]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/fs-writeback.c | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
6 | * Contains all the functions related to writing back and waiting | |
7 | * upon dirty inodes against superblocks, and writing back dirty | |
8 | * pages against inodes. ie: data writeback. Writeout of the | |
9 | * inode itself is not handled here. | |
10 | * | |
e1f8e874 | 11 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
12 | * Split out of fs/inode.c |
13 | * Additions for address_space-based writeback | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
f5ff8422 | 17 | #include <linux/module.h> |
1da177e4 | 18 | #include <linux/spinlock.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
1da177e4 LT |
20 | #include <linux/sched.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/mm.h> | |
03ba3782 JA |
23 | #include <linux/kthread.h> |
24 | #include <linux/freezer.h> | |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/buffer_head.h> | |
455b2864 | 29 | #include <linux/tracepoint.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 | 31 | |
c4a77a6c JA |
32 | /* |
33 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
34 | */ | |
83ba7b07 | 35 | struct wb_writeback_work { |
c4a77a6c JA |
36 | long nr_pages; |
37 | struct super_block *sb; | |
38 | enum writeback_sync_modes sync_mode; | |
6e6938b6 | 39 | unsigned int tagged_writepages:1; |
52957fe1 HS |
40 | unsigned int for_kupdate:1; |
41 | unsigned int range_cyclic:1; | |
42 | unsigned int for_background:1; | |
c4a77a6c | 43 | |
8010c3b6 | 44 | struct list_head list; /* pending work list */ |
83ba7b07 | 45 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
46 | }; |
47 | ||
455b2864 DC |
48 | /* |
49 | * Include the creation of the trace points after defining the | |
50 | * wb_writeback_work structure so that the definition remains local to this | |
51 | * file. | |
52 | */ | |
53 | #define CREATE_TRACE_POINTS | |
54 | #include <trace/events/writeback.h> | |
55 | ||
455b2864 DC |
56 | /* |
57 | * We don't actually have pdflush, but this one is exported though /proc... | |
58 | */ | |
59 | int nr_pdflush_threads; | |
60 | ||
f11b00f3 AB |
61 | /** |
62 | * writeback_in_progress - determine whether there is writeback in progress | |
63 | * @bdi: the device's backing_dev_info structure. | |
64 | * | |
03ba3782 JA |
65 | * Determine whether there is writeback waiting to be handled against a |
66 | * backing device. | |
f11b00f3 AB |
67 | */ |
68 | int writeback_in_progress(struct backing_dev_info *bdi) | |
69 | { | |
81d73a32 | 70 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 AB |
71 | } |
72 | ||
692ebd17 JK |
73 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
74 | { | |
75 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 76 | |
aaead25b CH |
77 | if (strcmp(sb->s_type->name, "bdev") == 0) |
78 | return inode->i_mapping->backing_dev_info; | |
79 | ||
80 | return sb->s_bdi; | |
692ebd17 JK |
81 | } |
82 | ||
7ccf19a8 NP |
83 | static inline struct inode *wb_inode(struct list_head *head) |
84 | { | |
85 | return list_entry(head, struct inode, i_wb_list); | |
86 | } | |
87 | ||
6585027a JK |
88 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
89 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) | |
03ba3782 | 90 | { |
fff5b85a AB |
91 | if (bdi->wb.task) { |
92 | wake_up_process(bdi->wb.task); | |
93 | } else { | |
94 | /* | |
95 | * The bdi thread isn't there, wake up the forker thread which | |
96 | * will create and run it. | |
97 | */ | |
03ba3782 | 98 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 99 | } |
6585027a JK |
100 | } |
101 | ||
102 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
103 | struct wb_writeback_work *work) | |
104 | { | |
105 | trace_writeback_queue(bdi, work); | |
106 | ||
107 | spin_lock_bh(&bdi->wb_lock); | |
108 | list_add_tail(&work->list, &bdi->work_list); | |
109 | if (!bdi->wb.task) | |
110 | trace_writeback_nothread(bdi, work); | |
111 | bdi_wakeup_flusher(bdi); | |
6467716a | 112 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
113 | } |
114 | ||
83ba7b07 CH |
115 | static void |
116 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
6585027a | 117 | bool range_cyclic) |
1da177e4 | 118 | { |
83ba7b07 | 119 | struct wb_writeback_work *work; |
03ba3782 | 120 | |
bcddc3f0 JA |
121 | /* |
122 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
123 | * wakeup the thread for old dirty data writeback | |
124 | */ | |
83ba7b07 CH |
125 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
126 | if (!work) { | |
455b2864 DC |
127 | if (bdi->wb.task) { |
128 | trace_writeback_nowork(bdi); | |
83ba7b07 | 129 | wake_up_process(bdi->wb.task); |
455b2864 | 130 | } |
83ba7b07 | 131 | return; |
bcddc3f0 | 132 | } |
03ba3782 | 133 | |
83ba7b07 CH |
134 | work->sync_mode = WB_SYNC_NONE; |
135 | work->nr_pages = nr_pages; | |
136 | work->range_cyclic = range_cyclic; | |
03ba3782 | 137 | |
83ba7b07 | 138 | bdi_queue_work(bdi, work); |
b6e51316 JA |
139 | } |
140 | ||
141 | /** | |
142 | * bdi_start_writeback - start writeback | |
143 | * @bdi: the backing device to write from | |
144 | * @nr_pages: the number of pages to write | |
145 | * | |
146 | * Description: | |
147 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 148 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 149 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
150 | * |
151 | */ | |
c5444198 | 152 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 153 | { |
6585027a | 154 | __bdi_start_writeback(bdi, nr_pages, true); |
c5444198 | 155 | } |
d3ddec76 | 156 | |
c5444198 CH |
157 | /** |
158 | * bdi_start_background_writeback - start background writeback | |
159 | * @bdi: the backing device to write from | |
160 | * | |
161 | * Description: | |
6585027a JK |
162 | * This makes sure WB_SYNC_NONE background writeback happens. When |
163 | * this function returns, it is only guaranteed that for given BDI | |
164 | * some IO is happening if we are over background dirty threshold. | |
165 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
166 | */ |
167 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
168 | { | |
6585027a JK |
169 | /* |
170 | * We just wake up the flusher thread. It will perform background | |
171 | * writeback as soon as there is no other work to do. | |
172 | */ | |
71927e84 | 173 | trace_writeback_wake_background(bdi); |
6585027a JK |
174 | spin_lock_bh(&bdi->wb_lock); |
175 | bdi_wakeup_flusher(bdi); | |
176 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
177 | } |
178 | ||
a66979ab DC |
179 | /* |
180 | * Remove the inode from the writeback list it is on. | |
181 | */ | |
182 | void inode_wb_list_del(struct inode *inode) | |
183 | { | |
f758eeab CH |
184 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
185 | ||
186 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 187 | list_del_init(&inode->i_wb_list); |
f758eeab | 188 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
189 | } |
190 | ||
6610a0bc AM |
191 | /* |
192 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
193 | * furthest end of its superblock's dirty-inode list. | |
194 | * | |
195 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 196 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
197 | * the case then the inode must have been redirtied while it was being written |
198 | * out and we don't reset its dirtied_when. | |
199 | */ | |
f758eeab | 200 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 201 | { |
f758eeab | 202 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 203 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 204 | struct inode *tail; |
6610a0bc | 205 | |
7ccf19a8 | 206 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 207 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
208 | inode->dirtied_when = jiffies; |
209 | } | |
7ccf19a8 | 210 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
211 | } |
212 | ||
c986d1e2 | 213 | /* |
66f3b8e2 | 214 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 215 | */ |
f758eeab | 216 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 217 | { |
f758eeab | 218 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 219 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
220 | } |
221 | ||
1c0eeaf5 JE |
222 | static void inode_sync_complete(struct inode *inode) |
223 | { | |
224 | /* | |
a66979ab | 225 | * Prevent speculative execution through |
f758eeab | 226 | * spin_unlock(&wb->list_lock); |
1c0eeaf5 | 227 | */ |
a66979ab | 228 | |
1c0eeaf5 JE |
229 | smp_mb(); |
230 | wake_up_bit(&inode->i_state, __I_SYNC); | |
231 | } | |
232 | ||
d2caa3c5 JL |
233 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
234 | { | |
235 | bool ret = time_after(inode->dirtied_when, t); | |
236 | #ifndef CONFIG_64BIT | |
237 | /* | |
238 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
239 | * It _appears_ to be in the future, but is actually in distant past. | |
240 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 241 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
242 | */ |
243 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
244 | #endif | |
245 | return ret; | |
246 | } | |
247 | ||
2c136579 FW |
248 | /* |
249 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
250 | */ | |
251 | static void move_expired_inodes(struct list_head *delaying_queue, | |
252 | struct list_head *dispatch_queue, | |
253 | unsigned long *older_than_this) | |
254 | { | |
5c03449d SL |
255 | LIST_HEAD(tmp); |
256 | struct list_head *pos, *node; | |
cf137307 | 257 | struct super_block *sb = NULL; |
5c03449d | 258 | struct inode *inode; |
cf137307 | 259 | int do_sb_sort = 0; |
5c03449d | 260 | |
2c136579 | 261 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 262 | inode = wb_inode(delaying_queue->prev); |
2c136579 | 263 | if (older_than_this && |
d2caa3c5 | 264 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 265 | break; |
cf137307 JA |
266 | if (sb && sb != inode->i_sb) |
267 | do_sb_sort = 1; | |
268 | sb = inode->i_sb; | |
7ccf19a8 | 269 | list_move(&inode->i_wb_list, &tmp); |
5c03449d SL |
270 | } |
271 | ||
cf137307 JA |
272 | /* just one sb in list, splice to dispatch_queue and we're done */ |
273 | if (!do_sb_sort) { | |
274 | list_splice(&tmp, dispatch_queue); | |
275 | return; | |
276 | } | |
277 | ||
5c03449d SL |
278 | /* Move inodes from one superblock together */ |
279 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 280 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 281 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 282 | inode = wb_inode(pos); |
5c03449d | 283 | if (inode->i_sb == sb) |
7ccf19a8 | 284 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 285 | } |
2c136579 FW |
286 | } |
287 | } | |
288 | ||
289 | /* | |
290 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
291 | * Before |
292 | * newly dirtied b_dirty b_io b_more_io | |
293 | * =============> gf edc BA | |
294 | * After | |
295 | * newly dirtied b_dirty b_io b_more_io | |
296 | * =============> g fBAedc | |
297 | * | | |
298 | * +--> dequeue for IO | |
2c136579 | 299 | */ |
03ba3782 | 300 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 301 | { |
f758eeab | 302 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 303 | list_splice_init(&wb->b_more_io, &wb->b_io); |
03ba3782 | 304 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
66f3b8e2 JA |
305 | } |
306 | ||
a9185b41 | 307 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 308 | { |
03ba3782 | 309 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 310 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 311 | return 0; |
08d8e974 | 312 | } |
08d8e974 | 313 | |
1da177e4 | 314 | /* |
01c03194 CH |
315 | * Wait for writeback on an inode to complete. |
316 | */ | |
f758eeab CH |
317 | static void inode_wait_for_writeback(struct inode *inode, |
318 | struct bdi_writeback *wb) | |
01c03194 CH |
319 | { |
320 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
321 | wait_queue_head_t *wqh; | |
322 | ||
323 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
324 | while (inode->i_state & I_SYNC) { |
325 | spin_unlock(&inode->i_lock); | |
f758eeab | 326 | spin_unlock(&wb->list_lock); |
01c03194 | 327 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); |
f758eeab | 328 | spin_lock(&wb->list_lock); |
250df6ed | 329 | spin_lock(&inode->i_lock); |
58a9d3d8 | 330 | } |
01c03194 CH |
331 | } |
332 | ||
333 | /* | |
f758eeab | 334 | * Write out an inode's dirty pages. Called under wb->list_lock and |
0f1b1fd8 DC |
335 | * inode->i_lock. Either the caller has an active reference on the inode or |
336 | * the inode has I_WILL_FREE set. | |
01c03194 | 337 | * |
1da177e4 LT |
338 | * If `wait' is set, wait on the writeout. |
339 | * | |
340 | * The whole writeout design is quite complex and fragile. We want to avoid | |
341 | * starvation of particular inodes when others are being redirtied, prevent | |
342 | * livelocks, etc. | |
1da177e4 LT |
343 | */ |
344 | static int | |
f758eeab CH |
345 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, |
346 | struct writeback_control *wbc) | |
1da177e4 | 347 | { |
1da177e4 | 348 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 349 | unsigned dirty; |
1da177e4 LT |
350 | int ret; |
351 | ||
f758eeab | 352 | assert_spin_locked(&wb->list_lock); |
0f1b1fd8 DC |
353 | assert_spin_locked(&inode->i_lock); |
354 | ||
01c03194 CH |
355 | if (!atomic_read(&inode->i_count)) |
356 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
357 | else | |
358 | WARN_ON(inode->i_state & I_WILL_FREE); | |
359 | ||
360 | if (inode->i_state & I_SYNC) { | |
361 | /* | |
362 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 363 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
364 | * writeback can proceed with the other inodes on s_io. |
365 | * | |
366 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 367 | * completed a full scan of b_io. |
01c03194 | 368 | */ |
a9185b41 | 369 | if (wbc->sync_mode != WB_SYNC_ALL) { |
f758eeab | 370 | requeue_io(inode, wb); |
01c03194 CH |
371 | return 0; |
372 | } | |
373 | ||
374 | /* | |
375 | * It's a data-integrity sync. We must wait. | |
376 | */ | |
f758eeab | 377 | inode_wait_for_writeback(inode, wb); |
01c03194 CH |
378 | } |
379 | ||
1c0eeaf5 | 380 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 381 | |
5547e8aa | 382 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 383 | inode->i_state |= I_SYNC; |
5547e8aa | 384 | inode->i_state &= ~I_DIRTY_PAGES; |
250df6ed | 385 | spin_unlock(&inode->i_lock); |
f758eeab | 386 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
387 | |
388 | ret = do_writepages(mapping, wbc); | |
389 | ||
26821ed4 CH |
390 | /* |
391 | * Make sure to wait on the data before writing out the metadata. | |
392 | * This is important for filesystems that modify metadata on data | |
393 | * I/O completion. | |
394 | */ | |
a9185b41 | 395 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 396 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
397 | if (ret == 0) |
398 | ret = err; | |
399 | } | |
400 | ||
5547e8aa DM |
401 | /* |
402 | * Some filesystems may redirty the inode during the writeback | |
403 | * due to delalloc, clear dirty metadata flags right before | |
404 | * write_inode() | |
405 | */ | |
250df6ed | 406 | spin_lock(&inode->i_lock); |
5547e8aa DM |
407 | dirty = inode->i_state & I_DIRTY; |
408 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 409 | spin_unlock(&inode->i_lock); |
26821ed4 CH |
410 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
411 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 412 | int err = write_inode(inode, wbc); |
1da177e4 LT |
413 | if (ret == 0) |
414 | ret = err; | |
415 | } | |
416 | ||
f758eeab | 417 | spin_lock(&wb->list_lock); |
250df6ed | 418 | spin_lock(&inode->i_lock); |
1c0eeaf5 | 419 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 420 | if (!(inode->i_state & I_FREEING)) { |
94c3dcbb WF |
421 | /* |
422 | * Sync livelock prevention. Each inode is tagged and synced in | |
423 | * one shot. If still dirty, it will be redirty_tail()'ed below. | |
424 | * Update the dirty time to prevent enqueue and sync it again. | |
425 | */ | |
426 | if ((inode->i_state & I_DIRTY) && | |
427 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
428 | inode->dirtied_when = jiffies; | |
429 | ||
23539afc | 430 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
431 | /* |
432 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 433 | * sometimes bales out without doing anything. |
1b43ef91 | 434 | */ |
a50aeb40 WF |
435 | inode->i_state |= I_DIRTY_PAGES; |
436 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 437 | /* |
a50aeb40 | 438 | * slice used up: queue for next turn |
1da177e4 | 439 | */ |
f758eeab | 440 | requeue_io(inode, wb); |
1da177e4 LT |
441 | } else { |
442 | /* | |
a50aeb40 WF |
443 | * Writeback blocked by something other than |
444 | * congestion. Delay the inode for some time to | |
445 | * avoid spinning on the CPU (100% iowait) | |
446 | * retrying writeback of the dirty page/inode | |
447 | * that cannot be performed immediately. | |
1da177e4 | 448 | */ |
f758eeab | 449 | redirty_tail(inode, wb); |
1da177e4 | 450 | } |
23539afc WF |
451 | } else if (inode->i_state & I_DIRTY) { |
452 | /* | |
453 | * Filesystems can dirty the inode during writeback | |
454 | * operations, such as delayed allocation during | |
455 | * submission or metadata updates after data IO | |
456 | * completion. | |
457 | */ | |
f758eeab | 458 | redirty_tail(inode, wb); |
1da177e4 LT |
459 | } else { |
460 | /* | |
9e38d86f NP |
461 | * The inode is clean. At this point we either have |
462 | * a reference to the inode or it's on it's way out. | |
463 | * No need to add it back to the LRU. | |
1da177e4 | 464 | */ |
7ccf19a8 | 465 | list_del_init(&inode->i_wb_list); |
cb9bd115 | 466 | wbc->inodes_written++; |
1da177e4 LT |
467 | } |
468 | } | |
1c0eeaf5 | 469 | inode_sync_complete(inode); |
1da177e4 LT |
470 | return ret; |
471 | } | |
472 | ||
03ba3782 | 473 | /* |
d19de7ed | 474 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
475 | * before calling writeback. So make sure that we do pin it, so it doesn't |
476 | * go away while we are writing inodes from it. | |
03ba3782 | 477 | */ |
d19de7ed | 478 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 479 | { |
03ba3782 | 480 | spin_lock(&sb_lock); |
29cb4859 CH |
481 | if (list_empty(&sb->s_instances)) { |
482 | spin_unlock(&sb_lock); | |
483 | return false; | |
484 | } | |
485 | ||
03ba3782 | 486 | sb->s_count++; |
29cb4859 CH |
487 | spin_unlock(&sb_lock); |
488 | ||
03ba3782 | 489 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 490 | if (sb->s_root) |
d19de7ed | 491 | return true; |
03ba3782 JA |
492 | up_read(&sb->s_umount); |
493 | } | |
29cb4859 CH |
494 | |
495 | put_super(sb); | |
d19de7ed | 496 | return false; |
03ba3782 JA |
497 | } |
498 | ||
f11c9c5c ES |
499 | /* |
500 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
501 | * |
502 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
503 | * inodes. Otherwise write only ones which go sequentially |
504 | * in reverse order. | |
edadfb10 | 505 | * |
f11c9c5c ES |
506 | * Return 1, if the caller writeback routine should be |
507 | * interrupted. Otherwise return 0. | |
508 | */ | |
edadfb10 CH |
509 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
510 | struct writeback_control *wbc, bool only_this_sb) | |
1da177e4 | 511 | { |
03ba3782 | 512 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 513 | long pages_skipped; |
7ccf19a8 | 514 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
515 | |
516 | if (inode->i_sb != sb) { | |
517 | if (only_this_sb) { | |
518 | /* | |
519 | * We only want to write back data for this | |
520 | * superblock, move all inodes not belonging | |
521 | * to it back onto the dirty list. | |
522 | */ | |
f758eeab | 523 | redirty_tail(inode, wb); |
edadfb10 CH |
524 | continue; |
525 | } | |
526 | ||
527 | /* | |
528 | * The inode belongs to a different superblock. | |
529 | * Bounce back to the caller to unpin this and | |
530 | * pin the next superblock. | |
531 | */ | |
f11c9c5c | 532 | return 0; |
edadfb10 CH |
533 | } |
534 | ||
9843b76a CH |
535 | /* |
536 | * Don't bother with new inodes or inodes beeing freed, first | |
537 | * kind does not need peridic writeout yet, and for the latter | |
538 | * kind writeout is handled by the freer. | |
539 | */ | |
250df6ed | 540 | spin_lock(&inode->i_lock); |
9843b76a | 541 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 542 | spin_unlock(&inode->i_lock); |
f758eeab | 543 | requeue_io(inode, wb); |
7ef0d737 NP |
544 | continue; |
545 | } | |
9843b76a | 546 | |
1da177e4 | 547 | __iget(inode); |
250df6ed | 548 | |
1da177e4 | 549 | pages_skipped = wbc->pages_skipped; |
f758eeab | 550 | writeback_single_inode(inode, wb, wbc); |
1da177e4 LT |
551 | if (wbc->pages_skipped != pages_skipped) { |
552 | /* | |
553 | * writeback is not making progress due to locked | |
554 | * buffers. Skip this inode for now. | |
555 | */ | |
f758eeab | 556 | redirty_tail(inode, wb); |
1da177e4 | 557 | } |
0f1b1fd8 | 558 | spin_unlock(&inode->i_lock); |
f758eeab | 559 | spin_unlock(&wb->list_lock); |
1da177e4 | 560 | iput(inode); |
4ffc8444 | 561 | cond_resched(); |
f758eeab | 562 | spin_lock(&wb->list_lock); |
b7a2441f | 563 | if (wbc->nr_to_write <= 0) |
f11c9c5c | 564 | return 1; |
1da177e4 | 565 | } |
f11c9c5c ES |
566 | /* b_io is empty */ |
567 | return 1; | |
568 | } | |
569 | ||
e8dfc305 WF |
570 | static void __writeback_inodes_wb(struct bdi_writeback *wb, |
571 | struct writeback_control *wbc) | |
f11c9c5c ES |
572 | { |
573 | int ret = 0; | |
574 | ||
f11c9c5c | 575 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 576 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 577 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 578 | |
edadfb10 | 579 | if (!pin_sb_for_writeback(sb)) { |
f758eeab | 580 | requeue_io(inode, wb); |
edadfb10 | 581 | continue; |
f11c9c5c | 582 | } |
edadfb10 CH |
583 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
584 | drop_super(sb); | |
f11c9c5c | 585 | |
f11c9c5c ES |
586 | if (ret) |
587 | break; | |
588 | } | |
66f3b8e2 JA |
589 | /* Leave any unwritten inodes on b_io */ |
590 | } | |
591 | ||
e8dfc305 WF |
592 | void writeback_inodes_wb(struct bdi_writeback *wb, |
593 | struct writeback_control *wbc) | |
edadfb10 | 594 | { |
f758eeab | 595 | spin_lock(&wb->list_lock); |
424b351f | 596 | if (list_empty(&wb->b_io)) |
edadfb10 | 597 | queue_io(wb, wbc->older_than_this); |
e8dfc305 | 598 | __writeback_inodes_wb(wb, wbc); |
f758eeab | 599 | spin_unlock(&wb->list_lock); |
edadfb10 CH |
600 | } |
601 | ||
66f3b8e2 | 602 | /* |
03ba3782 JA |
603 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
604 | * operation. We do this so we don't hold I_SYNC against an inode for | |
605 | * enormous amounts of time, which would block a userspace task which has | |
606 | * been forced to throttle against that inode. Also, the code reevaluates | |
607 | * the dirty each time it has written this many pages. | |
608 | */ | |
609 | #define MAX_WRITEBACK_PAGES 1024 | |
610 | ||
611 | static inline bool over_bground_thresh(void) | |
612 | { | |
613 | unsigned long background_thresh, dirty_thresh; | |
614 | ||
16c4042f | 615 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 JA |
616 | |
617 | return (global_page_state(NR_FILE_DIRTY) + | |
4cbec4c8 | 618 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
03ba3782 JA |
619 | } |
620 | ||
621 | /* | |
622 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 623 | * |
03ba3782 JA |
624 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
625 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
626 | * just walks the superblock inode list, writing back any inodes which are | |
627 | * older than a specific point in time. | |
66f3b8e2 | 628 | * |
03ba3782 JA |
629 | * Try to run once per dirty_writeback_interval. But if a writeback event |
630 | * takes longer than a dirty_writeback_interval interval, then leave a | |
631 | * one-second gap. | |
66f3b8e2 | 632 | * |
03ba3782 JA |
633 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
634 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 635 | */ |
c4a77a6c | 636 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 637 | struct wb_writeback_work *work) |
66f3b8e2 | 638 | { |
03ba3782 | 639 | struct writeback_control wbc = { |
83ba7b07 | 640 | .sync_mode = work->sync_mode, |
6e6938b6 | 641 | .tagged_writepages = work->tagged_writepages, |
03ba3782 | 642 | .older_than_this = NULL, |
83ba7b07 CH |
643 | .for_kupdate = work->for_kupdate, |
644 | .for_background = work->for_background, | |
645 | .range_cyclic = work->range_cyclic, | |
03ba3782 JA |
646 | }; |
647 | unsigned long oldest_jif; | |
648 | long wrote = 0; | |
6e6938b6 | 649 | long write_chunk = MAX_WRITEBACK_PAGES; |
a5989bdc | 650 | struct inode *inode; |
66f3b8e2 | 651 | |
c4a77a6c JA |
652 | if (!wbc.range_cyclic) { |
653 | wbc.range_start = 0; | |
654 | wbc.range_end = LLONG_MAX; | |
655 | } | |
38f21977 | 656 | |
b9543dac JK |
657 | /* |
658 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
659 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
660 | * here avoids calling into writeback_inodes_wb() more than once. | |
661 | * | |
662 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
663 | * | |
664 | * wb_writeback() | |
e8dfc305 | 665 | * writeback_sb_inodes() <== called only once |
b9543dac JK |
666 | * write_cache_pages() <== called once for each inode |
667 | * (quickly) tag currently dirty pages | |
668 | * (maybe slowly) sync all tagged pages | |
669 | */ | |
6e6938b6 | 670 | if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_writepages) |
b9543dac JK |
671 | write_chunk = LONG_MAX; |
672 | ||
e185dda8 WF |
673 | oldest_jif = jiffies; |
674 | wbc.older_than_this = &oldest_jif; | |
675 | ||
e8dfc305 | 676 | spin_lock(&wb->list_lock); |
03ba3782 JA |
677 | for (;;) { |
678 | /* | |
d3ddec76 | 679 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 680 | */ |
83ba7b07 | 681 | if (work->nr_pages <= 0) |
03ba3782 | 682 | break; |
66f3b8e2 | 683 | |
aa373cf5 JK |
684 | /* |
685 | * Background writeout and kupdate-style writeback may | |
686 | * run forever. Stop them if there is other work to do | |
687 | * so that e.g. sync can proceed. They'll be restarted | |
688 | * after the other works are all done. | |
689 | */ | |
690 | if ((work->for_background || work->for_kupdate) && | |
691 | !list_empty(&wb->bdi->work_list)) | |
692 | break; | |
693 | ||
38f21977 | 694 | /* |
d3ddec76 WF |
695 | * For background writeout, stop when we are below the |
696 | * background dirty threshold | |
38f21977 | 697 | */ |
83ba7b07 | 698 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 699 | break; |
38f21977 | 700 | |
ba9aa839 WF |
701 | if (work->for_kupdate) { |
702 | oldest_jif = jiffies - | |
703 | msecs_to_jiffies(dirty_expire_interval * 10); | |
704 | wbc.older_than_this = &oldest_jif; | |
705 | } | |
706 | ||
b9543dac | 707 | wbc.nr_to_write = write_chunk; |
03ba3782 | 708 | wbc.pages_skipped = 0; |
cb9bd115 | 709 | wbc.inodes_written = 0; |
028c2dd1 DC |
710 | |
711 | trace_wbc_writeback_start(&wbc, wb->bdi); | |
e8dfc305 WF |
712 | if (list_empty(&wb->b_io)) |
713 | queue_io(wb, wbc.older_than_this); | |
83ba7b07 | 714 | if (work->sb) |
e8dfc305 | 715 | writeback_sb_inodes(work->sb, wb, &wbc, true); |
edadfb10 | 716 | else |
e8dfc305 | 717 | __writeback_inodes_wb(wb, &wbc); |
028c2dd1 DC |
718 | trace_wbc_writeback_written(&wbc, wb->bdi); |
719 | ||
b9543dac JK |
720 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
721 | wrote += write_chunk - wbc.nr_to_write; | |
03ba3782 JA |
722 | |
723 | /* | |
e6fb6da2 WF |
724 | * Did we write something? Try for more |
725 | * | |
726 | * Dirty inodes are moved to b_io for writeback in batches. | |
727 | * The completion of the current batch does not necessarily | |
728 | * mean the overall work is done. So we keep looping as long | |
729 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 730 | */ |
e6fb6da2 | 731 | if (wbc.nr_to_write < write_chunk) |
71fd05a8 | 732 | continue; |
cb9bd115 WF |
733 | if (wbc.inodes_written) |
734 | continue; | |
71fd05a8 | 735 | /* |
e6fb6da2 | 736 | * No more inodes for IO, bail |
71fd05a8 | 737 | */ |
b7a2441f | 738 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 739 | break; |
71fd05a8 JA |
740 | /* |
741 | * Nothing written. Wait for some inode to | |
742 | * become available for writeback. Otherwise | |
743 | * we'll just busyloop. | |
744 | */ | |
71fd05a8 | 745 | if (!list_empty(&wb->b_more_io)) { |
7ccf19a8 | 746 | inode = wb_inode(wb->b_more_io.prev); |
028c2dd1 | 747 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
250df6ed | 748 | spin_lock(&inode->i_lock); |
f758eeab | 749 | inode_wait_for_writeback(inode, wb); |
250df6ed | 750 | spin_unlock(&inode->i_lock); |
03ba3782 JA |
751 | } |
752 | } | |
e8dfc305 | 753 | spin_unlock(&wb->list_lock); |
03ba3782 JA |
754 | |
755 | return wrote; | |
756 | } | |
757 | ||
758 | /* | |
83ba7b07 | 759 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 760 | */ |
83ba7b07 | 761 | static struct wb_writeback_work * |
08852b6d | 762 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 763 | { |
83ba7b07 | 764 | struct wb_writeback_work *work = NULL; |
03ba3782 | 765 | |
6467716a | 766 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
767 | if (!list_empty(&bdi->work_list)) { |
768 | work = list_entry(bdi->work_list.next, | |
769 | struct wb_writeback_work, list); | |
770 | list_del_init(&work->list); | |
03ba3782 | 771 | } |
6467716a | 772 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 773 | return work; |
03ba3782 JA |
774 | } |
775 | ||
cdf01dd5 LT |
776 | /* |
777 | * Add in the number of potentially dirty inodes, because each inode | |
778 | * write can dirty pagecache in the underlying blockdev. | |
779 | */ | |
780 | static unsigned long get_nr_dirty_pages(void) | |
781 | { | |
782 | return global_page_state(NR_FILE_DIRTY) + | |
783 | global_page_state(NR_UNSTABLE_NFS) + | |
784 | get_nr_dirty_inodes(); | |
785 | } | |
786 | ||
6585027a JK |
787 | static long wb_check_background_flush(struct bdi_writeback *wb) |
788 | { | |
789 | if (over_bground_thresh()) { | |
790 | ||
791 | struct wb_writeback_work work = { | |
792 | .nr_pages = LONG_MAX, | |
793 | .sync_mode = WB_SYNC_NONE, | |
794 | .for_background = 1, | |
795 | .range_cyclic = 1, | |
796 | }; | |
797 | ||
798 | return wb_writeback(wb, &work); | |
799 | } | |
800 | ||
801 | return 0; | |
802 | } | |
803 | ||
03ba3782 JA |
804 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
805 | { | |
806 | unsigned long expired; | |
807 | long nr_pages; | |
808 | ||
69b62d01 JA |
809 | /* |
810 | * When set to zero, disable periodic writeback | |
811 | */ | |
812 | if (!dirty_writeback_interval) | |
813 | return 0; | |
814 | ||
03ba3782 JA |
815 | expired = wb->last_old_flush + |
816 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
817 | if (time_before(jiffies, expired)) | |
818 | return 0; | |
819 | ||
820 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 821 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 822 | |
c4a77a6c | 823 | if (nr_pages) { |
83ba7b07 | 824 | struct wb_writeback_work work = { |
c4a77a6c JA |
825 | .nr_pages = nr_pages, |
826 | .sync_mode = WB_SYNC_NONE, | |
827 | .for_kupdate = 1, | |
828 | .range_cyclic = 1, | |
829 | }; | |
830 | ||
83ba7b07 | 831 | return wb_writeback(wb, &work); |
c4a77a6c | 832 | } |
03ba3782 JA |
833 | |
834 | return 0; | |
835 | } | |
836 | ||
837 | /* | |
838 | * Retrieve work items and do the writeback they describe | |
839 | */ | |
840 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
841 | { | |
842 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 843 | struct wb_writeback_work *work; |
c4a77a6c | 844 | long wrote = 0; |
03ba3782 | 845 | |
81d73a32 | 846 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 847 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
848 | /* |
849 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 850 | * because this thread is exiting now. |
03ba3782 JA |
851 | */ |
852 | if (force_wait) | |
83ba7b07 | 853 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 854 | |
455b2864 DC |
855 | trace_writeback_exec(bdi, work); |
856 | ||
83ba7b07 | 857 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
858 | |
859 | /* | |
83ba7b07 CH |
860 | * Notify the caller of completion if this is a synchronous |
861 | * work item, otherwise just free it. | |
03ba3782 | 862 | */ |
83ba7b07 CH |
863 | if (work->done) |
864 | complete(work->done); | |
865 | else | |
866 | kfree(work); | |
03ba3782 JA |
867 | } |
868 | ||
869 | /* | |
870 | * Check for periodic writeback, kupdated() style | |
871 | */ | |
872 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 873 | wrote += wb_check_background_flush(wb); |
81d73a32 | 874 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
875 | |
876 | return wrote; | |
877 | } | |
878 | ||
879 | /* | |
880 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
881 | * wakes up periodically and does kupdated style flushing. | |
882 | */ | |
08243900 | 883 | int bdi_writeback_thread(void *data) |
03ba3782 | 884 | { |
08243900 CH |
885 | struct bdi_writeback *wb = data; |
886 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
887 | long pages_written; |
888 | ||
766f9164 | 889 | current->flags |= PF_SWAPWRITE; |
08243900 | 890 | set_freezable(); |
ecd58403 | 891 | wb->last_active = jiffies; |
08243900 CH |
892 | |
893 | /* | |
894 | * Our parent may run at a different priority, just set us to normal | |
895 | */ | |
896 | set_user_nice(current, 0); | |
897 | ||
455b2864 DC |
898 | trace_writeback_thread_start(bdi); |
899 | ||
03ba3782 | 900 | while (!kthread_should_stop()) { |
6467716a AB |
901 | /* |
902 | * Remove own delayed wake-up timer, since we are already awake | |
903 | * and we'll take care of the preriodic write-back. | |
904 | */ | |
905 | del_timer(&wb->wakeup_timer); | |
906 | ||
03ba3782 JA |
907 | pages_written = wb_do_writeback(wb, 0); |
908 | ||
455b2864 DC |
909 | trace_writeback_pages_written(pages_written); |
910 | ||
03ba3782 | 911 | if (pages_written) |
ecd58403 | 912 | wb->last_active = jiffies; |
03ba3782 | 913 | |
297252c8 | 914 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 915 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 916 | __set_current_state(TASK_RUNNING); |
297252c8 | 917 | continue; |
03ba3782 JA |
918 | } |
919 | ||
253c34e9 | 920 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 921 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
922 | else { |
923 | /* | |
924 | * We have nothing to do, so can go sleep without any | |
925 | * timeout and save power. When a work is queued or | |
926 | * something is made dirty - we will be woken up. | |
927 | */ | |
297252c8 | 928 | schedule(); |
f9eadbbd | 929 | } |
69b62d01 | 930 | |
03ba3782 JA |
931 | try_to_freeze(); |
932 | } | |
933 | ||
fff5b85a | 934 | /* Flush any work that raced with us exiting */ |
08243900 CH |
935 | if (!list_empty(&bdi->work_list)) |
936 | wb_do_writeback(wb, 1); | |
455b2864 DC |
937 | |
938 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
939 | return 0; |
940 | } | |
941 | ||
08243900 | 942 | |
03ba3782 | 943 | /* |
b8c2f347 CH |
944 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
945 | * the whole world. | |
03ba3782 | 946 | */ |
b8c2f347 | 947 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 948 | { |
b8c2f347 | 949 | struct backing_dev_info *bdi; |
03ba3782 | 950 | |
83ba7b07 CH |
951 | if (!nr_pages) { |
952 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
953 | global_page_state(NR_UNSTABLE_NFS); |
954 | } | |
03ba3782 | 955 | |
b8c2f347 | 956 | rcu_read_lock(); |
cfc4ba53 | 957 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
958 | if (!bdi_has_dirty_io(bdi)) |
959 | continue; | |
6585027a | 960 | __bdi_start_writeback(bdi, nr_pages, false); |
03ba3782 | 961 | } |
cfc4ba53 | 962 | rcu_read_unlock(); |
1da177e4 LT |
963 | } |
964 | ||
03ba3782 JA |
965 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
966 | { | |
967 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
968 | struct dentry *dentry; | |
969 | const char *name = "?"; | |
970 | ||
971 | dentry = d_find_alias(inode); | |
972 | if (dentry) { | |
973 | spin_lock(&dentry->d_lock); | |
974 | name = (const char *) dentry->d_name.name; | |
975 | } | |
976 | printk(KERN_DEBUG | |
977 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
978 | current->comm, task_pid_nr(current), inode->i_ino, | |
979 | name, inode->i_sb->s_id); | |
980 | if (dentry) { | |
981 | spin_unlock(&dentry->d_lock); | |
982 | dput(dentry); | |
983 | } | |
984 | } | |
985 | } | |
986 | ||
987 | /** | |
988 | * __mark_inode_dirty - internal function | |
989 | * @inode: inode to mark | |
990 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
991 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
992 | * mark_inode_dirty_sync. | |
1da177e4 | 993 | * |
03ba3782 JA |
994 | * Put the inode on the super block's dirty list. |
995 | * | |
996 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
997 | * dirty list only if it is hashed or if it refers to a blockdev. | |
998 | * If it was not hashed, it will never be added to the dirty list | |
999 | * even if it is later hashed, as it will have been marked dirty already. | |
1000 | * | |
1001 | * In short, make sure you hash any inodes _before_ you start marking | |
1002 | * them dirty. | |
1da177e4 | 1003 | * |
03ba3782 JA |
1004 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1005 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1006 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1007 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1008 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1009 | * blockdev inode. | |
1da177e4 | 1010 | */ |
03ba3782 | 1011 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1012 | { |
03ba3782 | 1013 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1014 | struct backing_dev_info *bdi = NULL; |
1da177e4 | 1015 | |
03ba3782 JA |
1016 | /* |
1017 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1018 | * dirty the inode itself | |
1019 | */ | |
1020 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1021 | if (sb->s_op->dirty_inode) | |
aa385729 | 1022 | sb->s_op->dirty_inode(inode, flags); |
03ba3782 JA |
1023 | } |
1024 | ||
1025 | /* | |
1026 | * make sure that changes are seen by all cpus before we test i_state | |
1027 | * -- mikulas | |
1028 | */ | |
1029 | smp_mb(); | |
1030 | ||
1031 | /* avoid the locking if we can */ | |
1032 | if ((inode->i_state & flags) == flags) | |
1033 | return; | |
1034 | ||
1035 | if (unlikely(block_dump)) | |
1036 | block_dump___mark_inode_dirty(inode); | |
1037 | ||
250df6ed | 1038 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1039 | if ((inode->i_state & flags) != flags) { |
1040 | const int was_dirty = inode->i_state & I_DIRTY; | |
1041 | ||
1042 | inode->i_state |= flags; | |
1043 | ||
1044 | /* | |
1045 | * If the inode is being synced, just update its dirty state. | |
1046 | * The unlocker will place the inode on the appropriate | |
1047 | * superblock list, based upon its state. | |
1048 | */ | |
1049 | if (inode->i_state & I_SYNC) | |
250df6ed | 1050 | goto out_unlock_inode; |
03ba3782 JA |
1051 | |
1052 | /* | |
1053 | * Only add valid (hashed) inodes to the superblock's | |
1054 | * dirty list. Add blockdev inodes as well. | |
1055 | */ | |
1056 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1057 | if (inode_unhashed(inode)) |
250df6ed | 1058 | goto out_unlock_inode; |
03ba3782 | 1059 | } |
a4ffdde6 | 1060 | if (inode->i_state & I_FREEING) |
250df6ed | 1061 | goto out_unlock_inode; |
03ba3782 JA |
1062 | |
1063 | /* | |
1064 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1065 | * reposition it (that would break b_dirty time-ordering). | |
1066 | */ | |
1067 | if (!was_dirty) { | |
a66979ab | 1068 | bool wakeup_bdi = false; |
253c34e9 AB |
1069 | bdi = inode_to_bdi(inode); |
1070 | ||
1071 | if (bdi_cap_writeback_dirty(bdi)) { | |
1072 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1073 | "bdi-%s not registered\n", bdi->name); | |
1074 | ||
1075 | /* | |
1076 | * If this is the first dirty inode for this | |
1077 | * bdi, we have to wake-up the corresponding | |
1078 | * bdi thread to make sure background | |
1079 | * write-back happens later. | |
1080 | */ | |
1081 | if (!wb_has_dirty_io(&bdi->wb)) | |
1082 | wakeup_bdi = true; | |
500b067c | 1083 | } |
03ba3782 | 1084 | |
a66979ab | 1085 | spin_unlock(&inode->i_lock); |
f758eeab | 1086 | spin_lock(&bdi->wb.list_lock); |
03ba3782 | 1087 | inode->dirtied_when = jiffies; |
7ccf19a8 | 1088 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
f758eeab | 1089 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
1090 | |
1091 | if (wakeup_bdi) | |
1092 | bdi_wakeup_thread_delayed(bdi); | |
1093 | return; | |
1da177e4 | 1094 | } |
1da177e4 | 1095 | } |
250df6ed DC |
1096 | out_unlock_inode: |
1097 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1098 | |
03ba3782 JA |
1099 | } |
1100 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1101 | ||
1102 | /* | |
1103 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1104 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1105 | * | |
1106 | * If older_than_this is non-NULL, then only write out inodes which | |
1107 | * had their first dirtying at a time earlier than *older_than_this. | |
1108 | * | |
03ba3782 JA |
1109 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1110 | * This function assumes that the blockdev superblock's inodes are backed by | |
1111 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1112 | * assume that all inodes are backed by the same queue. | |
1113 | * | |
1114 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1115 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1116 | * on the writer throttling path, and we get decent balancing between many | |
1117 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1118 | */ | |
b6e51316 | 1119 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1120 | { |
1121 | struct inode *inode, *old_inode = NULL; | |
1122 | ||
1123 | /* | |
1124 | * We need to be protected against the filesystem going from | |
1125 | * r/o to r/w or vice versa. | |
1126 | */ | |
b6e51316 | 1127 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1128 | |
55fa6091 | 1129 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1130 | |
1131 | /* | |
1132 | * Data integrity sync. Must wait for all pages under writeback, | |
1133 | * because there may have been pages dirtied before our sync | |
1134 | * call, but which had writeout started before we write it out. | |
1135 | * In which case, the inode may not be on the dirty list, but | |
1136 | * we still have to wait for that writeout. | |
1137 | */ | |
b6e51316 | 1138 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1139 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1140 | |
250df6ed DC |
1141 | spin_lock(&inode->i_lock); |
1142 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1143 | (mapping->nrpages == 0)) { | |
1144 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1145 | continue; |
250df6ed | 1146 | } |
03ba3782 | 1147 | __iget(inode); |
250df6ed | 1148 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1149 | spin_unlock(&inode_sb_list_lock); |
1150 | ||
03ba3782 | 1151 | /* |
55fa6091 DC |
1152 | * We hold a reference to 'inode' so it couldn't have been |
1153 | * removed from s_inodes list while we dropped the | |
1154 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1155 | * be holding the last reference and we cannot iput it under | |
1156 | * inode_sb_list_lock. So we keep the reference and iput it | |
1157 | * later. | |
03ba3782 JA |
1158 | */ |
1159 | iput(old_inode); | |
1160 | old_inode = inode; | |
1161 | ||
1162 | filemap_fdatawait(mapping); | |
1163 | ||
1164 | cond_resched(); | |
1165 | ||
55fa6091 | 1166 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1167 | } |
55fa6091 | 1168 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1169 | iput(old_inode); |
1da177e4 LT |
1170 | } |
1171 | ||
d8a8559c | 1172 | /** |
3259f8be | 1173 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1174 | * @sb: the superblock |
3259f8be | 1175 | * @nr: the number of pages to write |
1da177e4 | 1176 | * |
d8a8559c JA |
1177 | * Start writeback on some inodes on this super_block. No guarantees are made |
1178 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1179 | * for IO completion of submitted IO. |
1da177e4 | 1180 | */ |
3259f8be | 1181 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
1da177e4 | 1182 | { |
83ba7b07 CH |
1183 | DECLARE_COMPLETION_ONSTACK(done); |
1184 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1185 | .sb = sb, |
1186 | .sync_mode = WB_SYNC_NONE, | |
1187 | .tagged_writepages = 1, | |
1188 | .done = &done, | |
1189 | .nr_pages = nr, | |
3c4d7165 | 1190 | }; |
d8a8559c | 1191 | |
cf37e972 | 1192 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1193 | bdi_queue_work(sb->s_bdi, &work); |
1194 | wait_for_completion(&done); | |
e913fc82 | 1195 | } |
3259f8be CM |
1196 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1197 | ||
1198 | /** | |
1199 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1200 | * @sb: the superblock | |
1201 | * | |
1202 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1203 | * on how many (if any) will be written, and this function does not wait | |
1204 | * for IO completion of submitted IO. | |
1205 | */ | |
1206 | void writeback_inodes_sb(struct super_block *sb) | |
1207 | { | |
925d169f | 1208 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
3259f8be | 1209 | } |
0e3c9a22 | 1210 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1211 | |
17bd55d0 ES |
1212 | /** |
1213 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1214 | * @sb: the superblock | |
1215 | * | |
1216 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1217 | * Returns 1 if writeback was started, 0 if not. | |
1218 | */ | |
1219 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1220 | { | |
1221 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1222 | down_read(&sb->s_umount); |
17bd55d0 | 1223 | writeback_inodes_sb(sb); |
cf37e972 | 1224 | up_read(&sb->s_umount); |
17bd55d0 ES |
1225 | return 1; |
1226 | } else | |
1227 | return 0; | |
1228 | } | |
1229 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1230 | ||
3259f8be CM |
1231 | /** |
1232 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1233 | * @sb: the superblock | |
1234 | * @nr: the number of pages to write | |
1235 | * | |
1236 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1237 | * Returns 1 if writeback was started, 0 if not. | |
1238 | */ | |
1239 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
1240 | unsigned long nr) | |
1241 | { | |
1242 | if (!writeback_in_progress(sb->s_bdi)) { | |
1243 | down_read(&sb->s_umount); | |
1244 | writeback_inodes_sb_nr(sb, nr); | |
1245 | up_read(&sb->s_umount); | |
1246 | return 1; | |
1247 | } else | |
1248 | return 0; | |
1249 | } | |
1250 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1251 | ||
d8a8559c JA |
1252 | /** |
1253 | * sync_inodes_sb - sync sb inode pages | |
1254 | * @sb: the superblock | |
1255 | * | |
1256 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1257 | * super_block. |
d8a8559c | 1258 | */ |
b6e51316 | 1259 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1260 | { |
83ba7b07 CH |
1261 | DECLARE_COMPLETION_ONSTACK(done); |
1262 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1263 | .sb = sb, |
1264 | .sync_mode = WB_SYNC_ALL, | |
1265 | .nr_pages = LONG_MAX, | |
1266 | .range_cyclic = 0, | |
83ba7b07 | 1267 | .done = &done, |
3c4d7165 CH |
1268 | }; |
1269 | ||
cf37e972 CH |
1270 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1271 | ||
83ba7b07 CH |
1272 | bdi_queue_work(sb->s_bdi, &work); |
1273 | wait_for_completion(&done); | |
1274 | ||
b6e51316 | 1275 | wait_sb_inodes(sb); |
1da177e4 | 1276 | } |
d8a8559c | 1277 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1278 | |
1da177e4 | 1279 | /** |
7f04c26d AA |
1280 | * write_inode_now - write an inode to disk |
1281 | * @inode: inode to write to disk | |
1282 | * @sync: whether the write should be synchronous or not | |
1283 | * | |
1284 | * This function commits an inode to disk immediately if it is dirty. This is | |
1285 | * primarily needed by knfsd. | |
1da177e4 | 1286 | * |
7f04c26d | 1287 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1288 | */ |
1da177e4 LT |
1289 | int write_inode_now(struct inode *inode, int sync) |
1290 | { | |
f758eeab | 1291 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1292 | int ret; |
1293 | struct writeback_control wbc = { | |
1294 | .nr_to_write = LONG_MAX, | |
18914b18 | 1295 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1296 | .range_start = 0, |
1297 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1298 | }; |
1299 | ||
1300 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1301 | wbc.nr_to_write = 0; |
1da177e4 LT |
1302 | |
1303 | might_sleep(); | |
f758eeab | 1304 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1305 | spin_lock(&inode->i_lock); |
f758eeab | 1306 | ret = writeback_single_inode(inode, wb, &wbc); |
0f1b1fd8 | 1307 | spin_unlock(&inode->i_lock); |
f758eeab | 1308 | spin_unlock(&wb->list_lock); |
1da177e4 | 1309 | if (sync) |
1c0eeaf5 | 1310 | inode_sync_wait(inode); |
1da177e4 LT |
1311 | return ret; |
1312 | } | |
1313 | EXPORT_SYMBOL(write_inode_now); | |
1314 | ||
1315 | /** | |
1316 | * sync_inode - write an inode and its pages to disk. | |
1317 | * @inode: the inode to sync | |
1318 | * @wbc: controls the writeback mode | |
1319 | * | |
1320 | * sync_inode() will write an inode and its pages to disk. It will also | |
1321 | * correctly update the inode on its superblock's dirty inode lists and will | |
1322 | * update inode->i_state. | |
1323 | * | |
1324 | * The caller must have a ref on the inode. | |
1325 | */ | |
1326 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1327 | { | |
f758eeab | 1328 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1329 | int ret; |
1330 | ||
f758eeab | 1331 | spin_lock(&wb->list_lock); |
0f1b1fd8 | 1332 | spin_lock(&inode->i_lock); |
f758eeab | 1333 | ret = writeback_single_inode(inode, wb, wbc); |
0f1b1fd8 | 1334 | spin_unlock(&inode->i_lock); |
f758eeab | 1335 | spin_unlock(&wb->list_lock); |
1da177e4 LT |
1336 | return ret; |
1337 | } | |
1338 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1339 | |
1340 | /** | |
c691b9d9 | 1341 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1342 | * @inode: the inode to sync |
1343 | * @wait: wait for I/O to complete. | |
1344 | * | |
c691b9d9 | 1345 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1346 | * |
1347 | * Note: only writes the actual inode, no associated data or other metadata. | |
1348 | */ | |
1349 | int sync_inode_metadata(struct inode *inode, int wait) | |
1350 | { | |
1351 | struct writeback_control wbc = { | |
1352 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1353 | .nr_to_write = 0, /* metadata-only */ | |
1354 | }; | |
1355 | ||
1356 | return sync_inode(inode, &wbc); | |
1357 | } | |
1358 | EXPORT_SYMBOL(sync_inode_metadata); |