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> | |
f5ff8422 | 17 | #include <linux/module.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> | |
03ba3782 JA |
23 | #include <linux/kthread.h> |
24 | #include <linux/freezer.h> | |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/buffer_head.h> | |
07f3f05c | 29 | #include "internal.h" |
1da177e4 | 30 | |
66f3b8e2 | 31 | #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) |
f11b00f3 | 32 | |
d0bceac7 JA |
33 | /* |
34 | * We don't actually have pdflush, but this one is exported though /proc... | |
35 | */ | |
36 | int nr_pdflush_threads; | |
37 | ||
c4a77a6c JA |
38 | /* |
39 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
40 | */ | |
83ba7b07 | 41 | struct wb_writeback_work { |
c4a77a6c JA |
42 | long nr_pages; |
43 | struct super_block *sb; | |
44 | enum writeback_sync_modes sync_mode; | |
52957fe1 HS |
45 | unsigned int for_kupdate:1; |
46 | unsigned int range_cyclic:1; | |
47 | unsigned int for_background:1; | |
c4a77a6c | 48 | |
8010c3b6 | 49 | struct list_head list; /* pending work list */ |
83ba7b07 | 50 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
51 | }; |
52 | ||
f11b00f3 AB |
53 | /** |
54 | * writeback_in_progress - determine whether there is writeback in progress | |
55 | * @bdi: the device's backing_dev_info structure. | |
56 | * | |
03ba3782 JA |
57 | * Determine whether there is writeback waiting to be handled against a |
58 | * backing device. | |
f11b00f3 AB |
59 | */ |
60 | int writeback_in_progress(struct backing_dev_info *bdi) | |
61 | { | |
03ba3782 | 62 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
63 | } |
64 | ||
83ba7b07 CH |
65 | static void bdi_queue_work(struct backing_dev_info *bdi, |
66 | struct wb_writeback_work *work) | |
03ba3782 | 67 | { |
bcddc3f0 | 68 | spin_lock(&bdi->wb_lock); |
83ba7b07 | 69 | list_add_tail(&work->list, &bdi->work_list); |
bcddc3f0 | 70 | spin_unlock(&bdi->wb_lock); |
03ba3782 JA |
71 | |
72 | /* | |
73 | * If the default thread isn't there, make sure we add it. When | |
74 | * it gets created and wakes up, we'll run this work. | |
75 | */ | |
76 | if (unlikely(list_empty_careful(&bdi->wb_list))) | |
77 | wake_up_process(default_backing_dev_info.wb.task); | |
78 | else { | |
79 | struct bdi_writeback *wb = &bdi->wb; | |
1da177e4 | 80 | |
1ef7d9aa | 81 | if (wb->task) |
03ba3782 | 82 | wake_up_process(wb->task); |
1da177e4 | 83 | } |
1da177e4 LT |
84 | } |
85 | ||
83ba7b07 CH |
86 | static void |
87 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
88 | bool range_cyclic, bool for_background) | |
1da177e4 | 89 | { |
83ba7b07 | 90 | struct wb_writeback_work *work; |
03ba3782 | 91 | |
bcddc3f0 JA |
92 | /* |
93 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
94 | * wakeup the thread for old dirty data writeback | |
95 | */ | |
83ba7b07 CH |
96 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
97 | if (!work) { | |
98 | if (bdi->wb.task) | |
99 | wake_up_process(bdi->wb.task); | |
100 | return; | |
bcddc3f0 | 101 | } |
03ba3782 | 102 | |
83ba7b07 CH |
103 | work->sync_mode = WB_SYNC_NONE; |
104 | work->nr_pages = nr_pages; | |
105 | work->range_cyclic = range_cyclic; | |
106 | work->for_background = for_background; | |
03ba3782 | 107 | |
83ba7b07 | 108 | bdi_queue_work(bdi, work); |
b6e51316 JA |
109 | } |
110 | ||
111 | /** | |
112 | * bdi_start_writeback - start writeback | |
113 | * @bdi: the backing device to write from | |
114 | * @nr_pages: the number of pages to write | |
115 | * | |
116 | * Description: | |
117 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
118 | * started when this function returns, we make no guarentees on | |
0e3c9a22 | 119 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
120 | * |
121 | */ | |
c5444198 | 122 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 123 | { |
83ba7b07 | 124 | __bdi_start_writeback(bdi, nr_pages, true, false); |
c5444198 | 125 | } |
d3ddec76 | 126 | |
c5444198 CH |
127 | /** |
128 | * bdi_start_background_writeback - start background writeback | |
129 | * @bdi: the backing device to write from | |
130 | * | |
131 | * Description: | |
132 | * This does WB_SYNC_NONE background writeback. The IO is only | |
133 | * started when this function returns, we make no guarentees on | |
134 | * completion. Caller need not hold sb s_umount semaphore. | |
135 | */ | |
136 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
137 | { | |
83ba7b07 | 138 | __bdi_start_writeback(bdi, LONG_MAX, true, true); |
1da177e4 LT |
139 | } |
140 | ||
6610a0bc AM |
141 | /* |
142 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
143 | * furthest end of its superblock's dirty-inode list. | |
144 | * | |
145 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 146 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
147 | * the case then the inode must have been redirtied while it was being written |
148 | * out and we don't reset its dirtied_when. | |
149 | */ | |
150 | static void redirty_tail(struct inode *inode) | |
151 | { | |
03ba3782 | 152 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 153 | |
03ba3782 | 154 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 155 | struct inode *tail; |
6610a0bc | 156 | |
03ba3782 | 157 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 158 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
159 | inode->dirtied_when = jiffies; |
160 | } | |
03ba3782 | 161 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
162 | } |
163 | ||
c986d1e2 | 164 | /* |
66f3b8e2 | 165 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 166 | */ |
0e0f4fc2 | 167 | static void requeue_io(struct inode *inode) |
c986d1e2 | 168 | { |
03ba3782 JA |
169 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
170 | ||
171 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
172 | } |
173 | ||
1c0eeaf5 JE |
174 | static void inode_sync_complete(struct inode *inode) |
175 | { | |
176 | /* | |
177 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
178 | */ | |
179 | smp_mb(); | |
180 | wake_up_bit(&inode->i_state, __I_SYNC); | |
181 | } | |
182 | ||
d2caa3c5 JL |
183 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
184 | { | |
185 | bool ret = time_after(inode->dirtied_when, t); | |
186 | #ifndef CONFIG_64BIT | |
187 | /* | |
188 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
189 | * It _appears_ to be in the future, but is actually in distant past. | |
190 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 191 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
192 | */ |
193 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
194 | #endif | |
195 | return ret; | |
196 | } | |
197 | ||
2c136579 FW |
198 | /* |
199 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
200 | */ | |
201 | static void move_expired_inodes(struct list_head *delaying_queue, | |
202 | struct list_head *dispatch_queue, | |
203 | unsigned long *older_than_this) | |
204 | { | |
5c03449d SL |
205 | LIST_HEAD(tmp); |
206 | struct list_head *pos, *node; | |
cf137307 | 207 | struct super_block *sb = NULL; |
5c03449d | 208 | struct inode *inode; |
cf137307 | 209 | int do_sb_sort = 0; |
5c03449d | 210 | |
2c136579 | 211 | while (!list_empty(delaying_queue)) { |
5c03449d | 212 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
2c136579 | 213 | if (older_than_this && |
d2caa3c5 | 214 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 215 | break; |
cf137307 JA |
216 | if (sb && sb != inode->i_sb) |
217 | do_sb_sort = 1; | |
218 | sb = inode->i_sb; | |
5c03449d SL |
219 | list_move(&inode->i_list, &tmp); |
220 | } | |
221 | ||
cf137307 JA |
222 | /* just one sb in list, splice to dispatch_queue and we're done */ |
223 | if (!do_sb_sort) { | |
224 | list_splice(&tmp, dispatch_queue); | |
225 | return; | |
226 | } | |
227 | ||
5c03449d SL |
228 | /* Move inodes from one superblock together */ |
229 | while (!list_empty(&tmp)) { | |
230 | inode = list_entry(tmp.prev, struct inode, i_list); | |
231 | sb = inode->i_sb; | |
232 | list_for_each_prev_safe(pos, node, &tmp) { | |
233 | inode = list_entry(pos, struct inode, i_list); | |
234 | if (inode->i_sb == sb) | |
235 | list_move(&inode->i_list, dispatch_queue); | |
236 | } | |
2c136579 FW |
237 | } |
238 | } | |
239 | ||
240 | /* | |
241 | * Queue all expired dirty inodes for io, eldest first. | |
242 | */ | |
03ba3782 | 243 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 244 | { |
03ba3782 JA |
245 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
246 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
247 | } |
248 | ||
a9185b41 | 249 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 250 | { |
03ba3782 | 251 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 252 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 253 | return 0; |
08d8e974 | 254 | } |
08d8e974 | 255 | |
1da177e4 | 256 | /* |
01c03194 CH |
257 | * Wait for writeback on an inode to complete. |
258 | */ | |
259 | static void inode_wait_for_writeback(struct inode *inode) | |
260 | { | |
261 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
262 | wait_queue_head_t *wqh; | |
263 | ||
264 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
58a9d3d8 | 265 | while (inode->i_state & I_SYNC) { |
01c03194 CH |
266 | spin_unlock(&inode_lock); |
267 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
268 | spin_lock(&inode_lock); | |
58a9d3d8 | 269 | } |
01c03194 CH |
270 | } |
271 | ||
272 | /* | |
273 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
274 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
275 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
276 | * | |
1da177e4 LT |
277 | * If `wait' is set, wait on the writeout. |
278 | * | |
279 | * The whole writeout design is quite complex and fragile. We want to avoid | |
280 | * starvation of particular inodes when others are being redirtied, prevent | |
281 | * livelocks, etc. | |
282 | * | |
283 | * Called under inode_lock. | |
284 | */ | |
285 | static int | |
01c03194 | 286 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 287 | { |
1da177e4 | 288 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 289 | unsigned dirty; |
1da177e4 LT |
290 | int ret; |
291 | ||
01c03194 CH |
292 | if (!atomic_read(&inode->i_count)) |
293 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
294 | else | |
295 | WARN_ON(inode->i_state & I_WILL_FREE); | |
296 | ||
297 | if (inode->i_state & I_SYNC) { | |
298 | /* | |
299 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 300 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
301 | * writeback can proceed with the other inodes on s_io. |
302 | * | |
303 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 304 | * completed a full scan of b_io. |
01c03194 | 305 | */ |
a9185b41 | 306 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
307 | requeue_io(inode); |
308 | return 0; | |
309 | } | |
310 | ||
311 | /* | |
312 | * It's a data-integrity sync. We must wait. | |
313 | */ | |
314 | inode_wait_for_writeback(inode); | |
315 | } | |
316 | ||
1c0eeaf5 | 317 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 318 | |
5547e8aa | 319 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 320 | inode->i_state |= I_SYNC; |
5547e8aa | 321 | inode->i_state &= ~I_DIRTY_PAGES; |
1da177e4 LT |
322 | spin_unlock(&inode_lock); |
323 | ||
324 | ret = do_writepages(mapping, wbc); | |
325 | ||
26821ed4 CH |
326 | /* |
327 | * Make sure to wait on the data before writing out the metadata. | |
328 | * This is important for filesystems that modify metadata on data | |
329 | * I/O completion. | |
330 | */ | |
a9185b41 | 331 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 332 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
333 | if (ret == 0) |
334 | ret = err; | |
335 | } | |
336 | ||
5547e8aa DM |
337 | /* |
338 | * Some filesystems may redirty the inode during the writeback | |
339 | * due to delalloc, clear dirty metadata flags right before | |
340 | * write_inode() | |
341 | */ | |
342 | spin_lock(&inode_lock); | |
343 | dirty = inode->i_state & I_DIRTY; | |
344 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
345 | spin_unlock(&inode_lock); | |
26821ed4 CH |
346 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
347 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 348 | int err = write_inode(inode, wbc); |
1da177e4 LT |
349 | if (ret == 0) |
350 | ret = err; | |
351 | } | |
352 | ||
353 | spin_lock(&inode_lock); | |
1c0eeaf5 | 354 | inode->i_state &= ~I_SYNC; |
84a89245 | 355 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
b3af9468 | 356 | if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { |
ae1b7f7d | 357 | /* |
b3af9468 WF |
358 | * More pages get dirtied by a fast dirtier. |
359 | */ | |
360 | goto select_queue; | |
361 | } else if (inode->i_state & I_DIRTY) { | |
362 | /* | |
363 | * At least XFS will redirty the inode during the | |
364 | * writeback (delalloc) and on io completion (isize). | |
ae1b7f7d WF |
365 | */ |
366 | redirty_tail(inode); | |
367 | } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
1da177e4 LT |
368 | /* |
369 | * We didn't write back all the pages. nfs_writepages() | |
370 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 371 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
372 | */ |
373 | /* | |
374 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 375 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
376 | * consideration. Otherwise, move it to the tail, for |
377 | * the reasons described there. I'm not really sure | |
378 | * how much sense this makes. Presumably I had a good | |
379 | * reasons for doing it this way, and I'd rather not | |
380 | * muck with it at present. | |
1da177e4 LT |
381 | */ |
382 | if (wbc->for_kupdate) { | |
383 | /* | |
2c136579 | 384 | * For the kupdate function we move the inode |
66f3b8e2 | 385 | * to b_more_io so it will get more writeout as |
2c136579 | 386 | * soon as the queue becomes uncongested. |
1da177e4 LT |
387 | */ |
388 | inode->i_state |= I_DIRTY_PAGES; | |
b3af9468 | 389 | select_queue: |
8bc3be27 FW |
390 | if (wbc->nr_to_write <= 0) { |
391 | /* | |
392 | * slice used up: queue for next turn | |
393 | */ | |
394 | requeue_io(inode); | |
395 | } else { | |
396 | /* | |
397 | * somehow blocked: retry later | |
398 | */ | |
399 | redirty_tail(inode); | |
400 | } | |
1da177e4 LT |
401 | } else { |
402 | /* | |
403 | * Otherwise fully redirty the inode so that | |
404 | * other inodes on this superblock will get some | |
405 | * writeout. Otherwise heavy writing to one | |
406 | * file would indefinitely suspend writeout of | |
407 | * all the other files. | |
408 | */ | |
409 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 410 | redirty_tail(inode); |
1da177e4 | 411 | } |
1da177e4 LT |
412 | } else if (atomic_read(&inode->i_count)) { |
413 | /* | |
414 | * The inode is clean, inuse | |
415 | */ | |
416 | list_move(&inode->i_list, &inode_in_use); | |
417 | } else { | |
418 | /* | |
419 | * The inode is clean, unused | |
420 | */ | |
421 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
422 | } |
423 | } | |
1c0eeaf5 | 424 | inode_sync_complete(inode); |
1da177e4 LT |
425 | return ret; |
426 | } | |
427 | ||
03ba3782 | 428 | /* |
d19de7ed | 429 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
430 | * before calling writeback. So make sure that we do pin it, so it doesn't |
431 | * go away while we are writing inodes from it. | |
03ba3782 | 432 | */ |
d19de7ed | 433 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 434 | { |
03ba3782 | 435 | spin_lock(&sb_lock); |
29cb4859 CH |
436 | if (list_empty(&sb->s_instances)) { |
437 | spin_unlock(&sb_lock); | |
438 | return false; | |
439 | } | |
440 | ||
03ba3782 | 441 | sb->s_count++; |
29cb4859 CH |
442 | spin_unlock(&sb_lock); |
443 | ||
03ba3782 | 444 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 445 | if (sb->s_root) |
d19de7ed | 446 | return true; |
03ba3782 JA |
447 | up_read(&sb->s_umount); |
448 | } | |
29cb4859 CH |
449 | |
450 | put_super(sb); | |
d19de7ed | 451 | return false; |
03ba3782 JA |
452 | } |
453 | ||
f11c9c5c ES |
454 | /* |
455 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
456 | * |
457 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
458 | * inodes. Otherwise write only ones which go sequentially |
459 | * in reverse order. | |
edadfb10 | 460 | * |
f11c9c5c ES |
461 | * Return 1, if the caller writeback routine should be |
462 | * interrupted. Otherwise return 0. | |
463 | */ | |
edadfb10 CH |
464 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
465 | struct writeback_control *wbc, bool only_this_sb) | |
1da177e4 | 466 | { |
03ba3782 | 467 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 468 | long pages_skipped; |
f11c9c5c ES |
469 | struct inode *inode = list_entry(wb->b_io.prev, |
470 | struct inode, i_list); | |
edadfb10 CH |
471 | |
472 | if (inode->i_sb != sb) { | |
473 | if (only_this_sb) { | |
474 | /* | |
475 | * We only want to write back data for this | |
476 | * superblock, move all inodes not belonging | |
477 | * to it back onto the dirty list. | |
478 | */ | |
479 | redirty_tail(inode); | |
480 | continue; | |
481 | } | |
482 | ||
483 | /* | |
484 | * The inode belongs to a different superblock. | |
485 | * Bounce back to the caller to unpin this and | |
486 | * pin the next superblock. | |
487 | */ | |
f11c9c5c | 488 | return 0; |
edadfb10 CH |
489 | } |
490 | ||
84a89245 | 491 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
492 | requeue_io(inode); |
493 | continue; | |
494 | } | |
d2caa3c5 JL |
495 | /* |
496 | * Was this inode dirtied after sync_sb_inodes was called? | |
497 | * This keeps sync from extra jobs and livelock. | |
498 | */ | |
f11c9c5c ES |
499 | if (inode_dirtied_after(inode, wbc->wb_start)) |
500 | return 1; | |
1da177e4 | 501 | |
84a89245 | 502 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
503 | __iget(inode); |
504 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 505 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
506 | if (wbc->pages_skipped != pages_skipped) { |
507 | /* | |
508 | * writeback is not making progress due to locked | |
509 | * buffers. Skip this inode for now. | |
510 | */ | |
f57b9b7b | 511 | redirty_tail(inode); |
1da177e4 LT |
512 | } |
513 | spin_unlock(&inode_lock); | |
1da177e4 | 514 | iput(inode); |
4ffc8444 | 515 | cond_resched(); |
1da177e4 | 516 | spin_lock(&inode_lock); |
8bc3be27 FW |
517 | if (wbc->nr_to_write <= 0) { |
518 | wbc->more_io = 1; | |
f11c9c5c | 519 | return 1; |
8bc3be27 | 520 | } |
03ba3782 | 521 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 522 | wbc->more_io = 1; |
1da177e4 | 523 | } |
f11c9c5c ES |
524 | /* b_io is empty */ |
525 | return 1; | |
526 | } | |
527 | ||
9c3a8ee8 CH |
528 | void writeback_inodes_wb(struct bdi_writeback *wb, |
529 | struct writeback_control *wbc) | |
f11c9c5c ES |
530 | { |
531 | int ret = 0; | |
532 | ||
533 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
534 | spin_lock(&inode_lock); | |
535 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
536 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 537 | |
f11c9c5c ES |
538 | while (!list_empty(&wb->b_io)) { |
539 | struct inode *inode = list_entry(wb->b_io.prev, | |
540 | struct inode, i_list); | |
541 | struct super_block *sb = inode->i_sb; | |
9ecc2738 | 542 | |
edadfb10 CH |
543 | if (!pin_sb_for_writeback(sb)) { |
544 | requeue_io(inode); | |
545 | continue; | |
f11c9c5c | 546 | } |
edadfb10 CH |
547 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
548 | drop_super(sb); | |
f11c9c5c | 549 | |
f11c9c5c ES |
550 | if (ret) |
551 | break; | |
552 | } | |
66f3b8e2 JA |
553 | spin_unlock(&inode_lock); |
554 | /* Leave any unwritten inodes on b_io */ | |
555 | } | |
556 | ||
edadfb10 CH |
557 | static void __writeback_inodes_sb(struct super_block *sb, |
558 | struct bdi_writeback *wb, struct writeback_control *wbc) | |
559 | { | |
560 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
561 | ||
562 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
563 | spin_lock(&inode_lock); | |
564 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
565 | queue_io(wb, wbc->older_than_this); | |
566 | writeback_sb_inodes(sb, wb, wbc, true); | |
567 | spin_unlock(&inode_lock); | |
568 | } | |
569 | ||
66f3b8e2 | 570 | /* |
03ba3782 JA |
571 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
572 | * operation. We do this so we don't hold I_SYNC against an inode for | |
573 | * enormous amounts of time, which would block a userspace task which has | |
574 | * been forced to throttle against that inode. Also, the code reevaluates | |
575 | * the dirty each time it has written this many pages. | |
576 | */ | |
577 | #define MAX_WRITEBACK_PAGES 1024 | |
578 | ||
579 | static inline bool over_bground_thresh(void) | |
580 | { | |
581 | unsigned long background_thresh, dirty_thresh; | |
582 | ||
583 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
584 | ||
585 | return (global_page_state(NR_FILE_DIRTY) + | |
586 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
587 | } | |
588 | ||
589 | /* | |
590 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 591 | * |
03ba3782 JA |
592 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
593 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
594 | * just walks the superblock inode list, writing back any inodes which are | |
595 | * older than a specific point in time. | |
66f3b8e2 | 596 | * |
03ba3782 JA |
597 | * Try to run once per dirty_writeback_interval. But if a writeback event |
598 | * takes longer than a dirty_writeback_interval interval, then leave a | |
599 | * one-second gap. | |
66f3b8e2 | 600 | * |
03ba3782 JA |
601 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
602 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 603 | */ |
c4a77a6c | 604 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 605 | struct wb_writeback_work *work) |
66f3b8e2 | 606 | { |
03ba3782 | 607 | struct writeback_control wbc = { |
83ba7b07 | 608 | .sync_mode = work->sync_mode, |
03ba3782 | 609 | .older_than_this = NULL, |
83ba7b07 CH |
610 | .for_kupdate = work->for_kupdate, |
611 | .for_background = work->for_background, | |
612 | .range_cyclic = work->range_cyclic, | |
03ba3782 JA |
613 | }; |
614 | unsigned long oldest_jif; | |
615 | long wrote = 0; | |
a5989bdc | 616 | struct inode *inode; |
66f3b8e2 | 617 | |
03ba3782 JA |
618 | if (wbc.for_kupdate) { |
619 | wbc.older_than_this = &oldest_jif; | |
620 | oldest_jif = jiffies - | |
621 | msecs_to_jiffies(dirty_expire_interval * 10); | |
622 | } | |
c4a77a6c JA |
623 | if (!wbc.range_cyclic) { |
624 | wbc.range_start = 0; | |
625 | wbc.range_end = LLONG_MAX; | |
626 | } | |
38f21977 | 627 | |
03ba3782 JA |
628 | for (;;) { |
629 | /* | |
d3ddec76 | 630 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 631 | */ |
83ba7b07 | 632 | if (work->nr_pages <= 0) |
03ba3782 | 633 | break; |
66f3b8e2 | 634 | |
38f21977 | 635 | /* |
d3ddec76 WF |
636 | * For background writeout, stop when we are below the |
637 | * background dirty threshold | |
38f21977 | 638 | */ |
83ba7b07 | 639 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 640 | break; |
38f21977 | 641 | |
03ba3782 | 642 | wbc.more_io = 0; |
03ba3782 JA |
643 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
644 | wbc.pages_skipped = 0; | |
83ba7b07 CH |
645 | if (work->sb) |
646 | __writeback_inodes_sb(work->sb, wb, &wbc); | |
edadfb10 CH |
647 | else |
648 | writeback_inodes_wb(wb, &wbc); | |
83ba7b07 | 649 | work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
650 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
651 | ||
652 | /* | |
71fd05a8 | 653 | * If we consumed everything, see if we have more |
03ba3782 | 654 | */ |
71fd05a8 JA |
655 | if (wbc.nr_to_write <= 0) |
656 | continue; | |
657 | /* | |
658 | * Didn't write everything and we don't have more IO, bail | |
659 | */ | |
660 | if (!wbc.more_io) | |
03ba3782 | 661 | break; |
71fd05a8 JA |
662 | /* |
663 | * Did we write something? Try for more | |
664 | */ | |
665 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
666 | continue; | |
667 | /* | |
668 | * Nothing written. Wait for some inode to | |
669 | * become available for writeback. Otherwise | |
670 | * we'll just busyloop. | |
671 | */ | |
672 | spin_lock(&inode_lock); | |
673 | if (!list_empty(&wb->b_more_io)) { | |
674 | inode = list_entry(wb->b_more_io.prev, | |
675 | struct inode, i_list); | |
676 | inode_wait_for_writeback(inode); | |
03ba3782 | 677 | } |
71fd05a8 | 678 | spin_unlock(&inode_lock); |
03ba3782 JA |
679 | } |
680 | ||
681 | return wrote; | |
682 | } | |
683 | ||
684 | /* | |
83ba7b07 | 685 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 686 | */ |
83ba7b07 CH |
687 | static struct wb_writeback_work * |
688 | get_next_work_item(struct backing_dev_info *bdi, struct bdi_writeback *wb) | |
03ba3782 | 689 | { |
83ba7b07 | 690 | struct wb_writeback_work *work = NULL; |
03ba3782 | 691 | |
83ba7b07 CH |
692 | spin_lock(&bdi->wb_lock); |
693 | if (!list_empty(&bdi->work_list)) { | |
694 | work = list_entry(bdi->work_list.next, | |
695 | struct wb_writeback_work, list); | |
696 | list_del_init(&work->list); | |
03ba3782 | 697 | } |
83ba7b07 CH |
698 | spin_unlock(&bdi->wb_lock); |
699 | return work; | |
03ba3782 JA |
700 | } |
701 | ||
702 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
703 | { | |
704 | unsigned long expired; | |
705 | long nr_pages; | |
706 | ||
69b62d01 JA |
707 | /* |
708 | * When set to zero, disable periodic writeback | |
709 | */ | |
710 | if (!dirty_writeback_interval) | |
711 | return 0; | |
712 | ||
03ba3782 JA |
713 | expired = wb->last_old_flush + |
714 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
715 | if (time_before(jiffies, expired)) | |
716 | return 0; | |
717 | ||
718 | wb->last_old_flush = jiffies; | |
719 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
720 | global_page_state(NR_UNSTABLE_NFS) + | |
721 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
722 | ||
c4a77a6c | 723 | if (nr_pages) { |
83ba7b07 | 724 | struct wb_writeback_work work = { |
c4a77a6c JA |
725 | .nr_pages = nr_pages, |
726 | .sync_mode = WB_SYNC_NONE, | |
727 | .for_kupdate = 1, | |
728 | .range_cyclic = 1, | |
729 | }; | |
730 | ||
83ba7b07 | 731 | return wb_writeback(wb, &work); |
c4a77a6c | 732 | } |
03ba3782 JA |
733 | |
734 | return 0; | |
735 | } | |
736 | ||
737 | /* | |
738 | * Retrieve work items and do the writeback they describe | |
739 | */ | |
740 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
741 | { | |
742 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 743 | struct wb_writeback_work *work; |
c4a77a6c | 744 | long wrote = 0; |
03ba3782 JA |
745 | |
746 | while ((work = get_next_work_item(bdi, wb)) != NULL) { | |
03ba3782 JA |
747 | /* |
748 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 749 | * because this thread is exiting now. |
03ba3782 JA |
750 | */ |
751 | if (force_wait) | |
83ba7b07 | 752 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 753 | |
83ba7b07 | 754 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
755 | |
756 | /* | |
83ba7b07 CH |
757 | * Notify the caller of completion if this is a synchronous |
758 | * work item, otherwise just free it. | |
03ba3782 | 759 | */ |
83ba7b07 CH |
760 | if (work->done) |
761 | complete(work->done); | |
762 | else | |
763 | kfree(work); | |
03ba3782 JA |
764 | } |
765 | ||
766 | /* | |
767 | * Check for periodic writeback, kupdated() style | |
768 | */ | |
769 | wrote += wb_check_old_data_flush(wb); | |
770 | ||
771 | return wrote; | |
772 | } | |
773 | ||
774 | /* | |
775 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
776 | * wakes up periodically and does kupdated style flushing. | |
777 | */ | |
778 | int bdi_writeback_task(struct bdi_writeback *wb) | |
779 | { | |
780 | unsigned long last_active = jiffies; | |
781 | unsigned long wait_jiffies = -1UL; | |
782 | long pages_written; | |
783 | ||
784 | while (!kthread_should_stop()) { | |
785 | pages_written = wb_do_writeback(wb, 0); | |
786 | ||
787 | if (pages_written) | |
788 | last_active = jiffies; | |
789 | else if (wait_jiffies != -1UL) { | |
790 | unsigned long max_idle; | |
791 | ||
38f21977 | 792 | /* |
03ba3782 JA |
793 | * Longest period of inactivity that we tolerate. If we |
794 | * see dirty data again later, the task will get | |
795 | * recreated automatically. | |
38f21977 | 796 | */ |
03ba3782 JA |
797 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
798 | if (time_after(jiffies, max_idle + last_active)) | |
799 | break; | |
800 | } | |
801 | ||
69b62d01 JA |
802 | if (dirty_writeback_interval) { |
803 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
804 | schedule_timeout_interruptible(wait_jiffies); | |
f9eadbbd JA |
805 | } else { |
806 | set_current_state(TASK_INTERRUPTIBLE); | |
807 | if (list_empty_careful(&wb->bdi->work_list) && | |
808 | !kthread_should_stop()) | |
809 | schedule(); | |
810 | __set_current_state(TASK_RUNNING); | |
811 | } | |
69b62d01 | 812 | |
03ba3782 JA |
813 | try_to_freeze(); |
814 | } | |
815 | ||
816 | return 0; | |
817 | } | |
818 | ||
819 | /* | |
b8c2f347 CH |
820 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
821 | * the whole world. | |
03ba3782 | 822 | */ |
b8c2f347 | 823 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 824 | { |
b8c2f347 | 825 | struct backing_dev_info *bdi; |
03ba3782 | 826 | |
83ba7b07 CH |
827 | if (!nr_pages) { |
828 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
829 | global_page_state(NR_UNSTABLE_NFS); |
830 | } | |
03ba3782 | 831 | |
b8c2f347 | 832 | rcu_read_lock(); |
cfc4ba53 | 833 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
834 | if (!bdi_has_dirty_io(bdi)) |
835 | continue; | |
83ba7b07 | 836 | __bdi_start_writeback(bdi, nr_pages, false, false); |
03ba3782 | 837 | } |
cfc4ba53 | 838 | rcu_read_unlock(); |
1da177e4 LT |
839 | } |
840 | ||
03ba3782 JA |
841 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
842 | { | |
843 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
844 | struct dentry *dentry; | |
845 | const char *name = "?"; | |
846 | ||
847 | dentry = d_find_alias(inode); | |
848 | if (dentry) { | |
849 | spin_lock(&dentry->d_lock); | |
850 | name = (const char *) dentry->d_name.name; | |
851 | } | |
852 | printk(KERN_DEBUG | |
853 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
854 | current->comm, task_pid_nr(current), inode->i_ino, | |
855 | name, inode->i_sb->s_id); | |
856 | if (dentry) { | |
857 | spin_unlock(&dentry->d_lock); | |
858 | dput(dentry); | |
859 | } | |
860 | } | |
861 | } | |
862 | ||
863 | /** | |
864 | * __mark_inode_dirty - internal function | |
865 | * @inode: inode to mark | |
866 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
867 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
868 | * mark_inode_dirty_sync. | |
1da177e4 | 869 | * |
03ba3782 JA |
870 | * Put the inode on the super block's dirty list. |
871 | * | |
872 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
873 | * dirty list only if it is hashed or if it refers to a blockdev. | |
874 | * If it was not hashed, it will never be added to the dirty list | |
875 | * even if it is later hashed, as it will have been marked dirty already. | |
876 | * | |
877 | * In short, make sure you hash any inodes _before_ you start marking | |
878 | * them dirty. | |
1da177e4 | 879 | * |
03ba3782 JA |
880 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
881 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 882 | * |
03ba3782 JA |
883 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
884 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
885 | * the kernel-internal blockdev inode represents the dirtying time of the | |
886 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
887 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
888 | * blockdev inode. | |
1da177e4 | 889 | */ |
03ba3782 | 890 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 891 | { |
03ba3782 | 892 | struct super_block *sb = inode->i_sb; |
1da177e4 | 893 | |
03ba3782 JA |
894 | /* |
895 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
896 | * dirty the inode itself | |
897 | */ | |
898 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
899 | if (sb->s_op->dirty_inode) | |
900 | sb->s_op->dirty_inode(inode); | |
901 | } | |
902 | ||
903 | /* | |
904 | * make sure that changes are seen by all cpus before we test i_state | |
905 | * -- mikulas | |
906 | */ | |
907 | smp_mb(); | |
908 | ||
909 | /* avoid the locking if we can */ | |
910 | if ((inode->i_state & flags) == flags) | |
911 | return; | |
912 | ||
913 | if (unlikely(block_dump)) | |
914 | block_dump___mark_inode_dirty(inode); | |
915 | ||
916 | spin_lock(&inode_lock); | |
917 | if ((inode->i_state & flags) != flags) { | |
918 | const int was_dirty = inode->i_state & I_DIRTY; | |
919 | ||
920 | inode->i_state |= flags; | |
921 | ||
922 | /* | |
923 | * If the inode is being synced, just update its dirty state. | |
924 | * The unlocker will place the inode on the appropriate | |
925 | * superblock list, based upon its state. | |
926 | */ | |
927 | if (inode->i_state & I_SYNC) | |
928 | goto out; | |
929 | ||
930 | /* | |
931 | * Only add valid (hashed) inodes to the superblock's | |
932 | * dirty list. Add blockdev inodes as well. | |
933 | */ | |
934 | if (!S_ISBLK(inode->i_mode)) { | |
935 | if (hlist_unhashed(&inode->i_hash)) | |
936 | goto out; | |
937 | } | |
938 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
939 | goto out; | |
940 | ||
941 | /* | |
942 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
943 | * reposition it (that would break b_dirty time-ordering). | |
944 | */ | |
945 | if (!was_dirty) { | |
946 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
947 | struct backing_dev_info *bdi = wb->bdi; |
948 | ||
949 | if (bdi_cap_writeback_dirty(bdi) && | |
950 | !test_bit(BDI_registered, &bdi->state)) { | |
951 | WARN_ON(1); | |
952 | printk(KERN_ERR "bdi-%s not registered\n", | |
953 | bdi->name); | |
954 | } | |
03ba3782 JA |
955 | |
956 | inode->dirtied_when = jiffies; | |
957 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 958 | } |
1da177e4 | 959 | } |
03ba3782 JA |
960 | out: |
961 | spin_unlock(&inode_lock); | |
962 | } | |
963 | EXPORT_SYMBOL(__mark_inode_dirty); | |
964 | ||
965 | /* | |
966 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
967 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
968 | * | |
969 | * If older_than_this is non-NULL, then only write out inodes which | |
970 | * had their first dirtying at a time earlier than *older_than_this. | |
971 | * | |
03ba3782 JA |
972 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
973 | * This function assumes that the blockdev superblock's inodes are backed by | |
974 | * a variety of queues, so all inodes are searched. For other superblocks, | |
975 | * assume that all inodes are backed by the same queue. | |
976 | * | |
977 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
978 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
979 | * on the writer throttling path, and we get decent balancing between many | |
980 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
981 | */ | |
b6e51316 | 982 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
983 | { |
984 | struct inode *inode, *old_inode = NULL; | |
985 | ||
986 | /* | |
987 | * We need to be protected against the filesystem going from | |
988 | * r/o to r/w or vice versa. | |
989 | */ | |
b6e51316 | 990 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
991 | |
992 | spin_lock(&inode_lock); | |
993 | ||
994 | /* | |
995 | * Data integrity sync. Must wait for all pages under writeback, | |
996 | * because there may have been pages dirtied before our sync | |
997 | * call, but which had writeout started before we write it out. | |
998 | * In which case, the inode may not be on the dirty list, but | |
999 | * we still have to wait for that writeout. | |
1000 | */ | |
b6e51316 | 1001 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1002 | struct address_space *mapping; |
1003 | ||
1004 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1005 | continue; | |
1006 | mapping = inode->i_mapping; | |
1007 | if (mapping->nrpages == 0) | |
1008 | continue; | |
1009 | __iget(inode); | |
1010 | spin_unlock(&inode_lock); | |
1011 | /* | |
1012 | * We hold a reference to 'inode' so it couldn't have | |
1013 | * been removed from s_inodes list while we dropped the | |
1014 | * inode_lock. We cannot iput the inode now as we can | |
1015 | * be holding the last reference and we cannot iput it | |
1016 | * under inode_lock. So we keep the reference and iput | |
1017 | * it later. | |
1018 | */ | |
1019 | iput(old_inode); | |
1020 | old_inode = inode; | |
1021 | ||
1022 | filemap_fdatawait(mapping); | |
1023 | ||
1024 | cond_resched(); | |
1025 | ||
1026 | spin_lock(&inode_lock); | |
1027 | } | |
1028 | spin_unlock(&inode_lock); | |
1029 | iput(old_inode); | |
1da177e4 LT |
1030 | } |
1031 | ||
d8a8559c JA |
1032 | /** |
1033 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1034 | * @sb: the superblock | |
1da177e4 | 1035 | * |
d8a8559c JA |
1036 | * Start writeback on some inodes on this super_block. No guarantees are made |
1037 | * on how many (if any) will be written, and this function does not wait | |
1038 | * for IO completion of submitted IO. The number of pages submitted is | |
1039 | * returned. | |
1da177e4 | 1040 | */ |
b6e51316 | 1041 | void writeback_inodes_sb(struct super_block *sb) |
1da177e4 | 1042 | { |
0e3c9a22 JA |
1043 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
1044 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
83ba7b07 CH |
1045 | DECLARE_COMPLETION_ONSTACK(done); |
1046 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1047 | .sb = sb, |
1048 | .sync_mode = WB_SYNC_NONE, | |
83ba7b07 | 1049 | .done = &done, |
3c4d7165 | 1050 | }; |
d8a8559c | 1051 | |
cf37e972 CH |
1052 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1053 | ||
83ba7b07 | 1054 | work.nr_pages = nr_dirty + nr_unstable + |
0e3c9a22 JA |
1055 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
1056 | ||
83ba7b07 CH |
1057 | bdi_queue_work(sb->s_bdi, &work); |
1058 | wait_for_completion(&done); | |
e913fc82 | 1059 | } |
0e3c9a22 | 1060 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1061 | |
17bd55d0 ES |
1062 | /** |
1063 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1064 | * @sb: the superblock | |
1065 | * | |
1066 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1067 | * Returns 1 if writeback was started, 0 if not. | |
1068 | */ | |
1069 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1070 | { | |
1071 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1072 | down_read(&sb->s_umount); |
17bd55d0 | 1073 | writeback_inodes_sb(sb); |
cf37e972 | 1074 | up_read(&sb->s_umount); |
17bd55d0 ES |
1075 | return 1; |
1076 | } else | |
1077 | return 0; | |
1078 | } | |
1079 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1080 | ||
d8a8559c JA |
1081 | /** |
1082 | * sync_inodes_sb - sync sb inode pages | |
1083 | * @sb: the superblock | |
1084 | * | |
1085 | * This function writes and waits on any dirty inode belonging to this | |
1086 | * super_block. The number of pages synced is returned. | |
1087 | */ | |
b6e51316 | 1088 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1089 | { |
83ba7b07 CH |
1090 | DECLARE_COMPLETION_ONSTACK(done); |
1091 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1092 | .sb = sb, |
1093 | .sync_mode = WB_SYNC_ALL, | |
1094 | .nr_pages = LONG_MAX, | |
1095 | .range_cyclic = 0, | |
83ba7b07 | 1096 | .done = &done, |
3c4d7165 CH |
1097 | }; |
1098 | ||
cf37e972 CH |
1099 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1100 | ||
83ba7b07 CH |
1101 | bdi_queue_work(sb->s_bdi, &work); |
1102 | wait_for_completion(&done); | |
1103 | ||
b6e51316 | 1104 | wait_sb_inodes(sb); |
1da177e4 | 1105 | } |
d8a8559c | 1106 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1107 | |
1da177e4 | 1108 | /** |
7f04c26d AA |
1109 | * write_inode_now - write an inode to disk |
1110 | * @inode: inode to write to disk | |
1111 | * @sync: whether the write should be synchronous or not | |
1112 | * | |
1113 | * This function commits an inode to disk immediately if it is dirty. This is | |
1114 | * primarily needed by knfsd. | |
1da177e4 | 1115 | * |
7f04c26d | 1116 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1117 | */ |
1da177e4 LT |
1118 | int write_inode_now(struct inode *inode, int sync) |
1119 | { | |
1120 | int ret; | |
1121 | struct writeback_control wbc = { | |
1122 | .nr_to_write = LONG_MAX, | |
18914b18 | 1123 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1124 | .range_start = 0, |
1125 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1126 | }; |
1127 | ||
1128 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1129 | wbc.nr_to_write = 0; |
1da177e4 LT |
1130 | |
1131 | might_sleep(); | |
1132 | spin_lock(&inode_lock); | |
01c03194 | 1133 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1134 | spin_unlock(&inode_lock); |
1135 | if (sync) | |
1c0eeaf5 | 1136 | inode_sync_wait(inode); |
1da177e4 LT |
1137 | return ret; |
1138 | } | |
1139 | EXPORT_SYMBOL(write_inode_now); | |
1140 | ||
1141 | /** | |
1142 | * sync_inode - write an inode and its pages to disk. | |
1143 | * @inode: the inode to sync | |
1144 | * @wbc: controls the writeback mode | |
1145 | * | |
1146 | * sync_inode() will write an inode and its pages to disk. It will also | |
1147 | * correctly update the inode on its superblock's dirty inode lists and will | |
1148 | * update inode->i_state. | |
1149 | * | |
1150 | * The caller must have a ref on the inode. | |
1151 | */ | |
1152 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1153 | { | |
1154 | int ret; | |
1155 | ||
1156 | spin_lock(&inode_lock); | |
01c03194 | 1157 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1158 | spin_unlock(&inode_lock); |
1159 | return ret; | |
1160 | } | |
1161 | EXPORT_SYMBOL(sync_inode); |