1/*
2 * QEMU throttling infrastructure
3 *
4 * Copyright (C) Nodalink, EURL. 2013-2014
5 * Copyright (C) Igalia, S.L. 2015
6 *
7 * Authors:
8 * Benoît Canet <benoit.canet@nodalink.com>
9 * Alberto Garcia <berto@igalia.com>
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2 or
14 * (at your option) version 3 of the License.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, see <http://www.gnu.org/licenses/>.
23 */
24
25#include "qemu/osdep.h"
26#include "qapi/error.h"
27#include "qemu/throttle.h"
28#include "qemu/timer.h"
29#include "block/aio.h"
30
31/* This function make a bucket leak
32 *
33 * @bkt: the bucket to make leak
34 * @delta_ns: the time delta
35 */
36void throttle_leak_bucket(LeakyBucket *bkt, int64_t delta_ns)
37{
38 double leak;
39
40 /* compute how much to leak */
41 leak = (bkt->avg * (double) delta_ns) / NANOSECONDS_PER_SECOND;
42
43 /* make the bucket leak */
44 bkt->level = MAX(bkt->level - leak, 0);
45
46 /* if we allow bursts for more than one second we also need to
47 * keep track of bkt->burst_level so the bkt->max goal per second
48 * is attained */
49 if (bkt->burst_length > 1) {
50 leak = (bkt->max * (double) delta_ns) / NANOSECONDS_PER_SECOND;
51 bkt->burst_level = MAX(bkt->burst_level - leak, 0);
52 }
53}
54
55/* Calculate the time delta since last leak and make proportionals leaks
56 *
57 * @now: the current timestamp in ns
58 */
59static void throttle_do_leak(ThrottleState *ts, int64_t now)
60{
61 /* compute the time elapsed since the last leak */
62 int64_t delta_ns = now - ts->previous_leak;
63 int i;
64
65 ts->previous_leak = now;
66
67 if (delta_ns <= 0) {
68 return;
69 }
70
71 /* make each bucket leak */
72 for (i = 0; i < BUCKETS_COUNT; i++) {
73 throttle_leak_bucket(&ts->cfg.buckets[i], delta_ns);
74 }
75}
76
77/* do the real job of computing the time to wait
78 *
79 * @limit: the throttling limit
80 * @extra: the number of operation to delay
81 * @ret: the time to wait in ns
82 */
83static int64_t throttle_do_compute_wait(double limit, double extra)
84{
85 double wait = extra * NANOSECONDS_PER_SECOND;
86 wait /= limit;
87 return wait;
88}
89
90/* This function compute the wait time in ns that a leaky bucket should trigger
91 *
92 * @bkt: the leaky bucket we operate on
93 * @ret: the resulting wait time in ns or 0 if the operation can go through
94 */
95int64_t throttle_compute_wait(LeakyBucket *bkt)
96{
97 double extra; /* the number of extra units blocking the io */
98 double bucket_size; /* I/O before throttling to bkt->avg */
99 double burst_bucket_size; /* Before throttling to bkt->max */
100
101 if (!bkt->avg) {
102 return 0;
103 }
104
105 if (!bkt->max) {
106 /* If bkt->max is 0 we still want to allow short bursts of I/O
107 * from the guest, otherwise every other request will be throttled
108 * and performance will suffer considerably. */
109 bucket_size = (double) bkt->avg / 10;
110 burst_bucket_size = 0;
111 } else {
112 /* If we have a burst limit then we have to wait until all I/O
113 * at burst rate has finished before throttling to bkt->avg */
114 bucket_size = bkt->max * bkt->burst_length;
115 burst_bucket_size = (double) bkt->max / 10;
116 }
117
118 /* If the main bucket is full then we have to wait */
119 extra = bkt->level - bucket_size;
120 if (extra > 0) {
121 return throttle_do_compute_wait(bkt->avg, extra);
122 }
123
124 /* If the main bucket is not full yet we still have to check the
125 * burst bucket in order to enforce the burst limit */
126 if (bkt->burst_length > 1) {
127 assert(bkt->max > 0); /* see throttle_is_valid() */
128 extra = bkt->burst_level - burst_bucket_size;
129 if (extra > 0) {
130 return throttle_do_compute_wait(bkt->max, extra);
131 }
132 }
133
134 return 0;
135}
136
137/* This function compute the time that must be waited while this IO
138 *
139 * @is_write: true if the current IO is a write, false if it's a read
140 * @ret: time to wait
141 */
142static int64_t throttle_compute_wait_for(ThrottleState *ts,
143 bool is_write)
144{
145 BucketType to_check[2][4] = { {THROTTLE_BPS_TOTAL,
146 THROTTLE_OPS_TOTAL,
147 THROTTLE_BPS_READ,
148 THROTTLE_OPS_READ},
149 {THROTTLE_BPS_TOTAL,
150 THROTTLE_OPS_TOTAL,
151 THROTTLE_BPS_WRITE,
152 THROTTLE_OPS_WRITE}, };
153 int64_t wait, max_wait = 0;
154 int i;
155
156 for (i = 0; i < 4; i++) {
157 BucketType index = to_check[is_write][i];
158 wait = throttle_compute_wait(&ts->cfg.buckets[index]);
159 if (wait > max_wait) {
160 max_wait = wait;
161 }
162 }
163
164 return max_wait;
165}
166
167/* compute the timer for this type of operation
168 *
169 * @is_write: the type of operation
170 * @now: the current clock timestamp
171 * @next_timestamp: the resulting timer
172 * @ret: true if a timer must be set
173 */
174static bool throttle_compute_timer(ThrottleState *ts,
175 bool is_write,
176 int64_t now,
177 int64_t *next_timestamp)
178{
179 int64_t wait;
180
181 /* leak proportionally to the time elapsed */
182 throttle_do_leak(ts, now);
183
184 /* compute the wait time if any */
185 wait = throttle_compute_wait_for(ts, is_write);
186
187 /* if the code must wait compute when the next timer should fire */
188 if (wait) {
189 *next_timestamp = now + wait;
190 return true;
191 }
192
193 /* else no need to wait at all */
194 *next_timestamp = now;
195 return false;
196}
197
198/* Add timers to event loop */
199void throttle_timers_attach_aio_context(ThrottleTimers *tt,
200 AioContext *new_context)
201{
202 tt->timers[0] = aio_timer_new(new_context, tt->clock_type, SCALE_NS,
203 tt->read_timer_cb, tt->timer_opaque);
204 tt->timers[1] = aio_timer_new(new_context, tt->clock_type, SCALE_NS,
205 tt->write_timer_cb, tt->timer_opaque);
206}
207
208/*
209 * Initialize the ThrottleConfig structure to a valid state
210 * @cfg: the config to initialize
211 */
212void throttle_config_init(ThrottleConfig *cfg)
213{
214 unsigned i;
215 memset(cfg, 0, sizeof(*cfg));
216 for (i = 0; i < BUCKETS_COUNT; i++) {
217 cfg->buckets[i].burst_length = 1;
218 }
219}
220
221/* To be called first on the ThrottleState */
222void throttle_init(ThrottleState *ts)
223{
224 memset(ts, 0, sizeof(ThrottleState));
225 throttle_config_init(&ts->cfg);
226}
227
228/* To be called first on the ThrottleTimers */
229void throttle_timers_init(ThrottleTimers *tt,
230 AioContext *aio_context,
231 QEMUClockType clock_type,
232 QEMUTimerCB *read_timer_cb,
233 QEMUTimerCB *write_timer_cb,
234 void *timer_opaque)
235{
236 memset(tt, 0, sizeof(ThrottleTimers));
237
238 tt->clock_type = clock_type;
239 tt->read_timer_cb = read_timer_cb;
240 tt->write_timer_cb = write_timer_cb;
241 tt->timer_opaque = timer_opaque;
242 throttle_timers_attach_aio_context(tt, aio_context);
243}
244
245/* destroy a timer */
246static void throttle_timer_destroy(QEMUTimer **timer)
247{
248 assert(*timer != NULL);
249
250 timer_del(*timer);
251 timer_free(*timer);
252 *timer = NULL;
253}
254
255/* Remove timers from event loop */
256void throttle_timers_detach_aio_context(ThrottleTimers *tt)
257{
258 int i;
259
260 for (i = 0; i < 2; i++) {
261 throttle_timer_destroy(&tt->timers[i]);
262 }
263}
264
265/* To be called last on the ThrottleTimers */
266void throttle_timers_destroy(ThrottleTimers *tt)
267{
268 throttle_timers_detach_aio_context(tt);
269}
270
271/* is any throttling timer configured */
272bool throttle_timers_are_initialized(ThrottleTimers *tt)
273{
274 if (tt->timers[0]) {
275 return true;
276 }
277
278 return false;
279}
280
281/* Does any throttling must be done
282 *
283 * @cfg: the throttling configuration to inspect
284 * @ret: true if throttling must be done else false
285 */
286bool throttle_enabled(ThrottleConfig *cfg)
287{
288 int i;
289
290 for (i = 0; i < BUCKETS_COUNT; i++) {
291 if (cfg->buckets[i].avg > 0) {
292 return true;
293 }
294 }
295
296 return false;
297}
298
299/* check if a throttling configuration is valid
300 * @cfg: the throttling configuration to inspect
301 * @ret: true if valid else false
302 * @errp: error object
303 */
304bool throttle_is_valid(ThrottleConfig *cfg, Error **errp)
305{
306 int i;
307 bool bps_flag, ops_flag;
308 bool bps_max_flag, ops_max_flag;
309
310 bps_flag = cfg->buckets[THROTTLE_BPS_TOTAL].avg &&
311 (cfg->buckets[THROTTLE_BPS_READ].avg ||
312 cfg->buckets[THROTTLE_BPS_WRITE].avg);
313
314 ops_flag = cfg->buckets[THROTTLE_OPS_TOTAL].avg &&
315 (cfg->buckets[THROTTLE_OPS_READ].avg ||
316 cfg->buckets[THROTTLE_OPS_WRITE].avg);
317
318 bps_max_flag = cfg->buckets[THROTTLE_BPS_TOTAL].max &&
319 (cfg->buckets[THROTTLE_BPS_READ].max ||
320 cfg->buckets[THROTTLE_BPS_WRITE].max);
321
322 ops_max_flag = cfg->buckets[THROTTLE_OPS_TOTAL].max &&
323 (cfg->buckets[THROTTLE_OPS_READ].max ||
324 cfg->buckets[THROTTLE_OPS_WRITE].max);
325
326 if (bps_flag || ops_flag || bps_max_flag || ops_max_flag) {
327 error_setg(errp, "bps/iops/max total values and read/write values"
328 " cannot be used at the same time");
329 return false;
330 }
331
332 if (cfg->op_size &&
333 !cfg->buckets[THROTTLE_OPS_TOTAL].avg &&
334 !cfg->buckets[THROTTLE_OPS_READ].avg &&
335 !cfg->buckets[THROTTLE_OPS_WRITE].avg) {
336 error_setg(errp, "iops size requires an iops value to be set");
337 return false;
338 }
339
340 for (i = 0; i < BUCKETS_COUNT; i++) {
341 LeakyBucket *bkt = &cfg->buckets[i];
342 if (bkt->avg > THROTTLE_VALUE_MAX || bkt->max > THROTTLE_VALUE_MAX) {
343 error_setg(errp, "bps/iops/max values must be within [0, %lld]",
344 THROTTLE_VALUE_MAX);
345 return false;
346 }
347
348 if (!bkt->burst_length) {
349 error_setg(errp, "the burst length cannot be 0");
350 return false;
351 }
352
353 if (bkt->burst_length > 1 && !bkt->max) {
354 error_setg(errp, "burst length set without burst rate");
355 return false;
356 }
357
358 if (bkt->max && bkt->burst_length > THROTTLE_VALUE_MAX / bkt->max) {
359 error_setg(errp, "burst length too high for this burst rate");
360 return false;
361 }
362
363 if (bkt->max && !bkt->avg) {
364 error_setg(errp, "bps_max/iops_max require corresponding"
365 " bps/iops values");
366 return false;
367 }
368
369 if (bkt->max && bkt->max < bkt->avg) {
370 error_setg(errp, "bps_max/iops_max cannot be lower than bps/iops");
371 return false;
372 }
373 }
374
375 return true;
376}
377
378/* Used to configure the throttle
379 *
380 * @ts: the throttle state we are working on
381 * @clock_type: the group's clock_type
382 * @cfg: the config to set
383 */
384void throttle_config(ThrottleState *ts,
385 QEMUClockType clock_type,
386 ThrottleConfig *cfg)
387{
388 int i;
389
390 ts->cfg = *cfg;
391
392 /* Zero bucket level */
393 for (i = 0; i < BUCKETS_COUNT; i++) {
394 ts->cfg.buckets[i].level = 0;
395 ts->cfg.buckets[i].burst_level = 0;
396 }
397
398 ts->previous_leak = qemu_clock_get_ns(clock_type);
399}
400
401/* used to get config
402 *
403 * @ts: the throttle state we are working on
404 * @cfg: the config to write
405 */
406void throttle_get_config(ThrottleState *ts, ThrottleConfig *cfg)
407{
408 *cfg = ts->cfg;
409}
410
411
412/* Schedule the read or write timer if needed
413 *
414 * NOTE: this function is not unit tested due to it's usage of timer_mod
415 *
416 * @tt: the timers structure
417 * @is_write: the type of operation (read/write)
418 * @ret: true if the timer has been scheduled else false
419 */
420bool throttle_schedule_timer(ThrottleState *ts,
421 ThrottleTimers *tt,
422 bool is_write)
423{
424 int64_t now = qemu_clock_get_ns(tt->clock_type);
425 int64_t next_timestamp;
426 bool must_wait;
427
428 must_wait = throttle_compute_timer(ts,
429 is_write,
430 now,
431 &next_timestamp);
432
433 /* request not throttled */
434 if (!must_wait) {
435 return false;
436 }
437
438 /* request throttled and timer pending -> do nothing */
439 if (timer_pending(tt->timers[is_write])) {
440 return true;
441 }
442
443 /* request throttled and timer not pending -> arm timer */
444 timer_mod(tt->timers[is_write], next_timestamp);
445 return true;
446}
447
448/* do the accounting for this operation
449 *
450 * @is_write: the type of operation (read/write)
451 * @size: the size of the operation
452 */
453void throttle_account(ThrottleState *ts, bool is_write, uint64_t size)
454{
455 const BucketType bucket_types_size[2][2] = {
456 { THROTTLE_BPS_TOTAL, THROTTLE_BPS_READ },
457 { THROTTLE_BPS_TOTAL, THROTTLE_BPS_WRITE }
458 };
459 const BucketType bucket_types_units[2][2] = {
460 { THROTTLE_OPS_TOTAL, THROTTLE_OPS_READ },
461 { THROTTLE_OPS_TOTAL, THROTTLE_OPS_WRITE }
462 };
463 double units = 1.0;
464 unsigned i;
465
466 /* if cfg.op_size is defined and smaller than size we compute unit count */
467 if (ts->cfg.op_size && size > ts->cfg.op_size) {
468 units = (double) size / ts->cfg.op_size;
469 }
470
471 for (i = 0; i < 2; i++) {
472 LeakyBucket *bkt;
473
474 bkt = &ts->cfg.buckets[bucket_types_size[is_write][i]];
475 bkt->level += size;
476 if (bkt->burst_length > 1) {
477 bkt->burst_level += size;
478 }
479
480 bkt = &ts->cfg.buckets[bucket_types_units[is_write][i]];
481 bkt->level += units;
482 if (bkt->burst_length > 1) {
483 bkt->burst_level += units;
484 }
485 }
486}
487
488/* return a ThrottleConfig based on the options in a ThrottleLimits
489 *
490 * @arg: the ThrottleLimits object to read from
491 * @cfg: the ThrottleConfig to edit
492 * @errp: error object
493 */
494void throttle_limits_to_config(ThrottleLimits *arg, ThrottleConfig *cfg,
495 Error **errp)
496{
497 if (arg->has_bps_total) {
498 cfg->buckets[THROTTLE_BPS_TOTAL].avg = arg->bps_total;
499 }
500 if (arg->has_bps_read) {
501 cfg->buckets[THROTTLE_BPS_READ].avg = arg->bps_read;
502 }
503 if (arg->has_bps_write) {
504 cfg->buckets[THROTTLE_BPS_WRITE].avg = arg->bps_write;
505 }
506
507 if (arg->has_iops_total) {
508 cfg->buckets[THROTTLE_OPS_TOTAL].avg = arg->iops_total;
509 }
510 if (arg->has_iops_read) {
511 cfg->buckets[THROTTLE_OPS_READ].avg = arg->iops_read;
512 }
513 if (arg->has_iops_write) {
514 cfg->buckets[THROTTLE_OPS_WRITE].avg = arg->iops_write;
515 }
516
517 if (arg->has_bps_total_max) {
518 cfg->buckets[THROTTLE_BPS_TOTAL].max = arg->bps_total_max;
519 }
520 if (arg->has_bps_read_max) {
521 cfg->buckets[THROTTLE_BPS_READ].max = arg->bps_read_max;
522 }
523 if (arg->has_bps_write_max) {
524 cfg->buckets[THROTTLE_BPS_WRITE].max = arg->bps_write_max;
525 }
526 if (arg->has_iops_total_max) {
527 cfg->buckets[THROTTLE_OPS_TOTAL].max = arg->iops_total_max;
528 }
529 if (arg->has_iops_read_max) {
530 cfg->buckets[THROTTLE_OPS_READ].max = arg->iops_read_max;
531 }
532 if (arg->has_iops_write_max) {
533 cfg->buckets[THROTTLE_OPS_WRITE].max = arg->iops_write_max;
534 }
535
536 if (arg->has_bps_total_max_length) {
537 if (arg->bps_total_max_length > UINT_MAX) {
538 error_setg(errp, "bps-total-max-length value must be in"
539 " the range [0, %u]", UINT_MAX);
540 return;
541 }
542 cfg->buckets[THROTTLE_BPS_TOTAL].burst_length = arg->bps_total_max_length;
543 }
544 if (arg->has_bps_read_max_length) {
545 if (arg->bps_read_max_length > UINT_MAX) {
546 error_setg(errp, "bps-read-max-length value must be in"
547 " the range [0, %u]", UINT_MAX);
548 return;
549 }
550 cfg->buckets[THROTTLE_BPS_READ].burst_length = arg->bps_read_max_length;
551 }
552 if (arg->has_bps_write_max_length) {
553 if (arg->bps_write_max_length > UINT_MAX) {
554 error_setg(errp, "bps-write-max-length value must be in"
555 " the range [0, %u]", UINT_MAX);
556 return;
557 }
558 cfg->buckets[THROTTLE_BPS_WRITE].burst_length = arg->bps_write_max_length;
559 }
560 if (arg->has_iops_total_max_length) {
561 if (arg->iops_total_max_length > UINT_MAX) {
562 error_setg(errp, "iops-total-max-length value must be in"
563 " the range [0, %u]", UINT_MAX);
564 return;
565 }
566 cfg->buckets[THROTTLE_OPS_TOTAL].burst_length = arg->iops_total_max_length;
567 }
568 if (arg->has_iops_read_max_length) {
569 if (arg->iops_read_max_length > UINT_MAX) {
570 error_setg(errp, "iops-read-max-length value must be in"
571 " the range [0, %u]", UINT_MAX);
572 return;
573 }
574 cfg->buckets[THROTTLE_OPS_READ].burst_length = arg->iops_read_max_length;
575 }
576 if (arg->has_iops_write_max_length) {
577 if (arg->iops_write_max_length > UINT_MAX) {
578 error_setg(errp, "iops-write-max-length value must be in"
579 " the range [0, %u]", UINT_MAX);
580 return;
581 }
582 cfg->buckets[THROTTLE_OPS_WRITE].burst_length = arg->iops_write_max_length;
583 }
584
585 if (arg->has_iops_size) {
586 cfg->op_size = arg->iops_size;
587 }
588
589 throttle_is_valid(cfg, errp);
590}
591
592/* write the options of a ThrottleConfig to a ThrottleLimits
593 *
594 * @cfg: the ThrottleConfig to read from
595 * @var: the ThrottleLimits to write to
596 */
597void throttle_config_to_limits(ThrottleConfig *cfg, ThrottleLimits *var)
598{
599 var->bps_total = cfg->buckets[THROTTLE_BPS_TOTAL].avg;
600 var->bps_read = cfg->buckets[THROTTLE_BPS_READ].avg;
601 var->bps_write = cfg->buckets[THROTTLE_BPS_WRITE].avg;
602 var->iops_total = cfg->buckets[THROTTLE_OPS_TOTAL].avg;
603 var->iops_read = cfg->buckets[THROTTLE_OPS_READ].avg;
604 var->iops_write = cfg->buckets[THROTTLE_OPS_WRITE].avg;
605 var->bps_total_max = cfg->buckets[THROTTLE_BPS_TOTAL].max;
606 var->bps_read_max = cfg->buckets[THROTTLE_BPS_READ].max;
607 var->bps_write_max = cfg->buckets[THROTTLE_BPS_WRITE].max;
608 var->iops_total_max = cfg->buckets[THROTTLE_OPS_TOTAL].max;
609 var->iops_read_max = cfg->buckets[THROTTLE_OPS_READ].max;
610 var->iops_write_max = cfg->buckets[THROTTLE_OPS_WRITE].max;
611 var->bps_total_max_length = cfg->buckets[THROTTLE_BPS_TOTAL].burst_length;
612 var->bps_read_max_length = cfg->buckets[THROTTLE_BPS_READ].burst_length;
613 var->bps_write_max_length = cfg->buckets[THROTTLE_BPS_WRITE].burst_length;
614 var->iops_total_max_length = cfg->buckets[THROTTLE_OPS_TOTAL].burst_length;
615 var->iops_read_max_length = cfg->buckets[THROTTLE_OPS_READ].burst_length;
616 var->iops_write_max_length = cfg->buckets[THROTTLE_OPS_WRITE].burst_length;
617 var->iops_size = cfg->op_size;
618
619 var->has_bps_total = true;
620 var->has_bps_read = true;
621 var->has_bps_write = true;
622 var->has_iops_total = true;
623 var->has_iops_read = true;
624 var->has_iops_write = true;
625 var->has_bps_total_max = true;
626 var->has_bps_read_max = true;
627 var->has_bps_write_max = true;
628 var->has_iops_total_max = true;
629 var->has_iops_read_max = true;
630 var->has_iops_write_max = true;
631 var->has_bps_read_max_length = true;
632 var->has_bps_total_max_length = true;
633 var->has_bps_write_max_length = true;
634 var->has_iops_total_max_length = true;
635 var->has_iops_read_max_length = true;
636 var->has_iops_write_max_length = true;
637 var->has_iops_size = true;
638}
639