Commit | Line | Data |
---|---|---|
3fcfab16 AM |
1 | |
2 | #include <linux/wait.h> | |
3 | #include <linux/backing-dev.h> | |
03ba3782 JA |
4 | #include <linux/kthread.h> |
5 | #include <linux/freezer.h> | |
3fcfab16 | 6 | #include <linux/fs.h> |
26160158 | 7 | #include <linux/pagemap.h> |
03ba3782 | 8 | #include <linux/mm.h> |
3fcfab16 AM |
9 | #include <linux/sched.h> |
10 | #include <linux/module.h> | |
cf0ca9fe PZ |
11 | #include <linux/writeback.h> |
12 | #include <linux/device.h> | |
455b2864 | 13 | #include <trace/events/writeback.h> |
cf0ca9fe | 14 | |
c3c53206 JA |
15 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); |
16 | ||
26160158 | 17 | struct backing_dev_info default_backing_dev_info = { |
d993831f | 18 | .name = "default", |
26160158 JA |
19 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
20 | .state = 0, | |
21 | .capabilities = BDI_CAP_MAP_COPY, | |
26160158 JA |
22 | }; |
23 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe | 24 | |
5129a469 JE |
25 | struct backing_dev_info noop_backing_dev_info = { |
26 | .name = "noop", | |
976e48f8 | 27 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
5129a469 JE |
28 | }; |
29 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | |
30 | ||
cf0ca9fe | 31 | static struct class *bdi_class; |
cfc4ba53 JA |
32 | |
33 | /* | |
181387da | 34 | * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side |
cfc4ba53 JA |
35 | * locking. |
36 | */ | |
03ba3782 | 37 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 38 | LIST_HEAD(bdi_list); |
03ba3782 | 39 | |
f758eeab CH |
40 | void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2) |
41 | { | |
42 | if (wb1 < wb2) { | |
43 | spin_lock(&wb1->list_lock); | |
44 | spin_lock_nested(&wb2->list_lock, 1); | |
45 | } else { | |
46 | spin_lock(&wb2->list_lock); | |
47 | spin_lock_nested(&wb1->list_lock, 1); | |
48 | } | |
49 | } | |
50 | ||
76f1418b MS |
51 | #ifdef CONFIG_DEBUG_FS |
52 | #include <linux/debugfs.h> | |
53 | #include <linux/seq_file.h> | |
54 | ||
55 | static struct dentry *bdi_debug_root; | |
56 | ||
57 | static void bdi_debug_init(void) | |
58 | { | |
59 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
60 | } | |
61 | ||
62 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
63 | { | |
64 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 65 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
66 | unsigned long background_thresh; |
67 | unsigned long dirty_thresh; | |
68 | unsigned long bdi_thresh; | |
345227d7 | 69 | unsigned long nr_dirty, nr_io, nr_more_io; |
f09b00d3 JA |
70 | struct inode *inode; |
71 | ||
345227d7 | 72 | nr_dirty = nr_io = nr_more_io = 0; |
f758eeab | 73 | spin_lock(&wb->list_lock); |
7ccf19a8 | 74 | list_for_each_entry(inode, &wb->b_dirty, i_wb_list) |
c1955ce3 | 75 | nr_dirty++; |
7ccf19a8 | 76 | list_for_each_entry(inode, &wb->b_io, i_wb_list) |
c1955ce3 | 77 | nr_io++; |
7ccf19a8 | 78 | list_for_each_entry(inode, &wb->b_more_io, i_wb_list) |
c1955ce3 | 79 | nr_more_io++; |
f758eeab | 80 | spin_unlock(&wb->list_lock); |
76f1418b | 81 | |
16c4042f WF |
82 | global_dirty_limits(&background_thresh, &dirty_thresh); |
83 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); | |
76f1418b MS |
84 | |
85 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
86 | seq_printf(m, | |
00821b00 WF |
87 | "BdiWriteback: %10lu kB\n" |
88 | "BdiReclaimable: %10lu kB\n" | |
89 | "BdiDirtyThresh: %10lu kB\n" | |
90 | "DirtyThresh: %10lu kB\n" | |
91 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 92 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
93 | "BdiWritten: %10lu kB\n" |
94 | "BdiWriteBandwidth: %10lu kBps\n" | |
95 | "b_dirty: %10lu\n" | |
96 | "b_io: %10lu\n" | |
97 | "b_more_io: %10lu\n" | |
98 | "bdi_list: %10u\n" | |
99 | "state: %10lx\n", | |
76f1418b MS |
100 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
101 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f7d2b1ec JK |
102 | K(bdi_thresh), |
103 | K(dirty_thresh), | |
104 | K(background_thresh), | |
c8e28ce0 | 105 | (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)), |
f7d2b1ec | 106 | (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)), |
00821b00 | 107 | (unsigned long) K(bdi->write_bandwidth), |
f7d2b1ec JK |
108 | nr_dirty, |
109 | nr_io, | |
110 | nr_more_io, | |
c1955ce3 | 111 | !list_empty(&bdi->bdi_list), bdi->state); |
76f1418b MS |
112 | #undef K |
113 | ||
114 | return 0; | |
115 | } | |
116 | ||
117 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
118 | { | |
119 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
120 | } | |
121 | ||
122 | static const struct file_operations bdi_debug_stats_fops = { | |
123 | .open = bdi_debug_stats_open, | |
124 | .read = seq_read, | |
125 | .llseek = seq_lseek, | |
126 | .release = single_release, | |
127 | }; | |
128 | ||
129 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
130 | { | |
131 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
132 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
133 | bdi, &bdi_debug_stats_fops); | |
134 | } | |
135 | ||
136 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
137 | { | |
138 | debugfs_remove(bdi->debug_stats); | |
139 | debugfs_remove(bdi->debug_dir); | |
140 | } | |
141 | #else | |
142 | static inline void bdi_debug_init(void) | |
143 | { | |
144 | } | |
145 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
146 | const char *name) | |
147 | { | |
148 | } | |
149 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
150 | { | |
151 | } | |
152 | #endif | |
153 | ||
cf0ca9fe PZ |
154 | static ssize_t read_ahead_kb_store(struct device *dev, |
155 | struct device_attribute *attr, | |
156 | const char *buf, size_t count) | |
157 | { | |
158 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 159 | unsigned long read_ahead_kb; |
7034ed13 | 160 | ssize_t ret; |
cf0ca9fe | 161 | |
7034ed13 NJ |
162 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
163 | if (ret < 0) | |
164 | return ret; | |
165 | ||
166 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
167 | ||
168 | return count; | |
cf0ca9fe PZ |
169 | } |
170 | ||
171 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
172 | ||
173 | #define BDI_SHOW(name, expr) \ | |
174 | static ssize_t name##_show(struct device *dev, \ | |
175 | struct device_attribute *attr, char *page) \ | |
176 | { \ | |
177 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
178 | \ | |
179 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
180 | } | |
181 | ||
182 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
183 | ||
189d3c4a PZ |
184 | static ssize_t min_ratio_store(struct device *dev, |
185 | struct device_attribute *attr, const char *buf, size_t count) | |
186 | { | |
187 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 188 | unsigned int ratio; |
7034ed13 NJ |
189 | ssize_t ret; |
190 | ||
191 | ret = kstrtouint(buf, 10, &ratio); | |
192 | if (ret < 0) | |
193 | return ret; | |
194 | ||
195 | ret = bdi_set_min_ratio(bdi, ratio); | |
196 | if (!ret) | |
197 | ret = count; | |
189d3c4a | 198 | |
189d3c4a PZ |
199 | return ret; |
200 | } | |
201 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
202 | ||
a42dde04 PZ |
203 | static ssize_t max_ratio_store(struct device *dev, |
204 | struct device_attribute *attr, const char *buf, size_t count) | |
205 | { | |
206 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 207 | unsigned int ratio; |
7034ed13 NJ |
208 | ssize_t ret; |
209 | ||
210 | ret = kstrtouint(buf, 10, &ratio); | |
211 | if (ret < 0) | |
212 | return ret; | |
213 | ||
214 | ret = bdi_set_max_ratio(bdi, ratio); | |
215 | if (!ret) | |
216 | ret = count; | |
a42dde04 | 217 | |
a42dde04 PZ |
218 | return ret; |
219 | } | |
220 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
221 | ||
7d311cda DW |
222 | static ssize_t stable_pages_required_show(struct device *dev, |
223 | struct device_attribute *attr, | |
224 | char *page) | |
225 | { | |
226 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
227 | ||
228 | return snprintf(page, PAGE_SIZE-1, "%d\n", | |
229 | bdi_cap_stable_pages_required(bdi) ? 1 : 0); | |
230 | } | |
231 | ||
cf0ca9fe PZ |
232 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
233 | ||
234 | static struct device_attribute bdi_dev_attrs[] = { | |
235 | __ATTR_RW(read_ahead_kb), | |
189d3c4a | 236 | __ATTR_RW(min_ratio), |
a42dde04 | 237 | __ATTR_RW(max_ratio), |
7d311cda | 238 | __ATTR_RO(stable_pages_required), |
cf0ca9fe PZ |
239 | __ATTR_NULL, |
240 | }; | |
241 | ||
242 | static __init int bdi_class_init(void) | |
243 | { | |
244 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
245 | if (IS_ERR(bdi_class)) |
246 | return PTR_ERR(bdi_class); | |
247 | ||
cf0ca9fe | 248 | bdi_class->dev_attrs = bdi_dev_attrs; |
76f1418b | 249 | bdi_debug_init(); |
cf0ca9fe PZ |
250 | return 0; |
251 | } | |
76f1418b | 252 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 253 | |
26160158 JA |
254 | static int __init default_bdi_init(void) |
255 | { | |
256 | int err; | |
257 | ||
258 | err = bdi_init(&default_backing_dev_info); | |
259 | if (!err) | |
260 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
976e48f8 | 261 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
262 | |
263 | return err; | |
264 | } | |
265 | subsys_initcall(default_bdi_init); | |
266 | ||
03ba3782 JA |
267 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
268 | { | |
269 | return wb_has_dirty_io(&bdi->wb); | |
270 | } | |
271 | ||
6467716a AB |
272 | static void wakeup_timer_fn(unsigned long data) |
273 | { | |
274 | struct backing_dev_info *bdi = (struct backing_dev_info *)data; | |
275 | ||
276 | spin_lock_bh(&bdi->wb_lock); | |
277 | if (bdi->wb.task) { | |
60332023 | 278 | trace_writeback_wake_thread(bdi); |
6467716a | 279 | wake_up_process(bdi->wb.task); |
2673b4cf | 280 | } else if (bdi->dev) { |
6467716a AB |
281 | /* |
282 | * When bdi tasks are inactive for long time, they are killed. | |
283 | * In this case we have to wake-up the forker thread which | |
284 | * should create and run the bdi thread. | |
285 | */ | |
60332023 | 286 | trace_writeback_wake_forker_thread(bdi); |
6467716a AB |
287 | wake_up_process(default_backing_dev_info.wb.task); |
288 | } | |
289 | spin_unlock_bh(&bdi->wb_lock); | |
290 | } | |
291 | ||
292 | /* | |
293 | * This function is used when the first inode for this bdi is marked dirty. It | |
294 | * wakes-up the corresponding bdi thread which should then take care of the | |
295 | * periodic background write-out of dirty inodes. Since the write-out would | |
296 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
297 | * set up a timer which wakes the bdi thread up later. | |
298 | * | |
299 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
300 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
301 | * by delaying the wake-up. | |
302 | */ | |
303 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | |
304 | { | |
305 | unsigned long timeout; | |
306 | ||
307 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
308 | mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout); | |
309 | } | |
310 | ||
fff5b85a AB |
311 | /* |
312 | * Calculate the longest interval (jiffies) bdi threads are allowed to be | |
313 | * inactive. | |
314 | */ | |
315 | static unsigned long bdi_longest_inactive(void) | |
316 | { | |
317 | unsigned long interval; | |
318 | ||
319 | interval = msecs_to_jiffies(dirty_writeback_interval * 10); | |
320 | return max(5UL * 60 * HZ, interval); | |
321 | } | |
322 | ||
5a042aa4 JK |
323 | /* |
324 | * Clear pending bit and wakeup anybody waiting for flusher thread creation or | |
325 | * shutdown | |
326 | */ | |
327 | static void bdi_clear_pending(struct backing_dev_info *bdi) | |
328 | { | |
329 | clear_bit(BDI_pending, &bdi->state); | |
330 | smp_mb__after_clear_bit(); | |
331 | wake_up_bit(&bdi->state, BDI_pending); | |
332 | } | |
333 | ||
6f904ff0 | 334 | static int bdi_forker_thread(void *ptr) |
03ba3782 JA |
335 | { |
336 | struct bdi_writeback *me = ptr; | |
337 | ||
766f9164 | 338 | current->flags |= PF_SWAPWRITE; |
c1955ce3 CH |
339 | set_freezable(); |
340 | ||
341 | /* | |
342 | * Our parent may run at a different priority, just set us to normal | |
343 | */ | |
344 | set_user_nice(current, 0); | |
03ba3782 JA |
345 | |
346 | for (;;) { | |
fff5b85a | 347 | struct task_struct *task = NULL; |
78c40cb6 | 348 | struct backing_dev_info *bdi; |
adf39240 AB |
349 | enum { |
350 | NO_ACTION, /* Nothing to do */ | |
351 | FORK_THREAD, /* Fork bdi thread */ | |
fff5b85a | 352 | KILL_THREAD, /* Kill inactive bdi thread */ |
adf39240 | 353 | } action = NO_ACTION; |
03ba3782 JA |
354 | |
355 | /* | |
356 | * Temporary measure, we want to make sure we don't see | |
357 | * dirty data on the default backing_dev_info | |
358 | */ | |
6467716a AB |
359 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) { |
360 | del_timer(&me->wakeup_timer); | |
03ba3782 | 361 | wb_do_writeback(me, 0); |
6467716a | 362 | } |
03ba3782 | 363 | |
cfc4ba53 | 364 | spin_lock_bh(&bdi_lock); |
09f40f98 JK |
365 | /* |
366 | * In the following loop we are going to check whether we have | |
367 | * some work to do without any synchronization with tasks | |
20c8c628 AM |
368 | * waking us up to do work for them. Set the task state here |
369 | * so that we don't miss wakeups after verifying conditions. | |
09f40f98 | 370 | */ |
c5f7ad23 | 371 | set_current_state(TASK_INTERRUPTIBLE); |
03ba3782 | 372 | |
78c40cb6 | 373 | list_for_each_entry(bdi, &bdi_list, bdi_list) { |
adf39240 AB |
374 | bool have_dirty_io; |
375 | ||
376 | if (!bdi_cap_writeback_dirty(bdi) || | |
377 | bdi_cap_flush_forker(bdi)) | |
03ba3782 JA |
378 | continue; |
379 | ||
080dcec4 AB |
380 | WARN(!test_bit(BDI_registered, &bdi->state), |
381 | "bdi %p/%s is not registered!\n", bdi, bdi->name); | |
382 | ||
adf39240 AB |
383 | have_dirty_io = !list_empty(&bdi->work_list) || |
384 | wb_has_dirty_io(&bdi->wb); | |
78c40cb6 AB |
385 | |
386 | /* | |
adf39240 AB |
387 | * If the bdi has work to do, but the thread does not |
388 | * exist - create it. | |
78c40cb6 | 389 | */ |
adf39240 AB |
390 | if (!bdi->wb.task && have_dirty_io) { |
391 | /* | |
392 | * Set the pending bit - if someone will try to | |
393 | * unregister this bdi - it'll wait on this bit. | |
394 | */ | |
395 | set_bit(BDI_pending, &bdi->state); | |
396 | action = FORK_THREAD; | |
397 | break; | |
398 | } | |
fff5b85a | 399 | |
6bf05d03 JA |
400 | spin_lock(&bdi->wb_lock); |
401 | ||
fff5b85a AB |
402 | /* |
403 | * If there is no work to do and the bdi thread was | |
404 | * inactive long enough - kill it. The wb_lock is taken | |
405 | * to make sure no-one adds more work to this bdi and | |
406 | * wakes the bdi thread up. | |
407 | */ | |
408 | if (bdi->wb.task && !have_dirty_io && | |
409 | time_after(jiffies, bdi->wb.last_active + | |
410 | bdi_longest_inactive())) { | |
411 | task = bdi->wb.task; | |
412 | bdi->wb.task = NULL; | |
413 | spin_unlock(&bdi->wb_lock); | |
414 | set_bit(BDI_pending, &bdi->state); | |
415 | action = KILL_THREAD; | |
416 | break; | |
417 | } | |
6bf05d03 | 418 | spin_unlock(&bdi->wb_lock); |
03ba3782 | 419 | } |
080dcec4 | 420 | spin_unlock_bh(&bdi_lock); |
03ba3782 | 421 | |
c4ec7908 AB |
422 | /* Keep working if default bdi still has things to do */ |
423 | if (!list_empty(&me->bdi->work_list)) | |
424 | __set_current_state(TASK_RUNNING); | |
425 | ||
adf39240 AB |
426 | switch (action) { |
427 | case FORK_THREAD: | |
428 | __set_current_state(TASK_RUNNING); | |
6628bc74 AB |
429 | task = kthread_create(bdi_writeback_thread, &bdi->wb, |
430 | "flush-%s", dev_name(bdi->dev)); | |
adf39240 AB |
431 | if (IS_ERR(task)) { |
432 | /* | |
433 | * If thread creation fails, force writeout of | |
d46db3d5 WF |
434 | * the bdi from the thread. Hopefully 1024 is |
435 | * large enough for efficient IO. | |
adf39240 | 436 | */ |
0e175a18 CW |
437 | writeback_inodes_wb(&bdi->wb, 1024, |
438 | WB_REASON_FORKER_THREAD); | |
fff5b85a AB |
439 | } else { |
440 | /* | |
441 | * The spinlock makes sure we do not lose | |
442 | * wake-ups when racing with 'bdi_queue_work()'. | |
6628bc74 AB |
443 | * And as soon as the bdi thread is visible, we |
444 | * can start it. | |
fff5b85a | 445 | */ |
6467716a | 446 | spin_lock_bh(&bdi->wb_lock); |
adf39240 | 447 | bdi->wb.task = task; |
6467716a | 448 | spin_unlock_bh(&bdi->wb_lock); |
6628bc74 | 449 | wake_up_process(task); |
fff5b85a | 450 | } |
5a042aa4 | 451 | bdi_clear_pending(bdi); |
fff5b85a AB |
452 | break; |
453 | ||
454 | case KILL_THREAD: | |
455 | __set_current_state(TASK_RUNNING); | |
456 | kthread_stop(task); | |
5a042aa4 | 457 | bdi_clear_pending(bdi); |
adf39240 | 458 | break; |
03ba3782 | 459 | |
adf39240 | 460 | case NO_ACTION: |
253c34e9 AB |
461 | if (!wb_has_dirty_io(me) || !dirty_writeback_interval) |
462 | /* | |
463 | * There are no dirty data. The only thing we | |
464 | * should now care about is checking for | |
465 | * inactive bdi threads and killing them. Thus, | |
466 | * let's sleep for longer time, save energy and | |
467 | * be friendly for battery-driven devices. | |
468 | */ | |
469 | schedule_timeout(bdi_longest_inactive()); | |
6423104b | 470 | else |
253c34e9 | 471 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
03ba3782 | 472 | try_to_freeze(); |
5a042aa4 | 473 | break; |
03ba3782 | 474 | } |
03ba3782 JA |
475 | } |
476 | ||
477 | return 0; | |
478 | } | |
479 | ||
cfc4ba53 JA |
480 | /* |
481 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
482 | */ | |
483 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
484 | { | |
485 | spin_lock_bh(&bdi_lock); | |
486 | list_del_rcu(&bdi->bdi_list); | |
487 | spin_unlock_bh(&bdi_lock); | |
488 | ||
ef323088 | 489 | synchronize_rcu_expedited(); |
cfc4ba53 JA |
490 | } |
491 | ||
cf0ca9fe PZ |
492 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
493 | const char *fmt, ...) | |
494 | { | |
cf0ca9fe | 495 | va_list args; |
cf0ca9fe PZ |
496 | struct device *dev; |
497 | ||
69fc208b | 498 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
c284de61 | 499 | return 0; |
f1d0b063 | 500 | |
cf0ca9fe | 501 | va_start(args, fmt); |
19051c50 | 502 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 503 | va_end(args); |
c284de61 AB |
504 | if (IS_ERR(dev)) |
505 | return PTR_ERR(dev); | |
66f3b8e2 | 506 | |
cf0ca9fe | 507 | bdi->dev = dev; |
cf0ca9fe | 508 | |
03ba3782 JA |
509 | /* |
510 | * Just start the forker thread for our default backing_dev_info, | |
511 | * and add other bdi's to the list. They will get a thread created | |
512 | * on-demand when they need it. | |
513 | */ | |
514 | if (bdi_cap_flush_forker(bdi)) { | |
515 | struct bdi_writeback *wb = &bdi->wb; | |
516 | ||
6f904ff0 | 517 | wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s", |
03ba3782 | 518 | dev_name(dev)); |
c284de61 AB |
519 | if (IS_ERR(wb->task)) |
520 | return PTR_ERR(wb->task); | |
03ba3782 JA |
521 | } |
522 | ||
523 | bdi_debug_register(bdi, dev_name(dev)); | |
500b067c | 524 | set_bit(BDI_registered, &bdi->state); |
c284de61 AB |
525 | |
526 | spin_lock_bh(&bdi_lock); | |
527 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
528 | spin_unlock_bh(&bdi_lock); | |
529 | ||
455b2864 | 530 | trace_writeback_bdi_register(bdi); |
c284de61 | 531 | return 0; |
cf0ca9fe PZ |
532 | } |
533 | EXPORT_SYMBOL(bdi_register); | |
534 | ||
535 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
536 | { | |
537 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
538 | } | |
539 | EXPORT_SYMBOL(bdi_register_dev); | |
540 | ||
03ba3782 JA |
541 | /* |
542 | * Remove bdi from the global list and shutdown any threads we have running | |
543 | */ | |
544 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 545 | { |
2673b4cf RV |
546 | struct task_struct *task; |
547 | ||
03ba3782 JA |
548 | if (!bdi_cap_writeback_dirty(bdi)) |
549 | return; | |
550 | ||
551 | /* | |
fff5b85a | 552 | * Make sure nobody finds us on the bdi_list anymore |
03ba3782 | 553 | */ |
fff5b85a | 554 | bdi_remove_from_list(bdi); |
03ba3782 JA |
555 | |
556 | /* | |
fff5b85a | 557 | * If setup is pending, wait for that to complete first |
03ba3782 | 558 | */ |
fff5b85a AB |
559 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, |
560 | TASK_UNINTERRUPTIBLE); | |
03ba3782 JA |
561 | |
562 | /* | |
c1955ce3 | 563 | * Finally, kill the kernel thread. We don't need to be RCU |
8a32c441 | 564 | * safe anymore, since the bdi is gone from visibility. |
03ba3782 | 565 | */ |
2673b4cf RV |
566 | spin_lock_bh(&bdi->wb_lock); |
567 | task = bdi->wb.task; | |
568 | bdi->wb.task = NULL; | |
569 | spin_unlock_bh(&bdi->wb_lock); | |
570 | ||
571 | if (task) | |
572 | kthread_stop(task); | |
66f3b8e2 JA |
573 | } |
574 | ||
592b09a4 JA |
575 | /* |
576 | * This bdi is going away now, make sure that no super_blocks point to it | |
577 | */ | |
578 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
579 | { | |
580 | struct super_block *sb; | |
581 | ||
582 | spin_lock(&sb_lock); | |
583 | list_for_each_entry(sb, &super_blocks, s_list) { | |
584 | if (sb->s_bdi == bdi) | |
95f28604 | 585 | sb->s_bdi = &default_backing_dev_info; |
592b09a4 JA |
586 | } |
587 | spin_unlock(&sb_lock); | |
588 | } | |
589 | ||
cf0ca9fe PZ |
590 | void bdi_unregister(struct backing_dev_info *bdi) |
591 | { | |
2673b4cf RV |
592 | struct device *dev = bdi->dev; |
593 | ||
594 | if (dev) { | |
ccb6108f | 595 | bdi_set_min_ratio(bdi, 0); |
455b2864 | 596 | trace_writeback_bdi_unregister(bdi); |
8c4db335 | 597 | bdi_prune_sb(bdi); |
6467716a | 598 | del_timer_sync(&bdi->wb.wakeup_timer); |
8c4db335 | 599 | |
03ba3782 JA |
600 | if (!bdi_cap_flush_forker(bdi)) |
601 | bdi_wb_shutdown(bdi); | |
76f1418b | 602 | bdi_debug_unregister(bdi); |
2673b4cf RV |
603 | |
604 | spin_lock_bh(&bdi->wb_lock); | |
cf0ca9fe | 605 | bdi->dev = NULL; |
2673b4cf RV |
606 | spin_unlock_bh(&bdi->wb_lock); |
607 | ||
608 | device_unregister(dev); | |
cf0ca9fe PZ |
609 | } |
610 | } | |
611 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 612 | |
6467716a AB |
613 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
614 | { | |
615 | memset(wb, 0, sizeof(*wb)); | |
616 | ||
617 | wb->bdi = bdi; | |
618 | wb->last_old_flush = jiffies; | |
619 | INIT_LIST_HEAD(&wb->b_dirty); | |
620 | INIT_LIST_HEAD(&wb->b_io); | |
621 | INIT_LIST_HEAD(&wb->b_more_io); | |
f758eeab | 622 | spin_lock_init(&wb->list_lock); |
6467716a AB |
623 | setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi); |
624 | } | |
625 | ||
e98be2d5 WF |
626 | /* |
627 | * Initial write bandwidth: 100 MB/s | |
628 | */ | |
629 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) | |
630 | ||
b2e8fb6e PZ |
631 | int bdi_init(struct backing_dev_info *bdi) |
632 | { | |
03ba3782 | 633 | int i, err; |
b2e8fb6e | 634 | |
cf0ca9fe PZ |
635 | bdi->dev = NULL; |
636 | ||
189d3c4a | 637 | bdi->min_ratio = 0; |
a42dde04 | 638 | bdi->max_ratio = 100; |
eb608e3a | 639 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
03ba3782 | 640 | spin_lock_init(&bdi->wb_lock); |
66f3b8e2 | 641 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
642 | INIT_LIST_HEAD(&bdi->work_list); |
643 | ||
644 | bdi_wb_init(&bdi->wb, bdi); | |
645 | ||
b2e8fb6e | 646 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 647 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
648 | if (err) |
649 | goto err; | |
650 | } | |
651 | ||
652 | bdi->dirty_exceeded = 0; | |
e98be2d5 WF |
653 | |
654 | bdi->bw_time_stamp = jiffies; | |
655 | bdi->written_stamp = 0; | |
656 | ||
7381131c | 657 | bdi->balanced_dirty_ratelimit = INIT_BW; |
be3ffa27 | 658 | bdi->dirty_ratelimit = INIT_BW; |
e98be2d5 WF |
659 | bdi->write_bandwidth = INIT_BW; |
660 | bdi->avg_write_bandwidth = INIT_BW; | |
661 | ||
eb608e3a | 662 | err = fprop_local_init_percpu(&bdi->completions); |
04fbfdc1 PZ |
663 | |
664 | if (err) { | |
665 | err: | |
4b01a0b1 | 666 | while (i--) |
04fbfdc1 | 667 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
668 | } |
669 | ||
670 | return err; | |
671 | } | |
672 | EXPORT_SYMBOL(bdi_init); | |
673 | ||
674 | void bdi_destroy(struct backing_dev_info *bdi) | |
675 | { | |
676 | int i; | |
677 | ||
ce5f8e77 JA |
678 | /* |
679 | * Splice our entries to the default_backing_dev_info, if this | |
680 | * bdi disappears | |
681 | */ | |
682 | if (bdi_has_dirty_io(bdi)) { | |
683 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
684 | ||
f758eeab | 685 | bdi_lock_two(&bdi->wb, dst); |
ce5f8e77 JA |
686 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
687 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
688 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
f758eeab CH |
689 | spin_unlock(&bdi->wb.list_lock); |
690 | spin_unlock(&dst->list_lock); | |
ce5f8e77 | 691 | } |
66f3b8e2 | 692 | |
cf0ca9fe PZ |
693 | bdi_unregister(bdi); |
694 | ||
7a401a97 RV |
695 | /* |
696 | * If bdi_unregister() had already been called earlier, the | |
697 | * wakeup_timer could still be armed because bdi_prune_sb() | |
698 | * can race with the bdi_wakeup_thread_delayed() calls from | |
699 | * __mark_inode_dirty(). | |
700 | */ | |
701 | del_timer_sync(&bdi->wb.wakeup_timer); | |
702 | ||
b2e8fb6e PZ |
703 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
704 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 | 705 | |
eb608e3a | 706 | fprop_local_destroy_percpu(&bdi->completions); |
b2e8fb6e PZ |
707 | } |
708 | EXPORT_SYMBOL(bdi_destroy); | |
709 | ||
c3c53206 JA |
710 | /* |
711 | * For use from filesystems to quickly init and register a bdi associated | |
712 | * with dirty writeback | |
713 | */ | |
714 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | |
715 | unsigned int cap) | |
716 | { | |
717 | char tmp[32]; | |
718 | int err; | |
719 | ||
720 | bdi->name = name; | |
721 | bdi->capabilities = cap; | |
722 | err = bdi_init(bdi); | |
723 | if (err) | |
724 | return err; | |
725 | ||
726 | sprintf(tmp, "%.28s%s", name, "-%d"); | |
727 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); | |
728 | if (err) { | |
729 | bdi_destroy(bdi); | |
730 | return err; | |
731 | } | |
732 | ||
733 | return 0; | |
734 | } | |
735 | EXPORT_SYMBOL(bdi_setup_and_register); | |
736 | ||
3fcfab16 AM |
737 | static wait_queue_head_t congestion_wqh[2] = { |
738 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
739 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
740 | }; | |
0e093d99 | 741 | static atomic_t nr_bdi_congested[2]; |
3fcfab16 | 742 | |
1faa16d2 | 743 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
744 | { |
745 | enum bdi_state bit; | |
1faa16d2 | 746 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 747 | |
1faa16d2 | 748 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
749 | if (test_and_clear_bit(bit, &bdi->state)) |
750 | atomic_dec(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
751 | smp_mb__after_clear_bit(); |
752 | if (waitqueue_active(wqh)) | |
753 | wake_up(wqh); | |
754 | } | |
755 | EXPORT_SYMBOL(clear_bdi_congested); | |
756 | ||
1faa16d2 | 757 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
758 | { |
759 | enum bdi_state bit; | |
760 | ||
1faa16d2 | 761 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
762 | if (!test_and_set_bit(bit, &bdi->state)) |
763 | atomic_inc(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
764 | } |
765 | EXPORT_SYMBOL(set_bdi_congested); | |
766 | ||
767 | /** | |
768 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 769 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
770 | * @timeout: timeout in jiffies |
771 | * | |
772 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
773 | * write congestion. If no backing_devs are congested then just wait for the | |
774 | * next write to be completed. | |
775 | */ | |
8aa7e847 | 776 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
777 | { |
778 | long ret; | |
52bb9198 | 779 | unsigned long start = jiffies; |
3fcfab16 | 780 | DEFINE_WAIT(wait); |
8aa7e847 | 781 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
782 | |
783 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
784 | ret = io_schedule_timeout(timeout); | |
785 | finish_wait(wqh, &wait); | |
52bb9198 MG |
786 | |
787 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
788 | jiffies_to_usecs(jiffies - start)); | |
789 | ||
3fcfab16 AM |
790 | return ret; |
791 | } | |
792 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 793 | |
0e093d99 MG |
794 | /** |
795 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
796 | * @zone: A zone to check if it is heavily congested | |
797 | * @sync: SYNC or ASYNC IO | |
798 | * @timeout: timeout in jiffies | |
799 | * | |
800 | * In the event of a congested backing_dev (any backing_dev) and the given | |
801 | * @zone has experienced recent congestion, this waits for up to @timeout | |
802 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
803 | * or a write to complete. | |
804 | * | |
25985edc | 805 | * In the absence of zone congestion, cond_resched() is called to yield |
0e093d99 MG |
806 | * the processor if necessary but otherwise does not sleep. |
807 | * | |
808 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
809 | * it is the number of jiffies that were still remaining when the function | |
810 | * returned. return_value == timeout implies the function did not sleep. | |
811 | */ | |
812 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
813 | { | |
814 | long ret; | |
815 | unsigned long start = jiffies; | |
816 | DEFINE_WAIT(wait); | |
817 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
818 | ||
819 | /* | |
820 | * If there is no congestion, or heavy congestion is not being | |
821 | * encountered in the current zone, yield if necessary instead | |
822 | * of sleeping on the congestion queue | |
823 | */ | |
824 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || | |
825 | !zone_is_reclaim_congested(zone)) { | |
826 | cond_resched(); | |
827 | ||
828 | /* In case we scheduled, work out time remaining */ | |
829 | ret = timeout - (jiffies - start); | |
830 | if (ret < 0) | |
831 | ret = 0; | |
832 | ||
833 | goto out; | |
834 | } | |
835 | ||
836 | /* Sleep until uncongested or a write happens */ | |
837 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
838 | ret = io_schedule_timeout(timeout); | |
839 | finish_wait(wqh, &wait); | |
840 | ||
841 | out: | |
842 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
843 | jiffies_to_usecs(jiffies - start)); | |
844 | ||
845 | return ret; | |
846 | } | |
847 | EXPORT_SYMBOL(wait_iff_congested); | |
3965c9ae WL |
848 | |
849 | int pdflush_proc_obsolete(struct ctl_table *table, int write, | |
850 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
851 | { | |
852 | char kbuf[] = "0\n"; | |
853 | ||
854 | if (*ppos) { | |
855 | *lenp = 0; | |
856 | return 0; | |
857 | } | |
858 | ||
859 | if (copy_to_user(buffer, kbuf, sizeof(kbuf))) | |
860 | return -EFAULT; | |
861 | printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n", | |
862 | table->procname); | |
863 | ||
864 | *lenp = 2; | |
865 | *ppos += *lenp; | |
866 | return 2; | |
867 | } |