job.c

   1 /*
   2  * Background jobs (long-running operations)
   3  *
   4  * Copyright (c) 2011 IBM Corp.
   5  * Copyright (c) 2012, 2018 Red Hat, Inc.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "qemu-common.h"
  28 #include "qapi/error.h"
  29 #include "qemu/job.h"
  30 #include "qemu/id.h"
  31 #include "qemu/main-loop.h"
  32 #include "trace-root.h"
  33 #include "qapi/qapi-events-job.h"
  34
  35 static QLIST_HEAD(, Job) jobs = QLIST_HEAD_INITIALIZER(jobs);
  36
  37 /* Job State Transition Table */
  38 bool JobSTT[JOB_STATUS__MAX][JOB_STATUS__MAX] = {
  39                                     /* U, C, R, P, Y, S, W, D, X, E, N */
  40     /* U: */ [JOB_STATUS_UNDEFINED] = {0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0},
  41     /* C: */ [JOB_STATUS_CREATED]   = {0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1},
  42     /* R: */ [JOB_STATUS_RUNNING]   = {0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0},
  43     /* P: */ [JOB_STATUS_PAUSED]    = {0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0},
  44     /* Y: */ [JOB_STATUS_READY]     = {0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0},
  45     /* S: */ [JOB_STATUS_STANDBY]   = {0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
  46     /* W: */ [JOB_STATUS_WAITING]   = {0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0},
  47     /* D: */ [JOB_STATUS_PENDING]   = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
  48     /* X: */ [JOB_STATUS_ABORTING]  = {0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0},
  49     /* E: */ [JOB_STATUS_CONCLUDED] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1},
  50     /* N: */ [JOB_STATUS_NULL]      = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
  51 };
  52
  53 bool JobVerbTable[JOB_VERB__MAX][JOB_STATUS__MAX] = {
  54                                     /* U, C, R, P, Y, S, W, D, X, E, N */
  55     [JOB_VERB_CANCEL]               = {0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0},
  56     [JOB_VERB_PAUSE]                = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
  57     [JOB_VERB_RESUME]               = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
  58     [JOB_VERB_SET_SPEED]            = {0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0},
  59     [JOB_VERB_COMPLETE]             = {0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0},
  60     [JOB_VERB_FINALIZE]             = {0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0},
  61     [JOB_VERB_DISMISS]              = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0},
  62 };
  63
  64 /* Transactional group of jobs */
  65 struct JobTxn {
  66
  67     /* Is this txn being cancelled? */
  68     bool aborting;
  69
  70     /* List of jobs */
  71     QLIST_HEAD(, Job) jobs;
  72
  73     /* Reference count */
  74     int refcnt;
  75 };
  76
  77 /* Right now, this mutex is only needed to synchronize accesses to job->busy
  78  * and job->sleep_timer, such as concurrent calls to job_do_yield and
  79  * job_enter. */
  80 static QemuMutex job_mutex;
  81
  82 static void job_lock(void)
  83 {
  84     qemu_mutex_lock(&job_mutex);
  85 }
  86
  87 static void job_unlock(void)
  88 {
  89     qemu_mutex_unlock(&job_mutex);
  90 }
  91
  92 static void __attribute__((__constructor__)) job_init(void)
  93 {
  94     qemu_mutex_init(&job_mutex);
  95 }
  96
  97 JobTxn *job_txn_new(void)
  98 {
  99     JobTxn *txn = g_new0(JobTxn, 1);
 100     QLIST_INIT(&txn->jobs);
 101     txn->refcnt = 1;
 102     return txn;
 103 }
 104
 105 static void job_txn_ref(JobTxn *txn)
 106 {
 107     txn->refcnt++;
 108 }
 109
 110 void job_txn_unref(JobTxn *txn)
 111 {
 112     if (txn && --txn->refcnt == 0) {
 113         g_free(txn);
 114     }
 115 }
 116
 117 void job_txn_add_job(JobTxn *txn, Job *job)
 118 {
 119     if (!txn) {
 120         return;
 121     }
 122
 123     assert(!job->txn);
 124     job->txn = txn;
 125
 126     QLIST_INSERT_HEAD(&txn->jobs, job, txn_list);
 127     job_txn_ref(txn);
 128 }
 129
 130 static void job_txn_del_job(Job *job)
 131 {
 132     if (job->txn) {
 133         QLIST_REMOVE(job, txn_list);
 134         job_txn_unref(job->txn);
 135         job->txn = NULL;
 136     }
 137 }
 138
 139 static int job_txn_apply(JobTxn *txn, int fn(Job *), bool lock)
 140 {
 141     AioContext *ctx;
 142     Job *job, *next;
 143     int rc = 0;
 144
 145     QLIST_FOREACH_SAFE(job, &txn->jobs, txn_list, next) {
 146         if (lock) {
 147             ctx = job->aio_context;
 148             aio_context_acquire(ctx);
 149         }
 150         rc = fn(job);
 151         if (lock) {
 152             aio_context_release(ctx);
 153         }
 154         if (rc) {
 155             break;
 156         }
 157     }
 158     return rc;
 159 }
 160
 161 bool job_is_internal(Job *job)
 162 {
 163     return (job->id == NULL);
 164 }
 165
 166 static void job_state_transition(Job *job, JobStatus s1)
 167 {
 168     JobStatus s0 = job->status;
 169     assert(s1 >= 0 && s1 <= JOB_STATUS__MAX);
 170     trace_job_state_transition(job, job->ret,
 171                                JobSTT[s0][s1] ? "allowed" : "disallowed",
 172                                JobStatus_str(s0), JobStatus_str(s1));
 173     assert(JobSTT[s0][s1]);
 174     job->status = s1;
 175
 176     if (!job_is_internal(job) && s1 != s0) {
 177         qapi_event_send_job_status_change(job->id, job->status);
 178     }
 179 }
 180
 181 int job_apply_verb(Job *job, JobVerb verb, Error **errp)
 182 {
 183     JobStatus s0 = job->status;
 184     assert(verb >= 0 && verb <= JOB_VERB__MAX);
 185     trace_job_apply_verb(job, JobStatus_str(s0), JobVerb_str(verb),
 186                          JobVerbTable[verb][s0] ? "allowed" : "prohibited");
 187     if (JobVerbTable[verb][s0]) {
 188         return 0;
 189     }
 190     error_setg(errp, "Job '%s' in state '%s' cannot accept command verb '%s'",
 191                job->id, JobStatus_str(s0), JobVerb_str(verb));
 192     return -EPERM;
 193 }
 194
 195 JobType job_type(const Job *job)
 196 {
 197     return job->driver->job_type;
 198 }
 199
 200 const char *job_type_str(const Job *job)
 201 {
 202     return JobType_str(job_type(job));
 203 }
 204
 205 bool job_is_cancelled(Job *job)
 206 {
 207     return job->cancelled;
 208 }
 209
 210 bool job_is_ready(Job *job)
 211 {
 212     switch (job->status) {
 213     case JOB_STATUS_UNDEFINED:
 214     case JOB_STATUS_CREATED:
 215     case JOB_STATUS_RUNNING:
 216     case JOB_STATUS_PAUSED:
 217     case JOB_STATUS_WAITING:
 218     case JOB_STATUS_PENDING:
 219     case JOB_STATUS_ABORTING:
 220     case JOB_STATUS_CONCLUDED:
 221     case JOB_STATUS_NULL:
 222         return false;
 223     case JOB_STATUS_READY:
 224     case JOB_STATUS_STANDBY:
 225         return true;
 226     default:
 227         g_assert_not_reached();
 228     }
 229     return false;
 230 }
 231
 232 bool job_is_completed(Job *job)
 233 {
 234     switch (job->status) {
 235     case JOB_STATUS_UNDEFINED:
 236     case JOB_STATUS_CREATED:
 237     case JOB_STATUS_RUNNING:
 238     case JOB_STATUS_PAUSED:
 239     case JOB_STATUS_READY:
 240     case JOB_STATUS_STANDBY:
 241         return false;
 242     case JOB_STATUS_WAITING:
 243     case JOB_STATUS_PENDING:
 244     case JOB_STATUS_ABORTING:
 245     case JOB_STATUS_CONCLUDED:
 246     case JOB_STATUS_NULL:
 247         return true;
 248     default:
 249         g_assert_not_reached();
 250     }
 251     return false;
 252 }
 253
 254 static bool job_started(Job *job)
 255 {
 256     return job->co;
 257 }
 258
 259 static bool job_should_pause(Job *job)
 260 {
 261     return job->pause_count > 0;
 262 }
 263
 264 Job *job_next(Job *job)
 265 {
 266     if (!job) {
 267         return QLIST_FIRST(&jobs);
 268     }
 269     return QLIST_NEXT(job, job_list);
 270 }
 271
 272 Job *job_get(const char *id)
 273 {
 274     Job *job;
 275
 276     QLIST_FOREACH(job, &jobs, job_list) {
 277         if (job->id && !strcmp(id, job->id)) {
 278             return job;
 279         }
 280     }
 281
 282     return NULL;
 283 }
 284
 285 static void job_sleep_timer_cb(void *opaque)
 286 {
 287     Job *job = opaque;
 288
 289     job_enter(job);
 290 }
 291
 292 void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn,
 293                  AioContext *ctx, int flags, BlockCompletionFunc *cb,
 294                  void *opaque, Error **errp)
 295 {
 296     Job *job;
 297
 298     if (job_id) {
 299         if (flags & JOB_INTERNAL) {
 300             error_setg(errp, "Cannot specify job ID for internal job");
 301             return NULL;
 302         }
 303         if (!id_wellformed(job_id)) {
 304             error_setg(errp, "Invalid job ID '%s'", job_id);
 305             return NULL;
 306         }
 307         if (job_get(job_id)) {
 308             error_setg(errp, "Job ID '%s' already in use", job_id);
 309             return NULL;
 310         }
 311     } else if (!(flags & JOB_INTERNAL)) {
 312         error_setg(errp, "An explicit job ID is required");
 313         return NULL;
 314     }
 315
 316     job = g_malloc0(driver->instance_size);
 317     job->driver        = driver;
 318     job->id            = g_strdup(job_id);
 319     job->refcnt        = 1;
 320     job->aio_context   = ctx;
 321     job->busy          = false;
 322     job->paused        = true;
 323     job->pause_count   = 1;
 324     job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE);
 325     job->auto_dismiss  = !(flags & JOB_MANUAL_DISMISS);
 326     job->cb            = cb;
 327     job->opaque        = opaque;
 328
 329     notifier_list_init(&job->on_finalize_cancelled);
 330     notifier_list_init(&job->on_finalize_completed);
 331     notifier_list_init(&job->on_pending);
 332     notifier_list_init(&job->on_ready);
 333
 334     job_state_transition(job, JOB_STATUS_CREATED);
 335     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
 336                    QEMU_CLOCK_REALTIME, SCALE_NS,
 337                    job_sleep_timer_cb, job);
 338
 339     QLIST_INSERT_HEAD(&jobs, job, job_list);
 340
 341     /* Single jobs are modeled as single-job transactions for sake of
 342      * consolidating the job management logic */
 343     if (!txn) {
 344         txn = job_txn_new();
 345         job_txn_add_job(txn, job);
 346         job_txn_unref(txn);
 347     } else {
 348         job_txn_add_job(txn, job);
 349     }
 350
 351     return job;
 352 }
 353
 354 void job_ref(Job *job)
 355 {
 356     ++job->refcnt;
 357 }
 358
 359 void job_unref(Job *job)
 360 {
 361     if (--job->refcnt == 0) {
 362         assert(job->status == JOB_STATUS_NULL);
 363         assert(!timer_pending(&job->sleep_timer));
 364         assert(!job->txn);
 365
 366         if (job->driver->free) {
 367             job->driver->free(job);
 368         }
 369
 370         QLIST_REMOVE(job, job_list);
 371
 372         error_free(job->err);
 373         g_free(job->id);
 374         g_free(job);
 375     }
 376 }
 377
 378 void job_progress_update(Job *job, uint64_t done)
 379 {
 380     job->progress_current += done;
 381 }
 382
 383 void job_progress_set_remaining(Job *job, uint64_t remaining)
 384 {
 385     job->progress_total = job->progress_current + remaining;
 386 }
 387
 388 void job_progress_increase_remaining(Job *job, uint64_t delta)
 389 {
 390     job->progress_total += delta;
 391 }
 392
 393 void job_event_cancelled(Job *job)
 394 {
 395     notifier_list_notify(&job->on_finalize_cancelled, job);
 396 }
 397
 398 void job_event_completed(Job *job)
 399 {
 400     notifier_list_notify(&job->on_finalize_completed, job);
 401 }
 402
 403 static void job_event_pending(Job *job)
 404 {
 405     notifier_list_notify(&job->on_pending, job);
 406 }
 407
 408 static void job_event_ready(Job *job)
 409 {
 410     notifier_list_notify(&job->on_ready, job);
 411 }
 412
 413 void job_enter_cond(Job *job, bool(*fn)(Job *job))
 414 {
 415     if (!job_started(job)) {
 416         return;
 417     }
 418     if (job->deferred_to_main_loop) {
 419         return;
 420     }
 421
 422     job_lock();
 423     if (job->busy) {
 424         job_unlock();
 425         return;
 426     }
 427
 428     if (fn && !fn(job)) {
 429         job_unlock();
 430         return;
 431     }
 432
 433     assert(!job->deferred_to_main_loop);
 434     timer_del(&job->sleep_timer);
 435     job->busy = true;
 436     job_unlock();
 437     aio_co_wake(job->co);
 438 }
 439
 440 void job_enter(Job *job)
 441 {
 442     job_enter_cond(job, NULL);
 443 }
 444
 445 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
 446  * Reentering the job coroutine with job_enter() before the timer has expired
 447  * is allowed and cancels the timer.
 448  *
 449  * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
 450  * called explicitly. */
 451 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
 452 {
 453     job_lock();
 454     if (ns != -1) {
 455         timer_mod(&job->sleep_timer, ns);
 456     }
 457     job->busy = false;
 458     job_unlock();
 459     qemu_coroutine_yield();
 460
 461     /* Set by job_enter_cond() before re-entering the coroutine.  */
 462     assert(job->busy);
 463 }
 464
 465 void coroutine_fn job_pause_point(Job *job)
 466 {
 467     assert(job && job_started(job));
 468
 469     if (!job_should_pause(job)) {
 470         return;
 471     }
 472     if (job_is_cancelled(job)) {
 473         return;
 474     }
 475
 476     if (job->driver->pause) {
 477         job->driver->pause(job);
 478     }
 479
 480     if (job_should_pause(job) && !job_is_cancelled(job)) {
 481         JobStatus status = job->status;
 482         job_state_transition(job, status == JOB_STATUS_READY
 483                                   ? JOB_STATUS_STANDBY
 484                                   : JOB_STATUS_PAUSED);
 485         job->paused = true;
 486         job_do_yield(job, -1);
 487         job->paused = false;
 488         job_state_transition(job, status);
 489     }
 490
 491     if (job->driver->resume) {
 492         job->driver->resume(job);
 493     }
 494 }
 495
 496 void job_yield(Job *job)
 497 {
 498     assert(job->busy);
 499
 500     /* Check cancellation *before* setting busy = false, too!  */
 501     if (job_is_cancelled(job)) {
 502         return;
 503     }
 504
 505     if (!job_should_pause(job)) {
 506         job_do_yield(job, -1);
 507     }
 508
 509     job_pause_point(job);
 510 }
 511
 512 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
 513 {
 514     assert(job->busy);
 515
 516     /* Check cancellation *before* setting busy = false, too!  */
 517     if (job_is_cancelled(job)) {
 518         return;
 519     }
 520
 521     if (!job_should_pause(job)) {
 522         job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
 523     }
 524
 525     job_pause_point(job);
 526 }
 527
 528 void job_drain(Job *job)
 529 {
 530     /* If job is !busy this kicks it into the next pause point. */
 531     job_enter(job);
 532
 533     if (job->driver->drain) {
 534         job->driver->drain(job);
 535     }
 536 }
 537
 538 static void job_completed(Job *job);
 539
 540 static void job_exit(void *opaque)
 541 {
 542     Job *job = (Job *)opaque;
 543     AioContext *aio_context = job->aio_context;
 544
 545     if (job->driver->exit) {
 546         aio_context_acquire(aio_context);
 547         job->driver->exit(job);
 548         aio_context_release(aio_context);
 549     }
 550     job_completed(job);
 551 }
 552
 553 /**
 554  * All jobs must allow a pause point before entering their job proper. This
 555  * ensures that jobs can be paused prior to being started, then resumed later.
 556  */
 557 static void coroutine_fn job_co_entry(void *opaque)
 558 {
 559     Job *job = opaque;
 560
 561     assert(job && job->driver && job->driver->run);
 562     job_pause_point(job);
 563     job->ret = job->driver->run(job, &job->err);
 564     job->deferred_to_main_loop = true;
 565     aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
 566 }
 567
 568
 569 void job_start(Job *job)
 570 {
 571     assert(job && !job_started(job) && job->paused &&
 572            job->driver && job->driver->run);
 573     job->co = qemu_coroutine_create(job_co_entry, job);
 574     job->pause_count--;
 575     job->busy = true;
 576     job->paused = false;
 577     job_state_transition(job, JOB_STATUS_RUNNING);
 578     aio_co_enter(job->aio_context, job->co);
 579 }
 580
 581 /* Assumes the block_job_mutex is held */
 582 static bool job_timer_not_pending(Job *job)
 583 {
 584     return !timer_pending(&job->sleep_timer);
 585 }
 586
 587 void job_pause(Job *job)
 588 {
 589     job->pause_count++;
 590 }
 591
 592 void job_resume(Job *job)
 593 {
 594     assert(job->pause_count > 0);
 595     job->pause_count--;
 596     if (job->pause_count) {
 597         return;
 598     }
 599
 600     /* kick only if no timer is pending */
 601     job_enter_cond(job, job_timer_not_pending);
 602 }
 603
 604 void job_user_pause(Job *job, Error **errp)
 605 {
 606     if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
 607         return;
 608     }
 609     if (job->user_paused) {
 610         error_setg(errp, "Job is already paused");
 611         return;
 612     }
 613     job->user_paused = true;
 614     job_pause(job);
 615 }
 616
 617 bool job_user_paused(Job *job)
 618 {
 619     return job->user_paused;
 620 }
 621
 622 void job_user_resume(Job *job, Error **errp)
 623 {
 624     assert(job);
 625     if (!job->user_paused || job->pause_count <= 0) {
 626         error_setg(errp, "Can't resume a job that was not paused");
 627         return;
 628     }
 629     if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
 630         return;
 631     }
 632     if (job->driver->user_resume) {
 633         job->driver->user_resume(job);
 634     }
 635     job->user_paused = false;
 636     job_resume(job);
 637 }
 638
 639 static void job_do_dismiss(Job *job)
 640 {
 641     assert(job);
 642     job->busy = false;
 643     job->paused = false;
 644     job->deferred_to_main_loop = true;
 645
 646     job_txn_del_job(job);
 647
 648     job_state_transition(job, JOB_STATUS_NULL);
 649     job_unref(job);
 650 }
 651
 652 void job_dismiss(Job **jobptr, Error **errp)
 653 {
 654     Job *job = *jobptr;
 655     /* similarly to _complete, this is QMP-interface only. */
 656     assert(job->id);
 657     if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
 658         return;
 659     }
 660
 661     job_do_dismiss(job);
 662     *jobptr = NULL;
 663 }
 664
 665 void job_early_fail(Job *job)
 666 {
 667     assert(job->status == JOB_STATUS_CREATED);
 668     job_do_dismiss(job);
 669 }
 670
 671 static void job_conclude(Job *job)
 672 {
 673     job_state_transition(job, JOB_STATUS_CONCLUDED);
 674     if (job->auto_dismiss || !job_started(job)) {
 675         job_do_dismiss(job);
 676     }
 677 }
 678
 679 static void job_update_rc(Job *job)
 680 {
 681     if (!job->ret && job_is_cancelled(job)) {
 682         job->ret = -ECANCELED;
 683     }
 684     if (job->ret) {
 685         if (!job->err) {
 686             error_setg(&job->err, "%s", strerror(-job->ret));
 687         }
 688         job_state_transition(job, JOB_STATUS_ABORTING);
 689     }
 690 }
 691
 692 static void job_commit(Job *job)
 693 {
 694     assert(!job->ret);
 695     if (job->driver->commit) {
 696         job->driver->commit(job);
 697     }
 698 }
 699
 700 static void job_abort(Job *job)
 701 {
 702     assert(job->ret);
 703     if (job->driver->abort) {
 704         job->driver->abort(job);
 705     }
 706 }
 707
 708 static void job_clean(Job *job)
 709 {
 710     if (job->driver->clean) {
 711         job->driver->clean(job);
 712     }
 713 }
 714
 715 static int job_finalize_single(Job *job)
 716 {
 717     assert(job_is_completed(job));
 718
 719     /* Ensure abort is called for late-transactional failures */
 720     job_update_rc(job);
 721
 722     if (!job->ret) {
 723         job_commit(job);
 724     } else {
 725         job_abort(job);
 726     }
 727     job_clean(job);
 728
 729     if (job->cb) {
 730         job->cb(job->opaque, job->ret);
 731     }
 732
 733     /* Emit events only if we actually started */
 734     if (job_started(job)) {
 735         if (job_is_cancelled(job)) {
 736             job_event_cancelled(job);
 737         } else {
 738             job_event_completed(job);
 739         }
 740     }
 741
 742     job_txn_del_job(job);
 743     job_conclude(job);
 744     return 0;
 745 }
 746
 747 static void job_cancel_async(Job *job, bool force)
 748 {
 749     if (job->user_paused) {
 750         /* Do not call job_enter here, the caller will handle it.  */
 751         if (job->driver->user_resume) {
 752             job->driver->user_resume(job);
 753         }
 754         job->user_paused = false;
 755         assert(job->pause_count > 0);
 756         job->pause_count--;
 757     }
 758     job->cancelled = true;
 759     /* To prevent 'force == false' overriding a previous 'force == true' */
 760     job->force_cancel |= force;
 761 }
 762
 763 static void job_completed_txn_abort(Job *job)
 764 {
 765     AioContext *ctx;
 766     JobTxn *txn = job->txn;
 767     Job *other_job;
 768
 769     if (txn->aborting) {
 770         /*
 771          * We are cancelled by another job, which will handle everything.
 772          */
 773         return;
 774     }
 775     txn->aborting = true;
 776     job_txn_ref(txn);
 777
 778     /* We are the first failed job. Cancel other jobs. */
 779     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 780         ctx = other_job->aio_context;
 781         aio_context_acquire(ctx);
 782     }
 783
 784     /* Other jobs are effectively cancelled by us, set the status for
 785      * them; this job, however, may or may not be cancelled, depending
 786      * on the caller, so leave it. */
 787     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 788         if (other_job != job) {
 789             job_cancel_async(other_job, false);
 790         }
 791     }
 792     while (!QLIST_EMPTY(&txn->jobs)) {
 793         other_job = QLIST_FIRST(&txn->jobs);
 794         ctx = other_job->aio_context;
 795         if (!job_is_completed(other_job)) {
 796             assert(job_is_cancelled(other_job));
 797             job_finish_sync(other_job, NULL, NULL);
 798         }
 799         job_finalize_single(other_job);
 800         aio_context_release(ctx);
 801     }
 802
 803     job_txn_unref(txn);
 804 }
 805
 806 static int job_prepare(Job *job)
 807 {
 808     if (job->ret == 0 && job->driver->prepare) {
 809         job->ret = job->driver->prepare(job);
 810         job_update_rc(job);
 811     }
 812     return job->ret;
 813 }
 814
 815 static int job_needs_finalize(Job *job)
 816 {
 817     return !job->auto_finalize;
 818 }
 819
 820 static void job_do_finalize(Job *job)
 821 {
 822     int rc;
 823     assert(job && job->txn);
 824
 825     /* prepare the transaction to complete */
 826     rc = job_txn_apply(job->txn, job_prepare, true);
 827     if (rc) {
 828         job_completed_txn_abort(job);
 829     } else {
 830         job_txn_apply(job->txn, job_finalize_single, true);
 831     }
 832 }
 833
 834 void job_finalize(Job *job, Error **errp)
 835 {
 836     assert(job && job->id);
 837     if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
 838         return;
 839     }
 840     job_do_finalize(job);
 841 }
 842
 843 static int job_transition_to_pending(Job *job)
 844 {
 845     job_state_transition(job, JOB_STATUS_PENDING);
 846     if (!job->auto_finalize) {
 847         job_event_pending(job);
 848     }
 849     return 0;
 850 }
 851
 852 void job_transition_to_ready(Job *job)
 853 {
 854     job_state_transition(job, JOB_STATUS_READY);
 855     job_event_ready(job);
 856 }
 857
 858 static void job_completed_txn_success(Job *job)
 859 {
 860     JobTxn *txn = job->txn;
 861     Job *other_job;
 862
 863     job_state_transition(job, JOB_STATUS_WAITING);
 864
 865     /*
 866      * Successful completion, see if there are other running jobs in this
 867      * txn.
 868      */
 869     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 870         if (!job_is_completed(other_job)) {
 871             return;
 872         }
 873         assert(other_job->ret == 0);
 874     }
 875
 876     job_txn_apply(txn, job_transition_to_pending, false);
 877
 878     /* If no jobs need manual finalization, automatically do so */
 879     if (job_txn_apply(txn, job_needs_finalize, false) == 0) {
 880         job_do_finalize(job);
 881     }
 882 }
 883
 884 static void job_completed(Job *job)
 885 {
 886     assert(job && job->txn && !job_is_completed(job));
 887
 888     job_update_rc(job);
 889     trace_job_completed(job, job->ret);
 890     if (job->ret) {
 891         job_completed_txn_abort(job);
 892     } else {
 893         job_completed_txn_success(job);
 894     }
 895 }
 896
 897 void job_cancel(Job *job, bool force)
 898 {
 899     if (job->status == JOB_STATUS_CONCLUDED) {
 900         job_do_dismiss(job);
 901         return;
 902     }
 903     job_cancel_async(job, force);
 904     if (!job_started(job)) {
 905         job_completed(job);
 906     } else if (job->deferred_to_main_loop) {
 907         job_completed_txn_abort(job);
 908     } else {
 909         job_enter(job);
 910     }
 911 }
 912
 913 void job_user_cancel(Job *job, bool force, Error **errp)
 914 {
 915     if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
 916         return;
 917     }
 918     job_cancel(job, force);
 919 }
 920
 921 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
 922  * used with job_finish_sync() without the need for (rather nasty) function
 923  * pointer casts there. */
 924 static void job_cancel_err(Job *job, Error **errp)
 925 {
 926     job_cancel(job, false);
 927 }
 928
 929 int job_cancel_sync(Job *job)
 930 {
 931     return job_finish_sync(job, &job_cancel_err, NULL);
 932 }
 933
 934 void job_cancel_sync_all(void)
 935 {
 936     Job *job;
 937     AioContext *aio_context;
 938
 939     while ((job = job_next(NULL))) {
 940         aio_context = job->aio_context;
 941         aio_context_acquire(aio_context);
 942         job_cancel_sync(job);
 943         aio_context_release(aio_context);
 944     }
 945 }
 946
 947 int job_complete_sync(Job *job, Error **errp)
 948 {
 949     return job_finish_sync(job, job_complete, errp);
 950 }
 951
 952 void job_complete(Job *job, Error **errp)
 953 {
 954     /* Should not be reachable via external interface for internal jobs */
 955     assert(job->id);
 956     if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
 957         return;
 958     }
 959     if (job->pause_count || job_is_cancelled(job) || !job->driver->complete) {
 960         error_setg(errp, "The active block job '%s' cannot be completed",
 961                    job->id);
 962         return;
 963     }
 964
 965     job->driver->complete(job, errp);
 966 }
 967
 968 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
 969 {
 970     Error *local_err = NULL;
 971     int ret;
 972
 973     job_ref(job);
 974
 975     if (finish) {
 976         finish(job, &local_err);
 977     }
 978     if (local_err) {
 979         error_propagate(errp, local_err);
 980         job_unref(job);
 981         return -EBUSY;
 982     }
 983     /* job_drain calls job_enter, and it should be enough to induce progress
 984      * until the job completes or moves to the main thread. */
 985     while (!job->deferred_to_main_loop && !job_is_completed(job)) {
 986         job_drain(job);
 987     }
 988     while (!job_is_completed(job)) {
 989         aio_poll(qemu_get_aio_context(), true);
 990     }
 991     ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
 992     job_unref(job);
 993     return ret;
 994 }