xref: /spdk/test/unit/lib/bdev/bdev.c/bdev_ut.c (revision cc6920a4763d4b9a43aa40583c8397d8f14fa100)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
6  *   Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "spdk_cunit.h"
36 
37 #include "common/lib/ut_multithread.c"
38 #include "unit/lib/json_mock.c"
39 
40 #include "spdk/config.h"
41 /* HACK: disable VTune integration so the unit test doesn't need VTune headers and libs to build */
42 #undef SPDK_CONFIG_VTUNE
43 
44 #include "bdev/bdev.c"
45 
46 DEFINE_STUB(spdk_notify_send, uint64_t, (const char *type, const char *ctx), 0);
47 DEFINE_STUB(spdk_notify_type_register, struct spdk_notify_type *, (const char *type), NULL);
48 
49 int g_status;
50 int g_count;
51 enum spdk_bdev_event_type g_event_type1;
52 enum spdk_bdev_event_type g_event_type2;
53 struct spdk_histogram_data *g_histogram;
54 void *g_unregister_arg;
55 int g_unregister_rc;
56 
57 void
58 spdk_scsi_nvme_translate(const struct spdk_bdev_io *bdev_io,
59 			 int *sc, int *sk, int *asc, int *ascq)
60 {
61 }
62 
63 static int
64 null_init(void)
65 {
66 	return 0;
67 }
68 
69 static int
70 null_clean(void)
71 {
72 	return 0;
73 }
74 
75 static int
76 stub_destruct(void *ctx)
77 {
78 	return 0;
79 }
80 
81 struct ut_expected_io {
82 	uint8_t				type;
83 	uint64_t			offset;
84 	uint64_t			length;
85 	int				iovcnt;
86 	struct iovec			iov[BDEV_IO_NUM_CHILD_IOV];
87 	void				*md_buf;
88 	struct spdk_bdev_ext_io_opts	*ext_io_opts;
89 	TAILQ_ENTRY(ut_expected_io)	link;
90 };
91 
92 struct bdev_ut_channel {
93 	TAILQ_HEAD(, spdk_bdev_io)	outstanding_io;
94 	uint32_t			outstanding_io_count;
95 	TAILQ_HEAD(, ut_expected_io)	expected_io;
96 };
97 
98 static bool g_io_done;
99 static struct spdk_bdev_io *g_bdev_io;
100 static enum spdk_bdev_io_status g_io_status;
101 static enum spdk_bdev_io_status g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
102 static uint32_t g_bdev_ut_io_device;
103 static struct bdev_ut_channel *g_bdev_ut_channel;
104 static void *g_compare_read_buf;
105 static uint32_t g_compare_read_buf_len;
106 static void *g_compare_write_buf;
107 static uint32_t g_compare_write_buf_len;
108 static bool g_abort_done;
109 static enum spdk_bdev_io_status g_abort_status;
110 static void *g_zcopy_read_buf;
111 static uint32_t g_zcopy_read_buf_len;
112 static void *g_zcopy_write_buf;
113 static uint32_t g_zcopy_write_buf_len;
114 static struct spdk_bdev_io *g_zcopy_bdev_io;
115 
116 static struct ut_expected_io *
117 ut_alloc_expected_io(uint8_t type, uint64_t offset, uint64_t length, int iovcnt)
118 {
119 	struct ut_expected_io *expected_io;
120 
121 	expected_io = calloc(1, sizeof(*expected_io));
122 	SPDK_CU_ASSERT_FATAL(expected_io != NULL);
123 
124 	expected_io->type = type;
125 	expected_io->offset = offset;
126 	expected_io->length = length;
127 	expected_io->iovcnt = iovcnt;
128 
129 	return expected_io;
130 }
131 
132 static void
133 ut_expected_io_set_iov(struct ut_expected_io *expected_io, int pos, void *base, size_t len)
134 {
135 	expected_io->iov[pos].iov_base = base;
136 	expected_io->iov[pos].iov_len = len;
137 }
138 
139 static void
140 stub_submit_request(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
141 {
142 	struct bdev_ut_channel *ch = spdk_io_channel_get_ctx(_ch);
143 	struct ut_expected_io *expected_io;
144 	struct iovec *iov, *expected_iov;
145 	struct spdk_bdev_io *bio_to_abort;
146 	int i;
147 
148 	g_bdev_io = bdev_io;
149 
150 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
151 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
152 
153 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
154 		CU_ASSERT(g_compare_read_buf_len == len);
155 		memcpy(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len);
156 	}
157 
158 	if (g_compare_write_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
159 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
160 
161 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
162 		CU_ASSERT(g_compare_write_buf_len == len);
163 		memcpy(g_compare_write_buf, bdev_io->u.bdev.iovs[0].iov_base, len);
164 	}
165 
166 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_COMPARE) {
167 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
168 
169 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
170 		CU_ASSERT(g_compare_read_buf_len == len);
171 		if (memcmp(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len)) {
172 			g_io_exp_status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
173 		}
174 	}
175 
176 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) {
177 		if (g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS) {
178 			TAILQ_FOREACH(bio_to_abort, &ch->outstanding_io, module_link) {
179 				if (bio_to_abort == bdev_io->u.abort.bio_to_abort) {
180 					TAILQ_REMOVE(&ch->outstanding_io, bio_to_abort, module_link);
181 					ch->outstanding_io_count--;
182 					spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_FAILED);
183 					break;
184 				}
185 			}
186 		}
187 	}
188 
189 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) {
190 		if (bdev_io->u.bdev.zcopy.start) {
191 			g_zcopy_bdev_io = bdev_io;
192 			if (bdev_io->u.bdev.zcopy.populate) {
193 				/* Start of a read */
194 				CU_ASSERT(g_zcopy_read_buf != NULL);
195 				CU_ASSERT(g_zcopy_read_buf_len > 0);
196 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_read_buf;
197 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_read_buf_len;
198 				bdev_io->u.bdev.iovcnt = 1;
199 			} else {
200 				/* Start of a write */
201 				CU_ASSERT(g_zcopy_write_buf != NULL);
202 				CU_ASSERT(g_zcopy_write_buf_len > 0);
203 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_write_buf;
204 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_write_buf_len;
205 				bdev_io->u.bdev.iovcnt = 1;
206 			}
207 		} else {
208 			if (bdev_io->u.bdev.zcopy.commit) {
209 				/* End of write */
210 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_write_buf);
211 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_write_buf_len);
212 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
213 				g_zcopy_write_buf = NULL;
214 				g_zcopy_write_buf_len = 0;
215 			} else {
216 				/* End of read */
217 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_read_buf);
218 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_read_buf_len);
219 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
220 				g_zcopy_read_buf = NULL;
221 				g_zcopy_read_buf_len = 0;
222 			}
223 		}
224 	}
225 
226 	TAILQ_INSERT_TAIL(&ch->outstanding_io, bdev_io, module_link);
227 	ch->outstanding_io_count++;
228 
229 	expected_io = TAILQ_FIRST(&ch->expected_io);
230 	if (expected_io == NULL) {
231 		return;
232 	}
233 	TAILQ_REMOVE(&ch->expected_io, expected_io, link);
234 
235 	if (expected_io->type != SPDK_BDEV_IO_TYPE_INVALID) {
236 		CU_ASSERT(bdev_io->type == expected_io->type);
237 	}
238 
239 	if (expected_io->md_buf != NULL) {
240 		CU_ASSERT(expected_io->md_buf == bdev_io->u.bdev.md_buf);
241 	}
242 
243 	if (expected_io->length == 0) {
244 		free(expected_io);
245 		return;
246 	}
247 
248 	CU_ASSERT(expected_io->offset == bdev_io->u.bdev.offset_blocks);
249 	CU_ASSERT(expected_io->length = bdev_io->u.bdev.num_blocks);
250 
251 	if (expected_io->iovcnt == 0) {
252 		free(expected_io);
253 		/* UNMAP, WRITE_ZEROES and FLUSH don't have iovs, so we can just return now. */
254 		return;
255 	}
256 
257 	CU_ASSERT(expected_io->iovcnt == bdev_io->u.bdev.iovcnt);
258 	for (i = 0; i < expected_io->iovcnt; i++) {
259 		iov = &bdev_io->u.bdev.iovs[i];
260 		expected_iov = &expected_io->iov[i];
261 		CU_ASSERT(iov->iov_len == expected_iov->iov_len);
262 		CU_ASSERT(iov->iov_base == expected_iov->iov_base);
263 	}
264 
265 	if (expected_io->ext_io_opts) {
266 		CU_ASSERT(expected_io->ext_io_opts == bdev_io->internal.ext_opts)
267 	}
268 
269 	free(expected_io);
270 }
271 
272 static void
273 stub_submit_request_get_buf_cb(struct spdk_io_channel *_ch,
274 			       struct spdk_bdev_io *bdev_io, bool success)
275 {
276 	CU_ASSERT(success == true);
277 
278 	stub_submit_request(_ch, bdev_io);
279 }
280 
281 static void
282 stub_submit_request_get_buf(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
283 {
284 	spdk_bdev_io_get_buf(bdev_io, stub_submit_request_get_buf_cb,
285 			     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
286 }
287 
288 static uint32_t
289 stub_complete_io(uint32_t num_to_complete)
290 {
291 	struct bdev_ut_channel *ch = g_bdev_ut_channel;
292 	struct spdk_bdev_io *bdev_io;
293 	static enum spdk_bdev_io_status io_status;
294 	uint32_t num_completed = 0;
295 
296 	while (num_completed < num_to_complete) {
297 		if (TAILQ_EMPTY(&ch->outstanding_io)) {
298 			break;
299 		}
300 		bdev_io = TAILQ_FIRST(&ch->outstanding_io);
301 		TAILQ_REMOVE(&ch->outstanding_io, bdev_io, module_link);
302 		ch->outstanding_io_count--;
303 		io_status = g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS ? SPDK_BDEV_IO_STATUS_SUCCESS :
304 			    g_io_exp_status;
305 		spdk_bdev_io_complete(bdev_io, io_status);
306 		num_completed++;
307 	}
308 
309 	return num_completed;
310 }
311 
312 static struct spdk_io_channel *
313 bdev_ut_get_io_channel(void *ctx)
314 {
315 	return spdk_get_io_channel(&g_bdev_ut_io_device);
316 }
317 
318 static bool g_io_types_supported[SPDK_BDEV_NUM_IO_TYPES] = {
319 	[SPDK_BDEV_IO_TYPE_READ]		= true,
320 	[SPDK_BDEV_IO_TYPE_WRITE]		= true,
321 	[SPDK_BDEV_IO_TYPE_COMPARE]		= true,
322 	[SPDK_BDEV_IO_TYPE_UNMAP]		= true,
323 	[SPDK_BDEV_IO_TYPE_FLUSH]		= true,
324 	[SPDK_BDEV_IO_TYPE_RESET]		= true,
325 	[SPDK_BDEV_IO_TYPE_NVME_ADMIN]		= true,
326 	[SPDK_BDEV_IO_TYPE_NVME_IO]		= true,
327 	[SPDK_BDEV_IO_TYPE_NVME_IO_MD]		= true,
328 	[SPDK_BDEV_IO_TYPE_WRITE_ZEROES]	= true,
329 	[SPDK_BDEV_IO_TYPE_ZCOPY]		= true,
330 	[SPDK_BDEV_IO_TYPE_ABORT]		= true,
331 };
332 
333 static void
334 ut_enable_io_type(enum spdk_bdev_io_type io_type, bool enable)
335 {
336 	g_io_types_supported[io_type] = enable;
337 }
338 
339 static bool
340 stub_io_type_supported(void *_bdev, enum spdk_bdev_io_type io_type)
341 {
342 	return g_io_types_supported[io_type];
343 }
344 
345 static struct spdk_bdev_fn_table fn_table = {
346 	.destruct = stub_destruct,
347 	.submit_request = stub_submit_request,
348 	.get_io_channel = bdev_ut_get_io_channel,
349 	.io_type_supported = stub_io_type_supported,
350 };
351 
352 static int
353 bdev_ut_create_ch(void *io_device, void *ctx_buf)
354 {
355 	struct bdev_ut_channel *ch = ctx_buf;
356 
357 	CU_ASSERT(g_bdev_ut_channel == NULL);
358 	g_bdev_ut_channel = ch;
359 
360 	TAILQ_INIT(&ch->outstanding_io);
361 	ch->outstanding_io_count = 0;
362 	TAILQ_INIT(&ch->expected_io);
363 	return 0;
364 }
365 
366 static void
367 bdev_ut_destroy_ch(void *io_device, void *ctx_buf)
368 {
369 	CU_ASSERT(g_bdev_ut_channel != NULL);
370 	g_bdev_ut_channel = NULL;
371 }
372 
373 struct spdk_bdev_module bdev_ut_if;
374 
375 static int
376 bdev_ut_module_init(void)
377 {
378 	spdk_io_device_register(&g_bdev_ut_io_device, bdev_ut_create_ch, bdev_ut_destroy_ch,
379 				sizeof(struct bdev_ut_channel), NULL);
380 	spdk_bdev_module_init_done(&bdev_ut_if);
381 	return 0;
382 }
383 
384 static void
385 bdev_ut_module_fini(void)
386 {
387 	spdk_io_device_unregister(&g_bdev_ut_io_device, NULL);
388 }
389 
390 struct spdk_bdev_module bdev_ut_if = {
391 	.name = "bdev_ut",
392 	.module_init = bdev_ut_module_init,
393 	.module_fini = bdev_ut_module_fini,
394 	.async_init = true,
395 };
396 
397 static void vbdev_ut_examine(struct spdk_bdev *bdev);
398 
399 static int
400 vbdev_ut_module_init(void)
401 {
402 	return 0;
403 }
404 
405 static void
406 vbdev_ut_module_fini(void)
407 {
408 }
409 
410 struct spdk_bdev_module vbdev_ut_if = {
411 	.name = "vbdev_ut",
412 	.module_init = vbdev_ut_module_init,
413 	.module_fini = vbdev_ut_module_fini,
414 	.examine_config = vbdev_ut_examine,
415 };
416 
417 SPDK_BDEV_MODULE_REGISTER(bdev_ut, &bdev_ut_if)
418 SPDK_BDEV_MODULE_REGISTER(vbdev_ut, &vbdev_ut_if)
419 
420 static void
421 vbdev_ut_examine(struct spdk_bdev *bdev)
422 {
423 	spdk_bdev_module_examine_done(&vbdev_ut_if);
424 }
425 
426 static struct spdk_bdev *
427 allocate_bdev(char *name)
428 {
429 	struct spdk_bdev *bdev;
430 	int rc;
431 
432 	bdev = calloc(1, sizeof(*bdev));
433 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
434 
435 	bdev->name = name;
436 	bdev->fn_table = &fn_table;
437 	bdev->module = &bdev_ut_if;
438 	bdev->blockcnt = 1024;
439 	bdev->blocklen = 512;
440 
441 	rc = spdk_bdev_register(bdev);
442 	CU_ASSERT(rc == 0);
443 
444 	return bdev;
445 }
446 
447 static struct spdk_bdev *
448 allocate_vbdev(char *name)
449 {
450 	struct spdk_bdev *bdev;
451 	int rc;
452 
453 	bdev = calloc(1, sizeof(*bdev));
454 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
455 
456 	bdev->name = name;
457 	bdev->fn_table = &fn_table;
458 	bdev->module = &vbdev_ut_if;
459 
460 	rc = spdk_bdev_register(bdev);
461 	CU_ASSERT(rc == 0);
462 
463 	return bdev;
464 }
465 
466 static void
467 free_bdev(struct spdk_bdev *bdev)
468 {
469 	spdk_bdev_unregister(bdev, NULL, NULL);
470 	poll_threads();
471 	memset(bdev, 0xFF, sizeof(*bdev));
472 	free(bdev);
473 }
474 
475 static void
476 free_vbdev(struct spdk_bdev *bdev)
477 {
478 	spdk_bdev_unregister(bdev, NULL, NULL);
479 	poll_threads();
480 	memset(bdev, 0xFF, sizeof(*bdev));
481 	free(bdev);
482 }
483 
484 static void
485 get_device_stat_cb(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat, void *cb_arg, int rc)
486 {
487 	const char *bdev_name;
488 
489 	CU_ASSERT(bdev != NULL);
490 	CU_ASSERT(rc == 0);
491 	bdev_name = spdk_bdev_get_name(bdev);
492 	CU_ASSERT_STRING_EQUAL(bdev_name, "bdev0");
493 
494 	free(stat);
495 
496 	*(bool *)cb_arg = true;
497 }
498 
499 static void
500 bdev_unregister_cb(void *cb_arg, int rc)
501 {
502 	g_unregister_arg = cb_arg;
503 	g_unregister_rc = rc;
504 }
505 
506 static void
507 bdev_ut_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
508 {
509 }
510 
511 static void
512 bdev_open_cb1(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
513 {
514 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
515 
516 	g_event_type1 = type;
517 	if (SPDK_BDEV_EVENT_REMOVE == type) {
518 		spdk_bdev_close(desc);
519 	}
520 }
521 
522 static void
523 bdev_open_cb2(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
524 {
525 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
526 
527 	g_event_type2 = type;
528 	if (SPDK_BDEV_EVENT_REMOVE == type) {
529 		spdk_bdev_close(desc);
530 	}
531 }
532 
533 static void
534 get_device_stat_test(void)
535 {
536 	struct spdk_bdev *bdev;
537 	struct spdk_bdev_io_stat *stat;
538 	bool done;
539 
540 	bdev = allocate_bdev("bdev0");
541 	stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
542 	if (stat == NULL) {
543 		free_bdev(bdev);
544 		return;
545 	}
546 
547 	done = false;
548 	spdk_bdev_get_device_stat(bdev, stat, get_device_stat_cb, &done);
549 	while (!done) { poll_threads(); }
550 
551 	free_bdev(bdev);
552 }
553 
554 static void
555 open_write_test(void)
556 {
557 	struct spdk_bdev *bdev[9];
558 	struct spdk_bdev_desc *desc[9] = {};
559 	int rc;
560 
561 	/*
562 	 * Create a tree of bdevs to test various open w/ write cases.
563 	 *
564 	 * bdev0 through bdev3 are physical block devices, such as NVMe
565 	 * namespaces or Ceph block devices.
566 	 *
567 	 * bdev4 is a virtual bdev with multiple base bdevs.  This models
568 	 * caching or RAID use cases.
569 	 *
570 	 * bdev5 through bdev7 are all virtual bdevs with the same base
571 	 * bdev (except bdev7). This models partitioning or logical volume
572 	 * use cases.
573 	 *
574 	 * bdev7 is a virtual bdev with multiple base bdevs. One of base bdevs
575 	 * (bdev2) is shared with other virtual bdevs: bdev5 and bdev6. This
576 	 * models caching, RAID, partitioning or logical volumes use cases.
577 	 *
578 	 * bdev8 is a virtual bdev with multiple base bdevs, but these
579 	 * base bdevs are themselves virtual bdevs.
580 	 *
581 	 *                bdev8
582 	 *                  |
583 	 *            +----------+
584 	 *            |          |
585 	 *          bdev4      bdev5   bdev6   bdev7
586 	 *            |          |       |       |
587 	 *        +---+---+      +---+   +   +---+---+
588 	 *        |       |           \  |  /         \
589 	 *      bdev0   bdev1          bdev2         bdev3
590 	 */
591 
592 	bdev[0] = allocate_bdev("bdev0");
593 	rc = spdk_bdev_module_claim_bdev(bdev[0], NULL, &bdev_ut_if);
594 	CU_ASSERT(rc == 0);
595 
596 	bdev[1] = allocate_bdev("bdev1");
597 	rc = spdk_bdev_module_claim_bdev(bdev[1], NULL, &bdev_ut_if);
598 	CU_ASSERT(rc == 0);
599 
600 	bdev[2] = allocate_bdev("bdev2");
601 	rc = spdk_bdev_module_claim_bdev(bdev[2], NULL, &bdev_ut_if);
602 	CU_ASSERT(rc == 0);
603 
604 	bdev[3] = allocate_bdev("bdev3");
605 	rc = spdk_bdev_module_claim_bdev(bdev[3], NULL, &bdev_ut_if);
606 	CU_ASSERT(rc == 0);
607 
608 	bdev[4] = allocate_vbdev("bdev4");
609 	rc = spdk_bdev_module_claim_bdev(bdev[4], NULL, &bdev_ut_if);
610 	CU_ASSERT(rc == 0);
611 
612 	bdev[5] = allocate_vbdev("bdev5");
613 	rc = spdk_bdev_module_claim_bdev(bdev[5], NULL, &bdev_ut_if);
614 	CU_ASSERT(rc == 0);
615 
616 	bdev[6] = allocate_vbdev("bdev6");
617 
618 	bdev[7] = allocate_vbdev("bdev7");
619 
620 	bdev[8] = allocate_vbdev("bdev8");
621 
622 	/* Open bdev0 read-only.  This should succeed. */
623 	rc = spdk_bdev_open_ext("bdev0", false, bdev_ut_event_cb, NULL, &desc[0]);
624 	CU_ASSERT(rc == 0);
625 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
626 	CU_ASSERT(bdev[0] == spdk_bdev_desc_get_bdev(desc[0]));
627 	spdk_bdev_close(desc[0]);
628 
629 	/*
630 	 * Open bdev1 read/write.  This should fail since bdev1 has been claimed
631 	 * by a vbdev module.
632 	 */
633 	rc = spdk_bdev_open_ext("bdev1", true, bdev_ut_event_cb, NULL, &desc[1]);
634 	CU_ASSERT(rc == -EPERM);
635 
636 	/*
637 	 * Open bdev4 read/write.  This should fail since bdev3 has been claimed
638 	 * by a vbdev module.
639 	 */
640 	rc = spdk_bdev_open_ext("bdev4", true, bdev_ut_event_cb, NULL, &desc[4]);
641 	CU_ASSERT(rc == -EPERM);
642 
643 	/* Open bdev4 read-only.  This should succeed. */
644 	rc = spdk_bdev_open_ext("bdev4", false, bdev_ut_event_cb, NULL, &desc[4]);
645 	CU_ASSERT(rc == 0);
646 	SPDK_CU_ASSERT_FATAL(desc[4] != NULL);
647 	CU_ASSERT(bdev[4] == spdk_bdev_desc_get_bdev(desc[4]));
648 	spdk_bdev_close(desc[4]);
649 
650 	/*
651 	 * Open bdev8 read/write.  This should succeed since it is a leaf
652 	 * bdev.
653 	 */
654 	rc = spdk_bdev_open_ext("bdev8", true, bdev_ut_event_cb, NULL, &desc[8]);
655 	CU_ASSERT(rc == 0);
656 	SPDK_CU_ASSERT_FATAL(desc[8] != NULL);
657 	CU_ASSERT(bdev[8] == spdk_bdev_desc_get_bdev(desc[8]));
658 	spdk_bdev_close(desc[8]);
659 
660 	/*
661 	 * Open bdev5 read/write.  This should fail since bdev4 has been claimed
662 	 * by a vbdev module.
663 	 */
664 	rc = spdk_bdev_open_ext("bdev5", true, bdev_ut_event_cb, NULL, &desc[5]);
665 	CU_ASSERT(rc == -EPERM);
666 
667 	/* Open bdev4 read-only.  This should succeed. */
668 	rc = spdk_bdev_open_ext("bdev5", false, bdev_ut_event_cb, NULL, &desc[5]);
669 	CU_ASSERT(rc == 0);
670 	SPDK_CU_ASSERT_FATAL(desc[5] != NULL);
671 	CU_ASSERT(bdev[5] == spdk_bdev_desc_get_bdev(desc[5]));
672 	spdk_bdev_close(desc[5]);
673 
674 	free_vbdev(bdev[8]);
675 
676 	free_vbdev(bdev[5]);
677 	free_vbdev(bdev[6]);
678 	free_vbdev(bdev[7]);
679 
680 	free_vbdev(bdev[4]);
681 
682 	free_bdev(bdev[0]);
683 	free_bdev(bdev[1]);
684 	free_bdev(bdev[2]);
685 	free_bdev(bdev[3]);
686 }
687 
688 static void
689 bytes_to_blocks_test(void)
690 {
691 	struct spdk_bdev bdev;
692 	uint64_t offset_blocks, num_blocks;
693 
694 	memset(&bdev, 0, sizeof(bdev));
695 
696 	bdev.blocklen = 512;
697 
698 	/* All parameters valid */
699 	offset_blocks = 0;
700 	num_blocks = 0;
701 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 1024, &num_blocks) == 0);
702 	CU_ASSERT(offset_blocks == 1);
703 	CU_ASSERT(num_blocks == 2);
704 
705 	/* Offset not a block multiple */
706 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 512, &num_blocks) != 0);
707 
708 	/* Length not a block multiple */
709 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 3, &num_blocks) != 0);
710 
711 	/* In case blocklen not the power of two */
712 	bdev.blocklen = 100;
713 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 200, &num_blocks) == 0);
714 	CU_ASSERT(offset_blocks == 1);
715 	CU_ASSERT(num_blocks == 2);
716 
717 	/* Offset not a block multiple */
718 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 100, &num_blocks) != 0);
719 
720 	/* Length not a block multiple */
721 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 3, &num_blocks) != 0);
722 }
723 
724 static void
725 num_blocks_test(void)
726 {
727 	struct spdk_bdev bdev;
728 	struct spdk_bdev_desc *desc = NULL;
729 	int rc;
730 
731 	memset(&bdev, 0, sizeof(bdev));
732 	bdev.name = "num_blocks";
733 	bdev.fn_table = &fn_table;
734 	bdev.module = &bdev_ut_if;
735 	spdk_bdev_register(&bdev);
736 	spdk_bdev_notify_blockcnt_change(&bdev, 50);
737 
738 	/* Growing block number */
739 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 70) == 0);
740 	/* Shrinking block number */
741 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 30) == 0);
742 
743 	rc = spdk_bdev_open_ext("num_blocks", false, bdev_open_cb1, &desc, &desc);
744 	CU_ASSERT(rc == 0);
745 	SPDK_CU_ASSERT_FATAL(desc != NULL);
746 	CU_ASSERT(&bdev == spdk_bdev_desc_get_bdev(desc));
747 
748 	/* Growing block number */
749 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 80) == 0);
750 	/* Shrinking block number */
751 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 20) != 0);
752 
753 	g_event_type1 = 0xFF;
754 	/* Growing block number */
755 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 90) == 0);
756 
757 	poll_threads();
758 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_RESIZE);
759 
760 	g_event_type1 = 0xFF;
761 	/* Growing block number and closing */
762 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 100) == 0);
763 
764 	spdk_bdev_close(desc);
765 	spdk_bdev_unregister(&bdev, NULL, NULL);
766 
767 	poll_threads();
768 
769 	/* Callback is not called for closed device */
770 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
771 }
772 
773 static void
774 io_valid_test(void)
775 {
776 	struct spdk_bdev bdev;
777 
778 	memset(&bdev, 0, sizeof(bdev));
779 
780 	bdev.blocklen = 512;
781 	CU_ASSERT(pthread_mutex_init(&bdev.internal.mutex, NULL) == 0);
782 
783 	spdk_bdev_notify_blockcnt_change(&bdev, 100);
784 
785 	/* All parameters valid */
786 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 1, 2) == true);
787 
788 	/* Last valid block */
789 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 1) == true);
790 
791 	/* Offset past end of bdev */
792 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 100, 1) == false);
793 
794 	/* Offset + length past end of bdev */
795 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 2) == false);
796 
797 	/* Offset near end of uint64_t range (2^64 - 1) */
798 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 18446744073709551615ULL, 1) == false);
799 
800 	CU_ASSERT(pthread_mutex_destroy(&bdev.internal.mutex) == 0);
801 }
802 
803 static void
804 alias_add_del_test(void)
805 {
806 	struct spdk_bdev *bdev[3];
807 	int rc;
808 
809 	/* Creating and registering bdevs */
810 	bdev[0] = allocate_bdev("bdev0");
811 	SPDK_CU_ASSERT_FATAL(bdev[0] != 0);
812 
813 	bdev[1] = allocate_bdev("bdev1");
814 	SPDK_CU_ASSERT_FATAL(bdev[1] != 0);
815 
816 	bdev[2] = allocate_bdev("bdev2");
817 	SPDK_CU_ASSERT_FATAL(bdev[2] != 0);
818 
819 	poll_threads();
820 
821 	/*
822 	 * Trying adding an alias identical to name.
823 	 * Alias is identical to name, so it can not be added to aliases list
824 	 */
825 	rc = spdk_bdev_alias_add(bdev[0], bdev[0]->name);
826 	CU_ASSERT(rc == -EEXIST);
827 
828 	/*
829 	 * Trying to add empty alias,
830 	 * this one should fail
831 	 */
832 	rc = spdk_bdev_alias_add(bdev[0], NULL);
833 	CU_ASSERT(rc == -EINVAL);
834 
835 	/* Trying adding same alias to two different registered bdevs */
836 
837 	/* Alias is used first time, so this one should pass */
838 	rc = spdk_bdev_alias_add(bdev[0], "proper alias 0");
839 	CU_ASSERT(rc == 0);
840 
841 	/* Alias was added to another bdev, so this one should fail */
842 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 0");
843 	CU_ASSERT(rc == -EEXIST);
844 
845 	/* Alias is used first time, so this one should pass */
846 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 1");
847 	CU_ASSERT(rc == 0);
848 
849 	/* Trying removing an alias from registered bdevs */
850 
851 	/* Alias is not on a bdev aliases list, so this one should fail */
852 	rc = spdk_bdev_alias_del(bdev[0], "not existing");
853 	CU_ASSERT(rc == -ENOENT);
854 
855 	/* Alias is present on a bdev aliases list, so this one should pass */
856 	rc = spdk_bdev_alias_del(bdev[0], "proper alias 0");
857 	CU_ASSERT(rc == 0);
858 
859 	/* Alias is present on a bdev aliases list, so this one should pass */
860 	rc = spdk_bdev_alias_del(bdev[1], "proper alias 1");
861 	CU_ASSERT(rc == 0);
862 
863 	/* Trying to remove name instead of alias, so this one should fail, name cannot be changed or removed */
864 	rc = spdk_bdev_alias_del(bdev[0], bdev[0]->name);
865 	CU_ASSERT(rc != 0);
866 
867 	/* Trying to del all alias from empty alias list */
868 	spdk_bdev_alias_del_all(bdev[2]);
869 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev[2]->aliases));
870 
871 	/* Trying to del all alias from non-empty alias list */
872 	rc = spdk_bdev_alias_add(bdev[2], "alias0");
873 	CU_ASSERT(rc == 0);
874 	rc = spdk_bdev_alias_add(bdev[2], "alias1");
875 	CU_ASSERT(rc == 0);
876 	spdk_bdev_alias_del_all(bdev[2]);
877 	CU_ASSERT(TAILQ_EMPTY(&bdev[2]->aliases));
878 
879 	/* Unregister and free bdevs */
880 	spdk_bdev_unregister(bdev[0], NULL, NULL);
881 	spdk_bdev_unregister(bdev[1], NULL, NULL);
882 	spdk_bdev_unregister(bdev[2], NULL, NULL);
883 
884 	poll_threads();
885 
886 	free(bdev[0]);
887 	free(bdev[1]);
888 	free(bdev[2]);
889 }
890 
891 static void
892 io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
893 {
894 	g_io_done = true;
895 	g_io_status = bdev_io->internal.status;
896 	if ((bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) &&
897 	    (bdev_io->u.bdev.zcopy.start)) {
898 		g_zcopy_bdev_io = bdev_io;
899 	} else {
900 		spdk_bdev_free_io(bdev_io);
901 		g_zcopy_bdev_io = NULL;
902 	}
903 }
904 
905 static void
906 bdev_init_cb(void *arg, int rc)
907 {
908 	CU_ASSERT(rc == 0);
909 }
910 
911 static void
912 bdev_fini_cb(void *arg)
913 {
914 }
915 
916 struct bdev_ut_io_wait_entry {
917 	struct spdk_bdev_io_wait_entry	entry;
918 	struct spdk_io_channel		*io_ch;
919 	struct spdk_bdev_desc		*desc;
920 	bool				submitted;
921 };
922 
923 static void
924 io_wait_cb(void *arg)
925 {
926 	struct bdev_ut_io_wait_entry *entry = arg;
927 	int rc;
928 
929 	rc = spdk_bdev_read_blocks(entry->desc, entry->io_ch, NULL, 0, 1, io_done, NULL);
930 	CU_ASSERT(rc == 0);
931 	entry->submitted = true;
932 }
933 
934 static void
935 bdev_io_types_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 	int rc;
942 
943 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
944 	bdev_opts.bdev_io_pool_size = 4;
945 	bdev_opts.bdev_io_cache_size = 2;
946 
947 	rc = spdk_bdev_set_opts(&bdev_opts);
948 	CU_ASSERT(rc == 0);
949 	spdk_bdev_initialize(bdev_init_cb, NULL);
950 	poll_threads();
951 
952 	bdev = allocate_bdev("bdev0");
953 
954 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
955 	CU_ASSERT(rc == 0);
956 	poll_threads();
957 	SPDK_CU_ASSERT_FATAL(desc != NULL);
958 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
959 	io_ch = spdk_bdev_get_io_channel(desc);
960 	CU_ASSERT(io_ch != NULL);
961 
962 	/* WRITE and WRITE ZEROES are not supported */
963 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
964 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, false);
965 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 0, 128, io_done, NULL);
966 	CU_ASSERT(rc == -ENOTSUP);
967 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
968 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, true);
969 
970 	spdk_put_io_channel(io_ch);
971 	spdk_bdev_close(desc);
972 	free_bdev(bdev);
973 	spdk_bdev_finish(bdev_fini_cb, NULL);
974 	poll_threads();
975 }
976 
977 static void
978 bdev_io_wait_test(void)
979 {
980 	struct spdk_bdev *bdev;
981 	struct spdk_bdev_desc *desc = NULL;
982 	struct spdk_io_channel *io_ch;
983 	struct spdk_bdev_opts bdev_opts = {};
984 	struct bdev_ut_io_wait_entry io_wait_entry;
985 	struct bdev_ut_io_wait_entry io_wait_entry2;
986 	int rc;
987 
988 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
989 	bdev_opts.bdev_io_pool_size = 4;
990 	bdev_opts.bdev_io_cache_size = 2;
991 
992 	rc = spdk_bdev_set_opts(&bdev_opts);
993 	CU_ASSERT(rc == 0);
994 	spdk_bdev_initialize(bdev_init_cb, NULL);
995 	poll_threads();
996 
997 	bdev = allocate_bdev("bdev0");
998 
999 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1000 	CU_ASSERT(rc == 0);
1001 	poll_threads();
1002 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1003 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1004 	io_ch = spdk_bdev_get_io_channel(desc);
1005 	CU_ASSERT(io_ch != NULL);
1006 
1007 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1008 	CU_ASSERT(rc == 0);
1009 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1010 	CU_ASSERT(rc == 0);
1011 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1012 	CU_ASSERT(rc == 0);
1013 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1014 	CU_ASSERT(rc == 0);
1015 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1016 
1017 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1018 	CU_ASSERT(rc == -ENOMEM);
1019 
1020 	io_wait_entry.entry.bdev = bdev;
1021 	io_wait_entry.entry.cb_fn = io_wait_cb;
1022 	io_wait_entry.entry.cb_arg = &io_wait_entry;
1023 	io_wait_entry.io_ch = io_ch;
1024 	io_wait_entry.desc = desc;
1025 	io_wait_entry.submitted = false;
1026 	/* Cannot use the same io_wait_entry for two different calls. */
1027 	memcpy(&io_wait_entry2, &io_wait_entry, sizeof(io_wait_entry));
1028 	io_wait_entry2.entry.cb_arg = &io_wait_entry2;
1029 
1030 	/* Queue two I/O waits. */
1031 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry.entry);
1032 	CU_ASSERT(rc == 0);
1033 	CU_ASSERT(io_wait_entry.submitted == false);
1034 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry2.entry);
1035 	CU_ASSERT(rc == 0);
1036 	CU_ASSERT(io_wait_entry2.submitted == false);
1037 
1038 	stub_complete_io(1);
1039 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1040 	CU_ASSERT(io_wait_entry.submitted == true);
1041 	CU_ASSERT(io_wait_entry2.submitted == false);
1042 
1043 	stub_complete_io(1);
1044 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1045 	CU_ASSERT(io_wait_entry2.submitted == true);
1046 
1047 	stub_complete_io(4);
1048 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1049 
1050 	spdk_put_io_channel(io_ch);
1051 	spdk_bdev_close(desc);
1052 	free_bdev(bdev);
1053 	spdk_bdev_finish(bdev_fini_cb, NULL);
1054 	poll_threads();
1055 }
1056 
1057 static void
1058 bdev_io_spans_split_test(void)
1059 {
1060 	struct spdk_bdev bdev;
1061 	struct spdk_bdev_io bdev_io;
1062 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV];
1063 
1064 	memset(&bdev, 0, sizeof(bdev));
1065 	bdev_io.u.bdev.iovs = iov;
1066 
1067 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1068 	bdev.optimal_io_boundary = 0;
1069 	bdev.max_segment_size = 0;
1070 	bdev.max_num_segments = 0;
1071 	bdev_io.bdev = &bdev;
1072 
1073 	/* bdev has no optimal_io_boundary and max_size set - so this should return false. */
1074 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1075 
1076 	bdev.split_on_optimal_io_boundary = true;
1077 	bdev.optimal_io_boundary = 32;
1078 	bdev_io.type = SPDK_BDEV_IO_TYPE_RESET;
1079 
1080 	/* RESETs are not based on LBAs - so this should return false. */
1081 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1082 
1083 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1084 	bdev_io.u.bdev.offset_blocks = 0;
1085 	bdev_io.u.bdev.num_blocks = 32;
1086 
1087 	/* This I/O run right up to, but does not cross, the boundary - so this should return false. */
1088 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1089 
1090 	bdev_io.u.bdev.num_blocks = 33;
1091 
1092 	/* This I/O spans a boundary. */
1093 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1094 
1095 	bdev_io.u.bdev.num_blocks = 32;
1096 	bdev.max_segment_size = 512 * 32;
1097 	bdev.max_num_segments = 1;
1098 	bdev_io.u.bdev.iovcnt = 1;
1099 	iov[0].iov_len = 512;
1100 
1101 	/* Does not cross and exceed max_size or max_segs */
1102 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1103 
1104 	bdev.split_on_optimal_io_boundary = false;
1105 	bdev.max_segment_size = 512;
1106 	bdev.max_num_segments = 1;
1107 	bdev_io.u.bdev.iovcnt = 2;
1108 
1109 	/* Exceed max_segs */
1110 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1111 
1112 	bdev.max_num_segments = 2;
1113 	iov[0].iov_len = 513;
1114 	iov[1].iov_len = 512;
1115 
1116 	/* Exceed max_sizes */
1117 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1118 }
1119 
1120 static void
1121 bdev_io_boundary_split_test(void)
1122 {
1123 	struct spdk_bdev *bdev;
1124 	struct spdk_bdev_desc *desc = NULL;
1125 	struct spdk_io_channel *io_ch;
1126 	struct spdk_bdev_opts bdev_opts = {};
1127 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1128 	struct ut_expected_io *expected_io;
1129 	void *md_buf = (void *)0xFF000000;
1130 	uint64_t i;
1131 	int rc;
1132 
1133 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1134 	bdev_opts.bdev_io_pool_size = 512;
1135 	bdev_opts.bdev_io_cache_size = 64;
1136 
1137 	rc = spdk_bdev_set_opts(&bdev_opts);
1138 	CU_ASSERT(rc == 0);
1139 	spdk_bdev_initialize(bdev_init_cb, NULL);
1140 
1141 	bdev = allocate_bdev("bdev0");
1142 
1143 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1144 	CU_ASSERT(rc == 0);
1145 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1146 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1147 	io_ch = spdk_bdev_get_io_channel(desc);
1148 	CU_ASSERT(io_ch != NULL);
1149 
1150 	bdev->optimal_io_boundary = 16;
1151 	bdev->split_on_optimal_io_boundary = false;
1152 
1153 	g_io_done = false;
1154 
1155 	/* First test that the I/O does not get split if split_on_optimal_io_boundary == false. */
1156 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 8, 1);
1157 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 8 * 512);
1158 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1159 
1160 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
1161 	CU_ASSERT(rc == 0);
1162 	CU_ASSERT(g_io_done == false);
1163 
1164 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1165 	stub_complete_io(1);
1166 	CU_ASSERT(g_io_done == true);
1167 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1168 
1169 	bdev->split_on_optimal_io_boundary = true;
1170 	bdev->md_interleave = false;
1171 	bdev->md_len = 8;
1172 
1173 	/* Now test that a single-vector command is split correctly.
1174 	 * Offset 14, length 8, payload 0xF000
1175 	 *  Child - Offset 14, length 2, payload 0xF000
1176 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
1177 	 *
1178 	 * Set up the expected values before calling spdk_bdev_read_blocks
1179 	 */
1180 	g_io_done = false;
1181 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
1182 	expected_io->md_buf = md_buf;
1183 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
1184 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1185 
1186 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
1187 	expected_io->md_buf = md_buf + 2 * 8;
1188 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
1189 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1190 
1191 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1192 	rc = spdk_bdev_read_blocks_with_md(desc, io_ch, (void *)0xF000, md_buf,
1193 					   14, 8, io_done, NULL);
1194 	CU_ASSERT(rc == 0);
1195 	CU_ASSERT(g_io_done == false);
1196 
1197 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1198 	stub_complete_io(2);
1199 	CU_ASSERT(g_io_done == true);
1200 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1201 
1202 	/* Now set up a more complex, multi-vector command that needs to be split,
1203 	 *  including splitting iovecs.
1204 	 */
1205 	iov[0].iov_base = (void *)0x10000;
1206 	iov[0].iov_len = 512;
1207 	iov[1].iov_base = (void *)0x20000;
1208 	iov[1].iov_len = 20 * 512;
1209 	iov[2].iov_base = (void *)0x30000;
1210 	iov[2].iov_len = 11 * 512;
1211 
1212 	g_io_done = false;
1213 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
1214 	expected_io->md_buf = md_buf;
1215 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
1216 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
1217 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1218 
1219 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
1220 	expected_io->md_buf = md_buf + 2 * 8;
1221 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
1222 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1223 
1224 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
1225 	expected_io->md_buf = md_buf + 18 * 8;
1226 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
1227 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
1228 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1229 
1230 	rc = spdk_bdev_writev_blocks_with_md(desc, io_ch, iov, 3, md_buf,
1231 					     14, 32, io_done, NULL);
1232 	CU_ASSERT(rc == 0);
1233 	CU_ASSERT(g_io_done == false);
1234 
1235 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
1236 	stub_complete_io(3);
1237 	CU_ASSERT(g_io_done == true);
1238 
1239 	/* Test multi vector command that needs to be split by strip and then needs to be
1240 	 * split further due to the capacity of child iovs.
1241 	 */
1242 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
1243 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1244 		iov[i].iov_len = 512;
1245 	}
1246 
1247 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1248 	g_io_done = false;
1249 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1250 					   BDEV_IO_NUM_CHILD_IOV);
1251 	expected_io->md_buf = md_buf;
1252 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1253 		ut_expected_io_set_iov(expected_io, i, (void *)((i + 1) * 0x10000), 512);
1254 	}
1255 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1256 
1257 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1258 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
1259 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1260 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1261 		ut_expected_io_set_iov(expected_io, i,
1262 				       (void *)((i + 1 + BDEV_IO_NUM_CHILD_IOV) * 0x10000), 512);
1263 	}
1264 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1265 
1266 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1267 					    0, BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1268 	CU_ASSERT(rc == 0);
1269 	CU_ASSERT(g_io_done == false);
1270 
1271 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1272 	stub_complete_io(1);
1273 	CU_ASSERT(g_io_done == false);
1274 
1275 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1276 	stub_complete_io(1);
1277 	CU_ASSERT(g_io_done == true);
1278 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1279 
1280 	/* Test multi vector command that needs to be split by strip and then needs to be
1281 	 * split further due to the capacity of child iovs. In this case, the length of
1282 	 * the rest of iovec array with an I/O boundary is the multiple of block size.
1283 	 */
1284 
1285 	/* Fill iovec array for exactly one boundary. The iovec cnt for this boundary
1286 	 * is BDEV_IO_NUM_CHILD_IOV + 1, which exceeds the capacity of child iovs.
1287 	 */
1288 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1289 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1290 		iov[i].iov_len = 512;
1291 	}
1292 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1293 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1294 		iov[i].iov_len = 256;
1295 	}
1296 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1297 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 512;
1298 
1299 	/* Add an extra iovec to trigger split */
1300 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1301 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1302 
1303 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1304 	g_io_done = false;
1305 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1306 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV);
1307 	expected_io->md_buf = md_buf;
1308 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1309 		ut_expected_io_set_iov(expected_io, i,
1310 				       (void *)((i + 1) * 0x10000), 512);
1311 	}
1312 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1313 		ut_expected_io_set_iov(expected_io, i,
1314 				       (void *)((i + 1) * 0x10000), 256);
1315 	}
1316 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1317 
1318 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1319 					   1, 1);
1320 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1321 	ut_expected_io_set_iov(expected_io, 0,
1322 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 512);
1323 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1324 
1325 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1326 					   1, 1);
1327 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1328 	ut_expected_io_set_iov(expected_io, 0,
1329 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1330 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1331 
1332 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, md_buf,
1333 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1334 	CU_ASSERT(rc == 0);
1335 	CU_ASSERT(g_io_done == false);
1336 
1337 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1338 	stub_complete_io(1);
1339 	CU_ASSERT(g_io_done == false);
1340 
1341 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1342 	stub_complete_io(2);
1343 	CU_ASSERT(g_io_done == true);
1344 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1345 
1346 	/* Test multi vector command that needs to be split by strip and then needs to be
1347 	 * split further due to the capacity of child iovs, the child request offset should
1348 	 * be rewind to last aligned offset and go success without error.
1349 	 */
1350 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1351 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1352 		iov[i].iov_len = 512;
1353 	}
1354 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1355 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1356 
1357 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1358 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1359 
1360 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1361 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1362 
1363 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1364 	g_io_done = false;
1365 	g_io_status = 0;
1366 	/* The first expected io should be start from offset 0 to BDEV_IO_NUM_CHILD_IOV - 1 */
1367 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1368 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV - 1);
1369 	expected_io->md_buf = md_buf;
1370 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1371 		ut_expected_io_set_iov(expected_io, i,
1372 				       (void *)((i + 1) * 0x10000), 512);
1373 	}
1374 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1375 	/* The second expected io should be start from offset BDEV_IO_NUM_CHILD_IOV - 1 to BDEV_IO_NUM_CHILD_IOV */
1376 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1377 					   1, 2);
1378 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1379 	ut_expected_io_set_iov(expected_io, 0,
1380 			       (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000), 256);
1381 	ut_expected_io_set_iov(expected_io, 1,
1382 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 256);
1383 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1384 	/* The third expected io should be start from offset BDEV_IO_NUM_CHILD_IOV to BDEV_IO_NUM_CHILD_IOV + 1 */
1385 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1386 					   1, 1);
1387 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1388 	ut_expected_io_set_iov(expected_io, 0,
1389 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1390 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1391 
1392 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1393 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1394 	CU_ASSERT(rc == 0);
1395 	CU_ASSERT(g_io_done == false);
1396 
1397 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1398 	stub_complete_io(1);
1399 	CU_ASSERT(g_io_done == false);
1400 
1401 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1402 	stub_complete_io(2);
1403 	CU_ASSERT(g_io_done == true);
1404 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1405 
1406 	/* Test multi vector command that needs to be split due to the IO boundary and
1407 	 * the capacity of child iovs. Especially test the case when the command is
1408 	 * split due to the capacity of child iovs, the tail address is not aligned with
1409 	 * block size and is rewinded to the aligned address.
1410 	 *
1411 	 * The iovecs used in read request is complex but is based on the data
1412 	 * collected in the real issue. We change the base addresses but keep the lengths
1413 	 * not to loose the credibility of the test.
1414 	 */
1415 	bdev->optimal_io_boundary = 128;
1416 	g_io_done = false;
1417 	g_io_status = 0;
1418 
1419 	for (i = 0; i < 31; i++) {
1420 		iov[i].iov_base = (void *)(0xFEED0000000 + (i << 20));
1421 		iov[i].iov_len = 1024;
1422 	}
1423 	iov[31].iov_base = (void *)0xFEED1F00000;
1424 	iov[31].iov_len = 32768;
1425 	iov[32].iov_base = (void *)0xFEED2000000;
1426 	iov[32].iov_len = 160;
1427 	iov[33].iov_base = (void *)0xFEED2100000;
1428 	iov[33].iov_len = 4096;
1429 	iov[34].iov_base = (void *)0xFEED2200000;
1430 	iov[34].iov_len = 4096;
1431 	iov[35].iov_base = (void *)0xFEED2300000;
1432 	iov[35].iov_len = 4096;
1433 	iov[36].iov_base = (void *)0xFEED2400000;
1434 	iov[36].iov_len = 4096;
1435 	iov[37].iov_base = (void *)0xFEED2500000;
1436 	iov[37].iov_len = 4096;
1437 	iov[38].iov_base = (void *)0xFEED2600000;
1438 	iov[38].iov_len = 4096;
1439 	iov[39].iov_base = (void *)0xFEED2700000;
1440 	iov[39].iov_len = 4096;
1441 	iov[40].iov_base = (void *)0xFEED2800000;
1442 	iov[40].iov_len = 4096;
1443 	iov[41].iov_base = (void *)0xFEED2900000;
1444 	iov[41].iov_len = 4096;
1445 	iov[42].iov_base = (void *)0xFEED2A00000;
1446 	iov[42].iov_len = 4096;
1447 	iov[43].iov_base = (void *)0xFEED2B00000;
1448 	iov[43].iov_len = 12288;
1449 	iov[44].iov_base = (void *)0xFEED2C00000;
1450 	iov[44].iov_len = 8192;
1451 	iov[45].iov_base = (void *)0xFEED2F00000;
1452 	iov[45].iov_len = 4096;
1453 	iov[46].iov_base = (void *)0xFEED3000000;
1454 	iov[46].iov_len = 4096;
1455 	iov[47].iov_base = (void *)0xFEED3100000;
1456 	iov[47].iov_len = 4096;
1457 	iov[48].iov_base = (void *)0xFEED3200000;
1458 	iov[48].iov_len = 24576;
1459 	iov[49].iov_base = (void *)0xFEED3300000;
1460 	iov[49].iov_len = 16384;
1461 	iov[50].iov_base = (void *)0xFEED3400000;
1462 	iov[50].iov_len = 12288;
1463 	iov[51].iov_base = (void *)0xFEED3500000;
1464 	iov[51].iov_len = 4096;
1465 	iov[52].iov_base = (void *)0xFEED3600000;
1466 	iov[52].iov_len = 4096;
1467 	iov[53].iov_base = (void *)0xFEED3700000;
1468 	iov[53].iov_len = 4096;
1469 	iov[54].iov_base = (void *)0xFEED3800000;
1470 	iov[54].iov_len = 28672;
1471 	iov[55].iov_base = (void *)0xFEED3900000;
1472 	iov[55].iov_len = 20480;
1473 	iov[56].iov_base = (void *)0xFEED3A00000;
1474 	iov[56].iov_len = 4096;
1475 	iov[57].iov_base = (void *)0xFEED3B00000;
1476 	iov[57].iov_len = 12288;
1477 	iov[58].iov_base = (void *)0xFEED3C00000;
1478 	iov[58].iov_len = 4096;
1479 	iov[59].iov_base = (void *)0xFEED3D00000;
1480 	iov[59].iov_len = 4096;
1481 	iov[60].iov_base = (void *)0xFEED3E00000;
1482 	iov[60].iov_len = 352;
1483 
1484 	/* The 1st child IO must be from iov[0] to iov[31] split by the capacity
1485 	 * of child iovs,
1486 	 */
1487 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 126, 32);
1488 	expected_io->md_buf = md_buf;
1489 	for (i = 0; i < 32; i++) {
1490 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1491 	}
1492 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1493 
1494 	/* The 2nd child IO must be from iov[32] to the first 864 bytes of iov[33]
1495 	 * split by the IO boundary requirement.
1496 	 */
1497 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 126, 2, 2);
1498 	expected_io->md_buf = md_buf + 126 * 8;
1499 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base, iov[32].iov_len);
1500 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 864);
1501 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1502 
1503 	/* The 3rd child IO must be from the remaining 3232 bytes of iov[33] to
1504 	 * the first 864 bytes of iov[46] split by the IO boundary requirement.
1505 	 */
1506 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 128, 128, 14);
1507 	expected_io->md_buf = md_buf + 128 * 8;
1508 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[33].iov_base + 864),
1509 			       iov[33].iov_len - 864);
1510 	ut_expected_io_set_iov(expected_io, 1, iov[34].iov_base, iov[34].iov_len);
1511 	ut_expected_io_set_iov(expected_io, 2, iov[35].iov_base, iov[35].iov_len);
1512 	ut_expected_io_set_iov(expected_io, 3, iov[36].iov_base, iov[36].iov_len);
1513 	ut_expected_io_set_iov(expected_io, 4, iov[37].iov_base, iov[37].iov_len);
1514 	ut_expected_io_set_iov(expected_io, 5, iov[38].iov_base, iov[38].iov_len);
1515 	ut_expected_io_set_iov(expected_io, 6, iov[39].iov_base, iov[39].iov_len);
1516 	ut_expected_io_set_iov(expected_io, 7, iov[40].iov_base, iov[40].iov_len);
1517 	ut_expected_io_set_iov(expected_io, 8, iov[41].iov_base, iov[41].iov_len);
1518 	ut_expected_io_set_iov(expected_io, 9, iov[42].iov_base, iov[42].iov_len);
1519 	ut_expected_io_set_iov(expected_io, 10, iov[43].iov_base, iov[43].iov_len);
1520 	ut_expected_io_set_iov(expected_io, 11, iov[44].iov_base, iov[44].iov_len);
1521 	ut_expected_io_set_iov(expected_io, 12, iov[45].iov_base, iov[45].iov_len);
1522 	ut_expected_io_set_iov(expected_io, 13, iov[46].iov_base, 864);
1523 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1524 
1525 	/* The 4th child IO must be from the remaining 3232 bytes of iov[46] to the
1526 	 * first 864 bytes of iov[52] split by the IO boundary requirement.
1527 	 */
1528 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 256, 128, 7);
1529 	expected_io->md_buf = md_buf + 256 * 8;
1530 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[46].iov_base + 864),
1531 			       iov[46].iov_len - 864);
1532 	ut_expected_io_set_iov(expected_io, 1, iov[47].iov_base, iov[47].iov_len);
1533 	ut_expected_io_set_iov(expected_io, 2, iov[48].iov_base, iov[48].iov_len);
1534 	ut_expected_io_set_iov(expected_io, 3, iov[49].iov_base, iov[49].iov_len);
1535 	ut_expected_io_set_iov(expected_io, 4, iov[50].iov_base, iov[50].iov_len);
1536 	ut_expected_io_set_iov(expected_io, 5, iov[51].iov_base, iov[51].iov_len);
1537 	ut_expected_io_set_iov(expected_io, 6, iov[52].iov_base, 864);
1538 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1539 
1540 	/* The 5th child IO must be from the remaining 3232 bytes of iov[52] to
1541 	 * the first 4096 bytes of iov[57] split by the IO boundary requirement.
1542 	 */
1543 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 384, 128, 6);
1544 	expected_io->md_buf = md_buf + 384 * 8;
1545 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[52].iov_base + 864),
1546 			       iov[52].iov_len - 864);
1547 	ut_expected_io_set_iov(expected_io, 1, iov[53].iov_base, iov[53].iov_len);
1548 	ut_expected_io_set_iov(expected_io, 2, iov[54].iov_base, iov[54].iov_len);
1549 	ut_expected_io_set_iov(expected_io, 3, iov[55].iov_base, iov[55].iov_len);
1550 	ut_expected_io_set_iov(expected_io, 4, iov[56].iov_base, iov[56].iov_len);
1551 	ut_expected_io_set_iov(expected_io, 5, iov[57].iov_base, 4960);
1552 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1553 
1554 	/* The 6th child IO must be from the remaining 7328 bytes of iov[57]
1555 	 * to the first 3936 bytes of iov[58] split by the capacity of child iovs.
1556 	 */
1557 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 512, 30, 3);
1558 	expected_io->md_buf = md_buf + 512 * 8;
1559 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[57].iov_base + 4960),
1560 			       iov[57].iov_len - 4960);
1561 	ut_expected_io_set_iov(expected_io, 1, iov[58].iov_base, iov[58].iov_len);
1562 	ut_expected_io_set_iov(expected_io, 2, iov[59].iov_base, 3936);
1563 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1564 
1565 	/* The 7th child IO is from the remaining 160 bytes of iov[59] and iov[60]. */
1566 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 542, 1, 2);
1567 	expected_io->md_buf = md_buf + 542 * 8;
1568 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[59].iov_base + 3936),
1569 			       iov[59].iov_len - 3936);
1570 	ut_expected_io_set_iov(expected_io, 1, iov[60].iov_base, iov[60].iov_len);
1571 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1572 
1573 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, 61, md_buf,
1574 					    0, 543, io_done, NULL);
1575 	CU_ASSERT(rc == 0);
1576 	CU_ASSERT(g_io_done == false);
1577 
1578 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1579 	stub_complete_io(1);
1580 	CU_ASSERT(g_io_done == false);
1581 
1582 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1583 	stub_complete_io(5);
1584 	CU_ASSERT(g_io_done == false);
1585 
1586 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1587 	stub_complete_io(1);
1588 	CU_ASSERT(g_io_done == true);
1589 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1590 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1591 
1592 	/* Test a WRITE_ZEROES that would span an I/O boundary.  WRITE_ZEROES should not be
1593 	 * split, so test that.
1594 	 */
1595 	bdev->optimal_io_boundary = 15;
1596 	g_io_done = false;
1597 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
1598 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1599 
1600 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
1601 	CU_ASSERT(rc == 0);
1602 	CU_ASSERT(g_io_done == false);
1603 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1604 	stub_complete_io(1);
1605 	CU_ASSERT(g_io_done == true);
1606 
1607 	/* Test an UNMAP.  This should also not be split. */
1608 	bdev->optimal_io_boundary = 16;
1609 	g_io_done = false;
1610 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 2, 0);
1611 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1612 
1613 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 2, io_done, NULL);
1614 	CU_ASSERT(rc == 0);
1615 	CU_ASSERT(g_io_done == false);
1616 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1617 	stub_complete_io(1);
1618 	CU_ASSERT(g_io_done == true);
1619 
1620 	/* Test a FLUSH.  This should also not be split. */
1621 	bdev->optimal_io_boundary = 16;
1622 	g_io_done = false;
1623 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 2, 0);
1624 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1625 
1626 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
1627 	CU_ASSERT(rc == 0);
1628 	CU_ASSERT(g_io_done == false);
1629 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1630 	stub_complete_io(1);
1631 	CU_ASSERT(g_io_done == true);
1632 
1633 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
1634 
1635 	/* Children requests return an error status */
1636 	bdev->optimal_io_boundary = 16;
1637 	iov[0].iov_base = (void *)0x10000;
1638 	iov[0].iov_len = 512 * 64;
1639 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1640 	g_io_done = false;
1641 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1642 
1643 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 1, 1, 64, io_done, NULL);
1644 	CU_ASSERT(rc == 0);
1645 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1646 	stub_complete_io(4);
1647 	CU_ASSERT(g_io_done == false);
1648 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1649 	stub_complete_io(1);
1650 	CU_ASSERT(g_io_done == true);
1651 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1652 
1653 	/* Test if a multi vector command terminated with failure before continueing
1654 	 * splitting process when one of child I/O failed.
1655 	 * The multi vector command is as same as the above that needs to be split by strip
1656 	 * and then needs to be split further due to the capacity of child iovs.
1657 	 */
1658 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1659 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1660 		iov[i].iov_len = 512;
1661 	}
1662 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1663 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1664 
1665 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1666 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1667 
1668 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1669 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1670 
1671 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1672 
1673 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1674 	g_io_done = false;
1675 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1676 
1677 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
1678 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1679 	CU_ASSERT(rc == 0);
1680 	CU_ASSERT(g_io_done == false);
1681 
1682 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1683 	stub_complete_io(1);
1684 	CU_ASSERT(g_io_done == true);
1685 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1686 
1687 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1688 
1689 	/* for this test we will create the following conditions to hit the code path where
1690 	 * we are trying to send and IO following a split that has no iovs because we had to
1691 	 * trim them for alignment reasons.
1692 	 *
1693 	 * - 16K boundary, our IO will start at offset 0 with a length of 0x4200
1694 	 * - Our IOVs are 0x212 in size so that we run into the 16K boundary at child IOV
1695 	 *   position 30 and overshoot by 0x2e.
1696 	 * - That means we'll send the IO and loop back to pick up the remaining bytes at
1697 	 *   child IOV index 31. When we do, we find that we have to shorten index 31 by 0x2e
1698 	 *   which eliniates that vector so we just send the first split IO with 30 vectors
1699 	 *   and let the completion pick up the last 2 vectors.
1700 	 */
1701 	bdev->optimal_io_boundary = 32;
1702 	bdev->split_on_optimal_io_boundary = true;
1703 	g_io_done = false;
1704 
1705 	/* Init all parent IOVs to 0x212 */
1706 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
1707 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1708 		iov[i].iov_len = 0x212;
1709 	}
1710 
1711 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1712 					   BDEV_IO_NUM_CHILD_IOV - 1);
1713 	/* expect 0-29 to be 1:1 with the parent iov */
1714 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1715 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1716 	}
1717 
1718 	/* expect index 30 to be shortened to 0x1e4 (0x212 - 0x1e) because of the alignment
1719 	 * where 0x1e is the amount we overshot the 16K boundary
1720 	 */
1721 	ut_expected_io_set_iov(expected_io, BDEV_IO_NUM_CHILD_IOV - 2,
1722 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base), 0x1e4);
1723 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1724 
1725 	/* 2nd child IO will have 2 remaining vectors, one to pick up from the one that was
1726 	 * shortened that take it to the next boundary and then a final one to get us to
1727 	 * 0x4200 bytes for the IO.
1728 	 */
1729 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1730 					   BDEV_IO_NUM_CHILD_IOV, 2);
1731 	/* position 30 picked up the remaining bytes to the next boundary */
1732 	ut_expected_io_set_iov(expected_io, 0,
1733 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base + 0x1e4), 0x2e);
1734 
1735 	/* position 31 picked the the rest of the trasnfer to get us to 0x4200 */
1736 	ut_expected_io_set_iov(expected_io, 1,
1737 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base), 0x1d2);
1738 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1739 
1740 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 1, 0,
1741 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1742 	CU_ASSERT(rc == 0);
1743 	CU_ASSERT(g_io_done == false);
1744 
1745 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1746 	stub_complete_io(1);
1747 	CU_ASSERT(g_io_done == false);
1748 
1749 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1750 	stub_complete_io(1);
1751 	CU_ASSERT(g_io_done == true);
1752 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1753 
1754 	spdk_put_io_channel(io_ch);
1755 	spdk_bdev_close(desc);
1756 	free_bdev(bdev);
1757 	spdk_bdev_finish(bdev_fini_cb, NULL);
1758 	poll_threads();
1759 }
1760 
1761 static void
1762 bdev_io_max_size_and_segment_split_test(void)
1763 {
1764 	struct spdk_bdev *bdev;
1765 	struct spdk_bdev_desc *desc = NULL;
1766 	struct spdk_io_channel *io_ch;
1767 	struct spdk_bdev_opts bdev_opts = {};
1768 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1769 	struct ut_expected_io *expected_io;
1770 	uint64_t i;
1771 	int rc;
1772 
1773 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1774 	bdev_opts.bdev_io_pool_size = 512;
1775 	bdev_opts.bdev_io_cache_size = 64;
1776 
1777 	bdev_opts.opts_size = sizeof(bdev_opts);
1778 	rc = spdk_bdev_set_opts(&bdev_opts);
1779 	CU_ASSERT(rc == 0);
1780 	spdk_bdev_initialize(bdev_init_cb, NULL);
1781 
1782 	bdev = allocate_bdev("bdev0");
1783 
1784 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
1785 	CU_ASSERT(rc == 0);
1786 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1787 	io_ch = spdk_bdev_get_io_channel(desc);
1788 	CU_ASSERT(io_ch != NULL);
1789 
1790 	bdev->split_on_optimal_io_boundary = false;
1791 	bdev->optimal_io_boundary = 0;
1792 
1793 	/* Case 0 max_num_segments == 0.
1794 	 * but segment size 2 * 512 > 512
1795 	 */
1796 	bdev->max_segment_size = 512;
1797 	bdev->max_num_segments = 0;
1798 	g_io_done = false;
1799 
1800 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
1801 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1802 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
1803 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1804 
1805 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1806 	CU_ASSERT(rc == 0);
1807 	CU_ASSERT(g_io_done == false);
1808 
1809 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1810 	stub_complete_io(1);
1811 	CU_ASSERT(g_io_done == true);
1812 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1813 
1814 	/* Case 1 max_segment_size == 0
1815 	 * but iov num 2 > 1.
1816 	 */
1817 	bdev->max_segment_size = 0;
1818 	bdev->max_num_segments = 1;
1819 	g_io_done = false;
1820 
1821 	iov[0].iov_base = (void *)0x10000;
1822 	iov[0].iov_len = 512;
1823 	iov[1].iov_base = (void *)0x20000;
1824 	iov[1].iov_len = 8 * 512;
1825 
1826 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1827 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, iov[0].iov_len);
1828 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1829 
1830 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 8, 1);
1831 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, iov[1].iov_len);
1832 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1833 
1834 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 14, 9, io_done, NULL);
1835 	CU_ASSERT(rc == 0);
1836 	CU_ASSERT(g_io_done == false);
1837 
1838 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1839 	stub_complete_io(2);
1840 	CU_ASSERT(g_io_done == true);
1841 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1842 
1843 	/* Test that a non-vector command is split correctly.
1844 	 * Set up the expected values before calling spdk_bdev_read_blocks
1845 	 */
1846 	bdev->max_segment_size = 512;
1847 	bdev->max_num_segments = 1;
1848 	g_io_done = false;
1849 
1850 	/* Child IO 0 */
1851 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1852 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1853 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1854 
1855 	/* Child IO 1 */
1856 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
1857 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 1 * 512), 512);
1858 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1859 
1860 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1861 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1862 	CU_ASSERT(rc == 0);
1863 	CU_ASSERT(g_io_done == false);
1864 
1865 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1866 	stub_complete_io(2);
1867 	CU_ASSERT(g_io_done == true);
1868 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1869 
1870 	/* Now set up a more complex, multi-vector command that needs to be split,
1871 	 * including splitting iovecs.
1872 	 */
1873 	bdev->max_segment_size = 2 * 512;
1874 	bdev->max_num_segments = 1;
1875 	g_io_done = false;
1876 
1877 	iov[0].iov_base = (void *)0x10000;
1878 	iov[0].iov_len = 2 * 512;
1879 	iov[1].iov_base = (void *)0x20000;
1880 	iov[1].iov_len = 4 * 512;
1881 	iov[2].iov_base = (void *)0x30000;
1882 	iov[2].iov_len = 6 * 512;
1883 
1884 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
1885 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 2);
1886 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1887 
1888 	/* Split iov[1].size to 2 iov entries then split the segments */
1889 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
1890 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, 512 * 2);
1891 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1892 
1893 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 18, 2, 1);
1894 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base + 512 * 2, 512 * 2);
1895 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1896 
1897 	/* Split iov[2].size to 3 iov entries then split the segments */
1898 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 20, 2, 1);
1899 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base, 512 * 2);
1900 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1901 
1902 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 22, 2, 1);
1903 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 2, 512 * 2);
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_WRITE, 24, 2, 1);
1907 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 4, 512 * 2);
1908 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1909 
1910 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 12, io_done, NULL);
1911 	CU_ASSERT(rc == 0);
1912 	CU_ASSERT(g_io_done == false);
1913 
1914 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
1915 	stub_complete_io(6);
1916 	CU_ASSERT(g_io_done == true);
1917 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1918 
1919 	/* Test multi vector command that needs to be split by strip and then needs to be
1920 	 * split further due to the capacity of parent IO child iovs.
1921 	 */
1922 	bdev->max_segment_size = 512;
1923 	bdev->max_num_segments = 1;
1924 	g_io_done = false;
1925 
1926 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1927 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1928 		iov[i].iov_len = 512 * 2;
1929 	}
1930 
1931 	/* Each input iov.size is split into 2 iovs,
1932 	 * half of the input iov can fill all child iov entries of a single IO.
1933 	 */
1934 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV / 2; i++) {
1935 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i, 1, 1);
1936 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1937 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1938 
1939 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i + 1, 1, 1);
1940 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1941 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1942 	}
1943 
1944 	/* The remaining iov is split in the second round */
1945 	for (i = BDEV_IO_NUM_CHILD_IOV / 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1946 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2, 1, 1);
1947 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1948 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1949 
1950 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2 + 1, 1, 1);
1951 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1952 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1953 	}
1954 
1955 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
1956 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1957 	CU_ASSERT(rc == 0);
1958 	CU_ASSERT(g_io_done == false);
1959 
1960 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1961 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1962 	CU_ASSERT(g_io_done == false);
1963 
1964 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1965 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1966 	CU_ASSERT(g_io_done == true);
1967 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1968 
1969 	/* A wrong case, a child IO that is divided does
1970 	 * not meet the principle of multiples of block size,
1971 	 * and exits with error
1972 	 */
1973 	bdev->max_segment_size = 512;
1974 	bdev->max_num_segments = 1;
1975 	g_io_done = false;
1976 
1977 	iov[0].iov_base = (void *)0x10000;
1978 	iov[0].iov_len = 512 + 256;
1979 	iov[1].iov_base = (void *)0x20000;
1980 	iov[1].iov_len = 256;
1981 
1982 	/* iov[0] is split to 512 and 256.
1983 	 * 256 is less than a block size, and it is found
1984 	 * in the next round of split that it is the first child IO smaller than
1985 	 * the block size, so the error exit
1986 	 */
1987 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 1, 1);
1988 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512);
1989 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1990 
1991 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 0, 2, io_done, NULL);
1992 	CU_ASSERT(rc == 0);
1993 	CU_ASSERT(g_io_done == false);
1994 
1995 	/* First child IO is OK */
1996 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1997 	stub_complete_io(1);
1998 	CU_ASSERT(g_io_done == true);
1999 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2000 
2001 	/* error exit */
2002 	stub_complete_io(1);
2003 	CU_ASSERT(g_io_done == true);
2004 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
2005 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2006 
2007 	/* Test multi vector command that needs to be split by strip and then needs to be
2008 	 * split further due to the capacity of child iovs.
2009 	 *
2010 	 * In this case, the last two iovs need to be split, but it will exceed the capacity
2011 	 * of child iovs, so it needs to wait until the first batch completed.
2012 	 */
2013 	bdev->max_segment_size = 512;
2014 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2015 	g_io_done = false;
2016 
2017 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2018 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2019 		iov[i].iov_len = 512;
2020 	}
2021 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
2022 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2023 		iov[i].iov_len = 512 * 2;
2024 	}
2025 
2026 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2027 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
2028 	/* 0 ~ (BDEV_IO_NUM_CHILD_IOV - 2) Will not be split */
2029 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2030 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2031 	}
2032 	/* (BDEV_IO_NUM_CHILD_IOV - 2) is split */
2033 	ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, 512);
2034 	ut_expected_io_set_iov(expected_io, i + 1, iov[i].iov_base + 512, 512);
2035 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2036 
2037 	/* Child iov entries exceed the max num of parent IO so split it in next round */
2038 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV, 2, 2);
2039 	ut_expected_io_set_iov(expected_io, 0, iov[i + 1].iov_base, 512);
2040 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base + 512, 512);
2041 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2042 
2043 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
2044 				    BDEV_IO_NUM_CHILD_IOV + 2, io_done, NULL);
2045 	CU_ASSERT(rc == 0);
2046 	CU_ASSERT(g_io_done == false);
2047 
2048 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2049 	stub_complete_io(1);
2050 	CU_ASSERT(g_io_done == false);
2051 
2052 	/* Next round */
2053 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2054 	stub_complete_io(1);
2055 	CU_ASSERT(g_io_done == true);
2056 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2057 
2058 	/* This case is similar to the previous one, but the io composed of
2059 	 * the last few entries of child iov is not enough for a blocklen, so they
2060 	 * cannot be put into this IO, but wait until the next time.
2061 	 */
2062 	bdev->max_segment_size = 512;
2063 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2064 	g_io_done = false;
2065 
2066 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2067 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2068 		iov[i].iov_len = 512;
2069 	}
2070 
2071 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2072 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2073 		iov[i].iov_len = 128;
2074 	}
2075 
2076 	/* First child iovcnt is't BDEV_IO_NUM_CHILD_IOV but BDEV_IO_NUM_CHILD_IOV - 2.
2077 	 * Because the left 2 iov is not enough for a blocklen.
2078 	 */
2079 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2080 					   BDEV_IO_NUM_CHILD_IOV - 2, BDEV_IO_NUM_CHILD_IOV - 2);
2081 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2082 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2083 	}
2084 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2085 
2086 	/* The second child io waits until the end of the first child io before executing.
2087 	 * Because the iovcnt of the two IOs exceeds the child iovcnt of the parent IO.
2088 	 * BDEV_IO_NUM_CHILD_IOV - 2 to BDEV_IO_NUM_CHILD_IOV + 2
2089 	 */
2090 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 2,
2091 					   1, 4);
2092 	ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, iov[i].iov_len);
2093 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base, iov[i + 1].iov_len);
2094 	ut_expected_io_set_iov(expected_io, 2, iov[i + 2].iov_base, iov[i + 2].iov_len);
2095 	ut_expected_io_set_iov(expected_io, 3, iov[i + 3].iov_base, iov[i + 3].iov_len);
2096 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2097 
2098 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2099 				    BDEV_IO_NUM_CHILD_IOV - 1, io_done, NULL);
2100 	CU_ASSERT(rc == 0);
2101 	CU_ASSERT(g_io_done == false);
2102 
2103 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2104 	stub_complete_io(1);
2105 	CU_ASSERT(g_io_done == false);
2106 
2107 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2108 	stub_complete_io(1);
2109 	CU_ASSERT(g_io_done == true);
2110 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2111 
2112 	/* A very complicated case. Each sg entry exceeds max_segment_size and
2113 	 * needs to be split. At the same time, child io must be a multiple of blocklen.
2114 	 * At the same time, child iovcnt exceeds parent iovcnt.
2115 	 */
2116 	bdev->max_segment_size = 512 + 128;
2117 	bdev->max_num_segments = 3;
2118 	g_io_done = false;
2119 
2120 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2121 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2122 		iov[i].iov_len = 512 + 256;
2123 	}
2124 
2125 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2126 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2127 		iov[i].iov_len = 512 + 128;
2128 	}
2129 
2130 	/* Child IOs use 9 entries per for() round and 3 * 9 = 27 child iov entries.
2131 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2132 	 * Generate 9 child IOs.
2133 	 */
2134 	for (i = 0; i < 3; i++) {
2135 		uint32_t j = i * 4;
2136 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6, 2, 3);
2137 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2138 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2139 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2140 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2141 
2142 		/* Child io must be a multiple of blocklen
2143 		 * iov[j + 2] must be split. If the third entry is also added,
2144 		 * the multiple of blocklen cannot be guaranteed. But it still
2145 		 * occupies one iov entry of the parent child iov.
2146 		 */
2147 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 2, 2, 2);
2148 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2149 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2150 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2151 
2152 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 4, 2, 3);
2153 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2154 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2155 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2156 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2157 	}
2158 
2159 	/* Child iov position at 27, the 10th child IO
2160 	 * iov entry index is 3 * 4 and offset is 3 * 6
2161 	 */
2162 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 18, 2, 3);
2163 	ut_expected_io_set_iov(expected_io, 0, iov[12].iov_base, 640);
2164 	ut_expected_io_set_iov(expected_io, 1, iov[12].iov_base + 640, 128);
2165 	ut_expected_io_set_iov(expected_io, 2, iov[13].iov_base, 256);
2166 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2167 
2168 	/* Child iov position at 30, the 11th child IO */
2169 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 20, 2, 2);
2170 	ut_expected_io_set_iov(expected_io, 0, iov[13].iov_base + 256, 512);
2171 	ut_expected_io_set_iov(expected_io, 1, iov[14].iov_base, 512);
2172 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2173 
2174 	/* The 2nd split round and iovpos is 0, the 12th child IO */
2175 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 22, 2, 3);
2176 	ut_expected_io_set_iov(expected_io, 0, iov[14].iov_base + 512, 256);
2177 	ut_expected_io_set_iov(expected_io, 1, iov[15].iov_base, 640);
2178 	ut_expected_io_set_iov(expected_io, 2, iov[15].iov_base + 640, 128);
2179 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2180 
2181 	/* Consume 9 child IOs and 27 child iov entries.
2182 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2183 	 * Parent IO iov index start from 16 and block offset start from 24
2184 	 */
2185 	for (i = 0; i < 3; i++) {
2186 		uint32_t j = i * 4 + 16;
2187 		uint32_t offset = i * 6 + 24;
2188 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, 2, 3);
2189 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2190 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2191 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2192 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2193 
2194 		/* Child io must be a multiple of blocklen
2195 		 * iov[j + 2] must be split. If the third entry is also added,
2196 		 * the multiple of blocklen cannot be guaranteed. But it still
2197 		 * occupies one iov entry of the parent child iov.
2198 		 */
2199 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 2, 2, 2);
2200 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2201 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2202 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2203 
2204 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 4, 2, 3);
2205 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2206 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2207 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2208 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2209 	}
2210 
2211 	/* The 22th child IO, child iov position at 30 */
2212 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 42, 1, 1);
2213 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base, 512);
2214 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2215 
2216 	/* The third round */
2217 	/* Here is the 23nd child IO and child iovpos is 0 */
2218 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 43, 2, 3);
2219 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base + 512, 256);
2220 	ut_expected_io_set_iov(expected_io, 1, iov[29].iov_base, 640);
2221 	ut_expected_io_set_iov(expected_io, 2, iov[29].iov_base + 640, 128);
2222 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2223 
2224 	/* The 24th child IO */
2225 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 45, 3, 3);
2226 	ut_expected_io_set_iov(expected_io, 0, iov[30].iov_base, 640);
2227 	ut_expected_io_set_iov(expected_io, 1, iov[31].iov_base, 640);
2228 	ut_expected_io_set_iov(expected_io, 2, iov[32].iov_base, 256);
2229 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2230 
2231 	/* The 25th child IO */
2232 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 48, 2, 2);
2233 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base + 256, 384);
2234 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 640);
2235 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2236 
2237 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2238 				    50, io_done, NULL);
2239 	CU_ASSERT(rc == 0);
2240 	CU_ASSERT(g_io_done == false);
2241 
2242 	/* Parent IO supports up to 32 child iovs, so it is calculated that
2243 	 * a maximum of 11 IOs can be split at a time, and the
2244 	 * splitting will continue after the first batch is over.
2245 	 */
2246 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2247 	stub_complete_io(11);
2248 	CU_ASSERT(g_io_done == false);
2249 
2250 	/* The 2nd round */
2251 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2252 	stub_complete_io(11);
2253 	CU_ASSERT(g_io_done == false);
2254 
2255 	/* The last round */
2256 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2257 	stub_complete_io(3);
2258 	CU_ASSERT(g_io_done == true);
2259 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2260 
2261 	/* Test an WRITE_ZEROES.  This should also not be split. */
2262 	bdev->max_segment_size = 512;
2263 	bdev->max_num_segments = 1;
2264 	g_io_done = false;
2265 
2266 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
2267 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2268 
2269 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
2270 	CU_ASSERT(rc == 0);
2271 	CU_ASSERT(g_io_done == false);
2272 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2273 	stub_complete_io(1);
2274 	CU_ASSERT(g_io_done == true);
2275 
2276 	/* Test an UNMAP.  This should also not be split. */
2277 	g_io_done = false;
2278 
2279 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 4, 0);
2280 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2281 
2282 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 4, io_done, NULL);
2283 	CU_ASSERT(rc == 0);
2284 	CU_ASSERT(g_io_done == false);
2285 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2286 	stub_complete_io(1);
2287 	CU_ASSERT(g_io_done == true);
2288 
2289 	/* Test a FLUSH.  This should also not be split. */
2290 	g_io_done = false;
2291 
2292 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 4, 0);
2293 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2294 
2295 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
2296 	CU_ASSERT(rc == 0);
2297 	CU_ASSERT(g_io_done == false);
2298 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2299 	stub_complete_io(1);
2300 	CU_ASSERT(g_io_done == true);
2301 
2302 	spdk_put_io_channel(io_ch);
2303 	spdk_bdev_close(desc);
2304 	free_bdev(bdev);
2305 	spdk_bdev_finish(bdev_fini_cb, NULL);
2306 	poll_threads();
2307 }
2308 
2309 static void
2310 bdev_io_mix_split_test(void)
2311 {
2312 	struct spdk_bdev *bdev;
2313 	struct spdk_bdev_desc *desc = NULL;
2314 	struct spdk_io_channel *io_ch;
2315 	struct spdk_bdev_opts bdev_opts = {};
2316 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
2317 	struct ut_expected_io *expected_io;
2318 	uint64_t i;
2319 	int rc;
2320 
2321 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2322 	bdev_opts.bdev_io_pool_size = 512;
2323 	bdev_opts.bdev_io_cache_size = 64;
2324 
2325 	rc = spdk_bdev_set_opts(&bdev_opts);
2326 	CU_ASSERT(rc == 0);
2327 	spdk_bdev_initialize(bdev_init_cb, NULL);
2328 
2329 	bdev = allocate_bdev("bdev0");
2330 
2331 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
2332 	CU_ASSERT(rc == 0);
2333 	SPDK_CU_ASSERT_FATAL(desc != NULL);
2334 	io_ch = spdk_bdev_get_io_channel(desc);
2335 	CU_ASSERT(io_ch != NULL);
2336 
2337 	/* First case optimal_io_boundary == max_segment_size * max_num_segments */
2338 	bdev->split_on_optimal_io_boundary = true;
2339 	bdev->optimal_io_boundary = 16;
2340 
2341 	bdev->max_segment_size = 512;
2342 	bdev->max_num_segments = 16;
2343 	g_io_done = false;
2344 
2345 	/* IO crossing the IO boundary requires split
2346 	 * Total 2 child IOs.
2347 	 */
2348 
2349 	/* The 1st child IO split the segment_size to multiple segment entry */
2350 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
2351 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
2352 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
2353 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2354 
2355 	/* The 2nd child IO split the segment_size to multiple segment entry */
2356 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 2);
2357 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 512);
2358 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 3 * 512), 512);
2359 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2360 
2361 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 4, io_done, NULL);
2362 	CU_ASSERT(rc == 0);
2363 	CU_ASSERT(g_io_done == false);
2364 
2365 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2366 	stub_complete_io(2);
2367 	CU_ASSERT(g_io_done == true);
2368 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2369 
2370 	/* Second case optimal_io_boundary > max_segment_size * max_num_segments */
2371 	bdev->max_segment_size = 15 * 512;
2372 	bdev->max_num_segments = 1;
2373 	g_io_done = false;
2374 
2375 	/* IO crossing the IO boundary requires split.
2376 	 * The 1st child IO segment size exceeds the max_segment_size,
2377 	 * So 1st child IO will be splitted to multiple segment entry.
2378 	 * Then it split to 2 child IOs because of the max_num_segments.
2379 	 * Total 3 child IOs.
2380 	 */
2381 
2382 	/* The first 2 IOs are in an IO boundary.
2383 	 * Because the optimal_io_boundary > max_segment_size * max_num_segments
2384 	 * So it split to the first 2 IOs.
2385 	 */
2386 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 15, 1);
2387 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 15);
2388 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2389 
2390 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
2391 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 15), 512);
2392 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2393 
2394 	/* The 3rd Child IO is because of the io boundary */
2395 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2396 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2397 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2398 
2399 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2400 	CU_ASSERT(rc == 0);
2401 	CU_ASSERT(g_io_done == false);
2402 
2403 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2404 	stub_complete_io(3);
2405 	CU_ASSERT(g_io_done == true);
2406 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2407 
2408 	/* Third case optimal_io_boundary < max_segment_size * max_num_segments */
2409 	bdev->max_segment_size = 17 * 512;
2410 	bdev->max_num_segments = 1;
2411 	g_io_done = false;
2412 
2413 	/* IO crossing the IO boundary requires split.
2414 	 * Child IO does not split.
2415 	 * Total 2 child IOs.
2416 	 */
2417 
2418 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 16, 1);
2419 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 16);
2420 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2421 
2422 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2423 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2424 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2425 
2426 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2427 	CU_ASSERT(rc == 0);
2428 	CU_ASSERT(g_io_done == false);
2429 
2430 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2431 	stub_complete_io(2);
2432 	CU_ASSERT(g_io_done == true);
2433 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2434 
2435 	/* Now set up a more complex, multi-vector command that needs to be split,
2436 	 * including splitting iovecs.
2437 	 * optimal_io_boundary < max_segment_size * max_num_segments
2438 	 */
2439 	bdev->max_segment_size = 3 * 512;
2440 	bdev->max_num_segments = 6;
2441 	g_io_done = false;
2442 
2443 	iov[0].iov_base = (void *)0x10000;
2444 	iov[0].iov_len = 4 * 512;
2445 	iov[1].iov_base = (void *)0x20000;
2446 	iov[1].iov_len = 4 * 512;
2447 	iov[2].iov_base = (void *)0x30000;
2448 	iov[2].iov_len = 10 * 512;
2449 
2450 	/* IO crossing the IO boundary requires split.
2451 	 * The 1st child IO segment size exceeds the max_segment_size and after
2452 	 * splitting segment_size, the num_segments exceeds max_num_segments.
2453 	 * So 1st child IO will be splitted to 2 child IOs.
2454 	 * Total 3 child IOs.
2455 	 */
2456 
2457 	/* The first 2 IOs are in an IO boundary.
2458 	 * After splitting segmemt size the segment num exceeds.
2459 	 * So it splits to 2 child IOs.
2460 	 */
2461 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 14, 6);
2462 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 3);
2463 	ut_expected_io_set_iov(expected_io, 1, iov[0].iov_base + 512 * 3, 512);
2464 	ut_expected_io_set_iov(expected_io, 2, iov[1].iov_base, 512 * 3);
2465 	ut_expected_io_set_iov(expected_io, 3, iov[1].iov_base + 512 * 3, 512);
2466 	ut_expected_io_set_iov(expected_io, 4, iov[2].iov_base, 512 * 3);
2467 	ut_expected_io_set_iov(expected_io, 5, iov[2].iov_base + 512 * 3, 512 * 3);
2468 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2469 
2470 	/* The 2nd child IO has the left segment entry */
2471 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
2472 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 6, 512 * 2);
2473 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2474 
2475 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
2476 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 8, 512 * 2);
2477 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2478 
2479 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 0, 18, io_done, NULL);
2480 	CU_ASSERT(rc == 0);
2481 	CU_ASSERT(g_io_done == false);
2482 
2483 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2484 	stub_complete_io(3);
2485 	CU_ASSERT(g_io_done == true);
2486 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2487 
2488 	/* A very complicated case. Each sg entry exceeds max_segment_size
2489 	 * and split on io boundary.
2490 	 * optimal_io_boundary < max_segment_size * max_num_segments
2491 	 */
2492 	bdev->max_segment_size = 3 * 512;
2493 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2494 	g_io_done = false;
2495 
2496 	for (i = 0; i < 20; i++) {
2497 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2498 		iov[i].iov_len = 512 * 4;
2499 	}
2500 
2501 	/* IO crossing the IO boundary requires split.
2502 	 * 80 block length can split 5 child IOs base on offset and IO boundary.
2503 	 * Each iov entry needs to be splitted to 2 entries because of max_segment_size
2504 	 * Total 5 child IOs.
2505 	 */
2506 
2507 	/* 4 iov entries are in an IO boundary and each iov entry splits to 2.
2508 	 * So each child IO occupies 8 child iov entries.
2509 	 */
2510 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 16, 8);
2511 	for (i = 0; i < 4; i++) {
2512 		int iovcnt = i * 2;
2513 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2514 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2515 	}
2516 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2517 
2518 	/* 2nd child IO and total 16 child iov entries of parent IO */
2519 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 8);
2520 	for (i = 4; i < 8; i++) {
2521 		int iovcnt = (i - 4) * 2;
2522 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2523 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2524 	}
2525 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2526 
2527 	/* 3rd child IO and total 24 child iov entries of parent IO */
2528 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 16, 8);
2529 	for (i = 8; i < 12; i++) {
2530 		int iovcnt = (i - 8) * 2;
2531 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2532 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2533 	}
2534 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2535 
2536 	/* 4th child IO and total 32 child iov entries of parent IO */
2537 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 48, 16, 8);
2538 	for (i = 12; i < 16; i++) {
2539 		int iovcnt = (i - 12) * 2;
2540 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2541 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2542 	}
2543 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2544 
2545 	/* 5th child IO and because of the child iov entry it should be splitted
2546 	 * in next round.
2547 	 */
2548 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 64, 16, 8);
2549 	for (i = 16; i < 20; i++) {
2550 		int iovcnt = (i - 16) * 2;
2551 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2552 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2553 	}
2554 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2555 
2556 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 20, 0, 80, io_done, NULL);
2557 	CU_ASSERT(rc == 0);
2558 	CU_ASSERT(g_io_done == false);
2559 
2560 	/* First split round */
2561 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
2562 	stub_complete_io(4);
2563 	CU_ASSERT(g_io_done == false);
2564 
2565 	/* Second split round */
2566 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2567 	stub_complete_io(1);
2568 	CU_ASSERT(g_io_done == true);
2569 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2570 
2571 	spdk_put_io_channel(io_ch);
2572 	spdk_bdev_close(desc);
2573 	free_bdev(bdev);
2574 	spdk_bdev_finish(bdev_fini_cb, NULL);
2575 	poll_threads();
2576 }
2577 
2578 static void
2579 bdev_io_split_with_io_wait(void)
2580 {
2581 	struct spdk_bdev *bdev;
2582 	struct spdk_bdev_desc *desc = NULL;
2583 	struct spdk_io_channel *io_ch;
2584 	struct spdk_bdev_channel *channel;
2585 	struct spdk_bdev_mgmt_channel *mgmt_ch;
2586 	struct spdk_bdev_opts bdev_opts = {};
2587 	struct iovec iov[3];
2588 	struct ut_expected_io *expected_io;
2589 	int rc;
2590 
2591 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2592 	bdev_opts.bdev_io_pool_size = 2;
2593 	bdev_opts.bdev_io_cache_size = 1;
2594 
2595 	rc = spdk_bdev_set_opts(&bdev_opts);
2596 	CU_ASSERT(rc == 0);
2597 	spdk_bdev_initialize(bdev_init_cb, NULL);
2598 
2599 	bdev = allocate_bdev("bdev0");
2600 
2601 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2602 	CU_ASSERT(rc == 0);
2603 	CU_ASSERT(desc != NULL);
2604 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2605 	io_ch = spdk_bdev_get_io_channel(desc);
2606 	CU_ASSERT(io_ch != NULL);
2607 	channel = spdk_io_channel_get_ctx(io_ch);
2608 	mgmt_ch = channel->shared_resource->mgmt_ch;
2609 
2610 	bdev->optimal_io_boundary = 16;
2611 	bdev->split_on_optimal_io_boundary = true;
2612 
2613 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
2614 	CU_ASSERT(rc == 0);
2615 
2616 	/* Now test that a single-vector command is split correctly.
2617 	 * Offset 14, length 8, payload 0xF000
2618 	 *  Child - Offset 14, length 2, payload 0xF000
2619 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
2620 	 *
2621 	 * Set up the expected values before calling spdk_bdev_read_blocks
2622 	 */
2623 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
2624 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
2625 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2626 
2627 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
2628 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
2629 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2630 
2631 	/* The following children will be submitted sequentially due to the capacity of
2632 	 * spdk_bdev_io.
2633 	 */
2634 
2635 	/* The first child I/O will be queued to wait until an spdk_bdev_io becomes available */
2636 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
2637 	CU_ASSERT(rc == 0);
2638 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2639 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2640 
2641 	/* Completing the first read I/O will submit the first child */
2642 	stub_complete_io(1);
2643 	CU_ASSERT(TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2644 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2645 
2646 	/* Completing the first child will submit the second child */
2647 	stub_complete_io(1);
2648 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2649 
2650 	/* Complete the second child I/O.  This should result in our callback getting
2651 	 * invoked since the parent I/O is now complete.
2652 	 */
2653 	stub_complete_io(1);
2654 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2655 
2656 	/* Now set up a more complex, multi-vector command that needs to be split,
2657 	 *  including splitting iovecs.
2658 	 */
2659 	iov[0].iov_base = (void *)0x10000;
2660 	iov[0].iov_len = 512;
2661 	iov[1].iov_base = (void *)0x20000;
2662 	iov[1].iov_len = 20 * 512;
2663 	iov[2].iov_base = (void *)0x30000;
2664 	iov[2].iov_len = 11 * 512;
2665 
2666 	g_io_done = false;
2667 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
2668 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
2669 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
2670 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2671 
2672 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
2673 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
2674 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2675 
2676 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
2677 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
2678 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
2679 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2680 
2681 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 32, io_done, NULL);
2682 	CU_ASSERT(rc == 0);
2683 	CU_ASSERT(g_io_done == false);
2684 
2685 	/* The following children will be submitted sequentially due to the capacity of
2686 	 * spdk_bdev_io.
2687 	 */
2688 
2689 	/* Completing the first child will submit the second child */
2690 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2691 	stub_complete_io(1);
2692 	CU_ASSERT(g_io_done == false);
2693 
2694 	/* Completing the second child will submit the third child */
2695 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2696 	stub_complete_io(1);
2697 	CU_ASSERT(g_io_done == false);
2698 
2699 	/* Completing the third child will result in our callback getting invoked
2700 	 * since the parent I/O is now complete.
2701 	 */
2702 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2703 	stub_complete_io(1);
2704 	CU_ASSERT(g_io_done == true);
2705 
2706 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
2707 
2708 	spdk_put_io_channel(io_ch);
2709 	spdk_bdev_close(desc);
2710 	free_bdev(bdev);
2711 	spdk_bdev_finish(bdev_fini_cb, NULL);
2712 	poll_threads();
2713 }
2714 
2715 static void
2716 bdev_io_alignment(void)
2717 {
2718 	struct spdk_bdev *bdev;
2719 	struct spdk_bdev_desc *desc = NULL;
2720 	struct spdk_io_channel *io_ch;
2721 	struct spdk_bdev_opts bdev_opts = {};
2722 	int rc;
2723 	void *buf = NULL;
2724 	struct iovec iovs[2];
2725 	int iovcnt;
2726 	uint64_t alignment;
2727 
2728 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2729 	bdev_opts.bdev_io_pool_size = 20;
2730 	bdev_opts.bdev_io_cache_size = 2;
2731 
2732 	rc = spdk_bdev_set_opts(&bdev_opts);
2733 	CU_ASSERT(rc == 0);
2734 	spdk_bdev_initialize(bdev_init_cb, NULL);
2735 
2736 	fn_table.submit_request = stub_submit_request_get_buf;
2737 	bdev = allocate_bdev("bdev0");
2738 
2739 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2740 	CU_ASSERT(rc == 0);
2741 	CU_ASSERT(desc != NULL);
2742 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2743 	io_ch = spdk_bdev_get_io_channel(desc);
2744 	CU_ASSERT(io_ch != NULL);
2745 
2746 	/* Create aligned buffer */
2747 	rc = posix_memalign(&buf, 4096, 8192);
2748 	SPDK_CU_ASSERT_FATAL(rc == 0);
2749 
2750 	/* Pass aligned single buffer with no alignment required */
2751 	alignment = 1;
2752 	bdev->required_alignment = spdk_u32log2(alignment);
2753 
2754 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2755 	CU_ASSERT(rc == 0);
2756 	stub_complete_io(1);
2757 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2758 				    alignment));
2759 
2760 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2761 	CU_ASSERT(rc == 0);
2762 	stub_complete_io(1);
2763 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2764 				    alignment));
2765 
2766 	/* Pass unaligned single buffer with no alignment required */
2767 	alignment = 1;
2768 	bdev->required_alignment = spdk_u32log2(alignment);
2769 
2770 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2771 	CU_ASSERT(rc == 0);
2772 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2773 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2774 	stub_complete_io(1);
2775 
2776 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2777 	CU_ASSERT(rc == 0);
2778 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2779 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2780 	stub_complete_io(1);
2781 
2782 	/* Pass unaligned single buffer with 512 alignment required */
2783 	alignment = 512;
2784 	bdev->required_alignment = spdk_u32log2(alignment);
2785 
2786 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2787 	CU_ASSERT(rc == 0);
2788 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2789 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2790 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2791 				    alignment));
2792 	stub_complete_io(1);
2793 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2794 
2795 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2796 	CU_ASSERT(rc == 0);
2797 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2798 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2799 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2800 				    alignment));
2801 	stub_complete_io(1);
2802 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2803 
2804 	/* Pass unaligned single buffer with 4096 alignment required */
2805 	alignment = 4096;
2806 	bdev->required_alignment = spdk_u32log2(alignment);
2807 
2808 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2809 	CU_ASSERT(rc == 0);
2810 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2811 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2812 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2813 				    alignment));
2814 	stub_complete_io(1);
2815 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2816 
2817 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2818 	CU_ASSERT(rc == 0);
2819 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2820 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2821 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2822 				    alignment));
2823 	stub_complete_io(1);
2824 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2825 
2826 	/* Pass aligned iovs with no alignment required */
2827 	alignment = 1;
2828 	bdev->required_alignment = spdk_u32log2(alignment);
2829 
2830 	iovcnt = 1;
2831 	iovs[0].iov_base = buf;
2832 	iovs[0].iov_len = 512;
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 == 0);
2837 	stub_complete_io(1);
2838 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2839 
2840 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2841 	CU_ASSERT(rc == 0);
2842 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2843 	stub_complete_io(1);
2844 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2845 
2846 	/* Pass unaligned iovs with no alignment required */
2847 	alignment = 1;
2848 	bdev->required_alignment = spdk_u32log2(alignment);
2849 
2850 	iovcnt = 2;
2851 	iovs[0].iov_base = buf + 16;
2852 	iovs[0].iov_len = 256;
2853 	iovs[1].iov_base = buf + 16 + 256 + 32;
2854 	iovs[1].iov_len = 256;
2855 
2856 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2857 	CU_ASSERT(rc == 0);
2858 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2859 	stub_complete_io(1);
2860 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2861 
2862 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2863 	CU_ASSERT(rc == 0);
2864 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2865 	stub_complete_io(1);
2866 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2867 
2868 	/* Pass unaligned iov with 2048 alignment required */
2869 	alignment = 2048;
2870 	bdev->required_alignment = spdk_u32log2(alignment);
2871 
2872 	iovcnt = 2;
2873 	iovs[0].iov_base = buf + 16;
2874 	iovs[0].iov_len = 256;
2875 	iovs[1].iov_base = buf + 16 + 256 + 32;
2876 	iovs[1].iov_len = 256;
2877 
2878 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2879 	CU_ASSERT(rc == 0);
2880 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2881 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2882 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2883 				    alignment));
2884 	stub_complete_io(1);
2885 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2886 
2887 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2888 	CU_ASSERT(rc == 0);
2889 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2890 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2891 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2892 				    alignment));
2893 	stub_complete_io(1);
2894 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2895 
2896 	/* Pass iov without allocated buffer without alignment required */
2897 	alignment = 1;
2898 	bdev->required_alignment = spdk_u32log2(alignment);
2899 
2900 	iovcnt = 1;
2901 	iovs[0].iov_base = NULL;
2902 	iovs[0].iov_len = 0;
2903 
2904 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2905 	CU_ASSERT(rc == 0);
2906 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2907 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2908 				    alignment));
2909 	stub_complete_io(1);
2910 
2911 	/* Pass iov without allocated buffer with 1024 alignment required */
2912 	alignment = 1024;
2913 	bdev->required_alignment = spdk_u32log2(alignment);
2914 
2915 	iovcnt = 1;
2916 	iovs[0].iov_base = NULL;
2917 	iovs[0].iov_len = 0;
2918 
2919 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2920 	CU_ASSERT(rc == 0);
2921 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2922 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2923 				    alignment));
2924 	stub_complete_io(1);
2925 
2926 	spdk_put_io_channel(io_ch);
2927 	spdk_bdev_close(desc);
2928 	free_bdev(bdev);
2929 	fn_table.submit_request = stub_submit_request;
2930 	spdk_bdev_finish(bdev_fini_cb, NULL);
2931 	poll_threads();
2932 
2933 	free(buf);
2934 }
2935 
2936 static void
2937 bdev_io_alignment_with_boundary(void)
2938 {
2939 	struct spdk_bdev *bdev;
2940 	struct spdk_bdev_desc *desc = NULL;
2941 	struct spdk_io_channel *io_ch;
2942 	struct spdk_bdev_opts bdev_opts = {};
2943 	int rc;
2944 	void *buf = NULL;
2945 	struct iovec iovs[2];
2946 	int iovcnt;
2947 	uint64_t alignment;
2948 
2949 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2950 	bdev_opts.bdev_io_pool_size = 20;
2951 	bdev_opts.bdev_io_cache_size = 2;
2952 
2953 	bdev_opts.opts_size = sizeof(bdev_opts);
2954 	rc = spdk_bdev_set_opts(&bdev_opts);
2955 	CU_ASSERT(rc == 0);
2956 	spdk_bdev_initialize(bdev_init_cb, NULL);
2957 
2958 	fn_table.submit_request = stub_submit_request_get_buf;
2959 	bdev = allocate_bdev("bdev0");
2960 
2961 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2962 	CU_ASSERT(rc == 0);
2963 	CU_ASSERT(desc != NULL);
2964 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2965 	io_ch = spdk_bdev_get_io_channel(desc);
2966 	CU_ASSERT(io_ch != NULL);
2967 
2968 	/* Create aligned buffer */
2969 	rc = posix_memalign(&buf, 4096, 131072);
2970 	SPDK_CU_ASSERT_FATAL(rc == 0);
2971 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2972 
2973 	/* 512 * 3 with 2 IO boundary, allocate small data buffer from bdev layer */
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 = 1;
2980 	iovs[0].iov_base = NULL;
2981 	iovs[0].iov_len = 512 * 3;
2982 
2983 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
2984 	CU_ASSERT(rc == 0);
2985 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2986 	stub_complete_io(2);
2987 
2988 	/* 8KiB with 16 IO boundary, allocate large data buffer from bdev layer */
2989 	alignment = 512;
2990 	bdev->required_alignment = spdk_u32log2(alignment);
2991 	bdev->optimal_io_boundary = 16;
2992 	bdev->split_on_optimal_io_boundary = true;
2993 
2994 	iovcnt = 1;
2995 	iovs[0].iov_base = NULL;
2996 	iovs[0].iov_len = 512 * 16;
2997 
2998 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 16, io_done, NULL);
2999 	CU_ASSERT(rc == 0);
3000 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3001 	stub_complete_io(2);
3002 
3003 	/* 512 * 160 with 128 IO boundary, 63.5KiB + 16.5KiB for the two children requests */
3004 	alignment = 512;
3005 	bdev->required_alignment = spdk_u32log2(alignment);
3006 	bdev->optimal_io_boundary = 128;
3007 	bdev->split_on_optimal_io_boundary = true;
3008 
3009 	iovcnt = 1;
3010 	iovs[0].iov_base = buf + 16;
3011 	iovs[0].iov_len = 512 * 160;
3012 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3013 	CU_ASSERT(rc == 0);
3014 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3015 	stub_complete_io(2);
3016 
3017 	/* 512 * 3 with 2 IO boundary */
3018 	alignment = 512;
3019 	bdev->required_alignment = spdk_u32log2(alignment);
3020 	bdev->optimal_io_boundary = 2;
3021 	bdev->split_on_optimal_io_boundary = true;
3022 
3023 	iovcnt = 2;
3024 	iovs[0].iov_base = buf + 16;
3025 	iovs[0].iov_len = 512;
3026 	iovs[1].iov_base = buf + 16 + 512 + 32;
3027 	iovs[1].iov_len = 1024;
3028 
3029 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3030 	CU_ASSERT(rc == 0);
3031 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3032 	stub_complete_io(2);
3033 
3034 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3035 	CU_ASSERT(rc == 0);
3036 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3037 	stub_complete_io(2);
3038 
3039 	/* 512 * 64 with 32 IO boundary */
3040 	bdev->optimal_io_boundary = 32;
3041 	iovcnt = 2;
3042 	iovs[0].iov_base = buf + 16;
3043 	iovs[0].iov_len = 16384;
3044 	iovs[1].iov_base = buf + 16 + 16384 + 32;
3045 	iovs[1].iov_len = 16384;
3046 
3047 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3048 	CU_ASSERT(rc == 0);
3049 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3050 	stub_complete_io(3);
3051 
3052 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3053 	CU_ASSERT(rc == 0);
3054 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3055 	stub_complete_io(3);
3056 
3057 	/* 512 * 160 with 32 IO boundary */
3058 	iovcnt = 1;
3059 	iovs[0].iov_base = buf + 16;
3060 	iovs[0].iov_len = 16384 + 65536;
3061 
3062 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3063 	CU_ASSERT(rc == 0);
3064 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
3065 	stub_complete_io(6);
3066 
3067 	spdk_put_io_channel(io_ch);
3068 	spdk_bdev_close(desc);
3069 	free_bdev(bdev);
3070 	fn_table.submit_request = stub_submit_request;
3071 	spdk_bdev_finish(bdev_fini_cb, NULL);
3072 	poll_threads();
3073 
3074 	free(buf);
3075 }
3076 
3077 static void
3078 histogram_status_cb(void *cb_arg, int status)
3079 {
3080 	g_status = status;
3081 }
3082 
3083 static void
3084 histogram_data_cb(void *cb_arg, int status, struct spdk_histogram_data *histogram)
3085 {
3086 	g_status = status;
3087 	g_histogram = histogram;
3088 }
3089 
3090 static void
3091 histogram_io_count(void *ctx, uint64_t start, uint64_t end, uint64_t count,
3092 		   uint64_t total, uint64_t so_far)
3093 {
3094 	g_count += count;
3095 }
3096 
3097 static void
3098 bdev_histograms(void)
3099 {
3100 	struct spdk_bdev *bdev;
3101 	struct spdk_bdev_desc *desc = NULL;
3102 	struct spdk_io_channel *ch;
3103 	struct spdk_histogram_data *histogram;
3104 	uint8_t buf[4096];
3105 	int rc;
3106 
3107 	spdk_bdev_initialize(bdev_init_cb, NULL);
3108 
3109 	bdev = allocate_bdev("bdev");
3110 
3111 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3112 	CU_ASSERT(rc == 0);
3113 	CU_ASSERT(desc != NULL);
3114 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3115 
3116 	ch = spdk_bdev_get_io_channel(desc);
3117 	CU_ASSERT(ch != NULL);
3118 
3119 	/* Enable histogram */
3120 	g_status = -1;
3121 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, true);
3122 	poll_threads();
3123 	CU_ASSERT(g_status == 0);
3124 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3125 
3126 	/* Allocate histogram */
3127 	histogram = spdk_histogram_data_alloc();
3128 	SPDK_CU_ASSERT_FATAL(histogram != NULL);
3129 
3130 	/* Check if histogram is zeroed */
3131 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3132 	poll_threads();
3133 	CU_ASSERT(g_status == 0);
3134 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3135 
3136 	g_count = 0;
3137 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3138 
3139 	CU_ASSERT(g_count == 0);
3140 
3141 	rc = spdk_bdev_write_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3142 	CU_ASSERT(rc == 0);
3143 
3144 	spdk_delay_us(10);
3145 	stub_complete_io(1);
3146 	poll_threads();
3147 
3148 	rc = spdk_bdev_read_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3149 	CU_ASSERT(rc == 0);
3150 
3151 	spdk_delay_us(10);
3152 	stub_complete_io(1);
3153 	poll_threads();
3154 
3155 	/* Check if histogram gathered data from all I/O channels */
3156 	g_histogram = NULL;
3157 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3158 	poll_threads();
3159 	CU_ASSERT(g_status == 0);
3160 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3161 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3162 
3163 	g_count = 0;
3164 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3165 	CU_ASSERT(g_count == 2);
3166 
3167 	/* Disable histogram */
3168 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, false);
3169 	poll_threads();
3170 	CU_ASSERT(g_status == 0);
3171 	CU_ASSERT(bdev->internal.histogram_enabled == false);
3172 
3173 	/* Try to run histogram commands on disabled bdev */
3174 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3175 	poll_threads();
3176 	CU_ASSERT(g_status == -EFAULT);
3177 
3178 	spdk_histogram_data_free(histogram);
3179 	spdk_put_io_channel(ch);
3180 	spdk_bdev_close(desc);
3181 	free_bdev(bdev);
3182 	spdk_bdev_finish(bdev_fini_cb, NULL);
3183 	poll_threads();
3184 }
3185 
3186 static void
3187 _bdev_compare(bool emulated)
3188 {
3189 	struct spdk_bdev *bdev;
3190 	struct spdk_bdev_desc *desc = NULL;
3191 	struct spdk_io_channel *ioch;
3192 	struct ut_expected_io *expected_io;
3193 	uint64_t offset, num_blocks;
3194 	uint32_t num_completed;
3195 	char aa_buf[512];
3196 	char bb_buf[512];
3197 	struct iovec compare_iov;
3198 	uint8_t io_type;
3199 	int rc;
3200 
3201 	if (emulated) {
3202 		io_type = SPDK_BDEV_IO_TYPE_READ;
3203 	} else {
3204 		io_type = SPDK_BDEV_IO_TYPE_COMPARE;
3205 	}
3206 
3207 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3208 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3209 
3210 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = !emulated;
3211 
3212 	spdk_bdev_initialize(bdev_init_cb, NULL);
3213 	fn_table.submit_request = stub_submit_request_get_buf;
3214 	bdev = allocate_bdev("bdev");
3215 
3216 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3217 	CU_ASSERT_EQUAL(rc, 0);
3218 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3219 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3220 	ioch = spdk_bdev_get_io_channel(desc);
3221 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3222 
3223 	fn_table.submit_request = stub_submit_request_get_buf;
3224 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3225 
3226 	offset = 50;
3227 	num_blocks = 1;
3228 	compare_iov.iov_base = aa_buf;
3229 	compare_iov.iov_len = sizeof(aa_buf);
3230 
3231 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3232 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3233 
3234 	g_io_done = false;
3235 	g_compare_read_buf = aa_buf;
3236 	g_compare_read_buf_len = sizeof(aa_buf);
3237 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3238 	CU_ASSERT_EQUAL(rc, 0);
3239 	num_completed = stub_complete_io(1);
3240 	CU_ASSERT_EQUAL(num_completed, 1);
3241 	CU_ASSERT(g_io_done == true);
3242 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3243 
3244 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3245 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3246 
3247 	g_io_done = false;
3248 	g_compare_read_buf = bb_buf;
3249 	g_compare_read_buf_len = sizeof(bb_buf);
3250 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3251 	CU_ASSERT_EQUAL(rc, 0);
3252 	num_completed = stub_complete_io(1);
3253 	CU_ASSERT_EQUAL(num_completed, 1);
3254 	CU_ASSERT(g_io_done == true);
3255 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3256 
3257 	spdk_put_io_channel(ioch);
3258 	spdk_bdev_close(desc);
3259 	free_bdev(bdev);
3260 	fn_table.submit_request = stub_submit_request;
3261 	spdk_bdev_finish(bdev_fini_cb, NULL);
3262 	poll_threads();
3263 
3264 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3265 
3266 	g_compare_read_buf = NULL;
3267 }
3268 
3269 static void
3270 bdev_compare(void)
3271 {
3272 	_bdev_compare(true);
3273 	_bdev_compare(false);
3274 }
3275 
3276 static void
3277 bdev_compare_and_write(void)
3278 {
3279 	struct spdk_bdev *bdev;
3280 	struct spdk_bdev_desc *desc = NULL;
3281 	struct spdk_io_channel *ioch;
3282 	struct ut_expected_io *expected_io;
3283 	uint64_t offset, num_blocks;
3284 	uint32_t num_completed;
3285 	char aa_buf[512];
3286 	char bb_buf[512];
3287 	char cc_buf[512];
3288 	char write_buf[512];
3289 	struct iovec compare_iov;
3290 	struct iovec write_iov;
3291 	int rc;
3292 
3293 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3294 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3295 	memset(cc_buf, 0xcc, sizeof(cc_buf));
3296 
3297 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = false;
3298 
3299 	spdk_bdev_initialize(bdev_init_cb, NULL);
3300 	fn_table.submit_request = stub_submit_request_get_buf;
3301 	bdev = allocate_bdev("bdev");
3302 
3303 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3304 	CU_ASSERT_EQUAL(rc, 0);
3305 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3306 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3307 	ioch = spdk_bdev_get_io_channel(desc);
3308 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3309 
3310 	fn_table.submit_request = stub_submit_request_get_buf;
3311 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3312 
3313 	offset = 50;
3314 	num_blocks = 1;
3315 	compare_iov.iov_base = aa_buf;
3316 	compare_iov.iov_len = sizeof(aa_buf);
3317 	write_iov.iov_base = bb_buf;
3318 	write_iov.iov_len = sizeof(bb_buf);
3319 
3320 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3321 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3322 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, offset, num_blocks, 0);
3323 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3324 
3325 	g_io_done = false;
3326 	g_compare_read_buf = aa_buf;
3327 	g_compare_read_buf_len = sizeof(aa_buf);
3328 	memset(write_buf, 0, sizeof(write_buf));
3329 	g_compare_write_buf = write_buf;
3330 	g_compare_write_buf_len = sizeof(write_buf);
3331 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3332 			offset, num_blocks, io_done, NULL);
3333 	/* Trigger range locking */
3334 	poll_threads();
3335 	CU_ASSERT_EQUAL(rc, 0);
3336 	num_completed = stub_complete_io(1);
3337 	CU_ASSERT_EQUAL(num_completed, 1);
3338 	CU_ASSERT(g_io_done == false);
3339 	num_completed = stub_complete_io(1);
3340 	/* Trigger range unlocking */
3341 	poll_threads();
3342 	CU_ASSERT_EQUAL(num_completed, 1);
3343 	CU_ASSERT(g_io_done == true);
3344 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3345 	CU_ASSERT(memcmp(write_buf, bb_buf, sizeof(write_buf)) == 0);
3346 
3347 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3348 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3349 
3350 	g_io_done = false;
3351 	g_compare_read_buf = cc_buf;
3352 	g_compare_read_buf_len = sizeof(cc_buf);
3353 	memset(write_buf, 0, sizeof(write_buf));
3354 	g_compare_write_buf = write_buf;
3355 	g_compare_write_buf_len = sizeof(write_buf);
3356 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3357 			offset, num_blocks, io_done, NULL);
3358 	/* Trigger range locking */
3359 	poll_threads();
3360 	CU_ASSERT_EQUAL(rc, 0);
3361 	num_completed = stub_complete_io(1);
3362 	/* Trigger range unlocking earlier because we expect error here */
3363 	poll_threads();
3364 	CU_ASSERT_EQUAL(num_completed, 1);
3365 	CU_ASSERT(g_io_done == true);
3366 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3367 	num_completed = stub_complete_io(1);
3368 	CU_ASSERT_EQUAL(num_completed, 0);
3369 
3370 	spdk_put_io_channel(ioch);
3371 	spdk_bdev_close(desc);
3372 	free_bdev(bdev);
3373 	fn_table.submit_request = stub_submit_request;
3374 	spdk_bdev_finish(bdev_fini_cb, NULL);
3375 	poll_threads();
3376 
3377 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3378 
3379 	g_compare_read_buf = NULL;
3380 	g_compare_write_buf = NULL;
3381 }
3382 
3383 static void
3384 bdev_write_zeroes(void)
3385 {
3386 	struct spdk_bdev *bdev;
3387 	struct spdk_bdev_desc *desc = NULL;
3388 	struct spdk_io_channel *ioch;
3389 	struct ut_expected_io *expected_io;
3390 	uint64_t offset, num_io_blocks, num_blocks;
3391 	uint32_t num_completed, num_requests;
3392 	int rc;
3393 
3394 	spdk_bdev_initialize(bdev_init_cb, NULL);
3395 	bdev = allocate_bdev("bdev");
3396 
3397 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3398 	CU_ASSERT_EQUAL(rc, 0);
3399 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3400 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3401 	ioch = spdk_bdev_get_io_channel(desc);
3402 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3403 
3404 	fn_table.submit_request = stub_submit_request;
3405 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3406 
3407 	/* First test that if the bdev supports write_zeroes, the request won't be split */
3408 	bdev->md_len = 0;
3409 	bdev->blocklen = 4096;
3410 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3411 
3412 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
3413 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3414 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3415 	CU_ASSERT_EQUAL(rc, 0);
3416 	num_completed = stub_complete_io(1);
3417 	CU_ASSERT_EQUAL(num_completed, 1);
3418 
3419 	/* Check that if write zeroes is not supported it'll be replaced by regular writes */
3420 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
3421 	num_io_blocks = ZERO_BUFFER_SIZE / bdev->blocklen;
3422 	num_requests = 2;
3423 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * num_requests;
3424 
3425 	for (offset = 0; offset < num_requests; ++offset) {
3426 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3427 						   offset * num_io_blocks, num_io_blocks, 0);
3428 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
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 	/* Check that the splitting is correct if bdev has interleaved metadata */
3437 	bdev->md_interleave = true;
3438 	bdev->md_len = 64;
3439 	bdev->blocklen = 4096 + 64;
3440 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3441 
3442 	num_requests = offset = 0;
3443 	while (offset < num_blocks) {
3444 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / bdev->blocklen, num_blocks - offset);
3445 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3446 						   offset, num_io_blocks, 0);
3447 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3448 		offset += num_io_blocks;
3449 		num_requests++;
3450 	}
3451 
3452 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3453 	CU_ASSERT_EQUAL(rc, 0);
3454 	num_completed = stub_complete_io(num_requests);
3455 	CU_ASSERT_EQUAL(num_completed, num_requests);
3456 	num_completed = stub_complete_io(num_requests);
3457 	assert(num_completed == 0);
3458 
3459 	/* Check the the same for separate metadata buffer */
3460 	bdev->md_interleave = false;
3461 	bdev->md_len = 64;
3462 	bdev->blocklen = 4096;
3463 
3464 	num_requests = offset = 0;
3465 	while (offset < num_blocks) {
3466 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / (bdev->blocklen + bdev->md_len), num_blocks);
3467 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3468 						   offset, num_io_blocks, 0);
3469 		expected_io->md_buf = (char *)g_bdev_mgr.zero_buffer + num_io_blocks * bdev->blocklen;
3470 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3471 		offset += num_io_blocks;
3472 		num_requests++;
3473 	}
3474 
3475 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3476 	CU_ASSERT_EQUAL(rc, 0);
3477 	num_completed = stub_complete_io(num_requests);
3478 	CU_ASSERT_EQUAL(num_completed, num_requests);
3479 
3480 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
3481 	spdk_put_io_channel(ioch);
3482 	spdk_bdev_close(desc);
3483 	free_bdev(bdev);
3484 	spdk_bdev_finish(bdev_fini_cb, NULL);
3485 	poll_threads();
3486 }
3487 
3488 static void
3489 bdev_zcopy_write(void)
3490 {
3491 	struct spdk_bdev *bdev;
3492 	struct spdk_bdev_desc *desc = NULL;
3493 	struct spdk_io_channel *ioch;
3494 	struct ut_expected_io *expected_io;
3495 	uint64_t offset, num_blocks;
3496 	uint32_t num_completed;
3497 	char aa_buf[512];
3498 	struct iovec iov;
3499 	int rc;
3500 	const bool populate = false;
3501 	const bool commit = true;
3502 
3503 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3504 
3505 	spdk_bdev_initialize(bdev_init_cb, NULL);
3506 	bdev = allocate_bdev("bdev");
3507 
3508 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3509 	CU_ASSERT_EQUAL(rc, 0);
3510 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3511 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3512 	ioch = spdk_bdev_get_io_channel(desc);
3513 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3514 
3515 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3516 
3517 	offset = 50;
3518 	num_blocks = 1;
3519 	iov.iov_base = NULL;
3520 	iov.iov_len = 0;
3521 
3522 	g_zcopy_read_buf = (void *) 0x1122334455667788UL;
3523 	g_zcopy_read_buf_len = (uint32_t) -1;
3524 	/* Do a zcopy start for a write (populate=false) */
3525 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3526 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3527 	g_io_done = false;
3528 	g_zcopy_write_buf = aa_buf;
3529 	g_zcopy_write_buf_len = sizeof(aa_buf);
3530 	g_zcopy_bdev_io = NULL;
3531 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3532 	CU_ASSERT_EQUAL(rc, 0);
3533 	num_completed = stub_complete_io(1);
3534 	CU_ASSERT_EQUAL(num_completed, 1);
3535 	CU_ASSERT(g_io_done == true);
3536 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3537 	/* Check that the iov has been set up */
3538 	CU_ASSERT(iov.iov_base == g_zcopy_write_buf);
3539 	CU_ASSERT(iov.iov_len == g_zcopy_write_buf_len);
3540 	/* Check that the bdev_io has been saved */
3541 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3542 	/* Now do the zcopy end for a write (commit=true) */
3543 	g_io_done = false;
3544 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3545 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3546 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3547 	CU_ASSERT_EQUAL(rc, 0);
3548 	num_completed = stub_complete_io(1);
3549 	CU_ASSERT_EQUAL(num_completed, 1);
3550 	CU_ASSERT(g_io_done == true);
3551 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3552 	/* Check the g_zcopy are reset by io_done */
3553 	CU_ASSERT(g_zcopy_write_buf == NULL);
3554 	CU_ASSERT(g_zcopy_write_buf_len == 0);
3555 	/* Check that io_done has freed the g_zcopy_bdev_io */
3556 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3557 
3558 	/* Check the zcopy read buffer has not been touched which
3559 	 * ensures that the correct buffers were used.
3560 	 */
3561 	CU_ASSERT(g_zcopy_read_buf == (void *) 0x1122334455667788UL);
3562 	CU_ASSERT(g_zcopy_read_buf_len == (uint32_t) -1);
3563 
3564 	spdk_put_io_channel(ioch);
3565 	spdk_bdev_close(desc);
3566 	free_bdev(bdev);
3567 	spdk_bdev_finish(bdev_fini_cb, NULL);
3568 	poll_threads();
3569 }
3570 
3571 static void
3572 bdev_zcopy_read(void)
3573 {
3574 	struct spdk_bdev *bdev;
3575 	struct spdk_bdev_desc *desc = NULL;
3576 	struct spdk_io_channel *ioch;
3577 	struct ut_expected_io *expected_io;
3578 	uint64_t offset, num_blocks;
3579 	uint32_t num_completed;
3580 	char aa_buf[512];
3581 	struct iovec iov;
3582 	int rc;
3583 	const bool populate = true;
3584 	const bool commit = false;
3585 
3586 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3587 
3588 	spdk_bdev_initialize(bdev_init_cb, NULL);
3589 	bdev = allocate_bdev("bdev");
3590 
3591 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3592 	CU_ASSERT_EQUAL(rc, 0);
3593 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3594 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3595 	ioch = spdk_bdev_get_io_channel(desc);
3596 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3597 
3598 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3599 
3600 	offset = 50;
3601 	num_blocks = 1;
3602 	iov.iov_base = NULL;
3603 	iov.iov_len = 0;
3604 
3605 	g_zcopy_write_buf = (void *) 0x1122334455667788UL;
3606 	g_zcopy_write_buf_len = (uint32_t) -1;
3607 
3608 	/* Do a zcopy start for a read (populate=true) */
3609 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3610 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3611 	g_io_done = false;
3612 	g_zcopy_read_buf = aa_buf;
3613 	g_zcopy_read_buf_len = sizeof(aa_buf);
3614 	g_zcopy_bdev_io = NULL;
3615 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3616 	CU_ASSERT_EQUAL(rc, 0);
3617 	num_completed = stub_complete_io(1);
3618 	CU_ASSERT_EQUAL(num_completed, 1);
3619 	CU_ASSERT(g_io_done == true);
3620 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3621 	/* Check that the iov has been set up */
3622 	CU_ASSERT(iov.iov_base == g_zcopy_read_buf);
3623 	CU_ASSERT(iov.iov_len == g_zcopy_read_buf_len);
3624 	/* Check that the bdev_io has been saved */
3625 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3626 
3627 	/* Now do the zcopy end for a read (commit=false) */
3628 	g_io_done = false;
3629 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3630 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3631 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3632 	CU_ASSERT_EQUAL(rc, 0);
3633 	num_completed = stub_complete_io(1);
3634 	CU_ASSERT_EQUAL(num_completed, 1);
3635 	CU_ASSERT(g_io_done == true);
3636 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3637 	/* Check the g_zcopy are reset by io_done */
3638 	CU_ASSERT(g_zcopy_read_buf == NULL);
3639 	CU_ASSERT(g_zcopy_read_buf_len == 0);
3640 	/* Check that io_done has freed the g_zcopy_bdev_io */
3641 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3642 
3643 	/* Check the zcopy write buffer has not been touched which
3644 	 * ensures that the correct buffers were used.
3645 	 */
3646 	CU_ASSERT(g_zcopy_write_buf == (void *) 0x1122334455667788UL);
3647 	CU_ASSERT(g_zcopy_write_buf_len == (uint32_t) -1);
3648 
3649 	spdk_put_io_channel(ioch);
3650 	spdk_bdev_close(desc);
3651 	free_bdev(bdev);
3652 	spdk_bdev_finish(bdev_fini_cb, NULL);
3653 	poll_threads();
3654 }
3655 
3656 static void
3657 bdev_open_while_hotremove(void)
3658 {
3659 	struct spdk_bdev *bdev;
3660 	struct spdk_bdev_desc *desc[2] = {};
3661 	int rc;
3662 
3663 	bdev = allocate_bdev("bdev");
3664 
3665 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[0]);
3666 	CU_ASSERT(rc == 0);
3667 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
3668 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc[0]));
3669 
3670 	spdk_bdev_unregister(bdev, NULL, NULL);
3671 
3672 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[1]);
3673 	CU_ASSERT(rc == -ENODEV);
3674 	SPDK_CU_ASSERT_FATAL(desc[1] == NULL);
3675 
3676 	spdk_bdev_close(desc[0]);
3677 	free_bdev(bdev);
3678 }
3679 
3680 static void
3681 bdev_close_while_hotremove(void)
3682 {
3683 	struct spdk_bdev *bdev;
3684 	struct spdk_bdev_desc *desc = NULL;
3685 	int rc = 0;
3686 
3687 	bdev = allocate_bdev("bdev");
3688 
3689 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc, &desc);
3690 	CU_ASSERT_EQUAL(rc, 0);
3691 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3692 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3693 
3694 	/* Simulate hot-unplug by unregistering bdev */
3695 	g_event_type1 = 0xFF;
3696 	g_unregister_arg = NULL;
3697 	g_unregister_rc = -1;
3698 	spdk_bdev_unregister(bdev, bdev_unregister_cb, (void *)0x12345678);
3699 	/* Close device while remove event is in flight */
3700 	spdk_bdev_close(desc);
3701 
3702 	/* Ensure that unregister callback is delayed */
3703 	CU_ASSERT_EQUAL(g_unregister_arg, NULL);
3704 	CU_ASSERT_EQUAL(g_unregister_rc, -1);
3705 
3706 	poll_threads();
3707 
3708 	/* Event callback shall not be issued because device was closed */
3709 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
3710 	/* Unregister callback is issued */
3711 	CU_ASSERT_EQUAL(g_unregister_arg, (void *)0x12345678);
3712 	CU_ASSERT_EQUAL(g_unregister_rc, 0);
3713 
3714 	free_bdev(bdev);
3715 }
3716 
3717 static void
3718 bdev_open_ext(void)
3719 {
3720 	struct spdk_bdev *bdev;
3721 	struct spdk_bdev_desc *desc1 = NULL;
3722 	struct spdk_bdev_desc *desc2 = NULL;
3723 	int rc = 0;
3724 
3725 	bdev = allocate_bdev("bdev");
3726 
3727 	rc = spdk_bdev_open_ext("bdev", true, NULL, NULL, &desc1);
3728 	CU_ASSERT_EQUAL(rc, -EINVAL);
3729 
3730 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc1, &desc1);
3731 	CU_ASSERT_EQUAL(rc, 0);
3732 
3733 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb2, &desc2, &desc2);
3734 	CU_ASSERT_EQUAL(rc, 0);
3735 
3736 	g_event_type1 = 0xFF;
3737 	g_event_type2 = 0xFF;
3738 
3739 	/* Simulate hot-unplug by unregistering bdev */
3740 	spdk_bdev_unregister(bdev, NULL, NULL);
3741 	poll_threads();
3742 
3743 	/* Check if correct events have been triggered in event callback fn */
3744 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_REMOVE);
3745 	CU_ASSERT_EQUAL(g_event_type2, SPDK_BDEV_EVENT_REMOVE);
3746 
3747 	free_bdev(bdev);
3748 	poll_threads();
3749 }
3750 
3751 struct timeout_io_cb_arg {
3752 	struct iovec iov;
3753 	uint8_t type;
3754 };
3755 
3756 static int
3757 bdev_channel_count_submitted_io(struct spdk_bdev_channel *ch)
3758 {
3759 	struct spdk_bdev_io *bdev_io;
3760 	int n = 0;
3761 
3762 	if (!ch) {
3763 		return -1;
3764 	}
3765 
3766 	TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
3767 		n++;
3768 	}
3769 
3770 	return n;
3771 }
3772 
3773 static void
3774 bdev_channel_io_timeout_cb(void *cb_arg, struct spdk_bdev_io *bdev_io)
3775 {
3776 	struct timeout_io_cb_arg *ctx = cb_arg;
3777 
3778 	ctx->type = bdev_io->type;
3779 	ctx->iov.iov_base = bdev_io->iov.iov_base;
3780 	ctx->iov.iov_len = bdev_io->iov.iov_len;
3781 }
3782 
3783 static void
3784 bdev_set_io_timeout(void)
3785 {
3786 	struct spdk_bdev *bdev;
3787 	struct spdk_bdev_desc *desc = NULL;
3788 	struct spdk_io_channel *io_ch = NULL;
3789 	struct spdk_bdev_channel *bdev_ch = NULL;
3790 	struct timeout_io_cb_arg cb_arg;
3791 
3792 	spdk_bdev_initialize(bdev_init_cb, NULL);
3793 
3794 	bdev = allocate_bdev("bdev");
3795 
3796 	CU_ASSERT(spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc) == 0);
3797 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3798 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3799 
3800 	io_ch = spdk_bdev_get_io_channel(desc);
3801 	CU_ASSERT(io_ch != NULL);
3802 
3803 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
3804 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
3805 
3806 	/* This is the part1.
3807 	 * We will check the bdev_ch->io_submitted list
3808 	 * TO make sure that it can link IOs and only the user submitted IOs
3809 	 */
3810 	CU_ASSERT(spdk_bdev_read(desc, io_ch, (void *)0x1000, 0, 4096, io_done, NULL) == 0);
3811 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3812 	CU_ASSERT(spdk_bdev_write(desc, io_ch, (void *)0x2000, 0, 4096, io_done, NULL) == 0);
3813 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3814 	stub_complete_io(1);
3815 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3816 	stub_complete_io(1);
3817 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3818 
3819 	/* Split IO */
3820 	bdev->optimal_io_boundary = 16;
3821 	bdev->split_on_optimal_io_boundary = true;
3822 
3823 	/* Now test that a single-vector command is split correctly.
3824 	 * Offset 14, length 8, payload 0xF000
3825 	 *  Child - Offset 14, length 2, payload 0xF000
3826 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
3827 	 *
3828 	 * Set up the expected values before calling spdk_bdev_read_blocks
3829 	 */
3830 	CU_ASSERT(spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3831 	/* We count all submitted IOs including IO that are generated by splitting. */
3832 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 3);
3833 	stub_complete_io(1);
3834 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3835 	stub_complete_io(1);
3836 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3837 
3838 	/* Also include the reset IO */
3839 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3840 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3841 	poll_threads();
3842 	stub_complete_io(1);
3843 	poll_threads();
3844 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3845 
3846 	/* This is part2
3847 	 * Test the desc timeout poller register
3848 	 */
3849 
3850 	/* Successfully set the timeout */
3851 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3852 	CU_ASSERT(desc->io_timeout_poller != NULL);
3853 	CU_ASSERT(desc->timeout_in_sec == 30);
3854 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3855 	CU_ASSERT(desc->cb_arg == &cb_arg);
3856 
3857 	/* Change the timeout limit */
3858 	CU_ASSERT(spdk_bdev_set_timeout(desc, 20, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3859 	CU_ASSERT(desc->io_timeout_poller != NULL);
3860 	CU_ASSERT(desc->timeout_in_sec == 20);
3861 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3862 	CU_ASSERT(desc->cb_arg == &cb_arg);
3863 
3864 	/* Disable the timeout */
3865 	CU_ASSERT(spdk_bdev_set_timeout(desc, 0, NULL, NULL) == 0);
3866 	CU_ASSERT(desc->io_timeout_poller == NULL);
3867 
3868 	/* This the part3
3869 	 * We will test to catch timeout IO and check whether the IO is
3870 	 * the submitted one.
3871 	 */
3872 	memset(&cb_arg, 0, sizeof(cb_arg));
3873 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3874 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0x1000, 0, 1, io_done, NULL) == 0);
3875 
3876 	/* Don't reach the limit */
3877 	spdk_delay_us(15 * spdk_get_ticks_hz());
3878 	poll_threads();
3879 	CU_ASSERT(cb_arg.type == 0);
3880 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3881 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3882 
3883 	/* 15 + 15 = 30 reach the limit */
3884 	spdk_delay_us(15 * spdk_get_ticks_hz());
3885 	poll_threads();
3886 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3887 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x1000);
3888 	CU_ASSERT(cb_arg.iov.iov_len == 1 * bdev->blocklen);
3889 	stub_complete_io(1);
3890 
3891 	/* Use the same split IO above and check the IO */
3892 	memset(&cb_arg, 0, sizeof(cb_arg));
3893 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3894 
3895 	/* The first child complete in time */
3896 	spdk_delay_us(15 * spdk_get_ticks_hz());
3897 	poll_threads();
3898 	stub_complete_io(1);
3899 	CU_ASSERT(cb_arg.type == 0);
3900 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3901 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3902 
3903 	/* The second child reach the limit */
3904 	spdk_delay_us(15 * spdk_get_ticks_hz());
3905 	poll_threads();
3906 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3907 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0xF000);
3908 	CU_ASSERT(cb_arg.iov.iov_len == 8 * bdev->blocklen);
3909 	stub_complete_io(1);
3910 
3911 	/* Also include the reset IO */
3912 	memset(&cb_arg, 0, sizeof(cb_arg));
3913 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3914 	spdk_delay_us(30 * spdk_get_ticks_hz());
3915 	poll_threads();
3916 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_RESET);
3917 	stub_complete_io(1);
3918 	poll_threads();
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 lba_range_overlap(void)
3929 {
3930 	struct lba_range r1, r2;
3931 
3932 	r1.offset = 100;
3933 	r1.length = 50;
3934 
3935 	r2.offset = 0;
3936 	r2.length = 1;
3937 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3938 
3939 	r2.offset = 0;
3940 	r2.length = 100;
3941 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3942 
3943 	r2.offset = 0;
3944 	r2.length = 110;
3945 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3946 
3947 	r2.offset = 100;
3948 	r2.length = 10;
3949 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3950 
3951 	r2.offset = 110;
3952 	r2.length = 20;
3953 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3954 
3955 	r2.offset = 140;
3956 	r2.length = 150;
3957 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3958 
3959 	r2.offset = 130;
3960 	r2.length = 200;
3961 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3962 
3963 	r2.offset = 150;
3964 	r2.length = 100;
3965 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3966 
3967 	r2.offset = 110;
3968 	r2.length = 0;
3969 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3970 }
3971 
3972 static bool g_lock_lba_range_done;
3973 static bool g_unlock_lba_range_done;
3974 
3975 static void
3976 lock_lba_range_done(void *ctx, int status)
3977 {
3978 	g_lock_lba_range_done = true;
3979 }
3980 
3981 static void
3982 unlock_lba_range_done(void *ctx, int status)
3983 {
3984 	g_unlock_lba_range_done = true;
3985 }
3986 
3987 static void
3988 lock_lba_range_check_ranges(void)
3989 {
3990 	struct spdk_bdev *bdev;
3991 	struct spdk_bdev_desc *desc = NULL;
3992 	struct spdk_io_channel *io_ch;
3993 	struct spdk_bdev_channel *channel;
3994 	struct lba_range *range;
3995 	int ctx1;
3996 	int rc;
3997 
3998 	spdk_bdev_initialize(bdev_init_cb, NULL);
3999 
4000 	bdev = allocate_bdev("bdev0");
4001 
4002 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4003 	CU_ASSERT(rc == 0);
4004 	CU_ASSERT(desc != NULL);
4005 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4006 	io_ch = spdk_bdev_get_io_channel(desc);
4007 	CU_ASSERT(io_ch != NULL);
4008 	channel = spdk_io_channel_get_ctx(io_ch);
4009 
4010 	g_lock_lba_range_done = false;
4011 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4012 	CU_ASSERT(rc == 0);
4013 	poll_threads();
4014 
4015 	CU_ASSERT(g_lock_lba_range_done == true);
4016 	range = TAILQ_FIRST(&channel->locked_ranges);
4017 	SPDK_CU_ASSERT_FATAL(range != NULL);
4018 	CU_ASSERT(range->offset == 20);
4019 	CU_ASSERT(range->length == 10);
4020 	CU_ASSERT(range->owner_ch == channel);
4021 
4022 	/* Unlocks must exactly match a lock. */
4023 	g_unlock_lba_range_done = false;
4024 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 1, unlock_lba_range_done, &ctx1);
4025 	CU_ASSERT(rc == -EINVAL);
4026 	CU_ASSERT(g_unlock_lba_range_done == false);
4027 
4028 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4029 	CU_ASSERT(rc == 0);
4030 	spdk_delay_us(100);
4031 	poll_threads();
4032 
4033 	CU_ASSERT(g_unlock_lba_range_done == true);
4034 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4035 
4036 	spdk_put_io_channel(io_ch);
4037 	spdk_bdev_close(desc);
4038 	free_bdev(bdev);
4039 	spdk_bdev_finish(bdev_fini_cb, NULL);
4040 	poll_threads();
4041 }
4042 
4043 static void
4044 lock_lba_range_with_io_outstanding(void)
4045 {
4046 	struct spdk_bdev *bdev;
4047 	struct spdk_bdev_desc *desc = NULL;
4048 	struct spdk_io_channel *io_ch;
4049 	struct spdk_bdev_channel *channel;
4050 	struct lba_range *range;
4051 	char buf[4096];
4052 	int ctx1;
4053 	int rc;
4054 
4055 	spdk_bdev_initialize(bdev_init_cb, NULL);
4056 
4057 	bdev = allocate_bdev("bdev0");
4058 
4059 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4060 	CU_ASSERT(rc == 0);
4061 	CU_ASSERT(desc != NULL);
4062 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4063 	io_ch = spdk_bdev_get_io_channel(desc);
4064 	CU_ASSERT(io_ch != NULL);
4065 	channel = spdk_io_channel_get_ctx(io_ch);
4066 
4067 	g_io_done = false;
4068 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4069 	CU_ASSERT(rc == 0);
4070 
4071 	g_lock_lba_range_done = false;
4072 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4073 	CU_ASSERT(rc == 0);
4074 	poll_threads();
4075 
4076 	/* The lock should immediately become valid, since there are no outstanding
4077 	 * write I/O.
4078 	 */
4079 	CU_ASSERT(g_io_done == false);
4080 	CU_ASSERT(g_lock_lba_range_done == true);
4081 	range = TAILQ_FIRST(&channel->locked_ranges);
4082 	SPDK_CU_ASSERT_FATAL(range != NULL);
4083 	CU_ASSERT(range->offset == 20);
4084 	CU_ASSERT(range->length == 10);
4085 	CU_ASSERT(range->owner_ch == channel);
4086 	CU_ASSERT(range->locked_ctx == &ctx1);
4087 
4088 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4089 	CU_ASSERT(rc == 0);
4090 	stub_complete_io(1);
4091 	spdk_delay_us(100);
4092 	poll_threads();
4093 
4094 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4095 
4096 	/* Now try again, but with a write I/O. */
4097 	g_io_done = false;
4098 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4099 	CU_ASSERT(rc == 0);
4100 
4101 	g_lock_lba_range_done = false;
4102 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4103 	CU_ASSERT(rc == 0);
4104 	poll_threads();
4105 
4106 	/* The lock should not be fully valid yet, since a write I/O is outstanding.
4107 	 * But note that the range should be on the channel's locked_list, to make sure no
4108 	 * new write I/O are started.
4109 	 */
4110 	CU_ASSERT(g_io_done == false);
4111 	CU_ASSERT(g_lock_lba_range_done == false);
4112 	range = TAILQ_FIRST(&channel->locked_ranges);
4113 	SPDK_CU_ASSERT_FATAL(range != NULL);
4114 	CU_ASSERT(range->offset == 20);
4115 	CU_ASSERT(range->length == 10);
4116 
4117 	/* Complete the write I/O.  This should make the lock valid (checked by confirming
4118 	 * our callback was invoked).
4119 	 */
4120 	stub_complete_io(1);
4121 	spdk_delay_us(100);
4122 	poll_threads();
4123 	CU_ASSERT(g_io_done == true);
4124 	CU_ASSERT(g_lock_lba_range_done == true);
4125 
4126 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4127 	CU_ASSERT(rc == 0);
4128 	poll_threads();
4129 
4130 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4131 
4132 	spdk_put_io_channel(io_ch);
4133 	spdk_bdev_close(desc);
4134 	free_bdev(bdev);
4135 	spdk_bdev_finish(bdev_fini_cb, NULL);
4136 	poll_threads();
4137 }
4138 
4139 static void
4140 lock_lba_range_overlapped(void)
4141 {
4142 	struct spdk_bdev *bdev;
4143 	struct spdk_bdev_desc *desc = NULL;
4144 	struct spdk_io_channel *io_ch;
4145 	struct spdk_bdev_channel *channel;
4146 	struct lba_range *range;
4147 	int ctx1;
4148 	int rc;
4149 
4150 	spdk_bdev_initialize(bdev_init_cb, NULL);
4151 
4152 	bdev = allocate_bdev("bdev0");
4153 
4154 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4155 	CU_ASSERT(rc == 0);
4156 	CU_ASSERT(desc != NULL);
4157 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4158 	io_ch = spdk_bdev_get_io_channel(desc);
4159 	CU_ASSERT(io_ch != NULL);
4160 	channel = spdk_io_channel_get_ctx(io_ch);
4161 
4162 	/* Lock range 20-29. */
4163 	g_lock_lba_range_done = false;
4164 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4165 	CU_ASSERT(rc == 0);
4166 	poll_threads();
4167 
4168 	CU_ASSERT(g_lock_lba_range_done == true);
4169 	range = TAILQ_FIRST(&channel->locked_ranges);
4170 	SPDK_CU_ASSERT_FATAL(range != NULL);
4171 	CU_ASSERT(range->offset == 20);
4172 	CU_ASSERT(range->length == 10);
4173 
4174 	/* Try to lock range 25-39.  It should not lock immediately, since it overlaps with
4175 	 * 20-29.
4176 	 */
4177 	g_lock_lba_range_done = false;
4178 	rc = bdev_lock_lba_range(desc, io_ch, 25, 15, lock_lba_range_done, &ctx1);
4179 	CU_ASSERT(rc == 0);
4180 	poll_threads();
4181 
4182 	CU_ASSERT(g_lock_lba_range_done == false);
4183 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4184 	SPDK_CU_ASSERT_FATAL(range != NULL);
4185 	CU_ASSERT(range->offset == 25);
4186 	CU_ASSERT(range->length == 15);
4187 
4188 	/* Unlock 20-29.  This should result in range 25-39 now getting locked since it
4189 	 * no longer overlaps with an active lock.
4190 	 */
4191 	g_unlock_lba_range_done = false;
4192 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4193 	CU_ASSERT(rc == 0);
4194 	poll_threads();
4195 
4196 	CU_ASSERT(g_unlock_lba_range_done == true);
4197 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4198 	range = TAILQ_FIRST(&channel->locked_ranges);
4199 	SPDK_CU_ASSERT_FATAL(range != NULL);
4200 	CU_ASSERT(range->offset == 25);
4201 	CU_ASSERT(range->length == 15);
4202 
4203 	/* Lock 40-59.  This should immediately lock since it does not overlap with the
4204 	 * currently active 25-39 lock.
4205 	 */
4206 	g_lock_lba_range_done = false;
4207 	rc = bdev_lock_lba_range(desc, io_ch, 40, 20, lock_lba_range_done, &ctx1);
4208 	CU_ASSERT(rc == 0);
4209 	poll_threads();
4210 
4211 	CU_ASSERT(g_lock_lba_range_done == true);
4212 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4213 	SPDK_CU_ASSERT_FATAL(range != NULL);
4214 	range = TAILQ_NEXT(range, tailq);
4215 	SPDK_CU_ASSERT_FATAL(range != NULL);
4216 	CU_ASSERT(range->offset == 40);
4217 	CU_ASSERT(range->length == 20);
4218 
4219 	/* Try to lock 35-44.  Note that this overlaps with both 25-39 and 40-59. */
4220 	g_lock_lba_range_done = false;
4221 	rc = bdev_lock_lba_range(desc, io_ch, 35, 10, lock_lba_range_done, &ctx1);
4222 	CU_ASSERT(rc == 0);
4223 	poll_threads();
4224 
4225 	CU_ASSERT(g_lock_lba_range_done == false);
4226 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4227 	SPDK_CU_ASSERT_FATAL(range != NULL);
4228 	CU_ASSERT(range->offset == 35);
4229 	CU_ASSERT(range->length == 10);
4230 
4231 	/* Unlock 25-39.  Make sure that 35-44 is still in the pending list, since
4232 	 * the 40-59 lock is still active.
4233 	 */
4234 	g_unlock_lba_range_done = false;
4235 	rc = bdev_unlock_lba_range(desc, io_ch, 25, 15, unlock_lba_range_done, &ctx1);
4236 	CU_ASSERT(rc == 0);
4237 	poll_threads();
4238 
4239 	CU_ASSERT(g_unlock_lba_range_done == true);
4240 	CU_ASSERT(g_lock_lba_range_done == false);
4241 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4242 	SPDK_CU_ASSERT_FATAL(range != NULL);
4243 	CU_ASSERT(range->offset == 35);
4244 	CU_ASSERT(range->length == 10);
4245 
4246 	/* Unlock 40-59.  This should result in 35-44 now getting locked, since there are
4247 	 * no longer any active overlapping locks.
4248 	 */
4249 	g_unlock_lba_range_done = false;
4250 	rc = bdev_unlock_lba_range(desc, io_ch, 40, 20, unlock_lba_range_done, &ctx1);
4251 	CU_ASSERT(rc == 0);
4252 	poll_threads();
4253 
4254 	CU_ASSERT(g_unlock_lba_range_done == true);
4255 	CU_ASSERT(g_lock_lba_range_done == true);
4256 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4257 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4258 	SPDK_CU_ASSERT_FATAL(range != NULL);
4259 	CU_ASSERT(range->offset == 35);
4260 	CU_ASSERT(range->length == 10);
4261 
4262 	/* Finally, unlock 35-44. */
4263 	g_unlock_lba_range_done = false;
4264 	rc = bdev_unlock_lba_range(desc, io_ch, 35, 10, unlock_lba_range_done, &ctx1);
4265 	CU_ASSERT(rc == 0);
4266 	poll_threads();
4267 
4268 	CU_ASSERT(g_unlock_lba_range_done == true);
4269 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.locked_ranges));
4270 
4271 	spdk_put_io_channel(io_ch);
4272 	spdk_bdev_close(desc);
4273 	free_bdev(bdev);
4274 	spdk_bdev_finish(bdev_fini_cb, NULL);
4275 	poll_threads();
4276 }
4277 
4278 static void
4279 abort_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
4280 {
4281 	g_abort_done = true;
4282 	g_abort_status = bdev_io->internal.status;
4283 	spdk_bdev_free_io(bdev_io);
4284 }
4285 
4286 static void
4287 bdev_io_abort(void)
4288 {
4289 	struct spdk_bdev *bdev;
4290 	struct spdk_bdev_desc *desc = NULL;
4291 	struct spdk_io_channel *io_ch;
4292 	struct spdk_bdev_channel *channel;
4293 	struct spdk_bdev_mgmt_channel *mgmt_ch;
4294 	struct spdk_bdev_opts bdev_opts = {};
4295 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
4296 	uint64_t io_ctx1 = 0, io_ctx2 = 0, i;
4297 	int rc;
4298 
4299 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4300 	bdev_opts.bdev_io_pool_size = 7;
4301 	bdev_opts.bdev_io_cache_size = 2;
4302 
4303 	rc = spdk_bdev_set_opts(&bdev_opts);
4304 	CU_ASSERT(rc == 0);
4305 	spdk_bdev_initialize(bdev_init_cb, NULL);
4306 
4307 	bdev = allocate_bdev("bdev0");
4308 
4309 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4310 	CU_ASSERT(rc == 0);
4311 	CU_ASSERT(desc != NULL);
4312 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4313 	io_ch = spdk_bdev_get_io_channel(desc);
4314 	CU_ASSERT(io_ch != NULL);
4315 	channel = spdk_io_channel_get_ctx(io_ch);
4316 	mgmt_ch = channel->shared_resource->mgmt_ch;
4317 
4318 	g_abort_done = false;
4319 
4320 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, false);
4321 
4322 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4323 	CU_ASSERT(rc == -ENOTSUP);
4324 
4325 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, true);
4326 
4327 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx2, abort_done, NULL);
4328 	CU_ASSERT(rc == 0);
4329 	CU_ASSERT(g_abort_done == true);
4330 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_FAILED);
4331 
4332 	/* Test the case that the target I/O was successfully aborted. */
4333 	g_io_done = false;
4334 
4335 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4336 	CU_ASSERT(rc == 0);
4337 	CU_ASSERT(g_io_done == false);
4338 
4339 	g_abort_done = false;
4340 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4341 
4342 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4343 	CU_ASSERT(rc == 0);
4344 	CU_ASSERT(g_io_done == true);
4345 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4346 	stub_complete_io(1);
4347 	CU_ASSERT(g_abort_done == true);
4348 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4349 
4350 	/* Test the case that the target I/O was not aborted because it completed
4351 	 * in the middle of execution of the abort.
4352 	 */
4353 	g_io_done = false;
4354 
4355 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4356 	CU_ASSERT(rc == 0);
4357 	CU_ASSERT(g_io_done == false);
4358 
4359 	g_abort_done = false;
4360 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4361 
4362 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4363 	CU_ASSERT(rc == 0);
4364 	CU_ASSERT(g_io_done == false);
4365 
4366 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4367 	stub_complete_io(1);
4368 	CU_ASSERT(g_io_done == true);
4369 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4370 
4371 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4372 	stub_complete_io(1);
4373 	CU_ASSERT(g_abort_done == true);
4374 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4375 
4376 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4377 
4378 	bdev->optimal_io_boundary = 16;
4379 	bdev->split_on_optimal_io_boundary = true;
4380 
4381 	/* Test that a single-vector command which is split is aborted correctly.
4382 	 * Offset 14, length 8, payload 0xF000
4383 	 *  Child - Offset 14, length 2, payload 0xF000
4384 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
4385 	 */
4386 	g_io_done = false;
4387 
4388 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, &io_ctx1);
4389 	CU_ASSERT(rc == 0);
4390 	CU_ASSERT(g_io_done == false);
4391 
4392 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4393 
4394 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4395 
4396 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4397 	CU_ASSERT(rc == 0);
4398 	CU_ASSERT(g_io_done == true);
4399 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4400 	stub_complete_io(2);
4401 	CU_ASSERT(g_abort_done == true);
4402 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4403 
4404 	/* Test that a multi-vector command that needs to be split by strip and then
4405 	 * needs to be split is aborted correctly. Abort is requested before the second
4406 	 * child I/O was submitted. The parent I/O should complete with failure without
4407 	 * submitting the second child I/O.
4408 	 */
4409 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
4410 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
4411 		iov[i].iov_len = 512;
4412 	}
4413 
4414 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
4415 	g_io_done = false;
4416 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
4417 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, &io_ctx1);
4418 	CU_ASSERT(rc == 0);
4419 	CU_ASSERT(g_io_done == false);
4420 
4421 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4422 
4423 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4424 
4425 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4426 	CU_ASSERT(rc == 0);
4427 	CU_ASSERT(g_io_done == true);
4428 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4429 	stub_complete_io(1);
4430 	CU_ASSERT(g_abort_done == true);
4431 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4432 
4433 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4434 
4435 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4436 
4437 	bdev->optimal_io_boundary = 16;
4438 	g_io_done = false;
4439 
4440 	/* Test that a ingle-vector command which is split is aborted correctly.
4441 	 * Differently from the above, the child abort request will be submitted
4442 	 * sequentially due to the capacity of spdk_bdev_io.
4443 	 */
4444 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 50, io_done, &io_ctx1);
4445 	CU_ASSERT(rc == 0);
4446 	CU_ASSERT(g_io_done == false);
4447 
4448 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4449 
4450 	g_abort_done = false;
4451 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4452 
4453 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4454 	CU_ASSERT(rc == 0);
4455 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
4456 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4457 
4458 	stub_complete_io(1);
4459 	CU_ASSERT(g_io_done == true);
4460 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4461 	stub_complete_io(3);
4462 	CU_ASSERT(g_abort_done == true);
4463 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4464 
4465 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4466 
4467 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4468 
4469 	spdk_put_io_channel(io_ch);
4470 	spdk_bdev_close(desc);
4471 	free_bdev(bdev);
4472 	spdk_bdev_finish(bdev_fini_cb, NULL);
4473 	poll_threads();
4474 }
4475 
4476 static void
4477 bdev_unmap(void)
4478 {
4479 	struct spdk_bdev *bdev;
4480 	struct spdk_bdev_desc *desc = NULL;
4481 	struct spdk_io_channel *ioch;
4482 	struct spdk_bdev_channel *bdev_ch;
4483 	struct ut_expected_io *expected_io;
4484 	struct spdk_bdev_opts bdev_opts = {};
4485 	uint32_t i, num_outstanding;
4486 	uint64_t offset, num_blocks, max_unmap_blocks, num_children;
4487 	int rc;
4488 
4489 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4490 	bdev_opts.bdev_io_pool_size = 512;
4491 	bdev_opts.bdev_io_cache_size = 64;
4492 	rc = spdk_bdev_set_opts(&bdev_opts);
4493 	CU_ASSERT(rc == 0);
4494 
4495 	spdk_bdev_initialize(bdev_init_cb, NULL);
4496 	bdev = allocate_bdev("bdev");
4497 
4498 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4499 	CU_ASSERT_EQUAL(rc, 0);
4500 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4501 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4502 	ioch = spdk_bdev_get_io_channel(desc);
4503 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4504 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4505 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4506 
4507 	fn_table.submit_request = stub_submit_request;
4508 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4509 
4510 	/* Case 1: First test the request won't be split */
4511 	num_blocks = 32;
4512 
4513 	g_io_done = false;
4514 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 0, num_blocks, 0);
4515 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4516 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4517 	CU_ASSERT_EQUAL(rc, 0);
4518 	CU_ASSERT(g_io_done == false);
4519 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4520 	stub_complete_io(1);
4521 	CU_ASSERT(g_io_done == true);
4522 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4523 
4524 	/* Case 2: Test the split with 2 children requests */
4525 	bdev->max_unmap = 8;
4526 	bdev->max_unmap_segments = 2;
4527 	max_unmap_blocks = bdev->max_unmap * bdev->max_unmap_segments;
4528 	num_blocks = max_unmap_blocks * 2;
4529 	offset = 0;
4530 
4531 	g_io_done = false;
4532 	for (i = 0; i < 2; i++) {
4533 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4534 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4535 		offset += max_unmap_blocks;
4536 	}
4537 
4538 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4539 	CU_ASSERT_EQUAL(rc, 0);
4540 	CU_ASSERT(g_io_done == false);
4541 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4542 	stub_complete_io(2);
4543 	CU_ASSERT(g_io_done == true);
4544 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4545 
4546 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4547 	num_children = 15;
4548 	num_blocks = max_unmap_blocks * num_children;
4549 	g_io_done = false;
4550 	offset = 0;
4551 	for (i = 0; i < num_children; i++) {
4552 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4553 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4554 		offset += max_unmap_blocks;
4555 	}
4556 
4557 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4558 	CU_ASSERT_EQUAL(rc, 0);
4559 	CU_ASSERT(g_io_done == false);
4560 
4561 	while (num_children > 0) {
4562 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4563 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4564 		stub_complete_io(num_outstanding);
4565 		num_children -= num_outstanding;
4566 	}
4567 	CU_ASSERT(g_io_done == true);
4568 
4569 	spdk_put_io_channel(ioch);
4570 	spdk_bdev_close(desc);
4571 	free_bdev(bdev);
4572 	spdk_bdev_finish(bdev_fini_cb, NULL);
4573 	poll_threads();
4574 }
4575 
4576 static void
4577 bdev_write_zeroes_split_test(void)
4578 {
4579 	struct spdk_bdev *bdev;
4580 	struct spdk_bdev_desc *desc = NULL;
4581 	struct spdk_io_channel *ioch;
4582 	struct spdk_bdev_channel *bdev_ch;
4583 	struct ut_expected_io *expected_io;
4584 	struct spdk_bdev_opts bdev_opts = {};
4585 	uint32_t i, num_outstanding;
4586 	uint64_t offset, num_blocks, max_write_zeroes_blocks, num_children;
4587 	int rc;
4588 
4589 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4590 	bdev_opts.bdev_io_pool_size = 512;
4591 	bdev_opts.bdev_io_cache_size = 64;
4592 	rc = spdk_bdev_set_opts(&bdev_opts);
4593 	CU_ASSERT(rc == 0);
4594 
4595 	spdk_bdev_initialize(bdev_init_cb, NULL);
4596 	bdev = allocate_bdev("bdev");
4597 
4598 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4599 	CU_ASSERT_EQUAL(rc, 0);
4600 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4601 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4602 	ioch = spdk_bdev_get_io_channel(desc);
4603 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4604 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4605 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4606 
4607 	fn_table.submit_request = stub_submit_request;
4608 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4609 
4610 	/* Case 1: First test the request won't be split */
4611 	num_blocks = 32;
4612 
4613 	g_io_done = false;
4614 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
4615 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4616 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4617 	CU_ASSERT_EQUAL(rc, 0);
4618 	CU_ASSERT(g_io_done == false);
4619 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4620 	stub_complete_io(1);
4621 	CU_ASSERT(g_io_done == true);
4622 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4623 
4624 	/* Case 2: Test the split with 2 children requests */
4625 	max_write_zeroes_blocks = 8;
4626 	bdev->max_write_zeroes = max_write_zeroes_blocks;
4627 	num_blocks = max_write_zeroes_blocks * 2;
4628 	offset = 0;
4629 
4630 	g_io_done = false;
4631 	for (i = 0; i < 2; i++) {
4632 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4633 						   0);
4634 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4635 		offset += max_write_zeroes_blocks;
4636 	}
4637 
4638 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4639 	CU_ASSERT_EQUAL(rc, 0);
4640 	CU_ASSERT(g_io_done == false);
4641 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4642 	stub_complete_io(2);
4643 	CU_ASSERT(g_io_done == true);
4644 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4645 
4646 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4647 	num_children = 15;
4648 	num_blocks = max_write_zeroes_blocks * num_children;
4649 	g_io_done = false;
4650 	offset = 0;
4651 	for (i = 0; i < num_children; i++) {
4652 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4653 						   0);
4654 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4655 		offset += max_write_zeroes_blocks;
4656 	}
4657 
4658 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4659 	CU_ASSERT_EQUAL(rc, 0);
4660 	CU_ASSERT(g_io_done == false);
4661 
4662 	while (num_children > 0) {
4663 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4664 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4665 		stub_complete_io(num_outstanding);
4666 		num_children -= num_outstanding;
4667 	}
4668 	CU_ASSERT(g_io_done == true);
4669 
4670 	spdk_put_io_channel(ioch);
4671 	spdk_bdev_close(desc);
4672 	free_bdev(bdev);
4673 	spdk_bdev_finish(bdev_fini_cb, NULL);
4674 	poll_threads();
4675 }
4676 
4677 static void
4678 bdev_set_options_test(void)
4679 {
4680 	struct spdk_bdev_opts bdev_opts = {};
4681 	int rc;
4682 
4683 	/* Case1: Do not set opts_size */
4684 	rc = spdk_bdev_set_opts(&bdev_opts);
4685 	CU_ASSERT(rc == -1);
4686 
4687 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4688 	bdev_opts.bdev_io_pool_size = 4;
4689 	bdev_opts.bdev_io_cache_size = 2;
4690 	bdev_opts.small_buf_pool_size = 4;
4691 
4692 	/* Case 2: Do not set valid small_buf_pool_size and large_buf_pool_size */
4693 	rc = spdk_bdev_set_opts(&bdev_opts);
4694 	CU_ASSERT(rc == -1);
4695 
4696 	/* Case 3: Do not set valid large_buf_pool_size */
4697 	bdev_opts.small_buf_pool_size = BUF_SMALL_POOL_SIZE;
4698 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE - 1;
4699 	rc = spdk_bdev_set_opts(&bdev_opts);
4700 	CU_ASSERT(rc == -1);
4701 
4702 	/* Case4: set valid large buf_pool_size */
4703 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE;
4704 	rc = spdk_bdev_set_opts(&bdev_opts);
4705 	CU_ASSERT(rc == 0);
4706 
4707 	/* Case5: Set different valid value for small and large buf pool */
4708 	bdev_opts.large_buf_pool_size = BUF_SMALL_POOL_SIZE + 3;
4709 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE + 3;
4710 	rc = spdk_bdev_set_opts(&bdev_opts);
4711 	CU_ASSERT(rc == 0);
4712 }
4713 
4714 static uint64_t
4715 get_ns_time(void)
4716 {
4717 	int rc;
4718 	struct timespec ts;
4719 
4720 	rc = clock_gettime(CLOCK_MONOTONIC, &ts);
4721 	CU_ASSERT(rc == 0);
4722 	return ts.tv_sec * 1000 * 1000 * 1000 + ts.tv_nsec;
4723 }
4724 
4725 static int
4726 rb_tree_get_height(struct spdk_bdev_name *bdev_name)
4727 {
4728 	int h1, h2;
4729 
4730 	if (bdev_name == NULL) {
4731 		return -1;
4732 	} else {
4733 		h1 = rb_tree_get_height(RB_LEFT(bdev_name, node));
4734 		h2 = rb_tree_get_height(RB_RIGHT(bdev_name, node));
4735 
4736 		return spdk_max(h1, h2) + 1;
4737 	}
4738 }
4739 
4740 static void
4741 bdev_multi_allocation(void)
4742 {
4743 	const int max_bdev_num = 1024 * 16;
4744 	char name[max_bdev_num][10];
4745 	char noexist_name[] = "invalid_bdev";
4746 	struct spdk_bdev *bdev[max_bdev_num];
4747 	int i, j;
4748 	uint64_t last_time;
4749 	int bdev_num;
4750 	int height;
4751 
4752 	for (j = 0; j < max_bdev_num; j++) {
4753 		snprintf(name[j], sizeof(name[j]), "bdev%d", j);
4754 	}
4755 
4756 	for (i = 0; i < 16; i++) {
4757 		last_time = get_ns_time();
4758 		bdev_num = 1024 * (i + 1);
4759 		for (j = 0; j < bdev_num; j++) {
4760 			bdev[j] = allocate_bdev(name[j]);
4761 			height = rb_tree_get_height(&bdev[j]->internal.bdev_name);
4762 			CU_ASSERT(height <= (int)(spdk_u32log2(j + 1)));
4763 		}
4764 		SPDK_NOTICELOG("alloc bdev num %d takes %" PRIu64 " ms\n", bdev_num,
4765 			       (get_ns_time() - last_time) / 1000 / 1000);
4766 		for (j = 0; j < bdev_num; j++) {
4767 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) != NULL);
4768 		}
4769 		CU_ASSERT(spdk_bdev_get_by_name(noexist_name) == NULL);
4770 
4771 		for (j = 0; j < bdev_num; j++) {
4772 			free_bdev(bdev[j]);
4773 		}
4774 		for (j = 0; j < bdev_num; j++) {
4775 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) == NULL);
4776 		}
4777 	}
4778 }
4779 
4780 static struct spdk_memory_domain *g_bdev_memory_domain = (struct spdk_memory_domain *) 0xf00df00d;
4781 
4782 static int
4783 test_bdev_get_supported_dma_device_types_op(void *ctx, struct spdk_memory_domain **domains,
4784 		int array_size)
4785 {
4786 	if (array_size > 0 && domains) {
4787 		domains[0] = g_bdev_memory_domain;
4788 	}
4789 
4790 	return 1;
4791 }
4792 
4793 static void
4794 bdev_get_memory_domains(void)
4795 {
4796 	struct spdk_bdev_fn_table fn_table = {
4797 		.get_memory_domains = test_bdev_get_supported_dma_device_types_op
4798 	};
4799 	struct spdk_bdev bdev = { .fn_table = &fn_table };
4800 	struct spdk_memory_domain *domains[2] = {};
4801 	int rc;
4802 
4803 	/* bdev is NULL */
4804 	rc = spdk_bdev_get_memory_domains(NULL, domains, 2);
4805 	CU_ASSERT(rc == -EINVAL);
4806 
4807 	/* domains is NULL */
4808 	rc = spdk_bdev_get_memory_domains(&bdev, NULL, 2);
4809 	CU_ASSERT(rc == 1);
4810 
4811 	/* array size is 0 */
4812 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 0);
4813 	CU_ASSERT(rc == 1);
4814 
4815 	/* get_supported_dma_device_types op is set */
4816 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4817 	CU_ASSERT(rc == 1);
4818 	CU_ASSERT(domains[0] == g_bdev_memory_domain);
4819 
4820 	/* get_supported_dma_device_types op is not set */
4821 	fn_table.get_memory_domains = NULL;
4822 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4823 	CU_ASSERT(rc == 0);
4824 }
4825 
4826 static void
4827 bdev_writev_readv_ext(void)
4828 {
4829 	struct spdk_bdev *bdev;
4830 	struct spdk_bdev_desc *desc = NULL;
4831 	struct spdk_io_channel *io_ch;
4832 	struct iovec iov = { .iov_base = (void *)0xbaaddead, .iov_len = 0x1000 };
4833 	struct ut_expected_io *expected_io;
4834 	struct spdk_bdev_ext_io_opts ext_io_opts = {
4835 		.metadata = (void *)0xFF000000
4836 	};
4837 	int rc;
4838 
4839 	spdk_bdev_initialize(bdev_init_cb, NULL);
4840 
4841 	bdev = allocate_bdev("bdev0");
4842 	bdev->md_interleave = false;
4843 	bdev->md_len = 8;
4844 
4845 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4846 	CU_ASSERT(rc == 0);
4847 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4848 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4849 	io_ch = spdk_bdev_get_io_channel(desc);
4850 	CU_ASSERT(io_ch != NULL);
4851 
4852 	g_io_done = false;
4853 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 32, 14, 1);
4854 	expected_io->md_buf = ext_io_opts.metadata;
4855 	expected_io->ext_io_opts = &ext_io_opts;
4856 	ut_expected_io_set_iov(expected_io, 0, iov.iov_base, iov.iov_len);
4857 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4858 
4859 	rc = spdk_bdev_readv_blocks_ext(desc, io_ch, &iov, 1, 32, 14, io_done, NULL, &ext_io_opts);
4860 
4861 	CU_ASSERT(rc == 0);
4862 	CU_ASSERT(g_io_done == false);
4863 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4864 	stub_complete_io(1);
4865 	CU_ASSERT(g_io_done == true);
4866 
4867 	g_io_done = false;
4868 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 1);
4869 	expected_io->md_buf = ext_io_opts.metadata;
4870 	expected_io->ext_io_opts = &ext_io_opts;
4871 	ut_expected_io_set_iov(expected_io, 0, iov.iov_base, iov.iov_len);
4872 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4873 
4874 	rc = spdk_bdev_writev_blocks_ext(desc, io_ch, &iov, 1, 32, 14, io_done, NULL, &ext_io_opts);
4875 
4876 	CU_ASSERT(rc == 0);
4877 	CU_ASSERT(g_io_done == false);
4878 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4879 	stub_complete_io(1);
4880 	CU_ASSERT(g_io_done == true);
4881 
4882 	spdk_put_io_channel(io_ch);
4883 	spdk_bdev_close(desc);
4884 	free_bdev(bdev);
4885 	spdk_bdev_finish(bdev_fini_cb, NULL);
4886 	poll_threads();
4887 }
4888 
4889 int
4890 main(int argc, char **argv)
4891 {
4892 	CU_pSuite		suite = NULL;
4893 	unsigned int		num_failures;
4894 
4895 	CU_set_error_action(CUEA_ABORT);
4896 	CU_initialize_registry();
4897 
4898 	suite = CU_add_suite("bdev", null_init, null_clean);
4899 
4900 	CU_ADD_TEST(suite, bytes_to_blocks_test);
4901 	CU_ADD_TEST(suite, num_blocks_test);
4902 	CU_ADD_TEST(suite, io_valid_test);
4903 	CU_ADD_TEST(suite, open_write_test);
4904 	CU_ADD_TEST(suite, alias_add_del_test);
4905 	CU_ADD_TEST(suite, get_device_stat_test);
4906 	CU_ADD_TEST(suite, bdev_io_types_test);
4907 	CU_ADD_TEST(suite, bdev_io_wait_test);
4908 	CU_ADD_TEST(suite, bdev_io_spans_split_test);
4909 	CU_ADD_TEST(suite, bdev_io_boundary_split_test);
4910 	CU_ADD_TEST(suite, bdev_io_max_size_and_segment_split_test);
4911 	CU_ADD_TEST(suite, bdev_io_mix_split_test);
4912 	CU_ADD_TEST(suite, bdev_io_split_with_io_wait);
4913 	CU_ADD_TEST(suite, bdev_io_alignment_with_boundary);
4914 	CU_ADD_TEST(suite, bdev_io_alignment);
4915 	CU_ADD_TEST(suite, bdev_histograms);
4916 	CU_ADD_TEST(suite, bdev_write_zeroes);
4917 	CU_ADD_TEST(suite, bdev_compare_and_write);
4918 	CU_ADD_TEST(suite, bdev_compare);
4919 	CU_ADD_TEST(suite, bdev_zcopy_write);
4920 	CU_ADD_TEST(suite, bdev_zcopy_read);
4921 	CU_ADD_TEST(suite, bdev_open_while_hotremove);
4922 	CU_ADD_TEST(suite, bdev_close_while_hotremove);
4923 	CU_ADD_TEST(suite, bdev_open_ext);
4924 	CU_ADD_TEST(suite, bdev_set_io_timeout);
4925 	CU_ADD_TEST(suite, lba_range_overlap);
4926 	CU_ADD_TEST(suite, lock_lba_range_check_ranges);
4927 	CU_ADD_TEST(suite, lock_lba_range_with_io_outstanding);
4928 	CU_ADD_TEST(suite, lock_lba_range_overlapped);
4929 	CU_ADD_TEST(suite, bdev_io_abort);
4930 	CU_ADD_TEST(suite, bdev_unmap);
4931 	CU_ADD_TEST(suite, bdev_write_zeroes_split_test);
4932 	CU_ADD_TEST(suite, bdev_set_options_test);
4933 	CU_ADD_TEST(suite, bdev_multi_allocation);
4934 	CU_ADD_TEST(suite, bdev_get_memory_domains);
4935 	CU_ADD_TEST(suite, bdev_writev_readv_ext);
4936 
4937 	allocate_cores(1);
4938 	allocate_threads(1);
4939 	set_thread(0);
4940 
4941 	CU_basic_set_mode(CU_BRM_VERBOSE);
4942 	CU_basic_run_tests();
4943 	num_failures = CU_get_number_of_failures();
4944 	CU_cleanup_registry();
4945 
4946 	free_threads();
4947 	free_cores();
4948 
4949 	return num_failures;
4950 }
4951