_bdi;
2365 
2366         if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info)
2367                 return;
2368         WARN_ON(!rwsem_is_locked(&sb->s_umount));
2369 
2370         bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy);
2371         wb_wait_for_completion(bdi, &done);
2372 }
2373 
2374 /**
2375  * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
2376  * @sb: the superblock
2377  * @nr: the number of pages to write
2378  * @reason: reason why some writeback work initiated
2379  *
2380  * Start writeback on some inodes on this super_block. No guarantees are made
2381  * on how many (if any) will be written, and this function does not wait
2382  * for IO completion of submitted IO.
2383  */
2384 void writeback_inodes_sb_nr(struct super_block *sb,
2385                             unsigned long nr,
2386                             enum wb_reason reason)
2387 {
2388         __writeback_inodes_sb_nr(sb, nr, reason, false);
2389 }
2390 EXPORT_SYMBOL(writeback_inodes_sb_nr);
2391 
2392 /**
2393  * writeback_inodes_sb  -       writeback dirty inodes from given super_block
2394  * @sb: the superblock
2395  * @reason: reason why some writeback work was initiated
2396  *
2397  * Start writeback on some inodes on this super_block. No guarantees are made
2398  * on how many (if any) will be written, and this function does not wait
2399  * for IO completion of submitted IO.
2400  */
2401 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
2402 {
2403         return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
2404 }
2405 EXPORT_SYMBOL(writeback_inodes_sb);
2406 
2407 /**
2408  * try_to_writeback_inodes_sb - try to start writeback if none underway
2409  * @sb: the superblock
2410  * @reason: reason why some writeback work was initiated
2411  *
2412  * Invoke __writeback_inodes_sb_nr if no writeback is currently underway.
2413  */
2414 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
2415 {
2416         if (!down_read_trylock(&sb->s_umount))
2417                 return;
2418 
2419         __writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason, true);
2420         up_read(&sb->s_umount);
2421 }
2422 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
2423 
2424 /**
2425  * sync_inodes_sb       -       sync sb inode pages
2426  * @sb: the superblock
2427  *
2428  * This function writes and waits on any dirty inode belonging to this
2429  * super_block.
2430  */
2431 void sync_inodes_sb(struct super_block *sb)
2432 {
2433         DEFINE_WB_COMPLETION_ONSTACK(done);
2434         struct wb_writeback_work work = {
2435                 .sb             = sb,
2436                 .sync_mode      = WB_SYNC_ALL,
2437                 .nr_pages       = LONG_MAX,
2438                 .range_cyclic   = 0,
2439                 .done           = &done,
2440                 .reason         = WB_REASON_SYNC,
2441                 .for_sync       = 1,
2442         };
2443         struct backing_dev_info *bdi = sb->s_bdi;
2444 
2445         /*
2446          * Can't skip on !bdi_has_dirty() because we should wait for !dirty
2447          * inodes under writeback and I_DIRTY_TIME inodes ignored by
2448          * bdi_has_dirty() need to be written out too.
2449          */
2450         if (bdi == &noop_backing_dev_info)
2451                 return;
2452         WARN_ON(!rwsem_is_locked(&sb->s_umount));
2453 
2454         /* protect against inode wb switch, see inode_switch_wbs_work_fn() */
2455         bdi_down_write_wb_switch_rwsem(bdi);
2456         bdi_split_work_to_wbs(bdi, &work, false);
2457         wb_wait_for_completion(bdi, &done);
2458         bdi_up_write_wb_switch_rwsem(bdi);
2459 
2460         wait_sb_inodes(sb);
2461 }
2462 EXPORT_SYMBOL(sync_inodes_sb);
2463 
2464 /**
2465  * write_inode_now      -       write an inode to disk
2466  * @inode: inode to write to disk
2467  * @sync: whether the write should be synchronous or not
2468  *
2469  * This function commits an inode to disk immediately if it is dirty. This is
2470  * primarily needed by knfsd.
2471  *
2472  * The caller must either have a ref on the inode or must have set I_WILL_FREE.
2473  */
2474 int write_inode_now(struct inode *inode, int sync)
2475 {
2476         struct writeback_control wbc = {
2477                 .nr_to_write = LONG_MAX,
2478                 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
2479                 .range_start = 0,
2480                 .range_end = LLONG_MAX,
2481         };
2482 
2483         if (!mapping_cap_writeback_dirty(inode->i_mapping))
2484                 wbc.nr_to_write = 0;
2485 
2486         might_sleep();
2487         return writeback_single_inode(inode, &wbc);
2488 }
2489 EXPORT_SYMBOL(write_inode_now);
2490 
2491 /**
2492  * sync_inode - write an inode and its pages to disk.
2493  * @inode: the inode to sync
2494  * @wbc: controls the writeback mode
2495  *
2496  * sync_inode() will write an inode and its pages to disk.  It will also
2497  * correctly update the inode on its superblock's dirty inode lists and will
2498  * update inode->i_state.
2499  *
2500  * The caller must have a ref on the inode.
2501  */
2502 int sync_inode(struct inode *inode, struct writeback_control *wbc)
2503 {
2504         return writeback_single_inode(inode, wbc);
2505 }
2506 EXPORT_SYMBOL(sync_inode);
2507 
2508 /**
2509  * sync_inode_metadata - write an inode to disk
2510  * @inode: the inode to sync
2511  * @wait: wait for I/O to complete.
2512  *
2513  * Write an inode to disk and adjust its dirty state after completion.
2514  *
2515  * Note: only writes the actual inode, no associated data or other metadata.
2516  */
2517 int sync_inode_metadata(struct inode *inode, int wait)
2518 {
2519         struct writeback_control wbc = {
2520                 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
2521                 .nr_to_write = 0, /* metadata-only */
2522         };
2523 
2524         return sync_inode(inode, &wbc);
2525 }
2526 EXPORT_SYMBOL(sync_inode_metadata);
2527 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp