Root/
1 | /* |
2 | * fs/fs-writeback.c |
3 | * |
4 | * Copyright (C) 2002, Linus Torvalds. |
5 | * |
6 | * Contains all the functions related to writing back and waiting |
7 | * upon dirty inodes against superblocks, and writing back dirty |
8 | * pages against inodes. ie: data writeback. Writeout of the |
9 | * inode itself is not handled here. |
10 | * |
11 | * 10Apr2002 Andrew Morton |
12 | * Split out of fs/inode.c |
13 | * Additions for address_space-based writeback |
14 | */ |
15 | |
16 | #include <linux/kernel.h> |
17 | #include <linux/export.h> |
18 | #include <linux/spinlock.h> |
19 | #include <linux/slab.h> |
20 | #include <linux/sched.h> |
21 | #include <linux/fs.h> |
22 | #include <linux/mm.h> |
23 | #include <linux/pagemap.h> |
24 | #include <linux/kthread.h> |
25 | #include <linux/freezer.h> |
26 | #include <linux/writeback.h> |
27 | #include <linux/blkdev.h> |
28 | #include <linux/backing-dev.h> |
29 | #include <linux/tracepoint.h> |
30 | #include "internal.h" |
31 | |
32 | /* |
33 | * 4MB minimal write chunk size |
34 | */ |
35 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) |
36 | |
37 | /* |
38 | * Passed into wb_writeback(), essentially a subset of writeback_control |
39 | */ |
40 | struct wb_writeback_work { |
41 | long nr_pages; |
42 | struct super_block *sb; |
43 | unsigned long *older_than_this; |
44 | enum writeback_sync_modes sync_mode; |
45 | unsigned int tagged_writepages:1; |
46 | unsigned int for_kupdate:1; |
47 | unsigned int range_cyclic:1; |
48 | unsigned int for_background:1; |
49 | enum wb_reason reason; /* why was writeback initiated? */ |
50 | |
51 | struct list_head list; /* pending work list */ |
52 | struct completion *done; /* set if the caller waits */ |
53 | }; |
54 | |
55 | /** |
56 | * writeback_in_progress - determine whether there is writeback in progress |
57 | * @bdi: the device's backing_dev_info structure. |
58 | * |
59 | * Determine whether there is writeback waiting to be handled against a |
60 | * backing device. |
61 | */ |
62 | int writeback_in_progress(struct backing_dev_info *bdi) |
63 | { |
64 | return test_bit(BDI_writeback_running, &bdi->state); |
65 | } |
66 | |
67 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
68 | { |
69 | struct super_block *sb = inode->i_sb; |
70 | |
71 | if (strcmp(sb->s_type->name, "bdev") == 0) |
72 | return inode->i_mapping->backing_dev_info; |
73 | |
74 | return sb->s_bdi; |
75 | } |
76 | |
77 | static inline struct inode *wb_inode(struct list_head *head) |
78 | { |
79 | return list_entry(head, struct inode, i_wb_list); |
80 | } |
81 | |
82 | /* |
83 | * Include the creation of the trace points after defining the |
84 | * wb_writeback_work structure and inline functions so that the definition |
85 | * remains local to this file. |
86 | */ |
87 | #define CREATE_TRACE_POINTS |
88 | #include <trace/events/writeback.h> |
89 | |
90 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
91 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) |
92 | { |
93 | if (bdi->wb.task) { |
94 | wake_up_process(bdi->wb.task); |
95 | } else { |
96 | /* |
97 | * The bdi thread isn't there, wake up the forker thread which |
98 | * will create and run it. |
99 | */ |
100 | wake_up_process(default_backing_dev_info.wb.task); |
101 | } |
102 | } |
103 | |
104 | static void bdi_queue_work(struct backing_dev_info *bdi, |
105 | struct wb_writeback_work *work) |
106 | { |
107 | trace_writeback_queue(bdi, work); |
108 | |
109 | spin_lock_bh(&bdi->wb_lock); |
110 | list_add_tail(&work->list, &bdi->work_list); |
111 | if (!bdi->wb.task) |
112 | trace_writeback_nothread(bdi, work); |
113 | bdi_wakeup_flusher(bdi); |
114 | spin_unlock_bh(&bdi->wb_lock); |
115 | } |
116 | |
117 | static void |
118 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
119 | bool range_cyclic, enum wb_reason reason) |
120 | { |
121 | struct wb_writeback_work *work; |
122 | |
123 | /* |
124 | * This is WB_SYNC_NONE writeback, so if allocation fails just |
125 | * wakeup the thread for old dirty data writeback |
126 | */ |
127 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
128 | if (!work) { |
129 | if (bdi->wb.task) { |
130 | trace_writeback_nowork(bdi); |
131 | wake_up_process(bdi->wb.task); |
132 | } |
133 | return; |
134 | } |
135 | |
136 | work->sync_mode = WB_SYNC_NONE; |
137 | work->nr_pages = nr_pages; |
138 | work->range_cyclic = range_cyclic; |
139 | work->reason = reason; |
140 | |
141 | bdi_queue_work(bdi, work); |
142 | } |
143 | |
144 | /** |
145 | * bdi_start_writeback - start writeback |
146 | * @bdi: the backing device to write from |
147 | * @nr_pages: the number of pages to write |
148 | * @reason: reason why some writeback work was initiated |
149 | * |
150 | * Description: |
151 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only |
152 | * started when this function returns, we make no guarantees on |
153 | * completion. Caller need not hold sb s_umount semaphore. |
154 | * |
155 | */ |
156 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
157 | enum wb_reason reason) |
158 | { |
159 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
160 | } |
161 | |
162 | /** |
163 | * bdi_start_background_writeback - start background writeback |
164 | * @bdi: the backing device to write from |
165 | * |
166 | * Description: |
167 | * This makes sure WB_SYNC_NONE background writeback happens. When |
168 | * this function returns, it is only guaranteed that for given BDI |
169 | * some IO is happening if we are over background dirty threshold. |
170 | * Caller need not hold sb s_umount semaphore. |
171 | */ |
172 | void bdi_start_background_writeback(struct backing_dev_info *bdi) |
173 | { |
174 | /* |
175 | * We just wake up the flusher thread. It will perform background |
176 | * writeback as soon as there is no other work to do. |
177 | */ |
178 | trace_writeback_wake_background(bdi); |
179 | spin_lock_bh(&bdi->wb_lock); |
180 | bdi_wakeup_flusher(bdi); |
181 | spin_unlock_bh(&bdi->wb_lock); |
182 | } |
183 | |
184 | /* |
185 | * Remove the inode from the writeback list it is on. |
186 | */ |
187 | void inode_wb_list_del(struct inode *inode) |
188 | { |
189 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
190 | |
191 | spin_lock(&bdi->wb.list_lock); |
192 | list_del_init(&inode->i_wb_list); |
193 | spin_unlock(&bdi->wb.list_lock); |
194 | } |
195 | |
196 | /* |
197 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the |
198 | * furthest end of its superblock's dirty-inode list. |
199 | * |
200 | * Before stamping the inode's ->dirtied_when, we check to see whether it is |
201 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
202 | * the case then the inode must have been redirtied while it was being written |
203 | * out and we don't reset its dirtied_when. |
204 | */ |
205 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
206 | { |
207 | assert_spin_locked(&wb->list_lock); |
208 | if (!list_empty(&wb->b_dirty)) { |
209 | struct inode *tail; |
210 | |
211 | tail = wb_inode(wb->b_dirty.next); |
212 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
213 | inode->dirtied_when = jiffies; |
214 | } |
215 | list_move(&inode->i_wb_list, &wb->b_dirty); |
216 | } |
217 | |
218 | /* |
219 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
220 | */ |
221 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
222 | { |
223 | assert_spin_locked(&wb->list_lock); |
224 | list_move(&inode->i_wb_list, &wb->b_more_io); |
225 | } |
226 | |
227 | static void inode_sync_complete(struct inode *inode) |
228 | { |
229 | inode->i_state &= ~I_SYNC; |
230 | /* Waiters must see I_SYNC cleared before being woken up */ |
231 | smp_mb(); |
232 | wake_up_bit(&inode->i_state, __I_SYNC); |
233 | } |
234 | |
235 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
236 | { |
237 | bool ret = time_after(inode->dirtied_when, t); |
238 | #ifndef CONFIG_64BIT |
239 | /* |
240 | * For inodes being constantly redirtied, dirtied_when can get stuck. |
241 | * It _appears_ to be in the future, but is actually in distant past. |
242 | * This test is necessary to prevent such wrapped-around relative times |
243 | * from permanently stopping the whole bdi writeback. |
244 | */ |
245 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); |
246 | #endif |
247 | return ret; |
248 | } |
249 | |
250 | /* |
251 | * Move expired (dirtied after work->older_than_this) dirty inodes from |
252 | * @delaying_queue to @dispatch_queue. |
253 | */ |
254 | static int move_expired_inodes(struct list_head *delaying_queue, |
255 | struct list_head *dispatch_queue, |
256 | struct wb_writeback_work *work) |
257 | { |
258 | LIST_HEAD(tmp); |
259 | struct list_head *pos, *node; |
260 | struct super_block *sb = NULL; |
261 | struct inode *inode; |
262 | int do_sb_sort = 0; |
263 | int moved = 0; |
264 | |
265 | while (!list_empty(delaying_queue)) { |
266 | inode = wb_inode(delaying_queue->prev); |
267 | if (work->older_than_this && |
268 | inode_dirtied_after(inode, *work->older_than_this)) |
269 | break; |
270 | if (sb && sb != inode->i_sb) |
271 | do_sb_sort = 1; |
272 | sb = inode->i_sb; |
273 | list_move(&inode->i_wb_list, &tmp); |
274 | moved++; |
275 | } |
276 | |
277 | /* just one sb in list, splice to dispatch_queue and we're done */ |
278 | if (!do_sb_sort) { |
279 | list_splice(&tmp, dispatch_queue); |
280 | goto out; |
281 | } |
282 | |
283 | /* Move inodes from one superblock together */ |
284 | while (!list_empty(&tmp)) { |
285 | sb = wb_inode(tmp.prev)->i_sb; |
286 | list_for_each_prev_safe(pos, node, &tmp) { |
287 | inode = wb_inode(pos); |
288 | if (inode->i_sb == sb) |
289 | list_move(&inode->i_wb_list, dispatch_queue); |
290 | } |
291 | } |
292 | out: |
293 | return moved; |
294 | } |
295 | |
296 | /* |
297 | * Queue all expired dirty inodes for io, eldest first. |
298 | * Before |
299 | * newly dirtied b_dirty b_io b_more_io |
300 | * =============> gf edc BA |
301 | * After |
302 | * newly dirtied b_dirty b_io b_more_io |
303 | * =============> g fBAedc |
304 | * | |
305 | * +--> dequeue for IO |
306 | */ |
307 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
308 | { |
309 | int moved; |
310 | assert_spin_locked(&wb->list_lock); |
311 | list_splice_init(&wb->b_more_io, &wb->b_io); |
312 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
313 | trace_writeback_queue_io(wb, work, moved); |
314 | } |
315 | |
316 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
317 | { |
318 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
319 | return inode->i_sb->s_op->write_inode(inode, wbc); |
320 | return 0; |
321 | } |
322 | |
323 | /* |
324 | * Wait for writeback on an inode to complete. Called with i_lock held. |
325 | * Caller must make sure inode cannot go away when we drop i_lock. |
326 | */ |
327 | static void __inode_wait_for_writeback(struct inode *inode) |
328 | __releases(inode->i_lock) |
329 | __acquires(inode->i_lock) |
330 | { |
331 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); |
332 | wait_queue_head_t *wqh; |
333 | |
334 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
335 | while (inode->i_state & I_SYNC) { |
336 | spin_unlock(&inode->i_lock); |
337 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
338 | spin_lock(&inode->i_lock); |
339 | } |
340 | } |
341 | |
342 | /* |
343 | * Wait for writeback on an inode to complete. Caller must have inode pinned. |
344 | */ |
345 | void inode_wait_for_writeback(struct inode *inode) |
346 | { |
347 | spin_lock(&inode->i_lock); |
348 | __inode_wait_for_writeback(inode); |
349 | spin_unlock(&inode->i_lock); |
350 | } |
351 | |
352 | /* |
353 | * Sleep until I_SYNC is cleared. This function must be called with i_lock |
354 | * held and drops it. It is aimed for callers not holding any inode reference |
355 | * so once i_lock is dropped, inode can go away. |
356 | */ |
357 | static void inode_sleep_on_writeback(struct inode *inode) |
358 | __releases(inode->i_lock) |
359 | { |
360 | DEFINE_WAIT(wait); |
361 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); |
362 | int sleep; |
363 | |
364 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
365 | sleep = inode->i_state & I_SYNC; |
366 | spin_unlock(&inode->i_lock); |
367 | if (sleep) |
368 | schedule(); |
369 | finish_wait(wqh, &wait); |
370 | } |
371 | |
372 | /* |
373 | * Find proper writeback list for the inode depending on its current state and |
374 | * possibly also change of its state while we were doing writeback. Here we |
375 | * handle things such as livelock prevention or fairness of writeback among |
376 | * inodes. This function can be called only by flusher thread - noone else |
377 | * processes all inodes in writeback lists and requeueing inodes behind flusher |
378 | * thread's back can have unexpected consequences. |
379 | */ |
380 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, |
381 | struct writeback_control *wbc) |
382 | { |
383 | if (inode->i_state & I_FREEING) |
384 | return; |
385 | |
386 | /* |
387 | * Sync livelock prevention. Each inode is tagged and synced in one |
388 | * shot. If still dirty, it will be redirty_tail()'ed below. Update |
389 | * the dirty time to prevent enqueue and sync it again. |
390 | */ |
391 | if ((inode->i_state & I_DIRTY) && |
392 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) |
393 | inode->dirtied_when = jiffies; |
394 | |
395 | if (wbc->pages_skipped) { |
396 | /* |
397 | * writeback is not making progress due to locked |
398 | * buffers. Skip this inode for now. |
399 | */ |
400 | redirty_tail(inode, wb); |
401 | return; |
402 | } |
403 | |
404 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
405 | /* |
406 | * We didn't write back all the pages. nfs_writepages() |
407 | * sometimes bales out without doing anything. |
408 | */ |
409 | if (wbc->nr_to_write <= 0) { |
410 | /* Slice used up. Queue for next turn. */ |
411 | requeue_io(inode, wb); |
412 | } else { |
413 | /* |
414 | * Writeback blocked by something other than |
415 | * congestion. Delay the inode for some time to |
416 | * avoid spinning on the CPU (100% iowait) |
417 | * retrying writeback of the dirty page/inode |
418 | * that cannot be performed immediately. |
419 | */ |
420 | redirty_tail(inode, wb); |
421 | } |
422 | } else if (inode->i_state & I_DIRTY) { |
423 | /* |
424 | * Filesystems can dirty the inode during writeback operations, |
425 | * such as delayed allocation during submission or metadata |
426 | * updates after data IO completion. |
427 | */ |
428 | redirty_tail(inode, wb); |
429 | } else { |
430 | /* The inode is clean. Remove from writeback lists. */ |
431 | list_del_init(&inode->i_wb_list); |
432 | } |
433 | } |
434 | |
435 | /* |
436 | * Write out an inode and its dirty pages. Do not update the writeback list |
437 | * linkage. That is left to the caller. The caller is also responsible for |
438 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. |
439 | */ |
440 | static int |
441 | __writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
442 | struct writeback_control *wbc) |
443 | { |
444 | struct address_space *mapping = inode->i_mapping; |
445 | long nr_to_write = wbc->nr_to_write; |
446 | unsigned dirty; |
447 | int ret; |
448 | |
449 | WARN_ON(!(inode->i_state & I_SYNC)); |
450 | |
451 | ret = do_writepages(mapping, wbc); |
452 | |
453 | /* |
454 | * Make sure to wait on the data before writing out the metadata. |
455 | * This is important for filesystems that modify metadata on data |
456 | * I/O completion. |
457 | */ |
458 | if (wbc->sync_mode == WB_SYNC_ALL) { |
459 | int err = filemap_fdatawait(mapping); |
460 | if (ret == 0) |
461 | ret = err; |
462 | } |
463 | |
464 | /* |
465 | * Some filesystems may redirty the inode during the writeback |
466 | * due to delalloc, clear dirty metadata flags right before |
467 | * write_inode() |
468 | */ |
469 | spin_lock(&inode->i_lock); |
470 | /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ |
471 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
472 | inode->i_state &= ~I_DIRTY_PAGES; |
473 | dirty = inode->i_state & I_DIRTY; |
474 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); |
475 | spin_unlock(&inode->i_lock); |
476 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
477 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
478 | int err = write_inode(inode, wbc); |
479 | if (ret == 0) |
480 | ret = err; |
481 | } |
482 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
483 | return ret; |
484 | } |
485 | |
486 | /* |
487 | * Write out an inode's dirty pages. Either the caller has an active reference |
488 | * on the inode or the inode has I_WILL_FREE set. |
489 | * |
490 | * This function is designed to be called for writing back one inode which |
491 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() |
492 | * and does more profound writeback list handling in writeback_sb_inodes(). |
493 | */ |
494 | static int |
495 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
496 | struct writeback_control *wbc) |
497 | { |
498 | int ret = 0; |
499 | |
500 | spin_lock(&inode->i_lock); |
501 | if (!atomic_read(&inode->i_count)) |
502 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
503 | else |
504 | WARN_ON(inode->i_state & I_WILL_FREE); |
505 | |
506 | if (inode->i_state & I_SYNC) { |
507 | if (wbc->sync_mode != WB_SYNC_ALL) |
508 | goto out; |
509 | /* |
510 | * It's a data-integrity sync. We must wait. Since callers hold |
511 | * inode reference or inode has I_WILL_FREE set, it cannot go |
512 | * away under us. |
513 | */ |
514 | __inode_wait_for_writeback(inode); |
515 | } |
516 | WARN_ON(inode->i_state & I_SYNC); |
517 | /* |
518 | * Skip inode if it is clean. We don't want to mess with writeback |
519 | * lists in this function since flusher thread may be doing for example |
520 | * sync in parallel and if we move the inode, it could get skipped. So |
521 | * here we make sure inode is on some writeback list and leave it there |
522 | * unless we have completely cleaned the inode. |
523 | */ |
524 | if (!(inode->i_state & I_DIRTY)) |
525 | goto out; |
526 | inode->i_state |= I_SYNC; |
527 | spin_unlock(&inode->i_lock); |
528 | |
529 | ret = __writeback_single_inode(inode, wb, wbc); |
530 | |
531 | spin_lock(&wb->list_lock); |
532 | spin_lock(&inode->i_lock); |
533 | /* |
534 | * If inode is clean, remove it from writeback lists. Otherwise don't |
535 | * touch it. See comment above for explanation. |
536 | */ |
537 | if (!(inode->i_state & I_DIRTY)) |
538 | list_del_init(&inode->i_wb_list); |
539 | spin_unlock(&wb->list_lock); |
540 | inode_sync_complete(inode); |
541 | out: |
542 | spin_unlock(&inode->i_lock); |
543 | return ret; |
544 | } |
545 | |
546 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
547 | struct wb_writeback_work *work) |
548 | { |
549 | long pages; |
550 | |
551 | /* |
552 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty |
553 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX |
554 | * here avoids calling into writeback_inodes_wb() more than once. |
555 | * |
556 | * The intended call sequence for WB_SYNC_ALL writeback is: |
557 | * |
558 | * wb_writeback() |
559 | * writeback_sb_inodes() <== called only once |
560 | * write_cache_pages() <== called once for each inode |
561 | * (quickly) tag currently dirty pages |
562 | * (maybe slowly) sync all tagged pages |
563 | */ |
564 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) |
565 | pages = LONG_MAX; |
566 | else { |
567 | pages = min(bdi->avg_write_bandwidth / 2, |
568 | global_dirty_limit / DIRTY_SCOPE); |
569 | pages = min(pages, work->nr_pages); |
570 | pages = round_down(pages + MIN_WRITEBACK_PAGES, |
571 | MIN_WRITEBACK_PAGES); |
572 | } |
573 | |
574 | return pages; |
575 | } |
576 | |
577 | /* |
578 | * Write a portion of b_io inodes which belong to @sb. |
579 | * |
580 | * If @only_this_sb is true, then find and write all such |
581 | * inodes. Otherwise write only ones which go sequentially |
582 | * in reverse order. |
583 | * |
584 | * Return the number of pages and/or inodes written. |
585 | */ |
586 | static long writeback_sb_inodes(struct super_block *sb, |
587 | struct bdi_writeback *wb, |
588 | struct wb_writeback_work *work) |
589 | { |
590 | struct writeback_control wbc = { |
591 | .sync_mode = work->sync_mode, |
592 | .tagged_writepages = work->tagged_writepages, |
593 | .for_kupdate = work->for_kupdate, |
594 | .for_background = work->for_background, |
595 | .range_cyclic = work->range_cyclic, |
596 | .range_start = 0, |
597 | .range_end = LLONG_MAX, |
598 | }; |
599 | unsigned long start_time = jiffies; |
600 | long write_chunk; |
601 | long wrote = 0; /* count both pages and inodes */ |
602 | |
603 | while (!list_empty(&wb->b_io)) { |
604 | struct inode *inode = wb_inode(wb->b_io.prev); |
605 | |
606 | if (inode->i_sb != sb) { |
607 | if (work->sb) { |
608 | /* |
609 | * We only want to write back data for this |
610 | * superblock, move all inodes not belonging |
611 | * to it back onto the dirty list. |
612 | */ |
613 | redirty_tail(inode, wb); |
614 | continue; |
615 | } |
616 | |
617 | /* |
618 | * The inode belongs to a different superblock. |
619 | * Bounce back to the caller to unpin this and |
620 | * pin the next superblock. |
621 | */ |
622 | break; |
623 | } |
624 | |
625 | /* |
626 | * Don't bother with new inodes or inodes being freed, first |
627 | * kind does not need periodic writeout yet, and for the latter |
628 | * kind writeout is handled by the freer. |
629 | */ |
630 | spin_lock(&inode->i_lock); |
631 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
632 | spin_unlock(&inode->i_lock); |
633 | redirty_tail(inode, wb); |
634 | continue; |
635 | } |
636 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
637 | /* |
638 | * If this inode is locked for writeback and we are not |
639 | * doing writeback-for-data-integrity, move it to |
640 | * b_more_io so that writeback can proceed with the |
641 | * other inodes on s_io. |
642 | * |
643 | * We'll have another go at writing back this inode |
644 | * when we completed a full scan of b_io. |
645 | */ |
646 | spin_unlock(&inode->i_lock); |
647 | requeue_io(inode, wb); |
648 | trace_writeback_sb_inodes_requeue(inode); |
649 | continue; |
650 | } |
651 | spin_unlock(&wb->list_lock); |
652 | |
653 | /* |
654 | * We already requeued the inode if it had I_SYNC set and we |
655 | * are doing WB_SYNC_NONE writeback. So this catches only the |
656 | * WB_SYNC_ALL case. |
657 | */ |
658 | if (inode->i_state & I_SYNC) { |
659 | /* Wait for I_SYNC. This function drops i_lock... */ |
660 | inode_sleep_on_writeback(inode); |
661 | /* Inode may be gone, start again */ |
662 | spin_lock(&wb->list_lock); |
663 | continue; |
664 | } |
665 | inode->i_state |= I_SYNC; |
666 | spin_unlock(&inode->i_lock); |
667 | |
668 | write_chunk = writeback_chunk_size(wb->bdi, work); |
669 | wbc.nr_to_write = write_chunk; |
670 | wbc.pages_skipped = 0; |
671 | |
672 | /* |
673 | * We use I_SYNC to pin the inode in memory. While it is set |
674 | * evict_inode() will wait so the inode cannot be freed. |
675 | */ |
676 | __writeback_single_inode(inode, wb, &wbc); |
677 | |
678 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
679 | wrote += write_chunk - wbc.nr_to_write; |
680 | spin_lock(&wb->list_lock); |
681 | spin_lock(&inode->i_lock); |
682 | if (!(inode->i_state & I_DIRTY)) |
683 | wrote++; |
684 | requeue_inode(inode, wb, &wbc); |
685 | inode_sync_complete(inode); |
686 | spin_unlock(&inode->i_lock); |
687 | cond_resched_lock(&wb->list_lock); |
688 | /* |
689 | * bail out to wb_writeback() often enough to check |
690 | * background threshold and other termination conditions. |
691 | */ |
692 | if (wrote) { |
693 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
694 | break; |
695 | if (work->nr_pages <= 0) |
696 | break; |
697 | } |
698 | } |
699 | return wrote; |
700 | } |
701 | |
702 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
703 | struct wb_writeback_work *work) |
704 | { |
705 | unsigned long start_time = jiffies; |
706 | long wrote = 0; |
707 | |
708 | while (!list_empty(&wb->b_io)) { |
709 | struct inode *inode = wb_inode(wb->b_io.prev); |
710 | struct super_block *sb = inode->i_sb; |
711 | |
712 | if (!grab_super_passive(sb)) { |
713 | /* |
714 | * grab_super_passive() may fail consistently due to |
715 | * s_umount being grabbed by someone else. Don't use |
716 | * requeue_io() to avoid busy retrying the inode/sb. |
717 | */ |
718 | redirty_tail(inode, wb); |
719 | continue; |
720 | } |
721 | wrote += writeback_sb_inodes(sb, wb, work); |
722 | drop_super(sb); |
723 | |
724 | /* refer to the same tests at the end of writeback_sb_inodes */ |
725 | if (wrote) { |
726 | if (time_is_before_jiffies(start_time + HZ / 10UL)) |
727 | break; |
728 | if (work->nr_pages <= 0) |
729 | break; |
730 | } |
731 | } |
732 | /* Leave any unwritten inodes on b_io */ |
733 | return wrote; |
734 | } |
735 | |
736 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
737 | enum wb_reason reason) |
738 | { |
739 | struct wb_writeback_work work = { |
740 | .nr_pages = nr_pages, |
741 | .sync_mode = WB_SYNC_NONE, |
742 | .range_cyclic = 1, |
743 | .reason = reason, |
744 | }; |
745 | |
746 | spin_lock(&wb->list_lock); |
747 | if (list_empty(&wb->b_io)) |
748 | queue_io(wb, &work); |
749 | __writeback_inodes_wb(wb, &work); |
750 | spin_unlock(&wb->list_lock); |
751 | |
752 | return nr_pages - work.nr_pages; |
753 | } |
754 | |
755 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
756 | { |
757 | unsigned long background_thresh, dirty_thresh; |
758 | |
759 | global_dirty_limits(&background_thresh, &dirty_thresh); |
760 | |
761 | if (global_page_state(NR_FILE_DIRTY) + |
762 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) |
763 | return true; |
764 | |
765 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > |
766 | bdi_dirty_limit(bdi, background_thresh)) |
767 | return true; |
768 | |
769 | return false; |
770 | } |
771 | |
772 | /* |
773 | * Called under wb->list_lock. If there are multiple wb per bdi, |
774 | * only the flusher working on the first wb should do it. |
775 | */ |
776 | static void wb_update_bandwidth(struct bdi_writeback *wb, |
777 | unsigned long start_time) |
778 | { |
779 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
780 | } |
781 | |
782 | /* |
783 | * Explicit flushing or periodic writeback of "old" data. |
784 | * |
785 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
786 | * dirtying-time in the inode's address_space. So this periodic writeback code |
787 | * just walks the superblock inode list, writing back any inodes which are |
788 | * older than a specific point in time. |
789 | * |
790 | * Try to run once per dirty_writeback_interval. But if a writeback event |
791 | * takes longer than a dirty_writeback_interval interval, then leave a |
792 | * one-second gap. |
793 | * |
794 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
795 | * all dirty pages if they are all attached to "old" mappings. |
796 | */ |
797 | static long wb_writeback(struct bdi_writeback *wb, |
798 | struct wb_writeback_work *work) |
799 | { |
800 | unsigned long wb_start = jiffies; |
801 | long nr_pages = work->nr_pages; |
802 | unsigned long oldest_jif; |
803 | struct inode *inode; |
804 | long progress; |
805 | |
806 | oldest_jif = jiffies; |
807 | work->older_than_this = &oldest_jif; |
808 | |
809 | spin_lock(&wb->list_lock); |
810 | for (;;) { |
811 | /* |
812 | * Stop writeback when nr_pages has been consumed |
813 | */ |
814 | if (work->nr_pages <= 0) |
815 | break; |
816 | |
817 | /* |
818 | * Background writeout and kupdate-style writeback may |
819 | * run forever. Stop them if there is other work to do |
820 | * so that e.g. sync can proceed. They'll be restarted |
821 | * after the other works are all done. |
822 | */ |
823 | if ((work->for_background || work->for_kupdate) && |
824 | !list_empty(&wb->bdi->work_list)) |
825 | break; |
826 | |
827 | /* |
828 | * For background writeout, stop when we are below the |
829 | * background dirty threshold |
830 | */ |
831 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
832 | break; |
833 | |
834 | /* |
835 | * Kupdate and background works are special and we want to |
836 | * include all inodes that need writing. Livelock avoidance is |
837 | * handled by these works yielding to any other work so we are |
838 | * safe. |
839 | */ |
840 | if (work->for_kupdate) { |
841 | oldest_jif = jiffies - |
842 | msecs_to_jiffies(dirty_expire_interval * 10); |
843 | } else if (work->for_background) |
844 | oldest_jif = jiffies; |
845 | |
846 | trace_writeback_start(wb->bdi, work); |
847 | if (list_empty(&wb->b_io)) |
848 | queue_io(wb, work); |
849 | if (work->sb) |
850 | progress = writeback_sb_inodes(work->sb, wb, work); |
851 | else |
852 | progress = __writeback_inodes_wb(wb, work); |
853 | trace_writeback_written(wb->bdi, work); |
854 | |
855 | wb_update_bandwidth(wb, wb_start); |
856 | |
857 | /* |
858 | * Did we write something? Try for more |
859 | * |
860 | * Dirty inodes are moved to b_io for writeback in batches. |
861 | * The completion of the current batch does not necessarily |
862 | * mean the overall work is done. So we keep looping as long |
863 | * as made some progress on cleaning pages or inodes. |
864 | */ |
865 | if (progress) |
866 | continue; |
867 | /* |
868 | * No more inodes for IO, bail |
869 | */ |
870 | if (list_empty(&wb->b_more_io)) |
871 | break; |
872 | /* |
873 | * Nothing written. Wait for some inode to |
874 | * become available for writeback. Otherwise |
875 | * we'll just busyloop. |
876 | */ |
877 | if (!list_empty(&wb->b_more_io)) { |
878 | trace_writeback_wait(wb->bdi, work); |
879 | inode = wb_inode(wb->b_more_io.prev); |
880 | spin_lock(&inode->i_lock); |
881 | spin_unlock(&wb->list_lock); |
882 | /* This function drops i_lock... */ |
883 | inode_sleep_on_writeback(inode); |
884 | spin_lock(&wb->list_lock); |
885 | } |
886 | } |
887 | spin_unlock(&wb->list_lock); |
888 | |
889 | return nr_pages - work->nr_pages; |
890 | } |
891 | |
892 | /* |
893 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
894 | */ |
895 | static struct wb_writeback_work * |
896 | get_next_work_item(struct backing_dev_info *bdi) |
897 | { |
898 | struct wb_writeback_work *work = NULL; |
899 | |
900 | spin_lock_bh(&bdi->wb_lock); |
901 | if (!list_empty(&bdi->work_list)) { |
902 | work = list_entry(bdi->work_list.next, |
903 | struct wb_writeback_work, list); |
904 | list_del_init(&work->list); |
905 | } |
906 | spin_unlock_bh(&bdi->wb_lock); |
907 | return work; |
908 | } |
909 | |
910 | /* |
911 | * Add in the number of potentially dirty inodes, because each inode |
912 | * write can dirty pagecache in the underlying blockdev. |
913 | */ |
914 | static unsigned long get_nr_dirty_pages(void) |
915 | { |
916 | return global_page_state(NR_FILE_DIRTY) + |
917 | global_page_state(NR_UNSTABLE_NFS) + |
918 | get_nr_dirty_inodes(); |
919 | } |
920 | |
921 | static long wb_check_background_flush(struct bdi_writeback *wb) |
922 | { |
923 | if (over_bground_thresh(wb->bdi)) { |
924 | |
925 | struct wb_writeback_work work = { |
926 | .nr_pages = LONG_MAX, |
927 | .sync_mode = WB_SYNC_NONE, |
928 | .for_background = 1, |
929 | .range_cyclic = 1, |
930 | .reason = WB_REASON_BACKGROUND, |
931 | }; |
932 | |
933 | return wb_writeback(wb, &work); |
934 | } |
935 | |
936 | return 0; |
937 | } |
938 | |
939 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
940 | { |
941 | unsigned long expired; |
942 | long nr_pages; |
943 | |
944 | /* |
945 | * When set to zero, disable periodic writeback |
946 | */ |
947 | if (!dirty_writeback_interval) |
948 | return 0; |
949 | |
950 | expired = wb->last_old_flush + |
951 | msecs_to_jiffies(dirty_writeback_interval * 10); |
952 | if (time_before(jiffies, expired)) |
953 | return 0; |
954 | |
955 | wb->last_old_flush = jiffies; |
956 | nr_pages = get_nr_dirty_pages(); |
957 | |
958 | if (nr_pages) { |
959 | struct wb_writeback_work work = { |
960 | .nr_pages = nr_pages, |
961 | .sync_mode = WB_SYNC_NONE, |
962 | .for_kupdate = 1, |
963 | .range_cyclic = 1, |
964 | .reason = WB_REASON_PERIODIC, |
965 | }; |
966 | |
967 | return wb_writeback(wb, &work); |
968 | } |
969 | |
970 | return 0; |
971 | } |
972 | |
973 | /* |
974 | * Retrieve work items and do the writeback they describe |
975 | */ |
976 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) |
977 | { |
978 | struct backing_dev_info *bdi = wb->bdi; |
979 | struct wb_writeback_work *work; |
980 | long wrote = 0; |
981 | |
982 | set_bit(BDI_writeback_running, &wb->bdi->state); |
983 | while ((work = get_next_work_item(bdi)) != NULL) { |
984 | /* |
985 | * Override sync mode, in case we must wait for completion |
986 | * because this thread is exiting now. |
987 | */ |
988 | if (force_wait) |
989 | work->sync_mode = WB_SYNC_ALL; |
990 | |
991 | trace_writeback_exec(bdi, work); |
992 | |
993 | wrote += wb_writeback(wb, work); |
994 | |
995 | /* |
996 | * Notify the caller of completion if this is a synchronous |
997 | * work item, otherwise just free it. |
998 | */ |
999 | if (work->done) |
1000 | complete(work->done); |
1001 | else |
1002 | kfree(work); |
1003 | } |
1004 | |
1005 | /* |
1006 | * Check for periodic writeback, kupdated() style |
1007 | */ |
1008 | wrote += wb_check_old_data_flush(wb); |
1009 | wrote += wb_check_background_flush(wb); |
1010 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
1011 | |
1012 | return wrote; |
1013 | } |
1014 | |
1015 | /* |
1016 | * Handle writeback of dirty data for the device backed by this bdi. Also |
1017 | * wakes up periodically and does kupdated style flushing. |
1018 | */ |
1019 | int bdi_writeback_thread(void *data) |
1020 | { |
1021 | struct bdi_writeback *wb = data; |
1022 | struct backing_dev_info *bdi = wb->bdi; |
1023 | long pages_written; |
1024 | |
1025 | current->flags |= PF_SWAPWRITE; |
1026 | set_freezable(); |
1027 | wb->last_active = jiffies; |
1028 | |
1029 | /* |
1030 | * Our parent may run at a different priority, just set us to normal |
1031 | */ |
1032 | set_user_nice(current, 0); |
1033 | |
1034 | trace_writeback_thread_start(bdi); |
1035 | |
1036 | while (!kthread_freezable_should_stop(NULL)) { |
1037 | /* |
1038 | * Remove own delayed wake-up timer, since we are already awake |
1039 | * and we'll take care of the preriodic write-back. |
1040 | */ |
1041 | del_timer(&wb->wakeup_timer); |
1042 | |
1043 | pages_written = wb_do_writeback(wb, 0); |
1044 | |
1045 | trace_writeback_pages_written(pages_written); |
1046 | |
1047 | if (pages_written) |
1048 | wb->last_active = jiffies; |
1049 | |
1050 | set_current_state(TASK_INTERRUPTIBLE); |
1051 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
1052 | __set_current_state(TASK_RUNNING); |
1053 | continue; |
1054 | } |
1055 | |
1056 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
1057 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
1058 | else { |
1059 | /* |
1060 | * We have nothing to do, so can go sleep without any |
1061 | * timeout and save power. When a work is queued or |
1062 | * something is made dirty - we will be woken up. |
1063 | */ |
1064 | schedule(); |
1065 | } |
1066 | } |
1067 | |
1068 | /* Flush any work that raced with us exiting */ |
1069 | if (!list_empty(&bdi->work_list)) |
1070 | wb_do_writeback(wb, 1); |
1071 | |
1072 | trace_writeback_thread_stop(bdi); |
1073 | return 0; |
1074 | } |
1075 | |
1076 | |
1077 | /* |
1078 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1079 | * the whole world. |
1080 | */ |
1081 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
1082 | { |
1083 | struct backing_dev_info *bdi; |
1084 | |
1085 | if (!nr_pages) { |
1086 | nr_pages = global_page_state(NR_FILE_DIRTY) + |
1087 | global_page_state(NR_UNSTABLE_NFS); |
1088 | } |
1089 | |
1090 | rcu_read_lock(); |
1091 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
1092 | if (!bdi_has_dirty_io(bdi)) |
1093 | continue; |
1094 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
1095 | } |
1096 | rcu_read_unlock(); |
1097 | } |
1098 | |
1099 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1100 | { |
1101 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { |
1102 | struct dentry *dentry; |
1103 | const char *name = "?"; |
1104 | |
1105 | dentry = d_find_alias(inode); |
1106 | if (dentry) { |
1107 | spin_lock(&dentry->d_lock); |
1108 | name = (const char *) dentry->d_name.name; |
1109 | } |
1110 | printk(KERN_DEBUG |
1111 | "%s(%d): dirtied inode %lu (%s) on %s\n", |
1112 | current->comm, task_pid_nr(current), inode->i_ino, |
1113 | name, inode->i_sb->s_id); |
1114 | if (dentry) { |
1115 | spin_unlock(&dentry->d_lock); |
1116 | dput(dentry); |
1117 | } |
1118 | } |
1119 | } |
1120 | |
1121 | /** |
1122 | * __mark_inode_dirty - internal function |
1123 | * @inode: inode to mark |
1124 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) |
1125 | * Mark an inode as dirty. Callers should use mark_inode_dirty or |
1126 | * mark_inode_dirty_sync. |
1127 | * |
1128 | * Put the inode on the super block's dirty list. |
1129 | * |
1130 | * CAREFUL! We mark it dirty unconditionally, but move it onto the |
1131 | * dirty list only if it is hashed or if it refers to a blockdev. |
1132 | * If it was not hashed, it will never be added to the dirty list |
1133 | * even if it is later hashed, as it will have been marked dirty already. |
1134 | * |
1135 | * In short, make sure you hash any inodes _before_ you start marking |
1136 | * them dirty. |
1137 | * |
1138 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1139 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of |
1140 | * the kernel-internal blockdev inode represents the dirtying time of the |
1141 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use |
1142 | * page->mapping->host, so the page-dirtying time is recorded in the internal |
1143 | * blockdev inode. |
1144 | */ |
1145 | void __mark_inode_dirty(struct inode *inode, int flags) |
1146 | { |
1147 | struct super_block *sb = inode->i_sb; |
1148 | struct backing_dev_info *bdi = NULL; |
1149 | |
1150 | /* |
1151 | * Don't do this for I_DIRTY_PAGES - that doesn't actually |
1152 | * dirty the inode itself |
1153 | */ |
1154 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { |
1155 | if (sb->s_op->dirty_inode) |
1156 | sb->s_op->dirty_inode(inode, flags); |
1157 | } |
1158 | |
1159 | /* |
1160 | * make sure that changes are seen by all cpus before we test i_state |
1161 | * -- mikulas |
1162 | */ |
1163 | smp_mb(); |
1164 | |
1165 | /* avoid the locking if we can */ |
1166 | if ((inode->i_state & flags) == flags) |
1167 | return; |
1168 | |
1169 | if (unlikely(block_dump)) |
1170 | block_dump___mark_inode_dirty(inode); |
1171 | |
1172 | spin_lock(&inode->i_lock); |
1173 | if ((inode->i_state & flags) != flags) { |
1174 | const int was_dirty = inode->i_state & I_DIRTY; |
1175 | |
1176 | inode->i_state |= flags; |
1177 | |
1178 | /* |
1179 | * If the inode is being synced, just update its dirty state. |
1180 | * The unlocker will place the inode on the appropriate |
1181 | * superblock list, based upon its state. |
1182 | */ |
1183 | if (inode->i_state & I_SYNC) |
1184 | goto out_unlock_inode; |
1185 | |
1186 | /* |
1187 | * Only add valid (hashed) inodes to the superblock's |
1188 | * dirty list. Add blockdev inodes as well. |
1189 | */ |
1190 | if (!S_ISBLK(inode->i_mode)) { |
1191 | if (inode_unhashed(inode)) |
1192 | goto out_unlock_inode; |
1193 | } |
1194 | if (inode->i_state & I_FREEING) |
1195 | goto out_unlock_inode; |
1196 | |
1197 | /* |
1198 | * If the inode was already on b_dirty/b_io/b_more_io, don't |
1199 | * reposition it (that would break b_dirty time-ordering). |
1200 | */ |
1201 | if (!was_dirty) { |
1202 | bool wakeup_bdi = false; |
1203 | bdi = inode_to_bdi(inode); |
1204 | |
1205 | if (bdi_cap_writeback_dirty(bdi)) { |
1206 | WARN(!test_bit(BDI_registered, &bdi->state), |
1207 | "bdi-%s not registered\n", bdi->name); |
1208 | |
1209 | /* |
1210 | * If this is the first dirty inode for this |
1211 | * bdi, we have to wake-up the corresponding |
1212 | * bdi thread to make sure background |
1213 | * write-back happens later. |
1214 | */ |
1215 | if (!wb_has_dirty_io(&bdi->wb)) |
1216 | wakeup_bdi = true; |
1217 | } |
1218 | |
1219 | spin_unlock(&inode->i_lock); |
1220 | spin_lock(&bdi->wb.list_lock); |
1221 | inode->dirtied_when = jiffies; |
1222 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
1223 | spin_unlock(&bdi->wb.list_lock); |
1224 | |
1225 | if (wakeup_bdi) |
1226 | bdi_wakeup_thread_delayed(bdi); |
1227 | return; |
1228 | } |
1229 | } |
1230 | out_unlock_inode: |
1231 | spin_unlock(&inode->i_lock); |
1232 | |
1233 | } |
1234 | EXPORT_SYMBOL(__mark_inode_dirty); |
1235 | |
1236 | static void wait_sb_inodes(struct super_block *sb) |
1237 | { |
1238 | struct inode *inode, *old_inode = NULL; |
1239 | |
1240 | /* |
1241 | * We need to be protected against the filesystem going from |
1242 | * r/o to r/w or vice versa. |
1243 | */ |
1244 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1245 | |
1246 | spin_lock(&inode_sb_list_lock); |
1247 | |
1248 | /* |
1249 | * Data integrity sync. Must wait for all pages under writeback, |
1250 | * because there may have been pages dirtied before our sync |
1251 | * call, but which had writeout started before we write it out. |
1252 | * In which case, the inode may not be on the dirty list, but |
1253 | * we still have to wait for that writeout. |
1254 | */ |
1255 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
1256 | struct address_space *mapping = inode->i_mapping; |
1257 | |
1258 | spin_lock(&inode->i_lock); |
1259 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || |
1260 | (mapping->nrpages == 0)) { |
1261 | spin_unlock(&inode->i_lock); |
1262 | continue; |
1263 | } |
1264 | __iget(inode); |
1265 | spin_unlock(&inode->i_lock); |
1266 | spin_unlock(&inode_sb_list_lock); |
1267 | |
1268 | /* |
1269 | * We hold a reference to 'inode' so it couldn't have been |
1270 | * removed from s_inodes list while we dropped the |
1271 | * inode_sb_list_lock. We cannot iput the inode now as we can |
1272 | * be holding the last reference and we cannot iput it under |
1273 | * inode_sb_list_lock. So we keep the reference and iput it |
1274 | * later. |
1275 | */ |
1276 | iput(old_inode); |
1277 | old_inode = inode; |
1278 | |
1279 | filemap_fdatawait(mapping); |
1280 | |
1281 | cond_resched(); |
1282 | |
1283 | spin_lock(&inode_sb_list_lock); |
1284 | } |
1285 | spin_unlock(&inode_sb_list_lock); |
1286 | iput(old_inode); |
1287 | } |
1288 | |
1289 | /** |
1290 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
1291 | * @sb: the superblock |
1292 | * @nr: the number of pages to write |
1293 | * @reason: reason why some writeback work initiated |
1294 | * |
1295 | * Start writeback on some inodes on this super_block. No guarantees are made |
1296 | * on how many (if any) will be written, and this function does not wait |
1297 | * for IO completion of submitted IO. |
1298 | */ |
1299 | void writeback_inodes_sb_nr(struct super_block *sb, |
1300 | unsigned long nr, |
1301 | enum wb_reason reason) |
1302 | { |
1303 | DECLARE_COMPLETION_ONSTACK(done); |
1304 | struct wb_writeback_work work = { |
1305 | .sb = sb, |
1306 | .sync_mode = WB_SYNC_NONE, |
1307 | .tagged_writepages = 1, |
1308 | .done = &done, |
1309 | .nr_pages = nr, |
1310 | .reason = reason, |
1311 | }; |
1312 | |
1313 | if (sb->s_bdi == &noop_backing_dev_info) |
1314 | return; |
1315 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1316 | bdi_queue_work(sb->s_bdi, &work); |
1317 | wait_for_completion(&done); |
1318 | } |
1319 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1320 | |
1321 | /** |
1322 | * writeback_inodes_sb - writeback dirty inodes from given super_block |
1323 | * @sb: the superblock |
1324 | * @reason: reason why some writeback work was initiated |
1325 | * |
1326 | * Start writeback on some inodes on this super_block. No guarantees are made |
1327 | * on how many (if any) will be written, and this function does not wait |
1328 | * for IO completion of submitted IO. |
1329 | */ |
1330 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
1331 | { |
1332 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
1333 | } |
1334 | EXPORT_SYMBOL(writeback_inodes_sb); |
1335 | |
1336 | /** |
1337 | * writeback_inodes_sb_if_idle - start writeback if none underway |
1338 | * @sb: the superblock |
1339 | * @reason: reason why some writeback work was initiated |
1340 | * |
1341 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
1342 | * Returns 1 if writeback was started, 0 if not. |
1343 | */ |
1344 | int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) |
1345 | { |
1346 | if (!writeback_in_progress(sb->s_bdi)) { |
1347 | down_read(&sb->s_umount); |
1348 | writeback_inodes_sb(sb, reason); |
1349 | up_read(&sb->s_umount); |
1350 | return 1; |
1351 | } else |
1352 | return 0; |
1353 | } |
1354 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); |
1355 | |
1356 | /** |
1357 | * writeback_inodes_sb_nr_if_idle - start writeback if none underway |
1358 | * @sb: the superblock |
1359 | * @nr: the number of pages to write |
1360 | * @reason: reason why some writeback work was initiated |
1361 | * |
1362 | * Invoke writeback_inodes_sb if no writeback is currently underway. |
1363 | * Returns 1 if writeback was started, 0 if not. |
1364 | */ |
1365 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, |
1366 | unsigned long nr, |
1367 | enum wb_reason reason) |
1368 | { |
1369 | if (!writeback_in_progress(sb->s_bdi)) { |
1370 | down_read(&sb->s_umount); |
1371 | writeback_inodes_sb_nr(sb, nr, reason); |
1372 | up_read(&sb->s_umount); |
1373 | return 1; |
1374 | } else |
1375 | return 0; |
1376 | } |
1377 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); |
1378 | |
1379 | /** |
1380 | * sync_inodes_sb - sync sb inode pages |
1381 | * @sb: the superblock |
1382 | * |
1383 | * This function writes and waits on any dirty inode belonging to this |
1384 | * super_block. |
1385 | */ |
1386 | void sync_inodes_sb(struct super_block *sb) |
1387 | { |
1388 | DECLARE_COMPLETION_ONSTACK(done); |
1389 | struct wb_writeback_work work = { |
1390 | .sb = sb, |
1391 | .sync_mode = WB_SYNC_ALL, |
1392 | .nr_pages = LONG_MAX, |
1393 | .range_cyclic = 0, |
1394 | .done = &done, |
1395 | .reason = WB_REASON_SYNC, |
1396 | }; |
1397 | |
1398 | /* Nothing to do? */ |
1399 | if (sb->s_bdi == &noop_backing_dev_info) |
1400 | return; |
1401 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1402 | |
1403 | bdi_queue_work(sb->s_bdi, &work); |
1404 | wait_for_completion(&done); |
1405 | |
1406 | wait_sb_inodes(sb); |
1407 | } |
1408 | EXPORT_SYMBOL(sync_inodes_sb); |
1409 | |
1410 | /** |
1411 | * write_inode_now - write an inode to disk |
1412 | * @inode: inode to write to disk |
1413 | * @sync: whether the write should be synchronous or not |
1414 | * |
1415 | * This function commits an inode to disk immediately if it is dirty. This is |
1416 | * primarily needed by knfsd. |
1417 | * |
1418 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1419 | */ |
1420 | int write_inode_now(struct inode *inode, int sync) |
1421 | { |
1422 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1423 | struct writeback_control wbc = { |
1424 | .nr_to_write = LONG_MAX, |
1425 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
1426 | .range_start = 0, |
1427 | .range_end = LLONG_MAX, |
1428 | }; |
1429 | |
1430 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) |
1431 | wbc.nr_to_write = 0; |
1432 | |
1433 | might_sleep(); |
1434 | return writeback_single_inode(inode, wb, &wbc); |
1435 | } |
1436 | EXPORT_SYMBOL(write_inode_now); |
1437 | |
1438 | /** |
1439 | * sync_inode - write an inode and its pages to disk. |
1440 | * @inode: the inode to sync |
1441 | * @wbc: controls the writeback mode |
1442 | * |
1443 | * sync_inode() will write an inode and its pages to disk. It will also |
1444 | * correctly update the inode on its superblock's dirty inode lists and will |
1445 | * update inode->i_state. |
1446 | * |
1447 | * The caller must have a ref on the inode. |
1448 | */ |
1449 | int sync_inode(struct inode *inode, struct writeback_control *wbc) |
1450 | { |
1451 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1452 | } |
1453 | EXPORT_SYMBOL(sync_inode); |
1454 | |
1455 | /** |
1456 | * sync_inode_metadata - write an inode to disk |
1457 | * @inode: the inode to sync |
1458 | * @wait: wait for I/O to complete. |
1459 | * |
1460 | * Write an inode to disk and adjust its dirty state after completion. |
1461 | * |
1462 | * Note: only writes the actual inode, no associated data or other metadata. |
1463 | */ |
1464 | int sync_inode_metadata(struct inode *inode, int wait) |
1465 | { |
1466 | struct writeback_control wbc = { |
1467 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, |
1468 | .nr_to_write = 0, /* metadata-only */ |
1469 | }; |
1470 | |
1471 | return sync_inode(inode, &wbc); |
1472 | } |
1473 | EXPORT_SYMBOL(sync_inode_metadata); |
1474 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9