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 | |
5129a469 JE |
15 | struct backing_dev_info noop_backing_dev_info = { |
16 | .name = "noop", | |
976e48f8 | 17 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
5129a469 | 18 | }; |
a212b105 | 19 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); |
5129a469 | 20 | |
cf0ca9fe | 21 | static struct class *bdi_class; |
cfc4ba53 JA |
22 | |
23 | /* | |
181387da | 24 | * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side |
cfc4ba53 JA |
25 | * locking. |
26 | */ | |
03ba3782 | 27 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 28 | LIST_HEAD(bdi_list); |
03ba3782 | 29 | |
839a8e86 TH |
30 | /* bdi_wq serves all asynchronous writeback tasks */ |
31 | struct workqueue_struct *bdi_wq; | |
32 | ||
76f1418b MS |
33 | #ifdef CONFIG_DEBUG_FS |
34 | #include <linux/debugfs.h> | |
35 | #include <linux/seq_file.h> | |
36 | ||
37 | static struct dentry *bdi_debug_root; | |
38 | ||
39 | static void bdi_debug_init(void) | |
40 | { | |
41 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
42 | } | |
43 | ||
44 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
45 | { | |
46 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 47 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
48 | unsigned long background_thresh; |
49 | unsigned long dirty_thresh; | |
0d960a38 | 50 | unsigned long wb_thresh; |
0ae45f63 | 51 | unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time; |
f09b00d3 JA |
52 | struct inode *inode; |
53 | ||
0ae45f63 | 54 | nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0; |
f758eeab | 55 | spin_lock(&wb->list_lock); |
c7f54084 | 56 | list_for_each_entry(inode, &wb->b_dirty, i_io_list) |
c1955ce3 | 57 | nr_dirty++; |
c7f54084 | 58 | list_for_each_entry(inode, &wb->b_io, i_io_list) |
c1955ce3 | 59 | nr_io++; |
c7f54084 | 60 | list_for_each_entry(inode, &wb->b_more_io, i_io_list) |
c1955ce3 | 61 | nr_more_io++; |
c7f54084 | 62 | list_for_each_entry(inode, &wb->b_dirty_time, i_io_list) |
0ae45f63 TT |
63 | if (inode->i_state & I_DIRTY_TIME) |
64 | nr_dirty_time++; | |
f758eeab | 65 | spin_unlock(&wb->list_lock); |
76f1418b | 66 | |
16c4042f | 67 | global_dirty_limits(&background_thresh, &dirty_thresh); |
0d960a38 | 68 | wb_thresh = wb_calc_thresh(wb, dirty_thresh); |
76f1418b MS |
69 | |
70 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
71 | seq_printf(m, | |
00821b00 WF |
72 | "BdiWriteback: %10lu kB\n" |
73 | "BdiReclaimable: %10lu kB\n" | |
74 | "BdiDirtyThresh: %10lu kB\n" | |
75 | "DirtyThresh: %10lu kB\n" | |
76 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 77 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
78 | "BdiWritten: %10lu kB\n" |
79 | "BdiWriteBandwidth: %10lu kBps\n" | |
80 | "b_dirty: %10lu\n" | |
81 | "b_io: %10lu\n" | |
82 | "b_more_io: %10lu\n" | |
0ae45f63 | 83 | "b_dirty_time: %10lu\n" |
00821b00 WF |
84 | "bdi_list: %10u\n" |
85 | "state: %10lx\n", | |
93f78d88 TH |
86 | (unsigned long) K(wb_stat(wb, WB_WRITEBACK)), |
87 | (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)), | |
0d960a38 | 88 | K(wb_thresh), |
f7d2b1ec JK |
89 | K(dirty_thresh), |
90 | K(background_thresh), | |
93f78d88 TH |
91 | (unsigned long) K(wb_stat(wb, WB_DIRTIED)), |
92 | (unsigned long) K(wb_stat(wb, WB_WRITTEN)), | |
a88a341a | 93 | (unsigned long) K(wb->write_bandwidth), |
f7d2b1ec JK |
94 | nr_dirty, |
95 | nr_io, | |
96 | nr_more_io, | |
0ae45f63 | 97 | nr_dirty_time, |
4452226e | 98 | !list_empty(&bdi->bdi_list), bdi->wb.state); |
76f1418b MS |
99 | #undef K |
100 | ||
101 | return 0; | |
102 | } | |
103 | ||
104 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
105 | { | |
106 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
107 | } | |
108 | ||
109 | static const struct file_operations bdi_debug_stats_fops = { | |
110 | .open = bdi_debug_stats_open, | |
111 | .read = seq_read, | |
112 | .llseek = seq_lseek, | |
113 | .release = single_release, | |
114 | }; | |
115 | ||
97f07697 | 116 | static int bdi_debug_register(struct backing_dev_info *bdi, const char *name) |
76f1418b | 117 | { |
97f07697 | 118 | if (!bdi_debug_root) |
119 | return -ENOMEM; | |
120 | ||
76f1418b | 121 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); |
97f07697 | 122 | if (!bdi->debug_dir) |
123 | return -ENOMEM; | |
124 | ||
76f1418b MS |
125 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, |
126 | bdi, &bdi_debug_stats_fops); | |
97f07697 | 127 | if (!bdi->debug_stats) { |
128 | debugfs_remove(bdi->debug_dir); | |
129 | return -ENOMEM; | |
130 | } | |
131 | ||
132 | return 0; | |
76f1418b MS |
133 | } |
134 | ||
135 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
136 | { | |
137 | debugfs_remove(bdi->debug_stats); | |
138 | debugfs_remove(bdi->debug_dir); | |
139 | } | |
140 | #else | |
141 | static inline void bdi_debug_init(void) | |
142 | { | |
143 | } | |
97f07697 | 144 | static inline int bdi_debug_register(struct backing_dev_info *bdi, |
76f1418b MS |
145 | const char *name) |
146 | { | |
97f07697 | 147 | return 0; |
76f1418b MS |
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); \ | |
d9e1241e GKH |
180 | } \ |
181 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
182 | |
183 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
184 | ||
189d3c4a PZ |
185 | static ssize_t min_ratio_store(struct device *dev, |
186 | struct device_attribute *attr, const char *buf, size_t count) | |
187 | { | |
188 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 189 | unsigned int ratio; |
7034ed13 NJ |
190 | ssize_t ret; |
191 | ||
192 | ret = kstrtouint(buf, 10, &ratio); | |
193 | if (ret < 0) | |
194 | return ret; | |
195 | ||
196 | ret = bdi_set_min_ratio(bdi, ratio); | |
197 | if (!ret) | |
198 | ret = count; | |
189d3c4a | 199 | |
189d3c4a PZ |
200 | return ret; |
201 | } | |
202 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
203 | ||
a42dde04 PZ |
204 | static ssize_t max_ratio_store(struct device *dev, |
205 | struct device_attribute *attr, const char *buf, size_t count) | |
206 | { | |
207 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 208 | unsigned int ratio; |
7034ed13 NJ |
209 | ssize_t ret; |
210 | ||
211 | ret = kstrtouint(buf, 10, &ratio); | |
212 | if (ret < 0) | |
213 | return ret; | |
214 | ||
215 | ret = bdi_set_max_ratio(bdi, ratio); | |
216 | if (!ret) | |
217 | ret = count; | |
a42dde04 | 218 | |
a42dde04 PZ |
219 | return ret; |
220 | } | |
221 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
222 | ||
7d311cda DW |
223 | static ssize_t stable_pages_required_show(struct device *dev, |
224 | struct device_attribute *attr, | |
225 | char *page) | |
226 | { | |
227 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
228 | ||
229 | return snprintf(page, PAGE_SIZE-1, "%d\n", | |
230 | bdi_cap_stable_pages_required(bdi) ? 1 : 0); | |
231 | } | |
d9e1241e GKH |
232 | static DEVICE_ATTR_RO(stable_pages_required); |
233 | ||
234 | static struct attribute *bdi_dev_attrs[] = { | |
235 | &dev_attr_read_ahead_kb.attr, | |
236 | &dev_attr_min_ratio.attr, | |
237 | &dev_attr_max_ratio.attr, | |
238 | &dev_attr_stable_pages_required.attr, | |
239 | NULL, | |
cf0ca9fe | 240 | }; |
d9e1241e | 241 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
242 | |
243 | static __init int bdi_class_init(void) | |
244 | { | |
245 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
246 | if (IS_ERR(bdi_class)) |
247 | return PTR_ERR(bdi_class); | |
248 | ||
d9e1241e | 249 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 250 | bdi_debug_init(); |
d03f6cdc | 251 | |
cf0ca9fe PZ |
252 | return 0; |
253 | } | |
76f1418b | 254 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 255 | |
2e82b84c JK |
256 | static int bdi_init(struct backing_dev_info *bdi); |
257 | ||
26160158 JA |
258 | static int __init default_bdi_init(void) |
259 | { | |
260 | int err; | |
261 | ||
839a8e86 | 262 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE | |
b5c872dd | 263 | WQ_UNBOUND | WQ_SYSFS, 0); |
839a8e86 TH |
264 | if (!bdi_wq) |
265 | return -ENOMEM; | |
266 | ||
976e48f8 | 267 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
268 | |
269 | return err; | |
270 | } | |
271 | subsys_initcall(default_bdi_init); | |
272 | ||
6467716a | 273 | /* |
f0054bb1 | 274 | * This function is used when the first inode for this wb is marked dirty. It |
6467716a AB |
275 | * wakes-up the corresponding bdi thread which should then take care of the |
276 | * periodic background write-out of dirty inodes. Since the write-out would | |
277 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
278 | * set up a timer which wakes the bdi thread up later. | |
279 | * | |
280 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
281 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
282 | * by delaying the wake-up. | |
6ca738d6 DB |
283 | * |
284 | * We have to be careful not to postpone flush work if it is scheduled for | |
285 | * earlier. Thus we use queue_delayed_work(). | |
6467716a | 286 | */ |
f0054bb1 | 287 | void wb_wakeup_delayed(struct bdi_writeback *wb) |
6467716a AB |
288 | { |
289 | unsigned long timeout; | |
290 | ||
291 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
f0054bb1 TH |
292 | spin_lock_bh(&wb->work_lock); |
293 | if (test_bit(WB_registered, &wb->state)) | |
294 | queue_delayed_work(bdi_wq, &wb->dwork, timeout); | |
295 | spin_unlock_bh(&wb->work_lock); | |
03ba3782 JA |
296 | } |
297 | ||
cfc4ba53 | 298 | /* |
a88a341a | 299 | * Initial write bandwidth: 100 MB/s |
cfc4ba53 | 300 | */ |
a88a341a | 301 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) |
cfc4ba53 | 302 | |
8395cd9f | 303 | static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi, |
a13f35e8 | 304 | int blkcg_id, gfp_t gfp) |
cf0ca9fe | 305 | { |
93f78d88 | 306 | int i, err; |
cf0ca9fe | 307 | |
6467716a | 308 | memset(wb, 0, sizeof(*wb)); |
f1d0b063 | 309 | |
810df54a JK |
310 | if (wb != &bdi->wb) |
311 | bdi_get(bdi); | |
6467716a AB |
312 | wb->bdi = bdi; |
313 | wb->last_old_flush = jiffies; | |
314 | INIT_LIST_HEAD(&wb->b_dirty); | |
315 | INIT_LIST_HEAD(&wb->b_io); | |
316 | INIT_LIST_HEAD(&wb->b_more_io); | |
0ae45f63 | 317 | INIT_LIST_HEAD(&wb->b_dirty_time); |
f758eeab | 318 | spin_lock_init(&wb->list_lock); |
66f3b8e2 | 319 | |
a88a341a TH |
320 | wb->bw_time_stamp = jiffies; |
321 | wb->balanced_dirty_ratelimit = INIT_BW; | |
322 | wb->dirty_ratelimit = INIT_BW; | |
323 | wb->write_bandwidth = INIT_BW; | |
324 | wb->avg_write_bandwidth = INIT_BW; | |
cf0ca9fe | 325 | |
f0054bb1 TH |
326 | spin_lock_init(&wb->work_lock); |
327 | INIT_LIST_HEAD(&wb->work_list); | |
328 | INIT_DELAYED_WORK(&wb->dwork, wb_workfn); | |
b57d74af | 329 | wb->dirty_sleep = jiffies; |
c284de61 | 330 | |
a13f35e8 | 331 | wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp); |
810df54a JK |
332 | if (!wb->congested) { |
333 | err = -ENOMEM; | |
334 | goto out_put_bdi; | |
335 | } | |
a13f35e8 | 336 | |
8395cd9f | 337 | err = fprop_local_init_percpu(&wb->completions, gfp); |
a88a341a | 338 | if (err) |
a13f35e8 | 339 | goto out_put_cong; |
c284de61 | 340 | |
93f78d88 | 341 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) { |
8395cd9f | 342 | err = percpu_counter_init(&wb->stat[i], 0, gfp); |
a13f35e8 TH |
343 | if (err) |
344 | goto out_destroy_stat; | |
93f78d88 | 345 | } |
cf0ca9fe | 346 | |
93f78d88 | 347 | return 0; |
a13f35e8 TH |
348 | |
349 | out_destroy_stat: | |
078c6c3a | 350 | while (i--) |
a13f35e8 TH |
351 | percpu_counter_destroy(&wb->stat[i]); |
352 | fprop_local_destroy_percpu(&wb->completions); | |
353 | out_put_cong: | |
354 | wb_congested_put(wb->congested); | |
810df54a JK |
355 | out_put_bdi: |
356 | if (wb != &bdi->wb) | |
357 | bdi_put(bdi); | |
a13f35e8 | 358 | return err; |
cf0ca9fe | 359 | } |
cf0ca9fe | 360 | |
e8cb72b3 JK |
361 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb); |
362 | ||
03ba3782 JA |
363 | /* |
364 | * Remove bdi from the global list and shutdown any threads we have running | |
365 | */ | |
46100071 | 366 | static void wb_shutdown(struct bdi_writeback *wb) |
66f3b8e2 | 367 | { |
c4db59d3 | 368 | /* Make sure nobody queues further work */ |
46100071 TH |
369 | spin_lock_bh(&wb->work_lock); |
370 | if (!test_and_clear_bit(WB_registered, &wb->state)) { | |
371 | spin_unlock_bh(&wb->work_lock); | |
5318ce7d JK |
372 | /* |
373 | * Wait for wb shutdown to finish if someone else is just | |
374 | * running wb_shutdown(). Otherwise we could proceed to wb / | |
375 | * bdi destruction before wb_shutdown() is finished. | |
376 | */ | |
377 | wait_on_bit(&wb->state, WB_shutting_down, TASK_UNINTERRUPTIBLE); | |
03ba3782 | 378 | return; |
c4db59d3 | 379 | } |
5318ce7d | 380 | set_bit(WB_shutting_down, &wb->state); |
46100071 | 381 | spin_unlock_bh(&wb->work_lock); |
03ba3782 | 382 | |
e8cb72b3 | 383 | cgwb_remove_from_bdi_list(wb); |
03ba3782 | 384 | /* |
46100071 TH |
385 | * Drain work list and shutdown the delayed_work. !WB_registered |
386 | * tells wb_workfn() that @wb is dying and its work_list needs to | |
387 | * be drained no matter what. | |
03ba3782 | 388 | */ |
46100071 TH |
389 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
390 | flush_delayed_work(&wb->dwork); | |
391 | WARN_ON(!list_empty(&wb->work_list)); | |
5318ce7d JK |
392 | /* |
393 | * Make sure bit gets cleared after shutdown is finished. Matches with | |
394 | * the barrier provided by test_and_clear_bit() above. | |
395 | */ | |
396 | smp_wmb(); | |
397 | clear_bit(WB_shutting_down, &wb->state); | |
46100071 TH |
398 | } |
399 | ||
f0054bb1 | 400 | static void wb_exit(struct bdi_writeback *wb) |
93f78d88 TH |
401 | { |
402 | int i; | |
403 | ||
404 | WARN_ON(delayed_work_pending(&wb->dwork)); | |
405 | ||
406 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) | |
407 | percpu_counter_destroy(&wb->stat[i]); | |
6467716a | 408 | |
a88a341a | 409 | fprop_local_destroy_percpu(&wb->completions); |
a13f35e8 | 410 | wb_congested_put(wb->congested); |
810df54a JK |
411 | if (wb != &wb->bdi->wb) |
412 | bdi_put(wb->bdi); | |
a88a341a | 413 | } |
e98be2d5 | 414 | |
52ebea74 TH |
415 | #ifdef CONFIG_CGROUP_WRITEBACK |
416 | ||
417 | #include <linux/memcontrol.h> | |
418 | ||
419 | /* | |
420 | * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree, | |
421 | * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU | |
4514451e | 422 | * protected. |
52ebea74 TH |
423 | */ |
424 | static DEFINE_SPINLOCK(cgwb_lock); | |
52ebea74 TH |
425 | |
426 | /** | |
427 | * wb_congested_get_create - get or create a wb_congested | |
428 | * @bdi: associated bdi | |
429 | * @blkcg_id: ID of the associated blkcg | |
430 | * @gfp: allocation mask | |
431 | * | |
432 | * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one. | |
433 | * The returned wb_congested has its reference count incremented. Returns | |
434 | * NULL on failure. | |
435 | */ | |
436 | struct bdi_writeback_congested * | |
437 | wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp) | |
438 | { | |
439 | struct bdi_writeback_congested *new_congested = NULL, *congested; | |
440 | struct rb_node **node, *parent; | |
441 | unsigned long flags; | |
52ebea74 TH |
442 | retry: |
443 | spin_lock_irqsave(&cgwb_lock, flags); | |
444 | ||
445 | node = &bdi->cgwb_congested_tree.rb_node; | |
446 | parent = NULL; | |
447 | ||
448 | while (*node != NULL) { | |
449 | parent = *node; | |
bc71226b GT |
450 | congested = rb_entry(parent, struct bdi_writeback_congested, |
451 | rb_node); | |
52ebea74 TH |
452 | if (congested->blkcg_id < blkcg_id) |
453 | node = &parent->rb_left; | |
454 | else if (congested->blkcg_id > blkcg_id) | |
455 | node = &parent->rb_right; | |
456 | else | |
457 | goto found; | |
458 | } | |
459 | ||
460 | if (new_congested) { | |
461 | /* !found and storage for new one already allocated, insert */ | |
462 | congested = new_congested; | |
463 | new_congested = NULL; | |
464 | rb_link_node(&congested->rb_node, parent, node); | |
465 | rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree); | |
52ebea74 TH |
466 | goto found; |
467 | } | |
468 | ||
469 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
470 | ||
471 | /* allocate storage for new one and retry */ | |
472 | new_congested = kzalloc(sizeof(*new_congested), gfp); | |
473 | if (!new_congested) | |
474 | return NULL; | |
475 | ||
476 | atomic_set(&new_congested->refcnt, 0); | |
b7d680d7 | 477 | new_congested->__bdi = bdi; |
52ebea74 TH |
478 | new_congested->blkcg_id = blkcg_id; |
479 | goto retry; | |
480 | ||
481 | found: | |
482 | atomic_inc(&congested->refcnt); | |
483 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
484 | kfree(new_congested); | |
485 | return congested; | |
486 | } | |
487 | ||
488 | /** | |
489 | * wb_congested_put - put a wb_congested | |
490 | * @congested: wb_congested to put | |
491 | * | |
492 | * Put @congested and destroy it if the refcnt reaches zero. | |
493 | */ | |
494 | void wb_congested_put(struct bdi_writeback_congested *congested) | |
495 | { | |
52ebea74 TH |
496 | unsigned long flags; |
497 | ||
52ebea74 TH |
498 | local_irq_save(flags); |
499 | if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) { | |
500 | local_irq_restore(flags); | |
501 | return; | |
502 | } | |
503 | ||
a20135ff | 504 | /* bdi might already have been destroyed leaving @congested unlinked */ |
b7d680d7 | 505 | if (congested->__bdi) { |
a20135ff | 506 | rb_erase(&congested->rb_node, |
b7d680d7 JK |
507 | &congested->__bdi->cgwb_congested_tree); |
508 | congested->__bdi = NULL; | |
a20135ff TH |
509 | } |
510 | ||
52ebea74 TH |
511 | spin_unlock_irqrestore(&cgwb_lock, flags); |
512 | kfree(congested); | |
52ebea74 TH |
513 | } |
514 | ||
515 | static void cgwb_release_workfn(struct work_struct *work) | |
516 | { | |
517 | struct bdi_writeback *wb = container_of(work, struct bdi_writeback, | |
518 | release_work); | |
52ebea74 TH |
519 | |
520 | wb_shutdown(wb); | |
521 | ||
522 | css_put(wb->memcg_css); | |
523 | css_put(wb->blkcg_css); | |
52ebea74 | 524 | |
841710aa | 525 | fprop_local_destroy_percpu(&wb->memcg_completions); |
52ebea74 TH |
526 | percpu_ref_exit(&wb->refcnt); |
527 | wb_exit(wb); | |
528 | kfree_rcu(wb, rcu); | |
52ebea74 TH |
529 | } |
530 | ||
531 | static void cgwb_release(struct percpu_ref *refcnt) | |
532 | { | |
533 | struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback, | |
534 | refcnt); | |
535 | schedule_work(&wb->release_work); | |
536 | } | |
537 | ||
538 | static void cgwb_kill(struct bdi_writeback *wb) | |
539 | { | |
540 | lockdep_assert_held(&cgwb_lock); | |
541 | ||
542 | WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id)); | |
543 | list_del(&wb->memcg_node); | |
544 | list_del(&wb->blkcg_node); | |
545 | percpu_ref_kill(&wb->refcnt); | |
546 | } | |
547 | ||
e8cb72b3 JK |
548 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) |
549 | { | |
550 | spin_lock_irq(&cgwb_lock); | |
551 | list_del_rcu(&wb->bdi_node); | |
552 | spin_unlock_irq(&cgwb_lock); | |
553 | } | |
554 | ||
52ebea74 TH |
555 | static int cgwb_create(struct backing_dev_info *bdi, |
556 | struct cgroup_subsys_state *memcg_css, gfp_t gfp) | |
557 | { | |
558 | struct mem_cgroup *memcg; | |
559 | struct cgroup_subsys_state *blkcg_css; | |
560 | struct blkcg *blkcg; | |
561 | struct list_head *memcg_cgwb_list, *blkcg_cgwb_list; | |
562 | struct bdi_writeback *wb; | |
563 | unsigned long flags; | |
564 | int ret = 0; | |
565 | ||
566 | memcg = mem_cgroup_from_css(memcg_css); | |
c165b3e3 | 567 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); |
52ebea74 TH |
568 | blkcg = css_to_blkcg(blkcg_css); |
569 | memcg_cgwb_list = mem_cgroup_cgwb_list(memcg); | |
570 | blkcg_cgwb_list = &blkcg->cgwb_list; | |
571 | ||
572 | /* look up again under lock and discard on blkcg mismatch */ | |
573 | spin_lock_irqsave(&cgwb_lock, flags); | |
574 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
575 | if (wb && wb->blkcg_css != blkcg_css) { | |
576 | cgwb_kill(wb); | |
577 | wb = NULL; | |
578 | } | |
579 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
580 | if (wb) | |
581 | goto out_put; | |
582 | ||
583 | /* need to create a new one */ | |
584 | wb = kmalloc(sizeof(*wb), gfp); | |
0b045bd1 CJ |
585 | if (!wb) { |
586 | ret = -ENOMEM; | |
587 | goto out_put; | |
588 | } | |
52ebea74 | 589 | |
a13f35e8 | 590 | ret = wb_init(wb, bdi, blkcg_css->id, gfp); |
52ebea74 TH |
591 | if (ret) |
592 | goto err_free; | |
593 | ||
594 | ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp); | |
595 | if (ret) | |
596 | goto err_wb_exit; | |
597 | ||
841710aa TH |
598 | ret = fprop_local_init_percpu(&wb->memcg_completions, gfp); |
599 | if (ret) | |
600 | goto err_ref_exit; | |
601 | ||
52ebea74 TH |
602 | wb->memcg_css = memcg_css; |
603 | wb->blkcg_css = blkcg_css; | |
604 | INIT_WORK(&wb->release_work, cgwb_release_workfn); | |
605 | set_bit(WB_registered, &wb->state); | |
03ba3782 JA |
606 | |
607 | /* | |
52ebea74 TH |
608 | * The root wb determines the registered state of the whole bdi and |
609 | * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate | |
610 | * whether they're still online. Don't link @wb if any is dead. | |
611 | * See wb_memcg_offline() and wb_blkcg_offline(). | |
03ba3782 | 612 | */ |
52ebea74 TH |
613 | ret = -ENODEV; |
614 | spin_lock_irqsave(&cgwb_lock, flags); | |
615 | if (test_bit(WB_registered, &bdi->wb.state) && | |
616 | blkcg_cgwb_list->next && memcg_cgwb_list->next) { | |
617 | /* we might have raced another instance of this function */ | |
618 | ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb); | |
619 | if (!ret) { | |
b817525a | 620 | list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list); |
52ebea74 TH |
621 | list_add(&wb->memcg_node, memcg_cgwb_list); |
622 | list_add(&wb->blkcg_node, blkcg_cgwb_list); | |
623 | css_get(memcg_css); | |
624 | css_get(blkcg_css); | |
625 | } | |
626 | } | |
627 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
628 | if (ret) { | |
629 | if (ret == -EEXIST) | |
630 | ret = 0; | |
a13f35e8 | 631 | goto err_fprop_exit; |
52ebea74 TH |
632 | } |
633 | goto out_put; | |
634 | ||
841710aa TH |
635 | err_fprop_exit: |
636 | fprop_local_destroy_percpu(&wb->memcg_completions); | |
52ebea74 TH |
637 | err_ref_exit: |
638 | percpu_ref_exit(&wb->refcnt); | |
639 | err_wb_exit: | |
640 | wb_exit(wb); | |
641 | err_free: | |
642 | kfree(wb); | |
643 | out_put: | |
644 | css_put(blkcg_css); | |
645 | return ret; | |
66f3b8e2 JA |
646 | } |
647 | ||
52ebea74 TH |
648 | /** |
649 | * wb_get_create - get wb for a given memcg, create if necessary | |
650 | * @bdi: target bdi | |
651 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
652 | * @gfp: allocation mask to use | |
653 | * | |
654 | * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to | |
655 | * create one. The returned wb has its refcount incremented. | |
656 | * | |
657 | * This function uses css_get() on @memcg_css and thus expects its refcnt | |
658 | * to be positive on invocation. IOW, rcu_read_lock() protection on | |
659 | * @memcg_css isn't enough. try_get it before calling this function. | |
660 | * | |
661 | * A wb is keyed by its associated memcg. As blkcg implicitly enables | |
662 | * memcg on the default hierarchy, memcg association is guaranteed to be | |
663 | * more specific (equal or descendant to the associated blkcg) and thus can | |
664 | * identify both the memcg and blkcg associations. | |
665 | * | |
666 | * Because the blkcg associated with a memcg may change as blkcg is enabled | |
667 | * and disabled closer to root in the hierarchy, each wb keeps track of | |
668 | * both the memcg and blkcg associated with it and verifies the blkcg on | |
669 | * each lookup. On mismatch, the existing wb is discarded and a new one is | |
670 | * created. | |
671 | */ | |
672 | struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, | |
673 | struct cgroup_subsys_state *memcg_css, | |
674 | gfp_t gfp) | |
6467716a | 675 | { |
52ebea74 TH |
676 | struct bdi_writeback *wb; |
677 | ||
d0164adc | 678 | might_sleep_if(gfpflags_allow_blocking(gfp)); |
52ebea74 TH |
679 | |
680 | if (!memcg_css->parent) | |
681 | return &bdi->wb; | |
682 | ||
683 | do { | |
684 | rcu_read_lock(); | |
685 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
686 | if (wb) { | |
687 | struct cgroup_subsys_state *blkcg_css; | |
688 | ||
689 | /* see whether the blkcg association has changed */ | |
690 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, | |
c165b3e3 | 691 | &io_cgrp_subsys); |
52ebea74 TH |
692 | if (unlikely(wb->blkcg_css != blkcg_css || |
693 | !wb_tryget(wb))) | |
694 | wb = NULL; | |
695 | css_put(blkcg_css); | |
696 | } | |
697 | rcu_read_unlock(); | |
698 | } while (!wb && !cgwb_create(bdi, memcg_css, gfp)); | |
699 | ||
700 | return wb; | |
701 | } | |
6467716a | 702 | |
a13f35e8 | 703 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
52ebea74 | 704 | { |
a13f35e8 TH |
705 | int ret; |
706 | ||
52ebea74 TH |
707 | INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC); |
708 | bdi->cgwb_congested_tree = RB_ROOT; | |
a13f35e8 TH |
709 | |
710 | ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL); | |
711 | if (!ret) { | |
7d828602 | 712 | bdi->wb.memcg_css = &root_mem_cgroup->css; |
a13f35e8 TH |
713 | bdi->wb.blkcg_css = blkcg_root_css; |
714 | } | |
715 | return ret; | |
6467716a AB |
716 | } |
717 | ||
b1c51afc | 718 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) |
52ebea74 TH |
719 | { |
720 | struct radix_tree_iter iter; | |
721 | void **slot; | |
5318ce7d | 722 | struct bdi_writeback *wb; |
52ebea74 TH |
723 | |
724 | WARN_ON(test_bit(WB_registered, &bdi->wb.state)); | |
725 | ||
726 | spin_lock_irq(&cgwb_lock); | |
727 | radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0) | |
728 | cgwb_kill(*slot); | |
5318ce7d JK |
729 | |
730 | while (!list_empty(&bdi->wb_list)) { | |
731 | wb = list_first_entry(&bdi->wb_list, struct bdi_writeback, | |
732 | bdi_node); | |
733 | spin_unlock_irq(&cgwb_lock); | |
734 | wb_shutdown(wb); | |
735 | spin_lock_irq(&cgwb_lock); | |
736 | } | |
52ebea74 | 737 | spin_unlock_irq(&cgwb_lock); |
52ebea74 TH |
738 | } |
739 | ||
740 | /** | |
741 | * wb_memcg_offline - kill all wb's associated with a memcg being offlined | |
742 | * @memcg: memcg being offlined | |
743 | * | |
744 | * Also prevents creation of any new wb's associated with @memcg. | |
e98be2d5 | 745 | */ |
52ebea74 TH |
746 | void wb_memcg_offline(struct mem_cgroup *memcg) |
747 | { | |
748 | LIST_HEAD(to_destroy); | |
749 | struct list_head *memcg_cgwb_list = mem_cgroup_cgwb_list(memcg); | |
750 | struct bdi_writeback *wb, *next; | |
751 | ||
752 | spin_lock_irq(&cgwb_lock); | |
753 | list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node) | |
754 | cgwb_kill(wb); | |
755 | memcg_cgwb_list->next = NULL; /* prevent new wb's */ | |
756 | spin_unlock_irq(&cgwb_lock); | |
757 | } | |
758 | ||
759 | /** | |
760 | * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined | |
761 | * @blkcg: blkcg being offlined | |
762 | * | |
763 | * Also prevents creation of any new wb's associated with @blkcg. | |
764 | */ | |
765 | void wb_blkcg_offline(struct blkcg *blkcg) | |
766 | { | |
767 | LIST_HEAD(to_destroy); | |
768 | struct bdi_writeback *wb, *next; | |
769 | ||
770 | spin_lock_irq(&cgwb_lock); | |
771 | list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node) | |
772 | cgwb_kill(wb); | |
773 | blkcg->cgwb_list.next = NULL; /* prevent new wb's */ | |
774 | spin_unlock_irq(&cgwb_lock); | |
775 | } | |
776 | ||
df23de55 JK |
777 | static void cgwb_bdi_exit(struct backing_dev_info *bdi) |
778 | { | |
779 | struct rb_node *rbn; | |
780 | ||
781 | spin_lock_irq(&cgwb_lock); | |
782 | while ((rbn = rb_first(&bdi->cgwb_congested_tree))) { | |
783 | struct bdi_writeback_congested *congested = | |
784 | rb_entry(rbn, struct bdi_writeback_congested, rb_node); | |
785 | ||
786 | rb_erase(rbn, &bdi->cgwb_congested_tree); | |
b7d680d7 | 787 | congested->__bdi = NULL; /* mark @congested unlinked */ |
df23de55 JK |
788 | } |
789 | spin_unlock_irq(&cgwb_lock); | |
790 | } | |
791 | ||
e8cb72b3 JK |
792 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
793 | { | |
794 | spin_lock_irq(&cgwb_lock); | |
795 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
796 | spin_unlock_irq(&cgwb_lock); | |
797 | } | |
798 | ||
52ebea74 TH |
799 | #else /* CONFIG_CGROUP_WRITEBACK */ |
800 | ||
a13f35e8 TH |
801 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
802 | { | |
803 | int err; | |
804 | ||
805 | bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL); | |
806 | if (!bdi->wb_congested) | |
807 | return -ENOMEM; | |
808 | ||
5f478e4e TH |
809 | atomic_set(&bdi->wb_congested->refcnt, 1); |
810 | ||
a13f35e8 TH |
811 | err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL); |
812 | if (err) { | |
5f478e4e | 813 | wb_congested_put(bdi->wb_congested); |
a13f35e8 TH |
814 | return err; |
815 | } | |
816 | return 0; | |
817 | } | |
818 | ||
b1c51afc | 819 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { } |
df23de55 JK |
820 | |
821 | static void cgwb_bdi_exit(struct backing_dev_info *bdi) | |
5f478e4e TH |
822 | { |
823 | wb_congested_put(bdi->wb_congested); | |
824 | } | |
52ebea74 | 825 | |
e8cb72b3 JK |
826 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
827 | { | |
828 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
829 | } | |
830 | ||
831 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) | |
832 | { | |
833 | list_del_rcu(&wb->bdi_node); | |
834 | } | |
835 | ||
52ebea74 | 836 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
e98be2d5 | 837 | |
2e82b84c | 838 | static int bdi_init(struct backing_dev_info *bdi) |
b2e8fb6e | 839 | { |
b817525a TH |
840 | int ret; |
841 | ||
cf0ca9fe PZ |
842 | bdi->dev = NULL; |
843 | ||
d03f6cdc | 844 | kref_init(&bdi->refcnt); |
189d3c4a | 845 | bdi->min_ratio = 0; |
a42dde04 | 846 | bdi->max_ratio = 100; |
eb608e3a | 847 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
66f3b8e2 | 848 | INIT_LIST_HEAD(&bdi->bdi_list); |
b817525a | 849 | INIT_LIST_HEAD(&bdi->wb_list); |
cc395d7f | 850 | init_waitqueue_head(&bdi->wb_waitq); |
03ba3782 | 851 | |
b817525a TH |
852 | ret = cgwb_bdi_init(bdi); |
853 | ||
b817525a | 854 | return ret; |
b2e8fb6e | 855 | } |
e98be2d5 | 856 | |
d03f6cdc JK |
857 | struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id) |
858 | { | |
859 | struct backing_dev_info *bdi; | |
860 | ||
861 | bdi = kmalloc_node(sizeof(struct backing_dev_info), | |
862 | gfp_mask | __GFP_ZERO, node_id); | |
863 | if (!bdi) | |
864 | return NULL; | |
865 | ||
866 | if (bdi_init(bdi)) { | |
867 | kfree(bdi); | |
868 | return NULL; | |
869 | } | |
870 | return bdi; | |
871 | } | |
62bf42ad | 872 | EXPORT_SYMBOL(bdi_alloc_node); |
d03f6cdc | 873 | |
7c4cc300 | 874 | int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args) |
46100071 | 875 | { |
46100071 | 876 | struct device *dev; |
e98be2d5 | 877 | |
46100071 TH |
878 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
879 | return 0; | |
e98be2d5 | 880 | |
7c4cc300 | 881 | dev = device_create_vargs(bdi_class, NULL, MKDEV(0, 0), bdi, fmt, args); |
46100071 TH |
882 | if (IS_ERR(dev)) |
883 | return PTR_ERR(dev); | |
04fbfdc1 | 884 | |
e8cb72b3 | 885 | cgwb_bdi_register(bdi); |
46100071 | 886 | bdi->dev = dev; |
b2e8fb6e | 887 | |
6d0e4827 | 888 | bdi_debug_register(bdi, dev_name(dev)); |
46100071 TH |
889 | set_bit(WB_registered, &bdi->wb.state); |
890 | ||
891 | spin_lock_bh(&bdi_lock); | |
892 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
893 | spin_unlock_bh(&bdi_lock); | |
894 | ||
895 | trace_writeback_bdi_register(bdi); | |
896 | return 0; | |
b2e8fb6e | 897 | } |
baf7a616 JK |
898 | EXPORT_SYMBOL(bdi_register_va); |
899 | ||
7c4cc300 | 900 | int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...) |
baf7a616 JK |
901 | { |
902 | va_list args; | |
903 | int ret; | |
904 | ||
905 | va_start(args, fmt); | |
7c4cc300 | 906 | ret = bdi_register_va(bdi, fmt, args); |
baf7a616 JK |
907 | va_end(args); |
908 | return ret; | |
909 | } | |
46100071 | 910 | EXPORT_SYMBOL(bdi_register); |
b2e8fb6e | 911 | |
df08c32c DW |
912 | int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner) |
913 | { | |
914 | int rc; | |
915 | ||
7c4cc300 | 916 | rc = bdi_register(bdi, "%u:%u", MAJOR(owner->devt), MINOR(owner->devt)); |
df08c32c DW |
917 | if (rc) |
918 | return rc; | |
b6f8fec4 JK |
919 | /* Leaking owner reference... */ |
920 | WARN_ON(bdi->owner); | |
df08c32c DW |
921 | bdi->owner = owner; |
922 | get_device(owner); | |
923 | return 0; | |
924 | } | |
925 | EXPORT_SYMBOL(bdi_register_owner); | |
926 | ||
46100071 TH |
927 | /* |
928 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
929 | */ | |
930 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
931 | { | |
932 | spin_lock_bh(&bdi_lock); | |
933 | list_del_rcu(&bdi->bdi_list); | |
934 | spin_unlock_bh(&bdi_lock); | |
b2e8fb6e | 935 | |
46100071 TH |
936 | synchronize_rcu_expedited(); |
937 | } | |
cf0ca9fe | 938 | |
b02176f3 | 939 | void bdi_unregister(struct backing_dev_info *bdi) |
b2e8fb6e | 940 | { |
f0054bb1 TH |
941 | /* make sure nobody finds us on the bdi_list anymore */ |
942 | bdi_remove_from_list(bdi); | |
943 | wb_shutdown(&bdi->wb); | |
b1c51afc | 944 | cgwb_bdi_unregister(bdi); |
7a401a97 | 945 | |
c4db59d3 CH |
946 | if (bdi->dev) { |
947 | bdi_debug_unregister(bdi); | |
948 | device_unregister(bdi->dev); | |
949 | bdi->dev = NULL; | |
950 | } | |
df08c32c DW |
951 | |
952 | if (bdi->owner) { | |
953 | put_device(bdi->owner); | |
954 | bdi->owner = NULL; | |
955 | } | |
b02176f3 | 956 | } |
c4db59d3 | 957 | |
d03f6cdc JK |
958 | static void release_bdi(struct kref *ref) |
959 | { | |
960 | struct backing_dev_info *bdi = | |
961 | container_of(ref, struct backing_dev_info, refcnt); | |
962 | ||
5af110b2 JK |
963 | if (test_bit(WB_registered, &bdi->wb.state)) |
964 | bdi_unregister(bdi); | |
2e82b84c JK |
965 | WARN_ON_ONCE(bdi->dev); |
966 | wb_exit(&bdi->wb); | |
967 | cgwb_bdi_exit(bdi); | |
d03f6cdc JK |
968 | kfree(bdi); |
969 | } | |
970 | ||
971 | void bdi_put(struct backing_dev_info *bdi) | |
972 | { | |
973 | kref_put(&bdi->refcnt, release_bdi); | |
974 | } | |
62bf42ad | 975 | EXPORT_SYMBOL(bdi_put); |
d03f6cdc | 976 | |
3fcfab16 AM |
977 | static wait_queue_head_t congestion_wqh[2] = { |
978 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
979 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
980 | }; | |
ec8a6f26 | 981 | static atomic_t nr_wb_congested[2]; |
3fcfab16 | 982 | |
ec8a6f26 | 983 | void clear_wb_congested(struct bdi_writeback_congested *congested, int sync) |
3fcfab16 | 984 | { |
1faa16d2 | 985 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
c877ef8a | 986 | enum wb_congested_state bit; |
3fcfab16 | 987 | |
4452226e | 988 | bit = sync ? WB_sync_congested : WB_async_congested; |
ec8a6f26 TH |
989 | if (test_and_clear_bit(bit, &congested->state)) |
990 | atomic_dec(&nr_wb_congested[sync]); | |
4e857c58 | 991 | smp_mb__after_atomic(); |
3fcfab16 AM |
992 | if (waitqueue_active(wqh)) |
993 | wake_up(wqh); | |
994 | } | |
ec8a6f26 | 995 | EXPORT_SYMBOL(clear_wb_congested); |
3fcfab16 | 996 | |
ec8a6f26 | 997 | void set_wb_congested(struct bdi_writeback_congested *congested, int sync) |
3fcfab16 | 998 | { |
c877ef8a | 999 | enum wb_congested_state bit; |
3fcfab16 | 1000 | |
4452226e | 1001 | bit = sync ? WB_sync_congested : WB_async_congested; |
ec8a6f26 TH |
1002 | if (!test_and_set_bit(bit, &congested->state)) |
1003 | atomic_inc(&nr_wb_congested[sync]); | |
3fcfab16 | 1004 | } |
ec8a6f26 | 1005 | EXPORT_SYMBOL(set_wb_congested); |
3fcfab16 AM |
1006 | |
1007 | /** | |
1008 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 1009 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
1010 | * @timeout: timeout in jiffies |
1011 | * | |
1012 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
1013 | * write congestion. If no backing_devs are congested then just wait for the | |
1014 | * next write to be completed. | |
1015 | */ | |
8aa7e847 | 1016 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
1017 | { |
1018 | long ret; | |
52bb9198 | 1019 | unsigned long start = jiffies; |
3fcfab16 | 1020 | DEFINE_WAIT(wait); |
8aa7e847 | 1021 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
1022 | |
1023 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1024 | ret = io_schedule_timeout(timeout); | |
1025 | finish_wait(wqh, &wait); | |
52bb9198 MG |
1026 | |
1027 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
1028 | jiffies_to_usecs(jiffies - start)); | |
1029 | ||
3fcfab16 AM |
1030 | return ret; |
1031 | } | |
1032 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 1033 | |
0e093d99 | 1034 | /** |
599d0c95 MG |
1035 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes |
1036 | * @pgdat: A pgdat to check if it is heavily congested | |
0e093d99 MG |
1037 | * @sync: SYNC or ASYNC IO |
1038 | * @timeout: timeout in jiffies | |
1039 | * | |
1040 | * In the event of a congested backing_dev (any backing_dev) and the given | |
599d0c95 | 1041 | * @pgdat has experienced recent congestion, this waits for up to @timeout |
0e093d99 MG |
1042 | * jiffies for either a BDI to exit congestion of the given @sync queue |
1043 | * or a write to complete. | |
1044 | * | |
599d0c95 | 1045 | * In the absence of pgdat congestion, cond_resched() is called to yield |
ede37713 | 1046 | * the processor if necessary but otherwise does not sleep. |
0e093d99 MG |
1047 | * |
1048 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
1049 | * it is the number of jiffies that were still remaining when the function | |
1050 | * returned. return_value == timeout implies the function did not sleep. | |
1051 | */ | |
599d0c95 | 1052 | long wait_iff_congested(struct pglist_data *pgdat, int sync, long timeout) |
0e093d99 MG |
1053 | { |
1054 | long ret; | |
1055 | unsigned long start = jiffies; | |
1056 | DEFINE_WAIT(wait); | |
1057 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
1058 | ||
1059 | /* | |
1060 | * If there is no congestion, or heavy congestion is not being | |
599d0c95 | 1061 | * encountered in the current pgdat, yield if necessary instead |
0e093d99 MG |
1062 | * of sleeping on the congestion queue |
1063 | */ | |
ec8a6f26 | 1064 | if (atomic_read(&nr_wb_congested[sync]) == 0 || |
599d0c95 | 1065 | !test_bit(PGDAT_CONGESTED, &pgdat->flags)) { |
ede37713 | 1066 | cond_resched(); |
599d0c95 | 1067 | |
0e093d99 MG |
1068 | /* In case we scheduled, work out time remaining */ |
1069 | ret = timeout - (jiffies - start); | |
1070 | if (ret < 0) | |
1071 | ret = 0; | |
1072 | ||
1073 | goto out; | |
1074 | } | |
1075 | ||
1076 | /* Sleep until uncongested or a write happens */ | |
1077 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1078 | ret = io_schedule_timeout(timeout); | |
1079 | finish_wait(wqh, &wait); | |
1080 | ||
1081 | out: | |
1082 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
1083 | jiffies_to_usecs(jiffies - start)); | |
1084 | ||
1085 | return ret; | |
1086 | } | |
1087 | EXPORT_SYMBOL(wait_iff_congested); |