]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * fs/fs-writeback.c | |
4 | * | |
5 | * Copyright (C) 2002, Linus Torvalds. | |
6 | * | |
7 | * Contains all the functions related to writing back and waiting | |
8 | * upon dirty inodes against superblocks, and writing back dirty | |
9 | * pages against inodes. ie: data writeback. Writeout of the | |
10 | * inode itself is not handled here. | |
11 | * | |
e1f8e874 | 12 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
13 | * Split out of fs/inode.c |
14 | * Additions for address_space-based writeback | |
15 | */ | |
16 | ||
17 | #include <linux/kernel.h> | |
630d9c47 | 18 | #include <linux/export.h> |
1da177e4 | 19 | #include <linux/spinlock.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
1da177e4 LT |
21 | #include <linux/sched.h> |
22 | #include <linux/fs.h> | |
23 | #include <linux/mm.h> | |
bc31b86a | 24 | #include <linux/pagemap.h> |
03ba3782 | 25 | #include <linux/kthread.h> |
1da177e4 LT |
26 | #include <linux/writeback.h> |
27 | #include <linux/blkdev.h> | |
28 | #include <linux/backing-dev.h> | |
455b2864 | 29 | #include <linux/tracepoint.h> |
719ea2fb | 30 | #include <linux/device.h> |
21c6321f | 31 | #include <linux/memcontrol.h> |
07f3f05c | 32 | #include "internal.h" |
1da177e4 | 33 | |
bc31b86a WF |
34 | /* |
35 | * 4MB minimal write chunk size | |
36 | */ | |
09cbfeaf | 37 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_SHIFT - 10)) |
bc31b86a | 38 | |
c4a77a6c JA |
39 | /* |
40 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
41 | */ | |
83ba7b07 | 42 | struct wb_writeback_work { |
c4a77a6c JA |
43 | long nr_pages; |
44 | struct super_block *sb; | |
0dc83bd3 | 45 | unsigned long *older_than_this; |
c4a77a6c | 46 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 47 | unsigned int tagged_writepages:1; |
52957fe1 HS |
48 | unsigned int for_kupdate:1; |
49 | unsigned int range_cyclic:1; | |
50 | unsigned int for_background:1; | |
7747bd4b | 51 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
ac7b19a3 | 52 | unsigned int auto_free:1; /* free on completion */ |
0e175a18 | 53 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 54 | |
8010c3b6 | 55 | struct list_head list; /* pending work list */ |
cc395d7f | 56 | struct wb_completion *done; /* set if the caller waits */ |
03ba3782 JA |
57 | }; |
58 | ||
a2f48706 TT |
59 | /* |
60 | * If an inode is constantly having its pages dirtied, but then the | |
61 | * updates stop dirtytime_expire_interval seconds in the past, it's | |
62 | * possible for the worst case time between when an inode has its | |
63 | * timestamps updated and when they finally get written out to be two | |
64 | * dirtytime_expire_intervals. We set the default to 12 hours (in | |
65 | * seconds), which means most of the time inodes will have their | |
66 | * timestamps written to disk after 12 hours, but in the worst case a | |
67 | * few inodes might not their timestamps updated for 24 hours. | |
68 | */ | |
69 | unsigned int dirtytime_expire_interval = 12 * 60 * 60; | |
70 | ||
7ccf19a8 NP |
71 | static inline struct inode *wb_inode(struct list_head *head) |
72 | { | |
c7f54084 | 73 | return list_entry(head, struct inode, i_io_list); |
7ccf19a8 NP |
74 | } |
75 | ||
15eb77a0 WF |
76 | /* |
77 | * Include the creation of the trace points after defining the | |
78 | * wb_writeback_work structure and inline functions so that the definition | |
79 | * remains local to this file. | |
80 | */ | |
81 | #define CREATE_TRACE_POINTS | |
82 | #include <trace/events/writeback.h> | |
83 | ||
774016b2 SW |
84 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
85 | ||
d6c10f1f TH |
86 | static bool wb_io_lists_populated(struct bdi_writeback *wb) |
87 | { | |
88 | if (wb_has_dirty_io(wb)) { | |
89 | return false; | |
90 | } else { | |
91 | set_bit(WB_has_dirty_io, &wb->state); | |
95a46c65 | 92 | WARN_ON_ONCE(!wb->avg_write_bandwidth); |
766a9d6e TH |
93 | atomic_long_add(wb->avg_write_bandwidth, |
94 | &wb->bdi->tot_write_bandwidth); | |
d6c10f1f TH |
95 | return true; |
96 | } | |
97 | } | |
98 | ||
99 | static void wb_io_lists_depopulated(struct bdi_writeback *wb) | |
100 | { | |
101 | if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && | |
766a9d6e | 102 | list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { |
d6c10f1f | 103 | clear_bit(WB_has_dirty_io, &wb->state); |
95a46c65 TH |
104 | WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, |
105 | &wb->bdi->tot_write_bandwidth) < 0); | |
766a9d6e | 106 | } |
d6c10f1f TH |
107 | } |
108 | ||
109 | /** | |
c7f54084 | 110 | * inode_io_list_move_locked - move an inode onto a bdi_writeback IO list |
d6c10f1f TH |
111 | * @inode: inode to be moved |
112 | * @wb: target bdi_writeback | |
bbbc3c1c | 113 | * @head: one of @wb->b_{dirty|io|more_io|dirty_time} |
d6c10f1f | 114 | * |
c7f54084 | 115 | * Move @inode->i_io_list to @list of @wb and set %WB_has_dirty_io. |
d6c10f1f TH |
116 | * Returns %true if @inode is the first occupant of the !dirty_time IO |
117 | * lists; otherwise, %false. | |
118 | */ | |
c7f54084 | 119 | static bool inode_io_list_move_locked(struct inode *inode, |
d6c10f1f TH |
120 | struct bdi_writeback *wb, |
121 | struct list_head *head) | |
122 | { | |
123 | assert_spin_locked(&wb->list_lock); | |
124 | ||
c7f54084 | 125 | list_move(&inode->i_io_list, head); |
d6c10f1f TH |
126 | |
127 | /* dirty_time doesn't count as dirty_io until expiration */ | |
128 | if (head != &wb->b_dirty_time) | |
129 | return wb_io_lists_populated(wb); | |
130 | ||
131 | wb_io_lists_depopulated(wb); | |
132 | return false; | |
133 | } | |
134 | ||
135 | /** | |
c7f54084 | 136 | * inode_io_list_del_locked - remove an inode from its bdi_writeback IO list |
d6c10f1f TH |
137 | * @inode: inode to be removed |
138 | * @wb: bdi_writeback @inode is being removed from | |
139 | * | |
140 | * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and | |
141 | * clear %WB_has_dirty_io if all are empty afterwards. | |
142 | */ | |
c7f54084 | 143 | static void inode_io_list_del_locked(struct inode *inode, |
d6c10f1f TH |
144 | struct bdi_writeback *wb) |
145 | { | |
146 | assert_spin_locked(&wb->list_lock); | |
147 | ||
c7f54084 | 148 | list_del_init(&inode->i_io_list); |
d6c10f1f TH |
149 | wb_io_lists_depopulated(wb); |
150 | } | |
151 | ||
f0054bb1 | 152 | static void wb_wakeup(struct bdi_writeback *wb) |
5acda9d1 | 153 | { |
f0054bb1 TH |
154 | spin_lock_bh(&wb->work_lock); |
155 | if (test_bit(WB_registered, &wb->state)) | |
156 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
157 | spin_unlock_bh(&wb->work_lock); | |
5acda9d1 JK |
158 | } |
159 | ||
4a3a485b TE |
160 | static void finish_writeback_work(struct bdi_writeback *wb, |
161 | struct wb_writeback_work *work) | |
162 | { | |
163 | struct wb_completion *done = work->done; | |
164 | ||
165 | if (work->auto_free) | |
166 | kfree(work); | |
8e00c4e9 TH |
167 | if (done) { |
168 | wait_queue_head_t *waitq = done->waitq; | |
169 | ||
170 | /* @done can't be accessed after the following dec */ | |
171 | if (atomic_dec_and_test(&done->cnt)) | |
172 | wake_up_all(waitq); | |
173 | } | |
4a3a485b TE |
174 | } |
175 | ||
f0054bb1 TH |
176 | static void wb_queue_work(struct bdi_writeback *wb, |
177 | struct wb_writeback_work *work) | |
6585027a | 178 | { |
5634cc2a | 179 | trace_writeback_queue(wb, work); |
6585027a | 180 | |
cc395d7f TH |
181 | if (work->done) |
182 | atomic_inc(&work->done->cnt); | |
4a3a485b TE |
183 | |
184 | spin_lock_bh(&wb->work_lock); | |
185 | ||
186 | if (test_bit(WB_registered, &wb->state)) { | |
187 | list_add_tail(&work->list, &wb->work_list); | |
188 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
189 | } else | |
190 | finish_writeback_work(wb, work); | |
191 | ||
f0054bb1 | 192 | spin_unlock_bh(&wb->work_lock); |
1da177e4 LT |
193 | } |
194 | ||
cc395d7f TH |
195 | /** |
196 | * wb_wait_for_completion - wait for completion of bdi_writeback_works | |
cc395d7f TH |
197 | * @done: target wb_completion |
198 | * | |
199 | * Wait for one or more work items issued to @bdi with their ->done field | |
5b9cce4c TH |
200 | * set to @done, which should have been initialized with |
201 | * DEFINE_WB_COMPLETION(). This function returns after all such work items | |
202 | * are completed. Work items which are waited upon aren't freed | |
cc395d7f TH |
203 | * automatically on completion. |
204 | */ | |
5b9cce4c | 205 | void wb_wait_for_completion(struct wb_completion *done) |
cc395d7f TH |
206 | { |
207 | atomic_dec(&done->cnt); /* put down the initial count */ | |
5b9cce4c | 208 | wait_event(*done->waitq, !atomic_read(&done->cnt)); |
cc395d7f TH |
209 | } |
210 | ||
703c2708 TH |
211 | #ifdef CONFIG_CGROUP_WRITEBACK |
212 | ||
55a694df TH |
213 | /* |
214 | * Parameters for foreign inode detection, see wbc_detach_inode() to see | |
215 | * how they're used. | |
216 | * | |
217 | * These paramters are inherently heuristical as the detection target | |
218 | * itself is fuzzy. All we want to do is detaching an inode from the | |
219 | * current owner if it's being written to by some other cgroups too much. | |
220 | * | |
221 | * The current cgroup writeback is built on the assumption that multiple | |
222 | * cgroups writing to the same inode concurrently is very rare and a mode | |
223 | * of operation which isn't well supported. As such, the goal is not | |
224 | * taking too long when a different cgroup takes over an inode while | |
225 | * avoiding too aggressive flip-flops from occasional foreign writes. | |
226 | * | |
227 | * We record, very roughly, 2s worth of IO time history and if more than | |
228 | * half of that is foreign, trigger the switch. The recording is quantized | |
229 | * to 16 slots. To avoid tiny writes from swinging the decision too much, | |
230 | * writes smaller than 1/8 of avg size are ignored. | |
231 | */ | |
2a814908 TH |
232 | #define WB_FRN_TIME_SHIFT 13 /* 1s = 2^13, upto 8 secs w/ 16bit */ |
233 | #define WB_FRN_TIME_AVG_SHIFT 3 /* avg = avg * 7/8 + new * 1/8 */ | |
55a694df | 234 | #define WB_FRN_TIME_CUT_DIV 8 /* ignore rounds < avg / 8 */ |
2a814908 TH |
235 | #define WB_FRN_TIME_PERIOD (2 * (1 << WB_FRN_TIME_SHIFT)) /* 2s */ |
236 | ||
237 | #define WB_FRN_HIST_SLOTS 16 /* inode->i_wb_frn_history is 16bit */ | |
238 | #define WB_FRN_HIST_UNIT (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS) | |
239 | /* each slot's duration is 2s / 16 */ | |
240 | #define WB_FRN_HIST_THR_SLOTS (WB_FRN_HIST_SLOTS / 2) | |
241 | /* if foreign slots >= 8, switch */ | |
242 | #define WB_FRN_HIST_MAX_SLOTS (WB_FRN_HIST_THR_SLOTS / 2 + 1) | |
243 | /* one round can affect upto 5 slots */ | |
6444f47e | 244 | #define WB_FRN_MAX_IN_FLIGHT 1024 /* don't queue too many concurrently */ |
2a814908 | 245 | |
a1a0e23e TH |
246 | static atomic_t isw_nr_in_flight = ATOMIC_INIT(0); |
247 | static struct workqueue_struct *isw_wq; | |
248 | ||
21c6321f TH |
249 | void __inode_attach_wb(struct inode *inode, struct page *page) |
250 | { | |
251 | struct backing_dev_info *bdi = inode_to_bdi(inode); | |
252 | struct bdi_writeback *wb = NULL; | |
253 | ||
254 | if (inode_cgwb_enabled(inode)) { | |
255 | struct cgroup_subsys_state *memcg_css; | |
256 | ||
257 | if (page) { | |
258 | memcg_css = mem_cgroup_css_from_page(page); | |
259 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
260 | } else { | |
261 | /* must pin memcg_css, see wb_get_create() */ | |
262 | memcg_css = task_get_css(current, memory_cgrp_id); | |
263 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
264 | css_put(memcg_css); | |
265 | } | |
266 | } | |
267 | ||
268 | if (!wb) | |
269 | wb = &bdi->wb; | |
270 | ||
271 | /* | |
272 | * There may be multiple instances of this function racing to | |
273 | * update the same inode. Use cmpxchg() to tell the winner. | |
274 | */ | |
275 | if (unlikely(cmpxchg(&inode->i_wb, NULL, wb))) | |
276 | wb_put(wb); | |
277 | } | |
9b0eb69b | 278 | EXPORT_SYMBOL_GPL(__inode_attach_wb); |
21c6321f | 279 | |
87e1d789 TH |
280 | /** |
281 | * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it | |
282 | * @inode: inode of interest with i_lock held | |
283 | * | |
284 | * Returns @inode's wb with its list_lock held. @inode->i_lock must be | |
285 | * held on entry and is released on return. The returned wb is guaranteed | |
286 | * to stay @inode's associated wb until its list_lock is released. | |
287 | */ | |
288 | static struct bdi_writeback * | |
289 | locked_inode_to_wb_and_lock_list(struct inode *inode) | |
290 | __releases(&inode->i_lock) | |
291 | __acquires(&wb->list_lock) | |
292 | { | |
293 | while (true) { | |
294 | struct bdi_writeback *wb = inode_to_wb(inode); | |
295 | ||
296 | /* | |
297 | * inode_to_wb() association is protected by both | |
298 | * @inode->i_lock and @wb->list_lock but list_lock nests | |
299 | * outside i_lock. Drop i_lock and verify that the | |
300 | * association hasn't changed after acquiring list_lock. | |
301 | */ | |
302 | wb_get(wb); | |
303 | spin_unlock(&inode->i_lock); | |
304 | spin_lock(&wb->list_lock); | |
87e1d789 | 305 | |
aaa2cacf | 306 | /* i_wb may have changed inbetween, can't use inode_to_wb() */ |
614a4e37 TH |
307 | if (likely(wb == inode->i_wb)) { |
308 | wb_put(wb); /* @inode already has ref */ | |
309 | return wb; | |
310 | } | |
87e1d789 TH |
311 | |
312 | spin_unlock(&wb->list_lock); | |
614a4e37 | 313 | wb_put(wb); |
87e1d789 TH |
314 | cpu_relax(); |
315 | spin_lock(&inode->i_lock); | |
316 | } | |
317 | } | |
318 | ||
319 | /** | |
320 | * inode_to_wb_and_lock_list - determine an inode's wb and lock it | |
321 | * @inode: inode of interest | |
322 | * | |
323 | * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held | |
324 | * on entry. | |
325 | */ | |
326 | static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) | |
327 | __acquires(&wb->list_lock) | |
328 | { | |
329 | spin_lock(&inode->i_lock); | |
330 | return locked_inode_to_wb_and_lock_list(inode); | |
331 | } | |
332 | ||
682aa8e1 TH |
333 | struct inode_switch_wbs_context { |
334 | struct inode *inode; | |
335 | struct bdi_writeback *new_wb; | |
336 | ||
337 | struct rcu_head rcu_head; | |
338 | struct work_struct work; | |
339 | }; | |
340 | ||
7fc5854f TH |
341 | static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) |
342 | { | |
343 | down_write(&bdi->wb_switch_rwsem); | |
344 | } | |
345 | ||
346 | static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) | |
347 | { | |
348 | up_write(&bdi->wb_switch_rwsem); | |
349 | } | |
350 | ||
682aa8e1 TH |
351 | static void inode_switch_wbs_work_fn(struct work_struct *work) |
352 | { | |
353 | struct inode_switch_wbs_context *isw = | |
354 | container_of(work, struct inode_switch_wbs_context, work); | |
355 | struct inode *inode = isw->inode; | |
7fc5854f | 356 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
d10c8095 TH |
357 | struct address_space *mapping = inode->i_mapping; |
358 | struct bdi_writeback *old_wb = inode->i_wb; | |
682aa8e1 | 359 | struct bdi_writeback *new_wb = isw->new_wb; |
04edf02c MW |
360 | XA_STATE(xas, &mapping->i_pages, 0); |
361 | struct page *page; | |
d10c8095 | 362 | bool switched = false; |
682aa8e1 | 363 | |
7fc5854f TH |
364 | /* |
365 | * If @inode switches cgwb membership while sync_inodes_sb() is | |
366 | * being issued, sync_inodes_sb() might miss it. Synchronize. | |
367 | */ | |
368 | down_read(&bdi->wb_switch_rwsem); | |
369 | ||
682aa8e1 TH |
370 | /* |
371 | * By the time control reaches here, RCU grace period has passed | |
372 | * since I_WB_SWITCH assertion and all wb stat update transactions | |
373 | * between unlocked_inode_to_wb_begin/end() are guaranteed to be | |
b93b0163 | 374 | * synchronizing against the i_pages lock. |
d10c8095 | 375 | * |
b93b0163 | 376 | * Grabbing old_wb->list_lock, inode->i_lock and the i_pages lock |
d10c8095 TH |
377 | * gives us exclusion against all wb related operations on @inode |
378 | * including IO list manipulations and stat updates. | |
682aa8e1 | 379 | */ |
d10c8095 TH |
380 | if (old_wb < new_wb) { |
381 | spin_lock(&old_wb->list_lock); | |
382 | spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING); | |
383 | } else { | |
384 | spin_lock(&new_wb->list_lock); | |
385 | spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING); | |
386 | } | |
682aa8e1 | 387 | spin_lock(&inode->i_lock); |
b93b0163 | 388 | xa_lock_irq(&mapping->i_pages); |
d10c8095 TH |
389 | |
390 | /* | |
391 | * Once I_FREEING is visible under i_lock, the eviction path owns | |
c7f54084 | 392 | * the inode and we shouldn't modify ->i_io_list. |
d10c8095 TH |
393 | */ |
394 | if (unlikely(inode->i_state & I_FREEING)) | |
395 | goto skip_switch; | |
396 | ||
3a8e9ac8 TH |
397 | trace_inode_switch_wbs(inode, old_wb, new_wb); |
398 | ||
d10c8095 TH |
399 | /* |
400 | * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points | |
401 | * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to | |
b93b0163 | 402 | * pages actually under writeback. |
d10c8095 | 403 | */ |
04edf02c MW |
404 | xas_for_each_marked(&xas, page, ULONG_MAX, PAGECACHE_TAG_DIRTY) { |
405 | if (PageDirty(page)) { | |
3e8f399d NB |
406 | dec_wb_stat(old_wb, WB_RECLAIMABLE); |
407 | inc_wb_stat(new_wb, WB_RECLAIMABLE); | |
d10c8095 TH |
408 | } |
409 | } | |
410 | ||
04edf02c MW |
411 | xas_set(&xas, 0); |
412 | xas_for_each_marked(&xas, page, ULONG_MAX, PAGECACHE_TAG_WRITEBACK) { | |
413 | WARN_ON_ONCE(!PageWriteback(page)); | |
414 | dec_wb_stat(old_wb, WB_WRITEBACK); | |
415 | inc_wb_stat(new_wb, WB_WRITEBACK); | |
d10c8095 TH |
416 | } |
417 | ||
418 | wb_get(new_wb); | |
419 | ||
420 | /* | |
421 | * Transfer to @new_wb's IO list if necessary. The specific list | |
422 | * @inode was on is ignored and the inode is put on ->b_dirty which | |
423 | * is always correct including from ->b_dirty_time. The transfer | |
424 | * preserves @inode->dirtied_when ordering. | |
425 | */ | |
c7f54084 | 426 | if (!list_empty(&inode->i_io_list)) { |
d10c8095 TH |
427 | struct inode *pos; |
428 | ||
c7f54084 | 429 | inode_io_list_del_locked(inode, old_wb); |
d10c8095 | 430 | inode->i_wb = new_wb; |
c7f54084 | 431 | list_for_each_entry(pos, &new_wb->b_dirty, i_io_list) |
d10c8095 TH |
432 | if (time_after_eq(inode->dirtied_when, |
433 | pos->dirtied_when)) | |
434 | break; | |
c7f54084 | 435 | inode_io_list_move_locked(inode, new_wb, pos->i_io_list.prev); |
d10c8095 TH |
436 | } else { |
437 | inode->i_wb = new_wb; | |
438 | } | |
682aa8e1 | 439 | |
d10c8095 | 440 | /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */ |
682aa8e1 TH |
441 | inode->i_wb_frn_winner = 0; |
442 | inode->i_wb_frn_avg_time = 0; | |
443 | inode->i_wb_frn_history = 0; | |
d10c8095 TH |
444 | switched = true; |
445 | skip_switch: | |
682aa8e1 TH |
446 | /* |
447 | * Paired with load_acquire in unlocked_inode_to_wb_begin() and | |
448 | * ensures that the new wb is visible if they see !I_WB_SWITCH. | |
449 | */ | |
450 | smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH); | |
451 | ||
b93b0163 | 452 | xa_unlock_irq(&mapping->i_pages); |
682aa8e1 | 453 | spin_unlock(&inode->i_lock); |
d10c8095 TH |
454 | spin_unlock(&new_wb->list_lock); |
455 | spin_unlock(&old_wb->list_lock); | |
682aa8e1 | 456 | |
7fc5854f TH |
457 | up_read(&bdi->wb_switch_rwsem); |
458 | ||
d10c8095 TH |
459 | if (switched) { |
460 | wb_wakeup(new_wb); | |
461 | wb_put(old_wb); | |
462 | } | |
682aa8e1 | 463 | wb_put(new_wb); |
d10c8095 TH |
464 | |
465 | iput(inode); | |
682aa8e1 | 466 | kfree(isw); |
a1a0e23e TH |
467 | |
468 | atomic_dec(&isw_nr_in_flight); | |
682aa8e1 TH |
469 | } |
470 | ||
471 | static void inode_switch_wbs_rcu_fn(struct rcu_head *rcu_head) | |
472 | { | |
473 | struct inode_switch_wbs_context *isw = container_of(rcu_head, | |
474 | struct inode_switch_wbs_context, rcu_head); | |
475 | ||
476 | /* needs to grab bh-unsafe locks, bounce to work item */ | |
477 | INIT_WORK(&isw->work, inode_switch_wbs_work_fn); | |
a1a0e23e | 478 | queue_work(isw_wq, &isw->work); |
682aa8e1 TH |
479 | } |
480 | ||
481 | /** | |
482 | * inode_switch_wbs - change the wb association of an inode | |
483 | * @inode: target inode | |
484 | * @new_wb_id: ID of the new wb | |
485 | * | |
486 | * Switch @inode's wb association to the wb identified by @new_wb_id. The | |
487 | * switching is performed asynchronously and may fail silently. | |
488 | */ | |
489 | static void inode_switch_wbs(struct inode *inode, int new_wb_id) | |
490 | { | |
491 | struct backing_dev_info *bdi = inode_to_bdi(inode); | |
492 | struct cgroup_subsys_state *memcg_css; | |
493 | struct inode_switch_wbs_context *isw; | |
494 | ||
495 | /* noop if seems to be already in progress */ | |
496 | if (inode->i_state & I_WB_SWITCH) | |
497 | return; | |
498 | ||
6444f47e TH |
499 | /* avoid queueing a new switch if too many are already in flight */ |
500 | if (atomic_read(&isw_nr_in_flight) > WB_FRN_MAX_IN_FLIGHT) | |
7fc5854f TH |
501 | return; |
502 | ||
682aa8e1 TH |
503 | isw = kzalloc(sizeof(*isw), GFP_ATOMIC); |
504 | if (!isw) | |
6444f47e | 505 | return; |
682aa8e1 TH |
506 | |
507 | /* find and pin the new wb */ | |
508 | rcu_read_lock(); | |
509 | memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys); | |
510 | if (memcg_css) | |
511 | isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
512 | rcu_read_unlock(); | |
513 | if (!isw->new_wb) | |
514 | goto out_free; | |
515 | ||
516 | /* while holding I_WB_SWITCH, no one else can update the association */ | |
517 | spin_lock(&inode->i_lock); | |
1751e8a6 | 518 | if (!(inode->i_sb->s_flags & SB_ACTIVE) || |
a1a0e23e TH |
519 | inode->i_state & (I_WB_SWITCH | I_FREEING) || |
520 | inode_to_wb(inode) == isw->new_wb) { | |
521 | spin_unlock(&inode->i_lock); | |
522 | goto out_free; | |
523 | } | |
682aa8e1 | 524 | inode->i_state |= I_WB_SWITCH; |
74524955 | 525 | __iget(inode); |
682aa8e1 TH |
526 | spin_unlock(&inode->i_lock); |
527 | ||
682aa8e1 TH |
528 | isw->inode = inode; |
529 | ||
530 | /* | |
531 | * In addition to synchronizing among switchers, I_WB_SWITCH tells | |
b93b0163 MW |
532 | * the RCU protected stat update paths to grab the i_page |
533 | * lock so that stat transfer can synchronize against them. | |
682aa8e1 TH |
534 | * Let's continue after I_WB_SWITCH is guaranteed to be visible. |
535 | */ | |
536 | call_rcu(&isw->rcu_head, inode_switch_wbs_rcu_fn); | |
ec084de9 JX |
537 | |
538 | atomic_inc(&isw_nr_in_flight); | |
6444f47e | 539 | return; |
682aa8e1 TH |
540 | |
541 | out_free: | |
542 | if (isw->new_wb) | |
543 | wb_put(isw->new_wb); | |
544 | kfree(isw); | |
545 | } | |
546 | ||
b16b1deb TH |
547 | /** |
548 | * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it | |
549 | * @wbc: writeback_control of interest | |
550 | * @inode: target inode | |
551 | * | |
552 | * @inode is locked and about to be written back under the control of @wbc. | |
553 | * Record @inode's writeback context into @wbc and unlock the i_lock. On | |
554 | * writeback completion, wbc_detach_inode() should be called. This is used | |
555 | * to track the cgroup writeback context. | |
556 | */ | |
557 | void wbc_attach_and_unlock_inode(struct writeback_control *wbc, | |
558 | struct inode *inode) | |
559 | { | |
dd73e4b7 TH |
560 | if (!inode_cgwb_enabled(inode)) { |
561 | spin_unlock(&inode->i_lock); | |
562 | return; | |
563 | } | |
564 | ||
b16b1deb | 565 | wbc->wb = inode_to_wb(inode); |
2a814908 TH |
566 | wbc->inode = inode; |
567 | ||
568 | wbc->wb_id = wbc->wb->memcg_css->id; | |
569 | wbc->wb_lcand_id = inode->i_wb_frn_winner; | |
570 | wbc->wb_tcand_id = 0; | |
571 | wbc->wb_bytes = 0; | |
572 | wbc->wb_lcand_bytes = 0; | |
573 | wbc->wb_tcand_bytes = 0; | |
574 | ||
b16b1deb TH |
575 | wb_get(wbc->wb); |
576 | spin_unlock(&inode->i_lock); | |
e8a7abf5 TH |
577 | |
578 | /* | |
579 | * A dying wb indicates that the memcg-blkcg mapping has changed | |
580 | * and a new wb is already serving the memcg. Switch immediately. | |
581 | */ | |
582 | if (unlikely(wb_dying(wbc->wb))) | |
583 | inode_switch_wbs(inode, wbc->wb_id); | |
b16b1deb | 584 | } |
9b0eb69b | 585 | EXPORT_SYMBOL_GPL(wbc_attach_and_unlock_inode); |
b16b1deb TH |
586 | |
587 | /** | |
2a814908 TH |
588 | * wbc_detach_inode - disassociate wbc from inode and perform foreign detection |
589 | * @wbc: writeback_control of the just finished writeback | |
b16b1deb TH |
590 | * |
591 | * To be called after a writeback attempt of an inode finishes and undoes | |
592 | * wbc_attach_and_unlock_inode(). Can be called under any context. | |
2a814908 TH |
593 | * |
594 | * As concurrent write sharing of an inode is expected to be very rare and | |
595 | * memcg only tracks page ownership on first-use basis severely confining | |
596 | * the usefulness of such sharing, cgroup writeback tracks ownership | |
597 | * per-inode. While the support for concurrent write sharing of an inode | |
598 | * is deemed unnecessary, an inode being written to by different cgroups at | |
599 | * different points in time is a lot more common, and, more importantly, | |
600 | * charging only by first-use can too readily lead to grossly incorrect | |
601 | * behaviors (single foreign page can lead to gigabytes of writeback to be | |
602 | * incorrectly attributed). | |
603 | * | |
604 | * To resolve this issue, cgroup writeback detects the majority dirtier of | |
605 | * an inode and transfers the ownership to it. To avoid unnnecessary | |
606 | * oscillation, the detection mechanism keeps track of history and gives | |
607 | * out the switch verdict only if the foreign usage pattern is stable over | |
608 | * a certain amount of time and/or writeback attempts. | |
609 | * | |
610 | * On each writeback attempt, @wbc tries to detect the majority writer | |
611 | * using Boyer-Moore majority vote algorithm. In addition to the byte | |
612 | * count from the majority voting, it also counts the bytes written for the | |
613 | * current wb and the last round's winner wb (max of last round's current | |
614 | * wb, the winner from two rounds ago, and the last round's majority | |
615 | * candidate). Keeping track of the historical winner helps the algorithm | |
616 | * to semi-reliably detect the most active writer even when it's not the | |
617 | * absolute majority. | |
618 | * | |
619 | * Once the winner of the round is determined, whether the winner is | |
620 | * foreign or not and how much IO time the round consumed is recorded in | |
621 | * inode->i_wb_frn_history. If the amount of recorded foreign IO time is | |
622 | * over a certain threshold, the switch verdict is given. | |
b16b1deb TH |
623 | */ |
624 | void wbc_detach_inode(struct writeback_control *wbc) | |
625 | { | |
2a814908 TH |
626 | struct bdi_writeback *wb = wbc->wb; |
627 | struct inode *inode = wbc->inode; | |
dd73e4b7 TH |
628 | unsigned long avg_time, max_bytes, max_time; |
629 | u16 history; | |
2a814908 TH |
630 | int max_id; |
631 | ||
dd73e4b7 TH |
632 | if (!wb) |
633 | return; | |
634 | ||
635 | history = inode->i_wb_frn_history; | |
636 | avg_time = inode->i_wb_frn_avg_time; | |
637 | ||
2a814908 TH |
638 | /* pick the winner of this round */ |
639 | if (wbc->wb_bytes >= wbc->wb_lcand_bytes && | |
640 | wbc->wb_bytes >= wbc->wb_tcand_bytes) { | |
641 | max_id = wbc->wb_id; | |
642 | max_bytes = wbc->wb_bytes; | |
643 | } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) { | |
644 | max_id = wbc->wb_lcand_id; | |
645 | max_bytes = wbc->wb_lcand_bytes; | |
646 | } else { | |
647 | max_id = wbc->wb_tcand_id; | |
648 | max_bytes = wbc->wb_tcand_bytes; | |
649 | } | |
650 | ||
651 | /* | |
652 | * Calculate the amount of IO time the winner consumed and fold it | |
653 | * into the running average kept per inode. If the consumed IO | |
654 | * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for | |
655 | * deciding whether to switch or not. This is to prevent one-off | |
656 | * small dirtiers from skewing the verdict. | |
657 | */ | |
658 | max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT, | |
659 | wb->avg_write_bandwidth); | |
660 | if (avg_time) | |
661 | avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) - | |
662 | (avg_time >> WB_FRN_TIME_AVG_SHIFT); | |
663 | else | |
664 | avg_time = max_time; /* immediate catch up on first run */ | |
665 | ||
666 | if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) { | |
667 | int slots; | |
668 | ||
669 | /* | |
670 | * The switch verdict is reached if foreign wb's consume | |
671 | * more than a certain proportion of IO time in a | |
672 | * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot | |
673 | * history mask where each bit represents one sixteenth of | |
674 | * the period. Determine the number of slots to shift into | |
675 | * history from @max_time. | |
676 | */ | |
677 | slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT), | |
678 | (unsigned long)WB_FRN_HIST_MAX_SLOTS); | |
679 | history <<= slots; | |
680 | if (wbc->wb_id != max_id) | |
681 | history |= (1U << slots) - 1; | |
682 | ||
3a8e9ac8 TH |
683 | if (history) |
684 | trace_inode_foreign_history(inode, wbc, history); | |
685 | ||
2a814908 TH |
686 | /* |
687 | * Switch if the current wb isn't the consistent winner. | |
688 | * If there are multiple closely competing dirtiers, the | |
689 | * inode may switch across them repeatedly over time, which | |
690 | * is okay. The main goal is avoiding keeping an inode on | |
691 | * the wrong wb for an extended period of time. | |
692 | */ | |
682aa8e1 TH |
693 | if (hweight32(history) > WB_FRN_HIST_THR_SLOTS) |
694 | inode_switch_wbs(inode, max_id); | |
2a814908 TH |
695 | } |
696 | ||
697 | /* | |
698 | * Multiple instances of this function may race to update the | |
699 | * following fields but we don't mind occassional inaccuracies. | |
700 | */ | |
701 | inode->i_wb_frn_winner = max_id; | |
702 | inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX); | |
703 | inode->i_wb_frn_history = history; | |
704 | ||
b16b1deb TH |
705 | wb_put(wbc->wb); |
706 | wbc->wb = NULL; | |
707 | } | |
9b0eb69b | 708 | EXPORT_SYMBOL_GPL(wbc_detach_inode); |
b16b1deb | 709 | |
2a814908 | 710 | /** |
34e51a5e | 711 | * wbc_account_cgroup_owner - account writeback to update inode cgroup ownership |
2a814908 TH |
712 | * @wbc: writeback_control of the writeback in progress |
713 | * @page: page being written out | |
714 | * @bytes: number of bytes being written out | |
715 | * | |
716 | * @bytes from @page are about to written out during the writeback | |
717 | * controlled by @wbc. Keep the book for foreign inode detection. See | |
718 | * wbc_detach_inode(). | |
719 | */ | |
34e51a5e TH |
720 | void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page, |
721 | size_t bytes) | |
2a814908 | 722 | { |
66311422 | 723 | struct cgroup_subsys_state *css; |
2a814908 TH |
724 | int id; |
725 | ||
726 | /* | |
727 | * pageout() path doesn't attach @wbc to the inode being written | |
728 | * out. This is intentional as we don't want the function to block | |
729 | * behind a slow cgroup. Ultimately, we want pageout() to kick off | |
730 | * regular writeback instead of writing things out itself. | |
731 | */ | |
27b36d8f | 732 | if (!wbc->wb || wbc->no_cgroup_owner) |
2a814908 TH |
733 | return; |
734 | ||
66311422 TH |
735 | css = mem_cgroup_css_from_page(page); |
736 | /* dead cgroups shouldn't contribute to inode ownership arbitration */ | |
737 | if (!(css->flags & CSS_ONLINE)) | |
738 | return; | |
739 | ||
740 | id = css->id; | |
2a814908 TH |
741 | |
742 | if (id == wbc->wb_id) { | |
743 | wbc->wb_bytes += bytes; | |
744 | return; | |
745 | } | |
746 | ||
747 | if (id == wbc->wb_lcand_id) | |
748 | wbc->wb_lcand_bytes += bytes; | |
749 | ||
750 | /* Boyer-Moore majority vote algorithm */ | |
751 | if (!wbc->wb_tcand_bytes) | |
752 | wbc->wb_tcand_id = id; | |
753 | if (id == wbc->wb_tcand_id) | |
754 | wbc->wb_tcand_bytes += bytes; | |
755 | else | |
756 | wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes); | |
757 | } | |
34e51a5e | 758 | EXPORT_SYMBOL_GPL(wbc_account_cgroup_owner); |
2a814908 | 759 | |
703c2708 TH |
760 | /** |
761 | * inode_congested - test whether an inode is congested | |
60292bcc | 762 | * @inode: inode to test for congestion (may be NULL) |
703c2708 TH |
763 | * @cong_bits: mask of WB_[a]sync_congested bits to test |
764 | * | |
765 | * Tests whether @inode is congested. @cong_bits is the mask of congestion | |
766 | * bits to test and the return value is the mask of set bits. | |
767 | * | |
768 | * If cgroup writeback is enabled for @inode, the congestion state is | |
769 | * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg | |
770 | * associated with @inode is congested; otherwise, the root wb's congestion | |
771 | * state is used. | |
60292bcc TH |
772 | * |
773 | * @inode is allowed to be NULL as this function is often called on | |
774 | * mapping->host which is NULL for the swapper space. | |
703c2708 TH |
775 | */ |
776 | int inode_congested(struct inode *inode, int cong_bits) | |
777 | { | |
5cb8b824 TH |
778 | /* |
779 | * Once set, ->i_wb never becomes NULL while the inode is alive. | |
780 | * Start transaction iff ->i_wb is visible. | |
781 | */ | |
aaa2cacf | 782 | if (inode && inode_to_wb_is_valid(inode)) { |
5cb8b824 | 783 | struct bdi_writeback *wb; |
2e898e4c GT |
784 | struct wb_lock_cookie lock_cookie = {}; |
785 | bool congested; | |
5cb8b824 | 786 | |
2e898e4c | 787 | wb = unlocked_inode_to_wb_begin(inode, &lock_cookie); |
5cb8b824 | 788 | congested = wb_congested(wb, cong_bits); |
2e898e4c | 789 | unlocked_inode_to_wb_end(inode, &lock_cookie); |
5cb8b824 | 790 | return congested; |
703c2708 TH |
791 | } |
792 | ||
793 | return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); | |
794 | } | |
795 | EXPORT_SYMBOL_GPL(inode_congested); | |
796 | ||
f2b65121 TH |
797 | /** |
798 | * wb_split_bdi_pages - split nr_pages to write according to bandwidth | |
799 | * @wb: target bdi_writeback to split @nr_pages to | |
800 | * @nr_pages: number of pages to write for the whole bdi | |
801 | * | |
802 | * Split @wb's portion of @nr_pages according to @wb's write bandwidth in | |
803 | * relation to the total write bandwidth of all wb's w/ dirty inodes on | |
804 | * @wb->bdi. | |
805 | */ | |
806 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
807 | { | |
808 | unsigned long this_bw = wb->avg_write_bandwidth; | |
809 | unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); | |
810 | ||
811 | if (nr_pages == LONG_MAX) | |
812 | return LONG_MAX; | |
813 | ||
814 | /* | |
815 | * This may be called on clean wb's and proportional distribution | |
816 | * may not make sense, just use the original @nr_pages in those | |
817 | * cases. In general, we wanna err on the side of writing more. | |
818 | */ | |
819 | if (!tot_bw || this_bw >= tot_bw) | |
820 | return nr_pages; | |
821 | else | |
822 | return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); | |
823 | } | |
824 | ||
db125360 TH |
825 | /** |
826 | * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi | |
827 | * @bdi: target backing_dev_info | |
828 | * @base_work: wb_writeback_work to issue | |
829 | * @skip_if_busy: skip wb's which already have writeback in progress | |
830 | * | |
831 | * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which | |
832 | * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's | |
833 | * distributed to the busy wbs according to each wb's proportion in the | |
834 | * total active write bandwidth of @bdi. | |
835 | */ | |
836 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, | |
837 | struct wb_writeback_work *base_work, | |
838 | bool skip_if_busy) | |
839 | { | |
b817525a | 840 | struct bdi_writeback *last_wb = NULL; |
b33e18f6 TH |
841 | struct bdi_writeback *wb = list_entry(&bdi->wb_list, |
842 | struct bdi_writeback, bdi_node); | |
db125360 TH |
843 | |
844 | might_sleep(); | |
db125360 TH |
845 | restart: |
846 | rcu_read_lock(); | |
b817525a | 847 | list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) { |
5b9cce4c | 848 | DEFINE_WB_COMPLETION(fallback_work_done, bdi); |
8a1270cd TH |
849 | struct wb_writeback_work fallback_work; |
850 | struct wb_writeback_work *work; | |
851 | long nr_pages; | |
852 | ||
b817525a TH |
853 | if (last_wb) { |
854 | wb_put(last_wb); | |
855 | last_wb = NULL; | |
856 | } | |
857 | ||
006a0973 TH |
858 | /* SYNC_ALL writes out I_DIRTY_TIME too */ |
859 | if (!wb_has_dirty_io(wb) && | |
860 | (base_work->sync_mode == WB_SYNC_NONE || | |
861 | list_empty(&wb->b_dirty_time))) | |
862 | continue; | |
863 | if (skip_if_busy && writeback_in_progress(wb)) | |
db125360 TH |
864 | continue; |
865 | ||
8a1270cd TH |
866 | nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages); |
867 | ||
868 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
869 | if (work) { | |
870 | *work = *base_work; | |
871 | work->nr_pages = nr_pages; | |
872 | work->auto_free = 1; | |
873 | wb_queue_work(wb, work); | |
874 | continue; | |
db125360 | 875 | } |
8a1270cd TH |
876 | |
877 | /* alloc failed, execute synchronously using on-stack fallback */ | |
878 | work = &fallback_work; | |
879 | *work = *base_work; | |
880 | work->nr_pages = nr_pages; | |
881 | work->auto_free = 0; | |
882 | work->done = &fallback_work_done; | |
883 | ||
884 | wb_queue_work(wb, work); | |
885 | ||
b817525a TH |
886 | /* |
887 | * Pin @wb so that it stays on @bdi->wb_list. This allows | |
888 | * continuing iteration from @wb after dropping and | |
889 | * regrabbing rcu read lock. | |
890 | */ | |
891 | wb_get(wb); | |
892 | last_wb = wb; | |
893 | ||
8a1270cd | 894 | rcu_read_unlock(); |
5b9cce4c | 895 | wb_wait_for_completion(&fallback_work_done); |
8a1270cd | 896 | goto restart; |
db125360 TH |
897 | } |
898 | rcu_read_unlock(); | |
b817525a TH |
899 | |
900 | if (last_wb) | |
901 | wb_put(last_wb); | |
db125360 TH |
902 | } |
903 | ||
d62241c7 TH |
904 | /** |
905 | * cgroup_writeback_by_id - initiate cgroup writeback from bdi and memcg IDs | |
906 | * @bdi_id: target bdi id | |
907 | * @memcg_id: target memcg css id | |
908 | * @nr_pages: number of pages to write, 0 for best-effort dirty flushing | |
909 | * @reason: reason why some writeback work initiated | |
910 | * @done: target wb_completion | |
911 | * | |
912 | * Initiate flush of the bdi_writeback identified by @bdi_id and @memcg_id | |
913 | * with the specified parameters. | |
914 | */ | |
915 | int cgroup_writeback_by_id(u64 bdi_id, int memcg_id, unsigned long nr, | |
916 | enum wb_reason reason, struct wb_completion *done) | |
917 | { | |
918 | struct backing_dev_info *bdi; | |
919 | struct cgroup_subsys_state *memcg_css; | |
920 | struct bdi_writeback *wb; | |
921 | struct wb_writeback_work *work; | |
922 | int ret; | |
923 | ||
924 | /* lookup bdi and memcg */ | |
925 | bdi = bdi_get_by_id(bdi_id); | |
926 | if (!bdi) | |
927 | return -ENOENT; | |
928 | ||
929 | rcu_read_lock(); | |
930 | memcg_css = css_from_id(memcg_id, &memory_cgrp_subsys); | |
931 | if (memcg_css && !css_tryget(memcg_css)) | |
932 | memcg_css = NULL; | |
933 | rcu_read_unlock(); | |
934 | if (!memcg_css) { | |
935 | ret = -ENOENT; | |
936 | goto out_bdi_put; | |
937 | } | |
938 | ||
939 | /* | |
940 | * And find the associated wb. If the wb isn't there already | |
941 | * there's nothing to flush, don't create one. | |
942 | */ | |
943 | wb = wb_get_lookup(bdi, memcg_css); | |
944 | if (!wb) { | |
945 | ret = -ENOENT; | |
946 | goto out_css_put; | |
947 | } | |
948 | ||
949 | /* | |
950 | * If @nr is zero, the caller is attempting to write out most of | |
951 | * the currently dirty pages. Let's take the current dirty page | |
952 | * count and inflate it by 25% which should be large enough to | |
953 | * flush out most dirty pages while avoiding getting livelocked by | |
954 | * concurrent dirtiers. | |
955 | */ | |
956 | if (!nr) { | |
957 | unsigned long filepages, headroom, dirty, writeback; | |
958 | ||
959 | mem_cgroup_wb_stats(wb, &filepages, &headroom, &dirty, | |
960 | &writeback); | |
961 | nr = dirty * 10 / 8; | |
962 | } | |
963 | ||
964 | /* issue the writeback work */ | |
965 | work = kzalloc(sizeof(*work), GFP_NOWAIT | __GFP_NOWARN); | |
966 | if (work) { | |
967 | work->nr_pages = nr; | |
968 | work->sync_mode = WB_SYNC_NONE; | |
969 | work->range_cyclic = 1; | |
970 | work->reason = reason; | |
971 | work->done = done; | |
972 | work->auto_free = 1; | |
973 | wb_queue_work(wb, work); | |
974 | ret = 0; | |
975 | } else { | |
976 | ret = -ENOMEM; | |
977 | } | |
978 | ||
979 | wb_put(wb); | |
980 | out_css_put: | |
981 | css_put(memcg_css); | |
982 | out_bdi_put: | |
983 | bdi_put(bdi); | |
984 | return ret; | |
985 | } | |
986 | ||
a1a0e23e TH |
987 | /** |
988 | * cgroup_writeback_umount - flush inode wb switches for umount | |
989 | * | |
990 | * This function is called when a super_block is about to be destroyed and | |
991 | * flushes in-flight inode wb switches. An inode wb switch goes through | |
992 | * RCU and then workqueue, so the two need to be flushed in order to ensure | |
993 | * that all previously scheduled switches are finished. As wb switches are | |
994 | * rare occurrences and synchronize_rcu() can take a while, perform | |
995 | * flushing iff wb switches are in flight. | |
996 | */ | |
997 | void cgroup_writeback_umount(void) | |
998 | { | |
999 | if (atomic_read(&isw_nr_in_flight)) { | |
ec084de9 JX |
1000 | /* |
1001 | * Use rcu_barrier() to wait for all pending callbacks to | |
1002 | * ensure that all in-flight wb switches are in the workqueue. | |
1003 | */ | |
1004 | rcu_barrier(); | |
a1a0e23e TH |
1005 | flush_workqueue(isw_wq); |
1006 | } | |
1007 | } | |
1008 | ||
1009 | static int __init cgroup_writeback_init(void) | |
1010 | { | |
1011 | isw_wq = alloc_workqueue("inode_switch_wbs", 0, 0); | |
1012 | if (!isw_wq) | |
1013 | return -ENOMEM; | |
1014 | return 0; | |
1015 | } | |
1016 | fs_initcall(cgroup_writeback_init); | |
1017 | ||
f2b65121 TH |
1018 | #else /* CONFIG_CGROUP_WRITEBACK */ |
1019 | ||
7fc5854f TH |
1020 | static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) { } |
1021 | static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) { } | |
1022 | ||
87e1d789 TH |
1023 | static struct bdi_writeback * |
1024 | locked_inode_to_wb_and_lock_list(struct inode *inode) | |
1025 | __releases(&inode->i_lock) | |
1026 | __acquires(&wb->list_lock) | |
1027 | { | |
1028 | struct bdi_writeback *wb = inode_to_wb(inode); | |
1029 | ||
1030 | spin_unlock(&inode->i_lock); | |
1031 | spin_lock(&wb->list_lock); | |
1032 | return wb; | |
1033 | } | |
1034 | ||
1035 | static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) | |
1036 | __acquires(&wb->list_lock) | |
1037 | { | |
1038 | struct bdi_writeback *wb = inode_to_wb(inode); | |
1039 | ||
1040 | spin_lock(&wb->list_lock); | |
1041 | return wb; | |
1042 | } | |
1043 | ||
f2b65121 TH |
1044 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) |
1045 | { | |
1046 | return nr_pages; | |
1047 | } | |
1048 | ||
db125360 TH |
1049 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, |
1050 | struct wb_writeback_work *base_work, | |
1051 | bool skip_if_busy) | |
1052 | { | |
1053 | might_sleep(); | |
1054 | ||
006a0973 | 1055 | if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) { |
db125360 | 1056 | base_work->auto_free = 0; |
db125360 TH |
1057 | wb_queue_work(&bdi->wb, base_work); |
1058 | } | |
1059 | } | |
1060 | ||
703c2708 TH |
1061 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
1062 | ||
e8e8a0c6 JA |
1063 | /* |
1064 | * Add in the number of potentially dirty inodes, because each inode | |
1065 | * write can dirty pagecache in the underlying blockdev. | |
1066 | */ | |
1067 | static unsigned long get_nr_dirty_pages(void) | |
1068 | { | |
1069 | return global_node_page_state(NR_FILE_DIRTY) + | |
1070 | global_node_page_state(NR_UNSTABLE_NFS) + | |
1071 | get_nr_dirty_inodes(); | |
1072 | } | |
1073 | ||
1074 | static void wb_start_writeback(struct bdi_writeback *wb, enum wb_reason reason) | |
b6e51316 | 1075 | { |
c00ddad3 TH |
1076 | if (!wb_has_dirty_io(wb)) |
1077 | return; | |
1078 | ||
aac8d41c JA |
1079 | /* |
1080 | * All callers of this function want to start writeback of all | |
1081 | * dirty pages. Places like vmscan can call this at a very | |
1082 | * high frequency, causing pointless allocations of tons of | |
1083 | * work items and keeping the flusher threads busy retrieving | |
1084 | * that work. Ensure that we only allow one of them pending and | |
85009b4f | 1085 | * inflight at the time. |
aac8d41c | 1086 | */ |
85009b4f JA |
1087 | if (test_bit(WB_start_all, &wb->state) || |
1088 | test_and_set_bit(WB_start_all, &wb->state)) | |
aac8d41c JA |
1089 | return; |
1090 | ||
85009b4f JA |
1091 | wb->start_all_reason = reason; |
1092 | wb_wakeup(wb); | |
c5444198 | 1093 | } |
d3ddec76 | 1094 | |
c5444198 | 1095 | /** |
9ecf4866 TH |
1096 | * wb_start_background_writeback - start background writeback |
1097 | * @wb: bdi_writback to write from | |
c5444198 CH |
1098 | * |
1099 | * Description: | |
6585027a | 1100 | * This makes sure WB_SYNC_NONE background writeback happens. When |
9ecf4866 | 1101 | * this function returns, it is only guaranteed that for given wb |
6585027a JK |
1102 | * some IO is happening if we are over background dirty threshold. |
1103 | * Caller need not hold sb s_umount semaphore. | |
c5444198 | 1104 | */ |
9ecf4866 | 1105 | void wb_start_background_writeback(struct bdi_writeback *wb) |
c5444198 | 1106 | { |
6585027a JK |
1107 | /* |
1108 | * We just wake up the flusher thread. It will perform background | |
1109 | * writeback as soon as there is no other work to do. | |
1110 | */ | |
5634cc2a | 1111 | trace_writeback_wake_background(wb); |
9ecf4866 | 1112 | wb_wakeup(wb); |
1da177e4 LT |
1113 | } |
1114 | ||
a66979ab DC |
1115 | /* |
1116 | * Remove the inode from the writeback list it is on. | |
1117 | */ | |
c7f54084 | 1118 | void inode_io_list_del(struct inode *inode) |
a66979ab | 1119 | { |
87e1d789 | 1120 | struct bdi_writeback *wb; |
f758eeab | 1121 | |
87e1d789 | 1122 | wb = inode_to_wb_and_lock_list(inode); |
c7f54084 | 1123 | inode_io_list_del_locked(inode, wb); |
52ebea74 | 1124 | spin_unlock(&wb->list_lock); |
a66979ab DC |
1125 | } |
1126 | ||
6c60d2b5 DC |
1127 | /* |
1128 | * mark an inode as under writeback on the sb | |
1129 | */ | |
1130 | void sb_mark_inode_writeback(struct inode *inode) | |
1131 | { | |
1132 | struct super_block *sb = inode->i_sb; | |
1133 | unsigned long flags; | |
1134 | ||
1135 | if (list_empty(&inode->i_wb_list)) { | |
1136 | spin_lock_irqsave(&sb->s_inode_wblist_lock, flags); | |
9a46b04f | 1137 | if (list_empty(&inode->i_wb_list)) { |
6c60d2b5 | 1138 | list_add_tail(&inode->i_wb_list, &sb->s_inodes_wb); |
9a46b04f BF |
1139 | trace_sb_mark_inode_writeback(inode); |
1140 | } | |
6c60d2b5 DC |
1141 | spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags); |
1142 | } | |
1143 | } | |
1144 | ||
1145 | /* | |
1146 | * clear an inode as under writeback on the sb | |
1147 | */ | |
1148 | void sb_clear_inode_writeback(struct inode *inode) | |
1149 | { | |
1150 | struct super_block *sb = inode->i_sb; | |
1151 | unsigned long flags; | |
1152 | ||
1153 | if (!list_empty(&inode->i_wb_list)) { | |
1154 | spin_lock_irqsave(&sb->s_inode_wblist_lock, flags); | |
9a46b04f BF |
1155 | if (!list_empty(&inode->i_wb_list)) { |
1156 | list_del_init(&inode->i_wb_list); | |
1157 | trace_sb_clear_inode_writeback(inode); | |
1158 | } | |
6c60d2b5 DC |
1159 | spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags); |
1160 | } | |
1161 | } | |
1162 | ||
6610a0bc AM |
1163 | /* |
1164 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
1165 | * furthest end of its superblock's dirty-inode list. | |
1166 | * | |
1167 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 1168 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
1169 | * the case then the inode must have been redirtied while it was being written |
1170 | * out and we don't reset its dirtied_when. | |
1171 | */ | |
f758eeab | 1172 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 1173 | { |
03ba3782 | 1174 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 1175 | struct inode *tail; |
6610a0bc | 1176 | |
7ccf19a8 | 1177 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 1178 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
1179 | inode->dirtied_when = jiffies; |
1180 | } | |
c7f54084 | 1181 | inode_io_list_move_locked(inode, wb, &wb->b_dirty); |
6610a0bc AM |
1182 | } |
1183 | ||
c986d1e2 | 1184 | /* |
66f3b8e2 | 1185 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 1186 | */ |
f758eeab | 1187 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 1188 | { |
c7f54084 | 1189 | inode_io_list_move_locked(inode, wb, &wb->b_more_io); |
c986d1e2 AM |
1190 | } |
1191 | ||
1c0eeaf5 JE |
1192 | static void inode_sync_complete(struct inode *inode) |
1193 | { | |
365b94ae | 1194 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
1195 | /* If inode is clean an unused, put it into LRU now... */ |
1196 | inode_add_lru(inode); | |
365b94ae | 1197 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
1198 | smp_mb(); |
1199 | wake_up_bit(&inode->i_state, __I_SYNC); | |
1200 | } | |
1201 | ||
d2caa3c5 JL |
1202 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
1203 | { | |
1204 | bool ret = time_after(inode->dirtied_when, t); | |
1205 | #ifndef CONFIG_64BIT | |
1206 | /* | |
1207 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
1208 | * It _appears_ to be in the future, but is actually in distant past. | |
1209 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 1210 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
1211 | */ |
1212 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
1213 | #endif | |
1214 | return ret; | |
1215 | } | |
1216 | ||
0ae45f63 TT |
1217 | #define EXPIRE_DIRTY_ATIME 0x0001 |
1218 | ||
2c136579 | 1219 | /* |
0e2f2b23 | 1220 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 1221 | * @delaying_queue to @dispatch_queue. |
2c136579 | 1222 | */ |
e84d0a4f | 1223 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 1224 | struct list_head *dispatch_queue, |
0ae45f63 | 1225 | int flags, |
ad4e38dd | 1226 | struct wb_writeback_work *work) |
2c136579 | 1227 | { |
0ae45f63 TT |
1228 | unsigned long *older_than_this = NULL; |
1229 | unsigned long expire_time; | |
5c03449d SL |
1230 | LIST_HEAD(tmp); |
1231 | struct list_head *pos, *node; | |
cf137307 | 1232 | struct super_block *sb = NULL; |
5c03449d | 1233 | struct inode *inode; |
cf137307 | 1234 | int do_sb_sort = 0; |
e84d0a4f | 1235 | int moved = 0; |
5c03449d | 1236 | |
0ae45f63 TT |
1237 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
1238 | older_than_this = work->older_than_this; | |
a2f48706 TT |
1239 | else if (!work->for_sync) { |
1240 | expire_time = jiffies - (dirtytime_expire_interval * HZ); | |
0ae45f63 TT |
1241 | older_than_this = &expire_time; |
1242 | } | |
2c136579 | 1243 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 1244 | inode = wb_inode(delaying_queue->prev); |
0ae45f63 TT |
1245 | if (older_than_this && |
1246 | inode_dirtied_after(inode, *older_than_this)) | |
2c136579 | 1247 | break; |
c7f54084 | 1248 | list_move(&inode->i_io_list, &tmp); |
a8855990 | 1249 | moved++; |
0ae45f63 TT |
1250 | if (flags & EXPIRE_DIRTY_ATIME) |
1251 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); | |
a8855990 JK |
1252 | if (sb_is_blkdev_sb(inode->i_sb)) |
1253 | continue; | |
cf137307 JA |
1254 | if (sb && sb != inode->i_sb) |
1255 | do_sb_sort = 1; | |
1256 | sb = inode->i_sb; | |
5c03449d SL |
1257 | } |
1258 | ||
cf137307 JA |
1259 | /* just one sb in list, splice to dispatch_queue and we're done */ |
1260 | if (!do_sb_sort) { | |
1261 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 1262 | goto out; |
cf137307 JA |
1263 | } |
1264 | ||
5c03449d SL |
1265 | /* Move inodes from one superblock together */ |
1266 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 1267 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 1268 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 1269 | inode = wb_inode(pos); |
5c03449d | 1270 | if (inode->i_sb == sb) |
c7f54084 | 1271 | list_move(&inode->i_io_list, dispatch_queue); |
5c03449d | 1272 | } |
2c136579 | 1273 | } |
e84d0a4f WF |
1274 | out: |
1275 | return moved; | |
2c136579 FW |
1276 | } |
1277 | ||
1278 | /* | |
1279 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
1280 | * Before |
1281 | * newly dirtied b_dirty b_io b_more_io | |
1282 | * =============> gf edc BA | |
1283 | * After | |
1284 | * newly dirtied b_dirty b_io b_more_io | |
1285 | * =============> g fBAedc | |
1286 | * | | |
1287 | * +--> dequeue for IO | |
2c136579 | 1288 | */ |
ad4e38dd | 1289 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 1290 | { |
e84d0a4f | 1291 | int moved; |
0ae45f63 | 1292 | |
f758eeab | 1293 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 1294 | list_splice_init(&wb->b_more_io, &wb->b_io); |
0ae45f63 TT |
1295 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
1296 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, | |
1297 | EXPIRE_DIRTY_ATIME, work); | |
d6c10f1f TH |
1298 | if (moved) |
1299 | wb_io_lists_populated(wb); | |
ad4e38dd | 1300 | trace_writeback_queue_io(wb, work, moved); |
66f3b8e2 JA |
1301 | } |
1302 | ||
a9185b41 | 1303 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 1304 | { |
9fb0a7da TH |
1305 | int ret; |
1306 | ||
1307 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
1308 | trace_writeback_write_inode_start(inode, wbc); | |
1309 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
1310 | trace_writeback_write_inode(inode, wbc); | |
1311 | return ret; | |
1312 | } | |
03ba3782 | 1313 | return 0; |
08d8e974 | 1314 | } |
08d8e974 | 1315 | |
1da177e4 | 1316 | /* |
169ebd90 JK |
1317 | * Wait for writeback on an inode to complete. Called with i_lock held. |
1318 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 1319 | */ |
169ebd90 JK |
1320 | static void __inode_wait_for_writeback(struct inode *inode) |
1321 | __releases(inode->i_lock) | |
1322 | __acquires(inode->i_lock) | |
01c03194 CH |
1323 | { |
1324 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
1325 | wait_queue_head_t *wqh; | |
1326 | ||
1327 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
1328 | while (inode->i_state & I_SYNC) { |
1329 | spin_unlock(&inode->i_lock); | |
74316201 N |
1330 | __wait_on_bit(wqh, &wq, bit_wait, |
1331 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 1332 | spin_lock(&inode->i_lock); |
58a9d3d8 | 1333 | } |
01c03194 CH |
1334 | } |
1335 | ||
169ebd90 JK |
1336 | /* |
1337 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
1338 | */ | |
1339 | void inode_wait_for_writeback(struct inode *inode) | |
1340 | { | |
1341 | spin_lock(&inode->i_lock); | |
1342 | __inode_wait_for_writeback(inode); | |
1343 | spin_unlock(&inode->i_lock); | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
1348 | * held and drops it. It is aimed for callers not holding any inode reference | |
1349 | * so once i_lock is dropped, inode can go away. | |
1350 | */ | |
1351 | static void inode_sleep_on_writeback(struct inode *inode) | |
1352 | __releases(inode->i_lock) | |
1353 | { | |
1354 | DEFINE_WAIT(wait); | |
1355 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
1356 | int sleep; | |
1357 | ||
1358 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1359 | sleep = inode->i_state & I_SYNC; | |
1360 | spin_unlock(&inode->i_lock); | |
1361 | if (sleep) | |
1362 | schedule(); | |
1363 | finish_wait(wqh, &wait); | |
1364 | } | |
1365 | ||
ccb26b5a JK |
1366 | /* |
1367 | * Find proper writeback list for the inode depending on its current state and | |
1368 | * possibly also change of its state while we were doing writeback. Here we | |
1369 | * handle things such as livelock prevention or fairness of writeback among | |
1370 | * inodes. This function can be called only by flusher thread - noone else | |
1371 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
1372 | * thread's back can have unexpected consequences. | |
1373 | */ | |
1374 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
1375 | struct writeback_control *wbc) | |
1376 | { | |
1377 | if (inode->i_state & I_FREEING) | |
1378 | return; | |
1379 | ||
1380 | /* | |
1381 | * Sync livelock prevention. Each inode is tagged and synced in one | |
1382 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
1383 | * the dirty time to prevent enqueue and sync it again. | |
1384 | */ | |
1385 | if ((inode->i_state & I_DIRTY) && | |
1386 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
1387 | inode->dirtied_when = jiffies; | |
1388 | ||
4f8ad655 JK |
1389 | if (wbc->pages_skipped) { |
1390 | /* | |
1391 | * writeback is not making progress due to locked | |
1392 | * buffers. Skip this inode for now. | |
1393 | */ | |
1394 | redirty_tail(inode, wb); | |
1395 | return; | |
1396 | } | |
1397 | ||
ccb26b5a JK |
1398 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
1399 | /* | |
1400 | * We didn't write back all the pages. nfs_writepages() | |
1401 | * sometimes bales out without doing anything. | |
1402 | */ | |
1403 | if (wbc->nr_to_write <= 0) { | |
1404 | /* Slice used up. Queue for next turn. */ | |
1405 | requeue_io(inode, wb); | |
1406 | } else { | |
1407 | /* | |
1408 | * Writeback blocked by something other than | |
1409 | * congestion. Delay the inode for some time to | |
1410 | * avoid spinning on the CPU (100% iowait) | |
1411 | * retrying writeback of the dirty page/inode | |
1412 | * that cannot be performed immediately. | |
1413 | */ | |
1414 | redirty_tail(inode, wb); | |
1415 | } | |
1416 | } else if (inode->i_state & I_DIRTY) { | |
1417 | /* | |
1418 | * Filesystems can dirty the inode during writeback operations, | |
1419 | * such as delayed allocation during submission or metadata | |
1420 | * updates after data IO completion. | |
1421 | */ | |
1422 | redirty_tail(inode, wb); | |
0ae45f63 | 1423 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 1424 | inode->dirtied_when = jiffies; |
c7f54084 | 1425 | inode_io_list_move_locked(inode, wb, &wb->b_dirty_time); |
ccb26b5a JK |
1426 | } else { |
1427 | /* The inode is clean. Remove from writeback lists. */ | |
c7f54084 | 1428 | inode_io_list_del_locked(inode, wb); |
ccb26b5a JK |
1429 | } |
1430 | } | |
1431 | ||
01c03194 | 1432 | /* |
4f8ad655 JK |
1433 | * Write out an inode and its dirty pages. Do not update the writeback list |
1434 | * linkage. That is left to the caller. The caller is also responsible for | |
1435 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
1436 | */ |
1437 | static int | |
cd8ed2a4 | 1438 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 1439 | { |
1da177e4 | 1440 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 1441 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 1442 | unsigned dirty; |
1da177e4 LT |
1443 | int ret; |
1444 | ||
4f8ad655 | 1445 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 1446 | |
9fb0a7da TH |
1447 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
1448 | ||
1da177e4 LT |
1449 | ret = do_writepages(mapping, wbc); |
1450 | ||
26821ed4 CH |
1451 | /* |
1452 | * Make sure to wait on the data before writing out the metadata. | |
1453 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
1454 | * I/O completion. We don't do it for sync(2) writeback because it has a |
1455 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
1456 | * inode metadata is written back correctly. | |
26821ed4 | 1457 | */ |
7747bd4b | 1458 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 1459 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
1460 | if (ret == 0) |
1461 | ret = err; | |
1462 | } | |
1463 | ||
5547e8aa DM |
1464 | /* |
1465 | * Some filesystems may redirty the inode during the writeback | |
1466 | * due to delalloc, clear dirty metadata flags right before | |
1467 | * write_inode() | |
1468 | */ | |
250df6ed | 1469 | spin_lock(&inode->i_lock); |
9c6ac78e | 1470 | |
5547e8aa | 1471 | dirty = inode->i_state & I_DIRTY; |
a2f48706 | 1472 | if (inode->i_state & I_DIRTY_TIME) { |
0e11f644 | 1473 | if ((dirty & I_DIRTY_INODE) || |
dc5ff2b1 | 1474 | wbc->sync_mode == WB_SYNC_ALL || |
a2f48706 TT |
1475 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || |
1476 | unlikely(time_after(jiffies, | |
1477 | (inode->dirtied_time_when + | |
1478 | dirtytime_expire_interval * HZ)))) { | |
1479 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; | |
1480 | trace_writeback_lazytime(inode); | |
1481 | } | |
1482 | } else | |
1483 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; | |
0ae45f63 | 1484 | inode->i_state &= ~dirty; |
9c6ac78e TH |
1485 | |
1486 | /* | |
1487 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
1488 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
1489 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
1490 | * inode. | |
1491 | * | |
1492 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
1493 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
1494 | * necessary. This guarantees that either __mark_inode_dirty() | |
1495 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
1496 | */ | |
1497 | smp_mb(); | |
1498 | ||
1499 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
1500 | inode->i_state |= I_DIRTY_PAGES; | |
1501 | ||
250df6ed | 1502 | spin_unlock(&inode->i_lock); |
9c6ac78e | 1503 | |
0ae45f63 TT |
1504 | if (dirty & I_DIRTY_TIME) |
1505 | mark_inode_dirty_sync(inode); | |
26821ed4 | 1506 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 1507 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 1508 | int err = write_inode(inode, wbc); |
1da177e4 LT |
1509 | if (ret == 0) |
1510 | ret = err; | |
1511 | } | |
4f8ad655 JK |
1512 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
1513 | return ret; | |
1514 | } | |
1515 | ||
1516 | /* | |
1517 | * Write out an inode's dirty pages. Either the caller has an active reference | |
1518 | * on the inode or the inode has I_WILL_FREE set. | |
1519 | * | |
1520 | * This function is designed to be called for writing back one inode which | |
1521 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
1522 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
1523 | */ | |
aaf25593 TH |
1524 | static int writeback_single_inode(struct inode *inode, |
1525 | struct writeback_control *wbc) | |
4f8ad655 | 1526 | { |
aaf25593 | 1527 | struct bdi_writeback *wb; |
4f8ad655 JK |
1528 | int ret = 0; |
1529 | ||
1530 | spin_lock(&inode->i_lock); | |
1531 | if (!atomic_read(&inode->i_count)) | |
1532 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
1533 | else | |
1534 | WARN_ON(inode->i_state & I_WILL_FREE); | |
1535 | ||
1536 | if (inode->i_state & I_SYNC) { | |
1537 | if (wbc->sync_mode != WB_SYNC_ALL) | |
1538 | goto out; | |
1539 | /* | |
169ebd90 JK |
1540 | * It's a data-integrity sync. We must wait. Since callers hold |
1541 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
1542 | * away under us. | |
4f8ad655 | 1543 | */ |
169ebd90 | 1544 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
1545 | } |
1546 | WARN_ON(inode->i_state & I_SYNC); | |
1547 | /* | |
f9b0e058 JK |
1548 | * Skip inode if it is clean and we have no outstanding writeback in |
1549 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
1550 | * function since flusher thread may be doing for example sync in | |
1551 | * parallel and if we move the inode, it could get skipped. So here we | |
1552 | * make sure inode is on some writeback list and leave it there unless | |
1553 | * we have completely cleaned the inode. | |
4f8ad655 | 1554 | */ |
0ae45f63 | 1555 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
1556 | (wbc->sync_mode != WB_SYNC_ALL || |
1557 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
1558 | goto out; |
1559 | inode->i_state |= I_SYNC; | |
b16b1deb | 1560 | wbc_attach_and_unlock_inode(wbc, inode); |
4f8ad655 | 1561 | |
cd8ed2a4 | 1562 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 1563 | |
b16b1deb | 1564 | wbc_detach_inode(wbc); |
aaf25593 TH |
1565 | |
1566 | wb = inode_to_wb_and_lock_list(inode); | |
250df6ed | 1567 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
1568 | /* |
1569 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
1570 | * touch it. See comment above for explanation. | |
1571 | */ | |
0ae45f63 | 1572 | if (!(inode->i_state & I_DIRTY_ALL)) |
c7f54084 | 1573 | inode_io_list_del_locked(inode, wb); |
4f8ad655 | 1574 | spin_unlock(&wb->list_lock); |
1c0eeaf5 | 1575 | inode_sync_complete(inode); |
4f8ad655 JK |
1576 | out: |
1577 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1578 | return ret; |
1579 | } | |
1580 | ||
a88a341a | 1581 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 1582 | struct wb_writeback_work *work) |
d46db3d5 WF |
1583 | { |
1584 | long pages; | |
1585 | ||
1586 | /* | |
1587 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
1588 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
1589 | * here avoids calling into writeback_inodes_wb() more than once. | |
1590 | * | |
1591 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
1592 | * | |
1593 | * wb_writeback() | |
1594 | * writeback_sb_inodes() <== called only once | |
1595 | * write_cache_pages() <== called once for each inode | |
1596 | * (quickly) tag currently dirty pages | |
1597 | * (maybe slowly) sync all tagged pages | |
1598 | */ | |
1599 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
1600 | pages = LONG_MAX; | |
1a12d8bd | 1601 | else { |
a88a341a | 1602 | pages = min(wb->avg_write_bandwidth / 2, |
dcc25ae7 | 1603 | global_wb_domain.dirty_limit / DIRTY_SCOPE); |
1a12d8bd WF |
1604 | pages = min(pages, work->nr_pages); |
1605 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
1606 | MIN_WRITEBACK_PAGES); | |
1607 | } | |
d46db3d5 WF |
1608 | |
1609 | return pages; | |
1610 | } | |
1611 | ||
f11c9c5c ES |
1612 | /* |
1613 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 1614 | * |
d46db3d5 | 1615 | * Return the number of pages and/or inodes written. |
0ba13fd1 LT |
1616 | * |
1617 | * NOTE! This is called with wb->list_lock held, and will | |
1618 | * unlock and relock that for each inode it ends up doing | |
1619 | * IO for. | |
f11c9c5c | 1620 | */ |
d46db3d5 WF |
1621 | static long writeback_sb_inodes(struct super_block *sb, |
1622 | struct bdi_writeback *wb, | |
1623 | struct wb_writeback_work *work) | |
1da177e4 | 1624 | { |
d46db3d5 WF |
1625 | struct writeback_control wbc = { |
1626 | .sync_mode = work->sync_mode, | |
1627 | .tagged_writepages = work->tagged_writepages, | |
1628 | .for_kupdate = work->for_kupdate, | |
1629 | .for_background = work->for_background, | |
7747bd4b | 1630 | .for_sync = work->for_sync, |
d46db3d5 WF |
1631 | .range_cyclic = work->range_cyclic, |
1632 | .range_start = 0, | |
1633 | .range_end = LLONG_MAX, | |
1634 | }; | |
1635 | unsigned long start_time = jiffies; | |
1636 | long write_chunk; | |
1637 | long wrote = 0; /* count both pages and inodes */ | |
1638 | ||
03ba3782 | 1639 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 1640 | struct inode *inode = wb_inode(wb->b_io.prev); |
aaf25593 | 1641 | struct bdi_writeback *tmp_wb; |
edadfb10 CH |
1642 | |
1643 | if (inode->i_sb != sb) { | |
d46db3d5 | 1644 | if (work->sb) { |
edadfb10 CH |
1645 | /* |
1646 | * We only want to write back data for this | |
1647 | * superblock, move all inodes not belonging | |
1648 | * to it back onto the dirty list. | |
1649 | */ | |
f758eeab | 1650 | redirty_tail(inode, wb); |
edadfb10 CH |
1651 | continue; |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * The inode belongs to a different superblock. | |
1656 | * Bounce back to the caller to unpin this and | |
1657 | * pin the next superblock. | |
1658 | */ | |
d46db3d5 | 1659 | break; |
edadfb10 CH |
1660 | } |
1661 | ||
9843b76a | 1662 | /* |
331cbdee WL |
1663 | * Don't bother with new inodes or inodes being freed, first |
1664 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
1665 | * kind writeout is handled by the freer. |
1666 | */ | |
250df6ed | 1667 | spin_lock(&inode->i_lock); |
9843b76a | 1668 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 1669 | spin_unlock(&inode->i_lock); |
fcc5c222 | 1670 | redirty_tail(inode, wb); |
7ef0d737 NP |
1671 | continue; |
1672 | } | |
cc1676d9 JK |
1673 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
1674 | /* | |
1675 | * If this inode is locked for writeback and we are not | |
1676 | * doing writeback-for-data-integrity, move it to | |
1677 | * b_more_io so that writeback can proceed with the | |
1678 | * other inodes on s_io. | |
1679 | * | |
1680 | * We'll have another go at writing back this inode | |
1681 | * when we completed a full scan of b_io. | |
1682 | */ | |
1683 | spin_unlock(&inode->i_lock); | |
1684 | requeue_io(inode, wb); | |
1685 | trace_writeback_sb_inodes_requeue(inode); | |
1686 | continue; | |
1687 | } | |
f0d07b7f JK |
1688 | spin_unlock(&wb->list_lock); |
1689 | ||
4f8ad655 JK |
1690 | /* |
1691 | * We already requeued the inode if it had I_SYNC set and we | |
1692 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
1693 | * WB_SYNC_ALL case. | |
1694 | */ | |
169ebd90 JK |
1695 | if (inode->i_state & I_SYNC) { |
1696 | /* Wait for I_SYNC. This function drops i_lock... */ | |
1697 | inode_sleep_on_writeback(inode); | |
1698 | /* Inode may be gone, start again */ | |
ead188f9 | 1699 | spin_lock(&wb->list_lock); |
169ebd90 JK |
1700 | continue; |
1701 | } | |
4f8ad655 | 1702 | inode->i_state |= I_SYNC; |
b16b1deb | 1703 | wbc_attach_and_unlock_inode(&wbc, inode); |
169ebd90 | 1704 | |
a88a341a | 1705 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
1706 | wbc.nr_to_write = write_chunk; |
1707 | wbc.pages_skipped = 0; | |
250df6ed | 1708 | |
169ebd90 JK |
1709 | /* |
1710 | * We use I_SYNC to pin the inode in memory. While it is set | |
1711 | * evict_inode() will wait so the inode cannot be freed. | |
1712 | */ | |
cd8ed2a4 | 1713 | __writeback_single_inode(inode, &wbc); |
250df6ed | 1714 | |
b16b1deb | 1715 | wbc_detach_inode(&wbc); |
d46db3d5 WF |
1716 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
1717 | wrote += write_chunk - wbc.nr_to_write; | |
590dca3a CM |
1718 | |
1719 | if (need_resched()) { | |
1720 | /* | |
1721 | * We're trying to balance between building up a nice | |
1722 | * long list of IOs to improve our merge rate, and | |
1723 | * getting those IOs out quickly for anyone throttling | |
1724 | * in balance_dirty_pages(). cond_resched() doesn't | |
1725 | * unplug, so get our IOs out the door before we | |
1726 | * give up the CPU. | |
1727 | */ | |
1728 | blk_flush_plug(current); | |
1729 | cond_resched(); | |
1730 | } | |
1731 | ||
aaf25593 TH |
1732 | /* |
1733 | * Requeue @inode if still dirty. Be careful as @inode may | |
1734 | * have been switched to another wb in the meantime. | |
1735 | */ | |
1736 | tmp_wb = inode_to_wb_and_lock_list(inode); | |
4f8ad655 | 1737 | spin_lock(&inode->i_lock); |
0ae45f63 | 1738 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 1739 | wrote++; |
aaf25593 | 1740 | requeue_inode(inode, tmp_wb, &wbc); |
4f8ad655 | 1741 | inode_sync_complete(inode); |
0f1b1fd8 | 1742 | spin_unlock(&inode->i_lock); |
590dca3a | 1743 | |
aaf25593 TH |
1744 | if (unlikely(tmp_wb != wb)) { |
1745 | spin_unlock(&tmp_wb->list_lock); | |
1746 | spin_lock(&wb->list_lock); | |
1747 | } | |
1748 | ||
d46db3d5 WF |
1749 | /* |
1750 | * bail out to wb_writeback() often enough to check | |
1751 | * background threshold and other termination conditions. | |
1752 | */ | |
1753 | if (wrote) { | |
1754 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1755 | break; | |
1756 | if (work->nr_pages <= 0) | |
1757 | break; | |
8bc3be27 | 1758 | } |
1da177e4 | 1759 | } |
d46db3d5 | 1760 | return wrote; |
f11c9c5c ES |
1761 | } |
1762 | ||
d46db3d5 WF |
1763 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
1764 | struct wb_writeback_work *work) | |
f11c9c5c | 1765 | { |
d46db3d5 WF |
1766 | unsigned long start_time = jiffies; |
1767 | long wrote = 0; | |
38f21977 | 1768 | |
f11c9c5c | 1769 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 1770 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 1771 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 1772 | |
eb6ef3df | 1773 | if (!trylock_super(sb)) { |
0e995816 | 1774 | /* |
eb6ef3df | 1775 | * trylock_super() may fail consistently due to |
0e995816 WF |
1776 | * s_umount being grabbed by someone else. Don't use |
1777 | * requeue_io() to avoid busy retrying the inode/sb. | |
1778 | */ | |
1779 | redirty_tail(inode, wb); | |
edadfb10 | 1780 | continue; |
f11c9c5c | 1781 | } |
d46db3d5 | 1782 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 1783 | up_read(&sb->s_umount); |
f11c9c5c | 1784 | |
d46db3d5 WF |
1785 | /* refer to the same tests at the end of writeback_sb_inodes */ |
1786 | if (wrote) { | |
1787 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1788 | break; | |
1789 | if (work->nr_pages <= 0) | |
1790 | break; | |
1791 | } | |
f11c9c5c | 1792 | } |
66f3b8e2 | 1793 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 1794 | return wrote; |
66f3b8e2 JA |
1795 | } |
1796 | ||
7d9f073b | 1797 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 1798 | enum wb_reason reason) |
edadfb10 | 1799 | { |
d46db3d5 WF |
1800 | struct wb_writeback_work work = { |
1801 | .nr_pages = nr_pages, | |
1802 | .sync_mode = WB_SYNC_NONE, | |
1803 | .range_cyclic = 1, | |
0e175a18 | 1804 | .reason = reason, |
d46db3d5 | 1805 | }; |
505a666e | 1806 | struct blk_plug plug; |
edadfb10 | 1807 | |
505a666e | 1808 | blk_start_plug(&plug); |
f758eeab | 1809 | spin_lock(&wb->list_lock); |
424b351f | 1810 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1811 | queue_io(wb, &work); |
d46db3d5 | 1812 | __writeback_inodes_wb(wb, &work); |
f758eeab | 1813 | spin_unlock(&wb->list_lock); |
505a666e | 1814 | blk_finish_plug(&plug); |
edadfb10 | 1815 | |
d46db3d5 WF |
1816 | return nr_pages - work.nr_pages; |
1817 | } | |
03ba3782 | 1818 | |
03ba3782 JA |
1819 | /* |
1820 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 1821 | * |
03ba3782 JA |
1822 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
1823 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
1824 | * just walks the superblock inode list, writing back any inodes which are | |
1825 | * older than a specific point in time. | |
66f3b8e2 | 1826 | * |
03ba3782 JA |
1827 | * Try to run once per dirty_writeback_interval. But if a writeback event |
1828 | * takes longer than a dirty_writeback_interval interval, then leave a | |
1829 | * one-second gap. | |
66f3b8e2 | 1830 | * |
03ba3782 JA |
1831 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
1832 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 1833 | */ |
c4a77a6c | 1834 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 1835 | struct wb_writeback_work *work) |
66f3b8e2 | 1836 | { |
e98be2d5 | 1837 | unsigned long wb_start = jiffies; |
d46db3d5 | 1838 | long nr_pages = work->nr_pages; |
0dc83bd3 | 1839 | unsigned long oldest_jif; |
a5989bdc | 1840 | struct inode *inode; |
d46db3d5 | 1841 | long progress; |
505a666e | 1842 | struct blk_plug plug; |
66f3b8e2 | 1843 | |
0dc83bd3 JK |
1844 | oldest_jif = jiffies; |
1845 | work->older_than_this = &oldest_jif; | |
38f21977 | 1846 | |
505a666e | 1847 | blk_start_plug(&plug); |
e8dfc305 | 1848 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1849 | for (;;) { |
1850 | /* | |
d3ddec76 | 1851 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 1852 | */ |
83ba7b07 | 1853 | if (work->nr_pages <= 0) |
03ba3782 | 1854 | break; |
66f3b8e2 | 1855 | |
aa373cf5 JK |
1856 | /* |
1857 | * Background writeout and kupdate-style writeback may | |
1858 | * run forever. Stop them if there is other work to do | |
1859 | * so that e.g. sync can proceed. They'll be restarted | |
1860 | * after the other works are all done. | |
1861 | */ | |
1862 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 1863 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
1864 | break; |
1865 | ||
38f21977 | 1866 | /* |
d3ddec76 WF |
1867 | * For background writeout, stop when we are below the |
1868 | * background dirty threshold | |
38f21977 | 1869 | */ |
aa661bbe | 1870 | if (work->for_background && !wb_over_bg_thresh(wb)) |
03ba3782 | 1871 | break; |
38f21977 | 1872 | |
1bc36b64 JK |
1873 | /* |
1874 | * Kupdate and background works are special and we want to | |
1875 | * include all inodes that need writing. Livelock avoidance is | |
1876 | * handled by these works yielding to any other work so we are | |
1877 | * safe. | |
1878 | */ | |
ba9aa839 | 1879 | if (work->for_kupdate) { |
0dc83bd3 | 1880 | oldest_jif = jiffies - |
ba9aa839 | 1881 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 1882 | } else if (work->for_background) |
0dc83bd3 | 1883 | oldest_jif = jiffies; |
028c2dd1 | 1884 | |
5634cc2a | 1885 | trace_writeback_start(wb, work); |
e8dfc305 | 1886 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1887 | queue_io(wb, work); |
83ba7b07 | 1888 | if (work->sb) |
d46db3d5 | 1889 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 1890 | else |
d46db3d5 | 1891 | progress = __writeback_inodes_wb(wb, work); |
5634cc2a | 1892 | trace_writeback_written(wb, work); |
028c2dd1 | 1893 | |
e98be2d5 | 1894 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
1895 | |
1896 | /* | |
e6fb6da2 WF |
1897 | * Did we write something? Try for more |
1898 | * | |
1899 | * Dirty inodes are moved to b_io for writeback in batches. | |
1900 | * The completion of the current batch does not necessarily | |
1901 | * mean the overall work is done. So we keep looping as long | |
1902 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 1903 | */ |
d46db3d5 | 1904 | if (progress) |
71fd05a8 JA |
1905 | continue; |
1906 | /* | |
e6fb6da2 | 1907 | * No more inodes for IO, bail |
71fd05a8 | 1908 | */ |
b7a2441f | 1909 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 1910 | break; |
71fd05a8 JA |
1911 | /* |
1912 | * Nothing written. Wait for some inode to | |
1913 | * become available for writeback. Otherwise | |
1914 | * we'll just busyloop. | |
1915 | */ | |
bace9248 TE |
1916 | trace_writeback_wait(wb, work); |
1917 | inode = wb_inode(wb->b_more_io.prev); | |
1918 | spin_lock(&inode->i_lock); | |
1919 | spin_unlock(&wb->list_lock); | |
1920 | /* This function drops i_lock... */ | |
1921 | inode_sleep_on_writeback(inode); | |
1922 | spin_lock(&wb->list_lock); | |
03ba3782 | 1923 | } |
e8dfc305 | 1924 | spin_unlock(&wb->list_lock); |
505a666e | 1925 | blk_finish_plug(&plug); |
03ba3782 | 1926 | |
d46db3d5 | 1927 | return nr_pages - work->nr_pages; |
03ba3782 JA |
1928 | } |
1929 | ||
1930 | /* | |
83ba7b07 | 1931 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 1932 | */ |
f0054bb1 | 1933 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 1934 | { |
83ba7b07 | 1935 | struct wb_writeback_work *work = NULL; |
03ba3782 | 1936 | |
f0054bb1 TH |
1937 | spin_lock_bh(&wb->work_lock); |
1938 | if (!list_empty(&wb->work_list)) { | |
1939 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
1940 | struct wb_writeback_work, list); |
1941 | list_del_init(&work->list); | |
03ba3782 | 1942 | } |
f0054bb1 | 1943 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 1944 | return work; |
03ba3782 JA |
1945 | } |
1946 | ||
6585027a JK |
1947 | static long wb_check_background_flush(struct bdi_writeback *wb) |
1948 | { | |
aa661bbe | 1949 | if (wb_over_bg_thresh(wb)) { |
6585027a JK |
1950 | |
1951 | struct wb_writeback_work work = { | |
1952 | .nr_pages = LONG_MAX, | |
1953 | .sync_mode = WB_SYNC_NONE, | |
1954 | .for_background = 1, | |
1955 | .range_cyclic = 1, | |
0e175a18 | 1956 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
1957 | }; |
1958 | ||
1959 | return wb_writeback(wb, &work); | |
1960 | } | |
1961 | ||
1962 | return 0; | |
1963 | } | |
1964 | ||
03ba3782 JA |
1965 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
1966 | { | |
1967 | unsigned long expired; | |
1968 | long nr_pages; | |
1969 | ||
69b62d01 JA |
1970 | /* |
1971 | * When set to zero, disable periodic writeback | |
1972 | */ | |
1973 | if (!dirty_writeback_interval) | |
1974 | return 0; | |
1975 | ||
03ba3782 JA |
1976 | expired = wb->last_old_flush + |
1977 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
1978 | if (time_before(jiffies, expired)) | |
1979 | return 0; | |
1980 | ||
1981 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 1982 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 1983 | |
c4a77a6c | 1984 | if (nr_pages) { |
83ba7b07 | 1985 | struct wb_writeback_work work = { |
c4a77a6c JA |
1986 | .nr_pages = nr_pages, |
1987 | .sync_mode = WB_SYNC_NONE, | |
1988 | .for_kupdate = 1, | |
1989 | .range_cyclic = 1, | |
0e175a18 | 1990 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
1991 | }; |
1992 | ||
83ba7b07 | 1993 | return wb_writeback(wb, &work); |
c4a77a6c | 1994 | } |
03ba3782 JA |
1995 | |
1996 | return 0; | |
1997 | } | |
1998 | ||
85009b4f JA |
1999 | static long wb_check_start_all(struct bdi_writeback *wb) |
2000 | { | |
2001 | long nr_pages; | |
2002 | ||
2003 | if (!test_bit(WB_start_all, &wb->state)) | |
2004 | return 0; | |
2005 | ||
2006 | nr_pages = get_nr_dirty_pages(); | |
2007 | if (nr_pages) { | |
2008 | struct wb_writeback_work work = { | |
2009 | .nr_pages = wb_split_bdi_pages(wb, nr_pages), | |
2010 | .sync_mode = WB_SYNC_NONE, | |
2011 | .range_cyclic = 1, | |
2012 | .reason = wb->start_all_reason, | |
2013 | }; | |
2014 | ||
2015 | nr_pages = wb_writeback(wb, &work); | |
2016 | } | |
2017 | ||
2018 | clear_bit(WB_start_all, &wb->state); | |
2019 | return nr_pages; | |
2020 | } | |
2021 | ||
2022 | ||
03ba3782 JA |
2023 | /* |
2024 | * Retrieve work items and do the writeback they describe | |
2025 | */ | |
25d130ba | 2026 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 2027 | { |
83ba7b07 | 2028 | struct wb_writeback_work *work; |
c4a77a6c | 2029 | long wrote = 0; |
03ba3782 | 2030 | |
4452226e | 2031 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 2032 | while ((work = get_next_work_item(wb)) != NULL) { |
5634cc2a | 2033 | trace_writeback_exec(wb, work); |
83ba7b07 | 2034 | wrote += wb_writeback(wb, work); |
4a3a485b | 2035 | finish_writeback_work(wb, work); |
03ba3782 JA |
2036 | } |
2037 | ||
85009b4f JA |
2038 | /* |
2039 | * Check for a flush-everything request | |
2040 | */ | |
2041 | wrote += wb_check_start_all(wb); | |
2042 | ||
03ba3782 JA |
2043 | /* |
2044 | * Check for periodic writeback, kupdated() style | |
2045 | */ | |
2046 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 2047 | wrote += wb_check_background_flush(wb); |
4452226e | 2048 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
2049 | |
2050 | return wrote; | |
2051 | } | |
2052 | ||
2053 | /* | |
2054 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 2055 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 2056 | */ |
f0054bb1 | 2057 | void wb_workfn(struct work_struct *work) |
03ba3782 | 2058 | { |
839a8e86 TH |
2059 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
2060 | struct bdi_writeback, dwork); | |
03ba3782 JA |
2061 | long pages_written; |
2062 | ||
f0054bb1 | 2063 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
766f9164 | 2064 | current->flags |= PF_SWAPWRITE; |
455b2864 | 2065 | |
839a8e86 | 2066 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 2067 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 2068 | /* |
f0054bb1 | 2069 | * The normal path. Keep writing back @wb until its |
839a8e86 | 2070 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 2071 | * if @wb is shutting down even when we're running off the |
839a8e86 | 2072 | * rescuer as work_list needs to be drained. |
6467716a | 2073 | */ |
839a8e86 | 2074 | do { |
25d130ba | 2075 | pages_written = wb_do_writeback(wb); |
839a8e86 | 2076 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 2077 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
2078 | } else { |
2079 | /* | |
2080 | * bdi_wq can't get enough workers and we're running off | |
2081 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
2082 | * enough for efficient IO. | |
2083 | */ | |
f0054bb1 | 2084 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 2085 | WB_REASON_FORKER_THREAD); |
455b2864 | 2086 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
2087 | } |
2088 | ||
f0054bb1 | 2089 | if (!list_empty(&wb->work_list)) |
b8b78495 | 2090 | wb_wakeup(wb); |
6ca738d6 | 2091 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
f0054bb1 | 2092 | wb_wakeup_delayed(wb); |
455b2864 | 2093 | |
839a8e86 | 2094 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
2095 | } |
2096 | ||
595043e5 JA |
2097 | /* |
2098 | * Start writeback of `nr_pages' pages on this bdi. If `nr_pages' is zero, | |
2099 | * write back the whole world. | |
2100 | */ | |
2101 | static void __wakeup_flusher_threads_bdi(struct backing_dev_info *bdi, | |
e8e8a0c6 | 2102 | enum wb_reason reason) |
595043e5 JA |
2103 | { |
2104 | struct bdi_writeback *wb; | |
2105 | ||
2106 | if (!bdi_has_dirty_io(bdi)) | |
2107 | return; | |
2108 | ||
2109 | list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) | |
e8e8a0c6 | 2110 | wb_start_writeback(wb, reason); |
595043e5 JA |
2111 | } |
2112 | ||
2113 | void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi, | |
2114 | enum wb_reason reason) | |
2115 | { | |
595043e5 | 2116 | rcu_read_lock(); |
e8e8a0c6 | 2117 | __wakeup_flusher_threads_bdi(bdi, reason); |
595043e5 JA |
2118 | rcu_read_unlock(); |
2119 | } | |
2120 | ||
03ba3782 | 2121 | /* |
9ba4b2df | 2122 | * Wakeup the flusher threads to start writeback of all currently dirty pages |
03ba3782 | 2123 | */ |
9ba4b2df | 2124 | void wakeup_flusher_threads(enum wb_reason reason) |
03ba3782 | 2125 | { |
b8c2f347 | 2126 | struct backing_dev_info *bdi; |
03ba3782 | 2127 | |
51350ea0 KK |
2128 | /* |
2129 | * If we are expecting writeback progress we must submit plugged IO. | |
2130 | */ | |
2131 | if (blk_needs_flush_plug(current)) | |
2132 | blk_schedule_flush_plug(current); | |
2133 | ||
b8c2f347 | 2134 | rcu_read_lock(); |
595043e5 | 2135 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) |
e8e8a0c6 | 2136 | __wakeup_flusher_threads_bdi(bdi, reason); |
cfc4ba53 | 2137 | rcu_read_unlock(); |
1da177e4 LT |
2138 | } |
2139 | ||
a2f48706 TT |
2140 | /* |
2141 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
2142 | * written back periodically. We deliberately do *not* check the | |
2143 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
2144 | * kernel to be constantly waking up once there are any dirtytime | |
2145 | * inodes on the system. So instead we define a separate delayed work | |
2146 | * function which gets called much more rarely. (By default, only | |
2147 | * once every 12 hours.) | |
2148 | * | |
2149 | * If there is any other write activity going on in the file system, | |
2150 | * this function won't be necessary. But if the only thing that has | |
2151 | * happened on the file system is a dirtytime inode caused by an atime | |
2152 | * update, we need this infrastructure below to make sure that inode | |
2153 | * eventually gets pushed out to disk. | |
2154 | */ | |
2155 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
2156 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
2157 | ||
2158 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
2159 | { | |
2160 | struct backing_dev_info *bdi; | |
2161 | ||
2162 | rcu_read_lock(); | |
2163 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
001fe6f6 | 2164 | struct bdi_writeback *wb; |
001fe6f6 | 2165 | |
b817525a | 2166 | list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) |
6fdf860f TH |
2167 | if (!list_empty(&wb->b_dirty_time)) |
2168 | wb_wakeup(wb); | |
a2f48706 TT |
2169 | } |
2170 | rcu_read_unlock(); | |
2171 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
2172 | } | |
2173 | ||
2174 | static int __init start_dirtytime_writeback(void) | |
2175 | { | |
2176 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
2177 | return 0; | |
2178 | } | |
2179 | __initcall(start_dirtytime_writeback); | |
2180 | ||
1efff914 TT |
2181 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
2182 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
2183 | { | |
2184 | int ret; | |
2185 | ||
2186 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
2187 | if (ret == 0 && write) | |
2188 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
2189 | return ret; | |
2190 | } | |
2191 | ||
03ba3782 JA |
2192 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
2193 | { | |
2194 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
2195 | struct dentry *dentry; | |
2196 | const char *name = "?"; | |
2197 | ||
2198 | dentry = d_find_alias(inode); | |
2199 | if (dentry) { | |
2200 | spin_lock(&dentry->d_lock); | |
2201 | name = (const char *) dentry->d_name.name; | |
2202 | } | |
2203 | printk(KERN_DEBUG | |
2204 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
2205 | current->comm, task_pid_nr(current), inode->i_ino, | |
2206 | name, inode->i_sb->s_id); | |
2207 | if (dentry) { | |
2208 | spin_unlock(&dentry->d_lock); | |
2209 | dput(dentry); | |
2210 | } | |
2211 | } | |
2212 | } | |
2213 | ||
2214 | /** | |
0117d427 MCC |
2215 | * __mark_inode_dirty - internal function |
2216 | * | |
2217 | * @inode: inode to mark | |
2218 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
2219 | * | |
2220 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
2221 | * mark_inode_dirty_sync. | |
1da177e4 | 2222 | * |
03ba3782 JA |
2223 | * Put the inode on the super block's dirty list. |
2224 | * | |
2225 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
2226 | * dirty list only if it is hashed or if it refers to a blockdev. | |
2227 | * If it was not hashed, it will never be added to the dirty list | |
2228 | * even if it is later hashed, as it will have been marked dirty already. | |
2229 | * | |
2230 | * In short, make sure you hash any inodes _before_ you start marking | |
2231 | * them dirty. | |
1da177e4 | 2232 | * |
03ba3782 JA |
2233 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
2234 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
2235 | * the kernel-internal blockdev inode represents the dirtying time of the | |
2236 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
2237 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
2238 | * blockdev inode. | |
1da177e4 | 2239 | */ |
03ba3782 | 2240 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 2241 | { |
03ba3782 | 2242 | struct super_block *sb = inode->i_sb; |
0ae45f63 TT |
2243 | int dirtytime; |
2244 | ||
2245 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 2246 | |
03ba3782 JA |
2247 | /* |
2248 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
2249 | * dirty the inode itself | |
2250 | */ | |
0e11f644 | 2251 | if (flags & (I_DIRTY_INODE | I_DIRTY_TIME)) { |
9fb0a7da TH |
2252 | trace_writeback_dirty_inode_start(inode, flags); |
2253 | ||
03ba3782 | 2254 | if (sb->s_op->dirty_inode) |
aa385729 | 2255 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
2256 | |
2257 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 | 2258 | } |
0ae45f63 TT |
2259 | if (flags & I_DIRTY_INODE) |
2260 | flags &= ~I_DIRTY_TIME; | |
2261 | dirtytime = flags & I_DIRTY_TIME; | |
03ba3782 JA |
2262 | |
2263 | /* | |
9c6ac78e TH |
2264 | * Paired with smp_mb() in __writeback_single_inode() for the |
2265 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
2266 | */ |
2267 | smp_mb(); | |
2268 | ||
0ae45f63 TT |
2269 | if (((inode->i_state & flags) == flags) || |
2270 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
2271 | return; |
2272 | ||
2273 | if (unlikely(block_dump)) | |
2274 | block_dump___mark_inode_dirty(inode); | |
2275 | ||
250df6ed | 2276 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
2277 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
2278 | goto out_unlock_inode; | |
03ba3782 JA |
2279 | if ((inode->i_state & flags) != flags) { |
2280 | const int was_dirty = inode->i_state & I_DIRTY; | |
2281 | ||
52ebea74 TH |
2282 | inode_attach_wb(inode, NULL); |
2283 | ||
0ae45f63 TT |
2284 | if (flags & I_DIRTY_INODE) |
2285 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
2286 | inode->i_state |= flags; |
2287 | ||
2288 | /* | |
2289 | * If the inode is being synced, just update its dirty state. | |
2290 | * The unlocker will place the inode on the appropriate | |
2291 | * superblock list, based upon its state. | |
2292 | */ | |
2293 | if (inode->i_state & I_SYNC) | |
250df6ed | 2294 | goto out_unlock_inode; |
03ba3782 JA |
2295 | |
2296 | /* | |
2297 | * Only add valid (hashed) inodes to the superblock's | |
2298 | * dirty list. Add blockdev inodes as well. | |
2299 | */ | |
2300 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 2301 | if (inode_unhashed(inode)) |
250df6ed | 2302 | goto out_unlock_inode; |
03ba3782 | 2303 | } |
a4ffdde6 | 2304 | if (inode->i_state & I_FREEING) |
250df6ed | 2305 | goto out_unlock_inode; |
03ba3782 JA |
2306 | |
2307 | /* | |
2308 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
2309 | * reposition it (that would break b_dirty time-ordering). | |
2310 | */ | |
2311 | if (!was_dirty) { | |
87e1d789 | 2312 | struct bdi_writeback *wb; |
d6c10f1f | 2313 | struct list_head *dirty_list; |
a66979ab | 2314 | bool wakeup_bdi = false; |
253c34e9 | 2315 | |
87e1d789 | 2316 | wb = locked_inode_to_wb_and_lock_list(inode); |
253c34e9 | 2317 | |
0747259d TH |
2318 | WARN(bdi_cap_writeback_dirty(wb->bdi) && |
2319 | !test_bit(WB_registered, &wb->state), | |
2320 | "bdi-%s not registered\n", wb->bdi->name); | |
03ba3782 JA |
2321 | |
2322 | inode->dirtied_when = jiffies; | |
a2f48706 TT |
2323 | if (dirtytime) |
2324 | inode->dirtied_time_when = jiffies; | |
d6c10f1f | 2325 | |
0e11f644 | 2326 | if (inode->i_state & I_DIRTY) |
0747259d | 2327 | dirty_list = &wb->b_dirty; |
a2f48706 | 2328 | else |
0747259d | 2329 | dirty_list = &wb->b_dirty_time; |
d6c10f1f | 2330 | |
c7f54084 | 2331 | wakeup_bdi = inode_io_list_move_locked(inode, wb, |
d6c10f1f TH |
2332 | dirty_list); |
2333 | ||
0747259d | 2334 | spin_unlock(&wb->list_lock); |
0ae45f63 | 2335 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab | 2336 | |
d6c10f1f TH |
2337 | /* |
2338 | * If this is the first dirty inode for this bdi, | |
2339 | * we have to wake-up the corresponding bdi thread | |
2340 | * to make sure background write-back happens | |
2341 | * later. | |
2342 | */ | |
0747259d TH |
2343 | if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi) |
2344 | wb_wakeup_delayed(wb); | |
a66979ab | 2345 | return; |
1da177e4 | 2346 | } |
1da177e4 | 2347 | } |
250df6ed DC |
2348 | out_unlock_inode: |
2349 | spin_unlock(&inode->i_lock); | |
03ba3782 JA |
2350 | } |
2351 | EXPORT_SYMBOL(__mark_inode_dirty); | |
2352 | ||
e97fedb9 DC |
2353 | /* |
2354 | * The @s_sync_lock is used to serialise concurrent sync operations | |
2355 | * to avoid lock contention problems with concurrent wait_sb_inodes() calls. | |
2356 | * Concurrent callers will block on the s_sync_lock rather than doing contending | |
2357 | * walks. The queueing maintains sync(2) required behaviour as all the IO that | |
2358 | * has been issued up to the time this function is enter is guaranteed to be | |
2359 | * completed by the time we have gained the lock and waited for all IO that is | |
2360 | * in progress regardless of the order callers are granted the lock. | |
2361 | */ | |
b6e51316 | 2362 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 | 2363 | { |
6c60d2b5 | 2364 | LIST_HEAD(sync_list); |
03ba3782 JA |
2365 | |
2366 | /* | |
2367 | * We need to be protected against the filesystem going from | |
2368 | * r/o to r/w or vice versa. | |
2369 | */ | |
b6e51316 | 2370 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 2371 | |
e97fedb9 | 2372 | mutex_lock(&sb->s_sync_lock); |
03ba3782 JA |
2373 | |
2374 | /* | |
6c60d2b5 DC |
2375 | * Splice the writeback list onto a temporary list to avoid waiting on |
2376 | * inodes that have started writeback after this point. | |
2377 | * | |
2378 | * Use rcu_read_lock() to keep the inodes around until we have a | |
2379 | * reference. s_inode_wblist_lock protects sb->s_inodes_wb as well as | |
2380 | * the local list because inodes can be dropped from either by writeback | |
2381 | * completion. | |
2382 | */ | |
2383 | rcu_read_lock(); | |
2384 | spin_lock_irq(&sb->s_inode_wblist_lock); | |
2385 | list_splice_init(&sb->s_inodes_wb, &sync_list); | |
2386 | ||
2387 | /* | |
2388 | * Data integrity sync. Must wait for all pages under writeback, because | |
2389 | * there may have been pages dirtied before our sync call, but which had | |
2390 | * writeout started before we write it out. In which case, the inode | |
2391 | * may not be on the dirty list, but we still have to wait for that | |
2392 | * writeout. | |
03ba3782 | 2393 | */ |
6c60d2b5 DC |
2394 | while (!list_empty(&sync_list)) { |
2395 | struct inode *inode = list_first_entry(&sync_list, struct inode, | |
2396 | i_wb_list); | |
250df6ed | 2397 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 2398 | |
6c60d2b5 DC |
2399 | /* |
2400 | * Move each inode back to the wb list before we drop the lock | |
2401 | * to preserve consistency between i_wb_list and the mapping | |
2402 | * writeback tag. Writeback completion is responsible to remove | |
2403 | * the inode from either list once the writeback tag is cleared. | |
2404 | */ | |
2405 | list_move_tail(&inode->i_wb_list, &sb->s_inodes_wb); | |
2406 | ||
2407 | /* | |
2408 | * The mapping can appear untagged while still on-list since we | |
2409 | * do not have the mapping lock. Skip it here, wb completion | |
2410 | * will remove it. | |
2411 | */ | |
2412 | if (!mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
2413 | continue; | |
2414 | ||
2415 | spin_unlock_irq(&sb->s_inode_wblist_lock); | |
2416 | ||
250df6ed | 2417 | spin_lock(&inode->i_lock); |
6c60d2b5 | 2418 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) { |
250df6ed | 2419 | spin_unlock(&inode->i_lock); |
6c60d2b5 DC |
2420 | |
2421 | spin_lock_irq(&sb->s_inode_wblist_lock); | |
03ba3782 | 2422 | continue; |
250df6ed | 2423 | } |
03ba3782 | 2424 | __iget(inode); |
250df6ed | 2425 | spin_unlock(&inode->i_lock); |
6c60d2b5 | 2426 | rcu_read_unlock(); |
03ba3782 | 2427 | |
aa750fd7 JN |
2428 | /* |
2429 | * We keep the error status of individual mapping so that | |
2430 | * applications can catch the writeback error using fsync(2). | |
2431 | * See filemap_fdatawait_keep_errors() for details. | |
2432 | */ | |
2433 | filemap_fdatawait_keep_errors(mapping); | |
03ba3782 JA |
2434 | |
2435 | cond_resched(); | |
2436 | ||
6c60d2b5 DC |
2437 | iput(inode); |
2438 | ||
2439 | rcu_read_lock(); | |
2440 | spin_lock_irq(&sb->s_inode_wblist_lock); | |
03ba3782 | 2441 | } |
6c60d2b5 DC |
2442 | spin_unlock_irq(&sb->s_inode_wblist_lock); |
2443 | rcu_read_unlock(); | |
e97fedb9 | 2444 | mutex_unlock(&sb->s_sync_lock); |
1da177e4 LT |
2445 | } |
2446 | ||
f30a7d0c TH |
2447 | static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
2448 | enum wb_reason reason, bool skip_if_busy) | |
1da177e4 | 2449 | { |
5b9cce4c TH |
2450 | struct backing_dev_info *bdi = sb->s_bdi; |
2451 | DEFINE_WB_COMPLETION(done, bdi); | |
83ba7b07 | 2452 | struct wb_writeback_work work = { |
6e6938b6 WF |
2453 | .sb = sb, |
2454 | .sync_mode = WB_SYNC_NONE, | |
2455 | .tagged_writepages = 1, | |
2456 | .done = &done, | |
2457 | .nr_pages = nr, | |
0e175a18 | 2458 | .reason = reason, |
3c4d7165 | 2459 | }; |
d8a8559c | 2460 | |
e7972912 | 2461 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 2462 | return; |
cf37e972 | 2463 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
f30a7d0c | 2464 | |
db125360 | 2465 | bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy); |
5b9cce4c | 2466 | wb_wait_for_completion(&done); |
e913fc82 | 2467 | } |
f30a7d0c TH |
2468 | |
2469 | /** | |
2470 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block | |
2471 | * @sb: the superblock | |
2472 | * @nr: the number of pages to write | |
2473 | * @reason: reason why some writeback work initiated | |
2474 | * | |
2475 | * Start writeback on some inodes on this super_block. No guarantees are made | |
2476 | * on how many (if any) will be written, and this function does not wait | |
2477 | * for IO completion of submitted IO. | |
2478 | */ | |
2479 | void writeback_inodes_sb_nr(struct super_block *sb, | |
2480 | unsigned long nr, | |
2481 | enum wb_reason reason) | |
2482 | { | |
2483 | __writeback_inodes_sb_nr(sb, nr, reason, false); | |
2484 | } | |
3259f8be CM |
2485 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
2486 | ||
2487 | /** | |
2488 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
2489 | * @sb: the superblock | |
786228ab | 2490 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
2491 | * |
2492 | * Start writeback on some inodes on this super_block. No guarantees are made | |
2493 | * on how many (if any) will be written, and this function does not wait | |
2494 | * for IO completion of submitted IO. | |
2495 | */ | |
0e175a18 | 2496 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 2497 | { |
0e175a18 | 2498 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 2499 | } |
0e3c9a22 | 2500 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 2501 | |
17bd55d0 | 2502 | /** |
8264c321 | 2503 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
17bd55d0 | 2504 | * @sb: the superblock |
8264c321 | 2505 | * @reason: reason why some writeback work was initiated |
17bd55d0 | 2506 | * |
8264c321 | 2507 | * Invoke __writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 | 2508 | */ |
8264c321 | 2509 | void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
17bd55d0 | 2510 | { |
10ee27a0 | 2511 | if (!down_read_trylock(&sb->s_umount)) |
8264c321 | 2512 | return; |
10ee27a0 | 2513 | |
8264c321 | 2514 | __writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason, true); |
10ee27a0 | 2515 | up_read(&sb->s_umount); |
3259f8be | 2516 | } |
10ee27a0 | 2517 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 2518 | |
d8a8559c JA |
2519 | /** |
2520 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 2521 | * @sb: the superblock |
d8a8559c JA |
2522 | * |
2523 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 2524 | * super_block. |
d8a8559c | 2525 | */ |
0dc83bd3 | 2526 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 2527 | { |
5b9cce4c TH |
2528 | struct backing_dev_info *bdi = sb->s_bdi; |
2529 | DEFINE_WB_COMPLETION(done, bdi); | |
83ba7b07 | 2530 | struct wb_writeback_work work = { |
3c4d7165 CH |
2531 | .sb = sb, |
2532 | .sync_mode = WB_SYNC_ALL, | |
2533 | .nr_pages = LONG_MAX, | |
2534 | .range_cyclic = 0, | |
83ba7b07 | 2535 | .done = &done, |
0e175a18 | 2536 | .reason = WB_REASON_SYNC, |
7747bd4b | 2537 | .for_sync = 1, |
3c4d7165 CH |
2538 | }; |
2539 | ||
006a0973 TH |
2540 | /* |
2541 | * Can't skip on !bdi_has_dirty() because we should wait for !dirty | |
2542 | * inodes under writeback and I_DIRTY_TIME inodes ignored by | |
2543 | * bdi_has_dirty() need to be written out too. | |
2544 | */ | |
2545 | if (bdi == &noop_backing_dev_info) | |
6eedc701 | 2546 | return; |
cf37e972 CH |
2547 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
2548 | ||
7fc5854f TH |
2549 | /* protect against inode wb switch, see inode_switch_wbs_work_fn() */ |
2550 | bdi_down_write_wb_switch_rwsem(bdi); | |
db125360 | 2551 | bdi_split_work_to_wbs(bdi, &work, false); |
5b9cce4c | 2552 | wb_wait_for_completion(&done); |
7fc5854f | 2553 | bdi_up_write_wb_switch_rwsem(bdi); |
83ba7b07 | 2554 | |
b6e51316 | 2555 | wait_sb_inodes(sb); |
1da177e4 | 2556 | } |
d8a8559c | 2557 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 2558 | |
1da177e4 | 2559 | /** |
7f04c26d AA |
2560 | * write_inode_now - write an inode to disk |
2561 | * @inode: inode to write to disk | |
2562 | * @sync: whether the write should be synchronous or not | |
2563 | * | |
2564 | * This function commits an inode to disk immediately if it is dirty. This is | |
2565 | * primarily needed by knfsd. | |
1da177e4 | 2566 | * |
7f04c26d | 2567 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 2568 | */ |
1da177e4 LT |
2569 | int write_inode_now(struct inode *inode, int sync) |
2570 | { | |
1da177e4 LT |
2571 | struct writeback_control wbc = { |
2572 | .nr_to_write = LONG_MAX, | |
18914b18 | 2573 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
2574 | .range_start = 0, |
2575 | .range_end = LLONG_MAX, | |
1da177e4 LT |
2576 | }; |
2577 | ||
2578 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 2579 | wbc.nr_to_write = 0; |
1da177e4 LT |
2580 | |
2581 | might_sleep(); | |
aaf25593 | 2582 | return writeback_single_inode(inode, &wbc); |
1da177e4 LT |
2583 | } |
2584 | EXPORT_SYMBOL(write_inode_now); | |
2585 | ||
2586 | /** | |
2587 | * sync_inode - write an inode and its pages to disk. | |
2588 | * @inode: the inode to sync | |
2589 | * @wbc: controls the writeback mode | |
2590 | * | |
2591 | * sync_inode() will write an inode and its pages to disk. It will also | |
2592 | * correctly update the inode on its superblock's dirty inode lists and will | |
2593 | * update inode->i_state. | |
2594 | * | |
2595 | * The caller must have a ref on the inode. | |
2596 | */ | |
2597 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
2598 | { | |
aaf25593 | 2599 | return writeback_single_inode(inode, wbc); |
1da177e4 LT |
2600 | } |
2601 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
2602 | |
2603 | /** | |
c691b9d9 | 2604 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
2605 | * @inode: the inode to sync |
2606 | * @wait: wait for I/O to complete. | |
2607 | * | |
c691b9d9 | 2608 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
2609 | * |
2610 | * Note: only writes the actual inode, no associated data or other metadata. | |
2611 | */ | |
2612 | int sync_inode_metadata(struct inode *inode, int wait) | |
2613 | { | |
2614 | struct writeback_control wbc = { | |
2615 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
2616 | .nr_to_write = 0, /* metadata-only */ | |
2617 | }; | |
2618 | ||
2619 | return sync_inode(inode, &wbc); | |
2620 | } | |
2621 | EXPORT_SYMBOL(sync_inode_metadata); |