Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
e1f8e874 | 11 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
12 | * Split out of fs/inode.c |
13 | * Additions for address_space-based writeback | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
630d9c47 | 17 | #include <linux/export.h> |
1da177e4 | 18 | #include <linux/spinlock.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
1da177e4 LT |
20 | #include <linux/sched.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/mm.h> | |
bc31b86a | 23 | #include <linux/pagemap.h> |
03ba3782 | 24 | #include <linux/kthread.h> |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
455b2864 | 28 | #include <linux/tracepoint.h> |
719ea2fb | 29 | #include <linux/device.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 | 31 | |
bc31b86a WF |
32 | /* |
33 | * 4MB minimal write chunk size | |
34 | */ | |
35 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) | |
36 | ||
c4a77a6c JA |
37 | /* |
38 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
39 | */ | |
83ba7b07 | 40 | struct wb_writeback_work { |
c4a77a6c JA |
41 | long nr_pages; |
42 | struct super_block *sb; | |
0dc83bd3 | 43 | unsigned long *older_than_this; |
c4a77a6c | 44 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 45 | unsigned int tagged_writepages:1; |
52957fe1 HS |
46 | unsigned int for_kupdate:1; |
47 | unsigned int range_cyclic:1; | |
48 | unsigned int for_background:1; | |
7747bd4b | 49 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
0e175a18 | 50 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 51 | |
8010c3b6 | 52 | struct list_head list; /* pending work list */ |
83ba7b07 | 53 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
54 | }; |
55 | ||
f11b00f3 AB |
56 | /** |
57 | * writeback_in_progress - determine whether there is writeback in progress | |
58 | * @bdi: the device's backing_dev_info structure. | |
59 | * | |
03ba3782 JA |
60 | * Determine whether there is writeback waiting to be handled against a |
61 | * backing device. | |
f11b00f3 AB |
62 | */ |
63 | int writeback_in_progress(struct backing_dev_info *bdi) | |
64 | { | |
81d73a32 | 65 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 | 66 | } |
00d4e736 | 67 | EXPORT_SYMBOL(writeback_in_progress); |
f11b00f3 | 68 | |
de1414a6 | 69 | struct backing_dev_info *inode_to_bdi(struct inode *inode) |
692ebd17 JK |
70 | { |
71 | struct super_block *sb = inode->i_sb; | |
495a276e | 72 | #ifdef CONFIG_BLOCK |
a8855990 | 73 | if (sb_is_blkdev_sb(sb)) |
495a276e CH |
74 | return blk_get_backing_dev_info(I_BDEV(inode)); |
75 | #endif | |
aaead25b | 76 | return sb->s_bdi; |
692ebd17 | 77 | } |
de1414a6 | 78 | EXPORT_SYMBOL_GPL(inode_to_bdi); |
692ebd17 | 79 | |
7ccf19a8 NP |
80 | static inline struct inode *wb_inode(struct list_head *head) |
81 | { | |
82 | return list_entry(head, struct inode, i_wb_list); | |
83 | } | |
84 | ||
15eb77a0 WF |
85 | /* |
86 | * Include the creation of the trace points after defining the | |
87 | * wb_writeback_work structure and inline functions so that the definition | |
88 | * remains local to this file. | |
89 | */ | |
90 | #define CREATE_TRACE_POINTS | |
91 | #include <trace/events/writeback.h> | |
92 | ||
774016b2 SW |
93 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
94 | ||
5acda9d1 JK |
95 | static void bdi_wakeup_thread(struct backing_dev_info *bdi) |
96 | { | |
97 | spin_lock_bh(&bdi->wb_lock); | |
98 | if (test_bit(BDI_registered, &bdi->state)) | |
99 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
100 | spin_unlock_bh(&bdi->wb_lock); | |
101 | } | |
102 | ||
6585027a JK |
103 | static void bdi_queue_work(struct backing_dev_info *bdi, |
104 | struct wb_writeback_work *work) | |
105 | { | |
106 | trace_writeback_queue(bdi, work); | |
107 | ||
108 | spin_lock_bh(&bdi->wb_lock); | |
5acda9d1 JK |
109 | if (!test_bit(BDI_registered, &bdi->state)) { |
110 | if (work->done) | |
111 | complete(work->done); | |
112 | goto out_unlock; | |
113 | } | |
6585027a | 114 | list_add_tail(&work->list, &bdi->work_list); |
839a8e86 | 115 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
5acda9d1 JK |
116 | out_unlock: |
117 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
118 | } |
119 | ||
83ba7b07 CH |
120 | static void |
121 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
0e175a18 | 122 | bool range_cyclic, enum wb_reason reason) |
1da177e4 | 123 | { |
83ba7b07 | 124 | struct wb_writeback_work *work; |
03ba3782 | 125 | |
bcddc3f0 JA |
126 | /* |
127 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
128 | * wakeup the thread for old dirty data writeback | |
129 | */ | |
83ba7b07 CH |
130 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
131 | if (!work) { | |
839a8e86 | 132 | trace_writeback_nowork(bdi); |
5acda9d1 | 133 | bdi_wakeup_thread(bdi); |
83ba7b07 | 134 | return; |
bcddc3f0 | 135 | } |
03ba3782 | 136 | |
83ba7b07 CH |
137 | work->sync_mode = WB_SYNC_NONE; |
138 | work->nr_pages = nr_pages; | |
139 | work->range_cyclic = range_cyclic; | |
0e175a18 | 140 | work->reason = reason; |
03ba3782 | 141 | |
83ba7b07 | 142 | bdi_queue_work(bdi, work); |
b6e51316 JA |
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 | |
786228ab | 149 | * @reason: reason why some writeback work was initiated |
b6e51316 JA |
150 | * |
151 | * Description: | |
152 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 153 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 154 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
155 | * |
156 | */ | |
0e175a18 CW |
157 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
158 | enum wb_reason reason) | |
b6e51316 | 159 | { |
0e175a18 | 160 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
c5444198 | 161 | } |
d3ddec76 | 162 | |
c5444198 CH |
163 | /** |
164 | * bdi_start_background_writeback - start background writeback | |
165 | * @bdi: the backing device to write from | |
166 | * | |
167 | * Description: | |
6585027a JK |
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. | |
c5444198 CH |
172 | */ |
173 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
174 | { | |
6585027a JK |
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 | */ | |
71927e84 | 179 | trace_writeback_wake_background(bdi); |
5acda9d1 | 180 | bdi_wakeup_thread(bdi); |
1da177e4 LT |
181 | } |
182 | ||
a66979ab DC |
183 | /* |
184 | * Remove the inode from the writeback list it is on. | |
185 | */ | |
186 | void inode_wb_list_del(struct inode *inode) | |
187 | { | |
f758eeab CH |
188 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
189 | ||
190 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 191 | list_del_init(&inode->i_wb_list); |
f758eeab | 192 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
193 | } |
194 | ||
6610a0bc AM |
195 | /* |
196 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
197 | * furthest end of its superblock's dirty-inode list. | |
198 | * | |
199 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 200 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
201 | * the case then the inode must have been redirtied while it was being written |
202 | * out and we don't reset its dirtied_when. | |
203 | */ | |
f758eeab | 204 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 205 | { |
f758eeab | 206 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 207 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 208 | struct inode *tail; |
6610a0bc | 209 | |
7ccf19a8 | 210 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 211 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
212 | inode->dirtied_when = jiffies; |
213 | } | |
7ccf19a8 | 214 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
215 | } |
216 | ||
c986d1e2 | 217 | /* |
66f3b8e2 | 218 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 219 | */ |
f758eeab | 220 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 221 | { |
f758eeab | 222 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 223 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
224 | } |
225 | ||
1c0eeaf5 JE |
226 | static void inode_sync_complete(struct inode *inode) |
227 | { | |
365b94ae | 228 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
229 | /* If inode is clean an unused, put it into LRU now... */ |
230 | inode_add_lru(inode); | |
365b94ae | 231 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
232 | smp_mb(); |
233 | wake_up_bit(&inode->i_state, __I_SYNC); | |
234 | } | |
235 | ||
d2caa3c5 JL |
236 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
237 | { | |
238 | bool ret = time_after(inode->dirtied_when, t); | |
239 | #ifndef CONFIG_64BIT | |
240 | /* | |
241 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
242 | * It _appears_ to be in the future, but is actually in distant past. | |
243 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 244 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
245 | */ |
246 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
247 | #endif | |
248 | return ret; | |
249 | } | |
250 | ||
2c136579 | 251 | /* |
0e2f2b23 | 252 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 253 | * @delaying_queue to @dispatch_queue. |
2c136579 | 254 | */ |
e84d0a4f | 255 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 256 | struct list_head *dispatch_queue, |
ad4e38dd | 257 | struct wb_writeback_work *work) |
2c136579 | 258 | { |
5c03449d SL |
259 | LIST_HEAD(tmp); |
260 | struct list_head *pos, *node; | |
cf137307 | 261 | struct super_block *sb = NULL; |
5c03449d | 262 | struct inode *inode; |
cf137307 | 263 | int do_sb_sort = 0; |
e84d0a4f | 264 | int moved = 0; |
5c03449d | 265 | |
2c136579 | 266 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 267 | inode = wb_inode(delaying_queue->prev); |
0dc83bd3 JK |
268 | if (work->older_than_this && |
269 | inode_dirtied_after(inode, *work->older_than_this)) | |
2c136579 | 270 | break; |
a8855990 JK |
271 | list_move(&inode->i_wb_list, &tmp); |
272 | moved++; | |
273 | if (sb_is_blkdev_sb(inode->i_sb)) | |
274 | continue; | |
cf137307 JA |
275 | if (sb && sb != inode->i_sb) |
276 | do_sb_sort = 1; | |
277 | sb = inode->i_sb; | |
5c03449d SL |
278 | } |
279 | ||
cf137307 JA |
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); | |
e84d0a4f | 283 | goto out; |
cf137307 JA |
284 | } |
285 | ||
5c03449d SL |
286 | /* Move inodes from one superblock together */ |
287 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 288 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 289 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 290 | inode = wb_inode(pos); |
5c03449d | 291 | if (inode->i_sb == sb) |
7ccf19a8 | 292 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 293 | } |
2c136579 | 294 | } |
e84d0a4f WF |
295 | out: |
296 | return moved; | |
2c136579 FW |
297 | } |
298 | ||
299 | /* | |
300 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
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 | |
2c136579 | 309 | */ |
ad4e38dd | 310 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 311 | { |
e84d0a4f | 312 | int moved; |
f758eeab | 313 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 314 | list_splice_init(&wb->b_more_io, &wb->b_io); |
ad4e38dd CW |
315 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
316 | trace_writeback_queue_io(wb, work, moved); | |
66f3b8e2 JA |
317 | } |
318 | ||
a9185b41 | 319 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 320 | { |
9fb0a7da TH |
321 | int ret; |
322 | ||
323 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
324 | trace_writeback_write_inode_start(inode, wbc); | |
325 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
326 | trace_writeback_write_inode(inode, wbc); | |
327 | return ret; | |
328 | } | |
03ba3782 | 329 | return 0; |
08d8e974 | 330 | } |
08d8e974 | 331 | |
1da177e4 | 332 | /* |
169ebd90 JK |
333 | * Wait for writeback on an inode to complete. Called with i_lock held. |
334 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 335 | */ |
169ebd90 JK |
336 | static void __inode_wait_for_writeback(struct inode *inode) |
337 | __releases(inode->i_lock) | |
338 | __acquires(inode->i_lock) | |
01c03194 CH |
339 | { |
340 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
341 | wait_queue_head_t *wqh; | |
342 | ||
343 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
344 | while (inode->i_state & I_SYNC) { |
345 | spin_unlock(&inode->i_lock); | |
74316201 N |
346 | __wait_on_bit(wqh, &wq, bit_wait, |
347 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 348 | spin_lock(&inode->i_lock); |
58a9d3d8 | 349 | } |
01c03194 CH |
350 | } |
351 | ||
169ebd90 JK |
352 | /* |
353 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
354 | */ | |
355 | void inode_wait_for_writeback(struct inode *inode) | |
356 | { | |
357 | spin_lock(&inode->i_lock); | |
358 | __inode_wait_for_writeback(inode); | |
359 | spin_unlock(&inode->i_lock); | |
360 | } | |
361 | ||
362 | /* | |
363 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
364 | * held and drops it. It is aimed for callers not holding any inode reference | |
365 | * so once i_lock is dropped, inode can go away. | |
366 | */ | |
367 | static void inode_sleep_on_writeback(struct inode *inode) | |
368 | __releases(inode->i_lock) | |
369 | { | |
370 | DEFINE_WAIT(wait); | |
371 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
372 | int sleep; | |
373 | ||
374 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
375 | sleep = inode->i_state & I_SYNC; | |
376 | spin_unlock(&inode->i_lock); | |
377 | if (sleep) | |
378 | schedule(); | |
379 | finish_wait(wqh, &wait); | |
380 | } | |
381 | ||
ccb26b5a JK |
382 | /* |
383 | * Find proper writeback list for the inode depending on its current state and | |
384 | * possibly also change of its state while we were doing writeback. Here we | |
385 | * handle things such as livelock prevention or fairness of writeback among | |
386 | * inodes. This function can be called only by flusher thread - noone else | |
387 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
388 | * thread's back can have unexpected consequences. | |
389 | */ | |
390 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
391 | struct writeback_control *wbc) | |
392 | { | |
393 | if (inode->i_state & I_FREEING) | |
394 | return; | |
395 | ||
396 | /* | |
397 | * Sync livelock prevention. Each inode is tagged and synced in one | |
398 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
399 | * the dirty time to prevent enqueue and sync it again. | |
400 | */ | |
401 | if ((inode->i_state & I_DIRTY) && | |
402 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
403 | inode->dirtied_when = jiffies; | |
404 | ||
4f8ad655 JK |
405 | if (wbc->pages_skipped) { |
406 | /* | |
407 | * writeback is not making progress due to locked | |
408 | * buffers. Skip this inode for now. | |
409 | */ | |
410 | redirty_tail(inode, wb); | |
411 | return; | |
412 | } | |
413 | ||
ccb26b5a JK |
414 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
415 | /* | |
416 | * We didn't write back all the pages. nfs_writepages() | |
417 | * sometimes bales out without doing anything. | |
418 | */ | |
419 | if (wbc->nr_to_write <= 0) { | |
420 | /* Slice used up. Queue for next turn. */ | |
421 | requeue_io(inode, wb); | |
422 | } else { | |
423 | /* | |
424 | * Writeback blocked by something other than | |
425 | * congestion. Delay the inode for some time to | |
426 | * avoid spinning on the CPU (100% iowait) | |
427 | * retrying writeback of the dirty page/inode | |
428 | * that cannot be performed immediately. | |
429 | */ | |
430 | redirty_tail(inode, wb); | |
431 | } | |
432 | } else if (inode->i_state & I_DIRTY) { | |
433 | /* | |
434 | * Filesystems can dirty the inode during writeback operations, | |
435 | * such as delayed allocation during submission or metadata | |
436 | * updates after data IO completion. | |
437 | */ | |
438 | redirty_tail(inode, wb); | |
439 | } else { | |
440 | /* The inode is clean. Remove from writeback lists. */ | |
441 | list_del_init(&inode->i_wb_list); | |
442 | } | |
443 | } | |
444 | ||
01c03194 | 445 | /* |
4f8ad655 JK |
446 | * Write out an inode and its dirty pages. Do not update the writeback list |
447 | * linkage. That is left to the caller. The caller is also responsible for | |
448 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
449 | */ |
450 | static int | |
cd8ed2a4 | 451 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 452 | { |
1da177e4 | 453 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 454 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 455 | unsigned dirty; |
1da177e4 LT |
456 | int ret; |
457 | ||
4f8ad655 | 458 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 459 | |
9fb0a7da TH |
460 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
461 | ||
1da177e4 LT |
462 | ret = do_writepages(mapping, wbc); |
463 | ||
26821ed4 CH |
464 | /* |
465 | * Make sure to wait on the data before writing out the metadata. | |
466 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
467 | * I/O completion. We don't do it for sync(2) writeback because it has a |
468 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
469 | * inode metadata is written back correctly. | |
26821ed4 | 470 | */ |
7747bd4b | 471 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 472 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
473 | if (ret == 0) |
474 | ret = err; | |
475 | } | |
476 | ||
5547e8aa DM |
477 | /* |
478 | * Some filesystems may redirty the inode during the writeback | |
479 | * due to delalloc, clear dirty metadata flags right before | |
480 | * write_inode() | |
481 | */ | |
250df6ed | 482 | spin_lock(&inode->i_lock); |
9c6ac78e | 483 | |
5547e8aa | 484 | dirty = inode->i_state & I_DIRTY; |
9c6ac78e TH |
485 | inode->i_state &= ~I_DIRTY; |
486 | ||
487 | /* | |
488 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
489 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
490 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
491 | * inode. | |
492 | * | |
493 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
494 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
495 | * necessary. This guarantees that either __mark_inode_dirty() | |
496 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
497 | */ | |
498 | smp_mb(); | |
499 | ||
500 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
501 | inode->i_state |= I_DIRTY_PAGES; | |
502 | ||
250df6ed | 503 | spin_unlock(&inode->i_lock); |
9c6ac78e | 504 | |
26821ed4 CH |
505 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
506 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 507 | int err = write_inode(inode, wbc); |
1da177e4 LT |
508 | if (ret == 0) |
509 | ret = err; | |
510 | } | |
4f8ad655 JK |
511 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
512 | return ret; | |
513 | } | |
514 | ||
515 | /* | |
516 | * Write out an inode's dirty pages. Either the caller has an active reference | |
517 | * on the inode or the inode has I_WILL_FREE set. | |
518 | * | |
519 | * This function is designed to be called for writing back one inode which | |
520 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
521 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
522 | */ | |
523 | static int | |
524 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
525 | struct writeback_control *wbc) | |
526 | { | |
527 | int ret = 0; | |
528 | ||
529 | spin_lock(&inode->i_lock); | |
530 | if (!atomic_read(&inode->i_count)) | |
531 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
532 | else | |
533 | WARN_ON(inode->i_state & I_WILL_FREE); | |
534 | ||
535 | if (inode->i_state & I_SYNC) { | |
536 | if (wbc->sync_mode != WB_SYNC_ALL) | |
537 | goto out; | |
538 | /* | |
169ebd90 JK |
539 | * It's a data-integrity sync. We must wait. Since callers hold |
540 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
541 | * away under us. | |
4f8ad655 | 542 | */ |
169ebd90 | 543 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
544 | } |
545 | WARN_ON(inode->i_state & I_SYNC); | |
546 | /* | |
f9b0e058 JK |
547 | * Skip inode if it is clean and we have no outstanding writeback in |
548 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
549 | * function since flusher thread may be doing for example sync in | |
550 | * parallel and if we move the inode, it could get skipped. So here we | |
551 | * make sure inode is on some writeback list and leave it there unless | |
552 | * we have completely cleaned the inode. | |
4f8ad655 | 553 | */ |
f9b0e058 JK |
554 | if (!(inode->i_state & I_DIRTY) && |
555 | (wbc->sync_mode != WB_SYNC_ALL || | |
556 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
557 | goto out; |
558 | inode->i_state |= I_SYNC; | |
559 | spin_unlock(&inode->i_lock); | |
560 | ||
cd8ed2a4 | 561 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 562 | |
f758eeab | 563 | spin_lock(&wb->list_lock); |
250df6ed | 564 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
565 | /* |
566 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
567 | * touch it. See comment above for explanation. | |
568 | */ | |
569 | if (!(inode->i_state & I_DIRTY)) | |
570 | list_del_init(&inode->i_wb_list); | |
571 | spin_unlock(&wb->list_lock); | |
1c0eeaf5 | 572 | inode_sync_complete(inode); |
4f8ad655 JK |
573 | out: |
574 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
575 | return ret; |
576 | } | |
577 | ||
1a12d8bd WF |
578 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
579 | struct wb_writeback_work *work) | |
d46db3d5 WF |
580 | { |
581 | long pages; | |
582 | ||
583 | /* | |
584 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
585 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
586 | * here avoids calling into writeback_inodes_wb() more than once. | |
587 | * | |
588 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
589 | * | |
590 | * wb_writeback() | |
591 | * writeback_sb_inodes() <== called only once | |
592 | * write_cache_pages() <== called once for each inode | |
593 | * (quickly) tag currently dirty pages | |
594 | * (maybe slowly) sync all tagged pages | |
595 | */ | |
596 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
597 | pages = LONG_MAX; | |
1a12d8bd WF |
598 | else { |
599 | pages = min(bdi->avg_write_bandwidth / 2, | |
600 | global_dirty_limit / DIRTY_SCOPE); | |
601 | pages = min(pages, work->nr_pages); | |
602 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
603 | MIN_WRITEBACK_PAGES); | |
604 | } | |
d46db3d5 WF |
605 | |
606 | return pages; | |
607 | } | |
608 | ||
f11c9c5c ES |
609 | /* |
610 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 611 | * |
d46db3d5 | 612 | * Return the number of pages and/or inodes written. |
f11c9c5c | 613 | */ |
d46db3d5 WF |
614 | static long writeback_sb_inodes(struct super_block *sb, |
615 | struct bdi_writeback *wb, | |
616 | struct wb_writeback_work *work) | |
1da177e4 | 617 | { |
d46db3d5 WF |
618 | struct writeback_control wbc = { |
619 | .sync_mode = work->sync_mode, | |
620 | .tagged_writepages = work->tagged_writepages, | |
621 | .for_kupdate = work->for_kupdate, | |
622 | .for_background = work->for_background, | |
7747bd4b | 623 | .for_sync = work->for_sync, |
d46db3d5 WF |
624 | .range_cyclic = work->range_cyclic, |
625 | .range_start = 0, | |
626 | .range_end = LLONG_MAX, | |
627 | }; | |
628 | unsigned long start_time = jiffies; | |
629 | long write_chunk; | |
630 | long wrote = 0; /* count both pages and inodes */ | |
631 | ||
03ba3782 | 632 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 633 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
634 | |
635 | if (inode->i_sb != sb) { | |
d46db3d5 | 636 | if (work->sb) { |
edadfb10 CH |
637 | /* |
638 | * We only want to write back data for this | |
639 | * superblock, move all inodes not belonging | |
640 | * to it back onto the dirty list. | |
641 | */ | |
f758eeab | 642 | redirty_tail(inode, wb); |
edadfb10 CH |
643 | continue; |
644 | } | |
645 | ||
646 | /* | |
647 | * The inode belongs to a different superblock. | |
648 | * Bounce back to the caller to unpin this and | |
649 | * pin the next superblock. | |
650 | */ | |
d46db3d5 | 651 | break; |
edadfb10 CH |
652 | } |
653 | ||
9843b76a | 654 | /* |
331cbdee WL |
655 | * Don't bother with new inodes or inodes being freed, first |
656 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
657 | * kind writeout is handled by the freer. |
658 | */ | |
250df6ed | 659 | spin_lock(&inode->i_lock); |
9843b76a | 660 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 661 | spin_unlock(&inode->i_lock); |
fcc5c222 | 662 | redirty_tail(inode, wb); |
7ef0d737 NP |
663 | continue; |
664 | } | |
cc1676d9 JK |
665 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
666 | /* | |
667 | * If this inode is locked for writeback and we are not | |
668 | * doing writeback-for-data-integrity, move it to | |
669 | * b_more_io so that writeback can proceed with the | |
670 | * other inodes on s_io. | |
671 | * | |
672 | * We'll have another go at writing back this inode | |
673 | * when we completed a full scan of b_io. | |
674 | */ | |
675 | spin_unlock(&inode->i_lock); | |
676 | requeue_io(inode, wb); | |
677 | trace_writeback_sb_inodes_requeue(inode); | |
678 | continue; | |
679 | } | |
f0d07b7f JK |
680 | spin_unlock(&wb->list_lock); |
681 | ||
4f8ad655 JK |
682 | /* |
683 | * We already requeued the inode if it had I_SYNC set and we | |
684 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
685 | * WB_SYNC_ALL case. | |
686 | */ | |
169ebd90 JK |
687 | if (inode->i_state & I_SYNC) { |
688 | /* Wait for I_SYNC. This function drops i_lock... */ | |
689 | inode_sleep_on_writeback(inode); | |
690 | /* Inode may be gone, start again */ | |
ead188f9 | 691 | spin_lock(&wb->list_lock); |
169ebd90 JK |
692 | continue; |
693 | } | |
4f8ad655 JK |
694 | inode->i_state |= I_SYNC; |
695 | spin_unlock(&inode->i_lock); | |
169ebd90 | 696 | |
1a12d8bd | 697 | write_chunk = writeback_chunk_size(wb->bdi, work); |
d46db3d5 WF |
698 | wbc.nr_to_write = write_chunk; |
699 | wbc.pages_skipped = 0; | |
250df6ed | 700 | |
169ebd90 JK |
701 | /* |
702 | * We use I_SYNC to pin the inode in memory. While it is set | |
703 | * evict_inode() will wait so the inode cannot be freed. | |
704 | */ | |
cd8ed2a4 | 705 | __writeback_single_inode(inode, &wbc); |
250df6ed | 706 | |
d46db3d5 WF |
707 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
708 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
709 | spin_lock(&wb->list_lock); |
710 | spin_lock(&inode->i_lock); | |
d46db3d5 WF |
711 | if (!(inode->i_state & I_DIRTY)) |
712 | wrote++; | |
4f8ad655 JK |
713 | requeue_inode(inode, wb, &wbc); |
714 | inode_sync_complete(inode); | |
0f1b1fd8 | 715 | spin_unlock(&inode->i_lock); |
169ebd90 | 716 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
717 | /* |
718 | * bail out to wb_writeback() often enough to check | |
719 | * background threshold and other termination conditions. | |
720 | */ | |
721 | if (wrote) { | |
722 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
723 | break; | |
724 | if (work->nr_pages <= 0) | |
725 | break; | |
8bc3be27 | 726 | } |
1da177e4 | 727 | } |
d46db3d5 | 728 | return wrote; |
f11c9c5c ES |
729 | } |
730 | ||
d46db3d5 WF |
731 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
732 | struct wb_writeback_work *work) | |
f11c9c5c | 733 | { |
d46db3d5 WF |
734 | unsigned long start_time = jiffies; |
735 | long wrote = 0; | |
38f21977 | 736 | |
f11c9c5c | 737 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 738 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 739 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 740 | |
12ad3ab6 | 741 | if (!grab_super_passive(sb)) { |
0e995816 WF |
742 | /* |
743 | * grab_super_passive() may fail consistently due to | |
744 | * s_umount being grabbed by someone else. Don't use | |
745 | * requeue_io() to avoid busy retrying the inode/sb. | |
746 | */ | |
747 | redirty_tail(inode, wb); | |
edadfb10 | 748 | continue; |
f11c9c5c | 749 | } |
d46db3d5 | 750 | wrote += writeback_sb_inodes(sb, wb, work); |
edadfb10 | 751 | drop_super(sb); |
f11c9c5c | 752 | |
d46db3d5 WF |
753 | /* refer to the same tests at the end of writeback_sb_inodes */ |
754 | if (wrote) { | |
755 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
756 | break; | |
757 | if (work->nr_pages <= 0) | |
758 | break; | |
759 | } | |
f11c9c5c | 760 | } |
66f3b8e2 | 761 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 762 | return wrote; |
66f3b8e2 JA |
763 | } |
764 | ||
7d9f073b | 765 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 766 | enum wb_reason reason) |
edadfb10 | 767 | { |
d46db3d5 WF |
768 | struct wb_writeback_work work = { |
769 | .nr_pages = nr_pages, | |
770 | .sync_mode = WB_SYNC_NONE, | |
771 | .range_cyclic = 1, | |
0e175a18 | 772 | .reason = reason, |
d46db3d5 | 773 | }; |
edadfb10 | 774 | |
f758eeab | 775 | spin_lock(&wb->list_lock); |
424b351f | 776 | if (list_empty(&wb->b_io)) |
ad4e38dd | 777 | queue_io(wb, &work); |
d46db3d5 | 778 | __writeback_inodes_wb(wb, &work); |
f758eeab | 779 | spin_unlock(&wb->list_lock); |
edadfb10 | 780 | |
d46db3d5 WF |
781 | return nr_pages - work.nr_pages; |
782 | } | |
03ba3782 | 783 | |
b00949aa | 784 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
03ba3782 JA |
785 | { |
786 | unsigned long background_thresh, dirty_thresh; | |
787 | ||
16c4042f | 788 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 789 | |
b00949aa WF |
790 | if (global_page_state(NR_FILE_DIRTY) + |
791 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
792 | return true; | |
793 | ||
794 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > | |
795 | bdi_dirty_limit(bdi, background_thresh)) | |
796 | return true; | |
797 | ||
798 | return false; | |
03ba3782 JA |
799 | } |
800 | ||
e98be2d5 WF |
801 | /* |
802 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
803 | * only the flusher working on the first wb should do it. | |
804 | */ | |
805 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
806 | unsigned long start_time) | |
807 | { | |
af6a3113 | 808 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
809 | } |
810 | ||
03ba3782 JA |
811 | /* |
812 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 813 | * |
03ba3782 JA |
814 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
815 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
816 | * just walks the superblock inode list, writing back any inodes which are | |
817 | * older than a specific point in time. | |
66f3b8e2 | 818 | * |
03ba3782 JA |
819 | * Try to run once per dirty_writeback_interval. But if a writeback event |
820 | * takes longer than a dirty_writeback_interval interval, then leave a | |
821 | * one-second gap. | |
66f3b8e2 | 822 | * |
03ba3782 JA |
823 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
824 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 825 | */ |
c4a77a6c | 826 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 827 | struct wb_writeback_work *work) |
66f3b8e2 | 828 | { |
e98be2d5 | 829 | unsigned long wb_start = jiffies; |
d46db3d5 | 830 | long nr_pages = work->nr_pages; |
0dc83bd3 | 831 | unsigned long oldest_jif; |
a5989bdc | 832 | struct inode *inode; |
d46db3d5 | 833 | long progress; |
66f3b8e2 | 834 | |
0dc83bd3 JK |
835 | oldest_jif = jiffies; |
836 | work->older_than_this = &oldest_jif; | |
38f21977 | 837 | |
e8dfc305 | 838 | spin_lock(&wb->list_lock); |
03ba3782 JA |
839 | for (;;) { |
840 | /* | |
d3ddec76 | 841 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 842 | */ |
83ba7b07 | 843 | if (work->nr_pages <= 0) |
03ba3782 | 844 | break; |
66f3b8e2 | 845 | |
aa373cf5 JK |
846 | /* |
847 | * Background writeout and kupdate-style writeback may | |
848 | * run forever. Stop them if there is other work to do | |
849 | * so that e.g. sync can proceed. They'll be restarted | |
850 | * after the other works are all done. | |
851 | */ | |
852 | if ((work->for_background || work->for_kupdate) && | |
853 | !list_empty(&wb->bdi->work_list)) | |
854 | break; | |
855 | ||
38f21977 | 856 | /* |
d3ddec76 WF |
857 | * For background writeout, stop when we are below the |
858 | * background dirty threshold | |
38f21977 | 859 | */ |
b00949aa | 860 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
03ba3782 | 861 | break; |
38f21977 | 862 | |
1bc36b64 JK |
863 | /* |
864 | * Kupdate and background works are special and we want to | |
865 | * include all inodes that need writing. Livelock avoidance is | |
866 | * handled by these works yielding to any other work so we are | |
867 | * safe. | |
868 | */ | |
ba9aa839 | 869 | if (work->for_kupdate) { |
0dc83bd3 | 870 | oldest_jif = jiffies - |
ba9aa839 | 871 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 872 | } else if (work->for_background) |
0dc83bd3 | 873 | oldest_jif = jiffies; |
028c2dd1 | 874 | |
d46db3d5 | 875 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 876 | if (list_empty(&wb->b_io)) |
ad4e38dd | 877 | queue_io(wb, work); |
83ba7b07 | 878 | if (work->sb) |
d46db3d5 | 879 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 880 | else |
d46db3d5 WF |
881 | progress = __writeback_inodes_wb(wb, work); |
882 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 883 | |
e98be2d5 | 884 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
885 | |
886 | /* | |
e6fb6da2 WF |
887 | * Did we write something? Try for more |
888 | * | |
889 | * Dirty inodes are moved to b_io for writeback in batches. | |
890 | * The completion of the current batch does not necessarily | |
891 | * mean the overall work is done. So we keep looping as long | |
892 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 893 | */ |
d46db3d5 | 894 | if (progress) |
71fd05a8 JA |
895 | continue; |
896 | /* | |
e6fb6da2 | 897 | * No more inodes for IO, bail |
71fd05a8 | 898 | */ |
b7a2441f | 899 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 900 | break; |
71fd05a8 JA |
901 | /* |
902 | * Nothing written. Wait for some inode to | |
903 | * become available for writeback. Otherwise | |
904 | * we'll just busyloop. | |
905 | */ | |
71fd05a8 | 906 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 907 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 908 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 909 | spin_lock(&inode->i_lock); |
f0d07b7f | 910 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
911 | /* This function drops i_lock... */ |
912 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 913 | spin_lock(&wb->list_lock); |
03ba3782 JA |
914 | } |
915 | } | |
e8dfc305 | 916 | spin_unlock(&wb->list_lock); |
03ba3782 | 917 | |
d46db3d5 | 918 | return nr_pages - work->nr_pages; |
03ba3782 JA |
919 | } |
920 | ||
921 | /* | |
83ba7b07 | 922 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 923 | */ |
83ba7b07 | 924 | static struct wb_writeback_work * |
08852b6d | 925 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 926 | { |
83ba7b07 | 927 | struct wb_writeback_work *work = NULL; |
03ba3782 | 928 | |
6467716a | 929 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
930 | if (!list_empty(&bdi->work_list)) { |
931 | work = list_entry(bdi->work_list.next, | |
932 | struct wb_writeback_work, list); | |
933 | list_del_init(&work->list); | |
03ba3782 | 934 | } |
6467716a | 935 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 936 | return work; |
03ba3782 JA |
937 | } |
938 | ||
cdf01dd5 LT |
939 | /* |
940 | * Add in the number of potentially dirty inodes, because each inode | |
941 | * write can dirty pagecache in the underlying blockdev. | |
942 | */ | |
943 | static unsigned long get_nr_dirty_pages(void) | |
944 | { | |
945 | return global_page_state(NR_FILE_DIRTY) + | |
946 | global_page_state(NR_UNSTABLE_NFS) + | |
947 | get_nr_dirty_inodes(); | |
948 | } | |
949 | ||
6585027a JK |
950 | static long wb_check_background_flush(struct bdi_writeback *wb) |
951 | { | |
b00949aa | 952 | if (over_bground_thresh(wb->bdi)) { |
6585027a JK |
953 | |
954 | struct wb_writeback_work work = { | |
955 | .nr_pages = LONG_MAX, | |
956 | .sync_mode = WB_SYNC_NONE, | |
957 | .for_background = 1, | |
958 | .range_cyclic = 1, | |
0e175a18 | 959 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
960 | }; |
961 | ||
962 | return wb_writeback(wb, &work); | |
963 | } | |
964 | ||
965 | return 0; | |
966 | } | |
967 | ||
03ba3782 JA |
968 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
969 | { | |
970 | unsigned long expired; | |
971 | long nr_pages; | |
972 | ||
69b62d01 JA |
973 | /* |
974 | * When set to zero, disable periodic writeback | |
975 | */ | |
976 | if (!dirty_writeback_interval) | |
977 | return 0; | |
978 | ||
03ba3782 JA |
979 | expired = wb->last_old_flush + |
980 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
981 | if (time_before(jiffies, expired)) | |
982 | return 0; | |
983 | ||
984 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 985 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 986 | |
c4a77a6c | 987 | if (nr_pages) { |
83ba7b07 | 988 | struct wb_writeback_work work = { |
c4a77a6c JA |
989 | .nr_pages = nr_pages, |
990 | .sync_mode = WB_SYNC_NONE, | |
991 | .for_kupdate = 1, | |
992 | .range_cyclic = 1, | |
0e175a18 | 993 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
994 | }; |
995 | ||
83ba7b07 | 996 | return wb_writeback(wb, &work); |
c4a77a6c | 997 | } |
03ba3782 JA |
998 | |
999 | return 0; | |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * Retrieve work items and do the writeback they describe | |
1004 | */ | |
25d130ba | 1005 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 JA |
1006 | { |
1007 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 1008 | struct wb_writeback_work *work; |
c4a77a6c | 1009 | long wrote = 0; |
03ba3782 | 1010 | |
81d73a32 | 1011 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 1012 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 | 1013 | |
455b2864 DC |
1014 | trace_writeback_exec(bdi, work); |
1015 | ||
83ba7b07 | 1016 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
1017 | |
1018 | /* | |
83ba7b07 CH |
1019 | * Notify the caller of completion if this is a synchronous |
1020 | * work item, otherwise just free it. | |
03ba3782 | 1021 | */ |
83ba7b07 CH |
1022 | if (work->done) |
1023 | complete(work->done); | |
1024 | else | |
1025 | kfree(work); | |
03ba3782 JA |
1026 | } |
1027 | ||
1028 | /* | |
1029 | * Check for periodic writeback, kupdated() style | |
1030 | */ | |
1031 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1032 | wrote += wb_check_background_flush(wb); |
81d73a32 | 1033 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
1034 | |
1035 | return wrote; | |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1040 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1041 | */ |
839a8e86 | 1042 | void bdi_writeback_workfn(struct work_struct *work) |
03ba3782 | 1043 | { |
839a8e86 TH |
1044 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1045 | struct bdi_writeback, dwork); | |
08243900 | 1046 | struct backing_dev_info *bdi = wb->bdi; |
03ba3782 JA |
1047 | long pages_written; |
1048 | ||
ef3b1019 | 1049 | set_worker_desc("flush-%s", dev_name(bdi->dev)); |
766f9164 | 1050 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1051 | |
839a8e86 | 1052 | if (likely(!current_is_workqueue_rescuer() || |
5acda9d1 | 1053 | !test_bit(BDI_registered, &bdi->state))) { |
6467716a | 1054 | /* |
839a8e86 TH |
1055 | * The normal path. Keep writing back @bdi until its |
1056 | * work_list is empty. Note that this path is also taken | |
1057 | * if @bdi is shutting down even when we're running off the | |
1058 | * rescuer as work_list needs to be drained. | |
6467716a | 1059 | */ |
839a8e86 | 1060 | do { |
25d130ba | 1061 | pages_written = wb_do_writeback(wb); |
839a8e86 TH |
1062 | trace_writeback_pages_written(pages_written); |
1063 | } while (!list_empty(&bdi->work_list)); | |
1064 | } else { | |
1065 | /* | |
1066 | * bdi_wq can't get enough workers and we're running off | |
1067 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1068 | * enough for efficient IO. | |
1069 | */ | |
1070 | pages_written = writeback_inodes_wb(&bdi->wb, 1024, | |
1071 | WB_REASON_FORKER_THREAD); | |
455b2864 | 1072 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1073 | } |
1074 | ||
6ca738d6 DB |
1075 | if (!list_empty(&bdi->work_list)) |
1076 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
1077 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
1078 | bdi_wakeup_thread_delayed(bdi); | |
455b2864 | 1079 | |
839a8e86 | 1080 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1081 | } |
1082 | ||
1083 | /* | |
b8c2f347 CH |
1084 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1085 | * the whole world. | |
03ba3782 | 1086 | */ |
0e175a18 | 1087 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1088 | { |
b8c2f347 | 1089 | struct backing_dev_info *bdi; |
03ba3782 | 1090 | |
47df3dde JK |
1091 | if (!nr_pages) |
1092 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1093 | |
b8c2f347 | 1094 | rcu_read_lock(); |
cfc4ba53 | 1095 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1096 | if (!bdi_has_dirty_io(bdi)) |
1097 | continue; | |
0e175a18 | 1098 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
03ba3782 | 1099 | } |
cfc4ba53 | 1100 | rcu_read_unlock(); |
1da177e4 LT |
1101 | } |
1102 | ||
03ba3782 JA |
1103 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1104 | { | |
1105 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1106 | struct dentry *dentry; | |
1107 | const char *name = "?"; | |
1108 | ||
1109 | dentry = d_find_alias(inode); | |
1110 | if (dentry) { | |
1111 | spin_lock(&dentry->d_lock); | |
1112 | name = (const char *) dentry->d_name.name; | |
1113 | } | |
1114 | printk(KERN_DEBUG | |
1115 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1116 | current->comm, task_pid_nr(current), inode->i_ino, | |
1117 | name, inode->i_sb->s_id); | |
1118 | if (dentry) { | |
1119 | spin_unlock(&dentry->d_lock); | |
1120 | dput(dentry); | |
1121 | } | |
1122 | } | |
1123 | } | |
1124 | ||
1125 | /** | |
1126 | * __mark_inode_dirty - internal function | |
1127 | * @inode: inode to mark | |
1128 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1129 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1130 | * mark_inode_dirty_sync. | |
1da177e4 | 1131 | * |
03ba3782 JA |
1132 | * Put the inode on the super block's dirty list. |
1133 | * | |
1134 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1135 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1136 | * If it was not hashed, it will never be added to the dirty list | |
1137 | * even if it is later hashed, as it will have been marked dirty already. | |
1138 | * | |
1139 | * In short, make sure you hash any inodes _before_ you start marking | |
1140 | * them dirty. | |
1da177e4 | 1141 | * |
03ba3782 JA |
1142 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1143 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1144 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1145 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1146 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1147 | * blockdev inode. | |
1da177e4 | 1148 | */ |
03ba3782 | 1149 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1150 | { |
03ba3782 | 1151 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1152 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1153 | |
03ba3782 JA |
1154 | /* |
1155 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1156 | * dirty the inode itself | |
1157 | */ | |
1158 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
9fb0a7da TH |
1159 | trace_writeback_dirty_inode_start(inode, flags); |
1160 | ||
03ba3782 | 1161 | if (sb->s_op->dirty_inode) |
aa385729 | 1162 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1163 | |
1164 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 JA |
1165 | } |
1166 | ||
1167 | /* | |
9c6ac78e TH |
1168 | * Paired with smp_mb() in __writeback_single_inode() for the |
1169 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
1170 | */ |
1171 | smp_mb(); | |
1172 | ||
03ba3782 JA |
1173 | if ((inode->i_state & flags) == flags) |
1174 | return; | |
1175 | ||
1176 | if (unlikely(block_dump)) | |
1177 | block_dump___mark_inode_dirty(inode); | |
1178 | ||
250df6ed | 1179 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1180 | if ((inode->i_state & flags) != flags) { |
1181 | const int was_dirty = inode->i_state & I_DIRTY; | |
1182 | ||
1183 | inode->i_state |= flags; | |
1184 | ||
1185 | /* | |
1186 | * If the inode is being synced, just update its dirty state. | |
1187 | * The unlocker will place the inode on the appropriate | |
1188 | * superblock list, based upon its state. | |
1189 | */ | |
1190 | if (inode->i_state & I_SYNC) | |
250df6ed | 1191 | goto out_unlock_inode; |
03ba3782 JA |
1192 | |
1193 | /* | |
1194 | * Only add valid (hashed) inodes to the superblock's | |
1195 | * dirty list. Add blockdev inodes as well. | |
1196 | */ | |
1197 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1198 | if (inode_unhashed(inode)) |
250df6ed | 1199 | goto out_unlock_inode; |
03ba3782 | 1200 | } |
a4ffdde6 | 1201 | if (inode->i_state & I_FREEING) |
250df6ed | 1202 | goto out_unlock_inode; |
03ba3782 JA |
1203 | |
1204 | /* | |
1205 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1206 | * reposition it (that would break b_dirty time-ordering). | |
1207 | */ | |
1208 | if (!was_dirty) { | |
a66979ab | 1209 | bool wakeup_bdi = false; |
253c34e9 AB |
1210 | bdi = inode_to_bdi(inode); |
1211 | ||
146d7009 JB |
1212 | spin_unlock(&inode->i_lock); |
1213 | spin_lock(&bdi->wb.list_lock); | |
253c34e9 AB |
1214 | if (bdi_cap_writeback_dirty(bdi)) { |
1215 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1216 | "bdi-%s not registered\n", bdi->name); | |
1217 | ||
1218 | /* | |
1219 | * If this is the first dirty inode for this | |
1220 | * bdi, we have to wake-up the corresponding | |
1221 | * bdi thread to make sure background | |
1222 | * write-back happens later. | |
1223 | */ | |
1224 | if (!wb_has_dirty_io(&bdi->wb)) | |
1225 | wakeup_bdi = true; | |
500b067c | 1226 | } |
03ba3782 JA |
1227 | |
1228 | inode->dirtied_when = jiffies; | |
7ccf19a8 | 1229 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1230 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1231 | |
1232 | if (wakeup_bdi) | |
1233 | bdi_wakeup_thread_delayed(bdi); | |
1234 | return; | |
1da177e4 | 1235 | } |
1da177e4 | 1236 | } |
250df6ed DC |
1237 | out_unlock_inode: |
1238 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1239 | |
03ba3782 JA |
1240 | } |
1241 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1242 | ||
b6e51316 | 1243 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1244 | { |
1245 | struct inode *inode, *old_inode = NULL; | |
1246 | ||
1247 | /* | |
1248 | * We need to be protected against the filesystem going from | |
1249 | * r/o to r/w or vice versa. | |
1250 | */ | |
b6e51316 | 1251 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1252 | |
55fa6091 | 1253 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1254 | |
1255 | /* | |
1256 | * Data integrity sync. Must wait for all pages under writeback, | |
1257 | * because there may have been pages dirtied before our sync | |
1258 | * call, but which had writeout started before we write it out. | |
1259 | * In which case, the inode may not be on the dirty list, but | |
1260 | * we still have to wait for that writeout. | |
1261 | */ | |
b6e51316 | 1262 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1263 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1264 | |
250df6ed DC |
1265 | spin_lock(&inode->i_lock); |
1266 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1267 | (mapping->nrpages == 0)) { | |
1268 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1269 | continue; |
250df6ed | 1270 | } |
03ba3782 | 1271 | __iget(inode); |
250df6ed | 1272 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1273 | spin_unlock(&inode_sb_list_lock); |
1274 | ||
03ba3782 | 1275 | /* |
55fa6091 DC |
1276 | * We hold a reference to 'inode' so it couldn't have been |
1277 | * removed from s_inodes list while we dropped the | |
1278 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1279 | * be holding the last reference and we cannot iput it under | |
1280 | * inode_sb_list_lock. So we keep the reference and iput it | |
1281 | * later. | |
03ba3782 JA |
1282 | */ |
1283 | iput(old_inode); | |
1284 | old_inode = inode; | |
1285 | ||
1286 | filemap_fdatawait(mapping); | |
1287 | ||
1288 | cond_resched(); | |
1289 | ||
55fa6091 | 1290 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1291 | } |
55fa6091 | 1292 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1293 | iput(old_inode); |
1da177e4 LT |
1294 | } |
1295 | ||
d8a8559c | 1296 | /** |
3259f8be | 1297 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1298 | * @sb: the superblock |
3259f8be | 1299 | * @nr: the number of pages to write |
786228ab | 1300 | * @reason: reason why some writeback work initiated |
1da177e4 | 1301 | * |
d8a8559c JA |
1302 | * Start writeback on some inodes on this super_block. No guarantees are made |
1303 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1304 | * for IO completion of submitted IO. |
1da177e4 | 1305 | */ |
0e175a18 CW |
1306 | void writeback_inodes_sb_nr(struct super_block *sb, |
1307 | unsigned long nr, | |
1308 | enum wb_reason reason) | |
1da177e4 | 1309 | { |
83ba7b07 CH |
1310 | DECLARE_COMPLETION_ONSTACK(done); |
1311 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1312 | .sb = sb, |
1313 | .sync_mode = WB_SYNC_NONE, | |
1314 | .tagged_writepages = 1, | |
1315 | .done = &done, | |
1316 | .nr_pages = nr, | |
0e175a18 | 1317 | .reason = reason, |
3c4d7165 | 1318 | }; |
d8a8559c | 1319 | |
6eedc701 JK |
1320 | if (sb->s_bdi == &noop_backing_dev_info) |
1321 | return; | |
cf37e972 | 1322 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1323 | bdi_queue_work(sb->s_bdi, &work); |
1324 | wait_for_completion(&done); | |
e913fc82 | 1325 | } |
3259f8be CM |
1326 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1327 | ||
1328 | /** | |
1329 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1330 | * @sb: the superblock | |
786228ab | 1331 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1332 | * |
1333 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1334 | * on how many (if any) will be written, and this function does not wait | |
1335 | * for IO completion of submitted IO. | |
1336 | */ | |
0e175a18 | 1337 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1338 | { |
0e175a18 | 1339 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1340 | } |
0e3c9a22 | 1341 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1342 | |
17bd55d0 | 1343 | /** |
10ee27a0 | 1344 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1345 | * @sb: the superblock |
10ee27a0 MX |
1346 | * @nr: the number of pages to write |
1347 | * @reason: the reason of writeback | |
17bd55d0 | 1348 | * |
10ee27a0 | 1349 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1350 | * Returns 1 if writeback was started, 0 if not. |
1351 | */ | |
10ee27a0 MX |
1352 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
1353 | unsigned long nr, | |
1354 | enum wb_reason reason) | |
17bd55d0 | 1355 | { |
10ee27a0 | 1356 | if (writeback_in_progress(sb->s_bdi)) |
17bd55d0 | 1357 | return 1; |
10ee27a0 MX |
1358 | |
1359 | if (!down_read_trylock(&sb->s_umount)) | |
17bd55d0 | 1360 | return 0; |
10ee27a0 MX |
1361 | |
1362 | writeback_inodes_sb_nr(sb, nr, reason); | |
1363 | up_read(&sb->s_umount); | |
1364 | return 1; | |
17bd55d0 | 1365 | } |
10ee27a0 | 1366 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1367 | |
3259f8be | 1368 | /** |
10ee27a0 | 1369 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1370 | * @sb: the superblock |
786228ab | 1371 | * @reason: reason why some writeback work was initiated |
3259f8be | 1372 | * |
10ee27a0 | 1373 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1374 | * Returns 1 if writeback was started, 0 if not. |
1375 | */ | |
10ee27a0 | 1376 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1377 | { |
10ee27a0 | 1378 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1379 | } |
10ee27a0 | 1380 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1381 | |
d8a8559c JA |
1382 | /** |
1383 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 1384 | * @sb: the superblock |
d8a8559c JA |
1385 | * |
1386 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 1387 | * super_block. |
d8a8559c | 1388 | */ |
0dc83bd3 | 1389 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1390 | { |
83ba7b07 CH |
1391 | DECLARE_COMPLETION_ONSTACK(done); |
1392 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1393 | .sb = sb, |
1394 | .sync_mode = WB_SYNC_ALL, | |
1395 | .nr_pages = LONG_MAX, | |
1396 | .range_cyclic = 0, | |
83ba7b07 | 1397 | .done = &done, |
0e175a18 | 1398 | .reason = WB_REASON_SYNC, |
7747bd4b | 1399 | .for_sync = 1, |
3c4d7165 CH |
1400 | }; |
1401 | ||
6eedc701 JK |
1402 | /* Nothing to do? */ |
1403 | if (sb->s_bdi == &noop_backing_dev_info) | |
1404 | return; | |
cf37e972 CH |
1405 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1406 | ||
83ba7b07 CH |
1407 | bdi_queue_work(sb->s_bdi, &work); |
1408 | wait_for_completion(&done); | |
1409 | ||
b6e51316 | 1410 | wait_sb_inodes(sb); |
1da177e4 | 1411 | } |
d8a8559c | 1412 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1413 | |
1da177e4 | 1414 | /** |
7f04c26d AA |
1415 | * write_inode_now - write an inode to disk |
1416 | * @inode: inode to write to disk | |
1417 | * @sync: whether the write should be synchronous or not | |
1418 | * | |
1419 | * This function commits an inode to disk immediately if it is dirty. This is | |
1420 | * primarily needed by knfsd. | |
1da177e4 | 1421 | * |
7f04c26d | 1422 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1423 | */ |
1da177e4 LT |
1424 | int write_inode_now(struct inode *inode, int sync) |
1425 | { | |
f758eeab | 1426 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1427 | struct writeback_control wbc = { |
1428 | .nr_to_write = LONG_MAX, | |
18914b18 | 1429 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1430 | .range_start = 0, |
1431 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1432 | }; |
1433 | ||
1434 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1435 | wbc.nr_to_write = 0; |
1da177e4 LT |
1436 | |
1437 | might_sleep(); | |
4f8ad655 | 1438 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1439 | } |
1440 | EXPORT_SYMBOL(write_inode_now); | |
1441 | ||
1442 | /** | |
1443 | * sync_inode - write an inode and its pages to disk. | |
1444 | * @inode: the inode to sync | |
1445 | * @wbc: controls the writeback mode | |
1446 | * | |
1447 | * sync_inode() will write an inode and its pages to disk. It will also | |
1448 | * correctly update the inode on its superblock's dirty inode lists and will | |
1449 | * update inode->i_state. | |
1450 | * | |
1451 | * The caller must have a ref on the inode. | |
1452 | */ | |
1453 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1454 | { | |
4f8ad655 | 1455 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1456 | } |
1457 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1458 | |
1459 | /** | |
c691b9d9 | 1460 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1461 | * @inode: the inode to sync |
1462 | * @wait: wait for I/O to complete. | |
1463 | * | |
c691b9d9 | 1464 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1465 | * |
1466 | * Note: only writes the actual inode, no associated data or other metadata. | |
1467 | */ | |
1468 | int sync_inode_metadata(struct inode *inode, int wait) | |
1469 | { | |
1470 | struct writeback_control wbc = { | |
1471 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1472 | .nr_to_write = 0, /* metadata-only */ | |
1473 | }; | |
1474 | ||
1475 | return sync_inode(inode, &wbc); | |
1476 | } | |
1477 | EXPORT_SYMBOL(sync_inode_metadata); |