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