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

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

Version: ~ [ linux-5.8-rc4 ] ~ [ linux-5.7.7 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.50 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.131 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.187 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.229 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.229 ] ~ [ 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-3.16.85 ] ~ [ 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-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * fs/fs-writeback.c
  3  *
  4  * Copyright (C) 2002, Linus Torvalds.
  5  *
  6  * Contains all the functions related to writing back and waiting
  7  * upon dirty inodes against superblocks, and writing back dirty
  8  * pages against inodes.  ie: data writeback.  Writeout of the
  9  * inode itself is not handled here.
 10  *
 11  * 10Apr2002    Andrew Morton
 12  *              Split out of fs/inode.c
 13  *              Additions for address_space-based writeback
 14  */
 15 
 16 #include <linux/kernel.h>
 17 #include <linux/export.h>
 18 #include <linux/spinlock.h>
 19 #include <linux/slab.h>
 20 #include <linux/sched.h>
 21 #include <linux/fs.h>
 22 #include <linux/mm.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/kthread.h>
 25 #include <linux/writeback.h>
 26 #include <linux/blkdev.h>
 27 #include <linux/backing-dev.h>
 28 #include <linux/tracepoint.h>
 29 #include "internal.h"
 30 
 31 /*
 32  * 4MB minimal write chunk size
 33  */
 34 #define MIN_WRITEBACK_PAGES     (4096UL >> (PAGE_CACHE_SHIFT - 10))
 35 
 36 /*
 37  * Passed into wb_writeback(), essentially a subset of writeback_control
 38  */
 39 struct wb_writeback_work {
 40         long nr_pages;
 41         struct super_block *sb;
 42         unsigned long *older_than_this;
 43         enum writeback_sync_modes sync_mode;
 44         unsigned int tagged_writepages:1;
 45         unsigned int for_kupdate:1;
 46         unsigned int range_cyclic:1;
 47         unsigned int for_background:1;
 48         unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
 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 (sb_is_blkdev_sb(sb))
 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 static void bdi_wakeup_thread(struct backing_dev_info *bdi)
 92 {
 93         spin_lock_bh(&bdi->wb_lock);
 94         if (test_bit(BDI_registered, &bdi->state))
 95                 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
 96         spin_unlock_bh(&bdi->wb_lock);
 97 }
 98 
 99 static void bdi_queue_work(struct backing_dev_info *bdi,
100                            struct wb_writeback_work *work)
101 {
102         trace_writeback_queue(bdi, work);
103 
104         spin_lock_bh(&bdi->wb_lock);
105         if (!test_bit(BDI_registered, &bdi->state)) {
106                 if (work->done)
107                         complete(work->done);
108                 goto out_unlock;
109         }
110         list_add_tail(&work->list, &bdi->work_list);
111         mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
112 out_unlock:
113         spin_unlock_bh(&bdi->wb_lock);
114 }
115 
116 static void
117 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
118                       bool range_cyclic, enum wb_reason reason)
119 {
120         struct wb_writeback_work *work;
121 
122         /*
123          * This is WB_SYNC_NONE writeback, so if allocation fails just
124          * wakeup the thread for old dirty data writeback
125          */
126         work = kzalloc(sizeof(*work), GFP_ATOMIC);
127         if (!work) {
128                 trace_writeback_nowork(bdi);
129                 bdi_wakeup_thread(bdi);
130                 return;
131         }
132 
133         work->sync_mode = WB_SYNC_NONE;
134         work->nr_pages  = nr_pages;
135         work->range_cyclic = range_cyclic;
136         work->reason    = reason;
137 
138         bdi_queue_work(bdi, work);
139 }
140 
141 /**
142  * bdi_start_writeback - start writeback
143  * @bdi: the backing device to write from
144  * @nr_pages: the number of pages to write
145  * @reason: reason why some writeback work was initiated
146  *
147  * Description:
148  *   This does WB_SYNC_NONE opportunistic writeback. The IO is only
149  *   started when this function returns, we make no guarantees on
150  *   completion. Caller need not hold sb s_umount semaphore.
151  *
152  */
153 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
154                         enum wb_reason reason)
155 {
156         __bdi_start_writeback(bdi, nr_pages, true, reason);
157 }
158 
159 /**
160  * bdi_start_background_writeback - start background writeback
161  * @bdi: the backing device to write from
162  *
163  * Description:
164  *   This makes sure WB_SYNC_NONE background writeback happens. When
165  *   this function returns, it is only guaranteed that for given BDI
166  *   some IO is happening if we are over background dirty threshold.
167  *   Caller need not hold sb s_umount semaphore.
168  */
169 void bdi_start_background_writeback(struct backing_dev_info *bdi)
170 {
171         /*
172          * We just wake up the flusher thread. It will perform background
173          * writeback as soon as there is no other work to do.
174          */
175         trace_writeback_wake_background(bdi);
176         bdi_wakeup_thread(bdi);
177 }
178 
179 /*
180  * Remove the inode from the writeback list it is on.
181  */
182 void inode_wb_list_del(struct inode *inode)
183 {
184         struct backing_dev_info *bdi = inode_to_bdi(inode);
185 
186         spin_lock(&bdi->wb.list_lock);
187         list_del_init(&inode->i_wb_list);
188         spin_unlock(&bdi->wb.list_lock);
189 }
190 
191 /*
192  * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
193  * furthest end of its superblock's dirty-inode list.
194  *
195  * Before stamping the inode's ->dirtied_when, we check to see whether it is
196  * already the most-recently-dirtied inode on the b_dirty list.  If that is
197  * the case then the inode must have been redirtied while it was being written
198  * out and we don't reset its dirtied_when.
199  */
200 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
201 {
202         assert_spin_locked(&wb->list_lock);
203         if (!list_empty(&wb->b_dirty)) {
204                 struct inode *tail;
205 
206                 tail = wb_inode(wb->b_dirty.next);
207                 if (time_before(inode->dirtied_when, tail->dirtied_when))
208                         inode->dirtied_when = jiffies;
209         }
210         list_move(&inode->i_wb_list, &wb->b_dirty);
211 }
212 
213 /*
214  * requeue inode for re-scanning after bdi->b_io list is exhausted.
215  */
216 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
217 {
218         assert_spin_locked(&wb->list_lock);
219         list_move(&inode->i_wb_list, &wb->b_more_io);
220 }
221 
222 static void inode_sync_complete(struct inode *inode)
223 {
224         inode->i_state &= ~I_SYNC;
225         /* If inode is clean an unused, put it into LRU now... */
226         inode_add_lru(inode);
227         /* Waiters must see I_SYNC cleared before being woken up */
228         smp_mb();
229         wake_up_bit(&inode->i_state, __I_SYNC);
230 }
231 
232 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
233 {
234         bool ret = time_after(inode->dirtied_when, t);
235 #ifndef CONFIG_64BIT
236         /*
237          * For inodes being constantly redirtied, dirtied_when can get stuck.
238          * It _appears_ to be in the future, but is actually in distant past.
239          * This test is necessary to prevent such wrapped-around relative times
240          * from permanently stopping the whole bdi writeback.
241          */
242         ret = ret && time_before_eq(inode->dirtied_when, jiffies);
243 #endif
244         return ret;
245 }
246 
247 /*
248  * Move expired (dirtied before work->older_than_this) dirty inodes from
249  * @delaying_queue to @dispatch_queue.
250  */
251 static int move_expired_inodes(struct list_head *delaying_queue,
252                                struct list_head *dispatch_queue,
253                                struct wb_writeback_work *work)
254 {
255         LIST_HEAD(tmp);
256         struct list_head *pos, *node;
257         struct super_block *sb = NULL;
258         struct inode *inode;
259         int do_sb_sort = 0;
260         int moved = 0;
261 
262         while (!list_empty(delaying_queue)) {
263                 inode = wb_inode(delaying_queue->prev);
264                 if (work->older_than_this &&
265                     inode_dirtied_after(inode, *work->older_than_this))
266                         break;
267                 list_move(&inode->i_wb_list, &tmp);
268                 moved++;
269                 if (sb_is_blkdev_sb(inode->i_sb))
270                         continue;
271                 if (sb && sb != inode->i_sb)
272                         do_sb_sort = 1;
273                 sb = inode->i_sb;
274         }
275 
276         /* just one sb in list, splice to dispatch_queue and we're done */
277         if (!do_sb_sort) {
278                 list_splice(&tmp, dispatch_queue);
279                 goto out;
280         }
281 
282         /* Move inodes from one superblock together */
283         while (!list_empty(&tmp)) {
284                 sb = wb_inode(tmp.prev)->i_sb;
285                 list_for_each_prev_safe(pos, node, &tmp) {
286                         inode = wb_inode(pos);
287                         if (inode->i_sb == sb)
288                                 list_move(&inode->i_wb_list, dispatch_queue);
289                 }
290         }
291 out:
292         return moved;
293 }
294 
295 /*
296  * Queue all expired dirty inodes for io, eldest first.
297  * Before
298  *         newly dirtied     b_dirty    b_io    b_more_io
299  *         =============>    gf         edc     BA
300  * After
301  *         newly dirtied     b_dirty    b_io    b_more_io
302  *         =============>    g          fBAedc
303  *                                           |
304  *                                           +--> dequeue for IO
305  */
306 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
307 {
308         int moved;
309         assert_spin_locked(&wb->list_lock);
310         list_splice_init(&wb->b_more_io, &wb->b_io);
311         moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
312         trace_writeback_queue_io(wb, work, moved);
313 }
314 
315 static int write_inode(struct inode *inode, struct writeback_control *wbc)
316 {
317         int ret;
318 
319         if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
320                 trace_writeback_write_inode_start(inode, wbc);
321                 ret = inode->i_sb->s_op->write_inode(inode, wbc);
322                 trace_writeback_write_inode(inode, wbc);
323                 return ret;
324         }
325         return 0;
326 }
327 
328 /*
329  * Wait for writeback on an inode to complete. Called with i_lock held.
330  * Caller must make sure inode cannot go away when we drop i_lock.
331  */
332 static void __inode_wait_for_writeback(struct inode *inode)
333         __releases(inode->i_lock)
334         __acquires(inode->i_lock)
335 {
336         DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
337         wait_queue_head_t *wqh;
338 
339         wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
340         while (inode->i_state & I_SYNC) {
341                 spin_unlock(&inode->i_lock);
342                 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
343                 spin_lock(&inode->i_lock);
344         }
345 }
346 
347 /*
348  * Wait for writeback on an inode to complete. Caller must have inode pinned.
349  */
350 void inode_wait_for_writeback(struct inode *inode)
351 {
352         spin_lock(&inode->i_lock);
353         __inode_wait_for_writeback(inode);
354         spin_unlock(&inode->i_lock);
355 }
356 
357 /*
358  * Sleep until I_SYNC is cleared. This function must be called with i_lock
359  * held and drops it. It is aimed for callers not holding any inode reference
360  * so once i_lock is dropped, inode can go away.
361  */
362 static void inode_sleep_on_writeback(struct inode *inode)
363         __releases(inode->i_lock)
364 {
365         DEFINE_WAIT(wait);
366         wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
367         int sleep;
368 
369         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
370         sleep = inode->i_state & I_SYNC;
371         spin_unlock(&inode->i_lock);
372         if (sleep)
373                 schedule();
374         finish_wait(wqh, &wait);
375 }
376 
377 /*
378  * Find proper writeback list for the inode depending on its current state and
379  * possibly also change of its state while we were doing writeback.  Here we
380  * handle things such as livelock prevention or fairness of writeback among
381  * inodes. This function can be called only by flusher thread - noone else
382  * processes all inodes in writeback lists and requeueing inodes behind flusher
383  * thread's back can have unexpected consequences.
384  */
385 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
386                           struct writeback_control *wbc)
387 {
388         if (inode->i_state & I_FREEING)
389                 return;
390 
391         /*
392          * Sync livelock prevention. Each inode is tagged and synced in one
393          * shot. If still dirty, it will be redirty_tail()'ed below.  Update
394          * the dirty time to prevent enqueue and sync it again.
395          */
396         if ((inode->i_state & I_DIRTY) &&
397             (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
398                 inode->dirtied_when = jiffies;
399 
400         if (wbc->pages_skipped) {
401                 /*
402                  * writeback is not making progress due to locked
403                  * buffers. Skip this inode for now.
404                  */
405                 redirty_tail(inode, wb);
406                 return;
407         }
408 
409         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
410                 /*
411                  * We didn't write back all the pages.  nfs_writepages()
412                  * sometimes bales out without doing anything.
413                  */
414                 if (wbc->nr_to_write <= 0) {
415                         /* Slice used up. Queue for next turn. */
416                         requeue_io(inode, wb);
417                 } else {
418                         /*
419                          * Writeback blocked by something other than
420                          * congestion. Delay the inode for some time to
421                          * avoid spinning on the CPU (100% iowait)
422                          * retrying writeback of the dirty page/inode
423                          * that cannot be performed immediately.
424                          */
425                         redirty_tail(inode, wb);
426                 }
427         } else if (inode->i_state & I_DIRTY) {
428                 /*
429                  * Filesystems can dirty the inode during writeback operations,
430                  * such as delayed allocation during submission or metadata
431                  * updates after data IO completion.
432                  */
433                 redirty_tail(inode, wb);
434         } else {
435                 /* The inode is clean. Remove from writeback lists. */
436                 list_del_init(&inode->i_wb_list);
437         }
438 }
439 
440 /*
441  * Write out an inode and its dirty pages. Do not update the writeback list
442  * linkage. That is left to the caller. The caller is also responsible for
443  * setting I_SYNC flag and calling inode_sync_complete() to clear it.
444  */
445 static int
446 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
447 {
448         struct address_space *mapping = inode->i_mapping;
449         long nr_to_write = wbc->nr_to_write;
450         unsigned dirty;
451         int ret;
452 
453         WARN_ON(!(inode->i_state & I_SYNC));
454 
455         trace_writeback_single_inode_start(inode, wbc, nr_to_write);
456 
457         ret = do_writepages(mapping, wbc);
458 
459         /*
460          * Make sure to wait on the data before writing out the metadata.
461          * This is important for filesystems that modify metadata on data
462          * I/O completion. We don't do it for sync(2) writeback because it has a
463          * separate, external IO completion path and ->sync_fs for guaranteeing
464          * inode metadata is written back correctly.
465          */
466         if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
467                 int err = filemap_fdatawait(mapping);
468                 if (ret == 0)
469                         ret = err;
470         }
471 
472         /*
473          * Some filesystems may redirty the inode during the writeback
474          * due to delalloc, clear dirty metadata flags right before
475          * write_inode()
476          */
477         spin_lock(&inode->i_lock);
478 
479         dirty = inode->i_state & I_DIRTY;
480         inode->i_state &= ~I_DIRTY;
481 
482         /*
483          * Paired with smp_mb() in __mark_inode_dirty().  This allows
484          * __mark_inode_dirty() to test i_state without grabbing i_lock -
485          * either they see the I_DIRTY bits cleared or we see the dirtied
486          * inode.
487          *
488          * I_DIRTY_PAGES is always cleared together above even if @mapping
489          * still has dirty pages.  The flag is reinstated after smp_mb() if
490          * necessary.  This guarantees that either __mark_inode_dirty()
491          * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
492          */
493         smp_mb();
494 
495         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
496                 inode->i_state |= I_DIRTY_PAGES;
497 
498         spin_unlock(&inode->i_lock);
499 
500         /* Don't write the inode if only I_DIRTY_PAGES was set */
501         if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
502                 int err = write_inode(inode, wbc);
503                 if (ret == 0)
504                         ret = err;
505         }
506         trace_writeback_single_inode(inode, wbc, nr_to_write);
507         return ret;
508 }
509 
510 /*
511  * Write out an inode's dirty pages. Either the caller has an active reference
512  * on the inode or the inode has I_WILL_FREE set.
513  *
514  * This function is designed to be called for writing back one inode which
515  * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
516  * and does more profound writeback list handling in writeback_sb_inodes().
517  */
518 static int
519 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
520                        struct writeback_control *wbc)
521 {
522         int ret = 0;
523 
524         spin_lock(&inode->i_lock);
525         if (!atomic_read(&inode->i_count))
526                 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
527         else
528                 WARN_ON(inode->i_state & I_WILL_FREE);
529 
530         if (inode->i_state & I_SYNC) {
531                 if (wbc->sync_mode != WB_SYNC_ALL)
532                         goto out;
533                 /*
534                  * It's a data-integrity sync. We must wait. Since callers hold
535                  * inode reference or inode has I_WILL_FREE set, it cannot go
536                  * away under us.
537                  */
538                 __inode_wait_for_writeback(inode);
539         }
540         WARN_ON(inode->i_state & I_SYNC);
541         /*
542          * Skip inode if it is clean and we have no outstanding writeback in
543          * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
544          * function since flusher thread may be doing for example sync in
545          * parallel and if we move the inode, it could get skipped. So here we
546          * make sure inode is on some writeback list and leave it there unless
547          * we have completely cleaned the inode.
548          */
549         if (!(inode->i_state & I_DIRTY) &&
550             (wbc->sync_mode != WB_SYNC_ALL ||
551              !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
552                 goto out;
553         inode->i_state |= I_SYNC;
554         spin_unlock(&inode->i_lock);
555 
556         ret = __writeback_single_inode(inode, wbc);
557 
558         spin_lock(&wb->list_lock);
559         spin_lock(&inode->i_lock);
560         /*
561          * If inode is clean, remove it from writeback lists. Otherwise don't
562          * touch it. See comment above for explanation.
563          */
564         if (!(inode->i_state & I_DIRTY))
565                 list_del_init(&inode->i_wb_list);
566         spin_unlock(&wb->list_lock);
567         inode_sync_complete(inode);
568 out:
569         spin_unlock(&inode->i_lock);
570         return ret;
571 }
572 
573 static long writeback_chunk_size(struct backing_dev_info *bdi,
574                                  struct wb_writeback_work *work)
575 {
576         long pages;
577 
578         /*
579          * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
580          * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
581          * here avoids calling into writeback_inodes_wb() more than once.
582          *
583          * The intended call sequence for WB_SYNC_ALL writeback is:
584          *
585          *      wb_writeback()
586          *          writeback_sb_inodes()       <== called only once
587          *              write_cache_pages()     <== called once for each inode
588          *                   (quickly) tag currently dirty pages
589          *                   (maybe slowly) sync all tagged pages
590          */
591         if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
592                 pages = LONG_MAX;
593         else {
594                 pages = min(bdi->avg_write_bandwidth / 2,
595                             global_dirty_limit / DIRTY_SCOPE);
596                 pages = min(pages, work->nr_pages);
597                 pages = round_down(pages + MIN_WRITEBACK_PAGES,
598                                    MIN_WRITEBACK_PAGES);
599         }
600 
601         return pages;
602 }
603 
604 /*
605  * Write a portion of b_io inodes which belong to @sb.
606  *
607  * Return the number of pages and/or inodes written.
608  */
609 static long writeback_sb_inodes(struct super_block *sb,
610                                 struct bdi_writeback *wb,
611                                 struct wb_writeback_work *work)
612 {
613         struct writeback_control wbc = {
614                 .sync_mode              = work->sync_mode,
615                 .tagged_writepages      = work->tagged_writepages,
616                 .for_kupdate            = work->for_kupdate,
617                 .for_background         = work->for_background,
618                 .for_sync               = work->for_sync,
619                 .range_cyclic           = work->range_cyclic,
620                 .range_start            = 0,
621                 .range_end              = LLONG_MAX,
622         };
623         unsigned long start_time = jiffies;
624         long write_chunk;
625         long wrote = 0;  /* count both pages and inodes */
626 
627         while (!list_empty(&wb->b_io)) {
628                 struct inode *inode = wb_inode(wb->b_io.prev);
629 
630                 if (inode->i_sb != sb) {
631                         if (work->sb) {
632                                 /*
633                                  * We only want to write back data for this
634                                  * superblock, move all inodes not belonging
635                                  * to it back onto the dirty list.
636                                  */
637                                 redirty_tail(inode, wb);
638                                 continue;
639                         }
640 
641                         /*
642                          * The inode belongs to a different superblock.
643                          * Bounce back to the caller to unpin this and
644                          * pin the next superblock.
645                          */
646                         break;
647                 }
648 
649                 /*
650                  * Don't bother with new inodes or inodes being freed, first
651                  * kind does not need periodic writeout yet, and for the latter
652                  * kind writeout is handled by the freer.
653                  */
654                 spin_lock(&inode->i_lock);
655                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
656                         spin_unlock(&inode->i_lock);
657                         redirty_tail(inode, wb);
658                         continue;
659                 }
660                 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
661                         /*
662                          * If this inode is locked for writeback and we are not
663                          * doing writeback-for-data-integrity, move it to
664                          * b_more_io so that writeback can proceed with the
665                          * other inodes on s_io.
666                          *
667                          * We'll have another go at writing back this inode
668                          * when we completed a full scan of b_io.
669                          */
670                         spin_unlock(&inode->i_lock);
671                         requeue_io(inode, wb);
672                         trace_writeback_sb_inodes_requeue(inode);
673                         continue;
674                 }
675                 spin_unlock(&wb->list_lock);
676 
677                 /*
678                  * We already requeued the inode if it had I_SYNC set and we
679                  * are doing WB_SYNC_NONE writeback. So this catches only the
680                  * WB_SYNC_ALL case.
681                  */
682                 if (inode->i_state & I_SYNC) {
683                         /* Wait for I_SYNC. This function drops i_lock... */
684                         inode_sleep_on_writeback(inode);
685                         /* Inode may be gone, start again */
686                         spin_lock(&wb->list_lock);
687                         continue;
688                 }
689                 inode->i_state |= I_SYNC;
690                 spin_unlock(&inode->i_lock);
691 
692                 write_chunk = writeback_chunk_size(wb->bdi, work);
693                 wbc.nr_to_write = write_chunk;
694                 wbc.pages_skipped = 0;
695 
696                 /*
697                  * We use I_SYNC to pin the inode in memory. While it is set
698                  * evict_inode() will wait so the inode cannot be freed.
699                  */
700                 __writeback_single_inode(inode, &wbc);
701 
702                 work->nr_pages -= write_chunk - wbc.nr_to_write;
703                 wrote += write_chunk - wbc.nr_to_write;
704                 spin_lock(&wb->list_lock);
705                 spin_lock(&inode->i_lock);
706                 if (!(inode->i_state & I_DIRTY))
707                         wrote++;
708                 requeue_inode(inode, wb, &wbc);
709                 inode_sync_complete(inode);
710                 spin_unlock(&inode->i_lock);
711                 cond_resched_lock(&wb->list_lock);
712                 /*
713                  * bail out to wb_writeback() often enough to check
714                  * background threshold and other termination conditions.
715                  */
716                 if (wrote) {
717                         if (time_is_before_jiffies(start_time + HZ / 10UL))
718                                 break;
719                         if (work->nr_pages <= 0)
720                                 break;
721                 }
722         }
723         return wrote;
724 }
725 
726 static long __writeback_inodes_wb(struct bdi_writeback *wb,
727                                   struct wb_writeback_work *work)
728 {
729         unsigned long start_time = jiffies;
730         long wrote = 0;
731 
732         while (!list_empty(&wb->b_io)) {
733                 struct inode *inode = wb_inode(wb->b_io.prev);
734                 struct super_block *sb = inode->i_sb;
735 
736                 if (!grab_super_passive(sb)) {
737                         /*
738                          * grab_super_passive() may fail consistently due to
739                          * s_umount being grabbed by someone else. Don't use
740                          * requeue_io() to avoid busy retrying the inode/sb.
741                          */
742                         redirty_tail(inode, wb);
743                         continue;
744                 }
745                 wrote += writeback_sb_inodes(sb, wb, work);
746                 drop_super(sb);
747 
748                 /* refer to the same tests at the end of writeback_sb_inodes */
749                 if (wrote) {
750                         if (time_is_before_jiffies(start_time + HZ / 10UL))
751                                 break;
752                         if (work->nr_pages <= 0)
753                                 break;
754                 }
755         }
756         /* Leave any unwritten inodes on b_io */
757         return wrote;
758 }
759 
760 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
761                                 enum wb_reason reason)
762 {
763         struct wb_writeback_work work = {
764                 .nr_pages       = nr_pages,
765                 .sync_mode      = WB_SYNC_NONE,
766                 .range_cyclic   = 1,
767                 .reason         = reason,
768         };
769 
770         spin_lock(&wb->list_lock);
771         if (list_empty(&wb->b_io))
772                 queue_io(wb, &work);
773         __writeback_inodes_wb(wb, &work);
774         spin_unlock(&wb->list_lock);
775 
776         return nr_pages - work.nr_pages;
777 }
778 
779 static bool over_bground_thresh(struct backing_dev_info *bdi)
780 {
781         unsigned long background_thresh, dirty_thresh;
782 
783         global_dirty_limits(&background_thresh, &dirty_thresh);
784 
785         if (global_page_state(NR_FILE_DIRTY) +
786             global_page_state(NR_UNSTABLE_NFS) > background_thresh)
787                 return true;
788 
789         if (bdi_stat(bdi, BDI_RECLAIMABLE) >
790                                 bdi_dirty_limit(bdi, background_thresh))
791                 return true;
792 
793         return false;
794 }
795 
796 /*
797  * Called under wb->list_lock. If there are multiple wb per bdi,
798  * only the flusher working on the first wb should do it.
799  */
800 static void wb_update_bandwidth(struct bdi_writeback *wb,
801                                 unsigned long start_time)
802 {
803         __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
804 }
805 
806 /*
807  * Explicit flushing or periodic writeback of "old" data.
808  *
809  * Define "old": the first time one of an inode's pages is dirtied, we mark the
810  * dirtying-time in the inode's address_space.  So this periodic writeback code
811  * just walks the superblock inode list, writing back any inodes which are
812  * older than a specific point in time.
813  *
814  * Try to run once per dirty_writeback_interval.  But if a writeback event
815  * takes longer than a dirty_writeback_interval interval, then leave a
816  * one-second gap.
817  *
818  * older_than_this takes precedence over nr_to_write.  So we'll only write back
819  * all dirty pages if they are all attached to "old" mappings.
820  */
821 static long wb_writeback(struct bdi_writeback *wb,
822                          struct wb_writeback_work *work)
823 {
824         unsigned long wb_start = jiffies;
825         long nr_pages = work->nr_pages;
826         unsigned long oldest_jif;
827         struct inode *inode;
828         long progress;
829 
830         oldest_jif = jiffies;
831         work->older_than_this = &oldest_jif;
832 
833         spin_lock(&wb->list_lock);
834         for (;;) {
835                 /*
836                  * Stop writeback when nr_pages has been consumed
837                  */
838                 if (work->nr_pages <= 0)
839                         break;
840 
841                 /*
842                  * Background writeout and kupdate-style writeback may
843                  * run forever. Stop them if there is other work to do
844                  * so that e.g. sync can proceed. They'll be restarted
845                  * after the other works are all done.
846                  */
847                 if ((work->for_background || work->for_kupdate) &&
848                     !list_empty(&wb->bdi->work_list))
849                         break;
850 
851                 /*
852                  * For background writeout, stop when we are below the
853                  * background dirty threshold
854                  */
855                 if (work->for_background && !over_bground_thresh(wb->bdi))
856                         break;
857 
858                 /*
859                  * Kupdate and background works are special and we want to
860                  * include all inodes that need writing. Livelock avoidance is
861                  * handled by these works yielding to any other work so we are
862                  * safe.
863                  */
864                 if (work->for_kupdate) {
865                         oldest_jif = jiffies -
866                                 msecs_to_jiffies(dirty_expire_interval * 10);
867                 } else if (work->for_background)
868                         oldest_jif = jiffies;
869 
870                 trace_writeback_start(wb->bdi, work);
871                 if (list_empty(&wb->b_io))
872                         queue_io(wb, work);
873                 if (work->sb)
874                         progress = writeback_sb_inodes(work->sb, wb, work);
875                 else
876                         progress = __writeback_inodes_wb(wb, work);
877                 trace_writeback_written(wb->bdi, work);
878 
879                 wb_update_bandwidth(wb, wb_start);
880 
881                 /*
882                  * Did we write something? Try for more
883                  *
884                  * Dirty inodes are moved to b_io for writeback in batches.
885                  * The completion of the current batch does not necessarily
886                  * mean the overall work is done. So we keep looping as long
887                  * as made some progress on cleaning pages or inodes.
888                  */
889                 if (progress)
890                         continue;
891                 /*
892                  * No more inodes for IO, bail
893                  */
894                 if (list_empty(&wb->b_more_io))
895                         break;
896                 /*
897                  * Nothing written. Wait for some inode to
898                  * become available for writeback. Otherwise
899                  * we'll just busyloop.
900                  */
901                 if (!list_empty(&wb->b_more_io))  {
902                         trace_writeback_wait(wb->bdi, work);
903                         inode = wb_inode(wb->b_more_io.prev);
904                         spin_lock(&inode->i_lock);
905                         spin_unlock(&wb->list_lock);
906                         /* This function drops i_lock... */
907                         inode_sleep_on_writeback(inode);
908                         spin_lock(&wb->list_lock);
909                 }
910         }
911         spin_unlock(&wb->list_lock);
912 
913         return nr_pages - work->nr_pages;
914 }
915 
916 /*
917  * Return the next wb_writeback_work struct that hasn't been processed yet.
918  */
919 static struct wb_writeback_work *
920 get_next_work_item(struct backing_dev_info *bdi)
921 {
922         struct wb_writeback_work *work = NULL;
923 
924         spin_lock_bh(&bdi->wb_lock);
925         if (!list_empty(&bdi->work_list)) {
926                 work = list_entry(bdi->work_list.next,
927                                   struct wb_writeback_work, list);
928                 list_del_init(&work->list);
929         }
930         spin_unlock_bh(&bdi->wb_lock);
931         return work;
932 }
933 
934 /*
935  * Add in the number of potentially dirty inodes, because each inode
936  * write can dirty pagecache in the underlying blockdev.
937  */
938 static unsigned long get_nr_dirty_pages(void)
939 {
940         return global_page_state(NR_FILE_DIRTY) +
941                 global_page_state(NR_UNSTABLE_NFS) +
942                 get_nr_dirty_inodes();
943 }
944 
945 static long wb_check_background_flush(struct bdi_writeback *wb)
946 {
947         if (over_bground_thresh(wb->bdi)) {
948 
949                 struct wb_writeback_work work = {
950                         .nr_pages       = LONG_MAX,
951                         .sync_mode      = WB_SYNC_NONE,
952                         .for_background = 1,
953                         .range_cyclic   = 1,
954                         .reason         = WB_REASON_BACKGROUND,
955                 };
956 
957                 return wb_writeback(wb, &work);
958         }
959 
960         return 0;
961 }
962 
963 static long wb_check_old_data_flush(struct bdi_writeback *wb)
964 {
965         unsigned long expired;
966         long nr_pages;
967 
968         /*
969          * When set to zero, disable periodic writeback
970          */
971         if (!dirty_writeback_interval)
972                 return 0;
973 
974         expired = wb->last_old_flush +
975                         msecs_to_jiffies(dirty_writeback_interval * 10);
976         if (time_before(jiffies, expired))
977                 return 0;
978 
979         wb->last_old_flush = jiffies;
980         nr_pages = get_nr_dirty_pages();
981 
982         if (nr_pages) {
983                 struct wb_writeback_work work = {
984                         .nr_pages       = nr_pages,
985                         .sync_mode      = WB_SYNC_NONE,
986                         .for_kupdate    = 1,
987                         .range_cyclic   = 1,
988                         .reason         = WB_REASON_PERIODIC,
989                 };
990 
991                 return wb_writeback(wb, &work);
992         }
993 
994         return 0;
995 }
996 
997 /*
998  * Retrieve work items and do the writeback they describe
999  */
1000 static long wb_do_writeback(struct bdi_writeback *wb)
1001 {
1002         struct backing_dev_info *bdi = wb->bdi;
1003         struct wb_writeback_work *work;
1004         long wrote = 0;
1005 
1006         set_bit(BDI_writeback_running, &wb->bdi->state);
1007         while ((work = get_next_work_item(bdi)) != NULL) {
1008 
1009                 trace_writeback_exec(bdi, work);
1010 
1011                 wrote += wb_writeback(wb, work);
1012 
1013                 /*
1014                  * Notify the caller of completion if this is a synchronous
1015                  * work item, otherwise just free it.
1016                  */
1017                 if (work->done)
1018                         complete(work->done);
1019                 else
1020                         kfree(work);
1021         }
1022 
1023         /*
1024          * Check for periodic writeback, kupdated() style
1025          */
1026         wrote += wb_check_old_data_flush(wb);
1027         wrote += wb_check_background_flush(wb);
1028         clear_bit(BDI_writeback_running, &wb->bdi->state);
1029 
1030         return wrote;
1031 }
1032 
1033 /*
1034  * Handle writeback of dirty data for the device backed by this bdi. Also
1035  * reschedules periodically and does kupdated style flushing.
1036  */
1037 void bdi_writeback_workfn(struct work_struct *work)
1038 {
1039         struct bdi_writeback *wb = container_of(to_delayed_work(work),
1040                                                 struct bdi_writeback, dwork);
1041         struct backing_dev_info *bdi = wb->bdi;
1042         long pages_written;
1043 
1044         set_worker_desc("flush-%s", dev_name(bdi->dev));
1045         current->flags |= PF_SWAPWRITE;
1046 
1047         if (likely(!current_is_workqueue_rescuer() ||
1048                    !test_bit(BDI_registered, &bdi->state))) {
1049                 /*
1050                  * The normal path.  Keep writing back @bdi until its
1051                  * work_list is empty.  Note that this path is also taken
1052                  * if @bdi is shutting down even when we're running off the
1053                  * rescuer as work_list needs to be drained.
1054                  */
1055                 do {
1056                         pages_written = wb_do_writeback(wb);
1057                         trace_writeback_pages_written(pages_written);
1058                 } while (!list_empty(&bdi->work_list));
1059         } else {
1060                 /*
1061                  * bdi_wq can't get enough workers and we're running off
1062                  * the emergency worker.  Don't hog it.  Hopefully, 1024 is
1063                  * enough for efficient IO.
1064                  */
1065                 pages_written = writeback_inodes_wb(&bdi->wb, 1024,
1066                                                     WB_REASON_FORKER_THREAD);
1067                 trace_writeback_pages_written(pages_written);
1068         }
1069 
1070         if (!list_empty(&bdi->work_list))
1071                 mod_delayed_work(bdi_wq, &wb->dwork, 0);
1072         else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1073                 bdi_wakeup_thread_delayed(bdi);
1074 
1075         current->flags &= ~PF_SWAPWRITE;
1076 }
1077 
1078 /*
1079  * Start writeback of `nr_pages' pages.  If `nr_pages' is zero, write back
1080  * the whole world.
1081  */
1082 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1083 {
1084         struct backing_dev_info *bdi;
1085 
1086         if (!nr_pages)
1087                 nr_pages = get_nr_dirty_pages();
1088 
1089         rcu_read_lock();
1090         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1091                 if (!bdi_has_dirty_io(bdi))
1092                         continue;
1093                 __bdi_start_writeback(bdi, nr_pages, false, reason);
1094         }
1095         rcu_read_unlock();
1096 }
1097 
1098 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1099 {
1100         if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1101                 struct dentry *dentry;
1102                 const char *name = "?";
1103 
1104                 dentry = d_find_alias(inode);
1105                 if (dentry) {
1106                         spin_lock(&dentry->d_lock);
1107                         name = (const char *) dentry->d_name.name;
1108                 }
1109                 printk(KERN_DEBUG
1110                        "%s(%d): dirtied inode %lu (%s) on %s\n",
1111                        current->comm, task_pid_nr(current), inode->i_ino,
1112                        name, inode->i_sb->s_id);
1113                 if (dentry) {
1114                         spin_unlock(&dentry->d_lock);
1115                         dput(dentry);
1116                 }
1117         }
1118 }
1119 
1120 /**
1121  *      __mark_inode_dirty -    internal function
1122  *      @inode: inode to mark
1123  *      @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1124  *      Mark an inode as dirty. Callers should use mark_inode_dirty or
1125  *      mark_inode_dirty_sync.
1126  *
1127  * Put the inode on the super block's dirty list.
1128  *
1129  * CAREFUL! We mark it dirty unconditionally, but move it onto the
1130  * dirty list only if it is hashed or if it refers to a blockdev.
1131  * If it was not hashed, it will never be added to the dirty list
1132  * even if it is later hashed, as it will have been marked dirty already.
1133  *
1134  * In short, make sure you hash any inodes _before_ you start marking
1135  * them dirty.
1136  *
1137  * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1138  * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
1139  * the kernel-internal blockdev inode represents the dirtying time of the
1140  * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
1141  * page->mapping->host, so the page-dirtying time is recorded in the internal
1142  * blockdev inode.
1143  */
1144 void __mark_inode_dirty(struct inode *inode, int flags)
1145 {
1146         struct super_block *sb = inode->i_sb;
1147         struct backing_dev_info *bdi = NULL;
1148 
1149         /*
1150          * Don't do this for I_DIRTY_PAGES - that doesn't actually
1151          * dirty the inode itself
1152          */
1153         if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1154                 trace_writeback_dirty_inode_start(inode, flags);
1155 
1156                 if (sb->s_op->dirty_inode)
1157                         sb->s_op->dirty_inode(inode, flags);
1158 
1159                 trace_writeback_dirty_inode(inode, flags);
1160         }
1161 
1162         /*
1163          * Paired with smp_mb() in __writeback_single_inode() for the
1164          * following lockless i_state test.  See there for details.
1165          */
1166         smp_mb();
1167 
1168         if ((inode->i_state & flags) == flags)
1169                 return;
1170 
1171         if (unlikely(block_dump))
1172                 block_dump___mark_inode_dirty(inode);
1173 
1174         spin_lock(&inode->i_lock);
1175         if ((inode->i_state & flags) != flags) {
1176                 const int was_dirty = inode->i_state & I_DIRTY;
1177 
1178                 inode->i_state |= flags;
1179 
1180                 /*
1181                  * If the inode is being synced, just update its dirty state.
1182                  * The unlocker will place the inode on the appropriate
1183                  * superblock list, based upon its state.
1184                  */
1185                 if (inode->i_state & I_SYNC)
1186                         goto out_unlock_inode;
1187 
1188                 /*
1189                  * Only add valid (hashed) inodes to the superblock's
1190                  * dirty list.  Add blockdev inodes as well.
1191                  */
1192                 if (!S_ISBLK(inode->i_mode)) {
1193                         if (inode_unhashed(inode))
1194                                 goto out_unlock_inode;
1195                 }
1196                 if (inode->i_state & I_FREEING)
1197                         goto out_unlock_inode;
1198 
1199                 /*
1200                  * If the inode was already on b_dirty/b_io/b_more_io, don't
1201                  * reposition it (that would break b_dirty time-ordering).
1202                  */
1203                 if (!was_dirty) {
1204                         bool wakeup_bdi = false;
1205                         bdi = inode_to_bdi(inode);
1206 
1207                         spin_unlock(&inode->i_lock);
1208                         spin_lock(&bdi->wb.list_lock);
1209                         if (bdi_cap_writeback_dirty(bdi)) {
1210                                 WARN(!test_bit(BDI_registered, &bdi->state),
1211                                      "bdi-%s not registered\n", bdi->name);
1212 
1213                                 /*
1214                                  * If this is the first dirty inode for this
1215                                  * bdi, we have to wake-up the corresponding
1216                                  * bdi thread to make sure background
1217                                  * write-back happens later.
1218                                  */
1219                                 if (!wb_has_dirty_io(&bdi->wb))
1220                                         wakeup_bdi = true;
1221                         }
1222 
1223                         inode->dirtied_when = jiffies;
1224                         list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1225                         spin_unlock(&bdi->wb.list_lock);
1226 
1227                         if (wakeup_bdi)
1228                                 bdi_wakeup_thread_delayed(bdi);
1229                         return;
1230                 }
1231         }
1232 out_unlock_inode:
1233         spin_unlock(&inode->i_lock);
1234 
1235 }
1236 EXPORT_SYMBOL(__mark_inode_dirty);
1237 
1238 static void wait_sb_inodes(struct super_block *sb)
1239 {
1240         struct inode *inode, *old_inode = NULL;
1241 
1242         /*
1243          * We need to be protected against the filesystem going from
1244          * r/o to r/w or vice versa.
1245          */
1246         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1247 
1248         spin_lock(&inode_sb_list_lock);
1249 
1250         /*
1251          * Data integrity sync. Must wait for all pages under writeback,
1252          * because there may have been pages dirtied before our sync
1253          * call, but which had writeout started before we write it out.
1254          * In which case, the inode may not be on the dirty list, but
1255          * we still have to wait for that writeout.
1256          */
1257         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1258                 struct address_space *mapping = inode->i_mapping;
1259 
1260                 spin_lock(&inode->i_lock);
1261                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1262                     (mapping->nrpages == 0)) {
1263                         spin_unlock(&inode->i_lock);
1264                         continue;
1265                 }
1266                 __iget(inode);
1267                 spin_unlock(&inode->i_lock);
1268                 spin_unlock(&inode_sb_list_lock);
1269 
1270                 /*
1271                  * We hold a reference to 'inode' so it couldn't have been
1272                  * removed from s_inodes list while we dropped the
1273                  * inode_sb_list_lock.  We cannot iput the inode now as we can
1274                  * be holding the last reference and we cannot iput it under
1275                  * inode_sb_list_lock. So we keep the reference and iput it
1276                  * later.
1277                  */
1278                 iput(old_inode);
1279                 old_inode = inode;
1280 
1281                 filemap_fdatawait(mapping);
1282 
1283                 cond_resched();
1284 
1285                 spin_lock(&inode_sb_list_lock);
1286         }
1287         spin_unlock(&inode_sb_list_lock);
1288         iput(old_inode);
1289 }
1290 
1291 /**
1292  * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
1293  * @sb: the superblock
1294  * @nr: the number of pages to write
1295  * @reason: reason why some writeback work initiated
1296  *
1297  * Start writeback on some inodes on this super_block. No guarantees are made
1298  * on how many (if any) will be written, and this function does not wait
1299  * for IO completion of submitted IO.
1300  */
1301 void writeback_inodes_sb_nr(struct super_block *sb,
1302                             unsigned long nr,
1303                             enum wb_reason reason)
1304 {
1305         DECLARE_COMPLETION_ONSTACK(done);
1306         struct wb_writeback_work work = {
1307                 .sb                     = sb,
1308                 .sync_mode              = WB_SYNC_NONE,
1309                 .tagged_writepages      = 1,
1310                 .done                   = &done,
1311                 .nr_pages               = nr,
1312                 .reason                 = reason,
1313         };
1314 
1315         if (sb->s_bdi == &noop_backing_dev_info)
1316                 return;
1317         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1318         bdi_queue_work(sb->s_bdi, &work);
1319         wait_for_completion(&done);
1320 }
1321 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1322 
1323 /**
1324  * writeback_inodes_sb  -       writeback dirty inodes from given super_block
1325  * @sb: the superblock
1326  * @reason: reason why some writeback work was initiated
1327  *
1328  * Start writeback on some inodes on this super_block. No guarantees are made
1329  * on how many (if any) will be written, and this function does not wait
1330  * for IO completion of submitted IO.
1331  */
1332 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1333 {
1334         return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1335 }
1336 EXPORT_SYMBOL(writeback_inodes_sb);
1337 
1338 /**
1339  * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1340  * @sb: the superblock
1341  * @nr: the number of pages to write
1342  * @reason: the reason of writeback
1343  *
1344  * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1345  * Returns 1 if writeback was started, 0 if not.
1346  */
1347 int try_to_writeback_inodes_sb_nr(struct super_block *sb,
1348                                   unsigned long nr,
1349                                   enum wb_reason reason)
1350 {
1351         if (writeback_in_progress(sb->s_bdi))
1352                 return 1;
1353 
1354         if (!down_read_trylock(&sb->s_umount))
1355                 return 0;
1356 
1357         writeback_inodes_sb_nr(sb, nr, reason);
1358         up_read(&sb->s_umount);
1359         return 1;
1360 }
1361 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr);
1362 
1363 /**
1364  * try_to_writeback_inodes_sb - try to start writeback if none underway
1365  * @sb: the superblock
1366  * @reason: reason why some writeback work was initiated
1367  *
1368  * Implement by try_to_writeback_inodes_sb_nr()
1369  * Returns 1 if writeback was started, 0 if not.
1370  */
1371 int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1372 {
1373         return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1374 }
1375 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
1376 
1377 /**
1378  * sync_inodes_sb       -       sync sb inode pages
1379  * @sb: the superblock
1380  *
1381  * This function writes and waits on any dirty inode belonging to this
1382  * super_block.
1383  */
1384 void sync_inodes_sb(struct super_block *sb)
1385 {
1386         DECLARE_COMPLETION_ONSTACK(done);
1387         struct wb_writeback_work work = {
1388                 .sb             = sb,
1389                 .sync_mode      = WB_SYNC_ALL,
1390                 .nr_pages       = LONG_MAX,
1391                 .range_cyclic   = 0,
1392                 .done           = &done,
1393                 .reason         = WB_REASON_SYNC,
1394                 .for_sync       = 1,
1395         };
1396 
1397         /* Nothing to do? */
1398         if (sb->s_bdi == &noop_backing_dev_info)
1399                 return;
1400         WARN_ON(!rwsem_is_locked(&sb->s_umount));
1401 
1402         bdi_queue_work(sb->s_bdi, &work);
1403         wait_for_completion(&done);
1404 
1405         wait_sb_inodes(sb);
1406 }
1407 EXPORT_SYMBOL(sync_inodes_sb);
1408 
1409 /**
1410  * write_inode_now      -       write an inode to disk
1411  * @inode: inode to write to disk
1412  * @sync: whether the write should be synchronous or not
1413  *
1414  * This function commits an inode to disk immediately if it is dirty. This is
1415  * primarily needed by knfsd.
1416  *
1417  * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1418  */
1419 int write_inode_now(struct inode *inode, int sync)
1420 {
1421         struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1422         struct writeback_control wbc = {
1423                 .nr_to_write = LONG_MAX,
1424                 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1425                 .range_start = 0,
1426                 .range_end = LLONG_MAX,
1427         };
1428 
1429         if (!mapping_cap_writeback_dirty(inode->i_mapping))
1430                 wbc.nr_to_write = 0;
1431 
1432         might_sleep();
1433         return writeback_single_inode(inode, wb, &wbc);
1434 }
1435 EXPORT_SYMBOL(write_inode_now);
1436 
1437 /**
1438  * sync_inode - write an inode and its pages to disk.
1439  * @inode: the inode to sync
1440  * @wbc: controls the writeback mode
1441  *
1442  * sync_inode() will write an inode and its pages to disk.  It will also
1443  * correctly update the inode on its superblock's dirty inode lists and will
1444  * update inode->i_state.
1445  *
1446  * The caller must have a ref on the inode.
1447  */
1448 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1449 {
1450         return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1451 }
1452 EXPORT_SYMBOL(sync_inode);
1453 
1454 /**
1455  * sync_inode_metadata - write an inode to disk
1456  * @inode: the inode to sync
1457  * @wait: wait for I/O to complete.
1458  *
1459  * Write an inode to disk and adjust its dirty state after completion.
1460  *
1461  * Note: only writes the actual inode, no associated data or other metadata.
1462  */
1463 int sync_inode_metadata(struct inode *inode, int wait)
1464 {
1465         struct writeback_control wbc = {
1466                 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1467                 .nr_to_write = 0, /* metadata-only */
1468         };
1469 
1470         return sync_inode(inode, &wbc);
1471 }
1472 EXPORT_SYMBOL(sync_inode_metadata);
1473 

~ [ 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