Version:  2.0.40 2.2.26 2.4.37 2.6.39 3.0 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15

Linux/fs/fs-writeback.c

  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 "internal.h"
 31 
 32 /*
 33  * 4MB minimal write chunk size
 34  */
 35 #define MIN_WRITEBACK_PAGES     (4096UL >> (PAGE_CACHE_SHIFT - 10))
 36 
 37 /*
 38  * Passed into wb_writeback(), essentially a subset of writeback_control
 39  */
 40 struct wb_writeback_work {
 41         long nr_pages;
 42         struct super_block *sb;
 43         unsigned long *older_than_this;
 44         enum writeback_sync_modes sync_mode;
 45         unsigned int tagged_writepages:1;
 46         unsigned int for_kupdate:1;
 47         unsigned int range_cyclic:1;
 48         unsigned int for_background:1;
 49         unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
 50         enum wb_reason reason;          /* why was writeback initiated? */
 51 
 52         struct list_head list;          /* pending work list */
 53         struct completion *done;        /* set if the caller waits */
 54 };
 55 
 56 /**
 57  * writeback_in_progress - determine whether there is writeback in progress
 58  * @bdi: the device's backing_dev_info structure.
 59  *
 60  * Determine whether there is writeback waiting to be handled against a
 61  * backing device.
 62  */
 63 int writeback_in_progress(struct backing_dev_info *bdi)
 64 {
 65         return test_bit(BDI_writeback_running, &bdi->state);
 66 }
 67 EXPORT_SYMBOL(writeback_in_progress);
 68 
 69 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
 70 {
 71         struct super_block *sb = inode->i_sb;
 72 
 73         if (sb_is_blkdev_sb(sb))
 74                 return inode->i_mapping->backing_dev_info;
 75 
 76         return sb->s_bdi;
 77 }
 78 
 79 static inline struct inode *wb_inode(struct list_head *head)
 80 {
 81         return list_entry(head, struct inode, i_wb_list);
 82 }
 83 
 84 /*
 85  * Include the creation of the trace points after defining the
 86  * wb_writeback_work structure and inline functions so that the definition
 87  * remains local to this file.
 88  */
 89 #define CREATE_TRACE_POINTS
 90 #include <trace/events/writeback.h>
 91 
 92 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
 93 
 94 static void bdi_wakeup_thread(struct backing_dev_info *bdi)
 95 {
 96         spin_lock_bh(&bdi->wb_lock);
 97         if (test_bit(BDI_registered, &bdi->state))
 98                 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 99         spin_unlock_bh(&bdi->wb_lock);
100 }
101 
102 static void bdi_queue_work(struct backing_dev_info *bdi,
103                            struct wb_writeback_work *work)
104 {
105         trace_writeback_queue(bdi, work);
106 
107         spin_lock_bh(&bdi->wb_lock);
108         if (!test_bit(BDI_registered, &bdi->state)) {
109                 if (work->done)
110                         complete(work->done);
111                 goto out_unlock;
112         }
113         list_add_tail(&work->list, &bdi->work_list);
114         mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
115 out_unlock:
116         spin_unlock_bh(&bdi->wb_lock);
117 }
118 
119 static void
120 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
121                       bool range_cyclic, enum wb_reason reason)
122 {
123         struct wb_writeback_work *work;
124 
125         /*
126          * This is WB_SYNC_NONE writeback, so if allocation fails just
127          * wakeup the thread for old dirty data writeback
128          */
129         work = kzalloc(sizeof(*work), GFP_ATOMIC);
130         if (!work) {
131                 trace_writeback_nowork(bdi);
132                 bdi_wakeup_thread(bdi);
133                 return;
134         }
135 
136         work->sync_mode = WB_SYNC_NONE;
137         work->nr_pages  = nr_pages;
138         work->range_cyclic = range_cyclic;
139         work->reason    = reason;
140 
141         bdi_queue_work(bdi, work);
142 }
143 
144 /**
145  * bdi_start_writeback - start writeback
146  * @bdi: the backing device to write from
147  * @nr_pages: the number of pages to write
148  * @reason: reason why some writeback work was initiated
149  *
150  * Description:
151  *   This does WB_SYNC_NONE opportunistic writeback. The IO is only
152  *   started when this function returns, we make no guarantees on
153  *   completion. Caller need not hold sb s_umount semaphore.
154  *
155  */
156 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
157                         enum wb_reason reason)
158 {
159         __bdi_start_writeback(bdi, nr_pages, true, reason);
160 }
161 
162 /**
163  * bdi_start_background_writeback - start background writeback
164  * @bdi: the backing device to write from
165  *
166  * Description:
167  *   This makes sure WB_SYNC_NONE background writeback happens. When
168  *   this function returns, it is only guaranteed that for given BDI
169  *   some IO is happening if we are over background dirty threshold.
170  *   Caller need not hold sb s_umount semaphore.
171  */
172 void bdi_start_background_writeback(struct backing_dev_info *bdi)
173 {
174         /*
175          * We just wake up the flusher thread. It will perform background
176          * writeback as soon as there is no other work to do.
177          */
178         trace_writeback_wake_background(bdi);
179         bdi_wakeup_thread(bdi);
180 }
181 
182 /*
183  * Remove the inode from the writeback list it is on.
184  */
185 void inode_wb_list_del(struct inode *inode)
186 {
187         struct backing_dev_info *bdi = inode_to_bdi(inode);
188 
189         spin_lock(&bdi->wb.list_lock);
190         list_del_init(&inode->i_wb_list);
191         spin_unlock(&bdi->wb.list_lock);
192 }
193 
194 /*
195  * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
196  * furthest end of its superblock's dirty-inode list.
197  *
198  * Before stamping the inode's ->dirtied_when, we check to see whether it is
199  * already the most-recently-dirtied inode on the b_dirty list.  If that is
200  * the case then the inode must have been redirtied while it was being written
201  * out and we don't reset its dirtied_when.
202  */
203 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
204 {
205         assert_spin_locked(&wb->list_lock);
206         if (!list_empty(&wb->b_dirty)) {
207                 struct inode *tail;
208 
209                 tail = wb_inode(wb->b_dirty.next);
210                 if (time_before(inode->dirtied_when, tail->dirtied_when))
211                         inode->dirtied_when = jiffies;
212         }
213         list_move(&inode->i_wb_list, &wb->b_dirty);
214 }
215 
216 /*
217  * requeue inode for re-scanning after bdi->b_io list is exhausted.
218  */
219 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
220 {
221         assert_spin_locked(&wb->list_lock);
222         list_move(&inode->i_wb_list, &wb->b_more_io);
223 }
224 
225 static void inode_sync_complete(struct inode *inode)
226 {
227         inode->i_state &= ~I_SYNC;
228         /* If inode is clean an unused, put it into LRU now... */
229         inode_add_lru(inode);
230         /* Waiters must see I_SYNC cleared before being woken up */
231         smp_mb();
232         wake_up_bit(&inode->i_state, __I_SYNC);
233 }
234 
235 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
236 {
237         bool ret = time_after(inode->dirtied_when, t);
238 #ifndef CONFIG_64BIT
239         /*
240          * For inodes being constantly redirtied, dirtied_when can get stuck.
241          * It _appears_ to be in the future, but is actually in distant past.
242          * This test is necessary to prevent such wrapped-around relative times
243          * from permanently stopping the whole bdi writeback.
244          */
245         ret = ret && time_before_eq(inode->dirtied_when, jiffies);
246 #endif
247         return ret;
248 }
249 
250 /*
251  * Move expired (dirtied before work->older_than_this) dirty inodes from
252  * @delaying_queue to @dispatch_queue.
253  */
254 static int move_expired_inodes(struct list_head *delaying_queue,
255                                struct list_head *dispatch_queue,
256                                struct wb_writeback_work *work)
257 {
258         LIST_HEAD(tmp);
259         struct list_head *pos, *node;
260         struct super_block *sb = NULL;
261         struct inode *inode;
262         int do_sb_sort = 0;
263         int moved = 0;
264 
265         while (!list_empty(delaying_queue)) {
266                 inode = wb_inode(delaying_queue->prev);
267                 if (work->older_than_this &&
268                     inode_dirtied_after(inode, *work->older_than_this))
269                         break;
270                 list_move(&inode->i_wb_list, &tmp);
271                 moved++;
272                 if (sb_is_blkdev_sb(inode->i_sb))
273                         continue;
274                 if (sb && sb != inode->i_sb)
275                         do_sb_sort = 1;
276                 sb = inode->i_sb;
277         }
278 
279         /* just one sb in list, splice to dispatch_queue and we're done */
280         if (!do_sb_sort) {
281                 list_splice(&tmp, dispatch_queue);
282                 goto out;
283         }
284 
285         /* Move inodes from one superblock together */
286         while (!list_empty(&tmp)) {
287                 sb = wb_inode(tmp.prev)->i_sb;
288                 list_for_each_prev_safe(pos, node, &tmp) {
289                         inode = wb_inode(pos);
290                         if (inode->i_sb == sb)
291                                 list_move(&inode->i_wb_list, dispatch_queue);
292                 }
293         }
294 out:
295         return moved;
296 }
297 
298 /*
299  * Queue all expired dirty inodes for io, eldest first.
300  * Before
301  *         newly dirtied     b_dirty    b_io    b_more_io
302  *         =============>    gf         edc     BA
303  * After
304  *         newly dirtied     b_dirty    b_io    b_more_io
305  *         =============>    g          fBAedc
306  *                                           |
307  *                                           +--> dequeue for IO
308  */
309 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
310 {
311         int moved;
312         assert_spin_locked(&wb->list_lock);
313         list_splice_init(&wb->b_more_io, &wb->b_io);
314         moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
315         trace_writeback_queue_io(wb, work, moved);
316 }
317 
318 static int write_inode(struct inode *inode, struct writeback_control *wbc)
319 {
320         int ret;
321 
322         if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
323                 trace_writeback_write_inode_start(inode, wbc);
324                 ret = inode->i_sb->s_op->write_inode(inode, wbc);
325                 trace_writeback_write_inode(inode, wbc);
326                 return ret;
327         }
328         return 0;
329 }
330 
331 /*
332  * Wait for writeback on an inode to complete. Called with i_lock held.
333  * Caller must make sure inode cannot go away when we drop i_lock.
334  */
335 static void __inode_wait_for_writeback(struct inode *inode)
336         __releases(inode->i_lock)
337         __acquires(inode->i_lock)
338 {
339         DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
340         wait_queue_head_t *wqh;
341 
342         wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
343         while (inode->i_state & I_SYNC) {
344                 spin_unlock(&inode->i_lock);
345                 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
346                 spin_lock(&inode->i_lock);
347         }
348 }
349 
350 /*
351  * Wait for writeback on an inode to complete. Caller must have inode pinned.
352  */
353 void inode_wait_for_writeback(struct inode *inode)
354 {
355         spin_lock(&inode->i_lock);
356         __inode_wait_for_writeback(inode);
357         spin_unlock(&inode->i_lock);
358 }
359 
360 /*
361  * Sleep until I_SYNC is cleared. This function must be called with i_lock
362  * held and drops it. It is aimed for callers not holding any inode reference
363  * so once i_lock is dropped, inode can go away.
364  */
365 static void inode_sleep_on_writeback(struct inode *inode)
366         __releases(inode->i_lock)
367 {
368         DEFINE_WAIT(wait);
369         wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
370         int sleep;
371 
372         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
373         sleep = inode->i_state & I_SYNC;
374         spin_unlock(&inode->i_lock);
375         if (sleep)
376                 schedule();
377         finish_wait(wqh, &wait);
378 }
379 
380 /*
381  * Find proper writeback list for the inode depending on its current state and
382  * possibly also change of its state while we were doing writeback.  Here we
383  * handle things such as livelock prevention or fairness of writeback among
384  * inodes. This function can be called only by flusher thread - noone else
385  * processes all inodes in writeback lists and requeueing inodes behind flusher
386  * thread's back can have unexpected consequences.
387  */
388 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
389                           struct writeback_control *wbc)
390 {
391         if (inode->i_state & I_FREEING)
392                 return;
393 
394         /*
395          * Sync livelock prevention. Each inode is tagged and synced in one
396          * shot. If still dirty, it will be redirty_tail()'ed below.  Update
397          * the dirty time to prevent enqueue and sync it again.
398          */
399         if ((inode->i_state & I_DIRTY) &&
400             (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
401                 inode->dirtied_when = jiffies;
402 
403         if (wbc->pages_skipped) {
404                 /*
405                  * writeback is not making progress due to locked
406                  * buffers. Skip this inode for now.
407                  */
408                 redirty_tail(inode, wb);
409                 return;
410         }
411 
412         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
413                 /*
414                  * We didn't write back all the pages.  nfs_writepages()
415                  * sometimes bales out without doing anything.
416                  */
417                 if (wbc->nr_to_write <= 0) {
418                         /* Slice used up. Queue for next turn. */
419                         requeue_io(inode, wb);
420                 } else {
421                         /*
422                          * Writeback blocked by something other than
423                          * congestion. Delay the inode for some time to
424                          * avoid spinning on the CPU (100% iowait)
425                          * retrying writeback of the dirty page/inode
426                          * that cannot be performed immediately.
427                          */
428                         redirty_tail(inode, wb);
429                 }
430         } else if (inode->i_state & I_DIRTY) {
431                 /*
432                  * Filesystems can dirty the inode during writeback operations,
433                  * such as delayed allocation during submission or metadata
434                  * updates after data IO completion.
435                  */
436                 redirty_tail(inode, wb);
437         } else {
438                 /* The inode is clean. Remove from writeback lists. */
439                 list_del_init(&inode->i_wb_list);
440         }
441 }
442 
443 /*
444  * Write out an inode and its dirty pages. Do not update the writeback list
445  * linkage. That is left to the caller. The caller is also responsible for
446  * setting I_SYNC flag and calling inode_sync_complete() to clear it.
447  */
448 static int
449 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
450 {
451         struct address_space *mapping = inode->i_mapping;
452         long nr_to_write = wbc->nr_to_write;
453         unsigned dirty;
454         int ret;
455 
456         WARN_ON(!(inode->i_state & I_SYNC));
457 
458         trace_writeback_single_inode_start(inode, wbc, nr_to_write);
459 
460         ret = do_writepages(mapping, wbc);
461 
462         /*
463          * Make sure to wait on the data before writing out the metadata.
464          * This is important for filesystems that modify metadata on data
465          * I/O completion. We don't do it for sync(2) writeback because it has a
466          * separate, external IO completion path and ->sync_fs for guaranteeing
467          * inode metadata is written back correctly.
468          */
469         if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
470                 int err = filemap_fdatawait(mapping);
471                 if (ret == 0)
472                         ret = err;
473         }
474 
475         /*
476          * Some filesystems may redirty the inode during the writeback
477          * due to delalloc, clear dirty metadata flags right before
478          * write_inode()
479          */
480         spin_lock(&inode->i_lock);
481         /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
482         if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
483                 inode->i_state &= ~I_DIRTY_PAGES;
484         dirty = inode->i_state & I_DIRTY;
485         inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
486         spin_unlock(&inode->i_lock);
487         /* Don't write the inode if only I_DIRTY_PAGES was set */
488         if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
489                 int err = write_inode(inode, wbc);
490                 if (ret == 0)
491                         ret = err;
492         }
493         trace_writeback_single_inode(inode, wbc, nr_to_write);
494         return ret;
495 }
496 
497 /*
498  * Write out an inode's dirty pages. Either the caller has an active reference
499  * on the inode or the inode has I_WILL_FREE set.
500  *
501  * This function is designed to be called for writing back one inode which
502  * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
503  * and does more profound writeback list handling in writeback_sb_inodes().
504  */
505 static int
506 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
507                        struct writeback_control *wbc)
508 {
509         int ret = 0;
510 
511         spin_lock(&inode->i_lock);
512         if (!atomic_read(&inode->i_count))
513                 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
514         else
515                 WARN_ON(inode->i_state & I_WILL_FREE);
516 
517         if (inode->i_state & I_SYNC) {
518                 if (wbc->sync_mode != WB_SYNC_ALL)
519                         goto out;
520                 /*
521                  * It's a data-integrity sync. We must wait. Since callers hold
522                  * inode reference or inode has I_WILL_FREE set, it cannot go
523                  * away under us.
524                  */
525                 __inode_wait_for_writeback(inode);
526         }
527         WARN_ON(inode->i_state & I_SYNC);
528         /*
529          * Skip inode if it is clean and we have no outstanding writeback in
530          * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
531          * function since flusher thread may be doing for example sync in
532          * parallel and if we move the inode, it could get skipped. So here we
533          * make sure inode is on some writeback list and leave it there unless
534          * we have completely cleaned the inode.
535          */
536         if (!(inode->i_state & I_DIRTY) &&
537             (wbc->sync_mode != WB_SYNC_ALL ||
538              !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
539                 goto out;
540         inode->i_state |= I_SYNC;
541         spin_unlock(&inode->i_lock);
542 
543         ret = __writeback_single_inode(inode, wbc);
544 
545         spin_lock(&wb->list_lock);
546         spin_lock(&inode->i_lock);
547         /*
548          * If inode is clean, remove it from writeback lists. Otherwise don't
549          * touch it. See comment above for explanation.
550          */
551         if (!(inode->i_state & I_DIRTY))
552                 list_del_init(&inode->i_wb_list);
553         spin_unlock(&wb->list_lock);
554         inode_sync_complete(inode);
555 out:
556         spin_unlock(&inode->i_lock);
557         return ret;
558 }
559 
560 static long writeback_chunk_size(struct backing_dev_info *bdi,
561                                  struct wb_writeback_work *work)
562 {
563         long pages;
564 
565         /*
566          * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
567          * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
568          * here avoids calling into writeback_inodes_wb() more than once.
569          *
570          * The intended call sequence for WB_SYNC_ALL writeback is:
571          *
572          *      wb_writeback()
573          *          writeback_sb_inodes()       <== called only once
574          *              write_cache_pages()     <== called once for each inode
575          *                   (quickly) tag currently dirty pages
576          *                   (maybe slowly) sync all tagged pages
577          */
578         if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
579                 pages = LONG_MAX;
580         else {
581                 pages = min(bdi->avg_write_bandwidth / 2,
582                             global_dirty_limit / DIRTY_SCOPE);
583                 pages = min(pages, work->nr_pages);
584                 pages = round_down(pages + MIN_WRITEBACK_PAGES,
585                                    MIN_WRITEBACK_PAGES);
586         }
587 
588         return pages;
589 }
590 
591 /*
592  * Write a portion of b_io inodes which belong to @sb.
593  *
594  * Return the number of pages and/or inodes written.
595  */
596 static long writeback_sb_inodes(struct super_block *sb,
597                                 struct bdi_writeback *wb,
598                                 struct wb_writeback_work *work)
599 {
600         struct writeback_control wbc = {
601                 .sync_mode              = work->sync_mode,
602                 .tagged_writepages      = work->tagged_writepages,
603                 .for_kupdate            = work->for_kupdate,
604                 .for_background         = work->for_background,
605                 .for_sync               = work->for_sync,
606                 .range_cyclic           = work->range_cyclic,
607                 .range_start            = 0,
608                 .range_end              = LLONG_MAX,
609         };
610         unsigned long start_time = jiffies;
611         long write_chunk;
612         long wrote = 0;  /* count both pages and inodes */
613 
614         while (!list_empty(&wb->b_io)) {
615                 struct inode *inode = wb_inode(wb->b_io.prev);
616 
617                 if (inode->i_sb != sb) {
618                         if (work->sb) {
619                                 /*
620                                  * We only want to write back data for this
621                                  * superblock, move all inodes not belonging
622                                  * to it back onto the dirty list.
623                                  */
624                                 redirty_tail(inode, wb);
625                                 continue;
626                         }
627 
628                         /*
629                          * The inode belongs to a different superblock.
630                          * Bounce back to the caller to unpin this and
631                          * pin the next superblock.
632                          */
633                         break;
634                 }
635 
636                 /*
637                  * Don't bother with new inodes or inodes being freed, first
638                  * kind does not need periodic writeout yet, and for the latter
639                  * kind writeout is handled by the freer.
640                  */
641                 spin_lock(&inode->i_lock);
642                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
643                         spin_unlock(&inode->i_lock);
644                         redirty_tail(inode, wb);
645                         continue;
646                 }
647                 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
648                         /*
649                          * If this inode is locked for writeback and we are not
650                          * doing writeback-for-data-integrity, move it to
651                          * b_more_io so that writeback can proceed with the
652                          * other inodes on s_io.
653                          *
654                          * We'll have another go at writing back this inode
655                          * when we completed a full scan of b_io.
656                          */
657                         spin_unlock(&inode->i_lock);
658                         requeue_io(inode, wb);
659                         trace_writeback_sb_inodes_requeue(inode);
660                         continue;
661                 }
662                 spin_unlock(&wb->list_lock);
663 
664                 /*
665                  * We already requeued the inode if it had I_SYNC set and we
666                  * are doing WB_SYNC_NONE writeback. So this catches only the
667                  * WB_SYNC_ALL case.
668                  */
669                 if (inode->i_state & I_SYNC) {
670                         /* Wait for I_SYNC. This function drops i_lock... */
671                         inode_sleep_on_writeback(inode);
672                         /* Inode may be gone, start again */
673                         spin_lock(&wb->list_lock);
674                         continue;
675                 }
676                 inode->i_state |= I_SYNC;
677                 spin_unlock(&inode->i_lock);
678 
679                 write_chunk = writeback_chunk_size(wb->bdi, work);
680                 wbc.nr_to_write = write_chunk;
681                 wbc.pages_skipped = 0;
682 
683                 /*
684                  * We use I_SYNC to pin the inode in memory. While it is set
685                  * evict_inode() will wait so the inode cannot be freed.
686                  */
687                 __writeback_single_inode(inode, &wbc);
688 
689                 work->nr_pages -= write_chunk - wbc.nr_to_write;
690                 wrote += write_chunk - wbc.nr_to_write;
691                 spin_lock(&wb->list_lock);
692                 spin_lock(&inode->i_lock);
693                 if (!(inode->i_state & I_DIRTY))
694                         wrote++;
695                 requeue_inode(inode, wb, &wbc);
696                 inode_sync_complete(inode);
697                 spin_unlock(&inode->i_lock);
698                 cond_resched_lock(&wb->list_lock);
699                 /*
700                  * bail out to wb_writeback() often enough to check
701                  * background threshold and other termination conditions.
702                  */
703                 if (wrote) {
704                         if (time_is_before_jiffies(start_time + HZ / 10UL))
705                                 break;
706                         if (work->nr_pages <= 0)
707                                 break;
708                 }
709         }
710         return wrote;
711 }
712 
713 static long __writeback_inodes_wb(struct bdi_writeback *wb,
714                                   struct wb_writeback_work *work)
715 {
716         unsigned long start_time = jiffies;
717         long wrote = 0;
718 
719         while (!list_empty(&wb->b_io)) {
720                 struct inode *inode = wb_inode(wb->b_io.prev);
721                 struct super_block *sb = inode->i_sb;
722 
723                 if (!grab_super_passive(sb)) {
724                         /*
725                          * grab_super_passive() may fail consistently due to
726                          * s_umount being grabbed by someone else. Don't use
727                          * requeue_io() to avoid busy retrying the inode/sb.
728                          */
729                         redirty_tail(inode, wb);
730                         continue;
731                 }
732                 wrote += writeback_sb_inodes(sb, wb, work);
733                 drop_super(sb);
734 
735                 /* refer to the same tests at the end of writeback_sb_inodes */
736                 if (wrote) {
737                         if (time_is_before_jiffies(start_time + HZ / 10UL))
738                                 break;
739                         if (work->nr_pages <= 0)
740                                 break;
741                 }
742         }
743         /* Leave any unwritten inodes on b_io */
744         return wrote;
745 }
746 
747 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
748                                 enum wb_reason reason)
749 {
750         struct wb_writeback_work work = {
751                 .nr_pages       = nr_pages,
752                 .sync_mode      = WB_SYNC_NONE,
753                 .range_cyclic   = 1,
754                 .reason         = reason,
755         };
756 
757         spin_lock(&wb->list_lock);
758         if (list_empty(&wb->b_io))
759                 queue_io(wb, &work);
760         __writeback_inodes_wb(wb, &work);
761         spin_unlock(&wb->list_lock);
762 
763         return nr_pages - work.nr_pages;
764 }
765 
766 static bool over_bground_thresh(struct backing_dev_info *bdi)
767 {
768         unsigned long background_thresh, dirty_thresh;
769 
770         global_dirty_limits(&background_thresh, &dirty_thresh);
771 
772         if (global_page_state(NR_FILE_DIRTY) +
773             global_page_state(NR_UNSTABLE_NFS) > background_thresh)
774                 return true;
775 
776         if (bdi_stat(bdi, BDI_RECLAIMABLE) >
777                                 bdi_dirty_limit(bdi, background_thresh))
778                 return true;
779 
780         return false;
781 }
782 
783 /*
784  * Called under wb->list_lock. If there are multiple wb per bdi,
785  * only the flusher working on the first wb should do it.
786  */
787 static void wb_update_bandwidth(struct bdi_writeback *wb,
788                                 unsigned long start_time)
789 {
790         __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
791 }
792 
793 /*
794  * Explicit flushing or periodic writeback of "old" data.
795  *
796  * Define "old": the first time one of an inode's pages is dirtied, we mark the
797  * dirtying-time in the inode's address_space.  So this periodic writeback code
798  * just walks the superblock inode list, writing back any inodes which are
799  * older than a specific point in time.
800  *
801  * Try to run once per dirty_writeback_interval.  But if a writeback event
802  * takes longer than a dirty_writeback_interval interval, then leave a
803  * one-second gap.
804  *
805  * older_than_this takes precedence over nr_to_write.  So we'll only write back
806  * all dirty pages if they are all attached to "old" mappings.
807  */
808 static long wb_writeback(struct bdi_writeback *wb,
809                          struct wb_writeback_work *work)
810 {
811         unsigned long wb_start = jiffies;
812         long nr_pages = work->nr_pages;
813         unsigned long oldest_jif;
814         struct inode *inode;
815         long progress;
816 
817         oldest_jif = jiffies;
818         work->older_than_this = &oldest_jif;
819 
820         spin_lock(&wb->list_lock);
821         for (;;) {
822                 /*
823                  * Stop writeback when nr_pages has been consumed
824                  */
825                 if (work->nr_pages <= 0)
826                         break;
827 
828                 /*
829                  * Background writeout and kupdate-style writeback may
830                  * run forever. Stop them if there is other work to do
831                  * so that e.g. sync can proceed. They'll be restarted
832                  * after the other works are all done.
833                  */
834                 if ((work->for_background || work->for_kupdate) &&
835                     !list_empty(&wb->bdi->work_list))
836                         break;
837 
838                 /*
839                  * For background writeout, stop when we are below the
840                  * background dirty threshold
841                  */
842                 if (work->for_background && !over_bground_thresh(wb->bdi))
843                         break;
844 
845                 /*
846                  * Kupdate and background works are special and we want to
847                  * include all inodes that need writing. Livelock avoidance is
848                  * handled by these works yielding to any other work so we are
849                  * safe.
850                  */
851                 if (work->for_kupdate) {
852                         oldest_jif = jiffies -
853                                 msecs_to_jiffies(dirty_expire_interval * 10);
854                 } else if (work->for_background)
855                         oldest_jif = jiffies;
856 
857                 trace_writeback_start(wb->bdi, work);
858                 if (list_empty(&wb->b_io))
859                         queue_io(wb, work);
860                 if (work->sb)
861                         progress = writeback_sb_inodes(work->sb, wb, work);
862                 else
863                         progress = __writeback_inodes_wb(wb, work);
864                 trace_writeback_written(wb->bdi, work);
865 
866                 wb_update_bandwidth(wb, wb_start);
867 
868                 /*
869                  * Did we write something? Try for more
870                  *
871                  * Dirty inodes are moved to b_io for writeback in batches.
872                  * The completion of the current batch does not necessarily
873                  * mean the overall work is done. So we keep looping as long
874                  * as made some progress on cleaning pages or inodes.
875                  */
876                 if (progress)
877                         continue;
878                 /*
879                  * No more inodes for IO, bail
880                  */
881                 if (list_empty(&wb->b_more_io))
882                         break;
883                 /*
884                  * Nothing written. Wait for some inode to
885                  * become available for writeback. Otherwise
886                  * we'll just busyloop.
887                  */
888                 if (!list_empty(&wb->b_more_io))  {
889                         trace_writeback_wait(wb->bdi, work);
890                         inode = wb_inode(wb->b_more_io.prev);
891                         spin_lock(&inode->i_lock);
892                         spin_unlock(&wb->list_lock);
893                         /* This function drops i_lock... */
894                         inode_sleep_on_writeback(inode);
895                         spin_lock(&wb->list_lock);
896                 }
897         }
898         spin_unlock(&wb->list_lock);
899 
900         return nr_pages - work->nr_pages;
901 }
902 
903 /*
904  * Return the next wb_writeback_work struct that hasn't been processed yet.
905  */
906 static struct wb_writeback_work *
907 get_next_work_item(struct backing_dev_info *bdi)
908 {
909         struct wb_writeback_work *work = NULL;
910 
911         spin_lock_bh(&bdi->wb_lock);
912         if (!list_empty(&bdi->work_list)) {
913                 work = list_entry(bdi->work_list.next,
914                                   struct wb_writeback_work, list);
915                 list_del_init(&work->list);
916         }
917         spin_unlock_bh(&bdi->wb_lock);
918         return work;
919 }
920 
921 /*
922  * Add in the number of potentially dirty inodes, because each inode
923  * write can dirty pagecache in the underlying blockdev.
924  */
925 static unsigned long get_nr_dirty_pages(void)
926 {
927         return global_page_state(NR_FILE_DIRTY) +
928                 global_page_state(NR_UNSTABLE_NFS) +
929                 get_nr_dirty_inodes();
930 }
931 
932 static long wb_check_background_flush(struct bdi_writeback *wb)
933 {
934         if (over_bground_thresh(wb->bdi)) {
935 
936                 struct wb_writeback_work work = {
937                         .nr_pages       = LONG_MAX,
938                         .sync_mode      = WB_SYNC_NONE,
939                         .for_background = 1,
940                         .range_cyclic   = 1,
941                         .reason         = WB_REASON_BACKGROUND,
942                 };
943 
944                 return wb_writeback(wb, &work);
945         }
946 
947         return 0;
948 }
949 
950 static long wb_check_old_data_flush(struct bdi_writeback *wb)
951 {
952         unsigned long expired;
953         long nr_pages;
954 
955         /*
956          * When set to zero, disable periodic writeback
957          */
958         if (!dirty_writeback_interval)
959                 return 0;
960 
961         expired = wb->last_old_flush +
962                         msecs_to_jiffies(dirty_writeback_interval * 10);
963         if (time_before(jiffies, expired))
964                 return 0;
965 
966         wb->last_old_flush = jiffies;
967         nr_pages = get_nr_dirty_pages();
968 
969         if (nr_pages) {
970                 struct wb_writeback_work work = {
971                         .nr_pages       = nr_pages,
972                         .sync_mode      = WB_SYNC_NONE,
973                         .for_kupdate    = 1,
974                         .range_cyclic   = 1,
975                         .reason         = WB_REASON_PERIODIC,
976                 };
977 
978                 return wb_writeback(wb, &work);
979         }
980 
981         return 0;
982 }
983 
984 /*
985  * Retrieve work items and do the writeback they describe
986  */
987 static long wb_do_writeback(struct bdi_writeback *wb)
988 {
989         struct backing_dev_info *bdi = wb->bdi;
990         struct wb_writeback_work *work;
991         long wrote = 0;
992 
993         set_bit(BDI_writeback_running, &wb->bdi->state);
994         while ((work = get_next_work_item(bdi)) != NULL) {
995 
996                 trace_writeback_exec(bdi, work);
997 
998                 wrote += wb_writeback(wb, work);
999 
1000                 /*
1001                  * Notify the caller of completion if this is a synchronous
1002                  * work item, otherwise just free it.
1003                  */
1004                 if (work->done)
1005                         complete(work->done);
1006                 else
1007                         kfree(work);
1008         }
1009 
1010         /*
1011          * Check for periodic writeback, kupdated() style
1012          */
1013         wrote += wb_check_old_data_flush(wb);
1014         wrote += wb_check_background_flush(wb);
1015         clear_bit(BDI_writeback_running, &wb->bdi->state);
1016 
1017         return wrote;
1018 }
1019 
1020 /*
1021  * Handle writeback of dirty data for the device backed by this bdi. Also
1022  * reschedules periodically and does kupdated style flushing.
1023  */
1024 void bdi_writeback_workfn(struct work_struct *work)
1025 {
1026         struct bdi_writeback *wb = container_of(to_delayed_work(work),
1027                                                 struct bdi_writeback, dwork);
1028         struct backing_dev_info *bdi = wb->bdi;
1029         long pages_written;
1030 
1031         set_worker_desc("flush-%s", dev_name(bdi->dev));
1032         current->flags |= PF_SWAPWRITE;
1033 
1034         if (likely(!current_is_workqueue_rescuer() ||
1035                    !test_bit(BDI_registered, &bdi->state))) {
1036                 /*
1037                  * The normal path.  Keep writing back @bdi until its
1038                  * work_list is empty.  Note that this path is also taken
1039                  * if @bdi is shutting down even when we're running off the
1040                  * rescuer as work_list needs to be drained.
1041                  */
1042                 do {
1043                         pages_written = wb_do_writeback(wb);
1044                         trace_writeback_pages_written(pages_written);
1045                 } while (!list_empty(&bdi->work_list));
1046         } else {
1047                 /*
1048                  * bdi_wq can't get enough workers and we're running off
1049                  * the emergency worker.  Don't hog it.  Hopefully, 1024 is
1050                  * enough for efficient IO.
1051                  */
1052                 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1053                                                     WB_REASON_FORKER_THREAD);
1054                 trace_writeback_pages_written(pages_written);
1055         }
1056 
1057         if (!list_empty(&bdi->work_list))
1058                 mod_delayed_work(bdi_wq, &wb->dwork, 0);
1059         else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1060                 bdi_wakeup_thread_delayed(bdi);
1061 
1062         current->flags &= ~PF_SWAPWRITE;
1063 }
1064 
1065 /*
1066  * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
1067  * the whole world.
1068  */
1069 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1070 {
1071         struct backing_dev_info *bdi;
1072 
1073         if (!nr_pages)
1074                 nr_pages = get_nr_dirty_pages();
1075 
1076         rcu_read_lock();
1077         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1078                 if (!bdi_has_dirty_io(bdi))
1079                         continue;
1080                 __bdi_start_writeback(bdi, nr_pages, false, reason);
1081         }
1082         rcu_read_unlock();
1083 }
1084 
1085 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1086 {
1087         if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1088                 struct dentry *dentry;
1089                 const char *name = "?";
1090 
1091                 dentry = d_find_alias(inode);
1092                 if (dentry) {
1093                         spin_lock(&dentry->d_lock);
1094                         name = (const char *) dentry->d_name.name;
1095                 }
1096                 printk(KERN_DEBUG
1097                        "%s(%d): dirtied inode %lu (%s) on %s\n",
1098                        current->comm, task_pid_nr(current), inode->i_ino,
1099                        name, inode->i_sb->s_id);
1100                 if (dentry) {
1101                         spin_unlock(&dentry->d_lock);
1102                         dput(dentry);
1103                 }
1104         }
1105 }
1106 
1107 /**
1108  *      __mark_inode_dirty -    internal function
1109  *      @inode: inode to mark
1110  *      @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1111  *      Mark an inode as dirty. Callers should use mark_inode_dirty or
1112  *      mark_inode_dirty_sync.
1113  *
1114  * Put the inode on the super block's dirty list.
1115  *
1116  * CAREFUL! We mark it dirty unconditionally, but move it onto the
1117  * dirty list only if it is hashed or if it refers to a blockdev.
1118  * If it was not hashed, it will never be added to the dirty list
1119  * even if it is later hashed, as it will have been marked dirty already.
1120  *
1121  * In short, make sure you hash any inodes _before_ you start marking
1122  * them dirty.
1123  *
1124  * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1125  * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
1126  * the kernel-internal blockdev inode represents the dirtying time of the
1127  * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
1128  * page->mapping->host, so the page-dirtying time is recorded in the internal
1129  * blockdev inode.
1130  */
1131 void __mark_inode_dirty(struct inode *inode, int flags)
1132 {
1133         struct super_block *sb = inode->i_sb;
1134         struct backing_dev_info *bdi = NULL;
1135 
1136         /*
1137          * Don't do this for I_DIRTY_PAGES - that doesn't actually
1138          * dirty the inode itself
1139          */
1140         if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1141                 trace_writeback_dirty_inode_start(inode, flags);
1142 
1143                 if (sb->s_op->dirty_inode)
1144                         sb->s_op->dirty_inode(inode, flags);
1145 
1146                 trace_writeback_dirty_inode(inode, flags);
1147         }
1148 
1149         /*
1150          * make sure that changes are seen by all cpus before we test i_state
1151          * -- mikulas
1152          */
1153         smp_mb();
1154 
1155         /* avoid the locking if we can */
1156         if ((inode->i_state & flags) == flags)
1157                 return;
1158 
1159         if (unlikely(block_dump))
1160                 block_dump___mark_inode_dirty(inode);
1161 
1162         spin_lock(&inode->i_lock);
1163         if ((inode->i_state & flags) != flags) {
1164                 const int was_dirty = inode->i_state & I_DIRTY;
1165 
1166                 inode->i_state |= flags;
1167 
1168                 /*
1169                  * If the inode is being synced, just update its dirty state.
1170                  * The unlocker will place the inode on the appropriate
1171                  * superblock list, based upon its state.
1172                  */
1173                 if (inode->i_state & I_SYNC)
1174                         goto out_unlock_inode;
1175 
1176                 /*
1177                  * Only add valid (hashed) inodes to the superblock's
1178                  * dirty list.  Add blockdev inodes as well.
1179                  */
1180                 if (!S_ISBLK(inode->i_mode)) {
1181                         if (inode_unhashed(inode))
1182                                 goto out_unlock_inode;
1183                 }
1184                 if (inode->i_state & I_FREEING)
1185                         goto out_unlock_inode;
1186 
1187                 /*
1188                  * If the inode was already on b_dirty/b_io/b_more_io, don't
1189                  * reposition it (that would break b_dirty time-ordering).
1190                  */
1191                 if (!was_dirty) {
1192                         bool wakeup_bdi = false;
1193                         bdi = inode_to_bdi(inode);
1194 
1195                         spin_unlock(&inode->i_lock);
1196                         spin_lock(&bdi->wb.list_lock);
1197                         if (bdi_cap_writeback_dirty(bdi)) {
1198                                 WARN(!test_bit(BDI_registered, &bdi->state),
1199                                      "bdi-%s not registered\n", bdi->name);
1200 
1201                                 /*
1202                                  * If this is the first dirty inode for this
1203                                  * bdi, we have to wake-up the corresponding
1204                                  * bdi thread to make sure background
1205                                  * write-back happens later.
1206                                  */
1207                                 if (!wb_has_dirty_io(&bdi->wb))
1208                                         wakeup_bdi = true;
1209                         }
1210 
1211                         inode->dirtied_when = jiffies;
1212                         list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1213                         spin_unlock(&bdi->wb.list_lock);
1214 
1215                         if (wakeup_bdi)
1216                                 bdi_wakeup_thread_delayed(bdi);
1217                         return;
1218                 }
1219         }
1220 out_unlock_inode:
1221         spin_unlock(&inode->i_lock);
1222 
1223 }
1224 EXPORT_SYMBOL(__mark_inode_dirty);
1225 
1226 static void wait_sb_inodes(struct super_block *sb)
1227 {
1228         struct inode *inode, *old_inode = NULL;
1229 
1230         /*
1231          * We need to be protected against the filesystem going from
1232          * r/o to r/w or vice versa.
1233          */
1234         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1235 
1236         spin_lock(&inode_sb_list_lock);
1237 
1238         /*
1239          * Data integrity sync. Must wait for all pages under writeback,
1240          * because there may have been pages dirtied before our sync
1241          * call, but which had writeout started before we write it out.
1242          * In which case, the inode may not be on the dirty list, but
1243          * we still have to wait for that writeout.
1244          */
1245         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1246                 struct address_space *mapping = inode->i_mapping;
1247 
1248                 spin_lock(&inode->i_lock);
1249                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1250                     (mapping->nrpages == 0)) {
1251                         spin_unlock(&inode->i_lock);
1252                         continue;
1253                 }
1254                 __iget(inode);
1255                 spin_unlock(&inode->i_lock);
1256                 spin_unlock(&inode_sb_list_lock);
1257 
1258                 /*
1259                  * We hold a reference to 'inode' so it couldn't have been
1260                  * removed from s_inodes list while we dropped the
1261                  * inode_sb_list_lock.  We cannot iput the inode now as we can
1262                  * be holding the last reference and we cannot iput it under
1263                  * inode_sb_list_lock. So we keep the reference and iput it
1264                  * later.
1265                  */
1266                 iput(old_inode);
1267                 old_inode = inode;
1268 
1269                 filemap_fdatawait(mapping);
1270 
1271                 cond_resched();
1272 
1273                 spin_lock(&inode_sb_list_lock);
1274         }
1275         spin_unlock(&inode_sb_list_lock);
1276         iput(old_inode);
1277 }
1278 
1279 /**
1280  * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
1281  * @sb: the superblock
1282  * @nr: the number of pages to write
1283  * @reason: reason why some writeback work initiated
1284  *
1285  * Start writeback on some inodes on this super_block. No guarantees are made
1286  * on how many (if any) will be written, and this function does not wait
1287  * for IO completion of submitted IO.
1288  */
1289 void writeback_inodes_sb_nr(struct super_block *sb,
1290                             unsigned long nr,
1291                             enum wb_reason reason)
1292 {
1293         DECLARE_COMPLETION_ONSTACK(done);
1294         struct wb_writeback_work work = {
1295                 .sb                     = sb,
1296                 .sync_mode              = WB_SYNC_NONE,
1297                 .tagged_writepages      = 1,
1298                 .done                   = &done,
1299                 .nr_pages               = nr,
1300                 .reason                 = reason,
1301         };
1302 
1303         if (sb->s_bdi == &noop_backing_dev_info)
1304                 return;
1305         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1306         bdi_queue_work(sb->s_bdi, &work);
1307         wait_for_completion(&done);
1308 }
1309 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1310 
1311 /**
1312  * writeback_inodes_sb  -       writeback dirty inodes from given super_block
1313  * @sb: the superblock
1314  * @reason: reason why some writeback work was initiated
1315  *
1316  * Start writeback on some inodes on this super_block. No guarantees are made
1317  * on how many (if any) will be written, and this function does not wait
1318  * for IO completion of submitted IO.
1319  */
1320 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1321 {
1322         return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1323 }
1324 EXPORT_SYMBOL(writeback_inodes_sb);
1325 
1326 /**
1327  * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1328  * @sb: the superblock
1329  * @nr: the number of pages to write
1330  * @reason: the reason of writeback
1331  *
1332  * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1333  * Returns 1 if writeback was started, 0 if not.
1334  */
1335 int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1336                                   unsigned long nr,
1337                                   enum wb_reason reason)
1338 {
1339         if (writeback_in_progress(sb->s_bdi))
1340                 return 1;
1341 
1342         if (!down_read_trylock(&sb->s_umount))
1343                 return 0;
1344 
1345         writeback_inodes_sb_nr(sb, nr, reason);
1346         up_read(&sb->s_umount);
1347         return 1;
1348 }
1349 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1350 
1351 /**
1352  * try_to_writeback_inodes_sb - try to start writeback if none underway
1353  * @sb: the superblock
1354  * @reason: reason why some writeback work was initiated
1355  *
1356  * Implement by try_to_writeback_inodes_sb_nr()
1357  * Returns 1 if writeback was started, 0 if not.
1358  */
1359 int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1360 {
1361         return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1362 }
1363 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
1364 
1365 /**
1366  * sync_inodes_sb       -       sync sb inode pages
1367  * @sb: the superblock
1368  *
1369  * This function writes and waits on any dirty inode belonging to this
1370  * super_block.
1371  */
1372 void sync_inodes_sb(struct super_block *sb)
1373 {
1374         DECLARE_COMPLETION_ONSTACK(done);
1375         struct wb_writeback_work work = {
1376                 .sb             = sb,
1377                 .sync_mode      = WB_SYNC_ALL,
1378                 .nr_pages       = LONG_MAX,
1379                 .range_cyclic   = 0,
1380                 .done           = &done,
1381                 .reason         = WB_REASON_SYNC,
1382                 .for_sync       = 1,
1383         };
1384 
1385         /* Nothing to do? */
1386         if (sb->s_bdi == &noop_backing_dev_info)
1387                 return;
1388         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1389 
1390         bdi_queue_work(sb->s_bdi, &work);
1391         wait_for_completion(&done);
1392 
1393         wait_sb_inodes(sb);
1394 }
1395 EXPORT_SYMBOL(sync_inodes_sb);
1396 
1397 /**
1398  * write_inode_now      -       write an inode to disk
1399  * @inode: inode to write to disk
1400  * @sync: whether the write should be synchronous or not
1401  *
1402  * This function commits an inode to disk immediately if it is dirty. This is
1403  * primarily needed by knfsd.
1404  *
1405  * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1406  */
1407 int write_inode_now(struct inode *inode, int sync)
1408 {
1409         struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1410         struct writeback_control wbc = {
1411                 .nr_to_write = LONG_MAX,
1412                 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1413                 .range_start = 0,
1414                 .range_end = LLONG_MAX,
1415         };
1416 
1417         if (!mapping_cap_writeback_dirty(inode->i_mapping))
1418                 wbc.nr_to_write = 0;
1419 
1420         might_sleep();
1421         return writeback_single_inode(inode, wb, &wbc);
1422 }
1423 EXPORT_SYMBOL(write_inode_now);
1424 
1425 /**
1426  * sync_inode - write an inode and its pages to disk.
1427  * @inode: the inode to sync
1428  * @wbc: controls the writeback mode
1429  *
1430  * sync_inode() will write an inode and its pages to disk.  It will also
1431  * correctly update the inode on its superblock's dirty inode lists and will
1432  * update inode->i_state.
1433  *
1434  * The caller must have a ref on the inode.
1435  */
1436 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1437 {
1438         return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1439 }
1440 EXPORT_SYMBOL(sync_inode);
1441 
1442 /**
1443  * sync_inode_metadata - write an inode to disk
1444  * @inode: the inode to sync
1445  * @wait: wait for I/O to complete.
1446  *
1447  * Write an inode to disk and adjust its dirty state after completion.
1448  *
1449  * Note: only writes the actual inode, no associated data or other metadata.
1450  */
1451 int sync_inode_metadata(struct inode *inode, int wait)
1452 {
1453         struct writeback_control wbc = {
1454                 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1455                 .nr_to_write = 0, /* metadata-only */
1456         };
1457 
1458         return sync_inode(inode, &wbc);
1459 }
1460 EXPORT_SYMBOL(sync_inode_metadata);
1461 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us