xref: /spdk/test/unit/lib/bdev/bdev.c/bdev_ut.c (revision 32999ab917f67af61872f868585fd3d78ad6fb8a)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019 Mellanox Technologies LTD. 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 struct spdk_trace_histories *g_trace_histories;
46 DEFINE_STUB_V(spdk_trace_add_register_fn, (struct spdk_trace_register_fn *reg_fn));
47 DEFINE_STUB_V(spdk_trace_register_owner, (uint8_t type, char id_prefix));
48 DEFINE_STUB_V(spdk_trace_register_object, (uint8_t type, char id_prefix));
49 DEFINE_STUB_V(spdk_trace_register_description, (const char *name,
50 		uint16_t tpoint_id, uint8_t owner_type,
51 		uint8_t object_type, uint8_t new_object,
52 		uint8_t arg1_type, const char *arg1_name));
53 DEFINE_STUB_V(_spdk_trace_record, (uint64_t tsc, uint16_t tpoint_id, uint16_t poller_id,
54 				   uint32_t size, uint64_t object_id, uint64_t arg1));
55 DEFINE_STUB(spdk_notify_send, uint64_t, (const char *type, const char *ctx), 0);
56 DEFINE_STUB(spdk_notify_type_register, struct spdk_notify_type *, (const char *type), NULL);
57 
58 
59 int g_status;
60 int g_count;
61 enum spdk_bdev_event_type g_event_type1;
62 enum spdk_bdev_event_type g_event_type2;
63 struct spdk_histogram_data *g_histogram;
64 void *g_unregister_arg;
65 int g_unregister_rc;
66 
67 void
68 spdk_scsi_nvme_translate(const struct spdk_bdev_io *bdev_io,
69 			 int *sc, int *sk, int *asc, int *ascq)
70 {
71 }
72 
73 static int
74 null_init(void)
75 {
76 	return 0;
77 }
78 
79 static int
80 null_clean(void)
81 {
82 	return 0;
83 }
84 
85 static int
86 stub_destruct(void *ctx)
87 {
88 	return 0;
89 }
90 
91 struct ut_expected_io {
92 	uint8_t				type;
93 	uint64_t			offset;
94 	uint64_t			length;
95 	int				iovcnt;
96 	struct iovec			iov[BDEV_IO_NUM_CHILD_IOV];
97 	void				*md_buf;
98 	TAILQ_ENTRY(ut_expected_io)	link;
99 };
100 
101 struct bdev_ut_channel {
102 	TAILQ_HEAD(, spdk_bdev_io)	outstanding_io;
103 	uint32_t			outstanding_io_count;
104 	TAILQ_HEAD(, ut_expected_io)	expected_io;
105 };
106 
107 static bool g_io_done;
108 static struct spdk_bdev_io *g_bdev_io;
109 static enum spdk_bdev_io_status g_io_status;
110 static enum spdk_bdev_io_status g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
111 static uint32_t g_bdev_ut_io_device;
112 static struct bdev_ut_channel *g_bdev_ut_channel;
113 static void *g_compare_read_buf;
114 static uint32_t g_compare_read_buf_len;
115 static void *g_compare_write_buf;
116 static uint32_t g_compare_write_buf_len;
117 static bool g_abort_done;
118 static enum spdk_bdev_io_status g_abort_status;
119 
120 static struct ut_expected_io *
121 ut_alloc_expected_io(uint8_t type, uint64_t offset, uint64_t length, int iovcnt)
122 {
123 	struct ut_expected_io *expected_io;
124 
125 	expected_io = calloc(1, sizeof(*expected_io));
126 	SPDK_CU_ASSERT_FATAL(expected_io != NULL);
127 
128 	expected_io->type = type;
129 	expected_io->offset = offset;
130 	expected_io->length = length;
131 	expected_io->iovcnt = iovcnt;
132 
133 	return expected_io;
134 }
135 
136 static void
137 ut_expected_io_set_iov(struct ut_expected_io *expected_io, int pos, void *base, size_t len)
138 {
139 	expected_io->iov[pos].iov_base = base;
140 	expected_io->iov[pos].iov_len = len;
141 }
142 
143 static void
144 stub_submit_request(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
145 {
146 	struct bdev_ut_channel *ch = spdk_io_channel_get_ctx(_ch);
147 	struct ut_expected_io *expected_io;
148 	struct iovec *iov, *expected_iov;
149 	struct spdk_bdev_io *bio_to_abort;
150 	int i;
151 
152 	g_bdev_io = bdev_io;
153 
154 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
155 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
156 
157 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
158 		CU_ASSERT(g_compare_read_buf_len == len);
159 		memcpy(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len);
160 	}
161 
162 	if (g_compare_write_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
163 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
164 
165 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
166 		CU_ASSERT(g_compare_write_buf_len == len);
167 		memcpy(g_compare_write_buf, bdev_io->u.bdev.iovs[0].iov_base, len);
168 	}
169 
170 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_COMPARE) {
171 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
172 
173 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
174 		CU_ASSERT(g_compare_read_buf_len == len);
175 		if (memcmp(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len)) {
176 			g_io_exp_status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
177 		}
178 	}
179 
180 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) {
181 		if (g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS) {
182 			TAILQ_FOREACH(bio_to_abort, &ch->outstanding_io, module_link) {
183 				if (bio_to_abort == bdev_io->u.abort.bio_to_abort) {
184 					TAILQ_REMOVE(&ch->outstanding_io, bio_to_abort, module_link);
185 					ch->outstanding_io_count--;
186 					spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_FAILED);
187 					break;
188 				}
189 			}
190 		}
191 	}
192 
193 	TAILQ_INSERT_TAIL(&ch->outstanding_io, bdev_io, module_link);
194 	ch->outstanding_io_count++;
195 
196 	expected_io = TAILQ_FIRST(&ch->expected_io);
197 	if (expected_io == NULL) {
198 		return;
199 	}
200 	TAILQ_REMOVE(&ch->expected_io, expected_io, link);
201 
202 	if (expected_io->type != SPDK_BDEV_IO_TYPE_INVALID) {
203 		CU_ASSERT(bdev_io->type == expected_io->type);
204 	}
205 
206 	if (expected_io->md_buf != NULL) {
207 		CU_ASSERT(expected_io->md_buf == bdev_io->u.bdev.md_buf);
208 	}
209 
210 	if (expected_io->length == 0) {
211 		free(expected_io);
212 		return;
213 	}
214 
215 	CU_ASSERT(expected_io->offset == bdev_io->u.bdev.offset_blocks);
216 	CU_ASSERT(expected_io->length = bdev_io->u.bdev.num_blocks);
217 
218 	if (expected_io->iovcnt == 0) {
219 		free(expected_io);
220 		/* UNMAP, WRITE_ZEROES and FLUSH don't have iovs, so we can just return now. */
221 		return;
222 	}
223 
224 	CU_ASSERT(expected_io->iovcnt == bdev_io->u.bdev.iovcnt);
225 	for (i = 0; i < expected_io->iovcnt; i++) {
226 		iov = &bdev_io->u.bdev.iovs[i];
227 		expected_iov = &expected_io->iov[i];
228 		CU_ASSERT(iov->iov_len == expected_iov->iov_len);
229 		CU_ASSERT(iov->iov_base == expected_iov->iov_base);
230 	}
231 
232 	free(expected_io);
233 }
234 
235 static void
236 stub_submit_request_get_buf_cb(struct spdk_io_channel *_ch,
237 			       struct spdk_bdev_io *bdev_io, bool success)
238 {
239 	CU_ASSERT(success == true);
240 
241 	stub_submit_request(_ch, bdev_io);
242 }
243 
244 static void
245 stub_submit_request_get_buf(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
246 {
247 	spdk_bdev_io_get_buf(bdev_io, stub_submit_request_get_buf_cb,
248 			     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
249 }
250 
251 static uint32_t
252 stub_complete_io(uint32_t num_to_complete)
253 {
254 	struct bdev_ut_channel *ch = g_bdev_ut_channel;
255 	struct spdk_bdev_io *bdev_io;
256 	static enum spdk_bdev_io_status io_status;
257 	uint32_t num_completed = 0;
258 
259 	while (num_completed < num_to_complete) {
260 		if (TAILQ_EMPTY(&ch->outstanding_io)) {
261 			break;
262 		}
263 		bdev_io = TAILQ_FIRST(&ch->outstanding_io);
264 		TAILQ_REMOVE(&ch->outstanding_io, bdev_io, module_link);
265 		ch->outstanding_io_count--;
266 		io_status = g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS ? SPDK_BDEV_IO_STATUS_SUCCESS :
267 			    g_io_exp_status;
268 		spdk_bdev_io_complete(bdev_io, io_status);
269 		num_completed++;
270 	}
271 
272 	return num_completed;
273 }
274 
275 static struct spdk_io_channel *
276 bdev_ut_get_io_channel(void *ctx)
277 {
278 	return spdk_get_io_channel(&g_bdev_ut_io_device);
279 }
280 
281 static bool g_io_types_supported[SPDK_BDEV_NUM_IO_TYPES] = {
282 	[SPDK_BDEV_IO_TYPE_READ]		= true,
283 	[SPDK_BDEV_IO_TYPE_WRITE]		= true,
284 	[SPDK_BDEV_IO_TYPE_COMPARE]		= true,
285 	[SPDK_BDEV_IO_TYPE_UNMAP]		= true,
286 	[SPDK_BDEV_IO_TYPE_FLUSH]		= true,
287 	[SPDK_BDEV_IO_TYPE_RESET]		= true,
288 	[SPDK_BDEV_IO_TYPE_NVME_ADMIN]		= true,
289 	[SPDK_BDEV_IO_TYPE_NVME_IO]		= true,
290 	[SPDK_BDEV_IO_TYPE_NVME_IO_MD]		= true,
291 	[SPDK_BDEV_IO_TYPE_WRITE_ZEROES]	= true,
292 	[SPDK_BDEV_IO_TYPE_ZCOPY]		= true,
293 	[SPDK_BDEV_IO_TYPE_ABORT]		= true,
294 };
295 
296 static void
297 ut_enable_io_type(enum spdk_bdev_io_type io_type, bool enable)
298 {
299 	g_io_types_supported[io_type] = enable;
300 }
301 
302 static bool
303 stub_io_type_supported(void *_bdev, enum spdk_bdev_io_type io_type)
304 {
305 	return g_io_types_supported[io_type];
306 }
307 
308 static struct spdk_bdev_fn_table fn_table = {
309 	.destruct = stub_destruct,
310 	.submit_request = stub_submit_request,
311 	.get_io_channel = bdev_ut_get_io_channel,
312 	.io_type_supported = stub_io_type_supported,
313 };
314 
315 static int
316 bdev_ut_create_ch(void *io_device, void *ctx_buf)
317 {
318 	struct bdev_ut_channel *ch = ctx_buf;
319 
320 	CU_ASSERT(g_bdev_ut_channel == NULL);
321 	g_bdev_ut_channel = ch;
322 
323 	TAILQ_INIT(&ch->outstanding_io);
324 	ch->outstanding_io_count = 0;
325 	TAILQ_INIT(&ch->expected_io);
326 	return 0;
327 }
328 
329 static void
330 bdev_ut_destroy_ch(void *io_device, void *ctx_buf)
331 {
332 	CU_ASSERT(g_bdev_ut_channel != NULL);
333 	g_bdev_ut_channel = NULL;
334 }
335 
336 struct spdk_bdev_module bdev_ut_if;
337 
338 static int
339 bdev_ut_module_init(void)
340 {
341 	spdk_io_device_register(&g_bdev_ut_io_device, bdev_ut_create_ch, bdev_ut_destroy_ch,
342 				sizeof(struct bdev_ut_channel), NULL);
343 	spdk_bdev_module_init_done(&bdev_ut_if);
344 	return 0;
345 }
346 
347 static void
348 bdev_ut_module_fini(void)
349 {
350 	spdk_io_device_unregister(&g_bdev_ut_io_device, NULL);
351 }
352 
353 struct spdk_bdev_module bdev_ut_if = {
354 	.name = "bdev_ut",
355 	.module_init = bdev_ut_module_init,
356 	.module_fini = bdev_ut_module_fini,
357 	.async_init = true,
358 };
359 
360 static void vbdev_ut_examine(struct spdk_bdev *bdev);
361 
362 static int
363 vbdev_ut_module_init(void)
364 {
365 	return 0;
366 }
367 
368 static void
369 vbdev_ut_module_fini(void)
370 {
371 }
372 
373 struct spdk_bdev_module vbdev_ut_if = {
374 	.name = "vbdev_ut",
375 	.module_init = vbdev_ut_module_init,
376 	.module_fini = vbdev_ut_module_fini,
377 	.examine_config = vbdev_ut_examine,
378 };
379 
380 SPDK_BDEV_MODULE_REGISTER(bdev_ut, &bdev_ut_if)
381 SPDK_BDEV_MODULE_REGISTER(vbdev_ut, &vbdev_ut_if)
382 
383 static void
384 vbdev_ut_examine(struct spdk_bdev *bdev)
385 {
386 	spdk_bdev_module_examine_done(&vbdev_ut_if);
387 }
388 
389 static struct spdk_bdev *
390 allocate_bdev(char *name)
391 {
392 	struct spdk_bdev *bdev;
393 	int rc;
394 
395 	bdev = calloc(1, sizeof(*bdev));
396 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
397 
398 	bdev->name = name;
399 	bdev->fn_table = &fn_table;
400 	bdev->module = &bdev_ut_if;
401 	bdev->blockcnt = 1024;
402 	bdev->blocklen = 512;
403 
404 	rc = spdk_bdev_register(bdev);
405 	CU_ASSERT(rc == 0);
406 
407 	return bdev;
408 }
409 
410 static struct spdk_bdev *
411 allocate_vbdev(char *name)
412 {
413 	struct spdk_bdev *bdev;
414 	int rc;
415 
416 	bdev = calloc(1, sizeof(*bdev));
417 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
418 
419 	bdev->name = name;
420 	bdev->fn_table = &fn_table;
421 	bdev->module = &vbdev_ut_if;
422 
423 	rc = spdk_bdev_register(bdev);
424 	CU_ASSERT(rc == 0);
425 
426 	return bdev;
427 }
428 
429 static void
430 free_bdev(struct spdk_bdev *bdev)
431 {
432 	spdk_bdev_unregister(bdev, NULL, NULL);
433 	poll_threads();
434 	memset(bdev, 0xFF, sizeof(*bdev));
435 	free(bdev);
436 }
437 
438 static void
439 free_vbdev(struct spdk_bdev *bdev)
440 {
441 	spdk_bdev_unregister(bdev, NULL, NULL);
442 	poll_threads();
443 	memset(bdev, 0xFF, sizeof(*bdev));
444 	free(bdev);
445 }
446 
447 static void
448 get_device_stat_cb(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat, void *cb_arg, int rc)
449 {
450 	const char *bdev_name;
451 
452 	CU_ASSERT(bdev != NULL);
453 	CU_ASSERT(rc == 0);
454 	bdev_name = spdk_bdev_get_name(bdev);
455 	CU_ASSERT_STRING_EQUAL(bdev_name, "bdev0");
456 
457 	free(stat);
458 
459 	*(bool *)cb_arg = true;
460 }
461 
462 static void
463 bdev_unregister_cb(void *cb_arg, int rc)
464 {
465 	g_unregister_arg = cb_arg;
466 	g_unregister_rc = rc;
467 }
468 
469 static void
470 bdev_ut_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
471 {
472 }
473 
474 static void
475 bdev_open_cb1(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
476 {
477 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
478 
479 	g_event_type1 = type;
480 	if (SPDK_BDEV_EVENT_REMOVE == type) {
481 		spdk_bdev_close(desc);
482 	}
483 }
484 
485 static void
486 bdev_open_cb2(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
487 {
488 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
489 
490 	g_event_type2 = type;
491 	if (SPDK_BDEV_EVENT_REMOVE == type) {
492 		spdk_bdev_close(desc);
493 	}
494 }
495 
496 static void
497 get_device_stat_test(void)
498 {
499 	struct spdk_bdev *bdev;
500 	struct spdk_bdev_io_stat *stat;
501 	bool done;
502 
503 	bdev = allocate_bdev("bdev0");
504 	stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
505 	if (stat == NULL) {
506 		free_bdev(bdev);
507 		return;
508 	}
509 
510 	done = false;
511 	spdk_bdev_get_device_stat(bdev, stat, get_device_stat_cb, &done);
512 	while (!done) { poll_threads(); }
513 
514 	free_bdev(bdev);
515 }
516 
517 static void
518 open_write_test(void)
519 {
520 	struct spdk_bdev *bdev[9];
521 	struct spdk_bdev_desc *desc[9] = {};
522 	int rc;
523 
524 	/*
525 	 * Create a tree of bdevs to test various open w/ write cases.
526 	 *
527 	 * bdev0 through bdev3 are physical block devices, such as NVMe
528 	 * namespaces or Ceph block devices.
529 	 *
530 	 * bdev4 is a virtual bdev with multiple base bdevs.  This models
531 	 * caching or RAID use cases.
532 	 *
533 	 * bdev5 through bdev7 are all virtual bdevs with the same base
534 	 * bdev (except bdev7). This models partitioning or logical volume
535 	 * use cases.
536 	 *
537 	 * bdev7 is a virtual bdev with multiple base bdevs. One of base bdevs
538 	 * (bdev2) is shared with other virtual bdevs: bdev5 and bdev6. This
539 	 * models caching, RAID, partitioning or logical volumes use cases.
540 	 *
541 	 * bdev8 is a virtual bdev with multiple base bdevs, but these
542 	 * base bdevs are themselves virtual bdevs.
543 	 *
544 	 *                bdev8
545 	 *                  |
546 	 *            +----------+
547 	 *            |          |
548 	 *          bdev4      bdev5   bdev6   bdev7
549 	 *            |          |       |       |
550 	 *        +---+---+      +---+   +   +---+---+
551 	 *        |       |           \  |  /         \
552 	 *      bdev0   bdev1          bdev2         bdev3
553 	 */
554 
555 	bdev[0] = allocate_bdev("bdev0");
556 	rc = spdk_bdev_module_claim_bdev(bdev[0], NULL, &bdev_ut_if);
557 	CU_ASSERT(rc == 0);
558 
559 	bdev[1] = allocate_bdev("bdev1");
560 	rc = spdk_bdev_module_claim_bdev(bdev[1], NULL, &bdev_ut_if);
561 	CU_ASSERT(rc == 0);
562 
563 	bdev[2] = allocate_bdev("bdev2");
564 	rc = spdk_bdev_module_claim_bdev(bdev[2], NULL, &bdev_ut_if);
565 	CU_ASSERT(rc == 0);
566 
567 	bdev[3] = allocate_bdev("bdev3");
568 	rc = spdk_bdev_module_claim_bdev(bdev[3], NULL, &bdev_ut_if);
569 	CU_ASSERT(rc == 0);
570 
571 	bdev[4] = allocate_vbdev("bdev4");
572 	rc = spdk_bdev_module_claim_bdev(bdev[4], NULL, &bdev_ut_if);
573 	CU_ASSERT(rc == 0);
574 
575 	bdev[5] = allocate_vbdev("bdev5");
576 	rc = spdk_bdev_module_claim_bdev(bdev[5], NULL, &bdev_ut_if);
577 	CU_ASSERT(rc == 0);
578 
579 	bdev[6] = allocate_vbdev("bdev6");
580 
581 	bdev[7] = allocate_vbdev("bdev7");
582 
583 	bdev[8] = allocate_vbdev("bdev8");
584 
585 	/* Open bdev0 read-only.  This should succeed. */
586 	rc = spdk_bdev_open_ext("bdev0", false, bdev_ut_event_cb, NULL, &desc[0]);
587 	CU_ASSERT(rc == 0);
588 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
589 	CU_ASSERT(bdev[0] == spdk_bdev_desc_get_bdev(desc[0]));
590 	spdk_bdev_close(desc[0]);
591 
592 	/*
593 	 * Open bdev1 read/write.  This should fail since bdev1 has been claimed
594 	 * by a vbdev module.
595 	 */
596 	rc = spdk_bdev_open_ext("bdev1", true, bdev_ut_event_cb, NULL, &desc[1]);
597 	CU_ASSERT(rc == -EPERM);
598 
599 	/*
600 	 * Open bdev4 read/write.  This should fail since bdev3 has been claimed
601 	 * by a vbdev module.
602 	 */
603 	rc = spdk_bdev_open_ext("bdev4", true, bdev_ut_event_cb, NULL, &desc[4]);
604 	CU_ASSERT(rc == -EPERM);
605 
606 	/* Open bdev4 read-only.  This should succeed. */
607 	rc = spdk_bdev_open_ext("bdev4", false, bdev_ut_event_cb, NULL, &desc[4]);
608 	CU_ASSERT(rc == 0);
609 	SPDK_CU_ASSERT_FATAL(desc[4] != NULL);
610 	CU_ASSERT(bdev[4] == spdk_bdev_desc_get_bdev(desc[4]));
611 	spdk_bdev_close(desc[4]);
612 
613 	/*
614 	 * Open bdev8 read/write.  This should succeed since it is a leaf
615 	 * bdev.
616 	 */
617 	rc = spdk_bdev_open_ext("bdev8", true, bdev_ut_event_cb, NULL, &desc[8]);
618 	CU_ASSERT(rc == 0);
619 	SPDK_CU_ASSERT_FATAL(desc[8] != NULL);
620 	CU_ASSERT(bdev[8] == spdk_bdev_desc_get_bdev(desc[8]));
621 	spdk_bdev_close(desc[8]);
622 
623 	/*
624 	 * Open bdev5 read/write.  This should fail since bdev4 has been claimed
625 	 * by a vbdev module.
626 	 */
627 	rc = spdk_bdev_open_ext("bdev5", true, bdev_ut_event_cb, NULL, &desc[5]);
628 	CU_ASSERT(rc == -EPERM);
629 
630 	/* Open bdev4 read-only.  This should succeed. */
631 	rc = spdk_bdev_open_ext("bdev5", false, bdev_ut_event_cb, NULL, &desc[5]);
632 	CU_ASSERT(rc == 0);
633 	SPDK_CU_ASSERT_FATAL(desc[5] != NULL);
634 	CU_ASSERT(bdev[5] == spdk_bdev_desc_get_bdev(desc[5]));
635 	spdk_bdev_close(desc[5]);
636 
637 	free_vbdev(bdev[8]);
638 
639 	free_vbdev(bdev[5]);
640 	free_vbdev(bdev[6]);
641 	free_vbdev(bdev[7]);
642 
643 	free_vbdev(bdev[4]);
644 
645 	free_bdev(bdev[0]);
646 	free_bdev(bdev[1]);
647 	free_bdev(bdev[2]);
648 	free_bdev(bdev[3]);
649 }
650 
651 static void
652 bytes_to_blocks_test(void)
653 {
654 	struct spdk_bdev bdev;
655 	uint64_t offset_blocks, num_blocks;
656 
657 	memset(&bdev, 0, sizeof(bdev));
658 
659 	bdev.blocklen = 512;
660 
661 	/* All parameters valid */
662 	offset_blocks = 0;
663 	num_blocks = 0;
664 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 1024, &num_blocks) == 0);
665 	CU_ASSERT(offset_blocks == 1);
666 	CU_ASSERT(num_blocks == 2);
667 
668 	/* Offset not a block multiple */
669 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 512, &num_blocks) != 0);
670 
671 	/* Length not a block multiple */
672 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 3, &num_blocks) != 0);
673 
674 	/* In case blocklen not the power of two */
675 	bdev.blocklen = 100;
676 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 200, &num_blocks) == 0);
677 	CU_ASSERT(offset_blocks == 1);
678 	CU_ASSERT(num_blocks == 2);
679 
680 	/* Offset not a block multiple */
681 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 100, &num_blocks) != 0);
682 
683 	/* Length not a block multiple */
684 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 3, &num_blocks) != 0);
685 }
686 
687 static void
688 num_blocks_test(void)
689 {
690 	struct spdk_bdev bdev;
691 	struct spdk_bdev_desc *desc = NULL;
692 	int rc;
693 
694 	memset(&bdev, 0, sizeof(bdev));
695 	bdev.name = "num_blocks";
696 	bdev.fn_table = &fn_table;
697 	bdev.module = &bdev_ut_if;
698 	spdk_bdev_register(&bdev);
699 	spdk_bdev_notify_blockcnt_change(&bdev, 50);
700 
701 	/* Growing block number */
702 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 70) == 0);
703 	/* Shrinking block number */
704 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 30) == 0);
705 
706 	rc = spdk_bdev_open_ext("num_blocks", false, bdev_open_cb1, &desc, &desc);
707 	CU_ASSERT(rc == 0);
708 	SPDK_CU_ASSERT_FATAL(desc != NULL);
709 	CU_ASSERT(&bdev == spdk_bdev_desc_get_bdev(desc));
710 
711 	/* Growing block number */
712 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 80) == 0);
713 	/* Shrinking block number */
714 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 20) != 0);
715 
716 	g_event_type1 = 0xFF;
717 	/* Growing block number */
718 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 90) == 0);
719 
720 	poll_threads();
721 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_RESIZE);
722 
723 	g_event_type1 = 0xFF;
724 	/* Growing block number and closing */
725 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 100) == 0);
726 
727 	spdk_bdev_close(desc);
728 	spdk_bdev_unregister(&bdev, NULL, NULL);
729 
730 	poll_threads();
731 
732 	/* Callback is not called for closed device */
733 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
734 }
735 
736 static void
737 io_valid_test(void)
738 {
739 	struct spdk_bdev bdev;
740 
741 	memset(&bdev, 0, sizeof(bdev));
742 
743 	bdev.blocklen = 512;
744 	CU_ASSERT(pthread_mutex_init(&bdev.internal.mutex, NULL) == 0);
745 
746 	spdk_bdev_notify_blockcnt_change(&bdev, 100);
747 
748 	/* All parameters valid */
749 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 1, 2) == true);
750 
751 	/* Last valid block */
752 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 1) == true);
753 
754 	/* Offset past end of bdev */
755 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 100, 1) == false);
756 
757 	/* Offset + length past end of bdev */
758 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 2) == false);
759 
760 	/* Offset near end of uint64_t range (2^64 - 1) */
761 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 18446744073709551615ULL, 1) == false);
762 
763 	CU_ASSERT(pthread_mutex_destroy(&bdev.internal.mutex) == 0);
764 }
765 
766 static void
767 alias_add_del_test(void)
768 {
769 	struct spdk_bdev *bdev[3];
770 	int rc;
771 
772 	/* Creating and registering bdevs */
773 	bdev[0] = allocate_bdev("bdev0");
774 	SPDK_CU_ASSERT_FATAL(bdev[0] != 0);
775 
776 	bdev[1] = allocate_bdev("bdev1");
777 	SPDK_CU_ASSERT_FATAL(bdev[1] != 0);
778 
779 	bdev[2] = allocate_bdev("bdev2");
780 	SPDK_CU_ASSERT_FATAL(bdev[2] != 0);
781 
782 	poll_threads();
783 
784 	/*
785 	 * Trying adding an alias identical to name.
786 	 * Alias is identical to name, so it can not be added to aliases list
787 	 */
788 	rc = spdk_bdev_alias_add(bdev[0], bdev[0]->name);
789 	CU_ASSERT(rc == -EEXIST);
790 
791 	/*
792 	 * Trying to add empty alias,
793 	 * this one should fail
794 	 */
795 	rc = spdk_bdev_alias_add(bdev[0], NULL);
796 	CU_ASSERT(rc == -EINVAL);
797 
798 	/* Trying adding same alias to two different registered bdevs */
799 
800 	/* Alias is used first time, so this one should pass */
801 	rc = spdk_bdev_alias_add(bdev[0], "proper alias 0");
802 	CU_ASSERT(rc == 0);
803 
804 	/* Alias was added to another bdev, so this one should fail */
805 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 0");
806 	CU_ASSERT(rc == -EEXIST);
807 
808 	/* Alias is used first time, so this one should pass */
809 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 1");
810 	CU_ASSERT(rc == 0);
811 
812 	/* Trying removing an alias from registered bdevs */
813 
814 	/* Alias is not on a bdev aliases list, so this one should fail */
815 	rc = spdk_bdev_alias_del(bdev[0], "not existing");
816 	CU_ASSERT(rc == -ENOENT);
817 
818 	/* Alias is present on a bdev aliases list, so this one should pass */
819 	rc = spdk_bdev_alias_del(bdev[0], "proper alias 0");
820 	CU_ASSERT(rc == 0);
821 
822 	/* Alias is present on a bdev aliases list, so this one should pass */
823 	rc = spdk_bdev_alias_del(bdev[1], "proper alias 1");
824 	CU_ASSERT(rc == 0);
825 
826 	/* Trying to remove name instead of alias, so this one should fail, name cannot be changed or removed */
827 	rc = spdk_bdev_alias_del(bdev[0], bdev[0]->name);
828 	CU_ASSERT(rc != 0);
829 
830 	/* Trying to del all alias from empty alias list */
831 	spdk_bdev_alias_del_all(bdev[2]);
832 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev[2]->aliases));
833 
834 	/* Trying to del all alias from non-empty alias list */
835 	rc = spdk_bdev_alias_add(bdev[2], "alias0");
836 	CU_ASSERT(rc == 0);
837 	rc = spdk_bdev_alias_add(bdev[2], "alias1");
838 	CU_ASSERT(rc == 0);
839 	spdk_bdev_alias_del_all(bdev[2]);
840 	CU_ASSERT(TAILQ_EMPTY(&bdev[2]->aliases));
841 
842 	/* Unregister and free bdevs */
843 	spdk_bdev_unregister(bdev[0], NULL, NULL);
844 	spdk_bdev_unregister(bdev[1], NULL, NULL);
845 	spdk_bdev_unregister(bdev[2], NULL, NULL);
846 
847 	poll_threads();
848 
849 	free(bdev[0]);
850 	free(bdev[1]);
851 	free(bdev[2]);
852 }
853 
854 static void
855 io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
856 {
857 	g_io_done = true;
858 	g_io_status = bdev_io->internal.status;
859 	spdk_bdev_free_io(bdev_io);
860 }
861 
862 static void
863 bdev_init_cb(void *arg, int rc)
864 {
865 	CU_ASSERT(rc == 0);
866 }
867 
868 static void
869 bdev_fini_cb(void *arg)
870 {
871 }
872 
873 struct bdev_ut_io_wait_entry {
874 	struct spdk_bdev_io_wait_entry	entry;
875 	struct spdk_io_channel		*io_ch;
876 	struct spdk_bdev_desc		*desc;
877 	bool				submitted;
878 };
879 
880 static void
881 io_wait_cb(void *arg)
882 {
883 	struct bdev_ut_io_wait_entry *entry = arg;
884 	int rc;
885 
886 	rc = spdk_bdev_read_blocks(entry->desc, entry->io_ch, NULL, 0, 1, io_done, NULL);
887 	CU_ASSERT(rc == 0);
888 	entry->submitted = true;
889 }
890 
891 static void
892 bdev_io_types_test(void)
893 {
894 	struct spdk_bdev *bdev;
895 	struct spdk_bdev_desc *desc = NULL;
896 	struct spdk_io_channel *io_ch;
897 	struct spdk_bdev_opts bdev_opts = {};
898 	int rc;
899 
900 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
901 	bdev_opts.bdev_io_pool_size = 4;
902 	bdev_opts.bdev_io_cache_size = 2;
903 
904 	rc = spdk_bdev_set_opts(&bdev_opts);
905 	CU_ASSERT(rc == 0);
906 	spdk_bdev_initialize(bdev_init_cb, NULL);
907 	poll_threads();
908 
909 	bdev = allocate_bdev("bdev0");
910 
911 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
912 	CU_ASSERT(rc == 0);
913 	poll_threads();
914 	SPDK_CU_ASSERT_FATAL(desc != NULL);
915 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
916 	io_ch = spdk_bdev_get_io_channel(desc);
917 	CU_ASSERT(io_ch != NULL);
918 
919 	/* WRITE and WRITE ZEROES are not supported */
920 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
921 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, false);
922 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 0, 128, io_done, NULL);
923 	CU_ASSERT(rc == -ENOTSUP);
924 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
925 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, true);
926 
927 	spdk_put_io_channel(io_ch);
928 	spdk_bdev_close(desc);
929 	free_bdev(bdev);
930 	spdk_bdev_finish(bdev_fini_cb, NULL);
931 	poll_threads();
932 }
933 
934 static void
935 bdev_io_wait_test(void)
936 {
937 	struct spdk_bdev *bdev;
938 	struct spdk_bdev_desc *desc = NULL;
939 	struct spdk_io_channel *io_ch;
940 	struct spdk_bdev_opts bdev_opts = {};
941 	struct bdev_ut_io_wait_entry io_wait_entry;
942 	struct bdev_ut_io_wait_entry io_wait_entry2;
943 	int rc;
944 
945 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
946 	bdev_opts.bdev_io_pool_size = 4;
947 	bdev_opts.bdev_io_cache_size = 2;
948 
949 	rc = spdk_bdev_set_opts(&bdev_opts);
950 	CU_ASSERT(rc == 0);
951 	spdk_bdev_initialize(bdev_init_cb, NULL);
952 	poll_threads();
953 
954 	bdev = allocate_bdev("bdev0");
955 
956 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
957 	CU_ASSERT(rc == 0);
958 	poll_threads();
959 	SPDK_CU_ASSERT_FATAL(desc != NULL);
960 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
961 	io_ch = spdk_bdev_get_io_channel(desc);
962 	CU_ASSERT(io_ch != NULL);
963 
964 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
965 	CU_ASSERT(rc == 0);
966 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
967 	CU_ASSERT(rc == 0);
968 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
969 	CU_ASSERT(rc == 0);
970 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
971 	CU_ASSERT(rc == 0);
972 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
973 
974 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
975 	CU_ASSERT(rc == -ENOMEM);
976 
977 	io_wait_entry.entry.bdev = bdev;
978 	io_wait_entry.entry.cb_fn = io_wait_cb;
979 	io_wait_entry.entry.cb_arg = &io_wait_entry;
980 	io_wait_entry.io_ch = io_ch;
981 	io_wait_entry.desc = desc;
982 	io_wait_entry.submitted = false;
983 	/* Cannot use the same io_wait_entry for two different calls. */
984 	memcpy(&io_wait_entry2, &io_wait_entry, sizeof(io_wait_entry));
985 	io_wait_entry2.entry.cb_arg = &io_wait_entry2;
986 
987 	/* Queue two I/O waits. */
988 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry.entry);
989 	CU_ASSERT(rc == 0);
990 	CU_ASSERT(io_wait_entry.submitted == false);
991 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry2.entry);
992 	CU_ASSERT(rc == 0);
993 	CU_ASSERT(io_wait_entry2.submitted == false);
994 
995 	stub_complete_io(1);
996 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
997 	CU_ASSERT(io_wait_entry.submitted == true);
998 	CU_ASSERT(io_wait_entry2.submitted == false);
999 
1000 	stub_complete_io(1);
1001 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1002 	CU_ASSERT(io_wait_entry2.submitted == true);
1003 
1004 	stub_complete_io(4);
1005 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1006 
1007 	spdk_put_io_channel(io_ch);
1008 	spdk_bdev_close(desc);
1009 	free_bdev(bdev);
1010 	spdk_bdev_finish(bdev_fini_cb, NULL);
1011 	poll_threads();
1012 }
1013 
1014 static void
1015 bdev_io_spans_split_test(void)
1016 {
1017 	struct spdk_bdev bdev;
1018 	struct spdk_bdev_io bdev_io;
1019 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV];
1020 
1021 	memset(&bdev, 0, sizeof(bdev));
1022 	bdev_io.u.bdev.iovs = iov;
1023 
1024 	bdev.optimal_io_boundary = 0;
1025 	bdev.max_segment_size = 0;
1026 	bdev.max_num_segments = 0;
1027 	bdev_io.bdev = &bdev;
1028 
1029 	/* bdev has no optimal_io_boundary and max_size set - so this should return false. */
1030 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1031 
1032 	bdev.split_on_optimal_io_boundary = true;
1033 	bdev.optimal_io_boundary = 32;
1034 	bdev_io.type = SPDK_BDEV_IO_TYPE_RESET;
1035 
1036 	/* RESETs are not based on LBAs - so this should return false. */
1037 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1038 
1039 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1040 	bdev_io.u.bdev.offset_blocks = 0;
1041 	bdev_io.u.bdev.num_blocks = 32;
1042 
1043 	/* This I/O run right up to, but does not cross, the boundary - so this should return false. */
1044 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1045 
1046 	bdev_io.u.bdev.num_blocks = 33;
1047 
1048 	/* This I/O spans a boundary. */
1049 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1050 
1051 	bdev_io.u.bdev.num_blocks = 32;
1052 	bdev.max_segment_size = 512 * 32;
1053 	bdev.max_num_segments = 1;
1054 	bdev_io.u.bdev.iovcnt = 1;
1055 	iov[0].iov_len = 512;
1056 
1057 	/* Does not cross and exceed max_size or max_segs */
1058 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1059 
1060 	bdev.split_on_optimal_io_boundary = false;
1061 	bdev.max_segment_size = 512;
1062 	bdev.max_num_segments = 1;
1063 	bdev_io.u.bdev.iovcnt = 2;
1064 
1065 	/* Exceed max_segs */
1066 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1067 
1068 	bdev.max_num_segments = 2;
1069 	iov[0].iov_len = 513;
1070 	iov[1].iov_len = 512;
1071 
1072 	/* Exceed max_sizes */
1073 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1074 }
1075 
1076 static void
1077 bdev_io_boundary_split_test(void)
1078 {
1079 	struct spdk_bdev *bdev;
1080 	struct spdk_bdev_desc *desc = NULL;
1081 	struct spdk_io_channel *io_ch;
1082 	struct spdk_bdev_opts bdev_opts = {};
1083 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1084 	struct ut_expected_io *expected_io;
1085 	void *md_buf = (void *)0xFF000000;
1086 	uint64_t i;
1087 	int rc;
1088 
1089 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1090 	bdev_opts.bdev_io_pool_size = 512;
1091 	bdev_opts.bdev_io_cache_size = 64;
1092 
1093 	rc = spdk_bdev_set_opts(&bdev_opts);
1094 	CU_ASSERT(rc == 0);
1095 	spdk_bdev_initialize(bdev_init_cb, NULL);
1096 
1097 	bdev = allocate_bdev("bdev0");
1098 
1099 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1100 	CU_ASSERT(rc == 0);
1101 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1102 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1103 	io_ch = spdk_bdev_get_io_channel(desc);
1104 	CU_ASSERT(io_ch != NULL);
1105 
1106 	bdev->optimal_io_boundary = 16;
1107 	bdev->split_on_optimal_io_boundary = false;
1108 
1109 	g_io_done = false;
1110 
1111 	/* First test that the I/O does not get split if split_on_optimal_io_boundary == false. */
1112 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 8, 1);
1113 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 8 * 512);
1114 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1115 
1116 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
1117 	CU_ASSERT(rc == 0);
1118 	CU_ASSERT(g_io_done == false);
1119 
1120 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1121 	stub_complete_io(1);
1122 	CU_ASSERT(g_io_done == true);
1123 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1124 
1125 	bdev->split_on_optimal_io_boundary = true;
1126 	bdev->md_interleave = false;
1127 	bdev->md_len = 8;
1128 
1129 	/* Now test that a single-vector command is split correctly.
1130 	 * Offset 14, length 8, payload 0xF000
1131 	 *  Child - Offset 14, length 2, payload 0xF000
1132 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
1133 	 *
1134 	 * Set up the expected values before calling spdk_bdev_read_blocks
1135 	 */
1136 	g_io_done = false;
1137 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
1138 	expected_io->md_buf = md_buf;
1139 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
1140 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1141 
1142 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
1143 	expected_io->md_buf = md_buf + 2 * 8;
1144 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
1145 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1146 
1147 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1148 	rc = spdk_bdev_read_blocks_with_md(desc, io_ch, (void *)0xF000, md_buf,
1149 					   14, 8, io_done, NULL);
1150 	CU_ASSERT(rc == 0);
1151 	CU_ASSERT(g_io_done == false);
1152 
1153 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1154 	stub_complete_io(2);
1155 	CU_ASSERT(g_io_done == true);
1156 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1157 
1158 	/* Now set up a more complex, multi-vector command that needs to be split,
1159 	 *  including splitting iovecs.
1160 	 */
1161 	iov[0].iov_base = (void *)0x10000;
1162 	iov[0].iov_len = 512;
1163 	iov[1].iov_base = (void *)0x20000;
1164 	iov[1].iov_len = 20 * 512;
1165 	iov[2].iov_base = (void *)0x30000;
1166 	iov[2].iov_len = 11 * 512;
1167 
1168 	g_io_done = false;
1169 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
1170 	expected_io->md_buf = md_buf;
1171 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
1172 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
1173 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1174 
1175 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
1176 	expected_io->md_buf = md_buf + 2 * 8;
1177 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
1178 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1179 
1180 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
1181 	expected_io->md_buf = md_buf + 18 * 8;
1182 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
1183 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
1184 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1185 
1186 	rc = spdk_bdev_writev_blocks_with_md(desc, io_ch, iov, 3, md_buf,
1187 					     14, 32, io_done, NULL);
1188 	CU_ASSERT(rc == 0);
1189 	CU_ASSERT(g_io_done == false);
1190 
1191 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
1192 	stub_complete_io(3);
1193 	CU_ASSERT(g_io_done == true);
1194 
1195 	/* Test multi vector command that needs to be split by strip and then needs to be
1196 	 * split further due to the capacity of child iovs.
1197 	 */
1198 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
1199 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1200 		iov[i].iov_len = 512;
1201 	}
1202 
1203 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1204 	g_io_done = false;
1205 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1206 					   BDEV_IO_NUM_CHILD_IOV);
1207 	expected_io->md_buf = md_buf;
1208 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1209 		ut_expected_io_set_iov(expected_io, i, (void *)((i + 1) * 0x10000), 512);
1210 	}
1211 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1212 
1213 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1214 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
1215 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1216 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1217 		ut_expected_io_set_iov(expected_io, i,
1218 				       (void *)((i + 1 + BDEV_IO_NUM_CHILD_IOV) * 0x10000), 512);
1219 	}
1220 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1221 
1222 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1223 					    0, BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1224 	CU_ASSERT(rc == 0);
1225 	CU_ASSERT(g_io_done == false);
1226 
1227 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1228 	stub_complete_io(1);
1229 	CU_ASSERT(g_io_done == false);
1230 
1231 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1232 	stub_complete_io(1);
1233 	CU_ASSERT(g_io_done == true);
1234 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1235 
1236 	/* Test multi vector command that needs to be split by strip and then needs to be
1237 	 * split further due to the capacity of child iovs. In this case, the length of
1238 	 * the rest of iovec array with an I/O boundary is the multiple of block size.
1239 	 */
1240 
1241 	/* Fill iovec array for exactly one boundary. The iovec cnt for this boundary
1242 	 * is BDEV_IO_NUM_CHILD_IOV + 1, which exceeds the capacity of child iovs.
1243 	 */
1244 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1245 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1246 		iov[i].iov_len = 512;
1247 	}
1248 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1249 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1250 		iov[i].iov_len = 256;
1251 	}
1252 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1253 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 512;
1254 
1255 	/* Add an extra iovec to trigger split */
1256 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1257 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1258 
1259 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1260 	g_io_done = false;
1261 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1262 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV);
1263 	expected_io->md_buf = md_buf;
1264 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1265 		ut_expected_io_set_iov(expected_io, i,
1266 				       (void *)((i + 1) * 0x10000), 512);
1267 	}
1268 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1269 		ut_expected_io_set_iov(expected_io, i,
1270 				       (void *)((i + 1) * 0x10000), 256);
1271 	}
1272 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1273 
1274 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1275 					   1, 1);
1276 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1277 	ut_expected_io_set_iov(expected_io, 0,
1278 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 512);
1279 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1280 
1281 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1282 					   1, 1);
1283 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1284 	ut_expected_io_set_iov(expected_io, 0,
1285 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1286 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1287 
1288 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, md_buf,
1289 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1290 	CU_ASSERT(rc == 0);
1291 	CU_ASSERT(g_io_done == false);
1292 
1293 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1294 	stub_complete_io(1);
1295 	CU_ASSERT(g_io_done == false);
1296 
1297 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1298 	stub_complete_io(2);
1299 	CU_ASSERT(g_io_done == true);
1300 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1301 
1302 	/* Test multi vector command that needs to be split by strip and then needs to be
1303 	 * split further due to the capacity of child iovs, the child request offset should
1304 	 * be rewind to last aligned offset and go success without error.
1305 	 */
1306 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1307 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1308 		iov[i].iov_len = 512;
1309 	}
1310 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1311 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1312 
1313 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1314 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1315 
1316 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1317 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1318 
1319 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1320 	g_io_done = false;
1321 	g_io_status = 0;
1322 	/* The first expected io should be start from offset 0 to BDEV_IO_NUM_CHILD_IOV - 1 */
1323 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1324 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV - 1);
1325 	expected_io->md_buf = md_buf;
1326 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1327 		ut_expected_io_set_iov(expected_io, i,
1328 				       (void *)((i + 1) * 0x10000), 512);
1329 	}
1330 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1331 	/* The second expected io should be start from offset BDEV_IO_NUM_CHILD_IOV - 1 to BDEV_IO_NUM_CHILD_IOV */
1332 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1333 					   1, 2);
1334 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1335 	ut_expected_io_set_iov(expected_io, 0,
1336 			       (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000), 256);
1337 	ut_expected_io_set_iov(expected_io, 1,
1338 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 256);
1339 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1340 	/* The third expected io should be start from offset BDEV_IO_NUM_CHILD_IOV to BDEV_IO_NUM_CHILD_IOV + 1 */
1341 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1342 					   1, 1);
1343 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1344 	ut_expected_io_set_iov(expected_io, 0,
1345 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1346 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1347 
1348 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1349 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1350 	CU_ASSERT(rc == 0);
1351 	CU_ASSERT(g_io_done == false);
1352 
1353 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1354 	stub_complete_io(1);
1355 	CU_ASSERT(g_io_done == false);
1356 
1357 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1358 	stub_complete_io(2);
1359 	CU_ASSERT(g_io_done == true);
1360 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1361 
1362 	/* Test multi vector command that needs to be split due to the IO boundary and
1363 	 * the capacity of child iovs. Especially test the case when the command is
1364 	 * split due to the capacity of child iovs, the tail address is not aligned with
1365 	 * block size and is rewinded to the aligned address.
1366 	 *
1367 	 * The iovecs used in read request is complex but is based on the data
1368 	 * collected in the real issue. We change the base addresses but keep the lengths
1369 	 * not to loose the credibility of the test.
1370 	 */
1371 	bdev->optimal_io_boundary = 128;
1372 	g_io_done = false;
1373 	g_io_status = 0;
1374 
1375 	for (i = 0; i < 31; i++) {
1376 		iov[i].iov_base = (void *)(0xFEED0000000 + (i << 20));
1377 		iov[i].iov_len = 1024;
1378 	}
1379 	iov[31].iov_base = (void *)0xFEED1F00000;
1380 	iov[31].iov_len = 32768;
1381 	iov[32].iov_base = (void *)0xFEED2000000;
1382 	iov[32].iov_len = 160;
1383 	iov[33].iov_base = (void *)0xFEED2100000;
1384 	iov[33].iov_len = 4096;
1385 	iov[34].iov_base = (void *)0xFEED2200000;
1386 	iov[34].iov_len = 4096;
1387 	iov[35].iov_base = (void *)0xFEED2300000;
1388 	iov[35].iov_len = 4096;
1389 	iov[36].iov_base = (void *)0xFEED2400000;
1390 	iov[36].iov_len = 4096;
1391 	iov[37].iov_base = (void *)0xFEED2500000;
1392 	iov[37].iov_len = 4096;
1393 	iov[38].iov_base = (void *)0xFEED2600000;
1394 	iov[38].iov_len = 4096;
1395 	iov[39].iov_base = (void *)0xFEED2700000;
1396 	iov[39].iov_len = 4096;
1397 	iov[40].iov_base = (void *)0xFEED2800000;
1398 	iov[40].iov_len = 4096;
1399 	iov[41].iov_base = (void *)0xFEED2900000;
1400 	iov[41].iov_len = 4096;
1401 	iov[42].iov_base = (void *)0xFEED2A00000;
1402 	iov[42].iov_len = 4096;
1403 	iov[43].iov_base = (void *)0xFEED2B00000;
1404 	iov[43].iov_len = 12288;
1405 	iov[44].iov_base = (void *)0xFEED2C00000;
1406 	iov[44].iov_len = 8192;
1407 	iov[45].iov_base = (void *)0xFEED2F00000;
1408 	iov[45].iov_len = 4096;
1409 	iov[46].iov_base = (void *)0xFEED3000000;
1410 	iov[46].iov_len = 4096;
1411 	iov[47].iov_base = (void *)0xFEED3100000;
1412 	iov[47].iov_len = 4096;
1413 	iov[48].iov_base = (void *)0xFEED3200000;
1414 	iov[48].iov_len = 24576;
1415 	iov[49].iov_base = (void *)0xFEED3300000;
1416 	iov[49].iov_len = 16384;
1417 	iov[50].iov_base = (void *)0xFEED3400000;
1418 	iov[50].iov_len = 12288;
1419 	iov[51].iov_base = (void *)0xFEED3500000;
1420 	iov[51].iov_len = 4096;
1421 	iov[52].iov_base = (void *)0xFEED3600000;
1422 	iov[52].iov_len = 4096;
1423 	iov[53].iov_base = (void *)0xFEED3700000;
1424 	iov[53].iov_len = 4096;
1425 	iov[54].iov_base = (void *)0xFEED3800000;
1426 	iov[54].iov_len = 28672;
1427 	iov[55].iov_base = (void *)0xFEED3900000;
1428 	iov[55].iov_len = 20480;
1429 	iov[56].iov_base = (void *)0xFEED3A00000;
1430 	iov[56].iov_len = 4096;
1431 	iov[57].iov_base = (void *)0xFEED3B00000;
1432 	iov[57].iov_len = 12288;
1433 	iov[58].iov_base = (void *)0xFEED3C00000;
1434 	iov[58].iov_len = 4096;
1435 	iov[59].iov_base = (void *)0xFEED3D00000;
1436 	iov[59].iov_len = 4096;
1437 	iov[60].iov_base = (void *)0xFEED3E00000;
1438 	iov[60].iov_len = 352;
1439 
1440 	/* The 1st child IO must be from iov[0] to iov[31] split by the capacity
1441 	 * of child iovs,
1442 	 */
1443 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 126, 32);
1444 	expected_io->md_buf = md_buf;
1445 	for (i = 0; i < 32; i++) {
1446 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1447 	}
1448 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1449 
1450 	/* The 2nd child IO must be from iov[32] to the first 864 bytes of iov[33]
1451 	 * split by the IO boundary requirement.
1452 	 */
1453 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 126, 2, 2);
1454 	expected_io->md_buf = md_buf + 126 * 8;
1455 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base, iov[32].iov_len);
1456 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 864);
1457 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1458 
1459 	/* The 3rd child IO must be from the remaining 3232 bytes of iov[33] to
1460 	 * the first 864 bytes of iov[46] split by the IO boundary requirement.
1461 	 */
1462 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 128, 128, 14);
1463 	expected_io->md_buf = md_buf + 128 * 8;
1464 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[33].iov_base + 864),
1465 			       iov[33].iov_len - 864);
1466 	ut_expected_io_set_iov(expected_io, 1, iov[34].iov_base, iov[34].iov_len);
1467 	ut_expected_io_set_iov(expected_io, 2, iov[35].iov_base, iov[35].iov_len);
1468 	ut_expected_io_set_iov(expected_io, 3, iov[36].iov_base, iov[36].iov_len);
1469 	ut_expected_io_set_iov(expected_io, 4, iov[37].iov_base, iov[37].iov_len);
1470 	ut_expected_io_set_iov(expected_io, 5, iov[38].iov_base, iov[38].iov_len);
1471 	ut_expected_io_set_iov(expected_io, 6, iov[39].iov_base, iov[39].iov_len);
1472 	ut_expected_io_set_iov(expected_io, 7, iov[40].iov_base, iov[40].iov_len);
1473 	ut_expected_io_set_iov(expected_io, 8, iov[41].iov_base, iov[41].iov_len);
1474 	ut_expected_io_set_iov(expected_io, 9, iov[42].iov_base, iov[42].iov_len);
1475 	ut_expected_io_set_iov(expected_io, 10, iov[43].iov_base, iov[43].iov_len);
1476 	ut_expected_io_set_iov(expected_io, 11, iov[44].iov_base, iov[44].iov_len);
1477 	ut_expected_io_set_iov(expected_io, 12, iov[45].iov_base, iov[45].iov_len);
1478 	ut_expected_io_set_iov(expected_io, 13, iov[46].iov_base, 864);
1479 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1480 
1481 	/* The 4th child IO must be from the remaining 3232 bytes of iov[46] to the
1482 	 * first 864 bytes of iov[52] split by the IO boundary requirement.
1483 	 */
1484 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 256, 128, 7);
1485 	expected_io->md_buf = md_buf + 256 * 8;
1486 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[46].iov_base + 864),
1487 			       iov[46].iov_len - 864);
1488 	ut_expected_io_set_iov(expected_io, 1, iov[47].iov_base, iov[47].iov_len);
1489 	ut_expected_io_set_iov(expected_io, 2, iov[48].iov_base, iov[48].iov_len);
1490 	ut_expected_io_set_iov(expected_io, 3, iov[49].iov_base, iov[49].iov_len);
1491 	ut_expected_io_set_iov(expected_io, 4, iov[50].iov_base, iov[50].iov_len);
1492 	ut_expected_io_set_iov(expected_io, 5, iov[51].iov_base, iov[51].iov_len);
1493 	ut_expected_io_set_iov(expected_io, 6, iov[52].iov_base, 864);
1494 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1495 
1496 	/* The 5th child IO must be from the remaining 3232 bytes of iov[52] to
1497 	 * the first 4096 bytes of iov[57] split by the IO boundary requirement.
1498 	 */
1499 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 384, 128, 6);
1500 	expected_io->md_buf = md_buf + 384 * 8;
1501 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[52].iov_base + 864),
1502 			       iov[52].iov_len - 864);
1503 	ut_expected_io_set_iov(expected_io, 1, iov[53].iov_base, iov[53].iov_len);
1504 	ut_expected_io_set_iov(expected_io, 2, iov[54].iov_base, iov[54].iov_len);
1505 	ut_expected_io_set_iov(expected_io, 3, iov[55].iov_base, iov[55].iov_len);
1506 	ut_expected_io_set_iov(expected_io, 4, iov[56].iov_base, iov[56].iov_len);
1507 	ut_expected_io_set_iov(expected_io, 5, iov[57].iov_base, 4960);
1508 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1509 
1510 	/* The 6th child IO must be from the remaining 7328 bytes of iov[57]
1511 	 * to the first 3936 bytes of iov[58] split by the capacity of child iovs.
1512 	 */
1513 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 512, 30, 3);
1514 	expected_io->md_buf = md_buf + 512 * 8;
1515 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[57].iov_base + 4960),
1516 			       iov[57].iov_len - 4960);
1517 	ut_expected_io_set_iov(expected_io, 1, iov[58].iov_base, iov[58].iov_len);
1518 	ut_expected_io_set_iov(expected_io, 2, iov[59].iov_base, 3936);
1519 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1520 
1521 	/* The 7th child IO is from the remaining 160 bytes of iov[59] and iov[60]. */
1522 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 542, 1, 2);
1523 	expected_io->md_buf = md_buf + 542 * 8;
1524 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[59].iov_base + 3936),
1525 			       iov[59].iov_len - 3936);
1526 	ut_expected_io_set_iov(expected_io, 1, iov[60].iov_base, iov[60].iov_len);
1527 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1528 
1529 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, 61, md_buf,
1530 					    0, 543, io_done, NULL);
1531 	CU_ASSERT(rc == 0);
1532 	CU_ASSERT(g_io_done == false);
1533 
1534 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1535 	stub_complete_io(1);
1536 	CU_ASSERT(g_io_done == false);
1537 
1538 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1539 	stub_complete_io(5);
1540 	CU_ASSERT(g_io_done == false);
1541 
1542 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1543 	stub_complete_io(1);
1544 	CU_ASSERT(g_io_done == true);
1545 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1546 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1547 
1548 	/* Test a WRITE_ZEROES that would span an I/O boundary.  WRITE_ZEROES should not be
1549 	 * split, so test that.
1550 	 */
1551 	bdev->optimal_io_boundary = 15;
1552 	g_io_done = false;
1553 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
1554 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1555 
1556 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
1557 	CU_ASSERT(rc == 0);
1558 	CU_ASSERT(g_io_done == false);
1559 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1560 	stub_complete_io(1);
1561 	CU_ASSERT(g_io_done == true);
1562 
1563 	/* Test an UNMAP.  This should also not be split. */
1564 	bdev->optimal_io_boundary = 16;
1565 	g_io_done = false;
1566 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 2, 0);
1567 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1568 
1569 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 2, io_done, NULL);
1570 	CU_ASSERT(rc == 0);
1571 	CU_ASSERT(g_io_done == false);
1572 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1573 	stub_complete_io(1);
1574 	CU_ASSERT(g_io_done == true);
1575 
1576 	/* Test a FLUSH.  This should also not be split. */
1577 	bdev->optimal_io_boundary = 16;
1578 	g_io_done = false;
1579 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 2, 0);
1580 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1581 
1582 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
1583 	CU_ASSERT(rc == 0);
1584 	CU_ASSERT(g_io_done == false);
1585 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1586 	stub_complete_io(1);
1587 	CU_ASSERT(g_io_done == true);
1588 
1589 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
1590 
1591 	/* Children requests return an error status */
1592 	bdev->optimal_io_boundary = 16;
1593 	iov[0].iov_base = (void *)0x10000;
1594 	iov[0].iov_len = 512 * 64;
1595 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1596 	g_io_done = false;
1597 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1598 
1599 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 1, 1, 64, io_done, NULL);
1600 	CU_ASSERT(rc == 0);
1601 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1602 	stub_complete_io(4);
1603 	CU_ASSERT(g_io_done == false);
1604 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1605 	stub_complete_io(1);
1606 	CU_ASSERT(g_io_done == true);
1607 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1608 
1609 	/* Test if a multi vector command terminated with failure before continueing
1610 	 * splitting process when one of child I/O failed.
1611 	 * The multi vector command is as same as the above that needs to be split by strip
1612 	 * and then needs to be split further due to the capacity of child iovs.
1613 	 */
1614 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1615 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1616 		iov[i].iov_len = 512;
1617 	}
1618 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1619 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1620 
1621 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1622 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1623 
1624 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1625 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1626 
1627 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1628 
1629 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1630 	g_io_done = false;
1631 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1632 
1633 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
1634 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1635 	CU_ASSERT(rc == 0);
1636 	CU_ASSERT(g_io_done == false);
1637 
1638 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1639 	stub_complete_io(1);
1640 	CU_ASSERT(g_io_done == true);
1641 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1642 
1643 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1644 
1645 	/* for this test we will create the following conditions to hit the code path where
1646 	 * we are trying to send and IO following a split that has no iovs because we had to
1647 	 * trim them for alignment reasons.
1648 	 *
1649 	 * - 16K boundary, our IO will start at offset 0 with a length of 0x4200
1650 	 * - Our IOVs are 0x212 in size so that we run into the 16K boundary at child IOV
1651 	 *   position 30 and overshoot by 0x2e.
1652 	 * - That means we'll send the IO and loop back to pick up the remaining bytes at
1653 	 *   child IOV index 31. When we do, we find that we have to shorten index 31 by 0x2e
1654 	 *   which eliniates that vector so we just send the first split IO with 30 vectors
1655 	 *   and let the completion pick up the last 2 vectors.
1656 	 */
1657 	bdev->optimal_io_boundary = 32;
1658 	bdev->split_on_optimal_io_boundary = true;
1659 	g_io_done = false;
1660 
1661 	/* Init all parent IOVs to 0x212 */
1662 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
1663 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1664 		iov[i].iov_len = 0x212;
1665 	}
1666 
1667 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1668 					   BDEV_IO_NUM_CHILD_IOV - 1);
1669 	/* expect 0-29 to be 1:1 with the parent iov */
1670 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1671 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1672 	}
1673 
1674 	/* expect index 30 to be shortened to 0x1e4 (0x212 - 0x1e) because of the alignment
1675 	 * where 0x1e is the amount we overshot the 16K boundary
1676 	 */
1677 	ut_expected_io_set_iov(expected_io, BDEV_IO_NUM_CHILD_IOV - 2,
1678 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base), 0x1e4);
1679 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1680 
1681 	/* 2nd child IO will have 2 remaining vectors, one to pick up from the one that was
1682 	 * shortened that take it to the next boundary and then a final one to get us to
1683 	 * 0x4200 bytes for the IO.
1684 	 */
1685 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1686 					   BDEV_IO_NUM_CHILD_IOV, 2);
1687 	/* position 30 picked up the remaining bytes to the next boundary */
1688 	ut_expected_io_set_iov(expected_io, 0,
1689 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base + 0x1e4), 0x2e);
1690 
1691 	/* position 31 picked the the rest of the trasnfer to get us to 0x4200 */
1692 	ut_expected_io_set_iov(expected_io, 1,
1693 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base), 0x1d2);
1694 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1695 
1696 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 1, 0,
1697 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1698 	CU_ASSERT(rc == 0);
1699 	CU_ASSERT(g_io_done == false);
1700 
1701 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1702 	stub_complete_io(1);
1703 	CU_ASSERT(g_io_done == false);
1704 
1705 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1706 	stub_complete_io(1);
1707 	CU_ASSERT(g_io_done == true);
1708 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1709 
1710 	spdk_put_io_channel(io_ch);
1711 	spdk_bdev_close(desc);
1712 	free_bdev(bdev);
1713 	spdk_bdev_finish(bdev_fini_cb, NULL);
1714 	poll_threads();
1715 }
1716 
1717 static void
1718 bdev_io_max_size_and_segment_split_test(void)
1719 {
1720 	struct spdk_bdev *bdev;
1721 	struct spdk_bdev_desc *desc = NULL;
1722 	struct spdk_io_channel *io_ch;
1723 	struct spdk_bdev_opts bdev_opts = {};
1724 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1725 	struct ut_expected_io *expected_io;
1726 	uint64_t i;
1727 	int rc;
1728 
1729 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1730 	bdev_opts.bdev_io_pool_size = 512;
1731 	bdev_opts.bdev_io_cache_size = 64;
1732 
1733 	bdev_opts.opts_size = sizeof(bdev_opts);
1734 	rc = spdk_bdev_set_opts(&bdev_opts);
1735 	CU_ASSERT(rc == 0);
1736 	spdk_bdev_initialize(bdev_init_cb, NULL);
1737 
1738 	bdev = allocate_bdev("bdev0");
1739 
1740 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
1741 	CU_ASSERT(rc == 0);
1742 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1743 	io_ch = spdk_bdev_get_io_channel(desc);
1744 	CU_ASSERT(io_ch != NULL);
1745 
1746 	bdev->split_on_optimal_io_boundary = false;
1747 	bdev->optimal_io_boundary = 0;
1748 
1749 	/* Case 0 max_num_segments == 0.
1750 	 * but segment size 2 * 512 > 512
1751 	 */
1752 	bdev->max_segment_size = 512;
1753 	bdev->max_num_segments = 0;
1754 	g_io_done = false;
1755 
1756 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
1757 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1758 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
1759 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1760 
1761 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1762 	CU_ASSERT(rc == 0);
1763 	CU_ASSERT(g_io_done == false);
1764 
1765 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1766 	stub_complete_io(1);
1767 	CU_ASSERT(g_io_done == true);
1768 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1769 
1770 	/* Case 1 max_segment_size == 0
1771 	 * but iov num 2 > 1.
1772 	 */
1773 	bdev->max_segment_size = 0;
1774 	bdev->max_num_segments = 1;
1775 	g_io_done = false;
1776 
1777 	iov[0].iov_base = (void *)0x10000;
1778 	iov[0].iov_len = 512;
1779 	iov[1].iov_base = (void *)0x20000;
1780 	iov[1].iov_len = 8 * 512;
1781 
1782 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1783 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, iov[0].iov_len);
1784 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1785 
1786 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 8, 1);
1787 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, iov[1].iov_len);
1788 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1789 
1790 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 14, 9, io_done, NULL);
1791 	CU_ASSERT(rc == 0);
1792 	CU_ASSERT(g_io_done == false);
1793 
1794 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1795 	stub_complete_io(2);
1796 	CU_ASSERT(g_io_done == true);
1797 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1798 
1799 	/* Test that a non-vector command is split correctly.
1800 	 * Set up the expected values before calling spdk_bdev_read_blocks
1801 	 */
1802 	bdev->max_segment_size = 512;
1803 	bdev->max_num_segments = 1;
1804 	g_io_done = false;
1805 
1806 	/* Child IO 0 */
1807 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1808 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1809 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1810 
1811 	/* Child IO 1 */
1812 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
1813 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 1 * 512), 512);
1814 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1815 
1816 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1817 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1818 	CU_ASSERT(rc == 0);
1819 	CU_ASSERT(g_io_done == false);
1820 
1821 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1822 	stub_complete_io(2);
1823 	CU_ASSERT(g_io_done == true);
1824 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1825 
1826 	/* Now set up a more complex, multi-vector command that needs to be split,
1827 	 * including splitting iovecs.
1828 	 */
1829 	bdev->max_segment_size = 2 * 512;
1830 	bdev->max_num_segments = 1;
1831 	g_io_done = false;
1832 
1833 	iov[0].iov_base = (void *)0x10000;
1834 	iov[0].iov_len = 2 * 512;
1835 	iov[1].iov_base = (void *)0x20000;
1836 	iov[1].iov_len = 4 * 512;
1837 	iov[2].iov_base = (void *)0x30000;
1838 	iov[2].iov_len = 6 * 512;
1839 
1840 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
1841 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 2);
1842 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1843 
1844 	/* Split iov[1].size to 2 iov entries then split the segments */
1845 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
1846 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, 512 * 2);
1847 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1848 
1849 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 18, 2, 1);
1850 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base + 512 * 2, 512 * 2);
1851 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1852 
1853 	/* Split iov[2].size to 3 iov entries then split the segments */
1854 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 20, 2, 1);
1855 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base, 512 * 2);
1856 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1857 
1858 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 22, 2, 1);
1859 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 2, 512 * 2);
1860 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1861 
1862 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 24, 2, 1);
1863 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 4, 512 * 2);
1864 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1865 
1866 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 12, io_done, NULL);
1867 	CU_ASSERT(rc == 0);
1868 	CU_ASSERT(g_io_done == false);
1869 
1870 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
1871 	stub_complete_io(6);
1872 	CU_ASSERT(g_io_done == true);
1873 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1874 
1875 	/* Test multi vector command that needs to be split by strip and then needs to be
1876 	 * split further due to the capacity of parent IO child iovs.
1877 	 */
1878 	bdev->max_segment_size = 512;
1879 	bdev->max_num_segments = 1;
1880 	g_io_done = false;
1881 
1882 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1883 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1884 		iov[i].iov_len = 512 * 2;
1885 	}
1886 
1887 	/* Each input iov.size is split into 2 iovs,
1888 	 * half of the input iov can fill all child iov entries of a single IO.
1889 	 */
1890 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV / 2; i++) {
1891 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i, 1, 1);
1892 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1893 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1894 
1895 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i + 1, 1, 1);
1896 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1897 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1898 	}
1899 
1900 	/* The remaining iov is split in the second round */
1901 	for (i = BDEV_IO_NUM_CHILD_IOV / 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1902 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2, 1, 1);
1903 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1904 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1905 
1906 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2 + 1, 1, 1);
1907 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1908 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1909 	}
1910 
1911 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
1912 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1913 	CU_ASSERT(rc == 0);
1914 	CU_ASSERT(g_io_done == false);
1915 
1916 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1917 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1918 	CU_ASSERT(g_io_done == false);
1919 
1920 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1921 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1922 	CU_ASSERT(g_io_done == true);
1923 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1924 
1925 	/* A wrong case, a child IO that is divided does
1926 	 * not meet the principle of multiples of block size,
1927 	 * and exits with error
1928 	 */
1929 	bdev->max_segment_size = 512;
1930 	bdev->max_num_segments = 1;
1931 	g_io_done = false;
1932 
1933 	iov[0].iov_base = (void *)0x10000;
1934 	iov[0].iov_len = 512 + 256;
1935 	iov[1].iov_base = (void *)0x20000;
1936 	iov[1].iov_len = 256;
1937 
1938 	/* iov[0] is split to 512 and 256.
1939 	 * 256 is less than a block size, and it is found
1940 	 * in the next round of split that it is the first child IO smaller than
1941 	 * the block size, so the error exit
1942 	 */
1943 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 1, 1);
1944 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512);
1945 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1946 
1947 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 0, 2, io_done, NULL);
1948 	CU_ASSERT(rc == 0);
1949 	CU_ASSERT(g_io_done == false);
1950 
1951 	/* First child IO is OK */
1952 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1953 	stub_complete_io(1);
1954 	CU_ASSERT(g_io_done == true);
1955 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1956 
1957 	/* error exit */
1958 	stub_complete_io(1);
1959 	CU_ASSERT(g_io_done == true);
1960 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1961 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1962 
1963 	/* Test multi vector command that needs to be split by strip and then needs to be
1964 	 * split further due to the capacity of child iovs.
1965 	 *
1966 	 * In this case, the last two iovs need to be split, but it will exceed the capacity
1967 	 * of child iovs, so it needs to wait until the first batch completed.
1968 	 */
1969 	bdev->max_segment_size = 512;
1970 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
1971 	g_io_done = false;
1972 
1973 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1974 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1975 		iov[i].iov_len = 512;
1976 	}
1977 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1978 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1979 		iov[i].iov_len = 512 * 2;
1980 	}
1981 
1982 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1983 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
1984 	/* 0 ~ (BDEV_IO_NUM_CHILD_IOV - 2) Will not be split */
1985 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1986 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1987 	}
1988 	/* (BDEV_IO_NUM_CHILD_IOV - 2) is split */
1989 	ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, 512);
1990 	ut_expected_io_set_iov(expected_io, i + 1, iov[i].iov_base + 512, 512);
1991 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1992 
1993 	/* Child iov entries exceed the max num of parent IO so split it in next round */
1994 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV, 2, 2);
1995 	ut_expected_io_set_iov(expected_io, 0, iov[i + 1].iov_base, 512);
1996 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base + 512, 512);
1997 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1998 
1999 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
2000 				    BDEV_IO_NUM_CHILD_IOV + 2, io_done, NULL);
2001 	CU_ASSERT(rc == 0);
2002 	CU_ASSERT(g_io_done == false);
2003 
2004 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2005 	stub_complete_io(1);
2006 	CU_ASSERT(g_io_done == false);
2007 
2008 	/* Next round */
2009 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2010 	stub_complete_io(1);
2011 	CU_ASSERT(g_io_done == true);
2012 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2013 
2014 	/* This case is similar to the previous one, but the io composed of
2015 	 * the last few entries of child iov is not enough for a blocklen, so they
2016 	 * cannot be put into this IO, but wait until the next time.
2017 	 */
2018 	bdev->max_segment_size = 512;
2019 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2020 	g_io_done = false;
2021 
2022 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2023 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2024 		iov[i].iov_len = 512;
2025 	}
2026 
2027 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2028 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2029 		iov[i].iov_len = 128;
2030 	}
2031 
2032 	/* First child iovcnt is't BDEV_IO_NUM_CHILD_IOV but BDEV_IO_NUM_CHILD_IOV - 2.
2033 	 * Because the left 2 iov is not enough for a blocklen.
2034 	 */
2035 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2036 					   BDEV_IO_NUM_CHILD_IOV - 2, BDEV_IO_NUM_CHILD_IOV - 2);
2037 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2038 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2039 	}
2040 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2041 
2042 	/* The second child io waits until the end of the first child io before executing.
2043 	 * Because the iovcnt of the two IOs exceeds the child iovcnt of the parent IO.
2044 	 * BDEV_IO_NUM_CHILD_IOV - 2 to BDEV_IO_NUM_CHILD_IOV + 2
2045 	 */
2046 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 2,
2047 					   1, 4);
2048 	ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, iov[i].iov_len);
2049 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base, iov[i + 1].iov_len);
2050 	ut_expected_io_set_iov(expected_io, 2, iov[i + 2].iov_base, iov[i + 2].iov_len);
2051 	ut_expected_io_set_iov(expected_io, 3, iov[i + 3].iov_base, iov[i + 3].iov_len);
2052 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2053 
2054 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2055 				    BDEV_IO_NUM_CHILD_IOV - 1, io_done, NULL);
2056 	CU_ASSERT(rc == 0);
2057 	CU_ASSERT(g_io_done == false);
2058 
2059 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2060 	stub_complete_io(1);
2061 	CU_ASSERT(g_io_done == false);
2062 
2063 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2064 	stub_complete_io(1);
2065 	CU_ASSERT(g_io_done == true);
2066 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2067 
2068 	/* A very complicated case. Each sg entry exceeds max_segment_size and
2069 	 * needs to be split. At the same time, child io must be a multiple of blocklen.
2070 	 * At the same time, child iovcnt exceeds parent iovcnt.
2071 	 */
2072 	bdev->max_segment_size = 512 + 128;
2073 	bdev->max_num_segments = 3;
2074 	g_io_done = false;
2075 
2076 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2077 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2078 		iov[i].iov_len = 512 + 256;
2079 	}
2080 
2081 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2082 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2083 		iov[i].iov_len = 512 + 128;
2084 	}
2085 
2086 	/* Child IOs use 9 entries per for() round and 3 * 9 = 27 child iov entries.
2087 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2088 	 * Generate 9 child IOs.
2089 	 */
2090 	for (i = 0; i < 3; i++) {
2091 		uint32_t j = i * 4;
2092 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6, 2, 3);
2093 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2094 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2095 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2096 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2097 
2098 		/* Child io must be a multiple of blocklen
2099 		 * iov[j + 2] must be split. If the third entry is also added,
2100 		 * the multiple of blocklen cannot be guaranteed. But it still
2101 		 * occupies one iov entry of the parent child iov.
2102 		 */
2103 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 2, 2, 2);
2104 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2105 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2106 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2107 
2108 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 4, 2, 3);
2109 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2110 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2111 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2112 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2113 	}
2114 
2115 	/* Child iov position at 27, the 10th child IO
2116 	 * iov entry index is 3 * 4 and offset is 3 * 6
2117 	 */
2118 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 18, 2, 3);
2119 	ut_expected_io_set_iov(expected_io, 0, iov[12].iov_base, 640);
2120 	ut_expected_io_set_iov(expected_io, 1, iov[12].iov_base + 640, 128);
2121 	ut_expected_io_set_iov(expected_io, 2, iov[13].iov_base, 256);
2122 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2123 
2124 	/* Child iov position at 30, the 11th child IO */
2125 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 20, 2, 2);
2126 	ut_expected_io_set_iov(expected_io, 0, iov[13].iov_base + 256, 512);
2127 	ut_expected_io_set_iov(expected_io, 1, iov[14].iov_base, 512);
2128 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2129 
2130 	/* The 2nd split round and iovpos is 0, the 12th child IO */
2131 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 22, 2, 3);
2132 	ut_expected_io_set_iov(expected_io, 0, iov[14].iov_base + 512, 256);
2133 	ut_expected_io_set_iov(expected_io, 1, iov[15].iov_base, 640);
2134 	ut_expected_io_set_iov(expected_io, 2, iov[15].iov_base + 640, 128);
2135 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2136 
2137 	/* Consume 9 child IOs and 27 child iov entries.
2138 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2139 	 * Parent IO iov index start from 16 and block offset start from 24
2140 	 */
2141 	for (i = 0; i < 3; i++) {
2142 		uint32_t j = i * 4 + 16;
2143 		uint32_t offset = i * 6 + 24;
2144 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, 2, 3);
2145 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2146 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2147 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2148 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2149 
2150 		/* Child io must be a multiple of blocklen
2151 		 * iov[j + 2] must be split. If the third entry is also added,
2152 		 * the multiple of blocklen cannot be guaranteed. But it still
2153 		 * occupies one iov entry of the parent child iov.
2154 		 */
2155 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 2, 2, 2);
2156 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2157 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2158 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2159 
2160 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 4, 2, 3);
2161 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2162 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2163 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2164 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2165 	}
2166 
2167 	/* The 22th child IO, child iov position at 30 */
2168 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 42, 1, 1);
2169 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base, 512);
2170 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2171 
2172 	/* The third round */
2173 	/* Here is the 23nd child IO and child iovpos is 0 */
2174 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 43, 2, 3);
2175 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base + 512, 256);
2176 	ut_expected_io_set_iov(expected_io, 1, iov[29].iov_base, 640);
2177 	ut_expected_io_set_iov(expected_io, 2, iov[29].iov_base + 640, 128);
2178 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2179 
2180 	/* The 24th child IO */
2181 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 45, 3, 3);
2182 	ut_expected_io_set_iov(expected_io, 0, iov[30].iov_base, 640);
2183 	ut_expected_io_set_iov(expected_io, 1, iov[31].iov_base, 640);
2184 	ut_expected_io_set_iov(expected_io, 2, iov[32].iov_base, 256);
2185 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2186 
2187 	/* The 25th child IO */
2188 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 48, 2, 2);
2189 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base + 256, 384);
2190 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 640);
2191 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2192 
2193 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2194 				    50, io_done, NULL);
2195 	CU_ASSERT(rc == 0);
2196 	CU_ASSERT(g_io_done == false);
2197 
2198 	/* Parent IO supports up to 32 child iovs, so it is calculated that
2199 	 * a maximum of 11 IOs can be split at a time, and the
2200 	 * splitting will continue after the first batch is over.
2201 	 */
2202 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2203 	stub_complete_io(11);
2204 	CU_ASSERT(g_io_done == false);
2205 
2206 	/* The 2nd round */
2207 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2208 	stub_complete_io(11);
2209 	CU_ASSERT(g_io_done == false);
2210 
2211 	/* The last round */
2212 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2213 	stub_complete_io(3);
2214 	CU_ASSERT(g_io_done == true);
2215 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2216 
2217 	/* Test an WRITE_ZEROES.  This should also not be split. */
2218 	bdev->max_segment_size = 512;
2219 	bdev->max_num_segments = 1;
2220 	g_io_done = false;
2221 
2222 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
2223 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2224 
2225 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
2226 	CU_ASSERT(rc == 0);
2227 	CU_ASSERT(g_io_done == false);
2228 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2229 	stub_complete_io(1);
2230 	CU_ASSERT(g_io_done == true);
2231 
2232 	/* Test an UNMAP.  This should also not be split. */
2233 	g_io_done = false;
2234 
2235 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 4, 0);
2236 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2237 
2238 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 4, io_done, NULL);
2239 	CU_ASSERT(rc == 0);
2240 	CU_ASSERT(g_io_done == false);
2241 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2242 	stub_complete_io(1);
2243 	CU_ASSERT(g_io_done == true);
2244 
2245 	/* Test a FLUSH.  This should also not be split. */
2246 	g_io_done = false;
2247 
2248 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 4, 0);
2249 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2250 
2251 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
2252 	CU_ASSERT(rc == 0);
2253 	CU_ASSERT(g_io_done == false);
2254 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2255 	stub_complete_io(1);
2256 	CU_ASSERT(g_io_done == true);
2257 
2258 	spdk_put_io_channel(io_ch);
2259 	spdk_bdev_close(desc);
2260 	free_bdev(bdev);
2261 	spdk_bdev_finish(bdev_fini_cb, NULL);
2262 	poll_threads();
2263 }
2264 
2265 static void
2266 bdev_io_mix_split_test(void)
2267 {
2268 	struct spdk_bdev *bdev;
2269 	struct spdk_bdev_desc *desc = NULL;
2270 	struct spdk_io_channel *io_ch;
2271 	struct spdk_bdev_opts bdev_opts = {};
2272 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
2273 	struct ut_expected_io *expected_io;
2274 	uint64_t i;
2275 	int rc;
2276 
2277 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2278 	bdev_opts.bdev_io_pool_size = 512;
2279 	bdev_opts.bdev_io_cache_size = 64;
2280 
2281 	rc = spdk_bdev_set_opts(&bdev_opts);
2282 	CU_ASSERT(rc == 0);
2283 	spdk_bdev_initialize(bdev_init_cb, NULL);
2284 
2285 	bdev = allocate_bdev("bdev0");
2286 
2287 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
2288 	CU_ASSERT(rc == 0);
2289 	SPDK_CU_ASSERT_FATAL(desc != NULL);
2290 	io_ch = spdk_bdev_get_io_channel(desc);
2291 	CU_ASSERT(io_ch != NULL);
2292 
2293 	/* First case optimal_io_boundary == max_segment_size * max_num_segments */
2294 	bdev->split_on_optimal_io_boundary = true;
2295 	bdev->optimal_io_boundary = 16;
2296 
2297 	bdev->max_segment_size = 512;
2298 	bdev->max_num_segments = 16;
2299 	g_io_done = false;
2300 
2301 	/* IO crossing the IO boundary requires split
2302 	 * Total 2 child IOs.
2303 	 */
2304 
2305 	/* The 1st child IO split the segment_size to multiple segment entry */
2306 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
2307 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
2308 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
2309 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2310 
2311 	/* The 2nd child IO split the segment_size to multiple segment entry */
2312 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 2);
2313 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 512);
2314 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 3 * 512), 512);
2315 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2316 
2317 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 4, io_done, NULL);
2318 	CU_ASSERT(rc == 0);
2319 	CU_ASSERT(g_io_done == false);
2320 
2321 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2322 	stub_complete_io(2);
2323 	CU_ASSERT(g_io_done == true);
2324 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2325 
2326 	/* Second case optimal_io_boundary > max_segment_size * max_num_segments */
2327 	bdev->max_segment_size = 15 * 512;
2328 	bdev->max_num_segments = 1;
2329 	g_io_done = false;
2330 
2331 	/* IO crossing the IO boundary requires split.
2332 	 * The 1st child IO segment size exceeds the max_segment_size,
2333 	 * So 1st child IO will be splitted to multiple segment entry.
2334 	 * Then it split to 2 child IOs because of the max_num_segments.
2335 	 * Total 3 child IOs.
2336 	 */
2337 
2338 	/* The first 2 IOs are in an IO boundary.
2339 	 * Because the optimal_io_boundary > max_segment_size * max_num_segments
2340 	 * So it split to the first 2 IOs.
2341 	 */
2342 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 15, 1);
2343 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 15);
2344 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2345 
2346 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
2347 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 15), 512);
2348 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2349 
2350 	/* The 3rd Child IO is because of the io boundary */
2351 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2352 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2353 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2354 
2355 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2356 	CU_ASSERT(rc == 0);
2357 	CU_ASSERT(g_io_done == false);
2358 
2359 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2360 	stub_complete_io(3);
2361 	CU_ASSERT(g_io_done == true);
2362 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2363 
2364 	/* Third case optimal_io_boundary < max_segment_size * max_num_segments */
2365 	bdev->max_segment_size = 17 * 512;
2366 	bdev->max_num_segments = 1;
2367 	g_io_done = false;
2368 
2369 	/* IO crossing the IO boundary requires split.
2370 	 * Child IO does not split.
2371 	 * Total 2 child IOs.
2372 	 */
2373 
2374 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 16, 1);
2375 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 16);
2376 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2377 
2378 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2379 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2380 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2381 
2382 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2383 	CU_ASSERT(rc == 0);
2384 	CU_ASSERT(g_io_done == false);
2385 
2386 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2387 	stub_complete_io(2);
2388 	CU_ASSERT(g_io_done == true);
2389 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2390 
2391 	/* Now set up a more complex, multi-vector command that needs to be split,
2392 	 * including splitting iovecs.
2393 	 * optimal_io_boundary < max_segment_size * max_num_segments
2394 	 */
2395 	bdev->max_segment_size = 3 * 512;
2396 	bdev->max_num_segments = 6;
2397 	g_io_done = false;
2398 
2399 	iov[0].iov_base = (void *)0x10000;
2400 	iov[0].iov_len = 4 * 512;
2401 	iov[1].iov_base = (void *)0x20000;
2402 	iov[1].iov_len = 4 * 512;
2403 	iov[2].iov_base = (void *)0x30000;
2404 	iov[2].iov_len = 10 * 512;
2405 
2406 	/* IO crossing the IO boundary requires split.
2407 	 * The 1st child IO segment size exceeds the max_segment_size and after
2408 	 * splitting segment_size, the num_segments exceeds max_num_segments.
2409 	 * So 1st child IO will be splitted to 2 child IOs.
2410 	 * Total 3 child IOs.
2411 	 */
2412 
2413 	/* The first 2 IOs are in an IO boundary.
2414 	 * After splitting segmemt size the segment num exceeds.
2415 	 * So it splits to 2 child IOs.
2416 	 */
2417 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 14, 6);
2418 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 3);
2419 	ut_expected_io_set_iov(expected_io, 1, iov[0].iov_base + 512 * 3, 512);
2420 	ut_expected_io_set_iov(expected_io, 2, iov[1].iov_base, 512 * 3);
2421 	ut_expected_io_set_iov(expected_io, 3, iov[1].iov_base + 512 * 3, 512);
2422 	ut_expected_io_set_iov(expected_io, 4, iov[2].iov_base, 512 * 3);
2423 	ut_expected_io_set_iov(expected_io, 5, iov[2].iov_base + 512 * 3, 512 * 3);
2424 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2425 
2426 	/* The 2nd child IO has the left segment entry */
2427 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
2428 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 6, 512 * 2);
2429 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2430 
2431 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
2432 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 8, 512 * 2);
2433 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2434 
2435 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 0, 18, io_done, NULL);
2436 	CU_ASSERT(rc == 0);
2437 	CU_ASSERT(g_io_done == false);
2438 
2439 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2440 	stub_complete_io(3);
2441 	CU_ASSERT(g_io_done == true);
2442 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2443 
2444 	/* A very complicated case. Each sg entry exceeds max_segment_size
2445 	 * and split on io boundary.
2446 	 * optimal_io_boundary < max_segment_size * max_num_segments
2447 	 */
2448 	bdev->max_segment_size = 3 * 512;
2449 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2450 	g_io_done = false;
2451 
2452 	for (i = 0; i < 20; i++) {
2453 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2454 		iov[i].iov_len = 512 * 4;
2455 	}
2456 
2457 	/* IO crossing the IO boundary requires split.
2458 	 * 80 block length can split 5 child IOs base on offset and IO boundary.
2459 	 * Each iov entry needs to be splitted to 2 entries because of max_segment_size
2460 	 * Total 5 child IOs.
2461 	 */
2462 
2463 	/* 4 iov entries are in an IO boundary and each iov entry splits to 2.
2464 	 * So each child IO occupies 8 child iov entries.
2465 	 */
2466 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 16, 8);
2467 	for (i = 0; i < 4; i++) {
2468 		int iovcnt = i * 2;
2469 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2470 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2471 	}
2472 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2473 
2474 	/* 2nd child IO and total 16 child iov entries of parent IO */
2475 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 8);
2476 	for (i = 4; i < 8; i++) {
2477 		int iovcnt = (i - 4) * 2;
2478 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2479 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2480 	}
2481 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2482 
2483 	/* 3rd child IO and total 24 child iov entries of parent IO */
2484 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 16, 8);
2485 	for (i = 8; i < 12; i++) {
2486 		int iovcnt = (i - 8) * 2;
2487 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2488 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2489 	}
2490 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2491 
2492 	/* 4th child IO and total 32 child iov entries of parent IO */
2493 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 48, 16, 8);
2494 	for (i = 12; i < 16; i++) {
2495 		int iovcnt = (i - 12) * 2;
2496 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2497 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2498 	}
2499 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2500 
2501 	/* 5th child IO and because of the child iov entry it should be splitted
2502 	 * in next round.
2503 	 */
2504 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 64, 16, 8);
2505 	for (i = 16; i < 20; i++) {
2506 		int iovcnt = (i - 16) * 2;
2507 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2508 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2509 	}
2510 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2511 
2512 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 20, 0, 80, io_done, NULL);
2513 	CU_ASSERT(rc == 0);
2514 	CU_ASSERT(g_io_done == false);
2515 
2516 	/* First split round */
2517 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
2518 	stub_complete_io(4);
2519 	CU_ASSERT(g_io_done == false);
2520 
2521 	/* Second split round */
2522 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2523 	stub_complete_io(1);
2524 	CU_ASSERT(g_io_done == true);
2525 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2526 
2527 	spdk_put_io_channel(io_ch);
2528 	spdk_bdev_close(desc);
2529 	free_bdev(bdev);
2530 	spdk_bdev_finish(bdev_fini_cb, NULL);
2531 	poll_threads();
2532 }
2533 
2534 static void
2535 bdev_io_split_with_io_wait(void)
2536 {
2537 	struct spdk_bdev *bdev;
2538 	struct spdk_bdev_desc *desc = NULL;
2539 	struct spdk_io_channel *io_ch;
2540 	struct spdk_bdev_channel *channel;
2541 	struct spdk_bdev_mgmt_channel *mgmt_ch;
2542 	struct spdk_bdev_opts bdev_opts = {};
2543 	struct iovec iov[3];
2544 	struct ut_expected_io *expected_io;
2545 	int rc;
2546 
2547 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2548 	bdev_opts.bdev_io_pool_size = 2;
2549 	bdev_opts.bdev_io_cache_size = 1;
2550 
2551 	rc = spdk_bdev_set_opts(&bdev_opts);
2552 	CU_ASSERT(rc == 0);
2553 	spdk_bdev_initialize(bdev_init_cb, NULL);
2554 
2555 	bdev = allocate_bdev("bdev0");
2556 
2557 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2558 	CU_ASSERT(rc == 0);
2559 	CU_ASSERT(desc != NULL);
2560 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2561 	io_ch = spdk_bdev_get_io_channel(desc);
2562 	CU_ASSERT(io_ch != NULL);
2563 	channel = spdk_io_channel_get_ctx(io_ch);
2564 	mgmt_ch = channel->shared_resource->mgmt_ch;
2565 
2566 	bdev->optimal_io_boundary = 16;
2567 	bdev->split_on_optimal_io_boundary = true;
2568 
2569 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
2570 	CU_ASSERT(rc == 0);
2571 
2572 	/* Now test that a single-vector command is split correctly.
2573 	 * Offset 14, length 8, payload 0xF000
2574 	 *  Child - Offset 14, length 2, payload 0xF000
2575 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
2576 	 *
2577 	 * Set up the expected values before calling spdk_bdev_read_blocks
2578 	 */
2579 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
2580 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
2581 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2582 
2583 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
2584 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
2585 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2586 
2587 	/* The following children will be submitted sequentially due to the capacity of
2588 	 * spdk_bdev_io.
2589 	 */
2590 
2591 	/* The first child I/O will be queued to wait until an spdk_bdev_io becomes available */
2592 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
2593 	CU_ASSERT(rc == 0);
2594 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2595 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2596 
2597 	/* Completing the first read I/O will submit the first child */
2598 	stub_complete_io(1);
2599 	CU_ASSERT(TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2600 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2601 
2602 	/* Completing the first child will submit the second child */
2603 	stub_complete_io(1);
2604 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2605 
2606 	/* Complete the second child I/O.  This should result in our callback getting
2607 	 * invoked since the parent I/O is now complete.
2608 	 */
2609 	stub_complete_io(1);
2610 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2611 
2612 	/* Now set up a more complex, multi-vector command that needs to be split,
2613 	 *  including splitting iovecs.
2614 	 */
2615 	iov[0].iov_base = (void *)0x10000;
2616 	iov[0].iov_len = 512;
2617 	iov[1].iov_base = (void *)0x20000;
2618 	iov[1].iov_len = 20 * 512;
2619 	iov[2].iov_base = (void *)0x30000;
2620 	iov[2].iov_len = 11 * 512;
2621 
2622 	g_io_done = false;
2623 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
2624 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
2625 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
2626 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2627 
2628 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
2629 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
2630 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2631 
2632 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
2633 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
2634 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
2635 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2636 
2637 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 32, io_done, NULL);
2638 	CU_ASSERT(rc == 0);
2639 	CU_ASSERT(g_io_done == false);
2640 
2641 	/* The following children will be submitted sequentially due to the capacity of
2642 	 * spdk_bdev_io.
2643 	 */
2644 
2645 	/* Completing the first child will submit the second child */
2646 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2647 	stub_complete_io(1);
2648 	CU_ASSERT(g_io_done == false);
2649 
2650 	/* Completing the second child will submit the third child */
2651 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2652 	stub_complete_io(1);
2653 	CU_ASSERT(g_io_done == false);
2654 
2655 	/* Completing the third child will result in our callback getting invoked
2656 	 * since the parent I/O is now complete.
2657 	 */
2658 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2659 	stub_complete_io(1);
2660 	CU_ASSERT(g_io_done == true);
2661 
2662 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
2663 
2664 	spdk_put_io_channel(io_ch);
2665 	spdk_bdev_close(desc);
2666 	free_bdev(bdev);
2667 	spdk_bdev_finish(bdev_fini_cb, NULL);
2668 	poll_threads();
2669 }
2670 
2671 static void
2672 bdev_io_alignment(void)
2673 {
2674 	struct spdk_bdev *bdev;
2675 	struct spdk_bdev_desc *desc = NULL;
2676 	struct spdk_io_channel *io_ch;
2677 	struct spdk_bdev_opts bdev_opts = {};
2678 	int rc;
2679 	void *buf = NULL;
2680 	struct iovec iovs[2];
2681 	int iovcnt;
2682 	uint64_t alignment;
2683 
2684 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2685 	bdev_opts.bdev_io_pool_size = 20;
2686 	bdev_opts.bdev_io_cache_size = 2;
2687 
2688 	rc = spdk_bdev_set_opts(&bdev_opts);
2689 	CU_ASSERT(rc == 0);
2690 	spdk_bdev_initialize(bdev_init_cb, NULL);
2691 
2692 	fn_table.submit_request = stub_submit_request_get_buf;
2693 	bdev = allocate_bdev("bdev0");
2694 
2695 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2696 	CU_ASSERT(rc == 0);
2697 	CU_ASSERT(desc != NULL);
2698 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2699 	io_ch = spdk_bdev_get_io_channel(desc);
2700 	CU_ASSERT(io_ch != NULL);
2701 
2702 	/* Create aligned buffer */
2703 	rc = posix_memalign(&buf, 4096, 8192);
2704 	SPDK_CU_ASSERT_FATAL(rc == 0);
2705 
2706 	/* Pass aligned single buffer with no alignment required */
2707 	alignment = 1;
2708 	bdev->required_alignment = spdk_u32log2(alignment);
2709 
2710 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2711 	CU_ASSERT(rc == 0);
2712 	stub_complete_io(1);
2713 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2714 				    alignment));
2715 
2716 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2717 	CU_ASSERT(rc == 0);
2718 	stub_complete_io(1);
2719 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2720 				    alignment));
2721 
2722 	/* Pass unaligned single buffer with no alignment required */
2723 	alignment = 1;
2724 	bdev->required_alignment = spdk_u32log2(alignment);
2725 
2726 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2727 	CU_ASSERT(rc == 0);
2728 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2729 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2730 	stub_complete_io(1);
2731 
2732 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2733 	CU_ASSERT(rc == 0);
2734 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2735 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2736 	stub_complete_io(1);
2737 
2738 	/* Pass unaligned single buffer with 512 alignment required */
2739 	alignment = 512;
2740 	bdev->required_alignment = spdk_u32log2(alignment);
2741 
2742 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2743 	CU_ASSERT(rc == 0);
2744 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2745 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2746 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2747 				    alignment));
2748 	stub_complete_io(1);
2749 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2750 
2751 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2752 	CU_ASSERT(rc == 0);
2753 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2754 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2755 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2756 				    alignment));
2757 	stub_complete_io(1);
2758 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2759 
2760 	/* Pass unaligned single buffer with 4096 alignment required */
2761 	alignment = 4096;
2762 	bdev->required_alignment = spdk_u32log2(alignment);
2763 
2764 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2765 	CU_ASSERT(rc == 0);
2766 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2767 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2768 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2769 				    alignment));
2770 	stub_complete_io(1);
2771 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2772 
2773 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2774 	CU_ASSERT(rc == 0);
2775 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2776 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2777 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2778 				    alignment));
2779 	stub_complete_io(1);
2780 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2781 
2782 	/* Pass aligned iovs with no alignment required */
2783 	alignment = 1;
2784 	bdev->required_alignment = spdk_u32log2(alignment);
2785 
2786 	iovcnt = 1;
2787 	iovs[0].iov_base = buf;
2788 	iovs[0].iov_len = 512;
2789 
2790 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2791 	CU_ASSERT(rc == 0);
2792 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2793 	stub_complete_io(1);
2794 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2795 
2796 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2797 	CU_ASSERT(rc == 0);
2798 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2799 	stub_complete_io(1);
2800 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2801 
2802 	/* Pass unaligned iovs with no alignment required */
2803 	alignment = 1;
2804 	bdev->required_alignment = spdk_u32log2(alignment);
2805 
2806 	iovcnt = 2;
2807 	iovs[0].iov_base = buf + 16;
2808 	iovs[0].iov_len = 256;
2809 	iovs[1].iov_base = buf + 16 + 256 + 32;
2810 	iovs[1].iov_len = 256;
2811 
2812 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2813 	CU_ASSERT(rc == 0);
2814 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2815 	stub_complete_io(1);
2816 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2817 
2818 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2819 	CU_ASSERT(rc == 0);
2820 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2821 	stub_complete_io(1);
2822 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2823 
2824 	/* Pass unaligned iov with 2048 alignment required */
2825 	alignment = 2048;
2826 	bdev->required_alignment = spdk_u32log2(alignment);
2827 
2828 	iovcnt = 2;
2829 	iovs[0].iov_base = buf + 16;
2830 	iovs[0].iov_len = 256;
2831 	iovs[1].iov_base = buf + 16 + 256 + 32;
2832 	iovs[1].iov_len = 256;
2833 
2834 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2835 	CU_ASSERT(rc == 0);
2836 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2837 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2838 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2839 				    alignment));
2840 	stub_complete_io(1);
2841 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2842 
2843 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2844 	CU_ASSERT(rc == 0);
2845 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2846 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2847 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2848 				    alignment));
2849 	stub_complete_io(1);
2850 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2851 
2852 	/* Pass iov without allocated buffer without alignment required */
2853 	alignment = 1;
2854 	bdev->required_alignment = spdk_u32log2(alignment);
2855 
2856 	iovcnt = 1;
2857 	iovs[0].iov_base = NULL;
2858 	iovs[0].iov_len = 0;
2859 
2860 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2861 	CU_ASSERT(rc == 0);
2862 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2863 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2864 				    alignment));
2865 	stub_complete_io(1);
2866 
2867 	/* Pass iov without allocated buffer with 1024 alignment required */
2868 	alignment = 1024;
2869 	bdev->required_alignment = spdk_u32log2(alignment);
2870 
2871 	iovcnt = 1;
2872 	iovs[0].iov_base = NULL;
2873 	iovs[0].iov_len = 0;
2874 
2875 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2876 	CU_ASSERT(rc == 0);
2877 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2878 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2879 				    alignment));
2880 	stub_complete_io(1);
2881 
2882 	spdk_put_io_channel(io_ch);
2883 	spdk_bdev_close(desc);
2884 	free_bdev(bdev);
2885 	fn_table.submit_request = stub_submit_request;
2886 	spdk_bdev_finish(bdev_fini_cb, NULL);
2887 	poll_threads();
2888 
2889 	free(buf);
2890 }
2891 
2892 static void
2893 bdev_io_alignment_with_boundary(void)
2894 {
2895 	struct spdk_bdev *bdev;
2896 	struct spdk_bdev_desc *desc = NULL;
2897 	struct spdk_io_channel *io_ch;
2898 	struct spdk_bdev_opts bdev_opts = {};
2899 	int rc;
2900 	void *buf = NULL;
2901 	struct iovec iovs[2];
2902 	int iovcnt;
2903 	uint64_t alignment;
2904 
2905 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2906 	bdev_opts.bdev_io_pool_size = 20;
2907 	bdev_opts.bdev_io_cache_size = 2;
2908 
2909 	bdev_opts.opts_size = sizeof(bdev_opts);
2910 	rc = spdk_bdev_set_opts(&bdev_opts);
2911 	CU_ASSERT(rc == 0);
2912 	spdk_bdev_initialize(bdev_init_cb, NULL);
2913 
2914 	fn_table.submit_request = stub_submit_request_get_buf;
2915 	bdev = allocate_bdev("bdev0");
2916 
2917 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2918 	CU_ASSERT(rc == 0);
2919 	CU_ASSERT(desc != NULL);
2920 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2921 	io_ch = spdk_bdev_get_io_channel(desc);
2922 	CU_ASSERT(io_ch != NULL);
2923 
2924 	/* Create aligned buffer */
2925 	rc = posix_memalign(&buf, 4096, 131072);
2926 	SPDK_CU_ASSERT_FATAL(rc == 0);
2927 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2928 
2929 	/* 512 * 3 with 2 IO boundary, allocate small data buffer from bdev layer */
2930 	alignment = 512;
2931 	bdev->required_alignment = spdk_u32log2(alignment);
2932 	bdev->optimal_io_boundary = 2;
2933 	bdev->split_on_optimal_io_boundary = true;
2934 
2935 	iovcnt = 1;
2936 	iovs[0].iov_base = NULL;
2937 	iovs[0].iov_len = 512 * 3;
2938 
2939 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
2940 	CU_ASSERT(rc == 0);
2941 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2942 	stub_complete_io(2);
2943 
2944 	/* 8KiB with 16 IO boundary, allocate large data buffer from bdev layer */
2945 	alignment = 512;
2946 	bdev->required_alignment = spdk_u32log2(alignment);
2947 	bdev->optimal_io_boundary = 16;
2948 	bdev->split_on_optimal_io_boundary = true;
2949 
2950 	iovcnt = 1;
2951 	iovs[0].iov_base = NULL;
2952 	iovs[0].iov_len = 512 * 16;
2953 
2954 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 16, io_done, NULL);
2955 	CU_ASSERT(rc == 0);
2956 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2957 	stub_complete_io(2);
2958 
2959 	/* 512 * 160 with 128 IO boundary, 63.5KiB + 16.5KiB for the two children requests */
2960 	alignment = 512;
2961 	bdev->required_alignment = spdk_u32log2(alignment);
2962 	bdev->optimal_io_boundary = 128;
2963 	bdev->split_on_optimal_io_boundary = true;
2964 
2965 	iovcnt = 1;
2966 	iovs[0].iov_base = buf + 16;
2967 	iovs[0].iov_len = 512 * 160;
2968 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
2969 	CU_ASSERT(rc == 0);
2970 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2971 	stub_complete_io(2);
2972 
2973 	/* 512 * 3 with 2 IO boundary */
2974 	alignment = 512;
2975 	bdev->required_alignment = spdk_u32log2(alignment);
2976 	bdev->optimal_io_boundary = 2;
2977 	bdev->split_on_optimal_io_boundary = true;
2978 
2979 	iovcnt = 2;
2980 	iovs[0].iov_base = buf + 16;
2981 	iovs[0].iov_len = 512;
2982 	iovs[1].iov_base = buf + 16 + 512 + 32;
2983 	iovs[1].iov_len = 1024;
2984 
2985 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
2986 	CU_ASSERT(rc == 0);
2987 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2988 	stub_complete_io(2);
2989 
2990 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
2991 	CU_ASSERT(rc == 0);
2992 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2993 	stub_complete_io(2);
2994 
2995 	/* 512 * 64 with 32 IO boundary */
2996 	bdev->optimal_io_boundary = 32;
2997 	iovcnt = 2;
2998 	iovs[0].iov_base = buf + 16;
2999 	iovs[0].iov_len = 16384;
3000 	iovs[1].iov_base = buf + 16 + 16384 + 32;
3001 	iovs[1].iov_len = 16384;
3002 
3003 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3004 	CU_ASSERT(rc == 0);
3005 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3006 	stub_complete_io(3);
3007 
3008 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3009 	CU_ASSERT(rc == 0);
3010 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3011 	stub_complete_io(3);
3012 
3013 	/* 512 * 160 with 32 IO boundary */
3014 	iovcnt = 1;
3015 	iovs[0].iov_base = buf + 16;
3016 	iovs[0].iov_len = 16384 + 65536;
3017 
3018 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3019 	CU_ASSERT(rc == 0);
3020 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
3021 	stub_complete_io(6);
3022 
3023 	spdk_put_io_channel(io_ch);
3024 	spdk_bdev_close(desc);
3025 	free_bdev(bdev);
3026 	fn_table.submit_request = stub_submit_request;
3027 	spdk_bdev_finish(bdev_fini_cb, NULL);
3028 	poll_threads();
3029 
3030 	free(buf);
3031 }
3032 
3033 static void
3034 histogram_status_cb(void *cb_arg, int status)
3035 {
3036 	g_status = status;
3037 }
3038 
3039 static void
3040 histogram_data_cb(void *cb_arg, int status, struct spdk_histogram_data *histogram)
3041 {
3042 	g_status = status;
3043 	g_histogram = histogram;
3044 }
3045 
3046 static void
3047 histogram_io_count(void *ctx, uint64_t start, uint64_t end, uint64_t count,
3048 		   uint64_t total, uint64_t so_far)
3049 {
3050 	g_count += count;
3051 }
3052 
3053 static void
3054 bdev_histograms(void)
3055 {
3056 	struct spdk_bdev *bdev;
3057 	struct spdk_bdev_desc *desc = NULL;
3058 	struct spdk_io_channel *ch;
3059 	struct spdk_histogram_data *histogram;
3060 	uint8_t buf[4096];
3061 	int rc;
3062 
3063 	spdk_bdev_initialize(bdev_init_cb, NULL);
3064 
3065 	bdev = allocate_bdev("bdev");
3066 
3067 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3068 	CU_ASSERT(rc == 0);
3069 	CU_ASSERT(desc != NULL);
3070 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3071 
3072 	ch = spdk_bdev_get_io_channel(desc);
3073 	CU_ASSERT(ch != NULL);
3074 
3075 	/* Enable histogram */
3076 	g_status = -1;
3077 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, true);
3078 	poll_threads();
3079 	CU_ASSERT(g_status == 0);
3080 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3081 
3082 	/* Allocate histogram */
3083 	histogram = spdk_histogram_data_alloc();
3084 	SPDK_CU_ASSERT_FATAL(histogram != NULL);
3085 
3086 	/* Check if histogram is zeroed */
3087 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3088 	poll_threads();
3089 	CU_ASSERT(g_status == 0);
3090 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3091 
3092 	g_count = 0;
3093 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3094 
3095 	CU_ASSERT(g_count == 0);
3096 
3097 	rc = spdk_bdev_write_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3098 	CU_ASSERT(rc == 0);
3099 
3100 	spdk_delay_us(10);
3101 	stub_complete_io(1);
3102 	poll_threads();
3103 
3104 	rc = spdk_bdev_read_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3105 	CU_ASSERT(rc == 0);
3106 
3107 	spdk_delay_us(10);
3108 	stub_complete_io(1);
3109 	poll_threads();
3110 
3111 	/* Check if histogram gathered data from all I/O channels */
3112 	g_histogram = NULL;
3113 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3114 	poll_threads();
3115 	CU_ASSERT(g_status == 0);
3116 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3117 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3118 
3119 	g_count = 0;
3120 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3121 	CU_ASSERT(g_count == 2);
3122 
3123 	/* Disable histogram */
3124 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, false);
3125 	poll_threads();
3126 	CU_ASSERT(g_status == 0);
3127 	CU_ASSERT(bdev->internal.histogram_enabled == false);
3128 
3129 	/* Try to run histogram commands on disabled bdev */
3130 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3131 	poll_threads();
3132 	CU_ASSERT(g_status == -EFAULT);
3133 
3134 	spdk_histogram_data_free(histogram);
3135 	spdk_put_io_channel(ch);
3136 	spdk_bdev_close(desc);
3137 	free_bdev(bdev);
3138 	spdk_bdev_finish(bdev_fini_cb, NULL);
3139 	poll_threads();
3140 }
3141 
3142 static void
3143 _bdev_compare(bool emulated)
3144 {
3145 	struct spdk_bdev *bdev;
3146 	struct spdk_bdev_desc *desc = NULL;
3147 	struct spdk_io_channel *ioch;
3148 	struct ut_expected_io *expected_io;
3149 	uint64_t offset, num_blocks;
3150 	uint32_t num_completed;
3151 	char aa_buf[512];
3152 	char bb_buf[512];
3153 	struct iovec compare_iov;
3154 	uint8_t io_type;
3155 	int rc;
3156 
3157 	if (emulated) {
3158 		io_type = SPDK_BDEV_IO_TYPE_READ;
3159 	} else {
3160 		io_type = SPDK_BDEV_IO_TYPE_COMPARE;
3161 	}
3162 
3163 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3164 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3165 
3166 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = !emulated;
3167 
3168 	spdk_bdev_initialize(bdev_init_cb, NULL);
3169 	fn_table.submit_request = stub_submit_request_get_buf;
3170 	bdev = allocate_bdev("bdev");
3171 
3172 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3173 	CU_ASSERT_EQUAL(rc, 0);
3174 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3175 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3176 	ioch = spdk_bdev_get_io_channel(desc);
3177 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3178 
3179 	fn_table.submit_request = stub_submit_request_get_buf;
3180 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3181 
3182 	offset = 50;
3183 	num_blocks = 1;
3184 	compare_iov.iov_base = aa_buf;
3185 	compare_iov.iov_len = sizeof(aa_buf);
3186 
3187 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3188 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3189 
3190 	g_io_done = false;
3191 	g_compare_read_buf = aa_buf;
3192 	g_compare_read_buf_len = sizeof(aa_buf);
3193 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3194 	CU_ASSERT_EQUAL(rc, 0);
3195 	num_completed = stub_complete_io(1);
3196 	CU_ASSERT_EQUAL(num_completed, 1);
3197 	CU_ASSERT(g_io_done == true);
3198 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3199 
3200 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3201 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3202 
3203 	g_io_done = false;
3204 	g_compare_read_buf = bb_buf;
3205 	g_compare_read_buf_len = sizeof(bb_buf);
3206 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3207 	CU_ASSERT_EQUAL(rc, 0);
3208 	num_completed = stub_complete_io(1);
3209 	CU_ASSERT_EQUAL(num_completed, 1);
3210 	CU_ASSERT(g_io_done == true);
3211 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3212 
3213 	spdk_put_io_channel(ioch);
3214 	spdk_bdev_close(desc);
3215 	free_bdev(bdev);
3216 	fn_table.submit_request = stub_submit_request;
3217 	spdk_bdev_finish(bdev_fini_cb, NULL);
3218 	poll_threads();
3219 
3220 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3221 
3222 	g_compare_read_buf = NULL;
3223 }
3224 
3225 static void
3226 bdev_compare(void)
3227 {
3228 	_bdev_compare(true);
3229 	_bdev_compare(false);
3230 }
3231 
3232 static void
3233 bdev_compare_and_write(void)
3234 {
3235 	struct spdk_bdev *bdev;
3236 	struct spdk_bdev_desc *desc = NULL;
3237 	struct spdk_io_channel *ioch;
3238 	struct ut_expected_io *expected_io;
3239 	uint64_t offset, num_blocks;
3240 	uint32_t num_completed;
3241 	char aa_buf[512];
3242 	char bb_buf[512];
3243 	char cc_buf[512];
3244 	char write_buf[512];
3245 	struct iovec compare_iov;
3246 	struct iovec write_iov;
3247 	int rc;
3248 
3249 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3250 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3251 	memset(cc_buf, 0xcc, sizeof(cc_buf));
3252 
3253 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = false;
3254 
3255 	spdk_bdev_initialize(bdev_init_cb, NULL);
3256 	fn_table.submit_request = stub_submit_request_get_buf;
3257 	bdev = allocate_bdev("bdev");
3258 
3259 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3260 	CU_ASSERT_EQUAL(rc, 0);
3261 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3262 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3263 	ioch = spdk_bdev_get_io_channel(desc);
3264 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3265 
3266 	fn_table.submit_request = stub_submit_request_get_buf;
3267 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3268 
3269 	offset = 50;
3270 	num_blocks = 1;
3271 	compare_iov.iov_base = aa_buf;
3272 	compare_iov.iov_len = sizeof(aa_buf);
3273 	write_iov.iov_base = bb_buf;
3274 	write_iov.iov_len = sizeof(bb_buf);
3275 
3276 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3277 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3278 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, offset, num_blocks, 0);
3279 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3280 
3281 	g_io_done = false;
3282 	g_compare_read_buf = aa_buf;
3283 	g_compare_read_buf_len = sizeof(aa_buf);
3284 	memset(write_buf, 0, sizeof(write_buf));
3285 	g_compare_write_buf = write_buf;
3286 	g_compare_write_buf_len = sizeof(write_buf);
3287 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3288 			offset, num_blocks, io_done, NULL);
3289 	/* Trigger range locking */
3290 	poll_threads();
3291 	CU_ASSERT_EQUAL(rc, 0);
3292 	num_completed = stub_complete_io(1);
3293 	CU_ASSERT_EQUAL(num_completed, 1);
3294 	CU_ASSERT(g_io_done == false);
3295 	num_completed = stub_complete_io(1);
3296 	/* Trigger range unlocking */
3297 	poll_threads();
3298 	CU_ASSERT_EQUAL(num_completed, 1);
3299 	CU_ASSERT(g_io_done == true);
3300 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3301 	CU_ASSERT(memcmp(write_buf, bb_buf, sizeof(write_buf)) == 0);
3302 
3303 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3304 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3305 
3306 	g_io_done = false;
3307 	g_compare_read_buf = cc_buf;
3308 	g_compare_read_buf_len = sizeof(cc_buf);
3309 	memset(write_buf, 0, sizeof(write_buf));
3310 	g_compare_write_buf = write_buf;
3311 	g_compare_write_buf_len = sizeof(write_buf);
3312 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3313 			offset, num_blocks, io_done, NULL);
3314 	/* Trigger range locking */
3315 	poll_threads();
3316 	CU_ASSERT_EQUAL(rc, 0);
3317 	num_completed = stub_complete_io(1);
3318 	/* Trigger range unlocking earlier because we expect error here */
3319 	poll_threads();
3320 	CU_ASSERT_EQUAL(num_completed, 1);
3321 	CU_ASSERT(g_io_done == true);
3322 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3323 	num_completed = stub_complete_io(1);
3324 	CU_ASSERT_EQUAL(num_completed, 0);
3325 
3326 	spdk_put_io_channel(ioch);
3327 	spdk_bdev_close(desc);
3328 	free_bdev(bdev);
3329 	fn_table.submit_request = stub_submit_request;
3330 	spdk_bdev_finish(bdev_fini_cb, NULL);
3331 	poll_threads();
3332 
3333 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3334 
3335 	g_compare_read_buf = NULL;
3336 	g_compare_write_buf = NULL;
3337 }
3338 
3339 static void
3340 bdev_write_zeroes(void)
3341 {
3342 	struct spdk_bdev *bdev;
3343 	struct spdk_bdev_desc *desc = NULL;
3344 	struct spdk_io_channel *ioch;
3345 	struct ut_expected_io *expected_io;
3346 	uint64_t offset, num_io_blocks, num_blocks;
3347 	uint32_t num_completed, num_requests;
3348 	int rc;
3349 
3350 	spdk_bdev_initialize(bdev_init_cb, NULL);
3351 	bdev = allocate_bdev("bdev");
3352 
3353 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3354 	CU_ASSERT_EQUAL(rc, 0);
3355 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3356 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3357 	ioch = spdk_bdev_get_io_channel(desc);
3358 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3359 
3360 	fn_table.submit_request = stub_submit_request;
3361 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3362 
3363 	/* First test that if the bdev supports write_zeroes, the request won't be split */
3364 	bdev->md_len = 0;
3365 	bdev->blocklen = 4096;
3366 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3367 
3368 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
3369 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3370 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3371 	CU_ASSERT_EQUAL(rc, 0);
3372 	num_completed = stub_complete_io(1);
3373 	CU_ASSERT_EQUAL(num_completed, 1);
3374 
3375 	/* Check that if write zeroes is not supported it'll be replaced by regular writes */
3376 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
3377 	num_io_blocks = ZERO_BUFFER_SIZE / bdev->blocklen;
3378 	num_requests = 2;
3379 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * num_requests;
3380 
3381 	for (offset = 0; offset < num_requests; ++offset) {
3382 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3383 						   offset * num_io_blocks, num_io_blocks, 0);
3384 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3385 	}
3386 
3387 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3388 	CU_ASSERT_EQUAL(rc, 0);
3389 	num_completed = stub_complete_io(num_requests);
3390 	CU_ASSERT_EQUAL(num_completed, num_requests);
3391 
3392 	/* Check that the splitting is correct if bdev has interleaved metadata */
3393 	bdev->md_interleave = true;
3394 	bdev->md_len = 64;
3395 	bdev->blocklen = 4096 + 64;
3396 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3397 
3398 	num_requests = offset = 0;
3399 	while (offset < num_blocks) {
3400 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / bdev->blocklen, num_blocks - offset);
3401 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3402 						   offset, num_io_blocks, 0);
3403 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3404 		offset += num_io_blocks;
3405 		num_requests++;
3406 	}
3407 
3408 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3409 	CU_ASSERT_EQUAL(rc, 0);
3410 	num_completed = stub_complete_io(num_requests);
3411 	CU_ASSERT_EQUAL(num_completed, num_requests);
3412 	num_completed = stub_complete_io(num_requests);
3413 	assert(num_completed == 0);
3414 
3415 	/* Check the the same for separate metadata buffer */
3416 	bdev->md_interleave = false;
3417 	bdev->md_len = 64;
3418 	bdev->blocklen = 4096;
3419 
3420 	num_requests = offset = 0;
3421 	while (offset < num_blocks) {
3422 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / (bdev->blocklen + bdev->md_len), num_blocks);
3423 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3424 						   offset, num_io_blocks, 0);
3425 		expected_io->md_buf = (char *)g_bdev_mgr.zero_buffer + num_io_blocks * bdev->blocklen;
3426 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3427 		offset += num_io_blocks;
3428 		num_requests++;
3429 	}
3430 
3431 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3432 	CU_ASSERT_EQUAL(rc, 0);
3433 	num_completed = stub_complete_io(num_requests);
3434 	CU_ASSERT_EQUAL(num_completed, num_requests);
3435 
3436 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
3437 	spdk_put_io_channel(ioch);
3438 	spdk_bdev_close(desc);
3439 	free_bdev(bdev);
3440 	spdk_bdev_finish(bdev_fini_cb, NULL);
3441 	poll_threads();
3442 }
3443 
3444 static void
3445 bdev_open_while_hotremove(void)
3446 {
3447 	struct spdk_bdev *bdev;
3448 	struct spdk_bdev_desc *desc[2] = {};
3449 	int rc;
3450 
3451 	bdev = allocate_bdev("bdev");
3452 
3453 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[0]);
3454 	CU_ASSERT(rc == 0);
3455 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
3456 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc[0]));
3457 
3458 	spdk_bdev_unregister(bdev, NULL, NULL);
3459 
3460 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[1]);
3461 	CU_ASSERT(rc == -ENODEV);
3462 	SPDK_CU_ASSERT_FATAL(desc[1] == NULL);
3463 
3464 	spdk_bdev_close(desc[0]);
3465 	free_bdev(bdev);
3466 }
3467 
3468 static void
3469 bdev_close_while_hotremove(void)
3470 {
3471 	struct spdk_bdev *bdev;
3472 	struct spdk_bdev_desc *desc = NULL;
3473 	int rc = 0;
3474 
3475 	bdev = allocate_bdev("bdev");
3476 
3477 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc, &desc);
3478 	CU_ASSERT_EQUAL(rc, 0);
3479 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3480 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3481 
3482 	/* Simulate hot-unplug by unregistering bdev */
3483 	g_event_type1 = 0xFF;
3484 	g_unregister_arg = NULL;
3485 	g_unregister_rc = -1;
3486 	spdk_bdev_unregister(bdev, bdev_unregister_cb, (void *)0x12345678);
3487 	/* Close device while remove event is in flight */
3488 	spdk_bdev_close(desc);
3489 
3490 	/* Ensure that unregister callback is delayed */
3491 	CU_ASSERT_EQUAL(g_unregister_arg, NULL);
3492 	CU_ASSERT_EQUAL(g_unregister_rc, -1);
3493 
3494 	poll_threads();
3495 
3496 	/* Event callback shall not be issued because device was closed */
3497 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
3498 	/* Unregister callback is issued */
3499 	CU_ASSERT_EQUAL(g_unregister_arg, (void *)0x12345678);
3500 	CU_ASSERT_EQUAL(g_unregister_rc, 0);
3501 
3502 	free_bdev(bdev);
3503 }
3504 
3505 static void
3506 bdev_open_ext(void)
3507 {
3508 	struct spdk_bdev *bdev;
3509 	struct spdk_bdev_desc *desc1 = NULL;
3510 	struct spdk_bdev_desc *desc2 = NULL;
3511 	int rc = 0;
3512 
3513 	bdev = allocate_bdev("bdev");
3514 
3515 	rc = spdk_bdev_open_ext("bdev", true, NULL, NULL, &desc1);
3516 	CU_ASSERT_EQUAL(rc, -EINVAL);
3517 
3518 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc1, &desc1);
3519 	CU_ASSERT_EQUAL(rc, 0);
3520 
3521 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb2, &desc2, &desc2);
3522 	CU_ASSERT_EQUAL(rc, 0);
3523 
3524 	g_event_type1 = 0xFF;
3525 	g_event_type2 = 0xFF;
3526 
3527 	/* Simulate hot-unplug by unregistering bdev */
3528 	spdk_bdev_unregister(bdev, NULL, NULL);
3529 	poll_threads();
3530 
3531 	/* Check if correct events have been triggered in event callback fn */
3532 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_REMOVE);
3533 	CU_ASSERT_EQUAL(g_event_type2, SPDK_BDEV_EVENT_REMOVE);
3534 
3535 	free_bdev(bdev);
3536 	poll_threads();
3537 }
3538 
3539 struct timeout_io_cb_arg {
3540 	struct iovec iov;
3541 	uint8_t type;
3542 };
3543 
3544 static int
3545 bdev_channel_count_submitted_io(struct spdk_bdev_channel *ch)
3546 {
3547 	struct spdk_bdev_io *bdev_io;
3548 	int n = 0;
3549 
3550 	if (!ch) {
3551 		return -1;
3552 	}
3553 
3554 	TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
3555 		n++;
3556 	}
3557 
3558 	return n;
3559 }
3560 
3561 static void
3562 bdev_channel_io_timeout_cb(void *cb_arg, struct spdk_bdev_io *bdev_io)
3563 {
3564 	struct timeout_io_cb_arg *ctx = cb_arg;
3565 
3566 	ctx->type = bdev_io->type;
3567 	ctx->iov.iov_base = bdev_io->iov.iov_base;
3568 	ctx->iov.iov_len = bdev_io->iov.iov_len;
3569 }
3570 
3571 static void
3572 bdev_set_io_timeout(void)
3573 {
3574 	struct spdk_bdev *bdev;
3575 	struct spdk_bdev_desc *desc = NULL;
3576 	struct spdk_io_channel *io_ch = NULL;
3577 	struct spdk_bdev_channel *bdev_ch = NULL;
3578 	struct timeout_io_cb_arg cb_arg;
3579 
3580 	spdk_bdev_initialize(bdev_init_cb, NULL);
3581 
3582 	bdev = allocate_bdev("bdev");
3583 
3584 	CU_ASSERT(spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc) == 0);
3585 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3586 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3587 
3588 	io_ch = spdk_bdev_get_io_channel(desc);
3589 	CU_ASSERT(io_ch != NULL);
3590 
3591 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
3592 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
3593 
3594 	/* This is the part1.
3595 	 * We will check the bdev_ch->io_submitted list
3596 	 * TO make sure that it can link IOs and only the user submitted IOs
3597 	 */
3598 	CU_ASSERT(spdk_bdev_read(desc, io_ch, (void *)0x1000, 0, 4096, io_done, NULL) == 0);
3599 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3600 	CU_ASSERT(spdk_bdev_write(desc, io_ch, (void *)0x2000, 0, 4096, io_done, NULL) == 0);
3601 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3602 	stub_complete_io(1);
3603 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3604 	stub_complete_io(1);
3605 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3606 
3607 	/* Split IO */
3608 	bdev->optimal_io_boundary = 16;
3609 	bdev->split_on_optimal_io_boundary = true;
3610 
3611 	/* Now test that a single-vector command is split correctly.
3612 	 * Offset 14, length 8, payload 0xF000
3613 	 *  Child - Offset 14, length 2, payload 0xF000
3614 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
3615 	 *
3616 	 * Set up the expected values before calling spdk_bdev_read_blocks
3617 	 */
3618 	CU_ASSERT(spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3619 	/* We count all submitted IOs including IO that are generated by splitting. */
3620 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 3);
3621 	stub_complete_io(1);
3622 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3623 	stub_complete_io(1);
3624 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3625 
3626 	/* Also include the reset IO */
3627 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3628 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3629 	poll_threads();
3630 	stub_complete_io(1);
3631 	poll_threads();
3632 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3633 
3634 	/* This is part2
3635 	 * Test the desc timeout poller register
3636 	 */
3637 
3638 	/* Successfully set the timeout */
3639 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3640 	CU_ASSERT(desc->io_timeout_poller != NULL);
3641 	CU_ASSERT(desc->timeout_in_sec == 30);
3642 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3643 	CU_ASSERT(desc->cb_arg == &cb_arg);
3644 
3645 	/* Change the timeout limit */
3646 	CU_ASSERT(spdk_bdev_set_timeout(desc, 20, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3647 	CU_ASSERT(desc->io_timeout_poller != NULL);
3648 	CU_ASSERT(desc->timeout_in_sec == 20);
3649 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3650 	CU_ASSERT(desc->cb_arg == &cb_arg);
3651 
3652 	/* Disable the timeout */
3653 	CU_ASSERT(spdk_bdev_set_timeout(desc, 0, NULL, NULL) == 0);
3654 	CU_ASSERT(desc->io_timeout_poller == NULL);
3655 
3656 	/* This the part3
3657 	 * We will test to catch timeout IO and check whether the IO is
3658 	 * the submitted one.
3659 	 */
3660 	memset(&cb_arg, 0, sizeof(cb_arg));
3661 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3662 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0x1000, 0, 1, io_done, NULL) == 0);
3663 
3664 	/* Don't reach the limit */
3665 	spdk_delay_us(15 * spdk_get_ticks_hz());
3666 	poll_threads();
3667 	CU_ASSERT(cb_arg.type == 0);
3668 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3669 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3670 
3671 	/* 15 + 15 = 30 reach the limit */
3672 	spdk_delay_us(15 * spdk_get_ticks_hz());
3673 	poll_threads();
3674 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3675 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x1000);
3676 	CU_ASSERT(cb_arg.iov.iov_len == 1 * bdev->blocklen);
3677 	stub_complete_io(1);
3678 
3679 	/* Use the same split IO above and check the IO */
3680 	memset(&cb_arg, 0, sizeof(cb_arg));
3681 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3682 
3683 	/* The first child complete in time */
3684 	spdk_delay_us(15 * spdk_get_ticks_hz());
3685 	poll_threads();
3686 	stub_complete_io(1);
3687 	CU_ASSERT(cb_arg.type == 0);
3688 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3689 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3690 
3691 	/* The second child reach the limit */
3692 	spdk_delay_us(15 * spdk_get_ticks_hz());
3693 	poll_threads();
3694 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3695 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0xF000);
3696 	CU_ASSERT(cb_arg.iov.iov_len == 8 * bdev->blocklen);
3697 	stub_complete_io(1);
3698 
3699 	/* Also include the reset IO */
3700 	memset(&cb_arg, 0, sizeof(cb_arg));
3701 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3702 	spdk_delay_us(30 * spdk_get_ticks_hz());
3703 	poll_threads();
3704 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_RESET);
3705 	stub_complete_io(1);
3706 	poll_threads();
3707 
3708 	spdk_put_io_channel(io_ch);
3709 	spdk_bdev_close(desc);
3710 	free_bdev(bdev);
3711 	spdk_bdev_finish(bdev_fini_cb, NULL);
3712 	poll_threads();
3713 }
3714 
3715 static void
3716 lba_range_overlap(void)
3717 {
3718 	struct lba_range r1, r2;
3719 
3720 	r1.offset = 100;
3721 	r1.length = 50;
3722 
3723 	r2.offset = 0;
3724 	r2.length = 1;
3725 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3726 
3727 	r2.offset = 0;
3728 	r2.length = 100;
3729 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3730 
3731 	r2.offset = 0;
3732 	r2.length = 110;
3733 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3734 
3735 	r2.offset = 100;
3736 	r2.length = 10;
3737 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3738 
3739 	r2.offset = 110;
3740 	r2.length = 20;
3741 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3742 
3743 	r2.offset = 140;
3744 	r2.length = 150;
3745 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3746 
3747 	r2.offset = 130;
3748 	r2.length = 200;
3749 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3750 
3751 	r2.offset = 150;
3752 	r2.length = 100;
3753 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3754 
3755 	r2.offset = 110;
3756 	r2.length = 0;
3757 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3758 }
3759 
3760 static bool g_lock_lba_range_done;
3761 static bool g_unlock_lba_range_done;
3762 
3763 static void
3764 lock_lba_range_done(void *ctx, int status)
3765 {
3766 	g_lock_lba_range_done = true;
3767 }
3768 
3769 static void
3770 unlock_lba_range_done(void *ctx, int status)
3771 {
3772 	g_unlock_lba_range_done = true;
3773 }
3774 
3775 static void
3776 lock_lba_range_check_ranges(void)
3777 {
3778 	struct spdk_bdev *bdev;
3779 	struct spdk_bdev_desc *desc = NULL;
3780 	struct spdk_io_channel *io_ch;
3781 	struct spdk_bdev_channel *channel;
3782 	struct lba_range *range;
3783 	int ctx1;
3784 	int rc;
3785 
3786 	spdk_bdev_initialize(bdev_init_cb, NULL);
3787 
3788 	bdev = allocate_bdev("bdev0");
3789 
3790 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
3791 	CU_ASSERT(rc == 0);
3792 	CU_ASSERT(desc != NULL);
3793 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3794 	io_ch = spdk_bdev_get_io_channel(desc);
3795 	CU_ASSERT(io_ch != NULL);
3796 	channel = spdk_io_channel_get_ctx(io_ch);
3797 
3798 	g_lock_lba_range_done = false;
3799 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
3800 	CU_ASSERT(rc == 0);
3801 	poll_threads();
3802 
3803 	CU_ASSERT(g_lock_lba_range_done == true);
3804 	range = TAILQ_FIRST(&channel->locked_ranges);
3805 	SPDK_CU_ASSERT_FATAL(range != NULL);
3806 	CU_ASSERT(range->offset == 20);
3807 	CU_ASSERT(range->length == 10);
3808 	CU_ASSERT(range->owner_ch == channel);
3809 
3810 	/* Unlocks must exactly match a lock. */
3811 	g_unlock_lba_range_done = false;
3812 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 1, unlock_lba_range_done, &ctx1);
3813 	CU_ASSERT(rc == -EINVAL);
3814 	CU_ASSERT(g_unlock_lba_range_done == false);
3815 
3816 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
3817 	CU_ASSERT(rc == 0);
3818 	spdk_delay_us(100);
3819 	poll_threads();
3820 
3821 	CU_ASSERT(g_unlock_lba_range_done == true);
3822 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
3823 
3824 	spdk_put_io_channel(io_ch);
3825 	spdk_bdev_close(desc);
3826 	free_bdev(bdev);
3827 	spdk_bdev_finish(bdev_fini_cb, NULL);
3828 	poll_threads();
3829 }
3830 
3831 static void
3832 lock_lba_range_with_io_outstanding(void)
3833 {
3834 	struct spdk_bdev *bdev;
3835 	struct spdk_bdev_desc *desc = NULL;
3836 	struct spdk_io_channel *io_ch;
3837 	struct spdk_bdev_channel *channel;
3838 	struct lba_range *range;
3839 	char buf[4096];
3840 	int ctx1;
3841 	int rc;
3842 
3843 	spdk_bdev_initialize(bdev_init_cb, NULL);
3844 
3845 	bdev = allocate_bdev("bdev0");
3846 
3847 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
3848 	CU_ASSERT(rc == 0);
3849 	CU_ASSERT(desc != NULL);
3850 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3851 	io_ch = spdk_bdev_get_io_channel(desc);
3852 	CU_ASSERT(io_ch != NULL);
3853 	channel = spdk_io_channel_get_ctx(io_ch);
3854 
3855 	g_io_done = false;
3856 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
3857 	CU_ASSERT(rc == 0);
3858 
3859 	g_lock_lba_range_done = false;
3860 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
3861 	CU_ASSERT(rc == 0);
3862 	poll_threads();
3863 
3864 	/* The lock should immediately become valid, since there are no outstanding
3865 	 * write I/O.
3866 	 */
3867 	CU_ASSERT(g_io_done == false);
3868 	CU_ASSERT(g_lock_lba_range_done == true);
3869 	range = TAILQ_FIRST(&channel->locked_ranges);
3870 	SPDK_CU_ASSERT_FATAL(range != NULL);
3871 	CU_ASSERT(range->offset == 20);
3872 	CU_ASSERT(range->length == 10);
3873 	CU_ASSERT(range->owner_ch == channel);
3874 	CU_ASSERT(range->locked_ctx == &ctx1);
3875 
3876 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
3877 	CU_ASSERT(rc == 0);
3878 	stub_complete_io(1);
3879 	spdk_delay_us(100);
3880 	poll_threads();
3881 
3882 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
3883 
3884 	/* Now try again, but with a write I/O. */
3885 	g_io_done = false;
3886 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
3887 	CU_ASSERT(rc == 0);
3888 
3889 	g_lock_lba_range_done = false;
3890 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
3891 	CU_ASSERT(rc == 0);
3892 	poll_threads();
3893 
3894 	/* The lock should not be fully valid yet, since a write I/O is outstanding.
3895 	 * But note that the range should be on the channel's locked_list, to make sure no
3896 	 * new write I/O are started.
3897 	 */
3898 	CU_ASSERT(g_io_done == false);
3899 	CU_ASSERT(g_lock_lba_range_done == false);
3900 	range = TAILQ_FIRST(&channel->locked_ranges);
3901 	SPDK_CU_ASSERT_FATAL(range != NULL);
3902 	CU_ASSERT(range->offset == 20);
3903 	CU_ASSERT(range->length == 10);
3904 
3905 	/* Complete the write I/O.  This should make the lock valid (checked by confirming
3906 	 * our callback was invoked).
3907 	 */
3908 	stub_complete_io(1);
3909 	spdk_delay_us(100);
3910 	poll_threads();
3911 	CU_ASSERT(g_io_done == true);
3912 	CU_ASSERT(g_lock_lba_range_done == true);
3913 
3914 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
3915 	CU_ASSERT(rc == 0);
3916 	poll_threads();
3917 
3918 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
3919 
3920 	spdk_put_io_channel(io_ch);
3921 	spdk_bdev_close(desc);
3922 	free_bdev(bdev);
3923 	spdk_bdev_finish(bdev_fini_cb, NULL);
3924 	poll_threads();
3925 }
3926 
3927 static void
3928 lock_lba_range_overlapped(void)
3929 {
3930 	struct spdk_bdev *bdev;
3931 	struct spdk_bdev_desc *desc = NULL;
3932 	struct spdk_io_channel *io_ch;
3933 	struct spdk_bdev_channel *channel;
3934 	struct lba_range *range;
3935 	int ctx1;
3936 	int rc;
3937 
3938 	spdk_bdev_initialize(bdev_init_cb, NULL);
3939 
3940 	bdev = allocate_bdev("bdev0");
3941 
3942 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
3943 	CU_ASSERT(rc == 0);
3944 	CU_ASSERT(desc != NULL);
3945 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3946 	io_ch = spdk_bdev_get_io_channel(desc);
3947 	CU_ASSERT(io_ch != NULL);
3948 	channel = spdk_io_channel_get_ctx(io_ch);
3949 
3950 	/* Lock range 20-29. */
3951 	g_lock_lba_range_done = false;
3952 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
3953 	CU_ASSERT(rc == 0);
3954 	poll_threads();
3955 
3956 	CU_ASSERT(g_lock_lba_range_done == true);
3957 	range = TAILQ_FIRST(&channel->locked_ranges);
3958 	SPDK_CU_ASSERT_FATAL(range != NULL);
3959 	CU_ASSERT(range->offset == 20);
3960 	CU_ASSERT(range->length == 10);
3961 
3962 	/* Try to lock range 25-39.  It should not lock immediately, since it overlaps with
3963 	 * 20-29.
3964 	 */
3965 	g_lock_lba_range_done = false;
3966 	rc = bdev_lock_lba_range(desc, io_ch, 25, 15, lock_lba_range_done, &ctx1);
3967 	CU_ASSERT(rc == 0);
3968 	poll_threads();
3969 
3970 	CU_ASSERT(g_lock_lba_range_done == false);
3971 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
3972 	SPDK_CU_ASSERT_FATAL(range != NULL);
3973 	CU_ASSERT(range->offset == 25);
3974 	CU_ASSERT(range->length == 15);
3975 
3976 	/* Unlock 20-29.  This should result in range 25-39 now getting locked since it
3977 	 * no longer overlaps with an active lock.
3978 	 */
3979 	g_unlock_lba_range_done = false;
3980 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
3981 	CU_ASSERT(rc == 0);
3982 	poll_threads();
3983 
3984 	CU_ASSERT(g_unlock_lba_range_done == true);
3985 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
3986 	range = TAILQ_FIRST(&channel->locked_ranges);
3987 	SPDK_CU_ASSERT_FATAL(range != NULL);
3988 	CU_ASSERT(range->offset == 25);
3989 	CU_ASSERT(range->length == 15);
3990 
3991 	/* Lock 40-59.  This should immediately lock since it does not overlap with the
3992 	 * currently active 25-39 lock.
3993 	 */
3994 	g_lock_lba_range_done = false;
3995 	rc = bdev_lock_lba_range(desc, io_ch, 40, 20, lock_lba_range_done, &ctx1);
3996 	CU_ASSERT(rc == 0);
3997 	poll_threads();
3998 
3999 	CU_ASSERT(g_lock_lba_range_done == true);
4000 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4001 	SPDK_CU_ASSERT_FATAL(range != NULL);
4002 	range = TAILQ_NEXT(range, tailq);
4003 	SPDK_CU_ASSERT_FATAL(range != NULL);
4004 	CU_ASSERT(range->offset == 40);
4005 	CU_ASSERT(range->length == 20);
4006 
4007 	/* Try to lock 35-44.  Note that this overlaps with both 25-39 and 40-59. */
4008 	g_lock_lba_range_done = false;
4009 	rc = bdev_lock_lba_range(desc, io_ch, 35, 10, lock_lba_range_done, &ctx1);
4010 	CU_ASSERT(rc == 0);
4011 	poll_threads();
4012 
4013 	CU_ASSERT(g_lock_lba_range_done == false);
4014 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4015 	SPDK_CU_ASSERT_FATAL(range != NULL);
4016 	CU_ASSERT(range->offset == 35);
4017 	CU_ASSERT(range->length == 10);
4018 
4019 	/* Unlock 25-39.  Make sure that 35-44 is still in the pending list, since
4020 	 * the 40-59 lock is still active.
4021 	 */
4022 	g_unlock_lba_range_done = false;
4023 	rc = bdev_unlock_lba_range(desc, io_ch, 25, 15, unlock_lba_range_done, &ctx1);
4024 	CU_ASSERT(rc == 0);
4025 	poll_threads();
4026 
4027 	CU_ASSERT(g_unlock_lba_range_done == true);
4028 	CU_ASSERT(g_lock_lba_range_done == false);
4029 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4030 	SPDK_CU_ASSERT_FATAL(range != NULL);
4031 	CU_ASSERT(range->offset == 35);
4032 	CU_ASSERT(range->length == 10);
4033 
4034 	/* Unlock 40-59.  This should result in 35-44 now getting locked, since there are
4035 	 * no longer any active overlapping locks.
4036 	 */
4037 	g_unlock_lba_range_done = false;
4038 	rc = bdev_unlock_lba_range(desc, io_ch, 40, 20, unlock_lba_range_done, &ctx1);
4039 	CU_ASSERT(rc == 0);
4040 	poll_threads();
4041 
4042 	CU_ASSERT(g_unlock_lba_range_done == true);
4043 	CU_ASSERT(g_lock_lba_range_done == true);
4044 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4045 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4046 	SPDK_CU_ASSERT_FATAL(range != NULL);
4047 	CU_ASSERT(range->offset == 35);
4048 	CU_ASSERT(range->length == 10);
4049 
4050 	/* Finally, unlock 35-44. */
4051 	g_unlock_lba_range_done = false;
4052 	rc = bdev_unlock_lba_range(desc, io_ch, 35, 10, unlock_lba_range_done, &ctx1);
4053 	CU_ASSERT(rc == 0);
4054 	poll_threads();
4055 
4056 	CU_ASSERT(g_unlock_lba_range_done == true);
4057 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.locked_ranges));
4058 
4059 	spdk_put_io_channel(io_ch);
4060 	spdk_bdev_close(desc);
4061 	free_bdev(bdev);
4062 	spdk_bdev_finish(bdev_fini_cb, NULL);
4063 	poll_threads();
4064 }
4065 
4066 static void
4067 abort_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
4068 {
4069 	g_abort_done = true;
4070 	g_abort_status = bdev_io->internal.status;
4071 	spdk_bdev_free_io(bdev_io);
4072 }
4073 
4074 static void
4075 bdev_io_abort(void)
4076 {
4077 	struct spdk_bdev *bdev;
4078 	struct spdk_bdev_desc *desc = NULL;
4079 	struct spdk_io_channel *io_ch;
4080 	struct spdk_bdev_channel *channel;
4081 	struct spdk_bdev_mgmt_channel *mgmt_ch;
4082 	struct spdk_bdev_opts bdev_opts = {};
4083 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
4084 	uint64_t io_ctx1 = 0, io_ctx2 = 0, i;
4085 	int rc;
4086 
4087 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4088 	bdev_opts.bdev_io_pool_size = 7;
4089 	bdev_opts.bdev_io_cache_size = 2;
4090 
4091 	rc = spdk_bdev_set_opts(&bdev_opts);
4092 	CU_ASSERT(rc == 0);
4093 	spdk_bdev_initialize(bdev_init_cb, NULL);
4094 
4095 	bdev = allocate_bdev("bdev0");
4096 
4097 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4098 	CU_ASSERT(rc == 0);
4099 	CU_ASSERT(desc != NULL);
4100 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4101 	io_ch = spdk_bdev_get_io_channel(desc);
4102 	CU_ASSERT(io_ch != NULL);
4103 	channel = spdk_io_channel_get_ctx(io_ch);
4104 	mgmt_ch = channel->shared_resource->mgmt_ch;
4105 
4106 	g_abort_done = false;
4107 
4108 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, false);
4109 
4110 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4111 	CU_ASSERT(rc == -ENOTSUP);
4112 
4113 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, true);
4114 
4115 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx2, abort_done, NULL);
4116 	CU_ASSERT(rc == 0);
4117 	CU_ASSERT(g_abort_done == true);
4118 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_FAILED);
4119 
4120 	/* Test the case that the target I/O was successfully aborted. */
4121 	g_io_done = false;
4122 
4123 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4124 	CU_ASSERT(rc == 0);
4125 	CU_ASSERT(g_io_done == false);
4126 
4127 	g_abort_done = false;
4128 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4129 
4130 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4131 	CU_ASSERT(rc == 0);
4132 	CU_ASSERT(g_io_done == true);
4133 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4134 	stub_complete_io(1);
4135 	CU_ASSERT(g_abort_done == true);
4136 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4137 
4138 	/* Test the case that the target I/O was not aborted because it completed
4139 	 * in the middle of execution of the abort.
4140 	 */
4141 	g_io_done = false;
4142 
4143 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4144 	CU_ASSERT(rc == 0);
4145 	CU_ASSERT(g_io_done == false);
4146 
4147 	g_abort_done = false;
4148 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4149 
4150 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4151 	CU_ASSERT(rc == 0);
4152 	CU_ASSERT(g_io_done == false);
4153 
4154 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4155 	stub_complete_io(1);
4156 	CU_ASSERT(g_io_done == true);
4157 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4158 
4159 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4160 	stub_complete_io(1);
4161 	CU_ASSERT(g_abort_done == true);
4162 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4163 
4164 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4165 
4166 	bdev->optimal_io_boundary = 16;
4167 	bdev->split_on_optimal_io_boundary = true;
4168 
4169 	/* Test that a single-vector command which is split is aborted correctly.
4170 	 * Offset 14, length 8, payload 0xF000
4171 	 *  Child - Offset 14, length 2, payload 0xF000
4172 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
4173 	 */
4174 	g_io_done = false;
4175 
4176 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, &io_ctx1);
4177 	CU_ASSERT(rc == 0);
4178 	CU_ASSERT(g_io_done == false);
4179 
4180 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4181 
4182 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4183 
4184 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4185 	CU_ASSERT(rc == 0);
4186 	CU_ASSERT(g_io_done == true);
4187 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4188 	stub_complete_io(2);
4189 	CU_ASSERT(g_abort_done == true);
4190 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4191 
4192 	/* Test that a multi-vector command that needs to be split by strip and then
4193 	 * needs to be split is aborted correctly. Abort is requested before the second
4194 	 * child I/O was submitted. The parent I/O should complete with failure without
4195 	 * submitting the second child I/O.
4196 	 */
4197 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
4198 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
4199 		iov[i].iov_len = 512;
4200 	}
4201 
4202 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
4203 	g_io_done = false;
4204 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
4205 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, &io_ctx1);
4206 	CU_ASSERT(rc == 0);
4207 	CU_ASSERT(g_io_done == false);
4208 
4209 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4210 
4211 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4212 
4213 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4214 	CU_ASSERT(rc == 0);
4215 	CU_ASSERT(g_io_done == true);
4216 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4217 	stub_complete_io(1);
4218 	CU_ASSERT(g_abort_done == true);
4219 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4220 
4221 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4222 
4223 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4224 
4225 	bdev->optimal_io_boundary = 16;
4226 	g_io_done = false;
4227 
4228 	/* Test that a ingle-vector command which is split is aborted correctly.
4229 	 * Differently from the above, the child abort request will be submitted
4230 	 * sequentially due to the capacity of spdk_bdev_io.
4231 	 */
4232 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 50, io_done, &io_ctx1);
4233 	CU_ASSERT(rc == 0);
4234 	CU_ASSERT(g_io_done == false);
4235 
4236 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4237 
4238 	g_abort_done = false;
4239 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4240 
4241 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4242 	CU_ASSERT(rc == 0);
4243 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
4244 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4245 
4246 	stub_complete_io(1);
4247 	CU_ASSERT(g_io_done == true);
4248 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4249 	stub_complete_io(3);
4250 	CU_ASSERT(g_abort_done == true);
4251 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4252 
4253 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4254 
4255 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4256 
4257 	spdk_put_io_channel(io_ch);
4258 	spdk_bdev_close(desc);
4259 	free_bdev(bdev);
4260 	spdk_bdev_finish(bdev_fini_cb, NULL);
4261 	poll_threads();
4262 }
4263 
4264 static void
4265 bdev_set_options_test(void)
4266 {
4267 	struct spdk_bdev_opts bdev_opts = {};
4268 	int rc;
4269 
4270 	/* Case1: Do not set opts_size */
4271 	rc = spdk_bdev_set_opts(&bdev_opts);
4272 	CU_ASSERT(rc == -1);
4273 
4274 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4275 	bdev_opts.bdev_io_pool_size = 4;
4276 	bdev_opts.bdev_io_cache_size = 2;
4277 	bdev_opts.small_buf_pool_size = 4;
4278 
4279 	/* Case 2: Do not set valid small_buf_pool_size and large_buf_pool_size */
4280 	rc = spdk_bdev_set_opts(&bdev_opts);
4281 	CU_ASSERT(rc == -1);
4282 
4283 	/* Case 3: Do not set valid large_buf_pool_size */
4284 	bdev_opts.small_buf_pool_size = BUF_SMALL_POOL_SIZE;
4285 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE - 1;
4286 	rc = spdk_bdev_set_opts(&bdev_opts);
4287 	CU_ASSERT(rc == -1);
4288 
4289 	/* Case4: set valid large buf_pool_size */
4290 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE;
4291 	rc = spdk_bdev_set_opts(&bdev_opts);
4292 	CU_ASSERT(rc == 0);
4293 
4294 	/* Case5: Set different valid value for small and large buf pool */
4295 	bdev_opts.large_buf_pool_size = BUF_SMALL_POOL_SIZE + 3;
4296 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE + 3;
4297 	rc = spdk_bdev_set_opts(&bdev_opts);
4298 	CU_ASSERT(rc == 0);
4299 }
4300 
4301 int
4302 main(int argc, char **argv)
4303 {
4304 	CU_pSuite		suite = NULL;
4305 	unsigned int		num_failures;
4306 
4307 	CU_set_error_action(CUEA_ABORT);
4308 	CU_initialize_registry();
4309 
4310 	suite = CU_add_suite("bdev", null_init, null_clean);
4311 
4312 	CU_ADD_TEST(suite, bytes_to_blocks_test);
4313 	CU_ADD_TEST(suite, num_blocks_test);
4314 	CU_ADD_TEST(suite, io_valid_test);
4315 	CU_ADD_TEST(suite, open_write_test);
4316 	CU_ADD_TEST(suite, alias_add_del_test);
4317 	CU_ADD_TEST(suite, get_device_stat_test);
4318 	CU_ADD_TEST(suite, bdev_io_types_test);
4319 	CU_ADD_TEST(suite, bdev_io_wait_test);
4320 	CU_ADD_TEST(suite, bdev_io_spans_split_test);
4321 	CU_ADD_TEST(suite, bdev_io_boundary_split_test);
4322 	CU_ADD_TEST(suite, bdev_io_max_size_and_segment_split_test);
4323 	CU_ADD_TEST(suite, bdev_io_mix_split_test);
4324 	CU_ADD_TEST(suite, bdev_io_split_with_io_wait);
4325 	CU_ADD_TEST(suite, bdev_io_alignment_with_boundary);
4326 	CU_ADD_TEST(suite, bdev_io_alignment);
4327 	CU_ADD_TEST(suite, bdev_histograms);
4328 	CU_ADD_TEST(suite, bdev_write_zeroes);
4329 	CU_ADD_TEST(suite, bdev_compare_and_write);
4330 	CU_ADD_TEST(suite, bdev_compare);
4331 	CU_ADD_TEST(suite, bdev_open_while_hotremove);
4332 	CU_ADD_TEST(suite, bdev_close_while_hotremove);
4333 	CU_ADD_TEST(suite, bdev_open_ext);
4334 	CU_ADD_TEST(suite, bdev_set_io_timeout);
4335 	CU_ADD_TEST(suite, lba_range_overlap);
4336 	CU_ADD_TEST(suite, lock_lba_range_check_ranges);
4337 	CU_ADD_TEST(suite, lock_lba_range_with_io_outstanding);
4338 	CU_ADD_TEST(suite, lock_lba_range_overlapped);
4339 	CU_ADD_TEST(suite, bdev_io_abort);
4340 	CU_ADD_TEST(suite, bdev_set_options_test);
4341 
4342 	allocate_cores(1);
4343 	allocate_threads(1);
4344 	set_thread(0);
4345 
4346 	CU_basic_set_mode(CU_BRM_VERBOSE);
4347 	CU_basic_run_tests();
4348 	num_failures = CU_get_number_of_failures();
4349 	CU_cleanup_registry();
4350 
4351 	free_threads();
4352 	free_cores();
4353 
4354 	return num_failures;
4355 }
4356