Commit | Line | Data |
---|---|---|
e43473b7 VG |
1 | /* |
2 | * Interface for controlling IO bandwidth on a request queue | |
3 | * | |
4 | * Copyright (C) 2010 Vivek Goyal <vgoyal@redhat.com> | |
5 | */ | |
6 | ||
7 | #include <linux/module.h> | |
8 | #include <linux/slab.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/bio.h> | |
11 | #include <linux/blktrace_api.h> | |
12 | #include "blk-cgroup.h" | |
13 | ||
14 | /* Max dispatch from a group in 1 round */ | |
15 | static int throtl_grp_quantum = 8; | |
16 | ||
17 | /* Total max dispatch from all groups in one round */ | |
18 | static int throtl_quantum = 32; | |
19 | ||
20 | /* Throttling is performed over 100ms slice and after that slice is renewed */ | |
21 | static unsigned long throtl_slice = HZ/10; /* 100 ms */ | |
22 | ||
23 | struct throtl_rb_root { | |
24 | struct rb_root rb; | |
25 | struct rb_node *left; | |
26 | unsigned int count; | |
27 | unsigned long min_disptime; | |
28 | }; | |
29 | ||
30 | #define THROTL_RB_ROOT (struct throtl_rb_root) { .rb = RB_ROOT, .left = NULL, \ | |
31 | .count = 0, .min_disptime = 0} | |
32 | ||
33 | #define rb_entry_tg(node) rb_entry((node), struct throtl_grp, rb_node) | |
34 | ||
35 | struct throtl_grp { | |
36 | /* List of throtl groups on the request queue*/ | |
37 | struct hlist_node tg_node; | |
38 | ||
39 | /* active throtl group service_tree member */ | |
40 | struct rb_node rb_node; | |
41 | ||
42 | /* | |
43 | * Dispatch time in jiffies. This is the estimated time when group | |
44 | * will unthrottle and is ready to dispatch more bio. It is used as | |
45 | * key to sort active groups in service tree. | |
46 | */ | |
47 | unsigned long disptime; | |
48 | ||
49 | struct blkio_group blkg; | |
50 | atomic_t ref; | |
51 | unsigned int flags; | |
52 | ||
53 | /* Two lists for READ and WRITE */ | |
54 | struct bio_list bio_lists[2]; | |
55 | ||
56 | /* Number of queued bios on READ and WRITE lists */ | |
57 | unsigned int nr_queued[2]; | |
58 | ||
59 | /* bytes per second rate limits */ | |
60 | uint64_t bps[2]; | |
61 | ||
62 | /* Number of bytes disptached in current slice */ | |
63 | uint64_t bytes_disp[2]; | |
64 | ||
65 | /* When did we start a new slice */ | |
66 | unsigned long slice_start[2]; | |
67 | unsigned long slice_end[2]; | |
68 | }; | |
69 | ||
70 | struct throtl_data | |
71 | { | |
72 | /* List of throtl groups */ | |
73 | struct hlist_head tg_list; | |
74 | ||
75 | /* service tree for active throtl groups */ | |
76 | struct throtl_rb_root tg_service_tree; | |
77 | ||
78 | struct throtl_grp root_tg; | |
79 | struct request_queue *queue; | |
80 | ||
81 | /* Total Number of queued bios on READ and WRITE lists */ | |
82 | unsigned int nr_queued[2]; | |
83 | ||
84 | /* | |
85 | * number of total undestroyed groups (excluding root group) | |
86 | */ | |
87 | unsigned int nr_undestroyed_grps; | |
88 | ||
89 | /* Work for dispatching throttled bios */ | |
90 | struct delayed_work throtl_work; | |
91 | }; | |
92 | ||
93 | enum tg_state_flags { | |
94 | THROTL_TG_FLAG_on_rr = 0, /* on round-robin busy list */ | |
95 | }; | |
96 | ||
97 | #define THROTL_TG_FNS(name) \ | |
98 | static inline void throtl_mark_tg_##name(struct throtl_grp *tg) \ | |
99 | { \ | |
100 | (tg)->flags |= (1 << THROTL_TG_FLAG_##name); \ | |
101 | } \ | |
102 | static inline void throtl_clear_tg_##name(struct throtl_grp *tg) \ | |
103 | { \ | |
104 | (tg)->flags &= ~(1 << THROTL_TG_FLAG_##name); \ | |
105 | } \ | |
106 | static inline int throtl_tg_##name(const struct throtl_grp *tg) \ | |
107 | { \ | |
108 | return ((tg)->flags & (1 << THROTL_TG_FLAG_##name)) != 0; \ | |
109 | } | |
110 | ||
111 | THROTL_TG_FNS(on_rr); | |
112 | ||
113 | #define throtl_log_tg(td, tg, fmt, args...) \ | |
114 | blk_add_trace_msg((td)->queue, "throtl %s " fmt, \ | |
115 | blkg_path(&(tg)->blkg), ##args); \ | |
116 | ||
117 | #define throtl_log(td, fmt, args...) \ | |
118 | blk_add_trace_msg((td)->queue, "throtl " fmt, ##args) | |
119 | ||
120 | static inline struct throtl_grp *tg_of_blkg(struct blkio_group *blkg) | |
121 | { | |
122 | if (blkg) | |
123 | return container_of(blkg, struct throtl_grp, blkg); | |
124 | ||
125 | return NULL; | |
126 | } | |
127 | ||
128 | static inline int total_nr_queued(struct throtl_data *td) | |
129 | { | |
130 | return (td->nr_queued[0] + td->nr_queued[1]); | |
131 | } | |
132 | ||
133 | static inline struct throtl_grp *throtl_ref_get_tg(struct throtl_grp *tg) | |
134 | { | |
135 | atomic_inc(&tg->ref); | |
136 | return tg; | |
137 | } | |
138 | ||
139 | static void throtl_put_tg(struct throtl_grp *tg) | |
140 | { | |
141 | BUG_ON(atomic_read(&tg->ref) <= 0); | |
142 | if (!atomic_dec_and_test(&tg->ref)) | |
143 | return; | |
144 | kfree(tg); | |
145 | } | |
146 | ||
147 | static struct throtl_grp * throtl_find_alloc_tg(struct throtl_data *td, | |
148 | struct cgroup *cgroup) | |
149 | { | |
150 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgroup); | |
151 | struct throtl_grp *tg = NULL; | |
152 | void *key = td; | |
153 | struct backing_dev_info *bdi = &td->queue->backing_dev_info; | |
154 | unsigned int major, minor; | |
155 | ||
156 | /* | |
157 | * TODO: Speed up blkiocg_lookup_group() by maintaining a radix | |
158 | * tree of blkg (instead of traversing through hash list all | |
159 | * the time. | |
160 | */ | |
161 | tg = tg_of_blkg(blkiocg_lookup_group(blkcg, key)); | |
162 | ||
163 | /* Fill in device details for root group */ | |
164 | if (tg && !tg->blkg.dev && bdi->dev && dev_name(bdi->dev)) { | |
165 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); | |
166 | tg->blkg.dev = MKDEV(major, minor); | |
167 | goto done; | |
168 | } | |
169 | ||
170 | if (tg) | |
171 | goto done; | |
172 | ||
173 | tg = kzalloc_node(sizeof(*tg), GFP_ATOMIC, td->queue->node); | |
174 | if (!tg) | |
175 | goto done; | |
176 | ||
177 | INIT_HLIST_NODE(&tg->tg_node); | |
178 | RB_CLEAR_NODE(&tg->rb_node); | |
179 | bio_list_init(&tg->bio_lists[0]); | |
180 | bio_list_init(&tg->bio_lists[1]); | |
181 | ||
182 | /* | |
183 | * Take the initial reference that will be released on destroy | |
184 | * This can be thought of a joint reference by cgroup and | |
185 | * request queue which will be dropped by either request queue | |
186 | * exit or cgroup deletion path depending on who is exiting first. | |
187 | */ | |
188 | atomic_set(&tg->ref, 1); | |
189 | ||
190 | /* Add group onto cgroup list */ | |
191 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); | |
192 | blkiocg_add_blkio_group(blkcg, &tg->blkg, (void *)td, | |
193 | MKDEV(major, minor), BLKIO_POLICY_THROTL); | |
194 | ||
195 | tg->bps[READ] = blkcg_get_read_bps(blkcg, tg->blkg.dev); | |
196 | tg->bps[WRITE] = blkcg_get_write_bps(blkcg, tg->blkg.dev); | |
197 | ||
198 | hlist_add_head(&tg->tg_node, &td->tg_list); | |
199 | td->nr_undestroyed_grps++; | |
200 | done: | |
201 | return tg; | |
202 | } | |
203 | ||
204 | static struct throtl_grp * throtl_get_tg(struct throtl_data *td) | |
205 | { | |
206 | struct cgroup *cgroup; | |
207 | struct throtl_grp *tg = NULL; | |
208 | ||
209 | rcu_read_lock(); | |
210 | cgroup = task_cgroup(current, blkio_subsys_id); | |
211 | tg = throtl_find_alloc_tg(td, cgroup); | |
212 | if (!tg) | |
213 | tg = &td->root_tg; | |
214 | rcu_read_unlock(); | |
215 | return tg; | |
216 | } | |
217 | ||
218 | static struct throtl_grp *throtl_rb_first(struct throtl_rb_root *root) | |
219 | { | |
220 | /* Service tree is empty */ | |
221 | if (!root->count) | |
222 | return NULL; | |
223 | ||
224 | if (!root->left) | |
225 | root->left = rb_first(&root->rb); | |
226 | ||
227 | if (root->left) | |
228 | return rb_entry_tg(root->left); | |
229 | ||
230 | return NULL; | |
231 | } | |
232 | ||
233 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) | |
234 | { | |
235 | rb_erase(n, root); | |
236 | RB_CLEAR_NODE(n); | |
237 | } | |
238 | ||
239 | static void throtl_rb_erase(struct rb_node *n, struct throtl_rb_root *root) | |
240 | { | |
241 | if (root->left == n) | |
242 | root->left = NULL; | |
243 | rb_erase_init(n, &root->rb); | |
244 | --root->count; | |
245 | } | |
246 | ||
247 | static void update_min_dispatch_time(struct throtl_rb_root *st) | |
248 | { | |
249 | struct throtl_grp *tg; | |
250 | ||
251 | tg = throtl_rb_first(st); | |
252 | if (!tg) | |
253 | return; | |
254 | ||
255 | st->min_disptime = tg->disptime; | |
256 | } | |
257 | ||
258 | static void | |
259 | tg_service_tree_add(struct throtl_rb_root *st, struct throtl_grp *tg) | |
260 | { | |
261 | struct rb_node **node = &st->rb.rb_node; | |
262 | struct rb_node *parent = NULL; | |
263 | struct throtl_grp *__tg; | |
264 | unsigned long key = tg->disptime; | |
265 | int left = 1; | |
266 | ||
267 | while (*node != NULL) { | |
268 | parent = *node; | |
269 | __tg = rb_entry_tg(parent); | |
270 | ||
271 | if (time_before(key, __tg->disptime)) | |
272 | node = &parent->rb_left; | |
273 | else { | |
274 | node = &parent->rb_right; | |
275 | left = 0; | |
276 | } | |
277 | } | |
278 | ||
279 | if (left) | |
280 | st->left = &tg->rb_node; | |
281 | ||
282 | rb_link_node(&tg->rb_node, parent, node); | |
283 | rb_insert_color(&tg->rb_node, &st->rb); | |
284 | } | |
285 | ||
286 | static void __throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg) | |
287 | { | |
288 | struct throtl_rb_root *st = &td->tg_service_tree; | |
289 | ||
290 | tg_service_tree_add(st, tg); | |
291 | throtl_mark_tg_on_rr(tg); | |
292 | st->count++; | |
293 | } | |
294 | ||
295 | static void throtl_enqueue_tg(struct throtl_data *td, struct throtl_grp *tg) | |
296 | { | |
297 | if (!throtl_tg_on_rr(tg)) | |
298 | __throtl_enqueue_tg(td, tg); | |
299 | } | |
300 | ||
301 | static void __throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg) | |
302 | { | |
303 | throtl_rb_erase(&tg->rb_node, &td->tg_service_tree); | |
304 | throtl_clear_tg_on_rr(tg); | |
305 | } | |
306 | ||
307 | static void throtl_dequeue_tg(struct throtl_data *td, struct throtl_grp *tg) | |
308 | { | |
309 | if (throtl_tg_on_rr(tg)) | |
310 | __throtl_dequeue_tg(td, tg); | |
311 | } | |
312 | ||
313 | static void throtl_schedule_next_dispatch(struct throtl_data *td) | |
314 | { | |
315 | struct throtl_rb_root *st = &td->tg_service_tree; | |
316 | ||
317 | /* | |
318 | * If there are more bios pending, schedule more work. | |
319 | */ | |
320 | if (!total_nr_queued(td)) | |
321 | return; | |
322 | ||
323 | BUG_ON(!st->count); | |
324 | ||
325 | update_min_dispatch_time(st); | |
326 | ||
327 | if (time_before_eq(st->min_disptime, jiffies)) | |
328 | throtl_schedule_delayed_work(td->queue, 0); | |
329 | else | |
330 | throtl_schedule_delayed_work(td->queue, | |
331 | (st->min_disptime - jiffies)); | |
332 | } | |
333 | ||
334 | static inline void | |
335 | throtl_start_new_slice(struct throtl_data *td, struct throtl_grp *tg, bool rw) | |
336 | { | |
337 | tg->bytes_disp[rw] = 0; | |
338 | tg->slice_start[rw] = jiffies; | |
339 | tg->slice_end[rw] = jiffies + throtl_slice; | |
340 | throtl_log_tg(td, tg, "[%c] new slice start=%lu end=%lu jiffies=%lu", | |
341 | rw == READ ? 'R' : 'W', tg->slice_start[rw], | |
342 | tg->slice_end[rw], jiffies); | |
343 | } | |
344 | ||
345 | static inline void throtl_extend_slice(struct throtl_data *td, | |
346 | struct throtl_grp *tg, bool rw, unsigned long jiffy_end) | |
347 | { | |
348 | tg->slice_end[rw] = roundup(jiffy_end, throtl_slice); | |
349 | throtl_log_tg(td, tg, "[%c] extend slice start=%lu end=%lu jiffies=%lu", | |
350 | rw == READ ? 'R' : 'W', tg->slice_start[rw], | |
351 | tg->slice_end[rw], jiffies); | |
352 | } | |
353 | ||
354 | /* Determine if previously allocated or extended slice is complete or not */ | |
355 | static bool | |
356 | throtl_slice_used(struct throtl_data *td, struct throtl_grp *tg, bool rw) | |
357 | { | |
358 | if (time_in_range(jiffies, tg->slice_start[rw], tg->slice_end[rw])) | |
359 | return 0; | |
360 | ||
361 | return 1; | |
362 | } | |
363 | ||
364 | /* Trim the used slices and adjust slice start accordingly */ | |
365 | static inline void | |
366 | throtl_trim_slice(struct throtl_data *td, struct throtl_grp *tg, bool rw) | |
367 | { | |
368 | unsigned long nr_slices, bytes_trim, time_elapsed; | |
369 | ||
370 | BUG_ON(time_before(tg->slice_end[rw], tg->slice_start[rw])); | |
371 | ||
372 | /* | |
373 | * If bps are unlimited (-1), then time slice don't get | |
374 | * renewed. Don't try to trim the slice if slice is used. A new | |
375 | * slice will start when appropriate. | |
376 | */ | |
377 | if (throtl_slice_used(td, tg, rw)) | |
378 | return; | |
379 | ||
380 | time_elapsed = jiffies - tg->slice_start[rw]; | |
381 | ||
382 | nr_slices = time_elapsed / throtl_slice; | |
383 | ||
384 | if (!nr_slices) | |
385 | return; | |
386 | ||
387 | bytes_trim = (tg->bps[rw] * throtl_slice * nr_slices)/HZ; | |
388 | ||
389 | if (!bytes_trim) | |
390 | return; | |
391 | ||
392 | if (tg->bytes_disp[rw] >= bytes_trim) | |
393 | tg->bytes_disp[rw] -= bytes_trim; | |
394 | else | |
395 | tg->bytes_disp[rw] = 0; | |
396 | ||
397 | tg->slice_start[rw] += nr_slices * throtl_slice; | |
398 | ||
399 | throtl_log_tg(td, tg, "[%c] trim slice nr=%lu bytes=%lu" | |
400 | " start=%lu end=%lu jiffies=%lu", | |
401 | rw == READ ? 'R' : 'W', nr_slices, bytes_trim, | |
402 | tg->slice_start[rw], tg->slice_end[rw], jiffies); | |
403 | } | |
404 | ||
405 | /* | |
406 | * Returns whether one can dispatch a bio or not. Also returns approx number | |
407 | * of jiffies to wait before this bio is with-in IO rate and can be dispatched | |
408 | */ | |
409 | static bool tg_may_dispatch(struct throtl_data *td, struct throtl_grp *tg, | |
410 | struct bio *bio, unsigned long *wait) | |
411 | { | |
412 | bool rw = bio_data_dir(bio); | |
413 | u64 bytes_allowed, extra_bytes; | |
414 | unsigned long jiffy_elapsed, jiffy_wait, jiffy_elapsed_rnd; | |
415 | ||
416 | /* | |
417 | * Currently whole state machine of group depends on first bio | |
418 | * queued in the group bio list. So one should not be calling | |
419 | * this function with a different bio if there are other bios | |
420 | * queued. | |
421 | */ | |
422 | BUG_ON(tg->nr_queued[rw] && bio != bio_list_peek(&tg->bio_lists[rw])); | |
423 | ||
424 | /* If tg->bps = -1, then BW is unlimited */ | |
425 | if (tg->bps[rw] == -1) { | |
426 | if (wait) | |
427 | *wait = 0; | |
428 | return 1; | |
429 | } | |
430 | ||
431 | /* | |
432 | * If previous slice expired, start a new one otherwise renew/extend | |
433 | * existing slice to make sure it is at least throtl_slice interval | |
434 | * long since now. | |
435 | */ | |
436 | if (throtl_slice_used(td, tg, rw)) | |
437 | throtl_start_new_slice(td, tg, rw); | |
438 | else { | |
439 | if (time_before(tg->slice_end[rw], jiffies + throtl_slice)) | |
440 | throtl_extend_slice(td, tg, rw, jiffies + throtl_slice); | |
441 | } | |
442 | ||
443 | jiffy_elapsed = jiffy_elapsed_rnd = jiffies - tg->slice_start[rw]; | |
444 | ||
445 | /* Slice has just started. Consider one slice interval */ | |
446 | if (!jiffy_elapsed) | |
447 | jiffy_elapsed_rnd = throtl_slice; | |
448 | ||
449 | jiffy_elapsed_rnd = roundup(jiffy_elapsed_rnd, throtl_slice); | |
450 | ||
451 | bytes_allowed = (tg->bps[rw] * jiffies_to_msecs(jiffy_elapsed_rnd)) | |
452 | / MSEC_PER_SEC; | |
453 | ||
454 | if (tg->bytes_disp[rw] + bio->bi_size <= bytes_allowed) { | |
455 | if (wait) | |
456 | *wait = 0; | |
457 | return 1; | |
458 | } | |
459 | ||
460 | /* Calc approx time to dispatch */ | |
461 | extra_bytes = tg->bytes_disp[rw] + bio->bi_size - bytes_allowed; | |
462 | jiffy_wait = div64_u64(extra_bytes * HZ, tg->bps[rw]); | |
463 | ||
464 | if (!jiffy_wait) | |
465 | jiffy_wait = 1; | |
466 | ||
467 | /* | |
468 | * This wait time is without taking into consideration the rounding | |
469 | * up we did. Add that time also. | |
470 | */ | |
471 | jiffy_wait = jiffy_wait + (jiffy_elapsed_rnd - jiffy_elapsed); | |
472 | ||
473 | if (wait) | |
474 | *wait = jiffy_wait; | |
475 | ||
476 | if (time_before(tg->slice_end[rw], jiffies + jiffy_wait)) | |
477 | throtl_extend_slice(td, tg, rw, jiffies + jiffy_wait); | |
478 | ||
479 | return 0; | |
480 | } | |
481 | ||
482 | static void throtl_charge_bio(struct throtl_grp *tg, struct bio *bio) | |
483 | { | |
484 | bool rw = bio_data_dir(bio); | |
485 | bool sync = bio->bi_rw & REQ_SYNC; | |
486 | ||
487 | /* Charge the bio to the group */ | |
488 | tg->bytes_disp[rw] += bio->bi_size; | |
489 | ||
490 | /* | |
491 | * TODO: This will take blkg->stats_lock. Figure out a way | |
492 | * to avoid this cost. | |
493 | */ | |
494 | blkiocg_update_dispatch_stats(&tg->blkg, bio->bi_size, rw, sync); | |
495 | ||
496 | } | |
497 | ||
498 | static void throtl_add_bio_tg(struct throtl_data *td, struct throtl_grp *tg, | |
499 | struct bio *bio) | |
500 | { | |
501 | bool rw = bio_data_dir(bio); | |
502 | ||
503 | bio_list_add(&tg->bio_lists[rw], bio); | |
504 | /* Take a bio reference on tg */ | |
505 | throtl_ref_get_tg(tg); | |
506 | tg->nr_queued[rw]++; | |
507 | td->nr_queued[rw]++; | |
508 | throtl_enqueue_tg(td, tg); | |
509 | } | |
510 | ||
511 | static void tg_update_disptime(struct throtl_data *td, struct throtl_grp *tg) | |
512 | { | |
513 | unsigned long read_wait = -1, write_wait = -1, min_wait = -1, disptime; | |
514 | struct bio *bio; | |
515 | ||
516 | if ((bio = bio_list_peek(&tg->bio_lists[READ]))) | |
517 | tg_may_dispatch(td, tg, bio, &read_wait); | |
518 | ||
519 | if ((bio = bio_list_peek(&tg->bio_lists[WRITE]))) | |
520 | tg_may_dispatch(td, tg, bio, &write_wait); | |
521 | ||
522 | min_wait = min(read_wait, write_wait); | |
523 | disptime = jiffies + min_wait; | |
524 | ||
525 | /* | |
526 | * If group is already on active tree, then update dispatch time | |
527 | * only if it is lesser than existing dispatch time. Otherwise | |
528 | * always update the dispatch time | |
529 | */ | |
530 | ||
531 | if (throtl_tg_on_rr(tg) && time_before(disptime, tg->disptime)) | |
532 | return; | |
533 | ||
534 | /* Update dispatch time */ | |
535 | throtl_dequeue_tg(td, tg); | |
536 | tg->disptime = disptime; | |
537 | throtl_enqueue_tg(td, tg); | |
538 | } | |
539 | ||
540 | static void tg_dispatch_one_bio(struct throtl_data *td, struct throtl_grp *tg, | |
541 | bool rw, struct bio_list *bl) | |
542 | { | |
543 | struct bio *bio; | |
544 | ||
545 | bio = bio_list_pop(&tg->bio_lists[rw]); | |
546 | tg->nr_queued[rw]--; | |
547 | /* Drop bio reference on tg */ | |
548 | throtl_put_tg(tg); | |
549 | ||
550 | BUG_ON(td->nr_queued[rw] <= 0); | |
551 | td->nr_queued[rw]--; | |
552 | ||
553 | throtl_charge_bio(tg, bio); | |
554 | bio_list_add(bl, bio); | |
555 | bio->bi_rw |= REQ_THROTTLED; | |
556 | ||
557 | throtl_trim_slice(td, tg, rw); | |
558 | } | |
559 | ||
560 | static int throtl_dispatch_tg(struct throtl_data *td, struct throtl_grp *tg, | |
561 | struct bio_list *bl) | |
562 | { | |
563 | unsigned int nr_reads = 0, nr_writes = 0; | |
564 | unsigned int max_nr_reads = throtl_grp_quantum*3/4; | |
565 | unsigned int max_nr_writes = throtl_grp_quantum - nr_reads; | |
566 | struct bio *bio; | |
567 | ||
568 | /* Try to dispatch 75% READS and 25% WRITES */ | |
569 | ||
570 | while ((bio = bio_list_peek(&tg->bio_lists[READ])) | |
571 | && tg_may_dispatch(td, tg, bio, NULL)) { | |
572 | ||
573 | tg_dispatch_one_bio(td, tg, bio_data_dir(bio), bl); | |
574 | nr_reads++; | |
575 | ||
576 | if (nr_reads >= max_nr_reads) | |
577 | break; | |
578 | } | |
579 | ||
580 | while ((bio = bio_list_peek(&tg->bio_lists[WRITE])) | |
581 | && tg_may_dispatch(td, tg, bio, NULL)) { | |
582 | ||
583 | tg_dispatch_one_bio(td, tg, bio_data_dir(bio), bl); | |
584 | nr_writes++; | |
585 | ||
586 | if (nr_writes >= max_nr_writes) | |
587 | break; | |
588 | } | |
589 | ||
590 | return nr_reads + nr_writes; | |
591 | } | |
592 | ||
593 | static int throtl_select_dispatch(struct throtl_data *td, struct bio_list *bl) | |
594 | { | |
595 | unsigned int nr_disp = 0; | |
596 | struct throtl_grp *tg; | |
597 | struct throtl_rb_root *st = &td->tg_service_tree; | |
598 | ||
599 | while (1) { | |
600 | tg = throtl_rb_first(st); | |
601 | ||
602 | if (!tg) | |
603 | break; | |
604 | ||
605 | if (time_before(jiffies, tg->disptime)) | |
606 | break; | |
607 | ||
608 | throtl_dequeue_tg(td, tg); | |
609 | ||
610 | nr_disp += throtl_dispatch_tg(td, tg, bl); | |
611 | ||
612 | if (tg->nr_queued[0] || tg->nr_queued[1]) { | |
613 | tg_update_disptime(td, tg); | |
614 | throtl_enqueue_tg(td, tg); | |
615 | } | |
616 | ||
617 | if (nr_disp >= throtl_quantum) | |
618 | break; | |
619 | } | |
620 | ||
621 | return nr_disp; | |
622 | } | |
623 | ||
624 | /* Dispatch throttled bios. Should be called without queue lock held. */ | |
625 | static int throtl_dispatch(struct request_queue *q) | |
626 | { | |
627 | struct throtl_data *td = q->td; | |
628 | unsigned int nr_disp = 0; | |
629 | struct bio_list bio_list_on_stack; | |
630 | struct bio *bio; | |
631 | ||
632 | spin_lock_irq(q->queue_lock); | |
633 | ||
634 | if (!total_nr_queued(td)) | |
635 | goto out; | |
636 | ||
637 | bio_list_init(&bio_list_on_stack); | |
638 | ||
639 | throtl_log(td, "dispatch nr_queued=%lu read=%u write=%u", | |
640 | total_nr_queued(td), td->nr_queued[READ], | |
641 | td->nr_queued[WRITE]); | |
642 | ||
643 | nr_disp = throtl_select_dispatch(td, &bio_list_on_stack); | |
644 | ||
645 | if (nr_disp) | |
646 | throtl_log(td, "bios disp=%u", nr_disp); | |
647 | ||
648 | throtl_schedule_next_dispatch(td); | |
649 | out: | |
650 | spin_unlock_irq(q->queue_lock); | |
651 | ||
652 | /* | |
653 | * If we dispatched some requests, unplug the queue to make sure | |
654 | * immediate dispatch | |
655 | */ | |
656 | if (nr_disp) { | |
657 | while((bio = bio_list_pop(&bio_list_on_stack))) | |
658 | generic_make_request(bio); | |
659 | blk_unplug(q); | |
660 | } | |
661 | return nr_disp; | |
662 | } | |
663 | ||
664 | void blk_throtl_work(struct work_struct *work) | |
665 | { | |
666 | struct throtl_data *td = container_of(work, struct throtl_data, | |
667 | throtl_work.work); | |
668 | struct request_queue *q = td->queue; | |
669 | ||
670 | throtl_dispatch(q); | |
671 | } | |
672 | ||
673 | /* Call with queue lock held */ | |
674 | void throtl_schedule_delayed_work(struct request_queue *q, unsigned long delay) | |
675 | { | |
676 | ||
677 | struct throtl_data *td = q->td; | |
678 | struct delayed_work *dwork = &td->throtl_work; | |
679 | ||
680 | if (total_nr_queued(td) > 0) { | |
681 | /* | |
682 | * We might have a work scheduled to be executed in future. | |
683 | * Cancel that and schedule a new one. | |
684 | */ | |
685 | __cancel_delayed_work(dwork); | |
686 | kblockd_schedule_delayed_work(q, dwork, delay); | |
687 | throtl_log(td, "schedule work. delay=%lu jiffies=%lu", | |
688 | delay, jiffies); | |
689 | } | |
690 | } | |
691 | EXPORT_SYMBOL(throtl_schedule_delayed_work); | |
692 | ||
693 | static void | |
694 | throtl_destroy_tg(struct throtl_data *td, struct throtl_grp *tg) | |
695 | { | |
696 | /* Something wrong if we are trying to remove same group twice */ | |
697 | BUG_ON(hlist_unhashed(&tg->tg_node)); | |
698 | ||
699 | hlist_del_init(&tg->tg_node); | |
700 | ||
701 | /* | |
702 | * Put the reference taken at the time of creation so that when all | |
703 | * queues are gone, group can be destroyed. | |
704 | */ | |
705 | throtl_put_tg(tg); | |
706 | td->nr_undestroyed_grps--; | |
707 | } | |
708 | ||
709 | static void throtl_release_tgs(struct throtl_data *td) | |
710 | { | |
711 | struct hlist_node *pos, *n; | |
712 | struct throtl_grp *tg; | |
713 | ||
714 | hlist_for_each_entry_safe(tg, pos, n, &td->tg_list, tg_node) { | |
715 | /* | |
716 | * If cgroup removal path got to blk_group first and removed | |
717 | * it from cgroup list, then it will take care of destroying | |
718 | * cfqg also. | |
719 | */ | |
720 | if (!blkiocg_del_blkio_group(&tg->blkg)) | |
721 | throtl_destroy_tg(td, tg); | |
722 | } | |
723 | } | |
724 | ||
725 | static void throtl_td_free(struct throtl_data *td) | |
726 | { | |
727 | kfree(td); | |
728 | } | |
729 | ||
730 | /* | |
731 | * Blk cgroup controller notification saying that blkio_group object is being | |
732 | * delinked as associated cgroup object is going away. That also means that | |
733 | * no new IO will come in this group. So get rid of this group as soon as | |
734 | * any pending IO in the group is finished. | |
735 | * | |
736 | * This function is called under rcu_read_lock(). key is the rcu protected | |
737 | * pointer. That means "key" is a valid throtl_data pointer as long as we are | |
738 | * rcu read lock. | |
739 | * | |
740 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means | |
741 | * it should not be NULL as even if queue was going away, cgroup deltion | |
742 | * path got to it first. | |
743 | */ | |
744 | void throtl_unlink_blkio_group(void *key, struct blkio_group *blkg) | |
745 | { | |
746 | unsigned long flags; | |
747 | struct throtl_data *td = key; | |
748 | ||
749 | spin_lock_irqsave(td->queue->queue_lock, flags); | |
750 | throtl_destroy_tg(td, tg_of_blkg(blkg)); | |
751 | spin_unlock_irqrestore(td->queue->queue_lock, flags); | |
752 | } | |
753 | ||
754 | static void throtl_update_blkio_group_read_bps (struct blkio_group *blkg, | |
755 | u64 read_bps) | |
756 | { | |
757 | tg_of_blkg(blkg)->bps[READ] = read_bps; | |
758 | } | |
759 | ||
760 | static void throtl_update_blkio_group_write_bps (struct blkio_group *blkg, | |
761 | u64 write_bps) | |
762 | { | |
763 | tg_of_blkg(blkg)->bps[WRITE] = write_bps; | |
764 | } | |
765 | ||
766 | void throtl_shutdown_timer_wq(struct request_queue *q) | |
767 | { | |
768 | struct throtl_data *td = q->td; | |
769 | ||
770 | cancel_delayed_work_sync(&td->throtl_work); | |
771 | } | |
772 | ||
773 | static struct blkio_policy_type blkio_policy_throtl = { | |
774 | .ops = { | |
775 | .blkio_unlink_group_fn = throtl_unlink_blkio_group, | |
776 | .blkio_update_group_read_bps_fn = | |
777 | throtl_update_blkio_group_read_bps, | |
778 | .blkio_update_group_write_bps_fn = | |
779 | throtl_update_blkio_group_write_bps, | |
780 | }, | |
781 | }; | |
782 | ||
783 | int blk_throtl_bio(struct request_queue *q, struct bio **biop) | |
784 | { | |
785 | struct throtl_data *td = q->td; | |
786 | struct throtl_grp *tg; | |
787 | struct bio *bio = *biop; | |
788 | bool rw = bio_data_dir(bio), update_disptime = true; | |
789 | ||
790 | if (bio->bi_rw & REQ_THROTTLED) { | |
791 | bio->bi_rw &= ~REQ_THROTTLED; | |
792 | return 0; | |
793 | } | |
794 | ||
795 | spin_lock_irq(q->queue_lock); | |
796 | tg = throtl_get_tg(td); | |
797 | ||
798 | if (tg->nr_queued[rw]) { | |
799 | /* | |
800 | * There is already another bio queued in same dir. No | |
801 | * need to update dispatch time. | |
802 | */ | |
803 | update_disptime = false; | |
804 | goto queue_bio; | |
805 | } | |
806 | ||
807 | /* Bio is with-in rate limit of group */ | |
808 | if (tg_may_dispatch(td, tg, bio, NULL)) { | |
809 | throtl_charge_bio(tg, bio); | |
810 | goto out; | |
811 | } | |
812 | ||
813 | queue_bio: | |
814 | throtl_log_tg(td, tg, "[%c] bio. disp=%u sz=%u bps=%llu" | |
815 | " queued=%d/%d", rw == READ ? 'R' : 'W', | |
816 | tg->bytes_disp[rw], bio->bi_size, tg->bps[rw], | |
817 | tg->nr_queued[READ], tg->nr_queued[WRITE]); | |
818 | ||
819 | throtl_add_bio_tg(q->td, tg, bio); | |
820 | *biop = NULL; | |
821 | ||
822 | if (update_disptime) { | |
823 | tg_update_disptime(td, tg); | |
824 | throtl_schedule_next_dispatch(td); | |
825 | } | |
826 | ||
827 | out: | |
828 | spin_unlock_irq(q->queue_lock); | |
829 | return 0; | |
830 | } | |
831 | ||
832 | int blk_throtl_init(struct request_queue *q) | |
833 | { | |
834 | struct throtl_data *td; | |
835 | struct throtl_grp *tg; | |
836 | ||
837 | td = kzalloc_node(sizeof(*td), GFP_KERNEL, q->node); | |
838 | if (!td) | |
839 | return -ENOMEM; | |
840 | ||
841 | INIT_HLIST_HEAD(&td->tg_list); | |
842 | td->tg_service_tree = THROTL_RB_ROOT; | |
843 | ||
844 | /* Init root group */ | |
845 | tg = &td->root_tg; | |
846 | INIT_HLIST_NODE(&tg->tg_node); | |
847 | RB_CLEAR_NODE(&tg->rb_node); | |
848 | bio_list_init(&tg->bio_lists[0]); | |
849 | bio_list_init(&tg->bio_lists[1]); | |
850 | ||
851 | /* Practically unlimited BW */ | |
852 | tg->bps[0] = tg->bps[1] = -1; | |
853 | atomic_set(&tg->ref, 1); | |
854 | ||
855 | INIT_DELAYED_WORK(&td->throtl_work, blk_throtl_work); | |
856 | ||
857 | rcu_read_lock(); | |
858 | blkiocg_add_blkio_group(&blkio_root_cgroup, &tg->blkg, (void *)td, | |
859 | 0, BLKIO_POLICY_THROTL); | |
860 | rcu_read_unlock(); | |
861 | ||
862 | /* Attach throtl data to request queue */ | |
863 | td->queue = q; | |
864 | q->td = td; | |
865 | return 0; | |
866 | } | |
867 | ||
868 | void blk_throtl_exit(struct request_queue *q) | |
869 | { | |
870 | struct throtl_data *td = q->td; | |
871 | bool wait = false; | |
872 | ||
873 | BUG_ON(!td); | |
874 | ||
875 | throtl_shutdown_timer_wq(q); | |
876 | ||
877 | spin_lock_irq(q->queue_lock); | |
878 | throtl_release_tgs(td); | |
879 | blkiocg_del_blkio_group(&td->root_tg.blkg); | |
880 | ||
881 | /* If there are other groups */ | |
882 | if (td->nr_undestroyed_grps >= 1) | |
883 | wait = true; | |
884 | ||
885 | spin_unlock_irq(q->queue_lock); | |
886 | ||
887 | /* | |
888 | * Wait for tg->blkg->key accessors to exit their grace periods. | |
889 | * Do this wait only if there are other undestroyed groups out | |
890 | * there (other than root group). This can happen if cgroup deletion | |
891 | * path claimed the responsibility of cleaning up a group before | |
892 | * queue cleanup code get to the group. | |
893 | * | |
894 | * Do not call synchronize_rcu() unconditionally as there are drivers | |
895 | * which create/delete request queue hundreds of times during scan/boot | |
896 | * and synchronize_rcu() can take significant time and slow down boot. | |
897 | */ | |
898 | if (wait) | |
899 | synchronize_rcu(); | |
900 | throtl_td_free(td); | |
901 | } | |
902 | ||
903 | static int __init throtl_init(void) | |
904 | { | |
905 | blkio_policy_register(&blkio_policy_throtl); | |
906 | return 0; | |
907 | } | |
908 | ||
909 | module_init(throtl_init); |