]>
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> | |
630d9c47 | 17 | #include <linux/export.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> | |
bc31b86a | 23 | #include <linux/pagemap.h> |
03ba3782 | 24 | #include <linux/kthread.h> |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
455b2864 | 28 | #include <linux/tracepoint.h> |
07f3f05c | 29 | #include "internal.h" |
1da177e4 | 30 | |
bc31b86a WF |
31 | /* |
32 | * 4MB minimal write chunk size | |
33 | */ | |
34 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) | |
35 | ||
c4a77a6c JA |
36 | /* |
37 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
38 | */ | |
83ba7b07 | 39 | struct wb_writeback_work { |
c4a77a6c JA |
40 | long nr_pages; |
41 | struct super_block *sb; | |
d46db3d5 | 42 | unsigned long *older_than_this; |
c4a77a6c | 43 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 44 | unsigned int tagged_writepages:1; |
52957fe1 HS |
45 | unsigned int for_kupdate:1; |
46 | unsigned int range_cyclic:1; | |
47 | unsigned int for_background:1; | |
0e175a18 | 48 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 49 | |
8010c3b6 | 50 | struct list_head list; /* pending work list */ |
83ba7b07 | 51 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
52 | }; |
53 | ||
f11b00f3 AB |
54 | /** |
55 | * writeback_in_progress - determine whether there is writeback in progress | |
56 | * @bdi: the device's backing_dev_info structure. | |
57 | * | |
03ba3782 JA |
58 | * Determine whether there is writeback waiting to be handled against a |
59 | * backing device. | |
f11b00f3 AB |
60 | */ |
61 | int writeback_in_progress(struct backing_dev_info *bdi) | |
62 | { | |
81d73a32 | 63 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 | 64 | } |
00d4e736 | 65 | EXPORT_SYMBOL(writeback_in_progress); |
f11b00f3 | 66 | |
692ebd17 JK |
67 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
68 | { | |
69 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 70 | |
a8855990 | 71 | if (sb_is_blkdev_sb(sb)) |
aaead25b CH |
72 | return inode->i_mapping->backing_dev_info; |
73 | ||
74 | return sb->s_bdi; | |
692ebd17 JK |
75 | } |
76 | ||
7ccf19a8 NP |
77 | static inline struct inode *wb_inode(struct list_head *head) |
78 | { | |
79 | return list_entry(head, struct inode, i_wb_list); | |
80 | } | |
81 | ||
15eb77a0 WF |
82 | /* |
83 | * Include the creation of the trace points after defining the | |
84 | * wb_writeback_work structure and inline functions so that the definition | |
85 | * remains local to this file. | |
86 | */ | |
87 | #define CREATE_TRACE_POINTS | |
88 | #include <trace/events/writeback.h> | |
89 | ||
6585027a JK |
90 | static void bdi_queue_work(struct backing_dev_info *bdi, |
91 | struct wb_writeback_work *work) | |
92 | { | |
93 | trace_writeback_queue(bdi, work); | |
94 | ||
95 | spin_lock_bh(&bdi->wb_lock); | |
96 | list_add_tail(&work->list, &bdi->work_list); | |
6467716a | 97 | spin_unlock_bh(&bdi->wb_lock); |
839a8e86 TH |
98 | |
99 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
1da177e4 LT |
100 | } |
101 | ||
83ba7b07 CH |
102 | static void |
103 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
0e175a18 | 104 | bool range_cyclic, enum wb_reason reason) |
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) { | |
839a8e86 TH |
114 | trace_writeback_nowork(bdi); |
115 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
83ba7b07 | 116 | return; |
bcddc3f0 | 117 | } |
03ba3782 | 118 | |
83ba7b07 CH |
119 | work->sync_mode = WB_SYNC_NONE; |
120 | work->nr_pages = nr_pages; | |
121 | work->range_cyclic = range_cyclic; | |
0e175a18 | 122 | work->reason = reason; |
03ba3782 | 123 | |
83ba7b07 | 124 | bdi_queue_work(bdi, work); |
b6e51316 JA |
125 | } |
126 | ||
127 | /** | |
128 | * bdi_start_writeback - start writeback | |
129 | * @bdi: the backing device to write from | |
130 | * @nr_pages: the number of pages to write | |
786228ab | 131 | * @reason: reason why some writeback work was initiated |
b6e51316 JA |
132 | * |
133 | * Description: | |
134 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 135 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 136 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
137 | * |
138 | */ | |
0e175a18 CW |
139 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
140 | enum wb_reason reason) | |
b6e51316 | 141 | { |
0e175a18 | 142 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
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: | |
6585027a JK |
150 | * This makes sure WB_SYNC_NONE background writeback happens. When |
151 | * this function returns, it is only guaranteed that for given BDI | |
152 | * some IO is happening if we are over background dirty threshold. | |
153 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
154 | */ |
155 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
156 | { | |
6585027a JK |
157 | /* |
158 | * We just wake up the flusher thread. It will perform background | |
159 | * writeback as soon as there is no other work to do. | |
160 | */ | |
71927e84 | 161 | trace_writeback_wake_background(bdi); |
839a8e86 | 162 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
1da177e4 LT |
163 | } |
164 | ||
a66979ab DC |
165 | /* |
166 | * Remove the inode from the writeback list it is on. | |
167 | */ | |
168 | void inode_wb_list_del(struct inode *inode) | |
169 | { | |
f758eeab CH |
170 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
171 | ||
172 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 173 | list_del_init(&inode->i_wb_list); |
f758eeab | 174 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
175 | } |
176 | ||
6610a0bc AM |
177 | /* |
178 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
179 | * furthest end of its superblock's dirty-inode list. | |
180 | * | |
181 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 182 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
183 | * the case then the inode must have been redirtied while it was being written |
184 | * out and we don't reset its dirtied_when. | |
185 | */ | |
f758eeab | 186 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 187 | { |
f758eeab | 188 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 189 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 190 | struct inode *tail; |
6610a0bc | 191 | |
7ccf19a8 | 192 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 193 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
194 | inode->dirtied_when = jiffies; |
195 | } | |
7ccf19a8 | 196 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
197 | } |
198 | ||
c986d1e2 | 199 | /* |
66f3b8e2 | 200 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 201 | */ |
f758eeab | 202 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 203 | { |
f758eeab | 204 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 205 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
206 | } |
207 | ||
1c0eeaf5 JE |
208 | static void inode_sync_complete(struct inode *inode) |
209 | { | |
365b94ae | 210 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
211 | /* If inode is clean an unused, put it into LRU now... */ |
212 | inode_add_lru(inode); | |
365b94ae | 213 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
214 | smp_mb(); |
215 | wake_up_bit(&inode->i_state, __I_SYNC); | |
216 | } | |
217 | ||
d2caa3c5 JL |
218 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
219 | { | |
220 | bool ret = time_after(inode->dirtied_when, t); | |
221 | #ifndef CONFIG_64BIT | |
222 | /* | |
223 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
224 | * It _appears_ to be in the future, but is actually in distant past. | |
225 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 226 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
227 | */ |
228 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
229 | #endif | |
230 | return ret; | |
231 | } | |
232 | ||
2c136579 | 233 | /* |
0e2f2b23 | 234 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 235 | * @delaying_queue to @dispatch_queue. |
2c136579 | 236 | */ |
e84d0a4f | 237 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 238 | struct list_head *dispatch_queue, |
ad4e38dd | 239 | struct wb_writeback_work *work) |
2c136579 | 240 | { |
5c03449d SL |
241 | LIST_HEAD(tmp); |
242 | struct list_head *pos, *node; | |
cf137307 | 243 | struct super_block *sb = NULL; |
5c03449d | 244 | struct inode *inode; |
cf137307 | 245 | int do_sb_sort = 0; |
e84d0a4f | 246 | int moved = 0; |
5c03449d | 247 | |
2c136579 | 248 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 249 | inode = wb_inode(delaying_queue->prev); |
ad4e38dd CW |
250 | if (work->older_than_this && |
251 | inode_dirtied_after(inode, *work->older_than_this)) | |
2c136579 | 252 | break; |
a8855990 JK |
253 | list_move(&inode->i_wb_list, &tmp); |
254 | moved++; | |
255 | if (sb_is_blkdev_sb(inode->i_sb)) | |
256 | continue; | |
cf137307 JA |
257 | if (sb && sb != inode->i_sb) |
258 | do_sb_sort = 1; | |
259 | sb = inode->i_sb; | |
5c03449d SL |
260 | } |
261 | ||
cf137307 JA |
262 | /* just one sb in list, splice to dispatch_queue and we're done */ |
263 | if (!do_sb_sort) { | |
264 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 265 | goto out; |
cf137307 JA |
266 | } |
267 | ||
5c03449d SL |
268 | /* Move inodes from one superblock together */ |
269 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 270 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 271 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 272 | inode = wb_inode(pos); |
5c03449d | 273 | if (inode->i_sb == sb) |
7ccf19a8 | 274 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 275 | } |
2c136579 | 276 | } |
e84d0a4f WF |
277 | out: |
278 | return moved; | |
2c136579 FW |
279 | } |
280 | ||
281 | /* | |
282 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
283 | * Before |
284 | * newly dirtied b_dirty b_io b_more_io | |
285 | * =============> gf edc BA | |
286 | * After | |
287 | * newly dirtied b_dirty b_io b_more_io | |
288 | * =============> g fBAedc | |
289 | * | | |
290 | * +--> dequeue for IO | |
2c136579 | 291 | */ |
ad4e38dd | 292 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 293 | { |
e84d0a4f | 294 | int moved; |
f758eeab | 295 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 296 | list_splice_init(&wb->b_more_io, &wb->b_io); |
ad4e38dd CW |
297 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
298 | trace_writeback_queue_io(wb, work, moved); | |
66f3b8e2 JA |
299 | } |
300 | ||
a9185b41 | 301 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 302 | { |
9fb0a7da TH |
303 | int ret; |
304 | ||
305 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
306 | trace_writeback_write_inode_start(inode, wbc); | |
307 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
308 | trace_writeback_write_inode(inode, wbc); | |
309 | return ret; | |
310 | } | |
03ba3782 | 311 | return 0; |
08d8e974 | 312 | } |
08d8e974 | 313 | |
1da177e4 | 314 | /* |
169ebd90 JK |
315 | * Wait for writeback on an inode to complete. Called with i_lock held. |
316 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 317 | */ |
169ebd90 JK |
318 | static void __inode_wait_for_writeback(struct inode *inode) |
319 | __releases(inode->i_lock) | |
320 | __acquires(inode->i_lock) | |
01c03194 CH |
321 | { |
322 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
323 | wait_queue_head_t *wqh; | |
324 | ||
325 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
326 | while (inode->i_state & I_SYNC) { |
327 | spin_unlock(&inode->i_lock); | |
01c03194 | 328 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
250df6ed | 329 | spin_lock(&inode->i_lock); |
58a9d3d8 | 330 | } |
01c03194 CH |
331 | } |
332 | ||
169ebd90 JK |
333 | /* |
334 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
335 | */ | |
336 | void inode_wait_for_writeback(struct inode *inode) | |
337 | { | |
338 | spin_lock(&inode->i_lock); | |
339 | __inode_wait_for_writeback(inode); | |
340 | spin_unlock(&inode->i_lock); | |
341 | } | |
342 | ||
343 | /* | |
344 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
345 | * held and drops it. It is aimed for callers not holding any inode reference | |
346 | * so once i_lock is dropped, inode can go away. | |
347 | */ | |
348 | static void inode_sleep_on_writeback(struct inode *inode) | |
349 | __releases(inode->i_lock) | |
350 | { | |
351 | DEFINE_WAIT(wait); | |
352 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
353 | int sleep; | |
354 | ||
355 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
356 | sleep = inode->i_state & I_SYNC; | |
357 | spin_unlock(&inode->i_lock); | |
358 | if (sleep) | |
359 | schedule(); | |
360 | finish_wait(wqh, &wait); | |
361 | } | |
362 | ||
ccb26b5a JK |
363 | /* |
364 | * Find proper writeback list for the inode depending on its current state and | |
365 | * possibly also change of its state while we were doing writeback. Here we | |
366 | * handle things such as livelock prevention or fairness of writeback among | |
367 | * inodes. This function can be called only by flusher thread - noone else | |
368 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
369 | * thread's back can have unexpected consequences. | |
370 | */ | |
371 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
372 | struct writeback_control *wbc) | |
373 | { | |
374 | if (inode->i_state & I_FREEING) | |
375 | return; | |
376 | ||
377 | /* | |
378 | * Sync livelock prevention. Each inode is tagged and synced in one | |
379 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
380 | * the dirty time to prevent enqueue and sync it again. | |
381 | */ | |
382 | if ((inode->i_state & I_DIRTY) && | |
383 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
384 | inode->dirtied_when = jiffies; | |
385 | ||
4f8ad655 JK |
386 | if (wbc->pages_skipped) { |
387 | /* | |
388 | * writeback is not making progress due to locked | |
389 | * buffers. Skip this inode for now. | |
390 | */ | |
391 | redirty_tail(inode, wb); | |
392 | return; | |
393 | } | |
394 | ||
ccb26b5a JK |
395 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
396 | /* | |
397 | * We didn't write back all the pages. nfs_writepages() | |
398 | * sometimes bales out without doing anything. | |
399 | */ | |
400 | if (wbc->nr_to_write <= 0) { | |
401 | /* Slice used up. Queue for next turn. */ | |
402 | requeue_io(inode, wb); | |
403 | } else { | |
404 | /* | |
405 | * Writeback blocked by something other than | |
406 | * congestion. Delay the inode for some time to | |
407 | * avoid spinning on the CPU (100% iowait) | |
408 | * retrying writeback of the dirty page/inode | |
409 | * that cannot be performed immediately. | |
410 | */ | |
411 | redirty_tail(inode, wb); | |
412 | } | |
413 | } else if (inode->i_state & I_DIRTY) { | |
414 | /* | |
415 | * Filesystems can dirty the inode during writeback operations, | |
416 | * such as delayed allocation during submission or metadata | |
417 | * updates after data IO completion. | |
418 | */ | |
419 | redirty_tail(inode, wb); | |
420 | } else { | |
421 | /* The inode is clean. Remove from writeback lists. */ | |
422 | list_del_init(&inode->i_wb_list); | |
423 | } | |
424 | } | |
425 | ||
01c03194 | 426 | /* |
4f8ad655 JK |
427 | * Write out an inode and its dirty pages. Do not update the writeback list |
428 | * linkage. That is left to the caller. The caller is also responsible for | |
429 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
430 | */ |
431 | static int | |
cd8ed2a4 | 432 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 433 | { |
1da177e4 | 434 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 435 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 436 | unsigned dirty; |
1da177e4 LT |
437 | int ret; |
438 | ||
4f8ad655 | 439 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 440 | |
9fb0a7da TH |
441 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
442 | ||
1da177e4 LT |
443 | ret = do_writepages(mapping, wbc); |
444 | ||
26821ed4 CH |
445 | /* |
446 | * Make sure to wait on the data before writing out the metadata. | |
447 | * This is important for filesystems that modify metadata on data | |
448 | * I/O completion. | |
449 | */ | |
a9185b41 | 450 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 451 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
452 | if (ret == 0) |
453 | ret = err; | |
454 | } | |
455 | ||
5547e8aa DM |
456 | /* |
457 | * Some filesystems may redirty the inode during the writeback | |
458 | * due to delalloc, clear dirty metadata flags right before | |
459 | * write_inode() | |
460 | */ | |
250df6ed | 461 | spin_lock(&inode->i_lock); |
6290be1c JK |
462 | /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ |
463 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
464 | inode->i_state &= ~I_DIRTY_PAGES; | |
5547e8aa DM |
465 | dirty = inode->i_state & I_DIRTY; |
466 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 467 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
468 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
469 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 470 | int err = write_inode(inode, wbc); |
1da177e4 LT |
471 | if (ret == 0) |
472 | ret = err; | |
473 | } | |
4f8ad655 JK |
474 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
475 | return ret; | |
476 | } | |
477 | ||
478 | /* | |
479 | * Write out an inode's dirty pages. Either the caller has an active reference | |
480 | * on the inode or the inode has I_WILL_FREE set. | |
481 | * | |
482 | * This function is designed to be called for writing back one inode which | |
483 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
484 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
485 | */ | |
486 | static int | |
487 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
488 | struct writeback_control *wbc) | |
489 | { | |
490 | int ret = 0; | |
491 | ||
492 | spin_lock(&inode->i_lock); | |
493 | if (!atomic_read(&inode->i_count)) | |
494 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
495 | else | |
496 | WARN_ON(inode->i_state & I_WILL_FREE); | |
497 | ||
498 | if (inode->i_state & I_SYNC) { | |
499 | if (wbc->sync_mode != WB_SYNC_ALL) | |
500 | goto out; | |
501 | /* | |
169ebd90 JK |
502 | * It's a data-integrity sync. We must wait. Since callers hold |
503 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
504 | * away under us. | |
4f8ad655 | 505 | */ |
169ebd90 | 506 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
507 | } |
508 | WARN_ON(inode->i_state & I_SYNC); | |
509 | /* | |
510 | * Skip inode if it is clean. We don't want to mess with writeback | |
511 | * lists in this function since flusher thread may be doing for example | |
512 | * sync in parallel and if we move the inode, it could get skipped. So | |
513 | * here we make sure inode is on some writeback list and leave it there | |
514 | * unless we have completely cleaned the inode. | |
515 | */ | |
516 | if (!(inode->i_state & I_DIRTY)) | |
517 | goto out; | |
518 | inode->i_state |= I_SYNC; | |
519 | spin_unlock(&inode->i_lock); | |
520 | ||
cd8ed2a4 | 521 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 522 | |
f758eeab | 523 | spin_lock(&wb->list_lock); |
250df6ed | 524 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
525 | /* |
526 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
527 | * touch it. See comment above for explanation. | |
528 | */ | |
529 | if (!(inode->i_state & I_DIRTY)) | |
530 | list_del_init(&inode->i_wb_list); | |
531 | spin_unlock(&wb->list_lock); | |
1c0eeaf5 | 532 | inode_sync_complete(inode); |
4f8ad655 JK |
533 | out: |
534 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
535 | return ret; |
536 | } | |
537 | ||
1a12d8bd WF |
538 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
539 | struct wb_writeback_work *work) | |
d46db3d5 WF |
540 | { |
541 | long pages; | |
542 | ||
543 | /* | |
544 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
545 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
546 | * here avoids calling into writeback_inodes_wb() more than once. | |
547 | * | |
548 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
549 | * | |
550 | * wb_writeback() | |
551 | * writeback_sb_inodes() <== called only once | |
552 | * write_cache_pages() <== called once for each inode | |
553 | * (quickly) tag currently dirty pages | |
554 | * (maybe slowly) sync all tagged pages | |
555 | */ | |
556 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
557 | pages = LONG_MAX; | |
1a12d8bd WF |
558 | else { |
559 | pages = min(bdi->avg_write_bandwidth / 2, | |
560 | global_dirty_limit / DIRTY_SCOPE); | |
561 | pages = min(pages, work->nr_pages); | |
562 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
563 | MIN_WRITEBACK_PAGES); | |
564 | } | |
d46db3d5 WF |
565 | |
566 | return pages; | |
567 | } | |
568 | ||
f11c9c5c ES |
569 | /* |
570 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 571 | * |
d46db3d5 | 572 | * Return the number of pages and/or inodes written. |
f11c9c5c | 573 | */ |
d46db3d5 WF |
574 | static long writeback_sb_inodes(struct super_block *sb, |
575 | struct bdi_writeback *wb, | |
576 | struct wb_writeback_work *work) | |
1da177e4 | 577 | { |
d46db3d5 WF |
578 | struct writeback_control wbc = { |
579 | .sync_mode = work->sync_mode, | |
580 | .tagged_writepages = work->tagged_writepages, | |
581 | .for_kupdate = work->for_kupdate, | |
582 | .for_background = work->for_background, | |
583 | .range_cyclic = work->range_cyclic, | |
584 | .range_start = 0, | |
585 | .range_end = LLONG_MAX, | |
586 | }; | |
587 | unsigned long start_time = jiffies; | |
588 | long write_chunk; | |
589 | long wrote = 0; /* count both pages and inodes */ | |
590 | ||
03ba3782 | 591 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 592 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
593 | |
594 | if (inode->i_sb != sb) { | |
d46db3d5 | 595 | if (work->sb) { |
edadfb10 CH |
596 | /* |
597 | * We only want to write back data for this | |
598 | * superblock, move all inodes not belonging | |
599 | * to it back onto the dirty list. | |
600 | */ | |
f758eeab | 601 | redirty_tail(inode, wb); |
edadfb10 CH |
602 | continue; |
603 | } | |
604 | ||
605 | /* | |
606 | * The inode belongs to a different superblock. | |
607 | * Bounce back to the caller to unpin this and | |
608 | * pin the next superblock. | |
609 | */ | |
d46db3d5 | 610 | break; |
edadfb10 CH |
611 | } |
612 | ||
9843b76a | 613 | /* |
331cbdee WL |
614 | * Don't bother with new inodes or inodes being freed, first |
615 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
616 | * kind writeout is handled by the freer. |
617 | */ | |
250df6ed | 618 | spin_lock(&inode->i_lock); |
9843b76a | 619 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 620 | spin_unlock(&inode->i_lock); |
fcc5c222 | 621 | redirty_tail(inode, wb); |
7ef0d737 NP |
622 | continue; |
623 | } | |
cc1676d9 JK |
624 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
625 | /* | |
626 | * If this inode is locked for writeback and we are not | |
627 | * doing writeback-for-data-integrity, move it to | |
628 | * b_more_io so that writeback can proceed with the | |
629 | * other inodes on s_io. | |
630 | * | |
631 | * We'll have another go at writing back this inode | |
632 | * when we completed a full scan of b_io. | |
633 | */ | |
634 | spin_unlock(&inode->i_lock); | |
635 | requeue_io(inode, wb); | |
636 | trace_writeback_sb_inodes_requeue(inode); | |
637 | continue; | |
638 | } | |
f0d07b7f JK |
639 | spin_unlock(&wb->list_lock); |
640 | ||
4f8ad655 JK |
641 | /* |
642 | * We already requeued the inode if it had I_SYNC set and we | |
643 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
644 | * WB_SYNC_ALL case. | |
645 | */ | |
169ebd90 JK |
646 | if (inode->i_state & I_SYNC) { |
647 | /* Wait for I_SYNC. This function drops i_lock... */ | |
648 | inode_sleep_on_writeback(inode); | |
649 | /* Inode may be gone, start again */ | |
ead188f9 | 650 | spin_lock(&wb->list_lock); |
169ebd90 JK |
651 | continue; |
652 | } | |
4f8ad655 JK |
653 | inode->i_state |= I_SYNC; |
654 | spin_unlock(&inode->i_lock); | |
169ebd90 | 655 | |
1a12d8bd | 656 | write_chunk = writeback_chunk_size(wb->bdi, work); |
d46db3d5 WF |
657 | wbc.nr_to_write = write_chunk; |
658 | wbc.pages_skipped = 0; | |
250df6ed | 659 | |
169ebd90 JK |
660 | /* |
661 | * We use I_SYNC to pin the inode in memory. While it is set | |
662 | * evict_inode() will wait so the inode cannot be freed. | |
663 | */ | |
cd8ed2a4 | 664 | __writeback_single_inode(inode, &wbc); |
250df6ed | 665 | |
d46db3d5 WF |
666 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
667 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
668 | spin_lock(&wb->list_lock); |
669 | spin_lock(&inode->i_lock); | |
d46db3d5 WF |
670 | if (!(inode->i_state & I_DIRTY)) |
671 | wrote++; | |
4f8ad655 JK |
672 | requeue_inode(inode, wb, &wbc); |
673 | inode_sync_complete(inode); | |
0f1b1fd8 | 674 | spin_unlock(&inode->i_lock); |
169ebd90 | 675 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
676 | /* |
677 | * bail out to wb_writeback() often enough to check | |
678 | * background threshold and other termination conditions. | |
679 | */ | |
680 | if (wrote) { | |
681 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
682 | break; | |
683 | if (work->nr_pages <= 0) | |
684 | break; | |
8bc3be27 | 685 | } |
1da177e4 | 686 | } |
d46db3d5 | 687 | return wrote; |
f11c9c5c ES |
688 | } |
689 | ||
d46db3d5 WF |
690 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
691 | struct wb_writeback_work *work) | |
f11c9c5c | 692 | { |
d46db3d5 WF |
693 | unsigned long start_time = jiffies; |
694 | long wrote = 0; | |
38f21977 | 695 | |
f11c9c5c | 696 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 697 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 698 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 699 | |
12ad3ab6 | 700 | if (!grab_super_passive(sb)) { |
0e995816 WF |
701 | /* |
702 | * grab_super_passive() may fail consistently due to | |
703 | * s_umount being grabbed by someone else. Don't use | |
704 | * requeue_io() to avoid busy retrying the inode/sb. | |
705 | */ | |
706 | redirty_tail(inode, wb); | |
edadfb10 | 707 | continue; |
f11c9c5c | 708 | } |
d46db3d5 | 709 | wrote += writeback_sb_inodes(sb, wb, work); |
edadfb10 | 710 | drop_super(sb); |
f11c9c5c | 711 | |
d46db3d5 WF |
712 | /* refer to the same tests at the end of writeback_sb_inodes */ |
713 | if (wrote) { | |
714 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
715 | break; | |
716 | if (work->nr_pages <= 0) | |
717 | break; | |
718 | } | |
f11c9c5c | 719 | } |
66f3b8e2 | 720 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 721 | return wrote; |
66f3b8e2 JA |
722 | } |
723 | ||
0e175a18 CW |
724 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
725 | enum wb_reason reason) | |
edadfb10 | 726 | { |
d46db3d5 WF |
727 | struct wb_writeback_work work = { |
728 | .nr_pages = nr_pages, | |
729 | .sync_mode = WB_SYNC_NONE, | |
730 | .range_cyclic = 1, | |
0e175a18 | 731 | .reason = reason, |
d46db3d5 | 732 | }; |
edadfb10 | 733 | |
f758eeab | 734 | spin_lock(&wb->list_lock); |
424b351f | 735 | if (list_empty(&wb->b_io)) |
ad4e38dd | 736 | queue_io(wb, &work); |
d46db3d5 | 737 | __writeback_inodes_wb(wb, &work); |
f758eeab | 738 | spin_unlock(&wb->list_lock); |
edadfb10 | 739 | |
d46db3d5 WF |
740 | return nr_pages - work.nr_pages; |
741 | } | |
03ba3782 | 742 | |
b00949aa | 743 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
03ba3782 JA |
744 | { |
745 | unsigned long background_thresh, dirty_thresh; | |
746 | ||
16c4042f | 747 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 748 | |
b00949aa WF |
749 | if (global_page_state(NR_FILE_DIRTY) + |
750 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
751 | return true; | |
752 | ||
753 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > | |
754 | bdi_dirty_limit(bdi, background_thresh)) | |
755 | return true; | |
756 | ||
757 | return false; | |
03ba3782 JA |
758 | } |
759 | ||
e98be2d5 WF |
760 | /* |
761 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
762 | * only the flusher working on the first wb should do it. | |
763 | */ | |
764 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
765 | unsigned long start_time) | |
766 | { | |
af6a3113 | 767 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
768 | } |
769 | ||
03ba3782 JA |
770 | /* |
771 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 772 | * |
03ba3782 JA |
773 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
774 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
775 | * just walks the superblock inode list, writing back any inodes which are | |
776 | * older than a specific point in time. | |
66f3b8e2 | 777 | * |
03ba3782 JA |
778 | * Try to run once per dirty_writeback_interval. But if a writeback event |
779 | * takes longer than a dirty_writeback_interval interval, then leave a | |
780 | * one-second gap. | |
66f3b8e2 | 781 | * |
03ba3782 JA |
782 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
783 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 784 | */ |
c4a77a6c | 785 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 786 | struct wb_writeback_work *work) |
66f3b8e2 | 787 | { |
e98be2d5 | 788 | unsigned long wb_start = jiffies; |
d46db3d5 | 789 | long nr_pages = work->nr_pages; |
03ba3782 | 790 | unsigned long oldest_jif; |
a5989bdc | 791 | struct inode *inode; |
d46db3d5 | 792 | long progress; |
66f3b8e2 | 793 | |
e185dda8 | 794 | oldest_jif = jiffies; |
d46db3d5 | 795 | work->older_than_this = &oldest_jif; |
38f21977 | 796 | |
e8dfc305 | 797 | spin_lock(&wb->list_lock); |
03ba3782 JA |
798 | for (;;) { |
799 | /* | |
d3ddec76 | 800 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 801 | */ |
83ba7b07 | 802 | if (work->nr_pages <= 0) |
03ba3782 | 803 | break; |
66f3b8e2 | 804 | |
aa373cf5 JK |
805 | /* |
806 | * Background writeout and kupdate-style writeback may | |
807 | * run forever. Stop them if there is other work to do | |
808 | * so that e.g. sync can proceed. They'll be restarted | |
809 | * after the other works are all done. | |
810 | */ | |
811 | if ((work->for_background || work->for_kupdate) && | |
812 | !list_empty(&wb->bdi->work_list)) | |
813 | break; | |
814 | ||
38f21977 | 815 | /* |
d3ddec76 WF |
816 | * For background writeout, stop when we are below the |
817 | * background dirty threshold | |
38f21977 | 818 | */ |
b00949aa | 819 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
03ba3782 | 820 | break; |
38f21977 | 821 | |
1bc36b64 JK |
822 | /* |
823 | * Kupdate and background works are special and we want to | |
824 | * include all inodes that need writing. Livelock avoidance is | |
825 | * handled by these works yielding to any other work so we are | |
826 | * safe. | |
827 | */ | |
ba9aa839 WF |
828 | if (work->for_kupdate) { |
829 | oldest_jif = jiffies - | |
830 | msecs_to_jiffies(dirty_expire_interval * 10); | |
1bc36b64 JK |
831 | } else if (work->for_background) |
832 | oldest_jif = jiffies; | |
028c2dd1 | 833 | |
d46db3d5 | 834 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 835 | if (list_empty(&wb->b_io)) |
ad4e38dd | 836 | queue_io(wb, work); |
83ba7b07 | 837 | if (work->sb) |
d46db3d5 | 838 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 839 | else |
d46db3d5 WF |
840 | progress = __writeback_inodes_wb(wb, work); |
841 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 842 | |
e98be2d5 | 843 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
844 | |
845 | /* | |
e6fb6da2 WF |
846 | * Did we write something? Try for more |
847 | * | |
848 | * Dirty inodes are moved to b_io for writeback in batches. | |
849 | * The completion of the current batch does not necessarily | |
850 | * mean the overall work is done. So we keep looping as long | |
851 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 852 | */ |
d46db3d5 | 853 | if (progress) |
71fd05a8 JA |
854 | continue; |
855 | /* | |
e6fb6da2 | 856 | * No more inodes for IO, bail |
71fd05a8 | 857 | */ |
b7a2441f | 858 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 859 | break; |
71fd05a8 JA |
860 | /* |
861 | * Nothing written. Wait for some inode to | |
862 | * become available for writeback. Otherwise | |
863 | * we'll just busyloop. | |
864 | */ | |
71fd05a8 | 865 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 866 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 867 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 868 | spin_lock(&inode->i_lock); |
f0d07b7f | 869 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
870 | /* This function drops i_lock... */ |
871 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 872 | spin_lock(&wb->list_lock); |
03ba3782 JA |
873 | } |
874 | } | |
e8dfc305 | 875 | spin_unlock(&wb->list_lock); |
03ba3782 | 876 | |
d46db3d5 | 877 | return nr_pages - work->nr_pages; |
03ba3782 JA |
878 | } |
879 | ||
880 | /* | |
83ba7b07 | 881 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 882 | */ |
83ba7b07 | 883 | static struct wb_writeback_work * |
08852b6d | 884 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 885 | { |
83ba7b07 | 886 | struct wb_writeback_work *work = NULL; |
03ba3782 | 887 | |
6467716a | 888 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
889 | if (!list_empty(&bdi->work_list)) { |
890 | work = list_entry(bdi->work_list.next, | |
891 | struct wb_writeback_work, list); | |
892 | list_del_init(&work->list); | |
03ba3782 | 893 | } |
6467716a | 894 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 895 | return work; |
03ba3782 JA |
896 | } |
897 | ||
cdf01dd5 LT |
898 | /* |
899 | * Add in the number of potentially dirty inodes, because each inode | |
900 | * write can dirty pagecache in the underlying blockdev. | |
901 | */ | |
902 | static unsigned long get_nr_dirty_pages(void) | |
903 | { | |
904 | return global_page_state(NR_FILE_DIRTY) + | |
905 | global_page_state(NR_UNSTABLE_NFS) + | |
906 | get_nr_dirty_inodes(); | |
907 | } | |
908 | ||
6585027a JK |
909 | static long wb_check_background_flush(struct bdi_writeback *wb) |
910 | { | |
b00949aa | 911 | if (over_bground_thresh(wb->bdi)) { |
6585027a JK |
912 | |
913 | struct wb_writeback_work work = { | |
914 | .nr_pages = LONG_MAX, | |
915 | .sync_mode = WB_SYNC_NONE, | |
916 | .for_background = 1, | |
917 | .range_cyclic = 1, | |
0e175a18 | 918 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
919 | }; |
920 | ||
921 | return wb_writeback(wb, &work); | |
922 | } | |
923 | ||
924 | return 0; | |
925 | } | |
926 | ||
03ba3782 JA |
927 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
928 | { | |
929 | unsigned long expired; | |
930 | long nr_pages; | |
931 | ||
69b62d01 JA |
932 | /* |
933 | * When set to zero, disable periodic writeback | |
934 | */ | |
935 | if (!dirty_writeback_interval) | |
936 | return 0; | |
937 | ||
03ba3782 JA |
938 | expired = wb->last_old_flush + |
939 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
940 | if (time_before(jiffies, expired)) | |
941 | return 0; | |
942 | ||
943 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 944 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 945 | |
c4a77a6c | 946 | if (nr_pages) { |
83ba7b07 | 947 | struct wb_writeback_work work = { |
c4a77a6c JA |
948 | .nr_pages = nr_pages, |
949 | .sync_mode = WB_SYNC_NONE, | |
950 | .for_kupdate = 1, | |
951 | .range_cyclic = 1, | |
0e175a18 | 952 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
953 | }; |
954 | ||
83ba7b07 | 955 | return wb_writeback(wb, &work); |
c4a77a6c | 956 | } |
03ba3782 JA |
957 | |
958 | return 0; | |
959 | } | |
960 | ||
961 | /* | |
962 | * Retrieve work items and do the writeback they describe | |
963 | */ | |
964 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
965 | { | |
966 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 967 | struct wb_writeback_work *work; |
c4a77a6c | 968 | long wrote = 0; |
03ba3782 | 969 | |
81d73a32 | 970 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 971 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
972 | /* |
973 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 974 | * because this thread is exiting now. |
03ba3782 JA |
975 | */ |
976 | if (force_wait) | |
83ba7b07 | 977 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 978 | |
455b2864 DC |
979 | trace_writeback_exec(bdi, work); |
980 | ||
83ba7b07 | 981 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
982 | |
983 | /* | |
83ba7b07 CH |
984 | * Notify the caller of completion if this is a synchronous |
985 | * work item, otherwise just free it. | |
03ba3782 | 986 | */ |
83ba7b07 CH |
987 | if (work->done) |
988 | complete(work->done); | |
989 | else | |
990 | kfree(work); | |
03ba3782 JA |
991 | } |
992 | ||
993 | /* | |
994 | * Check for periodic writeback, kupdated() style | |
995 | */ | |
996 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 997 | wrote += wb_check_background_flush(wb); |
81d73a32 | 998 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
999 | |
1000 | return wrote; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1005 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1006 | */ |
839a8e86 | 1007 | void bdi_writeback_workfn(struct work_struct *work) |
03ba3782 | 1008 | { |
839a8e86 TH |
1009 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1010 | struct bdi_writeback, dwork); | |
08243900 | 1011 | struct backing_dev_info *bdi = wb->bdi; |
03ba3782 JA |
1012 | long pages_written; |
1013 | ||
ef3b1019 | 1014 | set_worker_desc("flush-%s", dev_name(bdi->dev)); |
766f9164 | 1015 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1016 | |
839a8e86 TH |
1017 | if (likely(!current_is_workqueue_rescuer() || |
1018 | list_empty(&bdi->bdi_list))) { | |
6467716a | 1019 | /* |
839a8e86 TH |
1020 | * The normal path. Keep writing back @bdi until its |
1021 | * work_list is empty. Note that this path is also taken | |
1022 | * if @bdi is shutting down even when we're running off the | |
1023 | * rescuer as work_list needs to be drained. | |
6467716a | 1024 | */ |
839a8e86 TH |
1025 | do { |
1026 | pages_written = wb_do_writeback(wb, 0); | |
1027 | trace_writeback_pages_written(pages_written); | |
1028 | } while (!list_empty(&bdi->work_list)); | |
1029 | } else { | |
1030 | /* | |
1031 | * bdi_wq can't get enough workers and we're running off | |
1032 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1033 | * enough for efficient IO. | |
1034 | */ | |
1035 | pages_written = writeback_inodes_wb(&bdi->wb, 1024, | |
1036 | WB_REASON_FORKER_THREAD); | |
455b2864 | 1037 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1038 | } |
1039 | ||
839a8e86 TH |
1040 | if (!list_empty(&bdi->work_list) || |
1041 | (wb_has_dirty_io(wb) && dirty_writeback_interval)) | |
1042 | queue_delayed_work(bdi_wq, &wb->dwork, | |
1043 | msecs_to_jiffies(dirty_writeback_interval * 10)); | |
455b2864 | 1044 | |
839a8e86 | 1045 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1046 | } |
1047 | ||
1048 | /* | |
b8c2f347 CH |
1049 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1050 | * the whole world. | |
03ba3782 | 1051 | */ |
0e175a18 | 1052 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1053 | { |
b8c2f347 | 1054 | struct backing_dev_info *bdi; |
03ba3782 | 1055 | |
83ba7b07 CH |
1056 | if (!nr_pages) { |
1057 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
1058 | global_page_state(NR_UNSTABLE_NFS); |
1059 | } | |
03ba3782 | 1060 | |
b8c2f347 | 1061 | rcu_read_lock(); |
cfc4ba53 | 1062 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1063 | if (!bdi_has_dirty_io(bdi)) |
1064 | continue; | |
0e175a18 | 1065 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
03ba3782 | 1066 | } |
cfc4ba53 | 1067 | rcu_read_unlock(); |
1da177e4 LT |
1068 | } |
1069 | ||
03ba3782 JA |
1070 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1071 | { | |
1072 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1073 | struct dentry *dentry; | |
1074 | const char *name = "?"; | |
1075 | ||
1076 | dentry = d_find_alias(inode); | |
1077 | if (dentry) { | |
1078 | spin_lock(&dentry->d_lock); | |
1079 | name = (const char *) dentry->d_name.name; | |
1080 | } | |
1081 | printk(KERN_DEBUG | |
1082 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1083 | current->comm, task_pid_nr(current), inode->i_ino, | |
1084 | name, inode->i_sb->s_id); | |
1085 | if (dentry) { | |
1086 | spin_unlock(&dentry->d_lock); | |
1087 | dput(dentry); | |
1088 | } | |
1089 | } | |
1090 | } | |
1091 | ||
1092 | /** | |
1093 | * __mark_inode_dirty - internal function | |
1094 | * @inode: inode to mark | |
1095 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1096 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1097 | * mark_inode_dirty_sync. | |
1da177e4 | 1098 | * |
03ba3782 JA |
1099 | * Put the inode on the super block's dirty list. |
1100 | * | |
1101 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1102 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1103 | * If it was not hashed, it will never be added to the dirty list | |
1104 | * even if it is later hashed, as it will have been marked dirty already. | |
1105 | * | |
1106 | * In short, make sure you hash any inodes _before_ you start marking | |
1107 | * them dirty. | |
1da177e4 | 1108 | * |
03ba3782 JA |
1109 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1110 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1111 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1112 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1113 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1114 | * blockdev inode. | |
1da177e4 | 1115 | */ |
03ba3782 | 1116 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1117 | { |
03ba3782 | 1118 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1119 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1120 | |
03ba3782 JA |
1121 | /* |
1122 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1123 | * dirty the inode itself | |
1124 | */ | |
1125 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
9fb0a7da TH |
1126 | trace_writeback_dirty_inode_start(inode, flags); |
1127 | ||
03ba3782 | 1128 | if (sb->s_op->dirty_inode) |
aa385729 | 1129 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1130 | |
1131 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 JA |
1132 | } |
1133 | ||
1134 | /* | |
1135 | * make sure that changes are seen by all cpus before we test i_state | |
1136 | * -- mikulas | |
1137 | */ | |
1138 | smp_mb(); | |
1139 | ||
1140 | /* avoid the locking if we can */ | |
1141 | if ((inode->i_state & flags) == flags) | |
1142 | return; | |
1143 | ||
1144 | if (unlikely(block_dump)) | |
1145 | block_dump___mark_inode_dirty(inode); | |
1146 | ||
250df6ed | 1147 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1148 | if ((inode->i_state & flags) != flags) { |
1149 | const int was_dirty = inode->i_state & I_DIRTY; | |
1150 | ||
1151 | inode->i_state |= flags; | |
1152 | ||
1153 | /* | |
1154 | * If the inode is being synced, just update its dirty state. | |
1155 | * The unlocker will place the inode on the appropriate | |
1156 | * superblock list, based upon its state. | |
1157 | */ | |
1158 | if (inode->i_state & I_SYNC) | |
250df6ed | 1159 | goto out_unlock_inode; |
03ba3782 JA |
1160 | |
1161 | /* | |
1162 | * Only add valid (hashed) inodes to the superblock's | |
1163 | * dirty list. Add blockdev inodes as well. | |
1164 | */ | |
1165 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1166 | if (inode_unhashed(inode)) |
250df6ed | 1167 | goto out_unlock_inode; |
03ba3782 | 1168 | } |
a4ffdde6 | 1169 | if (inode->i_state & I_FREEING) |
250df6ed | 1170 | goto out_unlock_inode; |
03ba3782 JA |
1171 | |
1172 | /* | |
1173 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1174 | * reposition it (that would break b_dirty time-ordering). | |
1175 | */ | |
1176 | if (!was_dirty) { | |
a66979ab | 1177 | bool wakeup_bdi = false; |
253c34e9 AB |
1178 | bdi = inode_to_bdi(inode); |
1179 | ||
1180 | if (bdi_cap_writeback_dirty(bdi)) { | |
1181 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1182 | "bdi-%s not registered\n", bdi->name); | |
1183 | ||
1184 | /* | |
1185 | * If this is the first dirty inode for this | |
1186 | * bdi, we have to wake-up the corresponding | |
1187 | * bdi thread to make sure background | |
1188 | * write-back happens later. | |
1189 | */ | |
1190 | if (!wb_has_dirty_io(&bdi->wb)) | |
1191 | wakeup_bdi = true; | |
500b067c | 1192 | } |
03ba3782 | 1193 | |
a66979ab | 1194 | spin_unlock(&inode->i_lock); |
f758eeab | 1195 | spin_lock(&bdi->wb.list_lock); |
03ba3782 | 1196 | inode->dirtied_when = jiffies; |
7ccf19a8 | 1197 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1198 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1199 | |
1200 | if (wakeup_bdi) | |
1201 | bdi_wakeup_thread_delayed(bdi); | |
1202 | return; | |
1da177e4 | 1203 | } |
1da177e4 | 1204 | } |
250df6ed DC |
1205 | out_unlock_inode: |
1206 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1207 | |
03ba3782 JA |
1208 | } |
1209 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1210 | ||
b6e51316 | 1211 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1212 | { |
1213 | struct inode *inode, *old_inode = NULL; | |
1214 | ||
1215 | /* | |
1216 | * We need to be protected against the filesystem going from | |
1217 | * r/o to r/w or vice versa. | |
1218 | */ | |
b6e51316 | 1219 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1220 | |
55fa6091 | 1221 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1222 | |
1223 | /* | |
1224 | * Data integrity sync. Must wait for all pages under writeback, | |
1225 | * because there may have been pages dirtied before our sync | |
1226 | * call, but which had writeout started before we write it out. | |
1227 | * In which case, the inode may not be on the dirty list, but | |
1228 | * we still have to wait for that writeout. | |
1229 | */ | |
b6e51316 | 1230 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1231 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1232 | |
250df6ed DC |
1233 | spin_lock(&inode->i_lock); |
1234 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1235 | (mapping->nrpages == 0)) { | |
1236 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1237 | continue; |
250df6ed | 1238 | } |
03ba3782 | 1239 | __iget(inode); |
250df6ed | 1240 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1241 | spin_unlock(&inode_sb_list_lock); |
1242 | ||
03ba3782 | 1243 | /* |
55fa6091 DC |
1244 | * We hold a reference to 'inode' so it couldn't have been |
1245 | * removed from s_inodes list while we dropped the | |
1246 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1247 | * be holding the last reference and we cannot iput it under | |
1248 | * inode_sb_list_lock. So we keep the reference and iput it | |
1249 | * later. | |
03ba3782 JA |
1250 | */ |
1251 | iput(old_inode); | |
1252 | old_inode = inode; | |
1253 | ||
1254 | filemap_fdatawait(mapping); | |
1255 | ||
1256 | cond_resched(); | |
1257 | ||
55fa6091 | 1258 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1259 | } |
55fa6091 | 1260 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1261 | iput(old_inode); |
1da177e4 LT |
1262 | } |
1263 | ||
d8a8559c | 1264 | /** |
3259f8be | 1265 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1266 | * @sb: the superblock |
3259f8be | 1267 | * @nr: the number of pages to write |
786228ab | 1268 | * @reason: reason why some writeback work initiated |
1da177e4 | 1269 | * |
d8a8559c JA |
1270 | * Start writeback on some inodes on this super_block. No guarantees are made |
1271 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1272 | * for IO completion of submitted IO. |
1da177e4 | 1273 | */ |
0e175a18 CW |
1274 | void writeback_inodes_sb_nr(struct super_block *sb, |
1275 | unsigned long nr, | |
1276 | enum wb_reason reason) | |
1da177e4 | 1277 | { |
83ba7b07 CH |
1278 | DECLARE_COMPLETION_ONSTACK(done); |
1279 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1280 | .sb = sb, |
1281 | .sync_mode = WB_SYNC_NONE, | |
1282 | .tagged_writepages = 1, | |
1283 | .done = &done, | |
1284 | .nr_pages = nr, | |
0e175a18 | 1285 | .reason = reason, |
3c4d7165 | 1286 | }; |
d8a8559c | 1287 | |
6eedc701 JK |
1288 | if (sb->s_bdi == &noop_backing_dev_info) |
1289 | return; | |
cf37e972 | 1290 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1291 | bdi_queue_work(sb->s_bdi, &work); |
1292 | wait_for_completion(&done); | |
e913fc82 | 1293 | } |
3259f8be CM |
1294 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1295 | ||
1296 | /** | |
1297 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1298 | * @sb: the superblock | |
786228ab | 1299 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1300 | * |
1301 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1302 | * on how many (if any) will be written, and this function does not wait | |
1303 | * for IO completion of submitted IO. | |
1304 | */ | |
0e175a18 | 1305 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1306 | { |
0e175a18 | 1307 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1308 | } |
0e3c9a22 | 1309 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1310 | |
17bd55d0 | 1311 | /** |
10ee27a0 | 1312 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1313 | * @sb: the superblock |
10ee27a0 MX |
1314 | * @nr: the number of pages to write |
1315 | * @reason: the reason of writeback | |
17bd55d0 | 1316 | * |
10ee27a0 | 1317 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1318 | * Returns 1 if writeback was started, 0 if not. |
1319 | */ | |
10ee27a0 MX |
1320 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
1321 | unsigned long nr, | |
1322 | enum wb_reason reason) | |
17bd55d0 | 1323 | { |
10ee27a0 | 1324 | if (writeback_in_progress(sb->s_bdi)) |
17bd55d0 | 1325 | return 1; |
10ee27a0 MX |
1326 | |
1327 | if (!down_read_trylock(&sb->s_umount)) | |
17bd55d0 | 1328 | return 0; |
10ee27a0 MX |
1329 | |
1330 | writeback_inodes_sb_nr(sb, nr, reason); | |
1331 | up_read(&sb->s_umount); | |
1332 | return 1; | |
17bd55d0 | 1333 | } |
10ee27a0 | 1334 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1335 | |
3259f8be | 1336 | /** |
10ee27a0 | 1337 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1338 | * @sb: the superblock |
786228ab | 1339 | * @reason: reason why some writeback work was initiated |
3259f8be | 1340 | * |
10ee27a0 | 1341 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1342 | * Returns 1 if writeback was started, 0 if not. |
1343 | */ | |
10ee27a0 | 1344 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1345 | { |
10ee27a0 | 1346 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1347 | } |
10ee27a0 | 1348 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1349 | |
d8a8559c JA |
1350 | /** |
1351 | * sync_inodes_sb - sync sb inode pages | |
1352 | * @sb: the superblock | |
1353 | * | |
1354 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1355 | * super_block. |
d8a8559c | 1356 | */ |
b6e51316 | 1357 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1358 | { |
83ba7b07 CH |
1359 | DECLARE_COMPLETION_ONSTACK(done); |
1360 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1361 | .sb = sb, |
1362 | .sync_mode = WB_SYNC_ALL, | |
1363 | .nr_pages = LONG_MAX, | |
1364 | .range_cyclic = 0, | |
83ba7b07 | 1365 | .done = &done, |
0e175a18 | 1366 | .reason = WB_REASON_SYNC, |
3c4d7165 CH |
1367 | }; |
1368 | ||
6eedc701 JK |
1369 | /* Nothing to do? */ |
1370 | if (sb->s_bdi == &noop_backing_dev_info) | |
1371 | return; | |
cf37e972 CH |
1372 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1373 | ||
83ba7b07 CH |
1374 | bdi_queue_work(sb->s_bdi, &work); |
1375 | wait_for_completion(&done); | |
1376 | ||
b6e51316 | 1377 | wait_sb_inodes(sb); |
1da177e4 | 1378 | } |
d8a8559c | 1379 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1380 | |
1da177e4 | 1381 | /** |
7f04c26d AA |
1382 | * write_inode_now - write an inode to disk |
1383 | * @inode: inode to write to disk | |
1384 | * @sync: whether the write should be synchronous or not | |
1385 | * | |
1386 | * This function commits an inode to disk immediately if it is dirty. This is | |
1387 | * primarily needed by knfsd. | |
1da177e4 | 1388 | * |
7f04c26d | 1389 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1390 | */ |
1da177e4 LT |
1391 | int write_inode_now(struct inode *inode, int sync) |
1392 | { | |
f758eeab | 1393 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1394 | struct writeback_control wbc = { |
1395 | .nr_to_write = LONG_MAX, | |
18914b18 | 1396 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1397 | .range_start = 0, |
1398 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1399 | }; |
1400 | ||
1401 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1402 | wbc.nr_to_write = 0; |
1da177e4 LT |
1403 | |
1404 | might_sleep(); | |
4f8ad655 | 1405 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1406 | } |
1407 | EXPORT_SYMBOL(write_inode_now); | |
1408 | ||
1409 | /** | |
1410 | * sync_inode - write an inode and its pages to disk. | |
1411 | * @inode: the inode to sync | |
1412 | * @wbc: controls the writeback mode | |
1413 | * | |
1414 | * sync_inode() will write an inode and its pages to disk. It will also | |
1415 | * correctly update the inode on its superblock's dirty inode lists and will | |
1416 | * update inode->i_state. | |
1417 | * | |
1418 | * The caller must have a ref on the inode. | |
1419 | */ | |
1420 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1421 | { | |
4f8ad655 | 1422 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1423 | } |
1424 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1425 | |
1426 | /** | |
c691b9d9 | 1427 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1428 | * @inode: the inode to sync |
1429 | * @wait: wait for I/O to complete. | |
1430 | * | |
c691b9d9 | 1431 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1432 | * |
1433 | * Note: only writes the actual inode, no associated data or other metadata. | |
1434 | */ | |
1435 | int sync_inode_metadata(struct inode *inode, int wait) | |
1436 | { | |
1437 | struct writeback_control wbc = { | |
1438 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1439 | .nr_to_write = 0, /* metadata-only */ | |
1440 | }; | |
1441 | ||
1442 | return sync_inode(inode, &wbc); | |
1443 | } | |
1444 | EXPORT_SYMBOL(sync_inode_metadata); |