tests/unit/test-blockjob.c

   1 /*
   2  * Blockjob tests
   3  *
   4  * Copyright Igalia, S.L. 2016
   5  *
   6  * Authors:
   7  *  Alberto Garcia   <berto@igalia.com>
   8  *
   9  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
  10  * See the COPYING.LIB file in the top-level directory.
  11  */
  12
  13 #include "qemu/osdep.h"
  14 #include "qapi/error.h"
  15 #include "qemu/main-loop.h"
  16 #include "block/blockjob_int.h"
  17 #include "sysemu/block-backend.h"
  18 #include "qapi/qmp/qdict.h"
  19 #include "iothread.h"
  20
  21 static const BlockJobDriver test_block_job_driver = {
  22     .job_driver = {
  23         .instance_size = sizeof(BlockJob),
  24         .free          = block_job_free,
  25         .user_resume   = block_job_user_resume,
  26     },
  27 };
  28
  29 static void block_job_cb(void *opaque, int ret)
  30 {
  31 }
  32
  33 static BlockJob *mk_job(BlockBackend *blk, const char *id,
  34                         const BlockJobDriver *drv, bool should_succeed,
  35                         int flags)
  36 {
  37     BlockJob *job;
  38     Error *err = NULL;
  39
  40     job = block_job_create(id, drv, NULL, blk_bs(blk),
  41                            0, BLK_PERM_ALL, 0, flags, block_job_cb,
  42                            NULL, &err);
  43     if (should_succeed) {
  44         g_assert_null(err);
  45         g_assert_nonnull(job);
  46         if (id) {
  47             g_assert_cmpstr(job->job.id, ==, id);
  48         } else {
  49             g_assert_cmpstr(job->job.id, ==, blk_name(blk));
  50         }
  51     } else {
  52         error_free_or_abort(&err);
  53         g_assert_null(job);
  54     }
  55
  56     return job;
  57 }
  58
  59 static BlockJob *do_test_id(BlockBackend *blk, const char *id,
  60                             bool should_succeed)
  61 {
  62     return mk_job(blk, id, &test_block_job_driver,
  63                   should_succeed, JOB_DEFAULT);
  64 }
  65
  66 /* This creates a BlockBackend (optionally with a name) with a
  67  * BlockDriverState inserted. */
  68 static BlockBackend *create_blk(const char *name)
  69 {
  70     /* No I/O is performed on this device */
  71     BlockBackend *blk = blk_new(qemu_get_aio_context(), 0, BLK_PERM_ALL);
  72     BlockDriverState *bs;
  73
  74     QDict *opt = qdict_new();
  75     qdict_put_str(opt, "file.read-zeroes", "on");
  76     bs = bdrv_open("null-co://", NULL, opt, 0, &error_abort);
  77     g_assert_nonnull(bs);
  78
  79     blk_insert_bs(blk, bs, &error_abort);
  80     bdrv_unref(bs);
  81
  82     if (name) {
  83         Error *err = NULL;
  84         monitor_add_blk(blk, name, &err);
  85         g_assert_null(err);
  86     }
  87
  88     return blk;
  89 }
  90
  91 /* This destroys the backend */
  92 static void destroy_blk(BlockBackend *blk)
  93 {
  94     if (blk_name(blk)[0] != '\0') {
  95         monitor_remove_blk(blk);
  96     }
  97
  98     blk_remove_bs(blk);
  99     blk_unref(blk);
 100 }
 101
 102 static void test_job_ids(void)
 103 {
 104     BlockBackend *blk[3];
 105     BlockJob *job[3];
 106
 107     blk[0] = create_blk(NULL);
 108     blk[1] = create_blk("drive1");
 109     blk[2] = create_blk("drive2");
 110
 111     /* No job ID provided and the block backend has no name */
 112     job[0] = do_test_id(blk[0], NULL, false);
 113
 114     /* These are all invalid job IDs */
 115     job[0] = do_test_id(blk[0], "0id", false);
 116     job[0] = do_test_id(blk[0], "",    false);
 117     job[0] = do_test_id(blk[0], "   ", false);
 118     job[0] = do_test_id(blk[0], "123", false);
 119     job[0] = do_test_id(blk[0], "_id", false);
 120     job[0] = do_test_id(blk[0], "-id", false);
 121     job[0] = do_test_id(blk[0], ".id", false);
 122     job[0] = do_test_id(blk[0], "#id", false);
 123
 124     /* This one is valid */
 125     job[0] = do_test_id(blk[0], "id0", true);
 126
 127     /* We can have two jobs in the same BDS */
 128     job[1] = do_test_id(blk[0], "id1", true);
 129     job_early_fail(&job[1]->job);
 130
 131     /* Duplicate job IDs are not allowed */
 132     job[1] = do_test_id(blk[1], "id0", false);
 133
 134     /* But once job[0] finishes we can reuse its ID */
 135     job_early_fail(&job[0]->job);
 136     job[1] = do_test_id(blk[1], "id0", true);
 137
 138     /* No job ID specified, defaults to the backend name ('drive1') */
 139     job_early_fail(&job[1]->job);
 140     job[1] = do_test_id(blk[1], NULL, true);
 141
 142     /* Duplicate job ID */
 143     job[2] = do_test_id(blk[2], "drive1", false);
 144
 145     /* The ID of job[2] would default to 'drive2' but it is already in use */
 146     job[0] = do_test_id(blk[0], "drive2", true);
 147     job[2] = do_test_id(blk[2], NULL, false);
 148
 149     /* This one is valid */
 150     job[2] = do_test_id(blk[2], "id_2", true);
 151
 152     job_early_fail(&job[0]->job);
 153     job_early_fail(&job[1]->job);
 154     job_early_fail(&job[2]->job);
 155
 156     destroy_blk(blk[0]);
 157     destroy_blk(blk[1]);
 158     destroy_blk(blk[2]);
 159 }
 160
 161 typedef struct CancelJob {
 162     BlockJob common;
 163     BlockBackend *blk;
 164     bool should_converge;
 165     bool should_complete;
 166 } CancelJob;
 167
 168 static void cancel_job_complete(Job *job, Error **errp)
 169 {
 170     CancelJob *s = container_of(job, CancelJob, common.job);
 171     s->should_complete = true;
 172 }
 173
 174 static int coroutine_fn cancel_job_run(Job *job, Error **errp)
 175 {
 176     CancelJob *s = container_of(job, CancelJob, common.job);
 177
 178     while (!s->should_complete) {
 179         if (job_is_cancelled(&s->common.job)) {
 180             return 0;
 181         }
 182
 183         if (!job_is_ready(&s->common.job) && s->should_converge) {
 184             job_transition_to_ready(&s->common.job);
 185         }
 186
 187         job_sleep_ns(&s->common.job, 100000);
 188     }
 189
 190     return 0;
 191 }
 192
 193 static const BlockJobDriver test_cancel_driver = {
 194     .job_driver = {
 195         .instance_size = sizeof(CancelJob),
 196         .free          = block_job_free,
 197         .user_resume   = block_job_user_resume,
 198         .run           = cancel_job_run,
 199         .complete      = cancel_job_complete,
 200     },
 201 };
 202
 203 static CancelJob *create_common(Job **pjob)
 204 {
 205     BlockBackend *blk;
 206     Job *job;
 207     BlockJob *bjob;
 208     CancelJob *s;
 209
 210     blk = create_blk(NULL);
 211     bjob = mk_job(blk, "Steve", &test_cancel_driver, true,
 212                   JOB_MANUAL_FINALIZE | JOB_MANUAL_DISMISS);
 213     job = &bjob->job;
 214     job_ref(job);
 215     assert(job->status == JOB_STATUS_CREATED);
 216     s = container_of(bjob, CancelJob, common);
 217     s->blk = blk;
 218
 219     *pjob = job;
 220     return s;
 221 }
 222
 223 static void cancel_common(CancelJob *s)
 224 {
 225     BlockJob *job = &s->common;
 226     BlockBackend *blk = s->blk;
 227     JobStatus sts = job->job.status;
 228     AioContext *ctx;
 229
 230     ctx = job->job.aio_context;
 231     aio_context_acquire(ctx);
 232
 233     job_cancel_sync(&job->job, true);
 234     if (sts != JOB_STATUS_CREATED && sts != JOB_STATUS_CONCLUDED) {
 235         Job *dummy = &job->job;
 236         job_dismiss(&dummy, &error_abort);
 237     }
 238     assert(job->job.status == JOB_STATUS_NULL);
 239     job_unref(&job->job);
 240     destroy_blk(blk);
 241
 242     aio_context_release(ctx);
 243 }
 244
 245 static void test_cancel_created(void)
 246 {
 247     Job *job;
 248     CancelJob *s;
 249
 250     s = create_common(&job);
 251     cancel_common(s);
 252 }
 253
 254 static void test_cancel_running(void)
 255 {
 256     Job *job;
 257     CancelJob *s;
 258
 259     s = create_common(&job);
 260
 261     job_start(job);
 262     assert(job->status == JOB_STATUS_RUNNING);
 263
 264     cancel_common(s);
 265 }
 266
 267 static void test_cancel_paused(void)
 268 {
 269     Job *job;
 270     CancelJob *s;
 271
 272     s = create_common(&job);
 273
 274     job_start(job);
 275     assert(job->status == JOB_STATUS_RUNNING);
 276
 277     job_user_pause(job, &error_abort);
 278     job_enter(job);
 279     assert(job->status == JOB_STATUS_PAUSED);
 280
 281     cancel_common(s);
 282 }
 283
 284 static void test_cancel_ready(void)
 285 {
 286     Job *job;
 287     CancelJob *s;
 288
 289     s = create_common(&job);
 290
 291     job_start(job);
 292     assert(job->status == JOB_STATUS_RUNNING);
 293
 294     s->should_converge = true;
 295     job_enter(job);
 296     assert(job->status == JOB_STATUS_READY);
 297
 298     cancel_common(s);
 299 }
 300
 301 static void test_cancel_standby(void)
 302 {
 303     Job *job;
 304     CancelJob *s;
 305
 306     s = create_common(&job);
 307
 308     job_start(job);
 309     assert(job->status == JOB_STATUS_RUNNING);
 310
 311     s->should_converge = true;
 312     job_enter(job);
 313     assert(job->status == JOB_STATUS_READY);
 314
 315     job_user_pause(job, &error_abort);
 316     job_enter(job);
 317     assert(job->status == JOB_STATUS_STANDBY);
 318
 319     cancel_common(s);
 320 }
 321
 322 static void test_cancel_pending(void)
 323 {
 324     Job *job;
 325     CancelJob *s;
 326
 327     s = create_common(&job);
 328
 329     job_start(job);
 330     assert(job->status == JOB_STATUS_RUNNING);
 331
 332     s->should_converge = true;
 333     job_enter(job);
 334     assert(job->status == JOB_STATUS_READY);
 335
 336     job_complete(job, &error_abort);
 337     job_enter(job);
 338     while (!job->deferred_to_main_loop) {
 339         aio_poll(qemu_get_aio_context(), true);
 340     }
 341     assert(job->status == JOB_STATUS_READY);
 342     aio_poll(qemu_get_aio_context(), true);
 343     assert(job->status == JOB_STATUS_PENDING);
 344
 345     cancel_common(s);
 346 }
 347
 348 static void test_cancel_concluded(void)
 349 {
 350     Job *job;
 351     CancelJob *s;
 352
 353     s = create_common(&job);
 354
 355     job_start(job);
 356     assert(job->status == JOB_STATUS_RUNNING);
 357
 358     s->should_converge = true;
 359     job_enter(job);
 360     assert(job->status == JOB_STATUS_READY);
 361
 362     job_complete(job, &error_abort);
 363     job_enter(job);
 364     while (!job->deferred_to_main_loop) {
 365         aio_poll(qemu_get_aio_context(), true);
 366     }
 367     assert(job->status == JOB_STATUS_READY);
 368     aio_poll(qemu_get_aio_context(), true);
 369     assert(job->status == JOB_STATUS_PENDING);
 370
 371     aio_context_acquire(job->aio_context);
 372     job_finalize(job, &error_abort);
 373     aio_context_release(job->aio_context);
 374     assert(job->status == JOB_STATUS_CONCLUDED);
 375
 376     cancel_common(s);
 377 }
 378
 379 /* (See test_yielding_driver for the job description) */
 380 typedef struct YieldingJob {
 381     BlockJob common;
 382     bool should_complete;
 383 } YieldingJob;
 384
 385 static void yielding_job_complete(Job *job, Error **errp)
 386 {
 387     YieldingJob *s = container_of(job, YieldingJob, common.job);
 388     s->should_complete = true;
 389     job_enter(job);
 390 }
 391
 392 static int coroutine_fn yielding_job_run(Job *job, Error **errp)
 393 {
 394     YieldingJob *s = container_of(job, YieldingJob, common.job);
 395
 396     job_transition_to_ready(job);
 397
 398     while (!s->should_complete) {
 399         job_yield(job);
 400     }
 401
 402     return 0;
 403 }
 404
 405 /*
 406  * This job transitions immediately to the READY state, and then
 407  * yields until it is to complete.
 408  */
 409 static const BlockJobDriver test_yielding_driver = {
 410     .job_driver = {
 411         .instance_size  = sizeof(YieldingJob),
 412         .free           = block_job_free,
 413         .user_resume    = block_job_user_resume,
 414         .run            = yielding_job_run,
 415         .complete       = yielding_job_complete,
 416     },
 417 };
 418
 419 /*
 420  * Test that job_complete() works even on jobs that are in a paused
 421  * state (i.e., STANDBY).
 422  *
 423  * To do this, run YieldingJob in an IO thread, get it into the READY
 424  * state, then have a drained section.  Before ending the section,
 425  * acquire the context so the job will not be entered and will thus
 426  * remain on STANDBY.
 427  *
 428  * job_complete() should still work without error.
 429  *
 430  * Note that on the QMP interface, it is impossible to lock an IO
 431  * thread before a drained section ends.  In practice, the
 432  * bdrv_drain_all_end() and the aio_context_acquire() will be
 433  * reversed.  However, that makes for worse reproducibility here:
 434  * Sometimes, the job would no longer be in STANDBY then but already
 435  * be started.  We cannot prevent that, because the IO thread runs
 436  * concurrently.  We can only prevent it by taking the lock before
 437  * ending the drained section, so we do that.
 438  *
 439  * (You can reverse the order of operations and most of the time the
 440  * test will pass, but sometimes the assert(status == STANDBY) will
 441  * fail.)
 442  */
 443 static void test_complete_in_standby(void)
 444 {
 445     BlockBackend *blk;
 446     IOThread *iothread;
 447     AioContext *ctx;
 448     Job *job;
 449     BlockJob *bjob;
 450
 451     /* Create a test drive, move it to an IO thread */
 452     blk = create_blk(NULL);
 453     iothread = iothread_new();
 454
 455     ctx = iothread_get_aio_context(iothread);
 456     blk_set_aio_context(blk, ctx, &error_abort);
 457
 458     /* Create our test job */
 459     bjob = mk_job(blk, "job", &test_yielding_driver, true,
 460                   JOB_MANUAL_FINALIZE | JOB_MANUAL_DISMISS);
 461     job = &bjob->job;
 462     assert(job->status == JOB_STATUS_CREATED);
 463
 464     /* Wait for the job to become READY */
 465     job_start(job);
 466     aio_context_acquire(ctx);
 467     AIO_WAIT_WHILE(ctx, job->status != JOB_STATUS_READY);
 468     aio_context_release(ctx);
 469
 470     /* Begin the drained section, pausing the job */
 471     bdrv_drain_all_begin();
 472     assert(job->status == JOB_STATUS_STANDBY);
 473     /* Lock the IO thread to prevent the job from being run */
 474     aio_context_acquire(ctx);
 475     /* This will schedule the job to resume it */
 476     bdrv_drain_all_end();
 477
 478     /* But the job cannot run, so it will remain on standby */
 479     assert(job->status == JOB_STATUS_STANDBY);
 480
 481     /* Even though the job is on standby, this should work */
 482     job_complete(job, &error_abort);
 483
 484     /* The test is done now, clean up. */
 485     job_finish_sync(job, NULL, &error_abort);
 486     assert(job->status == JOB_STATUS_PENDING);
 487
 488     job_finalize(job, &error_abort);
 489     assert(job->status == JOB_STATUS_CONCLUDED);
 490
 491     job_dismiss(&job, &error_abort);
 492
 493     destroy_blk(blk);
 494     aio_context_release(ctx);
 495     iothread_join(iothread);
 496 }
 497
 498 int main(int argc, char **argv)
 499 {
 500     qemu_init_main_loop(&error_abort);
 501     bdrv_init();
 502
 503     g_test_init(&argc, &argv, NULL);
 504     g_test_add_func("/blockjob/ids", test_job_ids);
 505     g_test_add_func("/blockjob/cancel/created", test_cancel_created);
 506     g_test_add_func("/blockjob/cancel/running", test_cancel_running);
 507     g_test_add_func("/blockjob/cancel/paused", test_cancel_paused);
 508     g_test_add_func("/blockjob/cancel/ready", test_cancel_ready);
 509     g_test_add_func("/blockjob/cancel/standby", test_cancel_standby);
 510     g_test_add_func("/blockjob/cancel/pending", test_cancel_pending);
 511     g_test_add_func("/blockjob/cancel/concluded", test_cancel_concluded);
 512     g_test_add_func("/blockjob/complete_in_standby", test_complete_in_standby);
 513     return g_test_run();
 514 }