qemu/block/throttle-groups.c
<<
>>
Prefs
   1/*
   2 * QEMU block throttling group infrastructure
   3 *
   4 * Copyright (C) Nodalink, EURL. 2014
   5 * Copyright (C) Igalia, S.L. 2015
   6 *
   7 * Authors:
   8 *   BenoƮt 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 "sysemu/block-backend.h"
  27#include "block/throttle-groups.h"
  28#include "qemu/queue.h"
  29#include "qemu/thread.h"
  30#include "sysemu/qtest.h"
  31
  32/* The ThrottleGroup structure (with its ThrottleState) is shared
  33 * among different BlockBackends and it's independent from
  34 * AioContext, so in order to use it from different threads it needs
  35 * its own locking.
  36 *
  37 * This locking is however handled internally in this file, so it's
  38 * transparent to outside users.
  39 *
  40 * The whole ThrottleGroup structure is private and invisible to
  41 * outside users, that only use it through its ThrottleState.
  42 *
  43 * In addition to the ThrottleGroup structure, BlockBackendPublic has
  44 * fields that need to be accessed by other members of the group and
  45 * therefore also need to be protected by this lock. Once a
  46 * BlockBackend is registered in a group those fields can be accessed
  47 * by other threads any time.
  48 *
  49 * Again, all this is handled internally and is mostly transparent to
  50 * the outside. The 'throttle_timers' field however has an additional
  51 * constraint because it may be temporarily invalid (see for example
  52 * blk_set_aio_context()). Therefore in this file a thread will
  53 * access some other BlockBackend's timers only after verifying that
  54 * that BlockBackend has throttled requests in the queue.
  55 */
  56typedef struct ThrottleGroup {
  57    char *name; /* This is constant during the lifetime of the group */
  58
  59    QemuMutex lock; /* This lock protects the following four fields */
  60    ThrottleState ts;
  61    QLIST_HEAD(, BlockBackendPublic) head;
  62    BlockBackend *tokens[2];
  63    bool any_timer_armed[2];
  64    QEMUClockType clock_type;
  65
  66    /* These two are protected by the global throttle_groups_lock */
  67    unsigned refcount;
  68    QTAILQ_ENTRY(ThrottleGroup) list;
  69} ThrottleGroup;
  70
  71static QemuMutex throttle_groups_lock;
  72static QTAILQ_HEAD(, ThrottleGroup) throttle_groups =
  73    QTAILQ_HEAD_INITIALIZER(throttle_groups);
  74
  75/* Increments the reference count of a ThrottleGroup given its name.
  76 *
  77 * If no ThrottleGroup is found with the given name a new one is
  78 * created.
  79 *
  80 * @name: the name of the ThrottleGroup
  81 * @ret:  the ThrottleState member of the ThrottleGroup
  82 */
  83ThrottleState *throttle_group_incref(const char *name)
  84{
  85    ThrottleGroup *tg = NULL;
  86    ThrottleGroup *iter;
  87
  88    qemu_mutex_lock(&throttle_groups_lock);
  89
  90    /* Look for an existing group with that name */
  91    QTAILQ_FOREACH(iter, &throttle_groups, list) {
  92        if (!strcmp(name, iter->name)) {
  93            tg = iter;
  94            break;
  95        }
  96    }
  97
  98    /* Create a new one if not found */
  99    if (!tg) {
 100        tg = g_new0(ThrottleGroup, 1);
 101        tg->name = g_strdup(name);
 102        tg->clock_type = QEMU_CLOCK_REALTIME;
 103
 104        if (qtest_enabled()) {
 105            /* For testing block IO throttling only */
 106            tg->clock_type = QEMU_CLOCK_VIRTUAL;
 107        }
 108        qemu_mutex_init(&tg->lock);
 109        throttle_init(&tg->ts);
 110        QLIST_INIT(&tg->head);
 111
 112        QTAILQ_INSERT_TAIL(&throttle_groups, tg, list);
 113    }
 114
 115    tg->refcount++;
 116
 117    qemu_mutex_unlock(&throttle_groups_lock);
 118
 119    return &tg->ts;
 120}
 121
 122/* Decrease the reference count of a ThrottleGroup.
 123 *
 124 * When the reference count reaches zero the ThrottleGroup is
 125 * destroyed.
 126 *
 127 * @ts:  The ThrottleGroup to unref, given by its ThrottleState member
 128 */
 129void throttle_group_unref(ThrottleState *ts)
 130{
 131    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
 132
 133    qemu_mutex_lock(&throttle_groups_lock);
 134    if (--tg->refcount == 0) {
 135        QTAILQ_REMOVE(&throttle_groups, tg, list);
 136        qemu_mutex_destroy(&tg->lock);
 137        g_free(tg->name);
 138        g_free(tg);
 139    }
 140    qemu_mutex_unlock(&throttle_groups_lock);
 141}
 142
 143/* Get the name from a BlockBackend's ThrottleGroup. The name (and the pointer)
 144 * is guaranteed to remain constant during the lifetime of the group.
 145 *
 146 * @blk:  a BlockBackend that is member of a throttling group
 147 * @ret:  the name of the group.
 148 */
 149const char *throttle_group_get_name(BlockBackend *blk)
 150{
 151    BlockBackendPublic *blkp = blk_get_public(blk);
 152    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
 153    return tg->name;
 154}
 155
 156/* Return the next BlockBackend in the round-robin sequence, simulating a
 157 * circular list.
 158 *
 159 * This assumes that tg->lock is held.
 160 *
 161 * @blk: the current BlockBackend
 162 * @ret: the next BlockBackend in the sequence
 163 */
 164static BlockBackend *throttle_group_next_blk(BlockBackend *blk)
 165{
 166    BlockBackendPublic *blkp = blk_get_public(blk);
 167    ThrottleState *ts = blkp->throttle_state;
 168    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
 169    BlockBackendPublic *next = QLIST_NEXT(blkp, round_robin);
 170
 171    if (!next) {
 172        next = QLIST_FIRST(&tg->head);
 173    }
 174
 175    return blk_by_public(next);
 176}
 177
 178/*
 179 * Return whether a BlockBackend has pending requests.
 180 *
 181 * This assumes that tg->lock is held.
 182 *
 183 * @blk: the BlockBackend
 184 * @is_write:  the type of operation (read/write)
 185 * @ret:       whether the BlockBackend has pending requests.
 186 */
 187static inline bool blk_has_pending_reqs(BlockBackend *blk,
 188                                        bool is_write)
 189{
 190    const BlockBackendPublic *blkp = blk_get_public(blk);
 191    return blkp->pending_reqs[is_write];
 192}
 193
 194/* Return the next BlockBackend in the round-robin sequence with pending I/O
 195 * requests.
 196 *
 197 * This assumes that tg->lock is held.
 198 *
 199 * @blk:       the current BlockBackend
 200 * @is_write:  the type of operation (read/write)
 201 * @ret:       the next BlockBackend with pending requests, or blk if there is
 202 *             none.
 203 */
 204static BlockBackend *next_throttle_token(BlockBackend *blk, bool is_write)
 205{
 206    BlockBackendPublic *blkp = blk_get_public(blk);
 207    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
 208    BlockBackend *token, *start;
 209
 210    start = token = tg->tokens[is_write];
 211
 212    /* get next bs round in round robin style */
 213    token = throttle_group_next_blk(token);
 214    while (token != start && !blk_has_pending_reqs(token, is_write)) {
 215        token = throttle_group_next_blk(token);
 216    }
 217
 218    /* If no IO are queued for scheduling on the next round robin token
 219     * then decide the token is the current bs because chances are
 220     * the current bs get the current request queued.
 221     */
 222    if (token == start && !blk_has_pending_reqs(token, is_write)) {
 223        token = blk;
 224    }
 225
 226    /* Either we return the original BB, or one with pending requests */
 227    assert(token == blk || blk_has_pending_reqs(token, is_write));
 228
 229    return token;
 230}
 231
 232/* Check if the next I/O request for a BlockBackend needs to be throttled or
 233 * not. If there's no timer set in this group, set one and update the token
 234 * accordingly.
 235 *
 236 * This assumes that tg->lock is held.
 237 *
 238 * @blk:        the current BlockBackend
 239 * @is_write:   the type of operation (read/write)
 240 * @ret:        whether the I/O request needs to be throttled or not
 241 */
 242static bool throttle_group_schedule_timer(BlockBackend *blk, bool is_write)
 243{
 244    BlockBackendPublic *blkp = blk_get_public(blk);
 245    ThrottleState *ts = blkp->throttle_state;
 246    ThrottleTimers *tt = &blkp->throttle_timers;
 247    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
 248    bool must_wait;
 249
 250    if (atomic_read(&blkp->io_limits_disabled)) {
 251        return false;
 252    }
 253
 254    /* Check if any of the timers in this group is already armed */
 255    if (tg->any_timer_armed[is_write]) {
 256        return true;
 257    }
 258
 259    must_wait = throttle_schedule_timer(ts, tt, is_write);
 260
 261    /* If a timer just got armed, set blk as the current token */
 262    if (must_wait) {
 263        tg->tokens[is_write] = blk;
 264        tg->any_timer_armed[is_write] = true;
 265    }
 266
 267    return must_wait;
 268}
 269
 270/* Start the next pending I/O request for a BlockBackend.  Return whether
 271 * any request was actually pending.
 272 *
 273 * @blk:       the current BlockBackend
 274 * @is_write:  the type of operation (read/write)
 275 */
 276static bool coroutine_fn throttle_group_co_restart_queue(BlockBackend *blk,
 277                                                         bool is_write)
 278{
 279    BlockBackendPublic *blkp = blk_get_public(blk);
 280    bool ret;
 281
 282    qemu_co_mutex_lock(&blkp->throttled_reqs_lock);
 283    ret = qemu_co_queue_next(&blkp->throttled_reqs[is_write]);
 284    qemu_co_mutex_unlock(&blkp->throttled_reqs_lock);
 285
 286    return ret;
 287}
 288
 289/* Look for the next pending I/O request and schedule it.
 290 *
 291 * This assumes that tg->lock is held.
 292 *
 293 * @blk:       the current BlockBackend
 294 * @is_write:  the type of operation (read/write)
 295 */
 296static void schedule_next_request(BlockBackend *blk, bool is_write)
 297{
 298    BlockBackendPublic *blkp = blk_get_public(blk);
 299    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
 300    bool must_wait;
 301    BlockBackend *token;
 302
 303    /* Check if there's any pending request to schedule next */
 304    token = next_throttle_token(blk, is_write);
 305    if (!blk_has_pending_reqs(token, is_write)) {
 306        return;
 307    }
 308
 309    /* Set a timer for the request if it needs to be throttled */
 310    must_wait = throttle_group_schedule_timer(token, is_write);
 311
 312    /* If it doesn't have to wait, queue it for immediate execution */
 313    if (!must_wait) {
 314        /* Give preference to requests from the current blk */
 315        if (qemu_in_coroutine() &&
 316            throttle_group_co_restart_queue(blk, is_write)) {
 317            token = blk;
 318        } else {
 319            ThrottleTimers *tt = &blk_get_public(token)->throttle_timers;
 320            int64_t now = qemu_clock_get_ns(tg->clock_type);
 321            timer_mod(tt->timers[is_write], now);
 322            tg->any_timer_armed[is_write] = true;
 323        }
 324        tg->tokens[is_write] = token;
 325    }
 326}
 327
 328/* Check if an I/O request needs to be throttled, wait and set a timer
 329 * if necessary, and schedule the next request using a round robin
 330 * algorithm.
 331 *
 332 * @blk:       the current BlockBackend
 333 * @bytes:     the number of bytes for this I/O
 334 * @is_write:  the type of operation (read/write)
 335 */
 336void coroutine_fn throttle_group_co_io_limits_intercept(BlockBackend *blk,
 337                                                        unsigned int bytes,
 338                                                        bool is_write)
 339{
 340    bool must_wait;
 341    BlockBackend *token;
 342
 343    BlockBackendPublic *blkp = blk_get_public(blk);
 344    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
 345    qemu_mutex_lock(&tg->lock);
 346
 347    /* First we check if this I/O has to be throttled. */
 348    token = next_throttle_token(blk, is_write);
 349    must_wait = throttle_group_schedule_timer(token, is_write);
 350
 351    /* Wait if there's a timer set or queued requests of this type */
 352    if (must_wait || blkp->pending_reqs[is_write]) {
 353        blkp->pending_reqs[is_write]++;
 354        qemu_mutex_unlock(&tg->lock);
 355        qemu_co_mutex_lock(&blkp->throttled_reqs_lock);
 356        qemu_co_queue_wait(&blkp->throttled_reqs[is_write],
 357                           &blkp->throttled_reqs_lock);
 358        qemu_co_mutex_unlock(&blkp->throttled_reqs_lock);
 359        qemu_mutex_lock(&tg->lock);
 360        blkp->pending_reqs[is_write]--;
 361    }
 362
 363    /* The I/O will be executed, so do the accounting */
 364    throttle_account(blkp->throttle_state, is_write, bytes);
 365
 366    /* Schedule the next request */
 367    schedule_next_request(blk, is_write);
 368
 369    qemu_mutex_unlock(&tg->lock);
 370}
 371
 372typedef struct {
 373    BlockBackend *blk;
 374    bool is_write;
 375} RestartData;
 376
 377static void coroutine_fn throttle_group_restart_queue_entry(void *opaque)
 378{
 379    RestartData *data = opaque;
 380    BlockBackend *blk = data->blk;
 381    bool is_write = data->is_write;
 382    BlockBackendPublic *blkp = blk_get_public(blk);
 383    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
 384    bool empty_queue;
 385
 386    empty_queue = !throttle_group_co_restart_queue(blk, is_write);
 387
 388    /* If the request queue was empty then we have to take care of
 389     * scheduling the next one */
 390    if (empty_queue) {
 391        qemu_mutex_lock(&tg->lock);
 392        schedule_next_request(blk, is_write);
 393        qemu_mutex_unlock(&tg->lock);
 394    }
 395
 396    g_free(data);
 397}
 398
 399static void throttle_group_restart_queue(BlockBackend *blk, bool is_write)
 400{
 401    Coroutine *co;
 402    RestartData *rd = g_new0(RestartData, 1);
 403
 404    rd->blk = blk;
 405    rd->is_write = is_write;
 406
 407    co = qemu_coroutine_create(throttle_group_restart_queue_entry, rd);
 408    aio_co_enter(blk_get_aio_context(blk), co);
 409}
 410
 411void throttle_group_restart_blk(BlockBackend *blk)
 412{
 413    BlockBackendPublic *blkp = blk_get_public(blk);
 414
 415    if (blkp->throttle_state) {
 416        throttle_group_restart_queue(blk, 0);
 417        throttle_group_restart_queue(blk, 1);
 418    }
 419}
 420
 421/* Update the throttle configuration for a particular group. Similar
 422 * to throttle_config(), but guarantees atomicity within the
 423 * throttling group.
 424 *
 425 * @blk: a BlockBackend that is a member of the group
 426 * @cfg: the configuration to set
 427 */
 428void throttle_group_config(BlockBackend *blk, ThrottleConfig *cfg)
 429{
 430    BlockBackendPublic *blkp = blk_get_public(blk);
 431    ThrottleState *ts = blkp->throttle_state;
 432    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
 433    qemu_mutex_lock(&tg->lock);
 434    throttle_config(ts, tg->clock_type, cfg);
 435    qemu_mutex_unlock(&tg->lock);
 436
 437    throttle_group_restart_blk(blk);
 438}
 439
 440/* Get the throttle configuration from a particular group. Similar to
 441 * throttle_get_config(), but guarantees atomicity within the
 442 * throttling group.
 443 *
 444 * @blk: a BlockBackend that is a member of the group
 445 * @cfg: the configuration will be written here
 446 */
 447void throttle_group_get_config(BlockBackend *blk, ThrottleConfig *cfg)
 448{
 449    BlockBackendPublic *blkp = blk_get_public(blk);
 450    ThrottleState *ts = blkp->throttle_state;
 451    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
 452    qemu_mutex_lock(&tg->lock);
 453    throttle_get_config(ts, cfg);
 454    qemu_mutex_unlock(&tg->lock);
 455}
 456
 457/* ThrottleTimers callback. This wakes up a request that was waiting
 458 * because it had been throttled.
 459 *
 460 * @blk:       the BlockBackend whose request had been throttled
 461 * @is_write:  the type of operation (read/write)
 462 */
 463static void timer_cb(BlockBackend *blk, bool is_write)
 464{
 465    BlockBackendPublic *blkp = blk_get_public(blk);
 466    ThrottleState *ts = blkp->throttle_state;
 467    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
 468
 469    /* The timer has just been fired, so we can update the flag */
 470    qemu_mutex_lock(&tg->lock);
 471    tg->any_timer_armed[is_write] = false;
 472    qemu_mutex_unlock(&tg->lock);
 473
 474    /* Run the request that was waiting for this timer */
 475    throttle_group_restart_queue(blk, is_write);
 476}
 477
 478static void read_timer_cb(void *opaque)
 479{
 480    timer_cb(opaque, false);
 481}
 482
 483static void write_timer_cb(void *opaque)
 484{
 485    timer_cb(opaque, true);
 486}
 487
 488/* Register a BlockBackend in the throttling group, also initializing its
 489 * timers and updating its throttle_state pointer to point to it. If a
 490 * throttling group with that name does not exist yet, it will be created.
 491 *
 492 * @blk:       the BlockBackend to insert
 493 * @groupname: the name of the group
 494 */
 495void throttle_group_register_blk(BlockBackend *blk, const char *groupname)
 496{
 497    int i;
 498    BlockBackendPublic *blkp = blk_get_public(blk);
 499    ThrottleState *ts = throttle_group_incref(groupname);
 500    ThrottleGroup *tg = container_of(ts, ThrottleGroup, ts);
 501    blkp->throttle_state = ts;
 502
 503    qemu_mutex_lock(&tg->lock);
 504    /* If the ThrottleGroup is new set this BlockBackend as the token */
 505    for (i = 0; i < 2; i++) {
 506        if (!tg->tokens[i]) {
 507            tg->tokens[i] = blk;
 508        }
 509    }
 510
 511    QLIST_INSERT_HEAD(&tg->head, blkp, round_robin);
 512
 513    throttle_timers_init(&blkp->throttle_timers,
 514                         blk_get_aio_context(blk),
 515                         tg->clock_type,
 516                         read_timer_cb,
 517                         write_timer_cb,
 518                         blk);
 519
 520    qemu_mutex_unlock(&tg->lock);
 521}
 522
 523/* Unregister a BlockBackend from its group, removing it from the list,
 524 * destroying the timers and setting the throttle_state pointer to NULL.
 525 *
 526 * The BlockBackend must not have pending throttled requests, so the caller has
 527 * to drain them first.
 528 *
 529 * The group will be destroyed if it's empty after this operation.
 530 *
 531 * @blk: the BlockBackend to remove
 532 */
 533void throttle_group_unregister_blk(BlockBackend *blk)
 534{
 535    BlockBackendPublic *blkp = blk_get_public(blk);
 536    ThrottleGroup *tg = container_of(blkp->throttle_state, ThrottleGroup, ts);
 537    int i;
 538
 539    assert(blkp->pending_reqs[0] == 0 && blkp->pending_reqs[1] == 0);
 540    assert(qemu_co_queue_empty(&blkp->throttled_reqs[0]));
 541    assert(qemu_co_queue_empty(&blkp->throttled_reqs[1]));
 542
 543    qemu_mutex_lock(&tg->lock);
 544    for (i = 0; i < 2; i++) {
 545        if (tg->tokens[i] == blk) {
 546            BlockBackend *token = throttle_group_next_blk(blk);
 547            /* Take care of the case where this is the last blk in the group */
 548            if (token == blk) {
 549                token = NULL;
 550            }
 551            tg->tokens[i] = token;
 552        }
 553    }
 554
 555    /* remove the current blk from the list */
 556    QLIST_REMOVE(blkp, round_robin);
 557    throttle_timers_destroy(&blkp->throttle_timers);
 558    qemu_mutex_unlock(&tg->lock);
 559
 560    throttle_group_unref(&tg->ts);
 561    blkp->throttle_state = NULL;
 562}
 563
 564static void throttle_groups_init(void)
 565{
 566    qemu_mutex_init(&throttle_groups_lock);
 567}
 568
 569block_init(throttle_groups_init);
 570