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     /* force_cancel may be true only if cancelled is true, too */
 221     assert(job->cancelled || !job->force_cancel);
 222     return job->force_cancel;
 223 }
 224
 225 bool job_cancel_requested(Job *job)
 226 {
 227     return job->cancelled;
 228 }
 229
 230 bool job_is_ready(Job *job)
 231 {
 232     switch (job->status) {
 233     case JOB_STATUS_UNDEFINED:
 234     case JOB_STATUS_CREATED:
 235     case JOB_STATUS_RUNNING:
 236     case JOB_STATUS_PAUSED:
 237     case JOB_STATUS_WAITING:
 238     case JOB_STATUS_PENDING:
 239     case JOB_STATUS_ABORTING:
 240     case JOB_STATUS_CONCLUDED:
 241     case JOB_STATUS_NULL:
 242         return false;
 243     case JOB_STATUS_READY:
 244     case JOB_STATUS_STANDBY:
 245         return true;
 246     default:
 247         g_assert_not_reached();
 248     }
 249     return false;
 250 }
 251
 252 bool job_is_completed(Job *job)
 253 {
 254     switch (job->status) {
 255     case JOB_STATUS_UNDEFINED:
 256     case JOB_STATUS_CREATED:
 257     case JOB_STATUS_RUNNING:
 258     case JOB_STATUS_PAUSED:
 259     case JOB_STATUS_READY:
 260     case JOB_STATUS_STANDBY:
 261         return false;
 262     case JOB_STATUS_WAITING:
 263     case JOB_STATUS_PENDING:
 264     case JOB_STATUS_ABORTING:
 265     case JOB_STATUS_CONCLUDED:
 266     case JOB_STATUS_NULL:
 267         return true;
 268     default:
 269         g_assert_not_reached();
 270     }
 271     return false;
 272 }
 273
 274 static bool job_started(Job *job)
 275 {
 276     return job->co;
 277 }
 278
 279 static bool job_should_pause(Job *job)
 280 {
 281     return job->pause_count > 0;
 282 }
 283
 284 Job *job_next(Job *job)
 285 {
 286     if (!job) {
 287         return QLIST_FIRST(&jobs);
 288     }
 289     return QLIST_NEXT(job, job_list);
 290 }
 291
 292 Job *job_get(const char *id)
 293 {
 294     Job *job;
 295
 296     QLIST_FOREACH(job, &jobs, job_list) {
 297         if (job->id && !strcmp(id, job->id)) {
 298             return job;
 299         }
 300     }
 301
 302     return NULL;
 303 }
 304
 305 static void job_sleep_timer_cb(void *opaque)
 306 {
 307     Job *job = opaque;
 308
 309     job_enter(job);
 310 }
 311
 312 void *job_create(const char *job_id, const JobDriver *driver, JobTxn *txn,
 313                  AioContext *ctx, int flags, BlockCompletionFunc *cb,
 314                  void *opaque, Error **errp)
 315 {
 316     Job *job;
 317
 318     if (job_id) {
 319         if (flags & JOB_INTERNAL) {
 320             error_setg(errp, "Cannot specify job ID for internal job");
 321             return NULL;
 322         }
 323         if (!id_wellformed(job_id)) {
 324             error_setg(errp, "Invalid job ID '%s'", job_id);
 325             return NULL;
 326         }
 327         if (job_get(job_id)) {
 328             error_setg(errp, "Job ID '%s' already in use", job_id);
 329             return NULL;
 330         }
 331     } else if (!(flags & JOB_INTERNAL)) {
 332         error_setg(errp, "An explicit job ID is required");
 333         return NULL;
 334     }
 335
 336     job = g_malloc0(driver->instance_size);
 337     job->driver        = driver;
 338     job->id            = g_strdup(job_id);
 339     job->refcnt        = 1;
 340     job->aio_context   = ctx;
 341     job->busy          = false;
 342     job->paused        = true;
 343     job->pause_count   = 1;
 344     job->auto_finalize = !(flags & JOB_MANUAL_FINALIZE);
 345     job->auto_dismiss  = !(flags & JOB_MANUAL_DISMISS);
 346     job->cb            = cb;
 347     job->opaque        = opaque;
 348
 349     progress_init(&job->progress);
 350
 351     notifier_list_init(&job->on_finalize_cancelled);
 352     notifier_list_init(&job->on_finalize_completed);
 353     notifier_list_init(&job->on_pending);
 354     notifier_list_init(&job->on_ready);
 355
 356     job_state_transition(job, JOB_STATUS_CREATED);
 357     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
 358                    QEMU_CLOCK_REALTIME, SCALE_NS,
 359                    job_sleep_timer_cb, job);
 360
 361     QLIST_INSERT_HEAD(&jobs, job, job_list);
 362
 363     /* Single jobs are modeled as single-job transactions for sake of
 364      * consolidating the job management logic */
 365     if (!txn) {
 366         txn = job_txn_new();
 367         job_txn_add_job(txn, job);
 368         job_txn_unref(txn);
 369     } else {
 370         job_txn_add_job(txn, job);
 371     }
 372
 373     return job;
 374 }
 375
 376 void job_ref(Job *job)
 377 {
 378     ++job->refcnt;
 379 }
 380
 381 void job_unref(Job *job)
 382 {
 383     if (--job->refcnt == 0) {
 384         assert(job->status == JOB_STATUS_NULL);
 385         assert(!timer_pending(&job->sleep_timer));
 386         assert(!job->txn);
 387
 388         if (job->driver->free) {
 389             job->driver->free(job);
 390         }
 391
 392         QLIST_REMOVE(job, job_list);
 393
 394         progress_destroy(&job->progress);
 395         error_free(job->err);
 396         g_free(job->id);
 397         g_free(job);
 398     }
 399 }
 400
 401 void job_progress_update(Job *job, uint64_t done)
 402 {
 403     progress_work_done(&job->progress, done);
 404 }
 405
 406 void job_progress_set_remaining(Job *job, uint64_t remaining)
 407 {
 408     progress_set_remaining(&job->progress, remaining);
 409 }
 410
 411 void job_progress_increase_remaining(Job *job, uint64_t delta)
 412 {
 413     progress_increase_remaining(&job->progress, delta);
 414 }
 415
 416 void job_event_cancelled(Job *job)
 417 {
 418     notifier_list_notify(&job->on_finalize_cancelled, job);
 419 }
 420
 421 void job_event_completed(Job *job)
 422 {
 423     notifier_list_notify(&job->on_finalize_completed, job);
 424 }
 425
 426 static void job_event_pending(Job *job)
 427 {
 428     notifier_list_notify(&job->on_pending, job);
 429 }
 430
 431 static void job_event_ready(Job *job)
 432 {
 433     notifier_list_notify(&job->on_ready, job);
 434 }
 435
 436 static void job_event_idle(Job *job)
 437 {
 438     notifier_list_notify(&job->on_idle, job);
 439 }
 440
 441 void job_enter_cond(Job *job, bool(*fn)(Job *job))
 442 {
 443     if (!job_started(job)) {
 444         return;
 445     }
 446     if (job->deferred_to_main_loop) {
 447         return;
 448     }
 449
 450     job_lock();
 451     if (job->busy) {
 452         job_unlock();
 453         return;
 454     }
 455
 456     if (fn && !fn(job)) {
 457         job_unlock();
 458         return;
 459     }
 460
 461     assert(!job->deferred_to_main_loop);
 462     timer_del(&job->sleep_timer);
 463     job->busy = true;
 464     job_unlock();
 465     aio_co_enter(job->aio_context, job->co);
 466 }
 467
 468 void job_enter(Job *job)
 469 {
 470     job_enter_cond(job, NULL);
 471 }
 472
 473 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
 474  * Reentering the job coroutine with job_enter() before the timer has expired
 475  * is allowed and cancels the timer.
 476  *
 477  * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
 478  * called explicitly. */
 479 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
 480 {
 481     job_lock();
 482     if (ns != -1) {
 483         timer_mod(&job->sleep_timer, ns);
 484     }
 485     job->busy = false;
 486     job_event_idle(job);
 487     job_unlock();
 488     qemu_coroutine_yield();
 489
 490     /* Set by job_enter_cond() before re-entering the coroutine.  */
 491     assert(job->busy);
 492 }
 493
 494 void coroutine_fn job_pause_point(Job *job)
 495 {
 496     assert(job && job_started(job));
 497
 498     if (!job_should_pause(job)) {
 499         return;
 500     }
 501     if (job_is_cancelled(job)) {
 502         return;
 503     }
 504
 505     if (job->driver->pause) {
 506         job->driver->pause(job);
 507     }
 508
 509     if (job_should_pause(job) && !job_is_cancelled(job)) {
 510         JobStatus status = job->status;
 511         job_state_transition(job, status == JOB_STATUS_READY
 512                                   ? JOB_STATUS_STANDBY
 513                                   : JOB_STATUS_PAUSED);
 514         job->paused = true;
 515         job_do_yield(job, -1);
 516         job->paused = false;
 517         job_state_transition(job, status);
 518     }
 519
 520     if (job->driver->resume) {
 521         job->driver->resume(job);
 522     }
 523 }
 524
 525 void job_yield(Job *job)
 526 {
 527     assert(job->busy);
 528
 529     /* Check cancellation *before* setting busy = false, too!  */
 530     if (job_is_cancelled(job)) {
 531         return;
 532     }
 533
 534     if (!job_should_pause(job)) {
 535         job_do_yield(job, -1);
 536     }
 537
 538     job_pause_point(job);
 539 }
 540
 541 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
 542 {
 543     assert(job->busy);
 544
 545     /* Check cancellation *before* setting busy = false, too!  */
 546     if (job_is_cancelled(job)) {
 547         return;
 548     }
 549
 550     if (!job_should_pause(job)) {
 551         job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
 552     }
 553
 554     job_pause_point(job);
 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     if (!job->paused) {
 567         job_enter(job);
 568     }
 569 }
 570
 571 void job_resume(Job *job)
 572 {
 573     assert(job->pause_count > 0);
 574     job->pause_count--;
 575     if (job->pause_count) {
 576         return;
 577     }
 578
 579     /* kick only if no timer is pending */
 580     job_enter_cond(job, job_timer_not_pending);
 581 }
 582
 583 void job_user_pause(Job *job, Error **errp)
 584 {
 585     if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
 586         return;
 587     }
 588     if (job->user_paused) {
 589         error_setg(errp, "Job is already paused");
 590         return;
 591     }
 592     job->user_paused = true;
 593     job_pause(job);
 594 }
 595
 596 bool job_user_paused(Job *job)
 597 {
 598     return job->user_paused;
 599 }
 600
 601 void job_user_resume(Job *job, Error **errp)
 602 {
 603     assert(job);
 604     if (!job->user_paused || job->pause_count <= 0) {
 605         error_setg(errp, "Can't resume a job that was not paused");
 606         return;
 607     }
 608     if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
 609         return;
 610     }
 611     if (job->driver->user_resume) {
 612         job->driver->user_resume(job);
 613     }
 614     job->user_paused = false;
 615     job_resume(job);
 616 }
 617
 618 static void job_do_dismiss(Job *job)
 619 {
 620     assert(job);
 621     job->busy = false;
 622     job->paused = false;
 623     job->deferred_to_main_loop = true;
 624
 625     job_txn_del_job(job);
 626
 627     job_state_transition(job, JOB_STATUS_NULL);
 628     job_unref(job);
 629 }
 630
 631 void job_dismiss(Job **jobptr, Error **errp)
 632 {
 633     Job *job = *jobptr;
 634     /* similarly to _complete, this is QMP-interface only. */
 635     assert(job->id);
 636     if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
 637         return;
 638     }
 639
 640     job_do_dismiss(job);
 641     *jobptr = NULL;
 642 }
 643
 644 void job_early_fail(Job *job)
 645 {
 646     assert(job->status == JOB_STATUS_CREATED);
 647     job_do_dismiss(job);
 648 }
 649
 650 static void job_conclude(Job *job)
 651 {
 652     job_state_transition(job, JOB_STATUS_CONCLUDED);
 653     if (job->auto_dismiss || !job_started(job)) {
 654         job_do_dismiss(job);
 655     }
 656 }
 657
 658 static void job_update_rc(Job *job)
 659 {
 660     if (!job->ret && job_is_cancelled(job)) {
 661         job->ret = -ECANCELED;
 662     }
 663     if (job->ret) {
 664         if (!job->err) {
 665             error_setg(&job->err, "%s", strerror(-job->ret));
 666         }
 667         job_state_transition(job, JOB_STATUS_ABORTING);
 668     }
 669 }
 670
 671 static void job_commit(Job *job)
 672 {
 673     assert(!job->ret);
 674     if (job->driver->commit) {
 675         job->driver->commit(job);
 676     }
 677 }
 678
 679 static void job_abort(Job *job)
 680 {
 681     assert(job->ret);
 682     if (job->driver->abort) {
 683         job->driver->abort(job);
 684     }
 685 }
 686
 687 static void job_clean(Job *job)
 688 {
 689     if (job->driver->clean) {
 690         job->driver->clean(job);
 691     }
 692 }
 693
 694 static int job_finalize_single(Job *job)
 695 {
 696     assert(job_is_completed(job));
 697
 698     /* Ensure abort is called for late-transactional failures */
 699     job_update_rc(job);
 700
 701     if (!job->ret) {
 702         job_commit(job);
 703     } else {
 704         job_abort(job);
 705     }
 706     job_clean(job);
 707
 708     if (job->cb) {
 709         job->cb(job->opaque, job->ret);
 710     }
 711
 712     /* Emit events only if we actually started */
 713     if (job_started(job)) {
 714         if (job_is_cancelled(job)) {
 715             job_event_cancelled(job);
 716         } else {
 717             job_event_completed(job);
 718         }
 719     }
 720
 721     job_txn_del_job(job);
 722     job_conclude(job);
 723     return 0;
 724 }
 725
 726 static void job_cancel_async(Job *job, bool force)
 727 {
 728     if (job->driver->cancel) {
 729         force = job->driver->cancel(job, force);
 730     } else {
 731         /* No .cancel() means the job will behave as if force-cancelled */
 732         force = true;
 733     }
 734
 735     if (job->user_paused) {
 736         /* Do not call job_enter here, the caller will handle it.  */
 737         if (job->driver->user_resume) {
 738             job->driver->user_resume(job);
 739         }
 740         job->user_paused = false;
 741         assert(job->pause_count > 0);
 742         job->pause_count--;
 743     }
 744
 745     /*
 746      * Ignore soft cancel requests after the job is already done
 747      * (We will still invoke job->driver->cancel() above, but if the
 748      * job driver supports soft cancelling and the job is done, that
 749      * should be a no-op, too.  We still call it so it can override
 750      * @force.)
 751      */
 752     if (force || !job->deferred_to_main_loop) {
 753         job->cancelled = true;
 754         /* To prevent 'force == false' overriding a previous 'force == true' */
 755         job->force_cancel |= force;
 756     }
 757 }
 758
 759 static void job_completed_txn_abort(Job *job)
 760 {
 761     AioContext *ctx;
 762     JobTxn *txn = job->txn;
 763     Job *other_job;
 764
 765     if (txn->aborting) {
 766         /*
 767          * We are cancelled by another job, which will handle everything.
 768          */
 769         return;
 770     }
 771     txn->aborting = true;
 772     job_txn_ref(txn);
 773
 774     /*
 775      * We can only hold the single job's AioContext lock while calling
 776      * job_finalize_single() because the finalization callbacks can involve
 777      * calls of AIO_WAIT_WHILE(), which could deadlock otherwise.
 778      * Note that the job's AioContext may change when it is finalized.
 779      */
 780     job_ref(job);
 781     aio_context_release(job->aio_context);
 782
 783     /* Other jobs are effectively cancelled by us, set the status for
 784      * them; this job, however, may or may not be cancelled, depending
 785      * on the caller, so leave it. */
 786     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 787         if (other_job != job) {
 788             ctx = other_job->aio_context;
 789             aio_context_acquire(ctx);
 790             /*
 791              * This is a transaction: If one job failed, no result will matter.
 792              * Therefore, pass force=true to terminate all other jobs as quickly
 793              * as possible.
 794              */
 795             job_cancel_async(other_job, true);
 796             aio_context_release(ctx);
 797         }
 798     }
 799     while (!QLIST_EMPTY(&txn->jobs)) {
 800         other_job = QLIST_FIRST(&txn->jobs);
 801         /*
 802          * The job's AioContext may change, so store it in @ctx so we
 803          * release the same context that we have acquired before.
 804          */
 805         ctx = other_job->aio_context;
 806         aio_context_acquire(ctx);
 807         if (!job_is_completed(other_job)) {
 808             assert(job_cancel_requested(other_job));
 809             job_finish_sync(other_job, NULL, NULL);
 810         }
 811         job_finalize_single(other_job);
 812         aio_context_release(ctx);
 813     }
 814
 815     /*
 816      * Use job_ref()/job_unref() so we can read the AioContext here
 817      * even if the job went away during job_finalize_single().
 818      */
 819     aio_context_acquire(job->aio_context);
 820     job_unref(job);
 821
 822     job_txn_unref(txn);
 823 }
 824
 825 static int job_prepare(Job *job)
 826 {
 827     if (job->ret == 0 && job->driver->prepare) {
 828         job->ret = job->driver->prepare(job);
 829         job_update_rc(job);
 830     }
 831     return job->ret;
 832 }
 833
 834 static int job_needs_finalize(Job *job)
 835 {
 836     return !job->auto_finalize;
 837 }
 838
 839 static void job_do_finalize(Job *job)
 840 {
 841     int rc;
 842     assert(job && job->txn);
 843
 844     /* prepare the transaction to complete */
 845     rc = job_txn_apply(job, job_prepare);
 846     if (rc) {
 847         job_completed_txn_abort(job);
 848     } else {
 849         job_txn_apply(job, job_finalize_single);
 850     }
 851 }
 852
 853 void job_finalize(Job *job, Error **errp)
 854 {
 855     assert(job && job->id);
 856     if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
 857         return;
 858     }
 859     job_do_finalize(job);
 860 }
 861
 862 static int job_transition_to_pending(Job *job)
 863 {
 864     job_state_transition(job, JOB_STATUS_PENDING);
 865     if (!job->auto_finalize) {
 866         job_event_pending(job);
 867     }
 868     return 0;
 869 }
 870
 871 void job_transition_to_ready(Job *job)
 872 {
 873     job_state_transition(job, JOB_STATUS_READY);
 874     job_event_ready(job);
 875 }
 876
 877 static void job_completed_txn_success(Job *job)
 878 {
 879     JobTxn *txn = job->txn;
 880     Job *other_job;
 881
 882     job_state_transition(job, JOB_STATUS_WAITING);
 883
 884     /*
 885      * Successful completion, see if there are other running jobs in this
 886      * txn.
 887      */
 888     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 889         if (!job_is_completed(other_job)) {
 890             return;
 891         }
 892         assert(other_job->ret == 0);
 893     }
 894
 895     job_txn_apply(job, job_transition_to_pending);
 896
 897     /* If no jobs need manual finalization, automatically do so */
 898     if (job_txn_apply(job, job_needs_finalize) == 0) {
 899         job_do_finalize(job);
 900     }
 901 }
 902
 903 static void job_completed(Job *job)
 904 {
 905     assert(job && job->txn && !job_is_completed(job));
 906
 907     job_update_rc(job);
 908     trace_job_completed(job, job->ret);
 909     if (job->ret) {
 910         job_completed_txn_abort(job);
 911     } else {
 912         job_completed_txn_success(job);
 913     }
 914 }
 915
 916 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
 917 static void job_exit(void *opaque)
 918 {
 919     Job *job = (Job *)opaque;
 920     AioContext *ctx;
 921
 922     job_ref(job);
 923     aio_context_acquire(job->aio_context);
 924
 925     /* This is a lie, we're not quiescent, but still doing the completion
 926      * callbacks. However, completion callbacks tend to involve operations that
 927      * drain block nodes, and if .drained_poll still returned true, we would
 928      * deadlock. */
 929     job->busy = false;
 930     job_event_idle(job);
 931
 932     job_completed(job);
 933
 934     /*
 935      * Note that calling job_completed can move the job to a different
 936      * aio_context, so we cannot cache from above. job_txn_apply takes care of
 937      * acquiring the new lock, and we ref/unref to avoid job_completed freeing
 938      * the job underneath us.
 939      */
 940     ctx = job->aio_context;
 941     job_unref(job);
 942     aio_context_release(ctx);
 943 }
 944
 945 /**
 946  * All jobs must allow a pause point before entering their job proper. This
 947  * ensures that jobs can be paused prior to being started, then resumed later.
 948  */
 949 static void coroutine_fn job_co_entry(void *opaque)
 950 {
 951     Job *job = opaque;
 952
 953     assert(job && job->driver && job->driver->run);
 954     job_pause_point(job);
 955     job->ret = job->driver->run(job, &job->err);
 956     job->deferred_to_main_loop = true;
 957     job->busy = true;
 958     aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
 959 }
 960
 961 void job_start(Job *job)
 962 {
 963     assert(job && !job_started(job) && job->paused &&
 964            job->driver && job->driver->run);
 965     job->co = qemu_coroutine_create(job_co_entry, job);
 966     job->pause_count--;
 967     job->busy = true;
 968     job->paused = false;
 969     job_state_transition(job, JOB_STATUS_RUNNING);
 970     aio_co_enter(job->aio_context, job->co);
 971 }
 972
 973 void job_cancel(Job *job, bool force)
 974 {
 975     if (job->status == JOB_STATUS_CONCLUDED) {
 976         job_do_dismiss(job);
 977         return;
 978     }
 979     job_cancel_async(job, force);
 980     if (!job_started(job)) {
 981         job_completed(job);
 982     } else if (job->deferred_to_main_loop) {
 983         /*
 984          * job_cancel_async() ignores soft-cancel requests for jobs
 985          * that are already done (i.e. deferred to the main loop).  We
 986          * have to check again whether the job is really cancelled.
 987          * (job_cancel_requested() and job_is_cancelled() are equivalent
 988          * here, because job_cancel_async() will make soft-cancel
 989          * requests no-ops when deferred_to_main_loop is true.  We
 990          * choose to call job_is_cancelled() to show that we invoke
 991          * job_completed_txn_abort() only for force-cancelled jobs.)
 992          */
 993         if (job_is_cancelled(job)) {
 994             job_completed_txn_abort(job);
 995         }
 996     } else {
 997         job_enter(job);
 998     }
 999 }
1000
1001 void job_user_cancel(Job *job, bool force, Error **errp)
1002 {
1003     if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
1004         return;
1005     }
1006     job_cancel(job, force);
1007 }
1008
1009 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
1010  * used with job_finish_sync() without the need for (rather nasty) function
1011  * pointer casts there. */
1012 static void job_cancel_err(Job *job, Error **errp)
1013 {
1014     job_cancel(job, false);
1015 }
1016
1017 /**
1018  * Same as job_cancel_err(), but force-cancel.
1019  */
1020 static void job_force_cancel_err(Job *job, Error **errp)
1021 {
1022     job_cancel(job, true);
1023 }
1024
1025 int job_cancel_sync(Job *job, bool force)
1026 {
1027     if (force) {
1028         return job_finish_sync(job, &job_force_cancel_err, NULL);
1029     } else {
1030         return job_finish_sync(job, &job_cancel_err, NULL);
1031     }
1032 }
1033
1034 void job_cancel_sync_all(void)
1035 {
1036     Job *job;
1037     AioContext *aio_context;
1038
1039     while ((job = job_next(NULL))) {
1040         aio_context = job->aio_context;
1041         aio_context_acquire(aio_context);
1042         job_cancel_sync(job, true);
1043         aio_context_release(aio_context);
1044     }
1045 }
1046
1047 int job_complete_sync(Job *job, Error **errp)
1048 {
1049     return job_finish_sync(job, job_complete, errp);
1050 }
1051
1052 void job_complete(Job *job, Error **errp)
1053 {
1054     /* Should not be reachable via external interface for internal jobs */
1055     assert(job->id);
1056     if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
1057         return;
1058     }
1059     if (job_cancel_requested(job) || !job->driver->complete) {
1060         error_setg(errp, "The active block job '%s' cannot be completed",
1061                    job->id);
1062         return;
1063     }
1064
1065     job->driver->complete(job, errp);
1066 }
1067
1068 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
1069 {
1070     Error *local_err = NULL;
1071     int ret;
1072
1073     job_ref(job);
1074
1075     if (finish) {
1076         finish(job, &local_err);
1077     }
1078     if (local_err) {
1079         error_propagate(errp, local_err);
1080         job_unref(job);
1081         return -EBUSY;
1082     }
1083
1084     AIO_WAIT_WHILE(job->aio_context,
1085                    (job_enter(job), !job_is_completed(job)));
1086
1087     ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
1088     job_unref(job);
1089     return ret;
1090 }