]>
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 | ||
83ba7b07 CH |
82 | static void bdi_queue_work(struct backing_dev_info *bdi, |
83 | struct wb_writeback_work *work) | |
03ba3782 | 84 | { |
455b2864 | 85 | trace_writeback_queue(bdi, work); |
03ba3782 | 86 | |
6467716a | 87 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 | 88 | list_add_tail(&work->list, &bdi->work_list); |
fff5b85a AB |
89 | if (bdi->wb.task) { |
90 | wake_up_process(bdi->wb.task); | |
91 | } else { | |
92 | /* | |
93 | * The bdi thread isn't there, wake up the forker thread which | |
94 | * will create and run it. | |
95 | */ | |
455b2864 | 96 | trace_writeback_nothread(bdi, work); |
03ba3782 | 97 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 98 | } |
6467716a | 99 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
100 | } |
101 | ||
83ba7b07 CH |
102 | static void |
103 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
104 | bool range_cyclic, bool for_background) | |
1da177e4 | 105 | { |
83ba7b07 | 106 | struct wb_writeback_work *work; |
03ba3782 | 107 | |
bcddc3f0 JA |
108 | /* |
109 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
110 | * wakeup the thread for old dirty data writeback | |
111 | */ | |
83ba7b07 CH |
112 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
113 | if (!work) { | |
455b2864 DC |
114 | if (bdi->wb.task) { |
115 | trace_writeback_nowork(bdi); | |
83ba7b07 | 116 | wake_up_process(bdi->wb.task); |
455b2864 | 117 | } |
83ba7b07 | 118 | return; |
bcddc3f0 | 119 | } |
03ba3782 | 120 | |
83ba7b07 CH |
121 | work->sync_mode = WB_SYNC_NONE; |
122 | work->nr_pages = nr_pages; | |
123 | work->range_cyclic = range_cyclic; | |
124 | work->for_background = for_background; | |
03ba3782 | 125 | |
83ba7b07 | 126 | bdi_queue_work(bdi, work); |
b6e51316 JA |
127 | } |
128 | ||
129 | /** | |
130 | * bdi_start_writeback - start writeback | |
131 | * @bdi: the backing device to write from | |
132 | * @nr_pages: the number of pages to write | |
133 | * | |
134 | * Description: | |
135 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
136 | * started when this function returns, we make no guarentees on | |
0e3c9a22 | 137 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
138 | * |
139 | */ | |
c5444198 | 140 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 141 | { |
83ba7b07 | 142 | __bdi_start_writeback(bdi, nr_pages, true, false); |
c5444198 | 143 | } |
d3ddec76 | 144 | |
c5444198 CH |
145 | /** |
146 | * bdi_start_background_writeback - start background writeback | |
147 | * @bdi: the backing device to write from | |
148 | * | |
149 | * Description: | |
150 | * This does WB_SYNC_NONE background writeback. The IO is only | |
151 | * started when this function returns, we make no guarentees on | |
152 | * completion. Caller need not hold sb s_umount semaphore. | |
153 | */ | |
154 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
155 | { | |
83ba7b07 | 156 | __bdi_start_writeback(bdi, LONG_MAX, true, true); |
1da177e4 LT |
157 | } |
158 | ||
6610a0bc AM |
159 | /* |
160 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
161 | * furthest end of its superblock's dirty-inode list. | |
162 | * | |
163 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 164 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
165 | * the case then the inode must have been redirtied while it was being written |
166 | * out and we don't reset its dirtied_when. | |
167 | */ | |
168 | static void redirty_tail(struct inode *inode) | |
169 | { | |
03ba3782 | 170 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 171 | |
03ba3782 | 172 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 173 | struct inode *tail; |
6610a0bc | 174 | |
03ba3782 | 175 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 176 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
177 | inode->dirtied_when = jiffies; |
178 | } | |
03ba3782 | 179 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
180 | } |
181 | ||
c986d1e2 | 182 | /* |
66f3b8e2 | 183 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 184 | */ |
0e0f4fc2 | 185 | static void requeue_io(struct inode *inode) |
c986d1e2 | 186 | { |
03ba3782 JA |
187 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
188 | ||
189 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
190 | } |
191 | ||
1c0eeaf5 JE |
192 | static void inode_sync_complete(struct inode *inode) |
193 | { | |
194 | /* | |
195 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
196 | */ | |
197 | smp_mb(); | |
198 | wake_up_bit(&inode->i_state, __I_SYNC); | |
199 | } | |
200 | ||
d2caa3c5 JL |
201 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
202 | { | |
203 | bool ret = time_after(inode->dirtied_when, t); | |
204 | #ifndef CONFIG_64BIT | |
205 | /* | |
206 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
207 | * It _appears_ to be in the future, but is actually in distant past. | |
208 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 209 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
210 | */ |
211 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
212 | #endif | |
213 | return ret; | |
214 | } | |
215 | ||
2c136579 FW |
216 | /* |
217 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
218 | */ | |
219 | static void move_expired_inodes(struct list_head *delaying_queue, | |
220 | struct list_head *dispatch_queue, | |
221 | unsigned long *older_than_this) | |
222 | { | |
5c03449d SL |
223 | LIST_HEAD(tmp); |
224 | struct list_head *pos, *node; | |
cf137307 | 225 | struct super_block *sb = NULL; |
5c03449d | 226 | struct inode *inode; |
cf137307 | 227 | int do_sb_sort = 0; |
5c03449d | 228 | |
2c136579 | 229 | while (!list_empty(delaying_queue)) { |
5c03449d | 230 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
2c136579 | 231 | if (older_than_this && |
d2caa3c5 | 232 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 233 | break; |
cf137307 JA |
234 | if (sb && sb != inode->i_sb) |
235 | do_sb_sort = 1; | |
236 | sb = inode->i_sb; | |
5c03449d SL |
237 | list_move(&inode->i_list, &tmp); |
238 | } | |
239 | ||
cf137307 JA |
240 | /* just one sb in list, splice to dispatch_queue and we're done */ |
241 | if (!do_sb_sort) { | |
242 | list_splice(&tmp, dispatch_queue); | |
243 | return; | |
244 | } | |
245 | ||
5c03449d SL |
246 | /* Move inodes from one superblock together */ |
247 | while (!list_empty(&tmp)) { | |
248 | inode = list_entry(tmp.prev, struct inode, i_list); | |
249 | sb = inode->i_sb; | |
250 | list_for_each_prev_safe(pos, node, &tmp) { | |
251 | inode = list_entry(pos, struct inode, i_list); | |
252 | if (inode->i_sb == sb) | |
253 | list_move(&inode->i_list, dispatch_queue); | |
254 | } | |
2c136579 FW |
255 | } |
256 | } | |
257 | ||
258 | /* | |
259 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
260 | * Before |
261 | * newly dirtied b_dirty b_io b_more_io | |
262 | * =============> gf edc BA | |
263 | * After | |
264 | * newly dirtied b_dirty b_io b_more_io | |
265 | * =============> g fBAedc | |
266 | * | | |
267 | * +--> dequeue for IO | |
2c136579 | 268 | */ |
03ba3782 | 269 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 270 | { |
4ea879b9 | 271 | list_splice_init(&wb->b_more_io, &wb->b_io); |
03ba3782 | 272 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
66f3b8e2 JA |
273 | } |
274 | ||
a9185b41 | 275 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 276 | { |
03ba3782 | 277 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 278 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 279 | return 0; |
08d8e974 | 280 | } |
08d8e974 | 281 | |
1da177e4 | 282 | /* |
01c03194 CH |
283 | * Wait for writeback on an inode to complete. |
284 | */ | |
285 | static void inode_wait_for_writeback(struct inode *inode) | |
286 | { | |
287 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
288 | wait_queue_head_t *wqh; | |
289 | ||
290 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
58a9d3d8 | 291 | while (inode->i_state & I_SYNC) { |
01c03194 CH |
292 | spin_unlock(&inode_lock); |
293 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
294 | spin_lock(&inode_lock); | |
58a9d3d8 | 295 | } |
01c03194 CH |
296 | } |
297 | ||
298 | /* | |
299 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
300 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
301 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
302 | * | |
1da177e4 LT |
303 | * If `wait' is set, wait on the writeout. |
304 | * | |
305 | * The whole writeout design is quite complex and fragile. We want to avoid | |
306 | * starvation of particular inodes when others are being redirtied, prevent | |
307 | * livelocks, etc. | |
308 | * | |
309 | * Called under inode_lock. | |
310 | */ | |
311 | static int | |
01c03194 | 312 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 313 | { |
1da177e4 | 314 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 315 | unsigned dirty; |
1da177e4 LT |
316 | int ret; |
317 | ||
01c03194 CH |
318 | if (!atomic_read(&inode->i_count)) |
319 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
320 | else | |
321 | WARN_ON(inode->i_state & I_WILL_FREE); | |
322 | ||
323 | if (inode->i_state & I_SYNC) { | |
324 | /* | |
325 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 326 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
327 | * writeback can proceed with the other inodes on s_io. |
328 | * | |
329 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 330 | * completed a full scan of b_io. |
01c03194 | 331 | */ |
a9185b41 | 332 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
333 | requeue_io(inode); |
334 | return 0; | |
335 | } | |
336 | ||
337 | /* | |
338 | * It's a data-integrity sync. We must wait. | |
339 | */ | |
340 | inode_wait_for_writeback(inode); | |
341 | } | |
342 | ||
1c0eeaf5 | 343 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 344 | |
5547e8aa | 345 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 346 | inode->i_state |= I_SYNC; |
5547e8aa | 347 | inode->i_state &= ~I_DIRTY_PAGES; |
1da177e4 LT |
348 | spin_unlock(&inode_lock); |
349 | ||
350 | ret = do_writepages(mapping, wbc); | |
351 | ||
26821ed4 CH |
352 | /* |
353 | * Make sure to wait on the data before writing out the metadata. | |
354 | * This is important for filesystems that modify metadata on data | |
355 | * I/O completion. | |
356 | */ | |
a9185b41 | 357 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 358 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
359 | if (ret == 0) |
360 | ret = err; | |
361 | } | |
362 | ||
5547e8aa DM |
363 | /* |
364 | * Some filesystems may redirty the inode during the writeback | |
365 | * due to delalloc, clear dirty metadata flags right before | |
366 | * write_inode() | |
367 | */ | |
368 | spin_lock(&inode_lock); | |
369 | dirty = inode->i_state & I_DIRTY; | |
370 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
371 | spin_unlock(&inode_lock); | |
26821ed4 CH |
372 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
373 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 374 | int err = write_inode(inode, wbc); |
1da177e4 LT |
375 | if (ret == 0) |
376 | ret = err; | |
377 | } | |
378 | ||
379 | spin_lock(&inode_lock); | |
1c0eeaf5 | 380 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 381 | if (!(inode->i_state & I_FREEING)) { |
23539afc | 382 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
383 | /* |
384 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 385 | * sometimes bales out without doing anything. |
1b43ef91 | 386 | */ |
a50aeb40 WF |
387 | inode->i_state |= I_DIRTY_PAGES; |
388 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 389 | /* |
a50aeb40 | 390 | * slice used up: queue for next turn |
1da177e4 | 391 | */ |
a50aeb40 | 392 | requeue_io(inode); |
1da177e4 LT |
393 | } else { |
394 | /* | |
a50aeb40 WF |
395 | * Writeback blocked by something other than |
396 | * congestion. Delay the inode for some time to | |
397 | * avoid spinning on the CPU (100% iowait) | |
398 | * retrying writeback of the dirty page/inode | |
399 | * that cannot be performed immediately. | |
1da177e4 | 400 | */ |
1b43ef91 | 401 | redirty_tail(inode); |
1da177e4 | 402 | } |
23539afc WF |
403 | } else if (inode->i_state & I_DIRTY) { |
404 | /* | |
405 | * Filesystems can dirty the inode during writeback | |
406 | * operations, such as delayed allocation during | |
407 | * submission or metadata updates after data IO | |
408 | * completion. | |
409 | */ | |
410 | redirty_tail(inode); | |
1da177e4 LT |
411 | } else { |
412 | /* | |
9e38d86f NP |
413 | * The inode is clean. At this point we either have |
414 | * a reference to the inode or it's on it's way out. | |
415 | * No need to add it back to the LRU. | |
1da177e4 | 416 | */ |
9e38d86f | 417 | list_del_init(&inode->i_list); |
1da177e4 LT |
418 | } |
419 | } | |
1c0eeaf5 | 420 | inode_sync_complete(inode); |
1da177e4 LT |
421 | return ret; |
422 | } | |
423 | ||
03ba3782 | 424 | /* |
d19de7ed | 425 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
426 | * before calling writeback. So make sure that we do pin it, so it doesn't |
427 | * go away while we are writing inodes from it. | |
03ba3782 | 428 | */ |
d19de7ed | 429 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 430 | { |
03ba3782 | 431 | spin_lock(&sb_lock); |
29cb4859 CH |
432 | if (list_empty(&sb->s_instances)) { |
433 | spin_unlock(&sb_lock); | |
434 | return false; | |
435 | } | |
436 | ||
03ba3782 | 437 | sb->s_count++; |
29cb4859 CH |
438 | spin_unlock(&sb_lock); |
439 | ||
03ba3782 | 440 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 441 | if (sb->s_root) |
d19de7ed | 442 | return true; |
03ba3782 JA |
443 | up_read(&sb->s_umount); |
444 | } | |
29cb4859 CH |
445 | |
446 | put_super(sb); | |
d19de7ed | 447 | return false; |
03ba3782 JA |
448 | } |
449 | ||
f11c9c5c ES |
450 | /* |
451 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
452 | * |
453 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
454 | * inodes. Otherwise write only ones which go sequentially |
455 | * in reverse order. | |
edadfb10 | 456 | * |
f11c9c5c ES |
457 | * Return 1, if the caller writeback routine should be |
458 | * interrupted. Otherwise return 0. | |
459 | */ | |
edadfb10 CH |
460 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
461 | struct writeback_control *wbc, bool only_this_sb) | |
1da177e4 | 462 | { |
03ba3782 | 463 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 464 | long pages_skipped; |
f11c9c5c ES |
465 | struct inode *inode = list_entry(wb->b_io.prev, |
466 | struct inode, i_list); | |
edadfb10 CH |
467 | |
468 | if (inode->i_sb != sb) { | |
469 | if (only_this_sb) { | |
470 | /* | |
471 | * We only want to write back data for this | |
472 | * superblock, move all inodes not belonging | |
473 | * to it back onto the dirty list. | |
474 | */ | |
475 | redirty_tail(inode); | |
476 | continue; | |
477 | } | |
478 | ||
479 | /* | |
480 | * The inode belongs to a different superblock. | |
481 | * Bounce back to the caller to unpin this and | |
482 | * pin the next superblock. | |
483 | */ | |
f11c9c5c | 484 | return 0; |
edadfb10 CH |
485 | } |
486 | ||
84a89245 | 487 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
488 | requeue_io(inode); |
489 | continue; | |
490 | } | |
d2caa3c5 JL |
491 | /* |
492 | * Was this inode dirtied after sync_sb_inodes was called? | |
493 | * This keeps sync from extra jobs and livelock. | |
494 | */ | |
f11c9c5c ES |
495 | if (inode_dirtied_after(inode, wbc->wb_start)) |
496 | return 1; | |
1da177e4 | 497 | |
a4ffdde6 | 498 | BUG_ON(inode->i_state & I_FREEING); |
1da177e4 LT |
499 | __iget(inode); |
500 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 501 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
502 | if (wbc->pages_skipped != pages_skipped) { |
503 | /* | |
504 | * writeback is not making progress due to locked | |
505 | * buffers. Skip this inode for now. | |
506 | */ | |
f57b9b7b | 507 | redirty_tail(inode); |
1da177e4 LT |
508 | } |
509 | spin_unlock(&inode_lock); | |
1da177e4 | 510 | iput(inode); |
4ffc8444 | 511 | cond_resched(); |
1da177e4 | 512 | spin_lock(&inode_lock); |
8bc3be27 FW |
513 | if (wbc->nr_to_write <= 0) { |
514 | wbc->more_io = 1; | |
f11c9c5c | 515 | return 1; |
8bc3be27 | 516 | } |
03ba3782 | 517 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 518 | wbc->more_io = 1; |
1da177e4 | 519 | } |
f11c9c5c ES |
520 | /* b_io is empty */ |
521 | return 1; | |
522 | } | |
523 | ||
9c3a8ee8 CH |
524 | void writeback_inodes_wb(struct bdi_writeback *wb, |
525 | struct writeback_control *wbc) | |
f11c9c5c ES |
526 | { |
527 | int ret = 0; | |
528 | ||
7624ee72 JK |
529 | if (!wbc->wb_start) |
530 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
f11c9c5c ES |
531 | spin_lock(&inode_lock); |
532 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
533 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 534 | |
f11c9c5c ES |
535 | while (!list_empty(&wb->b_io)) { |
536 | struct inode *inode = list_entry(wb->b_io.prev, | |
537 | struct inode, i_list); | |
538 | struct super_block *sb = inode->i_sb; | |
9ecc2738 | 539 | |
edadfb10 CH |
540 | if (!pin_sb_for_writeback(sb)) { |
541 | requeue_io(inode); | |
542 | continue; | |
f11c9c5c | 543 | } |
edadfb10 CH |
544 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
545 | drop_super(sb); | |
f11c9c5c | 546 | |
f11c9c5c ES |
547 | if (ret) |
548 | break; | |
549 | } | |
66f3b8e2 JA |
550 | spin_unlock(&inode_lock); |
551 | /* Leave any unwritten inodes on b_io */ | |
552 | } | |
553 | ||
edadfb10 CH |
554 | static void __writeback_inodes_sb(struct super_block *sb, |
555 | struct bdi_writeback *wb, struct writeback_control *wbc) | |
556 | { | |
557 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
558 | ||
edadfb10 CH |
559 | spin_lock(&inode_lock); |
560 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
561 | queue_io(wb, wbc->older_than_this); | |
562 | writeback_sb_inodes(sb, wb, wbc, true); | |
563 | spin_unlock(&inode_lock); | |
564 | } | |
565 | ||
66f3b8e2 | 566 | /* |
03ba3782 JA |
567 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
568 | * operation. We do this so we don't hold I_SYNC against an inode for | |
569 | * enormous amounts of time, which would block a userspace task which has | |
570 | * been forced to throttle against that inode. Also, the code reevaluates | |
571 | * the dirty each time it has written this many pages. | |
572 | */ | |
573 | #define MAX_WRITEBACK_PAGES 1024 | |
574 | ||
575 | static inline bool over_bground_thresh(void) | |
576 | { | |
577 | unsigned long background_thresh, dirty_thresh; | |
578 | ||
16c4042f | 579 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 JA |
580 | |
581 | return (global_page_state(NR_FILE_DIRTY) + | |
582 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
583 | } | |
584 | ||
585 | /* | |
586 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 587 | * |
03ba3782 JA |
588 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
589 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
590 | * just walks the superblock inode list, writing back any inodes which are | |
591 | * older than a specific point in time. | |
66f3b8e2 | 592 | * |
03ba3782 JA |
593 | * Try to run once per dirty_writeback_interval. But if a writeback event |
594 | * takes longer than a dirty_writeback_interval interval, then leave a | |
595 | * one-second gap. | |
66f3b8e2 | 596 | * |
03ba3782 JA |
597 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
598 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 599 | */ |
c4a77a6c | 600 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 601 | struct wb_writeback_work *work) |
66f3b8e2 | 602 | { |
03ba3782 | 603 | struct writeback_control wbc = { |
83ba7b07 | 604 | .sync_mode = work->sync_mode, |
03ba3782 | 605 | .older_than_this = NULL, |
83ba7b07 CH |
606 | .for_kupdate = work->for_kupdate, |
607 | .for_background = work->for_background, | |
608 | .range_cyclic = work->range_cyclic, | |
03ba3782 JA |
609 | }; |
610 | unsigned long oldest_jif; | |
611 | long wrote = 0; | |
a5989bdc | 612 | struct inode *inode; |
66f3b8e2 | 613 | |
03ba3782 JA |
614 | if (wbc.for_kupdate) { |
615 | wbc.older_than_this = &oldest_jif; | |
616 | oldest_jif = jiffies - | |
617 | msecs_to_jiffies(dirty_expire_interval * 10); | |
618 | } | |
c4a77a6c JA |
619 | if (!wbc.range_cyclic) { |
620 | wbc.range_start = 0; | |
621 | wbc.range_end = LLONG_MAX; | |
622 | } | |
38f21977 | 623 | |
7624ee72 | 624 | wbc.wb_start = jiffies; /* livelock avoidance */ |
03ba3782 JA |
625 | for (;;) { |
626 | /* | |
d3ddec76 | 627 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 628 | */ |
83ba7b07 | 629 | if (work->nr_pages <= 0) |
03ba3782 | 630 | break; |
66f3b8e2 | 631 | |
38f21977 | 632 | /* |
d3ddec76 WF |
633 | * For background writeout, stop when we are below the |
634 | * background dirty threshold | |
38f21977 | 635 | */ |
83ba7b07 | 636 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 637 | break; |
38f21977 | 638 | |
03ba3782 | 639 | wbc.more_io = 0; |
03ba3782 JA |
640 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
641 | wbc.pages_skipped = 0; | |
028c2dd1 DC |
642 | |
643 | trace_wbc_writeback_start(&wbc, wb->bdi); | |
83ba7b07 CH |
644 | if (work->sb) |
645 | __writeback_inodes_sb(work->sb, wb, &wbc); | |
edadfb10 CH |
646 | else |
647 | writeback_inodes_wb(wb, &wbc); | |
028c2dd1 DC |
648 | trace_wbc_writeback_written(&wbc, wb->bdi); |
649 | ||
83ba7b07 | 650 | work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
651 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
652 | ||
653 | /* | |
71fd05a8 | 654 | * If we consumed everything, see if we have more |
03ba3782 | 655 | */ |
71fd05a8 JA |
656 | if (wbc.nr_to_write <= 0) |
657 | continue; | |
658 | /* | |
659 | * Didn't write everything and we don't have more IO, bail | |
660 | */ | |
661 | if (!wbc.more_io) | |
03ba3782 | 662 | break; |
71fd05a8 JA |
663 | /* |
664 | * Did we write something? Try for more | |
665 | */ | |
666 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
667 | continue; | |
668 | /* | |
669 | * Nothing written. Wait for some inode to | |
670 | * become available for writeback. Otherwise | |
671 | * we'll just busyloop. | |
672 | */ | |
673 | spin_lock(&inode_lock); | |
674 | if (!list_empty(&wb->b_more_io)) { | |
675 | inode = list_entry(wb->b_more_io.prev, | |
676 | struct inode, i_list); | |
028c2dd1 | 677 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
71fd05a8 | 678 | inode_wait_for_writeback(inode); |
03ba3782 | 679 | } |
71fd05a8 | 680 | spin_unlock(&inode_lock); |
03ba3782 JA |
681 | } |
682 | ||
683 | return wrote; | |
684 | } | |
685 | ||
686 | /* | |
83ba7b07 | 687 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 688 | */ |
83ba7b07 | 689 | static struct wb_writeback_work * |
08852b6d | 690 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 691 | { |
83ba7b07 | 692 | struct wb_writeback_work *work = NULL; |
03ba3782 | 693 | |
6467716a | 694 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
695 | if (!list_empty(&bdi->work_list)) { |
696 | work = list_entry(bdi->work_list.next, | |
697 | struct wb_writeback_work, list); | |
698 | list_del_init(&work->list); | |
03ba3782 | 699 | } |
6467716a | 700 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 701 | return work; |
03ba3782 JA |
702 | } |
703 | ||
704 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
705 | { | |
706 | unsigned long expired; | |
707 | long nr_pages; | |
708 | ||
69b62d01 JA |
709 | /* |
710 | * When set to zero, disable periodic writeback | |
711 | */ | |
712 | if (!dirty_writeback_interval) | |
713 | return 0; | |
714 | ||
03ba3782 JA |
715 | expired = wb->last_old_flush + |
716 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
717 | if (time_before(jiffies, expired)) | |
718 | return 0; | |
719 | ||
720 | wb->last_old_flush = jiffies; | |
721 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
722 | global_page_state(NR_UNSTABLE_NFS) + | |
cffbc8aa | 723 | get_nr_dirty_inodes(); |
03ba3782 | 724 | |
c4a77a6c | 725 | if (nr_pages) { |
83ba7b07 | 726 | struct wb_writeback_work work = { |
c4a77a6c JA |
727 | .nr_pages = nr_pages, |
728 | .sync_mode = WB_SYNC_NONE, | |
729 | .for_kupdate = 1, | |
730 | .range_cyclic = 1, | |
731 | }; | |
732 | ||
83ba7b07 | 733 | return wb_writeback(wb, &work); |
c4a77a6c | 734 | } |
03ba3782 JA |
735 | |
736 | return 0; | |
737 | } | |
738 | ||
739 | /* | |
740 | * Retrieve work items and do the writeback they describe | |
741 | */ | |
742 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
743 | { | |
744 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 745 | struct wb_writeback_work *work; |
c4a77a6c | 746 | long wrote = 0; |
03ba3782 | 747 | |
81d73a32 | 748 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 749 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
750 | /* |
751 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 752 | * because this thread is exiting now. |
03ba3782 JA |
753 | */ |
754 | if (force_wait) | |
83ba7b07 | 755 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 756 | |
455b2864 DC |
757 | trace_writeback_exec(bdi, work); |
758 | ||
83ba7b07 | 759 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
760 | |
761 | /* | |
83ba7b07 CH |
762 | * Notify the caller of completion if this is a synchronous |
763 | * work item, otherwise just free it. | |
03ba3782 | 764 | */ |
83ba7b07 CH |
765 | if (work->done) |
766 | complete(work->done); | |
767 | else | |
768 | kfree(work); | |
03ba3782 JA |
769 | } |
770 | ||
771 | /* | |
772 | * Check for periodic writeback, kupdated() style | |
773 | */ | |
774 | wrote += wb_check_old_data_flush(wb); | |
81d73a32 | 775 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
776 | |
777 | return wrote; | |
778 | } | |
779 | ||
780 | /* | |
781 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
782 | * wakes up periodically and does kupdated style flushing. | |
783 | */ | |
08243900 | 784 | int bdi_writeback_thread(void *data) |
03ba3782 | 785 | { |
08243900 CH |
786 | struct bdi_writeback *wb = data; |
787 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
788 | long pages_written; |
789 | ||
08243900 CH |
790 | current->flags |= PF_FLUSHER | PF_SWAPWRITE; |
791 | set_freezable(); | |
ecd58403 | 792 | wb->last_active = jiffies; |
08243900 CH |
793 | |
794 | /* | |
795 | * Our parent may run at a different priority, just set us to normal | |
796 | */ | |
797 | set_user_nice(current, 0); | |
798 | ||
455b2864 DC |
799 | trace_writeback_thread_start(bdi); |
800 | ||
03ba3782 | 801 | while (!kthread_should_stop()) { |
6467716a AB |
802 | /* |
803 | * Remove own delayed wake-up timer, since we are already awake | |
804 | * and we'll take care of the preriodic write-back. | |
805 | */ | |
806 | del_timer(&wb->wakeup_timer); | |
807 | ||
03ba3782 JA |
808 | pages_written = wb_do_writeback(wb, 0); |
809 | ||
455b2864 DC |
810 | trace_writeback_pages_written(pages_written); |
811 | ||
03ba3782 | 812 | if (pages_written) |
ecd58403 | 813 | wb->last_active = jiffies; |
03ba3782 | 814 | |
297252c8 | 815 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 816 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 817 | __set_current_state(TASK_RUNNING); |
297252c8 | 818 | continue; |
03ba3782 JA |
819 | } |
820 | ||
253c34e9 | 821 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 822 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
823 | else { |
824 | /* | |
825 | * We have nothing to do, so can go sleep without any | |
826 | * timeout and save power. When a work is queued or | |
827 | * something is made dirty - we will be woken up. | |
828 | */ | |
297252c8 | 829 | schedule(); |
f9eadbbd | 830 | } |
69b62d01 | 831 | |
03ba3782 JA |
832 | try_to_freeze(); |
833 | } | |
834 | ||
fff5b85a | 835 | /* Flush any work that raced with us exiting */ |
08243900 CH |
836 | if (!list_empty(&bdi->work_list)) |
837 | wb_do_writeback(wb, 1); | |
455b2864 DC |
838 | |
839 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
840 | return 0; |
841 | } | |
842 | ||
08243900 | 843 | |
03ba3782 | 844 | /* |
b8c2f347 CH |
845 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
846 | * the whole world. | |
03ba3782 | 847 | */ |
b8c2f347 | 848 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 849 | { |
b8c2f347 | 850 | struct backing_dev_info *bdi; |
03ba3782 | 851 | |
83ba7b07 CH |
852 | if (!nr_pages) { |
853 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
854 | global_page_state(NR_UNSTABLE_NFS); |
855 | } | |
03ba3782 | 856 | |
b8c2f347 | 857 | rcu_read_lock(); |
cfc4ba53 | 858 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
859 | if (!bdi_has_dirty_io(bdi)) |
860 | continue; | |
83ba7b07 | 861 | __bdi_start_writeback(bdi, nr_pages, false, false); |
03ba3782 | 862 | } |
cfc4ba53 | 863 | rcu_read_unlock(); |
1da177e4 LT |
864 | } |
865 | ||
03ba3782 JA |
866 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
867 | { | |
868 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
869 | struct dentry *dentry; | |
870 | const char *name = "?"; | |
871 | ||
872 | dentry = d_find_alias(inode); | |
873 | if (dentry) { | |
874 | spin_lock(&dentry->d_lock); | |
875 | name = (const char *) dentry->d_name.name; | |
876 | } | |
877 | printk(KERN_DEBUG | |
878 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
879 | current->comm, task_pid_nr(current), inode->i_ino, | |
880 | name, inode->i_sb->s_id); | |
881 | if (dentry) { | |
882 | spin_unlock(&dentry->d_lock); | |
883 | dput(dentry); | |
884 | } | |
885 | } | |
886 | } | |
887 | ||
888 | /** | |
889 | * __mark_inode_dirty - internal function | |
890 | * @inode: inode to mark | |
891 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
892 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
893 | * mark_inode_dirty_sync. | |
1da177e4 | 894 | * |
03ba3782 JA |
895 | * Put the inode on the super block's dirty list. |
896 | * | |
897 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
898 | * dirty list only if it is hashed or if it refers to a blockdev. | |
899 | * If it was not hashed, it will never be added to the dirty list | |
900 | * even if it is later hashed, as it will have been marked dirty already. | |
901 | * | |
902 | * In short, make sure you hash any inodes _before_ you start marking | |
903 | * them dirty. | |
1da177e4 | 904 | * |
03ba3782 JA |
905 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
906 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 907 | * |
03ba3782 JA |
908 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
909 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
910 | * the kernel-internal blockdev inode represents the dirtying time of the | |
911 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
912 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
913 | * blockdev inode. | |
1da177e4 | 914 | */ |
03ba3782 | 915 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 916 | { |
03ba3782 | 917 | struct super_block *sb = inode->i_sb; |
253c34e9 AB |
918 | struct backing_dev_info *bdi = NULL; |
919 | bool wakeup_bdi = false; | |
1da177e4 | 920 | |
03ba3782 JA |
921 | /* |
922 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
923 | * dirty the inode itself | |
924 | */ | |
925 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
926 | if (sb->s_op->dirty_inode) | |
927 | sb->s_op->dirty_inode(inode); | |
928 | } | |
929 | ||
930 | /* | |
931 | * make sure that changes are seen by all cpus before we test i_state | |
932 | * -- mikulas | |
933 | */ | |
934 | smp_mb(); | |
935 | ||
936 | /* avoid the locking if we can */ | |
937 | if ((inode->i_state & flags) == flags) | |
938 | return; | |
939 | ||
940 | if (unlikely(block_dump)) | |
941 | block_dump___mark_inode_dirty(inode); | |
942 | ||
943 | spin_lock(&inode_lock); | |
944 | if ((inode->i_state & flags) != flags) { | |
945 | const int was_dirty = inode->i_state & I_DIRTY; | |
946 | ||
947 | inode->i_state |= flags; | |
948 | ||
949 | /* | |
950 | * If the inode is being synced, just update its dirty state. | |
951 | * The unlocker will place the inode on the appropriate | |
952 | * superblock list, based upon its state. | |
953 | */ | |
954 | if (inode->i_state & I_SYNC) | |
955 | goto out; | |
956 | ||
957 | /* | |
958 | * Only add valid (hashed) inodes to the superblock's | |
959 | * dirty list. Add blockdev inodes as well. | |
960 | */ | |
961 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 962 | if (inode_unhashed(inode)) |
03ba3782 JA |
963 | goto out; |
964 | } | |
a4ffdde6 | 965 | if (inode->i_state & I_FREEING) |
03ba3782 JA |
966 | goto out; |
967 | ||
968 | /* | |
969 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
970 | * reposition it (that would break b_dirty time-ordering). | |
971 | */ | |
972 | if (!was_dirty) { | |
253c34e9 AB |
973 | bdi = inode_to_bdi(inode); |
974 | ||
975 | if (bdi_cap_writeback_dirty(bdi)) { | |
976 | WARN(!test_bit(BDI_registered, &bdi->state), | |
977 | "bdi-%s not registered\n", bdi->name); | |
978 | ||
979 | /* | |
980 | * If this is the first dirty inode for this | |
981 | * bdi, we have to wake-up the corresponding | |
982 | * bdi thread to make sure background | |
983 | * write-back happens later. | |
984 | */ | |
985 | if (!wb_has_dirty_io(&bdi->wb)) | |
986 | wakeup_bdi = true; | |
500b067c | 987 | } |
03ba3782 JA |
988 | |
989 | inode->dirtied_when = jiffies; | |
253c34e9 | 990 | list_move(&inode->i_list, &bdi->wb.b_dirty); |
1da177e4 | 991 | } |
1da177e4 | 992 | } |
03ba3782 JA |
993 | out: |
994 | spin_unlock(&inode_lock); | |
253c34e9 AB |
995 | |
996 | if (wakeup_bdi) | |
6467716a | 997 | bdi_wakeup_thread_delayed(bdi); |
03ba3782 JA |
998 | } |
999 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1000 | ||
1001 | /* | |
1002 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1003 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1004 | * | |
1005 | * If older_than_this is non-NULL, then only write out inodes which | |
1006 | * had their first dirtying at a time earlier than *older_than_this. | |
1007 | * | |
03ba3782 JA |
1008 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1009 | * This function assumes that the blockdev superblock's inodes are backed by | |
1010 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1011 | * assume that all inodes are backed by the same queue. | |
1012 | * | |
1013 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1014 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1015 | * on the writer throttling path, and we get decent balancing between many | |
1016 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1017 | */ | |
b6e51316 | 1018 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1019 | { |
1020 | struct inode *inode, *old_inode = NULL; | |
1021 | ||
1022 | /* | |
1023 | * We need to be protected against the filesystem going from | |
1024 | * r/o to r/w or vice versa. | |
1025 | */ | |
b6e51316 | 1026 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
1027 | |
1028 | spin_lock(&inode_lock); | |
1029 | ||
1030 | /* | |
1031 | * Data integrity sync. Must wait for all pages under writeback, | |
1032 | * because there may have been pages dirtied before our sync | |
1033 | * call, but which had writeout started before we write it out. | |
1034 | * In which case, the inode may not be on the dirty list, but | |
1035 | * we still have to wait for that writeout. | |
1036 | */ | |
b6e51316 | 1037 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1038 | struct address_space *mapping; |
1039 | ||
a4ffdde6 | 1040 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) |
03ba3782 JA |
1041 | continue; |
1042 | mapping = inode->i_mapping; | |
1043 | if (mapping->nrpages == 0) | |
1044 | continue; | |
1045 | __iget(inode); | |
1046 | spin_unlock(&inode_lock); | |
1047 | /* | |
1048 | * We hold a reference to 'inode' so it couldn't have | |
1049 | * been removed from s_inodes list while we dropped the | |
1050 | * inode_lock. We cannot iput the inode now as we can | |
1051 | * be holding the last reference and we cannot iput it | |
1052 | * under inode_lock. So we keep the reference and iput | |
1053 | * it later. | |
1054 | */ | |
1055 | iput(old_inode); | |
1056 | old_inode = inode; | |
1057 | ||
1058 | filemap_fdatawait(mapping); | |
1059 | ||
1060 | cond_resched(); | |
1061 | ||
1062 | spin_lock(&inode_lock); | |
1063 | } | |
1064 | spin_unlock(&inode_lock); | |
1065 | iput(old_inode); | |
1da177e4 LT |
1066 | } |
1067 | ||
d8a8559c JA |
1068 | /** |
1069 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1070 | * @sb: the superblock | |
1da177e4 | 1071 | * |
d8a8559c JA |
1072 | * Start writeback on some inodes on this super_block. No guarantees are made |
1073 | * on how many (if any) will be written, and this function does not wait | |
1074 | * for IO completion of submitted IO. The number of pages submitted is | |
1075 | * returned. | |
1da177e4 | 1076 | */ |
b6e51316 | 1077 | void writeback_inodes_sb(struct super_block *sb) |
1da177e4 | 1078 | { |
0e3c9a22 JA |
1079 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
1080 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
83ba7b07 CH |
1081 | DECLARE_COMPLETION_ONSTACK(done); |
1082 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1083 | .sb = sb, |
1084 | .sync_mode = WB_SYNC_NONE, | |
83ba7b07 | 1085 | .done = &done, |
3c4d7165 | 1086 | }; |
d8a8559c | 1087 | |
cf37e972 CH |
1088 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1089 | ||
cffbc8aa | 1090 | work.nr_pages = nr_dirty + nr_unstable + get_nr_dirty_inodes(); |
0e3c9a22 | 1091 | |
83ba7b07 CH |
1092 | bdi_queue_work(sb->s_bdi, &work); |
1093 | wait_for_completion(&done); | |
e913fc82 | 1094 | } |
0e3c9a22 | 1095 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1096 | |
17bd55d0 ES |
1097 | /** |
1098 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1099 | * @sb: the superblock | |
1100 | * | |
1101 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1102 | * Returns 1 if writeback was started, 0 if not. | |
1103 | */ | |
1104 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1105 | { | |
1106 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1107 | down_read(&sb->s_umount); |
17bd55d0 | 1108 | writeback_inodes_sb(sb); |
cf37e972 | 1109 | up_read(&sb->s_umount); |
17bd55d0 ES |
1110 | return 1; |
1111 | } else | |
1112 | return 0; | |
1113 | } | |
1114 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1115 | ||
d8a8559c JA |
1116 | /** |
1117 | * sync_inodes_sb - sync sb inode pages | |
1118 | * @sb: the superblock | |
1119 | * | |
1120 | * This function writes and waits on any dirty inode belonging to this | |
1121 | * super_block. The number of pages synced is returned. | |
1122 | */ | |
b6e51316 | 1123 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1124 | { |
83ba7b07 CH |
1125 | DECLARE_COMPLETION_ONSTACK(done); |
1126 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1127 | .sb = sb, |
1128 | .sync_mode = WB_SYNC_ALL, | |
1129 | .nr_pages = LONG_MAX, | |
1130 | .range_cyclic = 0, | |
83ba7b07 | 1131 | .done = &done, |
3c4d7165 CH |
1132 | }; |
1133 | ||
cf37e972 CH |
1134 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1135 | ||
83ba7b07 CH |
1136 | bdi_queue_work(sb->s_bdi, &work); |
1137 | wait_for_completion(&done); | |
1138 | ||
b6e51316 | 1139 | wait_sb_inodes(sb); |
1da177e4 | 1140 | } |
d8a8559c | 1141 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1142 | |
1da177e4 | 1143 | /** |
7f04c26d AA |
1144 | * write_inode_now - write an inode to disk |
1145 | * @inode: inode to write to disk | |
1146 | * @sync: whether the write should be synchronous or not | |
1147 | * | |
1148 | * This function commits an inode to disk immediately if it is dirty. This is | |
1149 | * primarily needed by knfsd. | |
1da177e4 | 1150 | * |
7f04c26d | 1151 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1152 | */ |
1da177e4 LT |
1153 | int write_inode_now(struct inode *inode, int sync) |
1154 | { | |
1155 | int ret; | |
1156 | struct writeback_control wbc = { | |
1157 | .nr_to_write = LONG_MAX, | |
18914b18 | 1158 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1159 | .range_start = 0, |
1160 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1161 | }; |
1162 | ||
1163 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1164 | wbc.nr_to_write = 0; |
1da177e4 LT |
1165 | |
1166 | might_sleep(); | |
1167 | spin_lock(&inode_lock); | |
01c03194 | 1168 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1169 | spin_unlock(&inode_lock); |
1170 | if (sync) | |
1c0eeaf5 | 1171 | inode_sync_wait(inode); |
1da177e4 LT |
1172 | return ret; |
1173 | } | |
1174 | EXPORT_SYMBOL(write_inode_now); | |
1175 | ||
1176 | /** | |
1177 | * sync_inode - write an inode and its pages to disk. | |
1178 | * @inode: the inode to sync | |
1179 | * @wbc: controls the writeback mode | |
1180 | * | |
1181 | * sync_inode() will write an inode and its pages to disk. It will also | |
1182 | * correctly update the inode on its superblock's dirty inode lists and will | |
1183 | * update inode->i_state. | |
1184 | * | |
1185 | * The caller must have a ref on the inode. | |
1186 | */ | |
1187 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1188 | { | |
1189 | int ret; | |
1190 | ||
1191 | spin_lock(&inode_lock); | |
01c03194 | 1192 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1193 | spin_unlock(&inode_lock); |
1194 | return ret; | |
1195 | } | |
1196 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1197 | |
1198 | /** | |
1199 | * sync_inode - write an inode to disk | |
1200 | * @inode: the inode to sync | |
1201 | * @wait: wait for I/O to complete. | |
1202 | * | |
1203 | * Write an inode to disk and adjust it's dirty state after completion. | |
1204 | * | |
1205 | * Note: only writes the actual inode, no associated data or other metadata. | |
1206 | */ | |
1207 | int sync_inode_metadata(struct inode *inode, int wait) | |
1208 | { | |
1209 | struct writeback_control wbc = { | |
1210 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1211 | .nr_to_write = 0, /* metadata-only */ | |
1212 | }; | |
1213 | ||
1214 | return sync_inode(inode, &wbc); | |
1215 | } | |
1216 | EXPORT_SYMBOL(sync_inode_metadata); |