]>
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 | ||
5129a469 JE |
17 | struct backing_dev_info noop_backing_dev_info = { |
18 | .name = "noop", | |
976e48f8 | 19 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
5129a469 | 20 | }; |
a212b105 | 21 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); |
5129a469 | 22 | |
cf0ca9fe | 23 | static struct class *bdi_class; |
cfc4ba53 JA |
24 | |
25 | /* | |
181387da | 26 | * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side |
cfc4ba53 JA |
27 | * locking. |
28 | */ | |
03ba3782 | 29 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 30 | LIST_HEAD(bdi_list); |
03ba3782 | 31 | |
839a8e86 TH |
32 | /* bdi_wq serves all asynchronous writeback tasks */ |
33 | struct workqueue_struct *bdi_wq; | |
34 | ||
76f1418b MS |
35 | #ifdef CONFIG_DEBUG_FS |
36 | #include <linux/debugfs.h> | |
37 | #include <linux/seq_file.h> | |
38 | ||
39 | static struct dentry *bdi_debug_root; | |
40 | ||
41 | static void bdi_debug_init(void) | |
42 | { | |
43 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
44 | } | |
45 | ||
46 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
47 | { | |
48 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 49 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
50 | unsigned long background_thresh; |
51 | unsigned long dirty_thresh; | |
0d960a38 | 52 | unsigned long wb_thresh; |
0ae45f63 | 53 | unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time; |
f09b00d3 JA |
54 | struct inode *inode; |
55 | ||
0ae45f63 | 56 | nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0; |
f758eeab | 57 | spin_lock(&wb->list_lock); |
c7f54084 | 58 | list_for_each_entry(inode, &wb->b_dirty, i_io_list) |
c1955ce3 | 59 | nr_dirty++; |
c7f54084 | 60 | list_for_each_entry(inode, &wb->b_io, i_io_list) |
c1955ce3 | 61 | nr_io++; |
c7f54084 | 62 | list_for_each_entry(inode, &wb->b_more_io, i_io_list) |
c1955ce3 | 63 | nr_more_io++; |
c7f54084 | 64 | list_for_each_entry(inode, &wb->b_dirty_time, i_io_list) |
0ae45f63 TT |
65 | if (inode->i_state & I_DIRTY_TIME) |
66 | nr_dirty_time++; | |
f758eeab | 67 | spin_unlock(&wb->list_lock); |
76f1418b | 68 | |
16c4042f | 69 | global_dirty_limits(&background_thresh, &dirty_thresh); |
0d960a38 | 70 | wb_thresh = wb_calc_thresh(wb, dirty_thresh); |
76f1418b MS |
71 | |
72 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
73 | seq_printf(m, | |
00821b00 WF |
74 | "BdiWriteback: %10lu kB\n" |
75 | "BdiReclaimable: %10lu kB\n" | |
76 | "BdiDirtyThresh: %10lu kB\n" | |
77 | "DirtyThresh: %10lu kB\n" | |
78 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 79 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
80 | "BdiWritten: %10lu kB\n" |
81 | "BdiWriteBandwidth: %10lu kBps\n" | |
82 | "b_dirty: %10lu\n" | |
83 | "b_io: %10lu\n" | |
84 | "b_more_io: %10lu\n" | |
0ae45f63 | 85 | "b_dirty_time: %10lu\n" |
00821b00 WF |
86 | "bdi_list: %10u\n" |
87 | "state: %10lx\n", | |
93f78d88 TH |
88 | (unsigned long) K(wb_stat(wb, WB_WRITEBACK)), |
89 | (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)), | |
0d960a38 | 90 | K(wb_thresh), |
f7d2b1ec JK |
91 | K(dirty_thresh), |
92 | K(background_thresh), | |
93f78d88 TH |
93 | (unsigned long) K(wb_stat(wb, WB_DIRTIED)), |
94 | (unsigned long) K(wb_stat(wb, WB_WRITTEN)), | |
a88a341a | 95 | (unsigned long) K(wb->write_bandwidth), |
f7d2b1ec JK |
96 | nr_dirty, |
97 | nr_io, | |
98 | nr_more_io, | |
0ae45f63 | 99 | nr_dirty_time, |
4452226e | 100 | !list_empty(&bdi->bdi_list), bdi->wb.state); |
76f1418b MS |
101 | #undef K |
102 | ||
103 | return 0; | |
104 | } | |
105 | ||
106 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
107 | { | |
108 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
109 | } | |
110 | ||
111 | static const struct file_operations bdi_debug_stats_fops = { | |
112 | .open = bdi_debug_stats_open, | |
113 | .read = seq_read, | |
114 | .llseek = seq_lseek, | |
115 | .release = single_release, | |
116 | }; | |
117 | ||
118 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
119 | { | |
120 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
121 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
122 | bdi, &bdi_debug_stats_fops); | |
123 | } | |
124 | ||
125 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
126 | { | |
127 | debugfs_remove(bdi->debug_stats); | |
128 | debugfs_remove(bdi->debug_dir); | |
129 | } | |
130 | #else | |
131 | static inline void bdi_debug_init(void) | |
132 | { | |
133 | } | |
134 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
135 | const char *name) | |
136 | { | |
137 | } | |
138 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
139 | { | |
140 | } | |
141 | #endif | |
142 | ||
cf0ca9fe PZ |
143 | static ssize_t read_ahead_kb_store(struct device *dev, |
144 | struct device_attribute *attr, | |
145 | const char *buf, size_t count) | |
146 | { | |
147 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 148 | unsigned long read_ahead_kb; |
7034ed13 | 149 | ssize_t ret; |
cf0ca9fe | 150 | |
7034ed13 NJ |
151 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
152 | if (ret < 0) | |
153 | return ret; | |
154 | ||
155 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
156 | ||
157 | return count; | |
cf0ca9fe PZ |
158 | } |
159 | ||
160 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
161 | ||
162 | #define BDI_SHOW(name, expr) \ | |
163 | static ssize_t name##_show(struct device *dev, \ | |
164 | struct device_attribute *attr, char *page) \ | |
165 | { \ | |
166 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
167 | \ | |
168 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
d9e1241e GKH |
169 | } \ |
170 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
171 | |
172 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
173 | ||
189d3c4a PZ |
174 | static ssize_t min_ratio_store(struct device *dev, |
175 | struct device_attribute *attr, const char *buf, size_t count) | |
176 | { | |
177 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 178 | unsigned int ratio; |
7034ed13 NJ |
179 | ssize_t ret; |
180 | ||
181 | ret = kstrtouint(buf, 10, &ratio); | |
182 | if (ret < 0) | |
183 | return ret; | |
184 | ||
185 | ret = bdi_set_min_ratio(bdi, ratio); | |
186 | if (!ret) | |
187 | ret = count; | |
189d3c4a | 188 | |
189d3c4a PZ |
189 | return ret; |
190 | } | |
191 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
192 | ||
a42dde04 PZ |
193 | static ssize_t max_ratio_store(struct device *dev, |
194 | struct device_attribute *attr, const char *buf, size_t count) | |
195 | { | |
196 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 197 | unsigned int ratio; |
7034ed13 NJ |
198 | ssize_t ret; |
199 | ||
200 | ret = kstrtouint(buf, 10, &ratio); | |
201 | if (ret < 0) | |
202 | return ret; | |
203 | ||
204 | ret = bdi_set_max_ratio(bdi, ratio); | |
205 | if (!ret) | |
206 | ret = count; | |
a42dde04 | 207 | |
a42dde04 PZ |
208 | return ret; |
209 | } | |
210 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
211 | ||
7d311cda DW |
212 | static ssize_t stable_pages_required_show(struct device *dev, |
213 | struct device_attribute *attr, | |
214 | char *page) | |
215 | { | |
216 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
217 | ||
218 | return snprintf(page, PAGE_SIZE-1, "%d\n", | |
219 | bdi_cap_stable_pages_required(bdi) ? 1 : 0); | |
220 | } | |
d9e1241e GKH |
221 | static DEVICE_ATTR_RO(stable_pages_required); |
222 | ||
223 | static struct attribute *bdi_dev_attrs[] = { | |
224 | &dev_attr_read_ahead_kb.attr, | |
225 | &dev_attr_min_ratio.attr, | |
226 | &dev_attr_max_ratio.attr, | |
227 | &dev_attr_stable_pages_required.attr, | |
228 | NULL, | |
cf0ca9fe | 229 | }; |
d9e1241e | 230 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
231 | |
232 | static __init int bdi_class_init(void) | |
233 | { | |
234 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
235 | if (IS_ERR(bdi_class)) |
236 | return PTR_ERR(bdi_class); | |
237 | ||
d9e1241e | 238 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 239 | bdi_debug_init(); |
1a1eab55 | 240 | |
cf0ca9fe PZ |
241 | return 0; |
242 | } | |
76f1418b | 243 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 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 | |
54a0f136 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); | |
c284de61 | 316 | |
a13f35e8 | 317 | wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp); |
54a0f136 JK |
318 | if (!wb->congested) { |
319 | err = -ENOMEM; | |
320 | goto out_put_bdi; | |
321 | } | |
a13f35e8 | 322 | |
8395cd9f | 323 | err = fprop_local_init_percpu(&wb->completions, gfp); |
a88a341a | 324 | if (err) |
a13f35e8 | 325 | goto out_put_cong; |
c284de61 | 326 | |
93f78d88 | 327 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) { |
8395cd9f | 328 | err = percpu_counter_init(&wb->stat[i], 0, gfp); |
a13f35e8 TH |
329 | if (err) |
330 | goto out_destroy_stat; | |
93f78d88 | 331 | } |
cf0ca9fe | 332 | |
93f78d88 | 333 | return 0; |
a13f35e8 TH |
334 | |
335 | out_destroy_stat: | |
336 | while (--i) | |
337 | percpu_counter_destroy(&wb->stat[i]); | |
338 | fprop_local_destroy_percpu(&wb->completions); | |
339 | out_put_cong: | |
340 | wb_congested_put(wb->congested); | |
54a0f136 JK |
341 | out_put_bdi: |
342 | if (wb != &bdi->wb) | |
343 | bdi_put(bdi); | |
a13f35e8 | 344 | return err; |
cf0ca9fe | 345 | } |
cf0ca9fe | 346 | |
63a7818e JK |
347 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb); |
348 | ||
03ba3782 JA |
349 | /* |
350 | * Remove bdi from the global list and shutdown any threads we have running | |
351 | */ | |
46100071 | 352 | static void wb_shutdown(struct bdi_writeback *wb) |
66f3b8e2 | 353 | { |
c4db59d3 | 354 | /* Make sure nobody queues further work */ |
46100071 TH |
355 | spin_lock_bh(&wb->work_lock); |
356 | if (!test_and_clear_bit(WB_registered, &wb->state)) { | |
357 | spin_unlock_bh(&wb->work_lock); | |
b608c987 JK |
358 | /* |
359 | * Wait for wb shutdown to finish if someone else is just | |
360 | * running wb_shutdown(). Otherwise we could proceed to wb / | |
361 | * bdi destruction before wb_shutdown() is finished. | |
362 | */ | |
363 | wait_on_bit(&wb->state, WB_shutting_down, TASK_UNINTERRUPTIBLE); | |
03ba3782 | 364 | return; |
c4db59d3 | 365 | } |
b608c987 | 366 | set_bit(WB_shutting_down, &wb->state); |
46100071 | 367 | spin_unlock_bh(&wb->work_lock); |
03ba3782 | 368 | |
63a7818e | 369 | cgwb_remove_from_bdi_list(wb); |
03ba3782 | 370 | /* |
46100071 TH |
371 | * Drain work list and shutdown the delayed_work. !WB_registered |
372 | * tells wb_workfn() that @wb is dying and its work_list needs to | |
373 | * be drained no matter what. | |
03ba3782 | 374 | */ |
46100071 TH |
375 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
376 | flush_delayed_work(&wb->dwork); | |
377 | WARN_ON(!list_empty(&wb->work_list)); | |
b608c987 JK |
378 | /* |
379 | * Make sure bit gets cleared after shutdown is finished. Matches with | |
380 | * the barrier provided by test_and_clear_bit() above. | |
381 | */ | |
382 | smp_wmb(); | |
383 | clear_bit(WB_shutting_down, &wb->state); | |
46100071 TH |
384 | } |
385 | ||
f0054bb1 | 386 | static void wb_exit(struct bdi_writeback *wb) |
93f78d88 TH |
387 | { |
388 | int i; | |
389 | ||
390 | WARN_ON(delayed_work_pending(&wb->dwork)); | |
391 | ||
392 | for (i = 0; i < NR_WB_STAT_ITEMS; i++) | |
393 | percpu_counter_destroy(&wb->stat[i]); | |
6467716a | 394 | |
a88a341a | 395 | fprop_local_destroy_percpu(&wb->completions); |
a13f35e8 | 396 | wb_congested_put(wb->congested); |
54a0f136 JK |
397 | if (wb != &wb->bdi->wb) |
398 | bdi_put(wb->bdi); | |
a88a341a | 399 | } |
e98be2d5 | 400 | |
52ebea74 TH |
401 | #ifdef CONFIG_CGROUP_WRITEBACK |
402 | ||
403 | #include <linux/memcontrol.h> | |
404 | ||
405 | /* | |
406 | * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree, | |
407 | * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU | |
0292e232 | 408 | * protected. |
52ebea74 TH |
409 | */ |
410 | static DEFINE_SPINLOCK(cgwb_lock); | |
52ebea74 TH |
411 | |
412 | /** | |
413 | * wb_congested_get_create - get or create a wb_congested | |
414 | * @bdi: associated bdi | |
415 | * @blkcg_id: ID of the associated blkcg | |
416 | * @gfp: allocation mask | |
417 | * | |
418 | * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one. | |
419 | * The returned wb_congested has its reference count incremented. Returns | |
420 | * NULL on failure. | |
421 | */ | |
422 | struct bdi_writeback_congested * | |
423 | wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp) | |
424 | { | |
425 | struct bdi_writeback_congested *new_congested = NULL, *congested; | |
426 | struct rb_node **node, *parent; | |
427 | unsigned long flags; | |
52ebea74 TH |
428 | retry: |
429 | spin_lock_irqsave(&cgwb_lock, flags); | |
430 | ||
431 | node = &bdi->cgwb_congested_tree.rb_node; | |
432 | parent = NULL; | |
433 | ||
434 | while (*node != NULL) { | |
435 | parent = *node; | |
436 | congested = container_of(parent, struct bdi_writeback_congested, | |
437 | rb_node); | |
438 | if (congested->blkcg_id < blkcg_id) | |
439 | node = &parent->rb_left; | |
440 | else if (congested->blkcg_id > blkcg_id) | |
441 | node = &parent->rb_right; | |
442 | else | |
443 | goto found; | |
444 | } | |
445 | ||
446 | if (new_congested) { | |
447 | /* !found and storage for new one already allocated, insert */ | |
448 | congested = new_congested; | |
449 | new_congested = NULL; | |
450 | rb_link_node(&congested->rb_node, parent, node); | |
451 | rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree); | |
52ebea74 TH |
452 | goto found; |
453 | } | |
454 | ||
455 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
456 | ||
457 | /* allocate storage for new one and retry */ | |
458 | new_congested = kzalloc(sizeof(*new_congested), gfp); | |
459 | if (!new_congested) | |
460 | return NULL; | |
461 | ||
462 | atomic_set(&new_congested->refcnt, 0); | |
e96540d5 | 463 | new_congested->__bdi = bdi; |
52ebea74 TH |
464 | new_congested->blkcg_id = blkcg_id; |
465 | goto retry; | |
466 | ||
467 | found: | |
468 | atomic_inc(&congested->refcnt); | |
469 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
470 | kfree(new_congested); | |
471 | return congested; | |
472 | } | |
473 | ||
474 | /** | |
475 | * wb_congested_put - put a wb_congested | |
476 | * @congested: wb_congested to put | |
477 | * | |
478 | * Put @congested and destroy it if the refcnt reaches zero. | |
479 | */ | |
480 | void wb_congested_put(struct bdi_writeback_congested *congested) | |
481 | { | |
52ebea74 TH |
482 | unsigned long flags; |
483 | ||
52ebea74 TH |
484 | local_irq_save(flags); |
485 | if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) { | |
486 | local_irq_restore(flags); | |
487 | return; | |
488 | } | |
489 | ||
a20135ff | 490 | /* bdi might already have been destroyed leaving @congested unlinked */ |
e96540d5 | 491 | if (congested->__bdi) { |
a20135ff | 492 | rb_erase(&congested->rb_node, |
e96540d5 JK |
493 | &congested->__bdi->cgwb_congested_tree); |
494 | congested->__bdi = NULL; | |
a20135ff TH |
495 | } |
496 | ||
52ebea74 TH |
497 | spin_unlock_irqrestore(&cgwb_lock, flags); |
498 | kfree(congested); | |
52ebea74 TH |
499 | } |
500 | ||
501 | static void cgwb_release_workfn(struct work_struct *work) | |
502 | { | |
503 | struct bdi_writeback *wb = container_of(work, struct bdi_writeback, | |
504 | release_work); | |
52ebea74 TH |
505 | |
506 | wb_shutdown(wb); | |
507 | ||
508 | css_put(wb->memcg_css); | |
509 | css_put(wb->blkcg_css); | |
52ebea74 | 510 | |
841710aa | 511 | fprop_local_destroy_percpu(&wb->memcg_completions); |
52ebea74 TH |
512 | percpu_ref_exit(&wb->refcnt); |
513 | wb_exit(wb); | |
514 | kfree_rcu(wb, rcu); | |
52ebea74 TH |
515 | } |
516 | ||
517 | static void cgwb_release(struct percpu_ref *refcnt) | |
518 | { | |
519 | struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback, | |
520 | refcnt); | |
521 | schedule_work(&wb->release_work); | |
522 | } | |
523 | ||
524 | static void cgwb_kill(struct bdi_writeback *wb) | |
525 | { | |
526 | lockdep_assert_held(&cgwb_lock); | |
527 | ||
528 | WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id)); | |
529 | list_del(&wb->memcg_node); | |
530 | list_del(&wb->blkcg_node); | |
531 | percpu_ref_kill(&wb->refcnt); | |
532 | } | |
533 | ||
63a7818e JK |
534 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) |
535 | { | |
536 | spin_lock_irq(&cgwb_lock); | |
537 | list_del_rcu(&wb->bdi_node); | |
538 | spin_unlock_irq(&cgwb_lock); | |
539 | } | |
540 | ||
52ebea74 TH |
541 | static int cgwb_create(struct backing_dev_info *bdi, |
542 | struct cgroup_subsys_state *memcg_css, gfp_t gfp) | |
543 | { | |
544 | struct mem_cgroup *memcg; | |
545 | struct cgroup_subsys_state *blkcg_css; | |
546 | struct blkcg *blkcg; | |
547 | struct list_head *memcg_cgwb_list, *blkcg_cgwb_list; | |
548 | struct bdi_writeback *wb; | |
549 | unsigned long flags; | |
550 | int ret = 0; | |
551 | ||
552 | memcg = mem_cgroup_from_css(memcg_css); | |
c165b3e3 | 553 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys); |
52ebea74 TH |
554 | blkcg = css_to_blkcg(blkcg_css); |
555 | memcg_cgwb_list = mem_cgroup_cgwb_list(memcg); | |
556 | blkcg_cgwb_list = &blkcg->cgwb_list; | |
557 | ||
558 | /* look up again under lock and discard on blkcg mismatch */ | |
559 | spin_lock_irqsave(&cgwb_lock, flags); | |
560 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
561 | if (wb && wb->blkcg_css != blkcg_css) { | |
562 | cgwb_kill(wb); | |
563 | wb = NULL; | |
564 | } | |
565 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
566 | if (wb) | |
567 | goto out_put; | |
568 | ||
569 | /* need to create a new one */ | |
570 | wb = kmalloc(sizeof(*wb), gfp); | |
571 | if (!wb) | |
572 | return -ENOMEM; | |
573 | ||
a13f35e8 | 574 | ret = wb_init(wb, bdi, blkcg_css->id, gfp); |
52ebea74 TH |
575 | if (ret) |
576 | goto err_free; | |
577 | ||
578 | ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp); | |
579 | if (ret) | |
580 | goto err_wb_exit; | |
581 | ||
841710aa TH |
582 | ret = fprop_local_init_percpu(&wb->memcg_completions, gfp); |
583 | if (ret) | |
584 | goto err_ref_exit; | |
585 | ||
52ebea74 TH |
586 | wb->memcg_css = memcg_css; |
587 | wb->blkcg_css = blkcg_css; | |
588 | INIT_WORK(&wb->release_work, cgwb_release_workfn); | |
589 | set_bit(WB_registered, &wb->state); | |
03ba3782 JA |
590 | |
591 | /* | |
52ebea74 TH |
592 | * The root wb determines the registered state of the whole bdi and |
593 | * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate | |
594 | * whether they're still online. Don't link @wb if any is dead. | |
595 | * See wb_memcg_offline() and wb_blkcg_offline(). | |
03ba3782 | 596 | */ |
52ebea74 TH |
597 | ret = -ENODEV; |
598 | spin_lock_irqsave(&cgwb_lock, flags); | |
599 | if (test_bit(WB_registered, &bdi->wb.state) && | |
600 | blkcg_cgwb_list->next && memcg_cgwb_list->next) { | |
601 | /* we might have raced another instance of this function */ | |
602 | ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb); | |
603 | if (!ret) { | |
b817525a | 604 | list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list); |
52ebea74 TH |
605 | list_add(&wb->memcg_node, memcg_cgwb_list); |
606 | list_add(&wb->blkcg_node, blkcg_cgwb_list); | |
607 | css_get(memcg_css); | |
608 | css_get(blkcg_css); | |
609 | } | |
610 | } | |
611 | spin_unlock_irqrestore(&cgwb_lock, flags); | |
612 | if (ret) { | |
613 | if (ret == -EEXIST) | |
614 | ret = 0; | |
a13f35e8 | 615 | goto err_fprop_exit; |
52ebea74 TH |
616 | } |
617 | goto out_put; | |
618 | ||
841710aa TH |
619 | err_fprop_exit: |
620 | fprop_local_destroy_percpu(&wb->memcg_completions); | |
52ebea74 TH |
621 | err_ref_exit: |
622 | percpu_ref_exit(&wb->refcnt); | |
623 | err_wb_exit: | |
624 | wb_exit(wb); | |
625 | err_free: | |
626 | kfree(wb); | |
627 | out_put: | |
628 | css_put(blkcg_css); | |
629 | return ret; | |
66f3b8e2 JA |
630 | } |
631 | ||
52ebea74 TH |
632 | /** |
633 | * wb_get_create - get wb for a given memcg, create if necessary | |
634 | * @bdi: target bdi | |
635 | * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref) | |
636 | * @gfp: allocation mask to use | |
637 | * | |
638 | * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to | |
639 | * create one. The returned wb has its refcount incremented. | |
640 | * | |
641 | * This function uses css_get() on @memcg_css and thus expects its refcnt | |
642 | * to be positive on invocation. IOW, rcu_read_lock() protection on | |
643 | * @memcg_css isn't enough. try_get it before calling this function. | |
644 | * | |
645 | * A wb is keyed by its associated memcg. As blkcg implicitly enables | |
646 | * memcg on the default hierarchy, memcg association is guaranteed to be | |
647 | * more specific (equal or descendant to the associated blkcg) and thus can | |
648 | * identify both the memcg and blkcg associations. | |
649 | * | |
650 | * Because the blkcg associated with a memcg may change as blkcg is enabled | |
651 | * and disabled closer to root in the hierarchy, each wb keeps track of | |
652 | * both the memcg and blkcg associated with it and verifies the blkcg on | |
653 | * each lookup. On mismatch, the existing wb is discarded and a new one is | |
654 | * created. | |
655 | */ | |
656 | struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi, | |
657 | struct cgroup_subsys_state *memcg_css, | |
658 | gfp_t gfp) | |
6467716a | 659 | { |
52ebea74 TH |
660 | struct bdi_writeback *wb; |
661 | ||
d0164adc | 662 | might_sleep_if(gfpflags_allow_blocking(gfp)); |
52ebea74 TH |
663 | |
664 | if (!memcg_css->parent) | |
665 | return &bdi->wb; | |
666 | ||
667 | do { | |
668 | rcu_read_lock(); | |
669 | wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id); | |
670 | if (wb) { | |
671 | struct cgroup_subsys_state *blkcg_css; | |
672 | ||
673 | /* see whether the blkcg association has changed */ | |
674 | blkcg_css = cgroup_get_e_css(memcg_css->cgroup, | |
c165b3e3 | 675 | &io_cgrp_subsys); |
52ebea74 TH |
676 | if (unlikely(wb->blkcg_css != blkcg_css || |
677 | !wb_tryget(wb))) | |
678 | wb = NULL; | |
679 | css_put(blkcg_css); | |
680 | } | |
681 | rcu_read_unlock(); | |
682 | } while (!wb && !cgwb_create(bdi, memcg_css, gfp)); | |
683 | ||
684 | return wb; | |
685 | } | |
6467716a | 686 | |
a13f35e8 | 687 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
52ebea74 | 688 | { |
a13f35e8 TH |
689 | int ret; |
690 | ||
52ebea74 TH |
691 | INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC); |
692 | bdi->cgwb_congested_tree = RB_ROOT; | |
a13f35e8 TH |
693 | |
694 | ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL); | |
695 | if (!ret) { | |
696 | bdi->wb.memcg_css = mem_cgroup_root_css; | |
697 | bdi->wb.blkcg_css = blkcg_root_css; | |
698 | } | |
699 | return ret; | |
6467716a AB |
700 | } |
701 | ||
f2f30f86 | 702 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) |
52ebea74 TH |
703 | { |
704 | struct radix_tree_iter iter; | |
705 | void **slot; | |
b608c987 | 706 | struct bdi_writeback *wb; |
52ebea74 TH |
707 | |
708 | WARN_ON(test_bit(WB_registered, &bdi->wb.state)); | |
709 | ||
710 | spin_lock_irq(&cgwb_lock); | |
711 | radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0) | |
712 | cgwb_kill(*slot); | |
b608c987 JK |
713 | |
714 | while (!list_empty(&bdi->wb_list)) { | |
715 | wb = list_first_entry(&bdi->wb_list, struct bdi_writeback, | |
716 | bdi_node); | |
717 | spin_unlock_irq(&cgwb_lock); | |
718 | wb_shutdown(wb); | |
719 | spin_lock_irq(&cgwb_lock); | |
720 | } | |
52ebea74 | 721 | spin_unlock_irq(&cgwb_lock); |
52ebea74 TH |
722 | } |
723 | ||
724 | /** | |
725 | * wb_memcg_offline - kill all wb's associated with a memcg being offlined | |
726 | * @memcg: memcg being offlined | |
727 | * | |
728 | * Also prevents creation of any new wb's associated with @memcg. | |
e98be2d5 | 729 | */ |
52ebea74 TH |
730 | void wb_memcg_offline(struct mem_cgroup *memcg) |
731 | { | |
732 | LIST_HEAD(to_destroy); | |
733 | struct list_head *memcg_cgwb_list = mem_cgroup_cgwb_list(memcg); | |
734 | struct bdi_writeback *wb, *next; | |
735 | ||
736 | spin_lock_irq(&cgwb_lock); | |
737 | list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node) | |
738 | cgwb_kill(wb); | |
739 | memcg_cgwb_list->next = NULL; /* prevent new wb's */ | |
740 | spin_unlock_irq(&cgwb_lock); | |
741 | } | |
742 | ||
743 | /** | |
744 | * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined | |
745 | * @blkcg: blkcg being offlined | |
746 | * | |
747 | * Also prevents creation of any new wb's associated with @blkcg. | |
748 | */ | |
749 | void wb_blkcg_offline(struct blkcg *blkcg) | |
750 | { | |
751 | LIST_HEAD(to_destroy); | |
752 | struct bdi_writeback *wb, *next; | |
753 | ||
754 | spin_lock_irq(&cgwb_lock); | |
755 | list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node) | |
756 | cgwb_kill(wb); | |
757 | blkcg->cgwb_list.next = NULL; /* prevent new wb's */ | |
758 | spin_unlock_irq(&cgwb_lock); | |
759 | } | |
760 | ||
5236015c JK |
761 | static void cgwb_bdi_exit(struct backing_dev_info *bdi) |
762 | { | |
763 | struct rb_node *rbn; | |
764 | ||
765 | spin_lock_irq(&cgwb_lock); | |
766 | while ((rbn = rb_first(&bdi->cgwb_congested_tree))) { | |
767 | struct bdi_writeback_congested *congested = | |
768 | rb_entry(rbn, struct bdi_writeback_congested, rb_node); | |
769 | ||
770 | rb_erase(rbn, &bdi->cgwb_congested_tree); | |
e96540d5 | 771 | congested->__bdi = NULL; /* mark @congested unlinked */ |
5236015c JK |
772 | } |
773 | spin_unlock_irq(&cgwb_lock); | |
774 | } | |
775 | ||
63a7818e JK |
776 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
777 | { | |
778 | spin_lock_irq(&cgwb_lock); | |
779 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
780 | spin_unlock_irq(&cgwb_lock); | |
781 | } | |
782 | ||
52ebea74 TH |
783 | #else /* CONFIG_CGROUP_WRITEBACK */ |
784 | ||
a13f35e8 TH |
785 | static int cgwb_bdi_init(struct backing_dev_info *bdi) |
786 | { | |
787 | int err; | |
788 | ||
789 | bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL); | |
790 | if (!bdi->wb_congested) | |
791 | return -ENOMEM; | |
792 | ||
e809b23e TH |
793 | atomic_set(&bdi->wb_congested->refcnt, 1); |
794 | ||
a13f35e8 TH |
795 | err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL); |
796 | if (err) { | |
e809b23e | 797 | wb_congested_put(bdi->wb_congested); |
a13f35e8 TH |
798 | return err; |
799 | } | |
800 | return 0; | |
801 | } | |
802 | ||
f2f30f86 | 803 | static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { } |
5236015c JK |
804 | |
805 | static void cgwb_bdi_exit(struct backing_dev_info *bdi) | |
e809b23e TH |
806 | { |
807 | wb_congested_put(bdi->wb_congested); | |
808 | } | |
52ebea74 | 809 | |
63a7818e JK |
810 | static void cgwb_bdi_register(struct backing_dev_info *bdi) |
811 | { | |
812 | list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list); | |
813 | } | |
814 | ||
815 | static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb) | |
816 | { | |
817 | list_del_rcu(&wb->bdi_node); | |
818 | } | |
819 | ||
52ebea74 | 820 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
e98be2d5 | 821 | |
b2e8fb6e PZ |
822 | int bdi_init(struct backing_dev_info *bdi) |
823 | { | |
b817525a TH |
824 | int ret; |
825 | ||
cf0ca9fe PZ |
826 | bdi->dev = NULL; |
827 | ||
1a1eab55 | 828 | kref_init(&bdi->refcnt); |
189d3c4a | 829 | bdi->min_ratio = 0; |
a42dde04 | 830 | bdi->max_ratio = 100; |
eb608e3a | 831 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
66f3b8e2 | 832 | INIT_LIST_HEAD(&bdi->bdi_list); |
b817525a | 833 | INIT_LIST_HEAD(&bdi->wb_list); |
cc395d7f | 834 | init_waitqueue_head(&bdi->wb_waitq); |
03ba3782 | 835 | |
b817525a TH |
836 | ret = cgwb_bdi_init(bdi); |
837 | ||
b817525a | 838 | return ret; |
b2e8fb6e PZ |
839 | } |
840 | EXPORT_SYMBOL(bdi_init); | |
e98be2d5 | 841 | |
1a1eab55 JK |
842 | struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id) |
843 | { | |
844 | struct backing_dev_info *bdi; | |
845 | ||
846 | bdi = kmalloc_node(sizeof(struct backing_dev_info), | |
847 | gfp_mask | __GFP_ZERO, node_id); | |
848 | if (!bdi) | |
849 | return NULL; | |
850 | ||
851 | if (bdi_init(bdi)) { | |
852 | kfree(bdi); | |
853 | return NULL; | |
854 | } | |
855 | return bdi; | |
856 | } | |
857 | ||
46100071 TH |
858 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
859 | const char *fmt, ...) | |
860 | { | |
861 | va_list args; | |
862 | struct device *dev; | |
e98be2d5 | 863 | |
46100071 TH |
864 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
865 | return 0; | |
e98be2d5 | 866 | |
46100071 TH |
867 | va_start(args, fmt); |
868 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); | |
869 | va_end(args); | |
870 | if (IS_ERR(dev)) | |
871 | return PTR_ERR(dev); | |
04fbfdc1 | 872 | |
63a7818e | 873 | cgwb_bdi_register(bdi); |
46100071 | 874 | bdi->dev = dev; |
b2e8fb6e | 875 | |
46100071 TH |
876 | bdi_debug_register(bdi, dev_name(dev)); |
877 | set_bit(WB_registered, &bdi->wb.state); | |
878 | ||
879 | spin_lock_bh(&bdi_lock); | |
880 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
881 | spin_unlock_bh(&bdi_lock); | |
882 | ||
883 | trace_writeback_bdi_register(bdi); | |
884 | return 0; | |
b2e8fb6e | 885 | } |
46100071 | 886 | EXPORT_SYMBOL(bdi_register); |
b2e8fb6e | 887 | |
46100071 | 888 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) |
b2e8fb6e | 889 | { |
46100071 TH |
890 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); |
891 | } | |
892 | EXPORT_SYMBOL(bdi_register_dev); | |
893 | ||
8adf1a71 DW |
894 | int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner) |
895 | { | |
896 | int rc; | |
897 | ||
898 | rc = bdi_register(bdi, NULL, "%u:%u", MAJOR(owner->devt), | |
899 | MINOR(owner->devt)); | |
900 | if (rc) | |
901 | return rc; | |
143e2f82 JK |
902 | /* Leaking owner reference... */ |
903 | WARN_ON(bdi->owner); | |
8adf1a71 DW |
904 | bdi->owner = owner; |
905 | get_device(owner); | |
906 | return 0; | |
907 | } | |
908 | EXPORT_SYMBOL(bdi_register_owner); | |
909 | ||
46100071 TH |
910 | /* |
911 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
912 | */ | |
913 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
914 | { | |
915 | spin_lock_bh(&bdi_lock); | |
916 | list_del_rcu(&bdi->bdi_list); | |
917 | spin_unlock_bh(&bdi_lock); | |
b2e8fb6e | 918 | |
46100071 TH |
919 | synchronize_rcu_expedited(); |
920 | } | |
cf0ca9fe | 921 | |
b02176f3 | 922 | void bdi_unregister(struct backing_dev_info *bdi) |
b2e8fb6e | 923 | { |
f0054bb1 TH |
924 | /* make sure nobody finds us on the bdi_list anymore */ |
925 | bdi_remove_from_list(bdi); | |
926 | wb_shutdown(&bdi->wb); | |
f2f30f86 | 927 | cgwb_bdi_unregister(bdi); |
7a401a97 | 928 | |
c4db59d3 CH |
929 | if (bdi->dev) { |
930 | bdi_debug_unregister(bdi); | |
931 | device_unregister(bdi->dev); | |
932 | bdi->dev = NULL; | |
933 | } | |
8adf1a71 DW |
934 | |
935 | if (bdi->owner) { | |
936 | put_device(bdi->owner); | |
937 | bdi->owner = NULL; | |
938 | } | |
b02176f3 | 939 | } |
c4db59d3 | 940 | |
1a1eab55 | 941 | static void bdi_exit(struct backing_dev_info *bdi) |
b02176f3 TH |
942 | { |
943 | WARN_ON_ONCE(bdi->dev); | |
f0054bb1 | 944 | wb_exit(&bdi->wb); |
5236015c | 945 | cgwb_bdi_exit(bdi); |
b2e8fb6e | 946 | } |
b02176f3 | 947 | |
1a1eab55 JK |
948 | static void release_bdi(struct kref *ref) |
949 | { | |
950 | struct backing_dev_info *bdi = | |
951 | container_of(ref, struct backing_dev_info, refcnt); | |
952 | ||
953 | bdi_exit(bdi); | |
954 | kfree(bdi); | |
955 | } | |
956 | ||
957 | void bdi_put(struct backing_dev_info *bdi) | |
958 | { | |
959 | kref_put(&bdi->refcnt, release_bdi); | |
960 | } | |
961 | ||
b02176f3 TH |
962 | void bdi_destroy(struct backing_dev_info *bdi) |
963 | { | |
964 | bdi_unregister(bdi); | |
965 | bdi_exit(bdi); | |
966 | } | |
b2e8fb6e PZ |
967 | EXPORT_SYMBOL(bdi_destroy); |
968 | ||
c3c53206 JA |
969 | /* |
970 | * For use from filesystems to quickly init and register a bdi associated | |
971 | * with dirty writeback | |
972 | */ | |
b4caecd4 | 973 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name) |
c3c53206 | 974 | { |
c3c53206 JA |
975 | int err; |
976 | ||
977 | bdi->name = name; | |
b4caecd4 | 978 | bdi->capabilities = 0; |
c3c53206 JA |
979 | err = bdi_init(bdi); |
980 | if (err) | |
981 | return err; | |
982 | ||
02aa2a37 KC |
983 | err = bdi_register(bdi, NULL, "%.28s-%ld", name, |
984 | atomic_long_inc_return(&bdi_seq)); | |
c3c53206 JA |
985 | if (err) { |
986 | bdi_destroy(bdi); | |
987 | return err; | |
988 | } | |
989 | ||
990 | return 0; | |
991 | } | |
992 | EXPORT_SYMBOL(bdi_setup_and_register); | |
993 | ||
3fcfab16 AM |
994 | static wait_queue_head_t congestion_wqh[2] = { |
995 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
996 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
997 | }; | |
ec8a6f26 | 998 | static atomic_t nr_wb_congested[2]; |
3fcfab16 | 999 | |
ec8a6f26 | 1000 | void clear_wb_congested(struct bdi_writeback_congested *congested, int sync) |
3fcfab16 | 1001 | { |
1faa16d2 | 1002 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
ec8a6f26 | 1003 | enum wb_state bit; |
3fcfab16 | 1004 | |
4452226e | 1005 | bit = sync ? WB_sync_congested : WB_async_congested; |
ec8a6f26 TH |
1006 | if (test_and_clear_bit(bit, &congested->state)) |
1007 | atomic_dec(&nr_wb_congested[sync]); | |
4e857c58 | 1008 | smp_mb__after_atomic(); |
3fcfab16 AM |
1009 | if (waitqueue_active(wqh)) |
1010 | wake_up(wqh); | |
1011 | } | |
ec8a6f26 | 1012 | EXPORT_SYMBOL(clear_wb_congested); |
3fcfab16 | 1013 | |
ec8a6f26 | 1014 | void set_wb_congested(struct bdi_writeback_congested *congested, int sync) |
3fcfab16 | 1015 | { |
4452226e | 1016 | enum wb_state bit; |
3fcfab16 | 1017 | |
4452226e | 1018 | bit = sync ? WB_sync_congested : WB_async_congested; |
ec8a6f26 TH |
1019 | if (!test_and_set_bit(bit, &congested->state)) |
1020 | atomic_inc(&nr_wb_congested[sync]); | |
3fcfab16 | 1021 | } |
ec8a6f26 | 1022 | EXPORT_SYMBOL(set_wb_congested); |
3fcfab16 AM |
1023 | |
1024 | /** | |
1025 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 1026 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
1027 | * @timeout: timeout in jiffies |
1028 | * | |
1029 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
1030 | * write congestion. If no backing_devs are congested then just wait for the | |
1031 | * next write to be completed. | |
1032 | */ | |
8aa7e847 | 1033 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
1034 | { |
1035 | long ret; | |
52bb9198 | 1036 | unsigned long start = jiffies; |
3fcfab16 | 1037 | DEFINE_WAIT(wait); |
8aa7e847 | 1038 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
1039 | |
1040 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1041 | ret = io_schedule_timeout(timeout); | |
1042 | finish_wait(wqh, &wait); | |
52bb9198 MG |
1043 | |
1044 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
1045 | jiffies_to_usecs(jiffies - start)); | |
1046 | ||
3fcfab16 AM |
1047 | return ret; |
1048 | } | |
1049 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 1050 | |
0e093d99 MG |
1051 | /** |
1052 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
1053 | * @zone: A zone to check if it is heavily congested | |
1054 | * @sync: SYNC or ASYNC IO | |
1055 | * @timeout: timeout in jiffies | |
1056 | * | |
1057 | * In the event of a congested backing_dev (any backing_dev) and the given | |
1058 | * @zone has experienced recent congestion, this waits for up to @timeout | |
1059 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
1060 | * or a write to complete. | |
1061 | * | |
57e9ef47 MH |
1062 | * In the absence of zone congestion, cond_resched() is called to yield |
1063 | * the processor if necessary but otherwise does not sleep. | |
0e093d99 MG |
1064 | * |
1065 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
1066 | * it is the number of jiffies that were still remaining when the function | |
1067 | * returned. return_value == timeout implies the function did not sleep. | |
1068 | */ | |
1069 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
1070 | { | |
1071 | long ret; | |
1072 | unsigned long start = jiffies; | |
1073 | DEFINE_WAIT(wait); | |
1074 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
1075 | ||
1076 | /* | |
1077 | * If there is no congestion, or heavy congestion is not being | |
1078 | * encountered in the current zone, yield if necessary instead | |
1079 | * of sleeping on the congestion queue | |
1080 | */ | |
ec8a6f26 | 1081 | if (atomic_read(&nr_wb_congested[sync]) == 0 || |
57054651 | 1082 | !test_bit(ZONE_CONGESTED, &zone->flags)) { |
57e9ef47 | 1083 | cond_resched(); |
0e093d99 MG |
1084 | /* In case we scheduled, work out time remaining */ |
1085 | ret = timeout - (jiffies - start); | |
1086 | if (ret < 0) | |
1087 | ret = 0; | |
1088 | ||
1089 | goto out; | |
1090 | } | |
1091 | ||
1092 | /* Sleep until uncongested or a write happens */ | |
1093 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
1094 | ret = io_schedule_timeout(timeout); | |
1095 | finish_wait(wqh, &wait); | |
1096 | ||
1097 | out: | |
1098 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
1099 | jiffies_to_usecs(jiffies - start)); | |
1100 | ||
1101 | return ret; | |
1102 | } | |
1103 | EXPORT_SYMBOL(wait_iff_congested); | |
3965c9ae WL |
1104 | |
1105 | int pdflush_proc_obsolete(struct ctl_table *table, int write, | |
1106 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1107 | { | |
1108 | char kbuf[] = "0\n"; | |
1109 | ||
4c3bffc2 | 1110 | if (*ppos || *lenp < sizeof(kbuf)) { |
3965c9ae WL |
1111 | *lenp = 0; |
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | if (copy_to_user(buffer, kbuf, sizeof(kbuf))) | |
1116 | return -EFAULT; | |
1117 | printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n", | |
1118 | table->procname); | |
1119 | ||
1120 | *lenp = 2; | |
1121 | *ppos += *lenp; | |
1122 | return 2; | |
1123 | } |