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 | |
c4a77a6c JA |
32 | /* |
33 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
34 | */ | |
83ba7b07 | 35 | struct wb_writeback_work { |
c4a77a6c JA |
36 | long nr_pages; |
37 | struct super_block *sb; | |
38 | enum writeback_sync_modes sync_mode; | |
52957fe1 HS |
39 | unsigned int for_kupdate:1; |
40 | unsigned int range_cyclic:1; | |
41 | unsigned int for_background:1; | |
c4a77a6c | 42 | |
8010c3b6 | 43 | struct list_head list; /* pending work list */ |
83ba7b07 | 44 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
45 | }; |
46 | ||
455b2864 DC |
47 | /* |
48 | * Include the creation of the trace points after defining the | |
49 | * wb_writeback_work structure so that the definition remains local to this | |
50 | * file. | |
51 | */ | |
52 | #define CREATE_TRACE_POINTS | |
53 | #include <trace/events/writeback.h> | |
54 | ||
455b2864 DC |
55 | /* |
56 | * We don't actually have pdflush, but this one is exported though /proc... | |
57 | */ | |
58 | int nr_pdflush_threads; | |
59 | ||
f11b00f3 AB |
60 | /** |
61 | * writeback_in_progress - determine whether there is writeback in progress | |
62 | * @bdi: the device's backing_dev_info structure. | |
63 | * | |
03ba3782 JA |
64 | * Determine whether there is writeback waiting to be handled against a |
65 | * backing device. | |
f11b00f3 AB |
66 | */ |
67 | int writeback_in_progress(struct backing_dev_info *bdi) | |
68 | { | |
81d73a32 | 69 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 AB |
70 | } |
71 | ||
692ebd17 JK |
72 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
73 | { | |
74 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 75 | |
aaead25b CH |
76 | if (strcmp(sb->s_type->name, "bdev") == 0) |
77 | return inode->i_mapping->backing_dev_info; | |
78 | ||
79 | return sb->s_bdi; | |
692ebd17 JK |
80 | } |
81 | ||
7ccf19a8 NP |
82 | static inline struct inode *wb_inode(struct list_head *head) |
83 | { | |
84 | return list_entry(head, struct inode, i_wb_list); | |
85 | } | |
86 | ||
6585027a JK |
87 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
88 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) | |
03ba3782 | 89 | { |
fff5b85a AB |
90 | if (bdi->wb.task) { |
91 | wake_up_process(bdi->wb.task); | |
92 | } else { | |
93 | /* | |
94 | * The bdi thread isn't there, wake up the forker thread which | |
95 | * will create and run it. | |
96 | */ | |
03ba3782 | 97 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 98 | } |
6585027a JK |
99 | } |
100 | ||
101 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
102 | struct wb_writeback_work *work) | |
103 | { | |
104 | trace_writeback_queue(bdi, work); | |
105 | ||
106 | spin_lock_bh(&bdi->wb_lock); | |
107 | list_add_tail(&work->list, &bdi->work_list); | |
108 | if (!bdi->wb.task) | |
109 | trace_writeback_nothread(bdi, work); | |
110 | bdi_wakeup_flusher(bdi); | |
6467716a | 111 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
112 | } |
113 | ||
83ba7b07 CH |
114 | static void |
115 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
6585027a | 116 | bool range_cyclic) |
1da177e4 | 117 | { |
83ba7b07 | 118 | struct wb_writeback_work *work; |
03ba3782 | 119 | |
bcddc3f0 JA |
120 | /* |
121 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
122 | * wakeup the thread for old dirty data writeback | |
123 | */ | |
83ba7b07 CH |
124 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
125 | if (!work) { | |
455b2864 DC |
126 | if (bdi->wb.task) { |
127 | trace_writeback_nowork(bdi); | |
83ba7b07 | 128 | wake_up_process(bdi->wb.task); |
455b2864 | 129 | } |
83ba7b07 | 130 | return; |
bcddc3f0 | 131 | } |
03ba3782 | 132 | |
83ba7b07 CH |
133 | work->sync_mode = WB_SYNC_NONE; |
134 | work->nr_pages = nr_pages; | |
135 | work->range_cyclic = range_cyclic; | |
03ba3782 | 136 | |
83ba7b07 | 137 | bdi_queue_work(bdi, work); |
b6e51316 JA |
138 | } |
139 | ||
140 | /** | |
141 | * bdi_start_writeback - start writeback | |
142 | * @bdi: the backing device to write from | |
143 | * @nr_pages: the number of pages to write | |
144 | * | |
145 | * Description: | |
146 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 147 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 148 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
149 | * |
150 | */ | |
c5444198 | 151 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 152 | { |
6585027a | 153 | __bdi_start_writeback(bdi, nr_pages, true); |
c5444198 | 154 | } |
d3ddec76 | 155 | |
c5444198 CH |
156 | /** |
157 | * bdi_start_background_writeback - start background writeback | |
158 | * @bdi: the backing device to write from | |
159 | * | |
160 | * Description: | |
6585027a JK |
161 | * This makes sure WB_SYNC_NONE background writeback happens. When |
162 | * this function returns, it is only guaranteed that for given BDI | |
163 | * some IO is happening if we are over background dirty threshold. | |
164 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
165 | */ |
166 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
167 | { | |
6585027a JK |
168 | /* |
169 | * We just wake up the flusher thread. It will perform background | |
170 | * writeback as soon as there is no other work to do. | |
171 | */ | |
71927e84 | 172 | trace_writeback_wake_background(bdi); |
6585027a JK |
173 | spin_lock_bh(&bdi->wb_lock); |
174 | bdi_wakeup_flusher(bdi); | |
175 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
176 | } |
177 | ||
a66979ab DC |
178 | /* |
179 | * Remove the inode from the writeback list it is on. | |
180 | */ | |
181 | void inode_wb_list_del(struct inode *inode) | |
182 | { | |
183 | spin_lock(&inode_wb_list_lock); | |
184 | list_del_init(&inode->i_wb_list); | |
185 | spin_unlock(&inode_wb_list_lock); | |
186 | } | |
187 | ||
188 | ||
6610a0bc AM |
189 | /* |
190 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
191 | * furthest end of its superblock's dirty-inode list. | |
192 | * | |
193 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 194 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
195 | * the case then the inode must have been redirtied while it was being written |
196 | * out and we don't reset its dirtied_when. | |
197 | */ | |
198 | static void redirty_tail(struct inode *inode) | |
199 | { | |
03ba3782 | 200 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 201 | |
a66979ab | 202 | assert_spin_locked(&inode_wb_list_lock); |
03ba3782 | 203 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 204 | struct inode *tail; |
6610a0bc | 205 | |
7ccf19a8 | 206 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 207 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
208 | inode->dirtied_when = jiffies; |
209 | } | |
7ccf19a8 | 210 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
211 | } |
212 | ||
c986d1e2 | 213 | /* |
66f3b8e2 | 214 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 215 | */ |
0e0f4fc2 | 216 | static void requeue_io(struct inode *inode) |
c986d1e2 | 217 | { |
03ba3782 JA |
218 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
219 | ||
a66979ab | 220 | assert_spin_locked(&inode_wb_list_lock); |
7ccf19a8 | 221 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
222 | } |
223 | ||
1c0eeaf5 JE |
224 | static void inode_sync_complete(struct inode *inode) |
225 | { | |
226 | /* | |
a66979ab DC |
227 | * Prevent speculative execution through |
228 | * spin_unlock(&inode_wb_list_lock); | |
1c0eeaf5 | 229 | */ |
a66979ab | 230 | |
1c0eeaf5 JE |
231 | smp_mb(); |
232 | wake_up_bit(&inode->i_state, __I_SYNC); | |
233 | } | |
234 | ||
d2caa3c5 JL |
235 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
236 | { | |
237 | bool ret = time_after(inode->dirtied_when, t); | |
238 | #ifndef CONFIG_64BIT | |
239 | /* | |
240 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
241 | * It _appears_ to be in the future, but is actually in distant past. | |
242 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 243 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
244 | */ |
245 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
246 | #endif | |
247 | return ret; | |
248 | } | |
249 | ||
2c136579 FW |
250 | /* |
251 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
252 | */ | |
253 | static void move_expired_inodes(struct list_head *delaying_queue, | |
254 | struct list_head *dispatch_queue, | |
255 | unsigned long *older_than_this) | |
256 | { | |
5c03449d SL |
257 | LIST_HEAD(tmp); |
258 | struct list_head *pos, *node; | |
cf137307 | 259 | struct super_block *sb = NULL; |
5c03449d | 260 | struct inode *inode; |
cf137307 | 261 | int do_sb_sort = 0; |
5c03449d | 262 | |
2c136579 | 263 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 264 | inode = wb_inode(delaying_queue->prev); |
2c136579 | 265 | if (older_than_this && |
d2caa3c5 | 266 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 267 | break; |
cf137307 JA |
268 | if (sb && sb != inode->i_sb) |
269 | do_sb_sort = 1; | |
270 | sb = inode->i_sb; | |
7ccf19a8 | 271 | list_move(&inode->i_wb_list, &tmp); |
5c03449d SL |
272 | } |
273 | ||
cf137307 JA |
274 | /* just one sb in list, splice to dispatch_queue and we're done */ |
275 | if (!do_sb_sort) { | |
276 | list_splice(&tmp, dispatch_queue); | |
277 | return; | |
278 | } | |
279 | ||
5c03449d SL |
280 | /* Move inodes from one superblock together */ |
281 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 282 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 283 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 284 | inode = wb_inode(pos); |
5c03449d | 285 | if (inode->i_sb == sb) |
7ccf19a8 | 286 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 287 | } |
2c136579 FW |
288 | } |
289 | } | |
290 | ||
291 | /* | |
292 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
293 | * Before |
294 | * newly dirtied b_dirty b_io b_more_io | |
295 | * =============> gf edc BA | |
296 | * After | |
297 | * newly dirtied b_dirty b_io b_more_io | |
298 | * =============> g fBAedc | |
299 | * | | |
300 | * +--> dequeue for IO | |
2c136579 | 301 | */ |
03ba3782 | 302 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 303 | { |
a66979ab | 304 | assert_spin_locked(&inode_wb_list_lock); |
4ea879b9 | 305 | list_splice_init(&wb->b_more_io, &wb->b_io); |
03ba3782 | 306 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
66f3b8e2 JA |
307 | } |
308 | ||
a9185b41 | 309 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 310 | { |
03ba3782 | 311 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 312 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 313 | return 0; |
08d8e974 | 314 | } |
08d8e974 | 315 | |
1da177e4 | 316 | /* |
01c03194 CH |
317 | * Wait for writeback on an inode to complete. |
318 | */ | |
319 | static void inode_wait_for_writeback(struct inode *inode) | |
320 | { | |
321 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
322 | wait_queue_head_t *wqh; | |
323 | ||
324 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
325 | while (inode->i_state & I_SYNC) { |
326 | spin_unlock(&inode->i_lock); | |
a66979ab | 327 | spin_unlock(&inode_wb_list_lock); |
01c03194 | 328 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
a66979ab | 329 | spin_lock(&inode_wb_list_lock); |
250df6ed | 330 | spin_lock(&inode->i_lock); |
58a9d3d8 | 331 | } |
01c03194 CH |
332 | } |
333 | ||
334 | /* | |
0f1b1fd8 DC |
335 | * Write out an inode's dirty pages. Called under inode_wb_list_lock and |
336 | * inode->i_lock. Either the caller has an active reference on the inode or | |
337 | * the inode has I_WILL_FREE set. | |
01c03194 | 338 | * |
1da177e4 LT |
339 | * If `wait' is set, wait on the writeout. |
340 | * | |
341 | * The whole writeout design is quite complex and fragile. We want to avoid | |
342 | * starvation of particular inodes when others are being redirtied, prevent | |
343 | * livelocks, etc. | |
1da177e4 LT |
344 | */ |
345 | static int | |
01c03194 | 346 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 347 | { |
1da177e4 | 348 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 349 | unsigned dirty; |
1da177e4 LT |
350 | int ret; |
351 | ||
0f1b1fd8 DC |
352 | assert_spin_locked(&inode_wb_list_lock); |
353 | assert_spin_locked(&inode->i_lock); | |
354 | ||
01c03194 CH |
355 | if (!atomic_read(&inode->i_count)) |
356 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
357 | else | |
358 | WARN_ON(inode->i_state & I_WILL_FREE); | |
359 | ||
360 | if (inode->i_state & I_SYNC) { | |
361 | /* | |
362 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 363 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
364 | * writeback can proceed with the other inodes on s_io. |
365 | * | |
366 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 367 | * completed a full scan of b_io. |
01c03194 | 368 | */ |
a9185b41 | 369 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
370 | requeue_io(inode); |
371 | return 0; | |
372 | } | |
373 | ||
374 | /* | |
375 | * It's a data-integrity sync. We must wait. | |
376 | */ | |
377 | inode_wait_for_writeback(inode); | |
378 | } | |
379 | ||
1c0eeaf5 | 380 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 381 | |
5547e8aa | 382 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 383 | inode->i_state |= I_SYNC; |
5547e8aa | 384 | inode->i_state &= ~I_DIRTY_PAGES; |
250df6ed | 385 | spin_unlock(&inode->i_lock); |
a66979ab | 386 | spin_unlock(&inode_wb_list_lock); |
1da177e4 LT |
387 | |
388 | ret = do_writepages(mapping, wbc); | |
389 | ||
26821ed4 CH |
390 | /* |
391 | * Make sure to wait on the data before writing out the metadata. | |
392 | * This is important for filesystems that modify metadata on data | |
393 | * I/O completion. | |
394 | */ | |
a9185b41 | 395 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 396 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
397 | if (ret == 0) |
398 | ret = err; | |
399 | } | |
400 | ||
5547e8aa DM |
401 | /* |
402 | * Some filesystems may redirty the inode during the writeback | |
403 | * due to delalloc, clear dirty metadata flags right before | |
404 | * write_inode() | |
405 | */ | |
250df6ed | 406 | spin_lock(&inode->i_lock); |
5547e8aa DM |
407 | dirty = inode->i_state & I_DIRTY; |
408 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 409 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
410 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
411 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 412 | int err = write_inode(inode, wbc); |
1da177e4 LT |
413 | if (ret == 0) |
414 | ret = err; | |
415 | } | |
416 | ||
a66979ab | 417 | spin_lock(&inode_wb_list_lock); |
250df6ed | 418 | spin_lock(&inode->i_lock); |
1c0eeaf5 | 419 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 420 | if (!(inode->i_state & I_FREEING)) { |
23539afc | 421 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
422 | /* |
423 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 424 | * sometimes bales out without doing anything. |
1b43ef91 | 425 | */ |
a50aeb40 WF |
426 | inode->i_state |= I_DIRTY_PAGES; |
427 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 428 | /* |
a50aeb40 | 429 | * slice used up: queue for next turn |
1da177e4 | 430 | */ |
a50aeb40 | 431 | requeue_io(inode); |
1da177e4 LT |
432 | } else { |
433 | /* | |
a50aeb40 WF |
434 | * Writeback blocked by something other than |
435 | * congestion. Delay the inode for some time to | |
436 | * avoid spinning on the CPU (100% iowait) | |
437 | * retrying writeback of the dirty page/inode | |
438 | * that cannot be performed immediately. | |
1da177e4 | 439 | */ |
1b43ef91 | 440 | redirty_tail(inode); |
1da177e4 | 441 | } |
23539afc WF |
442 | } else if (inode->i_state & I_DIRTY) { |
443 | /* | |
444 | * Filesystems can dirty the inode during writeback | |
445 | * operations, such as delayed allocation during | |
446 | * submission or metadata updates after data IO | |
447 | * completion. | |
448 | */ | |
449 | redirty_tail(inode); | |
1da177e4 LT |
450 | } else { |
451 | /* | |
9e38d86f NP |
452 | * The inode is clean. At this point we either have |
453 | * a reference to the inode or it's on it's way out. | |
454 | * No need to add it back to the LRU. | |
1da177e4 | 455 | */ |
7ccf19a8 | 456 | list_del_init(&inode->i_wb_list); |
1da177e4 LT |
457 | } |
458 | } | |
1c0eeaf5 | 459 | inode_sync_complete(inode); |
1da177e4 LT |
460 | return ret; |
461 | } | |
462 | ||
f11c9c5c ES |
463 | /* |
464 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
465 | * |
466 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
467 | * inodes. Otherwise write only ones which go sequentially |
468 | * in reverse order. | |
edadfb10 | 469 | * |
f11c9c5c ES |
470 | * Return 1, if the caller writeback routine should be |
471 | * interrupted. Otherwise return 0. | |
472 | */ | |
edadfb10 CH |
473 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
474 | struct writeback_control *wbc, bool only_this_sb) | |
1da177e4 | 475 | { |
03ba3782 | 476 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 477 | long pages_skipped; |
7ccf19a8 | 478 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
479 | |
480 | if (inode->i_sb != sb) { | |
481 | if (only_this_sb) { | |
482 | /* | |
483 | * We only want to write back data for this | |
484 | * superblock, move all inodes not belonging | |
485 | * to it back onto the dirty list. | |
486 | */ | |
487 | redirty_tail(inode); | |
488 | continue; | |
489 | } | |
490 | ||
491 | /* | |
492 | * The inode belongs to a different superblock. | |
493 | * Bounce back to the caller to unpin this and | |
494 | * pin the next superblock. | |
495 | */ | |
f11c9c5c | 496 | return 0; |
edadfb10 CH |
497 | } |
498 | ||
9843b76a CH |
499 | /* |
500 | * Don't bother with new inodes or inodes beeing freed, first | |
501 | * kind does not need peridic writeout yet, and for the latter | |
502 | * kind writeout is handled by the freer. | |
503 | */ | |
250df6ed | 504 | spin_lock(&inode->i_lock); |
9843b76a | 505 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 506 | spin_unlock(&inode->i_lock); |
7ef0d737 NP |
507 | requeue_io(inode); |
508 | continue; | |
509 | } | |
9843b76a | 510 | |
d2caa3c5 JL |
511 | /* |
512 | * Was this inode dirtied after sync_sb_inodes was called? | |
513 | * This keeps sync from extra jobs and livelock. | |
514 | */ | |
250df6ed DC |
515 | if (inode_dirtied_after(inode, wbc->wb_start)) { |
516 | spin_unlock(&inode->i_lock); | |
f11c9c5c | 517 | return 1; |
250df6ed | 518 | } |
1da177e4 | 519 | |
1da177e4 | 520 | __iget(inode); |
250df6ed | 521 | |
1da177e4 | 522 | pages_skipped = wbc->pages_skipped; |
01c03194 | 523 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
524 | if (wbc->pages_skipped != pages_skipped) { |
525 | /* | |
526 | * writeback is not making progress due to locked | |
527 | * buffers. Skip this inode for now. | |
528 | */ | |
f57b9b7b | 529 | redirty_tail(inode); |
1da177e4 | 530 | } |
0f1b1fd8 | 531 | spin_unlock(&inode->i_lock); |
a66979ab | 532 | spin_unlock(&inode_wb_list_lock); |
1da177e4 | 533 | iput(inode); |
4ffc8444 | 534 | cond_resched(); |
a66979ab | 535 | spin_lock(&inode_wb_list_lock); |
8bc3be27 FW |
536 | if (wbc->nr_to_write <= 0) { |
537 | wbc->more_io = 1; | |
f11c9c5c | 538 | return 1; |
8bc3be27 | 539 | } |
03ba3782 | 540 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 541 | wbc->more_io = 1; |
1da177e4 | 542 | } |
f11c9c5c ES |
543 | /* b_io is empty */ |
544 | return 1; | |
545 | } | |
546 | ||
9c3a8ee8 CH |
547 | void writeback_inodes_wb(struct bdi_writeback *wb, |
548 | struct writeback_control *wbc) | |
f11c9c5c ES |
549 | { |
550 | int ret = 0; | |
551 | ||
7624ee72 JK |
552 | if (!wbc->wb_start) |
553 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
a66979ab | 554 | spin_lock(&inode_wb_list_lock); |
f11c9c5c ES |
555 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
556 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 557 | |
f11c9c5c | 558 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 559 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 560 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 561 | |
12ad3ab6 | 562 | if (!grab_super_passive(sb)) { |
edadfb10 CH |
563 | requeue_io(inode); |
564 | continue; | |
f11c9c5c | 565 | } |
edadfb10 CH |
566 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
567 | drop_super(sb); | |
f11c9c5c | 568 | |
f11c9c5c ES |
569 | if (ret) |
570 | break; | |
571 | } | |
a66979ab | 572 | spin_unlock(&inode_wb_list_lock); |
66f3b8e2 JA |
573 | /* Leave any unwritten inodes on b_io */ |
574 | } | |
575 | ||
edadfb10 CH |
576 | static void __writeback_inodes_sb(struct super_block *sb, |
577 | struct bdi_writeback *wb, struct writeback_control *wbc) | |
578 | { | |
579 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
580 | ||
a66979ab | 581 | spin_lock(&inode_wb_list_lock); |
edadfb10 CH |
582 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
583 | queue_io(wb, wbc->older_than_this); | |
584 | writeback_sb_inodes(sb, wb, wbc, true); | |
a66979ab | 585 | spin_unlock(&inode_wb_list_lock); |
edadfb10 CH |
586 | } |
587 | ||
66f3b8e2 | 588 | /* |
03ba3782 JA |
589 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
590 | * operation. We do this so we don't hold I_SYNC against an inode for | |
591 | * enormous amounts of time, which would block a userspace task which has | |
592 | * been forced to throttle against that inode. Also, the code reevaluates | |
593 | * the dirty each time it has written this many pages. | |
594 | */ | |
595 | #define MAX_WRITEBACK_PAGES 1024 | |
596 | ||
597 | static inline bool over_bground_thresh(void) | |
598 | { | |
599 | unsigned long background_thresh, dirty_thresh; | |
600 | ||
16c4042f | 601 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 JA |
602 | |
603 | return (global_page_state(NR_FILE_DIRTY) + | |
4cbec4c8 | 604 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
03ba3782 JA |
605 | } |
606 | ||
607 | /* | |
608 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 609 | * |
03ba3782 JA |
610 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
611 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
612 | * just walks the superblock inode list, writing back any inodes which are | |
613 | * older than a specific point in time. | |
66f3b8e2 | 614 | * |
03ba3782 JA |
615 | * Try to run once per dirty_writeback_interval. But if a writeback event |
616 | * takes longer than a dirty_writeback_interval interval, then leave a | |
617 | * one-second gap. | |
66f3b8e2 | 618 | * |
03ba3782 JA |
619 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
620 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 621 | */ |
c4a77a6c | 622 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 623 | struct wb_writeback_work *work) |
66f3b8e2 | 624 | { |
03ba3782 | 625 | struct writeback_control wbc = { |
83ba7b07 | 626 | .sync_mode = work->sync_mode, |
03ba3782 | 627 | .older_than_this = NULL, |
83ba7b07 CH |
628 | .for_kupdate = work->for_kupdate, |
629 | .for_background = work->for_background, | |
630 | .range_cyclic = work->range_cyclic, | |
03ba3782 JA |
631 | }; |
632 | unsigned long oldest_jif; | |
633 | long wrote = 0; | |
b9543dac | 634 | long write_chunk; |
a5989bdc | 635 | struct inode *inode; |
66f3b8e2 | 636 | |
03ba3782 JA |
637 | if (wbc.for_kupdate) { |
638 | wbc.older_than_this = &oldest_jif; | |
639 | oldest_jif = jiffies - | |
640 | msecs_to_jiffies(dirty_expire_interval * 10); | |
641 | } | |
c4a77a6c JA |
642 | if (!wbc.range_cyclic) { |
643 | wbc.range_start = 0; | |
644 | wbc.range_end = LLONG_MAX; | |
645 | } | |
38f21977 | 646 | |
b9543dac JK |
647 | /* |
648 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
649 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
650 | * here avoids calling into writeback_inodes_wb() more than once. | |
651 | * | |
652 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
653 | * | |
654 | * wb_writeback() | |
655 | * __writeback_inodes_sb() <== called only once | |
656 | * write_cache_pages() <== called once for each inode | |
657 | * (quickly) tag currently dirty pages | |
658 | * (maybe slowly) sync all tagged pages | |
659 | */ | |
660 | if (wbc.sync_mode == WB_SYNC_NONE) | |
661 | write_chunk = MAX_WRITEBACK_PAGES; | |
662 | else | |
663 | write_chunk = LONG_MAX; | |
664 | ||
7624ee72 | 665 | wbc.wb_start = jiffies; /* livelock avoidance */ |
03ba3782 JA |
666 | for (;;) { |
667 | /* | |
d3ddec76 | 668 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 669 | */ |
83ba7b07 | 670 | if (work->nr_pages <= 0) |
03ba3782 | 671 | break; |
66f3b8e2 | 672 | |
aa373cf5 JK |
673 | /* |
674 | * Background writeout and kupdate-style writeback may | |
675 | * run forever. Stop them if there is other work to do | |
676 | * so that e.g. sync can proceed. They'll be restarted | |
677 | * after the other works are all done. | |
678 | */ | |
679 | if ((work->for_background || work->for_kupdate) && | |
680 | !list_empty(&wb->bdi->work_list)) | |
681 | break; | |
682 | ||
38f21977 | 683 | /* |
d3ddec76 WF |
684 | * For background writeout, stop when we are below the |
685 | * background dirty threshold | |
38f21977 | 686 | */ |
83ba7b07 | 687 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 688 | break; |
38f21977 | 689 | |
03ba3782 | 690 | wbc.more_io = 0; |
b9543dac | 691 | wbc.nr_to_write = write_chunk; |
03ba3782 | 692 | wbc.pages_skipped = 0; |
028c2dd1 DC |
693 | |
694 | trace_wbc_writeback_start(&wbc, wb->bdi); | |
83ba7b07 CH |
695 | if (work->sb) |
696 | __writeback_inodes_sb(work->sb, wb, &wbc); | |
edadfb10 CH |
697 | else |
698 | writeback_inodes_wb(wb, &wbc); | |
028c2dd1 DC |
699 | trace_wbc_writeback_written(&wbc, wb->bdi); |
700 | ||
b9543dac JK |
701 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
702 | wrote += write_chunk - wbc.nr_to_write; | |
03ba3782 JA |
703 | |
704 | /* | |
71fd05a8 | 705 | * If we consumed everything, see if we have more |
03ba3782 | 706 | */ |
71fd05a8 JA |
707 | if (wbc.nr_to_write <= 0) |
708 | continue; | |
709 | /* | |
710 | * Didn't write everything and we don't have more IO, bail | |
711 | */ | |
712 | if (!wbc.more_io) | |
03ba3782 | 713 | break; |
71fd05a8 JA |
714 | /* |
715 | * Did we write something? Try for more | |
716 | */ | |
b9543dac | 717 | if (wbc.nr_to_write < write_chunk) |
71fd05a8 JA |
718 | continue; |
719 | /* | |
720 | * Nothing written. Wait for some inode to | |
721 | * become available for writeback. Otherwise | |
722 | * we'll just busyloop. | |
723 | */ | |
a66979ab | 724 | spin_lock(&inode_wb_list_lock); |
71fd05a8 | 725 | if (!list_empty(&wb->b_more_io)) { |
7ccf19a8 | 726 | inode = wb_inode(wb->b_more_io.prev); |
028c2dd1 | 727 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
250df6ed | 728 | spin_lock(&inode->i_lock); |
71fd05a8 | 729 | inode_wait_for_writeback(inode); |
250df6ed | 730 | spin_unlock(&inode->i_lock); |
03ba3782 | 731 | } |
a66979ab | 732 | spin_unlock(&inode_wb_list_lock); |
03ba3782 JA |
733 | } |
734 | ||
735 | return wrote; | |
736 | } | |
737 | ||
738 | /* | |
83ba7b07 | 739 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 740 | */ |
83ba7b07 | 741 | static struct wb_writeback_work * |
08852b6d | 742 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 743 | { |
83ba7b07 | 744 | struct wb_writeback_work *work = NULL; |
03ba3782 | 745 | |
6467716a | 746 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
747 | if (!list_empty(&bdi->work_list)) { |
748 | work = list_entry(bdi->work_list.next, | |
749 | struct wb_writeback_work, list); | |
750 | list_del_init(&work->list); | |
03ba3782 | 751 | } |
6467716a | 752 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 753 | return work; |
03ba3782 JA |
754 | } |
755 | ||
cdf01dd5 LT |
756 | /* |
757 | * Add in the number of potentially dirty inodes, because each inode | |
758 | * write can dirty pagecache in the underlying blockdev. | |
759 | */ | |
760 | static unsigned long get_nr_dirty_pages(void) | |
761 | { | |
762 | return global_page_state(NR_FILE_DIRTY) + | |
763 | global_page_state(NR_UNSTABLE_NFS) + | |
764 | get_nr_dirty_inodes(); | |
765 | } | |
766 | ||
6585027a JK |
767 | static long wb_check_background_flush(struct bdi_writeback *wb) |
768 | { | |
769 | if (over_bground_thresh()) { | |
770 | ||
771 | struct wb_writeback_work work = { | |
772 | .nr_pages = LONG_MAX, | |
773 | .sync_mode = WB_SYNC_NONE, | |
774 | .for_background = 1, | |
775 | .range_cyclic = 1, | |
776 | }; | |
777 | ||
778 | return wb_writeback(wb, &work); | |
779 | } | |
780 | ||
781 | return 0; | |
782 | } | |
783 | ||
03ba3782 JA |
784 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
785 | { | |
786 | unsigned long expired; | |
787 | long nr_pages; | |
788 | ||
69b62d01 JA |
789 | /* |
790 | * When set to zero, disable periodic writeback | |
791 | */ | |
792 | if (!dirty_writeback_interval) | |
793 | return 0; | |
794 | ||
03ba3782 JA |
795 | expired = wb->last_old_flush + |
796 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
797 | if (time_before(jiffies, expired)) | |
798 | return 0; | |
799 | ||
800 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 801 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 802 | |
c4a77a6c | 803 | if (nr_pages) { |
83ba7b07 | 804 | struct wb_writeback_work work = { |
c4a77a6c JA |
805 | .nr_pages = nr_pages, |
806 | .sync_mode = WB_SYNC_NONE, | |
807 | .for_kupdate = 1, | |
808 | .range_cyclic = 1, | |
809 | }; | |
810 | ||
83ba7b07 | 811 | return wb_writeback(wb, &work); |
c4a77a6c | 812 | } |
03ba3782 JA |
813 | |
814 | return 0; | |
815 | } | |
816 | ||
817 | /* | |
818 | * Retrieve work items and do the writeback they describe | |
819 | */ | |
820 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
821 | { | |
822 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 823 | struct wb_writeback_work *work; |
c4a77a6c | 824 | long wrote = 0; |
03ba3782 | 825 | |
81d73a32 | 826 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 827 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
828 | /* |
829 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 830 | * because this thread is exiting now. |
03ba3782 JA |
831 | */ |
832 | if (force_wait) | |
83ba7b07 | 833 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 834 | |
455b2864 DC |
835 | trace_writeback_exec(bdi, work); |
836 | ||
83ba7b07 | 837 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
838 | |
839 | /* | |
83ba7b07 CH |
840 | * Notify the caller of completion if this is a synchronous |
841 | * work item, otherwise just free it. | |
03ba3782 | 842 | */ |
83ba7b07 CH |
843 | if (work->done) |
844 | complete(work->done); | |
845 | else | |
846 | kfree(work); | |
03ba3782 JA |
847 | } |
848 | ||
849 | /* | |
850 | * Check for periodic writeback, kupdated() style | |
851 | */ | |
852 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 853 | wrote += wb_check_background_flush(wb); |
81d73a32 | 854 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
855 | |
856 | return wrote; | |
857 | } | |
858 | ||
859 | /* | |
860 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
861 | * wakes up periodically and does kupdated style flushing. | |
862 | */ | |
08243900 | 863 | int bdi_writeback_thread(void *data) |
03ba3782 | 864 | { |
08243900 CH |
865 | struct bdi_writeback *wb = data; |
866 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
867 | long pages_written; |
868 | ||
766f9164 | 869 | current->flags |= PF_SWAPWRITE; |
08243900 | 870 | set_freezable(); |
ecd58403 | 871 | wb->last_active = jiffies; |
08243900 CH |
872 | |
873 | /* | |
874 | * Our parent may run at a different priority, just set us to normal | |
875 | */ | |
876 | set_user_nice(current, 0); | |
877 | ||
455b2864 DC |
878 | trace_writeback_thread_start(bdi); |
879 | ||
03ba3782 | 880 | while (!kthread_should_stop()) { |
6467716a AB |
881 | /* |
882 | * Remove own delayed wake-up timer, since we are already awake | |
883 | * and we'll take care of the preriodic write-back. | |
884 | */ | |
885 | del_timer(&wb->wakeup_timer); | |
886 | ||
03ba3782 JA |
887 | pages_written = wb_do_writeback(wb, 0); |
888 | ||
455b2864 DC |
889 | trace_writeback_pages_written(pages_written); |
890 | ||
03ba3782 | 891 | if (pages_written) |
ecd58403 | 892 | wb->last_active = jiffies; |
03ba3782 | 893 | |
297252c8 | 894 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 895 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 896 | __set_current_state(TASK_RUNNING); |
297252c8 | 897 | continue; |
03ba3782 JA |
898 | } |
899 | ||
253c34e9 | 900 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 901 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
902 | else { |
903 | /* | |
904 | * We have nothing to do, so can go sleep without any | |
905 | * timeout and save power. When a work is queued or | |
906 | * something is made dirty - we will be woken up. | |
907 | */ | |
297252c8 | 908 | schedule(); |
f9eadbbd | 909 | } |
69b62d01 | 910 | |
03ba3782 JA |
911 | try_to_freeze(); |
912 | } | |
913 | ||
fff5b85a | 914 | /* Flush any work that raced with us exiting */ |
08243900 CH |
915 | if (!list_empty(&bdi->work_list)) |
916 | wb_do_writeback(wb, 1); | |
455b2864 DC |
917 | |
918 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
919 | return 0; |
920 | } | |
921 | ||
08243900 | 922 | |
03ba3782 | 923 | /* |
b8c2f347 CH |
924 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
925 | * the whole world. | |
03ba3782 | 926 | */ |
b8c2f347 | 927 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 928 | { |
b8c2f347 | 929 | struct backing_dev_info *bdi; |
03ba3782 | 930 | |
83ba7b07 CH |
931 | if (!nr_pages) { |
932 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
933 | global_page_state(NR_UNSTABLE_NFS); |
934 | } | |
03ba3782 | 935 | |
b8c2f347 | 936 | rcu_read_lock(); |
cfc4ba53 | 937 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
938 | if (!bdi_has_dirty_io(bdi)) |
939 | continue; | |
6585027a | 940 | __bdi_start_writeback(bdi, nr_pages, false); |
03ba3782 | 941 | } |
cfc4ba53 | 942 | rcu_read_unlock(); |
1da177e4 LT |
943 | } |
944 | ||
03ba3782 JA |
945 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
946 | { | |
947 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
948 | struct dentry *dentry; | |
949 | const char *name = "?"; | |
950 | ||
951 | dentry = d_find_alias(inode); | |
952 | if (dentry) { | |
953 | spin_lock(&dentry->d_lock); | |
954 | name = (const char *) dentry->d_name.name; | |
955 | } | |
956 | printk(KERN_DEBUG | |
957 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
958 | current->comm, task_pid_nr(current), inode->i_ino, | |
959 | name, inode->i_sb->s_id); | |
960 | if (dentry) { | |
961 | spin_unlock(&dentry->d_lock); | |
962 | dput(dentry); | |
963 | } | |
964 | } | |
965 | } | |
966 | ||
967 | /** | |
968 | * __mark_inode_dirty - internal function | |
969 | * @inode: inode to mark | |
970 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
971 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
972 | * mark_inode_dirty_sync. | |
1da177e4 | 973 | * |
03ba3782 JA |
974 | * Put the inode on the super block's dirty list. |
975 | * | |
976 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
977 | * dirty list only if it is hashed or if it refers to a blockdev. | |
978 | * If it was not hashed, it will never be added to the dirty list | |
979 | * even if it is later hashed, as it will have been marked dirty already. | |
980 | * | |
981 | * In short, make sure you hash any inodes _before_ you start marking | |
982 | * them dirty. | |
1da177e4 | 983 | * |
03ba3782 JA |
984 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
985 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
986 | * the kernel-internal blockdev inode represents the dirtying time of the | |
987 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
988 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
989 | * blockdev inode. | |
1da177e4 | 990 | */ |
03ba3782 | 991 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 992 | { |
03ba3782 | 993 | struct super_block *sb = inode->i_sb; |
253c34e9 | 994 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 995 | |
03ba3782 JA |
996 | /* |
997 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
998 | * dirty the inode itself | |
999 | */ | |
1000 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1001 | if (sb->s_op->dirty_inode) | |
aa385729 | 1002 | sb->s_op->dirty_inode(inode, flags); |
03ba3782 JA |
1003 | } |
1004 | ||
1005 | /* | |
1006 | * make sure that changes are seen by all cpus before we test i_state | |
1007 | * -- mikulas | |
1008 | */ | |
1009 | smp_mb(); | |
1010 | ||
1011 | /* avoid the locking if we can */ | |
1012 | if ((inode->i_state & flags) == flags) | |
1013 | return; | |
1014 | ||
1015 | if (unlikely(block_dump)) | |
1016 | block_dump___mark_inode_dirty(inode); | |
1017 | ||
250df6ed | 1018 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1019 | if ((inode->i_state & flags) != flags) { |
1020 | const int was_dirty = inode->i_state & I_DIRTY; | |
1021 | ||
1022 | inode->i_state |= flags; | |
1023 | ||
1024 | /* | |
1025 | * If the inode is being synced, just update its dirty state. | |
1026 | * The unlocker will place the inode on the appropriate | |
1027 | * superblock list, based upon its state. | |
1028 | */ | |
1029 | if (inode->i_state & I_SYNC) | |
250df6ed | 1030 | goto out_unlock_inode; |
03ba3782 JA |
1031 | |
1032 | /* | |
1033 | * Only add valid (hashed) inodes to the superblock's | |
1034 | * dirty list. Add blockdev inodes as well. | |
1035 | */ | |
1036 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1037 | if (inode_unhashed(inode)) |
250df6ed | 1038 | goto out_unlock_inode; |
03ba3782 | 1039 | } |
a4ffdde6 | 1040 | if (inode->i_state & I_FREEING) |
250df6ed | 1041 | goto out_unlock_inode; |
03ba3782 JA |
1042 | |
1043 | /* | |
1044 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1045 | * reposition it (that would break b_dirty time-ordering). | |
1046 | */ | |
1047 | if (!was_dirty) { | |
a66979ab | 1048 | bool wakeup_bdi = false; |
253c34e9 AB |
1049 | bdi = inode_to_bdi(inode); |
1050 | ||
1051 | if (bdi_cap_writeback_dirty(bdi)) { | |
1052 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1053 | "bdi-%s not registered\n", bdi->name); | |
1054 | ||
1055 | /* | |
1056 | * If this is the first dirty inode for this | |
1057 | * bdi, we have to wake-up the corresponding | |
1058 | * bdi thread to make sure background | |
1059 | * write-back happens later. | |
1060 | */ | |
1061 | if (!wb_has_dirty_io(&bdi->wb)) | |
1062 | wakeup_bdi = true; | |
500b067c | 1063 | } |
03ba3782 | 1064 | |
a66979ab DC |
1065 | spin_unlock(&inode->i_lock); |
1066 | spin_lock(&inode_wb_list_lock); | |
03ba3782 | 1067 | inode->dirtied_when = jiffies; |
7ccf19a8 | 1068 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
a66979ab DC |
1069 | spin_unlock(&inode_wb_list_lock); |
1070 | ||
1071 | if (wakeup_bdi) | |
1072 | bdi_wakeup_thread_delayed(bdi); | |
1073 | return; | |
1da177e4 | 1074 | } |
1da177e4 | 1075 | } |
250df6ed DC |
1076 | out_unlock_inode: |
1077 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1078 | |
03ba3782 JA |
1079 | } |
1080 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1081 | ||
1082 | /* | |
1083 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1084 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1085 | * | |
1086 | * If older_than_this is non-NULL, then only write out inodes which | |
1087 | * had their first dirtying at a time earlier than *older_than_this. | |
1088 | * | |
03ba3782 JA |
1089 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1090 | * This function assumes that the blockdev superblock's inodes are backed by | |
1091 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1092 | * assume that all inodes are backed by the same queue. | |
1093 | * | |
1094 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1095 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1096 | * on the writer throttling path, and we get decent balancing between many | |
1097 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1098 | */ | |
b6e51316 | 1099 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1100 | { |
1101 | struct inode *inode, *old_inode = NULL; | |
1102 | ||
1103 | /* | |
1104 | * We need to be protected against the filesystem going from | |
1105 | * r/o to r/w or vice versa. | |
1106 | */ | |
b6e51316 | 1107 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1108 | |
55fa6091 | 1109 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1110 | |
1111 | /* | |
1112 | * Data integrity sync. Must wait for all pages under writeback, | |
1113 | * because there may have been pages dirtied before our sync | |
1114 | * call, but which had writeout started before we write it out. | |
1115 | * In which case, the inode may not be on the dirty list, but | |
1116 | * we still have to wait for that writeout. | |
1117 | */ | |
b6e51316 | 1118 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1119 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1120 | |
250df6ed DC |
1121 | spin_lock(&inode->i_lock); |
1122 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1123 | (mapping->nrpages == 0)) { | |
1124 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1125 | continue; |
250df6ed | 1126 | } |
03ba3782 | 1127 | __iget(inode); |
250df6ed | 1128 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1129 | spin_unlock(&inode_sb_list_lock); |
1130 | ||
03ba3782 | 1131 | /* |
55fa6091 DC |
1132 | * We hold a reference to 'inode' so it couldn't have been |
1133 | * removed from s_inodes list while we dropped the | |
1134 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1135 | * be holding the last reference and we cannot iput it under | |
1136 | * inode_sb_list_lock. So we keep the reference and iput it | |
1137 | * later. | |
03ba3782 JA |
1138 | */ |
1139 | iput(old_inode); | |
1140 | old_inode = inode; | |
1141 | ||
1142 | filemap_fdatawait(mapping); | |
1143 | ||
1144 | cond_resched(); | |
1145 | ||
55fa6091 | 1146 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1147 | } |
55fa6091 | 1148 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1149 | iput(old_inode); |
1da177e4 LT |
1150 | } |
1151 | ||
d8a8559c | 1152 | /** |
3259f8be | 1153 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1154 | * @sb: the superblock |
3259f8be | 1155 | * @nr: the number of pages to write |
1da177e4 | 1156 | * |
d8a8559c JA |
1157 | * Start writeback on some inodes on this super_block. No guarantees are made |
1158 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1159 | * for IO completion of submitted IO. |
1da177e4 | 1160 | */ |
3259f8be | 1161 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
1da177e4 | 1162 | { |
83ba7b07 CH |
1163 | DECLARE_COMPLETION_ONSTACK(done); |
1164 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1165 | .sb = sb, |
1166 | .sync_mode = WB_SYNC_NONE, | |
83ba7b07 | 1167 | .done = &done, |
3259f8be | 1168 | .nr_pages = nr, |
3c4d7165 | 1169 | }; |
d8a8559c | 1170 | |
cf37e972 | 1171 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1172 | bdi_queue_work(sb->s_bdi, &work); |
1173 | wait_for_completion(&done); | |
e913fc82 | 1174 | } |
3259f8be CM |
1175 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1176 | ||
1177 | /** | |
1178 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1179 | * @sb: the superblock | |
1180 | * | |
1181 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1182 | * on how many (if any) will be written, and this function does not wait | |
1183 | * for IO completion of submitted IO. | |
1184 | */ | |
1185 | void writeback_inodes_sb(struct super_block *sb) | |
1186 | { | |
925d169f | 1187 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
3259f8be | 1188 | } |
0e3c9a22 | 1189 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1190 | |
17bd55d0 ES |
1191 | /** |
1192 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1193 | * @sb: the superblock | |
1194 | * | |
1195 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1196 | * Returns 1 if writeback was started, 0 if not. | |
1197 | */ | |
1198 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1199 | { | |
1200 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1201 | down_read(&sb->s_umount); |
17bd55d0 | 1202 | writeback_inodes_sb(sb); |
cf37e972 | 1203 | up_read(&sb->s_umount); |
17bd55d0 ES |
1204 | return 1; |
1205 | } else | |
1206 | return 0; | |
1207 | } | |
1208 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1209 | ||
3259f8be CM |
1210 | /** |
1211 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1212 | * @sb: the superblock | |
1213 | * @nr: the number of pages to write | |
1214 | * | |
1215 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1216 | * Returns 1 if writeback was started, 0 if not. | |
1217 | */ | |
1218 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
1219 | unsigned long nr) | |
1220 | { | |
1221 | if (!writeback_in_progress(sb->s_bdi)) { | |
1222 | down_read(&sb->s_umount); | |
1223 | writeback_inodes_sb_nr(sb, nr); | |
1224 | up_read(&sb->s_umount); | |
1225 | return 1; | |
1226 | } else | |
1227 | return 0; | |
1228 | } | |
1229 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1230 | ||
d8a8559c JA |
1231 | /** |
1232 | * sync_inodes_sb - sync sb inode pages | |
1233 | * @sb: the superblock | |
1234 | * | |
1235 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1236 | * super_block. |
d8a8559c | 1237 | */ |
b6e51316 | 1238 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1239 | { |
83ba7b07 CH |
1240 | DECLARE_COMPLETION_ONSTACK(done); |
1241 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1242 | .sb = sb, |
1243 | .sync_mode = WB_SYNC_ALL, | |
1244 | .nr_pages = LONG_MAX, | |
1245 | .range_cyclic = 0, | |
83ba7b07 | 1246 | .done = &done, |
3c4d7165 CH |
1247 | }; |
1248 | ||
cf37e972 CH |
1249 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1250 | ||
83ba7b07 CH |
1251 | bdi_queue_work(sb->s_bdi, &work); |
1252 | wait_for_completion(&done); | |
1253 | ||
b6e51316 | 1254 | wait_sb_inodes(sb); |
1da177e4 | 1255 | } |
d8a8559c | 1256 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1257 | |
1da177e4 | 1258 | /** |
7f04c26d AA |
1259 | * write_inode_now - write an inode to disk |
1260 | * @inode: inode to write to disk | |
1261 | * @sync: whether the write should be synchronous or not | |
1262 | * | |
1263 | * This function commits an inode to disk immediately if it is dirty. This is | |
1264 | * primarily needed by knfsd. | |
1da177e4 | 1265 | * |
7f04c26d | 1266 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1267 | */ |
1da177e4 LT |
1268 | int write_inode_now(struct inode *inode, int sync) |
1269 | { | |
1270 | int ret; | |
1271 | struct writeback_control wbc = { | |
1272 | .nr_to_write = LONG_MAX, | |
18914b18 | 1273 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1274 | .range_start = 0, |
1275 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1276 | }; |
1277 | ||
1278 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1279 | wbc.nr_to_write = 0; |
1da177e4 LT |
1280 | |
1281 | might_sleep(); | |
a66979ab | 1282 | spin_lock(&inode_wb_list_lock); |
0f1b1fd8 | 1283 | spin_lock(&inode->i_lock); |
01c03194 | 1284 | ret = writeback_single_inode(inode, &wbc); |
0f1b1fd8 | 1285 | spin_unlock(&inode->i_lock); |
a66979ab | 1286 | spin_unlock(&inode_wb_list_lock); |
1da177e4 | 1287 | if (sync) |
1c0eeaf5 | 1288 | inode_sync_wait(inode); |
1da177e4 LT |
1289 | return ret; |
1290 | } | |
1291 | EXPORT_SYMBOL(write_inode_now); | |
1292 | ||
1293 | /** | |
1294 | * sync_inode - write an inode and its pages to disk. | |
1295 | * @inode: the inode to sync | |
1296 | * @wbc: controls the writeback mode | |
1297 | * | |
1298 | * sync_inode() will write an inode and its pages to disk. It will also | |
1299 | * correctly update the inode on its superblock's dirty inode lists and will | |
1300 | * update inode->i_state. | |
1301 | * | |
1302 | * The caller must have a ref on the inode. | |
1303 | */ | |
1304 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1305 | { | |
1306 | int ret; | |
1307 | ||
a66979ab | 1308 | spin_lock(&inode_wb_list_lock); |
0f1b1fd8 | 1309 | spin_lock(&inode->i_lock); |
01c03194 | 1310 | ret = writeback_single_inode(inode, wbc); |
0f1b1fd8 | 1311 | spin_unlock(&inode->i_lock); |
a66979ab | 1312 | spin_unlock(&inode_wb_list_lock); |
1da177e4 LT |
1313 | return ret; |
1314 | } | |
1315 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1316 | |
1317 | /** | |
c691b9d9 | 1318 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1319 | * @inode: the inode to sync |
1320 | * @wait: wait for I/O to complete. | |
1321 | * | |
c691b9d9 | 1322 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1323 | * |
1324 | * Note: only writes the actual inode, no associated data or other metadata. | |
1325 | */ | |
1326 | int sync_inode_metadata(struct inode *inode, int wait) | |
1327 | { | |
1328 | struct writeback_control wbc = { | |
1329 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1330 | .nr_to_write = 0, /* metadata-only */ | |
1331 | }; | |
1332 | ||
1333 | return sync_inode(inode, &wbc); | |
1334 | } | |
1335 | EXPORT_SYMBOL(sync_inode_metadata); |