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

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

Version: ~ [ linux-5.4.2 ] ~ [ linux-5.3.15 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.88 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.158 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.206 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.206 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.78 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * fs/fs-writeback.c
  3  *
  4  * Copyright (C) 2002, Linus Torvalds.
  5  *
  6  * Contains all the functions related to writing back and waiting
  7  * upon dirty inodes against superblocks, and writing back dirty
  8  * pages against inodes.  ie: data writeback.  Writeout of the
  9  * inode itself is not handled here.
 10  *
 11  * 10Apr2002    Andrew Morton
 12  *              Split out of fs/inode.c
 13  *              Additions for address_space-based writeback
 14  */
 15 
 16 #include <linux/kernel.h>
 17 #include <linux/export.h>
 18 #include <linux/spinlock.h>
 19 #include <linux/slab.h>
 20 #include <linux/sched.h>
 21 #include <linux/fs.h>
 22 #include <linux/mm.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/kthread.h>
 25 #include <linux/freezer.h>
 26 #include <linux/writeback.h>
 27 #include <linux/blkdev.h>
 28 #include <linux/backing-dev.h>
 29 #include <linux/tracepoint.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         enum wb_reason reason;          /* why was writeback initiated? */
 50 
 51         struct list_head list;          /* pending work list */
 52         struct completion *done;        /* set if the caller waits */
 53 };
 54 
 55 /**
 56  * writeback_in_progress - determine whether there is writeback in progress
 57  * @bdi: the device's backing_dev_info structure.
 58  *
 59  * Determine whether there is writeback waiting to be handled against a
 60  * backing device.
 61  */
 62 int writeback_in_progress(struct backing_dev_info *bdi)
 63 {
 64         return test_bit(BDI_writeback_running, &bdi->state);
 65 }
 66 EXPORT_SYMBOL(writeback_in_progress);
 67 
 68 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
 69 {
 70         struct super_block *sb = inode->i_sb;
 71 
 72         if (strcmp(sb->s_type->name, "bdev") == 0)
 73                 return inode->i_mapping->backing_dev_info;
 74 
 75         return sb->s_bdi;
 76 }
 77 
 78 static inline struct inode *wb_inode(struct list_head *head)
 79 {
 80         return list_entry(head, struct inode, i_wb_list);
 81 }
 82 
 83 /*
 84  * Include the creation of the trace points after defining the
 85  * wb_writeback_work structure and inline functions so that the definition
 86  * remains local to this file.
 87  */
 88 #define CREATE_TRACE_POINTS
 89 #include <trace/events/writeback.h>
 90 
 91 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
 92 static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
 93 {
 94         if (bdi->wb.task) {
 95                 wake_up_process(bdi->wb.task);
 96         } else {
 97                 /*
 98                  * The bdi thread isn't there, wake up the forker thread which
 99                  * will create and run it.
100                  */
101                 wake_up_process(default_backing_dev_info.wb.task);
102         }
103 }
104 
105 static void bdi_queue_work(struct backing_dev_info *bdi,
106                            struct wb_writeback_work *work)
107 {
108         trace_writeback_queue(bdi, work);
109 
110         spin_lock_bh(&bdi->wb_lock);
111         list_add_tail(&work->list, &bdi->work_list);
112         if (!bdi->wb.task)
113                 trace_writeback_nothread(bdi, work);
114         bdi_wakeup_flusher(bdi);
115         spin_unlock_bh(&bdi->wb_lock);
116 }
117 
118 static void
119 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
120                       bool range_cyclic, enum wb_reason reason)
121 {
122         struct wb_writeback_work *work;
123 
124         /*
125          * This is WB_SYNC_NONE writeback, so if allocation fails just
126          * wakeup the thread for old dirty data writeback
127          */
128         work = kzalloc(sizeof(*work), GFP_ATOMIC);
129         if (!work) {
130                 if (bdi->wb.task) {
131                         trace_writeback_nowork(bdi);
132                         wake_up_process(bdi->wb.task);
133                 }
134                 return;
135         }
136 
137         work->sync_mode = WB_SYNC_NONE;
138         work->nr_pages  = nr_pages;
139         work->range_cyclic = range_cyclic;
140         work->reason    = reason;
141 
142         bdi_queue_work(bdi, work);
143 }
144 
145 /**
146  * bdi_start_writeback - start writeback
147  * @bdi: the backing device to write from
148  * @nr_pages: the number of pages to write
149  * @reason: reason why some writeback work was initiated
150  *
151  * Description:
152  *   This does WB_SYNC_NONE opportunistic writeback. The IO is only
153  *   started when this function returns, we make no guarantees on
154  *   completion. Caller need not hold sb s_umount semaphore.
155  *
156  */
157 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
158                         enum wb_reason reason)
159 {
160         __bdi_start_writeback(bdi, nr_pages, true, reason);
161 }
162 
163 /**
164  * bdi_start_background_writeback - start background writeback
165  * @bdi: the backing device to write from
166  *
167  * Description:
168  *   This makes sure WB_SYNC_NONE background writeback happens. When
169  *   this function returns, it is only guaranteed that for given BDI
170  *   some IO is happening if we are over background dirty threshold.
171  *   Caller need not hold sb s_umount semaphore.
172  */
173 void bdi_start_background_writeback(struct backing_dev_info *bdi)
174 {
175         /*
176          * We just wake up the flusher thread. It will perform background
177          * writeback as soon as there is no other work to do.
178          */
179         trace_writeback_wake_background(bdi);
180         spin_lock_bh(&bdi->wb_lock);
181         bdi_wakeup_flusher(bdi);
182         spin_unlock_bh(&bdi->wb_lock);
183 }
184 
185 /*
186  * Remove the inode from the writeback list it is on.
187  */
188 void inode_wb_list_del(struct inode *inode)
189 {
190         struct backing_dev_info *bdi = inode_to_bdi(inode);
191 
192         spin_lock(&bdi->wb.list_lock);
193         list_del_init(&inode->i_wb_list);
194         spin_unlock(&bdi->wb.list_lock);
195 }
196 
197 /*
198  * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
199  * furthest end of its superblock's dirty-inode list.
200  *
201  * Before stamping the inode's ->dirtied_when, we check to see whether it is
202  * already the most-recently-dirtied inode on the b_dirty list.  If that is
203  * the case then the inode must have been redirtied while it was being written
204  * out and we don't reset its dirtied_when.
205  */
206 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
207 {
208         assert_spin_locked(&wb->list_lock);
209         if (!list_empty(&wb->b_dirty)) {
210                 struct inode *tail;
211 
212                 tail = wb_inode(wb->b_dirty.next);
213                 if (time_before(inode->dirtied_when, tail->dirtied_when))
214                         inode->dirtied_when = jiffies;
215         }
216         list_move(&inode->i_wb_list, &wb->b_dirty);
217 }
218 
219 /*
220  * requeue inode for re-scanning after bdi->b_io list is exhausted.
221  */
222 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
223 {
224         assert_spin_locked(&wb->list_lock);
225         list_move(&inode->i_wb_list, &wb->b_more_io);
226 }
227 
228 static void inode_sync_complete(struct inode *inode)
229 {
230         inode->i_state &= ~I_SYNC;
231         /* If inode is clean an unused, put it into LRU now... */
232         inode_add_lru(inode);
233         /* Waiters must see I_SYNC cleared before being woken up */
234         smp_mb();
235         wake_up_bit(&inode->i_state, __I_SYNC);
236 }
237 
238 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
239 {
240         bool ret = time_after(inode->dirtied_when, t);
241 #ifndef CONFIG_64BIT
242         /*
243          * For inodes being constantly redirtied, dirtied_when can get stuck.
244          * It _appears_ to be in the future, but is actually in distant past.
245          * This test is necessary to prevent such wrapped-around relative times
246          * from permanently stopping the whole bdi writeback.
247          */
248         ret = ret && time_before_eq(inode->dirtied_when, jiffies);
249 #endif
250         return ret;
251 }
252 
253 /*
254  * Move expired (dirtied before work->older_than_this) dirty inodes from
255  * @delaying_queue to @dispatch_queue.
256  */
257 static int move_expired_inodes(struct list_head *delaying_queue,
258                                struct list_head *dispatch_queue,
259                                struct wb_writeback_work *work)
260 {
261         LIST_HEAD(tmp);
262         struct list_head *pos, *node;
263         struct super_block *sb = NULL;
264         struct inode *inode;
265         int do_sb_sort = 0;
266         int moved = 0;
267 
268         while (!list_empty(delaying_queue)) {
269                 inode = wb_inode(delaying_queue->prev);
270                 if (work->older_than_this &&
271                     inode_dirtied_after(inode, *work->older_than_this))
272                         break;
273                 if (sb && sb != inode->i_sb)
274                         do_sb_sort = 1;
275                 sb = inode->i_sb;
276                 list_move(&inode->i_wb_list, &tmp);
277                 moved++;
278         }
279 
280         /* just one sb in list, splice to dispatch_queue and we're done */
281         if (!do_sb_sort) {
282                 list_splice(&tmp, dispatch_queue);
283                 goto out;
284         }
285 
286         /* Move inodes from one superblock together */
287         while (!list_empty(&tmp)) {
288                 sb = wb_inode(tmp.prev)->i_sb;
289                 list_for_each_prev_safe(pos, node, &tmp) {
290                         inode = wb_inode(pos);
291                         if (inode->i_sb == sb)
292                                 list_move(&inode->i_wb_list, dispatch_queue);
293                 }
294         }
295 out:
296         return moved;
297 }
298 
299 /*
300  * Queue all expired dirty inodes for io, eldest first.
301  * Before
302  *         newly dirtied     b_dirty    b_io    b_more_io
303  *         =============>    gf         edc     BA
304  * After
305  *         newly dirtied     b_dirty    b_io    b_more_io
306  *         =============>    g          fBAedc
307  *                                           |
308  *                                           +--> dequeue for IO
309  */
310 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
311 {
312         int moved;
313         assert_spin_locked(&wb->list_lock);
314         list_splice_init(&wb->b_more_io, &wb->b_io);
315         moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
316         trace_writeback_queue_io(wb, work, moved);
317 }
318 
319 static int write_inode(struct inode *inode, struct writeback_control *wbc)
320 {
321         int ret;
322 
323         if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
324                 trace_writeback_write_inode_start(inode, wbc);
325                 ret = inode->i_sb->s_op->write_inode(inode, wbc);
326                 trace_writeback_write_inode(inode, wbc);
327                 return ret;
328         }
329         return 0;
330 }
331 
332 /*
333  * Wait for writeback on an inode to complete. Called with i_lock held.
334  * Caller must make sure inode cannot go away when we drop i_lock.
335  */
336 static void __inode_wait_for_writeback(struct inode *inode)
337         __releases(inode->i_lock)
338         __acquires(inode->i_lock)
339 {
340         DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
341         wait_queue_head_t *wqh;
342 
343         wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
344         while (inode->i_state & I_SYNC) {
345                 spin_unlock(&inode->i_lock);
346                 __wait_on_bit(wqh, &wq, inode_wait, 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.
467          */
468         if (wbc->sync_mode == WB_SYNC_ALL) {
469                 int err = filemap_fdatawait(mapping);
470                 if (ret == 0)
471                         ret = err;
472         }
473 
474         /*
475          * Some filesystems may redirty the inode during the writeback
476          * due to delalloc, clear dirty metadata flags right before
477          * write_inode()
478          */
479         spin_lock(&inode->i_lock);
480         /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
481         if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
482                 inode->i_state &= ~I_DIRTY_PAGES;
483         dirty = inode->i_state & I_DIRTY;
484         inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
485         spin_unlock(&inode->i_lock);
486         /* Don't write the inode if only I_DIRTY_PAGES was set */
487         if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
488                 int err = write_inode(inode, wbc);
489                 if (ret == 0)
490                         ret = err;
491         }
492         trace_writeback_single_inode(inode, wbc, nr_to_write);
493         return ret;
494 }
495 
496 /*
497  * Write out an inode's dirty pages. Either the caller has an active reference
498  * on the inode or the inode has I_WILL_FREE set.
499  *
500  * This function is designed to be called for writing back one inode which
501  * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
502  * and does more profound writeback list handling in writeback_sb_inodes().
503  */
504 static int
505 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
506                        struct writeback_control *wbc)
507 {
508         int ret = 0;
509 
510         spin_lock(&inode->i_lock);
511         if (!atomic_read(&inode->i_count))
512                 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
513         else
514                 WARN_ON(inode->i_state & I_WILL_FREE);
515 
516         if (inode->i_state & I_SYNC) {
517                 if (wbc->sync_mode != WB_SYNC_ALL)
518                         goto out;
519                 /*
520                  * It's a data-integrity sync. We must wait. Since callers hold
521                  * inode reference or inode has I_WILL_FREE set, it cannot go
522                  * away under us.
523                  */
524                 __inode_wait_for_writeback(inode);
525         }
526         WARN_ON(inode->i_state & I_SYNC);
527         /*
528          * Skip inode if it is clean. We don't want to mess with writeback
529          * lists in this function since flusher thread may be doing for example
530          * sync in parallel and if we move the inode, it could get skipped. So
531          * here we make sure inode is on some writeback list and leave it there
532          * unless we have completely cleaned the inode.
533          */
534         if (!(inode->i_state & I_DIRTY))
535                 goto out;
536         inode->i_state |= I_SYNC;
537         spin_unlock(&inode->i_lock);
538 
539         ret = __writeback_single_inode(inode, wbc);
540 
541         spin_lock(&wb->list_lock);
542         spin_lock(&inode->i_lock);
543         /*
544          * If inode is clean, remove it from writeback lists. Otherwise don't
545          * touch it. See comment above for explanation.
546          */
547         if (!(inode->i_state & I_DIRTY))
548                 list_del_init(&inode->i_wb_list);
549         spin_unlock(&wb->list_lock);
550         inode_sync_complete(inode);
551 out:
552         spin_unlock(&inode->i_lock);
553         return ret;
554 }
555 
556 static long writeback_chunk_size(struct backing_dev_info *bdi,
557                                  struct wb_writeback_work *work)
558 {
559         long pages;
560 
561         /*
562          * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
563          * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
564          * here avoids calling into writeback_inodes_wb() more than once.
565          *
566          * The intended call sequence for WB_SYNC_ALL writeback is:
567          *
568          *      wb_writeback()
569          *          writeback_sb_inodes()       <== called only once
570          *              write_cache_pages()     <== called once for each inode
571          *                   (quickly) tag currently dirty pages
572          *                   (maybe slowly) sync all tagged pages
573          */
574         if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
575                 pages = LONG_MAX;
576         else {
577                 pages = min(bdi->avg_write_bandwidth / 2,
578                             global_dirty_limit / DIRTY_SCOPE);
579                 pages = min(pages, work->nr_pages);
580                 pages = round_down(pages + MIN_WRITEBACK_PAGES,
581                                    MIN_WRITEBACK_PAGES);
582         }
583 
584         return pages;
585 }
586 
587 /*
588  * Write a portion of b_io inodes which belong to @sb.
589  *
590  * Return the number of pages and/or inodes written.
591  */
592 static long writeback_sb_inodes(struct super_block *sb,
593                                 struct bdi_writeback *wb,
594                                 struct wb_writeback_work *work)
595 {
596         struct writeback_control wbc = {
597                 .sync_mode              = work->sync_mode,
598                 .tagged_writepages      = work->tagged_writepages,
599                 .for_kupdate            = work->for_kupdate,
600                 .for_background         = work->for_background,
601                 .range_cyclic           = work->range_cyclic,
602                 .range_start            = 0,
603                 .range_end              = LLONG_MAX,
604         };
605         unsigned long start_time = jiffies;
606         long write_chunk;
607         long wrote = 0;  /* count both pages and inodes */
608 
609         while (!list_empty(&wb->b_io)) {
610                 struct inode *inode = wb_inode(wb->b_io.prev);
611 
612                 if (inode->i_sb != sb) {
613                         if (work->sb) {
614                                 /*
615                                  * We only want to write back data for this
616                                  * superblock, move all inodes not belonging
617                                  * to it back onto the dirty list.
618                                  */
619                                 redirty_tail(inode, wb);
620                                 continue;
621                         }
622 
623                         /*
624                          * The inode belongs to a different superblock.
625                          * Bounce back to the caller to unpin this and
626                          * pin the next superblock.
627                          */
628                         break;
629                 }
630 
631                 /*
632                  * Don't bother with new inodes or inodes being freed, first
633                  * kind does not need periodic writeout yet, and for the latter
634                  * kind writeout is handled by the freer.
635                  */
636                 spin_lock(&inode->i_lock);
637                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
638                         spin_unlock(&inode->i_lock);
639                         redirty_tail(inode, wb);
640                         continue;
641                 }
642                 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
643                         /*
644                          * If this inode is locked for writeback and we are not
645                          * doing writeback-for-data-integrity, move it to
646                          * b_more_io so that writeback can proceed with the
647                          * other inodes on s_io.
648                          *
649                          * We'll have another go at writing back this inode
650                          * when we completed a full scan of b_io.
651                          */
652                         spin_unlock(&inode->i_lock);
653                         requeue_io(inode, wb);
654                         trace_writeback_sb_inodes_requeue(inode);
655                         continue;
656                 }
657                 spin_unlock(&wb->list_lock);
658 
659                 /*
660                  * We already requeued the inode if it had I_SYNC set and we
661                  * are doing WB_SYNC_NONE writeback. So this catches only the
662                  * WB_SYNC_ALL case.
663                  */
664                 if (inode->i_state & I_SYNC) {
665                         /* Wait for I_SYNC. This function drops i_lock... */
666                         inode_sleep_on_writeback(inode);
667                         /* Inode may be gone, start again */
668                         spin_lock(&wb->list_lock);
669                         continue;
670                 }
671                 inode->i_state |= I_SYNC;
672                 spin_unlock(&inode->i_lock);
673 
674                 write_chunk = writeback_chunk_size(wb->bdi, work);
675                 wbc.nr_to_write = write_chunk;
676                 wbc.pages_skipped = 0;
677 
678                 /*
679                  * We use I_SYNC to pin the inode in memory. While it is set
680                  * evict_inode() will wait so the inode cannot be freed.
681                  */
682                 __writeback_single_inode(inode, &wbc);
683 
684                 work->nr_pages -= write_chunk - wbc.nr_to_write;
685                 wrote += write_chunk - wbc.nr_to_write;
686                 spin_lock(&wb->list_lock);
687                 spin_lock(&inode->i_lock);
688                 if (!(inode->i_state & I_DIRTY))
689                         wrote++;
690                 requeue_inode(inode, wb, &wbc);
691                 inode_sync_complete(inode);
692                 spin_unlock(&inode->i_lock);
693                 cond_resched_lock(&wb->list_lock);
694                 /*
695                  * bail out to wb_writeback() often enough to check
696                  * background threshold and other termination conditions.
697                  */
698                 if (wrote) {
699                         if (time_is_before_jiffies(start_time + HZ / 10UL))
700                                 break;
701                         if (work->nr_pages <= 0)
702                                 break;
703                 }
704         }
705         return wrote;
706 }
707 
708 static long __writeback_inodes_wb(struct bdi_writeback *wb,
709                                   struct wb_writeback_work *work)
710 {
711         unsigned long start_time = jiffies;
712         long wrote = 0;
713 
714         while (!list_empty(&wb->b_io)) {
715                 struct inode *inode = wb_inode(wb->b_io.prev);
716                 struct super_block *sb = inode->i_sb;
717 
718                 if (!grab_super_passive(sb)) {
719                         /*
720                          * grab_super_passive() may fail consistently due to
721                          * s_umount being grabbed by someone else. Don't use
722                          * requeue_io() to avoid busy retrying the inode/sb.
723                          */
724                         redirty_tail(inode, wb);
725                         continue;
726                 }
727                 wrote += writeback_sb_inodes(sb, wb, work);
728                 drop_super(sb);
729 
730                 /* refer to the same tests at the end of writeback_sb_inodes */
731                 if (wrote) {
732                         if (time_is_before_jiffies(start_time + HZ / 10UL))
733                                 break;
734                         if (work->nr_pages <= 0)
735                                 break;
736                 }
737         }
738         /* Leave any unwritten inodes on b_io */
739         return wrote;
740 }
741 
742 long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
743                                 enum wb_reason reason)
744 {
745         struct wb_writeback_work work = {
746                 .nr_pages       = nr_pages,
747                 .sync_mode      = WB_SYNC_NONE,
748                 .range_cyclic   = 1,
749                 .reason         = reason,
750         };
751 
752         spin_lock(&wb->list_lock);
753         if (list_empty(&wb->b_io))
754                 queue_io(wb, &work);
755         __writeback_inodes_wb(wb, &work);
756         spin_unlock(&wb->list_lock);
757 
758         return nr_pages - work.nr_pages;
759 }
760 
761 static bool over_bground_thresh(struct backing_dev_info *bdi)
762 {
763         unsigned long background_thresh, dirty_thresh;
764 
765         global_dirty_limits(&background_thresh, &dirty_thresh);
766 
767         if (global_page_state(NR_FILE_DIRTY) +
768             global_page_state(NR_UNSTABLE_NFS) > background_thresh)
769                 return true;
770 
771         if (bdi_stat(bdi, BDI_RECLAIMABLE) >
772                                 bdi_dirty_limit(bdi, background_thresh))
773                 return true;
774 
775         return false;
776 }
777 
778 /*
779  * Called under wb->list_lock. If there are multiple wb per bdi,
780  * only the flusher working on the first wb should do it.
781  */
782 static void wb_update_bandwidth(struct bdi_writeback *wb,
783                                 unsigned long start_time)
784 {
785         __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
786 }
787 
788 /*
789  * Explicit flushing or periodic writeback of "old" data.
790  *
791  * Define "old": the first time one of an inode's pages is dirtied, we mark the
792  * dirtying-time in the inode's address_space.  So this periodic writeback code
793  * just walks the superblock inode list, writing back any inodes which are
794  * older than a specific point in time.
795  *
796  * Try to run once per dirty_writeback_interval.  But if a writeback event
797  * takes longer than a dirty_writeback_interval interval, then leave a
798  * one-second gap.
799  *
800  * older_than_this takes precedence over nr_to_write.  So we'll only write back
801  * all dirty pages if they are all attached to "old" mappings.
802  */
803 static long wb_writeback(struct bdi_writeback *wb,
804                          struct wb_writeback_work *work)
805 {
806         unsigned long wb_start = jiffies;
807         long nr_pages = work->nr_pages;
808         unsigned long oldest_jif;
809         struct inode *inode;
810         long progress;
811 
812         oldest_jif = jiffies;
813         work->older_than_this = &oldest_jif;
814 
815         spin_lock(&wb->list_lock);
816         for (;;) {
817                 /*
818                  * Stop writeback when nr_pages has been consumed
819                  */
820                 if (work->nr_pages <= 0)
821                         break;
822 
823                 /*
824                  * Background writeout and kupdate-style writeback may
825                  * run forever. Stop them if there is other work to do
826                  * so that e.g. sync can proceed. They'll be restarted
827                  * after the other works are all done.
828                  */
829                 if ((work->for_background || work->for_kupdate) &&
830                     !list_empty(&wb->bdi->work_list))
831                         break;
832 
833                 /*
834                  * For background writeout, stop when we are below the
835                  * background dirty threshold
836                  */
837                 if (work->for_background && !over_bground_thresh(wb->bdi))
838                         break;
839 
840                 /*
841                  * Kupdate and background works are special and we want to
842                  * include all inodes that need writing. Livelock avoidance is
843                  * handled by these works yielding to any other work so we are
844                  * safe.
845                  */
846                 if (work->for_kupdate) {
847                         oldest_jif = jiffies -
848                                 msecs_to_jiffies(dirty_expire_interval * 10);
849                 } else if (work->for_background)
850                         oldest_jif = jiffies;
851 
852                 trace_writeback_start(wb->bdi, work);
853                 if (list_empty(&wb->b_io))
854                         queue_io(wb, work);
855                 if (work->sb)
856                         progress = writeback_sb_inodes(work->sb, wb, work);
857                 else
858                         progress = __writeback_inodes_wb(wb, work);
859                 trace_writeback_written(wb->bdi, work);
860 
861                 wb_update_bandwidth(wb, wb_start);
862 
863                 /*
864                  * Did we write something? Try for more
865                  *
866                  * Dirty inodes are moved to b_io for writeback in batches.
867                  * The completion of the current batch does not necessarily
868                  * mean the overall work is done. So we keep looping as long
869                  * as made some progress on cleaning pages or inodes.
870                  */
871                 if (progress)
872                         continue;
873                 /*
874                  * No more inodes for IO, bail
875                  */
876                 if (list_empty(&wb->b_more_io))
877                         break;
878                 /*
879                  * Nothing written. Wait for some inode to
880                  * become available for writeback. Otherwise
881                  * we'll just busyloop.
882                  */
883                 if (!list_empty(&wb->b_more_io))  {
884                         trace_writeback_wait(wb->bdi, work);
885                         inode = wb_inode(wb->b_more_io.prev);
886                         spin_lock(&inode->i_lock);
887                         spin_unlock(&wb->list_lock);
888                         /* This function drops i_lock... */
889                         inode_sleep_on_writeback(inode);
890                         spin_lock(&wb->list_lock);
891                 }
892         }
893         spin_unlock(&wb->list_lock);
894 
895         return nr_pages - work->nr_pages;
896 }
897 
898 /*
899  * Return the next wb_writeback_work struct that hasn't been processed yet.
900  */
901 static struct wb_writeback_work *
902 get_next_work_item(struct backing_dev_info *bdi)
903 {
904         struct wb_writeback_work *work = NULL;
905 
906         spin_lock_bh(&bdi->wb_lock);
907         if (!list_empty(&bdi->work_list)) {
908                 work = list_entry(bdi->work_list.next,
909                                   struct wb_writeback_work, list);
910                 list_del_init(&work->list);
911         }
912         spin_unlock_bh(&bdi->wb_lock);
913         return work;
914 }
915 
916 /*
917  * Add in the number of potentially dirty inodes, because each inode
918  * write can dirty pagecache in the underlying blockdev.
919  */
920 static unsigned long get_nr_dirty_pages(void)
921 {
922         return global_page_state(NR_FILE_DIRTY) +
923                 global_page_state(NR_UNSTABLE_NFS) +
924                 get_nr_dirty_inodes();
925 }
926 
927 static long wb_check_background_flush(struct bdi_writeback *wb)
928 {
929         if (over_bground_thresh(wb->bdi)) {
930 
931                 struct wb_writeback_work work = {
932                         .nr_pages       = LONG_MAX,
933                         .sync_mode      = WB_SYNC_NONE,
934                         .for_background = 1,
935                         .range_cyclic   = 1,
936                         .reason         = WB_REASON_BACKGROUND,
937                 };
938 
939                 return wb_writeback(wb, &work);
940         }
941 
942         return 0;
943 }
944 
945 static long wb_check_old_data_flush(struct bdi_writeback *wb)
946 {
947         unsigned long expired;
948         long nr_pages;
949 
950         /*
951          * When set to zero, disable periodic writeback
952          */
953         if (!dirty_writeback_interval)
954                 return 0;
955 
956         expired = wb->last_old_flush +
957                         msecs_to_jiffies(dirty_writeback_interval * 10);
958         if (time_before(jiffies, expired))
959                 return 0;
960 
961         wb->last_old_flush = jiffies;
962         nr_pages = get_nr_dirty_pages();
963 
964         if (nr_pages) {
965                 struct wb_writeback_work work = {
966                         .nr_pages       = nr_pages,
967                         .sync_mode      = WB_SYNC_NONE,
968                         .for_kupdate    = 1,
969                         .range_cyclic   = 1,
970                         .reason         = WB_REASON_PERIODIC,
971                 };
972 
973                 return wb_writeback(wb, &work);
974         }
975 
976         return 0;
977 }
978 
979 /*
980  * Retrieve work items and do the writeback they describe
981  */
982 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
983 {
984         struct backing_dev_info *bdi = wb->bdi;
985         struct wb_writeback_work *work;
986         long wrote = 0;
987 
988         set_bit(BDI_writeback_running, &wb->bdi->state);
989         while ((work = get_next_work_item(bdi)) != NULL) {
990                 /*
991                  * Override sync mode, in case we must wait for completion
992                  * because this thread is exiting now.
993                  */
994                 if (force_wait)
995                         work->sync_mode = WB_SYNC_ALL;
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  * wakes up periodically and does kupdated style flushing.
1024  */
1025 int bdi_writeback_thread(void *data)
1026 {
1027         struct bdi_writeback *wb = data;
1028         struct backing_dev_info *bdi = wb->bdi;
1029         long pages_written;
1030 
1031         current->flags |= PF_SWAPWRITE;
1032         set_freezable();
1033         wb->last_active = jiffies;
1034 
1035         /*
1036          * Our parent may run at a different priority, just set us to normal
1037          */
1038         set_user_nice(current, 0);
1039 
1040         trace_writeback_thread_start(bdi);
1041 
1042         while (!kthread_freezable_should_stop(NULL)) {
1043                 /*
1044                  * Remove own delayed wake-up timer, since we are already awake
1045                  * and we'll take care of the periodic write-back.
1046                  */
1047                 del_timer(&wb->wakeup_timer);
1048 
1049                 pages_written = wb_do_writeback(wb, 0);
1050 
1051                 trace_writeback_pages_written(pages_written);
1052 
1053                 if (pages_written)
1054                         wb->last_active = jiffies;
1055 
1056                 set_current_state(TASK_INTERRUPTIBLE);
1057                 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
1058                         __set_current_state(TASK_RUNNING);
1059                         continue;
1060                 }
1061 
1062                 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1063                         schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
1064                 else {
1065                         /*
1066                          * We have nothing to do, so can go sleep without any
1067                          * timeout and save power. When a work is queued or
1068                          * something is made dirty - we will be woken up.
1069                          */
1070                         schedule();
1071                 }
1072         }
1073 
1074         /* Flush any work that raced with us exiting */
1075         if (!list_empty(&bdi->work_list))
1076                 wb_do_writeback(wb, 1);
1077 
1078         trace_writeback_thread_stop(bdi);
1079         return 0;
1080 }
1081 
1082 
1083 /*
1084  * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
1085  * the whole world.
1086  */
1087 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1088 {
1089         struct backing_dev_info *bdi;
1090 
1091         if (!nr_pages) {
1092                 nr_pages = global_page_state(NR_FILE_DIRTY) +
1093                                 global_page_state(NR_UNSTABLE_NFS);
1094         }
1095 
1096         rcu_read_lock();
1097         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1098                 if (!bdi_has_dirty_io(bdi))
1099                         continue;
1100                 __bdi_start_writeback(bdi, nr_pages, false, reason);
1101         }
1102         rcu_read_unlock();
1103 }
1104 
1105 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1106 {
1107         if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1108                 struct dentry *dentry;
1109                 const char *name = "?";
1110 
1111                 dentry = d_find_alias(inode);
1112                 if (dentry) {
1113                         spin_lock(&dentry->d_lock);
1114                         name = (const char *) dentry->d_name.name;
1115                 }
1116                 printk(KERN_DEBUG
1117                        "%s(%d): dirtied inode %lu (%s) on %s\n",
1118                        current->comm, task_pid_nr(current), inode->i_ino,
1119                        name, inode->i_sb->s_id);
1120                 if (dentry) {
1121                         spin_unlock(&dentry->d_lock);
1122                         dput(dentry);
1123                 }
1124         }
1125 }
1126 
1127 /**
1128  *      __mark_inode_dirty -    internal function
1129  *      @inode: inode to mark
1130  *      @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1131  *      Mark an inode as dirty. Callers should use mark_inode_dirty or
1132  *      mark_inode_dirty_sync.
1133  *
1134  * Put the inode on the super block's dirty list.
1135  *
1136  * CAREFUL! We mark it dirty unconditionally, but move it onto the
1137  * dirty list only if it is hashed or if it refers to a blockdev.
1138  * If it was not hashed, it will never be added to the dirty list
1139  * even if it is later hashed, as it will have been marked dirty already.
1140  *
1141  * In short, make sure you hash any inodes _before_ you start marking
1142  * them dirty.
1143  *
1144  * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1145  * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
1146  * the kernel-internal blockdev inode represents the dirtying time of the
1147  * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
1148  * page->mapping->host, so the page-dirtying time is recorded in the internal
1149  * blockdev inode.
1150  */
1151 void __mark_inode_dirty(struct inode *inode, int flags)
1152 {
1153         struct super_block *sb = inode->i_sb;
1154         struct backing_dev_info *bdi = NULL;
1155 
1156         /*
1157          * Don't do this for I_DIRTY_PAGES - that doesn't actually
1158          * dirty the inode itself
1159          */
1160         if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1161                 trace_writeback_dirty_inode_start(inode, flags);
1162 
1163                 if (sb->s_op->dirty_inode)
1164                         sb->s_op->dirty_inode(inode, flags);
1165 
1166                 trace_writeback_dirty_inode(inode, flags);
1167         }
1168 
1169         /*
1170          * make sure that changes are seen by all cpus before we test i_state
1171          * -- mikulas
1172          */
1173         smp_mb();
1174 
1175         /* avoid the locking if we can */
1176         if ((inode->i_state & flags) == flags)
1177                 return;
1178 
1179         if (unlikely(block_dump))
1180                 block_dump___mark_inode_dirty(inode);
1181 
1182         spin_lock(&inode->i_lock);
1183         if ((inode->i_state & flags) != flags) {
1184                 const int was_dirty = inode->i_state & I_DIRTY;
1185 
1186                 inode->i_state |= flags;
1187 
1188                 /*
1189                  * If the inode is being synced, just update its dirty state.
1190                  * The unlocker will place the inode on the appropriate
1191                  * superblock list, based upon its state.
1192                  */
1193                 if (inode->i_state & I_SYNC)
1194                         goto out_unlock_inode;
1195 
1196                 /*
1197                  * Only add valid (hashed) inodes to the superblock's
1198                  * dirty list.  Add blockdev inodes as well.
1199                  */
1200                 if (!S_ISBLK(inode->i_mode)) {
1201                         if (inode_unhashed(inode))
1202                                 goto out_unlock_inode;
1203                 }
1204                 if (inode->i_state & I_FREEING)
1205                         goto out_unlock_inode;
1206 
1207                 /*
1208                  * If the inode was already on b_dirty/b_io/b_more_io, don't
1209                  * reposition it (that would break b_dirty time-ordering).
1210                  */
1211                 if (!was_dirty) {
1212                         bool wakeup_bdi = false;
1213                         bdi = inode_to_bdi(inode);
1214 
1215                         if (bdi_cap_writeback_dirty(bdi)) {
1216                                 WARN(!test_bit(BDI_registered, &bdi->state),
1217                                      "bdi-%s not registered\n", bdi->name);
1218 
1219                                 /*
1220                                  * If this is the first dirty inode for this
1221                                  * bdi, we have to wake-up the corresponding
1222                                  * bdi thread to make sure background
1223                                  * write-back happens later.
1224                                  */
1225                                 if (!wb_has_dirty_io(&bdi->wb))
1226                                         wakeup_bdi = true;
1227                         }
1228 
1229                         spin_unlock(&inode->i_lock);
1230                         spin_lock(&bdi->wb.list_lock);
1231                         inode->dirtied_when = jiffies;
1232                         list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1233                         spin_unlock(&bdi->wb.list_lock);
1234 
1235                         if (wakeup_bdi)
1236                                 bdi_wakeup_thread_delayed(bdi);
1237                         return;
1238                 }
1239         }
1240 out_unlock_inode:
1241         spin_unlock(&inode->i_lock);
1242 
1243 }
1244 EXPORT_SYMBOL(__mark_inode_dirty);
1245 
1246 static void wait_sb_inodes(struct super_block *sb)
1247 {
1248         struct inode *inode, *old_inode = NULL;
1249 
1250         /*
1251          * We need to be protected against the filesystem going from
1252          * r/o to r/w or vice versa.
1253          */
1254         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1255 
1256         spin_lock(&inode_sb_list_lock);
1257 
1258         /*
1259          * Data integrity sync. Must wait for all pages under writeback,
1260          * because there may have been pages dirtied before our sync
1261          * call, but which had writeout started before we write it out.
1262          * In which case, the inode may not be on the dirty list, but
1263          * we still have to wait for that writeout.
1264          */
1265         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1266                 struct address_space *mapping = inode->i_mapping;
1267 
1268                 spin_lock(&inode->i_lock);
1269                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1270                     (mapping->nrpages == 0)) {
1271                         spin_unlock(&inode->i_lock);
1272                         continue;
1273                 }
1274                 __iget(inode);
1275                 spin_unlock(&inode->i_lock);
1276                 spin_unlock(&inode_sb_list_lock);
1277 
1278                 /*
1279                  * We hold a reference to 'inode' so it couldn't have been
1280                  * removed from s_inodes list while we dropped the
1281                  * inode_sb_list_lock.  We cannot iput the inode now as we can
1282                  * be holding the last reference and we cannot iput it under
1283                  * inode_sb_list_lock. So we keep the reference and iput it
1284                  * later.
1285                  */
1286                 iput(old_inode);
1287                 old_inode = inode;
1288 
1289                 filemap_fdatawait(mapping);
1290 
1291                 cond_resched();
1292 
1293                 spin_lock(&inode_sb_list_lock);
1294         }
1295         spin_unlock(&inode_sb_list_lock);
1296         iput(old_inode);
1297 }
1298 
1299 /**
1300  * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
1301  * @sb: the superblock
1302  * @nr: the number of pages to write
1303  * @reason: reason why some writeback work initiated
1304  *
1305  * Start writeback on some inodes on this super_block. No guarantees are made
1306  * on how many (if any) will be written, and this function does not wait
1307  * for IO completion of submitted IO.
1308  */
1309 void writeback_inodes_sb_nr(struct super_block *sb,
1310                             unsigned long nr,
1311                             enum wb_reason reason)
1312 {
1313         DECLARE_COMPLETION_ONSTACK(done);
1314         struct wb_writeback_work work = {
1315                 .sb                     = sb,
1316                 .sync_mode              = WB_SYNC_NONE,
1317                 .tagged_writepages      = 1,
1318                 .done                   = &done,
1319                 .nr_pages               = nr,
1320                 .reason                 = reason,
1321         };
1322 
1323         if (sb->s_bdi == &noop_backing_dev_info)
1324                 return;
1325         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1326         bdi_queue_work(sb->s_bdi, &work);
1327         wait_for_completion(&done);
1328 }
1329 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1330 
1331 /**
1332  * writeback_inodes_sb  -       writeback dirty inodes from given super_block
1333  * @sb: the superblock
1334  * @reason: reason why some writeback work was initiated
1335  *
1336  * Start writeback on some inodes on this super_block. No guarantees are made
1337  * on how many (if any) will be written, and this function does not wait
1338  * for IO completion of submitted IO.
1339  */
1340 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1341 {
1342         return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1343 }
1344 EXPORT_SYMBOL(writeback_inodes_sb);
1345 
1346 /**
1347  * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1348  * @sb: the superblock
1349  * @nr: the number of pages to write
1350  * @reason: the reason of writeback
1351  *
1352  * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1353  * Returns 1 if writeback was started, 0 if not.
1354  */
1355 int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1356                                   unsigned long nr,
1357                                   enum wb_reason reason)
1358 {
1359         if (writeback_in_progress(sb->s_bdi))
1360                 return 1;
1361 
1362         if (!down_read_trylock(&sb->s_umount))
1363                 return 0;
1364 
1365         writeback_inodes_sb_nr(sb, nr, reason);
1366         up_read(&sb->s_umount);
1367         return 1;
1368 }
1369 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1370 
1371 /**
1372  * try_to_writeback_inodes_sb - try to start writeback if none underway
1373  * @sb: the superblock
1374  * @reason: reason why some writeback work was initiated
1375  *
1376  * Implement by try_to_writeback_inodes_sb_nr()
1377  * Returns 1 if writeback was started, 0 if not.
1378  */
1379 int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1380 {
1381         return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1382 }
1383 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
1384 
1385 /**
1386  * sync_inodes_sb       -       sync sb inode pages
1387  * @sb: the superblock
1388  *
1389  * This function writes and waits on any dirty inode belonging to this
1390  * super_block.
1391  */
1392 void sync_inodes_sb(struct super_block *sb)
1393 {
1394         DECLARE_COMPLETION_ONSTACK(done);
1395         struct wb_writeback_work work = {
1396                 .sb             = sb,
1397                 .sync_mode      = WB_SYNC_ALL,
1398                 .nr_pages       = LONG_MAX,
1399                 .range_cyclic   = 0,
1400                 .done           = &done,
1401                 .reason         = WB_REASON_SYNC,
1402         };
1403 
1404         /* Nothing to do? */
1405         if (sb->s_bdi == &noop_backing_dev_info)
1406                 return;
1407         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1408 
1409         bdi_queue_work(sb->s_bdi, &work);
1410         wait_for_completion(&done);
1411 
1412         wait_sb_inodes(sb);
1413 }
1414 EXPORT_SYMBOL(sync_inodes_sb);
1415 
1416 /**
1417  * write_inode_now      -       write an inode to disk
1418  * @inode: inode to write to disk
1419  * @sync: whether the write should be synchronous or not
1420  *
1421  * This function commits an inode to disk immediately if it is dirty. This is
1422  * primarily needed by knfsd.
1423  *
1424  * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1425  */
1426 int write_inode_now(struct inode *inode, int sync)
1427 {
1428         struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1429         struct writeback_control wbc = {
1430                 .nr_to_write = LONG_MAX,
1431                 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1432                 .range_start = 0,
1433                 .range_end = LLONG_MAX,
1434         };
1435 
1436         if (!mapping_cap_writeback_dirty(inode->i_mapping))
1437                 wbc.nr_to_write = 0;
1438 
1439         might_sleep();
1440         return writeback_single_inode(inode, wb, &wbc);
1441 }
1442 EXPORT_SYMBOL(write_inode_now);
1443 
1444 /**
1445  * sync_inode - write an inode and its pages to disk.
1446  * @inode: the inode to sync
1447  * @wbc: controls the writeback mode
1448  *
1449  * sync_inode() will write an inode and its pages to disk.  It will also
1450  * correctly update the inode on its superblock's dirty inode lists and will
1451  * update inode->i_state.
1452  *
1453  * The caller must have a ref on the inode.
1454  */
1455 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1456 {
1457         return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1458 }
1459 EXPORT_SYMBOL(sync_inode);
1460 
1461 /**
1462  * sync_inode_metadata - write an inode to disk
1463  * @inode: the inode to sync
1464  * @wait: wait for I/O to complete.
1465  *
1466  * Write an inode to disk and adjust its dirty state after completion.
1467  *
1468  * Note: only writes the actual inode, no associated data or other metadata.
1469  */
1470 int sync_inode_metadata(struct inode *inode, int wait)
1471 {
1472         struct writeback_control wbc = {
1473                 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1474                 .nr_to_write = 0, /* metadata-only */
1475         };
1476 
1477         return sync_inode(inode, &wbc);
1478 }
1479 EXPORT_SYMBOL(sync_inode_metadata);
1480 

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

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

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

osdn.jp