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