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