xref: /openbmc/qemu/tests/unit/test-blockjob.c (revision f6b0de53)
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     WITH_JOB_LOCK_GUARD() {
215         job_ref_locked(job);
216         assert(job->status == JOB_STATUS_CREATED);
217     }
218 
219     s = container_of(bjob, CancelJob, common);
220     s->blk = blk;
221 
222     *pjob = job;
223     return s;
224 }
225 
226 static void cancel_common(CancelJob *s)
227 {
228     BlockJob *job = &s->common;
229     BlockBackend *blk = s->blk;
230     JobStatus sts = job->job.status;
231     AioContext *ctx = job->job.aio_context;
232 
233     job_cancel_sync(&job->job, true);
234     WITH_JOB_LOCK_GUARD() {
235         if (sts != JOB_STATUS_CREATED && sts != JOB_STATUS_CONCLUDED) {
236             Job *dummy = &job->job;
237             job_dismiss_locked(&dummy, &error_abort);
238         }
239         assert(job->job.status == JOB_STATUS_NULL);
240         job_unref_locked(&job->job);
241     }
242 
243     aio_context_acquire(ctx);
244     destroy_blk(blk);
245     aio_context_release(ctx);
246 
247 }
248 
249 static void test_cancel_created(void)
250 {
251     Job *job;
252     CancelJob *s;
253 
254     s = create_common(&job);
255     cancel_common(s);
256 }
257 
258 static void assert_job_status_is(Job *job, int status)
259 {
260     WITH_JOB_LOCK_GUARD() {
261         assert(job->status == status);
262     }
263 }
264 
265 static void test_cancel_running(void)
266 {
267     Job *job;
268     CancelJob *s;
269 
270     s = create_common(&job);
271 
272     job_start(job);
273     assert_job_status_is(job, JOB_STATUS_RUNNING);
274 
275     cancel_common(s);
276 }
277 
278 static void test_cancel_paused(void)
279 {
280     Job *job;
281     CancelJob *s;
282 
283     s = create_common(&job);
284 
285     job_start(job);
286     WITH_JOB_LOCK_GUARD() {
287         assert(job->status == JOB_STATUS_RUNNING);
288         job_user_pause_locked(job, &error_abort);
289     }
290     job_enter(job);
291     assert_job_status_is(job, JOB_STATUS_PAUSED);
292 
293     cancel_common(s);
294 }
295 
296 static void test_cancel_ready(void)
297 {
298     Job *job;
299     CancelJob *s;
300 
301     s = create_common(&job);
302 
303     job_start(job);
304     assert_job_status_is(job, JOB_STATUS_RUNNING);
305 
306     s->should_converge = true;
307     job_enter(job);
308     assert_job_status_is(job, JOB_STATUS_READY);
309 
310     cancel_common(s);
311 }
312 
313 static void test_cancel_standby(void)
314 {
315     Job *job;
316     CancelJob *s;
317 
318     s = create_common(&job);
319 
320     job_start(job);
321     assert_job_status_is(job, JOB_STATUS_RUNNING);
322 
323     s->should_converge = true;
324     job_enter(job);
325     WITH_JOB_LOCK_GUARD() {
326         assert(job->status == JOB_STATUS_READY);
327         job_user_pause_locked(job, &error_abort);
328     }
329     job_enter(job);
330     assert_job_status_is(job, JOB_STATUS_STANDBY);
331 
332     cancel_common(s);
333 }
334 
335 static void test_cancel_pending(void)
336 {
337     Job *job;
338     CancelJob *s;
339 
340     s = create_common(&job);
341 
342     job_start(job);
343     assert_job_status_is(job, JOB_STATUS_RUNNING);
344 
345     s->should_converge = true;
346     job_enter(job);
347     WITH_JOB_LOCK_GUARD() {
348         assert(job->status == JOB_STATUS_READY);
349         job_complete_locked(job, &error_abort);
350     }
351     job_enter(job);
352     while (!job->deferred_to_main_loop) {
353         aio_poll(qemu_get_aio_context(), true);
354     }
355     assert_job_status_is(job, JOB_STATUS_READY);
356     aio_poll(qemu_get_aio_context(), true);
357     assert_job_status_is(job, JOB_STATUS_PENDING);
358 
359     cancel_common(s);
360 }
361 
362 static void test_cancel_concluded(void)
363 {
364     Job *job;
365     CancelJob *s;
366 
367     s = create_common(&job);
368 
369     job_start(job);
370     assert_job_status_is(job, JOB_STATUS_RUNNING);
371 
372     s->should_converge = true;
373     job_enter(job);
374     WITH_JOB_LOCK_GUARD() {
375         assert(job->status == JOB_STATUS_READY);
376         job_complete_locked(job, &error_abort);
377     }
378     job_enter(job);
379     while (!job->deferred_to_main_loop) {
380         aio_poll(qemu_get_aio_context(), true);
381     }
382     assert_job_status_is(job, JOB_STATUS_READY);
383     aio_poll(qemu_get_aio_context(), true);
384     assert_job_status_is(job, JOB_STATUS_PENDING);
385 
386     WITH_JOB_LOCK_GUARD() {
387         job_finalize_locked(job, &error_abort);
388         assert(job->status == JOB_STATUS_CONCLUDED);
389     }
390 
391     cancel_common(s);
392 }
393 
394 /* (See test_yielding_driver for the job description) */
395 typedef struct YieldingJob {
396     BlockJob common;
397     bool should_complete;
398 } YieldingJob;
399 
400 static void yielding_job_complete(Job *job, Error **errp)
401 {
402     YieldingJob *s = container_of(job, YieldingJob, common.job);
403     s->should_complete = true;
404     job_enter(job);
405 }
406 
407 static int coroutine_fn yielding_job_run(Job *job, Error **errp)
408 {
409     YieldingJob *s = container_of(job, YieldingJob, common.job);
410 
411     job_transition_to_ready(job);
412 
413     while (!s->should_complete) {
414         job_yield(job);
415     }
416 
417     return 0;
418 }
419 
420 /*
421  * This job transitions immediately to the READY state, and then
422  * yields until it is to complete.
423  */
424 static const BlockJobDriver test_yielding_driver = {
425     .job_driver = {
426         .instance_size  = sizeof(YieldingJob),
427         .free           = block_job_free,
428         .user_resume    = block_job_user_resume,
429         .run            = yielding_job_run,
430         .complete       = yielding_job_complete,
431     },
432 };
433 
434 /*
435  * Test that job_complete_locked() works even on jobs that are in a paused
436  * state (i.e., STANDBY).
437  *
438  * To do this, run YieldingJob in an IO thread, get it into the READY
439  * state, then have a drained section.  Before ending the section,
440  * acquire the context so the job will not be entered and will thus
441  * remain on STANDBY.
442  *
443  * job_complete_locked() should still work without error.
444  *
445  * Note that on the QMP interface, it is impossible to lock an IO
446  * thread before a drained section ends.  In practice, the
447  * bdrv_drain_all_end() and the aio_context_acquire() will be
448  * reversed.  However, that makes for worse reproducibility here:
449  * Sometimes, the job would no longer be in STANDBY then but already
450  * be started.  We cannot prevent that, because the IO thread runs
451  * concurrently.  We can only prevent it by taking the lock before
452  * ending the drained section, so we do that.
453  *
454  * (You can reverse the order of operations and most of the time the
455  * test will pass, but sometimes the assert(status == STANDBY) will
456  * fail.)
457  */
458 static void test_complete_in_standby(void)
459 {
460     BlockBackend *blk;
461     IOThread *iothread;
462     AioContext *ctx;
463     Job *job;
464     BlockJob *bjob;
465 
466     /* Create a test drive, move it to an IO thread */
467     blk = create_blk(NULL);
468     iothread = iothread_new();
469 
470     ctx = iothread_get_aio_context(iothread);
471     blk_set_aio_context(blk, ctx, &error_abort);
472 
473     /* Create our test job */
474     bjob = mk_job(blk, "job", &test_yielding_driver, true,
475                   JOB_MANUAL_FINALIZE | JOB_MANUAL_DISMISS);
476     job = &bjob->job;
477     assert_job_status_is(job, JOB_STATUS_CREATED);
478 
479     /* Wait for the job to become READY */
480     job_start(job);
481     /*
482      * Here we are waiting for the status to change, so don't bother
483      * protecting the read every time.
484      */
485     AIO_WAIT_WHILE_UNLOCKED(ctx, job->status != JOB_STATUS_READY);
486 
487     /* Begin the drained section, pausing the job */
488     bdrv_drain_all_begin();
489     assert_job_status_is(job, JOB_STATUS_STANDBY);
490 
491     /* Lock the IO thread to prevent the job from being run */
492     aio_context_acquire(ctx);
493     /* This will schedule the job to resume it */
494     bdrv_drain_all_end();
495     aio_context_release(ctx);
496 
497     WITH_JOB_LOCK_GUARD() {
498         /* But the job cannot run, so it will remain on standby */
499         assert(job->status == JOB_STATUS_STANDBY);
500 
501         /* Even though the job is on standby, this should work */
502         job_complete_locked(job, &error_abort);
503 
504         /* The test is done now, clean up. */
505         job_finish_sync_locked(job, NULL, &error_abort);
506         assert(job->status == JOB_STATUS_PENDING);
507 
508         job_finalize_locked(job, &error_abort);
509         assert(job->status == JOB_STATUS_CONCLUDED);
510 
511         job_dismiss_locked(&job, &error_abort);
512     }
513 
514     aio_context_acquire(ctx);
515     destroy_blk(blk);
516     aio_context_release(ctx);
517     iothread_join(iothread);
518 }
519 
520 int main(int argc, char **argv)
521 {
522     qemu_init_main_loop(&error_abort);
523     bdrv_init();
524 
525     g_test_init(&argc, &argv, NULL);
526     g_test_add_func("/blockjob/ids", test_job_ids);
527     g_test_add_func("/blockjob/cancel/created", test_cancel_created);
528     g_test_add_func("/blockjob/cancel/running", test_cancel_running);
529     g_test_add_func("/blockjob/cancel/paused", test_cancel_paused);
530     g_test_add_func("/blockjob/cancel/ready", test_cancel_ready);
531     g_test_add_func("/blockjob/cancel/standby", test_cancel_standby);
532     g_test_add_func("/blockjob/cancel/pending", test_cancel_pending);
533     g_test_add_func("/blockjob/cancel/concluded", test_cancel_concluded);
534 
535     /*
536      * This test is flaky and sometimes fails in CI and otherwise:
537      * don't run unless user opts in via environment variable.
538      */
539     if (getenv("QEMU_TEST_FLAKY_TESTS")) {
540         g_test_add_func("/blockjob/complete_in_standby", test_complete_in_standby);
541     }
542     return g_test_run();
543 }
544