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> |
07f3f05c | 29 | #include "internal.h" |
1da177e4 | 30 | |
bc31b86a WF |
31 | /* |
32 | * 4MB minimal write chunk size | |
33 | */ | |
34 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) | |
35 | ||
c4a77a6c JA |
36 | /* |
37 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
38 | */ | |
83ba7b07 | 39 | struct wb_writeback_work { |
c4a77a6c JA |
40 | long nr_pages; |
41 | struct super_block *sb; | |
c4a391b5 JK |
42 | /* |
43 | * Write only inodes dirtied before this time. Don't forget to set | |
44 | * older_than_this_is_set when you set this. | |
45 | */ | |
46 | unsigned long older_than_this; | |
c4a77a6c | 47 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 48 | unsigned int tagged_writepages:1; |
52957fe1 HS |
49 | unsigned int for_kupdate:1; |
50 | unsigned int range_cyclic:1; | |
51 | unsigned int for_background:1; | |
7747bd4b | 52 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
c4a391b5 | 53 | unsigned int older_than_this_is_set:1; |
0e175a18 | 54 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 55 | |
8010c3b6 | 56 | struct list_head list; /* pending work list */ |
83ba7b07 | 57 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
58 | }; |
59 | ||
f11b00f3 AB |
60 | /** |
61 | * writeback_in_progress - determine whether there is writeback in progress | |
62 | * @bdi: the device's backing_dev_info structure. | |
63 | * | |
03ba3782 JA |
64 | * Determine whether there is writeback waiting to be handled against a |
65 | * backing device. | |
f11b00f3 AB |
66 | */ |
67 | int writeback_in_progress(struct backing_dev_info *bdi) | |
68 | { | |
81d73a32 | 69 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 | 70 | } |
00d4e736 | 71 | EXPORT_SYMBOL(writeback_in_progress); |
f11b00f3 | 72 | |
692ebd17 JK |
73 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
74 | { | |
75 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 76 | |
a8855990 | 77 | if (sb_is_blkdev_sb(sb)) |
aaead25b CH |
78 | return inode->i_mapping->backing_dev_info; |
79 | ||
80 | return sb->s_bdi; | |
692ebd17 JK |
81 | } |
82 | ||
7ccf19a8 NP |
83 | static inline struct inode *wb_inode(struct list_head *head) |
84 | { | |
85 | return list_entry(head, struct inode, i_wb_list); | |
86 | } | |
87 | ||
15eb77a0 WF |
88 | /* |
89 | * Include the creation of the trace points after defining the | |
90 | * wb_writeback_work structure and inline functions so that the definition | |
91 | * remains local to this file. | |
92 | */ | |
93 | #define CREATE_TRACE_POINTS | |
94 | #include <trace/events/writeback.h> | |
95 | ||
6585027a JK |
96 | static void bdi_queue_work(struct backing_dev_info *bdi, |
97 | struct wb_writeback_work *work) | |
98 | { | |
99 | trace_writeback_queue(bdi, work); | |
100 | ||
101 | spin_lock_bh(&bdi->wb_lock); | |
102 | list_add_tail(&work->list, &bdi->work_list); | |
6467716a | 103 | spin_unlock_bh(&bdi->wb_lock); |
839a8e86 TH |
104 | |
105 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
1da177e4 LT |
106 | } |
107 | ||
83ba7b07 CH |
108 | static void |
109 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
0e175a18 | 110 | bool range_cyclic, enum wb_reason reason) |
1da177e4 | 111 | { |
83ba7b07 | 112 | struct wb_writeback_work *work; |
03ba3782 | 113 | |
bcddc3f0 JA |
114 | /* |
115 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
116 | * wakeup the thread for old dirty data writeback | |
117 | */ | |
83ba7b07 CH |
118 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
119 | if (!work) { | |
839a8e86 TH |
120 | trace_writeback_nowork(bdi); |
121 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
83ba7b07 | 122 | return; |
bcddc3f0 | 123 | } |
03ba3782 | 124 | |
83ba7b07 CH |
125 | work->sync_mode = WB_SYNC_NONE; |
126 | work->nr_pages = nr_pages; | |
127 | work->range_cyclic = range_cyclic; | |
0e175a18 | 128 | work->reason = reason; |
03ba3782 | 129 | |
83ba7b07 | 130 | bdi_queue_work(bdi, work); |
b6e51316 JA |
131 | } |
132 | ||
133 | /** | |
134 | * bdi_start_writeback - start writeback | |
135 | * @bdi: the backing device to write from | |
136 | * @nr_pages: the number of pages to write | |
786228ab | 137 | * @reason: reason why some writeback work was initiated |
b6e51316 JA |
138 | * |
139 | * Description: | |
140 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 141 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 142 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
143 | * |
144 | */ | |
0e175a18 CW |
145 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
146 | enum wb_reason reason) | |
b6e51316 | 147 | { |
0e175a18 | 148 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
c5444198 | 149 | } |
d3ddec76 | 150 | |
c5444198 CH |
151 | /** |
152 | * bdi_start_background_writeback - start background writeback | |
153 | * @bdi: the backing device to write from | |
154 | * | |
155 | * Description: | |
6585027a JK |
156 | * This makes sure WB_SYNC_NONE background writeback happens. When |
157 | * this function returns, it is only guaranteed that for given BDI | |
158 | * some IO is happening if we are over background dirty threshold. | |
159 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
160 | */ |
161 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
162 | { | |
6585027a JK |
163 | /* |
164 | * We just wake up the flusher thread. It will perform background | |
165 | * writeback as soon as there is no other work to do. | |
166 | */ | |
71927e84 | 167 | trace_writeback_wake_background(bdi); |
839a8e86 | 168 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
1da177e4 LT |
169 | } |
170 | ||
a66979ab DC |
171 | /* |
172 | * Remove the inode from the writeback list it is on. | |
173 | */ | |
174 | void inode_wb_list_del(struct inode *inode) | |
175 | { | |
f758eeab CH |
176 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
177 | ||
178 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 179 | list_del_init(&inode->i_wb_list); |
f758eeab | 180 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
181 | } |
182 | ||
6610a0bc AM |
183 | /* |
184 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
185 | * furthest end of its superblock's dirty-inode list. | |
186 | * | |
187 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 188 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
189 | * the case then the inode must have been redirtied while it was being written |
190 | * out and we don't reset its dirtied_when. | |
191 | */ | |
f758eeab | 192 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 193 | { |
f758eeab | 194 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 195 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 196 | struct inode *tail; |
6610a0bc | 197 | |
7ccf19a8 | 198 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 199 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
200 | inode->dirtied_when = jiffies; |
201 | } | |
7ccf19a8 | 202 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
203 | } |
204 | ||
c986d1e2 | 205 | /* |
66f3b8e2 | 206 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 207 | */ |
f758eeab | 208 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 209 | { |
f758eeab | 210 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 211 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
212 | } |
213 | ||
1c0eeaf5 JE |
214 | static void inode_sync_complete(struct inode *inode) |
215 | { | |
365b94ae | 216 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
217 | /* If inode is clean an unused, put it into LRU now... */ |
218 | inode_add_lru(inode); | |
365b94ae | 219 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
220 | smp_mb(); |
221 | wake_up_bit(&inode->i_state, __I_SYNC); | |
222 | } | |
223 | ||
d2caa3c5 JL |
224 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
225 | { | |
226 | bool ret = time_after(inode->dirtied_when, t); | |
227 | #ifndef CONFIG_64BIT | |
228 | /* | |
229 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
230 | * It _appears_ to be in the future, but is actually in distant past. | |
231 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 232 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
233 | */ |
234 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
235 | #endif | |
236 | return ret; | |
237 | } | |
238 | ||
2c136579 | 239 | /* |
0e2f2b23 | 240 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 241 | * @delaying_queue to @dispatch_queue. |
2c136579 | 242 | */ |
e84d0a4f | 243 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 244 | struct list_head *dispatch_queue, |
ad4e38dd | 245 | struct wb_writeback_work *work) |
2c136579 | 246 | { |
5c03449d SL |
247 | LIST_HEAD(tmp); |
248 | struct list_head *pos, *node; | |
cf137307 | 249 | struct super_block *sb = NULL; |
5c03449d | 250 | struct inode *inode; |
cf137307 | 251 | int do_sb_sort = 0; |
e84d0a4f | 252 | int moved = 0; |
5c03449d | 253 | |
c4a391b5 | 254 | WARN_ON_ONCE(!work->older_than_this_is_set); |
2c136579 | 255 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 256 | inode = wb_inode(delaying_queue->prev); |
c4a391b5 | 257 | if (inode_dirtied_after(inode, work->older_than_this)) |
2c136579 | 258 | break; |
a8855990 JK |
259 | list_move(&inode->i_wb_list, &tmp); |
260 | moved++; | |
261 | if (sb_is_blkdev_sb(inode->i_sb)) | |
262 | continue; | |
cf137307 JA |
263 | if (sb && sb != inode->i_sb) |
264 | do_sb_sort = 1; | |
265 | sb = inode->i_sb; | |
5c03449d SL |
266 | } |
267 | ||
cf137307 JA |
268 | /* just one sb in list, splice to dispatch_queue and we're done */ |
269 | if (!do_sb_sort) { | |
270 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 271 | goto out; |
cf137307 JA |
272 | } |
273 | ||
5c03449d SL |
274 | /* Move inodes from one superblock together */ |
275 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 276 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 277 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 278 | inode = wb_inode(pos); |
5c03449d | 279 | if (inode->i_sb == sb) |
7ccf19a8 | 280 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 281 | } |
2c136579 | 282 | } |
e84d0a4f WF |
283 | out: |
284 | return moved; | |
2c136579 FW |
285 | } |
286 | ||
287 | /* | |
288 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
289 | * Before |
290 | * newly dirtied b_dirty b_io b_more_io | |
291 | * =============> gf edc BA | |
292 | * After | |
293 | * newly dirtied b_dirty b_io b_more_io | |
294 | * =============> g fBAedc | |
295 | * | | |
296 | * +--> dequeue for IO | |
2c136579 | 297 | */ |
ad4e38dd | 298 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 299 | { |
e84d0a4f | 300 | int moved; |
f758eeab | 301 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 302 | list_splice_init(&wb->b_more_io, &wb->b_io); |
ad4e38dd CW |
303 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
304 | trace_writeback_queue_io(wb, work, moved); | |
66f3b8e2 JA |
305 | } |
306 | ||
a9185b41 | 307 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 308 | { |
9fb0a7da TH |
309 | int ret; |
310 | ||
311 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
312 | trace_writeback_write_inode_start(inode, wbc); | |
313 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
314 | trace_writeback_write_inode(inode, wbc); | |
315 | return ret; | |
316 | } | |
03ba3782 | 317 | return 0; |
08d8e974 | 318 | } |
08d8e974 | 319 | |
1da177e4 | 320 | /* |
169ebd90 JK |
321 | * Wait for writeback on an inode to complete. Called with i_lock held. |
322 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 323 | */ |
169ebd90 JK |
324 | static void __inode_wait_for_writeback(struct inode *inode) |
325 | __releases(inode->i_lock) | |
326 | __acquires(inode->i_lock) | |
01c03194 CH |
327 | { |
328 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
329 | wait_queue_head_t *wqh; | |
330 | ||
331 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
332 | while (inode->i_state & I_SYNC) { |
333 | spin_unlock(&inode->i_lock); | |
01c03194 | 334 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 335 | spin_lock(&inode->i_lock); |
58a9d3d8 | 336 | } |
01c03194 CH |
337 | } |
338 | ||
169ebd90 JK |
339 | /* |
340 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
341 | */ | |
342 | void inode_wait_for_writeback(struct inode *inode) | |
343 | { | |
344 | spin_lock(&inode->i_lock); | |
345 | __inode_wait_for_writeback(inode); | |
346 | spin_unlock(&inode->i_lock); | |
347 | } | |
348 | ||
349 | /* | |
350 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
351 | * held and drops it. It is aimed for callers not holding any inode reference | |
352 | * so once i_lock is dropped, inode can go away. | |
353 | */ | |
354 | static void inode_sleep_on_writeback(struct inode *inode) | |
355 | __releases(inode->i_lock) | |
356 | { | |
357 | DEFINE_WAIT(wait); | |
358 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
359 | int sleep; | |
360 | ||
361 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
362 | sleep = inode->i_state & I_SYNC; | |
363 | spin_unlock(&inode->i_lock); | |
364 | if (sleep) | |
365 | schedule(); | |
366 | finish_wait(wqh, &wait); | |
367 | } | |
368 | ||
ccb26b5a JK |
369 | /* |
370 | * Find proper writeback list for the inode depending on its current state and | |
371 | * possibly also change of its state while we were doing writeback. Here we | |
372 | * handle things such as livelock prevention or fairness of writeback among | |
373 | * inodes. This function can be called only by flusher thread - noone else | |
374 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
375 | * thread's back can have unexpected consequences. | |
376 | */ | |
377 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
378 | struct writeback_control *wbc) | |
379 | { | |
380 | if (inode->i_state & I_FREEING) | |
381 | return; | |
382 | ||
383 | /* | |
384 | * Sync livelock prevention. Each inode is tagged and synced in one | |
385 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
386 | * the dirty time to prevent enqueue and sync it again. | |
387 | */ | |
388 | if ((inode->i_state & I_DIRTY) && | |
389 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
390 | inode->dirtied_when = jiffies; | |
391 | ||
4f8ad655 JK |
392 | if (wbc->pages_skipped) { |
393 | /* | |
394 | * writeback is not making progress due to locked | |
395 | * buffers. Skip this inode for now. | |
396 | */ | |
397 | redirty_tail(inode, wb); | |
398 | return; | |
399 | } | |
400 | ||
ccb26b5a JK |
401 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
402 | /* | |
403 | * We didn't write back all the pages. nfs_writepages() | |
404 | * sometimes bales out without doing anything. | |
405 | */ | |
406 | if (wbc->nr_to_write <= 0) { | |
407 | /* Slice used up. Queue for next turn. */ | |
408 | requeue_io(inode, wb); | |
409 | } else { | |
410 | /* | |
411 | * Writeback blocked by something other than | |
412 | * congestion. Delay the inode for some time to | |
413 | * avoid spinning on the CPU (100% iowait) | |
414 | * retrying writeback of the dirty page/inode | |
415 | * that cannot be performed immediately. | |
416 | */ | |
417 | redirty_tail(inode, wb); | |
418 | } | |
419 | } else if (inode->i_state & I_DIRTY) { | |
420 | /* | |
421 | * Filesystems can dirty the inode during writeback operations, | |
422 | * such as delayed allocation during submission or metadata | |
423 | * updates after data IO completion. | |
424 | */ | |
425 | redirty_tail(inode, wb); | |
426 | } else { | |
427 | /* The inode is clean. Remove from writeback lists. */ | |
428 | list_del_init(&inode->i_wb_list); | |
429 | } | |
430 | } | |
431 | ||
01c03194 | 432 | /* |
4f8ad655 JK |
433 | * Write out an inode and its dirty pages. Do not update the writeback list |
434 | * linkage. That is left to the caller. The caller is also responsible for | |
435 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
436 | */ |
437 | static int | |
cd8ed2a4 | 438 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 439 | { |
1da177e4 | 440 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 441 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 442 | unsigned dirty; |
1da177e4 LT |
443 | int ret; |
444 | ||
4f8ad655 | 445 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 446 | |
9fb0a7da TH |
447 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
448 | ||
1da177e4 LT |
449 | ret = do_writepages(mapping, wbc); |
450 | ||
26821ed4 CH |
451 | /* |
452 | * Make sure to wait on the data before writing out the metadata. | |
453 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
454 | * I/O completion. We don't do it for sync(2) writeback because it has a |
455 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
456 | * inode metadata is written back correctly. | |
26821ed4 | 457 | */ |
7747bd4b | 458 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 459 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
460 | if (ret == 0) |
461 | ret = err; | |
462 | } | |
463 | ||
5547e8aa DM |
464 | /* |
465 | * Some filesystems may redirty the inode during the writeback | |
466 | * due to delalloc, clear dirty metadata flags right before | |
467 | * write_inode() | |
468 | */ | |
250df6ed | 469 | spin_lock(&inode->i_lock); |
6290be1c JK |
470 | /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ |
471 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
472 | inode->i_state &= ~I_DIRTY_PAGES; | |
5547e8aa DM |
473 | dirty = inode->i_state & I_DIRTY; |
474 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 475 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
476 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
477 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 478 | int err = write_inode(inode, wbc); |
1da177e4 LT |
479 | if (ret == 0) |
480 | ret = err; | |
481 | } | |
4f8ad655 JK |
482 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
483 | return ret; | |
484 | } | |
485 | ||
486 | /* | |
487 | * Write out an inode's dirty pages. Either the caller has an active reference | |
488 | * on the inode or the inode has I_WILL_FREE set. | |
489 | * | |
490 | * This function is designed to be called for writing back one inode which | |
491 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
492 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
493 | */ | |
494 | static int | |
495 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
496 | struct writeback_control *wbc) | |
497 | { | |
498 | int ret = 0; | |
499 | ||
500 | spin_lock(&inode->i_lock); | |
501 | if (!atomic_read(&inode->i_count)) | |
502 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
503 | else | |
504 | WARN_ON(inode->i_state & I_WILL_FREE); | |
505 | ||
506 | if (inode->i_state & I_SYNC) { | |
507 | if (wbc->sync_mode != WB_SYNC_ALL) | |
508 | goto out; | |
509 | /* | |
169ebd90 JK |
510 | * It's a data-integrity sync. We must wait. Since callers hold |
511 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
512 | * away under us. | |
4f8ad655 | 513 | */ |
169ebd90 | 514 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
515 | } |
516 | WARN_ON(inode->i_state & I_SYNC); | |
517 | /* | |
518 | * Skip inode if it is clean. We don't want to mess with writeback | |
519 | * lists in this function since flusher thread may be doing for example | |
520 | * sync in parallel and if we move the inode, it could get skipped. So | |
521 | * here we make sure inode is on some writeback list and leave it there | |
522 | * unless we have completely cleaned the inode. | |
523 | */ | |
524 | if (!(inode->i_state & I_DIRTY)) | |
525 | goto out; | |
526 | inode->i_state |= I_SYNC; | |
527 | spin_unlock(&inode->i_lock); | |
528 | ||
cd8ed2a4 | 529 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 530 | |
f758eeab | 531 | spin_lock(&wb->list_lock); |
250df6ed | 532 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
533 | /* |
534 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
535 | * touch it. See comment above for explanation. | |
536 | */ | |
537 | if (!(inode->i_state & I_DIRTY)) | |
538 | list_del_init(&inode->i_wb_list); | |
539 | spin_unlock(&wb->list_lock); | |
1c0eeaf5 | 540 | inode_sync_complete(inode); |
4f8ad655 JK |
541 | out: |
542 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
543 | return ret; |
544 | } | |
545 | ||
1a12d8bd WF |
546 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
547 | struct wb_writeback_work *work) | |
d46db3d5 WF |
548 | { |
549 | long pages; | |
550 | ||
551 | /* | |
552 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
553 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
554 | * here avoids calling into writeback_inodes_wb() more than once. | |
555 | * | |
556 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
557 | * | |
558 | * wb_writeback() | |
559 | * writeback_sb_inodes() <== called only once | |
560 | * write_cache_pages() <== called once for each inode | |
561 | * (quickly) tag currently dirty pages | |
562 | * (maybe slowly) sync all tagged pages | |
563 | */ | |
564 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
565 | pages = LONG_MAX; | |
1a12d8bd WF |
566 | else { |
567 | pages = min(bdi->avg_write_bandwidth / 2, | |
568 | global_dirty_limit / DIRTY_SCOPE); | |
569 | pages = min(pages, work->nr_pages); | |
570 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
571 | MIN_WRITEBACK_PAGES); | |
572 | } | |
d46db3d5 WF |
573 | |
574 | return pages; | |
575 | } | |
576 | ||
f11c9c5c ES |
577 | /* |
578 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 579 | * |
d46db3d5 | 580 | * Return the number of pages and/or inodes written. |
f11c9c5c | 581 | */ |
d46db3d5 WF |
582 | static long writeback_sb_inodes(struct super_block *sb, |
583 | struct bdi_writeback *wb, | |
584 | struct wb_writeback_work *work) | |
1da177e4 | 585 | { |
d46db3d5 WF |
586 | struct writeback_control wbc = { |
587 | .sync_mode = work->sync_mode, | |
588 | .tagged_writepages = work->tagged_writepages, | |
589 | .for_kupdate = work->for_kupdate, | |
590 | .for_background = work->for_background, | |
7747bd4b | 591 | .for_sync = work->for_sync, |
d46db3d5 WF |
592 | .range_cyclic = work->range_cyclic, |
593 | .range_start = 0, | |
594 | .range_end = LLONG_MAX, | |
595 | }; | |
596 | unsigned long start_time = jiffies; | |
597 | long write_chunk; | |
598 | long wrote = 0; /* count both pages and inodes */ | |
599 | ||
03ba3782 | 600 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 601 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
602 | |
603 | if (inode->i_sb != sb) { | |
d46db3d5 | 604 | if (work->sb) { |
edadfb10 CH |
605 | /* |
606 | * We only want to write back data for this | |
607 | * superblock, move all inodes not belonging | |
608 | * to it back onto the dirty list. | |
609 | */ | |
f758eeab | 610 | redirty_tail(inode, wb); |
edadfb10 CH |
611 | continue; |
612 | } | |
613 | ||
614 | /* | |
615 | * The inode belongs to a different superblock. | |
616 | * Bounce back to the caller to unpin this and | |
617 | * pin the next superblock. | |
618 | */ | |
d46db3d5 | 619 | break; |
edadfb10 CH |
620 | } |
621 | ||
9843b76a | 622 | /* |
331cbdee WL |
623 | * Don't bother with new inodes or inodes being freed, first |
624 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
625 | * kind writeout is handled by the freer. |
626 | */ | |
250df6ed | 627 | spin_lock(&inode->i_lock); |
9843b76a | 628 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 629 | spin_unlock(&inode->i_lock); |
fcc5c222 | 630 | redirty_tail(inode, wb); |
7ef0d737 NP |
631 | continue; |
632 | } | |
cc1676d9 JK |
633 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
634 | /* | |
635 | * If this inode is locked for writeback and we are not | |
636 | * doing writeback-for-data-integrity, move it to | |
637 | * b_more_io so that writeback can proceed with the | |
638 | * other inodes on s_io. | |
639 | * | |
640 | * We'll have another go at writing back this inode | |
641 | * when we completed a full scan of b_io. | |
642 | */ | |
643 | spin_unlock(&inode->i_lock); | |
644 | requeue_io(inode, wb); | |
645 | trace_writeback_sb_inodes_requeue(inode); | |
646 | continue; | |
647 | } | |
f0d07b7f JK |
648 | spin_unlock(&wb->list_lock); |
649 | ||
4f8ad655 JK |
650 | /* |
651 | * We already requeued the inode if it had I_SYNC set and we | |
652 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
653 | * WB_SYNC_ALL case. | |
654 | */ | |
169ebd90 JK |
655 | if (inode->i_state & I_SYNC) { |
656 | /* Wait for I_SYNC. This function drops i_lock... */ | |
657 | inode_sleep_on_writeback(inode); | |
658 | /* Inode may be gone, start again */ | |
ead188f9 | 659 | spin_lock(&wb->list_lock); |
169ebd90 JK |
660 | continue; |
661 | } | |
4f8ad655 JK |
662 | inode->i_state |= I_SYNC; |
663 | spin_unlock(&inode->i_lock); | |
169ebd90 | 664 | |
1a12d8bd | 665 | write_chunk = writeback_chunk_size(wb->bdi, work); |
d46db3d5 WF |
666 | wbc.nr_to_write = write_chunk; |
667 | wbc.pages_skipped = 0; | |
250df6ed | 668 | |
169ebd90 JK |
669 | /* |
670 | * We use I_SYNC to pin the inode in memory. While it is set | |
671 | * evict_inode() will wait so the inode cannot be freed. | |
672 | */ | |
cd8ed2a4 | 673 | __writeback_single_inode(inode, &wbc); |
250df6ed | 674 | |
d46db3d5 WF |
675 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
676 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
677 | spin_lock(&wb->list_lock); |
678 | spin_lock(&inode->i_lock); | |
d46db3d5 WF |
679 | if (!(inode->i_state & I_DIRTY)) |
680 | wrote++; | |
4f8ad655 JK |
681 | requeue_inode(inode, wb, &wbc); |
682 | inode_sync_complete(inode); | |
0f1b1fd8 | 683 | spin_unlock(&inode->i_lock); |
169ebd90 | 684 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
685 | /* |
686 | * bail out to wb_writeback() often enough to check | |
687 | * background threshold and other termination conditions. | |
688 | */ | |
689 | if (wrote) { | |
690 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
691 | break; | |
692 | if (work->nr_pages <= 0) | |
693 | break; | |
8bc3be27 | 694 | } |
1da177e4 | 695 | } |
d46db3d5 | 696 | return wrote; |
f11c9c5c ES |
697 | } |
698 | ||
d46db3d5 WF |
699 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
700 | struct wb_writeback_work *work) | |
f11c9c5c | 701 | { |
d46db3d5 WF |
702 | unsigned long start_time = jiffies; |
703 | long wrote = 0; | |
38f21977 | 704 | |
f11c9c5c | 705 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 706 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 707 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 708 | |
12ad3ab6 | 709 | if (!grab_super_passive(sb)) { |
0e995816 WF |
710 | /* |
711 | * grab_super_passive() may fail consistently due to | |
712 | * s_umount being grabbed by someone else. Don't use | |
713 | * requeue_io() to avoid busy retrying the inode/sb. | |
714 | */ | |
715 | redirty_tail(inode, wb); | |
edadfb10 | 716 | continue; |
f11c9c5c | 717 | } |
d46db3d5 | 718 | wrote += writeback_sb_inodes(sb, wb, work); |
edadfb10 | 719 | drop_super(sb); |
f11c9c5c | 720 | |
d46db3d5 WF |
721 | /* refer to the same tests at the end of writeback_sb_inodes */ |
722 | if (wrote) { | |
723 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
724 | break; | |
725 | if (work->nr_pages <= 0) | |
726 | break; | |
727 | } | |
f11c9c5c | 728 | } |
66f3b8e2 | 729 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 730 | return wrote; |
66f3b8e2 JA |
731 | } |
732 | ||
7d9f073b | 733 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 734 | enum wb_reason reason) |
edadfb10 | 735 | { |
d46db3d5 WF |
736 | struct wb_writeback_work work = { |
737 | .nr_pages = nr_pages, | |
738 | .sync_mode = WB_SYNC_NONE, | |
739 | .range_cyclic = 1, | |
0e175a18 | 740 | .reason = reason, |
c4a391b5 JK |
741 | .older_than_this = jiffies, |
742 | .older_than_this_is_set = 1, | |
d46db3d5 | 743 | }; |
edadfb10 | 744 | |
f758eeab | 745 | spin_lock(&wb->list_lock); |
424b351f | 746 | if (list_empty(&wb->b_io)) |
ad4e38dd | 747 | queue_io(wb, &work); |
d46db3d5 | 748 | __writeback_inodes_wb(wb, &work); |
f758eeab | 749 | spin_unlock(&wb->list_lock); |
edadfb10 | 750 | |
d46db3d5 WF |
751 | return nr_pages - work.nr_pages; |
752 | } | |
03ba3782 | 753 | |
b00949aa | 754 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
03ba3782 JA |
755 | { |
756 | unsigned long background_thresh, dirty_thresh; | |
757 | ||
16c4042f | 758 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 759 | |
b00949aa WF |
760 | if (global_page_state(NR_FILE_DIRTY) + |
761 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
762 | return true; | |
763 | ||
764 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > | |
765 | bdi_dirty_limit(bdi, background_thresh)) | |
766 | return true; | |
767 | ||
768 | return false; | |
03ba3782 JA |
769 | } |
770 | ||
e98be2d5 WF |
771 | /* |
772 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
773 | * only the flusher working on the first wb should do it. | |
774 | */ | |
775 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
776 | unsigned long start_time) | |
777 | { | |
af6a3113 | 778 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
779 | } |
780 | ||
03ba3782 JA |
781 | /* |
782 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 783 | * |
03ba3782 JA |
784 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
785 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
786 | * just walks the superblock inode list, writing back any inodes which are | |
787 | * older than a specific point in time. | |
66f3b8e2 | 788 | * |
03ba3782 JA |
789 | * Try to run once per dirty_writeback_interval. But if a writeback event |
790 | * takes longer than a dirty_writeback_interval interval, then leave a | |
791 | * one-second gap. | |
66f3b8e2 | 792 | * |
03ba3782 JA |
793 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
794 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 795 | */ |
c4a77a6c | 796 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 797 | struct wb_writeback_work *work) |
66f3b8e2 | 798 | { |
e98be2d5 | 799 | unsigned long wb_start = jiffies; |
d46db3d5 | 800 | long nr_pages = work->nr_pages; |
a5989bdc | 801 | struct inode *inode; |
d46db3d5 | 802 | long progress; |
66f3b8e2 | 803 | |
c4a391b5 JK |
804 | if (!work->older_than_this_is_set) { |
805 | work->older_than_this = jiffies; | |
806 | work->older_than_this_is_set = 1; | |
807 | } | |
38f21977 | 808 | |
e8dfc305 | 809 | spin_lock(&wb->list_lock); |
03ba3782 JA |
810 | for (;;) { |
811 | /* | |
d3ddec76 | 812 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 813 | */ |
83ba7b07 | 814 | if (work->nr_pages <= 0) |
03ba3782 | 815 | break; |
66f3b8e2 | 816 | |
aa373cf5 JK |
817 | /* |
818 | * Background writeout and kupdate-style writeback may | |
819 | * run forever. Stop them if there is other work to do | |
820 | * so that e.g. sync can proceed. They'll be restarted | |
821 | * after the other works are all done. | |
822 | */ | |
823 | if ((work->for_background || work->for_kupdate) && | |
824 | !list_empty(&wb->bdi->work_list)) | |
825 | break; | |
826 | ||
38f21977 | 827 | /* |
d3ddec76 WF |
828 | * For background writeout, stop when we are below the |
829 | * background dirty threshold | |
38f21977 | 830 | */ |
b00949aa | 831 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
03ba3782 | 832 | break; |
38f21977 | 833 | |
1bc36b64 JK |
834 | /* |
835 | * Kupdate and background works are special and we want to | |
836 | * include all inodes that need writing. Livelock avoidance is | |
837 | * handled by these works yielding to any other work so we are | |
838 | * safe. | |
839 | */ | |
ba9aa839 | 840 | if (work->for_kupdate) { |
c4a391b5 | 841 | work->older_than_this = jiffies - |
ba9aa839 | 842 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 843 | } else if (work->for_background) |
c4a391b5 | 844 | work->older_than_this = jiffies; |
028c2dd1 | 845 | |
d46db3d5 | 846 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 847 | if (list_empty(&wb->b_io)) |
ad4e38dd | 848 | queue_io(wb, work); |
83ba7b07 | 849 | if (work->sb) |
d46db3d5 | 850 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 851 | else |
d46db3d5 WF |
852 | progress = __writeback_inodes_wb(wb, work); |
853 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 854 | |
e98be2d5 | 855 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
856 | |
857 | /* | |
e6fb6da2 WF |
858 | * Did we write something? Try for more |
859 | * | |
860 | * Dirty inodes are moved to b_io for writeback in batches. | |
861 | * The completion of the current batch does not necessarily | |
862 | * mean the overall work is done. So we keep looping as long | |
863 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 864 | */ |
d46db3d5 | 865 | if (progress) |
71fd05a8 JA |
866 | continue; |
867 | /* | |
e6fb6da2 | 868 | * No more inodes for IO, bail |
71fd05a8 | 869 | */ |
b7a2441f | 870 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 871 | break; |
71fd05a8 JA |
872 | /* |
873 | * Nothing written. Wait for some inode to | |
874 | * become available for writeback. Otherwise | |
875 | * we'll just busyloop. | |
876 | */ | |
71fd05a8 | 877 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 878 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 879 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 880 | spin_lock(&inode->i_lock); |
f0d07b7f | 881 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
882 | /* This function drops i_lock... */ |
883 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 884 | spin_lock(&wb->list_lock); |
03ba3782 JA |
885 | } |
886 | } | |
e8dfc305 | 887 | spin_unlock(&wb->list_lock); |
03ba3782 | 888 | |
d46db3d5 | 889 | return nr_pages - work->nr_pages; |
03ba3782 JA |
890 | } |
891 | ||
892 | /* | |
83ba7b07 | 893 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 894 | */ |
83ba7b07 | 895 | static struct wb_writeback_work * |
08852b6d | 896 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 897 | { |
83ba7b07 | 898 | struct wb_writeback_work *work = NULL; |
03ba3782 | 899 | |
6467716a | 900 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
901 | if (!list_empty(&bdi->work_list)) { |
902 | work = list_entry(bdi->work_list.next, | |
903 | struct wb_writeback_work, list); | |
904 | list_del_init(&work->list); | |
03ba3782 | 905 | } |
6467716a | 906 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 907 | return work; |
03ba3782 JA |
908 | } |
909 | ||
cdf01dd5 LT |
910 | /* |
911 | * Add in the number of potentially dirty inodes, because each inode | |
912 | * write can dirty pagecache in the underlying blockdev. | |
913 | */ | |
914 | static unsigned long get_nr_dirty_pages(void) | |
915 | { | |
916 | return global_page_state(NR_FILE_DIRTY) + | |
917 | global_page_state(NR_UNSTABLE_NFS) + | |
918 | get_nr_dirty_inodes(); | |
919 | } | |
920 | ||
6585027a JK |
921 | static long wb_check_background_flush(struct bdi_writeback *wb) |
922 | { | |
b00949aa | 923 | if (over_bground_thresh(wb->bdi)) { |
6585027a JK |
924 | |
925 | struct wb_writeback_work work = { | |
926 | .nr_pages = LONG_MAX, | |
927 | .sync_mode = WB_SYNC_NONE, | |
928 | .for_background = 1, | |
929 | .range_cyclic = 1, | |
0e175a18 | 930 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
931 | }; |
932 | ||
933 | return wb_writeback(wb, &work); | |
934 | } | |
935 | ||
936 | return 0; | |
937 | } | |
938 | ||
03ba3782 JA |
939 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
940 | { | |
941 | unsigned long expired; | |
942 | long nr_pages; | |
943 | ||
69b62d01 JA |
944 | /* |
945 | * When set to zero, disable periodic writeback | |
946 | */ | |
947 | if (!dirty_writeback_interval) | |
948 | return 0; | |
949 | ||
03ba3782 JA |
950 | expired = wb->last_old_flush + |
951 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
952 | if (time_before(jiffies, expired)) | |
953 | return 0; | |
954 | ||
955 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 956 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 957 | |
c4a77a6c | 958 | if (nr_pages) { |
83ba7b07 | 959 | struct wb_writeback_work work = { |
c4a77a6c JA |
960 | .nr_pages = nr_pages, |
961 | .sync_mode = WB_SYNC_NONE, | |
962 | .for_kupdate = 1, | |
963 | .range_cyclic = 1, | |
0e175a18 | 964 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
965 | }; |
966 | ||
83ba7b07 | 967 | return wb_writeback(wb, &work); |
c4a77a6c | 968 | } |
03ba3782 JA |
969 | |
970 | return 0; | |
971 | } | |
972 | ||
973 | /* | |
974 | * Retrieve work items and do the writeback they describe | |
975 | */ | |
25d130ba | 976 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 JA |
977 | { |
978 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 979 | struct wb_writeback_work *work; |
c4a77a6c | 980 | long wrote = 0; |
03ba3782 | 981 | |
81d73a32 | 982 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 983 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 | 984 | |
455b2864 DC |
985 | trace_writeback_exec(bdi, work); |
986 | ||
83ba7b07 | 987 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
988 | |
989 | /* | |
83ba7b07 CH |
990 | * Notify the caller of completion if this is a synchronous |
991 | * work item, otherwise just free it. | |
03ba3782 | 992 | */ |
83ba7b07 CH |
993 | if (work->done) |
994 | complete(work->done); | |
995 | else | |
996 | kfree(work); | |
03ba3782 JA |
997 | } |
998 | ||
999 | /* | |
1000 | * Check for periodic writeback, kupdated() style | |
1001 | */ | |
1002 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1003 | wrote += wb_check_background_flush(wb); |
81d73a32 | 1004 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
1005 | |
1006 | return wrote; | |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1011 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1012 | */ |
839a8e86 | 1013 | void bdi_writeback_workfn(struct work_struct *work) |
03ba3782 | 1014 | { |
839a8e86 TH |
1015 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1016 | struct bdi_writeback, dwork); | |
08243900 | 1017 | struct backing_dev_info *bdi = wb->bdi; |
03ba3782 JA |
1018 | long pages_written; |
1019 | ||
ef3b1019 | 1020 | set_worker_desc("flush-%s", dev_name(bdi->dev)); |
766f9164 | 1021 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1022 | |
839a8e86 TH |
1023 | if (likely(!current_is_workqueue_rescuer() || |
1024 | list_empty(&bdi->bdi_list))) { | |
6467716a | 1025 | /* |
839a8e86 TH |
1026 | * The normal path. Keep writing back @bdi until its |
1027 | * work_list is empty. Note that this path is also taken | |
1028 | * if @bdi is shutting down even when we're running off the | |
1029 | * rescuer as work_list needs to be drained. | |
6467716a | 1030 | */ |
839a8e86 | 1031 | do { |
25d130ba | 1032 | pages_written = wb_do_writeback(wb); |
839a8e86 TH |
1033 | trace_writeback_pages_written(pages_written); |
1034 | } while (!list_empty(&bdi->work_list)); | |
1035 | } else { | |
1036 | /* | |
1037 | * bdi_wq can't get enough workers and we're running off | |
1038 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1039 | * enough for efficient IO. | |
1040 | */ | |
1041 | pages_written = writeback_inodes_wb(&bdi->wb, 1024, | |
1042 | WB_REASON_FORKER_THREAD); | |
455b2864 | 1043 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1044 | } |
1045 | ||
839a8e86 TH |
1046 | if (!list_empty(&bdi->work_list) || |
1047 | (wb_has_dirty_io(wb) && dirty_writeback_interval)) | |
1048 | queue_delayed_work(bdi_wq, &wb->dwork, | |
1049 | msecs_to_jiffies(dirty_writeback_interval * 10)); | |
455b2864 | 1050 | |
839a8e86 | 1051 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1052 | } |
1053 | ||
1054 | /* | |
b8c2f347 CH |
1055 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1056 | * the whole world. | |
03ba3782 | 1057 | */ |
0e175a18 | 1058 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1059 | { |
b8c2f347 | 1060 | struct backing_dev_info *bdi; |
03ba3782 | 1061 | |
47df3dde JK |
1062 | if (!nr_pages) |
1063 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1064 | |
b8c2f347 | 1065 | rcu_read_lock(); |
cfc4ba53 | 1066 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1067 | if (!bdi_has_dirty_io(bdi)) |
1068 | continue; | |
0e175a18 | 1069 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
03ba3782 | 1070 | } |
cfc4ba53 | 1071 | rcu_read_unlock(); |
1da177e4 LT |
1072 | } |
1073 | ||
03ba3782 JA |
1074 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1075 | { | |
1076 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1077 | struct dentry *dentry; | |
1078 | const char *name = "?"; | |
1079 | ||
1080 | dentry = d_find_alias(inode); | |
1081 | if (dentry) { | |
1082 | spin_lock(&dentry->d_lock); | |
1083 | name = (const char *) dentry->d_name.name; | |
1084 | } | |
1085 | printk(KERN_DEBUG | |
1086 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1087 | current->comm, task_pid_nr(current), inode->i_ino, | |
1088 | name, inode->i_sb->s_id); | |
1089 | if (dentry) { | |
1090 | spin_unlock(&dentry->d_lock); | |
1091 | dput(dentry); | |
1092 | } | |
1093 | } | |
1094 | } | |
1095 | ||
1096 | /** | |
1097 | * __mark_inode_dirty - internal function | |
1098 | * @inode: inode to mark | |
1099 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1100 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1101 | * mark_inode_dirty_sync. | |
1da177e4 | 1102 | * |
03ba3782 JA |
1103 | * Put the inode on the super block's dirty list. |
1104 | * | |
1105 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1106 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1107 | * If it was not hashed, it will never be added to the dirty list | |
1108 | * even if it is later hashed, as it will have been marked dirty already. | |
1109 | * | |
1110 | * In short, make sure you hash any inodes _before_ you start marking | |
1111 | * them dirty. | |
1da177e4 | 1112 | * |
03ba3782 JA |
1113 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1114 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1115 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1116 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1117 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1118 | * blockdev inode. | |
1da177e4 | 1119 | */ |
03ba3782 | 1120 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1121 | { |
03ba3782 | 1122 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1123 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1124 | |
03ba3782 JA |
1125 | /* |
1126 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1127 | * dirty the inode itself | |
1128 | */ | |
1129 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
9fb0a7da TH |
1130 | trace_writeback_dirty_inode_start(inode, flags); |
1131 | ||
03ba3782 | 1132 | if (sb->s_op->dirty_inode) |
aa385729 | 1133 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1134 | |
1135 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 JA |
1136 | } |
1137 | ||
1138 | /* | |
1139 | * make sure that changes are seen by all cpus before we test i_state | |
1140 | * -- mikulas | |
1141 | */ | |
1142 | smp_mb(); | |
1143 | ||
1144 | /* avoid the locking if we can */ | |
1145 | if ((inode->i_state & flags) == flags) | |
1146 | return; | |
1147 | ||
1148 | if (unlikely(block_dump)) | |
1149 | block_dump___mark_inode_dirty(inode); | |
1150 | ||
250df6ed | 1151 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1152 | if ((inode->i_state & flags) != flags) { |
1153 | const int was_dirty = inode->i_state & I_DIRTY; | |
1154 | ||
1155 | inode->i_state |= flags; | |
1156 | ||
1157 | /* | |
1158 | * If the inode is being synced, just update its dirty state. | |
1159 | * The unlocker will place the inode on the appropriate | |
1160 | * superblock list, based upon its state. | |
1161 | */ | |
1162 | if (inode->i_state & I_SYNC) | |
250df6ed | 1163 | goto out_unlock_inode; |
03ba3782 JA |
1164 | |
1165 | /* | |
1166 | * Only add valid (hashed) inodes to the superblock's | |
1167 | * dirty list. Add blockdev inodes as well. | |
1168 | */ | |
1169 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1170 | if (inode_unhashed(inode)) |
250df6ed | 1171 | goto out_unlock_inode; |
03ba3782 | 1172 | } |
a4ffdde6 | 1173 | if (inode->i_state & I_FREEING) |
250df6ed | 1174 | goto out_unlock_inode; |
03ba3782 JA |
1175 | |
1176 | /* | |
1177 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1178 | * reposition it (that would break b_dirty time-ordering). | |
1179 | */ | |
1180 | if (!was_dirty) { | |
a66979ab | 1181 | bool wakeup_bdi = false; |
253c34e9 AB |
1182 | bdi = inode_to_bdi(inode); |
1183 | ||
146d7009 JB |
1184 | spin_unlock(&inode->i_lock); |
1185 | spin_lock(&bdi->wb.list_lock); | |
253c34e9 AB |
1186 | if (bdi_cap_writeback_dirty(bdi)) { |
1187 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1188 | "bdi-%s not registered\n", bdi->name); | |
1189 | ||
1190 | /* | |
1191 | * If this is the first dirty inode for this | |
1192 | * bdi, we have to wake-up the corresponding | |
1193 | * bdi thread to make sure background | |
1194 | * write-back happens later. | |
1195 | */ | |
1196 | if (!wb_has_dirty_io(&bdi->wb)) | |
1197 | wakeup_bdi = true; | |
500b067c | 1198 | } |
03ba3782 JA |
1199 | |
1200 | inode->dirtied_when = jiffies; | |
7ccf19a8 | 1201 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1202 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1203 | |
1204 | if (wakeup_bdi) | |
1205 | bdi_wakeup_thread_delayed(bdi); | |
1206 | return; | |
1da177e4 | 1207 | } |
1da177e4 | 1208 | } |
250df6ed DC |
1209 | out_unlock_inode: |
1210 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1211 | |
03ba3782 JA |
1212 | } |
1213 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1214 | ||
b6e51316 | 1215 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1216 | { |
1217 | struct inode *inode, *old_inode = NULL; | |
1218 | ||
1219 | /* | |
1220 | * We need to be protected against the filesystem going from | |
1221 | * r/o to r/w or vice versa. | |
1222 | */ | |
b6e51316 | 1223 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1224 | |
55fa6091 | 1225 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1226 | |
1227 | /* | |
1228 | * Data integrity sync. Must wait for all pages under writeback, | |
1229 | * because there may have been pages dirtied before our sync | |
1230 | * call, but which had writeout started before we write it out. | |
1231 | * In which case, the inode may not be on the dirty list, but | |
1232 | * we still have to wait for that writeout. | |
1233 | */ | |
b6e51316 | 1234 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1235 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1236 | |
250df6ed DC |
1237 | spin_lock(&inode->i_lock); |
1238 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1239 | (mapping->nrpages == 0)) { | |
1240 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1241 | continue; |
250df6ed | 1242 | } |
03ba3782 | 1243 | __iget(inode); |
250df6ed | 1244 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1245 | spin_unlock(&inode_sb_list_lock); |
1246 | ||
03ba3782 | 1247 | /* |
55fa6091 DC |
1248 | * We hold a reference to 'inode' so it couldn't have been |
1249 | * removed from s_inodes list while we dropped the | |
1250 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1251 | * be holding the last reference and we cannot iput it under | |
1252 | * inode_sb_list_lock. So we keep the reference and iput it | |
1253 | * later. | |
03ba3782 JA |
1254 | */ |
1255 | iput(old_inode); | |
1256 | old_inode = inode; | |
1257 | ||
1258 | filemap_fdatawait(mapping); | |
1259 | ||
1260 | cond_resched(); | |
1261 | ||
55fa6091 | 1262 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1263 | } |
55fa6091 | 1264 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1265 | iput(old_inode); |
1da177e4 LT |
1266 | } |
1267 | ||
d8a8559c | 1268 | /** |
3259f8be | 1269 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1270 | * @sb: the superblock |
3259f8be | 1271 | * @nr: the number of pages to write |
786228ab | 1272 | * @reason: reason why some writeback work initiated |
1da177e4 | 1273 | * |
d8a8559c JA |
1274 | * Start writeback on some inodes on this super_block. No guarantees are made |
1275 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1276 | * for IO completion of submitted IO. |
1da177e4 | 1277 | */ |
0e175a18 CW |
1278 | void writeback_inodes_sb_nr(struct super_block *sb, |
1279 | unsigned long nr, | |
1280 | enum wb_reason reason) | |
1da177e4 | 1281 | { |
83ba7b07 CH |
1282 | DECLARE_COMPLETION_ONSTACK(done); |
1283 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1284 | .sb = sb, |
1285 | .sync_mode = WB_SYNC_NONE, | |
1286 | .tagged_writepages = 1, | |
1287 | .done = &done, | |
1288 | .nr_pages = nr, | |
0e175a18 | 1289 | .reason = reason, |
3c4d7165 | 1290 | }; |
d8a8559c | 1291 | |
6eedc701 JK |
1292 | if (sb->s_bdi == &noop_backing_dev_info) |
1293 | return; | |
cf37e972 | 1294 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1295 | bdi_queue_work(sb->s_bdi, &work); |
1296 | wait_for_completion(&done); | |
e913fc82 | 1297 | } |
3259f8be CM |
1298 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1299 | ||
1300 | /** | |
1301 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1302 | * @sb: the superblock | |
786228ab | 1303 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1304 | * |
1305 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1306 | * on how many (if any) will be written, and this function does not wait | |
1307 | * for IO completion of submitted IO. | |
1308 | */ | |
0e175a18 | 1309 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1310 | { |
0e175a18 | 1311 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1312 | } |
0e3c9a22 | 1313 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1314 | |
17bd55d0 | 1315 | /** |
10ee27a0 | 1316 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1317 | * @sb: the superblock |
10ee27a0 MX |
1318 | * @nr: the number of pages to write |
1319 | * @reason: the reason of writeback | |
17bd55d0 | 1320 | * |
10ee27a0 | 1321 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1322 | * Returns 1 if writeback was started, 0 if not. |
1323 | */ | |
10ee27a0 MX |
1324 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
1325 | unsigned long nr, | |
1326 | enum wb_reason reason) | |
17bd55d0 | 1327 | { |
10ee27a0 | 1328 | if (writeback_in_progress(sb->s_bdi)) |
17bd55d0 | 1329 | return 1; |
10ee27a0 MX |
1330 | |
1331 | if (!down_read_trylock(&sb->s_umount)) | |
17bd55d0 | 1332 | return 0; |
10ee27a0 MX |
1333 | |
1334 | writeback_inodes_sb_nr(sb, nr, reason); | |
1335 | up_read(&sb->s_umount); | |
1336 | return 1; | |
17bd55d0 | 1337 | } |
10ee27a0 | 1338 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1339 | |
3259f8be | 1340 | /** |
10ee27a0 | 1341 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1342 | * @sb: the superblock |
786228ab | 1343 | * @reason: reason why some writeback work was initiated |
3259f8be | 1344 | * |
10ee27a0 | 1345 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1346 | * Returns 1 if writeback was started, 0 if not. |
1347 | */ | |
10ee27a0 | 1348 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1349 | { |
10ee27a0 | 1350 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1351 | } |
10ee27a0 | 1352 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1353 | |
d8a8559c JA |
1354 | /** |
1355 | * sync_inodes_sb - sync sb inode pages | |
c4a391b5 JK |
1356 | * @sb: the superblock |
1357 | * @older_than_this: timestamp | |
d8a8559c JA |
1358 | * |
1359 | * This function writes and waits on any dirty inode belonging to this | |
c4a391b5 | 1360 | * superblock that has been dirtied before given timestamp. |
d8a8559c | 1361 | */ |
c4a391b5 | 1362 | void sync_inodes_sb(struct super_block *sb, unsigned long older_than_this) |
d8a8559c | 1363 | { |
83ba7b07 CH |
1364 | DECLARE_COMPLETION_ONSTACK(done); |
1365 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1366 | .sb = sb, |
1367 | .sync_mode = WB_SYNC_ALL, | |
1368 | .nr_pages = LONG_MAX, | |
c4a391b5 JK |
1369 | .older_than_this = older_than_this, |
1370 | .older_than_this_is_set = 1, | |
3c4d7165 | 1371 | .range_cyclic = 0, |
83ba7b07 | 1372 | .done = &done, |
0e175a18 | 1373 | .reason = WB_REASON_SYNC, |
7747bd4b | 1374 | .for_sync = 1, |
3c4d7165 CH |
1375 | }; |
1376 | ||
6eedc701 JK |
1377 | /* Nothing to do? */ |
1378 | if (sb->s_bdi == &noop_backing_dev_info) | |
1379 | return; | |
cf37e972 CH |
1380 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1381 | ||
83ba7b07 CH |
1382 | bdi_queue_work(sb->s_bdi, &work); |
1383 | wait_for_completion(&done); | |
1384 | ||
b6e51316 | 1385 | wait_sb_inodes(sb); |
1da177e4 | 1386 | } |
d8a8559c | 1387 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1388 | |
1da177e4 | 1389 | /** |
7f04c26d AA |
1390 | * write_inode_now - write an inode to disk |
1391 | * @inode: inode to write to disk | |
1392 | * @sync: whether the write should be synchronous or not | |
1393 | * | |
1394 | * This function commits an inode to disk immediately if it is dirty. This is | |
1395 | * primarily needed by knfsd. | |
1da177e4 | 1396 | * |
7f04c26d | 1397 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1398 | */ |
1da177e4 LT |
1399 | int write_inode_now(struct inode *inode, int sync) |
1400 | { | |
f758eeab | 1401 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1402 | struct writeback_control wbc = { |
1403 | .nr_to_write = LONG_MAX, | |
18914b18 | 1404 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1405 | .range_start = 0, |
1406 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1407 | }; |
1408 | ||
1409 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1410 | wbc.nr_to_write = 0; |
1da177e4 LT |
1411 | |
1412 | might_sleep(); | |
4f8ad655 | 1413 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1414 | } |
1415 | EXPORT_SYMBOL(write_inode_now); | |
1416 | ||
1417 | /** | |
1418 | * sync_inode - write an inode and its pages to disk. | |
1419 | * @inode: the inode to sync | |
1420 | * @wbc: controls the writeback mode | |
1421 | * | |
1422 | * sync_inode() will write an inode and its pages to disk. It will also | |
1423 | * correctly update the inode on its superblock's dirty inode lists and will | |
1424 | * update inode->i_state. | |
1425 | * | |
1426 | * The caller must have a ref on the inode. | |
1427 | */ | |
1428 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1429 | { | |
4f8ad655 | 1430 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1431 | } |
1432 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1433 | |
1434 | /** | |
c691b9d9 | 1435 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1436 | * @inode: the inode to sync |
1437 | * @wait: wait for I/O to complete. | |
1438 | * | |
c691b9d9 | 1439 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1440 | * |
1441 | * Note: only writes the actual inode, no associated data or other metadata. | |
1442 | */ | |
1443 | int sync_inode_metadata(struct inode *inode, int wait) | |
1444 | { | |
1445 | struct writeback_control wbc = { | |
1446 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1447 | .nr_to_write = 0, /* metadata-only */ | |
1448 | }; | |
1449 | ||
1450 | return sync_inode(inode, &wbc); | |
1451 | } | |
1452 | EXPORT_SYMBOL(sync_inode_metadata); |