]>
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> | |
455b2864 | 29 | #include <linux/tracepoint.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 | 31 | |
d46db3d5 WF |
32 | /* |
33 | * The maximum number of pages to writeout in a single bdi flush/kupdate | |
34 | * operation. We do this so we don't hold I_SYNC against an inode for | |
35 | * enormous amounts of time, which would block a userspace task which has | |
36 | * been forced to throttle against that inode. Also, the code reevaluates | |
37 | * the dirty each time it has written this many pages. | |
38 | */ | |
39 | #define MAX_WRITEBACK_PAGES 1024L | |
40 | ||
c4a77a6c JA |
41 | /* |
42 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
43 | */ | |
83ba7b07 | 44 | struct wb_writeback_work { |
c4a77a6c JA |
45 | long nr_pages; |
46 | struct super_block *sb; | |
d46db3d5 | 47 | unsigned long *older_than_this; |
c4a77a6c | 48 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 49 | unsigned int tagged_writepages:1; |
52957fe1 HS |
50 | unsigned int for_kupdate:1; |
51 | unsigned int range_cyclic:1; | |
52 | unsigned int for_background:1; | |
c4a77a6c | 53 | |
8010c3b6 | 54 | struct list_head list; /* pending work list */ |
83ba7b07 | 55 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
56 | }; |
57 | ||
455b2864 DC |
58 | /* |
59 | * Include the creation of the trace points after defining the | |
60 | * wb_writeback_work structure so that the definition remains local to this | |
61 | * file. | |
62 | */ | |
63 | #define CREATE_TRACE_POINTS | |
64 | #include <trace/events/writeback.h> | |
65 | ||
455b2864 DC |
66 | /* |
67 | * We don't actually have pdflush, but this one is exported though /proc... | |
68 | */ | |
69 | int nr_pdflush_threads; | |
70 | ||
f11b00f3 AB |
71 | /** |
72 | * writeback_in_progress - determine whether there is writeback in progress | |
73 | * @bdi: the device's backing_dev_info structure. | |
74 | * | |
03ba3782 JA |
75 | * Determine whether there is writeback waiting to be handled against a |
76 | * backing device. | |
f11b00f3 AB |
77 | */ |
78 | int writeback_in_progress(struct backing_dev_info *bdi) | |
79 | { | |
81d73a32 | 80 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 AB |
81 | } |
82 | ||
692ebd17 JK |
83 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
84 | { | |
85 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 86 | |
aaead25b CH |
87 | if (strcmp(sb->s_type->name, "bdev") == 0) |
88 | return inode->i_mapping->backing_dev_info; | |
89 | ||
90 | return sb->s_bdi; | |
692ebd17 JK |
91 | } |
92 | ||
7ccf19a8 NP |
93 | static inline struct inode *wb_inode(struct list_head *head) |
94 | { | |
95 | return list_entry(head, struct inode, i_wb_list); | |
96 | } | |
97 | ||
6585027a JK |
98 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
99 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) | |
03ba3782 | 100 | { |
fff5b85a AB |
101 | if (bdi->wb.task) { |
102 | wake_up_process(bdi->wb.task); | |
103 | } else { | |
104 | /* | |
105 | * The bdi thread isn't there, wake up the forker thread which | |
106 | * will create and run it. | |
107 | */ | |
03ba3782 | 108 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 109 | } |
6585027a JK |
110 | } |
111 | ||
112 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
113 | struct wb_writeback_work *work) | |
114 | { | |
115 | trace_writeback_queue(bdi, work); | |
116 | ||
117 | spin_lock_bh(&bdi->wb_lock); | |
118 | list_add_tail(&work->list, &bdi->work_list); | |
119 | if (!bdi->wb.task) | |
120 | trace_writeback_nothread(bdi, work); | |
121 | bdi_wakeup_flusher(bdi); | |
6467716a | 122 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
123 | } |
124 | ||
83ba7b07 CH |
125 | static void |
126 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
6585027a | 127 | bool range_cyclic) |
1da177e4 | 128 | { |
83ba7b07 | 129 | struct wb_writeback_work *work; |
03ba3782 | 130 | |
bcddc3f0 JA |
131 | /* |
132 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
133 | * wakeup the thread for old dirty data writeback | |
134 | */ | |
83ba7b07 CH |
135 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
136 | if (!work) { | |
455b2864 DC |
137 | if (bdi->wb.task) { |
138 | trace_writeback_nowork(bdi); | |
83ba7b07 | 139 | wake_up_process(bdi->wb.task); |
455b2864 | 140 | } |
83ba7b07 | 141 | return; |
bcddc3f0 | 142 | } |
03ba3782 | 143 | |
83ba7b07 CH |
144 | work->sync_mode = WB_SYNC_NONE; |
145 | work->nr_pages = nr_pages; | |
146 | work->range_cyclic = range_cyclic; | |
03ba3782 | 147 | |
83ba7b07 | 148 | bdi_queue_work(bdi, work); |
b6e51316 JA |
149 | } |
150 | ||
151 | /** | |
152 | * bdi_start_writeback - start writeback | |
153 | * @bdi: the backing device to write from | |
154 | * @nr_pages: the number of pages to write | |
155 | * | |
156 | * Description: | |
157 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 158 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 159 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
160 | * |
161 | */ | |
c5444198 | 162 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 163 | { |
6585027a | 164 | __bdi_start_writeback(bdi, nr_pages, true); |
c5444198 | 165 | } |
d3ddec76 | 166 | |
c5444198 CH |
167 | /** |
168 | * bdi_start_background_writeback - start background writeback | |
169 | * @bdi: the backing device to write from | |
170 | * | |
171 | * Description: | |
6585027a JK |
172 | * This makes sure WB_SYNC_NONE background writeback happens. When |
173 | * this function returns, it is only guaranteed that for given BDI | |
174 | * some IO is happening if we are over background dirty threshold. | |
175 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
176 | */ |
177 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
178 | { | |
6585027a JK |
179 | /* |
180 | * We just wake up the flusher thread. It will perform background | |
181 | * writeback as soon as there is no other work to do. | |
182 | */ | |
71927e84 | 183 | trace_writeback_wake_background(bdi); |
6585027a JK |
184 | spin_lock_bh(&bdi->wb_lock); |
185 | bdi_wakeup_flusher(bdi); | |
186 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
187 | } |
188 | ||
a66979ab DC |
189 | /* |
190 | * Remove the inode from the writeback list it is on. | |
191 | */ | |
192 | void inode_wb_list_del(struct inode *inode) | |
193 | { | |
f758eeab CH |
194 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
195 | ||
196 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 197 | list_del_init(&inode->i_wb_list); |
f758eeab | 198 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
199 | } |
200 | ||
6610a0bc AM |
201 | /* |
202 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
203 | * furthest end of its superblock's dirty-inode list. | |
204 | * | |
205 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 206 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
207 | * the case then the inode must have been redirtied while it was being written |
208 | * out and we don't reset its dirtied_when. | |
209 | */ | |
f758eeab | 210 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 211 | { |
f758eeab | 212 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 213 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 214 | struct inode *tail; |
6610a0bc | 215 | |
7ccf19a8 | 216 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 217 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
218 | inode->dirtied_when = jiffies; |
219 | } | |
7ccf19a8 | 220 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
221 | } |
222 | ||
c986d1e2 | 223 | /* |
66f3b8e2 | 224 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 225 | */ |
f758eeab | 226 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 227 | { |
f758eeab | 228 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 229 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
230 | } |
231 | ||
1c0eeaf5 JE |
232 | static void inode_sync_complete(struct inode *inode) |
233 | { | |
234 | /* | |
a66979ab | 235 | * Prevent speculative execution through |
f758eeab | 236 | * spin_unlock(&wb->list_lock); |
1c0eeaf5 | 237 | */ |
a66979ab | 238 | |
1c0eeaf5 JE |
239 | smp_mb(); |
240 | wake_up_bit(&inode->i_state, __I_SYNC); | |
241 | } | |
242 | ||
d2caa3c5 JL |
243 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
244 | { | |
245 | bool ret = time_after(inode->dirtied_when, t); | |
246 | #ifndef CONFIG_64BIT | |
247 | /* | |
248 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
249 | * It _appears_ to be in the future, but is actually in distant past. | |
250 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 251 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
252 | */ |
253 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
254 | #endif | |
255 | return ret; | |
256 | } | |
257 | ||
2c136579 FW |
258 | /* |
259 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
260 | */ | |
e84d0a4f | 261 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 262 | struct list_head *dispatch_queue, |
e84d0a4f | 263 | unsigned long *older_than_this) |
2c136579 | 264 | { |
5c03449d SL |
265 | LIST_HEAD(tmp); |
266 | struct list_head *pos, *node; | |
cf137307 | 267 | struct super_block *sb = NULL; |
5c03449d | 268 | struct inode *inode; |
cf137307 | 269 | int do_sb_sort = 0; |
e84d0a4f | 270 | int moved = 0; |
5c03449d | 271 | |
2c136579 | 272 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 273 | inode = wb_inode(delaying_queue->prev); |
2c136579 | 274 | if (older_than_this && |
d2caa3c5 | 275 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 276 | break; |
cf137307 JA |
277 | if (sb && sb != inode->i_sb) |
278 | do_sb_sort = 1; | |
279 | sb = inode->i_sb; | |
7ccf19a8 | 280 | list_move(&inode->i_wb_list, &tmp); |
e84d0a4f | 281 | moved++; |
5c03449d SL |
282 | } |
283 | ||
cf137307 JA |
284 | /* just one sb in list, splice to dispatch_queue and we're done */ |
285 | if (!do_sb_sort) { | |
286 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 287 | goto out; |
cf137307 JA |
288 | } |
289 | ||
5c03449d SL |
290 | /* Move inodes from one superblock together */ |
291 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 292 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 293 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 294 | inode = wb_inode(pos); |
5c03449d | 295 | if (inode->i_sb == sb) |
7ccf19a8 | 296 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 297 | } |
2c136579 | 298 | } |
e84d0a4f WF |
299 | out: |
300 | return moved; | |
2c136579 FW |
301 | } |
302 | ||
303 | /* | |
304 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
305 | * Before |
306 | * newly dirtied b_dirty b_io b_more_io | |
307 | * =============> gf edc BA | |
308 | * After | |
309 | * newly dirtied b_dirty b_io b_more_io | |
310 | * =============> g fBAedc | |
311 | * | | |
312 | * +--> dequeue for IO | |
2c136579 | 313 | */ |
03ba3782 | 314 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 315 | { |
e84d0a4f | 316 | int moved; |
f758eeab | 317 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 318 | list_splice_init(&wb->b_more_io, &wb->b_io); |
e84d0a4f WF |
319 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
320 | trace_writeback_queue_io(wb, older_than_this, moved); | |
66f3b8e2 JA |
321 | } |
322 | ||
a9185b41 | 323 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 324 | { |
03ba3782 | 325 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 326 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 327 | return 0; |
08d8e974 | 328 | } |
08d8e974 | 329 | |
1da177e4 | 330 | /* |
01c03194 CH |
331 | * Wait for writeback on an inode to complete. |
332 | */ | |
f758eeab CH |
333 | static void inode_wait_for_writeback(struct inode *inode, |
334 | struct bdi_writeback *wb) | |
01c03194 CH |
335 | { |
336 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
337 | wait_queue_head_t *wqh; | |
338 | ||
339 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
340 | while (inode->i_state & I_SYNC) { |
341 | spin_unlock(&inode->i_lock); | |
f758eeab | 342 | spin_unlock(&wb->list_lock); |
01c03194 | 343 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
f758eeab | 344 | spin_lock(&wb->list_lock); |
250df6ed | 345 | spin_lock(&inode->i_lock); |
58a9d3d8 | 346 | } |
01c03194 CH |
347 | } |
348 | ||
349 | /* | |
f758eeab | 350 | * Write out an inode's dirty pages. Called under wb->list_lock and |
0f1b1fd8 DC |
351 | * inode->i_lock. Either the caller has an active reference on the inode or |
352 | * the inode has I_WILL_FREE set. | |
01c03194 | 353 | * |
1da177e4 LT |
354 | * If `wait' is set, wait on the writeout. |
355 | * | |
356 | * The whole writeout design is quite complex and fragile. We want to avoid | |
357 | * starvation of particular inodes when others are being redirtied, prevent | |
358 | * livelocks, etc. | |
1da177e4 LT |
359 | */ |
360 | static int | |
f758eeab CH |
361 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
362 | struct writeback_control *wbc) | |
1da177e4 | 363 | { |
1da177e4 | 364 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 365 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 366 | unsigned dirty; |
1da177e4 LT |
367 | int ret; |
368 | ||
f758eeab | 369 | assert_spin_locked(&wb->list_lock); |
0f1b1fd8 DC |
370 | assert_spin_locked(&inode->i_lock); |
371 | ||
01c03194 CH |
372 | if (!atomic_read(&inode->i_count)) |
373 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
374 | else | |
375 | WARN_ON(inode->i_state & I_WILL_FREE); | |
376 | ||
377 | if (inode->i_state & I_SYNC) { | |
378 | /* | |
379 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 380 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
381 | * writeback can proceed with the other inodes on s_io. |
382 | * | |
383 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 384 | * completed a full scan of b_io. |
01c03194 | 385 | */ |
a9185b41 | 386 | if (wbc->sync_mode != WB_SYNC_ALL) { |
f758eeab | 387 | requeue_io(inode, wb); |
251d6a47 WF |
388 | trace_writeback_single_inode_requeue(inode, wbc, |
389 | nr_to_write); | |
01c03194 CH |
390 | return 0; |
391 | } | |
392 | ||
393 | /* | |
394 | * It's a data-integrity sync. We must wait. | |
395 | */ | |
f758eeab | 396 | inode_wait_for_writeback(inode, wb); |
01c03194 CH |
397 | } |
398 | ||
1c0eeaf5 | 399 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 400 | |
5547e8aa | 401 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 402 | inode->i_state |= I_SYNC; |
5547e8aa | 403 | inode->i_state &= ~I_DIRTY_PAGES; |
250df6ed | 404 | spin_unlock(&inode->i_lock); |
f758eeab | 405 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
406 | |
407 | ret = do_writepages(mapping, wbc); | |
408 | ||
26821ed4 CH |
409 | /* |
410 | * Make sure to wait on the data before writing out the metadata. | |
411 | * This is important for filesystems that modify metadata on data | |
412 | * I/O completion. | |
413 | */ | |
a9185b41 | 414 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 415 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
416 | if (ret == 0) |
417 | ret = err; | |
418 | } | |
419 | ||
5547e8aa DM |
420 | /* |
421 | * Some filesystems may redirty the inode during the writeback | |
422 | * due to delalloc, clear dirty metadata flags right before | |
423 | * write_inode() | |
424 | */ | |
250df6ed | 425 | spin_lock(&inode->i_lock); |
5547e8aa DM |
426 | dirty = inode->i_state & I_DIRTY; |
427 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 428 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
429 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
430 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 431 | int err = write_inode(inode, wbc); |
1da177e4 LT |
432 | if (ret == 0) |
433 | ret = err; | |
434 | } | |
435 | ||
f758eeab | 436 | spin_lock(&wb->list_lock); |
250df6ed | 437 | spin_lock(&inode->i_lock); |
1c0eeaf5 | 438 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 439 | if (!(inode->i_state & I_FREEING)) { |
94c3dcbb WF |
440 | /* |
441 | * Sync livelock prevention. Each inode is tagged and synced in | |
442 | * one shot. If still dirty, it will be redirty_tail()'ed below. | |
443 | * Update the dirty time to prevent enqueue and sync it again. | |
444 | */ | |
445 | if ((inode->i_state & I_DIRTY) && | |
446 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
447 | inode->dirtied_when = jiffies; | |
448 | ||
23539afc | 449 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
450 | /* |
451 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 452 | * sometimes bales out without doing anything. |
1b43ef91 | 453 | */ |
a50aeb40 WF |
454 | inode->i_state |= I_DIRTY_PAGES; |
455 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 456 | /* |
a50aeb40 | 457 | * slice used up: queue for next turn |
1da177e4 | 458 | */ |
f758eeab | 459 | requeue_io(inode, wb); |
1da177e4 LT |
460 | } else { |
461 | /* | |
a50aeb40 WF |
462 | * Writeback blocked by something other than |
463 | * congestion. Delay the inode for some time to | |
464 | * avoid spinning on the CPU (100% iowait) | |
465 | * retrying writeback of the dirty page/inode | |
466 | * that cannot be performed immediately. | |
1da177e4 | 467 | */ |
f758eeab | 468 | redirty_tail(inode, wb); |
1da177e4 | 469 | } |
23539afc WF |
470 | } else if (inode->i_state & I_DIRTY) { |
471 | /* | |
472 | * Filesystems can dirty the inode during writeback | |
473 | * operations, such as delayed allocation during | |
474 | * submission or metadata updates after data IO | |
475 | * completion. | |
476 | */ | |
f758eeab | 477 | redirty_tail(inode, wb); |
1da177e4 LT |
478 | } else { |
479 | /* | |
9e38d86f NP |
480 | * The inode is clean. At this point we either have |
481 | * a reference to the inode or it's on it's way out. | |
482 | * No need to add it back to the LRU. | |
1da177e4 | 483 | */ |
7ccf19a8 | 484 | list_del_init(&inode->i_wb_list); |
1da177e4 LT |
485 | } |
486 | } | |
1c0eeaf5 | 487 | inode_sync_complete(inode); |
251d6a47 | 488 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
1da177e4 LT |
489 | return ret; |
490 | } | |
491 | ||
03ba3782 | 492 | /* |
d19de7ed | 493 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
494 | * before calling writeback. So make sure that we do pin it, so it doesn't |
495 | * go away while we are writing inodes from it. | |
03ba3782 | 496 | */ |
d19de7ed | 497 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 498 | { |
03ba3782 | 499 | spin_lock(&sb_lock); |
29cb4859 CH |
500 | if (list_empty(&sb->s_instances)) { |
501 | spin_unlock(&sb_lock); | |
502 | return false; | |
503 | } | |
504 | ||
03ba3782 | 505 | sb->s_count++; |
29cb4859 CH |
506 | spin_unlock(&sb_lock); |
507 | ||
03ba3782 | 508 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 509 | if (sb->s_root) |
d19de7ed | 510 | return true; |
03ba3782 JA |
511 | up_read(&sb->s_umount); |
512 | } | |
29cb4859 CH |
513 | |
514 | put_super(sb); | |
d19de7ed | 515 | return false; |
03ba3782 JA |
516 | } |
517 | ||
d46db3d5 WF |
518 | static long writeback_chunk_size(struct wb_writeback_work *work) |
519 | { | |
520 | long pages; | |
521 | ||
522 | /* | |
523 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
524 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
525 | * here avoids calling into writeback_inodes_wb() more than once. | |
526 | * | |
527 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
528 | * | |
529 | * wb_writeback() | |
530 | * writeback_sb_inodes() <== called only once | |
531 | * write_cache_pages() <== called once for each inode | |
532 | * (quickly) tag currently dirty pages | |
533 | * (maybe slowly) sync all tagged pages | |
534 | */ | |
535 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
536 | pages = LONG_MAX; | |
537 | else | |
538 | pages = min(MAX_WRITEBACK_PAGES, work->nr_pages); | |
539 | ||
540 | return pages; | |
541 | } | |
542 | ||
f11c9c5c ES |
543 | /* |
544 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
545 | * |
546 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
547 | * inodes. Otherwise write only ones which go sequentially |
548 | * in reverse order. | |
edadfb10 | 549 | * |
d46db3d5 | 550 | * Return the number of pages and/or inodes written. |
f11c9c5c | 551 | */ |
d46db3d5 WF |
552 | static long writeback_sb_inodes(struct super_block *sb, |
553 | struct bdi_writeback *wb, | |
554 | struct wb_writeback_work *work) | |
1da177e4 | 555 | { |
d46db3d5 WF |
556 | struct writeback_control wbc = { |
557 | .sync_mode = work->sync_mode, | |
558 | .tagged_writepages = work->tagged_writepages, | |
559 | .for_kupdate = work->for_kupdate, | |
560 | .for_background = work->for_background, | |
561 | .range_cyclic = work->range_cyclic, | |
562 | .range_start = 0, | |
563 | .range_end = LLONG_MAX, | |
564 | }; | |
565 | unsigned long start_time = jiffies; | |
566 | long write_chunk; | |
567 | long wrote = 0; /* count both pages and inodes */ | |
568 | ||
03ba3782 | 569 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 570 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
571 | |
572 | if (inode->i_sb != sb) { | |
d46db3d5 | 573 | if (work->sb) { |
edadfb10 CH |
574 | /* |
575 | * We only want to write back data for this | |
576 | * superblock, move all inodes not belonging | |
577 | * to it back onto the dirty list. | |
578 | */ | |
f758eeab | 579 | redirty_tail(inode, wb); |
edadfb10 CH |
580 | continue; |
581 | } | |
582 | ||
583 | /* | |
584 | * The inode belongs to a different superblock. | |
585 | * Bounce back to the caller to unpin this and | |
586 | * pin the next superblock. | |
587 | */ | |
d46db3d5 | 588 | break; |
edadfb10 CH |
589 | } |
590 | ||
9843b76a CH |
591 | /* |
592 | * Don't bother with new inodes or inodes beeing freed, first | |
593 | * kind does not need peridic writeout yet, and for the latter | |
594 | * kind writeout is handled by the freer. | |
595 | */ | |
250df6ed | 596 | spin_lock(&inode->i_lock); |
9843b76a | 597 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 598 | spin_unlock(&inode->i_lock); |
f758eeab | 599 | requeue_io(inode, wb); |
7ef0d737 NP |
600 | continue; |
601 | } | |
1da177e4 | 602 | __iget(inode); |
d46db3d5 WF |
603 | write_chunk = writeback_chunk_size(work); |
604 | wbc.nr_to_write = write_chunk; | |
605 | wbc.pages_skipped = 0; | |
606 | ||
607 | writeback_single_inode(inode, wb, &wbc); | |
250df6ed | 608 | |
d46db3d5 WF |
609 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
610 | wrote += write_chunk - wbc.nr_to_write; | |
611 | if (!(inode->i_state & I_DIRTY)) | |
612 | wrote++; | |
613 | if (wbc.pages_skipped) { | |
1da177e4 LT |
614 | /* |
615 | * writeback is not making progress due to locked | |
616 | * buffers. Skip this inode for now. | |
617 | */ | |
f758eeab | 618 | redirty_tail(inode, wb); |
1da177e4 | 619 | } |
0f1b1fd8 | 620 | spin_unlock(&inode->i_lock); |
f758eeab | 621 | spin_unlock(&wb->list_lock); |
1da177e4 | 622 | iput(inode); |
4ffc8444 | 623 | cond_resched(); |
f758eeab | 624 | spin_lock(&wb->list_lock); |
d46db3d5 WF |
625 | /* |
626 | * bail out to wb_writeback() often enough to check | |
627 | * background threshold and other termination conditions. | |
628 | */ | |
629 | if (wrote) { | |
630 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
631 | break; | |
632 | if (work->nr_pages <= 0) | |
633 | break; | |
634 | } | |
1da177e4 | 635 | } |
d46db3d5 | 636 | return wrote; |
f11c9c5c ES |
637 | } |
638 | ||
d46db3d5 WF |
639 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
640 | struct wb_writeback_work *work) | |
f11c9c5c | 641 | { |
d46db3d5 WF |
642 | unsigned long start_time = jiffies; |
643 | long wrote = 0; | |
f11c9c5c | 644 | |
f11c9c5c | 645 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 646 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 647 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 648 | |
edadfb10 | 649 | if (!pin_sb_for_writeback(sb)) { |
f758eeab | 650 | requeue_io(inode, wb); |
edadfb10 | 651 | continue; |
f11c9c5c | 652 | } |
d46db3d5 | 653 | wrote += writeback_sb_inodes(sb, wb, work); |
edadfb10 | 654 | drop_super(sb); |
f11c9c5c | 655 | |
d46db3d5 WF |
656 | /* refer to the same tests at the end of writeback_sb_inodes */ |
657 | if (wrote) { | |
658 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
659 | break; | |
660 | if (work->nr_pages <= 0) | |
661 | break; | |
662 | } | |
f11c9c5c | 663 | } |
66f3b8e2 | 664 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 665 | return wrote; |
66f3b8e2 JA |
666 | } |
667 | ||
d46db3d5 | 668 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages) |
edadfb10 | 669 | { |
d46db3d5 WF |
670 | struct wb_writeback_work work = { |
671 | .nr_pages = nr_pages, | |
672 | .sync_mode = WB_SYNC_NONE, | |
673 | .range_cyclic = 1, | |
674 | }; | |
675 | ||
f758eeab | 676 | spin_lock(&wb->list_lock); |
424b351f | 677 | if (list_empty(&wb->b_io)) |
d46db3d5 WF |
678 | queue_io(wb, NULL); |
679 | __writeback_inodes_wb(wb, &work); | |
f758eeab | 680 | spin_unlock(&wb->list_lock); |
edadfb10 | 681 | |
d46db3d5 WF |
682 | return nr_pages - work.nr_pages; |
683 | } | |
03ba3782 JA |
684 | |
685 | static inline bool over_bground_thresh(void) | |
686 | { | |
687 | unsigned long background_thresh, dirty_thresh; | |
688 | ||
16c4042f | 689 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 JA |
690 | |
691 | return (global_page_state(NR_FILE_DIRTY) + | |
4cbec4c8 | 692 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
03ba3782 JA |
693 | } |
694 | ||
e98be2d5 WF |
695 | /* |
696 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
697 | * only the flusher working on the first wb should do it. | |
698 | */ | |
699 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
700 | unsigned long start_time) | |
701 | { | |
c42843f2 | 702 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
703 | } |
704 | ||
03ba3782 JA |
705 | /* |
706 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 707 | * |
03ba3782 JA |
708 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
709 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
710 | * just walks the superblock inode list, writing back any inodes which are | |
711 | * older than a specific point in time. | |
66f3b8e2 | 712 | * |
03ba3782 JA |
713 | * Try to run once per dirty_writeback_interval. But if a writeback event |
714 | * takes longer than a dirty_writeback_interval interval, then leave a | |
715 | * one-second gap. | |
66f3b8e2 | 716 | * |
03ba3782 JA |
717 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
718 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 719 | */ |
c4a77a6c | 720 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 721 | struct wb_writeback_work *work) |
66f3b8e2 | 722 | { |
e98be2d5 | 723 | unsigned long wb_start = jiffies; |
d46db3d5 | 724 | long nr_pages = work->nr_pages; |
03ba3782 | 725 | unsigned long oldest_jif; |
a5989bdc | 726 | struct inode *inode; |
d46db3d5 | 727 | long progress; |
b9543dac | 728 | |
e185dda8 | 729 | oldest_jif = jiffies; |
d46db3d5 | 730 | work->older_than_this = &oldest_jif; |
e185dda8 | 731 | |
e8dfc305 | 732 | spin_lock(&wb->list_lock); |
03ba3782 JA |
733 | for (;;) { |
734 | /* | |
d3ddec76 | 735 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 736 | */ |
83ba7b07 | 737 | if (work->nr_pages <= 0) |
03ba3782 | 738 | break; |
66f3b8e2 | 739 | |
aa373cf5 JK |
740 | /* |
741 | * Background writeout and kupdate-style writeback may | |
742 | * run forever. Stop them if there is other work to do | |
743 | * so that e.g. sync can proceed. They'll be restarted | |
744 | * after the other works are all done. | |
745 | */ | |
746 | if ((work->for_background || work->for_kupdate) && | |
747 | !list_empty(&wb->bdi->work_list)) | |
748 | break; | |
749 | ||
38f21977 | 750 | /* |
d3ddec76 WF |
751 | * For background writeout, stop when we are below the |
752 | * background dirty threshold | |
38f21977 | 753 | */ |
83ba7b07 | 754 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 755 | break; |
38f21977 | 756 | |
ba9aa839 WF |
757 | if (work->for_kupdate) { |
758 | oldest_jif = jiffies - | |
759 | msecs_to_jiffies(dirty_expire_interval * 10); | |
d46db3d5 | 760 | work->older_than_this = &oldest_jif; |
ba9aa839 WF |
761 | } |
762 | ||
d46db3d5 | 763 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 764 | if (list_empty(&wb->b_io)) |
d46db3d5 | 765 | queue_io(wb, work->older_than_this); |
83ba7b07 | 766 | if (work->sb) |
d46db3d5 | 767 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 768 | else |
d46db3d5 WF |
769 | progress = __writeback_inodes_wb(wb, work); |
770 | trace_writeback_written(wb->bdi, work); | |
03ba3782 | 771 | |
e98be2d5 WF |
772 | wb_update_bandwidth(wb, wb_start); |
773 | ||
03ba3782 | 774 | /* |
e6fb6da2 WF |
775 | * Did we write something? Try for more |
776 | * | |
777 | * Dirty inodes are moved to b_io for writeback in batches. | |
778 | * The completion of the current batch does not necessarily | |
779 | * mean the overall work is done. So we keep looping as long | |
780 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 781 | */ |
d46db3d5 | 782 | if (progress) |
cb9bd115 | 783 | continue; |
71fd05a8 | 784 | /* |
e6fb6da2 | 785 | * No more inodes for IO, bail |
71fd05a8 | 786 | */ |
b7a2441f | 787 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 788 | break; |
71fd05a8 JA |
789 | /* |
790 | * Nothing written. Wait for some inode to | |
791 | * become available for writeback. Otherwise | |
792 | * we'll just busyloop. | |
793 | */ | |
71fd05a8 | 794 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 795 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 796 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 797 | spin_lock(&inode->i_lock); |
f758eeab | 798 | inode_wait_for_writeback(inode, wb); |
250df6ed | 799 | spin_unlock(&inode->i_lock); |
03ba3782 JA |
800 | } |
801 | } | |
e8dfc305 | 802 | spin_unlock(&wb->list_lock); |
03ba3782 | 803 | |
d46db3d5 | 804 | return nr_pages - work->nr_pages; |
03ba3782 JA |
805 | } |
806 | ||
807 | /* | |
83ba7b07 | 808 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 809 | */ |
83ba7b07 | 810 | static struct wb_writeback_work * |
08852b6d | 811 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 812 | { |
83ba7b07 | 813 | struct wb_writeback_work *work = NULL; |
03ba3782 | 814 | |
6467716a | 815 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
816 | if (!list_empty(&bdi->work_list)) { |
817 | work = list_entry(bdi->work_list.next, | |
818 | struct wb_writeback_work, list); | |
819 | list_del_init(&work->list); | |
03ba3782 | 820 | } |
6467716a | 821 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 822 | return work; |
03ba3782 JA |
823 | } |
824 | ||
cdf01dd5 LT |
825 | /* |
826 | * Add in the number of potentially dirty inodes, because each inode | |
827 | * write can dirty pagecache in the underlying blockdev. | |
828 | */ | |
829 | static unsigned long get_nr_dirty_pages(void) | |
830 | { | |
831 | return global_page_state(NR_FILE_DIRTY) + | |
832 | global_page_state(NR_UNSTABLE_NFS) + | |
833 | get_nr_dirty_inodes(); | |
834 | } | |
835 | ||
6585027a JK |
836 | static long wb_check_background_flush(struct bdi_writeback *wb) |
837 | { | |
838 | if (over_bground_thresh()) { | |
839 | ||
840 | struct wb_writeback_work work = { | |
841 | .nr_pages = LONG_MAX, | |
842 | .sync_mode = WB_SYNC_NONE, | |
843 | .for_background = 1, | |
844 | .range_cyclic = 1, | |
845 | }; | |
846 | ||
847 | return wb_writeback(wb, &work); | |
848 | } | |
849 | ||
850 | return 0; | |
851 | } | |
852 | ||
03ba3782 JA |
853 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
854 | { | |
855 | unsigned long expired; | |
856 | long nr_pages; | |
857 | ||
69b62d01 JA |
858 | /* |
859 | * When set to zero, disable periodic writeback | |
860 | */ | |
861 | if (!dirty_writeback_interval) | |
862 | return 0; | |
863 | ||
03ba3782 JA |
864 | expired = wb->last_old_flush + |
865 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
866 | if (time_before(jiffies, expired)) | |
867 | return 0; | |
868 | ||
869 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 870 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 871 | |
c4a77a6c | 872 | if (nr_pages) { |
83ba7b07 | 873 | struct wb_writeback_work work = { |
c4a77a6c JA |
874 | .nr_pages = nr_pages, |
875 | .sync_mode = WB_SYNC_NONE, | |
876 | .for_kupdate = 1, | |
877 | .range_cyclic = 1, | |
878 | }; | |
879 | ||
83ba7b07 | 880 | return wb_writeback(wb, &work); |
c4a77a6c | 881 | } |
03ba3782 JA |
882 | |
883 | return 0; | |
884 | } | |
885 | ||
886 | /* | |
887 | * Retrieve work items and do the writeback they describe | |
888 | */ | |
889 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
890 | { | |
891 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 892 | struct wb_writeback_work *work; |
c4a77a6c | 893 | long wrote = 0; |
03ba3782 | 894 | |
81d73a32 | 895 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 896 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
897 | /* |
898 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 899 | * because this thread is exiting now. |
03ba3782 JA |
900 | */ |
901 | if (force_wait) | |
83ba7b07 | 902 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 903 | |
455b2864 DC |
904 | trace_writeback_exec(bdi, work); |
905 | ||
83ba7b07 | 906 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
907 | |
908 | /* | |
83ba7b07 CH |
909 | * Notify the caller of completion if this is a synchronous |
910 | * work item, otherwise just free it. | |
03ba3782 | 911 | */ |
83ba7b07 CH |
912 | if (work->done) |
913 | complete(work->done); | |
914 | else | |
915 | kfree(work); | |
03ba3782 JA |
916 | } |
917 | ||
918 | /* | |
919 | * Check for periodic writeback, kupdated() style | |
920 | */ | |
921 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 922 | wrote += wb_check_background_flush(wb); |
81d73a32 | 923 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
924 | |
925 | return wrote; | |
926 | } | |
927 | ||
928 | /* | |
929 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
930 | * wakes up periodically and does kupdated style flushing. | |
931 | */ | |
08243900 | 932 | int bdi_writeback_thread(void *data) |
03ba3782 | 933 | { |
08243900 CH |
934 | struct bdi_writeback *wb = data; |
935 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
936 | long pages_written; |
937 | ||
766f9164 | 938 | current->flags |= PF_SWAPWRITE; |
08243900 | 939 | set_freezable(); |
ecd58403 | 940 | wb->last_active = jiffies; |
08243900 CH |
941 | |
942 | /* | |
943 | * Our parent may run at a different priority, just set us to normal | |
944 | */ | |
945 | set_user_nice(current, 0); | |
946 | ||
455b2864 DC |
947 | trace_writeback_thread_start(bdi); |
948 | ||
03ba3782 | 949 | while (!kthread_should_stop()) { |
6467716a AB |
950 | /* |
951 | * Remove own delayed wake-up timer, since we are already awake | |
952 | * and we'll take care of the preriodic write-back. | |
953 | */ | |
954 | del_timer(&wb->wakeup_timer); | |
955 | ||
03ba3782 JA |
956 | pages_written = wb_do_writeback(wb, 0); |
957 | ||
455b2864 DC |
958 | trace_writeback_pages_written(pages_written); |
959 | ||
03ba3782 | 960 | if (pages_written) |
ecd58403 | 961 | wb->last_active = jiffies; |
03ba3782 | 962 | |
297252c8 | 963 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 964 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 965 | __set_current_state(TASK_RUNNING); |
297252c8 | 966 | continue; |
03ba3782 JA |
967 | } |
968 | ||
253c34e9 | 969 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 970 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
971 | else { |
972 | /* | |
973 | * We have nothing to do, so can go sleep without any | |
974 | * timeout and save power. When a work is queued or | |
975 | * something is made dirty - we will be woken up. | |
976 | */ | |
297252c8 | 977 | schedule(); |
f9eadbbd | 978 | } |
69b62d01 | 979 | |
03ba3782 JA |
980 | try_to_freeze(); |
981 | } | |
982 | ||
fff5b85a | 983 | /* Flush any work that raced with us exiting */ |
08243900 CH |
984 | if (!list_empty(&bdi->work_list)) |
985 | wb_do_writeback(wb, 1); | |
455b2864 DC |
986 | |
987 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
988 | return 0; |
989 | } | |
990 | ||
08243900 | 991 | |
03ba3782 | 992 | /* |
b8c2f347 CH |
993 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
994 | * the whole world. | |
03ba3782 | 995 | */ |
b8c2f347 | 996 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 997 | { |
b8c2f347 | 998 | struct backing_dev_info *bdi; |
03ba3782 | 999 | |
83ba7b07 CH |
1000 | if (!nr_pages) { |
1001 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
1002 | global_page_state(NR_UNSTABLE_NFS); |
1003 | } | |
03ba3782 | 1004 | |
b8c2f347 | 1005 | rcu_read_lock(); |
cfc4ba53 | 1006 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1007 | if (!bdi_has_dirty_io(bdi)) |
1008 | continue; | |
6585027a | 1009 | __bdi_start_writeback(bdi, nr_pages, false); |
03ba3782 | 1010 | } |
cfc4ba53 | 1011 | rcu_read_unlock(); |
1da177e4 LT |
1012 | } |
1013 | ||
03ba3782 JA |
1014 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1015 | { | |
1016 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1017 | struct dentry *dentry; | |
1018 | const char *name = "?"; | |
1019 | ||
1020 | dentry = d_find_alias(inode); | |
1021 | if (dentry) { | |
1022 | spin_lock(&dentry->d_lock); | |
1023 | name = (const char *) dentry->d_name.name; | |
1024 | } | |
1025 | printk(KERN_DEBUG | |
1026 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1027 | current->comm, task_pid_nr(current), inode->i_ino, | |
1028 | name, inode->i_sb->s_id); | |
1029 | if (dentry) { | |
1030 | spin_unlock(&dentry->d_lock); | |
1031 | dput(dentry); | |
1032 | } | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | /** | |
1037 | * __mark_inode_dirty - internal function | |
1038 | * @inode: inode to mark | |
1039 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1040 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1041 | * mark_inode_dirty_sync. | |
1da177e4 | 1042 | * |
03ba3782 JA |
1043 | * Put the inode on the super block's dirty list. |
1044 | * | |
1045 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1046 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1047 | * If it was not hashed, it will never be added to the dirty list | |
1048 | * even if it is later hashed, as it will have been marked dirty already. | |
1049 | * | |
1050 | * In short, make sure you hash any inodes _before_ you start marking | |
1051 | * them dirty. | |
1da177e4 | 1052 | * |
03ba3782 JA |
1053 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1054 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1055 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1056 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1057 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1058 | * blockdev inode. | |
1da177e4 | 1059 | */ |
03ba3782 | 1060 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1061 | { |
03ba3782 | 1062 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1063 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1064 | |
03ba3782 JA |
1065 | /* |
1066 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1067 | * dirty the inode itself | |
1068 | */ | |
1069 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1070 | if (sb->s_op->dirty_inode) | |
aa385729 | 1071 | sb->s_op->dirty_inode(inode, flags); |
03ba3782 JA |
1072 | } |
1073 | ||
1074 | /* | |
1075 | * make sure that changes are seen by all cpus before we test i_state | |
1076 | * -- mikulas | |
1077 | */ | |
1078 | smp_mb(); | |
1079 | ||
1080 | /* avoid the locking if we can */ | |
1081 | if ((inode->i_state & flags) == flags) | |
1082 | return; | |
1083 | ||
1084 | if (unlikely(block_dump)) | |
1085 | block_dump___mark_inode_dirty(inode); | |
1086 | ||
250df6ed | 1087 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1088 | if ((inode->i_state & flags) != flags) { |
1089 | const int was_dirty = inode->i_state & I_DIRTY; | |
1090 | ||
1091 | inode->i_state |= flags; | |
1092 | ||
1093 | /* | |
1094 | * If the inode is being synced, just update its dirty state. | |
1095 | * The unlocker will place the inode on the appropriate | |
1096 | * superblock list, based upon its state. | |
1097 | */ | |
1098 | if (inode->i_state & I_SYNC) | |
250df6ed | 1099 | goto out_unlock_inode; |
03ba3782 JA |
1100 | |
1101 | /* | |
1102 | * Only add valid (hashed) inodes to the superblock's | |
1103 | * dirty list. Add blockdev inodes as well. | |
1104 | */ | |
1105 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1106 | if (inode_unhashed(inode)) |
250df6ed | 1107 | goto out_unlock_inode; |
03ba3782 | 1108 | } |
a4ffdde6 | 1109 | if (inode->i_state & I_FREEING) |
250df6ed | 1110 | goto out_unlock_inode; |
03ba3782 JA |
1111 | |
1112 | /* | |
1113 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1114 | * reposition it (that would break b_dirty time-ordering). | |
1115 | */ | |
1116 | if (!was_dirty) { | |
a66979ab | 1117 | bool wakeup_bdi = false; |
253c34e9 AB |
1118 | bdi = inode_to_bdi(inode); |
1119 | ||
1120 | if (bdi_cap_writeback_dirty(bdi)) { | |
1121 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1122 | "bdi-%s not registered\n", bdi->name); | |
1123 | ||
1124 | /* | |
1125 | * If this is the first dirty inode for this | |
1126 | * bdi, we have to wake-up the corresponding | |
1127 | * bdi thread to make sure background | |
1128 | * write-back happens later. | |
1129 | */ | |
1130 | if (!wb_has_dirty_io(&bdi->wb)) | |
1131 | wakeup_bdi = true; | |
500b067c | 1132 | } |
03ba3782 | 1133 | |
a66979ab | 1134 | spin_unlock(&inode->i_lock); |
f758eeab | 1135 | spin_lock(&bdi->wb.list_lock); |
03ba3782 | 1136 | inode->dirtied_when = jiffies; |
7ccf19a8 | 1137 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1138 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1139 | |
1140 | if (wakeup_bdi) | |
1141 | bdi_wakeup_thread_delayed(bdi); | |
1142 | return; | |
1da177e4 | 1143 | } |
1da177e4 | 1144 | } |
250df6ed DC |
1145 | out_unlock_inode: |
1146 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1147 | |
03ba3782 JA |
1148 | } |
1149 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1150 | ||
1151 | /* | |
1152 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1153 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1154 | * | |
1155 | * If older_than_this is non-NULL, then only write out inodes which | |
1156 | * had their first dirtying at a time earlier than *older_than_this. | |
1157 | * | |
03ba3782 JA |
1158 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1159 | * This function assumes that the blockdev superblock's inodes are backed by | |
1160 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1161 | * assume that all inodes are backed by the same queue. | |
1162 | * | |
1163 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1164 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1165 | * on the writer throttling path, and we get decent balancing between many | |
1166 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1167 | */ | |
b6e51316 | 1168 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1169 | { |
1170 | struct inode *inode, *old_inode = NULL; | |
1171 | ||
1172 | /* | |
1173 | * We need to be protected against the filesystem going from | |
1174 | * r/o to r/w or vice versa. | |
1175 | */ | |
b6e51316 | 1176 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1177 | |
55fa6091 | 1178 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1179 | |
1180 | /* | |
1181 | * Data integrity sync. Must wait for all pages under writeback, | |
1182 | * because there may have been pages dirtied before our sync | |
1183 | * call, but which had writeout started before we write it out. | |
1184 | * In which case, the inode may not be on the dirty list, but | |
1185 | * we still have to wait for that writeout. | |
1186 | */ | |
b6e51316 | 1187 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1188 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1189 | |
250df6ed DC |
1190 | spin_lock(&inode->i_lock); |
1191 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1192 | (mapping->nrpages == 0)) { | |
1193 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1194 | continue; |
250df6ed | 1195 | } |
03ba3782 | 1196 | __iget(inode); |
250df6ed | 1197 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1198 | spin_unlock(&inode_sb_list_lock); |
1199 | ||
03ba3782 | 1200 | /* |
55fa6091 DC |
1201 | * We hold a reference to 'inode' so it couldn't have been |
1202 | * removed from s_inodes list while we dropped the | |
1203 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1204 | * be holding the last reference and we cannot iput it under | |
1205 | * inode_sb_list_lock. So we keep the reference and iput it | |
1206 | * later. | |
03ba3782 JA |
1207 | */ |
1208 | iput(old_inode); | |
1209 | old_inode = inode; | |
1210 | ||
1211 | filemap_fdatawait(mapping); | |
1212 | ||
1213 | cond_resched(); | |
1214 | ||
55fa6091 | 1215 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1216 | } |
55fa6091 | 1217 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1218 | iput(old_inode); |
1da177e4 LT |
1219 | } |
1220 | ||
d8a8559c | 1221 | /** |
3259f8be | 1222 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1223 | * @sb: the superblock |
3259f8be | 1224 | * @nr: the number of pages to write |
1da177e4 | 1225 | * |
d8a8559c JA |
1226 | * Start writeback on some inodes on this super_block. No guarantees are made |
1227 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1228 | * for IO completion of submitted IO. |
1da177e4 | 1229 | */ |
3259f8be | 1230 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
1da177e4 | 1231 | { |
83ba7b07 CH |
1232 | DECLARE_COMPLETION_ONSTACK(done); |
1233 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1234 | .sb = sb, |
1235 | .sync_mode = WB_SYNC_NONE, | |
1236 | .tagged_writepages = 1, | |
1237 | .done = &done, | |
1238 | .nr_pages = nr, | |
3c4d7165 | 1239 | }; |
d8a8559c | 1240 | |
cf37e972 | 1241 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1242 | bdi_queue_work(sb->s_bdi, &work); |
1243 | wait_for_completion(&done); | |
e913fc82 | 1244 | } |
3259f8be CM |
1245 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1246 | ||
1247 | /** | |
1248 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1249 | * @sb: the superblock | |
1250 | * | |
1251 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1252 | * on how many (if any) will be written, and this function does not wait | |
1253 | * for IO completion of submitted IO. | |
1254 | */ | |
1255 | void writeback_inodes_sb(struct super_block *sb) | |
1256 | { | |
925d169f | 1257 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
3259f8be | 1258 | } |
0e3c9a22 | 1259 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1260 | |
17bd55d0 ES |
1261 | /** |
1262 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1263 | * @sb: the superblock | |
1264 | * | |
1265 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1266 | * Returns 1 if writeback was started, 0 if not. | |
1267 | */ | |
1268 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1269 | { | |
1270 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1271 | down_read(&sb->s_umount); |
17bd55d0 | 1272 | writeback_inodes_sb(sb); |
cf37e972 | 1273 | up_read(&sb->s_umount); |
17bd55d0 ES |
1274 | return 1; |
1275 | } else | |
1276 | return 0; | |
1277 | } | |
1278 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1279 | ||
3259f8be CM |
1280 | /** |
1281 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1282 | * @sb: the superblock | |
1283 | * @nr: the number of pages to write | |
1284 | * | |
1285 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1286 | * Returns 1 if writeback was started, 0 if not. | |
1287 | */ | |
1288 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
1289 | unsigned long nr) | |
1290 | { | |
1291 | if (!writeback_in_progress(sb->s_bdi)) { | |
1292 | down_read(&sb->s_umount); | |
1293 | writeback_inodes_sb_nr(sb, nr); | |
1294 | up_read(&sb->s_umount); | |
1295 | return 1; | |
1296 | } else | |
1297 | return 0; | |
1298 | } | |
1299 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1300 | ||
d8a8559c JA |
1301 | /** |
1302 | * sync_inodes_sb - sync sb inode pages | |
1303 | * @sb: the superblock | |
1304 | * | |
1305 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1306 | * super_block. |
d8a8559c | 1307 | */ |
b6e51316 | 1308 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1309 | { |
83ba7b07 CH |
1310 | DECLARE_COMPLETION_ONSTACK(done); |
1311 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1312 | .sb = sb, |
1313 | .sync_mode = WB_SYNC_ALL, | |
1314 | .nr_pages = LONG_MAX, | |
1315 | .range_cyclic = 0, | |
83ba7b07 | 1316 | .done = &done, |
3c4d7165 CH |
1317 | }; |
1318 | ||
cf37e972 CH |
1319 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1320 | ||
83ba7b07 CH |
1321 | bdi_queue_work(sb->s_bdi, &work); |
1322 | wait_for_completion(&done); | |
1323 | ||
b6e51316 | 1324 | wait_sb_inodes(sb); |
1da177e4 | 1325 | } |
d8a8559c | 1326 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1327 | |
1da177e4 | 1328 | /** |
7f04c26d AA |
1329 | * write_inode_now - write an inode to disk |
1330 | * @inode: inode to write to disk | |
1331 | * @sync: whether the write should be synchronous or not | |
1332 | * | |
1333 | * This function commits an inode to disk immediately if it is dirty. This is | |
1334 | * primarily needed by knfsd. | |
1da177e4 | 1335 | * |
7f04c26d | 1336 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1337 | */ |
1da177e4 LT |
1338 | int write_inode_now(struct inode *inode, int sync) |
1339 | { | |
f758eeab | 1340 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1341 | int ret; |
1342 | struct writeback_control wbc = { | |
1343 | .nr_to_write = LONG_MAX, | |
18914b18 | 1344 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1345 | .range_start = 0, |
1346 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1347 | }; |
1348 | ||
1349 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1350 | wbc.nr_to_write = 0; |
1da177e4 LT |
1351 | |
1352 | might_sleep(); | |
f758eeab | 1353 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1354 | spin_lock(&inode->i_lock); |
f758eeab | 1355 | ret = writeback_single_inode(inode, wb, &wbc); |
0f1b1fd8 | 1356 | spin_unlock(&inode->i_lock); |
f758eeab | 1357 | spin_unlock(&wb->list_lock); |
1da177e4 | 1358 | if (sync) |
1c0eeaf5 | 1359 | inode_sync_wait(inode); |
1da177e4 LT |
1360 | return ret; |
1361 | } | |
1362 | EXPORT_SYMBOL(write_inode_now); | |
1363 | ||
1364 | /** | |
1365 | * sync_inode - write an inode and its pages to disk. | |
1366 | * @inode: the inode to sync | |
1367 | * @wbc: controls the writeback mode | |
1368 | * | |
1369 | * sync_inode() will write an inode and its pages to disk. It will also | |
1370 | * correctly update the inode on its superblock's dirty inode lists and will | |
1371 | * update inode->i_state. | |
1372 | * | |
1373 | * The caller must have a ref on the inode. | |
1374 | */ | |
1375 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1376 | { | |
f758eeab | 1377 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1378 | int ret; |
1379 | ||
f758eeab | 1380 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1381 | spin_lock(&inode->i_lock); |
f758eeab | 1382 | ret = writeback_single_inode(inode, wb, wbc); |
0f1b1fd8 | 1383 | spin_unlock(&inode->i_lock); |
f758eeab | 1384 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
1385 | return ret; |
1386 | } | |
1387 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1388 | |
1389 | /** | |
c691b9d9 | 1390 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1391 | * @inode: the inode to sync |
1392 | * @wait: wait for I/O to complete. | |
1393 | * | |
c691b9d9 | 1394 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1395 | * |
1396 | * Note: only writes the actual inode, no associated data or other metadata. | |
1397 | */ | |
1398 | int sync_inode_metadata(struct inode *inode, int wait) | |
1399 | { | |
1400 | struct writeback_control wbc = { | |
1401 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1402 | .nr_to_write = 0, /* metadata-only */ | |
1403 | }; | |
1404 | ||
1405 | return sync_inode(inode, &wbc); | |
1406 | } | |
1407 | EXPORT_SYMBOL(sync_inode_metadata); |