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; | |
d46db3d5 | 38 | unsigned long *older_than_this; |
c4a77a6c | 39 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 40 | unsigned int tagged_writepages:1; |
52957fe1 HS |
41 | unsigned int for_kupdate:1; |
42 | unsigned int range_cyclic:1; | |
43 | unsigned int for_background:1; | |
0e175a18 | 44 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 45 | |
8010c3b6 | 46 | struct list_head list; /* pending work list */ |
83ba7b07 | 47 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
48 | }; |
49 | ||
455b2864 DC |
50 | /* |
51 | * Include the creation of the trace points after defining the | |
52 | * wb_writeback_work structure so that the definition remains local to this | |
53 | * file. | |
54 | */ | |
55 | #define CREATE_TRACE_POINTS | |
56 | #include <trace/events/writeback.h> | |
57 | ||
455b2864 DC |
58 | /* |
59 | * We don't actually have pdflush, but this one is exported though /proc... | |
60 | */ | |
61 | int nr_pdflush_threads; | |
62 | ||
f11b00f3 AB |
63 | /** |
64 | * writeback_in_progress - determine whether there is writeback in progress | |
65 | * @bdi: the device's backing_dev_info structure. | |
66 | * | |
03ba3782 JA |
67 | * Determine whether there is writeback waiting to be handled against a |
68 | * backing device. | |
f11b00f3 AB |
69 | */ |
70 | int writeback_in_progress(struct backing_dev_info *bdi) | |
71 | { | |
81d73a32 | 72 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 AB |
73 | } |
74 | ||
692ebd17 JK |
75 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
76 | { | |
77 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 78 | |
aaead25b CH |
79 | if (strcmp(sb->s_type->name, "bdev") == 0) |
80 | return inode->i_mapping->backing_dev_info; | |
81 | ||
82 | return sb->s_bdi; | |
692ebd17 JK |
83 | } |
84 | ||
7ccf19a8 NP |
85 | static inline struct inode *wb_inode(struct list_head *head) |
86 | { | |
87 | return list_entry(head, struct inode, i_wb_list); | |
88 | } | |
89 | ||
6585027a JK |
90 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
91 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) | |
03ba3782 | 92 | { |
fff5b85a AB |
93 | if (bdi->wb.task) { |
94 | wake_up_process(bdi->wb.task); | |
95 | } else { | |
96 | /* | |
97 | * The bdi thread isn't there, wake up the forker thread which | |
98 | * will create and run it. | |
99 | */ | |
03ba3782 | 100 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 101 | } |
6585027a JK |
102 | } |
103 | ||
104 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
105 | struct wb_writeback_work *work) | |
106 | { | |
107 | trace_writeback_queue(bdi, work); | |
108 | ||
109 | spin_lock_bh(&bdi->wb_lock); | |
110 | list_add_tail(&work->list, &bdi->work_list); | |
111 | if (!bdi->wb.task) | |
112 | trace_writeback_nothread(bdi, work); | |
113 | bdi_wakeup_flusher(bdi); | |
6467716a | 114 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
115 | } |
116 | ||
83ba7b07 CH |
117 | static void |
118 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
0e175a18 | 119 | bool range_cyclic, enum wb_reason reason) |
1da177e4 | 120 | { |
83ba7b07 | 121 | struct wb_writeback_work *work; |
03ba3782 | 122 | |
bcddc3f0 JA |
123 | /* |
124 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
125 | * wakeup the thread for old dirty data writeback | |
126 | */ | |
83ba7b07 CH |
127 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
128 | if (!work) { | |
455b2864 DC |
129 | if (bdi->wb.task) { |
130 | trace_writeback_nowork(bdi); | |
83ba7b07 | 131 | wake_up_process(bdi->wb.task); |
455b2864 | 132 | } |
83ba7b07 | 133 | return; |
bcddc3f0 | 134 | } |
03ba3782 | 135 | |
83ba7b07 CH |
136 | work->sync_mode = WB_SYNC_NONE; |
137 | work->nr_pages = nr_pages; | |
138 | work->range_cyclic = range_cyclic; | |
0e175a18 | 139 | work->reason = reason; |
03ba3782 | 140 | |
83ba7b07 | 141 | bdi_queue_work(bdi, work); |
b6e51316 JA |
142 | } |
143 | ||
144 | /** | |
145 | * bdi_start_writeback - start writeback | |
146 | * @bdi: the backing device to write from | |
147 | * @nr_pages: the number of pages to write | |
786228ab | 148 | * @reason: reason why some writeback work was initiated |
b6e51316 JA |
149 | * |
150 | * Description: | |
151 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 152 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 153 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
154 | * |
155 | */ | |
0e175a18 CW |
156 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
157 | enum wb_reason reason) | |
b6e51316 | 158 | { |
0e175a18 | 159 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
c5444198 | 160 | } |
d3ddec76 | 161 | |
c5444198 CH |
162 | /** |
163 | * bdi_start_background_writeback - start background writeback | |
164 | * @bdi: the backing device to write from | |
165 | * | |
166 | * Description: | |
6585027a JK |
167 | * This makes sure WB_SYNC_NONE background writeback happens. When |
168 | * this function returns, it is only guaranteed that for given BDI | |
169 | * some IO is happening if we are over background dirty threshold. | |
170 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
171 | */ |
172 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
173 | { | |
6585027a JK |
174 | /* |
175 | * We just wake up the flusher thread. It will perform background | |
176 | * writeback as soon as there is no other work to do. | |
177 | */ | |
71927e84 | 178 | trace_writeback_wake_background(bdi); |
6585027a JK |
179 | spin_lock_bh(&bdi->wb_lock); |
180 | bdi_wakeup_flusher(bdi); | |
181 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
182 | } |
183 | ||
a66979ab DC |
184 | /* |
185 | * Remove the inode from the writeback list it is on. | |
186 | */ | |
187 | void inode_wb_list_del(struct inode *inode) | |
188 | { | |
f758eeab CH |
189 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
190 | ||
191 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 192 | list_del_init(&inode->i_wb_list); |
f758eeab | 193 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
194 | } |
195 | ||
6610a0bc AM |
196 | /* |
197 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
198 | * furthest end of its superblock's dirty-inode list. | |
199 | * | |
200 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 201 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
202 | * the case then the inode must have been redirtied while it was being written |
203 | * out and we don't reset its dirtied_when. | |
204 | */ | |
f758eeab | 205 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 206 | { |
f758eeab | 207 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 208 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 209 | struct inode *tail; |
6610a0bc | 210 | |
7ccf19a8 | 211 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 212 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
213 | inode->dirtied_when = jiffies; |
214 | } | |
7ccf19a8 | 215 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
216 | } |
217 | ||
c986d1e2 | 218 | /* |
66f3b8e2 | 219 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 220 | */ |
f758eeab | 221 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 222 | { |
f758eeab | 223 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 224 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
225 | } |
226 | ||
1c0eeaf5 JE |
227 | static void inode_sync_complete(struct inode *inode) |
228 | { | |
229 | /* | |
a66979ab | 230 | * Prevent speculative execution through |
f758eeab | 231 | * spin_unlock(&wb->list_lock); |
1c0eeaf5 | 232 | */ |
a66979ab | 233 | |
1c0eeaf5 JE |
234 | smp_mb(); |
235 | wake_up_bit(&inode->i_state, __I_SYNC); | |
236 | } | |
237 | ||
d2caa3c5 JL |
238 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
239 | { | |
240 | bool ret = time_after(inode->dirtied_when, t); | |
241 | #ifndef CONFIG_64BIT | |
242 | /* | |
243 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
244 | * It _appears_ to be in the future, but is actually in distant past. | |
245 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 246 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
247 | */ |
248 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
249 | #endif | |
250 | return ret; | |
251 | } | |
252 | ||
2c136579 FW |
253 | /* |
254 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
255 | */ | |
e84d0a4f | 256 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 257 | struct list_head *dispatch_queue, |
ad4e38dd | 258 | struct wb_writeback_work *work) |
2c136579 | 259 | { |
5c03449d SL |
260 | LIST_HEAD(tmp); |
261 | struct list_head *pos, *node; | |
cf137307 | 262 | struct super_block *sb = NULL; |
5c03449d | 263 | struct inode *inode; |
cf137307 | 264 | int do_sb_sort = 0; |
e84d0a4f | 265 | int moved = 0; |
5c03449d | 266 | |
2c136579 | 267 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 268 | inode = wb_inode(delaying_queue->prev); |
ad4e38dd CW |
269 | if (work->older_than_this && |
270 | inode_dirtied_after(inode, *work->older_than_this)) | |
2c136579 | 271 | break; |
cf137307 JA |
272 | if (sb && sb != inode->i_sb) |
273 | do_sb_sort = 1; | |
274 | sb = inode->i_sb; | |
7ccf19a8 | 275 | list_move(&inode->i_wb_list, &tmp); |
e84d0a4f | 276 | moved++; |
5c03449d SL |
277 | } |
278 | ||
cf137307 JA |
279 | /* just one sb in list, splice to dispatch_queue and we're done */ |
280 | if (!do_sb_sort) { | |
281 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 282 | goto out; |
cf137307 JA |
283 | } |
284 | ||
5c03449d SL |
285 | /* Move inodes from one superblock together */ |
286 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 287 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 288 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 289 | inode = wb_inode(pos); |
5c03449d | 290 | if (inode->i_sb == sb) |
7ccf19a8 | 291 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 292 | } |
2c136579 | 293 | } |
e84d0a4f WF |
294 | out: |
295 | return moved; | |
2c136579 FW |
296 | } |
297 | ||
298 | /* | |
299 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
300 | * Before |
301 | * newly dirtied b_dirty b_io b_more_io | |
302 | * =============> gf edc BA | |
303 | * After | |
304 | * newly dirtied b_dirty b_io b_more_io | |
305 | * =============> g fBAedc | |
306 | * | | |
307 | * +--> dequeue for IO | |
2c136579 | 308 | */ |
ad4e38dd | 309 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 310 | { |
e84d0a4f | 311 | int moved; |
f758eeab | 312 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 313 | list_splice_init(&wb->b_more_io, &wb->b_io); |
ad4e38dd CW |
314 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
315 | trace_writeback_queue_io(wb, work, moved); | |
66f3b8e2 JA |
316 | } |
317 | ||
a9185b41 | 318 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 319 | { |
03ba3782 | 320 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 321 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 322 | return 0; |
08d8e974 | 323 | } |
08d8e974 | 324 | |
1da177e4 | 325 | /* |
01c03194 CH |
326 | * Wait for writeback on an inode to complete. |
327 | */ | |
f758eeab CH |
328 | static void inode_wait_for_writeback(struct inode *inode, |
329 | struct bdi_writeback *wb) | |
01c03194 CH |
330 | { |
331 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
332 | wait_queue_head_t *wqh; | |
333 | ||
334 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
335 | while (inode->i_state & I_SYNC) { |
336 | spin_unlock(&inode->i_lock); | |
f758eeab | 337 | spin_unlock(&wb->list_lock); |
01c03194 | 338 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
f758eeab | 339 | spin_lock(&wb->list_lock); |
250df6ed | 340 | spin_lock(&inode->i_lock); |
58a9d3d8 | 341 | } |
01c03194 CH |
342 | } |
343 | ||
344 | /* | |
f758eeab | 345 | * Write out an inode's dirty pages. Called under wb->list_lock and |
0f1b1fd8 DC |
346 | * inode->i_lock. Either the caller has an active reference on the inode or |
347 | * the inode has I_WILL_FREE set. | |
01c03194 | 348 | * |
1da177e4 LT |
349 | * If `wait' is set, wait on the writeout. |
350 | * | |
351 | * The whole writeout design is quite complex and fragile. We want to avoid | |
352 | * starvation of particular inodes when others are being redirtied, prevent | |
353 | * livelocks, etc. | |
1da177e4 LT |
354 | */ |
355 | static int | |
f758eeab CH |
356 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
357 | struct writeback_control *wbc) | |
1da177e4 | 358 | { |
1da177e4 | 359 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 360 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 361 | unsigned dirty; |
1da177e4 LT |
362 | int ret; |
363 | ||
f758eeab | 364 | assert_spin_locked(&wb->list_lock); |
0f1b1fd8 DC |
365 | assert_spin_locked(&inode->i_lock); |
366 | ||
01c03194 CH |
367 | if (!atomic_read(&inode->i_count)) |
368 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
369 | else | |
370 | WARN_ON(inode->i_state & I_WILL_FREE); | |
371 | ||
372 | if (inode->i_state & I_SYNC) { | |
373 | /* | |
374 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 375 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
376 | * writeback can proceed with the other inodes on s_io. |
377 | * | |
378 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 379 | * completed a full scan of b_io. |
01c03194 | 380 | */ |
a9185b41 | 381 | if (wbc->sync_mode != WB_SYNC_ALL) { |
f758eeab | 382 | requeue_io(inode, wb); |
251d6a47 WF |
383 | trace_writeback_single_inode_requeue(inode, wbc, |
384 | nr_to_write); | |
01c03194 CH |
385 | return 0; |
386 | } | |
387 | ||
388 | /* | |
389 | * It's a data-integrity sync. We must wait. | |
390 | */ | |
f758eeab | 391 | inode_wait_for_writeback(inode, wb); |
01c03194 CH |
392 | } |
393 | ||
1c0eeaf5 | 394 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 395 | |
5547e8aa | 396 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 397 | inode->i_state |= I_SYNC; |
5547e8aa | 398 | inode->i_state &= ~I_DIRTY_PAGES; |
250df6ed | 399 | spin_unlock(&inode->i_lock); |
f758eeab | 400 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
401 | |
402 | ret = do_writepages(mapping, wbc); | |
403 | ||
26821ed4 CH |
404 | /* |
405 | * Make sure to wait on the data before writing out the metadata. | |
406 | * This is important for filesystems that modify metadata on data | |
407 | * I/O completion. | |
408 | */ | |
a9185b41 | 409 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 410 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
411 | if (ret == 0) |
412 | ret = err; | |
413 | } | |
414 | ||
5547e8aa DM |
415 | /* |
416 | * Some filesystems may redirty the inode during the writeback | |
417 | * due to delalloc, clear dirty metadata flags right before | |
418 | * write_inode() | |
419 | */ | |
250df6ed | 420 | spin_lock(&inode->i_lock); |
5547e8aa DM |
421 | dirty = inode->i_state & I_DIRTY; |
422 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 423 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
424 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
425 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 426 | int err = write_inode(inode, wbc); |
1da177e4 LT |
427 | if (ret == 0) |
428 | ret = err; | |
429 | } | |
430 | ||
f758eeab | 431 | spin_lock(&wb->list_lock); |
250df6ed | 432 | spin_lock(&inode->i_lock); |
1c0eeaf5 | 433 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 434 | if (!(inode->i_state & I_FREEING)) { |
94c3dcbb WF |
435 | /* |
436 | * Sync livelock prevention. Each inode is tagged and synced in | |
437 | * one shot. If still dirty, it will be redirty_tail()'ed below. | |
438 | * Update the dirty time to prevent enqueue and sync it again. | |
439 | */ | |
440 | if ((inode->i_state & I_DIRTY) && | |
441 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
442 | inode->dirtied_when = jiffies; | |
443 | ||
23539afc | 444 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
445 | /* |
446 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 447 | * sometimes bales out without doing anything. |
1b43ef91 | 448 | */ |
a50aeb40 WF |
449 | inode->i_state |= I_DIRTY_PAGES; |
450 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 451 | /* |
a50aeb40 | 452 | * slice used up: queue for next turn |
1da177e4 | 453 | */ |
f758eeab | 454 | requeue_io(inode, wb); |
1da177e4 LT |
455 | } else { |
456 | /* | |
a50aeb40 WF |
457 | * Writeback blocked by something other than |
458 | * congestion. Delay the inode for some time to | |
459 | * avoid spinning on the CPU (100% iowait) | |
460 | * retrying writeback of the dirty page/inode | |
461 | * that cannot be performed immediately. | |
1da177e4 | 462 | */ |
f758eeab | 463 | redirty_tail(inode, wb); |
1da177e4 | 464 | } |
23539afc WF |
465 | } else if (inode->i_state & I_DIRTY) { |
466 | /* | |
467 | * Filesystems can dirty the inode during writeback | |
468 | * operations, such as delayed allocation during | |
469 | * submission or metadata updates after data IO | |
470 | * completion. | |
471 | */ | |
f758eeab | 472 | redirty_tail(inode, wb); |
1da177e4 LT |
473 | } else { |
474 | /* | |
9e38d86f NP |
475 | * The inode is clean. At this point we either have |
476 | * a reference to the inode or it's on it's way out. | |
477 | * No need to add it back to the LRU. | |
1da177e4 | 478 | */ |
7ccf19a8 | 479 | list_del_init(&inode->i_wb_list); |
1da177e4 LT |
480 | } |
481 | } | |
1c0eeaf5 | 482 | inode_sync_complete(inode); |
251d6a47 | 483 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
1da177e4 LT |
484 | return ret; |
485 | } | |
486 | ||
1a12d8bd WF |
487 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
488 | struct wb_writeback_work *work) | |
d46db3d5 WF |
489 | { |
490 | long pages; | |
491 | ||
492 | /* | |
493 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
494 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
495 | * here avoids calling into writeback_inodes_wb() more than once. | |
496 | * | |
497 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
498 | * | |
499 | * wb_writeback() | |
500 | * writeback_sb_inodes() <== called only once | |
501 | * write_cache_pages() <== called once for each inode | |
502 | * (quickly) tag currently dirty pages | |
503 | * (maybe slowly) sync all tagged pages | |
504 | */ | |
505 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
506 | pages = LONG_MAX; | |
1a12d8bd WF |
507 | else { |
508 | pages = min(bdi->avg_write_bandwidth / 2, | |
509 | global_dirty_limit / DIRTY_SCOPE); | |
510 | pages = min(pages, work->nr_pages); | |
511 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
512 | MIN_WRITEBACK_PAGES); | |
513 | } | |
d46db3d5 WF |
514 | |
515 | return pages; | |
516 | } | |
517 | ||
f11c9c5c ES |
518 | /* |
519 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
520 | * |
521 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
522 | * inodes. Otherwise write only ones which go sequentially |
523 | * in reverse order. | |
edadfb10 | 524 | * |
d46db3d5 | 525 | * Return the number of pages and/or inodes written. |
f11c9c5c | 526 | */ |
d46db3d5 WF |
527 | static long writeback_sb_inodes(struct super_block *sb, |
528 | struct bdi_writeback *wb, | |
529 | struct wb_writeback_work *work) | |
1da177e4 | 530 | { |
d46db3d5 WF |
531 | struct writeback_control wbc = { |
532 | .sync_mode = work->sync_mode, | |
533 | .tagged_writepages = work->tagged_writepages, | |
534 | .for_kupdate = work->for_kupdate, | |
535 | .for_background = work->for_background, | |
536 | .range_cyclic = work->range_cyclic, | |
537 | .range_start = 0, | |
538 | .range_end = LLONG_MAX, | |
539 | }; | |
540 | unsigned long start_time = jiffies; | |
541 | long write_chunk; | |
542 | long wrote = 0; /* count both pages and inodes */ | |
543 | ||
03ba3782 | 544 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 545 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
546 | |
547 | if (inode->i_sb != sb) { | |
d46db3d5 | 548 | if (work->sb) { |
edadfb10 CH |
549 | /* |
550 | * We only want to write back data for this | |
551 | * superblock, move all inodes not belonging | |
552 | * to it back onto the dirty list. | |
553 | */ | |
f758eeab | 554 | redirty_tail(inode, wb); |
edadfb10 CH |
555 | continue; |
556 | } | |
557 | ||
558 | /* | |
559 | * The inode belongs to a different superblock. | |
560 | * Bounce back to the caller to unpin this and | |
561 | * pin the next superblock. | |
562 | */ | |
d46db3d5 | 563 | break; |
edadfb10 CH |
564 | } |
565 | ||
9843b76a CH |
566 | /* |
567 | * Don't bother with new inodes or inodes beeing freed, first | |
568 | * kind does not need peridic writeout yet, and for the latter | |
569 | * kind writeout is handled by the freer. | |
570 | */ | |
250df6ed | 571 | spin_lock(&inode->i_lock); |
9843b76a | 572 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 573 | spin_unlock(&inode->i_lock); |
fcc5c222 | 574 | redirty_tail(inode, wb); |
7ef0d737 NP |
575 | continue; |
576 | } | |
1da177e4 | 577 | __iget(inode); |
1a12d8bd | 578 | write_chunk = writeback_chunk_size(wb->bdi, work); |
d46db3d5 WF |
579 | wbc.nr_to_write = write_chunk; |
580 | wbc.pages_skipped = 0; | |
250df6ed | 581 | |
d46db3d5 | 582 | writeback_single_inode(inode, wb, &wbc); |
250df6ed | 583 | |
d46db3d5 WF |
584 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
585 | wrote += write_chunk - wbc.nr_to_write; | |
586 | if (!(inode->i_state & I_DIRTY)) | |
587 | wrote++; | |
588 | if (wbc.pages_skipped) { | |
1da177e4 LT |
589 | /* |
590 | * writeback is not making progress due to locked | |
591 | * buffers. Skip this inode for now. | |
592 | */ | |
f758eeab | 593 | redirty_tail(inode, wb); |
1da177e4 | 594 | } |
0f1b1fd8 | 595 | spin_unlock(&inode->i_lock); |
f758eeab | 596 | spin_unlock(&wb->list_lock); |
1da177e4 | 597 | iput(inode); |
4ffc8444 | 598 | cond_resched(); |
f758eeab | 599 | spin_lock(&wb->list_lock); |
d46db3d5 WF |
600 | /* |
601 | * bail out to wb_writeback() often enough to check | |
602 | * background threshold and other termination conditions. | |
603 | */ | |
604 | if (wrote) { | |
605 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
606 | break; | |
607 | if (work->nr_pages <= 0) | |
608 | break; | |
8bc3be27 | 609 | } |
1da177e4 | 610 | } |
d46db3d5 | 611 | return wrote; |
f11c9c5c ES |
612 | } |
613 | ||
d46db3d5 WF |
614 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
615 | struct wb_writeback_work *work) | |
f11c9c5c | 616 | { |
d46db3d5 WF |
617 | unsigned long start_time = jiffies; |
618 | long wrote = 0; | |
38f21977 | 619 | |
f11c9c5c | 620 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 621 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 622 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 623 | |
12ad3ab6 | 624 | if (!grab_super_passive(sb)) { |
0e995816 WF |
625 | /* |
626 | * grab_super_passive() may fail consistently due to | |
627 | * s_umount being grabbed by someone else. Don't use | |
628 | * requeue_io() to avoid busy retrying the inode/sb. | |
629 | */ | |
630 | redirty_tail(inode, wb); | |
edadfb10 | 631 | continue; |
f11c9c5c | 632 | } |
d46db3d5 | 633 | wrote += writeback_sb_inodes(sb, wb, work); |
edadfb10 | 634 | drop_super(sb); |
f11c9c5c | 635 | |
d46db3d5 WF |
636 | /* refer to the same tests at the end of writeback_sb_inodes */ |
637 | if (wrote) { | |
638 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
639 | break; | |
640 | if (work->nr_pages <= 0) | |
641 | break; | |
642 | } | |
f11c9c5c | 643 | } |
66f3b8e2 | 644 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 645 | return wrote; |
66f3b8e2 JA |
646 | } |
647 | ||
0e175a18 CW |
648 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
649 | enum wb_reason reason) | |
edadfb10 | 650 | { |
d46db3d5 WF |
651 | struct wb_writeback_work work = { |
652 | .nr_pages = nr_pages, | |
653 | .sync_mode = WB_SYNC_NONE, | |
654 | .range_cyclic = 1, | |
0e175a18 | 655 | .reason = reason, |
d46db3d5 | 656 | }; |
edadfb10 | 657 | |
f758eeab | 658 | spin_lock(&wb->list_lock); |
424b351f | 659 | if (list_empty(&wb->b_io)) |
ad4e38dd | 660 | queue_io(wb, &work); |
d46db3d5 | 661 | __writeback_inodes_wb(wb, &work); |
f758eeab | 662 | spin_unlock(&wb->list_lock); |
edadfb10 | 663 | |
d46db3d5 WF |
664 | return nr_pages - work.nr_pages; |
665 | } | |
03ba3782 | 666 | |
b00949aa | 667 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
03ba3782 JA |
668 | { |
669 | unsigned long background_thresh, dirty_thresh; | |
670 | ||
16c4042f | 671 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 672 | |
b00949aa WF |
673 | if (global_page_state(NR_FILE_DIRTY) + |
674 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
675 | return true; | |
676 | ||
677 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > | |
678 | bdi_dirty_limit(bdi, background_thresh)) | |
679 | return true; | |
680 | ||
681 | return false; | |
03ba3782 JA |
682 | } |
683 | ||
e98be2d5 WF |
684 | /* |
685 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
686 | * only the flusher working on the first wb should do it. | |
687 | */ | |
688 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
689 | unsigned long start_time) | |
690 | { | |
af6a3113 | 691 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
692 | } |
693 | ||
03ba3782 JA |
694 | /* |
695 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 696 | * |
03ba3782 JA |
697 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
698 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
699 | * just walks the superblock inode list, writing back any inodes which are | |
700 | * older than a specific point in time. | |
66f3b8e2 | 701 | * |
03ba3782 JA |
702 | * Try to run once per dirty_writeback_interval. But if a writeback event |
703 | * takes longer than a dirty_writeback_interval interval, then leave a | |
704 | * one-second gap. | |
66f3b8e2 | 705 | * |
03ba3782 JA |
706 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
707 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 708 | */ |
c4a77a6c | 709 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 710 | struct wb_writeback_work *work) |
66f3b8e2 | 711 | { |
e98be2d5 | 712 | unsigned long wb_start = jiffies; |
d46db3d5 | 713 | long nr_pages = work->nr_pages; |
03ba3782 | 714 | unsigned long oldest_jif; |
a5989bdc | 715 | struct inode *inode; |
d46db3d5 | 716 | long progress; |
66f3b8e2 | 717 | |
e185dda8 | 718 | oldest_jif = jiffies; |
d46db3d5 | 719 | work->older_than_this = &oldest_jif; |
38f21977 | 720 | |
e8dfc305 | 721 | spin_lock(&wb->list_lock); |
03ba3782 JA |
722 | for (;;) { |
723 | /* | |
d3ddec76 | 724 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 725 | */ |
83ba7b07 | 726 | if (work->nr_pages <= 0) |
03ba3782 | 727 | break; |
66f3b8e2 | 728 | |
aa373cf5 JK |
729 | /* |
730 | * Background writeout and kupdate-style writeback may | |
731 | * run forever. Stop them if there is other work to do | |
732 | * so that e.g. sync can proceed. They'll be restarted | |
733 | * after the other works are all done. | |
734 | */ | |
735 | if ((work->for_background || work->for_kupdate) && | |
736 | !list_empty(&wb->bdi->work_list)) | |
737 | break; | |
738 | ||
38f21977 | 739 | /* |
d3ddec76 WF |
740 | * For background writeout, stop when we are below the |
741 | * background dirty threshold | |
38f21977 | 742 | */ |
b00949aa | 743 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
03ba3782 | 744 | break; |
38f21977 | 745 | |
ba9aa839 WF |
746 | if (work->for_kupdate) { |
747 | oldest_jif = jiffies - | |
748 | msecs_to_jiffies(dirty_expire_interval * 10); | |
d46db3d5 | 749 | work->older_than_this = &oldest_jif; |
ba9aa839 | 750 | } |
028c2dd1 | 751 | |
d46db3d5 | 752 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 753 | if (list_empty(&wb->b_io)) |
ad4e38dd | 754 | queue_io(wb, work); |
83ba7b07 | 755 | if (work->sb) |
d46db3d5 | 756 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 757 | else |
d46db3d5 WF |
758 | progress = __writeback_inodes_wb(wb, work); |
759 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 760 | |
e98be2d5 | 761 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
762 | |
763 | /* | |
e6fb6da2 WF |
764 | * Did we write something? Try for more |
765 | * | |
766 | * Dirty inodes are moved to b_io for writeback in batches. | |
767 | * The completion of the current batch does not necessarily | |
768 | * mean the overall work is done. So we keep looping as long | |
769 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 770 | */ |
d46db3d5 | 771 | if (progress) |
71fd05a8 JA |
772 | continue; |
773 | /* | |
e6fb6da2 | 774 | * No more inodes for IO, bail |
71fd05a8 | 775 | */ |
b7a2441f | 776 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 777 | break; |
71fd05a8 JA |
778 | /* |
779 | * Nothing written. Wait for some inode to | |
780 | * become available for writeback. Otherwise | |
781 | * we'll just busyloop. | |
782 | */ | |
71fd05a8 | 783 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 784 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 785 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 786 | spin_lock(&inode->i_lock); |
f758eeab | 787 | inode_wait_for_writeback(inode, wb); |
250df6ed | 788 | spin_unlock(&inode->i_lock); |
03ba3782 JA |
789 | } |
790 | } | |
e8dfc305 | 791 | spin_unlock(&wb->list_lock); |
03ba3782 | 792 | |
d46db3d5 | 793 | return nr_pages - work->nr_pages; |
03ba3782 JA |
794 | } |
795 | ||
796 | /* | |
83ba7b07 | 797 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 798 | */ |
83ba7b07 | 799 | static struct wb_writeback_work * |
08852b6d | 800 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 801 | { |
83ba7b07 | 802 | struct wb_writeback_work *work = NULL; |
03ba3782 | 803 | |
6467716a | 804 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
805 | if (!list_empty(&bdi->work_list)) { |
806 | work = list_entry(bdi->work_list.next, | |
807 | struct wb_writeback_work, list); | |
808 | list_del_init(&work->list); | |
03ba3782 | 809 | } |
6467716a | 810 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 811 | return work; |
03ba3782 JA |
812 | } |
813 | ||
cdf01dd5 LT |
814 | /* |
815 | * Add in the number of potentially dirty inodes, because each inode | |
816 | * write can dirty pagecache in the underlying blockdev. | |
817 | */ | |
818 | static unsigned long get_nr_dirty_pages(void) | |
819 | { | |
820 | return global_page_state(NR_FILE_DIRTY) + | |
821 | global_page_state(NR_UNSTABLE_NFS) + | |
822 | get_nr_dirty_inodes(); | |
823 | } | |
824 | ||
6585027a JK |
825 | static long wb_check_background_flush(struct bdi_writeback *wb) |
826 | { | |
b00949aa | 827 | if (over_bground_thresh(wb->bdi)) { |
6585027a JK |
828 | |
829 | struct wb_writeback_work work = { | |
830 | .nr_pages = LONG_MAX, | |
831 | .sync_mode = WB_SYNC_NONE, | |
832 | .for_background = 1, | |
833 | .range_cyclic = 1, | |
0e175a18 | 834 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
835 | }; |
836 | ||
837 | return wb_writeback(wb, &work); | |
838 | } | |
839 | ||
840 | return 0; | |
841 | } | |
842 | ||
03ba3782 JA |
843 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
844 | { | |
845 | unsigned long expired; | |
846 | long nr_pages; | |
847 | ||
69b62d01 JA |
848 | /* |
849 | * When set to zero, disable periodic writeback | |
850 | */ | |
851 | if (!dirty_writeback_interval) | |
852 | return 0; | |
853 | ||
03ba3782 JA |
854 | expired = wb->last_old_flush + |
855 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
856 | if (time_before(jiffies, expired)) | |
857 | return 0; | |
858 | ||
859 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 860 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 861 | |
c4a77a6c | 862 | if (nr_pages) { |
83ba7b07 | 863 | struct wb_writeback_work work = { |
c4a77a6c JA |
864 | .nr_pages = nr_pages, |
865 | .sync_mode = WB_SYNC_NONE, | |
866 | .for_kupdate = 1, | |
867 | .range_cyclic = 1, | |
0e175a18 | 868 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
869 | }; |
870 | ||
83ba7b07 | 871 | return wb_writeback(wb, &work); |
c4a77a6c | 872 | } |
03ba3782 JA |
873 | |
874 | return 0; | |
875 | } | |
876 | ||
877 | /* | |
878 | * Retrieve work items and do the writeback they describe | |
879 | */ | |
880 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
881 | { | |
882 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 883 | struct wb_writeback_work *work; |
c4a77a6c | 884 | long wrote = 0; |
03ba3782 | 885 | |
81d73a32 | 886 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 887 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
888 | /* |
889 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 890 | * because this thread is exiting now. |
03ba3782 JA |
891 | */ |
892 | if (force_wait) | |
83ba7b07 | 893 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 894 | |
455b2864 DC |
895 | trace_writeback_exec(bdi, work); |
896 | ||
83ba7b07 | 897 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
898 | |
899 | /* | |
83ba7b07 CH |
900 | * Notify the caller of completion if this is a synchronous |
901 | * work item, otherwise just free it. | |
03ba3782 | 902 | */ |
83ba7b07 CH |
903 | if (work->done) |
904 | complete(work->done); | |
905 | else | |
906 | kfree(work); | |
03ba3782 JA |
907 | } |
908 | ||
909 | /* | |
910 | * Check for periodic writeback, kupdated() style | |
911 | */ | |
912 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 913 | wrote += wb_check_background_flush(wb); |
81d73a32 | 914 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
915 | |
916 | return wrote; | |
917 | } | |
918 | ||
919 | /* | |
920 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
921 | * wakes up periodically and does kupdated style flushing. | |
922 | */ | |
08243900 | 923 | int bdi_writeback_thread(void *data) |
03ba3782 | 924 | { |
08243900 CH |
925 | struct bdi_writeback *wb = data; |
926 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
927 | long pages_written; |
928 | ||
766f9164 | 929 | current->flags |= PF_SWAPWRITE; |
08243900 | 930 | set_freezable(); |
ecd58403 | 931 | wb->last_active = jiffies; |
08243900 CH |
932 | |
933 | /* | |
934 | * Our parent may run at a different priority, just set us to normal | |
935 | */ | |
936 | set_user_nice(current, 0); | |
937 | ||
455b2864 DC |
938 | trace_writeback_thread_start(bdi); |
939 | ||
03ba3782 | 940 | while (!kthread_should_stop()) { |
6467716a AB |
941 | /* |
942 | * Remove own delayed wake-up timer, since we are already awake | |
943 | * and we'll take care of the preriodic write-back. | |
944 | */ | |
945 | del_timer(&wb->wakeup_timer); | |
946 | ||
03ba3782 JA |
947 | pages_written = wb_do_writeback(wb, 0); |
948 | ||
455b2864 DC |
949 | trace_writeback_pages_written(pages_written); |
950 | ||
03ba3782 | 951 | if (pages_written) |
ecd58403 | 952 | wb->last_active = jiffies; |
03ba3782 | 953 | |
297252c8 | 954 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 955 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 956 | __set_current_state(TASK_RUNNING); |
297252c8 | 957 | continue; |
03ba3782 JA |
958 | } |
959 | ||
253c34e9 | 960 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 961 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
962 | else { |
963 | /* | |
964 | * We have nothing to do, so can go sleep without any | |
965 | * timeout and save power. When a work is queued or | |
966 | * something is made dirty - we will be woken up. | |
967 | */ | |
297252c8 | 968 | schedule(); |
f9eadbbd | 969 | } |
69b62d01 | 970 | |
03ba3782 JA |
971 | try_to_freeze(); |
972 | } | |
973 | ||
fff5b85a | 974 | /* Flush any work that raced with us exiting */ |
08243900 CH |
975 | if (!list_empty(&bdi->work_list)) |
976 | wb_do_writeback(wb, 1); | |
455b2864 DC |
977 | |
978 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
979 | return 0; |
980 | } | |
981 | ||
08243900 | 982 | |
03ba3782 | 983 | /* |
b8c2f347 CH |
984 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
985 | * the whole world. | |
03ba3782 | 986 | */ |
0e175a18 | 987 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 988 | { |
b8c2f347 | 989 | struct backing_dev_info *bdi; |
03ba3782 | 990 | |
83ba7b07 CH |
991 | if (!nr_pages) { |
992 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
993 | global_page_state(NR_UNSTABLE_NFS); |
994 | } | |
03ba3782 | 995 | |
b8c2f347 | 996 | rcu_read_lock(); |
cfc4ba53 | 997 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
998 | if (!bdi_has_dirty_io(bdi)) |
999 | continue; | |
0e175a18 | 1000 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
03ba3782 | 1001 | } |
cfc4ba53 | 1002 | rcu_read_unlock(); |
1da177e4 LT |
1003 | } |
1004 | ||
03ba3782 JA |
1005 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1006 | { | |
1007 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1008 | struct dentry *dentry; | |
1009 | const char *name = "?"; | |
1010 | ||
1011 | dentry = d_find_alias(inode); | |
1012 | if (dentry) { | |
1013 | spin_lock(&dentry->d_lock); | |
1014 | name = (const char *) dentry->d_name.name; | |
1015 | } | |
1016 | printk(KERN_DEBUG | |
1017 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1018 | current->comm, task_pid_nr(current), inode->i_ino, | |
1019 | name, inode->i_sb->s_id); | |
1020 | if (dentry) { | |
1021 | spin_unlock(&dentry->d_lock); | |
1022 | dput(dentry); | |
1023 | } | |
1024 | } | |
1025 | } | |
1026 | ||
1027 | /** | |
1028 | * __mark_inode_dirty - internal function | |
1029 | * @inode: inode to mark | |
1030 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1031 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1032 | * mark_inode_dirty_sync. | |
1da177e4 | 1033 | * |
03ba3782 JA |
1034 | * Put the inode on the super block's dirty list. |
1035 | * | |
1036 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1037 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1038 | * If it was not hashed, it will never be added to the dirty list | |
1039 | * even if it is later hashed, as it will have been marked dirty already. | |
1040 | * | |
1041 | * In short, make sure you hash any inodes _before_ you start marking | |
1042 | * them dirty. | |
1da177e4 | 1043 | * |
03ba3782 JA |
1044 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1045 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1046 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1047 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1048 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1049 | * blockdev inode. | |
1da177e4 | 1050 | */ |
03ba3782 | 1051 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1052 | { |
03ba3782 | 1053 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1054 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1055 | |
03ba3782 JA |
1056 | /* |
1057 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1058 | * dirty the inode itself | |
1059 | */ | |
1060 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1061 | if (sb->s_op->dirty_inode) | |
aa385729 | 1062 | sb->s_op->dirty_inode(inode, flags); |
03ba3782 JA |
1063 | } |
1064 | ||
1065 | /* | |
1066 | * make sure that changes are seen by all cpus before we test i_state | |
1067 | * -- mikulas | |
1068 | */ | |
1069 | smp_mb(); | |
1070 | ||
1071 | /* avoid the locking if we can */ | |
1072 | if ((inode->i_state & flags) == flags) | |
1073 | return; | |
1074 | ||
1075 | if (unlikely(block_dump)) | |
1076 | block_dump___mark_inode_dirty(inode); | |
1077 | ||
250df6ed | 1078 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1079 | if ((inode->i_state & flags) != flags) { |
1080 | const int was_dirty = inode->i_state & I_DIRTY; | |
1081 | ||
1082 | inode->i_state |= flags; | |
1083 | ||
1084 | /* | |
1085 | * If the inode is being synced, just update its dirty state. | |
1086 | * The unlocker will place the inode on the appropriate | |
1087 | * superblock list, based upon its state. | |
1088 | */ | |
1089 | if (inode->i_state & I_SYNC) | |
250df6ed | 1090 | goto out_unlock_inode; |
03ba3782 JA |
1091 | |
1092 | /* | |
1093 | * Only add valid (hashed) inodes to the superblock's | |
1094 | * dirty list. Add blockdev inodes as well. | |
1095 | */ | |
1096 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1097 | if (inode_unhashed(inode)) |
250df6ed | 1098 | goto out_unlock_inode; |
03ba3782 | 1099 | } |
a4ffdde6 | 1100 | if (inode->i_state & I_FREEING) |
250df6ed | 1101 | goto out_unlock_inode; |
03ba3782 JA |
1102 | |
1103 | /* | |
1104 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1105 | * reposition it (that would break b_dirty time-ordering). | |
1106 | */ | |
1107 | if (!was_dirty) { | |
a66979ab | 1108 | bool wakeup_bdi = false; |
253c34e9 AB |
1109 | bdi = inode_to_bdi(inode); |
1110 | ||
1111 | if (bdi_cap_writeback_dirty(bdi)) { | |
1112 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1113 | "bdi-%s not registered\n", bdi->name); | |
1114 | ||
1115 | /* | |
1116 | * If this is the first dirty inode for this | |
1117 | * bdi, we have to wake-up the corresponding | |
1118 | * bdi thread to make sure background | |
1119 | * write-back happens later. | |
1120 | */ | |
1121 | if (!wb_has_dirty_io(&bdi->wb)) | |
1122 | wakeup_bdi = true; | |
500b067c | 1123 | } |
03ba3782 | 1124 | |
a66979ab | 1125 | spin_unlock(&inode->i_lock); |
f758eeab | 1126 | spin_lock(&bdi->wb.list_lock); |
03ba3782 | 1127 | inode->dirtied_when = jiffies; |
7ccf19a8 | 1128 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1129 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1130 | |
1131 | if (wakeup_bdi) | |
1132 | bdi_wakeup_thread_delayed(bdi); | |
1133 | return; | |
1da177e4 | 1134 | } |
1da177e4 | 1135 | } |
250df6ed DC |
1136 | out_unlock_inode: |
1137 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1138 | |
03ba3782 JA |
1139 | } |
1140 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1141 | ||
1142 | /* | |
1143 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1144 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1145 | * | |
1146 | * If older_than_this is non-NULL, then only write out inodes which | |
1147 | * had their first dirtying at a time earlier than *older_than_this. | |
1148 | * | |
03ba3782 JA |
1149 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1150 | * This function assumes that the blockdev superblock's inodes are backed by | |
1151 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1152 | * assume that all inodes are backed by the same queue. | |
1153 | * | |
1154 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1155 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1156 | * on the writer throttling path, and we get decent balancing between many | |
1157 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1158 | */ | |
b6e51316 | 1159 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1160 | { |
1161 | struct inode *inode, *old_inode = NULL; | |
1162 | ||
1163 | /* | |
1164 | * We need to be protected against the filesystem going from | |
1165 | * r/o to r/w or vice versa. | |
1166 | */ | |
b6e51316 | 1167 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1168 | |
55fa6091 | 1169 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1170 | |
1171 | /* | |
1172 | * Data integrity sync. Must wait for all pages under writeback, | |
1173 | * because there may have been pages dirtied before our sync | |
1174 | * call, but which had writeout started before we write it out. | |
1175 | * In which case, the inode may not be on the dirty list, but | |
1176 | * we still have to wait for that writeout. | |
1177 | */ | |
b6e51316 | 1178 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1179 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1180 | |
250df6ed DC |
1181 | spin_lock(&inode->i_lock); |
1182 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1183 | (mapping->nrpages == 0)) { | |
1184 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1185 | continue; |
250df6ed | 1186 | } |
03ba3782 | 1187 | __iget(inode); |
250df6ed | 1188 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1189 | spin_unlock(&inode_sb_list_lock); |
1190 | ||
03ba3782 | 1191 | /* |
55fa6091 DC |
1192 | * We hold a reference to 'inode' so it couldn't have been |
1193 | * removed from s_inodes list while we dropped the | |
1194 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1195 | * be holding the last reference and we cannot iput it under | |
1196 | * inode_sb_list_lock. So we keep the reference and iput it | |
1197 | * later. | |
03ba3782 JA |
1198 | */ |
1199 | iput(old_inode); | |
1200 | old_inode = inode; | |
1201 | ||
1202 | filemap_fdatawait(mapping); | |
1203 | ||
1204 | cond_resched(); | |
1205 | ||
55fa6091 | 1206 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1207 | } |
55fa6091 | 1208 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1209 | iput(old_inode); |
1da177e4 LT |
1210 | } |
1211 | ||
d8a8559c | 1212 | /** |
3259f8be | 1213 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1214 | * @sb: the superblock |
3259f8be | 1215 | * @nr: the number of pages to write |
786228ab | 1216 | * @reason: reason why some writeback work initiated |
1da177e4 | 1217 | * |
d8a8559c JA |
1218 | * Start writeback on some inodes on this super_block. No guarantees are made |
1219 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1220 | * for IO completion of submitted IO. |
1da177e4 | 1221 | */ |
0e175a18 CW |
1222 | void writeback_inodes_sb_nr(struct super_block *sb, |
1223 | unsigned long nr, | |
1224 | enum wb_reason reason) | |
1da177e4 | 1225 | { |
83ba7b07 CH |
1226 | DECLARE_COMPLETION_ONSTACK(done); |
1227 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1228 | .sb = sb, |
1229 | .sync_mode = WB_SYNC_NONE, | |
1230 | .tagged_writepages = 1, | |
1231 | .done = &done, | |
1232 | .nr_pages = nr, | |
0e175a18 | 1233 | .reason = reason, |
3c4d7165 | 1234 | }; |
d8a8559c | 1235 | |
cf37e972 | 1236 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1237 | bdi_queue_work(sb->s_bdi, &work); |
1238 | wait_for_completion(&done); | |
e913fc82 | 1239 | } |
3259f8be CM |
1240 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1241 | ||
1242 | /** | |
1243 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1244 | * @sb: the superblock | |
786228ab | 1245 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1246 | * |
1247 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1248 | * on how many (if any) will be written, and this function does not wait | |
1249 | * for IO completion of submitted IO. | |
1250 | */ | |
0e175a18 | 1251 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1252 | { |
0e175a18 | 1253 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1254 | } |
0e3c9a22 | 1255 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1256 | |
17bd55d0 ES |
1257 | /** |
1258 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1259 | * @sb: the superblock | |
786228ab | 1260 | * @reason: reason why some writeback work was initiated |
17bd55d0 ES |
1261 | * |
1262 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1263 | * Returns 1 if writeback was started, 0 if not. | |
1264 | */ | |
0e175a18 | 1265 | int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) |
17bd55d0 ES |
1266 | { |
1267 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1268 | down_read(&sb->s_umount); |
0e175a18 | 1269 | writeback_inodes_sb(sb, reason); |
cf37e972 | 1270 | up_read(&sb->s_umount); |
17bd55d0 ES |
1271 | return 1; |
1272 | } else | |
1273 | return 0; | |
1274 | } | |
1275 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1276 | ||
3259f8be CM |
1277 | /** |
1278 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1279 | * @sb: the superblock | |
1280 | * @nr: the number of pages to write | |
786228ab | 1281 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1282 | * |
1283 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1284 | * Returns 1 if writeback was started, 0 if not. | |
1285 | */ | |
1286 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
0e175a18 CW |
1287 | unsigned long nr, |
1288 | enum wb_reason reason) | |
3259f8be CM |
1289 | { |
1290 | if (!writeback_in_progress(sb->s_bdi)) { | |
1291 | down_read(&sb->s_umount); | |
0e175a18 | 1292 | writeback_inodes_sb_nr(sb, nr, reason); |
3259f8be CM |
1293 | up_read(&sb->s_umount); |
1294 | return 1; | |
1295 | } else | |
1296 | return 0; | |
1297 | } | |
1298 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1299 | ||
d8a8559c JA |
1300 | /** |
1301 | * sync_inodes_sb - sync sb inode pages | |
1302 | * @sb: the superblock | |
1303 | * | |
1304 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1305 | * super_block. |
d8a8559c | 1306 | */ |
b6e51316 | 1307 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1308 | { |
83ba7b07 CH |
1309 | DECLARE_COMPLETION_ONSTACK(done); |
1310 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1311 | .sb = sb, |
1312 | .sync_mode = WB_SYNC_ALL, | |
1313 | .nr_pages = LONG_MAX, | |
1314 | .range_cyclic = 0, | |
83ba7b07 | 1315 | .done = &done, |
0e175a18 | 1316 | .reason = WB_REASON_SYNC, |
3c4d7165 CH |
1317 | }; |
1318 | ||
cf37e972 CH |
1319 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1320 | ||
83ba7b07 CH |
1321 | bdi_queue_work(sb->s_bdi, &work); |
1322 | wait_for_completion(&done); | |
1323 | ||
b6e51316 | 1324 | wait_sb_inodes(sb); |
1da177e4 | 1325 | } |
d8a8559c | 1326 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1327 | |
1da177e4 | 1328 | /** |
7f04c26d AA |
1329 | * write_inode_now - write an inode to disk |
1330 | * @inode: inode to write to disk | |
1331 | * @sync: whether the write should be synchronous or not | |
1332 | * | |
1333 | * This function commits an inode to disk immediately if it is dirty. This is | |
1334 | * primarily needed by knfsd. | |
1da177e4 | 1335 | * |
7f04c26d | 1336 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1337 | */ |
1da177e4 LT |
1338 | int write_inode_now(struct inode *inode, int sync) |
1339 | { | |
f758eeab | 1340 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1341 | int ret; |
1342 | struct writeback_control wbc = { | |
1343 | .nr_to_write = LONG_MAX, | |
18914b18 | 1344 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1345 | .range_start = 0, |
1346 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1347 | }; |
1348 | ||
1349 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1350 | wbc.nr_to_write = 0; |
1da177e4 LT |
1351 | |
1352 | might_sleep(); | |
f758eeab | 1353 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1354 | spin_lock(&inode->i_lock); |
f758eeab | 1355 | ret = writeback_single_inode(inode, wb, &wbc); |
0f1b1fd8 | 1356 | spin_unlock(&inode->i_lock); |
f758eeab | 1357 | spin_unlock(&wb->list_lock); |
1da177e4 | 1358 | if (sync) |
1c0eeaf5 | 1359 | inode_sync_wait(inode); |
1da177e4 LT |
1360 | return ret; |
1361 | } | |
1362 | EXPORT_SYMBOL(write_inode_now); | |
1363 | ||
1364 | /** | |
1365 | * sync_inode - write an inode and its pages to disk. | |
1366 | * @inode: the inode to sync | |
1367 | * @wbc: controls the writeback mode | |
1368 | * | |
1369 | * sync_inode() will write an inode and its pages to disk. It will also | |
1370 | * correctly update the inode on its superblock's dirty inode lists and will | |
1371 | * update inode->i_state. | |
1372 | * | |
1373 | * The caller must have a ref on the inode. | |
1374 | */ | |
1375 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1376 | { | |
f758eeab | 1377 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1378 | int ret; |
1379 | ||
f758eeab | 1380 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1381 | spin_lock(&inode->i_lock); |
f758eeab | 1382 | ret = writeback_single_inode(inode, wb, wbc); |
0f1b1fd8 | 1383 | spin_unlock(&inode->i_lock); |
f758eeab | 1384 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
1385 | return ret; |
1386 | } | |
1387 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1388 | |
1389 | /** | |
c691b9d9 | 1390 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1391 | * @inode: the inode to sync |
1392 | * @wait: wait for I/O to complete. | |
1393 | * | |
c691b9d9 | 1394 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1395 | * |
1396 | * Note: only writes the actual inode, no associated data or other metadata. | |
1397 | */ | |
1398 | int sync_inode_metadata(struct inode *inode, int wait) | |
1399 | { | |
1400 | struct writeback_control wbc = { | |
1401 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1402 | .nr_to_write = 0, /* metadata-only */ | |
1403 | }; | |
1404 | ||
1405 | return sync_inode(inode, &wbc); | |
1406 | } | |
1407 | EXPORT_SYMBOL(sync_inode_metadata); |