]>
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; | |
c4a77a6c | 44 | |
8010c3b6 | 45 | struct list_head list; /* pending work list */ |
83ba7b07 | 46 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
47 | }; |
48 | ||
455b2864 DC |
49 | /* |
50 | * Include the creation of the trace points after defining the | |
51 | * wb_writeback_work structure so that the definition remains local to this | |
52 | * file. | |
53 | */ | |
54 | #define CREATE_TRACE_POINTS | |
55 | #include <trace/events/writeback.h> | |
56 | ||
455b2864 DC |
57 | /* |
58 | * We don't actually have pdflush, but this one is exported though /proc... | |
59 | */ | |
60 | int nr_pdflush_threads; | |
61 | ||
f11b00f3 AB |
62 | /** |
63 | * writeback_in_progress - determine whether there is writeback in progress | |
64 | * @bdi: the device's backing_dev_info structure. | |
65 | * | |
03ba3782 JA |
66 | * Determine whether there is writeback waiting to be handled against a |
67 | * backing device. | |
f11b00f3 AB |
68 | */ |
69 | int writeback_in_progress(struct backing_dev_info *bdi) | |
70 | { | |
81d73a32 | 71 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 AB |
72 | } |
73 | ||
692ebd17 JK |
74 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
75 | { | |
76 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 77 | |
aaead25b CH |
78 | if (strcmp(sb->s_type->name, "bdev") == 0) |
79 | return inode->i_mapping->backing_dev_info; | |
80 | ||
81 | return sb->s_bdi; | |
692ebd17 JK |
82 | } |
83 | ||
7ccf19a8 NP |
84 | static inline struct inode *wb_inode(struct list_head *head) |
85 | { | |
86 | return list_entry(head, struct inode, i_wb_list); | |
87 | } | |
88 | ||
6585027a JK |
89 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
90 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) | |
03ba3782 | 91 | { |
fff5b85a AB |
92 | if (bdi->wb.task) { |
93 | wake_up_process(bdi->wb.task); | |
94 | } else { | |
95 | /* | |
96 | * The bdi thread isn't there, wake up the forker thread which | |
97 | * will create and run it. | |
98 | */ | |
03ba3782 | 99 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 100 | } |
6585027a JK |
101 | } |
102 | ||
103 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
104 | struct wb_writeback_work *work) | |
105 | { | |
106 | trace_writeback_queue(bdi, work); | |
107 | ||
108 | spin_lock_bh(&bdi->wb_lock); | |
109 | list_add_tail(&work->list, &bdi->work_list); | |
110 | if (!bdi->wb.task) | |
111 | trace_writeback_nothread(bdi, work); | |
112 | bdi_wakeup_flusher(bdi); | |
6467716a | 113 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
114 | } |
115 | ||
83ba7b07 CH |
116 | static void |
117 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
6585027a | 118 | bool range_cyclic) |
1da177e4 | 119 | { |
83ba7b07 | 120 | struct wb_writeback_work *work; |
03ba3782 | 121 | |
bcddc3f0 JA |
122 | /* |
123 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
124 | * wakeup the thread for old dirty data writeback | |
125 | */ | |
83ba7b07 CH |
126 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
127 | if (!work) { | |
455b2864 DC |
128 | if (bdi->wb.task) { |
129 | trace_writeback_nowork(bdi); | |
83ba7b07 | 130 | wake_up_process(bdi->wb.task); |
455b2864 | 131 | } |
83ba7b07 | 132 | return; |
bcddc3f0 | 133 | } |
03ba3782 | 134 | |
83ba7b07 CH |
135 | work->sync_mode = WB_SYNC_NONE; |
136 | work->nr_pages = nr_pages; | |
137 | work->range_cyclic = range_cyclic; | |
03ba3782 | 138 | |
83ba7b07 | 139 | bdi_queue_work(bdi, work); |
b6e51316 JA |
140 | } |
141 | ||
142 | /** | |
143 | * bdi_start_writeback - start writeback | |
144 | * @bdi: the backing device to write from | |
145 | * @nr_pages: the number of pages to write | |
146 | * | |
147 | * Description: | |
148 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 149 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 150 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
151 | * |
152 | */ | |
c5444198 | 153 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 154 | { |
6585027a | 155 | __bdi_start_writeback(bdi, nr_pages, true); |
c5444198 | 156 | } |
d3ddec76 | 157 | |
c5444198 CH |
158 | /** |
159 | * bdi_start_background_writeback - start background writeback | |
160 | * @bdi: the backing device to write from | |
161 | * | |
162 | * Description: | |
6585027a JK |
163 | * This makes sure WB_SYNC_NONE background writeback happens. When |
164 | * this function returns, it is only guaranteed that for given BDI | |
165 | * some IO is happening if we are over background dirty threshold. | |
166 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
167 | */ |
168 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
169 | { | |
6585027a JK |
170 | /* |
171 | * We just wake up the flusher thread. It will perform background | |
172 | * writeback as soon as there is no other work to do. | |
173 | */ | |
71927e84 | 174 | trace_writeback_wake_background(bdi); |
6585027a JK |
175 | spin_lock_bh(&bdi->wb_lock); |
176 | bdi_wakeup_flusher(bdi); | |
177 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
178 | } |
179 | ||
a66979ab DC |
180 | /* |
181 | * Remove the inode from the writeback list it is on. | |
182 | */ | |
183 | void inode_wb_list_del(struct inode *inode) | |
184 | { | |
f758eeab CH |
185 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
186 | ||
187 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 188 | list_del_init(&inode->i_wb_list); |
f758eeab | 189 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
190 | } |
191 | ||
6610a0bc AM |
192 | /* |
193 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
194 | * furthest end of its superblock's dirty-inode list. | |
195 | * | |
196 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 197 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
198 | * the case then the inode must have been redirtied while it was being written |
199 | * out and we don't reset its dirtied_when. | |
200 | */ | |
f758eeab | 201 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 202 | { |
f758eeab | 203 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 204 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 205 | struct inode *tail; |
6610a0bc | 206 | |
7ccf19a8 | 207 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 208 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
209 | inode->dirtied_when = jiffies; |
210 | } | |
7ccf19a8 | 211 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
212 | } |
213 | ||
c986d1e2 | 214 | /* |
66f3b8e2 | 215 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 216 | */ |
f758eeab | 217 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 218 | { |
f758eeab | 219 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 220 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
221 | } |
222 | ||
1c0eeaf5 JE |
223 | static void inode_sync_complete(struct inode *inode) |
224 | { | |
225 | /* | |
a66979ab | 226 | * Prevent speculative execution through |
f758eeab | 227 | * spin_unlock(&wb->list_lock); |
1c0eeaf5 | 228 | */ |
a66979ab | 229 | |
1c0eeaf5 JE |
230 | smp_mb(); |
231 | wake_up_bit(&inode->i_state, __I_SYNC); | |
232 | } | |
233 | ||
d2caa3c5 JL |
234 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
235 | { | |
236 | bool ret = time_after(inode->dirtied_when, t); | |
237 | #ifndef CONFIG_64BIT | |
238 | /* | |
239 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
240 | * It _appears_ to be in the future, but is actually in distant past. | |
241 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 242 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
243 | */ |
244 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
245 | #endif | |
246 | return ret; | |
247 | } | |
248 | ||
2c136579 FW |
249 | /* |
250 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
251 | */ | |
e84d0a4f | 252 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 253 | struct list_head *dispatch_queue, |
e84d0a4f | 254 | unsigned long *older_than_this) |
2c136579 | 255 | { |
5c03449d SL |
256 | LIST_HEAD(tmp); |
257 | struct list_head *pos, *node; | |
cf137307 | 258 | struct super_block *sb = NULL; |
5c03449d | 259 | struct inode *inode; |
cf137307 | 260 | int do_sb_sort = 0; |
e84d0a4f | 261 | int moved = 0; |
5c03449d | 262 | |
2c136579 | 263 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 264 | inode = wb_inode(delaying_queue->prev); |
2c136579 | 265 | if (older_than_this && |
d2caa3c5 | 266 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 267 | break; |
cf137307 JA |
268 | if (sb && sb != inode->i_sb) |
269 | do_sb_sort = 1; | |
270 | sb = inode->i_sb; | |
7ccf19a8 | 271 | list_move(&inode->i_wb_list, &tmp); |
e84d0a4f | 272 | moved++; |
5c03449d SL |
273 | } |
274 | ||
cf137307 JA |
275 | /* just one sb in list, splice to dispatch_queue and we're done */ |
276 | if (!do_sb_sort) { | |
277 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 278 | goto out; |
cf137307 JA |
279 | } |
280 | ||
5c03449d SL |
281 | /* Move inodes from one superblock together */ |
282 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 283 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 284 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 285 | inode = wb_inode(pos); |
5c03449d | 286 | if (inode->i_sb == sb) |
7ccf19a8 | 287 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 288 | } |
2c136579 | 289 | } |
e84d0a4f WF |
290 | out: |
291 | return moved; | |
2c136579 FW |
292 | } |
293 | ||
294 | /* | |
295 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
296 | * Before |
297 | * newly dirtied b_dirty b_io b_more_io | |
298 | * =============> gf edc BA | |
299 | * After | |
300 | * newly dirtied b_dirty b_io b_more_io | |
301 | * =============> g fBAedc | |
302 | * | | |
303 | * +--> dequeue for IO | |
2c136579 | 304 | */ |
03ba3782 | 305 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 306 | { |
e84d0a4f | 307 | int moved; |
f758eeab | 308 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 309 | list_splice_init(&wb->b_more_io, &wb->b_io); |
e84d0a4f WF |
310 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
311 | trace_writeback_queue_io(wb, older_than_this, moved); | |
66f3b8e2 JA |
312 | } |
313 | ||
a9185b41 | 314 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 315 | { |
03ba3782 | 316 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 317 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 318 | return 0; |
08d8e974 | 319 | } |
08d8e974 | 320 | |
1da177e4 | 321 | /* |
01c03194 CH |
322 | * Wait for writeback on an inode to complete. |
323 | */ | |
f758eeab CH |
324 | static void inode_wait_for_writeback(struct inode *inode, |
325 | struct bdi_writeback *wb) | |
01c03194 CH |
326 | { |
327 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
328 | wait_queue_head_t *wqh; | |
329 | ||
330 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
331 | while (inode->i_state & I_SYNC) { |
332 | spin_unlock(&inode->i_lock); | |
f758eeab | 333 | spin_unlock(&wb->list_lock); |
01c03194 | 334 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
f758eeab | 335 | spin_lock(&wb->list_lock); |
250df6ed | 336 | spin_lock(&inode->i_lock); |
58a9d3d8 | 337 | } |
01c03194 CH |
338 | } |
339 | ||
340 | /* | |
f758eeab | 341 | * Write out an inode's dirty pages. Called under wb->list_lock and |
0f1b1fd8 DC |
342 | * inode->i_lock. Either the caller has an active reference on the inode or |
343 | * the inode has I_WILL_FREE set. | |
01c03194 | 344 | * |
1da177e4 LT |
345 | * If `wait' is set, wait on the writeout. |
346 | * | |
347 | * The whole writeout design is quite complex and fragile. We want to avoid | |
348 | * starvation of particular inodes when others are being redirtied, prevent | |
349 | * livelocks, etc. | |
1da177e4 LT |
350 | */ |
351 | static int | |
f758eeab CH |
352 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
353 | struct writeback_control *wbc) | |
1da177e4 | 354 | { |
1da177e4 | 355 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 356 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 357 | unsigned dirty; |
1da177e4 LT |
358 | int ret; |
359 | ||
f758eeab | 360 | assert_spin_locked(&wb->list_lock); |
0f1b1fd8 DC |
361 | assert_spin_locked(&inode->i_lock); |
362 | ||
01c03194 CH |
363 | if (!atomic_read(&inode->i_count)) |
364 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
365 | else | |
366 | WARN_ON(inode->i_state & I_WILL_FREE); | |
367 | ||
368 | if (inode->i_state & I_SYNC) { | |
369 | /* | |
370 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 371 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
372 | * writeback can proceed with the other inodes on s_io. |
373 | * | |
374 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 375 | * completed a full scan of b_io. |
01c03194 | 376 | */ |
a9185b41 | 377 | if (wbc->sync_mode != WB_SYNC_ALL) { |
f758eeab | 378 | requeue_io(inode, wb); |
251d6a47 WF |
379 | trace_writeback_single_inode_requeue(inode, wbc, |
380 | nr_to_write); | |
01c03194 CH |
381 | return 0; |
382 | } | |
383 | ||
384 | /* | |
385 | * It's a data-integrity sync. We must wait. | |
386 | */ | |
f758eeab | 387 | inode_wait_for_writeback(inode, wb); |
01c03194 CH |
388 | } |
389 | ||
1c0eeaf5 | 390 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 391 | |
5547e8aa | 392 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 393 | inode->i_state |= I_SYNC; |
5547e8aa | 394 | inode->i_state &= ~I_DIRTY_PAGES; |
250df6ed | 395 | spin_unlock(&inode->i_lock); |
f758eeab | 396 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
397 | |
398 | ret = do_writepages(mapping, wbc); | |
399 | ||
26821ed4 CH |
400 | /* |
401 | * Make sure to wait on the data before writing out the metadata. | |
402 | * This is important for filesystems that modify metadata on data | |
403 | * I/O completion. | |
404 | */ | |
a9185b41 | 405 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 406 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
407 | if (ret == 0) |
408 | ret = err; | |
409 | } | |
410 | ||
5547e8aa DM |
411 | /* |
412 | * Some filesystems may redirty the inode during the writeback | |
413 | * due to delalloc, clear dirty metadata flags right before | |
414 | * write_inode() | |
415 | */ | |
250df6ed | 416 | spin_lock(&inode->i_lock); |
5547e8aa DM |
417 | dirty = inode->i_state & I_DIRTY; |
418 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 419 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
420 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
421 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 422 | int err = write_inode(inode, wbc); |
1da177e4 LT |
423 | if (ret == 0) |
424 | ret = err; | |
425 | } | |
426 | ||
f758eeab | 427 | spin_lock(&wb->list_lock); |
250df6ed | 428 | spin_lock(&inode->i_lock); |
1c0eeaf5 | 429 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 430 | if (!(inode->i_state & I_FREEING)) { |
94c3dcbb WF |
431 | /* |
432 | * Sync livelock prevention. Each inode is tagged and synced in | |
433 | * one shot. If still dirty, it will be redirty_tail()'ed below. | |
434 | * Update the dirty time to prevent enqueue and sync it again. | |
435 | */ | |
436 | if ((inode->i_state & I_DIRTY) && | |
437 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
438 | inode->dirtied_when = jiffies; | |
439 | ||
23539afc | 440 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
441 | /* |
442 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 443 | * sometimes bales out without doing anything. |
1b43ef91 | 444 | */ |
a50aeb40 WF |
445 | inode->i_state |= I_DIRTY_PAGES; |
446 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 447 | /* |
a50aeb40 | 448 | * slice used up: queue for next turn |
1da177e4 | 449 | */ |
f758eeab | 450 | requeue_io(inode, wb); |
1da177e4 LT |
451 | } else { |
452 | /* | |
a50aeb40 WF |
453 | * Writeback blocked by something other than |
454 | * congestion. Delay the inode for some time to | |
455 | * avoid spinning on the CPU (100% iowait) | |
456 | * retrying writeback of the dirty page/inode | |
457 | * that cannot be performed immediately. | |
1da177e4 | 458 | */ |
f758eeab | 459 | redirty_tail(inode, wb); |
1da177e4 | 460 | } |
23539afc WF |
461 | } else if (inode->i_state & I_DIRTY) { |
462 | /* | |
463 | * Filesystems can dirty the inode during writeback | |
464 | * operations, such as delayed allocation during | |
465 | * submission or metadata updates after data IO | |
466 | * completion. | |
467 | */ | |
f758eeab | 468 | redirty_tail(inode, wb); |
1da177e4 LT |
469 | } else { |
470 | /* | |
9e38d86f NP |
471 | * The inode is clean. At this point we either have |
472 | * a reference to the inode or it's on it's way out. | |
473 | * No need to add it back to the LRU. | |
1da177e4 | 474 | */ |
7ccf19a8 | 475 | list_del_init(&inode->i_wb_list); |
1da177e4 LT |
476 | } |
477 | } | |
1c0eeaf5 | 478 | inode_sync_complete(inode); |
251d6a47 | 479 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
1da177e4 LT |
480 | return ret; |
481 | } | |
482 | ||
03ba3782 | 483 | /* |
d19de7ed | 484 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
485 | * before calling writeback. So make sure that we do pin it, so it doesn't |
486 | * go away while we are writing inodes from it. | |
03ba3782 | 487 | */ |
d19de7ed | 488 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 489 | { |
03ba3782 | 490 | spin_lock(&sb_lock); |
29cb4859 CH |
491 | if (list_empty(&sb->s_instances)) { |
492 | spin_unlock(&sb_lock); | |
493 | return false; | |
494 | } | |
495 | ||
03ba3782 | 496 | sb->s_count++; |
29cb4859 CH |
497 | spin_unlock(&sb_lock); |
498 | ||
03ba3782 | 499 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 500 | if (sb->s_root) |
d19de7ed | 501 | return true; |
03ba3782 JA |
502 | up_read(&sb->s_umount); |
503 | } | |
29cb4859 CH |
504 | |
505 | put_super(sb); | |
d19de7ed | 506 | return false; |
03ba3782 JA |
507 | } |
508 | ||
1a12d8bd WF |
509 | static long writeback_chunk_size(struct backing_dev_info *bdi, |
510 | struct wb_writeback_work *work) | |
d46db3d5 WF |
511 | { |
512 | long pages; | |
513 | ||
514 | /* | |
515 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
516 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
517 | * here avoids calling into writeback_inodes_wb() more than once. | |
518 | * | |
519 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
520 | * | |
521 | * wb_writeback() | |
522 | * writeback_sb_inodes() <== called only once | |
523 | * write_cache_pages() <== called once for each inode | |
524 | * (quickly) tag currently dirty pages | |
525 | * (maybe slowly) sync all tagged pages | |
526 | */ | |
527 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
528 | pages = LONG_MAX; | |
1a12d8bd WF |
529 | else { |
530 | pages = min(bdi->avg_write_bandwidth / 2, | |
531 | global_dirty_limit / DIRTY_SCOPE); | |
532 | pages = min(pages, work->nr_pages); | |
533 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
534 | MIN_WRITEBACK_PAGES); | |
535 | } | |
d46db3d5 WF |
536 | |
537 | return pages; | |
538 | } | |
539 | ||
f11c9c5c ES |
540 | /* |
541 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
542 | * |
543 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
544 | * inodes. Otherwise write only ones which go sequentially |
545 | * in reverse order. | |
edadfb10 | 546 | * |
d46db3d5 | 547 | * Return the number of pages and/or inodes written. |
f11c9c5c | 548 | */ |
d46db3d5 WF |
549 | static long writeback_sb_inodes(struct super_block *sb, |
550 | struct bdi_writeback *wb, | |
551 | struct wb_writeback_work *work) | |
1da177e4 | 552 | { |
d46db3d5 WF |
553 | struct writeback_control wbc = { |
554 | .sync_mode = work->sync_mode, | |
555 | .tagged_writepages = work->tagged_writepages, | |
556 | .for_kupdate = work->for_kupdate, | |
557 | .for_background = work->for_background, | |
558 | .range_cyclic = work->range_cyclic, | |
559 | .range_start = 0, | |
560 | .range_end = LLONG_MAX, | |
561 | }; | |
562 | unsigned long start_time = jiffies; | |
563 | long write_chunk; | |
564 | long wrote = 0; /* count both pages and inodes */ | |
565 | ||
03ba3782 | 566 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 567 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
568 | |
569 | if (inode->i_sb != sb) { | |
d46db3d5 | 570 | if (work->sb) { |
edadfb10 CH |
571 | /* |
572 | * We only want to write back data for this | |
573 | * superblock, move all inodes not belonging | |
574 | * to it back onto the dirty list. | |
575 | */ | |
f758eeab | 576 | redirty_tail(inode, wb); |
edadfb10 CH |
577 | continue; |
578 | } | |
579 | ||
580 | /* | |
581 | * The inode belongs to a different superblock. | |
582 | * Bounce back to the caller to unpin this and | |
583 | * pin the next superblock. | |
584 | */ | |
d46db3d5 | 585 | break; |
edadfb10 CH |
586 | } |
587 | ||
9843b76a CH |
588 | /* |
589 | * Don't bother with new inodes or inodes beeing freed, first | |
590 | * kind does not need peridic writeout yet, and for the latter | |
591 | * kind writeout is handled by the freer. | |
592 | */ | |
250df6ed | 593 | spin_lock(&inode->i_lock); |
9843b76a | 594 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 595 | spin_unlock(&inode->i_lock); |
f758eeab | 596 | requeue_io(inode, wb); |
7ef0d737 NP |
597 | continue; |
598 | } | |
1da177e4 | 599 | __iget(inode); |
1a12d8bd | 600 | write_chunk = writeback_chunk_size(wb->bdi, work); |
d46db3d5 WF |
601 | wbc.nr_to_write = write_chunk; |
602 | wbc.pages_skipped = 0; | |
603 | ||
604 | writeback_single_inode(inode, wb, &wbc); | |
250df6ed | 605 | |
d46db3d5 WF |
606 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
607 | wrote += write_chunk - wbc.nr_to_write; | |
608 | if (!(inode->i_state & I_DIRTY)) | |
609 | wrote++; | |
610 | if (wbc.pages_skipped) { | |
1da177e4 LT |
611 | /* |
612 | * writeback is not making progress due to locked | |
613 | * buffers. Skip this inode for now. | |
614 | */ | |
f758eeab | 615 | redirty_tail(inode, wb); |
1da177e4 | 616 | } |
0f1b1fd8 | 617 | spin_unlock(&inode->i_lock); |
f758eeab | 618 | spin_unlock(&wb->list_lock); |
1da177e4 | 619 | iput(inode); |
4ffc8444 | 620 | cond_resched(); |
f758eeab | 621 | spin_lock(&wb->list_lock); |
d46db3d5 WF |
622 | /* |
623 | * bail out to wb_writeback() often enough to check | |
624 | * background threshold and other termination conditions. | |
625 | */ | |
626 | if (wrote) { | |
627 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
628 | break; | |
629 | if (work->nr_pages <= 0) | |
630 | break; | |
631 | } | |
1da177e4 | 632 | } |
d46db3d5 | 633 | return wrote; |
f11c9c5c ES |
634 | } |
635 | ||
d46db3d5 WF |
636 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
637 | struct wb_writeback_work *work) | |
f11c9c5c | 638 | { |
d46db3d5 WF |
639 | unsigned long start_time = jiffies; |
640 | long wrote = 0; | |
f11c9c5c | 641 | |
f11c9c5c | 642 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 643 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 644 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 645 | |
edadfb10 | 646 | if (!pin_sb_for_writeback(sb)) { |
f758eeab | 647 | requeue_io(inode, wb); |
edadfb10 | 648 | continue; |
f11c9c5c | 649 | } |
d46db3d5 | 650 | wrote += writeback_sb_inodes(sb, wb, work); |
edadfb10 | 651 | drop_super(sb); |
f11c9c5c | 652 | |
d46db3d5 WF |
653 | /* refer to the same tests at the end of writeback_sb_inodes */ |
654 | if (wrote) { | |
655 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
656 | break; | |
657 | if (work->nr_pages <= 0) | |
658 | break; | |
659 | } | |
f11c9c5c | 660 | } |
66f3b8e2 | 661 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 662 | return wrote; |
66f3b8e2 JA |
663 | } |
664 | ||
d46db3d5 | 665 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages) |
edadfb10 | 666 | { |
d46db3d5 WF |
667 | struct wb_writeback_work work = { |
668 | .nr_pages = nr_pages, | |
669 | .sync_mode = WB_SYNC_NONE, | |
670 | .range_cyclic = 1, | |
671 | }; | |
672 | ||
f758eeab | 673 | spin_lock(&wb->list_lock); |
424b351f | 674 | if (list_empty(&wb->b_io)) |
d46db3d5 WF |
675 | queue_io(wb, NULL); |
676 | __writeback_inodes_wb(wb, &work); | |
f758eeab | 677 | spin_unlock(&wb->list_lock); |
edadfb10 | 678 | |
d46db3d5 WF |
679 | return nr_pages - work.nr_pages; |
680 | } | |
03ba3782 JA |
681 | |
682 | static inline bool over_bground_thresh(void) | |
683 | { | |
684 | unsigned long background_thresh, dirty_thresh; | |
685 | ||
16c4042f | 686 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 JA |
687 | |
688 | return (global_page_state(NR_FILE_DIRTY) + | |
4cbec4c8 | 689 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
03ba3782 JA |
690 | } |
691 | ||
e98be2d5 WF |
692 | /* |
693 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
694 | * only the flusher working on the first wb should do it. | |
695 | */ | |
696 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
697 | unsigned long start_time) | |
698 | { | |
c42843f2 | 699 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
700 | } |
701 | ||
03ba3782 JA |
702 | /* |
703 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 704 | * |
03ba3782 JA |
705 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
706 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
707 | * just walks the superblock inode list, writing back any inodes which are | |
708 | * older than a specific point in time. | |
66f3b8e2 | 709 | * |
03ba3782 JA |
710 | * Try to run once per dirty_writeback_interval. But if a writeback event |
711 | * takes longer than a dirty_writeback_interval interval, then leave a | |
712 | * one-second gap. | |
66f3b8e2 | 713 | * |
03ba3782 JA |
714 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
715 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 716 | */ |
c4a77a6c | 717 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 718 | struct wb_writeback_work *work) |
66f3b8e2 | 719 | { |
e98be2d5 | 720 | unsigned long wb_start = jiffies; |
d46db3d5 | 721 | long nr_pages = work->nr_pages; |
03ba3782 | 722 | unsigned long oldest_jif; |
a5989bdc | 723 | struct inode *inode; |
d46db3d5 | 724 | long progress; |
b9543dac | 725 | |
e185dda8 | 726 | oldest_jif = jiffies; |
d46db3d5 | 727 | work->older_than_this = &oldest_jif; |
e185dda8 | 728 | |
e8dfc305 | 729 | spin_lock(&wb->list_lock); |
03ba3782 JA |
730 | for (;;) { |
731 | /* | |
d3ddec76 | 732 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 733 | */ |
83ba7b07 | 734 | if (work->nr_pages <= 0) |
03ba3782 | 735 | break; |
66f3b8e2 | 736 | |
aa373cf5 JK |
737 | /* |
738 | * Background writeout and kupdate-style writeback may | |
739 | * run forever. Stop them if there is other work to do | |
740 | * so that e.g. sync can proceed. They'll be restarted | |
741 | * after the other works are all done. | |
742 | */ | |
743 | if ((work->for_background || work->for_kupdate) && | |
744 | !list_empty(&wb->bdi->work_list)) | |
745 | break; | |
746 | ||
38f21977 | 747 | /* |
d3ddec76 WF |
748 | * For background writeout, stop when we are below the |
749 | * background dirty threshold | |
38f21977 | 750 | */ |
83ba7b07 | 751 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 752 | break; |
38f21977 | 753 | |
ba9aa839 WF |
754 | if (work->for_kupdate) { |
755 | oldest_jif = jiffies - | |
756 | msecs_to_jiffies(dirty_expire_interval * 10); | |
d46db3d5 | 757 | work->older_than_this = &oldest_jif; |
ba9aa839 WF |
758 | } |
759 | ||
d46db3d5 | 760 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 761 | if (list_empty(&wb->b_io)) |
d46db3d5 | 762 | queue_io(wb, work->older_than_this); |
83ba7b07 | 763 | if (work->sb) |
d46db3d5 | 764 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 765 | else |
d46db3d5 WF |
766 | progress = __writeback_inodes_wb(wb, work); |
767 | trace_writeback_written(wb->bdi, work); | |
03ba3782 | 768 | |
e98be2d5 WF |
769 | wb_update_bandwidth(wb, wb_start); |
770 | ||
03ba3782 | 771 | /* |
e6fb6da2 WF |
772 | * Did we write something? Try for more |
773 | * | |
774 | * Dirty inodes are moved to b_io for writeback in batches. | |
775 | * The completion of the current batch does not necessarily | |
776 | * mean the overall work is done. So we keep looping as long | |
777 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 778 | */ |
d46db3d5 | 779 | if (progress) |
cb9bd115 | 780 | continue; |
71fd05a8 | 781 | /* |
e6fb6da2 | 782 | * No more inodes for IO, bail |
71fd05a8 | 783 | */ |
b7a2441f | 784 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 785 | break; |
71fd05a8 JA |
786 | /* |
787 | * Nothing written. Wait for some inode to | |
788 | * become available for writeback. Otherwise | |
789 | * we'll just busyloop. | |
790 | */ | |
71fd05a8 | 791 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 792 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 793 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 794 | spin_lock(&inode->i_lock); |
f758eeab | 795 | inode_wait_for_writeback(inode, wb); |
250df6ed | 796 | spin_unlock(&inode->i_lock); |
03ba3782 JA |
797 | } |
798 | } | |
e8dfc305 | 799 | spin_unlock(&wb->list_lock); |
03ba3782 | 800 | |
d46db3d5 | 801 | return nr_pages - work->nr_pages; |
03ba3782 JA |
802 | } |
803 | ||
804 | /* | |
83ba7b07 | 805 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 806 | */ |
83ba7b07 | 807 | static struct wb_writeback_work * |
08852b6d | 808 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 809 | { |
83ba7b07 | 810 | struct wb_writeback_work *work = NULL; |
03ba3782 | 811 | |
6467716a | 812 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
813 | if (!list_empty(&bdi->work_list)) { |
814 | work = list_entry(bdi->work_list.next, | |
815 | struct wb_writeback_work, list); | |
816 | list_del_init(&work->list); | |
03ba3782 | 817 | } |
6467716a | 818 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 819 | return work; |
03ba3782 JA |
820 | } |
821 | ||
cdf01dd5 LT |
822 | /* |
823 | * Add in the number of potentially dirty inodes, because each inode | |
824 | * write can dirty pagecache in the underlying blockdev. | |
825 | */ | |
826 | static unsigned long get_nr_dirty_pages(void) | |
827 | { | |
828 | return global_page_state(NR_FILE_DIRTY) + | |
829 | global_page_state(NR_UNSTABLE_NFS) + | |
830 | get_nr_dirty_inodes(); | |
831 | } | |
832 | ||
6585027a JK |
833 | static long wb_check_background_flush(struct bdi_writeback *wb) |
834 | { | |
835 | if (over_bground_thresh()) { | |
836 | ||
837 | struct wb_writeback_work work = { | |
838 | .nr_pages = LONG_MAX, | |
839 | .sync_mode = WB_SYNC_NONE, | |
840 | .for_background = 1, | |
841 | .range_cyclic = 1, | |
842 | }; | |
843 | ||
844 | return wb_writeback(wb, &work); | |
845 | } | |
846 | ||
847 | return 0; | |
848 | } | |
849 | ||
03ba3782 JA |
850 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
851 | { | |
852 | unsigned long expired; | |
853 | long nr_pages; | |
854 | ||
69b62d01 JA |
855 | /* |
856 | * When set to zero, disable periodic writeback | |
857 | */ | |
858 | if (!dirty_writeback_interval) | |
859 | return 0; | |
860 | ||
03ba3782 JA |
861 | expired = wb->last_old_flush + |
862 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
863 | if (time_before(jiffies, expired)) | |
864 | return 0; | |
865 | ||
866 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 867 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 868 | |
c4a77a6c | 869 | if (nr_pages) { |
83ba7b07 | 870 | struct wb_writeback_work work = { |
c4a77a6c JA |
871 | .nr_pages = nr_pages, |
872 | .sync_mode = WB_SYNC_NONE, | |
873 | .for_kupdate = 1, | |
874 | .range_cyclic = 1, | |
875 | }; | |
876 | ||
83ba7b07 | 877 | return wb_writeback(wb, &work); |
c4a77a6c | 878 | } |
03ba3782 JA |
879 | |
880 | return 0; | |
881 | } | |
882 | ||
883 | /* | |
884 | * Retrieve work items and do the writeback they describe | |
885 | */ | |
886 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
887 | { | |
888 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 889 | struct wb_writeback_work *work; |
c4a77a6c | 890 | long wrote = 0; |
03ba3782 | 891 | |
81d73a32 | 892 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 893 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
894 | /* |
895 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 896 | * because this thread is exiting now. |
03ba3782 JA |
897 | */ |
898 | if (force_wait) | |
83ba7b07 | 899 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 900 | |
455b2864 DC |
901 | trace_writeback_exec(bdi, work); |
902 | ||
83ba7b07 | 903 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
904 | |
905 | /* | |
83ba7b07 CH |
906 | * Notify the caller of completion if this is a synchronous |
907 | * work item, otherwise just free it. | |
03ba3782 | 908 | */ |
83ba7b07 CH |
909 | if (work->done) |
910 | complete(work->done); | |
911 | else | |
912 | kfree(work); | |
03ba3782 JA |
913 | } |
914 | ||
915 | /* | |
916 | * Check for periodic writeback, kupdated() style | |
917 | */ | |
918 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 919 | wrote += wb_check_background_flush(wb); |
81d73a32 | 920 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
921 | |
922 | return wrote; | |
923 | } | |
924 | ||
925 | /* | |
926 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
927 | * wakes up periodically and does kupdated style flushing. | |
928 | */ | |
08243900 | 929 | int bdi_writeback_thread(void *data) |
03ba3782 | 930 | { |
08243900 CH |
931 | struct bdi_writeback *wb = data; |
932 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
933 | long pages_written; |
934 | ||
766f9164 | 935 | current->flags |= PF_SWAPWRITE; |
08243900 | 936 | set_freezable(); |
ecd58403 | 937 | wb->last_active = jiffies; |
08243900 CH |
938 | |
939 | /* | |
940 | * Our parent may run at a different priority, just set us to normal | |
941 | */ | |
942 | set_user_nice(current, 0); | |
943 | ||
455b2864 DC |
944 | trace_writeback_thread_start(bdi); |
945 | ||
03ba3782 | 946 | while (!kthread_should_stop()) { |
6467716a AB |
947 | /* |
948 | * Remove own delayed wake-up timer, since we are already awake | |
949 | * and we'll take care of the preriodic write-back. | |
950 | */ | |
951 | del_timer(&wb->wakeup_timer); | |
952 | ||
03ba3782 JA |
953 | pages_written = wb_do_writeback(wb, 0); |
954 | ||
455b2864 DC |
955 | trace_writeback_pages_written(pages_written); |
956 | ||
03ba3782 | 957 | if (pages_written) |
ecd58403 | 958 | wb->last_active = jiffies; |
03ba3782 | 959 | |
297252c8 | 960 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 961 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 962 | __set_current_state(TASK_RUNNING); |
297252c8 | 963 | continue; |
03ba3782 JA |
964 | } |
965 | ||
253c34e9 | 966 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 967 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
968 | else { |
969 | /* | |
970 | * We have nothing to do, so can go sleep without any | |
971 | * timeout and save power. When a work is queued or | |
972 | * something is made dirty - we will be woken up. | |
973 | */ | |
297252c8 | 974 | schedule(); |
f9eadbbd | 975 | } |
69b62d01 | 976 | |
03ba3782 JA |
977 | try_to_freeze(); |
978 | } | |
979 | ||
fff5b85a | 980 | /* Flush any work that raced with us exiting */ |
08243900 CH |
981 | if (!list_empty(&bdi->work_list)) |
982 | wb_do_writeback(wb, 1); | |
455b2864 DC |
983 | |
984 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
985 | return 0; |
986 | } | |
987 | ||
08243900 | 988 | |
03ba3782 | 989 | /* |
b8c2f347 CH |
990 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
991 | * the whole world. | |
03ba3782 | 992 | */ |
b8c2f347 | 993 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 994 | { |
b8c2f347 | 995 | struct backing_dev_info *bdi; |
03ba3782 | 996 | |
83ba7b07 CH |
997 | if (!nr_pages) { |
998 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
999 | global_page_state(NR_UNSTABLE_NFS); |
1000 | } | |
03ba3782 | 1001 | |
b8c2f347 | 1002 | rcu_read_lock(); |
cfc4ba53 | 1003 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1004 | if (!bdi_has_dirty_io(bdi)) |
1005 | continue; | |
6585027a | 1006 | __bdi_start_writeback(bdi, nr_pages, false); |
03ba3782 | 1007 | } |
cfc4ba53 | 1008 | rcu_read_unlock(); |
1da177e4 LT |
1009 | } |
1010 | ||
03ba3782 JA |
1011 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1012 | { | |
1013 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1014 | struct dentry *dentry; | |
1015 | const char *name = "?"; | |
1016 | ||
1017 | dentry = d_find_alias(inode); | |
1018 | if (dentry) { | |
1019 | spin_lock(&dentry->d_lock); | |
1020 | name = (const char *) dentry->d_name.name; | |
1021 | } | |
1022 | printk(KERN_DEBUG | |
1023 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1024 | current->comm, task_pid_nr(current), inode->i_ino, | |
1025 | name, inode->i_sb->s_id); | |
1026 | if (dentry) { | |
1027 | spin_unlock(&dentry->d_lock); | |
1028 | dput(dentry); | |
1029 | } | |
1030 | } | |
1031 | } | |
1032 | ||
1033 | /** | |
1034 | * __mark_inode_dirty - internal function | |
1035 | * @inode: inode to mark | |
1036 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1037 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1038 | * mark_inode_dirty_sync. | |
1da177e4 | 1039 | * |
03ba3782 JA |
1040 | * Put the inode on the super block's dirty list. |
1041 | * | |
1042 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1043 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1044 | * If it was not hashed, it will never be added to the dirty list | |
1045 | * even if it is later hashed, as it will have been marked dirty already. | |
1046 | * | |
1047 | * In short, make sure you hash any inodes _before_ you start marking | |
1048 | * them dirty. | |
1da177e4 | 1049 | * |
03ba3782 JA |
1050 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1051 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1052 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1053 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1054 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1055 | * blockdev inode. | |
1da177e4 | 1056 | */ |
03ba3782 | 1057 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1058 | { |
03ba3782 | 1059 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1060 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1061 | |
03ba3782 JA |
1062 | /* |
1063 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1064 | * dirty the inode itself | |
1065 | */ | |
1066 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1067 | if (sb->s_op->dirty_inode) | |
aa385729 | 1068 | sb->s_op->dirty_inode(inode, flags); |
03ba3782 JA |
1069 | } |
1070 | ||
1071 | /* | |
1072 | * make sure that changes are seen by all cpus before we test i_state | |
1073 | * -- mikulas | |
1074 | */ | |
1075 | smp_mb(); | |
1076 | ||
1077 | /* avoid the locking if we can */ | |
1078 | if ((inode->i_state & flags) == flags) | |
1079 | return; | |
1080 | ||
1081 | if (unlikely(block_dump)) | |
1082 | block_dump___mark_inode_dirty(inode); | |
1083 | ||
250df6ed | 1084 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1085 | if ((inode->i_state & flags) != flags) { |
1086 | const int was_dirty = inode->i_state & I_DIRTY; | |
1087 | ||
1088 | inode->i_state |= flags; | |
1089 | ||
1090 | /* | |
1091 | * If the inode is being synced, just update its dirty state. | |
1092 | * The unlocker will place the inode on the appropriate | |
1093 | * superblock list, based upon its state. | |
1094 | */ | |
1095 | if (inode->i_state & I_SYNC) | |
250df6ed | 1096 | goto out_unlock_inode; |
03ba3782 JA |
1097 | |
1098 | /* | |
1099 | * Only add valid (hashed) inodes to the superblock's | |
1100 | * dirty list. Add blockdev inodes as well. | |
1101 | */ | |
1102 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1103 | if (inode_unhashed(inode)) |
250df6ed | 1104 | goto out_unlock_inode; |
03ba3782 | 1105 | } |
a4ffdde6 | 1106 | if (inode->i_state & I_FREEING) |
250df6ed | 1107 | goto out_unlock_inode; |
03ba3782 JA |
1108 | |
1109 | /* | |
1110 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1111 | * reposition it (that would break b_dirty time-ordering). | |
1112 | */ | |
1113 | if (!was_dirty) { | |
a66979ab | 1114 | bool wakeup_bdi = false; |
253c34e9 AB |
1115 | bdi = inode_to_bdi(inode); |
1116 | ||
1117 | if (bdi_cap_writeback_dirty(bdi)) { | |
1118 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1119 | "bdi-%s not registered\n", bdi->name); | |
1120 | ||
1121 | /* | |
1122 | * If this is the first dirty inode for this | |
1123 | * bdi, we have to wake-up the corresponding | |
1124 | * bdi thread to make sure background | |
1125 | * write-back happens later. | |
1126 | */ | |
1127 | if (!wb_has_dirty_io(&bdi->wb)) | |
1128 | wakeup_bdi = true; | |
500b067c | 1129 | } |
03ba3782 | 1130 | |
a66979ab | 1131 | spin_unlock(&inode->i_lock); |
f758eeab | 1132 | spin_lock(&bdi->wb.list_lock); |
03ba3782 | 1133 | inode->dirtied_when = jiffies; |
7ccf19a8 | 1134 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1135 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1136 | |
1137 | if (wakeup_bdi) | |
1138 | bdi_wakeup_thread_delayed(bdi); | |
1139 | return; | |
1da177e4 | 1140 | } |
1da177e4 | 1141 | } |
250df6ed DC |
1142 | out_unlock_inode: |
1143 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1144 | |
03ba3782 JA |
1145 | } |
1146 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1147 | ||
1148 | /* | |
1149 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1150 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1151 | * | |
1152 | * If older_than_this is non-NULL, then only write out inodes which | |
1153 | * had their first dirtying at a time earlier than *older_than_this. | |
1154 | * | |
03ba3782 JA |
1155 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1156 | * This function assumes that the blockdev superblock's inodes are backed by | |
1157 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1158 | * assume that all inodes are backed by the same queue. | |
1159 | * | |
1160 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1161 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1162 | * on the writer throttling path, and we get decent balancing between many | |
1163 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1164 | */ | |
b6e51316 | 1165 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1166 | { |
1167 | struct inode *inode, *old_inode = NULL; | |
1168 | ||
1169 | /* | |
1170 | * We need to be protected against the filesystem going from | |
1171 | * r/o to r/w or vice versa. | |
1172 | */ | |
b6e51316 | 1173 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1174 | |
55fa6091 | 1175 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1176 | |
1177 | /* | |
1178 | * Data integrity sync. Must wait for all pages under writeback, | |
1179 | * because there may have been pages dirtied before our sync | |
1180 | * call, but which had writeout started before we write it out. | |
1181 | * In which case, the inode may not be on the dirty list, but | |
1182 | * we still have to wait for that writeout. | |
1183 | */ | |
b6e51316 | 1184 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1185 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1186 | |
250df6ed DC |
1187 | spin_lock(&inode->i_lock); |
1188 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1189 | (mapping->nrpages == 0)) { | |
1190 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1191 | continue; |
250df6ed | 1192 | } |
03ba3782 | 1193 | __iget(inode); |
250df6ed | 1194 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1195 | spin_unlock(&inode_sb_list_lock); |
1196 | ||
03ba3782 | 1197 | /* |
55fa6091 DC |
1198 | * We hold a reference to 'inode' so it couldn't have been |
1199 | * removed from s_inodes list while we dropped the | |
1200 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1201 | * be holding the last reference and we cannot iput it under | |
1202 | * inode_sb_list_lock. So we keep the reference and iput it | |
1203 | * later. | |
03ba3782 JA |
1204 | */ |
1205 | iput(old_inode); | |
1206 | old_inode = inode; | |
1207 | ||
1208 | filemap_fdatawait(mapping); | |
1209 | ||
1210 | cond_resched(); | |
1211 | ||
55fa6091 | 1212 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1213 | } |
55fa6091 | 1214 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1215 | iput(old_inode); |
1da177e4 LT |
1216 | } |
1217 | ||
d8a8559c | 1218 | /** |
3259f8be | 1219 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1220 | * @sb: the superblock |
3259f8be | 1221 | * @nr: the number of pages to write |
1da177e4 | 1222 | * |
d8a8559c JA |
1223 | * Start writeback on some inodes on this super_block. No guarantees are made |
1224 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1225 | * for IO completion of submitted IO. |
1da177e4 | 1226 | */ |
3259f8be | 1227 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
1da177e4 | 1228 | { |
83ba7b07 CH |
1229 | DECLARE_COMPLETION_ONSTACK(done); |
1230 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1231 | .sb = sb, |
1232 | .sync_mode = WB_SYNC_NONE, | |
1233 | .tagged_writepages = 1, | |
1234 | .done = &done, | |
1235 | .nr_pages = nr, | |
3c4d7165 | 1236 | }; |
d8a8559c | 1237 | |
cf37e972 | 1238 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1239 | bdi_queue_work(sb->s_bdi, &work); |
1240 | wait_for_completion(&done); | |
e913fc82 | 1241 | } |
3259f8be CM |
1242 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1243 | ||
1244 | /** | |
1245 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1246 | * @sb: the superblock | |
1247 | * | |
1248 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1249 | * on how many (if any) will be written, and this function does not wait | |
1250 | * for IO completion of submitted IO. | |
1251 | */ | |
1252 | void writeback_inodes_sb(struct super_block *sb) | |
1253 | { | |
925d169f | 1254 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
3259f8be | 1255 | } |
0e3c9a22 | 1256 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1257 | |
17bd55d0 ES |
1258 | /** |
1259 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1260 | * @sb: the superblock | |
1261 | * | |
1262 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1263 | * Returns 1 if writeback was started, 0 if not. | |
1264 | */ | |
1265 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1266 | { | |
1267 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1268 | down_read(&sb->s_umount); |
17bd55d0 | 1269 | writeback_inodes_sb(sb); |
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 | |
1281 | * | |
1282 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1283 | * Returns 1 if writeback was started, 0 if not. | |
1284 | */ | |
1285 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
1286 | unsigned long nr) | |
1287 | { | |
1288 | if (!writeback_in_progress(sb->s_bdi)) { | |
1289 | down_read(&sb->s_umount); | |
1290 | writeback_inodes_sb_nr(sb, nr); | |
1291 | up_read(&sb->s_umount); | |
1292 | return 1; | |
1293 | } else | |
1294 | return 0; | |
1295 | } | |
1296 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1297 | ||
d8a8559c JA |
1298 | /** |
1299 | * sync_inodes_sb - sync sb inode pages | |
1300 | * @sb: the superblock | |
1301 | * | |
1302 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1303 | * super_block. |
d8a8559c | 1304 | */ |
b6e51316 | 1305 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1306 | { |
83ba7b07 CH |
1307 | DECLARE_COMPLETION_ONSTACK(done); |
1308 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1309 | .sb = sb, |
1310 | .sync_mode = WB_SYNC_ALL, | |
1311 | .nr_pages = LONG_MAX, | |
1312 | .range_cyclic = 0, | |
83ba7b07 | 1313 | .done = &done, |
3c4d7165 CH |
1314 | }; |
1315 | ||
cf37e972 CH |
1316 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1317 | ||
83ba7b07 CH |
1318 | bdi_queue_work(sb->s_bdi, &work); |
1319 | wait_for_completion(&done); | |
1320 | ||
b6e51316 | 1321 | wait_sb_inodes(sb); |
1da177e4 | 1322 | } |
d8a8559c | 1323 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1324 | |
1da177e4 | 1325 | /** |
7f04c26d AA |
1326 | * write_inode_now - write an inode to disk |
1327 | * @inode: inode to write to disk | |
1328 | * @sync: whether the write should be synchronous or not | |
1329 | * | |
1330 | * This function commits an inode to disk immediately if it is dirty. This is | |
1331 | * primarily needed by knfsd. | |
1da177e4 | 1332 | * |
7f04c26d | 1333 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1334 | */ |
1da177e4 LT |
1335 | int write_inode_now(struct inode *inode, int sync) |
1336 | { | |
f758eeab | 1337 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1338 | int ret; |
1339 | struct writeback_control wbc = { | |
1340 | .nr_to_write = LONG_MAX, | |
18914b18 | 1341 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1342 | .range_start = 0, |
1343 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1344 | }; |
1345 | ||
1346 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1347 | wbc.nr_to_write = 0; |
1da177e4 LT |
1348 | |
1349 | might_sleep(); | |
f758eeab | 1350 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1351 | spin_lock(&inode->i_lock); |
f758eeab | 1352 | ret = writeback_single_inode(inode, wb, &wbc); |
0f1b1fd8 | 1353 | spin_unlock(&inode->i_lock); |
f758eeab | 1354 | spin_unlock(&wb->list_lock); |
1da177e4 | 1355 | if (sync) |
1c0eeaf5 | 1356 | inode_sync_wait(inode); |
1da177e4 LT |
1357 | return ret; |
1358 | } | |
1359 | EXPORT_SYMBOL(write_inode_now); | |
1360 | ||
1361 | /** | |
1362 | * sync_inode - write an inode and its pages to disk. | |
1363 | * @inode: the inode to sync | |
1364 | * @wbc: controls the writeback mode | |
1365 | * | |
1366 | * sync_inode() will write an inode and its pages to disk. It will also | |
1367 | * correctly update the inode on its superblock's dirty inode lists and will | |
1368 | * update inode->i_state. | |
1369 | * | |
1370 | * The caller must have a ref on the inode. | |
1371 | */ | |
1372 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1373 | { | |
f758eeab | 1374 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1375 | int ret; |
1376 | ||
f758eeab | 1377 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1378 | spin_lock(&inode->i_lock); |
f758eeab | 1379 | ret = writeback_single_inode(inode, wb, wbc); |
0f1b1fd8 | 1380 | spin_unlock(&inode->i_lock); |
f758eeab | 1381 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
1382 | return ret; |
1383 | } | |
1384 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1385 | |
1386 | /** | |
c691b9d9 | 1387 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1388 | * @inode: the inode to sync |
1389 | * @wait: wait for I/O to complete. | |
1390 | * | |
c691b9d9 | 1391 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1392 | * |
1393 | * Note: only writes the actual inode, no associated data or other metadata. | |
1394 | */ | |
1395 | int sync_inode_metadata(struct inode *inode, int wait) | |
1396 | { | |
1397 | struct writeback_control wbc = { | |
1398 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1399 | .nr_to_write = 0, /* metadata-only */ | |
1400 | }; | |
1401 | ||
1402 | return sync_inode(inode, &wbc); | |
1403 | } | |
1404 | EXPORT_SYMBOL(sync_inode_metadata); |