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

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

Version: ~ [ linux-5.13-rc5 ] ~ [ linux-5.12.9 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.42 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.124 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.193 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.235 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.271 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.271 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * fs/fs-writeback.c
  3  *
  4  * Copyright (C) 2002, Linus Torvalds.
  5  *
  6  * Contains all the functions related to writing back and waiting
  7  * upon dirty inodes against superblocks, and writing back dirty
  8  * pages against inodes.  ie: data writeback.  Writeout of the
  9  * inode itself is not handled here.
 10  *
 11  * 10Apr2002    Andrew Morton
 12  *              Split out of fs/inode.c
 13  *              Additions for address_space-based writeback
 14  */
 15 
 16 #include <linux/kernel.h>
 17 #include <linux/export.h>
 18 #include <linux/spinlock.h>
 19 #include <linux/slab.h>
 20 #include <linux/sched.h>
 21 #include <linux/fs.h>
 22 #include <linux/mm.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/kthread.h>
 25 #include <linux/writeback.h>
 26 #include <linux/blkdev.h>
 27 #include <linux/backing-dev.h>
 28 #include <linux/tracepoint.h>
 29 #include <linux/device.h>
 30 #include <linux/memcontrol.h>
 31 #include "internal.h"
 32 
 33 /*
 34  * 4MB minimal write chunk size
 35  */
 36 #define MIN_WRITEBACK_PAGES     (4096UL >> (PAGE_SHIFT - 10))
 37 
 38 struct wb_completion {
 39         atomic_t                cnt;
 40 };
 41 
 42 /*
 43  * Passed into wb_writeback(), essentially a subset of writeback_control
 44  */
 45 struct wb_writeback_work {
 46         long nr_pages;
 47         struct super_block *sb;
 48         unsigned long *older_than_this;
 49         enum writeback_sync_modes sync_mode;
 50         unsigned int tagged_writepages:1;
 51         unsigned int for_kupdate:1;
 52         unsigned int range_cyclic:1;
 53         unsigned int for_background:1;
 54         unsigned int for_sync:1;        /* sync(2) WB_SYNC_ALL writeback */
 55         unsigned int auto_free:1;       /* free on completion */
 56         enum wb_reason reason;          /* why was writeback initiated? */
 57 
 58         struct list_head list;          /* pending work list */
 59         struct wb_completion *done;     /* set if the caller waits */
 60 };
 61 
 62 /*
 63  * If one wants to wait for one or more wb_writeback_works, each work's
 64  * ->done should be set to a wb_completion defined using the following
 65  * macro.  Once all work items are issued with wb_queue_work(), the caller
 66  * can wait for the completion of all using wb_wait_for_completion().  Work
 67  * items which are waited upon aren't freed automatically on completion.
 68  */
 69 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl)                              \
 70         struct wb_completion cmpl = {                                   \
 71                 .cnt            = ATOMIC_INIT(1),                       \
 72         }
 73 
 74 
 75 /*
 76  * If an inode is constantly having its pages dirtied, but then the
 77  * updates stop dirtytime_expire_interval seconds in the past, it's
 78  * possible for the worst case time between when an inode has its
 79  * timestamps updated and when they finally get written out to be two
 80  * dirtytime_expire_intervals.  We set the default to 12 hours (in
 81  * seconds), which means most of the time inodes will have their
 82  * timestamps written to disk after 12 hours, but in the worst case a
 83  * few inodes might not their timestamps updated for 24 hours.
 84  */
 85 unsigned int dirtytime_expire_interval = 12 * 60 * 60;
 86 
 87 static inline struct inode *wb_inode(struct list_head *head)
 88 {
 89         return list_entry(head, struct inode, i_io_list);
 90 }
 91 
 92 /*
 93  * Include the creation of the trace points after defining the
 94  * wb_writeback_work structure and inline functions so that the definition
 95  * remains local to this file.
 96  */
 97 #define CREATE_TRACE_POINTS
 98 #include <trace/events/writeback.h>
 99 
100 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage);
101 
102 static bool wb_io_lists_populated(struct bdi_writeback *wb)
103 {
104         if (wb_has_dirty_io(wb)) {
105                 return false;
106         } else {
107                 set_bit(WB_has_dirty_io, &wb->state);
108                 WARN_ON_ONCE(!wb->avg_write_bandwidth);
109                 atomic_long_add(wb->avg_write_bandwidth,
110                                 &wb->bdi->tot_write_bandwidth);
111                 return true;
112         }
113 }
114 
115 static void wb_io_lists_depopulated(struct bdi_writeback *wb)
116 {
117         if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) &&
118             list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) {
119                 clear_bit(WB_has_dirty_io, &wb->state);
120                 WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth,
121                                         &wb->bdi->tot_write_bandwidth) < 0);
122         }
123 }
124 
125 /**
126  * inode_io_list_move_locked - move an inode onto a bdi_writeback IO list
127  * @inode: inode to be moved
128  * @wb: target bdi_writeback
129  * @head: one of @wb->b_{dirty|io|more_io|dirty_time}
130  *
131  * Move @inode->i_io_list to @list of @wb and set %WB_has_dirty_io.
132  * Returns %true if @inode is the first occupant of the !dirty_time IO
133  * lists; otherwise, %false.
134  */
135 static bool inode_io_list_move_locked(struct inode *inode,
136                                       struct bdi_writeback *wb,
137                                       struct list_head *head)
138 {
139         assert_spin_locked(&wb->list_lock);
140 
141         list_move(&inode->i_io_list, head);
142 
143         /* dirty_time doesn't count as dirty_io until expiration */
144         if (head != &wb->b_dirty_time)
145                 return wb_io_lists_populated(wb);
146 
147         wb_io_lists_depopulated(wb);
148         return false;
149 }
150 
151 /**
152  * inode_io_list_del_locked - remove an inode from its bdi_writeback IO list
153  * @inode: inode to be removed
154  * @wb: bdi_writeback @inode is being removed from
155  *
156  * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
157  * clear %WB_has_dirty_io if all are empty afterwards.
158  */
159 static void inode_io_list_del_locked(struct inode *inode,
160                                      struct bdi_writeback *wb)
161 {
162         assert_spin_locked(&wb->list_lock);
163 
164         list_del_init(&inode->i_io_list);
165         wb_io_lists_depopulated(wb);
166 }
167 
168 static void wb_wakeup(struct bdi_writeback *wb)
169 {
170         spin_lock_bh(&wb->work_lock);
171         if (test_bit(WB_registered, &wb->state))
172                 mod_delayed_work(bdi_wq, &wb->dwork, 0);
173         spin_unlock_bh(&wb->work_lock);
174 }
175 
176 static void finish_writeback_work(struct bdi_writeback *wb,
177                                   struct wb_writeback_work *work)
178 {
179         struct wb_completion *done = work->done;
180 
181         if (work->auto_free)
182                 kfree(work);
183         if (done && atomic_dec_and_test(&done->cnt))
184                 wake_up_all(&wb->bdi->wb_waitq);
185 }
186 
187 static void wb_queue_work(struct bdi_writeback *wb,
188                           struct wb_writeback_work *work)
189 {
190         trace_writeback_queue(wb, work);
191 
192         if (work->done)
193                 atomic_inc(&work->done->cnt);
194 
195         spin_lock_bh(&wb->work_lock);
196 
197         if (test_bit(WB_registered, &wb->state)) {
198                 list_add_tail(&work->list, &wb->work_list);
199                 mod_delayed_work(bdi_wq, &wb->dwork, 0);
200         } else
201                 finish_writeback_work(wb, work);
202 
203         spin_unlock_bh(&wb->work_lock);
204 }
205 
206 /**
207  * wb_wait_for_completion - wait for completion of bdi_writeback_works
208  * @bdi: bdi work items were issued to
209  * @done: target wb_completion
210  *
211  * Wait for one or more work items issued to @bdi with their ->done field
212  * set to @done, which should have been defined with
213  * DEFINE_WB_COMPLETION_ONSTACK().  This function returns after all such
214  * work items are completed.  Work items which are waited upon aren't freed
215  * automatically on completion.
216  */
217 static void wb_wait_for_completion(struct backing_dev_info *bdi,
218                                    struct wb_completion *done)
219 {
220         atomic_dec(&done->cnt);         /* put down the initial count */
221         wait_event(bdi->wb_waitq, !atomic_read(&done->cnt));
222 }
223 
224 #ifdef CONFIG_CGROUP_WRITEBACK
225 
226 /* parameters for foreign inode detection, see wb_detach_inode() */
227 #define WB_FRN_TIME_SHIFT       13      /* 1s = 2^13, upto 8 secs w/ 16bit */
228 #define WB_FRN_TIME_AVG_SHIFT   3       /* avg = avg * 7/8 + new * 1/8 */
229 #define WB_FRN_TIME_CUT_DIV     2       /* ignore rounds < avg / 2 */
230 #define WB_FRN_TIME_PERIOD      (2 * (1 << WB_FRN_TIME_SHIFT))  /* 2s */
231 
232 #define WB_FRN_HIST_SLOTS       16      /* inode->i_wb_frn_history is 16bit */
233 #define WB_FRN_HIST_UNIT        (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS)
234                                         /* each slot's duration is 2s / 16 */
235 #define WB_FRN_HIST_THR_SLOTS   (WB_FRN_HIST_SLOTS / 2)
236                                         /* if foreign slots >= 8, switch */
237 #define WB_FRN_HIST_MAX_SLOTS   (WB_FRN_HIST_THR_SLOTS / 2 + 1)
238                                         /* one round can affect upto 5 slots */
239 
240 static atomic_t isw_nr_in_flight = ATOMIC_INIT(0);
241 static struct workqueue_struct *isw_wq;
242 
243 void __inode_attach_wb(struct inode *inode, struct page *page)
244 {
245         struct backing_dev_info *bdi = inode_to_bdi(inode);
246         struct bdi_writeback *wb = NULL;
247 
248         if (inode_cgwb_enabled(inode)) {
249                 struct cgroup_subsys_state *memcg_css;
250 
251                 if (page) {
252                         memcg_css = mem_cgroup_css_from_page(page);
253                         wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
254                 } else {
255                         /* must pin memcg_css, see wb_get_create() */
256                         memcg_css = task_get_css(current, memory_cgrp_id);
257                         wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
258                         css_put(memcg_css);
259                 }
260         }
261 
262         if (!wb)
263                 wb = &bdi->wb;
264 
265         /*
266          * There may be multiple instances of this function racing to
267          * update the same inode.  Use cmpxchg() to tell the winner.
268          */
269         if (unlikely(cmpxchg(&inode->i_wb, NULL, wb)))
270                 wb_put(wb);
271 }
272 
273 /**
274  * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it
275  * @inode: inode of interest with i_lock held
276  *
277  * Returns @inode's wb with its list_lock held.  @inode->i_lock must be
278  * held on entry and is released on return.  The returned wb is guaranteed
279  * to stay @inode's associated wb until its list_lock is released.
280  */
281 static struct bdi_writeback *
282 locked_inode_to_wb_and_lock_list(struct inode *inode)
283         __releases(&inode->i_lock)
284         __acquires(&wb->list_lock)
285 {
286         while (true) {
287                 struct bdi_writeback *wb = inode_to_wb(inode);
288 
289                 /*
290                  * inode_to_wb() association is protected by both
291                  * @inode->i_lock and @wb->list_lock but list_lock nests
292                  * outside i_lock.  Drop i_lock and verify that the
293                  * association hasn't changed after acquiring list_lock.
294                  */
295                 wb_get(wb);
296                 spin_unlock(&inode->i_lock);
297                 spin_lock(&wb->list_lock);
298 
299                 /* i_wb may have changed inbetween, can't use inode_to_wb() */
300                 if (likely(wb == inode->i_wb)) {
301                         wb_put(wb);     /* @inode already has ref */
302                         return wb;
303                 }
304 
305                 spin_unlock(&wb->list_lock);
306                 wb_put(wb);
307                 cpu_relax();
308                 spin_lock(&inode->i_lock);
309         }
310 }
311 
312 /**
313  * inode_to_wb_and_lock_list - determine an inode's wb and lock it
314  * @inode: inode of interest
315  *
316  * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held
317  * on entry.
318  */
319 static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode)
320         __acquires(&wb->list_lock)
321 {
322         spin_lock(&inode->i_lock);
323         return locked_inode_to_wb_and_lock_list(inode);
324 }
325 
326 struct inode_switch_wbs_context {
327         struct inode            *inode;
328         struct bdi_writeback    *new_wb;
329 
330         struct rcu_head         rcu_head;
331         struct work_struct      work;
332 };
333 
334 static void inode_switch_wbs_work_fn(struct work_struct *work)
335 {
336         struct inode_switch_wbs_context *isw =
337                 container_of(work, struct inode_switch_wbs_context, work);
338         struct inode *inode = isw->inode;
339         struct address_space *mapping = inode->i_mapping;
340         struct bdi_writeback *old_wb = inode->i_wb;
341         struct bdi_writeback *new_wb = isw->new_wb;
342         struct radix_tree_iter iter;
343         bool switched = false;
344         void **slot;
345 
346         /*
347          * By the time control reaches here, RCU grace period has passed
348          * since I_WB_SWITCH assertion and all wb stat update transactions
349          * between unlocked_inode_to_wb_begin/end() are guaranteed to be
350          * synchronizing against mapping->tree_lock.
351          *
352          * Grabbing old_wb->list_lock, inode->i_lock and mapping->tree_lock
353          * gives us exclusion against all wb related operations on @inode
354          * including IO list manipulations and stat updates.
355          */
356         if (old_wb < new_wb) {
357                 spin_lock(&old_wb->list_lock);
358                 spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING);
359         } else {
360                 spin_lock(&new_wb->list_lock);
361                 spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING);
362         }
363         spin_lock(&inode->i_lock);
364         spin_lock_irq(&mapping->tree_lock);
365 
366         /*
367          * Once I_FREEING is visible under i_lock, the eviction path owns
368          * the inode and we shouldn't modify ->i_io_list.
369          */
370         if (unlikely(inode->i_state & I_FREEING))
371                 goto skip_switch;
372 
373         /*
374          * Count and transfer stats.  Note that PAGECACHE_TAG_DIRTY points
375          * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to
376          * pages actually under underwriteback.
377          */
378         radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0,
379                                    PAGECACHE_TAG_DIRTY) {
380                 struct page *page = radix_tree_deref_slot_protected(slot,
381                                                         &mapping->tree_lock);
382                 if (likely(page) && PageDirty(page)) {
383                         dec_wb_stat(old_wb, WB_RECLAIMABLE);
384                         inc_wb_stat(new_wb, WB_RECLAIMABLE);
385                 }
386         }
387 
388         radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter, 0,
389                                    PAGECACHE_TAG_WRITEBACK) {
390                 struct page *page = radix_tree_deref_slot_protected(slot,
391                                                         &mapping->tree_lock);
392                 if (likely(page)) {
393                         WARN_ON_ONCE(!PageWriteback(page));
394                         dec_wb_stat(old_wb, WB_WRITEBACK);
395                         inc_wb_stat(new_wb, WB_WRITEBACK);
396                 }
397         }
398 
399         wb_get(new_wb);
400 
401         /*
402          * Transfer to @new_wb's IO list if necessary.  The specific list
403          * @inode was on is ignored and the inode is put on ->b_dirty which
404          * is always correct including from ->b_dirty_time.  The transfer
405          * preserves @inode->dirtied_when ordering.
406          */
407         if (!list_empty(&inode->i_io_list)) {
408                 struct inode *pos;
409 
410                 inode_io_list_del_locked(inode, old_wb);
411                 inode->i_wb = new_wb;
412                 list_for_each_entry(pos, &new_wb->b_dirty, i_io_list)
413                         if (time_after_eq(inode->dirtied_when,
414                                           pos->dirtied_when))
415                                 break;
416                 inode_io_list_move_locked(inode, new_wb, pos->i_io_list.prev);
417         } else {
418                 inode->i_wb = new_wb;
419         }
420 
421         /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */
422         inode->i_wb_frn_winner = 0;
423         inode->i_wb_frn_avg_time = 0;
424         inode->i_wb_frn_history = 0;
425         switched = true;
426 skip_switch:
427         /*
428          * Paired with load_acquire in unlocked_inode_to_wb_begin() and
429          * ensures that the new wb is visible if they see !I_WB_SWITCH.
430          */
431         smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH);
432 
433         spin_unlock_irq(&mapping->tree_lock);
434         spin_unlock(&inode->i_lock);
435         spin_unlock(&new_wb->list_lock);
436         spin_unlock(&old_wb->list_lock);
437 
438         if (switched) {
439                 wb_wakeup(new_wb);
440                 wb_put(old_wb);
441         }
442         wb_put(new_wb);
443 
444         iput(inode);
445         kfree(isw);
446 
447         atomic_dec(&isw_nr_in_flight);
448 }
449 
450 static void inode_switch_wbs_rcu_fn(struct rcu_head *rcu_head)
451 {
452         struct inode_switch_wbs_context *isw = container_of(rcu_head,
453                                 struct inode_switch_wbs_context, rcu_head);
454 
455         /* needs to grab bh-unsafe locks, bounce to work item */
456         INIT_WORK(&isw->work, inode_switch_wbs_work_fn);
457         queue_work(isw_wq, &isw->work);
458 }
459 
460 /**
461  * inode_switch_wbs - change the wb association of an inode
462  * @inode: target inode
463  * @new_wb_id: ID of the new wb
464  *
465  * Switch @inode's wb association to the wb identified by @new_wb_id.  The
466  * switching is performed asynchronously and may fail silently.
467  */
468 static void inode_switch_wbs(struct inode *inode, int new_wb_id)
469 {
470         struct backing_dev_info *bdi = inode_to_bdi(inode);
471         struct cgroup_subsys_state *memcg_css;
472         struct inode_switch_wbs_context *isw;
473 
474         /* noop if seems to be already in progress */
475         if (inode->i_state & I_WB_SWITCH)
476                 return;
477 
478         isw = kzalloc(sizeof(*isw), GFP_ATOMIC);
479         if (!isw)
480                 return;
481 
482         /* find and pin the new wb */
483         rcu_read_lock();
484         memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys);
485         if (memcg_css)
486                 isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC);
487         rcu_read_unlock();
488         if (!isw->new_wb)
489                 goto out_free;
490 
491         /* while holding I_WB_SWITCH, no one else can update the association */
492         spin_lock(&inode->i_lock);
493         if (!(inode->i_sb->s_flags & SB_ACTIVE) ||
494             inode->i_state & (I_WB_SWITCH | I_FREEING) ||
495             inode_to_wb(inode) == isw->new_wb) {
496                 spin_unlock(&inode->i_lock);
497                 goto out_free;
498         }
499         inode->i_state |= I_WB_SWITCH;
500         __iget(inode);
501         spin_unlock(&inode->i_lock);
502 
503         isw->inode = inode;
504 
505         atomic_inc(&isw_nr_in_flight);
506 
507         /*
508          * In addition to synchronizing among switchers, I_WB_SWITCH tells
509          * the RCU protected stat update paths to grab the mapping's
510          * tree_lock so that stat transfer can synchronize against them.
511          * Let's continue after I_WB_SWITCH is guaranteed to be visible.
512          */
513         call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn);
514         return;
515 
516 out_free:
517         if (isw->new_wb)
518                 wb_put(isw->new_wb);
519         kfree(isw);
520 }
521 
522 /**
523  * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it
524  * @wbc: writeback_control of interest
525  * @inode: target inode
526  *
527  * @inode is locked and about to be written back under the control of @wbc.
528  * Record @inode's writeback context into @wbc and unlock the i_lock.  On
529  * writeback completion, wbc_detach_inode() should be called.  This is used
530  * to track the cgroup writeback context.
531  */
532 void wbc_attach_and_unlock_inode(struct writeback_control *wbc,
533                                  struct inode *inode)
534 {
535         if (!inode_cgwb_enabled(inode)) {
536                 spin_unlock(&inode->i_lock);
537                 return;
538         }
539 
540         wbc->wb = inode_to_wb(inode);
541         wbc->inode = inode;
542 
543         wbc->wb_id = wbc->wb->memcg_css->id;
544         wbc->wb_lcand_id = inode->i_wb_frn_winner;
545         wbc->wb_tcand_id = 0;
546         wbc->wb_bytes = 0;
547         wbc->wb_lcand_bytes = 0;
548         wbc->wb_tcand_bytes = 0;
549 
550         wb_get(wbc->wb);
551         spin_unlock(&inode->i_lock);
552 
553         /*
554          * A dying wb indicates that the memcg-blkcg mapping has changed
555          * and a new wb is already serving the memcg.  Switch immediately.
556          */
557         if (unlikely(wb_dying(wbc->wb)))
558                 inode_switch_wbs(inode, wbc->wb_id);
559 }
560 
561 /**
562  * wbc_detach_inode - disassociate wbc from inode and perform foreign detection
563  * @wbc: writeback_control of the just finished writeback
564  *
565  * To be called after a writeback attempt of an inode finishes and undoes
566  * wbc_attach_and_unlock_inode().  Can be called under any context.
567  *
568  * As concurrent write sharing of an inode is expected to be very rare and
569  * memcg only tracks page ownership on first-use basis severely confining
570  * the usefulness of such sharing, cgroup writeback tracks ownership
571  * per-inode.  While the support for concurrent write sharing of an inode
572  * is deemed unnecessary, an inode being written to by different cgroups at
573  * different points in time is a lot more common, and, more importantly,
574  * charging only by first-use can too readily lead to grossly incorrect
575  * behaviors (single foreign page can lead to gigabytes of writeback to be
576  * incorrectly attributed).
577  *
578  * To resolve this issue, cgroup writeback detects the majority dirtier of
579  * an inode and transfers the ownership to it.  To avoid unnnecessary
580  * oscillation, the detection mechanism keeps track of history and gives
581  * out the switch verdict only if the foreign usage pattern is stable over
582  * a certain amount of time and/or writeback attempts.
583  *
584  * On each writeback attempt, @wbc tries to detect the majority writer
585  * using Boyer-Moore majority vote algorithm.  In addition to the byte
586  * count from the majority voting, it also counts the bytes written for the
587  * current wb and the last round's winner wb (max of last round's current
588  * wb, the winner from two rounds ago, and the last round's majority
589  * candidate).  Keeping track of the historical winner helps the algorithm
590  * to semi-reliably detect the most active writer even when it's not the
591  * absolute majority.
592  *
593  * Once the winner of the round is determined, whether the winner is
594  * foreign or not and how much IO time the round consumed is recorded in
595  * inode->i_wb_frn_history.  If the amount of recorded foreign IO time is
596  * over a certain threshold, the switch verdict is given.
597  */
598 void wbc_detach_inode(struct writeback_control *wbc)
599 {
600         struct bdi_writeback *wb = wbc->wb;
601         struct inode *inode = wbc->inode;
602         unsigned long avg_time, max_bytes, max_time;
603         u16 history;
604         int max_id;
605 
606         if (!wb)
607                 return;
608 
609         history = inode->i_wb_frn_history;
610         avg_time = inode->i_wb_frn_avg_time;
611 
612         /* pick the winner of this round */
613         if (wbc->wb_bytes >= wbc->wb_lcand_bytes &&
614             wbc->wb_bytes >= wbc->wb_tcand_bytes) {
615                 max_id = wbc->wb_id;
616                 max_bytes = wbc->wb_bytes;
617         } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) {
618                 max_id = wbc->wb_lcand_id;
619                 max_bytes = wbc->wb_lcand_bytes;
620         } else {
621                 max_id = wbc->wb_tcand_id;
622                 max_bytes = wbc->wb_tcand_bytes;
623         }
624 
625         /*
626          * Calculate the amount of IO time the winner consumed and fold it
627          * into the running average kept per inode.  If the consumed IO
628          * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for
629          * deciding whether to switch or not.  This is to prevent one-off
630          * small dirtiers from skewing the verdict.
631          */
632         max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT,
633                                 wb->avg_write_bandwidth);
634         if (avg_time)
635                 avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) -
636                             (avg_time >> WB_FRN_TIME_AVG_SHIFT);
637         else
638                 avg_time = max_time;    /* immediate catch up on first run */
639 
640         if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) {
641                 int slots;
642 
643                 /*
644                  * The switch verdict is reached if foreign wb's consume
645                  * more than a certain proportion of IO time in a
646                  * WB_FRN_TIME_PERIOD.  This is loosely tracked by 16 slot
647                  * history mask where each bit represents one sixteenth of
648                  * the period.  Determine the number of slots to shift into
649                  * history from @max_time.
650                  */
651                 slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT),
652                             (unsigned long)WB_FRN_HIST_MAX_SLOTS);
653                 history <<= slots;
654                 if (wbc->wb_id != max_id)
655                         history |= (1U << slots) - 1;
656 
657                 /*
658                  * Switch if the current wb isn't the consistent winner.
659                  * If there are multiple closely competing dirtiers, the
660                  * inode may switch across them repeatedly over time, which
661                  * is okay.  The main goal is avoiding keeping an inode on
662                  * the wrong wb for an extended period of time.
663                  */
664                 if (hweight32(history) > WB_FRN_HIST_THR_SLOTS)
665                         inode_switch_wbs(inode, max_id);
666         }
667 
668         /*
669          * Multiple instances of this function may race to update the
670          * following fields but we don't mind occassional inaccuracies.
671          */
672         inode->i_wb_frn_winner = max_id;
673         inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX);
674         inode->i_wb_frn_history = history;
675 
676         wb_put(wbc->wb);
677         wbc->wb = NULL;
678 }
679 
680 /**
681  * wbc_account_io - account IO issued during writeback
682  * @wbc: writeback_control of the writeback in progress
683  * @page: page being written out
684  * @bytes: number of bytes being written out
685  *
686  * @bytes from @page are about to written out during the writeback
687  * controlled by @wbc.  Keep the book for foreign inode detection.  See
688  * wbc_detach_inode().
689  */
690 void wbc_account_io(struct writeback_control *wbc, struct page *page,
691                     size_t bytes)
692 {
693         int id;
694 
695         /*
696          * pageout() path doesn't attach @wbc to the inode being written
697          * out.  This is intentional as we don't want the function to block
698          * behind a slow cgroup.  Ultimately, we want pageout() to kick off
699          * regular writeback instead of writing things out itself.
700          */
701         if (!wbc->wb)
702                 return;
703 
704         id = mem_cgroup_css_from_page(page)->id;
705 
706         if (id == wbc->wb_id) {
707                 wbc->wb_bytes += bytes;
708                 return;
709         }
710 
711         if (id == wbc->wb_lcand_id)
712                 wbc->wb_lcand_bytes += bytes;
713 
714         /* Boyer-Moore majority vote algorithm */
715         if (!wbc->wb_tcand_bytes)
716                 wbc->wb_tcand_id = id;
717         if (id == wbc->wb_tcand_id)
718                 wbc->wb_tcand_bytes += bytes;
719         else
720                 wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes);
721 }
722 EXPORT_SYMBOL_GPL(wbc_account_io);
723 
724 /**
725  * inode_congested - test whether an inode is congested
726  * @inode: inode to test for congestion (may be NULL)
727  * @cong_bits: mask of WB_[a]sync_congested bits to test
728  *
729  * Tests whether @inode is congested.  @cong_bits is the mask of congestion
730  * bits to test and the return value is the mask of set bits.
731  *
732  * If cgroup writeback is enabled for @inode, the congestion state is
733  * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
734  * associated with @inode is congested; otherwise, the root wb's congestion
735  * state is used.
736  *
737  * @inode is allowed to be NULL as this function is often called on
738  * mapping->host which is NULL for the swapper space.
739  */
740 int inode_congested(struct inode *inode, int cong_bits)
741 {
742         /*
743          * Once set, ->i_wb never becomes NULL while the inode is alive.
744          * Start transaction iff ->i_wb is visible.
745          */
746         if (inode && inode_to_wb_is_valid(inode)) {
747                 struct bdi_writeback *wb;
748                 struct wb_lock_cookie lock_cookie = {};
749                 bool congested;
750 
751                 wb = unlocked_inode_to_wb_begin(inode, &lock_cookie);
752                 congested = wb_congested(wb, cong_bits);
753                 unlocked_inode_to_wb_end(inode, &lock_cookie);
754                 return congested;
755         }
756 
757         return wb_congested(&inode_to_bdi(inode)->wb, cong_bits);
758 }
759 EXPORT_SYMBOL_GPL(inode_congested);
760 
761 /**
762  * wb_split_bdi_pages - split nr_pages to write according to bandwidth
763  * @wb: target bdi_writeback to split @nr_pages to
764  * @nr_pages: number of pages to write for the whole bdi
765  *
766  * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
767  * relation to the total write bandwidth of all wb's w/ dirty inodes on
768  * @wb->bdi.
769  */
770 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
771 {
772         unsigned long this_bw = wb->avg_write_bandwidth;
773         unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth);
774 
775         if (nr_pages == LONG_MAX)
776                 return LONG_MAX;
777 
778         /*
779          * This may be called on clean wb's and proportional distribution
780          * may not make sense, just use the original @nr_pages in those
781          * cases.  In general, we wanna err on the side of writing more.
782          */
783         if (!tot_bw || this_bw >= tot_bw)
784                 return nr_pages;
785         else
786                 return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw);
787 }
788 
789 /**
790  * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
791  * @bdi: target backing_dev_info
792  * @base_work: wb_writeback_work to issue
793  * @skip_if_busy: skip wb's which already have writeback in progress
794  *
795  * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
796  * have dirty inodes.  If @base_work->nr_page isn't %LONG_MAX, it's
797  * distributed to the busy wbs according to each wb's proportion in the
798  * total active write bandwidth of @bdi.
799  */
800 static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
801                                   struct wb_writeback_work *base_work,
802                                   bool skip_if_busy)
803 {
804         struct bdi_writeback *last_wb = NULL;
805         struct bdi_writeback *wb = list_entry(&bdi->wb_list,
806                                               struct bdi_writeback, bdi_node);
807 
808         might_sleep();
809 restart:
810         rcu_read_lock();
811         list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) {
812                 DEFINE_WB_COMPLETION_ONSTACK(fallback_work_done);
813                 struct wb_writeback_work fallback_work;
814                 struct wb_writeback_work *work;
815                 long nr_pages;
816 
817                 if (last_wb) {
818                         wb_put(last_wb);
819                         last_wb = NULL;
820                 }
821 
822                 /* SYNC_ALL writes out I_DIRTY_TIME too */
823                 if (!wb_has_dirty_io(wb) &&
824                     (base_work->sync_mode == WB_SYNC_NONE ||
825                      list_empty(&wb->b_dirty_time)))
826                         continue;
827                 if (skip_if_busy && writeback_in_progress(wb))
828                         continue;
829 
830                 nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages);
831 
832                 work = kmalloc(sizeof(*work), GFP_ATOMIC);
833                 if (work) {
834                         *work = *base_work;
835                         work->nr_pages = nr_pages;
836                         work->auto_free = 1;
837                         wb_queue_work(wb, work);
838                         continue;
839                 }
840 
841                 /* alloc failed, execute synchronously using on-stack fallback */
842                 work = &fallback_work;
843                 *work = *base_work;
844                 work->nr_pages = nr_pages;
845                 work->auto_free = 0;
846                 work->done = &fallback_work_done;
847 
848                 wb_queue_work(wb, work);
849 
850                 /*
851                  * Pin @wb so that it stays on @bdi->wb_list.  This allows
852                  * continuing iteration from @wb after dropping and
853                  * regrabbing rcu read lock.
854                  */
855                 wb_get(wb);
856                 last_wb = wb;
857 
858                 rcu_read_unlock();
859                 wb_wait_for_completion(bdi, &fallback_work_done);
860                 goto restart;
861         }
862         rcu_read_unlock();
863 
864         if (last_wb)
865                 wb_put(last_wb);
866 }
867 
868 /**
869  * cgroup_writeback_umount - flush inode wb switches for umount
870  *
871  * This function is called when a super_block is about to be destroyed and
872  * flushes in-flight inode wb switches.  An inode wb switch goes through
873  * RCU and then workqueue, so the two need to be flushed in order to ensure
874  * that all previously scheduled switches are finished.  As wb switches are
875  * rare occurrences and synchronize_rcu() can take a while, perform
876  * flushing iff wb switches are in flight.
877  */
878 void cgroup_writeback_umount(void)
879 {
880         if (atomic_read(&isw_nr_in_flight)) {
881                 synchronize_rcu();
882                 flush_workqueue(isw_wq);
883         }
884 }
885 
886 static int __init cgroup_writeback_init(void)
887 {
888         isw_wq = alloc_workqueue("inode_switch_wbs", 0, 0);
889         if (!isw_wq)
890                 return -ENOMEM;
891         return 0;
892 }
893 fs_initcall(cgroup_writeback_init);
894 
895 #else   /* CONFIG_CGROUP_WRITEBACK */
896 
897 static struct bdi_writeback *
898 locked_inode_to_wb_and_lock_list(struct inode *inode)
899         __releases(&inode->i_lock)
900         __acquires(&wb->list_lock)
901 {
902         struct bdi_writeback *wb = inode_to_wb(inode);
903 
904         spin_unlock(&inode->i_lock);
905         spin_lock(&wb->list_lock);
906         return wb;
907 }
908 
909 static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode)
910         __acquires(&wb->list_lock)
911 {
912         struct bdi_writeback *wb = inode_to_wb(inode);
913 
914         spin_lock(&wb->list_lock);
915         return wb;
916 }
917 
918 static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages)
919 {
920         return nr_pages;
921 }
922 
923 static void bdi_split_work_to_wbs(struct backing_dev_info *bdi,
924                                   struct wb_writeback_work *base_work,
925                                   bool skip_if_busy)
926 {
927         might_sleep();
928 
929         if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) {
930                 base_work->auto_free = 0;
931                 wb_queue_work(&bdi->wb, base_work);
932         }
933 }
934 
935 #endif  /* CONFIG_CGROUP_WRITEBACK */
936 
937 /*
938  * Add in the number of potentially dirty inodes, because each inode
939  * write can dirty pagecache in the underlying blockdev.
940  */
941 static unsigned long get_nr_dirty_pages(void)
942 {
943         return global_node_page_state(NR_FILE_DIRTY) +
944                 global_node_page_state(NR_UNSTABLE_NFS) +
945                 get_nr_dirty_inodes();
946 }
947 
948 static void wb_start_writeback(struct bdi_writeback *wb, enum wb_reason reason)
949 {
950         if (!wb_has_dirty_io(wb))
951                 return;
952 
953         /*
954          * All callers of this function want to start writeback of all
955          * dirty pages. Places like vmscan can call this at a very
956          * high frequency, causing pointless allocations of tons of
957          * work items and keeping the flusher threads busy retrieving
958          * that work. Ensure that we only allow one of them pending and
959          * inflight at the time.
960          */
961         if (test_bit(WB_start_all, &wb->state) ||
962             test_and_set_bit(WB_start_all, &wb->state))
963                 return;
964 
965         wb->start_all_reason = reason;
966         wb_wakeup(wb);
967 }
968 
969 /**
970  * wb_start_background_writeback - start background writeback
971  * @wb: bdi_writback to write from
972  *
973  * Description:
974  *   This makes sure WB_SYNC_NONE background writeback happens. When
975  *   this function returns, it is only guaranteed that for given wb
976  *   some IO is happening if we are over background dirty threshold.
977  *   Caller need not hold sb s_umount semaphore.
978  */
979 void wb_start_background_writeback(struct bdi_writeback *wb)
980 {
981         /*
982          * We just wake up the flusher thread. It will perform background
983          * writeback as soon as there is no other work to do.
984          */
985         trace_writeback_wake_background(wb);
986         wb_wakeup(wb);
987 }
988 
989 /*
990  * Remove the inode from the writeback list it is on.
991  */
992 void inode_io_list_del(struct inode *inode)
993 {
994         struct bdi_writeback *wb;
995 
996         wb = inode_to_wb_and_lock_list(inode);
997         inode_io_list_del_locked(inode, wb);
998         spin_unlock(&wb->list_lock);
999 }
1000 
1001 /*
1002  * mark an inode as under writeback on the sb
1003  */
1004 void sb_mark_inode_writeback(struct inode *inode)
1005 {
1006         struct super_block *sb = inode->i_sb;
1007         unsigned long flags;
1008 
1009         if (list_empty(&inode->i_wb_list)) {
1010                 spin_lock_irqsave(&sb->s_inode_wblist_lock, flags);
1011                 if (list_empty(&inode->i_wb_list)) {
1012                         list_add_tail(&inode->i_wb_list, &sb->s_inodes_wb);
1013                         trace_sb_mark_inode_writeback(inode);
1014                 }
1015                 spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags);
1016         }
1017 }
1018 
1019 /*
1020  * clear an inode as under writeback on the sb
1021  */
1022 void sb_clear_inode_writeback(struct inode *inode)
1023 {
1024         struct super_block *sb = inode->i_sb;
1025         unsigned long flags;
1026 
1027         if (!list_empty(&inode->i_wb_list)) {
1028                 spin_lock_irqsave(&sb->s_inode_wblist_lock, flags);
1029                 if (!list_empty(&inode->i_wb_list)) {
1030                         list_del_init(&inode->i_wb_list);
1031                         trace_sb_clear_inode_writeback(inode);
1032                 }
1033                 spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags);
1034         }
1035 }
1036 
1037 /*
1038  * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
1039  * furthest end of its superblock's dirty-inode list.
1040  *
1041  * Before stamping the inode's ->dirtied_when, we check to see whether it is
1042  * already the most-recently-dirtied inode on the b_dirty list.  If that is
1043  * the case then the inode must have been redirtied while it was being written
1044  * out and we don't reset its dirtied_when.
1045  */
1046 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
1047 {
1048         if (!list_empty(&wb->b_dirty)) {
1049                 struct inode *tail;
1050 
1051                 tail = wb_inode(wb->b_dirty.next);
1052                 if (time_before(inode->dirtied_when, tail->dirtied_when))
1053                         inode->dirtied_when = jiffies;
1054         }
1055         inode_io_list_move_locked(inode, wb, &wb->b_dirty);
1056 }
1057 
1058 /*
1059  * requeue inode for re-scanning after bdi->b_io list is exhausted.
1060  */
1061 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
1062 {
1063         inode_io_list_move_locked(inode, wb, &wb->b_more_io);
1064 }
1065 
1066 static void inode_sync_complete(struct inode *inode)
1067 {
1068         inode->i_state &= ~I_SYNC;
1069         /* If inode is clean an unused, put it into LRU now... */
1070         inode_add_lru(inode);
1071         /* Waiters must see I_SYNC cleared before being woken up */
1072         smp_mb();
1073         wake_up_bit(&inode->i_state, __I_SYNC);
1074 }
1075 
1076 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
1077 {
1078         bool ret = time_after(inode->dirtied_when, t);
1079 #ifndef CONFIG_64BIT
1080         /*
1081          * For inodes being constantly redirtied, dirtied_when can get stuck.
1082          * It _appears_ to be in the future, but is actually in distant past.
1083          * This test is necessary to prevent such wrapped-around relative times
1084          * from permanently stopping the whole bdi writeback.
1085          */
1086         ret = ret && time_before_eq(inode->dirtied_when, jiffies);
1087 #endif
1088         return ret;
1089 }
1090 
1091 #define EXPIRE_DIRTY_ATIME 0x0001
1092 
1093 /*
1094  * Move expired (dirtied before work->older_than_this) dirty inodes from
1095  * @delaying_queue to @dispatch_queue.
1096  */
1097 static int move_expired_inodes(struct list_head *delaying_queue,
1098                                struct list_head *dispatch_queue,
1099                                int flags,
1100                                struct wb_writeback_work *work)
1101 {
1102         unsigned long *older_than_this = NULL;
1103         unsigned long expire_time;
1104         LIST_HEAD(tmp);
1105         struct list_head *pos, *node;
1106         struct super_block *sb = NULL;
1107         struct inode *inode;
1108         int do_sb_sort = 0;
1109         int moved = 0;
1110 
1111         if ((flags & EXPIRE_DIRTY_ATIME) == 0)
1112                 older_than_this = work->older_than_this;
1113         else if (!work->for_sync) {
1114                 expire_time = jiffies - (dirtytime_expire_interval * HZ);
1115                 older_than_this = &expire_time;
1116         }
1117         while (!list_empty(delaying_queue)) {
1118                 inode = wb_inode(delaying_queue->prev);
1119                 if (older_than_this &&
1120                     inode_dirtied_after(inode, *older_than_this))
1121                         break;
1122                 list_move(&inode->i_io_list, &tmp);
1123                 moved++;
1124                 if (flags & EXPIRE_DIRTY_ATIME)
1125                         set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state);
1126                 if (sb_is_blkdev_sb(inode->i_sb))
1127                         continue;
1128                 if (sb && sb != inode->i_sb)
1129                         do_sb_sort = 1;
1130                 sb = inode->i_sb;
1131         }
1132 
1133         /* just one sb in list, splice to dispatch_queue and we're done */
1134         if (!do_sb_sort) {
1135                 list_splice(&tmp, dispatch_queue);
1136                 goto out;
1137         }
1138 
1139         /* Move inodes from one superblock together */
1140         while (!list_empty(&tmp)) {
1141                 sb = wb_inode(tmp.prev)->i_sb;
1142                 list_for_each_prev_safe(pos, node, &tmp) {
1143                         inode = wb_inode(pos);
1144                         if (inode->i_sb == sb)
1145                                 list_move(&inode->i_io_list, dispatch_queue);
1146                 }
1147         }
1148 out:
1149         return moved;
1150 }
1151 
1152 /*
1153  * Queue all expired dirty inodes for io, eldest first.
1154  * Before
1155  *         newly dirtied     b_dirty    b_io    b_more_io
1156  *         =============>    gf         edc     BA
1157  * After
1158  *         newly dirtied     b_dirty    b_io    b_more_io
1159  *         =============>    g          fBAedc
1160  *                                           |
1161  *                                           +--> dequeue for IO
1162  */
1163 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
1164 {
1165         int moved;
1166 
1167         assert_spin_locked(&wb->list_lock);
1168         list_splice_init(&wb->b_more_io, &wb->b_io);
1169         moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work);
1170         moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io,
1171                                      EXPIRE_DIRTY_ATIME, work);
1172         if (moved)
1173                 wb_io_lists_populated(wb);
1174         trace_writeback_queue_io(wb, work, moved);
1175 }
1176 
1177 static int write_inode(struct inode *inode, struct writeback_control *wbc)
1178 {
1179         int ret;
1180 
1181         if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) {
1182                 trace_writeback_write_inode_start(inode, wbc);
1183                 ret = inode->i_sb->s_op->write_inode(inode, wbc);
1184                 trace_writeback_write_inode(inode, wbc);
1185                 return ret;
1186         }
1187         return 0;
1188 }
1189 
1190 /*
1191  * Wait for writeback on an inode to complete. Called with i_lock held.
1192  * Caller must make sure inode cannot go away when we drop i_lock.
1193  */
1194 static void __inode_wait_for_writeback(struct inode *inode)
1195         __releases(inode->i_lock)
1196         __acquires(inode->i_lock)
1197 {
1198         DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
1199         wait_queue_head_t *wqh;
1200 
1201         wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
1202         while (inode->i_state & I_SYNC) {
1203                 spin_unlock(&inode->i_lock);
1204                 __wait_on_bit(wqh, &wq, bit_wait,
1205                               TASK_UNINTERRUPTIBLE);
1206                 spin_lock(&inode->i_lock);
1207         }
1208 }
1209 
1210 /*
1211  * Wait for writeback on an inode to complete. Caller must have inode pinned.
1212  */
1213 void inode_wait_for_writeback(struct inode *inode)
1214 {
1215         spin_lock(&inode->i_lock);
1216         __inode_wait_for_writeback(inode);
1217         spin_unlock(&inode->i_lock);
1218 }
1219 
1220 /*
1221  * Sleep until I_SYNC is cleared. This function must be called with i_lock
1222  * held and drops it. It is aimed for callers not holding any inode reference
1223  * so once i_lock is dropped, inode can go away.
1224  */
1225 static void inode_sleep_on_writeback(struct inode *inode)
1226         __releases(inode->i_lock)
1227 {
1228         DEFINE_WAIT(wait);
1229         wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
1230         int sleep;
1231 
1232         prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1233         sleep = inode->i_state & I_SYNC;
1234         spin_unlock(&inode->i_lock);
1235         if (sleep)
1236                 schedule();
1237         finish_wait(wqh, &wait);
1238 }
1239 
1240 /*
1241  * Find proper writeback list for the inode depending on its current state and
1242  * possibly also change of its state while we were doing writeback.  Here we
1243  * handle things such as livelock prevention or fairness of writeback among
1244  * inodes. This function can be called only by flusher thread - noone else
1245  * processes all inodes in writeback lists and requeueing inodes behind flusher
1246  * thread's back can have unexpected consequences.
1247  */
1248 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
1249                           struct writeback_control *wbc)
1250 {
1251         if (inode->i_state & I_FREEING)
1252                 return;
1253 
1254         /*
1255          * Sync livelock prevention. Each inode is tagged and synced in one
1256          * shot. If still dirty, it will be redirty_tail()'ed below.  Update
1257          * the dirty time to prevent enqueue and sync it again.
1258          */
1259         if ((inode->i_state & I_DIRTY) &&
1260             (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
1261                 inode->dirtied_when = jiffies;
1262 
1263         if (wbc->pages_skipped) {
1264                 /*
1265                  * writeback is not making progress due to locked
1266                  * buffers. Skip this inode for now.
1267                  */
1268                 redirty_tail(inode, wb);
1269                 return;
1270         }
1271 
1272         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
1273                 /*
1274                  * We didn't write back all the pages.  nfs_writepages()
1275                  * sometimes bales out without doing anything.
1276                  */
1277                 if (wbc->nr_to_write <= 0) {
1278                         /* Slice used up. Queue for next turn. */
1279                         requeue_io(inode, wb);
1280                 } else {
1281                         /*
1282                          * Writeback blocked by something other than
1283                          * congestion. Delay the inode for some time to
1284                          * avoid spinning on the CPU (100% iowait)
1285                          * retrying writeback of the dirty page/inode
1286                          * that cannot be performed immediately.
1287                          */
1288                         redirty_tail(inode, wb);
1289                 }
1290         } else if (inode->i_state & I_DIRTY) {
1291                 /*
1292                  * Filesystems can dirty the inode during writeback operations,
1293                  * such as delayed allocation during submission or metadata
1294                  * updates after data IO completion.
1295                  */
1296                 redirty_tail(inode, wb);
1297         } else if (inode->i_state & I_DIRTY_TIME) {
1298                 inode->dirtied_when = jiffies;
1299                 inode_io_list_move_locked(inode, wb, &wb->b_dirty_time);
1300         } else {
1301                 /* The inode is clean. Remove from writeback lists. */
1302                 inode_io_list_del_locked(inode, wb);
1303         }
1304 }
1305 
1306 /*
1307  * Write out an inode and its dirty pages. Do not update the writeback list
1308  * linkage. That is left to the caller. The caller is also responsible for
1309  * setting I_SYNC flag and calling inode_sync_complete() to clear it.
1310  */
1311 static int
1312 __writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
1313 {
1314         struct address_space *mapping = inode->i_mapping;
1315         long nr_to_write = wbc->nr_to_write;
1316         unsigned dirty;
1317         int ret;
1318 
1319         WARN_ON(!(inode->i_state & I_SYNC));
1320 
1321         trace_writeback_single_inode_start(inode, wbc, nr_to_write);
1322 
1323         ret = do_writepages(mapping, wbc);
1324 
1325         /*
1326          * Make sure to wait on the data before writing out the metadata.
1327          * This is important for filesystems that modify metadata on data
1328          * I/O completion. We don't do it for sync(2) writeback because it has a
1329          * separate, external IO completion path and ->sync_fs for guaranteeing
1330          * inode metadata is written back correctly.
1331          */
1332         if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) {
1333                 int err = filemap_fdatawait(mapping);
1334                 if (ret == 0)
1335                         ret = err;
1336         }
1337 
1338         /*
1339          * Some filesystems may redirty the inode during the writeback
1340          * due to delalloc, clear dirty metadata flags right before
1341          * write_inode()
1342          */
1343         spin_lock(&inode->i_lock);
1344 
1345         dirty = inode->i_state & I_DIRTY;
1346         if (inode->i_state & I_DIRTY_TIME) {
1347                 if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
1348                     wbc->sync_mode == WB_SYNC_ALL ||
1349                     unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) ||
1350                     unlikely(time_after(jiffies,
1351                                         (inode->dirtied_time_when +
1352                                          dirtytime_expire_interval * HZ)))) {
1353                         dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED;
1354                         trace_writeback_lazytime(inode);
1355                 }
1356         } else
1357                 inode->i_state &= ~I_DIRTY_TIME_EXPIRED;
1358         inode->i_state &= ~dirty;
1359 
1360         /*
1361          * Paired with smp_mb() in __mark_inode_dirty().  This allows
1362          * __mark_inode_dirty() to test i_state without grabbing i_lock -
1363          * either they see the I_DIRTY bits cleared or we see the dirtied
1364          * inode.
1365          *
1366          * I_DIRTY_PAGES is always cleared together above even if @mapping
1367          * still has dirty pages.  The flag is reinstated after smp_mb() if
1368          * necessary.  This guarantees that either __mark_inode_dirty()
1369          * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
1370          */
1371         smp_mb();
1372 
1373         if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
1374                 inode->i_state |= I_DIRTY_PAGES;
1375 
1376         spin_unlock(&inode->i_lock);
1377 
1378         if (dirty & I_DIRTY_TIME)
1379                 mark_inode_dirty_sync(inode);
1380         /* Don't write the inode if only I_DIRTY_PAGES was set */
1381         if (dirty & ~I_DIRTY_PAGES) {
1382                 int err = write_inode(inode, wbc);
1383                 if (ret == 0)
1384                         ret = err;
1385         }
1386         trace_writeback_single_inode(inode, wbc, nr_to_write);
1387         return ret;
1388 }
1389 
1390 /*
1391  * Write out an inode's dirty pages. Either the caller has an active reference
1392  * on the inode or the inode has I_WILL_FREE set.
1393  *
1394  * This function is designed to be called for writing back one inode which
1395  * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
1396  * and does more profound writeback list handling in writeback_sb_inodes().
1397  */
1398 static int writeback_single_inode(struct inode *inode,
1399                                   struct writeback_control *wbc)
1400 {
1401         struct bdi_writeback *wb;
1402         int ret = 0;
1403 
1404         spin_lock(&inode->i_lock);
1405         if (!atomic_read(&inode->i_count))
1406                 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
1407         else
1408                 WARN_ON(inode->i_state & I_WILL_FREE);
1409 
1410         if (inode->i_state & I_SYNC) {
1411                 if (wbc->sync_mode != WB_SYNC_ALL)
1412                         goto out;
1413                 /*
1414                  * It's a data-integrity sync. We must wait. Since callers hold
1415                  * inode reference or inode has I_WILL_FREE set, it cannot go
1416                  * away under us.
1417                  */
1418                 __inode_wait_for_writeback(inode);
1419         }
1420         WARN_ON(inode->i_state & I_SYNC);
1421         /*
1422          * Skip inode if it is clean and we have no outstanding writeback in
1423          * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
1424          * function since flusher thread may be doing for example sync in
1425          * parallel and if we move the inode, it could get skipped. So here we
1426          * make sure inode is on some writeback list and leave it there unless
1427          * we have completely cleaned the inode.
1428          */
1429         if (!(inode->i_state & I_DIRTY_ALL) &&
1430             (wbc->sync_mode != WB_SYNC_ALL ||
1431              !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK)))
1432                 goto out;
1433         inode->i_state |= I_SYNC;
1434         wbc_attach_and_unlock_inode(wbc, inode);
1435 
1436         ret = __writeback_single_inode(inode, wbc);
1437 
1438         wbc_detach_inode(wbc);
1439 
1440         wb = inode_to_wb_and_lock_list(inode);
1441         spin_lock(&inode->i_lock);
1442         /*
1443          * If inode is clean, remove it from writeback lists. Otherwise don't
1444          * touch it. See comment above for explanation.
1445          */
1446         if (!(inode->i_state & I_DIRTY_ALL))
1447                 inode_io_list_del_locked(inode, wb);
1448         spin_unlock(&wb->list_lock);
1449         inode_sync_complete(inode);
1450 out:
1451         spin_unlock(&inode->i_lock);
1452         return ret;
1453 }
1454 
1455 static long writeback_chunk_size(struct bdi_writeback *wb,
1456                                  struct wb_writeback_work *work)
1457 {
1458         long pages;
1459 
1460         /*
1461          * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
1462          * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
1463          * here avoids calling into writeback_inodes_wb() more than once.
1464          *
1465          * The intended call sequence for WB_SYNC_ALL writeback is:
1466          *
1467          *      wb_writeback()
1468          *          writeback_sb_inodes()       <== called only once
1469          *              write_cache_pages()     <== called once for each inode
1470          *                   (quickly) tag currently dirty pages
1471          *                   (maybe slowly) sync all tagged pages
1472          */
1473         if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
1474                 pages = LONG_MAX;
1475         else {
1476                 pages = min(wb->avg_write_bandwidth / 2,
1477                             global_wb_domain.dirty_limit / DIRTY_SCOPE);
1478                 pages = min(pages, work->nr_pages);
1479                 pages = round_down(pages + MIN_WRITEBACK_PAGES,
1480                                    MIN_WRITEBACK_PAGES);
1481         }
1482 
1483         return pages;
1484 }
1485 
1486 /*
1487  * Write a portion of b_io inodes which belong to @sb.
1488  *
1489  * Return the number of pages and/or inodes written.
1490  *
1491  * NOTE! This is called with wb->list_lock held, and will
1492  * unlock and relock that for each inode it ends up doing
1493  * IO for.
1494  */
1495 static long writeback_sb_inodes(struct super_block *sb,
1496                                 struct bdi_writeback *wb,
1497                                 struct wb_writeback_work *work)
1498 {
1499         struct writeback_control wbc = {
1500                 .sync_mode              = work->sync_mode,
1501                 .tagged_writepages      = work->tagged_writepages,
1502                 .for_kupdate            = work->for_kupdate,
1503                 .for_background         = work->for_background,
1504                 .for_sync               = work->for_sync,
1505                 .range_cyclic           = work->range_cyclic,
1506                 .range_start            = 0,
1507                 .range_end              = LLONG_MAX,
1508         };
1509         unsigned long start_time = jiffies;
1510         long write_chunk;
1511         long wrote = 0;  /* count both pages and inodes */
1512 
1513         while (!list_empty(&wb->b_io)) {
1514                 struct inode *inode = wb_inode(wb->b_io.prev);
1515                 struct bdi_writeback *tmp_wb;
1516 
1517                 if (inode->i_sb != sb) {
1518                         if (work->sb) {
1519                                 /*
1520                                  * We only want to write back data for this
1521                                  * superblock, move all inodes not belonging
1522                                  * to it back onto the dirty list.
1523                                  */
1524                                 redirty_tail(inode, wb);
1525                                 continue;
1526                         }
1527 
1528                         /*
1529                          * The inode belongs to a different superblock.
1530                          * Bounce back to the caller to unpin this and
1531                          * pin the next superblock.
1532                          */
1533                         break;
1534                 }
1535 
1536                 /*
1537                  * Don't bother with new inodes or inodes being freed, first
1538                  * kind does not need periodic writeout yet, and for the latter
1539                  * kind writeout is handled by the freer.
1540                  */
1541                 spin_lock(&inode->i_lock);
1542                 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
1543                         spin_unlock(&inode->i_lock);
1544                         redirty_tail(inode, wb);
1545                         continue;
1546                 }
1547                 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
1548                         /*
1549                          * If this inode is locked for writeback and we are not
1550                          * doing writeback-for-data-integrity, move it to
1551                          * b_more_io so that writeback can proceed with the
1552                          * other inodes on s_io.
1553                          *
1554                          * We'll have another go at writing back this inode
1555                          * when we completed a full scan of b_io.
1556                          */
1557                         spin_unlock(&inode->i_lock);
1558                         requeue_io(inode, wb);
1559                         trace_writeback_sb_inodes_requeue(inode);
1560                         continue;
1561                 }
1562                 spin_unlock(&wb->list_lock);
1563 
1564                 /*
1565                  * We already requeued the inode if it had I_SYNC set and we
1566                  * are doing WB_SYNC_NONE writeback. So this catches only the
1567                  * WB_SYNC_ALL case.
1568                  */
1569                 if (inode->i_state & I_SYNC) {
1570                         /* Wait for I_SYNC. This function drops i_lock... */
1571                         inode_sleep_on_writeback(inode);
1572                         /* Inode may be gone, start again */
1573                         spin_lock(&wb->list_lock);
1574                         continue;
1575                 }
1576                 inode->i_state |= I_SYNC;
1577                 wbc_attach_and_unlock_inode(&wbc, inode);
1578 
1579                 write_chunk = writeback_chunk_size(wb, work);
1580                 wbc.nr_to_write = write_chunk;
1581                 wbc.pages_skipped = 0;
1582 
1583                 /*
1584                  * We use I_SYNC to pin the inode in memory. While it is set
1585                  * evict_inode() will wait so the inode cannot be freed.
1586                  */
1587                 __writeback_single_inode(inode, &wbc);
1588 
1589                 wbc_detach_inode(&wbc);
1590                 work->nr_pages -= write_chunk - wbc.nr_to_write;
1591                 wrote += write_chunk - wbc.nr_to_write;
1592 
1593                 if (need_resched()) {
1594                         /*
1595                          * We're trying to balance between building up a nice
1596                          * long list of IOs to improve our merge rate, and
1597                          * getting those IOs out quickly for anyone throttling
1598                          * in balance_dirty_pages().  cond_resched() doesn't
1599                          * unplug, so get our IOs out the door before we
1600                          * give up the CPU.
1601                          */
1602                         blk_flush_plug(current);
1603                         cond_resched();
1604                 }
1605 
1606                 /*
1607                  * Requeue @inode if still dirty.  Be careful as @inode may
1608                  * have been switched to another wb in the meantime.
1609                  */
1610                 tmp_wb = inode_to_wb_and_lock_list(inode);
1611                 spin_lock(&inode->i_lock);
1612                 if (!(inode->i_state & I_DIRTY_ALL))
1613                         wrote++;
1614                 requeue_inode(inode, tmp_wb, &wbc);
1615                 inode_sync_complete(inode);
1616                 spin_unlock(&inode->i_lock);
1617 
1618                 if (unlikely(tmp_wb != wb)) {
1619                         spin_unlock(&tmp_wb->list_lock);
1620                         spin_lock(&wb->list_lock);
1621                 }
1622 
1623                 /*
1624                  * bail out to wb_writeback() often enough to check
1625                  * background threshold and other termination conditions.
1626                  */
1627                 if (wrote) {
1628                         if (time_is_before_jiffies(start_time + HZ / 10UL))
1629                                 break;
1630                         if (work->nr_pages <= 0)
1631                                 break;
1632                 }
1633         }
1634         return wrote;
1635 }
1636 
1637 static long __writeback_inodes_wb(struct bdi_writeback *wb,
1638                                   struct wb_writeback_work *work)
1639 {
1640         unsigned long start_time = jiffies;
1641         long wrote = 0;
1642 
1643         while (!list_empty(&wb->b_io)) {
1644                 struct inode *inode = wb_inode(wb->b_io.prev);
1645                 struct super_block *sb = inode->i_sb;
1646 
1647                 if (!trylock_super(sb)) {
1648                         /*
1649                          * trylock_super() may fail consistently due to
1650                          * s_umount being grabbed by someone else. Don't use
1651                          * requeue_io() to avoid busy retrying the inode/sb.
1652                          */
1653                         redirty_tail(inode, wb);
1654                         continue;
1655                 }
1656                 wrote += writeback_sb_inodes(sb, wb, work);
1657                 up_read(&sb->s_umount);
1658 
1659                 /* refer to the same tests at the end of writeback_sb_inodes */
1660                 if (wrote) {
1661                         if (time_is_before_jiffies(start_time + HZ / 10UL))
1662                                 break;
1663                         if (work->nr_pages <= 0)
1664                                 break;
1665                 }
1666         }
1667         /* Leave any unwritten inodes on b_io */
1668         return wrote;
1669 }
1670 
1671 static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
1672                                 enum wb_reason reason)
1673 {
1674         struct wb_writeback_work work = {
1675                 .nr_pages       = nr_pages,
1676                 .sync_mode      = WB_SYNC_NONE,
1677                 .range_cyclic   = 1,
1678                 .reason         = reason,
1679         };
1680         struct blk_plug plug;
1681 
1682         blk_start_plug(&plug);
1683         spin_lock(&wb->list_lock);
1684         if (list_empty(&wb->b_io))
1685                 queue_io(wb, &work);
1686         __writeback_inodes_wb(wb, &work);
1687         spin_unlock(&wb->list_lock);
1688         blk_finish_plug(&plug);
1689 
1690         return nr_pages - work.nr_pages;
1691 }
1692 
1693 /*
1694  * Explicit flushing or periodic writeback of "old" data.
1695  *
1696  * Define "old": the first time one of an inode's pages is dirtied, we mark the
1697  * dirtying-time in the inode's address_space.  So this periodic writeback code
1698  * just walks the superblock inode list, writing back any inodes which are
1699  * older than a specific point in time.
1700  *
1701  * Try to run once per dirty_writeback_interval.  But if a writeback event
1702  * takes longer than a dirty_writeback_interval interval, then leave a
1703  * one-second gap.
1704  *
1705  * older_than_this takes precedence over nr_to_write.  So we'll only write back
1706  * all dirty pages if they are all attached to "old" mappings.
1707  */
1708 static long wb_writeback(struct bdi_writeback *wb,
1709                          struct wb_writeback_work *work)
1710 {
1711         unsigned long wb_start = jiffies;
1712         long nr_pages = work->nr_pages;
1713         unsigned long oldest_jif;
1714         struct inode *inode;
1715         long progress;
1716         struct blk_plug plug;
1717 
1718         oldest_jif = jiffies;
1719         work->older_than_this = &oldest_jif;
1720 
1721         blk_start_plug(&plug);
1722         spin_lock(&wb->list_lock);
1723         for (;;) {
1724                 /*
1725                  * Stop writeback when nr_pages has been consumed
1726                  */
1727                 if (work->nr_pages <= 0)
1728                         break;
1729 
1730                 /*
1731                  * Background writeout and kupdate-style writeback may
1732                  * run forever. Stop them if there is other work to do
1733                  * so that e.g. sync can proceed. They'll be restarted
1734                  * after the other works are all done.
1735                  */
1736                 if ((work->for_background || work->for_kupdate) &&
1737                     !list_empty(&wb->work_list))
1738                         break;
1739 
1740                 /*
1741                  * For background writeout, stop when we are below the
1742                  * background dirty threshold
1743                  */
1744                 if (work->for_background && !wb_over_bg_thresh(wb))
1745                         break;
1746 
1747                 /*
1748                  * Kupdate and background works are special and we want to
1749                  * include all inodes that need writing. Livelock avoidance is
1750                  * handled by these works yielding to any other work so we are
1751                  * safe.
1752                  */
1753                 if (work->for_kupdate) {
1754                         oldest_jif = jiffies -
1755                                 msecs_to_jiffies(dirty_expire_interval * 10);
1756                 } else if (work->for_background)
1757                         oldest_jif = jiffies;
1758 
1759                 trace_writeback_start(wb, work);
1760                 if (list_empty(&wb->b_io))
1761                         queue_io(wb, work);
1762                 if (work->sb)
1763                         progress = writeback_sb_inodes(work->sb, wb, work);
1764                 else
1765                         progress = __writeback_inodes_wb(wb, work);
1766                 trace_writeback_written(wb, work);
1767 
1768                 wb_update_bandwidth(wb, wb_start);
1769 
1770                 /*
1771                  * Did we write something? Try for more
1772                  *
1773                  * Dirty inodes are moved to b_io for writeback in batches.
1774                  * The completion of the current batch does not necessarily
1775                  * mean the overall work is done. So we keep looping as long
1776                  * as made some progress on cleaning pages or inodes.
1777                  */
1778                 if (progress)
1779                         continue;
1780                 /*
1781                  * No more inodes for IO, bail
1782                  */
1783                 if (list_empty(&wb->b_more_io))
1784                         break;
1785                 /*
1786                  * Nothing written. Wait for some inode to
1787                  * become available for writeback. Otherwise
1788                  * we'll just busyloop.
1789                  */
1790                 trace_writeback_wait(wb, work);
1791                 inode = wb_inode(wb->b_more_io.prev);
1792                 spin_lock(&inode->i_lock);
1793                 spin_unlock(&wb->list_lock);
1794                 /* This function drops i_lock... */
1795                 inode_sleep_on_writeback(inode);
1796                 spin_lock(&wb->list_lock);
1797         }
1798         spin_unlock(&wb->list_lock);
1799         blk_finish_plug(&plug);
1800 
1801         return nr_pages - work->nr_pages;
1802 }
1803 
1804 /*
1805  * Return the next wb_writeback_work struct that hasn't been processed yet.
1806  */
1807 static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb)
1808 {
1809         struct wb_writeback_work *work = NULL;
1810 
1811         spin_lock_bh(&wb->work_lock);
1812         if (!list_empty(&wb->work_list)) {
1813                 work = list_entry(wb->work_list.next,
1814                                   struct wb_writeback_work, list);
1815                 list_del_init(&work->list);
1816         }
1817         spin_unlock_bh(&wb->work_lock);
1818         return work;
1819 }
1820 
1821 static long wb_check_background_flush(struct bdi_writeback *wb)
1822 {
1823         if (wb_over_bg_thresh(wb)) {
1824 
1825                 struct wb_writeback_work work = {
1826                         .nr_pages       = LONG_MAX,
1827                         .sync_mode      = WB_SYNC_NONE,
1828                         .for_background = 1,
1829                         .range_cyclic   = 1,
1830                         .reason         = WB_REASON_BACKGROUND,
1831                 };
1832 
1833                 return wb_writeback(wb, &work);
1834         }
1835 
1836         return 0;
1837 }
1838 
1839 static long wb_check_old_data_flush(struct bdi_writeback *wb)
1840 {
1841         unsigned long expired;
1842         long nr_pages;
1843 
1844         /*
1845          * When set to zero, disable periodic writeback
1846          */
1847         if (!dirty_writeback_interval)
1848                 return 0;
1849 
1850         expired = wb->last_old_flush +
1851                         msecs_to_jiffies(dirty_writeback_interval * 10);
1852         if (time_before(jiffies, expired))
1853                 return 0;
1854 
1855         wb->last_old_flush = jiffies;
1856         nr_pages = get_nr_dirty_pages();
1857 
1858         if (nr_pages) {
1859                 struct wb_writeback_work work = {
1860                         .nr_pages       = nr_pages,
1861                         .sync_mode      = WB_SYNC_NONE,
1862                         .for_kupdate    = 1,
1863                         .range_cyclic   = 1,
1864                         .reason         = WB_REASON_PERIODIC,
1865                 };
1866 
1867                 return wb_writeback(wb, &work);
1868         }
1869 
1870         return 0;
1871 }
1872 
1873 static long wb_check_start_all(struct bdi_writeback *wb)
1874 {
1875         long nr_pages;
1876 
1877         if (!test_bit(WB_start_all, &wb->state))
1878                 return 0;
1879 
1880         nr_pages = get_nr_dirty_pages();
1881         if (nr_pages) {
1882                 struct wb_writeback_work work = {
1883                         .nr_pages       = wb_split_bdi_pages(wb, nr_pages),
1884                         .sync_mode      = WB_SYNC_NONE,
1885                         .range_cyclic   = 1,
1886                         .reason         = wb->start_all_reason,
1887                 };
1888 
1889                 nr_pages = wb_writeback(wb, &work);
1890         }
1891 
1892         clear_bit(WB_start_all, &wb->state);
1893         return nr_pages;
1894 }
1895 
1896 
1897 /*
1898  * Retrieve work items and do the writeback they describe
1899  */
1900 static long wb_do_writeback(struct bdi_writeback *wb)
1901 {
1902         struct wb_writeback_work *work;
1903         long wrote = 0;
1904 
1905         set_bit(WB_writeback_running, &wb->state);
1906         while ((work = get_next_work_item(wb)) != NULL) {
1907                 trace_writeback_exec(wb, work);
1908                 wrote += wb_writeback(wb, work);
1909                 finish_writeback_work(wb, work);
1910         }
1911 
1912         /*
1913          * Check for a flush-everything request
1914          */
1915         wrote += wb_check_start_all(wb);
1916 
1917         /*
1918          * Check for periodic writeback, kupdated() style
1919          */
1920         wrote += wb_check_old_data_flush(wb);
1921         wrote += wb_check_background_flush(wb);
1922         clear_bit(WB_writeback_running, &wb->state);
1923 
1924         return wrote;
1925 }
1926 
1927 /*
1928  * Handle writeback of dirty data for the device backed by this bdi. Also
1929  * reschedules periodically and does kupdated style flushing.
1930  */
1931 void wb_workfn(struct work_struct *work)
1932 {
1933         struct bdi_writeback *wb = container_of(to_delayed_work(work),
1934                                                 struct bdi_writeback, dwork);
1935         long pages_written;
1936 
1937         set_worker_desc("flush-%s", dev_name(wb->bdi->dev));
1938         current->flags |= PF_SWAPWRITE;
1939 
1940         if (likely(!current_is_workqueue_rescuer() ||
1941                    !test_bit(WB_registered, &wb->state))) {
1942                 /*
1943                  * The normal path.  Keep writing back @wb until its
1944                  * work_list is empty.  Note that this path is also taken
1945                  * if @wb is shutting down even when we're running off the
1946                  * rescuer as work_list needs to be drained.
1947                  */
1948                 do {
1949                         pages_written = wb_do_writeback(wb);
1950                         trace_writeback_pages_written(pages_written);
1951                 } while (!list_empty(&wb->work_list));
1952         } else {
1953                 /*
1954                  * bdi_wq can't get enough workers and we're running off
1955                  * the emergency worker.  Don't hog it.  Hopefully, 1024 is
1956                  * enough for efficient IO.
1957                  */
1958                 pages_written = writeback_inodes_wb(wb, 1024,
1959                                                     WB_REASON_FORKER_THREAD);
1960                 trace_writeback_pages_written(pages_written);
1961         }
1962 
1963         if (!list_empty(&wb->work_list))
1964                 wb_wakeup(wb);
1965         else if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1966                 wb_wakeup_delayed(wb);
1967 
1968         current->flags &= ~PF_SWAPWRITE;
1969 }
1970 
1971 /*
1972  * Start writeback of `nr_pages' pages on this bdi. If `nr_pages' is zero,
1973  * write back the whole world.
1974  */
1975 static void __wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
1976                                          enum wb_reason reason)
1977 {
1978         struct bdi_writeback *wb;
1979 
1980         if (!bdi_has_dirty_io(bdi))
1981                 return;
1982 
1983         list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node)
1984                 wb_start_writeback(wb, reason);
1985 }
1986 
1987 void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi,
1988                                 enum wb_reason reason)
1989 {
1990         rcu_read_lock();
1991         __wakeup_flusher_threads_bdi(bdi, reason);
1992         rcu_read_unlock();
1993 }
1994 
1995 /*
1996  * Wakeup the flusher threads to start writeback of all currently dirty pages
1997  */
1998 void wakeup_flusher_threads(enum wb_reason reason)
1999 {
2000         struct backing_dev_info *bdi;
2001 
2002         /*
2003          * If we are expecting writeback progress we must submit plugged IO.
2004          */
2005         if (blk_needs_flush_plug(current))
2006                 blk_schedule_flush_plug(current);
2007 
2008         rcu_read_lock();
2009         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list)
2010                 __wakeup_flusher_threads_bdi(bdi, reason);
2011         rcu_read_unlock();
2012 }
2013 
2014 /*
2015  * Wake up bdi's periodically to make sure dirtytime inodes gets
2016  * written back periodically.  We deliberately do *not* check the
2017  * b_dirtytime list in wb_has_dirty_io(), since this would cause the
2018  * kernel to be constantly waking up once there are any dirtytime
2019  * inodes on the system.  So instead we define a separate delayed work
2020  * function which gets called much more rarely.  (By default, only
2021  * once every 12 hours.)
2022  *
2023  * If there is any other write activity going on in the file system,
2024  * this function won't be necessary.  But if the only thing that has
2025  * happened on the file system is a dirtytime inode caused by an atime
2026  * update, we need this infrastructure below to make sure that inode
2027  * eventually gets pushed out to disk.
2028  */
2029 static void wakeup_dirtytime_writeback(struct work_struct *w);
2030 static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback);
2031 
2032 static void wakeup_dirtytime_writeback(struct work_struct *w)
2033 {
2034         struct backing_dev_info *bdi;
2035 
2036         rcu_read_lock();
2037         list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
2038                 struct bdi_writeback *wb;
2039 
2040                 list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node)
2041                         if (!list_empty(&wb->b_dirty_time))
2042                                 wb_wakeup(wb);
2043         }
2044         rcu_read_unlock();
2045         schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
2046 }
2047 
2048 static int __init start_dirtytime_writeback(void)
2049 {
2050         schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ);
2051         return 0;
2052 }
2053 __initcall(start_dirtytime_writeback);
2054 
2055 int dirtytime_interval_handler(struct ctl_table *table, int write,
2056                                void __user *buffer, size_t *lenp, loff_t *ppos)
2057 {
2058         int ret;
2059 
2060         ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2061         if (ret == 0 && write)
2062                 mod_delayed_work(system_wq, &dirtytime_work, 0);
2063         return ret;
2064 }
2065 
2066 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
2067 {
2068         if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
2069                 struct dentry *dentry;
2070                 const char *name = "?";
2071 
2072                 dentry = d_find_alias(inode);
2073                 if (dentry) {
2074                         spin_lock(&dentry->d_lock);
2075                         name = (const char *) dentry->d_name.name;
2076                 }
2077                 printk(KERN_DEBUG
2078                        "%s(%d): dirtied inode %lu (%s) on %s\n",
2079                        current->comm, task_pid_nr(current), inode->i_ino,
2080                        name, inode->i_sb->s_id);
2081                 if (dentry) {
2082                         spin_unlock(&dentry->d_lock);
2083                         dput(dentry);
2084                 }
2085         }
2086 }
2087 
2088 /**
2089  * __mark_inode_dirty - internal function
2090  *
2091  * @inode: inode to mark
2092  * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
2093  *
2094  * Mark an inode as dirty. Callers should use mark_inode_dirty or
2095  * mark_inode_dirty_sync.
2096  *
2097  * Put the inode on the super block's dirty list.
2098  *
2099  * CAREFUL! We mark it dirty unconditionally, but move it onto the
2100  * dirty list only if it is hashed or if it refers to a blockdev.
2101  * If it was not hashed, it will never be added to the dirty list
2102  * even if it is later hashed, as it will have been marked dirty already.
2103  *
2104  * In short, make sure you hash any inodes _before_ you start marking
2105  * them dirty.
2106  *
2107  * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
2108  * the block-special inode (/dev/hda1) itself.  And the ->dirtied_when field of
2109  * the kernel-internal blockdev inode represents the dirtying time of the
2110  * blockdev's pages.  This is why for I_DIRTY_PAGES we always use
2111  * page->mapping->host, so the page-dirtying time is recorded in the internal
2112  * blockdev inode.
2113  */
2114 void __mark_inode_dirty(struct inode *inode, int flags)
2115 {
2116 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
2117         struct super_block *sb = inode->i_sb;
2118         int dirtytime;
2119 
2120         trace_writeback_mark_inode_dirty(inode, flags);
2121 
2122         /*
2123          * Don't do this for I_DIRTY_PAGES - that doesn't actually
2124          * dirty the inode itself
2125          */
2126         if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) {
2127                 trace_writeback_dirty_inode_start(inode, flags);
2128 
2129                 if (sb->s_op->dirty_inode)
2130                         sb->s_op->dirty_inode(inode, flags);
2131 
2132                 trace_writeback_dirty_inode(inode, flags);
2133         }
2134         if (flags & I_DIRTY_INODE)
2135                 flags &= ~I_DIRTY_TIME;
2136         dirtytime = flags & I_DIRTY_TIME;
2137 
2138         /*
2139          * Paired with smp_mb() in __writeback_single_inode() for the
2140          * following lockless i_state test.  See there for details.
2141          */
2142         smp_mb();
2143 
2144         if (((inode->i_state & flags) == flags) ||
2145             (dirtytime && (inode->i_state & I_DIRTY_INODE)))
2146                 return;
2147 
2148         if (unlikely(block_dump))
2149                 block_dump___mark_inode_dirty(inode);
2150 
2151         spin_lock(&inode->i_lock);
2152         if (dirtytime && (inode->i_state & I_DIRTY_INODE))
2153                 goto out_unlock_inode;
2154         if ((inode->i_state & flags) != flags) {
2155                 const int was_dirty = inode->i_state & I_DIRTY;
2156 
2157                 inode_attach_wb(inode, NULL);
2158 
2159                 if (flags & I_DIRTY_INODE)
2160                         inode->i_state &= ~I_DIRTY_TIME;
2161                 inode->i_state |= flags;
2162 
2163                 /*
2164                  * If the inode is being synced, just update its dirty state.
2165                  * The unlocker will place the inode on the appropriate
2166                  * superblock list, based upon its state.
2167                  */
2168                 if (inode->i_state & I_SYNC)
2169                         goto out_unlock_inode;
2170 
2171                 /*
2172                  * Only add valid (hashed) inodes to the superblock's
2173                  * dirty list.  Add blockdev inodes as well.
2174                  */
2175                 if (!S_ISBLK(inode->i_mode)) {
2176                         if (inode_unhashed(inode))
2177                                 goto out_unlock_inode;
2178                 }
2179                 if (inode->i_state & I_FREEING)
2180                         goto out_unlock_inode;
2181 
2182                 /*
2183                  * If the inode was already on b_dirty/b_io/b_more_io, don't
2184                  * reposition it (that would break b_dirty time-ordering).
2185                  */
2186                 if (!was_dirty) {
2187                         struct bdi_writeback *wb;
2188                         struct list_head *dirty_list;
2189                         bool wakeup_bdi = false;
2190 
2191                         wb = locked_inode_to_wb_and_lock_list(inode);
2192 
2193                         WARN(bdi_cap_writeback_dirty(wb->bdi) &&
2194                              !test_bit(WB_registered, &wb->state),
2195                              "bdi-%s not registered\n", wb->bdi->name);
2196 
2197                         inode->dirtied_when = jiffies;
2198                         if (dirtytime)
2199                                 inode->dirtied_time_when = jiffies;
2200 
2201                         if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES))
2202                                 dirty_list = &wb->b_dirty;
2203                         else
2204                                 dirty_list = &wb->b_dirty_time;
2205 
2206                         wakeup_bdi = inode_io_list_move_locked(inode, wb,
2207                                                                dirty_list);
2208 
2209                         spin_unlock(&wb->list_lock);
2210                         trace_writeback_dirty_inode_enqueue(inode);
2211 
2212                         /*
2213                          * If this is the first dirty inode for this bdi,
2214                          * we have to wake-up the corresponding bdi thread
2215                          * to make sure background write-back happens
2216                          * later.
2217                          */
2218                         if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi)
2219                                 wb_wakeup_delayed(wb);
2220                         return;
2221                 }
2222         }
2223 out_unlock_inode:
2224         spin_unlock(&inode->i_lock);
2225 
2226 #undef I_DIRTY_INODE
2227 }
2228 EXPORT_SYMBOL(__mark_inode_dirty);
2229 
2230 /*
2231  * The @s_sync_lock is used to serialise concurrent sync operations
2232  * to avoid lock contention problems with concurrent wait_sb_inodes() calls.
2233  * Concurrent callers will block on the s_sync_lock rather than doing contending
2234  * walks. The queueing maintains sync(2) required behaviour as all the IO that
2235  * has been issued up to the time this function is enter is guaranteed to be
2236  * completed by the time we have gained the lock and waited for all IO that is
2237  * in progress regardless of the order callers are granted the lock.
2238  */
2239 static void wait_sb_inodes(struct super_block *sb)
2240 {
2241         LIST_HEAD(sync_list);
2242 
2243         /*
2244          * We need to be protected against the filesystem going from
2245          * r/o to r/w or vice versa.
2246          */
2247         WARN_ON(!rwsem_is_locked(&sb->s_umount));
2248 
2249         mutex_lock(&sb->s_sync_lock);
2250 
2251         /*
2252          * Splice the writeback list onto a temporary list to avoid waiting on
2253          * inodes that have started writeback after this point.
2254          *
2255          * Use rcu_read_lock() to keep the inodes around until we have a
2256          * reference. s_inode_wblist_lock protects sb->s_inodes_wb as well as
2257          * the local list because inodes can be dropped from either by writeback
2258          * completion.
2259          */
2260         rcu_read_lock();
2261         spin_lock_irq(&sb->s_inode_wblist_lock);
2262         list_splice_init(&sb->s_inodes_wb, &sync_list);
2263 
2264         /*
2265          * Data integrity sync. Must wait for all pages under writeback, because
2266          * there may have been pages dirtied before our sync call, but which had
2267          * writeout started before we write it out.  In which case, the inode
2268          * may not be on the dirty list, but we still have to wait for that
2269          * writeout.
2270          */
2271         while (!list_empty(&sync_list)) {
2272                 struct inode *inode = list_first_entry(&sync_list, struct inode,
2273                                                        i_wb_list);
2274                 struct address_space *mapping = inode->i_mapping;
2275 
2276                 /*
2277                  * Move each inode back to the wb list before we drop the lock
2278                  * to preserve consistency between i_wb_list and the mapping
2279                  * writeback tag. Writeback completion is responsible to remove
2280                  * the inode from either list once the writeback tag is cleared.
2281                  */
2282                 list_move_tail(&inode->i_wb_list, &sb->s_inodes_wb);
2283 
2284                 /*
2285                  * The mapping can appear untagged while still on-list since we
2286                  * do not have the mapping lock. Skip it here, wb completion
2287                  * will remove it.
2288                  */
2289                 if (!mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
2290                         continue;
2291 
2292                 spin_unlock_irq(&sb->s_inode_wblist_lock);
2293 
2294                 spin_lock(&inode->i_lock);
2295                 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) {
2296                         spin_unlock(&inode->i_lock);
2297 
2298                         spin_lock_irq(&sb->s_inode_wblist_lock);
2299                         continue;
2300                 }
2301                 __iget(inode);
2302                 spin_unlock(&inode->i_lock);
2303                 rcu_read_unlock();
2304 
2305                 /*
2306                  * We keep the error status of individual mapping so that
2307                  * applications can catch the writeback error using fsync(2).
2308                  * See filemap_fdatawait_keep_errors() for details.
2309                  */
2310                 filemap_fdatawait_keep_errors(mapping);
2311 
2312                 cond_resched();
2313 
2314                 iput(inode);
2315 
2316                 rcu_read_lock();
2317                 spin_lock_irq(&sb->s_inode_wblist_lock);
2318         }
2319         spin_unlock_irq(&sb->s_inode_wblist_lock);
2320         rcu_read_unlock();
2321         mutex_unlock(&sb->s_sync_lock);
2322 }
2323 
2324 static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr,
2325                                      enum wb_reason reason, bool skip_if_busy)
2326 {
2327         DEFINE_WB_COMPLETION_ONSTACK(done);
2328         struct wb_writeback_work work = {
2329                 .sb                     = sb,
2330                 .sync_mode              = WB_SYNC_NONE,
2331                 .tagged_writepages      = 1,
2332                 .done                   = &done,
2333                 .nr_pages               = nr,
2334                 .reason                 = reason,
2335         };
2336         struct backing_dev_info *bdi = sb->s_bdi;
2337 
2338         if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info)
2339                 return;
2340         WARN_ON(!rwsem_is_locked(&sb->s_umount));
2341 
2342         bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy);
2343         wb_wait_for_completion(bdi, &done);
2344 }
2345 
2346 /**
2347  * writeback_inodes_sb_nr -     writeback dirty inodes from given super_block
2348  * @sb: the superblock
2349  * @nr: the number of pages to write
2350  * @reason: reason why some writeback work initiated
2351  *
2352  * Start writeback on some inodes on this super_block. No guarantees are made
2353  * on how many (if any) will be written, and this function does not wait
2354  * for IO completion of submitted IO.
2355  */
2356 void writeback_inodes_sb_nr(struct super_block *sb,
2357                             unsigned long nr,
2358                             enum wb_reason reason)
2359 {
2360         __writeback_inodes_sb_nr(sb, nr, reason, false);
2361 }
2362 EXPORT_SYMBOL(writeback_inodes_sb_nr);
2363 
2364 /**
2365  * writeback_inodes_sb  -       writeback dirty inodes from given super_block
2366  * @sb: the superblock
2367  * @reason: reason why some writeback work was initiated
2368  *
2369  * Start writeback on some inodes on this super_block. No guarantees are made
2370  * on how many (if any) will be written, and this function does not wait
2371  * for IO completion of submitted IO.
2372  */
2373 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
2374 {
2375         return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
2376 }
2377 EXPORT_SYMBOL(writeback_inodes_sb);
2378 
2379 /**
2380  * try_to_writeback_inodes_sb - try to start writeback if none underway
2381  * @sb: the superblock
2382  * @reason: reason why some writeback work was initiated
2383  *
2384  * Invoke __writeback_inodes_sb_nr if no writeback is currently underway.
2385  */
2386 void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
2387 {
2388         if (!down_read_trylock(&sb->s_umount))
2389                 return;
2390 
2391         __writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason, true);
2392         up_read(&sb->s_umount);
2393 }
2394 EXPORT_SYMBOL(try_to_writeback_inodes_sb);
2395 
2396 /**
2397  * sync_inodes_sb       -       sync sb inode pages
2398  * @sb: the superblock
2399  *
2400  * This function writes and waits on any dirty inode belonging to this
2401  * super_block.
2402  */
2403 void sync_inodes_sb(struct super_block *sb)
2404 {
2405         DEFINE_WB_COMPLETION_ONSTACK(done);
2406         struct wb_writeback_work work = {
2407                 .sb             = sb,
2408                 .sync_mode      = WB_SYNC_ALL,
2409                 .nr_pages       = LONG_MAX,
2410                 .range_cyclic   = 0,
2411                 .done           = &done,
2412                 .reason         = WB_REASON_SYNC,
2413                 .for_sync       = 1,
2414         };
2415         struct backing_dev_info *bdi = sb->s_bdi;
2416 
2417         /*
2418          * Can't skip on !bdi_has_dirty() because we should wait for !dirty
2419          * inodes under writeback and I_DIRTY_TIME inodes ignored by
2420          * bdi_has_dirty() need to be written out too.
2421          */
2422         if (bdi == &noop_backing_dev_info)
2423                 return;
2424         WARN_ON(!rwsem_is_locked(&sb->s_umount));
2425 
2426         bdi_split_work_to_wbs(bdi, &work, false);
2427         wb_wait_for_completion(bdi, &done);
2428 
2429         wait_sb_inodes(sb);
2430 }
2431 EXPORT_SYMBOL(sync_inodes_sb);
2432 
2433 /**
2434  * write_inode_now      -       write an inode to disk
2435  * @inode: inode to write to disk
2436  * @sync: whether the write should be synchronous or not
2437  *
2438  * This function commits an inode to disk immediately if it is dirty. This is
2439  * primarily needed by knfsd.
2440  *
2441  * The caller must either have a ref on the inode or must have set I_WILL_FREE.
2442  */
2443 int write_inode_now(struct inode *inode, int sync)
2444 {
2445         struct writeback_control wbc = {
2446                 .nr_to_write = LONG_MAX,
2447                 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
2448                 .range_start = 0,
2449                 .range_end = LLONG_MAX,
2450         };
2451 
2452         if (!mapping_cap_writeback_dirty(inode->i_mapping))
2453                 wbc.nr_to_write = 0;
2454 
2455         might_sleep();
2456         return writeback_single_inode(inode, &wbc);
2457 }
2458 EXPORT_SYMBOL(write_inode_now);
2459 
2460 /**
2461  * sync_inode - write an inode and its pages to disk.
2462  * @inode: the inode to sync
2463  * @wbc: controls the writeback mode
2464  *
2465  * sync_inode() will write an inode and its pages to disk.  It will also
2466  * correctly update the inode on its superblock's dirty inode lists and will
2467  * update inode->i_state.
2468  *
2469  * The caller must have a ref on the inode.
2470  */
2471 int sync_inode(struct inode *inode, struct writeback_control *wbc)
2472 {
2473         return writeback_single_inode(inode, wbc);
2474 }
2475 EXPORT_SYMBOL(sync_inode);
2476 
2477 /**
2478  * sync_inode_metadata - write an inode to disk
2479  * @inode: the inode to sync
2480  * @wait: wait for I/O to complete.
2481  *
2482  * Write an inode to disk and adjust its dirty state after completion.
2483  *
2484  * Note: only writes the actual inode, no associated data or other metadata.
2485  */
2486 int sync_inode_metadata(struct inode *inode, int wait)
2487 {
2488         struct writeback_control wbc = {
2489                 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
2490                 .nr_to_write = 0, /* metadata-only */
2491         };
2492 
2493         return sync_inode(inode, &wbc);
2494 }
2495 EXPORT_SYMBOL(sync_inode_metadata);
2496 

~ [ 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