xref: /spdk/test/unit/lib/bdev/mt/bdev.c/bdev_ut.c (revision 1fc4165fe9bf8512483356ad8e6d27f793f2e3db)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk_cunit.h"
35 
36 #include "common/lib/ut_multithread.c"
37 #include "unit/lib/json_mock.c"
38 
39 #include "spdk/config.h"
40 /* HACK: disable VTune integration so the unit test doesn't need VTune headers and libs to build */
41 #undef SPDK_CONFIG_VTUNE
42 
43 #include "bdev/bdev.c"
44 
45 #define BDEV_UT_NUM_THREADS 3
46 
47 DEFINE_STUB(spdk_conf_find_section, struct spdk_conf_section *, (struct spdk_conf *cp,
48 		const char *name), NULL);
49 DEFINE_STUB(spdk_conf_section_get_nmval, char *,
50 	    (struct spdk_conf_section *sp, const char *key, int idx1, int idx2), NULL);
51 DEFINE_STUB(spdk_conf_section_get_intval, int, (struct spdk_conf_section *sp, const char *key), -1);
52 
53 struct spdk_trace_histories *g_trace_histories;
54 DEFINE_STUB_V(spdk_trace_add_register_fn, (struct spdk_trace_register_fn *reg_fn));
55 DEFINE_STUB_V(spdk_trace_register_owner, (uint8_t type, char id_prefix));
56 DEFINE_STUB_V(spdk_trace_register_object, (uint8_t type, char id_prefix));
57 DEFINE_STUB_V(spdk_trace_register_description, (const char *name, const char *short_name,
58 		uint16_t tpoint_id, uint8_t owner_type,
59 		uint8_t object_type, uint8_t new_object,
60 		uint8_t arg1_is_ptr, const char *arg1_name));
61 DEFINE_STUB_V(_spdk_trace_record, (uint64_t tsc, uint16_t tpoint_id, uint16_t poller_id,
62 				   uint32_t size, uint64_t object_id, uint64_t arg1));
63 
64 struct ut_bdev {
65 	struct spdk_bdev	bdev;
66 	void			*io_target;
67 };
68 
69 struct ut_bdev_channel {
70 	TAILQ_HEAD(, spdk_bdev_io)	outstanding_io;
71 	uint32_t			outstanding_cnt;
72 	uint32_t			avail_cnt;
73 };
74 
75 int g_io_device;
76 struct ut_bdev g_bdev;
77 struct spdk_bdev_desc *g_desc;
78 bool g_teardown_done = false;
79 bool g_get_io_channel = true;
80 bool g_create_ch = true;
81 bool g_init_complete_called = false;
82 bool g_fini_start_called = true;
83 int g_status = 0;
84 int g_count = 0;
85 struct spdk_histogram_data *g_histogram = NULL;
86 
87 static int
88 stub_create_ch(void *io_device, void *ctx_buf)
89 {
90 	struct ut_bdev_channel *ch = ctx_buf;
91 
92 	if (g_create_ch == false) {
93 		return -1;
94 	}
95 
96 	TAILQ_INIT(&ch->outstanding_io);
97 	ch->outstanding_cnt = 0;
98 	/*
99 	 * When avail gets to 0, the submit_request function will return ENOMEM.
100 	 *  Most tests to not want ENOMEM to occur, so by default set this to a
101 	 *  big value that won't get hit.  The ENOMEM tests can then override this
102 	 *  value to something much smaller to induce ENOMEM conditions.
103 	 */
104 	ch->avail_cnt = 2048;
105 	return 0;
106 }
107 
108 static void
109 stub_destroy_ch(void *io_device, void *ctx_buf)
110 {
111 }
112 
113 static struct spdk_io_channel *
114 stub_get_io_channel(void *ctx)
115 {
116 	struct ut_bdev *ut_bdev = ctx;
117 
118 	if (g_get_io_channel == true) {
119 		return spdk_get_io_channel(ut_bdev->io_target);
120 	} else {
121 		return NULL;
122 	}
123 }
124 
125 static int
126 stub_destruct(void *ctx)
127 {
128 	return 0;
129 }
130 
131 static void
132 stub_submit_request(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
133 {
134 	struct ut_bdev_channel *ch = spdk_io_channel_get_ctx(_ch);
135 
136 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_RESET) {
137 		struct spdk_bdev_io *io;
138 
139 		while (!TAILQ_EMPTY(&ch->outstanding_io)) {
140 			io = TAILQ_FIRST(&ch->outstanding_io);
141 			TAILQ_REMOVE(&ch->outstanding_io, io, module_link);
142 			ch->outstanding_cnt--;
143 			spdk_bdev_io_complete(io, SPDK_BDEV_IO_STATUS_FAILED);
144 			ch->avail_cnt++;
145 		}
146 	}
147 
148 	if (ch->avail_cnt > 0) {
149 		TAILQ_INSERT_TAIL(&ch->outstanding_io, bdev_io, module_link);
150 		ch->outstanding_cnt++;
151 		ch->avail_cnt--;
152 	} else {
153 		spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM);
154 	}
155 }
156 
157 static uint32_t
158 stub_complete_io(void *io_target, uint32_t num_to_complete)
159 {
160 	struct spdk_io_channel *_ch = spdk_get_io_channel(io_target);
161 	struct ut_bdev_channel *ch = spdk_io_channel_get_ctx(_ch);
162 	struct spdk_bdev_io *io;
163 	bool complete_all = (num_to_complete == 0);
164 	uint32_t num_completed = 0;
165 
166 	while (complete_all || num_completed < num_to_complete) {
167 		if (TAILQ_EMPTY(&ch->outstanding_io)) {
168 			break;
169 		}
170 		io = TAILQ_FIRST(&ch->outstanding_io);
171 		TAILQ_REMOVE(&ch->outstanding_io, io, module_link);
172 		ch->outstanding_cnt--;
173 		spdk_bdev_io_complete(io, SPDK_BDEV_IO_STATUS_SUCCESS);
174 		ch->avail_cnt++;
175 		num_completed++;
176 	}
177 
178 	spdk_put_io_channel(_ch);
179 	return num_completed;
180 }
181 
182 static struct spdk_bdev_fn_table fn_table = {
183 	.get_io_channel =	stub_get_io_channel,
184 	.destruct =		stub_destruct,
185 	.submit_request =	stub_submit_request,
186 };
187 
188 static int
189 module_init(void)
190 {
191 	return 0;
192 }
193 
194 static void
195 module_fini(void)
196 {
197 }
198 
199 static void
200 init_complete(void)
201 {
202 	g_init_complete_called = true;
203 }
204 
205 static void
206 fini_start(void)
207 {
208 	g_fini_start_called = true;
209 }
210 
211 struct spdk_bdev_module bdev_ut_if = {
212 	.name = "bdev_ut",
213 	.module_init = module_init,
214 	.module_fini = module_fini,
215 	.init_complete = init_complete,
216 	.fini_start = fini_start,
217 };
218 
219 SPDK_BDEV_MODULE_REGISTER(bdev_ut, &bdev_ut_if)
220 
221 static void
222 register_bdev(struct ut_bdev *ut_bdev, char *name, void *io_target)
223 {
224 	memset(ut_bdev, 0, sizeof(*ut_bdev));
225 
226 	ut_bdev->io_target = io_target;
227 	ut_bdev->bdev.ctxt = ut_bdev;
228 	ut_bdev->bdev.name = name;
229 	ut_bdev->bdev.fn_table = &fn_table;
230 	ut_bdev->bdev.module = &bdev_ut_if;
231 	ut_bdev->bdev.blocklen = 4096;
232 	ut_bdev->bdev.blockcnt = 1024;
233 
234 	spdk_bdev_register(&ut_bdev->bdev);
235 }
236 
237 static void
238 unregister_bdev(struct ut_bdev *ut_bdev)
239 {
240 	/* Handle any deferred messages. */
241 	poll_threads();
242 	spdk_bdev_unregister(&ut_bdev->bdev, NULL, NULL);
243 }
244 
245 static void
246 bdev_init_cb(void *done, int rc)
247 {
248 	CU_ASSERT(rc == 0);
249 	*(bool *)done = true;
250 }
251 
252 static void
253 setup_test(void)
254 {
255 	bool done = false;
256 
257 	allocate_threads(BDEV_UT_NUM_THREADS);
258 	set_thread(0);
259 	spdk_bdev_initialize(bdev_init_cb, &done);
260 	spdk_io_device_register(&g_io_device, stub_create_ch, stub_destroy_ch,
261 				sizeof(struct ut_bdev_channel), NULL);
262 	register_bdev(&g_bdev, "ut_bdev", &g_io_device);
263 	spdk_bdev_open(&g_bdev.bdev, true, NULL, NULL, &g_desc);
264 }
265 
266 static void
267 finish_cb(void *cb_arg)
268 {
269 	g_teardown_done = true;
270 }
271 
272 static void
273 teardown_test(void)
274 {
275 	set_thread(0);
276 	g_teardown_done = false;
277 	spdk_bdev_close(g_desc);
278 	g_desc = NULL;
279 	unregister_bdev(&g_bdev);
280 	spdk_io_device_unregister(&g_io_device, NULL);
281 	spdk_bdev_finish(finish_cb, NULL);
282 	poll_threads();
283 	memset(&g_bdev, 0, sizeof(g_bdev));
284 	CU_ASSERT(g_teardown_done == true);
285 	g_teardown_done = false;
286 	free_threads();
287 }
288 
289 static uint32_t
290 bdev_io_tailq_cnt(bdev_io_tailq_t *tailq)
291 {
292 	struct spdk_bdev_io *io;
293 	uint32_t cnt = 0;
294 
295 	TAILQ_FOREACH(io, tailq, internal.link) {
296 		cnt++;
297 	}
298 
299 	return cnt;
300 }
301 
302 static void
303 basic(void)
304 {
305 	g_init_complete_called = false;
306 	setup_test();
307 	CU_ASSERT(g_init_complete_called == true);
308 
309 	set_thread(0);
310 
311 	g_get_io_channel = false;
312 	g_ut_threads[0].ch = spdk_bdev_get_io_channel(g_desc);
313 	CU_ASSERT(g_ut_threads[0].ch == NULL);
314 
315 	g_get_io_channel = true;
316 	g_create_ch = false;
317 	g_ut_threads[0].ch = spdk_bdev_get_io_channel(g_desc);
318 	CU_ASSERT(g_ut_threads[0].ch == NULL);
319 
320 	g_get_io_channel = true;
321 	g_create_ch = true;
322 	g_ut_threads[0].ch = spdk_bdev_get_io_channel(g_desc);
323 	CU_ASSERT(g_ut_threads[0].ch != NULL);
324 	spdk_put_io_channel(g_ut_threads[0].ch);
325 
326 	g_fini_start_called = false;
327 	teardown_test();
328 	CU_ASSERT(g_fini_start_called == true);
329 }
330 
331 static void
332 _bdev_removed(void *done)
333 {
334 	*(bool *)done = true;
335 }
336 
337 static void
338 _bdev_unregistered(void *done, int rc)
339 {
340 	CU_ASSERT(rc == 0);
341 	*(bool *)done = true;
342 }
343 
344 static void
345 unregister_and_close(void)
346 {
347 	bool done, remove_notify;
348 	struct spdk_bdev_desc *desc;
349 
350 	setup_test();
351 	set_thread(0);
352 
353 	/* setup_test() automatically opens the bdev,
354 	 * but this test needs to do that in a different
355 	 * way. */
356 	spdk_bdev_close(g_desc);
357 	poll_threads();
358 
359 	remove_notify = false;
360 	spdk_bdev_open(&g_bdev.bdev, true, _bdev_removed, &remove_notify, &desc);
361 	CU_ASSERT(remove_notify == false);
362 	CU_ASSERT(desc != NULL);
363 
364 	/* There is an open descriptor on the device. Unregister it,
365 	 * which can't proceed until the descriptor is closed. */
366 	done = false;
367 	spdk_bdev_unregister(&g_bdev.bdev, _bdev_unregistered, &done);
368 	/* No polling has occurred, so neither of these should execute */
369 	CU_ASSERT(remove_notify == false);
370 	CU_ASSERT(done == false);
371 
372 	/* Prior to the unregister completing, close the descriptor */
373 	spdk_bdev_close(desc);
374 
375 	/* Poll the threads to allow all events to be processed */
376 	poll_threads();
377 
378 	/* Remove notify should not have been called because the
379 	 * descriptor is already closed. */
380 	CU_ASSERT(remove_notify == false);
381 
382 	/* The unregister should have completed */
383 	CU_ASSERT(done == true);
384 
385 	spdk_bdev_finish(finish_cb, NULL);
386 	poll_threads();
387 	free_threads();
388 }
389 
390 static void
391 reset_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
392 {
393 	bool *done = cb_arg;
394 
395 	CU_ASSERT(success == true);
396 	*done = true;
397 	spdk_bdev_free_io(bdev_io);
398 }
399 
400 static void
401 put_channel_during_reset(void)
402 {
403 	struct spdk_io_channel *io_ch;
404 	bool done = false;
405 
406 	setup_test();
407 
408 	set_thread(0);
409 	io_ch = spdk_bdev_get_io_channel(g_desc);
410 	CU_ASSERT(io_ch != NULL);
411 
412 	/*
413 	 * Start a reset, but then put the I/O channel before
414 	 *  the deferred messages for the reset get a chance to
415 	 *  execute.
416 	 */
417 	spdk_bdev_reset(g_desc, io_ch, reset_done, &done);
418 	spdk_put_io_channel(io_ch);
419 	poll_threads();
420 	stub_complete_io(g_bdev.io_target, 0);
421 
422 	teardown_test();
423 }
424 
425 static void
426 aborted_reset_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
427 {
428 	enum spdk_bdev_io_status *status = cb_arg;
429 
430 	*status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
431 	spdk_bdev_free_io(bdev_io);
432 }
433 
434 static void
435 aborted_reset(void)
436 {
437 	struct spdk_io_channel *io_ch[2];
438 	enum spdk_bdev_io_status status1 = SPDK_BDEV_IO_STATUS_PENDING,
439 				 status2 = SPDK_BDEV_IO_STATUS_PENDING;
440 
441 	setup_test();
442 
443 	set_thread(0);
444 	io_ch[0] = spdk_bdev_get_io_channel(g_desc);
445 	CU_ASSERT(io_ch[0] != NULL);
446 	spdk_bdev_reset(g_desc, io_ch[0], aborted_reset_done, &status1);
447 	poll_threads();
448 	CU_ASSERT(g_bdev.bdev.internal.reset_in_progress != NULL);
449 
450 	/*
451 	 * First reset has been submitted on ch0.  Now submit a second
452 	 *  reset on ch1 which will get queued since there is already a
453 	 *  reset in progress.
454 	 */
455 	set_thread(1);
456 	io_ch[1] = spdk_bdev_get_io_channel(g_desc);
457 	CU_ASSERT(io_ch[1] != NULL);
458 	spdk_bdev_reset(g_desc, io_ch[1], aborted_reset_done, &status2);
459 	poll_threads();
460 	CU_ASSERT(g_bdev.bdev.internal.reset_in_progress != NULL);
461 
462 	/*
463 	 * Now destroy ch1.  This will abort the queued reset.  Check that
464 	 *  the second reset was completed with failed status.  Also check
465 	 *  that bdev->internal.reset_in_progress != NULL, since the
466 	 *  original reset has not been completed yet.  This ensures that
467 	 *  the bdev code is correctly noticing that the failed reset is
468 	 *  *not* the one that had been submitted to the bdev module.
469 	 */
470 	set_thread(1);
471 	spdk_put_io_channel(io_ch[1]);
472 	poll_threads();
473 	CU_ASSERT(status2 == SPDK_BDEV_IO_STATUS_FAILED);
474 	CU_ASSERT(g_bdev.bdev.internal.reset_in_progress != NULL);
475 
476 	/*
477 	 * Now complete the first reset, verify that it completed with SUCCESS
478 	 *  status and that bdev->internal.reset_in_progress is also set back to NULL.
479 	 */
480 	set_thread(0);
481 	spdk_put_io_channel(io_ch[0]);
482 	stub_complete_io(g_bdev.io_target, 0);
483 	poll_threads();
484 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_SUCCESS);
485 	CU_ASSERT(g_bdev.bdev.internal.reset_in_progress == NULL);
486 
487 	teardown_test();
488 }
489 
490 static void
491 io_during_io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
492 {
493 	enum spdk_bdev_io_status *status = cb_arg;
494 
495 	*status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
496 	spdk_bdev_free_io(bdev_io);
497 }
498 
499 static void
500 io_during_reset(void)
501 {
502 	struct spdk_io_channel *io_ch[2];
503 	struct spdk_bdev_channel *bdev_ch[2];
504 	enum spdk_bdev_io_status status0, status1, status_reset;
505 	int rc;
506 
507 	setup_test();
508 
509 	/*
510 	 * First test normal case - submit an I/O on each of two channels (with no resets)
511 	 *  and verify they complete successfully.
512 	 */
513 	set_thread(0);
514 	io_ch[0] = spdk_bdev_get_io_channel(g_desc);
515 	bdev_ch[0] = spdk_io_channel_get_ctx(io_ch[0]);
516 	CU_ASSERT(bdev_ch[0]->flags == 0);
517 	status0 = SPDK_BDEV_IO_STATUS_PENDING;
518 	rc = spdk_bdev_read_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &status0);
519 	CU_ASSERT(rc == 0);
520 
521 	set_thread(1);
522 	io_ch[1] = spdk_bdev_get_io_channel(g_desc);
523 	bdev_ch[1] = spdk_io_channel_get_ctx(io_ch[1]);
524 	CU_ASSERT(bdev_ch[1]->flags == 0);
525 	status1 = SPDK_BDEV_IO_STATUS_PENDING;
526 	rc = spdk_bdev_read_blocks(g_desc, io_ch[1], NULL, 0, 1, io_during_io_done, &status1);
527 	CU_ASSERT(rc == 0);
528 
529 	poll_threads();
530 	CU_ASSERT(status0 == SPDK_BDEV_IO_STATUS_PENDING);
531 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_PENDING);
532 
533 	set_thread(0);
534 	stub_complete_io(g_bdev.io_target, 0);
535 	CU_ASSERT(status0 == SPDK_BDEV_IO_STATUS_SUCCESS);
536 
537 	set_thread(1);
538 	stub_complete_io(g_bdev.io_target, 0);
539 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_SUCCESS);
540 
541 	/*
542 	 * Now submit a reset, and leave it pending while we submit I/O on two different
543 	 *  channels.  These I/O should be failed by the bdev layer since the reset is in
544 	 *  progress.
545 	 */
546 	set_thread(0);
547 	status_reset = SPDK_BDEV_IO_STATUS_PENDING;
548 	rc = spdk_bdev_reset(g_desc, io_ch[0], io_during_io_done, &status_reset);
549 	CU_ASSERT(rc == 0);
550 
551 	CU_ASSERT(bdev_ch[0]->flags == 0);
552 	CU_ASSERT(bdev_ch[1]->flags == 0);
553 	poll_threads();
554 	CU_ASSERT(bdev_ch[0]->flags == BDEV_CH_RESET_IN_PROGRESS);
555 	CU_ASSERT(bdev_ch[1]->flags == BDEV_CH_RESET_IN_PROGRESS);
556 
557 	set_thread(0);
558 	status0 = SPDK_BDEV_IO_STATUS_PENDING;
559 	rc = spdk_bdev_read_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &status0);
560 	CU_ASSERT(rc == 0);
561 
562 	set_thread(1);
563 	status1 = SPDK_BDEV_IO_STATUS_PENDING;
564 	rc = spdk_bdev_read_blocks(g_desc, io_ch[1], NULL, 0, 1, io_during_io_done, &status1);
565 	CU_ASSERT(rc == 0);
566 
567 	/*
568 	 * A reset is in progress so these read I/O should complete with failure.  Note that we
569 	 *  need to poll_threads() since I/O completed inline have their completion deferred.
570 	 */
571 	poll_threads();
572 	CU_ASSERT(status_reset == SPDK_BDEV_IO_STATUS_PENDING);
573 	CU_ASSERT(status0 == SPDK_BDEV_IO_STATUS_FAILED);
574 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_FAILED);
575 
576 	/*
577 	 * Complete the reset
578 	 */
579 	set_thread(0);
580 	stub_complete_io(g_bdev.io_target, 0);
581 
582 	/*
583 	 * Only poll thread 0. We should not get a completion.
584 	 */
585 	poll_thread(0);
586 	CU_ASSERT(status_reset == SPDK_BDEV_IO_STATUS_PENDING);
587 
588 	/*
589 	 * Poll both thread 0 and 1 so the messages can propagate and we
590 	 * get a completion.
591 	 */
592 	poll_threads();
593 	CU_ASSERT(status_reset == SPDK_BDEV_IO_STATUS_SUCCESS);
594 
595 	spdk_put_io_channel(io_ch[0]);
596 	set_thread(1);
597 	spdk_put_io_channel(io_ch[1]);
598 	poll_threads();
599 
600 	teardown_test();
601 }
602 
603 static void
604 basic_qos(void)
605 {
606 	struct spdk_io_channel *io_ch[2];
607 	struct spdk_bdev_channel *bdev_ch[2];
608 	struct spdk_bdev *bdev;
609 	enum spdk_bdev_io_status status;
610 	int rc;
611 
612 	setup_test();
613 
614 	/* Enable QoS */
615 	bdev = &g_bdev.bdev;
616 	bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
617 	SPDK_CU_ASSERT_FATAL(bdev->internal.qos != NULL);
618 	TAILQ_INIT(&bdev->internal.qos->queued);
619 	/*
620 	 * Enable read/write IOPS, read only byte per second and
621 	 * read/write byte per second rate limits.
622 	 * In this case, all rate limits will take equal effect.
623 	 */
624 	/* 2000 read/write I/O per second, or 2 per millisecond */
625 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT].limit = 2000;
626 	/* 8K read/write byte per millisecond with 4K block size */
627 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT].limit = 8192000;
628 	/* 8K read only byte per millisecond with 4K block size */
629 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_R_BPS_RATE_LIMIT].limit = 8192000;
630 
631 	g_get_io_channel = true;
632 
633 	set_thread(0);
634 	io_ch[0] = spdk_bdev_get_io_channel(g_desc);
635 	bdev_ch[0] = spdk_io_channel_get_ctx(io_ch[0]);
636 	CU_ASSERT(bdev_ch[0]->flags == BDEV_CH_QOS_ENABLED);
637 
638 	set_thread(1);
639 	io_ch[1] = spdk_bdev_get_io_channel(g_desc);
640 	bdev_ch[1] = spdk_io_channel_get_ctx(io_ch[1]);
641 	CU_ASSERT(bdev_ch[1]->flags == BDEV_CH_QOS_ENABLED);
642 
643 	/*
644 	 * Send an I/O on thread 0, which is where the QoS thread is running.
645 	 */
646 	set_thread(0);
647 	status = SPDK_BDEV_IO_STATUS_PENDING;
648 	rc = spdk_bdev_read_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &status);
649 	CU_ASSERT(rc == 0);
650 	CU_ASSERT(status == SPDK_BDEV_IO_STATUS_PENDING);
651 	poll_threads();
652 	stub_complete_io(g_bdev.io_target, 0);
653 	poll_threads();
654 	CU_ASSERT(status == SPDK_BDEV_IO_STATUS_SUCCESS);
655 
656 	/* Send an I/O on thread 1. The QoS thread is not running here. */
657 	status = SPDK_BDEV_IO_STATUS_PENDING;
658 	set_thread(1);
659 	rc = spdk_bdev_read_blocks(g_desc, io_ch[1], NULL, 0, 1, io_during_io_done, &status);
660 	CU_ASSERT(rc == 0);
661 	CU_ASSERT(status == SPDK_BDEV_IO_STATUS_PENDING);
662 	poll_threads();
663 	/* Complete I/O on thread 1. This should not complete the I/O we submitted */
664 	stub_complete_io(g_bdev.io_target, 0);
665 	poll_threads();
666 	CU_ASSERT(status == SPDK_BDEV_IO_STATUS_PENDING);
667 	/* Now complete I/O on thread 0 */
668 	set_thread(0);
669 	poll_threads();
670 	stub_complete_io(g_bdev.io_target, 0);
671 	poll_threads();
672 	CU_ASSERT(status == SPDK_BDEV_IO_STATUS_SUCCESS);
673 
674 	/*
675 	 * Close the descriptor only, which should stop the qos channel as
676 	 * the last descriptor removed.
677 	 */
678 	spdk_bdev_close(g_desc);
679 	poll_threads();
680 	CU_ASSERT(bdev->internal.qos->ch == NULL);
681 
682 	/*
683 	 * Open the bdev again which shall setup the qos channel as the
684 	 * channels are valid.
685 	 */
686 	spdk_bdev_open(bdev, true, NULL, NULL, &g_desc);
687 	poll_threads();
688 	CU_ASSERT(bdev->internal.qos->ch != NULL);
689 
690 	/* Tear down the channels */
691 	set_thread(0);
692 	spdk_put_io_channel(io_ch[0]);
693 	set_thread(1);
694 	spdk_put_io_channel(io_ch[1]);
695 	poll_threads();
696 	set_thread(0);
697 
698 	/* Close the descriptor, which should stop the qos channel */
699 	spdk_bdev_close(g_desc);
700 	poll_threads();
701 	CU_ASSERT(bdev->internal.qos->ch == NULL);
702 
703 	/* Open the bdev again, no qos channel setup without valid channels. */
704 	spdk_bdev_open(bdev, true, NULL, NULL, &g_desc);
705 	poll_threads();
706 	CU_ASSERT(bdev->internal.qos->ch == NULL);
707 
708 	/* Create the channels in reverse order. */
709 	set_thread(1);
710 	io_ch[1] = spdk_bdev_get_io_channel(g_desc);
711 	bdev_ch[1] = spdk_io_channel_get_ctx(io_ch[1]);
712 	CU_ASSERT(bdev_ch[1]->flags == BDEV_CH_QOS_ENABLED);
713 
714 	set_thread(0);
715 	io_ch[0] = spdk_bdev_get_io_channel(g_desc);
716 	bdev_ch[0] = spdk_io_channel_get_ctx(io_ch[0]);
717 	CU_ASSERT(bdev_ch[0]->flags == BDEV_CH_QOS_ENABLED);
718 
719 	/* Confirm that the qos thread is now thread 1 */
720 	CU_ASSERT(bdev->internal.qos->ch == bdev_ch[1]);
721 
722 	/* Tear down the channels */
723 	set_thread(0);
724 	spdk_put_io_channel(io_ch[0]);
725 	set_thread(1);
726 	spdk_put_io_channel(io_ch[1]);
727 	poll_threads();
728 
729 	set_thread(0);
730 
731 	teardown_test();
732 }
733 
734 static void
735 io_during_qos_queue(void)
736 {
737 	struct spdk_io_channel *io_ch[2];
738 	struct spdk_bdev_channel *bdev_ch[2];
739 	struct spdk_bdev *bdev;
740 	enum spdk_bdev_io_status status0, status1, status2;
741 	int rc;
742 
743 	setup_test();
744 	MOCK_SET(spdk_get_ticks, 0);
745 
746 	/* Enable QoS */
747 	bdev = &g_bdev.bdev;
748 	bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
749 	SPDK_CU_ASSERT_FATAL(bdev->internal.qos != NULL);
750 	TAILQ_INIT(&bdev->internal.qos->queued);
751 	/*
752 	 * Enable read/write IOPS, read only byte per sec, write only
753 	 * byte per sec and read/write byte per sec rate limits.
754 	 * In this case, both read only and write only byte per sec
755 	 * rate limit will take effect.
756 	 */
757 	/* 4000 read/write I/O per second, or 4 per millisecond */
758 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT].limit = 4000;
759 	/* 8K byte per millisecond with 4K block size */
760 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT].limit = 8192000;
761 	/* 4K byte per millisecond with 4K block size */
762 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_R_BPS_RATE_LIMIT].limit = 4096000;
763 	/* 4K byte per millisecond with 4K block size */
764 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_W_BPS_RATE_LIMIT].limit = 4096000;
765 
766 	g_get_io_channel = true;
767 
768 	/* Create channels */
769 	set_thread(0);
770 	io_ch[0] = spdk_bdev_get_io_channel(g_desc);
771 	bdev_ch[0] = spdk_io_channel_get_ctx(io_ch[0]);
772 	CU_ASSERT(bdev_ch[0]->flags == BDEV_CH_QOS_ENABLED);
773 
774 	set_thread(1);
775 	io_ch[1] = spdk_bdev_get_io_channel(g_desc);
776 	bdev_ch[1] = spdk_io_channel_get_ctx(io_ch[1]);
777 	CU_ASSERT(bdev_ch[1]->flags == BDEV_CH_QOS_ENABLED);
778 
779 	/* Send two read I/Os */
780 	status1 = SPDK_BDEV_IO_STATUS_PENDING;
781 	rc = spdk_bdev_read_blocks(g_desc, io_ch[1], NULL, 0, 1, io_during_io_done, &status1);
782 	CU_ASSERT(rc == 0);
783 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_PENDING);
784 	set_thread(0);
785 	status0 = SPDK_BDEV_IO_STATUS_PENDING;
786 	rc = spdk_bdev_read_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &status0);
787 	CU_ASSERT(rc == 0);
788 	CU_ASSERT(status0 == SPDK_BDEV_IO_STATUS_PENDING);
789 	/* Send one write I/O */
790 	status2 = SPDK_BDEV_IO_STATUS_PENDING;
791 	rc = spdk_bdev_write_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &status2);
792 	CU_ASSERT(rc == 0);
793 	CU_ASSERT(status2 == SPDK_BDEV_IO_STATUS_PENDING);
794 
795 	/* Complete any I/O that arrived at the disk */
796 	poll_threads();
797 	set_thread(1);
798 	stub_complete_io(g_bdev.io_target, 0);
799 	set_thread(0);
800 	stub_complete_io(g_bdev.io_target, 0);
801 	poll_threads();
802 
803 	/* Only one of the two read I/Os should complete. (logical XOR) */
804 	if (status0 == SPDK_BDEV_IO_STATUS_SUCCESS) {
805 		CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_PENDING);
806 	} else {
807 		CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_SUCCESS);
808 	}
809 	/* The write I/O should complete. */
810 	CU_ASSERT(status2 == SPDK_BDEV_IO_STATUS_SUCCESS);
811 
812 	/* Advance in time by a millisecond */
813 	spdk_delay_us(1000);
814 
815 	/* Complete more I/O */
816 	poll_threads();
817 	set_thread(1);
818 	stub_complete_io(g_bdev.io_target, 0);
819 	set_thread(0);
820 	stub_complete_io(g_bdev.io_target, 0);
821 	poll_threads();
822 
823 	/* Now the second read I/O should be done */
824 	CU_ASSERT(status0 == SPDK_BDEV_IO_STATUS_SUCCESS);
825 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_SUCCESS);
826 
827 	/* Tear down the channels */
828 	set_thread(1);
829 	spdk_put_io_channel(io_ch[1]);
830 	set_thread(0);
831 	spdk_put_io_channel(io_ch[0]);
832 	poll_threads();
833 
834 	teardown_test();
835 }
836 
837 static void
838 io_during_qos_reset(void)
839 {
840 	struct spdk_io_channel *io_ch[2];
841 	struct spdk_bdev_channel *bdev_ch[2];
842 	struct spdk_bdev *bdev;
843 	enum spdk_bdev_io_status status0, status1, reset_status;
844 	int rc;
845 
846 	setup_test();
847 	MOCK_SET(spdk_get_ticks, 0);
848 
849 	/* Enable QoS */
850 	bdev = &g_bdev.bdev;
851 	bdev->internal.qos = calloc(1, sizeof(*bdev->internal.qos));
852 	SPDK_CU_ASSERT_FATAL(bdev->internal.qos != NULL);
853 	TAILQ_INIT(&bdev->internal.qos->queued);
854 	/*
855 	 * Enable read/write IOPS, write only byte per sec and
856 	 * read/write byte per second rate limits.
857 	 * In this case, read/write byte per second rate limit will
858 	 * take effect first.
859 	 */
860 	/* 2000 read/write I/O per second, or 2 per millisecond */
861 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT].limit = 2000;
862 	/* 4K byte per millisecond with 4K block size */
863 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT].limit = 4096000;
864 	/* 8K byte per millisecond with 4K block size */
865 	bdev->internal.qos->rate_limits[SPDK_BDEV_QOS_W_BPS_RATE_LIMIT].limit = 8192000;
866 
867 	g_get_io_channel = true;
868 
869 	/* Create channels */
870 	set_thread(0);
871 	io_ch[0] = spdk_bdev_get_io_channel(g_desc);
872 	bdev_ch[0] = spdk_io_channel_get_ctx(io_ch[0]);
873 	CU_ASSERT(bdev_ch[0]->flags == BDEV_CH_QOS_ENABLED);
874 
875 	set_thread(1);
876 	io_ch[1] = spdk_bdev_get_io_channel(g_desc);
877 	bdev_ch[1] = spdk_io_channel_get_ctx(io_ch[1]);
878 	CU_ASSERT(bdev_ch[1]->flags == BDEV_CH_QOS_ENABLED);
879 
880 	/* Send two I/O. One of these gets queued by QoS. The other is sitting at the disk. */
881 	status1 = SPDK_BDEV_IO_STATUS_PENDING;
882 	rc = spdk_bdev_write_blocks(g_desc, io_ch[1], NULL, 0, 1, io_during_io_done, &status1);
883 	CU_ASSERT(rc == 0);
884 	set_thread(0);
885 	status0 = SPDK_BDEV_IO_STATUS_PENDING;
886 	rc = spdk_bdev_write_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &status0);
887 	CU_ASSERT(rc == 0);
888 
889 	poll_threads();
890 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_PENDING);
891 	CU_ASSERT(status0 == SPDK_BDEV_IO_STATUS_PENDING);
892 
893 	/* Reset the bdev. */
894 	reset_status = SPDK_BDEV_IO_STATUS_PENDING;
895 	rc = spdk_bdev_reset(g_desc, io_ch[0], io_during_io_done, &reset_status);
896 	CU_ASSERT(rc == 0);
897 
898 	/* Complete any I/O that arrived at the disk */
899 	poll_threads();
900 	set_thread(1);
901 	stub_complete_io(g_bdev.io_target, 0);
902 	set_thread(0);
903 	stub_complete_io(g_bdev.io_target, 0);
904 	poll_threads();
905 
906 	CU_ASSERT(reset_status == SPDK_BDEV_IO_STATUS_SUCCESS);
907 	CU_ASSERT(status0 == SPDK_BDEV_IO_STATUS_FAILED);
908 	CU_ASSERT(status1 == SPDK_BDEV_IO_STATUS_FAILED);
909 
910 	/* Tear down the channels */
911 	set_thread(1);
912 	spdk_put_io_channel(io_ch[1]);
913 	set_thread(0);
914 	spdk_put_io_channel(io_ch[0]);
915 	poll_threads();
916 
917 	teardown_test();
918 }
919 
920 static void
921 enomem_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
922 {
923 	enum spdk_bdev_io_status *status = cb_arg;
924 
925 	*status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED;
926 	spdk_bdev_free_io(bdev_io);
927 }
928 
929 static void
930 enomem(void)
931 {
932 	struct spdk_io_channel *io_ch;
933 	struct spdk_bdev_channel *bdev_ch;
934 	struct spdk_bdev_shared_resource *shared_resource;
935 	struct ut_bdev_channel *ut_ch;
936 	const uint32_t IO_ARRAY_SIZE = 64;
937 	const uint32_t AVAIL = 20;
938 	enum spdk_bdev_io_status status[IO_ARRAY_SIZE], status_reset;
939 	uint32_t nomem_cnt, i;
940 	struct spdk_bdev_io *first_io;
941 	int rc;
942 
943 	setup_test();
944 
945 	set_thread(0);
946 	io_ch = spdk_bdev_get_io_channel(g_desc);
947 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
948 	shared_resource = bdev_ch->shared_resource;
949 	ut_ch = spdk_io_channel_get_ctx(bdev_ch->channel);
950 	ut_ch->avail_cnt = AVAIL;
951 
952 	/* First submit a number of IOs equal to what the channel can support. */
953 	for (i = 0; i < AVAIL; i++) {
954 		status[i] = SPDK_BDEV_IO_STATUS_PENDING;
955 		rc = spdk_bdev_read_blocks(g_desc, io_ch, NULL, 0, 1, enomem_done, &status[i]);
956 		CU_ASSERT(rc == 0);
957 	}
958 	CU_ASSERT(TAILQ_EMPTY(&shared_resource->nomem_io));
959 
960 	/*
961 	 * Next, submit one additional I/O.  This one should fail with ENOMEM and then go onto
962 	 *  the enomem_io list.
963 	 */
964 	status[AVAIL] = SPDK_BDEV_IO_STATUS_PENDING;
965 	rc = spdk_bdev_read_blocks(g_desc, io_ch, NULL, 0, 1, enomem_done, &status[AVAIL]);
966 	CU_ASSERT(rc == 0);
967 	SPDK_CU_ASSERT_FATAL(!TAILQ_EMPTY(&shared_resource->nomem_io));
968 	first_io = TAILQ_FIRST(&shared_resource->nomem_io);
969 
970 	/*
971 	 * Now submit a bunch more I/O.  These should all fail with ENOMEM and get queued behind
972 	 *  the first_io above.
973 	 */
974 	for (i = AVAIL + 1; i < IO_ARRAY_SIZE; i++) {
975 		status[i] = SPDK_BDEV_IO_STATUS_PENDING;
976 		rc = spdk_bdev_read_blocks(g_desc, io_ch, NULL, 0, 1, enomem_done, &status[i]);
977 		CU_ASSERT(rc == 0);
978 	}
979 
980 	/* Assert that first_io is still at the head of the list. */
981 	CU_ASSERT(TAILQ_FIRST(&shared_resource->nomem_io) == first_io);
982 	CU_ASSERT(bdev_io_tailq_cnt(&shared_resource->nomem_io) == (IO_ARRAY_SIZE - AVAIL));
983 	nomem_cnt = bdev_io_tailq_cnt(&shared_resource->nomem_io);
984 	CU_ASSERT(shared_resource->nomem_threshold == (AVAIL - NOMEM_THRESHOLD_COUNT));
985 
986 	/*
987 	 * Complete 1 I/O only.  The key check here is bdev_io_tailq_cnt - this should not have
988 	 *  changed since completing just 1 I/O should not trigger retrying the queued nomem_io
989 	 *  list.
990 	 */
991 	stub_complete_io(g_bdev.io_target, 1);
992 	CU_ASSERT(bdev_io_tailq_cnt(&shared_resource->nomem_io) == nomem_cnt);
993 
994 	/*
995 	 * Complete enough I/O to hit the nomem_theshold.  This should trigger retrying nomem_io,
996 	 *  and we should see I/O get resubmitted to the test bdev module.
997 	 */
998 	stub_complete_io(g_bdev.io_target, NOMEM_THRESHOLD_COUNT - 1);
999 	CU_ASSERT(bdev_io_tailq_cnt(&shared_resource->nomem_io) < nomem_cnt);
1000 	nomem_cnt = bdev_io_tailq_cnt(&shared_resource->nomem_io);
1001 
1002 	/* Complete 1 I/O only.  This should not trigger retrying the queued nomem_io. */
1003 	stub_complete_io(g_bdev.io_target, 1);
1004 	CU_ASSERT(bdev_io_tailq_cnt(&shared_resource->nomem_io) == nomem_cnt);
1005 
1006 	/*
1007 	 * Send a reset and confirm that all I/O are completed, including the ones that
1008 	 *  were queued on the nomem_io list.
1009 	 */
1010 	status_reset = SPDK_BDEV_IO_STATUS_PENDING;
1011 	rc = spdk_bdev_reset(g_desc, io_ch, enomem_done, &status_reset);
1012 	poll_threads();
1013 	CU_ASSERT(rc == 0);
1014 	/* This will complete the reset. */
1015 	stub_complete_io(g_bdev.io_target, 0);
1016 
1017 	CU_ASSERT(bdev_io_tailq_cnt(&shared_resource->nomem_io) == 0);
1018 	CU_ASSERT(shared_resource->io_outstanding == 0);
1019 
1020 	spdk_put_io_channel(io_ch);
1021 	poll_threads();
1022 	teardown_test();
1023 }
1024 
1025 static void
1026 enomem_multi_bdev(void)
1027 {
1028 	struct spdk_io_channel *io_ch;
1029 	struct spdk_bdev_channel *bdev_ch;
1030 	struct spdk_bdev_shared_resource *shared_resource;
1031 	struct ut_bdev_channel *ut_ch;
1032 	const uint32_t IO_ARRAY_SIZE = 64;
1033 	const uint32_t AVAIL = 20;
1034 	enum spdk_bdev_io_status status[IO_ARRAY_SIZE];
1035 	uint32_t i;
1036 	struct ut_bdev *second_bdev;
1037 	struct spdk_bdev_desc *second_desc = NULL;
1038 	struct spdk_bdev_channel *second_bdev_ch;
1039 	struct spdk_io_channel *second_ch;
1040 	int rc;
1041 
1042 	setup_test();
1043 
1044 	/* Register second bdev with the same io_target  */
1045 	second_bdev = calloc(1, sizeof(*second_bdev));
1046 	SPDK_CU_ASSERT_FATAL(second_bdev != NULL);
1047 	register_bdev(second_bdev, "ut_bdev2", g_bdev.io_target);
1048 	spdk_bdev_open(&second_bdev->bdev, true, NULL, NULL, &second_desc);
1049 	SPDK_CU_ASSERT_FATAL(second_desc != NULL);
1050 
1051 	set_thread(0);
1052 	io_ch = spdk_bdev_get_io_channel(g_desc);
1053 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
1054 	shared_resource = bdev_ch->shared_resource;
1055 	ut_ch = spdk_io_channel_get_ctx(bdev_ch->channel);
1056 	ut_ch->avail_cnt = AVAIL;
1057 
1058 	second_ch = spdk_bdev_get_io_channel(second_desc);
1059 	second_bdev_ch = spdk_io_channel_get_ctx(second_ch);
1060 	SPDK_CU_ASSERT_FATAL(shared_resource == second_bdev_ch->shared_resource);
1061 
1062 	/* Saturate io_target through bdev A. */
1063 	for (i = 0; i < AVAIL; i++) {
1064 		status[i] = SPDK_BDEV_IO_STATUS_PENDING;
1065 		rc = spdk_bdev_read_blocks(g_desc, io_ch, NULL, 0, 1, enomem_done, &status[i]);
1066 		CU_ASSERT(rc == 0);
1067 	}
1068 	CU_ASSERT(TAILQ_EMPTY(&shared_resource->nomem_io));
1069 
1070 	/*
1071 	 * Now submit I/O through the second bdev. This should fail with ENOMEM
1072 	 * and then go onto the nomem_io list.
1073 	 */
1074 	status[AVAIL] = SPDK_BDEV_IO_STATUS_PENDING;
1075 	rc = spdk_bdev_read_blocks(second_desc, second_ch, NULL, 0, 1, enomem_done, &status[AVAIL]);
1076 	CU_ASSERT(rc == 0);
1077 	SPDK_CU_ASSERT_FATAL(!TAILQ_EMPTY(&shared_resource->nomem_io));
1078 
1079 	/* Complete first bdev's I/O. This should retry sending second bdev's nomem_io */
1080 	stub_complete_io(g_bdev.io_target, AVAIL);
1081 
1082 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&shared_resource->nomem_io));
1083 	CU_ASSERT(shared_resource->io_outstanding == 1);
1084 
1085 	/* Now complete our retried I/O  */
1086 	stub_complete_io(g_bdev.io_target, 1);
1087 	SPDK_CU_ASSERT_FATAL(shared_resource->io_outstanding == 0);
1088 
1089 	spdk_put_io_channel(io_ch);
1090 	spdk_put_io_channel(second_ch);
1091 	spdk_bdev_close(second_desc);
1092 	unregister_bdev(second_bdev);
1093 	poll_threads();
1094 	free(second_bdev);
1095 	teardown_test();
1096 }
1097 
1098 
1099 static void
1100 enomem_multi_io_target(void)
1101 {
1102 	struct spdk_io_channel *io_ch;
1103 	struct spdk_bdev_channel *bdev_ch;
1104 	struct ut_bdev_channel *ut_ch;
1105 	const uint32_t IO_ARRAY_SIZE = 64;
1106 	const uint32_t AVAIL = 20;
1107 	enum spdk_bdev_io_status status[IO_ARRAY_SIZE];
1108 	uint32_t i;
1109 	int new_io_device;
1110 	struct ut_bdev *second_bdev;
1111 	struct spdk_bdev_desc *second_desc = NULL;
1112 	struct spdk_bdev_channel *second_bdev_ch;
1113 	struct spdk_io_channel *second_ch;
1114 	int rc;
1115 
1116 	setup_test();
1117 
1118 	/* Create new io_target and a second bdev using it */
1119 	spdk_io_device_register(&new_io_device, stub_create_ch, stub_destroy_ch,
1120 				sizeof(struct ut_bdev_channel), NULL);
1121 	second_bdev = calloc(1, sizeof(*second_bdev));
1122 	SPDK_CU_ASSERT_FATAL(second_bdev != NULL);
1123 	register_bdev(second_bdev, "ut_bdev2", &new_io_device);
1124 	spdk_bdev_open(&second_bdev->bdev, true, NULL, NULL, &second_desc);
1125 	SPDK_CU_ASSERT_FATAL(second_desc != NULL);
1126 
1127 	set_thread(0);
1128 	io_ch = spdk_bdev_get_io_channel(g_desc);
1129 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
1130 	ut_ch = spdk_io_channel_get_ctx(bdev_ch->channel);
1131 	ut_ch->avail_cnt = AVAIL;
1132 
1133 	/* Different io_target should imply a different shared_resource */
1134 	second_ch = spdk_bdev_get_io_channel(second_desc);
1135 	second_bdev_ch = spdk_io_channel_get_ctx(second_ch);
1136 	SPDK_CU_ASSERT_FATAL(bdev_ch->shared_resource != second_bdev_ch->shared_resource);
1137 
1138 	/* Saturate io_target through bdev A. */
1139 	for (i = 0; i < AVAIL; i++) {
1140 		status[i] = SPDK_BDEV_IO_STATUS_PENDING;
1141 		rc = spdk_bdev_read_blocks(g_desc, io_ch, NULL, 0, 1, enomem_done, &status[i]);
1142 		CU_ASSERT(rc == 0);
1143 	}
1144 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->shared_resource->nomem_io));
1145 
1146 	/* Issue one more I/O to fill ENOMEM list. */
1147 	status[AVAIL] = SPDK_BDEV_IO_STATUS_PENDING;
1148 	rc = spdk_bdev_read_blocks(g_desc, io_ch, NULL, 0, 1, enomem_done, &status[AVAIL]);
1149 	CU_ASSERT(rc == 0);
1150 	SPDK_CU_ASSERT_FATAL(!TAILQ_EMPTY(&bdev_ch->shared_resource->nomem_io));
1151 
1152 	/*
1153 	 * Now submit I/O through the second bdev. This should go through and complete
1154 	 * successfully because we're using a different io_device underneath.
1155 	 */
1156 	status[AVAIL] = SPDK_BDEV_IO_STATUS_PENDING;
1157 	rc = spdk_bdev_read_blocks(second_desc, second_ch, NULL, 0, 1, enomem_done, &status[AVAIL]);
1158 	CU_ASSERT(rc == 0);
1159 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&second_bdev_ch->shared_resource->nomem_io));
1160 	stub_complete_io(second_bdev->io_target, 1);
1161 
1162 	/* Cleanup; Complete outstanding I/O. */
1163 	stub_complete_io(g_bdev.io_target, AVAIL);
1164 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev_ch->shared_resource->nomem_io));
1165 	/* Complete the ENOMEM I/O */
1166 	stub_complete_io(g_bdev.io_target, 1);
1167 	CU_ASSERT(bdev_ch->shared_resource->io_outstanding == 0);
1168 
1169 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev_ch->shared_resource->nomem_io));
1170 	CU_ASSERT(bdev_ch->shared_resource->io_outstanding == 0);
1171 	spdk_put_io_channel(io_ch);
1172 	spdk_put_io_channel(second_ch);
1173 	spdk_bdev_close(second_desc);
1174 	unregister_bdev(second_bdev);
1175 	spdk_io_device_unregister(&new_io_device, NULL);
1176 	poll_threads();
1177 	free(second_bdev);
1178 	teardown_test();
1179 }
1180 
1181 static void
1182 qos_dynamic_enable_done(void *cb_arg, int status)
1183 {
1184 	int *rc = cb_arg;
1185 	*rc = status;
1186 }
1187 
1188 static void
1189 qos_dynamic_enable(void)
1190 {
1191 	struct spdk_io_channel *io_ch[2];
1192 	struct spdk_bdev_channel *bdev_ch[2];
1193 	struct spdk_bdev *bdev;
1194 	enum spdk_bdev_io_status bdev_io_status[2];
1195 	uint64_t limits[SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES] = {};
1196 	int status, second_status, rc, i;
1197 
1198 	setup_test();
1199 	MOCK_SET(spdk_get_ticks, 0);
1200 
1201 	for (i = 0; i < SPDK_BDEV_QOS_NUM_RATE_LIMIT_TYPES; i++) {
1202 		limits[i] = UINT64_MAX;
1203 	}
1204 
1205 	bdev = &g_bdev.bdev;
1206 
1207 	g_get_io_channel = true;
1208 
1209 	/* Create channels */
1210 	set_thread(0);
1211 	io_ch[0] = spdk_bdev_get_io_channel(g_desc);
1212 	bdev_ch[0] = spdk_io_channel_get_ctx(io_ch[0]);
1213 	CU_ASSERT(bdev_ch[0]->flags == 0);
1214 
1215 	set_thread(1);
1216 	io_ch[1] = spdk_bdev_get_io_channel(g_desc);
1217 	bdev_ch[1] = spdk_io_channel_get_ctx(io_ch[1]);
1218 	CU_ASSERT(bdev_ch[1]->flags == 0);
1219 
1220 	set_thread(0);
1221 
1222 	/*
1223 	 * Enable QoS: Read/Write IOPS, Read/Write byte,
1224 	 * Read only byte and Write only byte per second
1225 	 * rate limits.
1226 	 * More than 10 I/Os allowed per timeslice.
1227 	 */
1228 	status = -1;
1229 	limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT] = 10000;
1230 	limits[SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT] = 100;
1231 	limits[SPDK_BDEV_QOS_R_BPS_RATE_LIMIT] = 100;
1232 	limits[SPDK_BDEV_QOS_W_BPS_RATE_LIMIT] = 10;
1233 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &status);
1234 	poll_threads();
1235 	CU_ASSERT(status == 0);
1236 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) != 0);
1237 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) != 0);
1238 
1239 	/*
1240 	 * Submit and complete 10 I/O to fill the QoS allotment for this timeslice.
1241 	 * Additional I/O will then be queued.
1242 	 */
1243 	set_thread(0);
1244 	for (i = 0; i < 10; i++) {
1245 		bdev_io_status[0] = SPDK_BDEV_IO_STATUS_PENDING;
1246 		rc = spdk_bdev_read_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &bdev_io_status[0]);
1247 		CU_ASSERT(rc == 0);
1248 		CU_ASSERT(bdev_io_status[0] == SPDK_BDEV_IO_STATUS_PENDING);
1249 		poll_thread(0);
1250 		stub_complete_io(g_bdev.io_target, 0);
1251 		CU_ASSERT(bdev_io_status[0] == SPDK_BDEV_IO_STATUS_SUCCESS);
1252 	}
1253 
1254 	/*
1255 	 * Send two more I/O.  These I/O will be queued since the current timeslice allotment has been
1256 	 * filled already.  We want to test that when QoS is disabled that these two I/O:
1257 	 *  1) are not aborted
1258 	 *  2) are sent back to their original thread for resubmission
1259 	 */
1260 	bdev_io_status[0] = SPDK_BDEV_IO_STATUS_PENDING;
1261 	rc = spdk_bdev_read_blocks(g_desc, io_ch[0], NULL, 0, 1, io_during_io_done, &bdev_io_status[0]);
1262 	CU_ASSERT(rc == 0);
1263 	CU_ASSERT(bdev_io_status[0] == SPDK_BDEV_IO_STATUS_PENDING);
1264 	set_thread(1);
1265 	bdev_io_status[1] = SPDK_BDEV_IO_STATUS_PENDING;
1266 	rc = spdk_bdev_read_blocks(g_desc, io_ch[1], NULL, 0, 1, io_during_io_done, &bdev_io_status[1]);
1267 	CU_ASSERT(rc == 0);
1268 	CU_ASSERT(bdev_io_status[1] == SPDK_BDEV_IO_STATUS_PENDING);
1269 	poll_threads();
1270 
1271 	/*
1272 	 * Disable QoS: Read/Write IOPS, Read/Write byte,
1273 	 * Read only byte rate limits
1274 	 */
1275 	status = -1;
1276 	limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT] = 0;
1277 	limits[SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT] = 0;
1278 	limits[SPDK_BDEV_QOS_R_BPS_RATE_LIMIT] = 0;
1279 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &status);
1280 	poll_threads();
1281 	CU_ASSERT(status == 0);
1282 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) != 0);
1283 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) != 0);
1284 
1285 	/* Disable QoS: Write only Byte per second rate limit */
1286 	status = -1;
1287 	limits[SPDK_BDEV_QOS_W_BPS_RATE_LIMIT] = 0;
1288 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &status);
1289 	poll_threads();
1290 	CU_ASSERT(status == 0);
1291 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) == 0);
1292 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) == 0);
1293 
1294 	/*
1295 	 * All I/O should have been resubmitted back on their original thread.  Complete
1296 	 *  all I/O on thread 0, and ensure that only the thread 0 I/O was completed.
1297 	 */
1298 	set_thread(0);
1299 	stub_complete_io(g_bdev.io_target, 0);
1300 	poll_threads();
1301 	CU_ASSERT(bdev_io_status[0] == SPDK_BDEV_IO_STATUS_SUCCESS);
1302 	CU_ASSERT(bdev_io_status[1] == SPDK_BDEV_IO_STATUS_PENDING);
1303 
1304 	/* Now complete all I/O on thread 1 and ensure the thread 1 I/O was completed. */
1305 	set_thread(1);
1306 	stub_complete_io(g_bdev.io_target, 0);
1307 	poll_threads();
1308 	CU_ASSERT(bdev_io_status[1] == SPDK_BDEV_IO_STATUS_SUCCESS);
1309 
1310 	/* Disable QoS again */
1311 	status = -1;
1312 	limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT] = 0;
1313 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &status);
1314 	poll_threads();
1315 	CU_ASSERT(status == 0); /* This should succeed */
1316 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) == 0);
1317 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) == 0);
1318 
1319 	/* Enable QoS on thread 0 */
1320 	status = -1;
1321 	limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT] = 10000;
1322 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &status);
1323 	poll_threads();
1324 	CU_ASSERT(status == 0);
1325 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) != 0);
1326 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) != 0);
1327 
1328 	/* Disable QoS on thread 1 */
1329 	set_thread(1);
1330 	status = -1;
1331 	limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT] = 0;
1332 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &status);
1333 	/* Don't poll yet. This should leave the channels with QoS enabled */
1334 	CU_ASSERT(status == -1);
1335 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) != 0);
1336 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) != 0);
1337 
1338 	/* Enable QoS. This should immediately fail because the previous disable QoS hasn't completed. */
1339 	second_status = 0;
1340 	limits[SPDK_BDEV_QOS_RW_BPS_RATE_LIMIT] = 10;
1341 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &second_status);
1342 	poll_threads();
1343 	CU_ASSERT(status == 0); /* The disable should succeed */
1344 	CU_ASSERT(second_status < 0); /* The enable should fail */
1345 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) == 0);
1346 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) == 0);
1347 
1348 	/* Enable QoS on thread 1. This should succeed now that the disable has completed. */
1349 	status = -1;
1350 	limits[SPDK_BDEV_QOS_RW_IOPS_RATE_LIMIT] = 10000;
1351 	spdk_bdev_set_qos_rate_limits(bdev, limits, qos_dynamic_enable_done, &status);
1352 	poll_threads();
1353 	CU_ASSERT(status == 0);
1354 	CU_ASSERT((bdev_ch[0]->flags & BDEV_CH_QOS_ENABLED) != 0);
1355 	CU_ASSERT((bdev_ch[1]->flags & BDEV_CH_QOS_ENABLED) != 0);
1356 
1357 	/* Tear down the channels */
1358 	set_thread(0);
1359 	spdk_put_io_channel(io_ch[0]);
1360 	set_thread(1);
1361 	spdk_put_io_channel(io_ch[1]);
1362 	poll_threads();
1363 
1364 	set_thread(0);
1365 	teardown_test();
1366 }
1367 
1368 static void
1369 histogram_status_cb(void *cb_arg, int status)
1370 {
1371 	g_status = status;
1372 }
1373 
1374 static void
1375 histogram_data_cb(void *cb_arg, int status, struct spdk_histogram_data *histogram)
1376 {
1377 	g_status = status;
1378 	g_histogram = histogram;
1379 }
1380 
1381 static void
1382 histogram_io_count(void *ctx, uint64_t start, uint64_t end, uint64_t count,
1383 		   uint64_t total, uint64_t so_far)
1384 {
1385 	g_count += count;
1386 }
1387 
1388 static void
1389 bdev_histograms_mt(void)
1390 {
1391 	struct spdk_io_channel *ch[2];
1392 	struct spdk_histogram_data *histogram;
1393 	uint8_t buf[4096];
1394 	int status = false;
1395 	int rc;
1396 
1397 
1398 	setup_test();
1399 
1400 	set_thread(0);
1401 	ch[0] = spdk_bdev_get_io_channel(g_desc);
1402 	CU_ASSERT(ch[0] != NULL);
1403 
1404 	set_thread(1);
1405 	ch[1] = spdk_bdev_get_io_channel(g_desc);
1406 	CU_ASSERT(ch[1] != NULL);
1407 
1408 
1409 	/* Enable histogram */
1410 	spdk_bdev_histogram_enable(&g_bdev.bdev, histogram_status_cb, NULL, true);
1411 	poll_threads();
1412 	CU_ASSERT(g_status == 0);
1413 	CU_ASSERT(g_bdev.bdev.internal.histogram_enabled == true);
1414 
1415 	/* Allocate histogram */
1416 	histogram = spdk_histogram_data_alloc();
1417 
1418 	/* Check if histogram is zeroed */
1419 	spdk_bdev_histogram_get(&g_bdev.bdev, histogram, histogram_data_cb, NULL);
1420 	poll_threads();
1421 	CU_ASSERT(g_status == 0);
1422 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
1423 
1424 	g_count = 0;
1425 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
1426 
1427 	CU_ASSERT(g_count == 0);
1428 
1429 	set_thread(0);
1430 	rc = spdk_bdev_write_blocks(g_desc, ch[0], &buf, 0, 1, io_during_io_done, &status);
1431 	CU_ASSERT(rc == 0);
1432 
1433 	spdk_delay_us(10);
1434 	stub_complete_io(g_bdev.io_target, 1);
1435 	poll_threads();
1436 	CU_ASSERT(status == true);
1437 
1438 
1439 	set_thread(1);
1440 	rc = spdk_bdev_read_blocks(g_desc, ch[1], &buf, 0, 1, io_during_io_done, &status);
1441 	CU_ASSERT(rc == 0);
1442 
1443 	spdk_delay_us(10);
1444 	stub_complete_io(g_bdev.io_target, 1);
1445 	poll_threads();
1446 	CU_ASSERT(status == true);
1447 
1448 	set_thread(0);
1449 
1450 	/* Check if histogram gathered data from all I/O channels */
1451 	spdk_bdev_histogram_get(&g_bdev.bdev, histogram, histogram_data_cb, NULL);
1452 	poll_threads();
1453 	CU_ASSERT(g_status == 0);
1454 	CU_ASSERT(g_bdev.bdev.internal.histogram_enabled == true);
1455 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
1456 
1457 	g_count = 0;
1458 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
1459 	CU_ASSERT(g_count == 2);
1460 
1461 	/* Disable histogram */
1462 	spdk_bdev_histogram_enable(&g_bdev.bdev, histogram_status_cb, NULL, false);
1463 	poll_threads();
1464 	CU_ASSERT(g_status == 0);
1465 	CU_ASSERT(g_bdev.bdev.internal.histogram_enabled == false);
1466 
1467 	spdk_histogram_data_free(g_histogram);
1468 }
1469 
1470 int
1471 main(int argc, char **argv)
1472 {
1473 	CU_pSuite	suite = NULL;
1474 	unsigned int	num_failures;
1475 
1476 	if (CU_initialize_registry() != CUE_SUCCESS) {
1477 		return CU_get_error();
1478 	}
1479 
1480 	suite = CU_add_suite("bdev", NULL, NULL);
1481 	if (suite == NULL) {
1482 		CU_cleanup_registry();
1483 		return CU_get_error();
1484 	}
1485 
1486 	if (
1487 		CU_add_test(suite, "basic", basic) == NULL ||
1488 		CU_add_test(suite, "unregister_and_close", unregister_and_close) == NULL ||
1489 		CU_add_test(suite, "basic_qos", basic_qos) == NULL ||
1490 		CU_add_test(suite, "put_channel_during_reset", put_channel_during_reset) == NULL ||
1491 		CU_add_test(suite, "aborted_reset", aborted_reset) == NULL ||
1492 		CU_add_test(suite, "io_during_reset", io_during_reset) == NULL ||
1493 		CU_add_test(suite, "io_during_qos_queue", io_during_qos_queue) == NULL ||
1494 		CU_add_test(suite, "io_during_qos_reset", io_during_qos_reset) == NULL ||
1495 		CU_add_test(suite, "enomem", enomem) == NULL ||
1496 		CU_add_test(suite, "enomem_multi_bdev", enomem_multi_bdev) == NULL ||
1497 		CU_add_test(suite, "enomem_multi_io_target", enomem_multi_io_target) == NULL ||
1498 		CU_add_test(suite, "qos_dynamic_enable", qos_dynamic_enable) == NULL ||
1499 		CU_add_test(suite, "bdev_histograms_mt", bdev_histograms_mt) == NULL
1500 	) {
1501 		CU_cleanup_registry();
1502 		return CU_get_error();
1503 	}
1504 
1505 	CU_basic_set_mode(CU_BRM_VERBOSE);
1506 	CU_basic_run_tests();
1507 	num_failures = CU_get_number_of_failures();
1508 	CU_cleanup_registry();
1509 	return num_failures;
1510 }
1511