aboutsummaryrefslogtreecommitdiff
path: root/block/throttle-groups.c
blob: da2b490c381abb08e4be2e24b8df1a4e93ff651a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
/*
 * QEMU block throttling group infrastructure
 *
 * Copyright (C) Nodalink, EURL. 2014
 * Copyright (C) Igalia, S.L. 2015
 *
 * Authors:
 *   BenoƮt Canet <benoit.canet@nodalink.com>
 *   Alberto Garcia <berto@igalia.com>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
 * published by the Free Software Foundation; either version 2 or
 * (at your option) version 3 of the License.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, see <http://www.gnu.org/licenses/>.
 */

#include "qemu/osdep.h"
#include "sysemu/block-backend.h"
#include "block/throttle-groups.h"
#include "qemu/queue.h"
#include "qemu/thread.h"
#include "sysemu/qtest.h"

/* The ThrottleGroup structure (with its ThrottleState) is shared
 * among different BlockBackends and it's independent from
 * AioContext, so in order to use it from different threads it needs
 * its own locking.
 *
 * This locking is however handled internally in this file, so it's
 * transparent to outside users.
 *
 * The whole ThrottleGroup structure is private and invisible to
 * outside users, that only use it through its ThrottleState.
 *
 * In addition to the ThrottleGroup structure, BlockBackendPublic has
 * fields that need to be accessed by other members of the group and
 * therefore also need to be protected by this lock. Once a
 * BlockBackend is registered in a group those fields can be accessed
 * by other threads any time.
 *
 * Again, all this is handled internally and is mostly transparent to
 * the outside. The 'throttle_timers' field however has an additional
 * constraint because it may be temporarily invalid (see for example
 * blk_set_aio_context()). Therefore in this file a thread will
 * access some other BlockBackend's timers only after verifying that
 * that BlockBackend has throttled requests in the queue.
 */
typedef struct ThrottleGroup {
    char *name; /* This is constant during the lifetime of the group */

    QemuMutex lock; /* This lock protects the following four fields */
    ThrottleState ts;
    QLIST_HEAD(, BlockBackendPublic) head;
    BlockBackend *tokens[2];
    bool any_timer_armed[2];

    /* These two are protected by the global throttle_groups_lock */
    unsigned refcount;
    QTAILQ_ENTRY(ThrottleGroup) list;
} ThrottleGroup;

static QemuMutex throttle_groups_lock;
static QTAILQ_HEAD(, ThrottleGroup) throttle_groups =
    QTAILQ_HEAD_INITIALIZER(throttle_groups);

/* Increments the reference count of a ThrottleGroup given its name.
 *
 * If no ThrottleGroup is found with the given name a new one is
 * created.
 *
 * @name: the name of the ThrottleGroup
 * @ret:  the ThrottleState member of the ThrottleGroup
 */
ThrottleState *throttle_group_incref(const char *name)
{
    ThrottleGroup *tg = NULL;
    ThrottleGroup *iter;

    qemu_mutex_lock(&throttle_groups_lock);

    /* Look for an existing group with that name */
    QTAILQ_FOREACH(iter, &throttle_groups, list) {
        if (!strcmp(name, iter->name)) {
            tg = iter;
            break;
        }
    }

    /* Create a new one if not found */
    if (!tg) {
        tg = g_new0(ThrottleGroup, 1);
        tg->name = g_strdup(name);
        qemu_mutex_init(&tg->lock);
        throttle_init(&tg->ts);
        QLIST_INIT(&tg->head);

        QTAILQ_INSERT_TAIL(&throttle_groups, tg, list);
    }

    tg->refcount++;

    qemu_mutex_unlock(&throttle_groups_lock);

    return &tg->ts;
}

/* Decrease the reference count of a ThrottleGroup.
 *
 * When the reference count reaches zero the ThrottleGroup is
 * destroyed.
 *
 * @ts:  The ThrottleGroup to unref, given by its ThrottleState member
 */
void throttle_group_unref(ThrottleState *ts)
{
    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);

    qemu_mutex_lock(&throttle_groups_lock);
    if (--tg->refcount == 0) {
        QTAILQ_REMOVE(&throttle_groups, tg, list);
        qemu_mutex_destroy(&tg->lock);
        g_free(tg->name);
        g_free(tg);
    }
    qemu_mutex_unlock(&throttle_groups_lock);
}

/* Get the name from a BlockBackend's ThrottleGroup. The name (and the pointer)
 * is guaranteed to remain constant during the lifetime of the group.
 *
 * @blk:  a BlockBackend that is member of a throttling group
 * @ret:  the name of the group.
 */
const char *throttle_group_get_name(BlockBackend *blk)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
    return tg->name;
}

/* Return the next BlockBackend in the round-robin sequence, simulating a
 * circular list.
 *
 * This assumes that tg->lock is held.
 *
 * @blk: the current BlockBackend
 * @ret: the next BlockBackend in the sequence
 */
static BlockBackend *throttle_group_next_blk(BlockBackend *blk)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleState *ts = blkp->throttle_state;
    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
    BlockBackendPublic *next = QLIST_NEXT(blkp, round_robin);

    if (!next) {
        next = QLIST_FIRST(&tg->head);
    }

    return blk_by_public(next);
}

/*
 * Return whether a BlockBackend has pending requests.
 *
 * This assumes that tg->lock is held.
 *
 * @blk: the BlockBackend
 * @is_write:  the type of operation (read/write)
 * @ret:       whether the BlockBackend has pending requests.
 */
static inline bool blk_has_pending_reqs(BlockBackend *blk,
                                        bool is_write)
{
    const BlockBackendPublic *blkp = blk_get_public(blk);
    return blkp->pending_reqs[is_write];
}

/* Return the next BlockBackend in the round-robin sequence with pending I/O
 * requests.
 *
 * This assumes that tg->lock is held.
 *
 * @blk:       the current BlockBackend
 * @is_write:  the type of operation (read/write)
 * @ret:       the next BlockBackend with pending requests, or blk if there is
 *             none.
 */
static BlockBackend *next_throttle_token(BlockBackend *blk, bool is_write)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
    BlockBackend *token, *start;

    start = token = tg->tokens[is_write];

    /* get next bs round in round robin style */
    token = throttle_group_next_blk(token);
    while (token != start && !blk_has_pending_reqs(token, is_write)) {
        token = throttle_group_next_blk(token);
    }

    /* If no IO are queued for scheduling on the next round robin token
     * then decide the token is the current bs because chances are
     * the current bs get the current request queued.
     */
    if (token == start && !blk_has_pending_reqs(token, is_write)) {
        token = blk;
    }

    /* Either we return the original BB, or one with pending requests */
    assert(token == blk || blk_has_pending_reqs(token, is_write));

    return token;
}

/* Check if the next I/O request for a BlockBackend needs to be throttled or
 * not. If there's no timer set in this group, set one and update the token
 * accordingly.
 *
 * This assumes that tg->lock is held.
 *
 * @blk:        the current BlockBackend
 * @is_write:   the type of operation (read/write)
 * @ret:        whether the I/O request needs to be throttled or not
 */
static bool throttle_group_schedule_timer(BlockBackend *blk, bool is_write)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleState *ts = blkp->throttle_state;
    ThrottleTimers *tt = &blkp->throttle_timers;
    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
    bool must_wait;

    if (atomic_read(&blkp->io_limits_disabled)) {
        return false;
    }

    /* Check if any of the timers in this group is already armed */
    if (tg->any_timer_armed[is_write]) {
        return true;
    }

    must_wait = throttle_schedule_timer(ts, tt, is_write);

    /* If a timer just got armed, set blk as the current token */
    if (must_wait) {
        tg->tokens[is_write] = blk;
        tg->any_timer_armed[is_write] = true;
    }

    return must_wait;
}

/* Start the next pending I/O request for a BlockBackend.  Return whether
 * any request was actually pending.
 *
 * @blk:       the current BlockBackend
 * @is_write:  the type of operation (read/write)
 */
static bool coroutine_fn throttle_group_co_restart_queue(BlockBackend *blk,
                                                         bool is_write)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    bool ret;

    qemu_co_mutex_lock(&blkp->throttled_reqs_lock);
    ret = qemu_co_queue_next(&blkp->throttled_reqs[is_write]);
    qemu_co_mutex_unlock(&blkp->throttled_reqs_lock);

    return ret;
}

/* Look for the next pending I/O request and schedule it.
 *
 * This assumes that tg->lock is held.
 *
 * @blk:       the current BlockBackend
 * @is_write:  the type of operation (read/write)
 */
static void schedule_next_request(BlockBackend *blk, bool is_write)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
    bool must_wait;
    BlockBackend *token;

    /* Check if there's any pending request to schedule next */
    token = next_throttle_token(blk, is_write);
    if (!blk_has_pending_reqs(token, is_write)) {
        return;
    }

    /* Set a timer for the request if it needs to be throttled */
    must_wait = throttle_group_schedule_timer(token, is_write);

    /* If it doesn't have to wait, queue it for immediate execution */
    if (!must_wait) {
        /* Give preference to requests from the current blk */
        if (qemu_in_coroutine() &&
            throttle_group_co_restart_queue(blk, is_write)) {
            token = blk;
        } else {
            ThrottleTimers *tt = &blk_get_public(token)->throttle_timers;
            int64_t now = qemu_clock_get_ns(tt->clock_type);
            timer_mod(tt->timers[is_write], now);
            tg->any_timer_armed[is_write] = true;
        }
        tg->tokens[is_write] = token;
    }
}

/* Check if an I/O request needs to be throttled, wait and set a timer
 * if necessary, and schedule the next request using a round robin
 * algorithm.
 *
 * @blk:       the current BlockBackend
 * @bytes:     the number of bytes for this I/O
 * @is_write:  the type of operation (read/write)
 */
void coroutine_fn throttle_group_co_io_limits_intercept(BlockBackend *blk,
                                                        unsigned int bytes,
                                                        bool is_write)
{
    bool must_wait;
    BlockBackend *token;

    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
    qemu_mutex_lock(&tg->lock);

    /* First we check if this I/O has to be throttled. */
    token = next_throttle_token(blk, is_write);
    must_wait = throttle_group_schedule_timer(token, is_write);

    /* Wait if there's a timer set or queued requests of this type */
    if (must_wait || blkp->pending_reqs[is_write]) {
        blkp->pending_reqs[is_write]++;
        qemu_mutex_unlock(&tg->lock);
        qemu_co_mutex_lock(&blkp->throttled_reqs_lock);
        qemu_co_queue_wait(&blkp->throttled_reqs[is_write],
                           &blkp->throttled_reqs_lock);
        qemu_co_mutex_unlock(&blkp->throttled_reqs_lock);
        qemu_mutex_lock(&tg->lock);
        blkp->pending_reqs[is_write]--;
    }

    /* The I/O will be executed, so do the accounting */
    throttle_account(blkp->throttle_state, is_write, bytes);

    /* Schedule the next request */
    schedule_next_request(blk, is_write);

    qemu_mutex_unlock(&tg->lock);
}

typedef struct {
    BlockBackend *blk;
    bool is_write;
} RestartData;

static void coroutine_fn throttle_group_restart_queue_entry(void *opaque)
{
    RestartData *data = opaque;
    BlockBackend *blk = data->blk;
    bool is_write = data->is_write;
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
    bool empty_queue;

    empty_queue = !throttle_group_co_restart_queue(blk, is_write);

    /* If the request queue was empty then we have to take care of
     * scheduling the next one */
    if (empty_queue) {
        qemu_mutex_lock(&tg->lock);
        schedule_next_request(blk, is_write);
        qemu_mutex_unlock(&tg->lock);
    }
}

static void throttle_group_restart_queue(BlockBackend *blk, bool is_write)
{
    Coroutine *co;
    RestartData rd = {
        .blk = blk,
        .is_write = is_write
    };

    co = qemu_coroutine_create(throttle_group_restart_queue_entry, &rd);
    aio_co_enter(blk_get_aio_context(blk), co);
}

void throttle_group_restart_blk(BlockBackend *blk)
{
    BlockBackendPublic *blkp = blk_get_public(blk);

    if (blkp->throttle_state) {
        throttle_group_restart_queue(blk, 0);
        throttle_group_restart_queue(blk, 1);
    }
}

/* Update the throttle configuration for a particular group. Similar
 * to throttle_config(), but guarantees atomicity within the
 * throttling group.
 *
 * @blk: a BlockBackend that is a member of the group
 * @cfg: the configuration to set
 */
void throttle_group_config(BlockBackend *blk, ThrottleConfig *cfg)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleTimers *tt = &blkp->throttle_timers;
    ThrottleState *ts = blkp->throttle_state;
    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
    qemu_mutex_lock(&tg->lock);
    /* throttle_config() cancels the timers */
    if (timer_pending(tt->timers[0])) {
        tg->any_timer_armed[0] = false;
    }
    if (timer_pending(tt->timers[1])) {
        tg->any_timer_armed[1] = false;
    }
    throttle_config(ts, tt, cfg);
    qemu_mutex_unlock(&tg->lock);

    throttle_group_restart_blk(blk);
}

/* Get the throttle configuration from a particular group. Similar to
 * throttle_get_config(), but guarantees atomicity within the
 * throttling group.
 *
 * @blk: a BlockBackend that is a member of the group
 * @cfg: the configuration will be written here
 */
void throttle_group_get_config(BlockBackend *blk, ThrottleConfig *cfg)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleState *ts = blkp->throttle_state;
    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
    qemu_mutex_lock(&tg->lock);
    throttle_get_config(ts, cfg);
    qemu_mutex_unlock(&tg->lock);
}

/* ThrottleTimers callback. This wakes up a request that was waiting
 * because it had been throttled.
 *
 * @blk:       the BlockBackend whose request had been throttled
 * @is_write:  the type of operation (read/write)
 */
static void timer_cb(BlockBackend *blk, bool is_write)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleState *ts = blkp->throttle_state;
    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);

    /* The timer has just been fired, so we can update the flag */
    qemu_mutex_lock(&tg->lock);
    tg->any_timer_armed[is_write] = false;
    qemu_mutex_unlock(&tg->lock);

    /* Run the request that was waiting for this timer */
    throttle_group_restart_queue(blk, is_write);
}

static void read_timer_cb(void *opaque)
{
    timer_cb(opaque, false);
}

static void write_timer_cb(void *opaque)
{
    timer_cb(opaque, true);
}

/* Register a BlockBackend in the throttling group, also initializing its
 * timers and updating its throttle_state pointer to point to it. If a
 * throttling group with that name does not exist yet, it will be created.
 *
 * @blk:       the BlockBackend to insert
 * @groupname: the name of the group
 */
void throttle_group_register_blk(BlockBackend *blk, const char *groupname)
{
    int i;
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleState *ts = throttle_group_incref(groupname);
    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
    int clock_type = QEMU_CLOCK_REALTIME;

    if (qtest_enabled()) {
        /* For testing block IO throttling only */
        clock_type = QEMU_CLOCK_VIRTUAL;
    }

    blkp->throttle_state = ts;

    qemu_mutex_lock(&tg->lock);
    /* If the ThrottleGroup is new set this BlockBackend as the token */
    for (i = 0; i < 2; i++) {
        if (!tg->tokens[i]) {
            tg->tokens[i] = blk;
        }
    }

    QLIST_INSERT_HEAD(&tg->head, blkp, round_robin);

    throttle_timers_init(&blkp->throttle_timers,
                         blk_get_aio_context(blk),
                         clock_type,
                         read_timer_cb,
                         write_timer_cb,
                         blk);

    qemu_mutex_unlock(&tg->lock);
}

/* Unregister a BlockBackend from its group, removing it from the list,
 * destroying the timers and setting the throttle_state pointer to NULL.
 *
 * The BlockBackend must not have pending throttled requests, so the caller has
 * to drain them first.
 *
 * The group will be destroyed if it's empty after this operation.
 *
 * @blk: the BlockBackend to remove
 */
void throttle_group_unregister_blk(BlockBackend *blk)
{
    BlockBackendPublic *blkp = blk_get_public(blk);
    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
    int i;

    assert(blkp->pending_reqs[0] == 0 && blkp->pending_reqs[1] == 0);
    assert(qemu_co_queue_empty(&blkp->throttled_reqs[0]));
    assert(qemu_co_queue_empty(&blkp->throttled_reqs[1]));

    qemu_mutex_lock(&tg->lock);
    for (i = 0; i < 2; i++) {
        if (tg->tokens[i] == blk) {
            BlockBackend *token = throttle_group_next_blk(blk);
            /* Take care of the case where this is the last blk in the group */
            if (token == blk) {
                token = NULL;
            }
            tg->tokens[i] = token;
        }
    }

    /* remove the current blk from the list */
    QLIST_REMOVE(blkp, round_robin);
    throttle_timers_destroy(&blkp->throttle_timers);
    qemu_mutex_unlock(&tg->lock);

    throttle_group_unref(&tg->ts);
    blkp->throttle_state = NULL;
}

static void throttle_groups_init(void)
{
    qemu_mutex_init(&throttle_groups_lock);
}

block_init(throttle_groups_init);