Version:  2.0.40 2.2.26 2.4.37 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 3.16 3.17 3.18

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, bit_wait,
346                               TASK_UNINTERRUPTIBLE);
347                 spin_lock(&inode->i_lock);
348         }
349 }
350 
351 /*
352  * Wait for writeback on an inode to complete. Caller must have inode pinned.
353  */
354 void inode_wait_for_writeback(struct inode *inode)
355 {
356         spin_lock(&inode->i_lock);
357         __inode_wait_for_writeback(inode);
358         spin_unlock(&inode->i_lock);
359 }
360 
361 /*
362  * Sleep until I_SYNC is cleared. This function must be called with i_lock
363  * held and drops it. It is aimed for callers not holding any inode reference
364  * so once i_lock is dropped, inode can go away.
365  */
366 static void inode_sleep_on_writeback(struct inode *inode)
367         __releases(inode->i_lock)
368 {
369         DEFINE_WAIT(wait);
370         wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
371         int sleep;
372 
373         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
374         sleep = inode->i_state & I_SYNC;
375         spin_unlock(&inode->i_lock);
376         if (sleep)
377                 schedule();
378         finish_wait(wqh, &wait);
379 }
380 
381 /*
382  * Find proper writeback list for the inode depending on its current state and
383  * possibly also change of its state while we were doing writeback.  Here we
384  * handle things such as livelock prevention or fairness of writeback among
385  * inodes. This function can be called only by flusher thread - noone else
386  * processes all inodes in writeback lists and requeueing inodes behind flusher
387  * thread's back can have unexpected consequences.
388  */
389 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
390                           struct writeback_control *wbc)
391 {
392         if (inode->i_state & I_FREEING)
393                 return;
394 
395         /*
396          * Sync livelock prevention. Each inode is tagged and synced in one
397          * shot. If still dirty, it will be redirty_tail()'ed below.  Update
398          * the dirty time to prevent enqueue and sync it again.
399          */
400         if ((inode->i_state & I_DIRTY) &&
401             (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
402                 inode->dirtied_when = jiffies;
403 
404         if (wbc->pages_skipped) {
405                 /*
406                  * writeback is not making progress due to locked
407                  * buffers. Skip this inode for now.
408                  */
409                 redirty_tail(inode, wb);
410                 return;
411         }
412 
413         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
414                 /*
415                  * We didn't write back all the pages.  nfs_writepages()
416                  * sometimes bales out without doing anything.
417                  */
418                 if (wbc->nr_to_write <= 0) {
419                         /* Slice used up. Queue for next turn. */
420                         requeue_io(inode, wb);
421                 } else {
422                         /*
423                          * Writeback blocked by something other than
424                          * congestion. Delay the inode for some time to
425                          * avoid spinning on the CPU (100% iowait)
426                          * retrying writeback of the dirty page/inode
427                          * that cannot be performed immediately.
428                          */
429                         redirty_tail(inode, wb);
430                 }
431         } else if (inode->i_state & I_DIRTY) {
432                 /*
433                  * Filesystems can dirty the inode during writeback operations,
434                  * such as delayed allocation during submission or metadata
435                  * updates after data IO completion.
436                  */
437                 redirty_tail(inode, wb);
438         } else {
439                 /* The inode is clean. Remove from writeback lists. */
440                 list_del_init(&inode->i_wb_list);
441         }
442 }
443 
444 /*
445  * Write out an inode and its dirty pages. Do not update the writeback list
446  * linkage. That is left to the caller. The caller is also responsible for
447  * setting I_SYNC flag and calling inode_sync_complete() to clear it.
448  */
449 static int
450 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
451 {
452         struct address_space *mapping = inode->i_mapping;
453         long nr_to_write = wbc->nr_to_write;
454         unsigned dirty;
455         int ret;
456 
457         WARN_ON(!(inode->i_state & I_SYNC));
458 
459         trace_writeback_single_inode_start(inode, wbc, nr_to_write);
460 
461         ret = do_writepages(mapping, wbc);
462 
463         /*
464          * Make sure to wait on the data before writing out the metadata.
465          * This is important for filesystems that modify metadata on data
466          * I/O completion. We don't do it for sync(2) writeback because it has a
467          * separate, external IO completion path and ->sync_fs for guaranteeing
468          * inode metadata is written back correctly.
469          */
470         if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
471                 int err = filemap_fdatawait(mapping);
472                 if (ret == 0)
473                         ret = err;
474         }
475 
476         /*
477          * Some filesystems may redirty the inode during the writeback
478          * due to delalloc, clear dirty metadata flags right before
479          * write_inode()
480          */
481         spin_lock(&inode->i_lock);
482         /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
483         if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
484                 inode->i_state &= ~I_DIRTY_PAGES;
485         dirty = inode->i_state & I_DIRTY;
486         inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
487         spin_unlock(&inode->i_lock);
488         /* Don't write the inode if only I_DIRTY_PAGES was set */
489         if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
490                 int err = write_inode(inode, wbc);
491                 if (ret == 0)
492                         ret = err;
493         }
494         trace_writeback_single_inode(inode, wbc, nr_to_write);
495         return ret;
496 }
497 
498 /*
499  * Write out an inode's dirty pages. Either the caller has an active reference
500  * on the inode or the inode has I_WILL_FREE set.
501  *
502  * This function is designed to be called for writing back one inode which
503  * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
504  * and does more profound writeback list handling in writeback_sb_inodes().
505  */
506 static int
507 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
508                        struct writeback_control *wbc)
509 {
510         int ret = 0;
511 
512         spin_lock(&inode->i_lock);
513         if (!atomic_read(&inode->i_count))
514                 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
515         else
516                 WARN_ON(inode->i_state & I_WILL_FREE);
517 
518         if (inode->i_state & I_SYNC) {
519                 if (wbc->sync_mode != WB_SYNC_ALL)
520                         goto out;
521                 /*
522                  * It's a data-integrity sync. We must wait. Since callers hold
523                  * inode reference or inode has I_WILL_FREE set, it cannot go
524                  * away under us.
525                  */
526                 __inode_wait_for_writeback(inode);
527         }
528         WARN_ON(inode->i_state & I_SYNC);
529         /*
530          * Skip inode if it is clean and we have no outstanding writeback in
531          * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
532          * function since flusher thread may be doing for example sync in
533          * parallel and if we move the inode, it could get skipped. So here we
534          * make sure inode is on some writeback list and leave it there unless
535          * we have completely cleaned the inode.
536          */
537         if (!(inode->i_state & I_DIRTY) &&
538             (wbc->sync_mode != WB_SYNC_ALL ||
539              !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
540                 goto out;
541         inode->i_state |= I_SYNC;
542         spin_unlock(&inode->i_lock);
543 
544         ret = __writeback_single_inode(inode, wbc);
545 
546         spin_lock(&wb->list_lock);
547         spin_lock(&inode->i_lock);
548         /*
549          * If inode is clean, remove it from writeback lists. Otherwise don't
550          * touch it. See comment above for explanation.
551          */
552         if (!(inode->i_state & I_DIRTY))
553                 list_del_init(&inode->i_wb_list);
554         spin_unlock(&wb->list_lock);
555         inode_sync_complete(inode);
556 out:
557         spin_unlock(&inode->i_lock);
558         return ret;
559 }
560 
561 static long writeback_chunk_size(struct backing_dev_info *bdi,
562                                  struct wb_writeback_work *work)
563 {
564         long pages;
565 
566         /*
567          * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
568          * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
569          * here avoids calling into writeback_inodes_wb() more than once.
570          *
571          * The intended call sequence for WB_SYNC_ALL writeback is:
572          *
573          *      wb_writeback()
574          *          writeback_sb_inodes()       <== called only once
575          *              write_cache_pages()     <== called once for each inode
576          *                   (quickly) tag currently dirty pages
577          *                   (maybe slowly) sync all tagged pages
578          */
579         if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
580                 pages = LONG_MAX;
581         else {
582                 pages = min(bdi->avg_write_bandwidth / 2,
583                             global_dirty_limit / DIRTY_SCOPE);
584                 pages = min(pages, work->nr_pages);
585                 pages = round_down(pages + MIN_WRITEBACK_PAGES,
586                                    MIN_WRITEBACK_PAGES);
587         }
588 
589         return pages;
590 }
591 
592 /*
593  * Write a portion of b_io inodes which belong to @sb.
594  *
595  * Return the number of pages and/or inodes written.
596  */
597 static long writeback_sb_inodes(struct super_block *sb,
598                                 struct bdi_writeback *wb,
599                                 struct wb_writeback_work *work)
600 {
601         struct writeback_control wbc = {
602                 .sync_mode              = work->sync_mode,
603                 .tagged_writepages      = work->tagged_writepages,
604                 .for_kupdate            = work->for_kupdate,
605                 .for_background         = work->for_background,
606                 .for_sync               = work->for_sync,
607                 .range_cyclic           = work->range_cyclic,
608                 .range_start            = 0,
609                 .range_end              = LLONG_MAX,
610         };
611         unsigned long start_time = jiffies;
612         long write_chunk;
613         long wrote = 0;  /* count both pages and inodes */
614 
615         while (!list_empty(&wb->b_io)) {
616                 struct inode *inode = wb_inode(wb->b_io.prev);
617 
618                 if (inode->i_sb != sb) {
619                         if (work->sb) {
620                                 /*
621                                  * We only want to write back data for this
622                                  * superblock, move all inodes not belonging
623                                  * to it back onto the dirty list.
624                                  */
625                                 redirty_tail(inode, wb);
626                                 continue;
627                         }
628 
629                         /*
630                          * The inode belongs to a different superblock.
631                          * Bounce back to the caller to unpin this and
632                          * pin the next superblock.
633                          */
634                         break;
635                 }
636 
637                 /*
638                  * Don't bother with new inodes or inodes being freed, first
639                  * kind does not need periodic writeout yet, and for the latter
640                  * kind writeout is handled by the freer.
641                  */
642                 spin_lock(&inode->i_lock);
643                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
644                         spin_unlock(&inode->i_lock);
645                         redirty_tail(inode, wb);
646                         continue;
647                 }
648                 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
649                         /*
650                          * If this inode is locked for writeback and we are not
651                          * doing writeback-for-data-integrity, move it to
652                          * b_more_io so that writeback can proceed with the
653                          * other inodes on s_io.
654                          *
655                          * We'll have another go at writing back this inode
656                          * when we completed a full scan of b_io.
657                          */
658                         spin_unlock(&inode->i_lock);
659                         requeue_io(inode, wb);
660                         trace_writeback_sb_inodes_requeue(inode);
661                         continue;
662                 }
663                 spin_unlock(&wb->list_lock);
664 
665                 /*
666                  * We already requeued the inode if it had I_SYNC set and we
667                  * are doing WB_SYNC_NONE writeback. So this catches only the
668                  * WB_SYNC_ALL case.
669                  */
670                 if (inode->i_state & I_SYNC) {
671                         /* Wait for I_SYNC. This function drops i_lock... */
672                         inode_sleep_on_writeback(inode);
673                         /* Inode may be gone, start again */
674                         spin_lock(&wb->list_lock);
675                         continue;
676                 }
677                 inode->i_state |= I_SYNC;
678                 spin_unlock(&inode->i_lock);
679 
680                 write_chunk = writeback_chunk_size(wb->bdi, work);
681                 wbc.nr_to_write = write_chunk;
682                 wbc.pages_skipped = 0;
683 
684                 /*
685                  * We use I_SYNC to pin the inode in memory. While it is set
686                  * evict_inode() will wait so the inode cannot be freed.
687                  */
688                 __writeback_single_inode(inode, &wbc);
689 
690                 work->nr_pages -= write_chunk - wbc.nr_to_write;
691                 wrote += write_chunk - wbc.nr_to_write;
692                 spin_lock(&wb->list_lock);
693                 spin_lock(&inode->i_lock);
694                 if (!(inode->i_state & I_DIRTY))
695                         wrote++;
696                 requeue_inode(inode, wb, &wbc);
697                 inode_sync_complete(inode);
698                 spin_unlock(&inode->i_lock);
699                 cond_resched_lock(&wb->list_lock);
700                 /*
701                  * bail out to wb_writeback() often enough to check
702                  * background threshold and other termination conditions.
703                  */
704                 if (wrote) {
705                         if (time_is_before_jiffies(start_time + HZ / 10UL))
706                                 break;
707                         if (work->nr_pages <= 0)
708                                 break;
709                 }
710         }
711         return wrote;
712 }
713 
714 static long __writeback_inodes_wb(struct bdi_writeback *wb,
715                                   struct wb_writeback_work *work)
716 {
717         unsigned long start_time = jiffies;
718         long wrote = 0;
719 
720         while (!list_empty(&wb->b_io)) {
721                 struct inode *inode = wb_inode(wb->b_io.prev);
722                 struct super_block *sb = inode->i_sb;
723 
724                 if (!grab_super_passive(sb)) {
725                         /*
726                          * grab_super_passive() may fail consistently due to
727                          * s_umount being grabbed by someone else. Don't use
728                          * requeue_io() to avoid busy retrying the inode/sb.
729                          */
730                         redirty_tail(inode, wb);
731                         continue;
732                 }
733                 wrote += writeback_sb_inodes(sb, wb, work);
734                 drop_super(sb);
735 
736                 /* refer to the same tests at the end of writeback_sb_inodes */
737                 if (wrote) {
738                         if (time_is_before_jiffies(start_time + HZ / 10UL))
739                                 break;
740                         if (work->nr_pages <= 0)
741                                 break;
742                 }
743         }
744         /* Leave any unwritten inodes on b_io */
745         return wrote;
746 }
747 
748 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
749                                 enum wb_reason reason)
750 {
751         struct wb_writeback_work work = {
752                 .nr_pages       = nr_pages,
753                 .sync_mode      = WB_SYNC_NONE,
754                 .range_cyclic   = 1,
755                 .reason         = reason,
756         };
757 
758         spin_lock(&wb->list_lock);
759         if (list_empty(&wb->b_io))
760                 queue_io(wb, &work);
761         __writeback_inodes_wb(wb, &work);
762         spin_unlock(&wb->list_lock);
763 
764         return nr_pages - work.nr_pages;
765 }
766 
767 static bool over_bground_thresh(struct backing_dev_info *bdi)
768 {
769         unsigned long background_thresh, dirty_thresh;
770 
771         global_dirty_limits(&background_thresh, &dirty_thresh);
772 
773         if (global_page_state(NR_FILE_DIRTY) +
774             global_page_state(NR_UNSTABLE_NFS) > background_thresh)
775                 return true;
776 
777         if (bdi_stat(bdi, BDI_RECLAIMABLE) >
778                                 bdi_dirty_limit(bdi, background_thresh))
779                 return true;
780 
781         return false;
782 }
783 
784 /*
785  * Called under wb->list_lock. If there are multiple wb per bdi,
786  * only the flusher working on the first wb should do it.
787  */
788 static void wb_update_bandwidth(struct bdi_writeback *wb,
789                                 unsigned long start_time)
790 {
791         __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
792 }
793 
794 /*
795  * Explicit flushing or periodic writeback of "old" data.
796  *
797  * Define "old": the first time one of an inode's pages is dirtied, we mark the
798  * dirtying-time in the inode's address_space.  So this periodic writeback code
799  * just walks the superblock inode list, writing back any inodes which are
800  * older than a specific point in time.
801  *
802  * Try to run once per dirty_writeback_interval.  But if a writeback event
803  * takes longer than a dirty_writeback_interval interval, then leave a
804  * one-second gap.
805  *
806  * older_than_this takes precedence over nr_to_write.  So we'll only write back
807  * all dirty pages if they are all attached to "old" mappings.
808  */
809 static long wb_writeback(struct bdi_writeback *wb,
810                          struct wb_writeback_work *work)
811 {
812         unsigned long wb_start = jiffies;
813         long nr_pages = work->nr_pages;
814         unsigned long oldest_jif;
815         struct inode *inode;
816         long progress;
817 
818         oldest_jif = jiffies;
819         work->older_than_this = &oldest_jif;
820 
821         spin_lock(&wb->list_lock);
822         for (;;) {
823                 /*
824                  * Stop writeback when nr_pages has been consumed
825                  */
826                 if (work->nr_pages <= 0)
827                         break;
828 
829                 /*
830                  * Background writeout and kupdate-style writeback may
831                  * run forever. Stop them if there is other work to do
832                  * so that e.g. sync can proceed. They'll be restarted
833                  * after the other works are all done.
834                  */
835                 if ((work->for_background || work->for_kupdate) &&
836                     !list_empty(&wb->bdi->work_list))
837                         break;
838 
839                 /*
840                  * For background writeout, stop when we are below the
841                  * background dirty threshold
842                  */
843                 if (work->for_background && !over_bground_thresh(wb->bdi))
844                         break;
845 
846                 /*
847                  * Kupdate and background works are special and we want to
848                  * include all inodes that need writing. Livelock avoidance is
849                  * handled by these works yielding to any other work so we are
850                  * safe.
851                  */
852                 if (work->for_kupdate) {
853                         oldest_jif = jiffies -
854                                 msecs_to_jiffies(dirty_expire_interval * 10);
855                 } else if (work->for_background)
856                         oldest_jif = jiffies;
857 
858                 trace_writeback_start(wb->bdi, work);
859                 if (list_empty(&wb->b_io))
860                         queue_io(wb, work);
861                 if (work->sb)
862                         progress = writeback_sb_inodes(work->sb, wb, work);
863                 else
864                         progress = __writeback_inodes_wb(wb, work);
865                 trace_writeback_written(wb->bdi, work);
866 
867                 wb_update_bandwidth(wb, wb_start);
868 
869                 /*
870                  * Did we write something? Try for more
871                  *
872                  * Dirty inodes are moved to b_io for writeback in batches.
873                  * The completion of the current batch does not necessarily
874                  * mean the overall work is done. So we keep looping as long
875                  * as made some progress on cleaning pages or inodes.
876                  */
877                 if (progress)
878                         continue;
879                 /*
880                  * No more inodes for IO, bail
881                  */
882                 if (list_empty(&wb->b_more_io))
883                         break;
884                 /*
885                  * Nothing written. Wait for some inode to
886                  * become available for writeback. Otherwise
887                  * we'll just busyloop.
888                  */
889                 if (!list_empty(&wb->b_more_io))  {
890                         trace_writeback_wait(wb->bdi, work);
891                         inode = wb_inode(wb->b_more_io.prev);
892                         spin_lock(&inode->i_lock);
893                         spin_unlock(&wb->list_lock);
894                         /* This function drops i_lock... */
895                         inode_sleep_on_writeback(inode);
896                         spin_lock(&wb->list_lock);
897                 }
898         }
899         spin_unlock(&wb->list_lock);
900 
901         return nr_pages - work->nr_pages;
902 }
903 
904 /*
905  * Return the next wb_writeback_work struct that hasn't been processed yet.
906  */
907 static struct wb_writeback_work *
908 get_next_work_item(struct backing_dev_info *bdi)
909 {
910         struct wb_writeback_work *work = NULL;
911 
912         spin_lock_bh(&bdi->wb_lock);
913         if (!list_empty(&bdi->work_list)) {
914                 work = list_entry(bdi->work_list.next,
915                                   struct wb_writeback_work, list);
916                 list_del_init(&work->list);
917         }
918         spin_unlock_bh(&bdi->wb_lock);
919         return work;
920 }
921 
922 /*
923  * Add in the number of potentially dirty inodes, because each inode
924  * write can dirty pagecache in the underlying blockdev.
925  */
926 static unsigned long get_nr_dirty_pages(void)
927 {
928         return global_page_state(NR_FILE_DIRTY) +
929                 global_page_state(NR_UNSTABLE_NFS) +
930                 get_nr_dirty_inodes();
931 }
932 
933 static long wb_check_background_flush(struct bdi_writeback *wb)
934 {
935         if (over_bground_thresh(wb->bdi)) {
936 
937                 struct wb_writeback_work work = {
938                         .nr_pages       = LONG_MAX,
939                         .sync_mode      = WB_SYNC_NONE,
940                         .for_background = 1,
941                         .range_cyclic   = 1,
942                         .reason         = WB_REASON_BACKGROUND,
943                 };
944 
945                 return wb_writeback(wb, &work);
946         }
947 
948         return 0;
949 }
950 
951 static long wb_check_old_data_flush(struct bdi_writeback *wb)
952 {
953         unsigned long expired;
954         long nr_pages;
955 
956         /*
957          * When set to zero, disable periodic writeback
958          */
959         if (!dirty_writeback_interval)
960                 return 0;
961 
962         expired = wb->last_old_flush +
963                         msecs_to_jiffies(dirty_writeback_interval * 10);
964         if (time_before(jiffies, expired))
965                 return 0;
966 
967         wb->last_old_flush = jiffies;
968         nr_pages = get_nr_dirty_pages();
969 
970         if (nr_pages) {
971                 struct wb_writeback_work work = {
972                         .nr_pages       = nr_pages,
973                         .sync_mode      = WB_SYNC_NONE,
974                         .for_kupdate    = 1,
975                         .range_cyclic   = 1,
976                         .reason         = WB_REASON_PERIODIC,
977                 };
978 
979                 return wb_writeback(wb, &work);
980         }
981 
982         return 0;
983 }
984 
985 /*
986  * Retrieve work items and do the writeback they describe
987  */
988 static long wb_do_writeback(struct bdi_writeback *wb)
989 {
990         struct backing_dev_info *bdi = wb->bdi;
991         struct wb_writeback_work *work;
992         long wrote = 0;
993 
994         set_bit(BDI_writeback_running, &wb->bdi->state);
995         while ((work = get_next_work_item(bdi)) != NULL) {
996 
997                 trace_writeback_exec(bdi, work);
998 
999                 wrote += wb_writeback(wb, work);
1000 
1001                 /*
1002                  * Notify the caller of completion if this is a synchronous
1003                  * work item, otherwise just free it.
1004                  */
1005                 if (work->done)
1006                         complete(work->done);
1007                 else
1008                         kfree(work);
1009         }
1010 
1011         /*
1012          * Check for periodic writeback, kupdated() style
1013          */
1014         wrote += wb_check_old_data_flush(wb);
1015         wrote += wb_check_background_flush(wb);
1016         clear_bit(BDI_writeback_running, &wb->bdi->state);
1017 
1018         return wrote;
1019 }
1020 
1021 /*
1022  * Handle writeback of dirty data for the device backed by this bdi. Also
1023  * reschedules periodically and does kupdated style flushing.
1024  */
1025 void bdi_writeback_workfn(struct work_struct *work)
1026 {
1027         struct bdi_writeback *wb = container_of(to_delayed_work(work),
1028                                                 struct bdi_writeback, dwork);
1029         struct backing_dev_info *bdi = wb->bdi;
1030         long pages_written;
1031 
1032         set_worker_desc("flush-%s", dev_name(bdi->dev));
1033         current->flags |= PF_SWAPWRITE;
1034 
1035         if (likely(!current_is_workqueue_rescuer() ||
1036                    !test_bit(BDI_registered, &bdi->state))) {
1037                 /*
1038                  * The normal path.  Keep writing back @bdi until its
1039                  * work_list is empty.  Note that this path is also taken
1040                  * if @bdi is shutting down even when we're running off the
1041                  * rescuer as work_list needs to be drained.
1042                  */
1043                 do {
1044                         pages_written = wb_do_writeback(wb);
1045                         trace_writeback_pages_written(pages_written);
1046                 } while (!list_empty(&bdi->work_list));
1047         } else {
1048                 /*
1049                  * bdi_wq can't get enough workers and we're running off
1050                  * the emergency worker.  Don't hog it.  Hopefully, 1024 is
1051                  * enough for efficient IO.
1052                  */
1053                 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1054                                                     WB_REASON_FORKER_THREAD);
1055                 trace_writeback_pages_written(pages_written);
1056         }
1057 
1058         if (!list_empty(&bdi->work_list))
1059                 mod_delayed_work(bdi_wq, &wb->dwork, 0);
1060         else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1061                 bdi_wakeup_thread_delayed(bdi);
1062 
1063         current->flags &= ~PF_SWAPWRITE;
1064 }
1065 
1066 /*
1067  * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
1068  * the whole world.
1069  */
1070 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1071 {
1072         struct backing_dev_info *bdi;
1073 
1074         if (!nr_pages)
1075                 nr_pages = get_nr_dirty_pages();
1076 
1077         rcu_read_lock();
1078         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1079                 if (!bdi_has_dirty_io(bdi))
1080                         continue;
1081                 __bdi_start_writeback(bdi, nr_pages, false, reason);
1082         }
1083         rcu_read_unlock();
1084 }
1085 
1086 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1087 {
1088         if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1089                 struct dentry *dentry;
1090                 const char *name = "?";
1091 
1092                 dentry = d_find_alias(inode);
1093                 if (dentry) {
1094                         spin_lock(&dentry->d_lock);
1095                         name = (const char *) dentry->d_name.name;
1096                 }
1097                 printk(KERN_DEBUG
1098                        "%s(%d): dirtied inode %lu (%s) on %s\n",
1099                        current->comm, task_pid_nr(current), inode->i_ino,
1100                        name, inode->i_sb->s_id);
1101                 if (dentry) {
1102                         spin_unlock(&dentry->d_lock);
1103                         dput(dentry);
1104                 }
1105         }
1106 }
1107 
1108 /**
1109  *      __mark_inode_dirty -    internal function
1110  *      @inode: inode to mark
1111  *      @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1112  *      Mark an inode as dirty. Callers should use mark_inode_dirty or
1113  *      mark_inode_dirty_sync.
1114  *
1115  * Put the inode on the super block's dirty list.
1116  *
1117  * CAREFUL! We mark it dirty unconditionally, but move it onto the
1118  * dirty list only if it is hashed or if it refers to a blockdev.
1119  * If it was not hashed, it will never be added to the dirty list
1120  * even if it is later hashed, as it will have been marked dirty already.
1121  *
1122  * In short, make sure you hash any inodes _before_ you start marking
1123  * them dirty.
1124  *
1125  * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1126  * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
1127  * the kernel-internal blockdev inode represents the dirtying time of the
1128  * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
1129  * page->mapping->host, so the page-dirtying time is recorded in the internal
1130  * blockdev inode.
1131  */
1132 void __mark_inode_dirty(struct inode *inode, int flags)
1133 {
1134         struct super_block *sb = inode->i_sb;
1135         struct backing_dev_info *bdi = NULL;
1136 
1137         /*
1138          * Don't do this for I_DIRTY_PAGES - that doesn't actually
1139          * dirty the inode itself
1140          */
1141         if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1142                 trace_writeback_dirty_inode_start(inode, flags);
1143 
1144                 if (sb->s_op->dirty_inode)
1145                         sb->s_op->dirty_inode(inode, flags);
1146 
1147                 trace_writeback_dirty_inode(inode, flags);
1148         }
1149 
1150         /*
1151          * make sure that changes are seen by all cpus before we test i_state
1152          * -- mikulas
1153          */
1154         smp_mb();
1155 
1156         /* avoid the locking if we can */
1157         if ((inode->i_state & flags) == flags)
1158                 return;
1159 
1160         if (unlikely(block_dump))
1161                 block_dump___mark_inode_dirty(inode);
1162 
1163         spin_lock(&inode->i_lock);
1164         if ((inode->i_state & flags) != flags) {
1165                 const int was_dirty = inode->i_state & I_DIRTY;
1166 
1167                 inode->i_state |= flags;
1168 
1169                 /*
1170                  * If the inode is being synced, just update its dirty state.
1171                  * The unlocker will place the inode on the appropriate
1172                  * superblock list, based upon its state.
1173                  */
1174                 if (inode->i_state & I_SYNC)
1175                         goto out_unlock_inode;
1176 
1177                 /*
1178                  * Only add valid (hashed) inodes to the superblock's
1179                  * dirty list.  Add blockdev inodes as well.
1180                  */
1181                 if (!S_ISBLK(inode->i_mode)) {
1182                         if (inode_unhashed(inode))
1183                                 goto out_unlock_inode;
1184                 }
1185                 if (inode->i_state & I_FREEING)
1186                         goto out_unlock_inode;
1187 
1188                 /*
1189                  * If the inode was already on b_dirty/b_io/b_more_io, don't
1190                  * reposition it (that would break b_dirty time-ordering).
1191                  */
1192                 if (!was_dirty) {
1193                         bool wakeup_bdi = false;
1194                         bdi = inode_to_bdi(inode);
1195 
1196                         spin_unlock(&inode->i_lock);
1197                         spin_lock(&bdi->wb.list_lock);
1198                         if (bdi_cap_writeback_dirty(bdi)) {
1199                                 WARN(!test_bit(BDI_registered, &bdi->state),
1200                                      "bdi-%s not registered\n", bdi->name);
1201 
1202                                 /*
1203                                  * If this is the first dirty inode for this
1204                                  * bdi, we have to wake-up the corresponding
1205                                  * bdi thread to make sure background
1206                                  * write-back happens later.
1207                                  */
1208                                 if (!wb_has_dirty_io(&bdi->wb))
1209                                         wakeup_bdi = true;
1210                         }
1211 
1212                         inode->dirtied_when = jiffies;
1213                         list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1214                         spin_unlock(&bdi->wb.list_lock);
1215 
1216                         if (wakeup_bdi)
1217                                 bdi_wakeup_thread_delayed(bdi);
1218                         return;
1219                 }
1220         }
1221 out_unlock_inode:
1222         spin_unlock(&inode->i_lock);
1223 
1224 }
1225 EXPORT_SYMBOL(__mark_inode_dirty);
1226 
1227 static void wait_sb_inodes(struct super_block *sb)
1228 {
1229         struct inode *inode, *old_inode = NULL;
1230 
1231         /*
1232          * We need to be protected against the filesystem going from
1233          * r/o to r/w or vice versa.
1234          */
1235         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1236 
1237         spin_lock(&inode_sb_list_lock);
1238 
1239         /*
1240          * Data integrity sync. Must wait for all pages under writeback,
1241          * because there may have been pages dirtied before our sync
1242          * call, but which had writeout started before we write it out.
1243          * In which case, the inode may not be on the dirty list, but
1244          * we still have to wait for that writeout.
1245          */
1246         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1247                 struct address_space *mapping = inode->i_mapping;
1248 
1249                 spin_lock(&inode->i_lock);
1250                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1251                     (mapping->nrpages == 0)) {
1252                         spin_unlock(&inode->i_lock);
1253                         continue;
1254                 }
1255                 __iget(inode);
1256                 spin_unlock(&inode->i_lock);
1257                 spin_unlock(&inode_sb_list_lock);
1258 
1259                 /*
1260                  * We hold a reference to 'inode' so it couldn't have been
1261                  * removed from s_inodes list while we dropped the
1262                  * inode_sb_list_lock.  We cannot iput the inode now as we can
1263                  * be holding the last reference and we cannot iput it under
1264                  * inode_sb_list_lock. So we keep the reference and iput it
1265                  * later.
1266                  */
1267                 iput(old_inode);
1268                 old_inode = inode;
1269 
1270                 filemap_fdatawait(mapping);
1271 
1272                 cond_resched();
1273 
1274                 spin_lock(&inode_sb_list_lock);
1275         }
1276         spin_unlock(&inode_sb_list_lock);
1277         iput(old_inode);
1278 }
1279 
1280 /**
1281  * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
1282  * @sb: the superblock
1283  * @nr: the number of pages to write
1284  * @reason: reason why some writeback work initiated
1285  *
1286  * Start writeback on some inodes on this super_block. No guarantees are made
1287  * on how many (if any) will be written, and this function does not wait
1288  * for IO completion of submitted IO.
1289  */
1290 void writeback_inodes_sb_nr(struct super_block *sb,
1291                             unsigned long nr,
1292                             enum wb_reason reason)
1293 {
1294         DECLARE_COMPLETION_ONSTACK(done);
1295         struct wb_writeback_work work = {
1296                 .sb                     = sb,
1297                 .sync_mode              = WB_SYNC_NONE,
1298                 .tagged_writepages      = 1,
1299                 .done                   = &done,
1300                 .nr_pages               = nr,
1301                 .reason                 = reason,
1302         };
1303 
1304         if (sb->s_bdi == &noop_backing_dev_info)
1305                 return;
1306         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1307         bdi_queue_work(sb->s_bdi, &work);
1308         wait_for_completion(&done);
1309 }
1310 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1311 
1312 /**
1313  * writeback_inodes_sb  -       writeback dirty inodes from given super_block
1314  * @sb: the superblock
1315  * @reason: reason why some writeback work was initiated
1316  *
1317  * Start writeback on some inodes on this super_block. No guarantees are made
1318  * on how many (if any) will be written, and this function does not wait
1319  * for IO completion of submitted IO.
1320  */
1321 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1322 {
1323         return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1324 }
1325 EXPORT_SYMBOL(writeback_inodes_sb);
1326 
1327 /**
1328  * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1329  * @sb: the superblock
1330  * @nr: the number of pages to write
1331  * @reason: the reason of writeback
1332  *
1333  * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1334  * Returns 1 if writeback was started, 0 if not.
1335  */
1336 int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1337                                   unsigned long nr,
1338                                   enum wb_reason reason)
1339 {
1340         if (writeback_in_progress(sb->s_bdi))
1341                 return 1;
1342 
1343         if (!down_read_trylock(&sb->s_umount))
1344                 return 0;
1345 
1346         writeback_inodes_sb_nr(sb, nr, reason);
1347         up_read(&sb->s_umount);
1348         return 1;
1349 }
1350 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1351 
1352 /**
1353  * try_to_writeback_inodes_sb - try to start writeback if none underway
1354  * @sb: the superblock
1355  * @reason: reason why some writeback work was initiated
1356  *
1357  * Implement by try_to_writeback_inodes_sb_nr()
1358  * Returns 1 if writeback was started, 0 if not.
1359  */
1360 int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1361 {
1362         return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1363 }
1364 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
1365 
1366 /**
1367  * sync_inodes_sb       -       sync sb inode pages
1368  * @sb: the superblock
1369  *
1370  * This function writes and waits on any dirty inode belonging to this
1371  * super_block.
1372  */
1373 void sync_inodes_sb(struct super_block *sb)
1374 {
1375         DECLARE_COMPLETION_ONSTACK(done);
1376         struct wb_writeback_work work = {
1377                 .sb             = sb,
1378                 .sync_mode      = WB_SYNC_ALL,
1379                 .nr_pages       = LONG_MAX,
1380                 .range_cyclic   = 0,
1381                 .done           = &done,
1382                 .reason         = WB_REASON_SYNC,
1383                 .for_sync       = 1,
1384         };
1385 
1386         /* Nothing to do? */
1387         if (sb->s_bdi == &noop_backing_dev_info)
1388                 return;
1389         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1390 
1391         bdi_queue_work(sb->s_bdi, &work);
1392         wait_for_completion(&done);
1393 
1394         wait_sb_inodes(sb);
1395 }
1396 EXPORT_SYMBOL(sync_inodes_sb);
1397 
1398 /**
1399  * write_inode_now      -       write an inode to disk
1400  * @inode: inode to write to disk
1401  * @sync: whether the write should be synchronous or not
1402  *
1403  * This function commits an inode to disk immediately if it is dirty. This is
1404  * primarily needed by knfsd.
1405  *
1406  * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1407  */
1408 int write_inode_now(struct inode *inode, int sync)
1409 {
1410         struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1411         struct writeback_control wbc = {
1412                 .nr_to_write = LONG_MAX,
1413                 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1414                 .range_start = 0,
1415                 .range_end = LLONG_MAX,
1416         };
1417 
1418         if (!mapping_cap_writeback_dirty(inode->i_mapping))
1419                 wbc.nr_to_write = 0;
1420 
1421         might_sleep();
1422         return writeback_single_inode(inode, wb, &wbc);
1423 }
1424 EXPORT_SYMBOL(write_inode_now);
1425 
1426 /**
1427  * sync_inode - write an inode and its pages to disk.
1428  * @inode: the inode to sync
1429  * @wbc: controls the writeback mode
1430  *
1431  * sync_inode() will write an inode and its pages to disk.  It will also
1432  * correctly update the inode on its superblock's dirty inode lists and will
1433  * update inode->i_state.
1434  *
1435  * The caller must have a ref on the inode.
1436  */
1437 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1438 {
1439         return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1440 }
1441 EXPORT_SYMBOL(sync_inode);
1442 
1443 /**
1444  * sync_inode_metadata - write an inode to disk
1445  * @inode: the inode to sync
1446  * @wait: wait for I/O to complete.
1447  *
1448  * Write an inode to disk and adjust its dirty state after completion.
1449  *
1450  * Note: only writes the actual inode, no associated data or other metadata.
1451  */
1452 int sync_inode_metadata(struct inode *inode, int wait)
1453 {
1454         struct writeback_control wbc = {
1455                 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1456                 .nr_to_write = 0, /* metadata-only */
1457         };
1458 
1459         return sync_inode(inode, &wbc);
1460 }
1461 EXPORT_SYMBOL(sync_inode_metadata);
1462 

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