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 JA |
24 | #include <linux/kthread.h> |
25 | #include <linux/freezer.h> | |
1da177e4 LT |
26 | #include <linux/writeback.h> |
27 | #include <linux/blkdev.h> | |
28 | #include <linux/backing-dev.h> | |
455b2864 | 29 | #include <linux/tracepoint.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 | 31 | |
bc31b86a WF |
32 | /* |
33 | * 4MB minimal write chunk size | |
34 | */ | |
35 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) | |
36 | ||
c4a77a6c JA |
37 | /* |
38 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
39 | */ | |
83ba7b07 | 40 | struct wb_writeback_work { |
c4a77a6c JA |
41 | long nr_pages; |
42 | struct super_block *sb; | |
d46db3d5 | 43 | unsigned long *older_than_this; |
c4a77a6c | 44 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 45 | unsigned int tagged_writepages:1; |
52957fe1 HS |
46 | unsigned int for_kupdate:1; |
47 | unsigned int range_cyclic:1; | |
48 | unsigned int for_background:1; | |
0e175a18 | 49 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 50 | |
8010c3b6 | 51 | struct list_head list; /* pending work list */ |
83ba7b07 | 52 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
53 | }; |
54 | ||
455b2864 DC |
55 | /* |
56 | * We don't actually have pdflush, but this one is exported though /proc... | |
57 | */ | |
58 | int nr_pdflush_threads; | |
59 | ||
f11b00f3 AB |
60 | /** |
61 | * writeback_in_progress - determine whether there is writeback in progress | |
62 | * @bdi: the device's backing_dev_info structure. | |
63 | * | |
03ba3782 JA |
64 | * Determine whether there is writeback waiting to be handled against a |
65 | * backing device. | |
f11b00f3 AB |
66 | */ |
67 | int writeback_in_progress(struct backing_dev_info *bdi) | |
68 | { | |
81d73a32 | 69 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 AB |
70 | } |
71 | ||
692ebd17 JK |
72 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
73 | { | |
74 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 75 | |
aaead25b CH |
76 | if (strcmp(sb->s_type->name, "bdev") == 0) |
77 | return inode->i_mapping->backing_dev_info; | |
78 | ||
79 | return sb->s_bdi; | |
692ebd17 JK |
80 | } |
81 | ||
7ccf19a8 NP |
82 | static inline struct inode *wb_inode(struct list_head *head) |
83 | { | |
84 | return list_entry(head, struct inode, i_wb_list); | |
85 | } | |
86 | ||
15eb77a0 WF |
87 | /* |
88 | * Include the creation of the trace points after defining the | |
89 | * wb_writeback_work structure and inline functions so that the definition | |
90 | * remains local to this file. | |
91 | */ | |
92 | #define CREATE_TRACE_POINTS | |
93 | #include <trace/events/writeback.h> | |
94 | ||
6585027a JK |
95 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
96 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) | |
03ba3782 | 97 | { |
fff5b85a AB |
98 | if (bdi->wb.task) { |
99 | wake_up_process(bdi->wb.task); | |
100 | } else { | |
101 | /* | |
102 | * The bdi thread isn't there, wake up the forker thread which | |
103 | * will create and run it. | |
104 | */ | |
03ba3782 | 105 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 106 | } |
6585027a JK |
107 | } |
108 | ||
109 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
110 | struct wb_writeback_work *work) | |
111 | { | |
112 | trace_writeback_queue(bdi, work); | |
113 | ||
114 | spin_lock_bh(&bdi->wb_lock); | |
115 | list_add_tail(&work->list, &bdi->work_list); | |
116 | if (!bdi->wb.task) | |
117 | trace_writeback_nothread(bdi, work); | |
118 | bdi_wakeup_flusher(bdi); | |
6467716a | 119 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
120 | } |
121 | ||
83ba7b07 CH |
122 | static void |
123 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
0e175a18 | 124 | bool range_cyclic, enum wb_reason reason) |
1da177e4 | 125 | { |
83ba7b07 | 126 | struct wb_writeback_work *work; |
03ba3782 | 127 | |
bcddc3f0 JA |
128 | /* |
129 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
130 | * wakeup the thread for old dirty data writeback | |
131 | */ | |
83ba7b07 CH |
132 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
133 | if (!work) { | |
455b2864 DC |
134 | if (bdi->wb.task) { |
135 | trace_writeback_nowork(bdi); | |
83ba7b07 | 136 | wake_up_process(bdi->wb.task); |
455b2864 | 137 | } |
83ba7b07 | 138 | return; |
bcddc3f0 | 139 | } |
03ba3782 | 140 | |
83ba7b07 CH |
141 | work->sync_mode = WB_SYNC_NONE; |
142 | work->nr_pages = nr_pages; | |
143 | work->range_cyclic = range_cyclic; | |
0e175a18 | 144 | work->reason = reason; |
03ba3782 | 145 | |
83ba7b07 | 146 | bdi_queue_work(bdi, work); |
b6e51316 JA |
147 | } |
148 | ||
149 | /** | |
150 | * bdi_start_writeback - start writeback | |
151 | * @bdi: the backing device to write from | |
152 | * @nr_pages: the number of pages to write | |
786228ab | 153 | * @reason: reason why some writeback work was initiated |
b6e51316 JA |
154 | * |
155 | * Description: | |
156 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 157 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 158 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
159 | * |
160 | */ | |
0e175a18 CW |
161 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
162 | enum wb_reason reason) | |
b6e51316 | 163 | { |
0e175a18 | 164 | __bdi_start_writeback(bdi, nr_pages, true, reason); |
c5444198 | 165 | } |
d3ddec76 | 166 | |
c5444198 CH |
167 | /** |
168 | * bdi_start_background_writeback - start background writeback | |
169 | * @bdi: the backing device to write from | |
170 | * | |
171 | * Description: | |
6585027a JK |
172 | * This makes sure WB_SYNC_NONE background writeback happens. When |
173 | * this function returns, it is only guaranteed that for given BDI | |
174 | * some IO is happening if we are over background dirty threshold. | |
175 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
176 | */ |
177 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
178 | { | |
6585027a JK |
179 | /* |
180 | * We just wake up the flusher thread. It will perform background | |
181 | * writeback as soon as there is no other work to do. | |
182 | */ | |
71927e84 | 183 | trace_writeback_wake_background(bdi); |
6585027a JK |
184 | spin_lock_bh(&bdi->wb_lock); |
185 | bdi_wakeup_flusher(bdi); | |
186 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
187 | } |
188 | ||
a66979ab DC |
189 | /* |
190 | * Remove the inode from the writeback list it is on. | |
191 | */ | |
192 | void inode_wb_list_del(struct inode *inode) | |
193 | { | |
f758eeab CH |
194 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
195 | ||
196 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 197 | list_del_init(&inode->i_wb_list); |
f758eeab | 198 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
199 | } |
200 | ||
6610a0bc AM |
201 | /* |
202 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
203 | * furthest end of its superblock's dirty-inode list. | |
204 | * | |
205 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 206 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
207 | * the case then the inode must have been redirtied while it was being written |
208 | * out and we don't reset its dirtied_when. | |
209 | */ | |
f758eeab | 210 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 211 | { |
f758eeab | 212 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 213 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 214 | struct inode *tail; |
6610a0bc | 215 | |
7ccf19a8 | 216 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 217 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
218 | inode->dirtied_when = jiffies; |
219 | } | |
7ccf19a8 | 220 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
221 | } |
222 | ||
c986d1e2 | 223 | /* |
66f3b8e2 | 224 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 225 | */ |
f758eeab | 226 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 227 | { |
f758eeab | 228 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 229 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
230 | } |
231 | ||
1c0eeaf5 JE |
232 | static void inode_sync_complete(struct inode *inode) |
233 | { | |
365b94ae JK |
234 | inode->i_state &= ~I_SYNC; |
235 | /* Waiters must see I_SYNC cleared before being woken up */ | |
1c0eeaf5 JE |
236 | smp_mb(); |
237 | wake_up_bit(&inode->i_state, __I_SYNC); | |
238 | } | |
239 | ||
d2caa3c5 JL |
240 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
241 | { | |
242 | bool ret = time_after(inode->dirtied_when, t); | |
243 | #ifndef CONFIG_64BIT | |
244 | /* | |
245 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
246 | * It _appears_ to be in the future, but is actually in distant past. | |
247 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 248 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
249 | */ |
250 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
251 | #endif | |
252 | return ret; | |
253 | } | |
254 | ||
2c136579 | 255 | /* |
697e6fed JK |
256 | * Move expired (dirtied after work->older_than_this) dirty inodes from |
257 | * @delaying_queue to @dispatch_queue. | |
2c136579 | 258 | */ |
e84d0a4f | 259 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 260 | struct list_head *dispatch_queue, |
ad4e38dd | 261 | struct wb_writeback_work *work) |
2c136579 | 262 | { |
5c03449d SL |
263 | LIST_HEAD(tmp); |
264 | struct list_head *pos, *node; | |
cf137307 | 265 | struct super_block *sb = NULL; |
5c03449d | 266 | struct inode *inode; |
cf137307 | 267 | int do_sb_sort = 0; |
e84d0a4f | 268 | int moved = 0; |
5c03449d | 269 | |
2c136579 | 270 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 271 | inode = wb_inode(delaying_queue->prev); |
ad4e38dd CW |
272 | if (work->older_than_this && |
273 | inode_dirtied_after(inode, *work->older_than_this)) | |
2c136579 | 274 | break; |
cf137307 JA |
275 | if (sb && sb != inode->i_sb) |
276 | do_sb_sort = 1; | |
277 | sb = inode->i_sb; | |
7ccf19a8 | 278 | list_move(&inode->i_wb_list, &tmp); |
e84d0a4f | 279 | moved++; |
5c03449d SL |
280 | } |
281 | ||
cf137307 JA |
282 | /* just one sb in list, splice to dispatch_queue and we're done */ |
283 | if (!do_sb_sort) { | |
284 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 285 | goto out; |
cf137307 JA |
286 | } |
287 | ||
5c03449d SL |
288 | /* Move inodes from one superblock together */ |
289 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 290 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 291 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 292 | inode = wb_inode(pos); |
5c03449d | 293 | if (inode->i_sb == sb) |
7ccf19a8 | 294 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 295 | } |
2c136579 | 296 | } |
e84d0a4f WF |
297 | out: |
298 | return moved; | |
2c136579 FW |
299 | } |
300 | ||
301 | /* | |
302 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
303 | * Before |
304 | * newly dirtied b_dirty b_io b_more_io | |
305 | * =============> gf edc BA | |
306 | * After | |
307 | * newly dirtied b_dirty b_io b_more_io | |
308 | * =============> g fBAedc | |
309 | * | | |
310 | * +--> dequeue for IO | |
2c136579 | 311 | */ |
ad4e38dd | 312 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 313 | { |
e84d0a4f | 314 | int moved; |
f758eeab | 315 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 316 | list_splice_init(&wb->b_more_io, &wb->b_io); |
ad4e38dd CW |
317 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); |
318 | trace_writeback_queue_io(wb, work, moved); | |
66f3b8e2 JA |
319 | } |
320 | ||
a9185b41 | 321 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 322 | { |
03ba3782 | 323 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 324 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 325 | return 0; |
08d8e974 | 326 | } |
08d8e974 | 327 | |
1da177e4 | 328 | /* |
01c03194 CH |
329 | * Wait for writeback on an inode to complete. |
330 | */ | |
f758eeab CH |
331 | static void inode_wait_for_writeback(struct inode *inode, |
332 | struct bdi_writeback *wb) | |
01c03194 CH |
333 | { |
334 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
335 | wait_queue_head_t *wqh; | |
336 | ||
337 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
338 | while (inode->i_state & I_SYNC) { |
339 | spin_unlock(&inode->i_lock); | |
f758eeab | 340 | spin_unlock(&wb->list_lock); |
01c03194 | 341 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
f758eeab | 342 | spin_lock(&wb->list_lock); |
250df6ed | 343 | spin_lock(&inode->i_lock); |
58a9d3d8 | 344 | } |
01c03194 CH |
345 | } |
346 | ||
347 | /* | |
f758eeab | 348 | * Write out an inode's dirty pages. Called under wb->list_lock and |
0f1b1fd8 DC |
349 | * inode->i_lock. Either the caller has an active reference on the inode or |
350 | * the inode has I_WILL_FREE set. | |
01c03194 | 351 | * |
1da177e4 LT |
352 | * If `wait' is set, wait on the writeout. |
353 | * | |
354 | * The whole writeout design is quite complex and fragile. We want to avoid | |
355 | * starvation of particular inodes when others are being redirtied, prevent | |
356 | * livelocks, etc. | |
1da177e4 LT |
357 | */ |
358 | static int | |
f758eeab CH |
359 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
360 | struct writeback_control *wbc) | |
1da177e4 | 361 | { |
1da177e4 | 362 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 363 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 364 | unsigned dirty; |
1da177e4 LT |
365 | int ret; |
366 | ||
f758eeab | 367 | assert_spin_locked(&wb->list_lock); |
0f1b1fd8 DC |
368 | assert_spin_locked(&inode->i_lock); |
369 | ||
01c03194 CH |
370 | if (!atomic_read(&inode->i_count)) |
371 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
372 | else | |
373 | WARN_ON(inode->i_state & I_WILL_FREE); | |
374 | ||
375 | if (inode->i_state & I_SYNC) { | |
cc1676d9 | 376 | if (wbc->sync_mode != WB_SYNC_ALL) |
01c03194 | 377 | return 0; |
01c03194 CH |
378 | /* |
379 | * It's a data-integrity sync. We must wait. | |
380 | */ | |
f758eeab | 381 | inode_wait_for_writeback(inode, wb); |
01c03194 CH |
382 | } |
383 | ||
1c0eeaf5 | 384 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 385 | |
5547e8aa | 386 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 387 | inode->i_state |= I_SYNC; |
250df6ed | 388 | spin_unlock(&inode->i_lock); |
f758eeab | 389 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
390 | |
391 | ret = do_writepages(mapping, wbc); | |
392 | ||
26821ed4 CH |
393 | /* |
394 | * Make sure to wait on the data before writing out the metadata. | |
395 | * This is important for filesystems that modify metadata on data | |
396 | * I/O completion. | |
397 | */ | |
a9185b41 | 398 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 399 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
400 | if (ret == 0) |
401 | ret = err; | |
402 | } | |
403 | ||
5547e8aa DM |
404 | /* |
405 | * Some filesystems may redirty the inode during the writeback | |
406 | * due to delalloc, clear dirty metadata flags right before | |
407 | * write_inode() | |
408 | */ | |
250df6ed | 409 | spin_lock(&inode->i_lock); |
6290be1c JK |
410 | /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ |
411 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
412 | inode->i_state &= ~I_DIRTY_PAGES; | |
5547e8aa DM |
413 | dirty = inode->i_state & I_DIRTY; |
414 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 415 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
416 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
417 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 418 | int err = write_inode(inode, wbc); |
1da177e4 LT |
419 | if (ret == 0) |
420 | ret = err; | |
421 | } | |
422 | ||
f758eeab | 423 | spin_lock(&wb->list_lock); |
250df6ed | 424 | spin_lock(&inode->i_lock); |
a4ffdde6 | 425 | if (!(inode->i_state & I_FREEING)) { |
94c3dcbb WF |
426 | /* |
427 | * Sync livelock prevention. Each inode is tagged and synced in | |
428 | * one shot. If still dirty, it will be redirty_tail()'ed below. | |
429 | * Update the dirty time to prevent enqueue and sync it again. | |
430 | */ | |
431 | if ((inode->i_state & I_DIRTY) && | |
432 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
433 | inode->dirtied_when = jiffies; | |
434 | ||
23539afc | 435 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
436 | /* |
437 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 438 | * sometimes bales out without doing anything. |
1b43ef91 | 439 | */ |
a50aeb40 | 440 | if (wbc->nr_to_write <= 0) { |
1da177e4 | 441 | /* |
a50aeb40 | 442 | * slice used up: queue for next turn |
1da177e4 | 443 | */ |
f758eeab | 444 | requeue_io(inode, wb); |
1da177e4 LT |
445 | } else { |
446 | /* | |
a50aeb40 WF |
447 | * Writeback blocked by something other than |
448 | * congestion. Delay the inode for some time to | |
449 | * avoid spinning on the CPU (100% iowait) | |
450 | * retrying writeback of the dirty page/inode | |
451 | * that cannot be performed immediately. | |
1da177e4 | 452 | */ |
f758eeab | 453 | redirty_tail(inode, wb); |
1da177e4 | 454 | } |
23539afc WF |
455 | } else if (inode->i_state & I_DIRTY) { |
456 | /* | |
457 | * Filesystems can dirty the inode during writeback | |
458 | * operations, such as delayed allocation during | |
459 | * submission or metadata updates after data IO | |
460 | * completion. | |
461 | */ | |
f758eeab | 462 | redirty_tail(inode, wb); |
1da177e4 LT |
463 | } else { |
464 | /* | |
9e38d86f NP |
465 | * The inode is clean. At this point we either have |
466 | * a reference to the inode or it's on it's way out. | |
467 | * No need to add it back to the LRU. | |
1da177e4 | 468 | */ |
7ccf19a8 | 469 | list_del_init(&inode->i_wb_list); |
1da177e4 LT |
470 | } |
471 | } | |
1c0eeaf5 | 472 | inode_sync_complete(inode); |
251d6a47 | 473 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
1da177e4 LT |
474 | return ret; |
475 | } | |
476 | ||
1a12d8bd WF |
477 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
478 | struct wb_writeback_work *work) | |
d46db3d5 WF |
479 | { |
480 | long pages; | |
481 | ||
482 | /* | |
483 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
484 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
485 | * here avoids calling into writeback_inodes_wb() more than once. | |
486 | * | |
487 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
488 | * | |
489 | * wb_writeback() | |
490 | * writeback_sb_inodes() <== called only once | |
491 | * write_cache_pages() <== called once for each inode | |
492 | * (quickly) tag currently dirty pages | |
493 | * (maybe slowly) sync all tagged pages | |
494 | */ | |
495 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
496 | pages = LONG_MAX; | |
1a12d8bd WF |
497 | else { |
498 | pages = min(bdi->avg_write_bandwidth / 2, | |
499 | global_dirty_limit / DIRTY_SCOPE); | |
500 | pages = min(pages, work->nr_pages); | |
501 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
502 | MIN_WRITEBACK_PAGES); | |
503 | } | |
d46db3d5 WF |
504 | |
505 | return pages; | |
506 | } | |
507 | ||
f11c9c5c ES |
508 | /* |
509 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
510 | * |
511 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
512 | * inodes. Otherwise write only ones which go sequentially |
513 | * in reverse order. | |
edadfb10 | 514 | * |
d46db3d5 | 515 | * Return the number of pages and/or inodes written. |
f11c9c5c | 516 | */ |
d46db3d5 WF |
517 | static long writeback_sb_inodes(struct super_block *sb, |
518 | struct bdi_writeback *wb, | |
519 | struct wb_writeback_work *work) | |
1da177e4 | 520 | { |
d46db3d5 WF |
521 | struct writeback_control wbc = { |
522 | .sync_mode = work->sync_mode, | |
523 | .tagged_writepages = work->tagged_writepages, | |
524 | .for_kupdate = work->for_kupdate, | |
525 | .for_background = work->for_background, | |
526 | .range_cyclic = work->range_cyclic, | |
527 | .range_start = 0, | |
528 | .range_end = LLONG_MAX, | |
529 | }; | |
530 | unsigned long start_time = jiffies; | |
531 | long write_chunk; | |
532 | long wrote = 0; /* count both pages and inodes */ | |
533 | ||
03ba3782 | 534 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 535 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
536 | |
537 | if (inode->i_sb != sb) { | |
d46db3d5 | 538 | if (work->sb) { |
edadfb10 CH |
539 | /* |
540 | * We only want to write back data for this | |
541 | * superblock, move all inodes not belonging | |
542 | * to it back onto the dirty list. | |
543 | */ | |
f758eeab | 544 | redirty_tail(inode, wb); |
edadfb10 CH |
545 | continue; |
546 | } | |
547 | ||
548 | /* | |
549 | * The inode belongs to a different superblock. | |
550 | * Bounce back to the caller to unpin this and | |
551 | * pin the next superblock. | |
552 | */ | |
d46db3d5 | 553 | break; |
edadfb10 CH |
554 | } |
555 | ||
9843b76a CH |
556 | /* |
557 | * Don't bother with new inodes or inodes beeing freed, first | |
558 | * kind does not need peridic writeout yet, and for the latter | |
559 | * kind writeout is handled by the freer. | |
560 | */ | |
250df6ed | 561 | spin_lock(&inode->i_lock); |
9843b76a | 562 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 563 | spin_unlock(&inode->i_lock); |
fcc5c222 | 564 | redirty_tail(inode, wb); |
7ef0d737 NP |
565 | continue; |
566 | } | |
cc1676d9 JK |
567 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
568 | /* | |
569 | * If this inode is locked for writeback and we are not | |
570 | * doing writeback-for-data-integrity, move it to | |
571 | * b_more_io so that writeback can proceed with the | |
572 | * other inodes on s_io. | |
573 | * | |
574 | * We'll have another go at writing back this inode | |
575 | * when we completed a full scan of b_io. | |
576 | */ | |
577 | spin_unlock(&inode->i_lock); | |
578 | requeue_io(inode, wb); | |
579 | trace_writeback_sb_inodes_requeue(inode); | |
580 | continue; | |
581 | } | |
1da177e4 | 582 | __iget(inode); |
1a12d8bd | 583 | write_chunk = writeback_chunk_size(wb->bdi, work); |
d46db3d5 WF |
584 | wbc.nr_to_write = write_chunk; |
585 | wbc.pages_skipped = 0; | |
250df6ed | 586 | |
d46db3d5 | 587 | writeback_single_inode(inode, wb, &wbc); |
250df6ed | 588 | |
d46db3d5 WF |
589 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
590 | wrote += write_chunk - wbc.nr_to_write; | |
591 | if (!(inode->i_state & I_DIRTY)) | |
592 | wrote++; | |
593 | if (wbc.pages_skipped) { | |
1da177e4 LT |
594 | /* |
595 | * writeback is not making progress due to locked | |
596 | * buffers. Skip this inode for now. | |
597 | */ | |
f758eeab | 598 | redirty_tail(inode, wb); |
1da177e4 | 599 | } |
0f1b1fd8 | 600 | spin_unlock(&inode->i_lock); |
f758eeab | 601 | spin_unlock(&wb->list_lock); |
1da177e4 | 602 | iput(inode); |
4ffc8444 | 603 | cond_resched(); |
f758eeab | 604 | spin_lock(&wb->list_lock); |
d46db3d5 WF |
605 | /* |
606 | * bail out to wb_writeback() often enough to check | |
607 | * background threshold and other termination conditions. | |
608 | */ | |
609 | if (wrote) { | |
610 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
611 | break; | |
612 | if (work->nr_pages <= 0) | |
613 | break; | |
8bc3be27 | 614 | } |
1da177e4 | 615 | } |
d46db3d5 | 616 | return wrote; |
f11c9c5c ES |
617 | } |
618 | ||
d46db3d5 WF |
619 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
620 | struct wb_writeback_work *work) | |
f11c9c5c | 621 | { |
d46db3d5 WF |
622 | unsigned long start_time = jiffies; |
623 | long wrote = 0; | |
38f21977 | 624 | |
f11c9c5c | 625 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 626 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 627 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 628 | |
12ad3ab6 | 629 | if (!grab_super_passive(sb)) { |
0e995816 WF |
630 | /* |
631 | * grab_super_passive() may fail consistently due to | |
632 | * s_umount being grabbed by someone else. Don't use | |
633 | * requeue_io() to avoid busy retrying the inode/sb. | |
634 | */ | |
635 | redirty_tail(inode, wb); | |
edadfb10 | 636 | continue; |
f11c9c5c | 637 | } |
d46db3d5 | 638 | wrote += writeback_sb_inodes(sb, wb, work); |
edadfb10 | 639 | drop_super(sb); |
f11c9c5c | 640 | |
d46db3d5 WF |
641 | /* refer to the same tests at the end of writeback_sb_inodes */ |
642 | if (wrote) { | |
643 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
644 | break; | |
645 | if (work->nr_pages <= 0) | |
646 | break; | |
647 | } | |
f11c9c5c | 648 | } |
66f3b8e2 | 649 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 650 | return wrote; |
66f3b8e2 JA |
651 | } |
652 | ||
0e175a18 CW |
653 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
654 | enum wb_reason reason) | |
edadfb10 | 655 | { |
d46db3d5 WF |
656 | struct wb_writeback_work work = { |
657 | .nr_pages = nr_pages, | |
658 | .sync_mode = WB_SYNC_NONE, | |
659 | .range_cyclic = 1, | |
0e175a18 | 660 | .reason = reason, |
d46db3d5 | 661 | }; |
edadfb10 | 662 | |
f758eeab | 663 | spin_lock(&wb->list_lock); |
424b351f | 664 | if (list_empty(&wb->b_io)) |
ad4e38dd | 665 | queue_io(wb, &work); |
d46db3d5 | 666 | __writeback_inodes_wb(wb, &work); |
f758eeab | 667 | spin_unlock(&wb->list_lock); |
edadfb10 | 668 | |
d46db3d5 WF |
669 | return nr_pages - work.nr_pages; |
670 | } | |
03ba3782 | 671 | |
b00949aa | 672 | static bool over_bground_thresh(struct backing_dev_info *bdi) |
03ba3782 JA |
673 | { |
674 | unsigned long background_thresh, dirty_thresh; | |
675 | ||
16c4042f | 676 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 677 | |
b00949aa WF |
678 | if (global_page_state(NR_FILE_DIRTY) + |
679 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
680 | return true; | |
681 | ||
682 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > | |
683 | bdi_dirty_limit(bdi, background_thresh)) | |
684 | return true; | |
685 | ||
686 | return false; | |
03ba3782 JA |
687 | } |
688 | ||
e98be2d5 WF |
689 | /* |
690 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
691 | * only the flusher working on the first wb should do it. | |
692 | */ | |
693 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
694 | unsigned long start_time) | |
695 | { | |
af6a3113 | 696 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
697 | } |
698 | ||
03ba3782 JA |
699 | /* |
700 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 701 | * |
03ba3782 JA |
702 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
703 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
704 | * just walks the superblock inode list, writing back any inodes which are | |
705 | * older than a specific point in time. | |
66f3b8e2 | 706 | * |
03ba3782 JA |
707 | * Try to run once per dirty_writeback_interval. But if a writeback event |
708 | * takes longer than a dirty_writeback_interval interval, then leave a | |
709 | * one-second gap. | |
66f3b8e2 | 710 | * |
03ba3782 JA |
711 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
712 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 713 | */ |
c4a77a6c | 714 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 715 | struct wb_writeback_work *work) |
66f3b8e2 | 716 | { |
e98be2d5 | 717 | unsigned long wb_start = jiffies; |
d46db3d5 | 718 | long nr_pages = work->nr_pages; |
03ba3782 | 719 | unsigned long oldest_jif; |
a5989bdc | 720 | struct inode *inode; |
d46db3d5 | 721 | long progress; |
66f3b8e2 | 722 | |
e185dda8 | 723 | oldest_jif = jiffies; |
d46db3d5 | 724 | work->older_than_this = &oldest_jif; |
38f21977 | 725 | |
e8dfc305 | 726 | spin_lock(&wb->list_lock); |
03ba3782 JA |
727 | for (;;) { |
728 | /* | |
d3ddec76 | 729 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 730 | */ |
83ba7b07 | 731 | if (work->nr_pages <= 0) |
03ba3782 | 732 | break; |
66f3b8e2 | 733 | |
aa373cf5 JK |
734 | /* |
735 | * Background writeout and kupdate-style writeback may | |
736 | * run forever. Stop them if there is other work to do | |
737 | * so that e.g. sync can proceed. They'll be restarted | |
738 | * after the other works are all done. | |
739 | */ | |
740 | if ((work->for_background || work->for_kupdate) && | |
741 | !list_empty(&wb->bdi->work_list)) | |
742 | break; | |
743 | ||
38f21977 | 744 | /* |
d3ddec76 WF |
745 | * For background writeout, stop when we are below the |
746 | * background dirty threshold | |
38f21977 | 747 | */ |
b00949aa | 748 | if (work->for_background && !over_bground_thresh(wb->bdi)) |
03ba3782 | 749 | break; |
38f21977 | 750 | |
1bc36b64 JK |
751 | /* |
752 | * Kupdate and background works are special and we want to | |
753 | * include all inodes that need writing. Livelock avoidance is | |
754 | * handled by these works yielding to any other work so we are | |
755 | * safe. | |
756 | */ | |
ba9aa839 WF |
757 | if (work->for_kupdate) { |
758 | oldest_jif = jiffies - | |
759 | msecs_to_jiffies(dirty_expire_interval * 10); | |
1bc36b64 JK |
760 | } else if (work->for_background) |
761 | oldest_jif = jiffies; | |
028c2dd1 | 762 | |
d46db3d5 | 763 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 764 | if (list_empty(&wb->b_io)) |
ad4e38dd | 765 | queue_io(wb, work); |
83ba7b07 | 766 | if (work->sb) |
d46db3d5 | 767 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 768 | else |
d46db3d5 WF |
769 | progress = __writeback_inodes_wb(wb, work); |
770 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 771 | |
e98be2d5 | 772 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
773 | |
774 | /* | |
e6fb6da2 WF |
775 | * Did we write something? Try for more |
776 | * | |
777 | * Dirty inodes are moved to b_io for writeback in batches. | |
778 | * The completion of the current batch does not necessarily | |
779 | * mean the overall work is done. So we keep looping as long | |
780 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 781 | */ |
d46db3d5 | 782 | if (progress) |
71fd05a8 JA |
783 | continue; |
784 | /* | |
e6fb6da2 | 785 | * No more inodes for IO, bail |
71fd05a8 | 786 | */ |
b7a2441f | 787 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 788 | break; |
71fd05a8 JA |
789 | /* |
790 | * Nothing written. Wait for some inode to | |
791 | * become available for writeback. Otherwise | |
792 | * we'll just busyloop. | |
793 | */ | |
71fd05a8 | 794 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 795 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 796 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 797 | spin_lock(&inode->i_lock); |
f758eeab | 798 | inode_wait_for_writeback(inode, wb); |
250df6ed | 799 | spin_unlock(&inode->i_lock); |
03ba3782 JA |
800 | } |
801 | } | |
e8dfc305 | 802 | spin_unlock(&wb->list_lock); |
03ba3782 | 803 | |
d46db3d5 | 804 | return nr_pages - work->nr_pages; |
03ba3782 JA |
805 | } |
806 | ||
807 | /* | |
83ba7b07 | 808 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 809 | */ |
83ba7b07 | 810 | static struct wb_writeback_work * |
08852b6d | 811 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 812 | { |
83ba7b07 | 813 | struct wb_writeback_work *work = NULL; |
03ba3782 | 814 | |
6467716a | 815 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
816 | if (!list_empty(&bdi->work_list)) { |
817 | work = list_entry(bdi->work_list.next, | |
818 | struct wb_writeback_work, list); | |
819 | list_del_init(&work->list); | |
03ba3782 | 820 | } |
6467716a | 821 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 822 | return work; |
03ba3782 JA |
823 | } |
824 | ||
cdf01dd5 LT |
825 | /* |
826 | * Add in the number of potentially dirty inodes, because each inode | |
827 | * write can dirty pagecache in the underlying blockdev. | |
828 | */ | |
829 | static unsigned long get_nr_dirty_pages(void) | |
830 | { | |
831 | return global_page_state(NR_FILE_DIRTY) + | |
832 | global_page_state(NR_UNSTABLE_NFS) + | |
833 | get_nr_dirty_inodes(); | |
834 | } | |
835 | ||
6585027a JK |
836 | static long wb_check_background_flush(struct bdi_writeback *wb) |
837 | { | |
b00949aa | 838 | if (over_bground_thresh(wb->bdi)) { |
6585027a JK |
839 | |
840 | struct wb_writeback_work work = { | |
841 | .nr_pages = LONG_MAX, | |
842 | .sync_mode = WB_SYNC_NONE, | |
843 | .for_background = 1, | |
844 | .range_cyclic = 1, | |
0e175a18 | 845 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
846 | }; |
847 | ||
848 | return wb_writeback(wb, &work); | |
849 | } | |
850 | ||
851 | return 0; | |
852 | } | |
853 | ||
03ba3782 JA |
854 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
855 | { | |
856 | unsigned long expired; | |
857 | long nr_pages; | |
858 | ||
69b62d01 JA |
859 | /* |
860 | * When set to zero, disable periodic writeback | |
861 | */ | |
862 | if (!dirty_writeback_interval) | |
863 | return 0; | |
864 | ||
03ba3782 JA |
865 | expired = wb->last_old_flush + |
866 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
867 | if (time_before(jiffies, expired)) | |
868 | return 0; | |
869 | ||
870 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 871 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 872 | |
c4a77a6c | 873 | if (nr_pages) { |
83ba7b07 | 874 | struct wb_writeback_work work = { |
c4a77a6c JA |
875 | .nr_pages = nr_pages, |
876 | .sync_mode = WB_SYNC_NONE, | |
877 | .for_kupdate = 1, | |
878 | .range_cyclic = 1, | |
0e175a18 | 879 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
880 | }; |
881 | ||
83ba7b07 | 882 | return wb_writeback(wb, &work); |
c4a77a6c | 883 | } |
03ba3782 JA |
884 | |
885 | return 0; | |
886 | } | |
887 | ||
888 | /* | |
889 | * Retrieve work items and do the writeback they describe | |
890 | */ | |
891 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
892 | { | |
893 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 894 | struct wb_writeback_work *work; |
c4a77a6c | 895 | long wrote = 0; |
03ba3782 | 896 | |
81d73a32 | 897 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 898 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
899 | /* |
900 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 901 | * because this thread is exiting now. |
03ba3782 JA |
902 | */ |
903 | if (force_wait) | |
83ba7b07 | 904 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 905 | |
455b2864 DC |
906 | trace_writeback_exec(bdi, work); |
907 | ||
83ba7b07 | 908 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
909 | |
910 | /* | |
83ba7b07 CH |
911 | * Notify the caller of completion if this is a synchronous |
912 | * work item, otherwise just free it. | |
03ba3782 | 913 | */ |
83ba7b07 CH |
914 | if (work->done) |
915 | complete(work->done); | |
916 | else | |
917 | kfree(work); | |
03ba3782 JA |
918 | } |
919 | ||
920 | /* | |
921 | * Check for periodic writeback, kupdated() style | |
922 | */ | |
923 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 924 | wrote += wb_check_background_flush(wb); |
81d73a32 | 925 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
926 | |
927 | return wrote; | |
928 | } | |
929 | ||
930 | /* | |
931 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
932 | * wakes up periodically and does kupdated style flushing. | |
933 | */ | |
08243900 | 934 | int bdi_writeback_thread(void *data) |
03ba3782 | 935 | { |
08243900 CH |
936 | struct bdi_writeback *wb = data; |
937 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
938 | long pages_written; |
939 | ||
766f9164 | 940 | current->flags |= PF_SWAPWRITE; |
08243900 | 941 | set_freezable(); |
ecd58403 | 942 | wb->last_active = jiffies; |
08243900 CH |
943 | |
944 | /* | |
945 | * Our parent may run at a different priority, just set us to normal | |
946 | */ | |
947 | set_user_nice(current, 0); | |
948 | ||
455b2864 DC |
949 | trace_writeback_thread_start(bdi); |
950 | ||
8a32c441 | 951 | while (!kthread_freezable_should_stop(NULL)) { |
6467716a AB |
952 | /* |
953 | * Remove own delayed wake-up timer, since we are already awake | |
954 | * and we'll take care of the preriodic write-back. | |
955 | */ | |
956 | del_timer(&wb->wakeup_timer); | |
957 | ||
03ba3782 JA |
958 | pages_written = wb_do_writeback(wb, 0); |
959 | ||
455b2864 DC |
960 | trace_writeback_pages_written(pages_written); |
961 | ||
03ba3782 | 962 | if (pages_written) |
ecd58403 | 963 | wb->last_active = jiffies; |
03ba3782 | 964 | |
297252c8 | 965 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 966 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 967 | __set_current_state(TASK_RUNNING); |
297252c8 | 968 | continue; |
03ba3782 JA |
969 | } |
970 | ||
253c34e9 | 971 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 972 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
973 | else { |
974 | /* | |
975 | * We have nothing to do, so can go sleep without any | |
976 | * timeout and save power. When a work is queued or | |
977 | * something is made dirty - we will be woken up. | |
978 | */ | |
297252c8 | 979 | schedule(); |
f9eadbbd | 980 | } |
03ba3782 JA |
981 | } |
982 | ||
fff5b85a | 983 | /* Flush any work that raced with us exiting */ |
08243900 CH |
984 | if (!list_empty(&bdi->work_list)) |
985 | wb_do_writeback(wb, 1); | |
455b2864 DC |
986 | |
987 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
988 | return 0; |
989 | } | |
990 | ||
08243900 | 991 | |
03ba3782 | 992 | /* |
b8c2f347 CH |
993 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
994 | * the whole world. | |
03ba3782 | 995 | */ |
0e175a18 | 996 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 997 | { |
b8c2f347 | 998 | struct backing_dev_info *bdi; |
03ba3782 | 999 | |
83ba7b07 CH |
1000 | if (!nr_pages) { |
1001 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
1002 | global_page_state(NR_UNSTABLE_NFS); |
1003 | } | |
03ba3782 | 1004 | |
b8c2f347 | 1005 | rcu_read_lock(); |
cfc4ba53 | 1006 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1007 | if (!bdi_has_dirty_io(bdi)) |
1008 | continue; | |
0e175a18 | 1009 | __bdi_start_writeback(bdi, nr_pages, false, reason); |
03ba3782 | 1010 | } |
cfc4ba53 | 1011 | rcu_read_unlock(); |
1da177e4 LT |
1012 | } |
1013 | ||
03ba3782 JA |
1014 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1015 | { | |
1016 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1017 | struct dentry *dentry; | |
1018 | const char *name = "?"; | |
1019 | ||
1020 | dentry = d_find_alias(inode); | |
1021 | if (dentry) { | |
1022 | spin_lock(&dentry->d_lock); | |
1023 | name = (const char *) dentry->d_name.name; | |
1024 | } | |
1025 | printk(KERN_DEBUG | |
1026 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1027 | current->comm, task_pid_nr(current), inode->i_ino, | |
1028 | name, inode->i_sb->s_id); | |
1029 | if (dentry) { | |
1030 | spin_unlock(&dentry->d_lock); | |
1031 | dput(dentry); | |
1032 | } | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | /** | |
1037 | * __mark_inode_dirty - internal function | |
1038 | * @inode: inode to mark | |
1039 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1040 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1041 | * mark_inode_dirty_sync. | |
1da177e4 | 1042 | * |
03ba3782 JA |
1043 | * Put the inode on the super block's dirty list. |
1044 | * | |
1045 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1046 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1047 | * If it was not hashed, it will never be added to the dirty list | |
1048 | * even if it is later hashed, as it will have been marked dirty already. | |
1049 | * | |
1050 | * In short, make sure you hash any inodes _before_ you start marking | |
1051 | * them dirty. | |
1da177e4 | 1052 | * |
03ba3782 JA |
1053 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1054 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1055 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1056 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1057 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1058 | * blockdev inode. | |
1da177e4 | 1059 | */ |
03ba3782 | 1060 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1061 | { |
03ba3782 | 1062 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1063 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1064 | |
03ba3782 JA |
1065 | /* |
1066 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1067 | * dirty the inode itself | |
1068 | */ | |
1069 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1070 | if (sb->s_op->dirty_inode) | |
aa385729 | 1071 | sb->s_op->dirty_inode(inode, flags); |
03ba3782 JA |
1072 | } |
1073 | ||
1074 | /* | |
1075 | * make sure that changes are seen by all cpus before we test i_state | |
1076 | * -- mikulas | |
1077 | */ | |
1078 | smp_mb(); | |
1079 | ||
1080 | /* avoid the locking if we can */ | |
1081 | if ((inode->i_state & flags) == flags) | |
1082 | return; | |
1083 | ||
1084 | if (unlikely(block_dump)) | |
1085 | block_dump___mark_inode_dirty(inode); | |
1086 | ||
250df6ed | 1087 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1088 | if ((inode->i_state & flags) != flags) { |
1089 | const int was_dirty = inode->i_state & I_DIRTY; | |
1090 | ||
1091 | inode->i_state |= flags; | |
1092 | ||
1093 | /* | |
1094 | * If the inode is being synced, just update its dirty state. | |
1095 | * The unlocker will place the inode on the appropriate | |
1096 | * superblock list, based upon its state. | |
1097 | */ | |
1098 | if (inode->i_state & I_SYNC) | |
250df6ed | 1099 | goto out_unlock_inode; |
03ba3782 JA |
1100 | |
1101 | /* | |
1102 | * Only add valid (hashed) inodes to the superblock's | |
1103 | * dirty list. Add blockdev inodes as well. | |
1104 | */ | |
1105 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1106 | if (inode_unhashed(inode)) |
250df6ed | 1107 | goto out_unlock_inode; |
03ba3782 | 1108 | } |
a4ffdde6 | 1109 | if (inode->i_state & I_FREEING) |
250df6ed | 1110 | goto out_unlock_inode; |
03ba3782 JA |
1111 | |
1112 | /* | |
1113 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1114 | * reposition it (that would break b_dirty time-ordering). | |
1115 | */ | |
1116 | if (!was_dirty) { | |
a66979ab | 1117 | bool wakeup_bdi = false; |
253c34e9 AB |
1118 | bdi = inode_to_bdi(inode); |
1119 | ||
1120 | if (bdi_cap_writeback_dirty(bdi)) { | |
1121 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1122 | "bdi-%s not registered\n", bdi->name); | |
1123 | ||
1124 | /* | |
1125 | * If this is the first dirty inode for this | |
1126 | * bdi, we have to wake-up the corresponding | |
1127 | * bdi thread to make sure background | |
1128 | * write-back happens later. | |
1129 | */ | |
1130 | if (!wb_has_dirty_io(&bdi->wb)) | |
1131 | wakeup_bdi = true; | |
500b067c | 1132 | } |
03ba3782 | 1133 | |
a66979ab | 1134 | spin_unlock(&inode->i_lock); |
f758eeab | 1135 | spin_lock(&bdi->wb.list_lock); |
03ba3782 | 1136 | inode->dirtied_when = jiffies; |
7ccf19a8 | 1137 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1138 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1139 | |
1140 | if (wakeup_bdi) | |
1141 | bdi_wakeup_thread_delayed(bdi); | |
1142 | return; | |
1da177e4 | 1143 | } |
1da177e4 | 1144 | } |
250df6ed DC |
1145 | out_unlock_inode: |
1146 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1147 | |
03ba3782 JA |
1148 | } |
1149 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1150 | ||
b6e51316 | 1151 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1152 | { |
1153 | struct inode *inode, *old_inode = NULL; | |
1154 | ||
1155 | /* | |
1156 | * We need to be protected against the filesystem going from | |
1157 | * r/o to r/w or vice versa. | |
1158 | */ | |
b6e51316 | 1159 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1160 | |
55fa6091 | 1161 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1162 | |
1163 | /* | |
1164 | * Data integrity sync. Must wait for all pages under writeback, | |
1165 | * because there may have been pages dirtied before our sync | |
1166 | * call, but which had writeout started before we write it out. | |
1167 | * In which case, the inode may not be on the dirty list, but | |
1168 | * we still have to wait for that writeout. | |
1169 | */ | |
b6e51316 | 1170 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1171 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1172 | |
250df6ed DC |
1173 | spin_lock(&inode->i_lock); |
1174 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1175 | (mapping->nrpages == 0)) { | |
1176 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1177 | continue; |
250df6ed | 1178 | } |
03ba3782 | 1179 | __iget(inode); |
250df6ed | 1180 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1181 | spin_unlock(&inode_sb_list_lock); |
1182 | ||
03ba3782 | 1183 | /* |
55fa6091 DC |
1184 | * We hold a reference to 'inode' so it couldn't have been |
1185 | * removed from s_inodes list while we dropped the | |
1186 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1187 | * be holding the last reference and we cannot iput it under | |
1188 | * inode_sb_list_lock. So we keep the reference and iput it | |
1189 | * later. | |
03ba3782 JA |
1190 | */ |
1191 | iput(old_inode); | |
1192 | old_inode = inode; | |
1193 | ||
1194 | filemap_fdatawait(mapping); | |
1195 | ||
1196 | cond_resched(); | |
1197 | ||
55fa6091 | 1198 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1199 | } |
55fa6091 | 1200 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1201 | iput(old_inode); |
1da177e4 LT |
1202 | } |
1203 | ||
d8a8559c | 1204 | /** |
3259f8be | 1205 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1206 | * @sb: the superblock |
3259f8be | 1207 | * @nr: the number of pages to write |
786228ab | 1208 | * @reason: reason why some writeback work initiated |
1da177e4 | 1209 | * |
d8a8559c JA |
1210 | * Start writeback on some inodes on this super_block. No guarantees are made |
1211 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1212 | * for IO completion of submitted IO. |
1da177e4 | 1213 | */ |
0e175a18 CW |
1214 | void writeback_inodes_sb_nr(struct super_block *sb, |
1215 | unsigned long nr, | |
1216 | enum wb_reason reason) | |
1da177e4 | 1217 | { |
83ba7b07 CH |
1218 | DECLARE_COMPLETION_ONSTACK(done); |
1219 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1220 | .sb = sb, |
1221 | .sync_mode = WB_SYNC_NONE, | |
1222 | .tagged_writepages = 1, | |
1223 | .done = &done, | |
1224 | .nr_pages = nr, | |
0e175a18 | 1225 | .reason = reason, |
3c4d7165 | 1226 | }; |
d8a8559c | 1227 | |
cf37e972 | 1228 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1229 | bdi_queue_work(sb->s_bdi, &work); |
1230 | wait_for_completion(&done); | |
e913fc82 | 1231 | } |
3259f8be CM |
1232 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1233 | ||
1234 | /** | |
1235 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1236 | * @sb: the superblock | |
786228ab | 1237 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1238 | * |
1239 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1240 | * on how many (if any) will be written, and this function does not wait | |
1241 | * for IO completion of submitted IO. | |
1242 | */ | |
0e175a18 | 1243 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1244 | { |
0e175a18 | 1245 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1246 | } |
0e3c9a22 | 1247 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1248 | |
17bd55d0 ES |
1249 | /** |
1250 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1251 | * @sb: the superblock | |
786228ab | 1252 | * @reason: reason why some writeback work was initiated |
17bd55d0 ES |
1253 | * |
1254 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1255 | * Returns 1 if writeback was started, 0 if not. | |
1256 | */ | |
0e175a18 | 1257 | int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason) |
17bd55d0 ES |
1258 | { |
1259 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1260 | down_read(&sb->s_umount); |
0e175a18 | 1261 | writeback_inodes_sb(sb, reason); |
cf37e972 | 1262 | up_read(&sb->s_umount); |
17bd55d0 ES |
1263 | return 1; |
1264 | } else | |
1265 | return 0; | |
1266 | } | |
1267 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1268 | ||
3259f8be | 1269 | /** |
c097b2ca | 1270 | * writeback_inodes_sb_nr_if_idle - start writeback if none underway |
3259f8be CM |
1271 | * @sb: the superblock |
1272 | * @nr: the number of pages to write | |
786228ab | 1273 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1274 | * |
1275 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1276 | * Returns 1 if writeback was started, 0 if not. | |
1277 | */ | |
1278 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
0e175a18 CW |
1279 | unsigned long nr, |
1280 | enum wb_reason reason) | |
3259f8be CM |
1281 | { |
1282 | if (!writeback_in_progress(sb->s_bdi)) { | |
1283 | down_read(&sb->s_umount); | |
0e175a18 | 1284 | writeback_inodes_sb_nr(sb, nr, reason); |
3259f8be CM |
1285 | up_read(&sb->s_umount); |
1286 | return 1; | |
1287 | } else | |
1288 | return 0; | |
1289 | } | |
1290 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1291 | ||
d8a8559c JA |
1292 | /** |
1293 | * sync_inodes_sb - sync sb inode pages | |
1294 | * @sb: the superblock | |
1295 | * | |
1296 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1297 | * super_block. |
d8a8559c | 1298 | */ |
b6e51316 | 1299 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1300 | { |
83ba7b07 CH |
1301 | DECLARE_COMPLETION_ONSTACK(done); |
1302 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1303 | .sb = sb, |
1304 | .sync_mode = WB_SYNC_ALL, | |
1305 | .nr_pages = LONG_MAX, | |
1306 | .range_cyclic = 0, | |
83ba7b07 | 1307 | .done = &done, |
0e175a18 | 1308 | .reason = WB_REASON_SYNC, |
3c4d7165 CH |
1309 | }; |
1310 | ||
cf37e972 CH |
1311 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1312 | ||
83ba7b07 CH |
1313 | bdi_queue_work(sb->s_bdi, &work); |
1314 | wait_for_completion(&done); | |
1315 | ||
b6e51316 | 1316 | wait_sb_inodes(sb); |
1da177e4 | 1317 | } |
d8a8559c | 1318 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1319 | |
1da177e4 | 1320 | /** |
7f04c26d AA |
1321 | * write_inode_now - write an inode to disk |
1322 | * @inode: inode to write to disk | |
1323 | * @sync: whether the write should be synchronous or not | |
1324 | * | |
1325 | * This function commits an inode to disk immediately if it is dirty. This is | |
1326 | * primarily needed by knfsd. | |
1da177e4 | 1327 | * |
7f04c26d | 1328 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1329 | */ |
1da177e4 LT |
1330 | int write_inode_now(struct inode *inode, int sync) |
1331 | { | |
f758eeab | 1332 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1333 | int ret; |
1334 | struct writeback_control wbc = { | |
1335 | .nr_to_write = LONG_MAX, | |
18914b18 | 1336 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1337 | .range_start = 0, |
1338 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1339 | }; |
1340 | ||
1341 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1342 | wbc.nr_to_write = 0; |
1da177e4 LT |
1343 | |
1344 | might_sleep(); | |
f758eeab | 1345 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1346 | spin_lock(&inode->i_lock); |
f758eeab | 1347 | ret = writeback_single_inode(inode, wb, &wbc); |
0f1b1fd8 | 1348 | spin_unlock(&inode->i_lock); |
f758eeab | 1349 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
1350 | return ret; |
1351 | } | |
1352 | EXPORT_SYMBOL(write_inode_now); | |
1353 | ||
1354 | /** | |
1355 | * sync_inode - write an inode and its pages to disk. | |
1356 | * @inode: the inode to sync | |
1357 | * @wbc: controls the writeback mode | |
1358 | * | |
1359 | * sync_inode() will write an inode and its pages to disk. It will also | |
1360 | * correctly update the inode on its superblock's dirty inode lists and will | |
1361 | * update inode->i_state. | |
1362 | * | |
1363 | * The caller must have a ref on the inode. | |
1364 | */ | |
1365 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1366 | { | |
f758eeab | 1367 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1368 | int ret; |
1369 | ||
f758eeab | 1370 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1371 | spin_lock(&inode->i_lock); |
f758eeab | 1372 | ret = writeback_single_inode(inode, wb, wbc); |
0f1b1fd8 | 1373 | spin_unlock(&inode->i_lock); |
f758eeab | 1374 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
1375 | return ret; |
1376 | } | |
1377 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1378 | |
1379 | /** | |
c691b9d9 | 1380 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1381 | * @inode: the inode to sync |
1382 | * @wait: wait for I/O to complete. | |
1383 | * | |
c691b9d9 | 1384 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1385 | * |
1386 | * Note: only writes the actual inode, no associated data or other metadata. | |
1387 | */ | |
1388 | int sync_inode_metadata(struct inode *inode, int wait) | |
1389 | { | |
1390 | struct writeback_control wbc = { | |
1391 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1392 | .nr_to_write = 0, /* metadata-only */ | |
1393 | }; | |
1394 | ||
1395 | return sync_inode(inode, &wbc); | |
1396 | } | |
1397 | EXPORT_SYMBOL(sync_inode_metadata); |