]>
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; | |
45 | enum writeback_sync_modes sync_mode; | |
6e6938b6 | 46 | unsigned int tagged_writepages:1; |
52957fe1 HS |
47 | unsigned int for_kupdate:1; |
48 | unsigned int range_cyclic:1; | |
49 | unsigned int for_background:1; | |
7747bd4b | 50 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
ac7b19a3 | 51 | unsigned int auto_free:1; /* free on completion */ |
0e175a18 | 52 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 53 | |
8010c3b6 | 54 | struct list_head list; /* pending work list */ |
cc395d7f | 55 | struct wb_completion *done; /* set if the caller waits */ |
03ba3782 JA |
56 | }; |
57 | ||
a2f48706 TT |
58 | /* |
59 | * If an inode is constantly having its pages dirtied, but then the | |
60 | * updates stop dirtytime_expire_interval seconds in the past, it's | |
61 | * possible for the worst case time between when an inode has its | |
62 | * timestamps updated and when they finally get written out to be two | |
63 | * dirtytime_expire_intervals. We set the default to 12 hours (in | |
64 | * seconds), which means most of the time inodes will have their | |
65 | * timestamps written to disk after 12 hours, but in the worst case a | |
66 | * few inodes might not their timestamps updated for 24 hours. | |
67 | */ | |
68 | unsigned int dirtytime_expire_interval = 12 * 60 * 60; | |
69 | ||
7ccf19a8 NP |
70 | static inline struct inode *wb_inode(struct list_head *head) |
71 | { | |
c7f54084 | 72 | return list_entry(head, struct inode, i_io_list); |
7ccf19a8 NP |
73 | } |
74 | ||
15eb77a0 WF |
75 | /* |
76 | * Include the creation of the trace points after defining the | |
77 | * wb_writeback_work structure and inline functions so that the definition | |
78 | * remains local to this file. | |
79 | */ | |
80 | #define CREATE_TRACE_POINTS | |
81 | #include <trace/events/writeback.h> | |
82 | ||
774016b2 SW |
83 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
84 | ||
d6c10f1f TH |
85 | static bool wb_io_lists_populated(struct bdi_writeback *wb) |
86 | { | |
87 | if (wb_has_dirty_io(wb)) { | |
88 | return false; | |
89 | } else { | |
90 | set_bit(WB_has_dirty_io, &wb->state); | |
95a46c65 | 91 | WARN_ON_ONCE(!wb->avg_write_bandwidth); |
766a9d6e TH |
92 | atomic_long_add(wb->avg_write_bandwidth, |
93 | &wb->bdi->tot_write_bandwidth); | |
d6c10f1f TH |
94 | return true; |
95 | } | |
96 | } | |
97 | ||
98 | static void wb_io_lists_depopulated(struct bdi_writeback *wb) | |
99 | { | |
100 | if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && | |
766a9d6e | 101 | list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { |
d6c10f1f | 102 | clear_bit(WB_has_dirty_io, &wb->state); |
95a46c65 TH |
103 | WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, |
104 | &wb->bdi->tot_write_bandwidth) < 0); | |
766a9d6e | 105 | } |
d6c10f1f TH |
106 | } |
107 | ||
108 | /** | |
c7f54084 | 109 | * inode_io_list_move_locked - move an inode onto a bdi_writeback IO list |
d6c10f1f TH |
110 | * @inode: inode to be moved |
111 | * @wb: target bdi_writeback | |
bbbc3c1c | 112 | * @head: one of @wb->b_{dirty|io|more_io|dirty_time} |
d6c10f1f | 113 | * |
c7f54084 | 114 | * Move @inode->i_io_list to @list of @wb and set %WB_has_dirty_io. |
d6c10f1f TH |
115 | * Returns %true if @inode is the first occupant of the !dirty_time IO |
116 | * lists; otherwise, %false. | |
117 | */ | |
c7f54084 | 118 | static bool inode_io_list_move_locked(struct inode *inode, |
d6c10f1f TH |
119 | struct bdi_writeback *wb, |
120 | struct list_head *head) | |
121 | { | |
122 | assert_spin_locked(&wb->list_lock); | |
123 | ||
c7f54084 | 124 | list_move(&inode->i_io_list, head); |
d6c10f1f TH |
125 | |
126 | /* dirty_time doesn't count as dirty_io until expiration */ | |
127 | if (head != &wb->b_dirty_time) | |
128 | return wb_io_lists_populated(wb); | |
129 | ||
130 | wb_io_lists_depopulated(wb); | |
131 | return false; | |
132 | } | |
133 | ||
f0054bb1 | 134 | static void wb_wakeup(struct bdi_writeback *wb) |
5acda9d1 | 135 | { |
f0054bb1 TH |
136 | spin_lock_bh(&wb->work_lock); |
137 | if (test_bit(WB_registered, &wb->state)) | |
138 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
139 | spin_unlock_bh(&wb->work_lock); | |
5acda9d1 JK |
140 | } |
141 | ||
4a3a485b TE |
142 | static void finish_writeback_work(struct bdi_writeback *wb, |
143 | struct wb_writeback_work *work) | |
144 | { | |
145 | struct wb_completion *done = work->done; | |
146 | ||
147 | if (work->auto_free) | |
148 | kfree(work); | |
8e00c4e9 TH |
149 | if (done) { |
150 | wait_queue_head_t *waitq = done->waitq; | |
151 | ||
152 | /* @done can't be accessed after the following dec */ | |
153 | if (atomic_dec_and_test(&done->cnt)) | |
154 | wake_up_all(waitq); | |
155 | } | |
4a3a485b TE |
156 | } |
157 | ||
f0054bb1 TH |
158 | static void wb_queue_work(struct bdi_writeback *wb, |
159 | struct wb_writeback_work *work) | |
6585027a | 160 | { |
5634cc2a | 161 | trace_writeback_queue(wb, work); |
6585027a | 162 | |
cc395d7f TH |
163 | if (work->done) |
164 | atomic_inc(&work->done->cnt); | |
4a3a485b TE |
165 | |
166 | spin_lock_bh(&wb->work_lock); | |
167 | ||
168 | if (test_bit(WB_registered, &wb->state)) { | |
169 | list_add_tail(&work->list, &wb->work_list); | |
170 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
171 | } else | |
172 | finish_writeback_work(wb, work); | |
173 | ||
f0054bb1 | 174 | spin_unlock_bh(&wb->work_lock); |
1da177e4 LT |
175 | } |
176 | ||
cc395d7f TH |
177 | /** |
178 | * wb_wait_for_completion - wait for completion of bdi_writeback_works | |
cc395d7f TH |
179 | * @done: target wb_completion |
180 | * | |
181 | * Wait for one or more work items issued to @bdi with their ->done field | |
5b9cce4c TH |
182 | * set to @done, which should have been initialized with |
183 | * DEFINE_WB_COMPLETION(). This function returns after all such work items | |
184 | * are completed. Work items which are waited upon aren't freed | |
cc395d7f TH |
185 | * automatically on completion. |
186 | */ | |
5b9cce4c | 187 | void wb_wait_for_completion(struct wb_completion *done) |
cc395d7f TH |
188 | { |
189 | atomic_dec(&done->cnt); /* put down the initial count */ | |
5b9cce4c | 190 | wait_event(*done->waitq, !atomic_read(&done->cnt)); |
cc395d7f TH |
191 | } |
192 | ||
703c2708 TH |
193 | #ifdef CONFIG_CGROUP_WRITEBACK |
194 | ||
55a694df TH |
195 | /* |
196 | * Parameters for foreign inode detection, see wbc_detach_inode() to see | |
197 | * how they're used. | |
198 | * | |
199 | * These paramters are inherently heuristical as the detection target | |
200 | * itself is fuzzy. All we want to do is detaching an inode from the | |
201 | * current owner if it's being written to by some other cgroups too much. | |
202 | * | |
203 | * The current cgroup writeback is built on the assumption that multiple | |
204 | * cgroups writing to the same inode concurrently is very rare and a mode | |
205 | * of operation which isn't well supported. As such, the goal is not | |
206 | * taking too long when a different cgroup takes over an inode while | |
207 | * avoiding too aggressive flip-flops from occasional foreign writes. | |
208 | * | |
209 | * We record, very roughly, 2s worth of IO time history and if more than | |
210 | * half of that is foreign, trigger the switch. The recording is quantized | |
211 | * to 16 slots. To avoid tiny writes from swinging the decision too much, | |
212 | * writes smaller than 1/8 of avg size are ignored. | |
213 | */ | |
2a814908 TH |
214 | #define WB_FRN_TIME_SHIFT 13 /* 1s = 2^13, upto 8 secs w/ 16bit */ |
215 | #define WB_FRN_TIME_AVG_SHIFT 3 /* avg = avg * 7/8 + new * 1/8 */ | |
55a694df | 216 | #define WB_FRN_TIME_CUT_DIV 8 /* ignore rounds < avg / 8 */ |
2a814908 TH |
217 | #define WB_FRN_TIME_PERIOD (2 * (1 << WB_FRN_TIME_SHIFT)) /* 2s */ |
218 | ||
219 | #define WB_FRN_HIST_SLOTS 16 /* inode->i_wb_frn_history is 16bit */ | |
220 | #define WB_FRN_HIST_UNIT (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS) | |
221 | /* each slot's duration is 2s / 16 */ | |
222 | #define WB_FRN_HIST_THR_SLOTS (WB_FRN_HIST_SLOTS / 2) | |
223 | /* if foreign slots >= 8, switch */ | |
224 | #define WB_FRN_HIST_MAX_SLOTS (WB_FRN_HIST_THR_SLOTS / 2 + 1) | |
225 | /* one round can affect upto 5 slots */ | |
6444f47e | 226 | #define WB_FRN_MAX_IN_FLIGHT 1024 /* don't queue too many concurrently */ |
2a814908 | 227 | |
a1a0e23e TH |
228 | static atomic_t isw_nr_in_flight = ATOMIC_INIT(0); |
229 | static struct workqueue_struct *isw_wq; | |
230 | ||
21c6321f TH |
231 | void __inode_attach_wb(struct inode *inode, struct page *page) |
232 | { | |
233 | struct backing_dev_info *bdi = inode_to_bdi(inode); | |
234 | struct bdi_writeback *wb = NULL; | |
235 | ||
236 | if (inode_cgwb_enabled(inode)) { | |
237 | struct cgroup_subsys_state *memcg_css; | |
238 | ||
239 | if (page) { | |
240 | memcg_css = mem_cgroup_css_from_page(page); | |
241 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
242 | } else { | |
243 | /* must pin memcg_css, see wb_get_create() */ | |
244 | memcg_css = task_get_css(current, memory_cgrp_id); | |
245 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
246 | css_put(memcg_css); | |
247 | } | |
248 | } | |
249 | ||
250 | if (!wb) | |
251 | wb = &bdi->wb; | |
252 | ||
253 | /* | |
254 | * There may be multiple instances of this function racing to | |
255 | * update the same inode. Use cmpxchg() to tell the winner. | |
256 | */ | |
257 | if (unlikely(cmpxchg(&inode->i_wb, NULL, wb))) | |
258 | wb_put(wb); | |
259 | } | |
9b0eb69b | 260 | EXPORT_SYMBOL_GPL(__inode_attach_wb); |
21c6321f | 261 | |
f3b6a6df RG |
262 | /** |
263 | * inode_cgwb_move_to_attached - put the inode onto wb->b_attached list | |
264 | * @inode: inode of interest with i_lock held | |
265 | * @wb: target bdi_writeback | |
266 | * | |
267 | * Remove the inode from wb's io lists and if necessarily put onto b_attached | |
268 | * list. Only inodes attached to cgwb's are kept on this list. | |
269 | */ | |
270 | static void inode_cgwb_move_to_attached(struct inode *inode, | |
271 | struct bdi_writeback *wb) | |
272 | { | |
273 | assert_spin_locked(&wb->list_lock); | |
274 | assert_spin_locked(&inode->i_lock); | |
275 | ||
276 | inode->i_state &= ~I_SYNC_QUEUED; | |
277 | if (wb != &wb->bdi->wb) | |
278 | list_move(&inode->i_io_list, &wb->b_attached); | |
279 | else | |
280 | list_del_init(&inode->i_io_list); | |
281 | wb_io_lists_depopulated(wb); | |
282 | } | |
283 | ||
87e1d789 TH |
284 | /** |
285 | * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it | |
286 | * @inode: inode of interest with i_lock held | |
287 | * | |
288 | * Returns @inode's wb with its list_lock held. @inode->i_lock must be | |
289 | * held on entry and is released on return. The returned wb is guaranteed | |
290 | * to stay @inode's associated wb until its list_lock is released. | |
291 | */ | |
292 | static struct bdi_writeback * | |
293 | locked_inode_to_wb_and_lock_list(struct inode *inode) | |
294 | __releases(&inode->i_lock) | |
295 | __acquires(&wb->list_lock) | |
296 | { | |
297 | while (true) { | |
298 | struct bdi_writeback *wb = inode_to_wb(inode); | |
299 | ||
300 | /* | |
301 | * inode_to_wb() association is protected by both | |
302 | * @inode->i_lock and @wb->list_lock but list_lock nests | |
303 | * outside i_lock. Drop i_lock and verify that the | |
304 | * association hasn't changed after acquiring list_lock. | |
305 | */ | |
306 | wb_get(wb); | |
307 | spin_unlock(&inode->i_lock); | |
308 | spin_lock(&wb->list_lock); | |
87e1d789 | 309 | |
aaa2cacf | 310 | /* i_wb may have changed inbetween, can't use inode_to_wb() */ |
614a4e37 TH |
311 | if (likely(wb == inode->i_wb)) { |
312 | wb_put(wb); /* @inode already has ref */ | |
313 | return wb; | |
314 | } | |
87e1d789 TH |
315 | |
316 | spin_unlock(&wb->list_lock); | |
614a4e37 | 317 | wb_put(wb); |
87e1d789 TH |
318 | cpu_relax(); |
319 | spin_lock(&inode->i_lock); | |
320 | } | |
321 | } | |
322 | ||
323 | /** | |
324 | * inode_to_wb_and_lock_list - determine an inode's wb and lock it | |
325 | * @inode: inode of interest | |
326 | * | |
327 | * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held | |
328 | * on entry. | |
329 | */ | |
330 | static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) | |
331 | __acquires(&wb->list_lock) | |
332 | { | |
333 | spin_lock(&inode->i_lock); | |
334 | return locked_inode_to_wb_and_lock_list(inode); | |
335 | } | |
336 | ||
682aa8e1 TH |
337 | struct inode_switch_wbs_context { |
338 | struct inode *inode; | |
339 | struct bdi_writeback *new_wb; | |
340 | ||
29264d92 | 341 | struct rcu_work work; |
682aa8e1 TH |
342 | }; |
343 | ||
7fc5854f TH |
344 | static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) |
345 | { | |
346 | down_write(&bdi->wb_switch_rwsem); | |
347 | } | |
348 | ||
349 | static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) | |
350 | { | |
351 | up_write(&bdi->wb_switch_rwsem); | |
352 | } | |
353 | ||
682aa8e1 TH |
354 | static void inode_switch_wbs_work_fn(struct work_struct *work) |
355 | { | |
356 | struct inode_switch_wbs_context *isw = | |
29264d92 | 357 | container_of(to_rcu_work(work), struct inode_switch_wbs_context, work); |
682aa8e1 | 358 | struct inode *inode = isw->inode; |
7fc5854f | 359 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
d10c8095 TH |
360 | struct address_space *mapping = inode->i_mapping; |
361 | struct bdi_writeback *old_wb = inode->i_wb; | |
682aa8e1 | 362 | struct bdi_writeback *new_wb = isw->new_wb; |
04edf02c MW |
363 | XA_STATE(xas, &mapping->i_pages, 0); |
364 | struct page *page; | |
d10c8095 | 365 | bool switched = false; |
682aa8e1 | 366 | |
7fc5854f TH |
367 | /* |
368 | * If @inode switches cgwb membership while sync_inodes_sb() is | |
369 | * being issued, sync_inodes_sb() might miss it. Synchronize. | |
370 | */ | |
371 | down_read(&bdi->wb_switch_rwsem); | |
372 | ||
682aa8e1 TH |
373 | /* |
374 | * By the time control reaches here, RCU grace period has passed | |
375 | * since I_WB_SWITCH assertion and all wb stat update transactions | |
376 | * between unlocked_inode_to_wb_begin/end() are guaranteed to be | |
b93b0163 | 377 | * synchronizing against the i_pages lock. |
d10c8095 | 378 | * |
b93b0163 | 379 | * Grabbing old_wb->list_lock, inode->i_lock and the i_pages lock |
d10c8095 TH |
380 | * gives us exclusion against all wb related operations on @inode |
381 | * including IO list manipulations and stat updates. | |
682aa8e1 | 382 | */ |
d10c8095 TH |
383 | if (old_wb < new_wb) { |
384 | spin_lock(&old_wb->list_lock); | |
385 | spin_lock_nested(&new_wb->list_lock, SINGLE_DEPTH_NESTING); | |
386 | } else { | |
387 | spin_lock(&new_wb->list_lock); | |
388 | spin_lock_nested(&old_wb->list_lock, SINGLE_DEPTH_NESTING); | |
389 | } | |
682aa8e1 | 390 | spin_lock(&inode->i_lock); |
b93b0163 | 391 | xa_lock_irq(&mapping->i_pages); |
d10c8095 TH |
392 | |
393 | /* | |
4ade5867 RG |
394 | * Once I_FREEING or I_WILL_FREE are visible under i_lock, the eviction |
395 | * path owns the inode and we shouldn't modify ->i_io_list. | |
d10c8095 | 396 | */ |
4ade5867 | 397 | if (unlikely(inode->i_state & (I_FREEING | I_WILL_FREE))) |
d10c8095 TH |
398 | goto skip_switch; |
399 | ||
3a8e9ac8 TH |
400 | trace_inode_switch_wbs(inode, old_wb, new_wb); |
401 | ||
d10c8095 TH |
402 | /* |
403 | * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points | |
404 | * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to | |
b93b0163 | 405 | * pages actually under writeback. |
d10c8095 | 406 | */ |
04edf02c MW |
407 | xas_for_each_marked(&xas, page, ULONG_MAX, PAGECACHE_TAG_DIRTY) { |
408 | if (PageDirty(page)) { | |
3e8f399d NB |
409 | dec_wb_stat(old_wb, WB_RECLAIMABLE); |
410 | inc_wb_stat(new_wb, WB_RECLAIMABLE); | |
d10c8095 TH |
411 | } |
412 | } | |
413 | ||
04edf02c MW |
414 | xas_set(&xas, 0); |
415 | xas_for_each_marked(&xas, page, ULONG_MAX, PAGECACHE_TAG_WRITEBACK) { | |
416 | WARN_ON_ONCE(!PageWriteback(page)); | |
417 | dec_wb_stat(old_wb, WB_WRITEBACK); | |
418 | inc_wb_stat(new_wb, WB_WRITEBACK); | |
d10c8095 TH |
419 | } |
420 | ||
421 | wb_get(new_wb); | |
422 | ||
423 | /* | |
f3b6a6df RG |
424 | * Transfer to @new_wb's IO list if necessary. If the @inode is dirty, |
425 | * the specific list @inode was on is ignored and the @inode is put on | |
426 | * ->b_dirty which is always correct including from ->b_dirty_time. | |
427 | * The transfer preserves @inode->dirtied_when ordering. If the @inode | |
428 | * was clean, it means it was on the b_attached list, so move it onto | |
429 | * the b_attached list of @new_wb. | |
d10c8095 | 430 | */ |
c7f54084 | 431 | if (!list_empty(&inode->i_io_list)) { |
d10c8095 | 432 | inode->i_wb = new_wb; |
f3b6a6df RG |
433 | |
434 | if (inode->i_state & I_DIRTY_ALL) { | |
435 | struct inode *pos; | |
436 | ||
437 | list_for_each_entry(pos, &new_wb->b_dirty, i_io_list) | |
438 | if (time_after_eq(inode->dirtied_when, | |
439 | pos->dirtied_when)) | |
440 | break; | |
441 | inode_io_list_move_locked(inode, new_wb, | |
442 | pos->i_io_list.prev); | |
443 | } else { | |
444 | inode_cgwb_move_to_attached(inode, new_wb); | |
445 | } | |
d10c8095 TH |
446 | } else { |
447 | inode->i_wb = new_wb; | |
448 | } | |
682aa8e1 | 449 | |
d10c8095 | 450 | /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */ |
682aa8e1 TH |
451 | inode->i_wb_frn_winner = 0; |
452 | inode->i_wb_frn_avg_time = 0; | |
453 | inode->i_wb_frn_history = 0; | |
d10c8095 TH |
454 | switched = true; |
455 | skip_switch: | |
682aa8e1 TH |
456 | /* |
457 | * Paired with load_acquire in unlocked_inode_to_wb_begin() and | |
458 | * ensures that the new wb is visible if they see !I_WB_SWITCH. | |
459 | */ | |
460 | smp_store_release(&inode->i_state, inode->i_state & ~I_WB_SWITCH); | |
461 | ||
b93b0163 | 462 | xa_unlock_irq(&mapping->i_pages); |
682aa8e1 | 463 | spin_unlock(&inode->i_lock); |
d10c8095 TH |
464 | spin_unlock(&new_wb->list_lock); |
465 | spin_unlock(&old_wb->list_lock); | |
682aa8e1 | 466 | |
7fc5854f TH |
467 | up_read(&bdi->wb_switch_rwsem); |
468 | ||
d10c8095 TH |
469 | if (switched) { |
470 | wb_wakeup(new_wb); | |
471 | wb_put(old_wb); | |
472 | } | |
682aa8e1 | 473 | wb_put(new_wb); |
d10c8095 TH |
474 | |
475 | iput(inode); | |
682aa8e1 | 476 | kfree(isw); |
a1a0e23e TH |
477 | |
478 | atomic_dec(&isw_nr_in_flight); | |
682aa8e1 TH |
479 | } |
480 | ||
682aa8e1 TH |
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 | 506 | |
8826ee4f RG |
507 | atomic_inc(&isw_nr_in_flight); |
508 | ||
682aa8e1 TH |
509 | /* find and pin the new wb */ |
510 | rcu_read_lock(); | |
511 | memcg_css = css_from_id(new_wb_id, &memory_cgrp_subsys); | |
512 | if (memcg_css) | |
513 | isw->new_wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
514 | rcu_read_unlock(); | |
515 | if (!isw->new_wb) | |
516 | goto out_free; | |
517 | ||
518 | /* while holding I_WB_SWITCH, no one else can update the association */ | |
519 | spin_lock(&inode->i_lock); | |
1751e8a6 | 520 | if (!(inode->i_sb->s_flags & SB_ACTIVE) || |
4ade5867 | 521 | inode->i_state & (I_WB_SWITCH | I_FREEING | I_WILL_FREE) || |
a1a0e23e TH |
522 | inode_to_wb(inode) == isw->new_wb) { |
523 | spin_unlock(&inode->i_lock); | |
524 | goto out_free; | |
525 | } | |
682aa8e1 | 526 | inode->i_state |= I_WB_SWITCH; |
74524955 | 527 | __iget(inode); |
682aa8e1 TH |
528 | spin_unlock(&inode->i_lock); |
529 | ||
682aa8e1 TH |
530 | isw->inode = inode; |
531 | ||
532 | /* | |
533 | * In addition to synchronizing among switchers, I_WB_SWITCH tells | |
b93b0163 MW |
534 | * the RCU protected stat update paths to grab the i_page |
535 | * lock so that stat transfer can synchronize against them. | |
682aa8e1 TH |
536 | * Let's continue after I_WB_SWITCH is guaranteed to be visible. |
537 | */ | |
29264d92 RG |
538 | INIT_RCU_WORK(&isw->work, inode_switch_wbs_work_fn); |
539 | queue_rcu_work(isw_wq, &isw->work); | |
6444f47e | 540 | return; |
682aa8e1 TH |
541 | |
542 | out_free: | |
8826ee4f | 543 | atomic_dec(&isw_nr_in_flight); |
682aa8e1 TH |
544 | if (isw->new_wb) |
545 | wb_put(isw->new_wb); | |
546 | kfree(isw); | |
547 | } | |
548 | ||
b16b1deb TH |
549 | /** |
550 | * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it | |
551 | * @wbc: writeback_control of interest | |
552 | * @inode: target inode | |
553 | * | |
554 | * @inode is locked and about to be written back under the control of @wbc. | |
555 | * Record @inode's writeback context into @wbc and unlock the i_lock. On | |
556 | * writeback completion, wbc_detach_inode() should be called. This is used | |
557 | * to track the cgroup writeback context. | |
558 | */ | |
559 | void wbc_attach_and_unlock_inode(struct writeback_control *wbc, | |
560 | struct inode *inode) | |
561 | { | |
dd73e4b7 TH |
562 | if (!inode_cgwb_enabled(inode)) { |
563 | spin_unlock(&inode->i_lock); | |
564 | return; | |
565 | } | |
566 | ||
b16b1deb | 567 | wbc->wb = inode_to_wb(inode); |
2a814908 TH |
568 | wbc->inode = inode; |
569 | ||
570 | wbc->wb_id = wbc->wb->memcg_css->id; | |
571 | wbc->wb_lcand_id = inode->i_wb_frn_winner; | |
572 | wbc->wb_tcand_id = 0; | |
573 | wbc->wb_bytes = 0; | |
574 | wbc->wb_lcand_bytes = 0; | |
575 | wbc->wb_tcand_bytes = 0; | |
576 | ||
b16b1deb TH |
577 | wb_get(wbc->wb); |
578 | spin_unlock(&inode->i_lock); | |
e8a7abf5 TH |
579 | |
580 | /* | |
65de03e2 TH |
581 | * A dying wb indicates that either the blkcg associated with the |
582 | * memcg changed or the associated memcg is dying. In the first | |
583 | * case, a replacement wb should already be available and we should | |
584 | * refresh the wb immediately. In the second case, trying to | |
585 | * refresh will keep failing. | |
e8a7abf5 | 586 | */ |
65de03e2 | 587 | if (unlikely(wb_dying(wbc->wb) && !css_is_dying(wbc->wb->memcg_css))) |
e8a7abf5 | 588 | inode_switch_wbs(inode, wbc->wb_id); |
b16b1deb | 589 | } |
9b0eb69b | 590 | EXPORT_SYMBOL_GPL(wbc_attach_and_unlock_inode); |
b16b1deb TH |
591 | |
592 | /** | |
2a814908 TH |
593 | * wbc_detach_inode - disassociate wbc from inode and perform foreign detection |
594 | * @wbc: writeback_control of the just finished writeback | |
b16b1deb TH |
595 | * |
596 | * To be called after a writeback attempt of an inode finishes and undoes | |
597 | * wbc_attach_and_unlock_inode(). Can be called under any context. | |
2a814908 TH |
598 | * |
599 | * As concurrent write sharing of an inode is expected to be very rare and | |
600 | * memcg only tracks page ownership on first-use basis severely confining | |
601 | * the usefulness of such sharing, cgroup writeback tracks ownership | |
602 | * per-inode. While the support for concurrent write sharing of an inode | |
603 | * is deemed unnecessary, an inode being written to by different cgroups at | |
604 | * different points in time is a lot more common, and, more importantly, | |
605 | * charging only by first-use can too readily lead to grossly incorrect | |
606 | * behaviors (single foreign page can lead to gigabytes of writeback to be | |
607 | * incorrectly attributed). | |
608 | * | |
609 | * To resolve this issue, cgroup writeback detects the majority dirtier of | |
610 | * an inode and transfers the ownership to it. To avoid unnnecessary | |
611 | * oscillation, the detection mechanism keeps track of history and gives | |
612 | * out the switch verdict only if the foreign usage pattern is stable over | |
613 | * a certain amount of time and/or writeback attempts. | |
614 | * | |
615 | * On each writeback attempt, @wbc tries to detect the majority writer | |
616 | * using Boyer-Moore majority vote algorithm. In addition to the byte | |
617 | * count from the majority voting, it also counts the bytes written for the | |
618 | * current wb and the last round's winner wb (max of last round's current | |
619 | * wb, the winner from two rounds ago, and the last round's majority | |
620 | * candidate). Keeping track of the historical winner helps the algorithm | |
621 | * to semi-reliably detect the most active writer even when it's not the | |
622 | * absolute majority. | |
623 | * | |
624 | * Once the winner of the round is determined, whether the winner is | |
625 | * foreign or not and how much IO time the round consumed is recorded in | |
626 | * inode->i_wb_frn_history. If the amount of recorded foreign IO time is | |
627 | * over a certain threshold, the switch verdict is given. | |
b16b1deb TH |
628 | */ |
629 | void wbc_detach_inode(struct writeback_control *wbc) | |
630 | { | |
2a814908 TH |
631 | struct bdi_writeback *wb = wbc->wb; |
632 | struct inode *inode = wbc->inode; | |
dd73e4b7 TH |
633 | unsigned long avg_time, max_bytes, max_time; |
634 | u16 history; | |
2a814908 TH |
635 | int max_id; |
636 | ||
dd73e4b7 TH |
637 | if (!wb) |
638 | return; | |
639 | ||
640 | history = inode->i_wb_frn_history; | |
641 | avg_time = inode->i_wb_frn_avg_time; | |
642 | ||
2a814908 TH |
643 | /* pick the winner of this round */ |
644 | if (wbc->wb_bytes >= wbc->wb_lcand_bytes && | |
645 | wbc->wb_bytes >= wbc->wb_tcand_bytes) { | |
646 | max_id = wbc->wb_id; | |
647 | max_bytes = wbc->wb_bytes; | |
648 | } else if (wbc->wb_lcand_bytes >= wbc->wb_tcand_bytes) { | |
649 | max_id = wbc->wb_lcand_id; | |
650 | max_bytes = wbc->wb_lcand_bytes; | |
651 | } else { | |
652 | max_id = wbc->wb_tcand_id; | |
653 | max_bytes = wbc->wb_tcand_bytes; | |
654 | } | |
655 | ||
656 | /* | |
657 | * Calculate the amount of IO time the winner consumed and fold it | |
658 | * into the running average kept per inode. If the consumed IO | |
659 | * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for | |
660 | * deciding whether to switch or not. This is to prevent one-off | |
661 | * small dirtiers from skewing the verdict. | |
662 | */ | |
663 | max_time = DIV_ROUND_UP((max_bytes >> PAGE_SHIFT) << WB_FRN_TIME_SHIFT, | |
664 | wb->avg_write_bandwidth); | |
665 | if (avg_time) | |
666 | avg_time += (max_time >> WB_FRN_TIME_AVG_SHIFT) - | |
667 | (avg_time >> WB_FRN_TIME_AVG_SHIFT); | |
668 | else | |
669 | avg_time = max_time; /* immediate catch up on first run */ | |
670 | ||
671 | if (max_time >= avg_time / WB_FRN_TIME_CUT_DIV) { | |
672 | int slots; | |
673 | ||
674 | /* | |
675 | * The switch verdict is reached if foreign wb's consume | |
676 | * more than a certain proportion of IO time in a | |
677 | * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot | |
678 | * history mask where each bit represents one sixteenth of | |
679 | * the period. Determine the number of slots to shift into | |
680 | * history from @max_time. | |
681 | */ | |
682 | slots = min(DIV_ROUND_UP(max_time, WB_FRN_HIST_UNIT), | |
683 | (unsigned long)WB_FRN_HIST_MAX_SLOTS); | |
684 | history <<= slots; | |
685 | if (wbc->wb_id != max_id) | |
686 | history |= (1U << slots) - 1; | |
687 | ||
3a8e9ac8 TH |
688 | if (history) |
689 | trace_inode_foreign_history(inode, wbc, history); | |
690 | ||
2a814908 TH |
691 | /* |
692 | * Switch if the current wb isn't the consistent winner. | |
693 | * If there are multiple closely competing dirtiers, the | |
694 | * inode may switch across them repeatedly over time, which | |
695 | * is okay. The main goal is avoiding keeping an inode on | |
696 | * the wrong wb for an extended period of time. | |
697 | */ | |
682aa8e1 TH |
698 | if (hweight32(history) > WB_FRN_HIST_THR_SLOTS) |
699 | inode_switch_wbs(inode, max_id); | |
2a814908 TH |
700 | } |
701 | ||
702 | /* | |
703 | * Multiple instances of this function may race to update the | |
704 | * following fields but we don't mind occassional inaccuracies. | |
705 | */ | |
706 | inode->i_wb_frn_winner = max_id; | |
707 | inode->i_wb_frn_avg_time = min(avg_time, (unsigned long)U16_MAX); | |
708 | inode->i_wb_frn_history = history; | |
709 | ||
b16b1deb TH |
710 | wb_put(wbc->wb); |
711 | wbc->wb = NULL; | |
712 | } | |
9b0eb69b | 713 | EXPORT_SYMBOL_GPL(wbc_detach_inode); |
b16b1deb | 714 | |
2a814908 | 715 | /** |
34e51a5e | 716 | * wbc_account_cgroup_owner - account writeback to update inode cgroup ownership |
2a814908 TH |
717 | * @wbc: writeback_control of the writeback in progress |
718 | * @page: page being written out | |
719 | * @bytes: number of bytes being written out | |
720 | * | |
721 | * @bytes from @page are about to written out during the writeback | |
722 | * controlled by @wbc. Keep the book for foreign inode detection. See | |
723 | * wbc_detach_inode(). | |
724 | */ | |
34e51a5e TH |
725 | void wbc_account_cgroup_owner(struct writeback_control *wbc, struct page *page, |
726 | size_t bytes) | |
2a814908 | 727 | { |
66311422 | 728 | struct cgroup_subsys_state *css; |
2a814908 TH |
729 | int id; |
730 | ||
731 | /* | |
732 | * pageout() path doesn't attach @wbc to the inode being written | |
733 | * out. This is intentional as we don't want the function to block | |
734 | * behind a slow cgroup. Ultimately, we want pageout() to kick off | |
735 | * regular writeback instead of writing things out itself. | |
736 | */ | |
27b36d8f | 737 | if (!wbc->wb || wbc->no_cgroup_owner) |
2a814908 TH |
738 | return; |
739 | ||
66311422 TH |
740 | css = mem_cgroup_css_from_page(page); |
741 | /* dead cgroups shouldn't contribute to inode ownership arbitration */ | |
742 | if (!(css->flags & CSS_ONLINE)) | |
743 | return; | |
744 | ||
745 | id = css->id; | |
2a814908 TH |
746 | |
747 | if (id == wbc->wb_id) { | |
748 | wbc->wb_bytes += bytes; | |
749 | return; | |
750 | } | |
751 | ||
752 | if (id == wbc->wb_lcand_id) | |
753 | wbc->wb_lcand_bytes += bytes; | |
754 | ||
755 | /* Boyer-Moore majority vote algorithm */ | |
756 | if (!wbc->wb_tcand_bytes) | |
757 | wbc->wb_tcand_id = id; | |
758 | if (id == wbc->wb_tcand_id) | |
759 | wbc->wb_tcand_bytes += bytes; | |
760 | else | |
761 | wbc->wb_tcand_bytes -= min(bytes, wbc->wb_tcand_bytes); | |
762 | } | |
34e51a5e | 763 | EXPORT_SYMBOL_GPL(wbc_account_cgroup_owner); |
2a814908 | 764 | |
703c2708 TH |
765 | /** |
766 | * inode_congested - test whether an inode is congested | |
60292bcc | 767 | * @inode: inode to test for congestion (may be NULL) |
703c2708 TH |
768 | * @cong_bits: mask of WB_[a]sync_congested bits to test |
769 | * | |
770 | * Tests whether @inode is congested. @cong_bits is the mask of congestion | |
771 | * bits to test and the return value is the mask of set bits. | |
772 | * | |
773 | * If cgroup writeback is enabled for @inode, the congestion state is | |
774 | * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg | |
775 | * associated with @inode is congested; otherwise, the root wb's congestion | |
776 | * state is used. | |
60292bcc TH |
777 | * |
778 | * @inode is allowed to be NULL as this function is often called on | |
779 | * mapping->host which is NULL for the swapper space. | |
703c2708 TH |
780 | */ |
781 | int inode_congested(struct inode *inode, int cong_bits) | |
782 | { | |
5cb8b824 TH |
783 | /* |
784 | * Once set, ->i_wb never becomes NULL while the inode is alive. | |
785 | * Start transaction iff ->i_wb is visible. | |
786 | */ | |
aaa2cacf | 787 | if (inode && inode_to_wb_is_valid(inode)) { |
5cb8b824 | 788 | struct bdi_writeback *wb; |
2e898e4c GT |
789 | struct wb_lock_cookie lock_cookie = {}; |
790 | bool congested; | |
5cb8b824 | 791 | |
2e898e4c | 792 | wb = unlocked_inode_to_wb_begin(inode, &lock_cookie); |
5cb8b824 | 793 | congested = wb_congested(wb, cong_bits); |
2e898e4c | 794 | unlocked_inode_to_wb_end(inode, &lock_cookie); |
5cb8b824 | 795 | return congested; |
703c2708 TH |
796 | } |
797 | ||
798 | return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); | |
799 | } | |
800 | EXPORT_SYMBOL_GPL(inode_congested); | |
801 | ||
f2b65121 TH |
802 | /** |
803 | * wb_split_bdi_pages - split nr_pages to write according to bandwidth | |
804 | * @wb: target bdi_writeback to split @nr_pages to | |
805 | * @nr_pages: number of pages to write for the whole bdi | |
806 | * | |
807 | * Split @wb's portion of @nr_pages according to @wb's write bandwidth in | |
808 | * relation to the total write bandwidth of all wb's w/ dirty inodes on | |
809 | * @wb->bdi. | |
810 | */ | |
811 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
812 | { | |
813 | unsigned long this_bw = wb->avg_write_bandwidth; | |
814 | unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); | |
815 | ||
816 | if (nr_pages == LONG_MAX) | |
817 | return LONG_MAX; | |
818 | ||
819 | /* | |
820 | * This may be called on clean wb's and proportional distribution | |
821 | * may not make sense, just use the original @nr_pages in those | |
822 | * cases. In general, we wanna err on the side of writing more. | |
823 | */ | |
824 | if (!tot_bw || this_bw >= tot_bw) | |
825 | return nr_pages; | |
826 | else | |
827 | return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); | |
828 | } | |
829 | ||
db125360 TH |
830 | /** |
831 | * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi | |
832 | * @bdi: target backing_dev_info | |
833 | * @base_work: wb_writeback_work to issue | |
834 | * @skip_if_busy: skip wb's which already have writeback in progress | |
835 | * | |
836 | * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which | |
837 | * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's | |
838 | * distributed to the busy wbs according to each wb's proportion in the | |
839 | * total active write bandwidth of @bdi. | |
840 | */ | |
841 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, | |
842 | struct wb_writeback_work *base_work, | |
843 | bool skip_if_busy) | |
844 | { | |
b817525a | 845 | struct bdi_writeback *last_wb = NULL; |
b33e18f6 TH |
846 | struct bdi_writeback *wb = list_entry(&bdi->wb_list, |
847 | struct bdi_writeback, bdi_node); | |
db125360 TH |
848 | |
849 | might_sleep(); | |
db125360 TH |
850 | restart: |
851 | rcu_read_lock(); | |
b817525a | 852 | list_for_each_entry_continue_rcu(wb, &bdi->wb_list, bdi_node) { |
5b9cce4c | 853 | DEFINE_WB_COMPLETION(fallback_work_done, bdi); |
8a1270cd TH |
854 | struct wb_writeback_work fallback_work; |
855 | struct wb_writeback_work *work; | |
856 | long nr_pages; | |
857 | ||
b817525a TH |
858 | if (last_wb) { |
859 | wb_put(last_wb); | |
860 | last_wb = NULL; | |
861 | } | |
862 | ||
006a0973 TH |
863 | /* SYNC_ALL writes out I_DIRTY_TIME too */ |
864 | if (!wb_has_dirty_io(wb) && | |
865 | (base_work->sync_mode == WB_SYNC_NONE || | |
866 | list_empty(&wb->b_dirty_time))) | |
867 | continue; | |
868 | if (skip_if_busy && writeback_in_progress(wb)) | |
db125360 TH |
869 | continue; |
870 | ||
8a1270cd TH |
871 | nr_pages = wb_split_bdi_pages(wb, base_work->nr_pages); |
872 | ||
873 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
874 | if (work) { | |
875 | *work = *base_work; | |
876 | work->nr_pages = nr_pages; | |
877 | work->auto_free = 1; | |
878 | wb_queue_work(wb, work); | |
879 | continue; | |
db125360 | 880 | } |
8a1270cd TH |
881 | |
882 | /* alloc failed, execute synchronously using on-stack fallback */ | |
883 | work = &fallback_work; | |
884 | *work = *base_work; | |
885 | work->nr_pages = nr_pages; | |
886 | work->auto_free = 0; | |
887 | work->done = &fallback_work_done; | |
888 | ||
889 | wb_queue_work(wb, work); | |
890 | ||
b817525a TH |
891 | /* |
892 | * Pin @wb so that it stays on @bdi->wb_list. This allows | |
893 | * continuing iteration from @wb after dropping and | |
894 | * regrabbing rcu read lock. | |
895 | */ | |
896 | wb_get(wb); | |
897 | last_wb = wb; | |
898 | ||
8a1270cd | 899 | rcu_read_unlock(); |
5b9cce4c | 900 | wb_wait_for_completion(&fallback_work_done); |
8a1270cd | 901 | goto restart; |
db125360 TH |
902 | } |
903 | rcu_read_unlock(); | |
b817525a TH |
904 | |
905 | if (last_wb) | |
906 | wb_put(last_wb); | |
db125360 TH |
907 | } |
908 | ||
d62241c7 TH |
909 | /** |
910 | * cgroup_writeback_by_id - initiate cgroup writeback from bdi and memcg IDs | |
911 | * @bdi_id: target bdi id | |
912 | * @memcg_id: target memcg css id | |
b46ec1da | 913 | * @nr: number of pages to write, 0 for best-effort dirty flushing |
d62241c7 TH |
914 | * @reason: reason why some writeback work initiated |
915 | * @done: target wb_completion | |
916 | * | |
917 | * Initiate flush of the bdi_writeback identified by @bdi_id and @memcg_id | |
918 | * with the specified parameters. | |
919 | */ | |
920 | int cgroup_writeback_by_id(u64 bdi_id, int memcg_id, unsigned long nr, | |
921 | enum wb_reason reason, struct wb_completion *done) | |
922 | { | |
923 | struct backing_dev_info *bdi; | |
924 | struct cgroup_subsys_state *memcg_css; | |
925 | struct bdi_writeback *wb; | |
926 | struct wb_writeback_work *work; | |
927 | int ret; | |
928 | ||
929 | /* lookup bdi and memcg */ | |
930 | bdi = bdi_get_by_id(bdi_id); | |
931 | if (!bdi) | |
932 | return -ENOENT; | |
933 | ||
934 | rcu_read_lock(); | |
935 | memcg_css = css_from_id(memcg_id, &memory_cgrp_subsys); | |
936 | if (memcg_css && !css_tryget(memcg_css)) | |
937 | memcg_css = NULL; | |
938 | rcu_read_unlock(); | |
939 | if (!memcg_css) { | |
940 | ret = -ENOENT; | |
941 | goto out_bdi_put; | |
942 | } | |
943 | ||
944 | /* | |
945 | * And find the associated wb. If the wb isn't there already | |
946 | * there's nothing to flush, don't create one. | |
947 | */ | |
948 | wb = wb_get_lookup(bdi, memcg_css); | |
949 | if (!wb) { | |
950 | ret = -ENOENT; | |
951 | goto out_css_put; | |
952 | } | |
953 | ||
954 | /* | |
955 | * If @nr is zero, the caller is attempting to write out most of | |
956 | * the currently dirty pages. Let's take the current dirty page | |
957 | * count and inflate it by 25% which should be large enough to | |
958 | * flush out most dirty pages while avoiding getting livelocked by | |
959 | * concurrent dirtiers. | |
960 | */ | |
961 | if (!nr) { | |
962 | unsigned long filepages, headroom, dirty, writeback; | |
963 | ||
964 | mem_cgroup_wb_stats(wb, &filepages, &headroom, &dirty, | |
965 | &writeback); | |
966 | nr = dirty * 10 / 8; | |
967 | } | |
968 | ||
969 | /* issue the writeback work */ | |
970 | work = kzalloc(sizeof(*work), GFP_NOWAIT | __GFP_NOWARN); | |
971 | if (work) { | |
972 | work->nr_pages = nr; | |
973 | work->sync_mode = WB_SYNC_NONE; | |
974 | work->range_cyclic = 1; | |
975 | work->reason = reason; | |
976 | work->done = done; | |
977 | work->auto_free = 1; | |
978 | wb_queue_work(wb, work); | |
979 | ret = 0; | |
980 | } else { | |
981 | ret = -ENOMEM; | |
982 | } | |
983 | ||
984 | wb_put(wb); | |
985 | out_css_put: | |
986 | css_put(memcg_css); | |
987 | out_bdi_put: | |
988 | bdi_put(bdi); | |
989 | return ret; | |
990 | } | |
991 | ||
a1a0e23e TH |
992 | /** |
993 | * cgroup_writeback_umount - flush inode wb switches for umount | |
994 | * | |
995 | * This function is called when a super_block is about to be destroyed and | |
996 | * flushes in-flight inode wb switches. An inode wb switch goes through | |
997 | * RCU and then workqueue, so the two need to be flushed in order to ensure | |
998 | * that all previously scheduled switches are finished. As wb switches are | |
999 | * rare occurrences and synchronize_rcu() can take a while, perform | |
1000 | * flushing iff wb switches are in flight. | |
1001 | */ | |
1002 | void cgroup_writeback_umount(void) | |
1003 | { | |
592fa002 RG |
1004 | /* |
1005 | * SB_ACTIVE should be reliably cleared before checking | |
1006 | * isw_nr_in_flight, see generic_shutdown_super(). | |
1007 | */ | |
1008 | smp_mb(); | |
1009 | ||
a1a0e23e | 1010 | if (atomic_read(&isw_nr_in_flight)) { |
ec084de9 JX |
1011 | /* |
1012 | * Use rcu_barrier() to wait for all pending callbacks to | |
1013 | * ensure that all in-flight wb switches are in the workqueue. | |
1014 | */ | |
1015 | rcu_barrier(); | |
a1a0e23e TH |
1016 | flush_workqueue(isw_wq); |
1017 | } | |
1018 | } | |
1019 | ||
1020 | static int __init cgroup_writeback_init(void) | |
1021 | { | |
1022 | isw_wq = alloc_workqueue("inode_switch_wbs", 0, 0); | |
1023 | if (!isw_wq) | |
1024 | return -ENOMEM; | |
1025 | return 0; | |
1026 | } | |
1027 | fs_initcall(cgroup_writeback_init); | |
1028 | ||
f2b65121 TH |
1029 | #else /* CONFIG_CGROUP_WRITEBACK */ |
1030 | ||
7fc5854f TH |
1031 | static void bdi_down_write_wb_switch_rwsem(struct backing_dev_info *bdi) { } |
1032 | static void bdi_up_write_wb_switch_rwsem(struct backing_dev_info *bdi) { } | |
1033 | ||
f3b6a6df RG |
1034 | static void inode_cgwb_move_to_attached(struct inode *inode, |
1035 | struct bdi_writeback *wb) | |
1036 | { | |
1037 | assert_spin_locked(&wb->list_lock); | |
1038 | assert_spin_locked(&inode->i_lock); | |
1039 | ||
1040 | inode->i_state &= ~I_SYNC_QUEUED; | |
1041 | list_del_init(&inode->i_io_list); | |
1042 | wb_io_lists_depopulated(wb); | |
1043 | } | |
1044 | ||
87e1d789 TH |
1045 | static struct bdi_writeback * |
1046 | locked_inode_to_wb_and_lock_list(struct inode *inode) | |
1047 | __releases(&inode->i_lock) | |
1048 | __acquires(&wb->list_lock) | |
1049 | { | |
1050 | struct bdi_writeback *wb = inode_to_wb(inode); | |
1051 | ||
1052 | spin_unlock(&inode->i_lock); | |
1053 | spin_lock(&wb->list_lock); | |
1054 | return wb; | |
1055 | } | |
1056 | ||
1057 | static struct bdi_writeback *inode_to_wb_and_lock_list(struct inode *inode) | |
1058 | __acquires(&wb->list_lock) | |
1059 | { | |
1060 | struct bdi_writeback *wb = inode_to_wb(inode); | |
1061 | ||
1062 | spin_lock(&wb->list_lock); | |
1063 | return wb; | |
1064 | } | |
1065 | ||
f2b65121 TH |
1066 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) |
1067 | { | |
1068 | return nr_pages; | |
1069 | } | |
1070 | ||
db125360 TH |
1071 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, |
1072 | struct wb_writeback_work *base_work, | |
1073 | bool skip_if_busy) | |
1074 | { | |
1075 | might_sleep(); | |
1076 | ||
006a0973 | 1077 | if (!skip_if_busy || !writeback_in_progress(&bdi->wb)) { |
db125360 | 1078 | base_work->auto_free = 0; |
db125360 TH |
1079 | wb_queue_work(&bdi->wb, base_work); |
1080 | } | |
1081 | } | |
1082 | ||
703c2708 TH |
1083 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
1084 | ||
e8e8a0c6 JA |
1085 | /* |
1086 | * Add in the number of potentially dirty inodes, because each inode | |
1087 | * write can dirty pagecache in the underlying blockdev. | |
1088 | */ | |
1089 | static unsigned long get_nr_dirty_pages(void) | |
1090 | { | |
1091 | return global_node_page_state(NR_FILE_DIRTY) + | |
e8e8a0c6 JA |
1092 | get_nr_dirty_inodes(); |
1093 | } | |
1094 | ||
1095 | static void wb_start_writeback(struct bdi_writeback *wb, enum wb_reason reason) | |
b6e51316 | 1096 | { |
c00ddad3 TH |
1097 | if (!wb_has_dirty_io(wb)) |
1098 | return; | |
1099 | ||
aac8d41c JA |
1100 | /* |
1101 | * All callers of this function want to start writeback of all | |
1102 | * dirty pages. Places like vmscan can call this at a very | |
1103 | * high frequency, causing pointless allocations of tons of | |
1104 | * work items and keeping the flusher threads busy retrieving | |
1105 | * that work. Ensure that we only allow one of them pending and | |
85009b4f | 1106 | * inflight at the time. |
aac8d41c | 1107 | */ |
85009b4f JA |
1108 | if (test_bit(WB_start_all, &wb->state) || |
1109 | test_and_set_bit(WB_start_all, &wb->state)) | |
aac8d41c JA |
1110 | return; |
1111 | ||
85009b4f JA |
1112 | wb->start_all_reason = reason; |
1113 | wb_wakeup(wb); | |
c5444198 | 1114 | } |
d3ddec76 | 1115 | |
c5444198 | 1116 | /** |
9ecf4866 TH |
1117 | * wb_start_background_writeback - start background writeback |
1118 | * @wb: bdi_writback to write from | |
c5444198 CH |
1119 | * |
1120 | * Description: | |
6585027a | 1121 | * This makes sure WB_SYNC_NONE background writeback happens. When |
9ecf4866 | 1122 | * this function returns, it is only guaranteed that for given wb |
6585027a JK |
1123 | * some IO is happening if we are over background dirty threshold. |
1124 | * Caller need not hold sb s_umount semaphore. | |
c5444198 | 1125 | */ |
9ecf4866 | 1126 | void wb_start_background_writeback(struct bdi_writeback *wb) |
c5444198 | 1127 | { |
6585027a JK |
1128 | /* |
1129 | * We just wake up the flusher thread. It will perform background | |
1130 | * writeback as soon as there is no other work to do. | |
1131 | */ | |
5634cc2a | 1132 | trace_writeback_wake_background(wb); |
9ecf4866 | 1133 | wb_wakeup(wb); |
1da177e4 LT |
1134 | } |
1135 | ||
a66979ab DC |
1136 | /* |
1137 | * Remove the inode from the writeback list it is on. | |
1138 | */ | |
c7f54084 | 1139 | void inode_io_list_del(struct inode *inode) |
a66979ab | 1140 | { |
87e1d789 | 1141 | struct bdi_writeback *wb; |
f758eeab | 1142 | |
87e1d789 | 1143 | wb = inode_to_wb_and_lock_list(inode); |
b35250c0 | 1144 | spin_lock(&inode->i_lock); |
f3b6a6df RG |
1145 | |
1146 | inode->i_state &= ~I_SYNC_QUEUED; | |
1147 | list_del_init(&inode->i_io_list); | |
1148 | wb_io_lists_depopulated(wb); | |
1149 | ||
b35250c0 | 1150 | spin_unlock(&inode->i_lock); |
52ebea74 | 1151 | spin_unlock(&wb->list_lock); |
a66979ab | 1152 | } |
4301efa4 | 1153 | EXPORT_SYMBOL(inode_io_list_del); |
a66979ab | 1154 | |
6c60d2b5 DC |
1155 | /* |
1156 | * mark an inode as under writeback on the sb | |
1157 | */ | |
1158 | void sb_mark_inode_writeback(struct inode *inode) | |
1159 | { | |
1160 | struct super_block *sb = inode->i_sb; | |
1161 | unsigned long flags; | |
1162 | ||
1163 | if (list_empty(&inode->i_wb_list)) { | |
1164 | spin_lock_irqsave(&sb->s_inode_wblist_lock, flags); | |
9a46b04f | 1165 | if (list_empty(&inode->i_wb_list)) { |
6c60d2b5 | 1166 | list_add_tail(&inode->i_wb_list, &sb->s_inodes_wb); |
9a46b04f BF |
1167 | trace_sb_mark_inode_writeback(inode); |
1168 | } | |
6c60d2b5 DC |
1169 | spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags); |
1170 | } | |
1171 | } | |
1172 | ||
1173 | /* | |
1174 | * clear an inode as under writeback on the sb | |
1175 | */ | |
1176 | void sb_clear_inode_writeback(struct inode *inode) | |
1177 | { | |
1178 | struct super_block *sb = inode->i_sb; | |
1179 | unsigned long flags; | |
1180 | ||
1181 | if (!list_empty(&inode->i_wb_list)) { | |
1182 | spin_lock_irqsave(&sb->s_inode_wblist_lock, flags); | |
9a46b04f BF |
1183 | if (!list_empty(&inode->i_wb_list)) { |
1184 | list_del_init(&inode->i_wb_list); | |
1185 | trace_sb_clear_inode_writeback(inode); | |
1186 | } | |
6c60d2b5 DC |
1187 | spin_unlock_irqrestore(&sb->s_inode_wblist_lock, flags); |
1188 | } | |
1189 | } | |
1190 | ||
6610a0bc AM |
1191 | /* |
1192 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
1193 | * furthest end of its superblock's dirty-inode list. | |
1194 | * | |
1195 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 1196 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
1197 | * the case then the inode must have been redirtied while it was being written |
1198 | * out and we don't reset its dirtied_when. | |
1199 | */ | |
b35250c0 | 1200 | static void redirty_tail_locked(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 1201 | { |
b35250c0 JK |
1202 | assert_spin_locked(&inode->i_lock); |
1203 | ||
03ba3782 | 1204 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 1205 | struct inode *tail; |
6610a0bc | 1206 | |
7ccf19a8 | 1207 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 1208 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
1209 | inode->dirtied_when = jiffies; |
1210 | } | |
c7f54084 | 1211 | inode_io_list_move_locked(inode, wb, &wb->b_dirty); |
5afced3b | 1212 | inode->i_state &= ~I_SYNC_QUEUED; |
6610a0bc AM |
1213 | } |
1214 | ||
b35250c0 JK |
1215 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
1216 | { | |
1217 | spin_lock(&inode->i_lock); | |
1218 | redirty_tail_locked(inode, wb); | |
1219 | spin_unlock(&inode->i_lock); | |
1220 | } | |
1221 | ||
c986d1e2 | 1222 | /* |
66f3b8e2 | 1223 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 1224 | */ |
f758eeab | 1225 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 1226 | { |
c7f54084 | 1227 | inode_io_list_move_locked(inode, wb, &wb->b_more_io); |
c986d1e2 AM |
1228 | } |
1229 | ||
1c0eeaf5 JE |
1230 | static void inode_sync_complete(struct inode *inode) |
1231 | { | |
365b94ae | 1232 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
1233 | /* If inode is clean an unused, put it into LRU now... */ |
1234 | inode_add_lru(inode); | |
365b94ae | 1235 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
1236 | smp_mb(); |
1237 | wake_up_bit(&inode->i_state, __I_SYNC); | |
1238 | } | |
1239 | ||
d2caa3c5 JL |
1240 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
1241 | { | |
1242 | bool ret = time_after(inode->dirtied_when, t); | |
1243 | #ifndef CONFIG_64BIT | |
1244 | /* | |
1245 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
1246 | * It _appears_ to be in the future, but is actually in distant past. | |
1247 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 1248 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
1249 | */ |
1250 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
1251 | #endif | |
1252 | return ret; | |
1253 | } | |
1254 | ||
0ae45f63 TT |
1255 | #define EXPIRE_DIRTY_ATIME 0x0001 |
1256 | ||
2c136579 | 1257 | /* |
f9cae926 | 1258 | * Move expired (dirtied before dirtied_before) dirty inodes from |
697e6fed | 1259 | * @delaying_queue to @dispatch_queue. |
2c136579 | 1260 | */ |
e84d0a4f | 1261 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 1262 | struct list_head *dispatch_queue, |
5fcd5750 | 1263 | unsigned long dirtied_before) |
2c136579 | 1264 | { |
5c03449d SL |
1265 | LIST_HEAD(tmp); |
1266 | struct list_head *pos, *node; | |
cf137307 | 1267 | struct super_block *sb = NULL; |
5c03449d | 1268 | struct inode *inode; |
cf137307 | 1269 | int do_sb_sort = 0; |
e84d0a4f | 1270 | int moved = 0; |
5c03449d | 1271 | |
2c136579 | 1272 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 1273 | inode = wb_inode(delaying_queue->prev); |
f9cae926 | 1274 | if (inode_dirtied_after(inode, dirtied_before)) |
2c136579 | 1275 | break; |
c7f54084 | 1276 | list_move(&inode->i_io_list, &tmp); |
a8855990 | 1277 | moved++; |
5afced3b | 1278 | spin_lock(&inode->i_lock); |
5afced3b JK |
1279 | inode->i_state |= I_SYNC_QUEUED; |
1280 | spin_unlock(&inode->i_lock); | |
a8855990 JK |
1281 | if (sb_is_blkdev_sb(inode->i_sb)) |
1282 | continue; | |
cf137307 JA |
1283 | if (sb && sb != inode->i_sb) |
1284 | do_sb_sort = 1; | |
1285 | sb = inode->i_sb; | |
5c03449d SL |
1286 | } |
1287 | ||
cf137307 JA |
1288 | /* just one sb in list, splice to dispatch_queue and we're done */ |
1289 | if (!do_sb_sort) { | |
1290 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 1291 | goto out; |
cf137307 JA |
1292 | } |
1293 | ||
5c03449d SL |
1294 | /* Move inodes from one superblock together */ |
1295 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 1296 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 1297 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 1298 | inode = wb_inode(pos); |
5c03449d | 1299 | if (inode->i_sb == sb) |
c7f54084 | 1300 | list_move(&inode->i_io_list, dispatch_queue); |
5c03449d | 1301 | } |
2c136579 | 1302 | } |
e84d0a4f WF |
1303 | out: |
1304 | return moved; | |
2c136579 FW |
1305 | } |
1306 | ||
1307 | /* | |
1308 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
1309 | * Before |
1310 | * newly dirtied b_dirty b_io b_more_io | |
1311 | * =============> gf edc BA | |
1312 | * After | |
1313 | * newly dirtied b_dirty b_io b_more_io | |
1314 | * =============> g fBAedc | |
1315 | * | | |
1316 | * +--> dequeue for IO | |
2c136579 | 1317 | */ |
f9cae926 JK |
1318 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work, |
1319 | unsigned long dirtied_before) | |
66f3b8e2 | 1320 | { |
e84d0a4f | 1321 | int moved; |
f9cae926 | 1322 | unsigned long time_expire_jif = dirtied_before; |
0ae45f63 | 1323 | |
f758eeab | 1324 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 1325 | list_splice_init(&wb->b_more_io, &wb->b_io); |
5fcd5750 | 1326 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, dirtied_before); |
f9cae926 JK |
1327 | if (!work->for_sync) |
1328 | time_expire_jif = jiffies - dirtytime_expire_interval * HZ; | |
0ae45f63 | 1329 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, |
5fcd5750 | 1330 | time_expire_jif); |
d6c10f1f TH |
1331 | if (moved) |
1332 | wb_io_lists_populated(wb); | |
f9cae926 | 1333 | trace_writeback_queue_io(wb, work, dirtied_before, moved); |
66f3b8e2 JA |
1334 | } |
1335 | ||
a9185b41 | 1336 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 1337 | { |
9fb0a7da TH |
1338 | int ret; |
1339 | ||
1340 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
1341 | trace_writeback_write_inode_start(inode, wbc); | |
1342 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
1343 | trace_writeback_write_inode(inode, wbc); | |
1344 | return ret; | |
1345 | } | |
03ba3782 | 1346 | return 0; |
08d8e974 | 1347 | } |
08d8e974 | 1348 | |
1da177e4 | 1349 | /* |
169ebd90 JK |
1350 | * Wait for writeback on an inode to complete. Called with i_lock held. |
1351 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 1352 | */ |
169ebd90 JK |
1353 | static void __inode_wait_for_writeback(struct inode *inode) |
1354 | __releases(inode->i_lock) | |
1355 | __acquires(inode->i_lock) | |
01c03194 CH |
1356 | { |
1357 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
1358 | wait_queue_head_t *wqh; | |
1359 | ||
1360 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
1361 | while (inode->i_state & I_SYNC) { |
1362 | spin_unlock(&inode->i_lock); | |
74316201 N |
1363 | __wait_on_bit(wqh, &wq, bit_wait, |
1364 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 1365 | spin_lock(&inode->i_lock); |
58a9d3d8 | 1366 | } |
01c03194 CH |
1367 | } |
1368 | ||
169ebd90 JK |
1369 | /* |
1370 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
1371 | */ | |
1372 | void inode_wait_for_writeback(struct inode *inode) | |
1373 | { | |
1374 | spin_lock(&inode->i_lock); | |
1375 | __inode_wait_for_writeback(inode); | |
1376 | spin_unlock(&inode->i_lock); | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
1381 | * held and drops it. It is aimed for callers not holding any inode reference | |
1382 | * so once i_lock is dropped, inode can go away. | |
1383 | */ | |
1384 | static void inode_sleep_on_writeback(struct inode *inode) | |
1385 | __releases(inode->i_lock) | |
1386 | { | |
1387 | DEFINE_WAIT(wait); | |
1388 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
1389 | int sleep; | |
1390 | ||
1391 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1392 | sleep = inode->i_state & I_SYNC; | |
1393 | spin_unlock(&inode->i_lock); | |
1394 | if (sleep) | |
1395 | schedule(); | |
1396 | finish_wait(wqh, &wait); | |
1397 | } | |
1398 | ||
ccb26b5a JK |
1399 | /* |
1400 | * Find proper writeback list for the inode depending on its current state and | |
1401 | * possibly also change of its state while we were doing writeback. Here we | |
1402 | * handle things such as livelock prevention or fairness of writeback among | |
1403 | * inodes. This function can be called only by flusher thread - noone else | |
1404 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
1405 | * thread's back can have unexpected consequences. | |
1406 | */ | |
1407 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
1408 | struct writeback_control *wbc) | |
1409 | { | |
1410 | if (inode->i_state & I_FREEING) | |
1411 | return; | |
1412 | ||
1413 | /* | |
1414 | * Sync livelock prevention. Each inode is tagged and synced in one | |
1415 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
1416 | * the dirty time to prevent enqueue and sync it again. | |
1417 | */ | |
1418 | if ((inode->i_state & I_DIRTY) && | |
1419 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
1420 | inode->dirtied_when = jiffies; | |
1421 | ||
4f8ad655 JK |
1422 | if (wbc->pages_skipped) { |
1423 | /* | |
1424 | * writeback is not making progress due to locked | |
1425 | * buffers. Skip this inode for now. | |
1426 | */ | |
b35250c0 | 1427 | redirty_tail_locked(inode, wb); |
4f8ad655 JK |
1428 | return; |
1429 | } | |
1430 | ||
ccb26b5a JK |
1431 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
1432 | /* | |
1433 | * We didn't write back all the pages. nfs_writepages() | |
1434 | * sometimes bales out without doing anything. | |
1435 | */ | |
1436 | if (wbc->nr_to_write <= 0) { | |
1437 | /* Slice used up. Queue for next turn. */ | |
1438 | requeue_io(inode, wb); | |
1439 | } else { | |
1440 | /* | |
1441 | * Writeback blocked by something other than | |
1442 | * congestion. Delay the inode for some time to | |
1443 | * avoid spinning on the CPU (100% iowait) | |
1444 | * retrying writeback of the dirty page/inode | |
1445 | * that cannot be performed immediately. | |
1446 | */ | |
b35250c0 | 1447 | redirty_tail_locked(inode, wb); |
ccb26b5a JK |
1448 | } |
1449 | } else if (inode->i_state & I_DIRTY) { | |
1450 | /* | |
1451 | * Filesystems can dirty the inode during writeback operations, | |
1452 | * such as delayed allocation during submission or metadata | |
1453 | * updates after data IO completion. | |
1454 | */ | |
b35250c0 | 1455 | redirty_tail_locked(inode, wb); |
0ae45f63 | 1456 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 1457 | inode->dirtied_when = jiffies; |
c7f54084 | 1458 | inode_io_list_move_locked(inode, wb, &wb->b_dirty_time); |
5afced3b | 1459 | inode->i_state &= ~I_SYNC_QUEUED; |
ccb26b5a JK |
1460 | } else { |
1461 | /* The inode is clean. Remove from writeback lists. */ | |
f3b6a6df | 1462 | inode_cgwb_move_to_attached(inode, wb); |
ccb26b5a JK |
1463 | } |
1464 | } | |
1465 | ||
01c03194 | 1466 | /* |
da0c4c60 EB |
1467 | * Write out an inode and its dirty pages (or some of its dirty pages, depending |
1468 | * on @wbc->nr_to_write), and clear the relevant dirty flags from i_state. | |
1469 | * | |
1470 | * This doesn't remove the inode from the writeback list it is on, except | |
1471 | * potentially to move it from b_dirty_time to b_dirty due to timestamp | |
1472 | * expiration. The caller is otherwise responsible for writeback list handling. | |
1473 | * | |
1474 | * The caller is also responsible for setting the I_SYNC flag beforehand and | |
1475 | * calling inode_sync_complete() to clear it afterwards. | |
1da177e4 LT |
1476 | */ |
1477 | static int | |
cd8ed2a4 | 1478 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 1479 | { |
1da177e4 | 1480 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 1481 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 1482 | unsigned dirty; |
1da177e4 LT |
1483 | int ret; |
1484 | ||
4f8ad655 | 1485 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 1486 | |
9fb0a7da TH |
1487 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
1488 | ||
1da177e4 LT |
1489 | ret = do_writepages(mapping, wbc); |
1490 | ||
26821ed4 CH |
1491 | /* |
1492 | * Make sure to wait on the data before writing out the metadata. | |
1493 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
1494 | * I/O completion. We don't do it for sync(2) writeback because it has a |
1495 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
1496 | * inode metadata is written back correctly. | |
26821ed4 | 1497 | */ |
7747bd4b | 1498 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 1499 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
1500 | if (ret == 0) |
1501 | ret = err; | |
1502 | } | |
1503 | ||
5547e8aa | 1504 | /* |
1e249cb5 EB |
1505 | * If the inode has dirty timestamps and we need to write them, call |
1506 | * mark_inode_dirty_sync() to notify the filesystem about it and to | |
1507 | * change I_DIRTY_TIME into I_DIRTY_SYNC. | |
5547e8aa | 1508 | */ |
5fcd5750 | 1509 | if ((inode->i_state & I_DIRTY_TIME) && |
83dc881d | 1510 | (wbc->sync_mode == WB_SYNC_ALL || |
5fcd5750 JK |
1511 | time_after(jiffies, inode->dirtied_time_when + |
1512 | dirtytime_expire_interval * HZ))) { | |
5fcd5750 | 1513 | trace_writeback_lazytime(inode); |
1e249cb5 | 1514 | mark_inode_dirty_sync(inode); |
5fcd5750 | 1515 | } |
1e249cb5 EB |
1516 | |
1517 | /* | |
da0c4c60 EB |
1518 | * Get and clear the dirty flags from i_state. This needs to be done |
1519 | * after calling writepages because some filesystems may redirty the | |
1520 | * inode during writepages due to delalloc. It also needs to be done | |
1521 | * after handling timestamp expiration, as that may dirty the inode too. | |
1e249cb5 EB |
1522 | */ |
1523 | spin_lock(&inode->i_lock); | |
1524 | dirty = inode->i_state & I_DIRTY; | |
0ae45f63 | 1525 | inode->i_state &= ~dirty; |
9c6ac78e TH |
1526 | |
1527 | /* | |
1528 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
1529 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
1530 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
1531 | * inode. | |
1532 | * | |
1533 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
1534 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
1535 | * necessary. This guarantees that either __mark_inode_dirty() | |
1536 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
1537 | */ | |
1538 | smp_mb(); | |
1539 | ||
1540 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
1541 | inode->i_state |= I_DIRTY_PAGES; | |
1542 | ||
250df6ed | 1543 | spin_unlock(&inode->i_lock); |
9c6ac78e | 1544 | |
26821ed4 | 1545 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 1546 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 1547 | int err = write_inode(inode, wbc); |
1da177e4 LT |
1548 | if (ret == 0) |
1549 | ret = err; | |
1550 | } | |
4f8ad655 JK |
1551 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
1552 | return ret; | |
1553 | } | |
1554 | ||
1555 | /* | |
da0c4c60 EB |
1556 | * Write out an inode's dirty data and metadata on-demand, i.e. separately from |
1557 | * the regular batched writeback done by the flusher threads in | |
1558 | * writeback_sb_inodes(). @wbc controls various aspects of the write, such as | |
1559 | * whether it is a data-integrity sync (%WB_SYNC_ALL) or not (%WB_SYNC_NONE). | |
4f8ad655 | 1560 | * |
da0c4c60 EB |
1561 | * To prevent the inode from going away, either the caller must have a reference |
1562 | * to the inode, or the inode must have I_WILL_FREE or I_FREEING set. | |
4f8ad655 | 1563 | */ |
aaf25593 TH |
1564 | static int writeback_single_inode(struct inode *inode, |
1565 | struct writeback_control *wbc) | |
4f8ad655 | 1566 | { |
aaf25593 | 1567 | struct bdi_writeback *wb; |
4f8ad655 JK |
1568 | int ret = 0; |
1569 | ||
1570 | spin_lock(&inode->i_lock); | |
1571 | if (!atomic_read(&inode->i_count)) | |
1572 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
1573 | else | |
1574 | WARN_ON(inode->i_state & I_WILL_FREE); | |
1575 | ||
1576 | if (inode->i_state & I_SYNC) { | |
4f8ad655 | 1577 | /* |
da0c4c60 EB |
1578 | * Writeback is already running on the inode. For WB_SYNC_NONE, |
1579 | * that's enough and we can just return. For WB_SYNC_ALL, we | |
1580 | * must wait for the existing writeback to complete, then do | |
1581 | * writeback again if there's anything left. | |
4f8ad655 | 1582 | */ |
da0c4c60 EB |
1583 | if (wbc->sync_mode != WB_SYNC_ALL) |
1584 | goto out; | |
169ebd90 | 1585 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
1586 | } |
1587 | WARN_ON(inode->i_state & I_SYNC); | |
1588 | /* | |
da0c4c60 EB |
1589 | * If the inode is already fully clean, then there's nothing to do. |
1590 | * | |
1591 | * For data-integrity syncs we also need to check whether any pages are | |
1592 | * still under writeback, e.g. due to prior WB_SYNC_NONE writeback. If | |
1593 | * there are any such pages, we'll need to wait for them. | |
4f8ad655 | 1594 | */ |
0ae45f63 | 1595 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
1596 | (wbc->sync_mode != WB_SYNC_ALL || |
1597 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
1598 | goto out; |
1599 | inode->i_state |= I_SYNC; | |
b16b1deb | 1600 | wbc_attach_and_unlock_inode(wbc, inode); |
4f8ad655 | 1601 | |
cd8ed2a4 | 1602 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 1603 | |
b16b1deb | 1604 | wbc_detach_inode(wbc); |
aaf25593 TH |
1605 | |
1606 | wb = inode_to_wb_and_lock_list(inode); | |
250df6ed | 1607 | spin_lock(&inode->i_lock); |
4f8ad655 | 1608 | /* |
da0c4c60 EB |
1609 | * If the inode is now fully clean, then it can be safely removed from |
1610 | * its writeback list (if any). Otherwise the flusher threads are | |
1611 | * responsible for the writeback lists. | |
4f8ad655 | 1612 | */ |
0ae45f63 | 1613 | if (!(inode->i_state & I_DIRTY_ALL)) |
f3b6a6df | 1614 | inode_cgwb_move_to_attached(inode, wb); |
4f8ad655 | 1615 | spin_unlock(&wb->list_lock); |
1c0eeaf5 | 1616 | inode_sync_complete(inode); |
4f8ad655 JK |
1617 | out: |
1618 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1619 | return ret; |
1620 | } | |
1621 | ||
a88a341a | 1622 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 1623 | struct wb_writeback_work *work) |
d46db3d5 WF |
1624 | { |
1625 | long pages; | |
1626 | ||
1627 | /* | |
1628 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
1629 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
1630 | * here avoids calling into writeback_inodes_wb() more than once. | |
1631 | * | |
1632 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
1633 | * | |
1634 | * wb_writeback() | |
1635 | * writeback_sb_inodes() <== called only once | |
1636 | * write_cache_pages() <== called once for each inode | |
1637 | * (quickly) tag currently dirty pages | |
1638 | * (maybe slowly) sync all tagged pages | |
1639 | */ | |
1640 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
1641 | pages = LONG_MAX; | |
1a12d8bd | 1642 | else { |
a88a341a | 1643 | pages = min(wb->avg_write_bandwidth / 2, |
dcc25ae7 | 1644 | global_wb_domain.dirty_limit / DIRTY_SCOPE); |
1a12d8bd WF |
1645 | pages = min(pages, work->nr_pages); |
1646 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
1647 | MIN_WRITEBACK_PAGES); | |
1648 | } | |
d46db3d5 WF |
1649 | |
1650 | return pages; | |
1651 | } | |
1652 | ||
f11c9c5c ES |
1653 | /* |
1654 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 1655 | * |
d46db3d5 | 1656 | * Return the number of pages and/or inodes written. |
0ba13fd1 LT |
1657 | * |
1658 | * NOTE! This is called with wb->list_lock held, and will | |
1659 | * unlock and relock that for each inode it ends up doing | |
1660 | * IO for. | |
f11c9c5c | 1661 | */ |
d46db3d5 WF |
1662 | static long writeback_sb_inodes(struct super_block *sb, |
1663 | struct bdi_writeback *wb, | |
1664 | struct wb_writeback_work *work) | |
1da177e4 | 1665 | { |
d46db3d5 WF |
1666 | struct writeback_control wbc = { |
1667 | .sync_mode = work->sync_mode, | |
1668 | .tagged_writepages = work->tagged_writepages, | |
1669 | .for_kupdate = work->for_kupdate, | |
1670 | .for_background = work->for_background, | |
7747bd4b | 1671 | .for_sync = work->for_sync, |
d46db3d5 WF |
1672 | .range_cyclic = work->range_cyclic, |
1673 | .range_start = 0, | |
1674 | .range_end = LLONG_MAX, | |
1675 | }; | |
1676 | unsigned long start_time = jiffies; | |
1677 | long write_chunk; | |
1678 | long wrote = 0; /* count both pages and inodes */ | |
1679 | ||
03ba3782 | 1680 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 1681 | struct inode *inode = wb_inode(wb->b_io.prev); |
aaf25593 | 1682 | struct bdi_writeback *tmp_wb; |
edadfb10 CH |
1683 | |
1684 | if (inode->i_sb != sb) { | |
d46db3d5 | 1685 | if (work->sb) { |
edadfb10 CH |
1686 | /* |
1687 | * We only want to write back data for this | |
1688 | * superblock, move all inodes not belonging | |
1689 | * to it back onto the dirty list. | |
1690 | */ | |
f758eeab | 1691 | redirty_tail(inode, wb); |
edadfb10 CH |
1692 | continue; |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * The inode belongs to a different superblock. | |
1697 | * Bounce back to the caller to unpin this and | |
1698 | * pin the next superblock. | |
1699 | */ | |
d46db3d5 | 1700 | break; |
edadfb10 CH |
1701 | } |
1702 | ||
9843b76a | 1703 | /* |
331cbdee WL |
1704 | * Don't bother with new inodes or inodes being freed, first |
1705 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
1706 | * kind writeout is handled by the freer. |
1707 | */ | |
250df6ed | 1708 | spin_lock(&inode->i_lock); |
9843b76a | 1709 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
b35250c0 | 1710 | redirty_tail_locked(inode, wb); |
250df6ed | 1711 | spin_unlock(&inode->i_lock); |
7ef0d737 NP |
1712 | continue; |
1713 | } | |
cc1676d9 JK |
1714 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
1715 | /* | |
1716 | * If this inode is locked for writeback and we are not | |
1717 | * doing writeback-for-data-integrity, move it to | |
1718 | * b_more_io so that writeback can proceed with the | |
1719 | * other inodes on s_io. | |
1720 | * | |
1721 | * We'll have another go at writing back this inode | |
1722 | * when we completed a full scan of b_io. | |
1723 | */ | |
1724 | spin_unlock(&inode->i_lock); | |
1725 | requeue_io(inode, wb); | |
1726 | trace_writeback_sb_inodes_requeue(inode); | |
1727 | continue; | |
1728 | } | |
f0d07b7f JK |
1729 | spin_unlock(&wb->list_lock); |
1730 | ||
4f8ad655 JK |
1731 | /* |
1732 | * We already requeued the inode if it had I_SYNC set and we | |
1733 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
1734 | * WB_SYNC_ALL case. | |
1735 | */ | |
169ebd90 JK |
1736 | if (inode->i_state & I_SYNC) { |
1737 | /* Wait for I_SYNC. This function drops i_lock... */ | |
1738 | inode_sleep_on_writeback(inode); | |
1739 | /* Inode may be gone, start again */ | |
ead188f9 | 1740 | spin_lock(&wb->list_lock); |
169ebd90 JK |
1741 | continue; |
1742 | } | |
4f8ad655 | 1743 | inode->i_state |= I_SYNC; |
b16b1deb | 1744 | wbc_attach_and_unlock_inode(&wbc, inode); |
169ebd90 | 1745 | |
a88a341a | 1746 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
1747 | wbc.nr_to_write = write_chunk; |
1748 | wbc.pages_skipped = 0; | |
250df6ed | 1749 | |
169ebd90 JK |
1750 | /* |
1751 | * We use I_SYNC to pin the inode in memory. While it is set | |
1752 | * evict_inode() will wait so the inode cannot be freed. | |
1753 | */ | |
cd8ed2a4 | 1754 | __writeback_single_inode(inode, &wbc); |
250df6ed | 1755 | |
b16b1deb | 1756 | wbc_detach_inode(&wbc); |
d46db3d5 WF |
1757 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
1758 | wrote += write_chunk - wbc.nr_to_write; | |
590dca3a CM |
1759 | |
1760 | if (need_resched()) { | |
1761 | /* | |
1762 | * We're trying to balance between building up a nice | |
1763 | * long list of IOs to improve our merge rate, and | |
1764 | * getting those IOs out quickly for anyone throttling | |
1765 | * in balance_dirty_pages(). cond_resched() doesn't | |
1766 | * unplug, so get our IOs out the door before we | |
1767 | * give up the CPU. | |
1768 | */ | |
1769 | blk_flush_plug(current); | |
1770 | cond_resched(); | |
1771 | } | |
1772 | ||
aaf25593 TH |
1773 | /* |
1774 | * Requeue @inode if still dirty. Be careful as @inode may | |
1775 | * have been switched to another wb in the meantime. | |
1776 | */ | |
1777 | tmp_wb = inode_to_wb_and_lock_list(inode); | |
4f8ad655 | 1778 | spin_lock(&inode->i_lock); |
0ae45f63 | 1779 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 1780 | wrote++; |
aaf25593 | 1781 | requeue_inode(inode, tmp_wb, &wbc); |
4f8ad655 | 1782 | inode_sync_complete(inode); |
0f1b1fd8 | 1783 | spin_unlock(&inode->i_lock); |
590dca3a | 1784 | |
aaf25593 TH |
1785 | if (unlikely(tmp_wb != wb)) { |
1786 | spin_unlock(&tmp_wb->list_lock); | |
1787 | spin_lock(&wb->list_lock); | |
1788 | } | |
1789 | ||
d46db3d5 WF |
1790 | /* |
1791 | * bail out to wb_writeback() often enough to check | |
1792 | * background threshold and other termination conditions. | |
1793 | */ | |
1794 | if (wrote) { | |
1795 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1796 | break; | |
1797 | if (work->nr_pages <= 0) | |
1798 | break; | |
8bc3be27 | 1799 | } |
1da177e4 | 1800 | } |
d46db3d5 | 1801 | return wrote; |
f11c9c5c ES |
1802 | } |
1803 | ||
d46db3d5 WF |
1804 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
1805 | struct wb_writeback_work *work) | |
f11c9c5c | 1806 | { |
d46db3d5 WF |
1807 | unsigned long start_time = jiffies; |
1808 | long wrote = 0; | |
38f21977 | 1809 | |
f11c9c5c | 1810 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 1811 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 1812 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 1813 | |
eb6ef3df | 1814 | if (!trylock_super(sb)) { |
0e995816 | 1815 | /* |
eb6ef3df | 1816 | * trylock_super() may fail consistently due to |
0e995816 WF |
1817 | * s_umount being grabbed by someone else. Don't use |
1818 | * requeue_io() to avoid busy retrying the inode/sb. | |
1819 | */ | |
1820 | redirty_tail(inode, wb); | |
edadfb10 | 1821 | continue; |
f11c9c5c | 1822 | } |
d46db3d5 | 1823 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 1824 | up_read(&sb->s_umount); |
f11c9c5c | 1825 | |
d46db3d5 WF |
1826 | /* refer to the same tests at the end of writeback_sb_inodes */ |
1827 | if (wrote) { | |
1828 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1829 | break; | |
1830 | if (work->nr_pages <= 0) | |
1831 | break; | |
1832 | } | |
f11c9c5c | 1833 | } |
66f3b8e2 | 1834 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 1835 | return wrote; |
66f3b8e2 JA |
1836 | } |
1837 | ||
7d9f073b | 1838 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 1839 | enum wb_reason reason) |
edadfb10 | 1840 | { |
d46db3d5 WF |
1841 | struct wb_writeback_work work = { |
1842 | .nr_pages = nr_pages, | |
1843 | .sync_mode = WB_SYNC_NONE, | |
1844 | .range_cyclic = 1, | |
0e175a18 | 1845 | .reason = reason, |
d46db3d5 | 1846 | }; |
505a666e | 1847 | struct blk_plug plug; |
edadfb10 | 1848 | |
505a666e | 1849 | blk_start_plug(&plug); |
f758eeab | 1850 | spin_lock(&wb->list_lock); |
424b351f | 1851 | if (list_empty(&wb->b_io)) |
f9cae926 | 1852 | queue_io(wb, &work, jiffies); |
d46db3d5 | 1853 | __writeback_inodes_wb(wb, &work); |
f758eeab | 1854 | spin_unlock(&wb->list_lock); |
505a666e | 1855 | blk_finish_plug(&plug); |
edadfb10 | 1856 | |
d46db3d5 WF |
1857 | return nr_pages - work.nr_pages; |
1858 | } | |
03ba3782 | 1859 | |
03ba3782 JA |
1860 | /* |
1861 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 1862 | * |
03ba3782 JA |
1863 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
1864 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
1865 | * just walks the superblock inode list, writing back any inodes which are | |
1866 | * older than a specific point in time. | |
66f3b8e2 | 1867 | * |
03ba3782 JA |
1868 | * Try to run once per dirty_writeback_interval. But if a writeback event |
1869 | * takes longer than a dirty_writeback_interval interval, then leave a | |
1870 | * one-second gap. | |
66f3b8e2 | 1871 | * |
f9cae926 | 1872 | * dirtied_before takes precedence over nr_to_write. So we'll only write back |
03ba3782 | 1873 | * all dirty pages if they are all attached to "old" mappings. |
66f3b8e2 | 1874 | */ |
c4a77a6c | 1875 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 1876 | struct wb_writeback_work *work) |
66f3b8e2 | 1877 | { |
e98be2d5 | 1878 | unsigned long wb_start = jiffies; |
d46db3d5 | 1879 | long nr_pages = work->nr_pages; |
f9cae926 | 1880 | unsigned long dirtied_before = jiffies; |
a5989bdc | 1881 | struct inode *inode; |
d46db3d5 | 1882 | long progress; |
505a666e | 1883 | struct blk_plug plug; |
66f3b8e2 | 1884 | |
505a666e | 1885 | blk_start_plug(&plug); |
e8dfc305 | 1886 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1887 | for (;;) { |
1888 | /* | |
d3ddec76 | 1889 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 1890 | */ |
83ba7b07 | 1891 | if (work->nr_pages <= 0) |
03ba3782 | 1892 | break; |
66f3b8e2 | 1893 | |
aa373cf5 JK |
1894 | /* |
1895 | * Background writeout and kupdate-style writeback may | |
1896 | * run forever. Stop them if there is other work to do | |
1897 | * so that e.g. sync can proceed. They'll be restarted | |
1898 | * after the other works are all done. | |
1899 | */ | |
1900 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 1901 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
1902 | break; |
1903 | ||
38f21977 | 1904 | /* |
d3ddec76 WF |
1905 | * For background writeout, stop when we are below the |
1906 | * background dirty threshold | |
38f21977 | 1907 | */ |
aa661bbe | 1908 | if (work->for_background && !wb_over_bg_thresh(wb)) |
03ba3782 | 1909 | break; |
38f21977 | 1910 | |
1bc36b64 JK |
1911 | /* |
1912 | * Kupdate and background works are special and we want to | |
1913 | * include all inodes that need writing. Livelock avoidance is | |
1914 | * handled by these works yielding to any other work so we are | |
1915 | * safe. | |
1916 | */ | |
ba9aa839 | 1917 | if (work->for_kupdate) { |
f9cae926 | 1918 | dirtied_before = jiffies - |
ba9aa839 | 1919 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 1920 | } else if (work->for_background) |
f9cae926 | 1921 | dirtied_before = jiffies; |
028c2dd1 | 1922 | |
5634cc2a | 1923 | trace_writeback_start(wb, work); |
e8dfc305 | 1924 | if (list_empty(&wb->b_io)) |
f9cae926 | 1925 | queue_io(wb, work, dirtied_before); |
83ba7b07 | 1926 | if (work->sb) |
d46db3d5 | 1927 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 1928 | else |
d46db3d5 | 1929 | progress = __writeback_inodes_wb(wb, work); |
5634cc2a | 1930 | trace_writeback_written(wb, work); |
028c2dd1 | 1931 | |
e98be2d5 | 1932 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
1933 | |
1934 | /* | |
e6fb6da2 WF |
1935 | * Did we write something? Try for more |
1936 | * | |
1937 | * Dirty inodes are moved to b_io for writeback in batches. | |
1938 | * The completion of the current batch does not necessarily | |
1939 | * mean the overall work is done. So we keep looping as long | |
1940 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 1941 | */ |
d46db3d5 | 1942 | if (progress) |
71fd05a8 JA |
1943 | continue; |
1944 | /* | |
e6fb6da2 | 1945 | * No more inodes for IO, bail |
71fd05a8 | 1946 | */ |
b7a2441f | 1947 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 1948 | break; |
71fd05a8 JA |
1949 | /* |
1950 | * Nothing written. Wait for some inode to | |
1951 | * become available for writeback. Otherwise | |
1952 | * we'll just busyloop. | |
1953 | */ | |
bace9248 TE |
1954 | trace_writeback_wait(wb, work); |
1955 | inode = wb_inode(wb->b_more_io.prev); | |
1956 | spin_lock(&inode->i_lock); | |
1957 | spin_unlock(&wb->list_lock); | |
1958 | /* This function drops i_lock... */ | |
1959 | inode_sleep_on_writeback(inode); | |
1960 | spin_lock(&wb->list_lock); | |
03ba3782 | 1961 | } |
e8dfc305 | 1962 | spin_unlock(&wb->list_lock); |
505a666e | 1963 | blk_finish_plug(&plug); |
03ba3782 | 1964 | |
d46db3d5 | 1965 | return nr_pages - work->nr_pages; |
03ba3782 JA |
1966 | } |
1967 | ||
1968 | /* | |
83ba7b07 | 1969 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 1970 | */ |
f0054bb1 | 1971 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 1972 | { |
83ba7b07 | 1973 | struct wb_writeback_work *work = NULL; |
03ba3782 | 1974 | |
f0054bb1 TH |
1975 | spin_lock_bh(&wb->work_lock); |
1976 | if (!list_empty(&wb->work_list)) { | |
1977 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
1978 | struct wb_writeback_work, list); |
1979 | list_del_init(&work->list); | |
03ba3782 | 1980 | } |
f0054bb1 | 1981 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 1982 | return work; |
03ba3782 JA |
1983 | } |
1984 | ||
6585027a JK |
1985 | static long wb_check_background_flush(struct bdi_writeback *wb) |
1986 | { | |
aa661bbe | 1987 | if (wb_over_bg_thresh(wb)) { |
6585027a JK |
1988 | |
1989 | struct wb_writeback_work work = { | |
1990 | .nr_pages = LONG_MAX, | |
1991 | .sync_mode = WB_SYNC_NONE, | |
1992 | .for_background = 1, | |
1993 | .range_cyclic = 1, | |
0e175a18 | 1994 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
1995 | }; |
1996 | ||
1997 | return wb_writeback(wb, &work); | |
1998 | } | |
1999 | ||
2000 | return 0; | |
2001 | } | |
2002 | ||
03ba3782 JA |
2003 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
2004 | { | |
2005 | unsigned long expired; | |
2006 | long nr_pages; | |
2007 | ||
69b62d01 JA |
2008 | /* |
2009 | * When set to zero, disable periodic writeback | |
2010 | */ | |
2011 | if (!dirty_writeback_interval) | |
2012 | return 0; | |
2013 | ||
03ba3782 JA |
2014 | expired = wb->last_old_flush + |
2015 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
2016 | if (time_before(jiffies, expired)) | |
2017 | return 0; | |
2018 | ||
2019 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 2020 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 2021 | |
c4a77a6c | 2022 | if (nr_pages) { |
83ba7b07 | 2023 | struct wb_writeback_work work = { |
c4a77a6c JA |
2024 | .nr_pages = nr_pages, |
2025 | .sync_mode = WB_SYNC_NONE, | |
2026 | .for_kupdate = 1, | |
2027 | .range_cyclic = 1, | |
0e175a18 | 2028 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
2029 | }; |
2030 | ||
83ba7b07 | 2031 | return wb_writeback(wb, &work); |
c4a77a6c | 2032 | } |
03ba3782 JA |
2033 | |
2034 | return 0; | |
2035 | } | |
2036 | ||
85009b4f JA |
2037 | static long wb_check_start_all(struct bdi_writeback *wb) |
2038 | { | |
2039 | long nr_pages; | |
2040 | ||
2041 | if (!test_bit(WB_start_all, &wb->state)) | |
2042 | return 0; | |
2043 | ||
2044 | nr_pages = get_nr_dirty_pages(); | |
2045 | if (nr_pages) { | |
2046 | struct wb_writeback_work work = { | |
2047 | .nr_pages = wb_split_bdi_pages(wb, nr_pages), | |
2048 | .sync_mode = WB_SYNC_NONE, | |
2049 | .range_cyclic = 1, | |
2050 | .reason = wb->start_all_reason, | |
2051 | }; | |
2052 | ||
2053 | nr_pages = wb_writeback(wb, &work); | |
2054 | } | |
2055 | ||
2056 | clear_bit(WB_start_all, &wb->state); | |
2057 | return nr_pages; | |
2058 | } | |
2059 | ||
2060 | ||
03ba3782 JA |
2061 | /* |
2062 | * Retrieve work items and do the writeback they describe | |
2063 | */ | |
25d130ba | 2064 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 2065 | { |
83ba7b07 | 2066 | struct wb_writeback_work *work; |
c4a77a6c | 2067 | long wrote = 0; |
03ba3782 | 2068 | |
4452226e | 2069 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 2070 | while ((work = get_next_work_item(wb)) != NULL) { |
5634cc2a | 2071 | trace_writeback_exec(wb, work); |
83ba7b07 | 2072 | wrote += wb_writeback(wb, work); |
4a3a485b | 2073 | finish_writeback_work(wb, work); |
03ba3782 JA |
2074 | } |
2075 | ||
85009b4f JA |
2076 | /* |
2077 | * Check for a flush-everything request | |
2078 | */ | |
2079 | wrote += wb_check_start_all(wb); | |
2080 | ||
03ba3782 JA |
2081 | /* |
2082 | * Check for periodic writeback, kupdated() style | |
2083 | */ | |
2084 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 2085 | wrote += wb_check_background_flush(wb); |
4452226e | 2086 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
2087 | |
2088 | return wrote; | |
2089 | } | |
2090 | ||
2091 | /* | |
2092 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 2093 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 2094 | */ |
f0054bb1 | 2095 | void wb_workfn(struct work_struct *work) |
03ba3782 | 2096 | { |
839a8e86 TH |
2097 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
2098 | struct bdi_writeback, dwork); | |
03ba3782 JA |
2099 | long pages_written; |
2100 | ||
68f23b89 | 2101 | set_worker_desc("flush-%s", bdi_dev_name(wb->bdi)); |
766f9164 | 2102 | current->flags |= PF_SWAPWRITE; |
455b2864 | 2103 | |
839a8e86 | 2104 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 2105 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 2106 | /* |
f0054bb1 | 2107 | * The normal path. Keep writing back @wb until its |
839a8e86 | 2108 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 2109 | * if @wb is shutting down even when we're running off the |
839a8e86 | 2110 | * rescuer as work_list needs to be drained. |
6467716a | 2111 | */ |
839a8e86 | 2112 | do { |
25d130ba | 2113 | pages_written = wb_do_writeback(wb); |
839a8e86 | 2114 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 2115 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
2116 | } else { |
2117 | /* | |
2118 | * bdi_wq can't get enough workers and we're running off | |
2119 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
2120 | * enough for efficient IO. | |
2121 | */ | |
f0054bb1 | 2122 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 2123 | WB_REASON_FORKER_THREAD); |
455b2864 | 2124 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
2125 | } |
2126 | ||
f0054bb1 | 2127 | if (!list_empty(&wb->work_list)) |
b8b78495 | 2128 | wb_wakeup(wb); |
6ca738d6 | 2129 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
f0054bb1 | 2130 | wb_wakeup_delayed(wb); |
455b2864 | 2131 | |
839a8e86 | 2132 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
2133 | } |
2134 | ||
595043e5 JA |
2135 | /* |
2136 | * Start writeback of `nr_pages' pages on this bdi. If `nr_pages' is zero, | |
2137 | * write back the whole world. | |
2138 | */ | |
2139 | static void __wakeup_flusher_threads_bdi(struct backing_dev_info *bdi, | |
e8e8a0c6 | 2140 | enum wb_reason reason) |
595043e5 JA |
2141 | { |
2142 | struct bdi_writeback *wb; | |
2143 | ||
2144 | if (!bdi_has_dirty_io(bdi)) | |
2145 | return; | |
2146 | ||
2147 | list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) | |
e8e8a0c6 | 2148 | wb_start_writeback(wb, reason); |
595043e5 JA |
2149 | } |
2150 | ||
2151 | void wakeup_flusher_threads_bdi(struct backing_dev_info *bdi, | |
2152 | enum wb_reason reason) | |
2153 | { | |
595043e5 | 2154 | rcu_read_lock(); |
e8e8a0c6 | 2155 | __wakeup_flusher_threads_bdi(bdi, reason); |
595043e5 JA |
2156 | rcu_read_unlock(); |
2157 | } | |
2158 | ||
03ba3782 | 2159 | /* |
9ba4b2df | 2160 | * Wakeup the flusher threads to start writeback of all currently dirty pages |
03ba3782 | 2161 | */ |
9ba4b2df | 2162 | void wakeup_flusher_threads(enum wb_reason reason) |
03ba3782 | 2163 | { |
b8c2f347 | 2164 | struct backing_dev_info *bdi; |
03ba3782 | 2165 | |
51350ea0 KK |
2166 | /* |
2167 | * If we are expecting writeback progress we must submit plugged IO. | |
2168 | */ | |
2169 | if (blk_needs_flush_plug(current)) | |
2170 | blk_schedule_flush_plug(current); | |
2171 | ||
b8c2f347 | 2172 | rcu_read_lock(); |
595043e5 | 2173 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) |
e8e8a0c6 | 2174 | __wakeup_flusher_threads_bdi(bdi, reason); |
cfc4ba53 | 2175 | rcu_read_unlock(); |
1da177e4 LT |
2176 | } |
2177 | ||
a2f48706 TT |
2178 | /* |
2179 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
2180 | * written back periodically. We deliberately do *not* check the | |
2181 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
2182 | * kernel to be constantly waking up once there are any dirtytime | |
2183 | * inodes on the system. So instead we define a separate delayed work | |
2184 | * function which gets called much more rarely. (By default, only | |
2185 | * once every 12 hours.) | |
2186 | * | |
2187 | * If there is any other write activity going on in the file system, | |
2188 | * this function won't be necessary. But if the only thing that has | |
2189 | * happened on the file system is a dirtytime inode caused by an atime | |
2190 | * update, we need this infrastructure below to make sure that inode | |
2191 | * eventually gets pushed out to disk. | |
2192 | */ | |
2193 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
2194 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
2195 | ||
2196 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
2197 | { | |
2198 | struct backing_dev_info *bdi; | |
2199 | ||
2200 | rcu_read_lock(); | |
2201 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
001fe6f6 | 2202 | struct bdi_writeback *wb; |
001fe6f6 | 2203 | |
b817525a | 2204 | list_for_each_entry_rcu(wb, &bdi->wb_list, bdi_node) |
6fdf860f TH |
2205 | if (!list_empty(&wb->b_dirty_time)) |
2206 | wb_wakeup(wb); | |
a2f48706 TT |
2207 | } |
2208 | rcu_read_unlock(); | |
2209 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
2210 | } | |
2211 | ||
2212 | static int __init start_dirtytime_writeback(void) | |
2213 | { | |
2214 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
2215 | return 0; | |
2216 | } | |
2217 | __initcall(start_dirtytime_writeback); | |
2218 | ||
1efff914 | 2219 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
9ca48e20 | 2220 | void *buffer, size_t *lenp, loff_t *ppos) |
1efff914 TT |
2221 | { |
2222 | int ret; | |
2223 | ||
2224 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
2225 | if (ret == 0 && write) | |
2226 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
2227 | return ret; | |
2228 | } | |
2229 | ||
03ba3782 JA |
2230 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
2231 | { | |
2232 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
2233 | struct dentry *dentry; | |
2234 | const char *name = "?"; | |
2235 | ||
2236 | dentry = d_find_alias(inode); | |
2237 | if (dentry) { | |
2238 | spin_lock(&dentry->d_lock); | |
2239 | name = (const char *) dentry->d_name.name; | |
2240 | } | |
2241 | printk(KERN_DEBUG | |
2242 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
2243 | current->comm, task_pid_nr(current), inode->i_ino, | |
2244 | name, inode->i_sb->s_id); | |
2245 | if (dentry) { | |
2246 | spin_unlock(&dentry->d_lock); | |
2247 | dput(dentry); | |
2248 | } | |
2249 | } | |
2250 | } | |
2251 | ||
2252 | /** | |
35d14f27 | 2253 | * __mark_inode_dirty - internal function to mark an inode dirty |
0117d427 MCC |
2254 | * |
2255 | * @inode: inode to mark | |
35d14f27 EB |
2256 | * @flags: what kind of dirty, e.g. I_DIRTY_SYNC. This can be a combination of |
2257 | * multiple I_DIRTY_* flags, except that I_DIRTY_TIME can't be combined | |
2258 | * with I_DIRTY_PAGES. | |
0117d427 | 2259 | * |
35d14f27 EB |
2260 | * Mark an inode as dirty. We notify the filesystem, then update the inode's |
2261 | * dirty flags. Then, if needed we add the inode to the appropriate dirty list. | |
1da177e4 | 2262 | * |
35d14f27 EB |
2263 | * Most callers should use mark_inode_dirty() or mark_inode_dirty_sync() |
2264 | * instead of calling this directly. | |
03ba3782 | 2265 | * |
35d14f27 EB |
2266 | * CAREFUL! We only add the inode to the dirty list if it is hashed or if it |
2267 | * refers to a blockdev. Unhashed inodes will never be added to the dirty list | |
2268 | * even if they are later hashed, as they will have been marked dirty already. | |
03ba3782 | 2269 | * |
35d14f27 | 2270 | * In short, ensure you hash any inodes _before_ you start marking them dirty. |
1da177e4 | 2271 | * |
03ba3782 JA |
2272 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
2273 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
2274 | * the kernel-internal blockdev inode represents the dirtying time of the | |
2275 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
2276 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
2277 | * blockdev inode. | |
1da177e4 | 2278 | */ |
03ba3782 | 2279 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 2280 | { |
03ba3782 | 2281 | struct super_block *sb = inode->i_sb; |
35d14f27 | 2282 | int dirtytime = 0; |
0ae45f63 TT |
2283 | |
2284 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 2285 | |
e2728c56 | 2286 | if (flags & I_DIRTY_INODE) { |
35d14f27 EB |
2287 | /* |
2288 | * Notify the filesystem about the inode being dirtied, so that | |
2289 | * (if needed) it can update on-disk fields and journal the | |
2290 | * inode. This is only needed when the inode itself is being | |
2291 | * dirtied now. I.e. it's only needed for I_DIRTY_INODE, not | |
2292 | * for just I_DIRTY_PAGES or I_DIRTY_TIME. | |
2293 | */ | |
9fb0a7da | 2294 | trace_writeback_dirty_inode_start(inode, flags); |
03ba3782 | 2295 | if (sb->s_op->dirty_inode) |
a38ed483 | 2296 | sb->s_op->dirty_inode(inode, flags & I_DIRTY_INODE); |
9fb0a7da | 2297 | trace_writeback_dirty_inode(inode, flags); |
e2728c56 | 2298 | |
35d14f27 | 2299 | /* I_DIRTY_INODE supersedes I_DIRTY_TIME. */ |
0ae45f63 | 2300 | flags &= ~I_DIRTY_TIME; |
35d14f27 EB |
2301 | } else { |
2302 | /* | |
2303 | * Else it's either I_DIRTY_PAGES, I_DIRTY_TIME, or nothing. | |
2304 | * (We don't support setting both I_DIRTY_PAGES and I_DIRTY_TIME | |
2305 | * in one call to __mark_inode_dirty().) | |
2306 | */ | |
2307 | dirtytime = flags & I_DIRTY_TIME; | |
2308 | WARN_ON_ONCE(dirtytime && flags != I_DIRTY_TIME); | |
e2728c56 | 2309 | } |
03ba3782 JA |
2310 | |
2311 | /* | |
9c6ac78e TH |
2312 | * Paired with smp_mb() in __writeback_single_inode() for the |
2313 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
2314 | */ |
2315 | smp_mb(); | |
2316 | ||
0ae45f63 TT |
2317 | if (((inode->i_state & flags) == flags) || |
2318 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
2319 | return; |
2320 | ||
2321 | if (unlikely(block_dump)) | |
2322 | block_dump___mark_inode_dirty(inode); | |
2323 | ||
250df6ed | 2324 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
2325 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
2326 | goto out_unlock_inode; | |
03ba3782 JA |
2327 | if ((inode->i_state & flags) != flags) { |
2328 | const int was_dirty = inode->i_state & I_DIRTY; | |
2329 | ||
52ebea74 TH |
2330 | inode_attach_wb(inode, NULL); |
2331 | ||
35d14f27 | 2332 | /* I_DIRTY_INODE supersedes I_DIRTY_TIME. */ |
0ae45f63 TT |
2333 | if (flags & I_DIRTY_INODE) |
2334 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
2335 | inode->i_state |= flags; |
2336 | ||
2337 | /* | |
5afced3b JK |
2338 | * If the inode is queued for writeback by flush worker, just |
2339 | * update its dirty state. Once the flush worker is done with | |
2340 | * the inode it will place it on the appropriate superblock | |
2341 | * list, based upon its state. | |
03ba3782 | 2342 | */ |
5afced3b | 2343 | if (inode->i_state & I_SYNC_QUEUED) |
250df6ed | 2344 | goto out_unlock_inode; |
03ba3782 JA |
2345 | |
2346 | /* | |
2347 | * Only add valid (hashed) inodes to the superblock's | |
2348 | * dirty list. Add blockdev inodes as well. | |
2349 | */ | |
2350 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 2351 | if (inode_unhashed(inode)) |
250df6ed | 2352 | goto out_unlock_inode; |
03ba3782 | 2353 | } |
a4ffdde6 | 2354 | if (inode->i_state & I_FREEING) |
250df6ed | 2355 | goto out_unlock_inode; |
03ba3782 JA |
2356 | |
2357 | /* | |
2358 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
2359 | * reposition it (that would break b_dirty time-ordering). | |
2360 | */ | |
2361 | if (!was_dirty) { | |
87e1d789 | 2362 | struct bdi_writeback *wb; |
d6c10f1f | 2363 | struct list_head *dirty_list; |
a66979ab | 2364 | bool wakeup_bdi = false; |
253c34e9 | 2365 | |
87e1d789 | 2366 | wb = locked_inode_to_wb_and_lock_list(inode); |
253c34e9 | 2367 | |
03ba3782 | 2368 | inode->dirtied_when = jiffies; |
a2f48706 TT |
2369 | if (dirtytime) |
2370 | inode->dirtied_time_when = jiffies; | |
d6c10f1f | 2371 | |
0e11f644 | 2372 | if (inode->i_state & I_DIRTY) |
0747259d | 2373 | dirty_list = &wb->b_dirty; |
a2f48706 | 2374 | else |
0747259d | 2375 | dirty_list = &wb->b_dirty_time; |
d6c10f1f | 2376 | |
c7f54084 | 2377 | wakeup_bdi = inode_io_list_move_locked(inode, wb, |
d6c10f1f TH |
2378 | dirty_list); |
2379 | ||
0747259d | 2380 | spin_unlock(&wb->list_lock); |
0ae45f63 | 2381 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab | 2382 | |
d6c10f1f TH |
2383 | /* |
2384 | * If this is the first dirty inode for this bdi, | |
2385 | * we have to wake-up the corresponding bdi thread | |
2386 | * to make sure background write-back happens | |
2387 | * later. | |
2388 | */ | |
f56753ac CH |
2389 | if (wakeup_bdi && |
2390 | (wb->bdi->capabilities & BDI_CAP_WRITEBACK)) | |
0747259d | 2391 | wb_wakeup_delayed(wb); |
a66979ab | 2392 | return; |
1da177e4 | 2393 | } |
1da177e4 | 2394 | } |
250df6ed DC |
2395 | out_unlock_inode: |
2396 | spin_unlock(&inode->i_lock); | |
03ba3782 JA |
2397 | } |
2398 | EXPORT_SYMBOL(__mark_inode_dirty); | |
2399 | ||
e97fedb9 DC |
2400 | /* |
2401 | * The @s_sync_lock is used to serialise concurrent sync operations | |
2402 | * to avoid lock contention problems with concurrent wait_sb_inodes() calls. | |
2403 | * Concurrent callers will block on the s_sync_lock rather than doing contending | |
2404 | * walks. The queueing maintains sync(2) required behaviour as all the IO that | |
2405 | * has been issued up to the time this function is enter is guaranteed to be | |
2406 | * completed by the time we have gained the lock and waited for all IO that is | |
2407 | * in progress regardless of the order callers are granted the lock. | |
2408 | */ | |
b6e51316 | 2409 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 | 2410 | { |
6c60d2b5 | 2411 | LIST_HEAD(sync_list); |
03ba3782 JA |
2412 | |
2413 | /* | |
2414 | * We need to be protected against the filesystem going from | |
2415 | * r/o to r/w or vice versa. | |
2416 | */ | |
b6e51316 | 2417 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 2418 | |
e97fedb9 | 2419 | mutex_lock(&sb->s_sync_lock); |
03ba3782 JA |
2420 | |
2421 | /* | |
6c60d2b5 DC |
2422 | * Splice the writeback list onto a temporary list to avoid waiting on |
2423 | * inodes that have started writeback after this point. | |
2424 | * | |
2425 | * Use rcu_read_lock() to keep the inodes around until we have a | |
2426 | * reference. s_inode_wblist_lock protects sb->s_inodes_wb as well as | |
2427 | * the local list because inodes can be dropped from either by writeback | |
2428 | * completion. | |
2429 | */ | |
2430 | rcu_read_lock(); | |
2431 | spin_lock_irq(&sb->s_inode_wblist_lock); | |
2432 | list_splice_init(&sb->s_inodes_wb, &sync_list); | |
2433 | ||
2434 | /* | |
2435 | * Data integrity sync. Must wait for all pages under writeback, because | |
2436 | * there may have been pages dirtied before our sync call, but which had | |
2437 | * writeout started before we write it out. In which case, the inode | |
2438 | * may not be on the dirty list, but we still have to wait for that | |
2439 | * writeout. | |
03ba3782 | 2440 | */ |
6c60d2b5 DC |
2441 | while (!list_empty(&sync_list)) { |
2442 | struct inode *inode = list_first_entry(&sync_list, struct inode, | |
2443 | i_wb_list); | |
250df6ed | 2444 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 2445 | |
6c60d2b5 DC |
2446 | /* |
2447 | * Move each inode back to the wb list before we drop the lock | |
2448 | * to preserve consistency between i_wb_list and the mapping | |
2449 | * writeback tag. Writeback completion is responsible to remove | |
2450 | * the inode from either list once the writeback tag is cleared. | |
2451 | */ | |
2452 | list_move_tail(&inode->i_wb_list, &sb->s_inodes_wb); | |
2453 | ||
2454 | /* | |
2455 | * The mapping can appear untagged while still on-list since we | |
2456 | * do not have the mapping lock. Skip it here, wb completion | |
2457 | * will remove it. | |
2458 | */ | |
2459 | if (!mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
2460 | continue; | |
2461 | ||
2462 | spin_unlock_irq(&sb->s_inode_wblist_lock); | |
2463 | ||
250df6ed | 2464 | spin_lock(&inode->i_lock); |
6c60d2b5 | 2465 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) { |
250df6ed | 2466 | spin_unlock(&inode->i_lock); |
6c60d2b5 DC |
2467 | |
2468 | spin_lock_irq(&sb->s_inode_wblist_lock); | |
03ba3782 | 2469 | continue; |
250df6ed | 2470 | } |
03ba3782 | 2471 | __iget(inode); |
250df6ed | 2472 | spin_unlock(&inode->i_lock); |
6c60d2b5 | 2473 | rcu_read_unlock(); |
03ba3782 | 2474 | |
aa750fd7 JN |
2475 | /* |
2476 | * We keep the error status of individual mapping so that | |
2477 | * applications can catch the writeback error using fsync(2). | |
2478 | * See filemap_fdatawait_keep_errors() for details. | |
2479 | */ | |
2480 | filemap_fdatawait_keep_errors(mapping); | |
03ba3782 JA |
2481 | |
2482 | cond_resched(); | |
2483 | ||
6c60d2b5 DC |
2484 | iput(inode); |
2485 | ||
2486 | rcu_read_lock(); | |
2487 | spin_lock_irq(&sb->s_inode_wblist_lock); | |
03ba3782 | 2488 | } |
6c60d2b5 DC |
2489 | spin_unlock_irq(&sb->s_inode_wblist_lock); |
2490 | rcu_read_unlock(); | |
e97fedb9 | 2491 | mutex_unlock(&sb->s_sync_lock); |
1da177e4 LT |
2492 | } |
2493 | ||
f30a7d0c TH |
2494 | static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
2495 | enum wb_reason reason, bool skip_if_busy) | |
1da177e4 | 2496 | { |
5b9cce4c TH |
2497 | struct backing_dev_info *bdi = sb->s_bdi; |
2498 | DEFINE_WB_COMPLETION(done, bdi); | |
83ba7b07 | 2499 | struct wb_writeback_work work = { |
6e6938b6 WF |
2500 | .sb = sb, |
2501 | .sync_mode = WB_SYNC_NONE, | |
2502 | .tagged_writepages = 1, | |
2503 | .done = &done, | |
2504 | .nr_pages = nr, | |
0e175a18 | 2505 | .reason = reason, |
3c4d7165 | 2506 | }; |
d8a8559c | 2507 | |
e7972912 | 2508 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 2509 | return; |
cf37e972 | 2510 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
f30a7d0c | 2511 | |
db125360 | 2512 | bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy); |
5b9cce4c | 2513 | wb_wait_for_completion(&done); |
e913fc82 | 2514 | } |
f30a7d0c TH |
2515 | |
2516 | /** | |
2517 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block | |
2518 | * @sb: the superblock | |
2519 | * @nr: the number of pages to write | |
2520 | * @reason: reason why some writeback work initiated | |
2521 | * | |
2522 | * Start writeback on some inodes on this super_block. No guarantees are made | |
2523 | * on how many (if any) will be written, and this function does not wait | |
2524 | * for IO completion of submitted IO. | |
2525 | */ | |
2526 | void writeback_inodes_sb_nr(struct super_block *sb, | |
2527 | unsigned long nr, | |
2528 | enum wb_reason reason) | |
2529 | { | |
2530 | __writeback_inodes_sb_nr(sb, nr, reason, false); | |
2531 | } | |
3259f8be CM |
2532 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
2533 | ||
2534 | /** | |
2535 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
2536 | * @sb: the superblock | |
786228ab | 2537 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
2538 | * |
2539 | * Start writeback on some inodes on this super_block. No guarantees are made | |
2540 | * on how many (if any) will be written, and this function does not wait | |
2541 | * for IO completion of submitted IO. | |
2542 | */ | |
0e175a18 | 2543 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 2544 | { |
0e175a18 | 2545 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 2546 | } |
0e3c9a22 | 2547 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 2548 | |
17bd55d0 | 2549 | /** |
8264c321 | 2550 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
17bd55d0 | 2551 | * @sb: the superblock |
8264c321 | 2552 | * @reason: reason why some writeback work was initiated |
17bd55d0 | 2553 | * |
8264c321 | 2554 | * Invoke __writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 | 2555 | */ |
8264c321 | 2556 | void try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
17bd55d0 | 2557 | { |
10ee27a0 | 2558 | if (!down_read_trylock(&sb->s_umount)) |
8264c321 | 2559 | return; |
10ee27a0 | 2560 | |
8264c321 | 2561 | __writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason, true); |
10ee27a0 | 2562 | up_read(&sb->s_umount); |
3259f8be | 2563 | } |
10ee27a0 | 2564 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 2565 | |
d8a8559c JA |
2566 | /** |
2567 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 2568 | * @sb: the superblock |
d8a8559c JA |
2569 | * |
2570 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 2571 | * super_block. |
d8a8559c | 2572 | */ |
0dc83bd3 | 2573 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 2574 | { |
5b9cce4c TH |
2575 | struct backing_dev_info *bdi = sb->s_bdi; |
2576 | DEFINE_WB_COMPLETION(done, bdi); | |
83ba7b07 | 2577 | struct wb_writeback_work work = { |
3c4d7165 CH |
2578 | .sb = sb, |
2579 | .sync_mode = WB_SYNC_ALL, | |
2580 | .nr_pages = LONG_MAX, | |
2581 | .range_cyclic = 0, | |
83ba7b07 | 2582 | .done = &done, |
0e175a18 | 2583 | .reason = WB_REASON_SYNC, |
7747bd4b | 2584 | .for_sync = 1, |
3c4d7165 CH |
2585 | }; |
2586 | ||
006a0973 TH |
2587 | /* |
2588 | * Can't skip on !bdi_has_dirty() because we should wait for !dirty | |
2589 | * inodes under writeback and I_DIRTY_TIME inodes ignored by | |
2590 | * bdi_has_dirty() need to be written out too. | |
2591 | */ | |
2592 | if (bdi == &noop_backing_dev_info) | |
6eedc701 | 2593 | return; |
cf37e972 CH |
2594 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
2595 | ||
7fc5854f TH |
2596 | /* protect against inode wb switch, see inode_switch_wbs_work_fn() */ |
2597 | bdi_down_write_wb_switch_rwsem(bdi); | |
db125360 | 2598 | bdi_split_work_to_wbs(bdi, &work, false); |
5b9cce4c | 2599 | wb_wait_for_completion(&done); |
7fc5854f | 2600 | bdi_up_write_wb_switch_rwsem(bdi); |
83ba7b07 | 2601 | |
b6e51316 | 2602 | wait_sb_inodes(sb); |
1da177e4 | 2603 | } |
d8a8559c | 2604 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 2605 | |
1da177e4 | 2606 | /** |
7f04c26d AA |
2607 | * write_inode_now - write an inode to disk |
2608 | * @inode: inode to write to disk | |
2609 | * @sync: whether the write should be synchronous or not | |
2610 | * | |
2611 | * This function commits an inode to disk immediately if it is dirty. This is | |
2612 | * primarily needed by knfsd. | |
1da177e4 | 2613 | * |
7f04c26d | 2614 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 2615 | */ |
1da177e4 LT |
2616 | int write_inode_now(struct inode *inode, int sync) |
2617 | { | |
1da177e4 LT |
2618 | struct writeback_control wbc = { |
2619 | .nr_to_write = LONG_MAX, | |
18914b18 | 2620 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
2621 | .range_start = 0, |
2622 | .range_end = LLONG_MAX, | |
1da177e4 LT |
2623 | }; |
2624 | ||
f56753ac | 2625 | if (!mapping_can_writeback(inode->i_mapping)) |
49364ce2 | 2626 | wbc.nr_to_write = 0; |
1da177e4 LT |
2627 | |
2628 | might_sleep(); | |
aaf25593 | 2629 | return writeback_single_inode(inode, &wbc); |
1da177e4 LT |
2630 | } |
2631 | EXPORT_SYMBOL(write_inode_now); | |
2632 | ||
2633 | /** | |
2634 | * sync_inode - write an inode and its pages to disk. | |
2635 | * @inode: the inode to sync | |
2636 | * @wbc: controls the writeback mode | |
2637 | * | |
2638 | * sync_inode() will write an inode and its pages to disk. It will also | |
2639 | * correctly update the inode on its superblock's dirty inode lists and will | |
2640 | * update inode->i_state. | |
2641 | * | |
2642 | * The caller must have a ref on the inode. | |
2643 | */ | |
2644 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
2645 | { | |
aaf25593 | 2646 | return writeback_single_inode(inode, wbc); |
1da177e4 LT |
2647 | } |
2648 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
2649 | |
2650 | /** | |
c691b9d9 | 2651 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
2652 | * @inode: the inode to sync |
2653 | * @wait: wait for I/O to complete. | |
2654 | * | |
c691b9d9 | 2655 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
2656 | * |
2657 | * Note: only writes the actual inode, no associated data or other metadata. | |
2658 | */ | |
2659 | int sync_inode_metadata(struct inode *inode, int wait) | |
2660 | { | |
2661 | struct writeback_control wbc = { | |
2662 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
2663 | .nr_to_write = 0, /* metadata-only */ | |
2664 | }; | |
2665 | ||
2666 | return sync_inode(inode, &wbc); | |
2667 | } | |
2668 | EXPORT_SYMBOL(sync_inode_metadata); |