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