Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/fs-writeback.c | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
6 | * Contains all the functions related to writing back and waiting | |
7 | * upon dirty inodes against superblocks, and writing back dirty | |
8 | * pages against inodes. ie: data writeback. Writeout of the | |
9 | * inode itself is not handled here. | |
10 | * | |
e1f8e874 | 11 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
12 | * Split out of fs/inode.c |
13 | * Additions for address_space-based writeback | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
f5ff8422 | 17 | #include <linux/module.h> |
1da177e4 | 18 | #include <linux/spinlock.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
1da177e4 LT |
20 | #include <linux/sched.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/mm.h> | |
03ba3782 JA |
23 | #include <linux/kthread.h> |
24 | #include <linux/freezer.h> | |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/buffer_head.h> | |
07f3f05c | 29 | #include "internal.h" |
1da177e4 | 30 | |
66f3b8e2 | 31 | #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) |
f11b00f3 | 32 | |
d0bceac7 JA |
33 | /* |
34 | * We don't actually have pdflush, but this one is exported though /proc... | |
35 | */ | |
36 | int nr_pdflush_threads; | |
37 | ||
c4a77a6c JA |
38 | /* |
39 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
40 | */ | |
41 | struct wb_writeback_args { | |
42 | long nr_pages; | |
43 | struct super_block *sb; | |
44 | enum writeback_sync_modes sync_mode; | |
52957fe1 HS |
45 | unsigned int for_kupdate:1; |
46 | unsigned int range_cyclic:1; | |
47 | unsigned int for_background:1; | |
e8bebe2f | 48 | unsigned int sb_pinned:1; |
c4a77a6c JA |
49 | }; |
50 | ||
03ba3782 JA |
51 | /* |
52 | * Work items for the bdi_writeback threads | |
f11b00f3 | 53 | */ |
03ba3782 | 54 | struct bdi_work { |
8010c3b6 JA |
55 | struct list_head list; /* pending work list */ |
56 | struct rcu_head rcu_head; /* for RCU free/clear of work */ | |
03ba3782 | 57 | |
8010c3b6 JA |
58 | unsigned long seen; /* threads that have seen this work */ |
59 | atomic_t pending; /* number of threads still to do work */ | |
03ba3782 | 60 | |
8010c3b6 | 61 | struct wb_writeback_args args; /* writeback arguments */ |
03ba3782 | 62 | |
8010c3b6 | 63 | unsigned long state; /* flag bits, see WS_* */ |
03ba3782 JA |
64 | }; |
65 | ||
66 | enum { | |
67 | WS_USED_B = 0, | |
68 | WS_ONSTACK_B, | |
69 | }; | |
70 | ||
71 | #define WS_USED (1 << WS_USED_B) | |
72 | #define WS_ONSTACK (1 << WS_ONSTACK_B) | |
73 | ||
74 | static inline bool bdi_work_on_stack(struct bdi_work *work) | |
75 | { | |
76 | return test_bit(WS_ONSTACK_B, &work->state); | |
77 | } | |
78 | ||
79 | static inline void bdi_work_init(struct bdi_work *work, | |
b6e51316 | 80 | struct wb_writeback_args *args) |
03ba3782 JA |
81 | { |
82 | INIT_RCU_HEAD(&work->rcu_head); | |
b6e51316 | 83 | work->args = *args; |
03ba3782 JA |
84 | work->state = WS_USED; |
85 | } | |
86 | ||
f11b00f3 AB |
87 | /** |
88 | * writeback_in_progress - determine whether there is writeback in progress | |
89 | * @bdi: the device's backing_dev_info structure. | |
90 | * | |
03ba3782 JA |
91 | * Determine whether there is writeback waiting to be handled against a |
92 | * backing device. | |
f11b00f3 AB |
93 | */ |
94 | int writeback_in_progress(struct backing_dev_info *bdi) | |
95 | { | |
03ba3782 | 96 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
97 | } |
98 | ||
03ba3782 | 99 | static void bdi_work_clear(struct bdi_work *work) |
f11b00f3 | 100 | { |
03ba3782 JA |
101 | clear_bit(WS_USED_B, &work->state); |
102 | smp_mb__after_clear_bit(); | |
1ef7d9aa NP |
103 | /* |
104 | * work can have disappeared at this point. bit waitq functions | |
105 | * should be able to tolerate this, provided bdi_sched_wait does | |
106 | * not dereference it's pointer argument. | |
107 | */ | |
03ba3782 | 108 | wake_up_bit(&work->state, WS_USED_B); |
f11b00f3 AB |
109 | } |
110 | ||
03ba3782 | 111 | static void bdi_work_free(struct rcu_head *head) |
4195f73d | 112 | { |
03ba3782 | 113 | struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); |
4195f73d | 114 | |
03ba3782 JA |
115 | if (!bdi_work_on_stack(work)) |
116 | kfree(work); | |
117 | else | |
118 | bdi_work_clear(work); | |
4195f73d NP |
119 | } |
120 | ||
03ba3782 | 121 | static void wb_work_complete(struct bdi_work *work) |
1da177e4 | 122 | { |
c4a77a6c | 123 | const enum writeback_sync_modes sync_mode = work->args.sync_mode; |
77b9d059 | 124 | int onstack = bdi_work_on_stack(work); |
1da177e4 LT |
125 | |
126 | /* | |
03ba3782 JA |
127 | * For allocated work, we can clear the done/seen bit right here. |
128 | * For on-stack work, we need to postpone both the clear and free | |
129 | * to after the RCU grace period, since the stack could be invalidated | |
130 | * as soon as bdi_work_clear() has done the wakeup. | |
1da177e4 | 131 | */ |
77b9d059 | 132 | if (!onstack) |
03ba3782 | 133 | bdi_work_clear(work); |
77b9d059 | 134 | if (sync_mode == WB_SYNC_NONE || onstack) |
03ba3782 JA |
135 | call_rcu(&work->rcu_head, bdi_work_free); |
136 | } | |
1da177e4 | 137 | |
03ba3782 JA |
138 | static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) |
139 | { | |
1da177e4 | 140 | /* |
03ba3782 JA |
141 | * The caller has retrieved the work arguments from this work, |
142 | * drop our reference. If this is the last ref, delete and free it | |
1da177e4 | 143 | */ |
03ba3782 JA |
144 | if (atomic_dec_and_test(&work->pending)) { |
145 | struct backing_dev_info *bdi = wb->bdi; | |
1da177e4 | 146 | |
03ba3782 JA |
147 | spin_lock(&bdi->wb_lock); |
148 | list_del_rcu(&work->list); | |
149 | spin_unlock(&bdi->wb_lock); | |
1da177e4 | 150 | |
03ba3782 JA |
151 | wb_work_complete(work); |
152 | } | |
153 | } | |
1da177e4 | 154 | |
03ba3782 JA |
155 | static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) |
156 | { | |
bcddc3f0 JA |
157 | work->seen = bdi->wb_mask; |
158 | BUG_ON(!work->seen); | |
159 | atomic_set(&work->pending, bdi->wb_cnt); | |
160 | BUG_ON(!bdi->wb_cnt); | |
1da177e4 | 161 | |
bcddc3f0 | 162 | /* |
deed62ed NP |
163 | * list_add_tail_rcu() contains the necessary barriers to |
164 | * make sure the above stores are seen before the item is | |
165 | * noticed on the list | |
bcddc3f0 | 166 | */ |
bcddc3f0 JA |
167 | spin_lock(&bdi->wb_lock); |
168 | list_add_tail_rcu(&work->list, &bdi->work_list); | |
169 | spin_unlock(&bdi->wb_lock); | |
03ba3782 JA |
170 | |
171 | /* | |
172 | * If the default thread isn't there, make sure we add it. When | |
173 | * it gets created and wakes up, we'll run this work. | |
174 | */ | |
175 | if (unlikely(list_empty_careful(&bdi->wb_list))) | |
176 | wake_up_process(default_backing_dev_info.wb.task); | |
177 | else { | |
178 | struct bdi_writeback *wb = &bdi->wb; | |
1da177e4 | 179 | |
1ef7d9aa | 180 | if (wb->task) |
03ba3782 | 181 | wake_up_process(wb->task); |
1da177e4 | 182 | } |
1da177e4 LT |
183 | } |
184 | ||
03ba3782 JA |
185 | /* |
186 | * Used for on-stack allocated work items. The caller needs to wait until | |
187 | * the wb threads have acked the work before it's safe to continue. | |
188 | */ | |
189 | static void bdi_wait_on_work_clear(struct bdi_work *work) | |
190 | { | |
191 | wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, | |
192 | TASK_UNINTERRUPTIBLE); | |
193 | } | |
1da177e4 | 194 | |
f11fcae8 | 195 | static void bdi_alloc_queue_work(struct backing_dev_info *bdi, |
7c8a3554 JA |
196 | struct wb_writeback_args *args, |
197 | int wait) | |
1da177e4 | 198 | { |
03ba3782 JA |
199 | struct bdi_work *work; |
200 | ||
bcddc3f0 JA |
201 | /* |
202 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
203 | * wakeup the thread for old dirty data writeback | |
204 | */ | |
03ba3782 | 205 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
bcddc3f0 | 206 | if (work) { |
b6e51316 | 207 | bdi_work_init(work, args); |
bcddc3f0 | 208 | bdi_queue_work(bdi, work); |
7c8a3554 JA |
209 | if (wait) |
210 | bdi_wait_on_work_clear(work); | |
bcddc3f0 JA |
211 | } else { |
212 | struct bdi_writeback *wb = &bdi->wb; | |
03ba3782 | 213 | |
bcddc3f0 JA |
214 | if (wb->task) |
215 | wake_up_process(wb->task); | |
216 | } | |
03ba3782 JA |
217 | } |
218 | ||
b6e51316 JA |
219 | /** |
220 | * bdi_sync_writeback - start and wait for writeback | |
221 | * @bdi: the backing device to write from | |
222 | * @sb: write inodes from this super_block | |
223 | * | |
224 | * Description: | |
225 | * This does WB_SYNC_ALL data integrity writeback and waits for the | |
226 | * IO to complete. Callers must hold the sb s_umount semaphore for | |
227 | * reading, to avoid having the super disappear before we are done. | |
228 | */ | |
229 | static void bdi_sync_writeback(struct backing_dev_info *bdi, | |
230 | struct super_block *sb) | |
03ba3782 | 231 | { |
b6e51316 JA |
232 | struct wb_writeback_args args = { |
233 | .sb = sb, | |
234 | .sync_mode = WB_SYNC_ALL, | |
235 | .nr_pages = LONG_MAX, | |
236 | .range_cyclic = 0, | |
e913fc82 JA |
237 | /* |
238 | * Setting sb_pinned is not necessary for WB_SYNC_ALL, but | |
239 | * lets make it explicitly clear. | |
240 | */ | |
241 | .sb_pinned = 1, | |
b6e51316 JA |
242 | }; |
243 | struct bdi_work work; | |
03ba3782 | 244 | |
b6e51316 JA |
245 | bdi_work_init(&work, &args); |
246 | work.state |= WS_ONSTACK; | |
03ba3782 | 247 | |
b6e51316 JA |
248 | bdi_queue_work(bdi, &work); |
249 | bdi_wait_on_work_clear(&work); | |
250 | } | |
251 | ||
252 | /** | |
253 | * bdi_start_writeback - start writeback | |
254 | * @bdi: the backing device to write from | |
4b6764fa | 255 | * @sb: write inodes from this super_block |
b6e51316 | 256 | * @nr_pages: the number of pages to write |
e913fc82 | 257 | * @sb_locked: caller already holds sb umount sem. |
b6e51316 JA |
258 | * |
259 | * Description: | |
260 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
261 | * started when this function returns, we make no guarentees on | |
e913fc82 | 262 | * completion. Caller specifies whether sb umount sem is held already or not. |
b6e51316 JA |
263 | * |
264 | */ | |
a72bfd4d | 265 | void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb, |
e913fc82 | 266 | long nr_pages, int sb_locked) |
b6e51316 JA |
267 | { |
268 | struct wb_writeback_args args = { | |
a72bfd4d | 269 | .sb = sb, |
b6e51316 JA |
270 | .sync_mode = WB_SYNC_NONE, |
271 | .nr_pages = nr_pages, | |
272 | .range_cyclic = 1, | |
e913fc82 | 273 | .sb_pinned = sb_locked, |
b6e51316 JA |
274 | }; |
275 | ||
d3ddec76 WF |
276 | /* |
277 | * We treat @nr_pages=0 as the special case to do background writeback, | |
278 | * ie. to sync pages until the background dirty threshold is reached. | |
279 | */ | |
280 | if (!nr_pages) { | |
281 | args.nr_pages = LONG_MAX; | |
282 | args.for_background = 1; | |
283 | } | |
284 | ||
7c8a3554 | 285 | bdi_alloc_queue_work(bdi, &args, sb_locked); |
1da177e4 LT |
286 | } |
287 | ||
6610a0bc AM |
288 | /* |
289 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
290 | * furthest end of its superblock's dirty-inode list. | |
291 | * | |
292 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 293 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
294 | * the case then the inode must have been redirtied while it was being written |
295 | * out and we don't reset its dirtied_when. | |
296 | */ | |
297 | static void redirty_tail(struct inode *inode) | |
298 | { | |
03ba3782 | 299 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 300 | |
03ba3782 | 301 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 302 | struct inode *tail; |
6610a0bc | 303 | |
03ba3782 | 304 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 305 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
306 | inode->dirtied_when = jiffies; |
307 | } | |
03ba3782 | 308 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
309 | } |
310 | ||
c986d1e2 | 311 | /* |
66f3b8e2 | 312 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 313 | */ |
0e0f4fc2 | 314 | static void requeue_io(struct inode *inode) |
c986d1e2 | 315 | { |
03ba3782 JA |
316 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
317 | ||
318 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
319 | } |
320 | ||
1c0eeaf5 JE |
321 | static void inode_sync_complete(struct inode *inode) |
322 | { | |
323 | /* | |
324 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
325 | */ | |
326 | smp_mb(); | |
327 | wake_up_bit(&inode->i_state, __I_SYNC); | |
328 | } | |
329 | ||
d2caa3c5 JL |
330 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
331 | { | |
332 | bool ret = time_after(inode->dirtied_when, t); | |
333 | #ifndef CONFIG_64BIT | |
334 | /* | |
335 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
336 | * It _appears_ to be in the future, but is actually in distant past. | |
337 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 338 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
339 | */ |
340 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
341 | #endif | |
342 | return ret; | |
343 | } | |
344 | ||
2c136579 FW |
345 | /* |
346 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
347 | */ | |
348 | static void move_expired_inodes(struct list_head *delaying_queue, | |
349 | struct list_head *dispatch_queue, | |
350 | unsigned long *older_than_this) | |
351 | { | |
5c03449d SL |
352 | LIST_HEAD(tmp); |
353 | struct list_head *pos, *node; | |
cf137307 | 354 | struct super_block *sb = NULL; |
5c03449d | 355 | struct inode *inode; |
cf137307 | 356 | int do_sb_sort = 0; |
5c03449d | 357 | |
2c136579 | 358 | while (!list_empty(delaying_queue)) { |
5c03449d | 359 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
2c136579 | 360 | if (older_than_this && |
d2caa3c5 | 361 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 362 | break; |
cf137307 JA |
363 | if (sb && sb != inode->i_sb) |
364 | do_sb_sort = 1; | |
365 | sb = inode->i_sb; | |
5c03449d SL |
366 | list_move(&inode->i_list, &tmp); |
367 | } | |
368 | ||
cf137307 JA |
369 | /* just one sb in list, splice to dispatch_queue and we're done */ |
370 | if (!do_sb_sort) { | |
371 | list_splice(&tmp, dispatch_queue); | |
372 | return; | |
373 | } | |
374 | ||
5c03449d SL |
375 | /* Move inodes from one superblock together */ |
376 | while (!list_empty(&tmp)) { | |
377 | inode = list_entry(tmp.prev, struct inode, i_list); | |
378 | sb = inode->i_sb; | |
379 | list_for_each_prev_safe(pos, node, &tmp) { | |
380 | inode = list_entry(pos, struct inode, i_list); | |
381 | if (inode->i_sb == sb) | |
382 | list_move(&inode->i_list, dispatch_queue); | |
383 | } | |
2c136579 FW |
384 | } |
385 | } | |
386 | ||
387 | /* | |
388 | * Queue all expired dirty inodes for io, eldest first. | |
389 | */ | |
03ba3782 | 390 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 391 | { |
03ba3782 JA |
392 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
393 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
394 | } |
395 | ||
a9185b41 | 396 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 397 | { |
03ba3782 | 398 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 399 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 400 | return 0; |
08d8e974 | 401 | } |
08d8e974 | 402 | |
1da177e4 | 403 | /* |
01c03194 CH |
404 | * Wait for writeback on an inode to complete. |
405 | */ | |
406 | static void inode_wait_for_writeback(struct inode *inode) | |
407 | { | |
408 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
409 | wait_queue_head_t *wqh; | |
410 | ||
411 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
412 | do { | |
413 | spin_unlock(&inode_lock); | |
414 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
415 | spin_lock(&inode_lock); | |
416 | } while (inode->i_state & I_SYNC); | |
417 | } | |
418 | ||
419 | /* | |
420 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
421 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
422 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
423 | * | |
1da177e4 LT |
424 | * If `wait' is set, wait on the writeout. |
425 | * | |
426 | * The whole writeout design is quite complex and fragile. We want to avoid | |
427 | * starvation of particular inodes when others are being redirtied, prevent | |
428 | * livelocks, etc. | |
429 | * | |
430 | * Called under inode_lock. | |
431 | */ | |
432 | static int | |
01c03194 | 433 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 434 | { |
1da177e4 | 435 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 436 | unsigned dirty; |
1da177e4 LT |
437 | int ret; |
438 | ||
01c03194 CH |
439 | if (!atomic_read(&inode->i_count)) |
440 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
441 | else | |
442 | WARN_ON(inode->i_state & I_WILL_FREE); | |
443 | ||
444 | if (inode->i_state & I_SYNC) { | |
445 | /* | |
446 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 447 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
448 | * writeback can proceed with the other inodes on s_io. |
449 | * | |
450 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 451 | * completed a full scan of b_io. |
01c03194 | 452 | */ |
a9185b41 | 453 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
454 | requeue_io(inode); |
455 | return 0; | |
456 | } | |
457 | ||
458 | /* | |
459 | * It's a data-integrity sync. We must wait. | |
460 | */ | |
461 | inode_wait_for_writeback(inode); | |
462 | } | |
463 | ||
1c0eeaf5 | 464 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 465 | |
5547e8aa | 466 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 467 | inode->i_state |= I_SYNC; |
5547e8aa | 468 | inode->i_state &= ~I_DIRTY_PAGES; |
1da177e4 LT |
469 | spin_unlock(&inode_lock); |
470 | ||
471 | ret = do_writepages(mapping, wbc); | |
472 | ||
26821ed4 CH |
473 | /* |
474 | * Make sure to wait on the data before writing out the metadata. | |
475 | * This is important for filesystems that modify metadata on data | |
476 | * I/O completion. | |
477 | */ | |
a9185b41 | 478 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 479 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
480 | if (ret == 0) |
481 | ret = err; | |
482 | } | |
483 | ||
5547e8aa DM |
484 | /* |
485 | * Some filesystems may redirty the inode during the writeback | |
486 | * due to delalloc, clear dirty metadata flags right before | |
487 | * write_inode() | |
488 | */ | |
489 | spin_lock(&inode_lock); | |
490 | dirty = inode->i_state & I_DIRTY; | |
491 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
492 | spin_unlock(&inode_lock); | |
26821ed4 CH |
493 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
494 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 495 | int err = write_inode(inode, wbc); |
1da177e4 LT |
496 | if (ret == 0) |
497 | ret = err; | |
498 | } | |
499 | ||
500 | spin_lock(&inode_lock); | |
1c0eeaf5 | 501 | inode->i_state &= ~I_SYNC; |
84a89245 | 502 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
b3af9468 | 503 | if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { |
ae1b7f7d | 504 | /* |
b3af9468 WF |
505 | * More pages get dirtied by a fast dirtier. |
506 | */ | |
507 | goto select_queue; | |
508 | } else if (inode->i_state & I_DIRTY) { | |
509 | /* | |
510 | * At least XFS will redirty the inode during the | |
511 | * writeback (delalloc) and on io completion (isize). | |
ae1b7f7d WF |
512 | */ |
513 | redirty_tail(inode); | |
514 | } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
1da177e4 LT |
515 | /* |
516 | * We didn't write back all the pages. nfs_writepages() | |
517 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 518 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
519 | */ |
520 | /* | |
521 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 522 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
523 | * consideration. Otherwise, move it to the tail, for |
524 | * the reasons described there. I'm not really sure | |
525 | * how much sense this makes. Presumably I had a good | |
526 | * reasons for doing it this way, and I'd rather not | |
527 | * muck with it at present. | |
1da177e4 LT |
528 | */ |
529 | if (wbc->for_kupdate) { | |
530 | /* | |
2c136579 | 531 | * For the kupdate function we move the inode |
66f3b8e2 | 532 | * to b_more_io so it will get more writeout as |
2c136579 | 533 | * soon as the queue becomes uncongested. |
1da177e4 LT |
534 | */ |
535 | inode->i_state |= I_DIRTY_PAGES; | |
b3af9468 | 536 | select_queue: |
8bc3be27 FW |
537 | if (wbc->nr_to_write <= 0) { |
538 | /* | |
539 | * slice used up: queue for next turn | |
540 | */ | |
541 | requeue_io(inode); | |
542 | } else { | |
543 | /* | |
544 | * somehow blocked: retry later | |
545 | */ | |
546 | redirty_tail(inode); | |
547 | } | |
1da177e4 LT |
548 | } else { |
549 | /* | |
550 | * Otherwise fully redirty the inode so that | |
551 | * other inodes on this superblock will get some | |
552 | * writeout. Otherwise heavy writing to one | |
553 | * file would indefinitely suspend writeout of | |
554 | * all the other files. | |
555 | */ | |
556 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 557 | redirty_tail(inode); |
1da177e4 | 558 | } |
1da177e4 LT |
559 | } else if (atomic_read(&inode->i_count)) { |
560 | /* | |
561 | * The inode is clean, inuse | |
562 | */ | |
563 | list_move(&inode->i_list, &inode_in_use); | |
564 | } else { | |
565 | /* | |
566 | * The inode is clean, unused | |
567 | */ | |
568 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
569 | } |
570 | } | |
1c0eeaf5 | 571 | inode_sync_complete(inode); |
1da177e4 LT |
572 | return ret; |
573 | } | |
574 | ||
f11c9c5c | 575 | static void unpin_sb_for_writeback(struct super_block *sb) |
9ecc2738 | 576 | { |
f11c9c5c ES |
577 | up_read(&sb->s_umount); |
578 | put_super(sb); | |
9ecc2738 JA |
579 | } |
580 | ||
f11c9c5c ES |
581 | enum sb_pin_state { |
582 | SB_PINNED, | |
583 | SB_NOT_PINNED, | |
584 | SB_PIN_FAILED | |
585 | }; | |
586 | ||
03ba3782 JA |
587 | /* |
588 | * For WB_SYNC_NONE writeback, the caller does not have the sb pinned | |
589 | * before calling writeback. So make sure that we do pin it, so it doesn't | |
590 | * go away while we are writing inodes from it. | |
03ba3782 | 591 | */ |
f11c9c5c ES |
592 | static enum sb_pin_state pin_sb_for_writeback(struct writeback_control *wbc, |
593 | struct super_block *sb) | |
03ba3782 | 594 | { |
03ba3782 JA |
595 | /* |
596 | * Caller must already hold the ref for this | |
597 | */ | |
e913fc82 | 598 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->sb_pinned) { |
03ba3782 | 599 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
f11c9c5c | 600 | return SB_NOT_PINNED; |
03ba3782 | 601 | } |
03ba3782 JA |
602 | spin_lock(&sb_lock); |
603 | sb->s_count++; | |
604 | if (down_read_trylock(&sb->s_umount)) { | |
605 | if (sb->s_root) { | |
606 | spin_unlock(&sb_lock); | |
f11c9c5c | 607 | return SB_PINNED; |
03ba3782 JA |
608 | } |
609 | /* | |
610 | * umounted, drop rwsem again and fall through to failure | |
611 | */ | |
612 | up_read(&sb->s_umount); | |
613 | } | |
03ba3782 JA |
614 | sb->s_count--; |
615 | spin_unlock(&sb_lock); | |
f11c9c5c | 616 | return SB_PIN_FAILED; |
03ba3782 JA |
617 | } |
618 | ||
f11c9c5c ES |
619 | /* |
620 | * Write a portion of b_io inodes which belong to @sb. | |
621 | * If @wbc->sb != NULL, then find and write all such | |
622 | * inodes. Otherwise write only ones which go sequentially | |
623 | * in reverse order. | |
624 | * Return 1, if the caller writeback routine should be | |
625 | * interrupted. Otherwise return 0. | |
626 | */ | |
627 | static int writeback_sb_inodes(struct super_block *sb, | |
628 | struct bdi_writeback *wb, | |
629 | struct writeback_control *wbc) | |
1da177e4 | 630 | { |
03ba3782 | 631 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 632 | long pages_skipped; |
f11c9c5c ES |
633 | struct inode *inode = list_entry(wb->b_io.prev, |
634 | struct inode, i_list); | |
635 | if (wbc->sb && sb != inode->i_sb) { | |
636 | /* super block given and doesn't | |
637 | match, skip this inode */ | |
66f3b8e2 JA |
638 | redirty_tail(inode); |
639 | continue; | |
640 | } | |
f11c9c5c ES |
641 | if (sb != inode->i_sb) |
642 | /* finish with this superblock */ | |
643 | return 0; | |
84a89245 | 644 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
645 | requeue_io(inode); |
646 | continue; | |
647 | } | |
d2caa3c5 JL |
648 | /* |
649 | * Was this inode dirtied after sync_sb_inodes was called? | |
650 | * This keeps sync from extra jobs and livelock. | |
651 | */ | |
f11c9c5c ES |
652 | if (inode_dirtied_after(inode, wbc->wb_start)) |
653 | return 1; | |
1da177e4 | 654 | |
84a89245 | 655 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
656 | __iget(inode); |
657 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 658 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
659 | if (wbc->pages_skipped != pages_skipped) { |
660 | /* | |
661 | * writeback is not making progress due to locked | |
662 | * buffers. Skip this inode for now. | |
663 | */ | |
f57b9b7b | 664 | redirty_tail(inode); |
1da177e4 LT |
665 | } |
666 | spin_unlock(&inode_lock); | |
1da177e4 | 667 | iput(inode); |
4ffc8444 | 668 | cond_resched(); |
1da177e4 | 669 | spin_lock(&inode_lock); |
8bc3be27 FW |
670 | if (wbc->nr_to_write <= 0) { |
671 | wbc->more_io = 1; | |
f11c9c5c | 672 | return 1; |
8bc3be27 | 673 | } |
03ba3782 | 674 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 675 | wbc->more_io = 1; |
1da177e4 | 676 | } |
f11c9c5c ES |
677 | /* b_io is empty */ |
678 | return 1; | |
679 | } | |
680 | ||
681 | static void writeback_inodes_wb(struct bdi_writeback *wb, | |
682 | struct writeback_control *wbc) | |
683 | { | |
684 | int ret = 0; | |
685 | ||
686 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
687 | spin_lock(&inode_lock); | |
688 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
689 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 690 | |
f11c9c5c ES |
691 | while (!list_empty(&wb->b_io)) { |
692 | struct inode *inode = list_entry(wb->b_io.prev, | |
693 | struct inode, i_list); | |
694 | struct super_block *sb = inode->i_sb; | |
695 | enum sb_pin_state state; | |
9ecc2738 | 696 | |
f11c9c5c ES |
697 | if (wbc->sb && sb != wbc->sb) { |
698 | /* super block given and doesn't | |
699 | match, skip this inode */ | |
700 | redirty_tail(inode); | |
701 | continue; | |
702 | } | |
703 | state = pin_sb_for_writeback(wbc, sb); | |
704 | ||
705 | if (state == SB_PIN_FAILED) { | |
706 | requeue_io(inode); | |
707 | continue; | |
708 | } | |
709 | ret = writeback_sb_inodes(sb, wb, wbc); | |
710 | ||
711 | if (state == SB_PINNED) | |
712 | unpin_sb_for_writeback(sb); | |
713 | if (ret) | |
714 | break; | |
715 | } | |
66f3b8e2 JA |
716 | spin_unlock(&inode_lock); |
717 | /* Leave any unwritten inodes on b_io */ | |
718 | } | |
719 | ||
03ba3782 JA |
720 | void writeback_inodes_wbc(struct writeback_control *wbc) |
721 | { | |
722 | struct backing_dev_info *bdi = wbc->bdi; | |
723 | ||
724 | writeback_inodes_wb(&bdi->wb, wbc); | |
725 | } | |
726 | ||
66f3b8e2 | 727 | /* |
03ba3782 JA |
728 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
729 | * operation. We do this so we don't hold I_SYNC against an inode for | |
730 | * enormous amounts of time, which would block a userspace task which has | |
731 | * been forced to throttle against that inode. Also, the code reevaluates | |
732 | * the dirty each time it has written this many pages. | |
733 | */ | |
734 | #define MAX_WRITEBACK_PAGES 1024 | |
735 | ||
736 | static inline bool over_bground_thresh(void) | |
737 | { | |
738 | unsigned long background_thresh, dirty_thresh; | |
739 | ||
740 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
741 | ||
742 | return (global_page_state(NR_FILE_DIRTY) + | |
743 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
744 | } | |
745 | ||
746 | /* | |
747 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 748 | * |
03ba3782 JA |
749 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
750 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
751 | * just walks the superblock inode list, writing back any inodes which are | |
752 | * older than a specific point in time. | |
66f3b8e2 | 753 | * |
03ba3782 JA |
754 | * Try to run once per dirty_writeback_interval. But if a writeback event |
755 | * takes longer than a dirty_writeback_interval interval, then leave a | |
756 | * one-second gap. | |
66f3b8e2 | 757 | * |
03ba3782 JA |
758 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
759 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 760 | */ |
c4a77a6c JA |
761 | static long wb_writeback(struct bdi_writeback *wb, |
762 | struct wb_writeback_args *args) | |
66f3b8e2 | 763 | { |
03ba3782 JA |
764 | struct writeback_control wbc = { |
765 | .bdi = wb->bdi, | |
c4a77a6c JA |
766 | .sb = args->sb, |
767 | .sync_mode = args->sync_mode, | |
03ba3782 | 768 | .older_than_this = NULL, |
c4a77a6c | 769 | .for_kupdate = args->for_kupdate, |
b17621fe | 770 | .for_background = args->for_background, |
c4a77a6c | 771 | .range_cyclic = args->range_cyclic, |
e913fc82 | 772 | .sb_pinned = args->sb_pinned, |
03ba3782 JA |
773 | }; |
774 | unsigned long oldest_jif; | |
775 | long wrote = 0; | |
a5989bdc | 776 | struct inode *inode; |
66f3b8e2 | 777 | |
03ba3782 JA |
778 | if (wbc.for_kupdate) { |
779 | wbc.older_than_this = &oldest_jif; | |
780 | oldest_jif = jiffies - | |
781 | msecs_to_jiffies(dirty_expire_interval * 10); | |
782 | } | |
c4a77a6c JA |
783 | if (!wbc.range_cyclic) { |
784 | wbc.range_start = 0; | |
785 | wbc.range_end = LLONG_MAX; | |
786 | } | |
38f21977 | 787 | |
03ba3782 JA |
788 | for (;;) { |
789 | /* | |
d3ddec76 | 790 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 791 | */ |
d3ddec76 | 792 | if (args->nr_pages <= 0) |
03ba3782 | 793 | break; |
66f3b8e2 | 794 | |
38f21977 | 795 | /* |
d3ddec76 WF |
796 | * For background writeout, stop when we are below the |
797 | * background dirty threshold | |
38f21977 | 798 | */ |
d3ddec76 | 799 | if (args->for_background && !over_bground_thresh()) |
03ba3782 | 800 | break; |
38f21977 | 801 | |
03ba3782 | 802 | wbc.more_io = 0; |
03ba3782 JA |
803 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
804 | wbc.pages_skipped = 0; | |
805 | writeback_inodes_wb(wb, &wbc); | |
c4a77a6c | 806 | args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
807 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
808 | ||
809 | /* | |
71fd05a8 | 810 | * If we consumed everything, see if we have more |
03ba3782 | 811 | */ |
71fd05a8 JA |
812 | if (wbc.nr_to_write <= 0) |
813 | continue; | |
814 | /* | |
815 | * Didn't write everything and we don't have more IO, bail | |
816 | */ | |
817 | if (!wbc.more_io) | |
03ba3782 | 818 | break; |
71fd05a8 JA |
819 | /* |
820 | * Did we write something? Try for more | |
821 | */ | |
822 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
823 | continue; | |
824 | /* | |
825 | * Nothing written. Wait for some inode to | |
826 | * become available for writeback. Otherwise | |
827 | * we'll just busyloop. | |
828 | */ | |
829 | spin_lock(&inode_lock); | |
830 | if (!list_empty(&wb->b_more_io)) { | |
831 | inode = list_entry(wb->b_more_io.prev, | |
832 | struct inode, i_list); | |
833 | inode_wait_for_writeback(inode); | |
03ba3782 | 834 | } |
71fd05a8 | 835 | spin_unlock(&inode_lock); |
03ba3782 JA |
836 | } |
837 | ||
838 | return wrote; | |
839 | } | |
840 | ||
841 | /* | |
842 | * Return the next bdi_work struct that hasn't been processed by this | |
8010c3b6 JA |
843 | * wb thread yet. ->seen is initially set for each thread that exists |
844 | * for this device, when a thread first notices a piece of work it | |
845 | * clears its bit. Depending on writeback type, the thread will notify | |
846 | * completion on either receiving the work (WB_SYNC_NONE) or after | |
847 | * it is done (WB_SYNC_ALL). | |
03ba3782 JA |
848 | */ |
849 | static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, | |
850 | struct bdi_writeback *wb) | |
851 | { | |
852 | struct bdi_work *work, *ret = NULL; | |
853 | ||
854 | rcu_read_lock(); | |
855 | ||
856 | list_for_each_entry_rcu(work, &bdi->work_list, list) { | |
77fad5e6 | 857 | if (!test_bit(wb->nr, &work->seen)) |
03ba3782 | 858 | continue; |
77fad5e6 | 859 | clear_bit(wb->nr, &work->seen); |
03ba3782 JA |
860 | |
861 | ret = work; | |
862 | break; | |
863 | } | |
864 | ||
865 | rcu_read_unlock(); | |
866 | return ret; | |
867 | } | |
868 | ||
869 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
870 | { | |
871 | unsigned long expired; | |
872 | long nr_pages; | |
873 | ||
69b62d01 JA |
874 | /* |
875 | * When set to zero, disable periodic writeback | |
876 | */ | |
877 | if (!dirty_writeback_interval) | |
878 | return 0; | |
879 | ||
03ba3782 JA |
880 | expired = wb->last_old_flush + |
881 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
882 | if (time_before(jiffies, expired)) | |
883 | return 0; | |
884 | ||
885 | wb->last_old_flush = jiffies; | |
886 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
887 | global_page_state(NR_UNSTABLE_NFS) + | |
888 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
889 | ||
c4a77a6c JA |
890 | if (nr_pages) { |
891 | struct wb_writeback_args args = { | |
892 | .nr_pages = nr_pages, | |
893 | .sync_mode = WB_SYNC_NONE, | |
894 | .for_kupdate = 1, | |
895 | .range_cyclic = 1, | |
896 | }; | |
897 | ||
898 | return wb_writeback(wb, &args); | |
899 | } | |
03ba3782 JA |
900 | |
901 | return 0; | |
902 | } | |
903 | ||
904 | /* | |
905 | * Retrieve work items and do the writeback they describe | |
906 | */ | |
907 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
908 | { | |
909 | struct backing_dev_info *bdi = wb->bdi; | |
910 | struct bdi_work *work; | |
c4a77a6c | 911 | long wrote = 0; |
03ba3782 JA |
912 | |
913 | while ((work = get_next_work_item(bdi, wb)) != NULL) { | |
c4a77a6c | 914 | struct wb_writeback_args args = work->args; |
7c8a3554 | 915 | int post_clear; |
03ba3782 JA |
916 | |
917 | /* | |
918 | * Override sync mode, in case we must wait for completion | |
919 | */ | |
920 | if (force_wait) | |
c4a77a6c | 921 | work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; |
03ba3782 | 922 | |
7c8a3554 JA |
923 | post_clear = WB_SYNC_ALL || args.sb_pinned; |
924 | ||
03ba3782 JA |
925 | /* |
926 | * If this isn't a data integrity operation, just notify | |
927 | * that we have seen this work and we are now starting it. | |
928 | */ | |
7c8a3554 | 929 | if (!post_clear) |
03ba3782 JA |
930 | wb_clear_pending(wb, work); |
931 | ||
c4a77a6c | 932 | wrote += wb_writeback(wb, &args); |
03ba3782 JA |
933 | |
934 | /* | |
935 | * This is a data integrity writeback, so only do the | |
936 | * notification when we have completed the work. | |
937 | */ | |
7c8a3554 | 938 | if (post_clear) |
03ba3782 JA |
939 | wb_clear_pending(wb, work); |
940 | } | |
941 | ||
942 | /* | |
943 | * Check for periodic writeback, kupdated() style | |
944 | */ | |
945 | wrote += wb_check_old_data_flush(wb); | |
946 | ||
947 | return wrote; | |
948 | } | |
949 | ||
950 | /* | |
951 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
952 | * wakes up periodically and does kupdated style flushing. | |
953 | */ | |
954 | int bdi_writeback_task(struct bdi_writeback *wb) | |
955 | { | |
956 | unsigned long last_active = jiffies; | |
957 | unsigned long wait_jiffies = -1UL; | |
958 | long pages_written; | |
959 | ||
960 | while (!kthread_should_stop()) { | |
961 | pages_written = wb_do_writeback(wb, 0); | |
962 | ||
963 | if (pages_written) | |
964 | last_active = jiffies; | |
965 | else if (wait_jiffies != -1UL) { | |
966 | unsigned long max_idle; | |
967 | ||
38f21977 | 968 | /* |
03ba3782 JA |
969 | * Longest period of inactivity that we tolerate. If we |
970 | * see dirty data again later, the task will get | |
971 | * recreated automatically. | |
38f21977 | 972 | */ |
03ba3782 JA |
973 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
974 | if (time_after(jiffies, max_idle + last_active)) | |
975 | break; | |
976 | } | |
977 | ||
69b62d01 JA |
978 | if (dirty_writeback_interval) { |
979 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
980 | schedule_timeout_interruptible(wait_jiffies); | |
f9eadbbd JA |
981 | } else { |
982 | set_current_state(TASK_INTERRUPTIBLE); | |
983 | if (list_empty_careful(&wb->bdi->work_list) && | |
984 | !kthread_should_stop()) | |
985 | schedule(); | |
986 | __set_current_state(TASK_RUNNING); | |
987 | } | |
69b62d01 | 988 | |
03ba3782 JA |
989 | try_to_freeze(); |
990 | } | |
991 | ||
992 | return 0; | |
993 | } | |
994 | ||
995 | /* | |
b6e51316 JA |
996 | * Schedule writeback for all backing devices. This does WB_SYNC_NONE |
997 | * writeback, for integrity writeback see bdi_sync_writeback(). | |
03ba3782 | 998 | */ |
b6e51316 | 999 | static void bdi_writeback_all(struct super_block *sb, long nr_pages) |
03ba3782 | 1000 | { |
b6e51316 JA |
1001 | struct wb_writeback_args args = { |
1002 | .sb = sb, | |
1003 | .nr_pages = nr_pages, | |
1004 | .sync_mode = WB_SYNC_NONE, | |
1005 | }; | |
03ba3782 | 1006 | struct backing_dev_info *bdi; |
03ba3782 | 1007 | |
cfc4ba53 | 1008 | rcu_read_lock(); |
03ba3782 | 1009 | |
cfc4ba53 | 1010 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1011 | if (!bdi_has_dirty_io(bdi)) |
1012 | continue; | |
38f21977 | 1013 | |
7c8a3554 | 1014 | bdi_alloc_queue_work(bdi, &args, 0); |
03ba3782 JA |
1015 | } |
1016 | ||
cfc4ba53 | 1017 | rcu_read_unlock(); |
1da177e4 LT |
1018 | } |
1019 | ||
1020 | /* | |
03ba3782 JA |
1021 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1022 | * the whole world. | |
1023 | */ | |
1024 | void wakeup_flusher_threads(long nr_pages) | |
1025 | { | |
03ba3782 JA |
1026 | if (nr_pages == 0) |
1027 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
1028 | global_page_state(NR_UNSTABLE_NFS); | |
b6e51316 | 1029 | bdi_writeback_all(NULL, nr_pages); |
03ba3782 JA |
1030 | } |
1031 | ||
1032 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
1033 | { | |
1034 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1035 | struct dentry *dentry; | |
1036 | const char *name = "?"; | |
1037 | ||
1038 | dentry = d_find_alias(inode); | |
1039 | if (dentry) { | |
1040 | spin_lock(&dentry->d_lock); | |
1041 | name = (const char *) dentry->d_name.name; | |
1042 | } | |
1043 | printk(KERN_DEBUG | |
1044 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1045 | current->comm, task_pid_nr(current), inode->i_ino, | |
1046 | name, inode->i_sb->s_id); | |
1047 | if (dentry) { | |
1048 | spin_unlock(&dentry->d_lock); | |
1049 | dput(dentry); | |
1050 | } | |
1051 | } | |
1052 | } | |
1053 | ||
1054 | /** | |
1055 | * __mark_inode_dirty - internal function | |
1056 | * @inode: inode to mark | |
1057 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1058 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1059 | * mark_inode_dirty_sync. | |
1da177e4 | 1060 | * |
03ba3782 JA |
1061 | * Put the inode on the super block's dirty list. |
1062 | * | |
1063 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1064 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1065 | * If it was not hashed, it will never be added to the dirty list | |
1066 | * even if it is later hashed, as it will have been marked dirty already. | |
1067 | * | |
1068 | * In short, make sure you hash any inodes _before_ you start marking | |
1069 | * them dirty. | |
1da177e4 | 1070 | * |
03ba3782 JA |
1071 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
1072 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 1073 | * |
03ba3782 JA |
1074 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1075 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1076 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1077 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1078 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1079 | * blockdev inode. | |
1da177e4 | 1080 | */ |
03ba3782 | 1081 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1082 | { |
03ba3782 | 1083 | struct super_block *sb = inode->i_sb; |
1da177e4 | 1084 | |
03ba3782 JA |
1085 | /* |
1086 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1087 | * dirty the inode itself | |
1088 | */ | |
1089 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1090 | if (sb->s_op->dirty_inode) | |
1091 | sb->s_op->dirty_inode(inode); | |
1092 | } | |
1093 | ||
1094 | /* | |
1095 | * make sure that changes are seen by all cpus before we test i_state | |
1096 | * -- mikulas | |
1097 | */ | |
1098 | smp_mb(); | |
1099 | ||
1100 | /* avoid the locking if we can */ | |
1101 | if ((inode->i_state & flags) == flags) | |
1102 | return; | |
1103 | ||
1104 | if (unlikely(block_dump)) | |
1105 | block_dump___mark_inode_dirty(inode); | |
1106 | ||
1107 | spin_lock(&inode_lock); | |
1108 | if ((inode->i_state & flags) != flags) { | |
1109 | const int was_dirty = inode->i_state & I_DIRTY; | |
1110 | ||
1111 | inode->i_state |= flags; | |
1112 | ||
1113 | /* | |
1114 | * If the inode is being synced, just update its dirty state. | |
1115 | * The unlocker will place the inode on the appropriate | |
1116 | * superblock list, based upon its state. | |
1117 | */ | |
1118 | if (inode->i_state & I_SYNC) | |
1119 | goto out; | |
1120 | ||
1121 | /* | |
1122 | * Only add valid (hashed) inodes to the superblock's | |
1123 | * dirty list. Add blockdev inodes as well. | |
1124 | */ | |
1125 | if (!S_ISBLK(inode->i_mode)) { | |
1126 | if (hlist_unhashed(&inode->i_hash)) | |
1127 | goto out; | |
1128 | } | |
1129 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
1130 | goto out; | |
1131 | ||
1132 | /* | |
1133 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1134 | * reposition it (that would break b_dirty time-ordering). | |
1135 | */ | |
1136 | if (!was_dirty) { | |
1137 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
1138 | struct backing_dev_info *bdi = wb->bdi; |
1139 | ||
1140 | if (bdi_cap_writeback_dirty(bdi) && | |
1141 | !test_bit(BDI_registered, &bdi->state)) { | |
1142 | WARN_ON(1); | |
1143 | printk(KERN_ERR "bdi-%s not registered\n", | |
1144 | bdi->name); | |
1145 | } | |
03ba3782 JA |
1146 | |
1147 | inode->dirtied_when = jiffies; | |
1148 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 1149 | } |
1da177e4 | 1150 | } |
03ba3782 JA |
1151 | out: |
1152 | spin_unlock(&inode_lock); | |
1153 | } | |
1154 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1155 | ||
1156 | /* | |
1157 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1158 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1159 | * | |
1160 | * If older_than_this is non-NULL, then only write out inodes which | |
1161 | * had their first dirtying at a time earlier than *older_than_this. | |
1162 | * | |
03ba3782 JA |
1163 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1164 | * This function assumes that the blockdev superblock's inodes are backed by | |
1165 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1166 | * assume that all inodes are backed by the same queue. | |
1167 | * | |
1168 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1169 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1170 | * on the writer throttling path, and we get decent balancing between many | |
1171 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1172 | */ | |
b6e51316 | 1173 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1174 | { |
1175 | struct inode *inode, *old_inode = NULL; | |
1176 | ||
1177 | /* | |
1178 | * We need to be protected against the filesystem going from | |
1179 | * r/o to r/w or vice versa. | |
1180 | */ | |
b6e51316 | 1181 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
1182 | |
1183 | spin_lock(&inode_lock); | |
1184 | ||
1185 | /* | |
1186 | * Data integrity sync. Must wait for all pages under writeback, | |
1187 | * because there may have been pages dirtied before our sync | |
1188 | * call, but which had writeout started before we write it out. | |
1189 | * In which case, the inode may not be on the dirty list, but | |
1190 | * we still have to wait for that writeout. | |
1191 | */ | |
b6e51316 | 1192 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1193 | struct address_space *mapping; |
1194 | ||
1195 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1196 | continue; | |
1197 | mapping = inode->i_mapping; | |
1198 | if (mapping->nrpages == 0) | |
1199 | continue; | |
1200 | __iget(inode); | |
1201 | spin_unlock(&inode_lock); | |
1202 | /* | |
1203 | * We hold a reference to 'inode' so it couldn't have | |
1204 | * been removed from s_inodes list while we dropped the | |
1205 | * inode_lock. We cannot iput the inode now as we can | |
1206 | * be holding the last reference and we cannot iput it | |
1207 | * under inode_lock. So we keep the reference and iput | |
1208 | * it later. | |
1209 | */ | |
1210 | iput(old_inode); | |
1211 | old_inode = inode; | |
1212 | ||
1213 | filemap_fdatawait(mapping); | |
1214 | ||
1215 | cond_resched(); | |
1216 | ||
1217 | spin_lock(&inode_lock); | |
1218 | } | |
1219 | spin_unlock(&inode_lock); | |
1220 | iput(old_inode); | |
1da177e4 LT |
1221 | } |
1222 | ||
e913fc82 JA |
1223 | static void __writeback_inodes_sb(struct super_block *sb, int sb_locked) |
1224 | { | |
1225 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); | |
1226 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
1227 | long nr_to_write; | |
1228 | ||
1229 | nr_to_write = nr_dirty + nr_unstable + | |
1230 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
1231 | ||
1232 | bdi_start_writeback(sb->s_bdi, sb, nr_to_write, sb_locked); | |
1233 | } | |
1234 | ||
d8a8559c JA |
1235 | /** |
1236 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1237 | * @sb: the superblock | |
1da177e4 | 1238 | * |
d8a8559c JA |
1239 | * Start writeback on some inodes on this super_block. No guarantees are made |
1240 | * on how many (if any) will be written, and this function does not wait | |
1241 | * for IO completion of submitted IO. The number of pages submitted is | |
1242 | * returned. | |
1da177e4 | 1243 | */ |
b6e51316 | 1244 | void writeback_inodes_sb(struct super_block *sb) |
1da177e4 | 1245 | { |
e913fc82 | 1246 | __writeback_inodes_sb(sb, 0); |
d8a8559c JA |
1247 | } |
1248 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1249 | ||
e913fc82 JA |
1250 | /** |
1251 | * writeback_inodes_sb_locked - writeback dirty inodes from given super_block | |
1252 | * @sb: the superblock | |
1253 | * | |
1254 | * Like writeback_inodes_sb(), except the caller already holds the | |
1255 | * sb umount sem. | |
1256 | */ | |
1257 | void writeback_inodes_sb_locked(struct super_block *sb) | |
1258 | { | |
1259 | __writeback_inodes_sb(sb, 1); | |
1260 | } | |
1261 | ||
17bd55d0 ES |
1262 | /** |
1263 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1264 | * @sb: the superblock | |
1265 | * | |
1266 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1267 | * Returns 1 if writeback was started, 0 if not. | |
1268 | */ | |
1269 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1270 | { | |
1271 | if (!writeback_in_progress(sb->s_bdi)) { | |
1272 | writeback_inodes_sb(sb); | |
1273 | return 1; | |
1274 | } else | |
1275 | return 0; | |
1276 | } | |
1277 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1278 | ||
d8a8559c JA |
1279 | /** |
1280 | * sync_inodes_sb - sync sb inode pages | |
1281 | * @sb: the superblock | |
1282 | * | |
1283 | * This function writes and waits on any dirty inode belonging to this | |
1284 | * super_block. The number of pages synced is returned. | |
1285 | */ | |
b6e51316 | 1286 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1287 | { |
b6e51316 JA |
1288 | bdi_sync_writeback(sb->s_bdi, sb); |
1289 | wait_sb_inodes(sb); | |
1da177e4 | 1290 | } |
d8a8559c | 1291 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1292 | |
1da177e4 | 1293 | /** |
7f04c26d AA |
1294 | * write_inode_now - write an inode to disk |
1295 | * @inode: inode to write to disk | |
1296 | * @sync: whether the write should be synchronous or not | |
1297 | * | |
1298 | * This function commits an inode to disk immediately if it is dirty. This is | |
1299 | * primarily needed by knfsd. | |
1da177e4 | 1300 | * |
7f04c26d | 1301 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1302 | */ |
1da177e4 LT |
1303 | int write_inode_now(struct inode *inode, int sync) |
1304 | { | |
1305 | int ret; | |
1306 | struct writeback_control wbc = { | |
1307 | .nr_to_write = LONG_MAX, | |
18914b18 | 1308 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1309 | .range_start = 0, |
1310 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1311 | }; |
1312 | ||
1313 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1314 | wbc.nr_to_write = 0; |
1da177e4 LT |
1315 | |
1316 | might_sleep(); | |
1317 | spin_lock(&inode_lock); | |
01c03194 | 1318 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1319 | spin_unlock(&inode_lock); |
1320 | if (sync) | |
1c0eeaf5 | 1321 | inode_sync_wait(inode); |
1da177e4 LT |
1322 | return ret; |
1323 | } | |
1324 | EXPORT_SYMBOL(write_inode_now); | |
1325 | ||
1326 | /** | |
1327 | * sync_inode - write an inode and its pages to disk. | |
1328 | * @inode: the inode to sync | |
1329 | * @wbc: controls the writeback mode | |
1330 | * | |
1331 | * sync_inode() will write an inode and its pages to disk. It will also | |
1332 | * correctly update the inode on its superblock's dirty inode lists and will | |
1333 | * update inode->i_state. | |
1334 | * | |
1335 | * The caller must have a ref on the inode. | |
1336 | */ | |
1337 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1338 | { | |
1339 | int ret; | |
1340 | ||
1341 | spin_lock(&inode_lock); | |
01c03194 | 1342 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1343 | spin_unlock(&inode_lock); |
1344 | return ret; | |
1345 | } | |
1346 | EXPORT_SYMBOL(sync_inode); |