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, 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(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     notifier_list_init(&job->on_finalize_cancelled);
 343     notifier_list_init(&job->on_finalize_completed);
 344     notifier_list_init(&job->on_pending);
 345     notifier_list_init(&job->on_ready);
 346
 347     job_state_transition(job, JOB_STATUS_CREATED);
 348     aio_timer_init(qemu_get_aio_context(), &job->sleep_timer,
 349                    QEMU_CLOCK_REALTIME, SCALE_NS,
 350                    job_sleep_timer_cb, job);
 351
 352     QLIST_INSERT_HEAD(&jobs, job, job_list);
 353
 354     /* Single jobs are modeled as single-job transactions for sake of
 355      * consolidating the job management logic */
 356     if (!txn) {
 357         txn = job_txn_new();
 358         job_txn_add_job(txn, job);
 359         job_txn_unref(txn);
 360     } else {
 361         job_txn_add_job(txn, job);
 362     }
 363
 364     return job;
 365 }
 366
 367 void job_ref(Job *job)
 368 {
 369     ++job->refcnt;
 370 }
 371
 372 void job_unref(Job *job)
 373 {
 374     if (--job->refcnt == 0) {
 375         assert(job->status == JOB_STATUS_NULL);
 376         assert(!timer_pending(&job->sleep_timer));
 377         assert(!job->txn);
 378
 379         if (job->driver->free) {
 380             job->driver->free(job);
 381         }
 382
 383         QLIST_REMOVE(job, job_list);
 384
 385         error_free(job->err);
 386         g_free(job->id);
 387         g_free(job);
 388     }
 389 }
 390
 391 void job_progress_update(Job *job, uint64_t done)
 392 {
 393     progress_work_done(&job->progress, done);
 394 }
 395
 396 void job_progress_set_remaining(Job *job, uint64_t remaining)
 397 {
 398     progress_set_remaining(&job->progress, remaining);
 399 }
 400
 401 void job_progress_increase_remaining(Job *job, uint64_t delta)
 402 {
 403     progress_increase_remaining(&job->progress, delta);
 404 }
 405
 406 void job_event_cancelled(Job *job)
 407 {
 408     notifier_list_notify(&job->on_finalize_cancelled, job);
 409 }
 410
 411 void job_event_completed(Job *job)
 412 {
 413     notifier_list_notify(&job->on_finalize_completed, job);
 414 }
 415
 416 static void job_event_pending(Job *job)
 417 {
 418     notifier_list_notify(&job->on_pending, job);
 419 }
 420
 421 static void job_event_ready(Job *job)
 422 {
 423     notifier_list_notify(&job->on_ready, job);
 424 }
 425
 426 static void job_event_idle(Job *job)
 427 {
 428     notifier_list_notify(&job->on_idle, job);
 429 }
 430
 431 void job_enter_cond(Job *job, bool(*fn)(Job *job))
 432 {
 433     if (!job_started(job)) {
 434         return;
 435     }
 436     if (job->deferred_to_main_loop) {
 437         return;
 438     }
 439
 440     job_lock();
 441     if (job->busy) {
 442         job_unlock();
 443         return;
 444     }
 445
 446     if (fn && !fn(job)) {
 447         job_unlock();
 448         return;
 449     }
 450
 451     assert(!job->deferred_to_main_loop);
 452     timer_del(&job->sleep_timer);
 453     job->busy = true;
 454     job_unlock();
 455     aio_co_enter(job->aio_context, job->co);
 456 }
 457
 458 void job_enter(Job *job)
 459 {
 460     job_enter_cond(job, NULL);
 461 }
 462
 463 /* Yield, and schedule a timer to reenter the coroutine after @ns nanoseconds.
 464  * Reentering the job coroutine with job_enter() before the timer has expired
 465  * is allowed and cancels the timer.
 466  *
 467  * If @ns is (uint64_t) -1, no timer is scheduled and job_enter() must be
 468  * called explicitly. */
 469 static void coroutine_fn job_do_yield(Job *job, uint64_t ns)
 470 {
 471     job_lock();
 472     if (ns != -1) {
 473         timer_mod(&job->sleep_timer, ns);
 474     }
 475     job->busy = false;
 476     job_event_idle(job);
 477     job_unlock();
 478     qemu_coroutine_yield();
 479
 480     /* Set by job_enter_cond() before re-entering the coroutine.  */
 481     assert(job->busy);
 482 }
 483
 484 void coroutine_fn job_pause_point(Job *job)
 485 {
 486     assert(job && job_started(job));
 487
 488     if (!job_should_pause(job)) {
 489         return;
 490     }
 491     if (job_is_cancelled(job)) {
 492         return;
 493     }
 494
 495     if (job->driver->pause) {
 496         job->driver->pause(job);
 497     }
 498
 499     if (job_should_pause(job) && !job_is_cancelled(job)) {
 500         JobStatus status = job->status;
 501         job_state_transition(job, status == JOB_STATUS_READY
 502                                   ? JOB_STATUS_STANDBY
 503                                   : JOB_STATUS_PAUSED);
 504         job->paused = true;
 505         job_do_yield(job, -1);
 506         job->paused = false;
 507         job_state_transition(job, status);
 508     }
 509
 510     if (job->driver->resume) {
 511         job->driver->resume(job);
 512     }
 513 }
 514
 515 void job_yield(Job *job)
 516 {
 517     assert(job->busy);
 518
 519     /* Check cancellation *before* setting busy = false, too!  */
 520     if (job_is_cancelled(job)) {
 521         return;
 522     }
 523
 524     if (!job_should_pause(job)) {
 525         job_do_yield(job, -1);
 526     }
 527
 528     job_pause_point(job);
 529 }
 530
 531 void coroutine_fn job_sleep_ns(Job *job, int64_t ns)
 532 {
 533     assert(job->busy);
 534
 535     /* Check cancellation *before* setting busy = false, too!  */
 536     if (job_is_cancelled(job)) {
 537         return;
 538     }
 539
 540     if (!job_should_pause(job)) {
 541         job_do_yield(job, qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + ns);
 542     }
 543
 544     job_pause_point(job);
 545 }
 546
 547 /* Assumes the block_job_mutex is held */
 548 static bool job_timer_not_pending(Job *job)
 549 {
 550     return !timer_pending(&job->sleep_timer);
 551 }
 552
 553 void job_pause(Job *job)
 554 {
 555     job->pause_count++;
 556     if (!job->paused) {
 557         job_enter(job);
 558     }
 559 }
 560
 561 void job_resume(Job *job)
 562 {
 563     assert(job->pause_count > 0);
 564     job->pause_count--;
 565     if (job->pause_count) {
 566         return;
 567     }
 568
 569     /* kick only if no timer is pending */
 570     job_enter_cond(job, job_timer_not_pending);
 571 }
 572
 573 void job_user_pause(Job *job, Error **errp)
 574 {
 575     if (job_apply_verb(job, JOB_VERB_PAUSE, errp)) {
 576         return;
 577     }
 578     if (job->user_paused) {
 579         error_setg(errp, "Job is already paused");
 580         return;
 581     }
 582     job->user_paused = true;
 583     job_pause(job);
 584 }
 585
 586 bool job_user_paused(Job *job)
 587 {
 588     return job->user_paused;
 589 }
 590
 591 void job_user_resume(Job *job, Error **errp)
 592 {
 593     assert(job);
 594     if (!job->user_paused || job->pause_count <= 0) {
 595         error_setg(errp, "Can't resume a job that was not paused");
 596         return;
 597     }
 598     if (job_apply_verb(job, JOB_VERB_RESUME, errp)) {
 599         return;
 600     }
 601     if (job->driver->user_resume) {
 602         job->driver->user_resume(job);
 603     }
 604     job->user_paused = false;
 605     job_resume(job);
 606 }
 607
 608 static void job_do_dismiss(Job *job)
 609 {
 610     assert(job);
 611     job->busy = false;
 612     job->paused = false;
 613     job->deferred_to_main_loop = true;
 614
 615     job_txn_del_job(job);
 616
 617     job_state_transition(job, JOB_STATUS_NULL);
 618     job_unref(job);
 619 }
 620
 621 void job_dismiss(Job **jobptr, Error **errp)
 622 {
 623     Job *job = *jobptr;
 624     /* similarly to _complete, this is QMP-interface only. */
 625     assert(job->id);
 626     if (job_apply_verb(job, JOB_VERB_DISMISS, errp)) {
 627         return;
 628     }
 629
 630     job_do_dismiss(job);
 631     *jobptr = NULL;
 632 }
 633
 634 void job_early_fail(Job *job)
 635 {
 636     assert(job->status == JOB_STATUS_CREATED);
 637     job_do_dismiss(job);
 638 }
 639
 640 static void job_conclude(Job *job)
 641 {
 642     job_state_transition(job, JOB_STATUS_CONCLUDED);
 643     if (job->auto_dismiss || !job_started(job)) {
 644         job_do_dismiss(job);
 645     }
 646 }
 647
 648 static void job_update_rc(Job *job)
 649 {
 650     if (!job->ret && job_is_cancelled(job)) {
 651         job->ret = -ECANCELED;
 652     }
 653     if (job->ret) {
 654         if (!job->err) {
 655             error_setg(&job->err, "%s", strerror(-job->ret));
 656         }
 657         job_state_transition(job, JOB_STATUS_ABORTING);
 658     }
 659 }
 660
 661 static void job_commit(Job *job)
 662 {
 663     assert(!job->ret);
 664     if (job->driver->commit) {
 665         job->driver->commit(job);
 666     }
 667 }
 668
 669 static void job_abort(Job *job)
 670 {
 671     assert(job->ret);
 672     if (job->driver->abort) {
 673         job->driver->abort(job);
 674     }
 675 }
 676
 677 static void job_clean(Job *job)
 678 {
 679     if (job->driver->clean) {
 680         job->driver->clean(job);
 681     }
 682 }
 683
 684 static int job_finalize_single(Job *job)
 685 {
 686     assert(job_is_completed(job));
 687
 688     /* Ensure abort is called for late-transactional failures */
 689     job_update_rc(job);
 690
 691     if (!job->ret) {
 692         job_commit(job);
 693     } else {
 694         job_abort(job);
 695     }
 696     job_clean(job);
 697
 698     if (job->cb) {
 699         job->cb(job->opaque, job->ret);
 700     }
 701
 702     /* Emit events only if we actually started */
 703     if (job_started(job)) {
 704         if (job_is_cancelled(job)) {
 705             job_event_cancelled(job);
 706         } else {
 707             job_event_completed(job);
 708         }
 709     }
 710
 711     job_txn_del_job(job);
 712     job_conclude(job);
 713     return 0;
 714 }
 715
 716 static void job_cancel_async(Job *job, bool force)
 717 {
 718     if (job->user_paused) {
 719         /* Do not call job_enter here, the caller will handle it.  */
 720         if (job->driver->user_resume) {
 721             job->driver->user_resume(job);
 722         }
 723         job->user_paused = false;
 724         assert(job->pause_count > 0);
 725         job->pause_count--;
 726     }
 727     job->cancelled = true;
 728     /* To prevent 'force == false' overriding a previous 'force == true' */
 729     job->force_cancel |= force;
 730 }
 731
 732 static void job_completed_txn_abort(Job *job)
 733 {
 734     AioContext *outer_ctx = job->aio_context;
 735     AioContext *ctx;
 736     JobTxn *txn = job->txn;
 737     Job *other_job;
 738
 739     if (txn->aborting) {
 740         /*
 741          * We are cancelled by another job, which will handle everything.
 742          */
 743         return;
 744     }
 745     txn->aborting = true;
 746     job_txn_ref(txn);
 747
 748     /* We can only hold the single job's AioContext lock while calling
 749      * job_finalize_single() because the finalization callbacks can involve
 750      * calls of AIO_WAIT_WHILE(), which could deadlock otherwise. */
 751     aio_context_release(outer_ctx);
 752
 753     /* Other jobs are effectively cancelled by us, set the status for
 754      * them; this job, however, may or may not be cancelled, depending
 755      * on the caller, so leave it. */
 756     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 757         if (other_job != job) {
 758             ctx = other_job->aio_context;
 759             aio_context_acquire(ctx);
 760             job_cancel_async(other_job, false);
 761             aio_context_release(ctx);
 762         }
 763     }
 764     while (!QLIST_EMPTY(&txn->jobs)) {
 765         other_job = QLIST_FIRST(&txn->jobs);
 766         ctx = other_job->aio_context;
 767         aio_context_acquire(ctx);
 768         if (!job_is_completed(other_job)) {
 769             assert(job_is_cancelled(other_job));
 770             job_finish_sync(other_job, NULL, NULL);
 771         }
 772         job_finalize_single(other_job);
 773         aio_context_release(ctx);
 774     }
 775
 776     aio_context_acquire(outer_ctx);
 777
 778     job_txn_unref(txn);
 779 }
 780
 781 static int job_prepare(Job *job)
 782 {
 783     if (job->ret == 0 && job->driver->prepare) {
 784         job->ret = job->driver->prepare(job);
 785         job_update_rc(job);
 786     }
 787     return job->ret;
 788 }
 789
 790 static int job_needs_finalize(Job *job)
 791 {
 792     return !job->auto_finalize;
 793 }
 794
 795 static void job_do_finalize(Job *job)
 796 {
 797     int rc;
 798     assert(job && job->txn);
 799
 800     /* prepare the transaction to complete */
 801     rc = job_txn_apply(job, job_prepare);
 802     if (rc) {
 803         job_completed_txn_abort(job);
 804     } else {
 805         job_txn_apply(job, job_finalize_single);
 806     }
 807 }
 808
 809 void job_finalize(Job *job, Error **errp)
 810 {
 811     assert(job && job->id);
 812     if (job_apply_verb(job, JOB_VERB_FINALIZE, errp)) {
 813         return;
 814     }
 815     job_do_finalize(job);
 816 }
 817
 818 static int job_transition_to_pending(Job *job)
 819 {
 820     job_state_transition(job, JOB_STATUS_PENDING);
 821     if (!job->auto_finalize) {
 822         job_event_pending(job);
 823     }
 824     return 0;
 825 }
 826
 827 void job_transition_to_ready(Job *job)
 828 {
 829     job_state_transition(job, JOB_STATUS_READY);
 830     job_event_ready(job);
 831 }
 832
 833 static void job_completed_txn_success(Job *job)
 834 {
 835     JobTxn *txn = job->txn;
 836     Job *other_job;
 837
 838     job_state_transition(job, JOB_STATUS_WAITING);
 839
 840     /*
 841      * Successful completion, see if there are other running jobs in this
 842      * txn.
 843      */
 844     QLIST_FOREACH(other_job, &txn->jobs, txn_list) {
 845         if (!job_is_completed(other_job)) {
 846             return;
 847         }
 848         assert(other_job->ret == 0);
 849     }
 850
 851     job_txn_apply(job, job_transition_to_pending);
 852
 853     /* If no jobs need manual finalization, automatically do so */
 854     if (job_txn_apply(job, job_needs_finalize) == 0) {
 855         job_do_finalize(job);
 856     }
 857 }
 858
 859 static void job_completed(Job *job)
 860 {
 861     assert(job && job->txn && !job_is_completed(job));
 862
 863     job_update_rc(job);
 864     trace_job_completed(job, job->ret);
 865     if (job->ret) {
 866         job_completed_txn_abort(job);
 867     } else {
 868         job_completed_txn_success(job);
 869     }
 870 }
 871
 872 /** Useful only as a type shim for aio_bh_schedule_oneshot. */
 873 static void job_exit(void *opaque)
 874 {
 875     Job *job = (Job *)opaque;
 876     AioContext *ctx;
 877
 878     job_ref(job);
 879     aio_context_acquire(job->aio_context);
 880
 881     /* This is a lie, we're not quiescent, but still doing the completion
 882      * callbacks. However, completion callbacks tend to involve operations that
 883      * drain block nodes, and if .drained_poll still returned true, we would
 884      * deadlock. */
 885     job->busy = false;
 886     job_event_idle(job);
 887
 888     job_completed(job);
 889
 890     /*
 891      * Note that calling job_completed can move the job to a different
 892      * aio_context, so we cannot cache from above. job_txn_apply takes care of
 893      * acquiring the new lock, and we ref/unref to avoid job_completed freeing
 894      * the job underneath us.
 895      */
 896     ctx = job->aio_context;
 897     job_unref(job);
 898     aio_context_release(ctx);
 899 }
 900
 901 /**
 902  * All jobs must allow a pause point before entering their job proper. This
 903  * ensures that jobs can be paused prior to being started, then resumed later.
 904  */
 905 static void coroutine_fn job_co_entry(void *opaque)
 906 {
 907     Job *job = opaque;
 908
 909     assert(job && job->driver && job->driver->run);
 910     job_pause_point(job);
 911     job->ret = job->driver->run(job, &job->err);
 912     job->deferred_to_main_loop = true;
 913     job->busy = true;
 914     aio_bh_schedule_oneshot(qemu_get_aio_context(), job_exit, job);
 915 }
 916
 917 void job_start(Job *job)
 918 {
 919     assert(job && !job_started(job) && job->paused &&
 920            job->driver && job->driver->run);
 921     job->co = qemu_coroutine_create(job_co_entry, job);
 922     job->pause_count--;
 923     job->busy = true;
 924     job->paused = false;
 925     job_state_transition(job, JOB_STATUS_RUNNING);
 926     aio_co_enter(job->aio_context, job->co);
 927 }
 928
 929 void job_cancel(Job *job, bool force)
 930 {
 931     if (job->status == JOB_STATUS_CONCLUDED) {
 932         job_do_dismiss(job);
 933         return;
 934     }
 935     job_cancel_async(job, force);
 936     if (!job_started(job)) {
 937         job_completed(job);
 938     } else if (job->deferred_to_main_loop) {
 939         job_completed_txn_abort(job);
 940     } else {
 941         job_enter(job);
 942     }
 943 }
 944
 945 void job_user_cancel(Job *job, bool force, Error **errp)
 946 {
 947     if (job_apply_verb(job, JOB_VERB_CANCEL, errp)) {
 948         return;
 949     }
 950     job_cancel(job, force);
 951 }
 952
 953 /* A wrapper around job_cancel() taking an Error ** parameter so it may be
 954  * used with job_finish_sync() without the need for (rather nasty) function
 955  * pointer casts there. */
 956 static void job_cancel_err(Job *job, Error **errp)
 957 {
 958     job_cancel(job, false);
 959 }
 960
 961 int job_cancel_sync(Job *job)
 962 {
 963     return job_finish_sync(job, &job_cancel_err, NULL);
 964 }
 965
 966 void job_cancel_sync_all(void)
 967 {
 968     Job *job;
 969     AioContext *aio_context;
 970
 971     while ((job = job_next(NULL))) {
 972         aio_context = job->aio_context;
 973         aio_context_acquire(aio_context);
 974         job_cancel_sync(job);
 975         aio_context_release(aio_context);
 976     }
 977 }
 978
 979 int job_complete_sync(Job *job, Error **errp)
 980 {
 981     return job_finish_sync(job, job_complete, errp);
 982 }
 983
 984 void job_complete(Job *job, Error **errp)
 985 {
 986     /* Should not be reachable via external interface for internal jobs */
 987     assert(job->id);
 988     if (job_apply_verb(job, JOB_VERB_COMPLETE, errp)) {
 989         return;
 990     }
 991     if (job->pause_count || job_is_cancelled(job) || !job->driver->complete) {
 992         error_setg(errp, "The active block job '%s' cannot be completed",
 993                    job->id);
 994         return;
 995     }
 996
 997     job->driver->complete(job, errp);
 998 }
 999
1000 int job_finish_sync(Job *job, void (*finish)(Job *, Error **errp), Error **errp)
1001 {
1002     Error *local_err = NULL;
1003     int ret;
1004
1005     job_ref(job);
1006
1007     if (finish) {
1008         finish(job, &local_err);
1009     }
1010     if (local_err) {
1011         error_propagate(errp, local_err);
1012         job_unref(job);
1013         return -EBUSY;
1014     }
1015
1016     AIO_WAIT_WHILE(job->aio_context,
1017                    (job_enter(job), !job_is_completed(job)));
1018
1019     ret = (job_is_cancelled(job) && job->ret == 0) ? -ECANCELED : job->ret;
1020     job_unref(job);
1021     return ret;
1022 }