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 | ||
116 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
117 | { | |
118 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
119 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
120 | bdi, &bdi_debug_stats_fops); | |
121 | } | |
122 | ||
123 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
124 | { | |
125 | debugfs_remove(bdi->debug_stats); | |
126 | debugfs_remove(bdi->debug_dir); | |
127 | } | |
128 | #else | |
129 | static inline void bdi_debug_init(void) | |
130 | { | |
131 | } | |
132 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
133 | const char *name) | |
134 | { | |
135 | } | |
136 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
137 | { | |
138 | } | |
139 | #endif | |
140 | ||
cf0ca9fe PZ |
141 | static ssize_t read_ahead_kb_store(struct device *dev, |
142 | struct device_attribute *attr, | |
143 | const char *buf, size_t count) | |
144 | { | |
145 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 146 | unsigned long read_ahead_kb; |
7034ed13 | 147 | ssize_t ret; |
cf0ca9fe | 148 | |
7034ed13 NJ |
149 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
150 | if (ret < 0) | |
151 | return ret; | |
152 | ||
153 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
154 | ||
155 | return count; | |
cf0ca9fe PZ |
156 | } |
157 | ||
158 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
159 | ||
160 | #define BDI_SHOW(name, expr) \ | |
161 | static ssize_t name##_show(struct device *dev, \ | |
162 | struct device_attribute *attr, char *page) \ | |
163 | { \ | |
164 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
165 | \ | |
166 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
d9e1241e GKH |
167 | } \ |
168 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
169 | |
170 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
171 | ||
189d3c4a PZ |
172 | static ssize_t min_ratio_store(struct device *dev, |
173 | struct device_attribute *attr, const char *buf, size_t count) | |
174 | { | |
175 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 176 | unsigned int ratio; |
7034ed13 NJ |
177 | ssize_t ret; |
178 | ||
179 | ret = kstrtouint(buf, 10, &ratio); | |
180 | if (ret < 0) | |
181 | return ret; | |
182 | ||
183 | ret = bdi_set_min_ratio(bdi, ratio); | |
184 | if (!ret) | |
185 | ret = count; | |
189d3c4a | 186 | |
189d3c4a PZ |
187 | return ret; |
188 | } | |
189 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
190 | ||
a42dde04 PZ |
191 | static ssize_t max_ratio_store(struct device *dev, |
192 | struct device_attribute *attr, const char *buf, size_t count) | |
193 | { | |
194 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 195 | unsigned int ratio; |
7034ed13 NJ |
196 | ssize_t ret; |
197 | ||
198 | ret = kstrtouint(buf, 10, &ratio); | |
199 | if (ret < 0) | |
200 | return ret; | |
201 | ||
202 | ret = bdi_set_max_ratio(bdi, ratio); | |
203 | if (!ret) | |
204 | ret = count; | |
a42dde04 | 205 | |
a42dde04 PZ |
206 | return ret; |
207 | } | |
208 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
209 | ||
7d311cda DW |
210 | static ssize_t stable_pages_required_show(struct device *dev, |
211 | struct device_attribute *attr, | |
212 | char *page) | |
213 | { | |
214 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
215 | ||
216 | return snprintf(page, PAGE_SIZE-1, "%d\n", | |
217 | bdi_cap_stable_pages_required(bdi) ? 1 : 0); | |
218 | } | |
d9e1241e GKH |
219 | static DEVICE_ATTR_RO(stable_pages_required); |
220 | ||
221 | static struct attribute *bdi_dev_attrs[] = { | |
222 | &dev_attr_read_ahead_kb.attr, | |
223 | &dev_attr_min_ratio.attr, | |
224 | &dev_attr_max_ratio.attr, | |
225 | &dev_attr_stable_pages_required.attr, | |
226 | NULL, | |
cf0ca9fe | 227 | }; |
d9e1241e | 228 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
229 | |
230 | static __init int bdi_class_init(void) | |
231 | { | |
232 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
233 | if (IS_ERR(bdi_class)) |
234 | return PTR_ERR(bdi_class); | |
235 | ||
d9e1241e | 236 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 237 | bdi_debug_init(); |
d03f6cdc | 238 | |
cf0ca9fe PZ |
239 | return 0; |
240 | } | |
76f1418b | 241 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 242 | |
2e82b84c JK |
243 | static int bdi_init(struct backing_dev_info *bdi); |
244 | ||
26160158 JA |
245 | static int __init default_bdi_init(void) |
246 | { | |
247 | int err; | |
248 | ||
839a8e86 | 249 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE | |
b5c872dd | 250 | WQ_UNBOUND | WQ_SYSFS, 0); |
839a8e86 TH |
251 | if (!bdi_wq) |
252 | return -ENOMEM; | |
253 | ||
976e48f8 | 254 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
255 | |
256 | return err; | |
257 | } | |
258 | subsys_initcall(default_bdi_init); | |
259 | ||
6467716a | 260 | /* |
f0054bb1 | 261 | * This function is used when the first inode for this wb is marked dirty. It |
6467716a AB |
262 | * wakes-up the corresponding bdi thread which should then take care of the |
263 | * periodic background write-out of dirty inodes. Since the write-out would | |
264 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
265 | * set up a timer which wakes the bdi thread up later. | |
266 | * | |
267 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
268 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
269 | * by delaying the wake-up. | |
6ca738d6 DB |
270 | * |
271 | * We have to be careful not to postpone flush work if it is scheduled for | |
272 | * earlier. Thus we use queue_delayed_work(). | |
6467716a | 273 | */ |
f0054bb1 | 274 | void wb_wakeup_delayed(struct bdi_writeback *wb) |
6467716a AB |
275 | { |
276 | unsigned long timeout; | |
277 | ||
278 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
f0054bb1 TH |
279 | spin_lock_bh(&wb->work_lock); |
280 | if (test_bit(WB_registered, &wb->state)) | |
281 | queue_delayed_work(bdi_wq, &wb->dwork, timeout); | |
282 | spin_unlock_bh(&wb->work_lock); | |
03ba3782 JA |
283 | } |
284 | ||
cfc4ba53 | 285 | /* |
a88a341a | 286 | * Initial write bandwidth: 100 MB/s |
cfc4ba53 | 287 | */ |
a88a341a | 288 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) |
cfc4ba53 | 289 | |
8395cd9f | 290 | static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi, |
a13f35e8 | 291 | int blkcg_id, gfp_t gfp) |
cf0ca9fe | 292 | { |
93f78d88 | 293 | int i, err; |
cf0ca9fe | 294 | |
6467716a | 295 | memset(wb, 0, sizeof(*wb)); |
f1d0b063 | 296 | |
810df54a JK |
297 | if (wb != &bdi->wb) |
298 | bdi_get(bdi); | |
6467716a AB |
299 | wb->bdi = bdi; |
300 | wb->last_old_flush = jiffies; | |
301 | INIT_LIST_HEAD(&wb->b_dirty); | |
302 | INIT_LIST_HEAD(&wb->b_io); | |
303 | INIT_LIST_HEAD(&wb->b_more_io); | |
0ae45f63 | 304 | INIT_LIST_HEAD(&wb->b_dirty_time); |
f758eeab | 305 | spin_lock_init(&wb->list_lock); |
66f3b8e2 | 306 | |
a88a341a TH |
307 | wb->bw_time_stamp = jiffies; |
308 | wb->balanced_dirty_ratelimit = INIT_BW; | |
309 | wb->dirty_ratelimit = INIT_BW; | |
310 | wb->write_bandwidth = INIT_BW; | |
311 | wb->avg_write_bandwidth = INIT_BW; | |
cf0ca9fe | 312 | |
f0054bb1 TH |
313 | spin_lock_init(&wb->work_lock); |
314 | INIT_LIST_HEAD(&wb->work_list); | |
315 | INIT_DELAYED_WORK(&wb->dwork, wb_workfn); | |
b57d74af | 316 | wb->dirty_sleep = jiffies; |
c284de61 | 317 | |
a13f35e8 | 318 | wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp); |
810df54a JK |
319 | if (!wb->congested) { |
320 | err = -ENOMEM; | |
321 | goto out_put_bdi; | |
322 | } | |
a13f35e8 | 323 | |
8395cd9f | 324 | err = fprop_local_init_percpu(&wb->completions, gfp); |
a88a341a | 325 | if (err) |
a13f35e8 | 326 | goto out_put_cong; |
c284de61 | 327 | |
93f78d88 | 328 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) { |
8395cd9f | 329 | err = percpu_counter_init(&wb->stat[i], 0, gfp); |
a13f35e8 TH |
330 | if (err) |
331 | goto out_destroy_stat; | |
93f78d88 | 332 | } |
cf0ca9fe | 333 | |
93f78d88 | 334 | return 0; |
a13f35e8 TH |
335 | |
336 | out_destroy_stat: | |
078c6c3a | 337 | while (i--) |
a13f35e8 TH |
338 | percpu_counter_destroy(&wb->stat[i]); |
339 | fprop_local_destroy_percpu(&wb->completions); | |
340 | out_put_cong: | |
341 | wb_congested_put(wb->congested); | |
810df54a JK |
342 | out_put_bdi: |
343 | if (wb != &bdi->wb) | |
344 | bdi_put(bdi); | |
a13f35e8 | 345 | return err; |
cf0ca9fe | 346 | } |
cf0ca9fe | 347 | |
e8cb72b3 JK |
348 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb); |
349 | ||
03ba3782 JA |
350 | /* |
351 | * Remove bdi from the global list and shutdown any threads we have running | |
352 | */ | |
46100071 | 353 | static void wb_shutdown(struct bdi_writeback *wb) |
66f3b8e2 | 354 | { |
c4db59d3 | 355 | /* Make sure nobody queues further work */ |
46100071 TH |
356 | spin_lock_bh(&wb->work_lock); |
357 | if (!test_and_clear_bit(WB_registered, &wb->state)) { | |
358 | spin_unlock_bh(&wb->work_lock); | |
5318ce7d JK |
359 | /* |
360 | * Wait for wb shutdown to finish if someone else is just | |
361 | * running wb_shutdown(). Otherwise we could proceed to wb / | |
362 | * bdi destruction before wb_shutdown() is finished. | |
363 | */ | |
364 | wait_on_bit(&wb->state, WB_shutting_down, TASK_UNINTERRUPTIBLE); | |
03ba3782 | 365 | return; |
c4db59d3 | 366 | } |
5318ce7d | 367 | set_bit(WB_shutting_down, &wb->state); |
46100071 | 368 | spin_unlock_bh(&wb->work_lock); |
03ba3782 | 369 | |
e8cb72b3 | 370 | cgwb_remove_from_bdi_list(wb); |
03ba3782 | 371 | /* |
46100071 TH |
372 | * Drain work list and shutdown the delayed_work. !WB_registered |
373 | * tells wb_workfn() that @wb is dying and its work_list needs to | |
374 | * be drained no matter what. | |
03ba3782 | 375 | */ |
46100071 TH |
376 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
377 | flush_delayed_work(&wb->dwork); | |
378 | WARN_ON(!list_empty(&wb->work_list)); | |
5318ce7d JK |
379 | /* |
380 | * Make sure bit gets cleared after shutdown is finished. Matches with | |
381 | * the barrier provided by test_and_clear_bit() above. | |
382 | */ | |
383 | smp_wmb(); | |
384 | clear_bit(WB_shutting_down, &wb->state); | |
46100071 TH |
385 | } |
386 | ||
f0054bb1 | 387 | static void wb_exit(struct bdi_writeback *wb) |
93f78d88 TH |
388 | { |
389 | int i; | |
390 | ||
391 | WARN_ON(delayed_work_pending(&wb->dwork)); | |
392 | ||
393 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) | |
394 | percpu_counter_destroy(&wb->stat[i]); | |
6467716a | 395 | |
a88a341a | 396 | fprop_local_destroy_percpu(&wb->completions); |
a13f35e8 | 397 | wb_congested_put(wb->congested); |
810df54a JK |
398 | if (wb != &wb->bdi->wb) |
399 | bdi_put(wb->bdi); | |
a88a341a | 400 | } |
e98be2d5 | 401 | |
52ebea74 TH |
402 | #ifdef CONFIG_CGROUP_WRITEBACK |
403 | ||
404 | #include <linux/memcontrol.h> | |
405 | ||
406 | /* | |
407 | * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree, | |
408 | * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU | |
4514451e | 409 | * protected. |
52ebea74 TH |
410 | */ |
411 | static DEFINE_SPINLOCK(cgwb_lock); | |
52ebea74 TH |
412 | |
413 | /** | |
414 | * wb_congested_get_create - get or create a wb_congested | |
415 | * @bdi: associated bdi | |
416 | * @blkcg_id: ID of the associated blkcg | |
417 | * @gfp: allocation mask | |
418 | * | |
419 | * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one. | |
420 | * The returned wb_congested has its reference count incremented. Returns | |
421 | * NULL on failure. | |
422 | */ | |
423 | struct bdi_writeback_congested * | |
424 | wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp) | |
425 | { | |
426 | struct bdi_writeback_congested *new_congested = NULL, *congested; | |
427 | struct rb_node **node, *parent; | |
428 | unsigned long flags; | |
52ebea74 TH |
429 | retry: |
430 | spin_lock_irqsave(&cgwb_lock, flags); | |
431 | ||
432 | node = &bdi->cgwb_congested_tree.rb_node; | |
433 | parent = NULL; | |
434 | ||
435 | while (*node != NULL) { | |
436 | parent = *node; | |
bc71226b GT |
437 | congested = rb_entry(parent, struct bdi_writeback_congested, |
438 | rb_node); | |
52ebea74 TH |
439 | if (congested->blkcg_id < blkcg_id) |
440 | node = &parent->rb_left; | |
441 | else if (congested->blkcg_id > blkcg_id) | |
442 | node = &parent->rb_right; | |
443 | else | |
444 | goto found; | |
445 | } | |
446 | ||
447 | if (new_congested) { | |
448 | /* !found and storage for new one already allocated, insert */ | |
449 | congested = new_congested; | |
450 | new_congested = NULL; | |
451 | rb_link_node(&congested->rb_node, parent, node); | |
452 | rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree); | |
52ebea74 TH |
453 | goto found; |
454 | } | |
455 | ||
456 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
457 | ||
458 | /* allocate storage for new one and retry */ | |
459 | new_congested = kzalloc(sizeof(*new_congested), gfp); | |
460 | if (!new_congested) | |
461 | return NULL; | |
462 | ||
463 | atomic_set(&new_congested->refcnt, 0); | |
b7d680d7 | 464 | new_congested->__bdi = bdi; |
52ebea74 TH |
465 | new_congested->blkcg_id = blkcg_id; |
466 | goto retry; | |
467 | ||
468 | found: | |
469 | atomic_inc(&congested->refcnt); | |
470 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
471 | kfree(new_congested); | |
472 | return congested; | |
473 | } | |
474 | ||
475 | /** | |
476 | * wb_congested_put - put a wb_congested | |
477 | * @congested: wb_congested to put | |
478 | * | |
479 | * Put @congested and destroy it if the refcnt reaches zero. | |
480 | */ | |
481 | void wb_congested_put(struct bdi_writeback_congested *congested) | |
482 | { | |
52ebea74 TH |
483 | unsigned long flags; |
484 | ||
52ebea74 TH |
485 | local_irq_save(flags); |
486 | if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) { | |
487 | local_irq_restore(flags); | |
488 | return; | |
489 | } | |
490 | ||
a20135ff | 491 | /* bdi might already have been destroyed leaving @congested unlinked */ |
b7d680d7 | 492 | if (congested->__bdi) { |
a20135ff | 493 | rb_erase(&congested->rb_node, |
b7d680d7 JK |
494 | &congested->__bdi->cgwb_congested_tree); |
495 | congested->__bdi = NULL; | |
a20135ff TH |
496 | } |
497 | ||
52ebea74 TH |
498 | spin_unlock_irqrestore(&cgwb_lock, flags); |
499 | kfree(congested); | |
52ebea74 TH |
500 | } |
501 | ||
502 | static void cgwb_release_workfn(struct work_struct *work) | |
503 | { | |
504 | struct bdi_writeback *wb = container_of(work, struct bdi_writeback, | |
505 | release_work); | |
52ebea74 TH |
506 | |
507 | wb_shutdown(wb); | |
508 | ||
509 | css_put(wb->memcg_css); | |
510 | css_put(wb->blkcg_css); | |
52ebea74 | 511 | |
841710aa | 512 | fprop_local_destroy_percpu(&wb->memcg_completions); |
52ebea74 TH |
513 | percpu_ref_exit(&wb->refcnt); |
514 | wb_exit(wb); | |
515 | kfree_rcu(wb, rcu); | |
52ebea74 TH |
516 | } |
517 | ||
518 | static void cgwb_release(struct percpu_ref *refcnt) | |
519 | { | |
520 | struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback, | |
521 | refcnt); | |
522 | schedule_work(&wb->release_work); | |
523 | } | |
524 | ||
525 | static void cgwb_kill(struct bdi_writeback *wb) | |
526 | { | |
527 | lockdep_assert_held(&cgwb_lock); | |
528 | ||
529 | WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id)); | |
530 | list_del(&wb->memcg_node); | |
531 | list_del(&wb->blkcg_node); | |
532 | percpu_ref_kill(&wb->refcnt); | |
533 | } | |
534 | ||
e8cb72b3 JK |
535 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) |
536 | { | |
537 | spin_lock_irq(&cgwb_lock); | |
538 | list_del_rcu(&wb->bdi_node); | |
539 | spin_unlock_irq(&cgwb_lock); | |
540 | } | |
541 | ||
52ebea74 TH |
542 | static int cgwb_create(struct backing_dev_info *bdi, |
543 | struct cgroup_subsys_state *memcg_css, gfp_t gfp) | |
544 | { | |
545 | struct mem_cgroup *memcg; | |
546 | struct cgroup_subsys_state *blkcg_css; | |
547 | struct blkcg *blkcg; | |
548 | struct list_head *memcg_cgwb_list, *blkcg_cgwb_list; | |
549 | struct bdi_writeback *wb; | |
550 | unsigned long flags; | |
551 | int ret = 0; | |
552 | ||
553 | memcg = mem_cgroup_from_css(memcg_css); | |
c165b3e3 | 554 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); |
52ebea74 TH |
555 | blkcg = css_to_blkcg(blkcg_css); |
556 | memcg_cgwb_list = mem_cgroup_cgwb_list(memcg); | |
557 | blkcg_cgwb_list = &blkcg->cgwb_list; | |
558 | ||
559 | /* look up again under lock and discard on blkcg mismatch */ | |
560 | spin_lock_irqsave(&cgwb_lock, flags); | |
561 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
562 | if (wb && wb->blkcg_css != blkcg_css) { | |
563 | cgwb_kill(wb); | |
564 | wb = NULL; | |
565 | } | |
566 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
567 | if (wb) | |
568 | goto out_put; | |
569 | ||
570 | /* need to create a new one */ | |
571 | wb = kmalloc(sizeof(*wb), gfp); | |
0b045bd1 CJ |
572 | if (!wb) { |
573 | ret = -ENOMEM; | |
574 | goto out_put; | |
575 | } | |
52ebea74 | 576 | |
a13f35e8 | 577 | ret = wb_init(wb, bdi, blkcg_css->id, gfp); |
52ebea74 TH |
578 | if (ret) |
579 | goto err_free; | |
580 | ||
581 | ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp); | |
582 | if (ret) | |
583 | goto err_wb_exit; | |
584 | ||
841710aa TH |
585 | ret = fprop_local_init_percpu(&wb->memcg_completions, gfp); |
586 | if (ret) | |
587 | goto err_ref_exit; | |
588 | ||
52ebea74 TH |
589 | wb->memcg_css = memcg_css; |
590 | wb->blkcg_css = blkcg_css; | |
591 | INIT_WORK(&wb->release_work, cgwb_release_workfn); | |
592 | set_bit(WB_registered, &wb->state); | |
03ba3782 JA |
593 | |
594 | /* | |
52ebea74 TH |
595 | * The root wb determines the registered state of the whole bdi and |
596 | * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate | |
597 | * whether they're still online. Don't link @wb if any is dead. | |
598 | * See wb_memcg_offline() and wb_blkcg_offline(). | |
03ba3782 | 599 | */ |
52ebea74 TH |
600 | ret = -ENODEV; |
601 | spin_lock_irqsave(&cgwb_lock, flags); | |
602 | if (test_bit(WB_registered, &bdi->wb.state) && | |
603 | blkcg_cgwb_list->next && memcg_cgwb_list->next) { | |
604 | /* we might have raced another instance of this function */ | |
605 | ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb); | |
606 | if (!ret) { | |
b817525a | 607 | list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list); |
52ebea74 TH |
608 | list_add(&wb->memcg_node, memcg_cgwb_list); |
609 | list_add(&wb->blkcg_node, blkcg_cgwb_list); | |
610 | css_get(memcg_css); | |
611 | css_get(blkcg_css); | |
612 | } | |
613 | } | |
614 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
615 | if (ret) { | |
616 | if (ret == -EEXIST) | |
617 | ret = 0; | |
a13f35e8 | 618 | goto err_fprop_exit; |
52ebea74 TH |
619 | } |
620 | goto out_put; | |
621 | ||
841710aa TH |
622 | err_fprop_exit: |
623 | fprop_local_destroy_percpu(&wb->memcg_completions); | |
52ebea74 TH |
624 | err_ref_exit: |
625 | percpu_ref_exit(&wb->refcnt); | |
626 | err_wb_exit: | |
627 | wb_exit(wb); | |
628 | err_free: | |
629 | kfree(wb); | |
630 | out_put: | |
631 | css_put(blkcg_css); | |
632 | return ret; | |
66f3b8e2 JA |
633 | } |
634 | ||
52ebea74 TH |
635 | /** |
636 | * wb_get_create - get wb for a given memcg, create if necessary | |
637 | * @bdi: target bdi | |
638 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
639 | * @gfp: allocation mask to use | |
640 | * | |
641 | * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to | |
642 | * create one. The returned wb has its refcount incremented. | |
643 | * | |
644 | * This function uses css_get() on @memcg_css and thus expects its refcnt | |
645 | * to be positive on invocation. IOW, rcu_read_lock() protection on | |
646 | * @memcg_css isn't enough. try_get it before calling this function. | |
647 | * | |
648 | * A wb is keyed by its associated memcg. As blkcg implicitly enables | |
649 | * memcg on the default hierarchy, memcg association is guaranteed to be | |
650 | * more specific (equal or descendant to the associated blkcg) and thus can | |
651 | * identify both the memcg and blkcg associations. | |
652 | * | |
653 | * Because the blkcg associated with a memcg may change as blkcg is enabled | |
654 | * and disabled closer to root in the hierarchy, each wb keeps track of | |
655 | * both the memcg and blkcg associated with it and verifies the blkcg on | |
656 | * each lookup. On mismatch, the existing wb is discarded and a new one is | |
657 | * created. | |
658 | */ | |
659 | struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, | |
660 | struct cgroup_subsys_state *memcg_css, | |
661 | gfp_t gfp) | |
6467716a | 662 | { |
52ebea74 TH |
663 | struct bdi_writeback *wb; |
664 | ||
d0164adc | 665 | might_sleep_if(gfpflags_allow_blocking(gfp)); |
52ebea74 TH |
666 | |
667 | if (!memcg_css->parent) | |
668 | return &bdi->wb; | |
669 | ||
670 | do { | |
671 | rcu_read_lock(); | |
672 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
673 | if (wb) { | |
674 | struct cgroup_subsys_state *blkcg_css; | |
675 | ||
676 | /* see whether the blkcg association has changed */ | |
677 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, | |
c165b3e3 | 678 | &io_cgrp_subsys); |
52ebea74 TH |
679 | if (unlikely(wb->blkcg_css != blkcg_css || |
680 | !wb_tryget(wb))) | |
681 | wb = NULL; | |
682 | css_put(blkcg_css); | |
683 | } | |
684 | rcu_read_unlock(); | |
685 | } while (!wb && !cgwb_create(bdi, memcg_css, gfp)); | |
686 | ||
687 | return wb; | |
688 | } | |
6467716a | 689 | |
a13f35e8 | 690 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
52ebea74 | 691 | { |
a13f35e8 TH |
692 | int ret; |
693 | ||
52ebea74 TH |
694 | INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC); |
695 | bdi->cgwb_congested_tree = RB_ROOT; | |
a13f35e8 TH |
696 | |
697 | ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL); | |
698 | if (!ret) { | |
7d828602 | 699 | bdi->wb.memcg_css = &root_mem_cgroup->css; |
a13f35e8 TH |
700 | bdi->wb.blkcg_css = blkcg_root_css; |
701 | } | |
702 | return ret; | |
6467716a AB |
703 | } |
704 | ||
b1c51afc | 705 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) |
52ebea74 TH |
706 | { |
707 | struct radix_tree_iter iter; | |
708 | void **slot; | |
5318ce7d | 709 | struct bdi_writeback *wb; |
52ebea74 TH |
710 | |
711 | WARN_ON(test_bit(WB_registered, &bdi->wb.state)); | |
712 | ||
713 | spin_lock_irq(&cgwb_lock); | |
714 | radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0) | |
715 | cgwb_kill(*slot); | |
5318ce7d JK |
716 | |
717 | while (!list_empty(&bdi->wb_list)) { | |
718 | wb = list_first_entry(&bdi->wb_list, struct bdi_writeback, | |
719 | bdi_node); | |
720 | spin_unlock_irq(&cgwb_lock); | |
721 | wb_shutdown(wb); | |
722 | spin_lock_irq(&cgwb_lock); | |
723 | } | |
52ebea74 | 724 | spin_unlock_irq(&cgwb_lock); |
52ebea74 TH |
725 | } |
726 | ||
727 | /** | |
728 | * wb_memcg_offline - kill all wb's associated with a memcg being offlined | |
729 | * @memcg: memcg being offlined | |
730 | * | |
731 | * Also prevents creation of any new wb's associated with @memcg. | |
e98be2d5 | 732 | */ |
52ebea74 TH |
733 | void wb_memcg_offline(struct mem_cgroup *memcg) |
734 | { | |
735 | LIST_HEAD(to_destroy); | |
736 | struct list_head *memcg_cgwb_list = mem_cgroup_cgwb_list(memcg); | |
737 | struct bdi_writeback *wb, *next; | |
738 | ||
739 | spin_lock_irq(&cgwb_lock); | |
740 | list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node) | |
741 | cgwb_kill(wb); | |
742 | memcg_cgwb_list->next = NULL; /* prevent new wb's */ | |
743 | spin_unlock_irq(&cgwb_lock); | |
744 | } | |
745 | ||
746 | /** | |
747 | * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined | |
748 | * @blkcg: blkcg being offlined | |
749 | * | |
750 | * Also prevents creation of any new wb's associated with @blkcg. | |
751 | */ | |
752 | void wb_blkcg_offline(struct blkcg *blkcg) | |
753 | { | |
754 | LIST_HEAD(to_destroy); | |
755 | struct bdi_writeback *wb, *next; | |
756 | ||
757 | spin_lock_irq(&cgwb_lock); | |
758 | list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node) | |
759 | cgwb_kill(wb); | |
760 | blkcg->cgwb_list.next = NULL; /* prevent new wb's */ | |
761 | spin_unlock_irq(&cgwb_lock); | |
762 | } | |
763 | ||
df23de55 JK |
764 | static void cgwb_bdi_exit(struct backing_dev_info *bdi) |
765 | { | |
766 | struct rb_node *rbn; | |
767 | ||
768 | spin_lock_irq(&cgwb_lock); | |
769 | while ((rbn = rb_first(&bdi->cgwb_congested_tree))) { | |
770 | struct bdi_writeback_congested *congested = | |
771 | rb_entry(rbn, struct bdi_writeback_congested, rb_node); | |
772 | ||
773 | rb_erase(rbn, &bdi->cgwb_congested_tree); | |
b7d680d7 | 774 | congested->__bdi = NULL; /* mark @congested unlinked */ |
df23de55 JK |
775 | } |
776 | spin_unlock_irq(&cgwb_lock); | |
777 | } | |
778 | ||
e8cb72b3 JK |
779 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
780 | { | |
781 | spin_lock_irq(&cgwb_lock); | |
782 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
783 | spin_unlock_irq(&cgwb_lock); | |
784 | } | |
785 | ||
52ebea74 TH |
786 | #else /* CONFIG_CGROUP_WRITEBACK */ |
787 | ||
a13f35e8 TH |
788 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
789 | { | |
790 | int err; | |
791 | ||
792 | bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL); | |
793 | if (!bdi->wb_congested) | |
794 | return -ENOMEM; | |
795 | ||
5f478e4e TH |
796 | atomic_set(&bdi->wb_congested->refcnt, 1); |
797 | ||
a13f35e8 TH |
798 | err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL); |
799 | if (err) { | |
5f478e4e | 800 | wb_congested_put(bdi->wb_congested); |
a13f35e8 TH |
801 | return err; |
802 | } | |
803 | return 0; | |
804 | } | |
805 | ||
b1c51afc | 806 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { } |
df23de55 JK |
807 | |
808 | static void cgwb_bdi_exit(struct backing_dev_info *bdi) | |
5f478e4e TH |
809 | { |
810 | wb_congested_put(bdi->wb_congested); | |
811 | } | |
52ebea74 | 812 | |
e8cb72b3 JK |
813 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
814 | { | |
815 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
816 | } | |
817 | ||
818 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) | |
819 | { | |
820 | list_del_rcu(&wb->bdi_node); | |
821 | } | |
822 | ||
52ebea74 | 823 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
e98be2d5 | 824 | |
2e82b84c | 825 | static int bdi_init(struct backing_dev_info *bdi) |
b2e8fb6e | 826 | { |
b817525a TH |
827 | int ret; |
828 | ||
cf0ca9fe PZ |
829 | bdi->dev = NULL; |
830 | ||
d03f6cdc | 831 | kref_init(&bdi->refcnt); |
189d3c4a | 832 | bdi->min_ratio = 0; |
a42dde04 | 833 | bdi->max_ratio = 100; |
eb608e3a | 834 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
66f3b8e2 | 835 | INIT_LIST_HEAD(&bdi->bdi_list); |
b817525a | 836 | INIT_LIST_HEAD(&bdi->wb_list); |
cc395d7f | 837 | init_waitqueue_head(&bdi->wb_waitq); |
03ba3782 | 838 | |
b817525a TH |
839 | ret = cgwb_bdi_init(bdi); |
840 | ||
b817525a | 841 | return ret; |
b2e8fb6e | 842 | } |
e98be2d5 | 843 | |
d03f6cdc JK |
844 | struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id) |
845 | { | |
846 | struct backing_dev_info *bdi; | |
847 | ||
848 | bdi = kmalloc_node(sizeof(struct backing_dev_info), | |
849 | gfp_mask | __GFP_ZERO, node_id); | |
850 | if (!bdi) | |
851 | return NULL; | |
852 | ||
853 | if (bdi_init(bdi)) { | |
854 | kfree(bdi); | |
855 | return NULL; | |
856 | } | |
857 | return bdi; | |
858 | } | |
62bf42ad | 859 | EXPORT_SYMBOL(bdi_alloc_node); |
d03f6cdc | 860 | |
7c4cc300 | 861 | int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args) |
46100071 | 862 | { |
46100071 | 863 | struct device *dev; |
e98be2d5 | 864 | |
46100071 TH |
865 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
866 | return 0; | |
e98be2d5 | 867 | |
7c4cc300 | 868 | dev = device_create_vargs(bdi_class, NULL, MKDEV(0, 0), bdi, fmt, args); |
46100071 TH |
869 | if (IS_ERR(dev)) |
870 | return PTR_ERR(dev); | |
04fbfdc1 | 871 | |
e8cb72b3 | 872 | cgwb_bdi_register(bdi); |
46100071 | 873 | bdi->dev = dev; |
b2e8fb6e | 874 | |
46100071 TH |
875 | bdi_debug_register(bdi, dev_name(dev)); |
876 | set_bit(WB_registered, &bdi->wb.state); | |
877 | ||
878 | spin_lock_bh(&bdi_lock); | |
879 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
880 | spin_unlock_bh(&bdi_lock); | |
881 | ||
882 | trace_writeback_bdi_register(bdi); | |
883 | return 0; | |
b2e8fb6e | 884 | } |
baf7a616 JK |
885 | EXPORT_SYMBOL(bdi_register_va); |
886 | ||
7c4cc300 | 887 | int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...) |
baf7a616 JK |
888 | { |
889 | va_list args; | |
890 | int ret; | |
891 | ||
892 | va_start(args, fmt); | |
7c4cc300 | 893 | ret = bdi_register_va(bdi, fmt, args); |
baf7a616 JK |
894 | va_end(args); |
895 | return ret; | |
896 | } | |
46100071 | 897 | EXPORT_SYMBOL(bdi_register); |
b2e8fb6e | 898 | |
df08c32c DW |
899 | int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner) |
900 | { | |
901 | int rc; | |
902 | ||
7c4cc300 | 903 | rc = bdi_register(bdi, "%u:%u", MAJOR(owner->devt), MINOR(owner->devt)); |
df08c32c DW |
904 | if (rc) |
905 | return rc; | |
b6f8fec4 JK |
906 | /* Leaking owner reference... */ |
907 | WARN_ON(bdi->owner); | |
df08c32c DW |
908 | bdi->owner = owner; |
909 | get_device(owner); | |
910 | return 0; | |
911 | } | |
912 | EXPORT_SYMBOL(bdi_register_owner); | |
913 | ||
46100071 TH |
914 | /* |
915 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
916 | */ | |
917 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
918 | { | |
919 | spin_lock_bh(&bdi_lock); | |
920 | list_del_rcu(&bdi->bdi_list); | |
921 | spin_unlock_bh(&bdi_lock); | |
b2e8fb6e | 922 | |
46100071 TH |
923 | synchronize_rcu_expedited(); |
924 | } | |
cf0ca9fe | 925 | |
b02176f3 | 926 | void bdi_unregister(struct backing_dev_info *bdi) |
b2e8fb6e | 927 | { |
f0054bb1 TH |
928 | /* make sure nobody finds us on the bdi_list anymore */ |
929 | bdi_remove_from_list(bdi); | |
930 | wb_shutdown(&bdi->wb); | |
b1c51afc | 931 | cgwb_bdi_unregister(bdi); |
7a401a97 | 932 | |
c4db59d3 CH |
933 | if (bdi->dev) { |
934 | bdi_debug_unregister(bdi); | |
935 | device_unregister(bdi->dev); | |
936 | bdi->dev = NULL; | |
937 | } | |
df08c32c DW |
938 | |
939 | if (bdi->owner) { | |
940 | put_device(bdi->owner); | |
941 | bdi->owner = NULL; | |
942 | } | |
b02176f3 | 943 | } |
c4db59d3 | 944 | |
d03f6cdc JK |
945 | static void release_bdi(struct kref *ref) |
946 | { | |
947 | struct backing_dev_info *bdi = | |
948 | container_of(ref, struct backing_dev_info, refcnt); | |
949 | ||
5af110b2 JK |
950 | if (test_bit(WB_registered, &bdi->wb.state)) |
951 | bdi_unregister(bdi); | |
2e82b84c JK |
952 | WARN_ON_ONCE(bdi->dev); |
953 | wb_exit(&bdi->wb); | |
954 | cgwb_bdi_exit(bdi); | |
d03f6cdc JK |
955 | kfree(bdi); |
956 | } | |
957 | ||
958 | void bdi_put(struct backing_dev_info *bdi) | |
959 | { | |
960 | kref_put(&bdi->refcnt, release_bdi); | |
961 | } | |
62bf42ad | 962 | EXPORT_SYMBOL(bdi_put); |
d03f6cdc | 963 | |
3fcfab16 AM |
964 | static wait_queue_head_t congestion_wqh[2] = { |
965 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
966 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
967 | }; | |
ec8a6f26 | 968 | static atomic_t nr_wb_congested[2]; |
3fcfab16 | 969 | |
ec8a6f26 | 970 | void clear_wb_congested(struct bdi_writeback_congested *congested, int sync) |
3fcfab16 | 971 | { |
1faa16d2 | 972 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
c877ef8a | 973 | enum wb_congested_state bit; |
3fcfab16 | 974 | |
4452226e | 975 | bit = sync ? WB_sync_congested : WB_async_congested; |
ec8a6f26 TH |
976 | if (test_and_clear_bit(bit, &congested->state)) |
977 | atomic_dec(&nr_wb_congested[sync]); | |
4e857c58 | 978 | smp_mb__after_atomic(); |
3fcfab16 AM |
979 | if (waitqueue_active(wqh)) |
980 | wake_up(wqh); | |
981 | } | |
ec8a6f26 | 982 | EXPORT_SYMBOL(clear_wb_congested); |
3fcfab16 | 983 | |
ec8a6f26 | 984 | void set_wb_congested(struct bdi_writeback_congested *congested, int sync) |
3fcfab16 | 985 | { |
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_set_bit(bit, &congested->state)) |
990 | atomic_inc(&nr_wb_congested[sync]); | |
3fcfab16 | 991 | } |
ec8a6f26 | 992 | EXPORT_SYMBOL(set_wb_congested); |
3fcfab16 AM |
993 | |
994 | /** | |
995 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 996 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
997 | * @timeout: timeout in jiffies |
998 | * | |
999 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
1000 | * write congestion. If no backing_devs are congested then just wait for the | |
1001 | * next write to be completed. | |
1002 | */ | |
8aa7e847 | 1003 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
1004 | { |
1005 | long ret; | |
52bb9198 | 1006 | unsigned long start = jiffies; |
3fcfab16 | 1007 | DEFINE_WAIT(wait); |
8aa7e847 | 1008 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
1009 | |
1010 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1011 | ret = io_schedule_timeout(timeout); | |
1012 | finish_wait(wqh, &wait); | |
52bb9198 MG |
1013 | |
1014 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
1015 | jiffies_to_usecs(jiffies - start)); | |
1016 | ||
3fcfab16 AM |
1017 | return ret; |
1018 | } | |
1019 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 1020 | |
0e093d99 | 1021 | /** |
599d0c95 MG |
1022 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes |
1023 | * @pgdat: A pgdat to check if it is heavily congested | |
0e093d99 MG |
1024 | * @sync: SYNC or ASYNC IO |
1025 | * @timeout: timeout in jiffies | |
1026 | * | |
1027 | * In the event of a congested backing_dev (any backing_dev) and the given | |
599d0c95 | 1028 | * @pgdat has experienced recent congestion, this waits for up to @timeout |
0e093d99 MG |
1029 | * jiffies for either a BDI to exit congestion of the given @sync queue |
1030 | * or a write to complete. | |
1031 | * | |
599d0c95 | 1032 | * In the absence of pgdat congestion, cond_resched() is called to yield |
ede37713 | 1033 | * the processor if necessary but otherwise does not sleep. |
0e093d99 MG |
1034 | * |
1035 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
1036 | * it is the number of jiffies that were still remaining when the function | |
1037 | * returned. return_value == timeout implies the function did not sleep. | |
1038 | */ | |
599d0c95 | 1039 | long wait_iff_congested(struct pglist_data *pgdat, int sync, long timeout) |
0e093d99 MG |
1040 | { |
1041 | long ret; | |
1042 | unsigned long start = jiffies; | |
1043 | DEFINE_WAIT(wait); | |
1044 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
1045 | ||
1046 | /* | |
1047 | * If there is no congestion, or heavy congestion is not being | |
599d0c95 | 1048 | * encountered in the current pgdat, yield if necessary instead |
0e093d99 MG |
1049 | * of sleeping on the congestion queue |
1050 | */ | |
ec8a6f26 | 1051 | if (atomic_read(&nr_wb_congested[sync]) == 0 || |
599d0c95 | 1052 | !test_bit(PGDAT_CONGESTED, &pgdat->flags)) { |
ede37713 | 1053 | cond_resched(); |
599d0c95 | 1054 | |
0e093d99 MG |
1055 | /* In case we scheduled, work out time remaining */ |
1056 | ret = timeout - (jiffies - start); | |
1057 | if (ret < 0) | |
1058 | ret = 0; | |
1059 | ||
1060 | goto out; | |
1061 | } | |
1062 | ||
1063 | /* Sleep until uncongested or a write happens */ | |
1064 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1065 | ret = io_schedule_timeout(timeout); | |
1066 | finish_wait(wqh, &wait); | |
1067 | ||
1068 | out: | |
1069 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
1070 | jiffies_to_usecs(jiffies - start)); | |
1071 | ||
1072 | return ret; | |
1073 | } | |
1074 | EXPORT_SYMBOL(wait_iff_congested); |