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 "qapi/error.h"
  28 #include "qemu/job.h"
  29 #include "qemu/id.h"
  30 #include "qemu/main-loop.h"
  31 #include "block/aio-wait.h"
  32 #include "trace/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, 1, 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(Job *job, int fn(Job *))
 140 {
 141     AioContext *inner_ctx;
 142     Job *other_job, *next;
 143     JobTxn *txn = job->txn;
 144     int rc = 0;
 145
 146     /*
 147      * Similar to job_completed_txn_abort, we take each job's lock before
 148      * applying fn, but since we assume that outer_ctx is held by the caller,
 149      * we need to release it here to avoid holding the lock twice - which would
 150      * break AIO_WAIT_WHILE from within fn.
 151      */
 152     job_ref(job);
 153     aio_context_release(job->aio_context);
 154
 155     QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) {
 156         inner_ctx = other_job->aio_context;
 157         aio_context_acquire(inner_ctx);
 158         rc = fn(other_job);
 159         aio_context_release(inner_ctx);
 160         if (rc) {
 161             break;
 162         }
 163     }
 164
 165     /*
 166      * Note that job->aio_context might have been changed by calling fn, so we
 167      * can't use a local variable to cache it.
 168      */
 169     aio_context_acquire(job->aio_context);
 170     job_unref(job);
 171     return rc;
 172 }
 173
 174 bool job_is_internal(Job *job)
 175 {
 176     return (job->id == NULL);
 177 }
 178
 179 static void job_state_transition(Job *job, JobStatus s1)
 180 {
 181     JobStatus s0 = job->status;
 182     assert(s1 >= 0 && s1 < JOB_STATUS__MAX);
 183     trace_job_state_transition(job, job->ret,
 184                                JobSTT[s0][s1] ? "allowed" : "disallowed",
 185                                JobStatus_str(s0), JobStatus_str(s1));
 186     assert(JobSTT[s0][s1]);
 187     job->status = s1;
 188
 189     if (!job_is_internal(job) && s1 != s0) {
 190         qapi_event_send_job_status_change(job->id, job->status);
 191     }
 192 }
 193
 194 int job_apply_verb(Job *job, JobVerb verb, Error **errp)
 195 {
 196     JobStatus s0 = job->status;
 197     assert(verb >= 0 && verb < JOB_VERB__MAX);
 198     trace_job_apply_verb(job, JobStatus_str(s0), JobVerb_str(verb),
 199                          JobVerbTable[verb][s0] ? "allowed" : "prohibited");
 200     if (JobVerbTable[verb][s0]) {
 201         return 0;
 202     }
 203     error_setg(errp, "Job '%s' in state '%s' cannot accept command verb '%s'",
 204                job->id, JobStatus_str(s0), JobVerb_str(verb));
 205     return -EPERM;
 206 }
 207
 208 JobType job_type(const Job *job)
 209 {
 210     return job->driver->job_type;
 211 }
 212
 213 const char *job_type_str(const Job *job)
 214 {
 215     return JobType_str(job_type(job));
 216 }
 217
 218 bool job_is_cancelled(Job *job)
 219 {
 220     return job->cancelled;
 221 }
 222
 223 bool job_is_ready(Job *job)
 224 {
 225     switch (job->status) {
 226     case JOB_STATUS_UNDEFINED:
 227     case JOB_STATUS_CREATED:
 228     case JOB_STATUS_RUNNING:
 229     case JOB_STATUS_PAUSED:
 230     case JOB_STATUS_WAITING:
 231     case JOB_STATUS_PENDING:
 232     case JOB_STATUS_ABORTING:
 233     case JOB_STATUS_CONCLUDED:
 234     case JOB_STATUS_NULL:
 235         return false;
 236     case JOB_STATUS_READY:
 237     case JOB_STATUS_STANDBY:
 238         return true;
 239     default:
 240         g_assert_not_reached();
 241     }
 242     return false;
 243 }
 244
 245 bool job_is_completed(Job *job)
 246 {
 247     switch (job->status) {
 248     case JOB_STATUS_UNDEFINED:
 249     case JOB_STATUS_CREATED:
 250     case JOB_STATUS_RUNNING:
 251     case JOB_STATUS_PAUSED:
 252     case JOB_STATUS_READY:
 253     case JOB_STATUS_STANDBY:
 254         return false;
 255     case JOB_STATUS_WAITING:
 256     case JOB_STATUS_PENDING:
 257     case JOB_STATUS_ABORTING:
 258     case JOB_STATUS_CONCLUDED:
 259     case JOB_STATUS_NULL:
 260         return true;
 261     default:
 262         g_assert_not_reached();
 263     }
 264     return false;
 265 }
 266
 267 static bool job_started(Job *job)
 268 {
 269     return job->co;
 270 }
 271
 272 static bool job_should_pause(Job *job)
 273 {
 274     return job->pause_count > 0;
 275 }
 276
 277 Job *job_next(Job *job)
 278 {
 279     if (!job) {
 280         return QLIST_FIRST(&jobs);
 281     }
 282     return QLIST_NEXT(job, job_list);
 283 }
 284
 285 Job *job_get(const char *id)
 286 {
 287     Job *job;
 288
 289     QLIST_FOREACH(job, &jobs, job_list) {
 290         if (job->id && !strcmp(id, job->id)) {
 291             return job;
 292         }
 293     }
 294
 295     return NULL;
 296 }
 297
 298 static void job_sleep_timer_cb(void *opaque)
 299 {
 300     Job *job = opaque;
 301
 302     job_enter(job);
 303 }
 304
 305 void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn,
 306                  AioContext *ctx, int flags, BlockCompletionFunc *cb,
 307                  void *opaque, Error **errp)
 308 {
 309     Job *job;
 310
 311     if (job_id) {
 312         if (flags & JOB_INTERNAL) {
 313             error_setg(errp, "Cannot specify job ID for internal job");
 314             return NULL;
 315         }
 316         if (!id_wellformed(job_id)) {
 317             error_setg(errp, "Invalid job ID '%s'", job_id);
 318             return NULL;
 319         }
 320         if (job_get(job_id)) {
 321             error_setg(errp, "Job ID '%s' already in use", job_id);
 322             return NULL;
 323         }
 324     } else if (!(flags & JOB_INTERNAL)) {
 325         error_setg(errp, "An explicit job ID is required");
 326         return NULL;
 327     }
 328
 329     job = g_malloc0(driver->instance_size);
 330     job->driver        = driver;
 331     job->id            = g_strdup(job_id);
 332     job->refcnt        = 1;
 333     job->aio_context   = ctx;
 334     job->busy          = false;
 335     job->paused        = true;
 336     job->pause_count   = 1;
 337     job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE);
 338     job->auto_dismiss  = !(flags & JOB_MANUAL_DISMISS);
 339     job->cb            = cb;
 340     job->opaque        = opaque;
 341
 342     progress_init(&job->progress);
 343
 344     notifier_list_init(&job->on_finalize_cancelled);
 345     notifier_list_init(&job->on_finalize_completed);
 346     notifier_list_init(&job->on_pending);
 347     notifier_list_init(&job->on_ready);
 348
 349     job_state_transition(job, JOB_STATUS_CREATED);
 350     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
 351                    QEMU_CLOCK_REALTIME, SCALE_NS,
 352                    job_sleep_timer_cb, job);
 353
 354     QLIST_INSERT_HEAD(&jobs, job, job_list);
 355
 356     /* Single jobs are modeled as single-job transactions for sake of
 357      * consolidating the job management logic */
 358     if (!txn) {
 359         txn = job_txn_new();
 360         job_txn_add_job(txn, job);
 361         job_txn_unref(txn);
 362     } else {
 363         job_txn_add_job(txn, job);
 364     }
 365
 366     return job;
 367 }
 368
 369 void job_ref(Job *job)
 370 {
 371     ++job->refcnt;
 372 }
 373
 374 void job_unref(Job *job)
 375 {
 376     if (--job->refcnt == 0) {
 377         assert(job->status == JOB_STATUS_NULL);
 378         assert(!timer_pending(&job->sleep_timer));
 379         assert(!job->txn);
 380
 381         if (job->driver->free) {
 382             job->driver->free(job);
 383         }
 384
 385         QLIST_REMOVE(job, job_list);
 386
 387         progress_destroy(&job->progress);
 388         error_free(job->err);
 389         g_free(job->id);
 390         g_free(job);
 391     }
 392 }
 393
 394 void job_progress_update(Job *job, uint64_t done)
 395 {
 396     progress_work_done(&job->progress, done);
 397 }
 398
 399 void job_progress_set_remaining(Job *job, uint64_t remaining)
 400 {
 401     progress_set_remaining(&job->progress, remaining);
 402 }
 403
 404 void job_progress_increase_remaining(Job *job, uint64_t delta)
 405 {
 406     progress_increase_remaining(&job->progress, delta);
 407 }
 408
 409 void job_event_cancelled(Job *job)
 410 {
 411     notifier_list_notify(&job->on_finalize_cancelled, job);
 412 }
 413
 414 void job_event_completed(Job *job)
 415 {
 416     notifier_list_notify(&job->on_finalize_completed, job);
 417 }
 418
 419 static void job_event_pending(Job *job)
 420 {
 421     notifier_list_notify(&job->on_pending, job);
 422 }
 423
 424 static void job_event_ready(Job *job)
 425 {
 426     notifier_list_notify(&job->on_ready, job);
 427 }
 428
 429 static void job_event_idle(Job *job)
 430 {
 431     notifier_list_notify(&job->on_idle, job);
 432 }
 433
 434 void job_enter_cond(Job *job, bool(*fn)(Job *job))
 435 {
 436     if (!job_started(job)) {
 437         return;
 438     }
 439     if (job->deferred_to_main_loop) {
 440         return;
 441     }
 442
 443     job_lock();
 444     if (job->busy) {
 445         job_unlock();
 446         return;
 447     }
 448
 449     if (fn && !fn(job)) {
 450         job_unlock();
 451         return;
 452     }
 453
 454     assert(!job->deferred_to_main_loop);
 455     timer_del(&job->sleep_timer);
 456     job->busy = true;
 457     job_unlock();
 458     aio_co_enter(job->aio_context, job->co);
 459 }
 460
 461 void job_enter(Job *job)
 462 {
 463     job_enter_cond(job, NULL);
 464 }
 465
 466 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
 467  * Reentering the job coroutine with job_enter() before the timer has expired
 468  * is allowed and cancels the timer.
 469  *
 470  * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
 471  * called explicitly. */
 472 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
 473 {
 474     job_lock();
 475     if (ns != -1) {
 476         timer_mod(&job->sleep_timer, ns);
 477     }
 478     job->busy = false;
 479     job_event_idle(job);
 480     job_unlock();
 481     qemu_coroutine_yield();
 482
 483     /* Set by job_enter_cond() before re-entering the coroutine.  */
 484     assert(job->busy);
 485 }
 486
 487 void coroutine_fn job_pause_point(Job *job)
 488 {
 489     assert(job && job_started(job));
 490
 491     if (!job_should_pause(job)) {
 492         return;
 493     }
 494     if (job_is_cancelled(job)) {
 495         return;
 496     }
 497
 498     if (job->driver->pause) {
 499         job->driver->pause(job);
 500     }
 501
 502     if (job_should_pause(job) && !job_is_cancelled(job)) {
 503         JobStatus status = job->status;
 504         job_state_transition(job, status == JOB_STATUS_READY
 505                                   ? JOB_STATUS_STANDBY
 506                                   : JOB_STATUS_PAUSED);
 507         job->paused = true;
 508         job_do_yield(job, -1);
 509         job->paused = false;
 510         job_state_transition(job, status);
 511     }
 512
 513     if (job->driver->resume) {
 514         job->driver->resume(job);
 515     }
 516 }
 517
 518 void job_yield(Job *job)
 519 {
 520     assert(job->busy);
 521
 522     /* Check cancellation *before* setting busy = false, too!  */
 523     if (job_is_cancelled(job)) {
 524         return;
 525     }
 526
 527     if (!job_should_pause(job)) {
 528         job_do_yield(job, -1);
 529     }
 530
 531     job_pause_point(job);
 532 }
 533
 534 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
 535 {
 536     assert(job->busy);
 537
 538     /* Check cancellation *before* setting busy = false, too!  */
 539     if (job_is_cancelled(job)) {
 540         return;
 541     }
 542
 543     if (!job_should_pause(job)) {
 544         job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
 545     }
 546
 547     job_pause_point(job);
 548 }
 549
 550 /* Assumes the block_job_mutex is held */
 551 static bool job_timer_not_pending(Job *job)
 552 {
 553     return !timer_pending(&job->sleep_timer);
 554 }
 555
 556 void job_pause(Job *job)
 557 {
 558     job->pause_count++;
 559     if (!job->paused) {
 560         job_enter(job);
 561     }
 562 }
 563
 564 void job_resume(Job *job)
 565 {
 566     assert(job->pause_count > 0);
 567     job->pause_count--;
 568     if (job->pause_count) {
 569         return;
 570     }
 571
 572     /* kick only if no timer is pending */
 573     job_enter_cond(job, job_timer_not_pending);
 574 }
 575
 576 void job_user_pause(Job *job, Error **errp)
 577 {
 578     if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
 579         return;
 580     }
 581     if (job->user_paused) {
 582         error_setg(errp, "Job is already paused");
 583         return;
 584     }
 585     job->user_paused = true;
 586     job_pause(job);
 587 }
 588
 589 bool job_user_paused(Job *job)
 590 {
 591     return job->user_paused;
 592 }
 593
 594 void job_user_resume(Job *job, Error **errp)
 595 {
 596     assert(job);
 597     if (!job->user_paused || job->pause_count <= 0) {
 598         error_setg(errp, "Can't resume a job that was not paused");
 599         return;
 600     }
 601     if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
 602         return;
 603     }
 604     if (job->driver->user_resume) {
 605         job->driver->user_resume(job);
 606     }
 607     job->user_paused = false;
 608     job_resume(job);
 609 }
 610
 611 static void job_do_dismiss(Job *job)
 612 {
 613     assert(job);
 614     job->busy = false;
 615     job->paused = false;
 616     job->deferred_to_main_loop = true;
 617
 618     job_txn_del_job(job);
 619
 620     job_state_transition(job, JOB_STATUS_NULL);
 621     job_unref(job);
 622 }
 623
 624 void job_dismiss(Job **jobptr, Error **errp)
 625 {
 626     Job *job = *jobptr;
 627     /* similarly to _complete, this is QMP-interface only. */
 628     assert(job->id);
 629     if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
 630         return;
 631     }
 632
 633     job_do_dismiss(job);
 634     *jobptr = NULL;
 635 }
 636
 637 void job_early_fail(Job *job)
 638 {
 639     assert(job->status == JOB_STATUS_CREATED);
 640     job_do_dismiss(job);
 641 }
 642
 643 static void job_conclude(Job *job)
 644 {
 645     job_state_transition(job, JOB_STATUS_CONCLUDED);
 646     if (job->auto_dismiss || !job_started(job)) {
 647         job_do_dismiss(job);
 648     }
 649 }
 650
 651 static void job_update_rc(Job *job)
 652 {
 653     if (!job->ret && job_is_cancelled(job)) {
 654         job->ret = -ECANCELED;
 655     }
 656     if (job->ret) {
 657         if (!job->err) {
 658             error_setg(&job->err, "%s", strerror(-job->ret));
 659         }
 660         job_state_transition(job, JOB_STATUS_ABORTING);
 661     }
 662 }
 663
 664 static void job_commit(Job *job)
 665 {
 666     assert(!job->ret);
 667     if (job->driver->commit) {
 668         job->driver->commit(job);
 669     }
 670 }
 671
 672 static void job_abort(Job *job)
 673 {
 674     assert(job->ret);
 675     if (job->driver->abort) {
 676         job->driver->abort(job);
 677     }
 678 }
 679
 680 static void job_clean(Job *job)
 681 {
 682     if (job->driver->clean) {
 683         job->driver->clean(job);
 684     }
 685 }
 686
 687 static int job_finalize_single(Job *job)
 688 {
 689     assert(job_is_completed(job));
 690
 691     /* Ensure abort is called for late-transactional failures */
 692     job_update_rc(job);
 693
 694     if (!job->ret) {
 695         job_commit(job);
 696     } else {
 697         job_abort(job);
 698     }
 699     job_clean(job);
 700
 701     if (job->cb) {
 702         job->cb(job->opaque, job->ret);
 703     }
 704
 705     /* Emit events only if we actually started */
 706     if (job_started(job)) {
 707         if (job_is_cancelled(job)) {
 708             job_event_cancelled(job);
 709         } else {
 710             job_event_completed(job);
 711         }
 712     }
 713
 714     job_txn_del_job(job);
 715     job_conclude(job);
 716     return 0;
 717 }
 718
 719 static void job_cancel_async(Job *job, bool force)
 720 {
 721     if (job->driver->cancel) {
 722         job->driver->cancel(job, force);
 723     }
 724     if (job->user_paused) {
 725         /* Do not call job_enter here, the caller will handle it.  */
 726         if (job->driver->user_resume) {
 727             job->driver->user_resume(job);
 728         }
 729         job->user_paused = false;
 730         assert(job->pause_count > 0);
 731         job->pause_count--;
 732     }
 733     job->cancelled = true;
 734     /* To prevent 'force == false' overriding a previous 'force == true' */
 735     job->force_cancel |= force;
 736 }
 737
 738 static void job_completed_txn_abort(Job *job)
 739 {
 740     AioContext *outer_ctx = job->aio_context;
 741     AioContext *ctx;
 742     JobTxn *txn = job->txn;
 743     Job *other_job;
 744
 745     if (txn->aborting) {
 746         /*
 747          * We are cancelled by another job, which will handle everything.
 748          */
 749         return;
 750     }
 751     txn->aborting = true;
 752     job_txn_ref(txn);
 753
 754     /* We can only hold the single job's AioContext lock while calling
 755      * job_finalize_single() because the finalization callbacks can involve
 756      * calls of AIO_WAIT_WHILE(), which could deadlock otherwise. */
 757     aio_context_release(outer_ctx);
 758
 759     /* Other jobs are effectively cancelled by us, set the status for
 760      * them; this job, however, may or may not be cancelled, depending
 761      * on the caller, so leave it. */
 762     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 763         if (other_job != job) {
 764             ctx = other_job->aio_context;
 765             aio_context_acquire(ctx);
 766             job_cancel_async(other_job, false);
 767             aio_context_release(ctx);
 768         }
 769     }
 770     while (!QLIST_EMPTY(&txn->jobs)) {
 771         other_job = QLIST_FIRST(&txn->jobs);
 772         ctx = other_job->aio_context;
 773         aio_context_acquire(ctx);
 774         if (!job_is_completed(other_job)) {
 775             assert(job_is_cancelled(other_job));
 776             job_finish_sync(other_job, NULL, NULL);
 777         }
 778         job_finalize_single(other_job);
 779         aio_context_release(ctx);
 780     }
 781
 782     aio_context_acquire(outer_ctx);
 783
 784     job_txn_unref(txn);
 785 }
 786
 787 static int job_prepare(Job *job)
 788 {
 789     if (job->ret == 0 && job->driver->prepare) {
 790         job->ret = job->driver->prepare(job);
 791         job_update_rc(job);
 792     }
 793     return job->ret;
 794 }
 795
 796 static int job_needs_finalize(Job *job)
 797 {
 798     return !job->auto_finalize;
 799 }
 800
 801 static void job_do_finalize(Job *job)
 802 {
 803     int rc;
 804     assert(job && job->txn);
 805
 806     /* prepare the transaction to complete */
 807     rc = job_txn_apply(job, job_prepare);
 808     if (rc) {
 809         job_completed_txn_abort(job);
 810     } else {
 811         job_txn_apply(job, job_finalize_single);
 812     }
 813 }
 814
 815 void job_finalize(Job *job, Error **errp)
 816 {
 817     assert(job && job->id);
 818     if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
 819         return;
 820     }
 821     job_do_finalize(job);
 822 }
 823
 824 static int job_transition_to_pending(Job *job)
 825 {
 826     job_state_transition(job, JOB_STATUS_PENDING);
 827     if (!job->auto_finalize) {
 828         job_event_pending(job);
 829     }
 830     return 0;
 831 }
 832
 833 void job_transition_to_ready(Job *job)
 834 {
 835     job_state_transition(job, JOB_STATUS_READY);
 836     job_event_ready(job);
 837 }
 838
 839 static void job_completed_txn_success(Job *job)
 840 {
 841     JobTxn *txn = job->txn;
 842     Job *other_job;
 843
 844     job_state_transition(job, JOB_STATUS_WAITING);
 845
 846     /*
 847      * Successful completion, see if there are other running jobs in this
 848      * txn.
 849      */
 850     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 851         if (!job_is_completed(other_job)) {
 852             return;
 853         }
 854         assert(other_job->ret == 0);
 855     }
 856
 857     job_txn_apply(job, job_transition_to_pending);
 858
 859     /* If no jobs need manual finalization, automatically do so */
 860     if (job_txn_apply(job, job_needs_finalize) == 0) {
 861         job_do_finalize(job);
 862     }
 863 }
 864
 865 static void job_completed(Job *job)
 866 {
 867     assert(job && job->txn && !job_is_completed(job));
 868
 869     job_update_rc(job);
 870     trace_job_completed(job, job->ret);
 871     if (job->ret) {
 872         job_completed_txn_abort(job);
 873     } else {
 874         job_completed_txn_success(job);
 875     }
 876 }
 877
 878 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
 879 static void job_exit(void *opaque)
 880 {
 881     Job *job = (Job *)opaque;
 882     AioContext *ctx;
 883
 884     job_ref(job);
 885     aio_context_acquire(job->aio_context);
 886
 887     /* This is a lie, we're not quiescent, but still doing the completion
 888      * callbacks. However, completion callbacks tend to involve operations that
 889      * drain block nodes, and if .drained_poll still returned true, we would
 890      * deadlock. */
 891     job->busy = false;
 892     job_event_idle(job);
 893
 894     job_completed(job);
 895
 896     /*
 897      * Note that calling job_completed can move the job to a different
 898      * aio_context, so we cannot cache from above. job_txn_apply takes care of
 899      * acquiring the new lock, and we ref/unref to avoid job_completed freeing
 900      * the job underneath us.
 901      */
 902     ctx = job->aio_context;
 903     job_unref(job);
 904     aio_context_release(ctx);
 905 }
 906
 907 /**
 908  * All jobs must allow a pause point before entering their job proper. This
 909  * ensures that jobs can be paused prior to being started, then resumed later.
 910  */
 911 static void coroutine_fn job_co_entry(void *opaque)
 912 {
 913     Job *job = opaque;
 914
 915     assert(job && job->driver && job->driver->run);
 916     job_pause_point(job);
 917     job->ret = job->driver->run(job, &job->err);
 918     job->deferred_to_main_loop = true;
 919     job->busy = true;
 920     aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
 921 }
 922
 923 void job_start(Job *job)
 924 {
 925     assert(job && !job_started(job) && job->paused &&
 926            job->driver && job->driver->run);
 927     job->co = qemu_coroutine_create(job_co_entry, job);
 928     job->pause_count--;
 929     job->busy = true;
 930     job->paused = false;
 931     job_state_transition(job, JOB_STATUS_RUNNING);
 932     aio_co_enter(job->aio_context, job->co);
 933 }
 934
 935 void job_cancel(Job *job, bool force)
 936 {
 937     if (job->status == JOB_STATUS_CONCLUDED) {
 938         job_do_dismiss(job);
 939         return;
 940     }
 941     job_cancel_async(job, force);
 942     if (!job_started(job)) {
 943         job_completed(job);
 944     } else if (job->deferred_to_main_loop) {
 945         job_completed_txn_abort(job);
 946     } else {
 947         job_enter(job);
 948     }
 949 }
 950
 951 void job_user_cancel(Job *job, bool force, Error **errp)
 952 {
 953     if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
 954         return;
 955     }
 956     job_cancel(job, force);
 957 }
 958
 959 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
 960  * used with job_finish_sync() without the need for (rather nasty) function
 961  * pointer casts there. */
 962 static void job_cancel_err(Job *job, Error **errp)
 963 {
 964     job_cancel(job, false);
 965 }
 966
 967 int job_cancel_sync(Job *job)
 968 {
 969     return job_finish_sync(job, &job_cancel_err, NULL);
 970 }
 971
 972 void job_cancel_sync_all(void)
 973 {
 974     Job *job;
 975     AioContext *aio_context;
 976
 977     while ((job = job_next(NULL))) {
 978         aio_context = job->aio_context;
 979         aio_context_acquire(aio_context);
 980         job_cancel_sync(job);
 981         aio_context_release(aio_context);
 982     }
 983 }
 984
 985 int job_complete_sync(Job *job, Error **errp)
 986 {
 987     return job_finish_sync(job, job_complete, errp);
 988 }
 989
 990 void job_complete(Job *job, Error **errp)
 991 {
 992     /* Should not be reachable via external interface for internal jobs */
 993     assert(job->id);
 994     if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
 995         return;
 996     }
 997     if (job_is_cancelled(job) || !job->driver->complete) {
 998         error_setg(errp, "The active block job '%s' cannot be completed",
 999                    job->id);
1000         return;
1001     }
1002
1003     job->driver->complete(job, errp);
1004 }
1005
1006 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
1007 {
1008     Error *local_err = NULL;
1009     int ret;
1010
1011     job_ref(job);
1012
1013     if (finish) {
1014         finish(job, &local_err);
1015     }
1016     if (local_err) {
1017         error_propagate(errp, local_err);
1018         job_unref(job);
1019         return -EBUSY;
1020     }
1021
1022     AIO_WAIT_WHILE(job->aio_context,
1023                    (job_enter(job), !job_is_completed(job)));
1024
1025     ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
1026     job_unref(job);
1027     return ret;
1028 }