]>
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/writeback.h> | |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/tracepoint.h> | |
29 | #include <linux/device.h> | |
30 | #include "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 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ | |
50 | enum wb_reason reason; /* why was writeback initiated? */ | |
51 | ||
52 | struct list_head list; /* pending work list */ | |
53 | struct completion *done; /* set if the caller waits */ | |
54 | }; | |
55 | ||
56 | /** | |
57 | * writeback_in_progress - determine whether there is writeback in progress | |
58 | * @bdi: the device's backing_dev_info structure. | |
59 | * | |
60 | * Determine whether there is writeback waiting to be handled against a | |
61 | * backing device. | |
62 | */ | |
63 | int writeback_in_progress(struct backing_dev_info *bdi) | |
64 | { | |
65 | return test_bit(BDI_writeback_running, &bdi->state); | |
66 | } | |
67 | EXPORT_SYMBOL(writeback_in_progress); | |
68 | ||
69 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) | |
70 | { | |
71 | struct super_block *sb = inode->i_sb; | |
72 | ||
73 | if (sb_is_blkdev_sb(sb)) | |
74 | return inode->i_mapping->backing_dev_info; | |
75 | ||
76 | return sb->s_bdi; | |
77 | } | |
78 | ||
79 | static inline struct inode *wb_inode(struct list_head *head) | |
80 | { | |
81 | return list_entry(head, struct inode, i_wb_list); | |
82 | } | |
83 | ||
84 | /* | |
85 | * Include the creation of the trace points after defining the | |
86 | * wb_writeback_work structure and inline functions so that the definition | |
87 | * remains local to this file. | |
88 | */ | |
89 | #define CREATE_TRACE_POINTS | |
90 | #include <trace/events/writeback.h> | |
91 | ||
92 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); | |
93 | ||
94 | static void bdi_wakeup_thread(struct backing_dev_info *bdi) | |
95 | { | |
96 | spin_lock_bh(&bdi->wb_lock); | |
97 | if (test_bit(BDI_registered, &bdi->state)) | |
98 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
99 | spin_unlock_bh(&bdi->wb_lock); | |
100 | } | |
101 | ||
102 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
103 | struct wb_writeback_work *work) | |
104 | { | |
105 | trace_writeback_queue(bdi, work); | |
106 | ||
107 | spin_lock_bh(&bdi->wb_lock); | |
108 | if (!test_bit(BDI_registered, &bdi->state)) { | |
109 | if (work->done) | |
110 | complete(work->done); | |
111 | goto out_unlock; | |
112 | } | |
113 | list_add_tail(&work->list, &bdi->work_list); | |
114 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
115 | out_unlock: | |
116 | spin_unlock_bh(&bdi->wb_lock); | |
117 | } | |
118 | ||
119 | static void | |
120 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
121 | bool range_cyclic, enum wb_reason reason) | |
122 | { | |
123 | struct wb_writeback_work *work; | |
124 | ||
125 | /* | |
126 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
127 | * wakeup the thread for old dirty data writeback | |
128 | */ | |
129 | work = kzalloc(sizeof(*work), GFP_ATOMIC); | |
130 | if (!work) { | |
131 | trace_writeback_nowork(bdi); | |
132 | bdi_wakeup_thread(bdi); | |
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 | bdi_wakeup_thread(bdi); | |
180 | } | |
181 | ||
182 | /* | |
183 | * Remove the inode from the writeback list it is on. | |
184 | */ | |
185 | void inode_wb_list_del(struct inode *inode) | |
186 | { | |
187 | struct backing_dev_info *bdi = inode_to_bdi(inode); | |
188 | ||
189 | spin_lock(&bdi->wb.list_lock); | |
190 | list_del_init(&inode->i_wb_list); | |
191 | spin_unlock(&bdi->wb.list_lock); | |
192 | } | |
193 | ||
194 | /* | |
195 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
196 | * furthest end of its superblock's dirty-inode list. | |
197 | * | |
198 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
199 | * already the most-recently-dirtied inode on the b_dirty list. If that is | |
200 | * the case then the inode must have been redirtied while it was being written | |
201 | * out and we don't reset its dirtied_when. | |
202 | */ | |
203 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) | |
204 | { | |
205 | assert_spin_locked(&wb->list_lock); | |
206 | if (!list_empty(&wb->b_dirty)) { | |
207 | struct inode *tail; | |
208 | ||
209 | tail = wb_inode(wb->b_dirty.next); | |
210 | if (time_before(inode->dirtied_when, tail->dirtied_when)) | |
211 | inode->dirtied_when = jiffies; | |
212 | } | |
213 | list_move(&inode->i_wb_list, &wb->b_dirty); | |
214 | } | |
215 | ||
216 | /* | |
217 | * requeue inode for re-scanning after bdi->b_io list is exhausted. | |
218 | */ | |
219 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) | |
220 | { | |
221 | assert_spin_locked(&wb->list_lock); | |
222 | list_move(&inode->i_wb_list, &wb->b_more_io); | |
223 | } | |
224 | ||
225 | static void inode_sync_complete(struct inode *inode) | |
226 | { | |
227 | inode->i_state &= ~I_SYNC; | |
228 | /* If inode is clean an unused, put it into LRU now... */ | |
229 | inode_add_lru(inode); | |
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 before 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 | list_move(&inode->i_wb_list, &tmp); | |
271 | moved++; | |
272 | if (sb_is_blkdev_sb(inode->i_sb)) | |
273 | continue; | |
274 | if (sb && sb != inode->i_sb) | |
275 | do_sb_sort = 1; | |
276 | sb = inode->i_sb; | |
277 | } | |
278 | ||
279 | /* just one sb in list, splice to dispatch_queue and we're done */ | |
280 | if (!do_sb_sort) { | |
281 | list_splice(&tmp, dispatch_queue); | |
282 | goto out; | |
283 | } | |
284 | ||
285 | /* Move inodes from one superblock together */ | |
286 | while (!list_empty(&tmp)) { | |
287 | sb = wb_inode(tmp.prev)->i_sb; | |
288 | list_for_each_prev_safe(pos, node, &tmp) { | |
289 | inode = wb_inode(pos); | |
290 | if (inode->i_sb == sb) | |
291 | list_move(&inode->i_wb_list, dispatch_queue); | |
292 | } | |
293 | } | |
294 | out: | |
295 | return moved; | |
296 | } | |
297 | ||
298 | /* | |
299 | * Queue all expired dirty inodes for io, eldest first. | |
300 | * Before | |
301 | * newly dirtied b_dirty b_io b_more_io | |
302 | * =============> gf edc BA | |
303 | * After | |
304 | * newly dirtied b_dirty b_io b_more_io | |
305 | * =============> g fBAedc | |
306 | * | | |
307 | * +--> dequeue for IO | |
308 | */ | |
309 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) | |
310 | { | |
311 | int moved; | |
312 | assert_spin_locked(&wb->list_lock); | |
313 | list_splice_init(&wb->b_more_io, &wb->b_io); | |
314 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); | |
315 | trace_writeback_queue_io(wb, work, moved); | |
316 | } | |
317 | ||
318 | static int write_inode(struct inode *inode, struct writeback_control *wbc) | |
319 | { | |
320 | int ret; | |
321 | ||
322 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
323 | trace_writeback_write_inode_start(inode, wbc); | |
324 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
325 | trace_writeback_write_inode(inode, wbc); | |
326 | return ret; | |
327 | } | |
328 | return 0; | |
329 | } | |
330 | ||
331 | /* | |
332 | * Wait for writeback on an inode to complete. Called with i_lock held. | |
333 | * Caller must make sure inode cannot go away when we drop i_lock. | |
334 | */ | |
335 | static void __inode_wait_for_writeback(struct inode *inode) | |
336 | __releases(inode->i_lock) | |
337 | __acquires(inode->i_lock) | |
338 | { | |
339 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
340 | wait_queue_head_t *wqh; | |
341 | ||
342 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
343 | while (inode->i_state & I_SYNC) { | |
344 | spin_unlock(&inode->i_lock); | |
345 | __wait_on_bit(wqh, &wq, bit_wait, | |
346 | TASK_UNINTERRUPTIBLE); | |
347 | spin_lock(&inode->i_lock); | |
348 | } | |
349 | } | |
350 | ||
351 | /* | |
352 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
353 | */ | |
354 | void inode_wait_for_writeback(struct inode *inode) | |
355 | { | |
356 | spin_lock(&inode->i_lock); | |
357 | __inode_wait_for_writeback(inode); | |
358 | spin_unlock(&inode->i_lock); | |
359 | } | |
360 | ||
361 | /* | |
362 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
363 | * held and drops it. It is aimed for callers not holding any inode reference | |
364 | * so once i_lock is dropped, inode can go away. | |
365 | */ | |
366 | static void inode_sleep_on_writeback(struct inode *inode) | |
367 | __releases(inode->i_lock) | |
368 | { | |
369 | DEFINE_WAIT(wait); | |
370 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
371 | int sleep; | |
372 | ||
373 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
374 | sleep = inode->i_state & I_SYNC; | |
375 | spin_unlock(&inode->i_lock); | |
376 | if (sleep) | |
377 | schedule(); | |
378 | finish_wait(wqh, &wait); | |
379 | } | |
380 | ||
381 | /* | |
382 | * Find proper writeback list for the inode depending on its current state and | |
383 | * possibly also change of its state while we were doing writeback. Here we | |
384 | * handle things such as livelock prevention or fairness of writeback among | |
385 | * inodes. This function can be called only by flusher thread - noone else | |
386 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
387 | * thread's back can have unexpected consequences. | |
388 | */ | |
389 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
390 | struct writeback_control *wbc) | |
391 | { | |
392 | if (inode->i_state & I_FREEING) | |
393 | return; | |
394 | ||
395 | /* | |
396 | * Sync livelock prevention. Each inode is tagged and synced in one | |
397 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
398 | * the dirty time to prevent enqueue and sync it again. | |
399 | */ | |
400 | if ((inode->i_state & I_DIRTY) && | |
401 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
402 | inode->dirtied_when = jiffies; | |
403 | ||
404 | if (wbc->pages_skipped) { | |
405 | /* | |
406 | * writeback is not making progress due to locked | |
407 | * buffers. Skip this inode for now. | |
408 | */ | |
409 | redirty_tail(inode, wb); | |
410 | return; | |
411 | } | |
412 | ||
413 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { | |
414 | /* | |
415 | * We didn't write back all the pages. nfs_writepages() | |
416 | * sometimes bales out without doing anything. | |
417 | */ | |
418 | if (wbc->nr_to_write <= 0) { | |
419 | /* Slice used up. Queue for next turn. */ | |
420 | requeue_io(inode, wb); | |
421 | } else { | |
422 | /* | |
423 | * Writeback blocked by something other than | |
424 | * congestion. Delay the inode for some time to | |
425 | * avoid spinning on the CPU (100% iowait) | |
426 | * retrying writeback of the dirty page/inode | |
427 | * that cannot be performed immediately. | |
428 | */ | |
429 | redirty_tail(inode, wb); | |
430 | } | |
431 | } else if (inode->i_state & I_DIRTY) { | |
432 | /* | |
433 | * Filesystems can dirty the inode during writeback operations, | |
434 | * such as delayed allocation during submission or metadata | |
435 | * updates after data IO completion. | |
436 | */ | |
437 | redirty_tail(inode, wb); | |
438 | } else { | |
439 | /* The inode is clean. Remove from writeback lists. */ | |
440 | list_del_init(&inode->i_wb_list); | |
441 | } | |
442 | } | |
443 | ||
444 | /* | |
445 | * Write out an inode and its dirty pages. Do not update the writeback list | |
446 | * linkage. That is left to the caller. The caller is also responsible for | |
447 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
448 | */ | |
449 | static int | |
450 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) | |
451 | { | |
452 | struct address_space *mapping = inode->i_mapping; | |
453 | long nr_to_write = wbc->nr_to_write; | |
454 | unsigned dirty; | |
455 | int ret; | |
456 | ||
457 | WARN_ON(!(inode->i_state & I_SYNC)); | |
458 | ||
459 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); | |
460 | ||
461 | ret = do_writepages(mapping, wbc); | |
462 | ||
463 | /* | |
464 | * Make sure to wait on the data before writing out the metadata. | |
465 | * This is important for filesystems that modify metadata on data | |
466 | * I/O completion. We don't do it for sync(2) writeback because it has a | |
467 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
468 | * inode metadata is written back correctly. | |
469 | */ | |
470 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { | |
471 | int err = filemap_fdatawait(mapping); | |
472 | if (ret == 0) | |
473 | ret = err; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Some filesystems may redirty the inode during the writeback | |
478 | * due to delalloc, clear dirty metadata flags right before | |
479 | * write_inode() | |
480 | */ | |
481 | spin_lock(&inode->i_lock); | |
482 | ||
483 | dirty = inode->i_state & I_DIRTY; | |
484 | inode->i_state &= ~I_DIRTY; | |
485 | ||
486 | /* | |
487 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
488 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
489 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
490 | * inode. | |
491 | * | |
492 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
493 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
494 | * necessary. This guarantees that either __mark_inode_dirty() | |
495 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
496 | */ | |
497 | smp_mb(); | |
498 | ||
499 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
500 | inode->i_state |= I_DIRTY_PAGES; | |
501 | ||
502 | spin_unlock(&inode->i_lock); | |
503 | ||
504 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | |
505 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
506 | int err = write_inode(inode, wbc); | |
507 | if (ret == 0) | |
508 | ret = err; | |
509 | } | |
510 | trace_writeback_single_inode(inode, wbc, nr_to_write); | |
511 | return ret; | |
512 | } | |
513 | ||
514 | /* | |
515 | * Write out an inode's dirty pages. Either the caller has an active reference | |
516 | * on the inode or the inode has I_WILL_FREE set. | |
517 | * | |
518 | * This function is designed to be called for writing back one inode which | |
519 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
520 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
521 | */ | |
522 | static int | |
523 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
524 | struct writeback_control *wbc) | |
525 | { | |
526 | int ret = 0; | |
527 | ||
528 | spin_lock(&inode->i_lock); | |
529 | if (!atomic_read(&inode->i_count)) | |
530 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
531 | else | |
532 | WARN_ON(inode->i_state & I_WILL_FREE); | |
533 | ||
534 | if (inode->i_state & I_SYNC) { | |
535 | if (wbc->sync_mode != WB_SYNC_ALL) | |
536 | goto out; | |
537 | /* | |
538 | * It's a data-integrity sync. We must wait. Since callers hold | |
539 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
540 | * away under us. | |
541 | */ | |
542 | __inode_wait_for_writeback(inode); | |
543 | } | |
544 | WARN_ON(inode->i_state & I_SYNC); | |
545 | /* | |
546 | * Skip inode if it is clean and we have no outstanding writeback in | |
547 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
548 | * function since flusher thread may be doing for example sync in | |
549 | * parallel and if we move the inode, it could get skipped. So here we | |
550 | * make sure inode is on some writeback list and leave it there unless | |
551 | * we have completely cleaned the inode. | |
552 | */ | |
553 | if (!(inode->i_state & I_DIRTY) && | |
554 | (wbc->sync_mode != WB_SYNC_ALL || | |
555 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
556 | goto out; | |
557 | inode->i_state |= I_SYNC; | |
558 | spin_unlock(&inode->i_lock); | |
559 | ||
560 | ret = __writeback_single_inode(inode, wbc); | |
561 | ||
562 | spin_lock(&wb->list_lock); | |
563 | spin_lock(&inode->i_lock); | |
564 | /* | |
565 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
566 | * touch it. See comment above for explanation. | |
567 | */ | |
568 | if (!(inode->i_state & I_DIRTY)) | |
569 | list_del_init(&inode->i_wb_list); | |
570 | spin_unlock(&wb->list_lock); | |
571 | inode_sync_complete(inode); | |
572 | out: | |
573 | spin_unlock(&inode->i_lock); | |
574 | return ret; | |
575 | } | |
576 | ||
577 | static long writeback_chunk_size(struct backing_dev_info *bdi, | |
578 | struct wb_writeback_work *work) | |
579 | { | |
580 | long pages; | |
581 | ||
582 | /* | |
583 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
584 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
585 | * here avoids calling into writeback_inodes_wb() more than once. | |
586 | * | |
587 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
588 | * | |
589 | * wb_writeback() | |
590 | * writeback_sb_inodes() <== called only once | |
591 | * write_cache_pages() <== called once for each inode | |
592 | * (quickly) tag currently dirty pages | |
593 | * (maybe slowly) sync all tagged pages | |
594 | */ | |
595 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
596 | pages = LONG_MAX; | |
597 | else { | |
598 | pages = min(bdi->avg_write_bandwidth / 2, | |
599 | global_dirty_limit / DIRTY_SCOPE); | |
600 | pages = min(pages, work->nr_pages); | |
601 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
602 | MIN_WRITEBACK_PAGES); | |
603 | } | |
604 | ||
605 | return pages; | |
606 | } | |
607 | ||
608 | /* | |
609 | * Write a portion of b_io inodes which belong to @sb. | |
610 | * | |
611 | * Return the number of pages and/or inodes written. | |
612 | */ | |
613 | static long writeback_sb_inodes(struct super_block *sb, | |
614 | struct bdi_writeback *wb, | |
615 | struct wb_writeback_work *work) | |
616 | { | |
617 | struct writeback_control wbc = { | |
618 | .sync_mode = work->sync_mode, | |
619 | .tagged_writepages = work->tagged_writepages, | |
620 | .for_kupdate = work->for_kupdate, | |
621 | .for_background = work->for_background, | |
622 | .for_sync = work->for_sync, | |
623 | .range_cyclic = work->range_cyclic, | |
624 | .range_start = 0, | |
625 | .range_end = LLONG_MAX, | |
626 | }; | |
627 | unsigned long start_time = jiffies; | |
628 | long write_chunk; | |
629 | long wrote = 0; /* count both pages and inodes */ | |
630 | ||
631 | while (!list_empty(&wb->b_io)) { | |
632 | struct inode *inode = wb_inode(wb->b_io.prev); | |
633 | ||
634 | if (inode->i_sb != sb) { | |
635 | if (work->sb) { | |
636 | /* | |
637 | * We only want to write back data for this | |
638 | * superblock, move all inodes not belonging | |
639 | * to it back onto the dirty list. | |
640 | */ | |
641 | redirty_tail(inode, wb); | |
642 | continue; | |
643 | } | |
644 | ||
645 | /* | |
646 | * The inode belongs to a different superblock. | |
647 | * Bounce back to the caller to unpin this and | |
648 | * pin the next superblock. | |
649 | */ | |
650 | break; | |
651 | } | |
652 | ||
653 | /* | |
654 | * Don't bother with new inodes or inodes being freed, first | |
655 | * kind does not need periodic writeout yet, and for the latter | |
656 | * kind writeout is handled by the freer. | |
657 | */ | |
658 | spin_lock(&inode->i_lock); | |
659 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
660 | spin_unlock(&inode->i_lock); | |
661 | redirty_tail(inode, wb); | |
662 | continue; | |
663 | } | |
664 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { | |
665 | /* | |
666 | * If this inode is locked for writeback and we are not | |
667 | * doing writeback-for-data-integrity, move it to | |
668 | * b_more_io so that writeback can proceed with the | |
669 | * other inodes on s_io. | |
670 | * | |
671 | * We'll have another go at writing back this inode | |
672 | * when we completed a full scan of b_io. | |
673 | */ | |
674 | spin_unlock(&inode->i_lock); | |
675 | requeue_io(inode, wb); | |
676 | trace_writeback_sb_inodes_requeue(inode); | |
677 | continue; | |
678 | } | |
679 | spin_unlock(&wb->list_lock); | |
680 | ||
681 | /* | |
682 | * We already requeued the inode if it had I_SYNC set and we | |
683 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
684 | * WB_SYNC_ALL case. | |
685 | */ | |
686 | if (inode->i_state & I_SYNC) { | |
687 | /* Wait for I_SYNC. This function drops i_lock... */ | |
688 | inode_sleep_on_writeback(inode); | |
689 | /* Inode may be gone, start again */ | |
690 | spin_lock(&wb->list_lock); | |
691 | continue; | |
692 | } | |
693 | inode->i_state |= I_SYNC; | |
694 | spin_unlock(&inode->i_lock); | |
695 | ||
696 | write_chunk = writeback_chunk_size(wb->bdi, work); | |
697 | wbc.nr_to_write = write_chunk; | |
698 | wbc.pages_skipped = 0; | |
699 | ||
700 | /* | |
701 | * We use I_SYNC to pin the inode in memory. While it is set | |
702 | * evict_inode() will wait so the inode cannot be freed. | |
703 | */ | |
704 | __writeback_single_inode(inode, &wbc); | |
705 | ||
706 | work->nr_pages -= write_chunk - wbc.nr_to_write; | |
707 | wrote += write_chunk - wbc.nr_to_write; | |
708 | spin_lock(&wb->list_lock); | |
709 | spin_lock(&inode->i_lock); | |
710 | if (!(inode->i_state & I_DIRTY)) | |
711 | wrote++; | |
712 | requeue_inode(inode, wb, &wbc); | |
713 | inode_sync_complete(inode); | |
714 | spin_unlock(&inode->i_lock); | |
715 | cond_resched_lock(&wb->list_lock); | |
716 | /* | |
717 | * bail out to wb_writeback() often enough to check | |
718 | * background threshold and other termination conditions. | |
719 | */ | |
720 | if (wrote) { | |
721 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
722 | break; | |
723 | if (work->nr_pages <= 0) | |
724 | break; | |
725 | } | |
726 | } | |
727 | return wrote; | |
728 | } | |
729 | ||
730 | static long __writeback_inodes_wb(struct bdi_writeback *wb, | |
731 | struct wb_writeback_work *work) | |
732 | { | |
733 | unsigned long start_time = jiffies; | |
734 | long wrote = 0; | |
735 | ||
736 | while (!list_empty(&wb->b_io)) { | |
737 | struct inode *inode = wb_inode(wb->b_io.prev); | |
738 | struct super_block *sb = inode->i_sb; | |
739 | ||
740 | if (!grab_super_passive(sb)) { | |
741 | /* | |
742 | * grab_super_passive() may fail consistently due to | |
743 | * s_umount being grabbed by someone else. Don't use | |
744 | * requeue_io() to avoid busy retrying the inode/sb. | |
745 | */ | |
746 | redirty_tail(inode, wb); | |
747 | continue; | |
748 | } | |
749 | wrote += writeback_sb_inodes(sb, wb, work); | |
750 | drop_super(sb); | |
751 | ||
752 | /* refer to the same tests at the end of writeback_sb_inodes */ | |
753 | if (wrote) { | |
754 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
755 | break; | |
756 | if (work->nr_pages <= 0) | |
757 | break; | |
758 | } | |
759 | } | |
760 | /* Leave any unwritten inodes on b_io */ | |
761 | return wrote; | |
762 | } | |
763 | ||
764 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, | |
765 | enum wb_reason reason) | |
766 | { | |
767 | struct wb_writeback_work work = { | |
768 | .nr_pages = nr_pages, | |
769 | .sync_mode = WB_SYNC_NONE, | |
770 | .range_cyclic = 1, | |
771 | .reason = reason, | |
772 | }; | |
773 | ||
774 | spin_lock(&wb->list_lock); | |
775 | if (list_empty(&wb->b_io)) | |
776 | queue_io(wb, &work); | |
777 | __writeback_inodes_wb(wb, &work); | |
778 | spin_unlock(&wb->list_lock); | |
779 | ||
780 | return nr_pages - work.nr_pages; | |
781 | } | |
782 | ||
783 | static bool over_bground_thresh(struct backing_dev_info *bdi) | |
784 | { | |
785 | unsigned long background_thresh, dirty_thresh; | |
786 | ||
787 | global_dirty_limits(&background_thresh, &dirty_thresh); | |
788 | ||
789 | if (global_page_state(NR_FILE_DIRTY) + | |
790 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
791 | return true; | |
792 | ||
793 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > | |
794 | bdi_dirty_limit(bdi, background_thresh)) | |
795 | return true; | |
796 | ||
797 | return false; | |
798 | } | |
799 | ||
800 | /* | |
801 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
802 | * only the flusher working on the first wb should do it. | |
803 | */ | |
804 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
805 | unsigned long start_time) | |
806 | { | |
807 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); | |
808 | } | |
809 | ||
810 | /* | |
811 | * Explicit flushing or periodic writeback of "old" data. | |
812 | * | |
813 | * Define "old": the first time one of an inode's pages is dirtied, we mark the | |
814 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
815 | * just walks the superblock inode list, writing back any inodes which are | |
816 | * older than a specific point in time. | |
817 | * | |
818 | * Try to run once per dirty_writeback_interval. But if a writeback event | |
819 | * takes longer than a dirty_writeback_interval interval, then leave a | |
820 | * one-second gap. | |
821 | * | |
822 | * older_than_this takes precedence over nr_to_write. So we'll only write back | |
823 | * all dirty pages if they are all attached to "old" mappings. | |
824 | */ | |
825 | static long wb_writeback(struct bdi_writeback *wb, | |
826 | struct wb_writeback_work *work) | |
827 | { | |
828 | unsigned long wb_start = jiffies; | |
829 | long nr_pages = work->nr_pages; | |
830 | unsigned long oldest_jif; | |
831 | struct inode *inode; | |
832 | long progress; | |
833 | ||
834 | oldest_jif = jiffies; | |
835 | work->older_than_this = &oldest_jif; | |
836 | ||
837 | spin_lock(&wb->list_lock); | |
838 | for (;;) { | |
839 | /* | |
840 | * Stop writeback when nr_pages has been consumed | |
841 | */ | |
842 | if (work->nr_pages <= 0) | |
843 | break; | |
844 | ||
845 | /* | |
846 | * Background writeout and kupdate-style writeback may | |
847 | * run forever. Stop them if there is other work to do | |
848 | * so that e.g. sync can proceed. They'll be restarted | |
849 | * after the other works are all done. | |
850 | */ | |
851 | if ((work->for_background || work->for_kupdate) && | |
852 | !list_empty(&wb->bdi->work_list)) | |
853 | break; | |
854 | ||
855 | /* | |
856 | * For background writeout, stop when we are below the | |
857 | * background dirty threshold | |
858 | */ | |
859 | if (work->for_background && !over_bground_thresh(wb->bdi)) | |
860 | break; | |
861 | ||
862 | /* | |
863 | * Kupdate and background works are special and we want to | |
864 | * include all inodes that need writing. Livelock avoidance is | |
865 | * handled by these works yielding to any other work so we are | |
866 | * safe. | |
867 | */ | |
868 | if (work->for_kupdate) { | |
869 | oldest_jif = jiffies - | |
870 | msecs_to_jiffies(dirty_expire_interval * 10); | |
871 | } else if (work->for_background) | |
872 | oldest_jif = jiffies; | |
873 | ||
874 | trace_writeback_start(wb->bdi, work); | |
875 | if (list_empty(&wb->b_io)) | |
876 | queue_io(wb, work); | |
877 | if (work->sb) | |
878 | progress = writeback_sb_inodes(work->sb, wb, work); | |
879 | else | |
880 | progress = __writeback_inodes_wb(wb, work); | |
881 | trace_writeback_written(wb->bdi, work); | |
882 | ||
883 | wb_update_bandwidth(wb, wb_start); | |
884 | ||
885 | /* | |
886 | * Did we write something? Try for more | |
887 | * | |
888 | * Dirty inodes are moved to b_io for writeback in batches. | |
889 | * The completion of the current batch does not necessarily | |
890 | * mean the overall work is done. So we keep looping as long | |
891 | * as made some progress on cleaning pages or inodes. | |
892 | */ | |
893 | if (progress) | |
894 | continue; | |
895 | /* | |
896 | * No more inodes for IO, bail | |
897 | */ | |
898 | if (list_empty(&wb->b_more_io)) | |
899 | break; | |
900 | /* | |
901 | * Nothing written. Wait for some inode to | |
902 | * become available for writeback. Otherwise | |
903 | * we'll just busyloop. | |
904 | */ | |
905 | if (!list_empty(&wb->b_more_io)) { | |
906 | trace_writeback_wait(wb->bdi, work); | |
907 | inode = wb_inode(wb->b_more_io.prev); | |
908 | spin_lock(&inode->i_lock); | |
909 | spin_unlock(&wb->list_lock); | |
910 | /* This function drops i_lock... */ | |
911 | inode_sleep_on_writeback(inode); | |
912 | spin_lock(&wb->list_lock); | |
913 | } | |
914 | } | |
915 | spin_unlock(&wb->list_lock); | |
916 | ||
917 | return nr_pages - work->nr_pages; | |
918 | } | |
919 | ||
920 | /* | |
921 | * Return the next wb_writeback_work struct that hasn't been processed yet. | |
922 | */ | |
923 | static struct wb_writeback_work * | |
924 | get_next_work_item(struct backing_dev_info *bdi) | |
925 | { | |
926 | struct wb_writeback_work *work = NULL; | |
927 | ||
928 | spin_lock_bh(&bdi->wb_lock); | |
929 | if (!list_empty(&bdi->work_list)) { | |
930 | work = list_entry(bdi->work_list.next, | |
931 | struct wb_writeback_work, list); | |
932 | list_del_init(&work->list); | |
933 | } | |
934 | spin_unlock_bh(&bdi->wb_lock); | |
935 | return work; | |
936 | } | |
937 | ||
938 | /* | |
939 | * Add in the number of potentially dirty inodes, because each inode | |
940 | * write can dirty pagecache in the underlying blockdev. | |
941 | */ | |
942 | static unsigned long get_nr_dirty_pages(void) | |
943 | { | |
944 | return global_page_state(NR_FILE_DIRTY) + | |
945 | global_page_state(NR_UNSTABLE_NFS) + | |
946 | get_nr_dirty_inodes(); | |
947 | } | |
948 | ||
949 | static long wb_check_background_flush(struct bdi_writeback *wb) | |
950 | { | |
951 | if (over_bground_thresh(wb->bdi)) { | |
952 | ||
953 | struct wb_writeback_work work = { | |
954 | .nr_pages = LONG_MAX, | |
955 | .sync_mode = WB_SYNC_NONE, | |
956 | .for_background = 1, | |
957 | .range_cyclic = 1, | |
958 | .reason = WB_REASON_BACKGROUND, | |
959 | }; | |
960 | ||
961 | return wb_writeback(wb, &work); | |
962 | } | |
963 | ||
964 | return 0; | |
965 | } | |
966 | ||
967 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
968 | { | |
969 | unsigned long expired; | |
970 | long nr_pages; | |
971 | ||
972 | /* | |
973 | * When set to zero, disable periodic writeback | |
974 | */ | |
975 | if (!dirty_writeback_interval) | |
976 | return 0; | |
977 | ||
978 | expired = wb->last_old_flush + | |
979 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
980 | if (time_before(jiffies, expired)) | |
981 | return 0; | |
982 | ||
983 | wb->last_old_flush = jiffies; | |
984 | nr_pages = get_nr_dirty_pages(); | |
985 | ||
986 | if (nr_pages) { | |
987 | struct wb_writeback_work work = { | |
988 | .nr_pages = nr_pages, | |
989 | .sync_mode = WB_SYNC_NONE, | |
990 | .for_kupdate = 1, | |
991 | .range_cyclic = 1, | |
992 | .reason = WB_REASON_PERIODIC, | |
993 | }; | |
994 | ||
995 | return wb_writeback(wb, &work); | |
996 | } | |
997 | ||
998 | return 0; | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * Retrieve work items and do the writeback they describe | |
1003 | */ | |
1004 | static long wb_do_writeback(struct bdi_writeback *wb) | |
1005 | { | |
1006 | struct backing_dev_info *bdi = wb->bdi; | |
1007 | struct wb_writeback_work *work; | |
1008 | long wrote = 0; | |
1009 | ||
1010 | set_bit(BDI_writeback_running, &wb->bdi->state); | |
1011 | while ((work = get_next_work_item(bdi)) != NULL) { | |
1012 | ||
1013 | trace_writeback_exec(bdi, work); | |
1014 | ||
1015 | wrote += wb_writeback(wb, work); | |
1016 | ||
1017 | /* | |
1018 | * Notify the caller of completion if this is a synchronous | |
1019 | * work item, otherwise just free it. | |
1020 | */ | |
1021 | if (work->done) | |
1022 | complete(work->done); | |
1023 | else | |
1024 | kfree(work); | |
1025 | } | |
1026 | ||
1027 | /* | |
1028 | * Check for periodic writeback, kupdated() style | |
1029 | */ | |
1030 | wrote += wb_check_old_data_flush(wb); | |
1031 | wrote += wb_check_background_flush(wb); | |
1032 | clear_bit(BDI_writeback_running, &wb->bdi->state); | |
1033 | ||
1034 | return wrote; | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
1039 | * reschedules periodically and does kupdated style flushing. | |
1040 | */ | |
1041 | void bdi_writeback_workfn(struct work_struct *work) | |
1042 | { | |
1043 | struct bdi_writeback *wb = container_of(to_delayed_work(work), | |
1044 | struct bdi_writeback, dwork); | |
1045 | struct backing_dev_info *bdi = wb->bdi; | |
1046 | long pages_written; | |
1047 | ||
1048 | set_worker_desc("flush-%s", dev_name(bdi->dev)); | |
1049 | current->flags |= PF_SWAPWRITE; | |
1050 | ||
1051 | if (likely(!current_is_workqueue_rescuer() || | |
1052 | !test_bit(BDI_registered, &bdi->state))) { | |
1053 | /* | |
1054 | * The normal path. Keep writing back @bdi until its | |
1055 | * work_list is empty. Note that this path is also taken | |
1056 | * if @bdi is shutting down even when we're running off the | |
1057 | * rescuer as work_list needs to be drained. | |
1058 | */ | |
1059 | do { | |
1060 | pages_written = wb_do_writeback(wb); | |
1061 | trace_writeback_pages_written(pages_written); | |
1062 | } while (!list_empty(&bdi->work_list)); | |
1063 | } else { | |
1064 | /* | |
1065 | * bdi_wq can't get enough workers and we're running off | |
1066 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1067 | * enough for efficient IO. | |
1068 | */ | |
1069 | pages_written = writeback_inodes_wb(&bdi->wb, 1024, | |
1070 | WB_REASON_FORKER_THREAD); | |
1071 | trace_writeback_pages_written(pages_written); | |
1072 | } | |
1073 | ||
1074 | if (!list_empty(&bdi->work_list)) | |
1075 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
1076 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
1077 | bdi_wakeup_thread_delayed(bdi); | |
1078 | ||
1079 | current->flags &= ~PF_SWAPWRITE; | |
1080 | } | |
1081 | ||
1082 | /* | |
1083 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back | |
1084 | * the whole world. | |
1085 | */ | |
1086 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) | |
1087 | { | |
1088 | struct backing_dev_info *bdi; | |
1089 | ||
1090 | if (!nr_pages) | |
1091 | nr_pages = get_nr_dirty_pages(); | |
1092 | ||
1093 | rcu_read_lock(); | |
1094 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
1095 | if (!bdi_has_dirty_io(bdi)) | |
1096 | continue; | |
1097 | __bdi_start_writeback(bdi, nr_pages, false, reason); | |
1098 | } | |
1099 | rcu_read_unlock(); | |
1100 | } | |
1101 | ||
1102 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
1103 | { | |
1104 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1105 | struct dentry *dentry; | |
1106 | const char *name = "?"; | |
1107 | ||
1108 | dentry = d_find_alias(inode); | |
1109 | if (dentry) { | |
1110 | spin_lock(&dentry->d_lock); | |
1111 | name = (const char *) dentry->d_name.name; | |
1112 | } | |
1113 | printk(KERN_DEBUG | |
1114 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1115 | current->comm, task_pid_nr(current), inode->i_ino, | |
1116 | name, inode->i_sb->s_id); | |
1117 | if (dentry) { | |
1118 | spin_unlock(&dentry->d_lock); | |
1119 | dput(dentry); | |
1120 | } | |
1121 | } | |
1122 | } | |
1123 | ||
1124 | /** | |
1125 | * __mark_inode_dirty - internal function | |
1126 | * @inode: inode to mark | |
1127 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1128 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1129 | * mark_inode_dirty_sync. | |
1130 | * | |
1131 | * Put the inode on the super block's dirty list. | |
1132 | * | |
1133 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1134 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1135 | * If it was not hashed, it will never be added to the dirty list | |
1136 | * even if it is later hashed, as it will have been marked dirty already. | |
1137 | * | |
1138 | * In short, make sure you hash any inodes _before_ you start marking | |
1139 | * them dirty. | |
1140 | * | |
1141 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of | |
1142 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1143 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1144 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1145 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1146 | * blockdev inode. | |
1147 | */ | |
1148 | void __mark_inode_dirty(struct inode *inode, int flags) | |
1149 | { | |
1150 | struct super_block *sb = inode->i_sb; | |
1151 | struct backing_dev_info *bdi = NULL; | |
1152 | ||
1153 | /* | |
1154 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1155 | * dirty the inode itself | |
1156 | */ | |
1157 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1158 | trace_writeback_dirty_inode_start(inode, flags); | |
1159 | ||
1160 | if (sb->s_op->dirty_inode) | |
1161 | sb->s_op->dirty_inode(inode, flags); | |
1162 | ||
1163 | trace_writeback_dirty_inode(inode, flags); | |
1164 | } | |
1165 | ||
1166 | /* | |
1167 | * Paired with smp_mb() in __writeback_single_inode() for the | |
1168 | * following lockless i_state test. See there for details. | |
1169 | */ | |
1170 | smp_mb(); | |
1171 | ||
1172 | if ((inode->i_state & flags) == flags) | |
1173 | return; | |
1174 | ||
1175 | if (unlikely(block_dump)) | |
1176 | block_dump___mark_inode_dirty(inode); | |
1177 | ||
1178 | spin_lock(&inode->i_lock); | |
1179 | if ((inode->i_state & flags) != flags) { | |
1180 | const int was_dirty = inode->i_state & I_DIRTY; | |
1181 | ||
1182 | inode->i_state |= flags; | |
1183 | ||
1184 | /* | |
1185 | * If the inode is being synced, just update its dirty state. | |
1186 | * The unlocker will place the inode on the appropriate | |
1187 | * superblock list, based upon its state. | |
1188 | */ | |
1189 | if (inode->i_state & I_SYNC) | |
1190 | goto out_unlock_inode; | |
1191 | ||
1192 | /* | |
1193 | * Only add valid (hashed) inodes to the superblock's | |
1194 | * dirty list. Add blockdev inodes as well. | |
1195 | */ | |
1196 | if (!S_ISBLK(inode->i_mode)) { | |
1197 | if (inode_unhashed(inode)) | |
1198 | goto out_unlock_inode; | |
1199 | } | |
1200 | if (inode->i_state & I_FREEING) | |
1201 | goto out_unlock_inode; | |
1202 | ||
1203 | /* | |
1204 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1205 | * reposition it (that would break b_dirty time-ordering). | |
1206 | */ | |
1207 | if (!was_dirty) { | |
1208 | bool wakeup_bdi = false; | |
1209 | bdi = inode_to_bdi(inode); | |
1210 | ||
1211 | spin_unlock(&inode->i_lock); | |
1212 | spin_lock(&bdi->wb.list_lock); | |
1213 | if (bdi_cap_writeback_dirty(bdi)) { | |
1214 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1215 | "bdi-%s not registered\n", bdi->name); | |
1216 | ||
1217 | /* | |
1218 | * If this is the first dirty inode for this | |
1219 | * bdi, we have to wake-up the corresponding | |
1220 | * bdi thread to make sure background | |
1221 | * write-back happens later. | |
1222 | */ | |
1223 | if (!wb_has_dirty_io(&bdi->wb)) | |
1224 | wakeup_bdi = true; | |
1225 | } | |
1226 | ||
1227 | inode->dirtied_when = jiffies; | |
1228 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); | |
1229 | spin_unlock(&bdi->wb.list_lock); | |
1230 | ||
1231 | if (wakeup_bdi) | |
1232 | bdi_wakeup_thread_delayed(bdi); | |
1233 | return; | |
1234 | } | |
1235 | } | |
1236 | out_unlock_inode: | |
1237 | spin_unlock(&inode->i_lock); | |
1238 | ||
1239 | } | |
1240 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1241 | ||
1242 | static void wait_sb_inodes(struct super_block *sb) | |
1243 | { | |
1244 | struct inode *inode, *old_inode = NULL; | |
1245 | ||
1246 | /* | |
1247 | * We need to be protected against the filesystem going from | |
1248 | * r/o to r/w or vice versa. | |
1249 | */ | |
1250 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
1251 | ||
1252 | spin_lock(&inode_sb_list_lock); | |
1253 | ||
1254 | /* | |
1255 | * Data integrity sync. Must wait for all pages under writeback, | |
1256 | * because there may have been pages dirtied before our sync | |
1257 | * call, but which had writeout started before we write it out. | |
1258 | * In which case, the inode may not be on the dirty list, but | |
1259 | * we still have to wait for that writeout. | |
1260 | */ | |
1261 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
1262 | struct address_space *mapping = inode->i_mapping; | |
1263 | ||
1264 | spin_lock(&inode->i_lock); | |
1265 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1266 | (mapping->nrpages == 0)) { | |
1267 | spin_unlock(&inode->i_lock); | |
1268 | continue; | |
1269 | } | |
1270 | __iget(inode); | |
1271 | spin_unlock(&inode->i_lock); | |
1272 | spin_unlock(&inode_sb_list_lock); | |
1273 | ||
1274 | /* | |
1275 | * We hold a reference to 'inode' so it couldn't have been | |
1276 | * removed from s_inodes list while we dropped the | |
1277 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1278 | * be holding the last reference and we cannot iput it under | |
1279 | * inode_sb_list_lock. So we keep the reference and iput it | |
1280 | * later. | |
1281 | */ | |
1282 | iput(old_inode); | |
1283 | old_inode = inode; | |
1284 | ||
1285 | filemap_fdatawait(mapping); | |
1286 | ||
1287 | cond_resched(); | |
1288 | ||
1289 | spin_lock(&inode_sb_list_lock); | |
1290 | } | |
1291 | spin_unlock(&inode_sb_list_lock); | |
1292 | iput(old_inode); | |
1293 | } | |
1294 | ||
1295 | /** | |
1296 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block | |
1297 | * @sb: the superblock | |
1298 | * @nr: the number of pages to write | |
1299 | * @reason: reason why some writeback work initiated | |
1300 | * | |
1301 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1302 | * on how many (if any) will be written, and this function does not wait | |
1303 | * for IO completion of submitted IO. | |
1304 | */ | |
1305 | void writeback_inodes_sb_nr(struct super_block *sb, | |
1306 | unsigned long nr, | |
1307 | enum wb_reason reason) | |
1308 | { | |
1309 | DECLARE_COMPLETION_ONSTACK(done); | |
1310 | struct wb_writeback_work work = { | |
1311 | .sb = sb, | |
1312 | .sync_mode = WB_SYNC_NONE, | |
1313 | .tagged_writepages = 1, | |
1314 | .done = &done, | |
1315 | .nr_pages = nr, | |
1316 | .reason = reason, | |
1317 | }; | |
1318 | ||
1319 | if (sb->s_bdi == &noop_backing_dev_info) | |
1320 | return; | |
1321 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
1322 | bdi_queue_work(sb->s_bdi, &work); | |
1323 | wait_for_completion(&done); | |
1324 | } | |
1325 | EXPORT_SYMBOL(writeback_inodes_sb_nr); | |
1326 | ||
1327 | /** | |
1328 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1329 | * @sb: the superblock | |
1330 | * @reason: reason why some writeback work was initiated | |
1331 | * | |
1332 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1333 | * on how many (if any) will be written, and this function does not wait | |
1334 | * for IO completion of submitted IO. | |
1335 | */ | |
1336 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) | |
1337 | { | |
1338 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); | |
1339 | } | |
1340 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1341 | ||
1342 | /** | |
1343 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway | |
1344 | * @sb: the superblock | |
1345 | * @nr: the number of pages to write | |
1346 | * @reason: the reason of writeback | |
1347 | * | |
1348 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. | |
1349 | * Returns 1 if writeback was started, 0 if not. | |
1350 | */ | |
1351 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, | |
1352 | unsigned long nr, | |
1353 | enum wb_reason reason) | |
1354 | { | |
1355 | if (writeback_in_progress(sb->s_bdi)) | |
1356 | return 1; | |
1357 | ||
1358 | if (!down_read_trylock(&sb->s_umount)) | |
1359 | return 0; | |
1360 | ||
1361 | writeback_inodes_sb_nr(sb, nr, reason); | |
1362 | up_read(&sb->s_umount); | |
1363 | return 1; | |
1364 | } | |
1365 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); | |
1366 | ||
1367 | /** | |
1368 | * try_to_writeback_inodes_sb - try to start writeback if none underway | |
1369 | * @sb: the superblock | |
1370 | * @reason: reason why some writeback work was initiated | |
1371 | * | |
1372 | * Implement by try_to_writeback_inodes_sb_nr() | |
1373 | * Returns 1 if writeback was started, 0 if not. | |
1374 | */ | |
1375 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) | |
1376 | { | |
1377 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); | |
1378 | } | |
1379 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); | |
1380 | ||
1381 | /** | |
1382 | * sync_inodes_sb - sync sb inode pages | |
1383 | * @sb: the superblock | |
1384 | * | |
1385 | * This function writes and waits on any dirty inode belonging to this | |
1386 | * super_block. | |
1387 | */ | |
1388 | void sync_inodes_sb(struct super_block *sb) | |
1389 | { | |
1390 | DECLARE_COMPLETION_ONSTACK(done); | |
1391 | struct wb_writeback_work work = { | |
1392 | .sb = sb, | |
1393 | .sync_mode = WB_SYNC_ALL, | |
1394 | .nr_pages = LONG_MAX, | |
1395 | .range_cyclic = 0, | |
1396 | .done = &done, | |
1397 | .reason = WB_REASON_SYNC, | |
1398 | .for_sync = 1, | |
1399 | }; | |
1400 | ||
1401 | /* Nothing to do? */ | |
1402 | if (sb->s_bdi == &noop_backing_dev_info) | |
1403 | return; | |
1404 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
1405 | ||
1406 | bdi_queue_work(sb->s_bdi, &work); | |
1407 | wait_for_completion(&done); | |
1408 | ||
1409 | wait_sb_inodes(sb); | |
1410 | } | |
1411 | EXPORT_SYMBOL(sync_inodes_sb); | |
1412 | ||
1413 | /** | |
1414 | * write_inode_now - write an inode to disk | |
1415 | * @inode: inode to write to disk | |
1416 | * @sync: whether the write should be synchronous or not | |
1417 | * | |
1418 | * This function commits an inode to disk immediately if it is dirty. This is | |
1419 | * primarily needed by knfsd. | |
1420 | * | |
1421 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. | |
1422 | */ | |
1423 | int write_inode_now(struct inode *inode, int sync) | |
1424 | { | |
1425 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
1426 | struct writeback_control wbc = { | |
1427 | .nr_to_write = LONG_MAX, | |
1428 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1429 | .range_start = 0, | |
1430 | .range_end = LLONG_MAX, | |
1431 | }; | |
1432 | ||
1433 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
1434 | wbc.nr_to_write = 0; | |
1435 | ||
1436 | might_sleep(); | |
1437 | return writeback_single_inode(inode, wb, &wbc); | |
1438 | } | |
1439 | EXPORT_SYMBOL(write_inode_now); | |
1440 | ||
1441 | /** | |
1442 | * sync_inode - write an inode and its pages to disk. | |
1443 | * @inode: the inode to sync | |
1444 | * @wbc: controls the writeback mode | |
1445 | * | |
1446 | * sync_inode() will write an inode and its pages to disk. It will also | |
1447 | * correctly update the inode on its superblock's dirty inode lists and will | |
1448 | * update inode->i_state. | |
1449 | * | |
1450 | * The caller must have a ref on the inode. | |
1451 | */ | |
1452 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1453 | { | |
1454 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); | |
1455 | } | |
1456 | EXPORT_SYMBOL(sync_inode); | |
1457 | ||
1458 | /** | |
1459 | * sync_inode_metadata - write an inode to disk | |
1460 | * @inode: the inode to sync | |
1461 | * @wait: wait for I/O to complete. | |
1462 | * | |
1463 | * Write an inode to disk and adjust its dirty state after completion. | |
1464 | * | |
1465 | * Note: only writes the actual inode, no associated data or other metadata. | |
1466 | */ | |
1467 | int sync_inode_metadata(struct inode *inode, int wait) | |
1468 | { | |
1469 | struct writeback_control wbc = { | |
1470 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1471 | .nr_to_write = 0, /* metadata-only */ | |
1472 | }; | |
1473 | ||
1474 | return sync_inode(inode, &wbc); | |
1475 | } | |
1476 | EXPORT_SYMBOL(sync_inode_metadata); |