]>
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 LT |
18 | #include <linux/spinlock.h> |
19 | #include <linux/sched.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/mm.h> | |
03ba3782 JA |
22 | #include <linux/kthread.h> |
23 | #include <linux/freezer.h> | |
1da177e4 LT |
24 | #include <linux/writeback.h> |
25 | #include <linux/blkdev.h> | |
26 | #include <linux/backing-dev.h> | |
27 | #include <linux/buffer_head.h> | |
07f3f05c | 28 | #include "internal.h" |
1da177e4 | 29 | |
66f3b8e2 | 30 | #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) |
f11b00f3 | 31 | |
d0bceac7 JA |
32 | /* |
33 | * We don't actually have pdflush, but this one is exported though /proc... | |
34 | */ | |
35 | int nr_pdflush_threads; | |
36 | ||
c4a77a6c JA |
37 | /* |
38 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
39 | */ | |
40 | struct wb_writeback_args { | |
41 | long nr_pages; | |
42 | struct super_block *sb; | |
43 | enum writeback_sync_modes sync_mode; | |
d3ddec76 WF |
44 | int for_kupdate:1; |
45 | int range_cyclic:1; | |
46 | int for_background:1; | |
c4a77a6c JA |
47 | }; |
48 | ||
03ba3782 JA |
49 | /* |
50 | * Work items for the bdi_writeback threads | |
f11b00f3 | 51 | */ |
03ba3782 | 52 | struct bdi_work { |
8010c3b6 JA |
53 | struct list_head list; /* pending work list */ |
54 | struct rcu_head rcu_head; /* for RCU free/clear of work */ | |
03ba3782 | 55 | |
8010c3b6 JA |
56 | unsigned long seen; /* threads that have seen this work */ |
57 | atomic_t pending; /* number of threads still to do work */ | |
03ba3782 | 58 | |
8010c3b6 | 59 | struct wb_writeback_args args; /* writeback arguments */ |
03ba3782 | 60 | |
8010c3b6 | 61 | unsigned long state; /* flag bits, see WS_* */ |
03ba3782 JA |
62 | }; |
63 | ||
64 | enum { | |
65 | WS_USED_B = 0, | |
66 | WS_ONSTACK_B, | |
67 | }; | |
68 | ||
69 | #define WS_USED (1 << WS_USED_B) | |
70 | #define WS_ONSTACK (1 << WS_ONSTACK_B) | |
71 | ||
72 | static inline bool bdi_work_on_stack(struct bdi_work *work) | |
73 | { | |
74 | return test_bit(WS_ONSTACK_B, &work->state); | |
75 | } | |
76 | ||
77 | static inline void bdi_work_init(struct bdi_work *work, | |
b6e51316 | 78 | struct wb_writeback_args *args) |
03ba3782 JA |
79 | { |
80 | INIT_RCU_HEAD(&work->rcu_head); | |
b6e51316 | 81 | work->args = *args; |
03ba3782 JA |
82 | work->state = WS_USED; |
83 | } | |
84 | ||
f11b00f3 AB |
85 | /** |
86 | * writeback_in_progress - determine whether there is writeback in progress | |
87 | * @bdi: the device's backing_dev_info structure. | |
88 | * | |
03ba3782 JA |
89 | * Determine whether there is writeback waiting to be handled against a |
90 | * backing device. | |
f11b00f3 AB |
91 | */ |
92 | int writeback_in_progress(struct backing_dev_info *bdi) | |
93 | { | |
03ba3782 | 94 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
95 | } |
96 | ||
03ba3782 | 97 | static void bdi_work_clear(struct bdi_work *work) |
f11b00f3 | 98 | { |
03ba3782 JA |
99 | clear_bit(WS_USED_B, &work->state); |
100 | smp_mb__after_clear_bit(); | |
1ef7d9aa NP |
101 | /* |
102 | * work can have disappeared at this point. bit waitq functions | |
103 | * should be able to tolerate this, provided bdi_sched_wait does | |
104 | * not dereference it's pointer argument. | |
105 | */ | |
03ba3782 | 106 | wake_up_bit(&work->state, WS_USED_B); |
f11b00f3 AB |
107 | } |
108 | ||
03ba3782 | 109 | static void bdi_work_free(struct rcu_head *head) |
4195f73d | 110 | { |
03ba3782 | 111 | struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); |
4195f73d | 112 | |
03ba3782 JA |
113 | if (!bdi_work_on_stack(work)) |
114 | kfree(work); | |
115 | else | |
116 | bdi_work_clear(work); | |
4195f73d NP |
117 | } |
118 | ||
03ba3782 | 119 | static void wb_work_complete(struct bdi_work *work) |
1da177e4 | 120 | { |
c4a77a6c | 121 | const enum writeback_sync_modes sync_mode = work->args.sync_mode; |
77b9d059 | 122 | int onstack = bdi_work_on_stack(work); |
1da177e4 LT |
123 | |
124 | /* | |
03ba3782 JA |
125 | * For allocated work, we can clear the done/seen bit right here. |
126 | * For on-stack work, we need to postpone both the clear and free | |
127 | * to after the RCU grace period, since the stack could be invalidated | |
128 | * as soon as bdi_work_clear() has done the wakeup. | |
1da177e4 | 129 | */ |
77b9d059 | 130 | if (!onstack) |
03ba3782 | 131 | bdi_work_clear(work); |
77b9d059 | 132 | if (sync_mode == WB_SYNC_NONE || onstack) |
03ba3782 JA |
133 | call_rcu(&work->rcu_head, bdi_work_free); |
134 | } | |
1da177e4 | 135 | |
03ba3782 JA |
136 | static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) |
137 | { | |
1da177e4 | 138 | /* |
03ba3782 JA |
139 | * The caller has retrieved the work arguments from this work, |
140 | * drop our reference. If this is the last ref, delete and free it | |
1da177e4 | 141 | */ |
03ba3782 JA |
142 | if (atomic_dec_and_test(&work->pending)) { |
143 | struct backing_dev_info *bdi = wb->bdi; | |
1da177e4 | 144 | |
03ba3782 JA |
145 | spin_lock(&bdi->wb_lock); |
146 | list_del_rcu(&work->list); | |
147 | spin_unlock(&bdi->wb_lock); | |
1da177e4 | 148 | |
03ba3782 JA |
149 | wb_work_complete(work); |
150 | } | |
151 | } | |
1da177e4 | 152 | |
03ba3782 JA |
153 | static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) |
154 | { | |
bcddc3f0 JA |
155 | work->seen = bdi->wb_mask; |
156 | BUG_ON(!work->seen); | |
157 | atomic_set(&work->pending, bdi->wb_cnt); | |
158 | BUG_ON(!bdi->wb_cnt); | |
1da177e4 | 159 | |
bcddc3f0 | 160 | /* |
deed62ed NP |
161 | * list_add_tail_rcu() contains the necessary barriers to |
162 | * make sure the above stores are seen before the item is | |
163 | * noticed on the list | |
bcddc3f0 | 164 | */ |
bcddc3f0 JA |
165 | spin_lock(&bdi->wb_lock); |
166 | list_add_tail_rcu(&work->list, &bdi->work_list); | |
167 | spin_unlock(&bdi->wb_lock); | |
03ba3782 JA |
168 | |
169 | /* | |
170 | * If the default thread isn't there, make sure we add it. When | |
171 | * it gets created and wakes up, we'll run this work. | |
172 | */ | |
173 | if (unlikely(list_empty_careful(&bdi->wb_list))) | |
174 | wake_up_process(default_backing_dev_info.wb.task); | |
175 | else { | |
176 | struct bdi_writeback *wb = &bdi->wb; | |
1da177e4 | 177 | |
1ef7d9aa | 178 | if (wb->task) |
03ba3782 | 179 | wake_up_process(wb->task); |
1da177e4 | 180 | } |
1da177e4 LT |
181 | } |
182 | ||
03ba3782 JA |
183 | /* |
184 | * Used for on-stack allocated work items. The caller needs to wait until | |
185 | * the wb threads have acked the work before it's safe to continue. | |
186 | */ | |
187 | static void bdi_wait_on_work_clear(struct bdi_work *work) | |
188 | { | |
189 | wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, | |
190 | TASK_UNINTERRUPTIBLE); | |
191 | } | |
1da177e4 | 192 | |
f11fcae8 | 193 | static void bdi_alloc_queue_work(struct backing_dev_info *bdi, |
b6e51316 | 194 | struct wb_writeback_args *args) |
1da177e4 | 195 | { |
03ba3782 JA |
196 | struct bdi_work *work; |
197 | ||
bcddc3f0 JA |
198 | /* |
199 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
200 | * wakeup the thread for old dirty data writeback | |
201 | */ | |
03ba3782 | 202 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
bcddc3f0 | 203 | if (work) { |
b6e51316 | 204 | bdi_work_init(work, args); |
bcddc3f0 JA |
205 | bdi_queue_work(bdi, work); |
206 | } else { | |
207 | struct bdi_writeback *wb = &bdi->wb; | |
03ba3782 | 208 | |
bcddc3f0 JA |
209 | if (wb->task) |
210 | wake_up_process(wb->task); | |
211 | } | |
03ba3782 JA |
212 | } |
213 | ||
b6e51316 JA |
214 | /** |
215 | * bdi_sync_writeback - start and wait for writeback | |
216 | * @bdi: the backing device to write from | |
217 | * @sb: write inodes from this super_block | |
218 | * | |
219 | * Description: | |
220 | * This does WB_SYNC_ALL data integrity writeback and waits for the | |
221 | * IO to complete. Callers must hold the sb s_umount semaphore for | |
222 | * reading, to avoid having the super disappear before we are done. | |
223 | */ | |
224 | static void bdi_sync_writeback(struct backing_dev_info *bdi, | |
225 | struct super_block *sb) | |
03ba3782 | 226 | { |
b6e51316 JA |
227 | struct wb_writeback_args args = { |
228 | .sb = sb, | |
229 | .sync_mode = WB_SYNC_ALL, | |
230 | .nr_pages = LONG_MAX, | |
231 | .range_cyclic = 0, | |
232 | }; | |
233 | struct bdi_work work; | |
03ba3782 | 234 | |
b6e51316 JA |
235 | bdi_work_init(&work, &args); |
236 | work.state |= WS_ONSTACK; | |
03ba3782 | 237 | |
b6e51316 JA |
238 | bdi_queue_work(bdi, &work); |
239 | bdi_wait_on_work_clear(&work); | |
240 | } | |
241 | ||
242 | /** | |
243 | * bdi_start_writeback - start writeback | |
244 | * @bdi: the backing device to write from | |
245 | * @nr_pages: the number of pages to write | |
246 | * | |
247 | * Description: | |
248 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
249 | * started when this function returns, we make no guarentees on | |
250 | * completion. Caller need not hold sb s_umount semaphore. | |
251 | * | |
252 | */ | |
253 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) | |
254 | { | |
255 | struct wb_writeback_args args = { | |
256 | .sync_mode = WB_SYNC_NONE, | |
257 | .nr_pages = nr_pages, | |
258 | .range_cyclic = 1, | |
259 | }; | |
260 | ||
d3ddec76 WF |
261 | /* |
262 | * We treat @nr_pages=0 as the special case to do background writeback, | |
263 | * ie. to sync pages until the background dirty threshold is reached. | |
264 | */ | |
265 | if (!nr_pages) { | |
266 | args.nr_pages = LONG_MAX; | |
267 | args.for_background = 1; | |
268 | } | |
269 | ||
b6e51316 | 270 | bdi_alloc_queue_work(bdi, &args); |
1da177e4 LT |
271 | } |
272 | ||
6610a0bc AM |
273 | /* |
274 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
275 | * furthest end of its superblock's dirty-inode list. | |
276 | * | |
277 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 278 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
279 | * the case then the inode must have been redirtied while it was being written |
280 | * out and we don't reset its dirtied_when. | |
281 | */ | |
282 | static void redirty_tail(struct inode *inode) | |
283 | { | |
03ba3782 | 284 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 285 | |
03ba3782 | 286 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 287 | struct inode *tail; |
6610a0bc | 288 | |
03ba3782 | 289 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 290 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
291 | inode->dirtied_when = jiffies; |
292 | } | |
03ba3782 | 293 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
294 | } |
295 | ||
c986d1e2 | 296 | /* |
66f3b8e2 | 297 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 298 | */ |
0e0f4fc2 | 299 | static void requeue_io(struct inode *inode) |
c986d1e2 | 300 | { |
03ba3782 JA |
301 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
302 | ||
303 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
304 | } |
305 | ||
1c0eeaf5 JE |
306 | static void inode_sync_complete(struct inode *inode) |
307 | { | |
308 | /* | |
309 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
310 | */ | |
311 | smp_mb(); | |
312 | wake_up_bit(&inode->i_state, __I_SYNC); | |
313 | } | |
314 | ||
d2caa3c5 JL |
315 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
316 | { | |
317 | bool ret = time_after(inode->dirtied_when, t); | |
318 | #ifndef CONFIG_64BIT | |
319 | /* | |
320 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
321 | * It _appears_ to be in the future, but is actually in distant past. | |
322 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 323 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
324 | */ |
325 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
326 | #endif | |
327 | return ret; | |
328 | } | |
329 | ||
2c136579 FW |
330 | /* |
331 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
332 | */ | |
333 | static void move_expired_inodes(struct list_head *delaying_queue, | |
334 | struct list_head *dispatch_queue, | |
335 | unsigned long *older_than_this) | |
336 | { | |
5c03449d SL |
337 | LIST_HEAD(tmp); |
338 | struct list_head *pos, *node; | |
cf137307 | 339 | struct super_block *sb = NULL; |
5c03449d | 340 | struct inode *inode; |
cf137307 | 341 | int do_sb_sort = 0; |
5c03449d | 342 | |
2c136579 | 343 | while (!list_empty(delaying_queue)) { |
5c03449d | 344 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
2c136579 | 345 | if (older_than_this && |
d2caa3c5 | 346 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 347 | break; |
cf137307 JA |
348 | if (sb && sb != inode->i_sb) |
349 | do_sb_sort = 1; | |
350 | sb = inode->i_sb; | |
5c03449d SL |
351 | list_move(&inode->i_list, &tmp); |
352 | } | |
353 | ||
cf137307 JA |
354 | /* just one sb in list, splice to dispatch_queue and we're done */ |
355 | if (!do_sb_sort) { | |
356 | list_splice(&tmp, dispatch_queue); | |
357 | return; | |
358 | } | |
359 | ||
5c03449d SL |
360 | /* Move inodes from one superblock together */ |
361 | while (!list_empty(&tmp)) { | |
362 | inode = list_entry(tmp.prev, struct inode, i_list); | |
363 | sb = inode->i_sb; | |
364 | list_for_each_prev_safe(pos, node, &tmp) { | |
365 | inode = list_entry(pos, struct inode, i_list); | |
366 | if (inode->i_sb == sb) | |
367 | list_move(&inode->i_list, dispatch_queue); | |
368 | } | |
2c136579 FW |
369 | } |
370 | } | |
371 | ||
372 | /* | |
373 | * Queue all expired dirty inodes for io, eldest first. | |
374 | */ | |
03ba3782 | 375 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 376 | { |
03ba3782 JA |
377 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
378 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
379 | } |
380 | ||
03ba3782 | 381 | static int write_inode(struct inode *inode, int sync) |
08d8e974 | 382 | { |
03ba3782 JA |
383 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
384 | return inode->i_sb->s_op->write_inode(inode, sync); | |
385 | return 0; | |
08d8e974 | 386 | } |
08d8e974 | 387 | |
1da177e4 | 388 | /* |
01c03194 CH |
389 | * Wait for writeback on an inode to complete. |
390 | */ | |
391 | static void inode_wait_for_writeback(struct inode *inode) | |
392 | { | |
393 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
394 | wait_queue_head_t *wqh; | |
395 | ||
396 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
397 | do { | |
398 | spin_unlock(&inode_lock); | |
399 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
400 | spin_lock(&inode_lock); | |
401 | } while (inode->i_state & I_SYNC); | |
402 | } | |
403 | ||
404 | /* | |
405 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
406 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
407 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
408 | * | |
1da177e4 LT |
409 | * If `wait' is set, wait on the writeout. |
410 | * | |
411 | * The whole writeout design is quite complex and fragile. We want to avoid | |
412 | * starvation of particular inodes when others are being redirtied, prevent | |
413 | * livelocks, etc. | |
414 | * | |
415 | * Called under inode_lock. | |
416 | */ | |
417 | static int | |
01c03194 | 418 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 419 | { |
1da177e4 | 420 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 421 | int wait = wbc->sync_mode == WB_SYNC_ALL; |
01c03194 | 422 | unsigned dirty; |
1da177e4 LT |
423 | int ret; |
424 | ||
01c03194 CH |
425 | if (!atomic_read(&inode->i_count)) |
426 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
427 | else | |
428 | WARN_ON(inode->i_state & I_WILL_FREE); | |
429 | ||
430 | if (inode->i_state & I_SYNC) { | |
431 | /* | |
432 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 433 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
434 | * writeback can proceed with the other inodes on s_io. |
435 | * | |
436 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 437 | * completed a full scan of b_io. |
01c03194 CH |
438 | */ |
439 | if (!wait) { | |
440 | requeue_io(inode); | |
441 | return 0; | |
442 | } | |
443 | ||
444 | /* | |
445 | * It's a data-integrity sync. We must wait. | |
446 | */ | |
447 | inode_wait_for_writeback(inode); | |
448 | } | |
449 | ||
1c0eeaf5 | 450 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 451 | |
1c0eeaf5 | 452 | /* Set I_SYNC, reset I_DIRTY */ |
1da177e4 | 453 | dirty = inode->i_state & I_DIRTY; |
1c0eeaf5 | 454 | inode->i_state |= I_SYNC; |
1da177e4 LT |
455 | inode->i_state &= ~I_DIRTY; |
456 | ||
457 | spin_unlock(&inode_lock); | |
458 | ||
459 | ret = do_writepages(mapping, wbc); | |
460 | ||
461 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | |
462 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
463 | int err = write_inode(inode, wait); | |
464 | if (ret == 0) | |
465 | ret = err; | |
466 | } | |
467 | ||
468 | if (wait) { | |
469 | int err = filemap_fdatawait(mapping); | |
470 | if (ret == 0) | |
471 | ret = err; | |
472 | } | |
473 | ||
474 | spin_lock(&inode_lock); | |
1c0eeaf5 | 475 | inode->i_state &= ~I_SYNC; |
84a89245 | 476 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
b3af9468 | 477 | if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { |
ae1b7f7d | 478 | /* |
b3af9468 WF |
479 | * More pages get dirtied by a fast dirtier. |
480 | */ | |
481 | goto select_queue; | |
482 | } else if (inode->i_state & I_DIRTY) { | |
483 | /* | |
484 | * At least XFS will redirty the inode during the | |
485 | * writeback (delalloc) and on io completion (isize). | |
ae1b7f7d WF |
486 | */ |
487 | redirty_tail(inode); | |
488 | } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
1da177e4 LT |
489 | /* |
490 | * We didn't write back all the pages. nfs_writepages() | |
491 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 492 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
493 | */ |
494 | /* | |
495 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 496 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
497 | * consideration. Otherwise, move it to the tail, for |
498 | * the reasons described there. I'm not really sure | |
499 | * how much sense this makes. Presumably I had a good | |
500 | * reasons for doing it this way, and I'd rather not | |
501 | * muck with it at present. | |
1da177e4 LT |
502 | */ |
503 | if (wbc->for_kupdate) { | |
504 | /* | |
2c136579 | 505 | * For the kupdate function we move the inode |
66f3b8e2 | 506 | * to b_more_io so it will get more writeout as |
2c136579 | 507 | * soon as the queue becomes uncongested. |
1da177e4 LT |
508 | */ |
509 | inode->i_state |= I_DIRTY_PAGES; | |
b3af9468 | 510 | select_queue: |
8bc3be27 FW |
511 | if (wbc->nr_to_write <= 0) { |
512 | /* | |
513 | * slice used up: queue for next turn | |
514 | */ | |
515 | requeue_io(inode); | |
516 | } else { | |
517 | /* | |
518 | * somehow blocked: retry later | |
519 | */ | |
520 | redirty_tail(inode); | |
521 | } | |
1da177e4 LT |
522 | } else { |
523 | /* | |
524 | * Otherwise fully redirty the inode so that | |
525 | * other inodes on this superblock will get some | |
526 | * writeout. Otherwise heavy writing to one | |
527 | * file would indefinitely suspend writeout of | |
528 | * all the other files. | |
529 | */ | |
530 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 531 | redirty_tail(inode); |
1da177e4 | 532 | } |
1da177e4 LT |
533 | } else if (atomic_read(&inode->i_count)) { |
534 | /* | |
535 | * The inode is clean, inuse | |
536 | */ | |
537 | list_move(&inode->i_list, &inode_in_use); | |
538 | } else { | |
539 | /* | |
540 | * The inode is clean, unused | |
541 | */ | |
542 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
543 | } |
544 | } | |
1c0eeaf5 | 545 | inode_sync_complete(inode); |
1da177e4 LT |
546 | return ret; |
547 | } | |
548 | ||
9ecc2738 JA |
549 | static void unpin_sb_for_writeback(struct super_block **psb) |
550 | { | |
551 | struct super_block *sb = *psb; | |
552 | ||
553 | if (sb) { | |
554 | up_read(&sb->s_umount); | |
555 | put_super(sb); | |
556 | *psb = NULL; | |
557 | } | |
558 | } | |
559 | ||
03ba3782 JA |
560 | /* |
561 | * For WB_SYNC_NONE writeback, the caller does not have the sb pinned | |
562 | * before calling writeback. So make sure that we do pin it, so it doesn't | |
563 | * go away while we are writing inodes from it. | |
564 | * | |
565 | * Returns 0 if the super was successfully pinned (or pinning wasn't needed), | |
566 | * 1 if we failed. | |
567 | */ | |
568 | static int pin_sb_for_writeback(struct writeback_control *wbc, | |
9ecc2738 | 569 | struct inode *inode, struct super_block **psb) |
03ba3782 JA |
570 | { |
571 | struct super_block *sb = inode->i_sb; | |
572 | ||
9ecc2738 JA |
573 | /* |
574 | * If this sb is already pinned, nothing more to do. If not and | |
575 | * *psb is non-NULL, unpin the old one first | |
576 | */ | |
577 | if (sb == *psb) | |
578 | return 0; | |
579 | else if (*psb) | |
580 | unpin_sb_for_writeback(psb); | |
581 | ||
03ba3782 JA |
582 | /* |
583 | * Caller must already hold the ref for this | |
584 | */ | |
585 | if (wbc->sync_mode == WB_SYNC_ALL) { | |
586 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
587 | return 0; | |
588 | } | |
589 | ||
590 | spin_lock(&sb_lock); | |
591 | sb->s_count++; | |
592 | if (down_read_trylock(&sb->s_umount)) { | |
593 | if (sb->s_root) { | |
594 | spin_unlock(&sb_lock); | |
9ecc2738 | 595 | goto pinned; |
03ba3782 JA |
596 | } |
597 | /* | |
598 | * umounted, drop rwsem again and fall through to failure | |
599 | */ | |
600 | up_read(&sb->s_umount); | |
601 | } | |
602 | ||
603 | sb->s_count--; | |
604 | spin_unlock(&sb_lock); | |
605 | return 1; | |
9ecc2738 JA |
606 | pinned: |
607 | *psb = sb; | |
608 | return 0; | |
03ba3782 JA |
609 | } |
610 | ||
611 | static void writeback_inodes_wb(struct bdi_writeback *wb, | |
612 | struct writeback_control *wbc) | |
1da177e4 | 613 | { |
9ecc2738 | 614 | struct super_block *sb = wbc->sb, *pin_sb = NULL; |
66f3b8e2 | 615 | const int is_blkdev_sb = sb_is_blkdev_sb(sb); |
1da177e4 LT |
616 | const unsigned long start = jiffies; /* livelock avoidance */ |
617 | ||
ae8547b0 | 618 | spin_lock(&inode_lock); |
1da177e4 | 619 | |
03ba3782 JA |
620 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
621 | queue_io(wb, wbc->older_than_this); | |
66f3b8e2 | 622 | |
03ba3782 JA |
623 | while (!list_empty(&wb->b_io)) { |
624 | struct inode *inode = list_entry(wb->b_io.prev, | |
1da177e4 | 625 | struct inode, i_list); |
1da177e4 LT |
626 | long pages_skipped; |
627 | ||
66f3b8e2 JA |
628 | /* |
629 | * super block given and doesn't match, skip this inode | |
630 | */ | |
631 | if (sb && sb != inode->i_sb) { | |
632 | redirty_tail(inode); | |
633 | continue; | |
634 | } | |
635 | ||
03ba3782 | 636 | if (!bdi_cap_writeback_dirty(wb->bdi)) { |
9852a0e7 | 637 | redirty_tail(inode); |
66f3b8e2 | 638 | if (is_blkdev_sb) { |
1da177e4 LT |
639 | /* |
640 | * Dirty memory-backed blockdev: the ramdisk | |
641 | * driver does this. Skip just this inode | |
642 | */ | |
643 | continue; | |
644 | } | |
645 | /* | |
646 | * Dirty memory-backed inode against a filesystem other | |
647 | * than the kernel-internal bdev filesystem. Skip the | |
648 | * entire superblock. | |
649 | */ | |
650 | break; | |
651 | } | |
652 | ||
84a89245 | 653 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
654 | requeue_io(inode); |
655 | continue; | |
656 | } | |
657 | ||
03ba3782 | 658 | if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { |
1da177e4 | 659 | wbc->encountered_congestion = 1; |
66f3b8e2 | 660 | if (!is_blkdev_sb) |
1da177e4 | 661 | break; /* Skip a congested fs */ |
0e0f4fc2 | 662 | requeue_io(inode); |
1da177e4 LT |
663 | continue; /* Skip a congested blockdev */ |
664 | } | |
665 | ||
d2caa3c5 JL |
666 | /* |
667 | * Was this inode dirtied after sync_sb_inodes was called? | |
668 | * This keeps sync from extra jobs and livelock. | |
669 | */ | |
670 | if (inode_dirtied_after(inode, start)) | |
1da177e4 LT |
671 | break; |
672 | ||
9ecc2738 | 673 | if (pin_sb_for_writeback(wbc, inode, &pin_sb)) { |
03ba3782 JA |
674 | requeue_io(inode); |
675 | continue; | |
676 | } | |
1da177e4 | 677 | |
84a89245 | 678 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
679 | __iget(inode); |
680 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 681 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
682 | if (wbc->pages_skipped != pages_skipped) { |
683 | /* | |
684 | * writeback is not making progress due to locked | |
685 | * buffers. Skip this inode for now. | |
686 | */ | |
f57b9b7b | 687 | redirty_tail(inode); |
1da177e4 LT |
688 | } |
689 | spin_unlock(&inode_lock); | |
1da177e4 | 690 | iput(inode); |
4ffc8444 | 691 | cond_resched(); |
1da177e4 | 692 | spin_lock(&inode_lock); |
8bc3be27 FW |
693 | if (wbc->nr_to_write <= 0) { |
694 | wbc->more_io = 1; | |
1da177e4 | 695 | break; |
8bc3be27 | 696 | } |
03ba3782 | 697 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 698 | wbc->more_io = 1; |
1da177e4 | 699 | } |
38f21977 | 700 | |
9ecc2738 JA |
701 | unpin_sb_for_writeback(&pin_sb); |
702 | ||
66f3b8e2 JA |
703 | spin_unlock(&inode_lock); |
704 | /* Leave any unwritten inodes on b_io */ | |
705 | } | |
706 | ||
03ba3782 JA |
707 | void writeback_inodes_wbc(struct writeback_control *wbc) |
708 | { | |
709 | struct backing_dev_info *bdi = wbc->bdi; | |
710 | ||
711 | writeback_inodes_wb(&bdi->wb, wbc); | |
712 | } | |
713 | ||
66f3b8e2 | 714 | /* |
03ba3782 JA |
715 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
716 | * operation. We do this so we don't hold I_SYNC against an inode for | |
717 | * enormous amounts of time, which would block a userspace task which has | |
718 | * been forced to throttle against that inode. Also, the code reevaluates | |
719 | * the dirty each time it has written this many pages. | |
720 | */ | |
721 | #define MAX_WRITEBACK_PAGES 1024 | |
722 | ||
723 | static inline bool over_bground_thresh(void) | |
724 | { | |
725 | unsigned long background_thresh, dirty_thresh; | |
726 | ||
727 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
728 | ||
729 | return (global_page_state(NR_FILE_DIRTY) + | |
730 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
731 | } | |
732 | ||
733 | /* | |
734 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 735 | * |
03ba3782 JA |
736 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
737 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
738 | * just walks the superblock inode list, writing back any inodes which are | |
739 | * older than a specific point in time. | |
66f3b8e2 | 740 | * |
03ba3782 JA |
741 | * Try to run once per dirty_writeback_interval. But if a writeback event |
742 | * takes longer than a dirty_writeback_interval interval, then leave a | |
743 | * one-second gap. | |
66f3b8e2 | 744 | * |
03ba3782 JA |
745 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
746 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 747 | */ |
c4a77a6c JA |
748 | static long wb_writeback(struct bdi_writeback *wb, |
749 | struct wb_writeback_args *args) | |
66f3b8e2 | 750 | { |
03ba3782 JA |
751 | struct writeback_control wbc = { |
752 | .bdi = wb->bdi, | |
c4a77a6c JA |
753 | .sb = args->sb, |
754 | .sync_mode = args->sync_mode, | |
03ba3782 | 755 | .older_than_this = NULL, |
c4a77a6c JA |
756 | .for_kupdate = args->for_kupdate, |
757 | .range_cyclic = args->range_cyclic, | |
03ba3782 JA |
758 | }; |
759 | unsigned long oldest_jif; | |
760 | long wrote = 0; | |
a5989bdc | 761 | struct inode *inode; |
66f3b8e2 | 762 | |
03ba3782 JA |
763 | if (wbc.for_kupdate) { |
764 | wbc.older_than_this = &oldest_jif; | |
765 | oldest_jif = jiffies - | |
766 | msecs_to_jiffies(dirty_expire_interval * 10); | |
767 | } | |
c4a77a6c JA |
768 | if (!wbc.range_cyclic) { |
769 | wbc.range_start = 0; | |
770 | wbc.range_end = LLONG_MAX; | |
771 | } | |
38f21977 | 772 | |
03ba3782 JA |
773 | for (;;) { |
774 | /* | |
d3ddec76 | 775 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 776 | */ |
d3ddec76 | 777 | if (args->nr_pages <= 0) |
03ba3782 | 778 | break; |
66f3b8e2 | 779 | |
38f21977 | 780 | /* |
d3ddec76 WF |
781 | * For background writeout, stop when we are below the |
782 | * background dirty threshold | |
38f21977 | 783 | */ |
d3ddec76 | 784 | if (args->for_background && !over_bground_thresh()) |
03ba3782 | 785 | break; |
38f21977 | 786 | |
03ba3782 JA |
787 | wbc.more_io = 0; |
788 | wbc.encountered_congestion = 0; | |
789 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; | |
790 | wbc.pages_skipped = 0; | |
791 | writeback_inodes_wb(wb, &wbc); | |
c4a77a6c | 792 | args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
793 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
794 | ||
795 | /* | |
71fd05a8 | 796 | * If we consumed everything, see if we have more |
03ba3782 | 797 | */ |
71fd05a8 JA |
798 | if (wbc.nr_to_write <= 0) |
799 | continue; | |
800 | /* | |
801 | * Didn't write everything and we don't have more IO, bail | |
802 | */ | |
803 | if (!wbc.more_io) | |
03ba3782 | 804 | break; |
71fd05a8 JA |
805 | /* |
806 | * Did we write something? Try for more | |
807 | */ | |
808 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
809 | continue; | |
810 | /* | |
811 | * Nothing written. Wait for some inode to | |
812 | * become available for writeback. Otherwise | |
813 | * we'll just busyloop. | |
814 | */ | |
815 | spin_lock(&inode_lock); | |
816 | if (!list_empty(&wb->b_more_io)) { | |
817 | inode = list_entry(wb->b_more_io.prev, | |
818 | struct inode, i_list); | |
819 | inode_wait_for_writeback(inode); | |
03ba3782 | 820 | } |
71fd05a8 | 821 | spin_unlock(&inode_lock); |
03ba3782 JA |
822 | } |
823 | ||
824 | return wrote; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Return the next bdi_work struct that hasn't been processed by this | |
8010c3b6 JA |
829 | * wb thread yet. ->seen is initially set for each thread that exists |
830 | * for this device, when a thread first notices a piece of work it | |
831 | * clears its bit. Depending on writeback type, the thread will notify | |
832 | * completion on either receiving the work (WB_SYNC_NONE) or after | |
833 | * it is done (WB_SYNC_ALL). | |
03ba3782 JA |
834 | */ |
835 | static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, | |
836 | struct bdi_writeback *wb) | |
837 | { | |
838 | struct bdi_work *work, *ret = NULL; | |
839 | ||
840 | rcu_read_lock(); | |
841 | ||
842 | list_for_each_entry_rcu(work, &bdi->work_list, list) { | |
77fad5e6 | 843 | if (!test_bit(wb->nr, &work->seen)) |
03ba3782 | 844 | continue; |
77fad5e6 | 845 | clear_bit(wb->nr, &work->seen); |
03ba3782 JA |
846 | |
847 | ret = work; | |
848 | break; | |
849 | } | |
850 | ||
851 | rcu_read_unlock(); | |
852 | return ret; | |
853 | } | |
854 | ||
855 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
856 | { | |
857 | unsigned long expired; | |
858 | long nr_pages; | |
859 | ||
860 | expired = wb->last_old_flush + | |
861 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
862 | if (time_before(jiffies, expired)) | |
863 | return 0; | |
864 | ||
865 | wb->last_old_flush = jiffies; | |
866 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
867 | global_page_state(NR_UNSTABLE_NFS) + | |
868 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
869 | ||
c4a77a6c JA |
870 | if (nr_pages) { |
871 | struct wb_writeback_args args = { | |
872 | .nr_pages = nr_pages, | |
873 | .sync_mode = WB_SYNC_NONE, | |
874 | .for_kupdate = 1, | |
875 | .range_cyclic = 1, | |
876 | }; | |
877 | ||
878 | return wb_writeback(wb, &args); | |
879 | } | |
03ba3782 JA |
880 | |
881 | return 0; | |
882 | } | |
883 | ||
884 | /* | |
885 | * Retrieve work items and do the writeback they describe | |
886 | */ | |
887 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
888 | { | |
889 | struct backing_dev_info *bdi = wb->bdi; | |
890 | struct bdi_work *work; | |
c4a77a6c | 891 | long wrote = 0; |
03ba3782 JA |
892 | |
893 | while ((work = get_next_work_item(bdi, wb)) != NULL) { | |
c4a77a6c | 894 | struct wb_writeback_args args = work->args; |
03ba3782 JA |
895 | |
896 | /* | |
897 | * Override sync mode, in case we must wait for completion | |
898 | */ | |
899 | if (force_wait) | |
c4a77a6c | 900 | work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; |
03ba3782 JA |
901 | |
902 | /* | |
903 | * If this isn't a data integrity operation, just notify | |
904 | * that we have seen this work and we are now starting it. | |
905 | */ | |
c4a77a6c | 906 | if (args.sync_mode == WB_SYNC_NONE) |
03ba3782 JA |
907 | wb_clear_pending(wb, work); |
908 | ||
c4a77a6c | 909 | wrote += wb_writeback(wb, &args); |
03ba3782 JA |
910 | |
911 | /* | |
912 | * This is a data integrity writeback, so only do the | |
913 | * notification when we have completed the work. | |
914 | */ | |
c4a77a6c | 915 | if (args.sync_mode == WB_SYNC_ALL) |
03ba3782 JA |
916 | wb_clear_pending(wb, work); |
917 | } | |
918 | ||
919 | /* | |
920 | * Check for periodic writeback, kupdated() style | |
921 | */ | |
922 | wrote += wb_check_old_data_flush(wb); | |
923 | ||
924 | return wrote; | |
925 | } | |
926 | ||
927 | /* | |
928 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
929 | * wakes up periodically and does kupdated style flushing. | |
930 | */ | |
931 | int bdi_writeback_task(struct bdi_writeback *wb) | |
932 | { | |
933 | unsigned long last_active = jiffies; | |
934 | unsigned long wait_jiffies = -1UL; | |
935 | long pages_written; | |
936 | ||
937 | while (!kthread_should_stop()) { | |
938 | pages_written = wb_do_writeback(wb, 0); | |
939 | ||
940 | if (pages_written) | |
941 | last_active = jiffies; | |
942 | else if (wait_jiffies != -1UL) { | |
943 | unsigned long max_idle; | |
944 | ||
38f21977 | 945 | /* |
03ba3782 JA |
946 | * Longest period of inactivity that we tolerate. If we |
947 | * see dirty data again later, the task will get | |
948 | * recreated automatically. | |
38f21977 | 949 | */ |
03ba3782 JA |
950 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
951 | if (time_after(jiffies, max_idle + last_active)) | |
952 | break; | |
953 | } | |
954 | ||
955 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
49db0414 | 956 | schedule_timeout_interruptible(wait_jiffies); |
03ba3782 JA |
957 | try_to_freeze(); |
958 | } | |
959 | ||
960 | return 0; | |
961 | } | |
962 | ||
963 | /* | |
b6e51316 JA |
964 | * Schedule writeback for all backing devices. This does WB_SYNC_NONE |
965 | * writeback, for integrity writeback see bdi_sync_writeback(). | |
03ba3782 | 966 | */ |
b6e51316 | 967 | static void bdi_writeback_all(struct super_block *sb, long nr_pages) |
03ba3782 | 968 | { |
b6e51316 JA |
969 | struct wb_writeback_args args = { |
970 | .sb = sb, | |
971 | .nr_pages = nr_pages, | |
972 | .sync_mode = WB_SYNC_NONE, | |
973 | }; | |
03ba3782 | 974 | struct backing_dev_info *bdi; |
03ba3782 | 975 | |
cfc4ba53 | 976 | rcu_read_lock(); |
03ba3782 | 977 | |
cfc4ba53 | 978 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
979 | if (!bdi_has_dirty_io(bdi)) |
980 | continue; | |
38f21977 | 981 | |
b6e51316 | 982 | bdi_alloc_queue_work(bdi, &args); |
03ba3782 JA |
983 | } |
984 | ||
cfc4ba53 | 985 | rcu_read_unlock(); |
1da177e4 LT |
986 | } |
987 | ||
988 | /* | |
03ba3782 JA |
989 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
990 | * the whole world. | |
991 | */ | |
992 | void wakeup_flusher_threads(long nr_pages) | |
993 | { | |
03ba3782 JA |
994 | if (nr_pages == 0) |
995 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
996 | global_page_state(NR_UNSTABLE_NFS); | |
b6e51316 | 997 | bdi_writeback_all(NULL, nr_pages); |
03ba3782 JA |
998 | } |
999 | ||
1000 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
1001 | { | |
1002 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1003 | struct dentry *dentry; | |
1004 | const char *name = "?"; | |
1005 | ||
1006 | dentry = d_find_alias(inode); | |
1007 | if (dentry) { | |
1008 | spin_lock(&dentry->d_lock); | |
1009 | name = (const char *) dentry->d_name.name; | |
1010 | } | |
1011 | printk(KERN_DEBUG | |
1012 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1013 | current->comm, task_pid_nr(current), inode->i_ino, | |
1014 | name, inode->i_sb->s_id); | |
1015 | if (dentry) { | |
1016 | spin_unlock(&dentry->d_lock); | |
1017 | dput(dentry); | |
1018 | } | |
1019 | } | |
1020 | } | |
1021 | ||
1022 | /** | |
1023 | * __mark_inode_dirty - internal function | |
1024 | * @inode: inode to mark | |
1025 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1026 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1027 | * mark_inode_dirty_sync. | |
1da177e4 | 1028 | * |
03ba3782 JA |
1029 | * Put the inode on the super block's dirty list. |
1030 | * | |
1031 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1032 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1033 | * If it was not hashed, it will never be added to the dirty list | |
1034 | * even if it is later hashed, as it will have been marked dirty already. | |
1035 | * | |
1036 | * In short, make sure you hash any inodes _before_ you start marking | |
1037 | * them dirty. | |
1da177e4 | 1038 | * |
03ba3782 JA |
1039 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
1040 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 1041 | * |
03ba3782 JA |
1042 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1043 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1044 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1045 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1046 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1047 | * blockdev inode. | |
1da177e4 | 1048 | */ |
03ba3782 | 1049 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1050 | { |
03ba3782 | 1051 | struct super_block *sb = inode->i_sb; |
1da177e4 | 1052 | |
03ba3782 JA |
1053 | /* |
1054 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1055 | * dirty the inode itself | |
1056 | */ | |
1057 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1058 | if (sb->s_op->dirty_inode) | |
1059 | sb->s_op->dirty_inode(inode); | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * make sure that changes are seen by all cpus before we test i_state | |
1064 | * -- mikulas | |
1065 | */ | |
1066 | smp_mb(); | |
1067 | ||
1068 | /* avoid the locking if we can */ | |
1069 | if ((inode->i_state & flags) == flags) | |
1070 | return; | |
1071 | ||
1072 | if (unlikely(block_dump)) | |
1073 | block_dump___mark_inode_dirty(inode); | |
1074 | ||
1075 | spin_lock(&inode_lock); | |
1076 | if ((inode->i_state & flags) != flags) { | |
1077 | const int was_dirty = inode->i_state & I_DIRTY; | |
1078 | ||
1079 | inode->i_state |= flags; | |
1080 | ||
1081 | /* | |
1082 | * If the inode is being synced, just update its dirty state. | |
1083 | * The unlocker will place the inode on the appropriate | |
1084 | * superblock list, based upon its state. | |
1085 | */ | |
1086 | if (inode->i_state & I_SYNC) | |
1087 | goto out; | |
1088 | ||
1089 | /* | |
1090 | * Only add valid (hashed) inodes to the superblock's | |
1091 | * dirty list. Add blockdev inodes as well. | |
1092 | */ | |
1093 | if (!S_ISBLK(inode->i_mode)) { | |
1094 | if (hlist_unhashed(&inode->i_hash)) | |
1095 | goto out; | |
1096 | } | |
1097 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
1098 | goto out; | |
1099 | ||
1100 | /* | |
1101 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1102 | * reposition it (that would break b_dirty time-ordering). | |
1103 | */ | |
1104 | if (!was_dirty) { | |
1105 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
1106 | struct backing_dev_info *bdi = wb->bdi; |
1107 | ||
1108 | if (bdi_cap_writeback_dirty(bdi) && | |
1109 | !test_bit(BDI_registered, &bdi->state)) { | |
1110 | WARN_ON(1); | |
1111 | printk(KERN_ERR "bdi-%s not registered\n", | |
1112 | bdi->name); | |
1113 | } | |
03ba3782 JA |
1114 | |
1115 | inode->dirtied_when = jiffies; | |
1116 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 1117 | } |
1da177e4 | 1118 | } |
03ba3782 JA |
1119 | out: |
1120 | spin_unlock(&inode_lock); | |
1121 | } | |
1122 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1123 | ||
1124 | /* | |
1125 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1126 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1127 | * | |
1128 | * If older_than_this is non-NULL, then only write out inodes which | |
1129 | * had their first dirtying at a time earlier than *older_than_this. | |
1130 | * | |
03ba3782 JA |
1131 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1132 | * This function assumes that the blockdev superblock's inodes are backed by | |
1133 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1134 | * assume that all inodes are backed by the same queue. | |
1135 | * | |
1136 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1137 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1138 | * on the writer throttling path, and we get decent balancing between many | |
1139 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1140 | */ | |
b6e51316 | 1141 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1142 | { |
1143 | struct inode *inode, *old_inode = NULL; | |
1144 | ||
1145 | /* | |
1146 | * We need to be protected against the filesystem going from | |
1147 | * r/o to r/w or vice versa. | |
1148 | */ | |
b6e51316 | 1149 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
1150 | |
1151 | spin_lock(&inode_lock); | |
1152 | ||
1153 | /* | |
1154 | * Data integrity sync. Must wait for all pages under writeback, | |
1155 | * because there may have been pages dirtied before our sync | |
1156 | * call, but which had writeout started before we write it out. | |
1157 | * In which case, the inode may not be on the dirty list, but | |
1158 | * we still have to wait for that writeout. | |
1159 | */ | |
b6e51316 | 1160 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1161 | struct address_space *mapping; |
1162 | ||
1163 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1164 | continue; | |
1165 | mapping = inode->i_mapping; | |
1166 | if (mapping->nrpages == 0) | |
1167 | continue; | |
1168 | __iget(inode); | |
1169 | spin_unlock(&inode_lock); | |
1170 | /* | |
1171 | * We hold a reference to 'inode' so it couldn't have | |
1172 | * been removed from s_inodes list while we dropped the | |
1173 | * inode_lock. We cannot iput the inode now as we can | |
1174 | * be holding the last reference and we cannot iput it | |
1175 | * under inode_lock. So we keep the reference and iput | |
1176 | * it later. | |
1177 | */ | |
1178 | iput(old_inode); | |
1179 | old_inode = inode; | |
1180 | ||
1181 | filemap_fdatawait(mapping); | |
1182 | ||
1183 | cond_resched(); | |
1184 | ||
1185 | spin_lock(&inode_lock); | |
1186 | } | |
1187 | spin_unlock(&inode_lock); | |
1188 | iput(old_inode); | |
1da177e4 LT |
1189 | } |
1190 | ||
d8a8559c JA |
1191 | /** |
1192 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1193 | * @sb: the superblock | |
1da177e4 | 1194 | * |
d8a8559c JA |
1195 | * Start writeback on some inodes on this super_block. No guarantees are made |
1196 | * on how many (if any) will be written, and this function does not wait | |
1197 | * for IO completion of submitted IO. The number of pages submitted is | |
1198 | * returned. | |
1da177e4 | 1199 | */ |
b6e51316 | 1200 | void writeback_inodes_sb(struct super_block *sb) |
1da177e4 | 1201 | { |
d8a8559c JA |
1202 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
1203 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
1204 | long nr_to_write; | |
1da177e4 | 1205 | |
d8a8559c | 1206 | nr_to_write = nr_dirty + nr_unstable + |
38f21977 | 1207 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
38f21977 | 1208 | |
56a131dc | 1209 | bdi_start_writeback(sb->s_bdi, nr_to_write); |
d8a8559c JA |
1210 | } |
1211 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1212 | ||
1213 | /** | |
1214 | * sync_inodes_sb - sync sb inode pages | |
1215 | * @sb: the superblock | |
1216 | * | |
1217 | * This function writes and waits on any dirty inode belonging to this | |
1218 | * super_block. The number of pages synced is returned. | |
1219 | */ | |
b6e51316 | 1220 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1221 | { |
b6e51316 JA |
1222 | bdi_sync_writeback(sb->s_bdi, sb); |
1223 | wait_sb_inodes(sb); | |
1da177e4 | 1224 | } |
d8a8559c | 1225 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1226 | |
1da177e4 | 1227 | /** |
7f04c26d AA |
1228 | * write_inode_now - write an inode to disk |
1229 | * @inode: inode to write to disk | |
1230 | * @sync: whether the write should be synchronous or not | |
1231 | * | |
1232 | * This function commits an inode to disk immediately if it is dirty. This is | |
1233 | * primarily needed by knfsd. | |
1da177e4 | 1234 | * |
7f04c26d | 1235 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1236 | */ |
1da177e4 LT |
1237 | int write_inode_now(struct inode *inode, int sync) |
1238 | { | |
1239 | int ret; | |
1240 | struct writeback_control wbc = { | |
1241 | .nr_to_write = LONG_MAX, | |
18914b18 | 1242 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1243 | .range_start = 0, |
1244 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1245 | }; |
1246 | ||
1247 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1248 | wbc.nr_to_write = 0; |
1da177e4 LT |
1249 | |
1250 | might_sleep(); | |
1251 | spin_lock(&inode_lock); | |
01c03194 | 1252 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1253 | spin_unlock(&inode_lock); |
1254 | if (sync) | |
1c0eeaf5 | 1255 | inode_sync_wait(inode); |
1da177e4 LT |
1256 | return ret; |
1257 | } | |
1258 | EXPORT_SYMBOL(write_inode_now); | |
1259 | ||
1260 | /** | |
1261 | * sync_inode - write an inode and its pages to disk. | |
1262 | * @inode: the inode to sync | |
1263 | * @wbc: controls the writeback mode | |
1264 | * | |
1265 | * sync_inode() will write an inode and its pages to disk. It will also | |
1266 | * correctly update the inode on its superblock's dirty inode lists and will | |
1267 | * update inode->i_state. | |
1268 | * | |
1269 | * The caller must have a ref on the inode. | |
1270 | */ | |
1271 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1272 | { | |
1273 | int ret; | |
1274 | ||
1275 | spin_lock(&inode_lock); | |
01c03194 | 1276 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1277 | spin_unlock(&inode_lock); |
1278 | return ret; | |
1279 | } | |
1280 | EXPORT_SYMBOL(sync_inode); |