xref: /spdk/test/unit/lib/bdev/bdev.c/bdev_ut.c (revision c3a5848966ab086e75cf67833d82d7278cc92ad6)
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 	TAILQ_ENTRY(ut_expected_io)	link;
89 };
90 
91 struct bdev_ut_channel {
92 	TAILQ_HEAD(, spdk_bdev_io)	outstanding_io;
93 	uint32_t			outstanding_io_count;
94 	TAILQ_HEAD(, ut_expected_io)	expected_io;
95 };
96 
97 static bool g_io_done;
98 static struct spdk_bdev_io *g_bdev_io;
99 static enum spdk_bdev_io_status g_io_status;
100 static enum spdk_bdev_io_status g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
101 static uint32_t g_bdev_ut_io_device;
102 static struct bdev_ut_channel *g_bdev_ut_channel;
103 static void *g_compare_read_buf;
104 static uint32_t g_compare_read_buf_len;
105 static void *g_compare_write_buf;
106 static uint32_t g_compare_write_buf_len;
107 static bool g_abort_done;
108 static enum spdk_bdev_io_status g_abort_status;
109 static void *g_zcopy_read_buf;
110 static uint32_t g_zcopy_read_buf_len;
111 static void *g_zcopy_write_buf;
112 static uint32_t g_zcopy_write_buf_len;
113 static struct spdk_bdev_io *g_zcopy_bdev_io;
114 
115 static struct ut_expected_io *
116 ut_alloc_expected_io(uint8_t type, uint64_t offset, uint64_t length, int iovcnt)
117 {
118 	struct ut_expected_io *expected_io;
119 
120 	expected_io = calloc(1, sizeof(*expected_io));
121 	SPDK_CU_ASSERT_FATAL(expected_io != NULL);
122 
123 	expected_io->type = type;
124 	expected_io->offset = offset;
125 	expected_io->length = length;
126 	expected_io->iovcnt = iovcnt;
127 
128 	return expected_io;
129 }
130 
131 static void
132 ut_expected_io_set_iov(struct ut_expected_io *expected_io, int pos, void *base, size_t len)
133 {
134 	expected_io->iov[pos].iov_base = base;
135 	expected_io->iov[pos].iov_len = len;
136 }
137 
138 static void
139 stub_submit_request(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
140 {
141 	struct bdev_ut_channel *ch = spdk_io_channel_get_ctx(_ch);
142 	struct ut_expected_io *expected_io;
143 	struct iovec *iov, *expected_iov;
144 	struct spdk_bdev_io *bio_to_abort;
145 	int i;
146 
147 	g_bdev_io = bdev_io;
148 
149 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
150 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
151 
152 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
153 		CU_ASSERT(g_compare_read_buf_len == len);
154 		memcpy(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len);
155 	}
156 
157 	if (g_compare_write_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
158 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
159 
160 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
161 		CU_ASSERT(g_compare_write_buf_len == len);
162 		memcpy(g_compare_write_buf, bdev_io->u.bdev.iovs[0].iov_base, len);
163 	}
164 
165 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_COMPARE) {
166 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
167 
168 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
169 		CU_ASSERT(g_compare_read_buf_len == len);
170 		if (memcmp(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len)) {
171 			g_io_exp_status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
172 		}
173 	}
174 
175 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) {
176 		if (g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS) {
177 			TAILQ_FOREACH(bio_to_abort, &ch->outstanding_io, module_link) {
178 				if (bio_to_abort == bdev_io->u.abort.bio_to_abort) {
179 					TAILQ_REMOVE(&ch->outstanding_io, bio_to_abort, module_link);
180 					ch->outstanding_io_count--;
181 					spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_FAILED);
182 					break;
183 				}
184 			}
185 		}
186 	}
187 
188 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) {
189 		if (bdev_io->u.bdev.zcopy.start) {
190 			g_zcopy_bdev_io = bdev_io;
191 			if (bdev_io->u.bdev.zcopy.populate) {
192 				/* Start of a read */
193 				CU_ASSERT(g_zcopy_read_buf != NULL);
194 				CU_ASSERT(g_zcopy_read_buf_len > 0);
195 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_read_buf;
196 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_read_buf_len;
197 				bdev_io->u.bdev.iovcnt = 1;
198 			} else {
199 				/* Start of a write */
200 				CU_ASSERT(g_zcopy_write_buf != NULL);
201 				CU_ASSERT(g_zcopy_write_buf_len > 0);
202 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_write_buf;
203 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_write_buf_len;
204 				bdev_io->u.bdev.iovcnt = 1;
205 			}
206 		} else {
207 			if (bdev_io->u.bdev.zcopy.commit) {
208 				/* End of write */
209 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_write_buf);
210 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_write_buf_len);
211 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
212 				g_zcopy_write_buf = NULL;
213 				g_zcopy_write_buf_len = 0;
214 			} else {
215 				/* End of read */
216 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_read_buf);
217 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_read_buf_len);
218 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
219 				g_zcopy_read_buf = NULL;
220 				g_zcopy_read_buf_len = 0;
221 			}
222 		}
223 	}
224 
225 	TAILQ_INSERT_TAIL(&ch->outstanding_io, bdev_io, module_link);
226 	ch->outstanding_io_count++;
227 
228 	expected_io = TAILQ_FIRST(&ch->expected_io);
229 	if (expected_io == NULL) {
230 		return;
231 	}
232 	TAILQ_REMOVE(&ch->expected_io, expected_io, link);
233 
234 	if (expected_io->type != SPDK_BDEV_IO_TYPE_INVALID) {
235 		CU_ASSERT(bdev_io->type == expected_io->type);
236 	}
237 
238 	if (expected_io->md_buf != NULL) {
239 		CU_ASSERT(expected_io->md_buf == bdev_io->u.bdev.md_buf);
240 	}
241 
242 	if (expected_io->length == 0) {
243 		free(expected_io);
244 		return;
245 	}
246 
247 	CU_ASSERT(expected_io->offset == bdev_io->u.bdev.offset_blocks);
248 	CU_ASSERT(expected_io->length = bdev_io->u.bdev.num_blocks);
249 
250 	if (expected_io->iovcnt == 0) {
251 		free(expected_io);
252 		/* UNMAP, WRITE_ZEROES and FLUSH don't have iovs, so we can just return now. */
253 		return;
254 	}
255 
256 	CU_ASSERT(expected_io->iovcnt == bdev_io->u.bdev.iovcnt);
257 	for (i = 0; i < expected_io->iovcnt; i++) {
258 		iov = &bdev_io->u.bdev.iovs[i];
259 		expected_iov = &expected_io->iov[i];
260 		CU_ASSERT(iov->iov_len == expected_iov->iov_len);
261 		CU_ASSERT(iov->iov_base == expected_iov->iov_base);
262 	}
263 
264 	free(expected_io);
265 }
266 
267 static void
268 stub_submit_request_get_buf_cb(struct spdk_io_channel *_ch,
269 			       struct spdk_bdev_io *bdev_io, bool success)
270 {
271 	CU_ASSERT(success == true);
272 
273 	stub_submit_request(_ch, bdev_io);
274 }
275 
276 static void
277 stub_submit_request_get_buf(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
278 {
279 	spdk_bdev_io_get_buf(bdev_io, stub_submit_request_get_buf_cb,
280 			     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
281 }
282 
283 static uint32_t
284 stub_complete_io(uint32_t num_to_complete)
285 {
286 	struct bdev_ut_channel *ch = g_bdev_ut_channel;
287 	struct spdk_bdev_io *bdev_io;
288 	static enum spdk_bdev_io_status io_status;
289 	uint32_t num_completed = 0;
290 
291 	while (num_completed < num_to_complete) {
292 		if (TAILQ_EMPTY(&ch->outstanding_io)) {
293 			break;
294 		}
295 		bdev_io = TAILQ_FIRST(&ch->outstanding_io);
296 		TAILQ_REMOVE(&ch->outstanding_io, bdev_io, module_link);
297 		ch->outstanding_io_count--;
298 		io_status = g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS ? SPDK_BDEV_IO_STATUS_SUCCESS :
299 			    g_io_exp_status;
300 		spdk_bdev_io_complete(bdev_io, io_status);
301 		num_completed++;
302 	}
303 
304 	return num_completed;
305 }
306 
307 static struct spdk_io_channel *
308 bdev_ut_get_io_channel(void *ctx)
309 {
310 	return spdk_get_io_channel(&g_bdev_ut_io_device);
311 }
312 
313 static bool g_io_types_supported[SPDK_BDEV_NUM_IO_TYPES] = {
314 	[SPDK_BDEV_IO_TYPE_READ]		= true,
315 	[SPDK_BDEV_IO_TYPE_WRITE]		= true,
316 	[SPDK_BDEV_IO_TYPE_COMPARE]		= true,
317 	[SPDK_BDEV_IO_TYPE_UNMAP]		= true,
318 	[SPDK_BDEV_IO_TYPE_FLUSH]		= true,
319 	[SPDK_BDEV_IO_TYPE_RESET]		= true,
320 	[SPDK_BDEV_IO_TYPE_NVME_ADMIN]		= true,
321 	[SPDK_BDEV_IO_TYPE_NVME_IO]		= true,
322 	[SPDK_BDEV_IO_TYPE_NVME_IO_MD]		= true,
323 	[SPDK_BDEV_IO_TYPE_WRITE_ZEROES]	= true,
324 	[SPDK_BDEV_IO_TYPE_ZCOPY]		= true,
325 	[SPDK_BDEV_IO_TYPE_ABORT]		= true,
326 };
327 
328 static void
329 ut_enable_io_type(enum spdk_bdev_io_type io_type, bool enable)
330 {
331 	g_io_types_supported[io_type] = enable;
332 }
333 
334 static bool
335 stub_io_type_supported(void *_bdev, enum spdk_bdev_io_type io_type)
336 {
337 	return g_io_types_supported[io_type];
338 }
339 
340 static struct spdk_bdev_fn_table fn_table = {
341 	.destruct = stub_destruct,
342 	.submit_request = stub_submit_request,
343 	.get_io_channel = bdev_ut_get_io_channel,
344 	.io_type_supported = stub_io_type_supported,
345 };
346 
347 static int
348 bdev_ut_create_ch(void *io_device, void *ctx_buf)
349 {
350 	struct bdev_ut_channel *ch = ctx_buf;
351 
352 	CU_ASSERT(g_bdev_ut_channel == NULL);
353 	g_bdev_ut_channel = ch;
354 
355 	TAILQ_INIT(&ch->outstanding_io);
356 	ch->outstanding_io_count = 0;
357 	TAILQ_INIT(&ch->expected_io);
358 	return 0;
359 }
360 
361 static void
362 bdev_ut_destroy_ch(void *io_device, void *ctx_buf)
363 {
364 	CU_ASSERT(g_bdev_ut_channel != NULL);
365 	g_bdev_ut_channel = NULL;
366 }
367 
368 struct spdk_bdev_module bdev_ut_if;
369 
370 static int
371 bdev_ut_module_init(void)
372 {
373 	spdk_io_device_register(&g_bdev_ut_io_device, bdev_ut_create_ch, bdev_ut_destroy_ch,
374 				sizeof(struct bdev_ut_channel), NULL);
375 	spdk_bdev_module_init_done(&bdev_ut_if);
376 	return 0;
377 }
378 
379 static void
380 bdev_ut_module_fini(void)
381 {
382 	spdk_io_device_unregister(&g_bdev_ut_io_device, NULL);
383 }
384 
385 struct spdk_bdev_module bdev_ut_if = {
386 	.name = "bdev_ut",
387 	.module_init = bdev_ut_module_init,
388 	.module_fini = bdev_ut_module_fini,
389 	.async_init = true,
390 };
391 
392 static void vbdev_ut_examine(struct spdk_bdev *bdev);
393 
394 static int
395 vbdev_ut_module_init(void)
396 {
397 	return 0;
398 }
399 
400 static void
401 vbdev_ut_module_fini(void)
402 {
403 }
404 
405 struct spdk_bdev_module vbdev_ut_if = {
406 	.name = "vbdev_ut",
407 	.module_init = vbdev_ut_module_init,
408 	.module_fini = vbdev_ut_module_fini,
409 	.examine_config = vbdev_ut_examine,
410 };
411 
412 SPDK_BDEV_MODULE_REGISTER(bdev_ut, &bdev_ut_if)
413 SPDK_BDEV_MODULE_REGISTER(vbdev_ut, &vbdev_ut_if)
414 
415 static void
416 vbdev_ut_examine(struct spdk_bdev *bdev)
417 {
418 	spdk_bdev_module_examine_done(&vbdev_ut_if);
419 }
420 
421 static struct spdk_bdev *
422 allocate_bdev(char *name)
423 {
424 	struct spdk_bdev *bdev;
425 	int rc;
426 
427 	bdev = calloc(1, sizeof(*bdev));
428 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
429 
430 	bdev->name = name;
431 	bdev->fn_table = &fn_table;
432 	bdev->module = &bdev_ut_if;
433 	bdev->blockcnt = 1024;
434 	bdev->blocklen = 512;
435 
436 	rc = spdk_bdev_register(bdev);
437 	CU_ASSERT(rc == 0);
438 
439 	return bdev;
440 }
441 
442 static struct spdk_bdev *
443 allocate_vbdev(char *name)
444 {
445 	struct spdk_bdev *bdev;
446 	int rc;
447 
448 	bdev = calloc(1, sizeof(*bdev));
449 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
450 
451 	bdev->name = name;
452 	bdev->fn_table = &fn_table;
453 	bdev->module = &vbdev_ut_if;
454 
455 	rc = spdk_bdev_register(bdev);
456 	CU_ASSERT(rc == 0);
457 
458 	return bdev;
459 }
460 
461 static void
462 free_bdev(struct spdk_bdev *bdev)
463 {
464 	spdk_bdev_unregister(bdev, NULL, NULL);
465 	poll_threads();
466 	memset(bdev, 0xFF, sizeof(*bdev));
467 	free(bdev);
468 }
469 
470 static void
471 free_vbdev(struct spdk_bdev *bdev)
472 {
473 	spdk_bdev_unregister(bdev, NULL, NULL);
474 	poll_threads();
475 	memset(bdev, 0xFF, sizeof(*bdev));
476 	free(bdev);
477 }
478 
479 static void
480 get_device_stat_cb(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat, void *cb_arg, int rc)
481 {
482 	const char *bdev_name;
483 
484 	CU_ASSERT(bdev != NULL);
485 	CU_ASSERT(rc == 0);
486 	bdev_name = spdk_bdev_get_name(bdev);
487 	CU_ASSERT_STRING_EQUAL(bdev_name, "bdev0");
488 
489 	free(stat);
490 
491 	*(bool *)cb_arg = true;
492 }
493 
494 static void
495 bdev_unregister_cb(void *cb_arg, int rc)
496 {
497 	g_unregister_arg = cb_arg;
498 	g_unregister_rc = rc;
499 }
500 
501 static void
502 bdev_ut_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
503 {
504 }
505 
506 static void
507 bdev_open_cb1(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
508 {
509 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
510 
511 	g_event_type1 = type;
512 	if (SPDK_BDEV_EVENT_REMOVE == type) {
513 		spdk_bdev_close(desc);
514 	}
515 }
516 
517 static void
518 bdev_open_cb2(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
519 {
520 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
521 
522 	g_event_type2 = type;
523 	if (SPDK_BDEV_EVENT_REMOVE == type) {
524 		spdk_bdev_close(desc);
525 	}
526 }
527 
528 static void
529 get_device_stat_test(void)
530 {
531 	struct spdk_bdev *bdev;
532 	struct spdk_bdev_io_stat *stat;
533 	bool done;
534 
535 	bdev = allocate_bdev("bdev0");
536 	stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
537 	if (stat == NULL) {
538 		free_bdev(bdev);
539 		return;
540 	}
541 
542 	done = false;
543 	spdk_bdev_get_device_stat(bdev, stat, get_device_stat_cb, &done);
544 	while (!done) { poll_threads(); }
545 
546 	free_bdev(bdev);
547 }
548 
549 static void
550 open_write_test(void)
551 {
552 	struct spdk_bdev *bdev[9];
553 	struct spdk_bdev_desc *desc[9] = {};
554 	int rc;
555 
556 	/*
557 	 * Create a tree of bdevs to test various open w/ write cases.
558 	 *
559 	 * bdev0 through bdev3 are physical block devices, such as NVMe
560 	 * namespaces or Ceph block devices.
561 	 *
562 	 * bdev4 is a virtual bdev with multiple base bdevs.  This models
563 	 * caching or RAID use cases.
564 	 *
565 	 * bdev5 through bdev7 are all virtual bdevs with the same base
566 	 * bdev (except bdev7). This models partitioning or logical volume
567 	 * use cases.
568 	 *
569 	 * bdev7 is a virtual bdev with multiple base bdevs. One of base bdevs
570 	 * (bdev2) is shared with other virtual bdevs: bdev5 and bdev6. This
571 	 * models caching, RAID, partitioning or logical volumes use cases.
572 	 *
573 	 * bdev8 is a virtual bdev with multiple base bdevs, but these
574 	 * base bdevs are themselves virtual bdevs.
575 	 *
576 	 *                bdev8
577 	 *                  |
578 	 *            +----------+
579 	 *            |          |
580 	 *          bdev4      bdev5   bdev6   bdev7
581 	 *            |          |       |       |
582 	 *        +---+---+      +---+   +   +---+---+
583 	 *        |       |           \  |  /         \
584 	 *      bdev0   bdev1          bdev2         bdev3
585 	 */
586 
587 	bdev[0] = allocate_bdev("bdev0");
588 	rc = spdk_bdev_module_claim_bdev(bdev[0], NULL, &bdev_ut_if);
589 	CU_ASSERT(rc == 0);
590 
591 	bdev[1] = allocate_bdev("bdev1");
592 	rc = spdk_bdev_module_claim_bdev(bdev[1], NULL, &bdev_ut_if);
593 	CU_ASSERT(rc == 0);
594 
595 	bdev[2] = allocate_bdev("bdev2");
596 	rc = spdk_bdev_module_claim_bdev(bdev[2], NULL, &bdev_ut_if);
597 	CU_ASSERT(rc == 0);
598 
599 	bdev[3] = allocate_bdev("bdev3");
600 	rc = spdk_bdev_module_claim_bdev(bdev[3], NULL, &bdev_ut_if);
601 	CU_ASSERT(rc == 0);
602 
603 	bdev[4] = allocate_vbdev("bdev4");
604 	rc = spdk_bdev_module_claim_bdev(bdev[4], NULL, &bdev_ut_if);
605 	CU_ASSERT(rc == 0);
606 
607 	bdev[5] = allocate_vbdev("bdev5");
608 	rc = spdk_bdev_module_claim_bdev(bdev[5], NULL, &bdev_ut_if);
609 	CU_ASSERT(rc == 0);
610 
611 	bdev[6] = allocate_vbdev("bdev6");
612 
613 	bdev[7] = allocate_vbdev("bdev7");
614 
615 	bdev[8] = allocate_vbdev("bdev8");
616 
617 	/* Open bdev0 read-only.  This should succeed. */
618 	rc = spdk_bdev_open_ext("bdev0", false, bdev_ut_event_cb, NULL, &desc[0]);
619 	CU_ASSERT(rc == 0);
620 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
621 	CU_ASSERT(bdev[0] == spdk_bdev_desc_get_bdev(desc[0]));
622 	spdk_bdev_close(desc[0]);
623 
624 	/*
625 	 * Open bdev1 read/write.  This should fail since bdev1 has been claimed
626 	 * by a vbdev module.
627 	 */
628 	rc = spdk_bdev_open_ext("bdev1", true, bdev_ut_event_cb, NULL, &desc[1]);
629 	CU_ASSERT(rc == -EPERM);
630 
631 	/*
632 	 * Open bdev4 read/write.  This should fail since bdev3 has been claimed
633 	 * by a vbdev module.
634 	 */
635 	rc = spdk_bdev_open_ext("bdev4", true, bdev_ut_event_cb, NULL, &desc[4]);
636 	CU_ASSERT(rc == -EPERM);
637 
638 	/* Open bdev4 read-only.  This should succeed. */
639 	rc = spdk_bdev_open_ext("bdev4", false, bdev_ut_event_cb, NULL, &desc[4]);
640 	CU_ASSERT(rc == 0);
641 	SPDK_CU_ASSERT_FATAL(desc[4] != NULL);
642 	CU_ASSERT(bdev[4] == spdk_bdev_desc_get_bdev(desc[4]));
643 	spdk_bdev_close(desc[4]);
644 
645 	/*
646 	 * Open bdev8 read/write.  This should succeed since it is a leaf
647 	 * bdev.
648 	 */
649 	rc = spdk_bdev_open_ext("bdev8", true, bdev_ut_event_cb, NULL, &desc[8]);
650 	CU_ASSERT(rc == 0);
651 	SPDK_CU_ASSERT_FATAL(desc[8] != NULL);
652 	CU_ASSERT(bdev[8] == spdk_bdev_desc_get_bdev(desc[8]));
653 	spdk_bdev_close(desc[8]);
654 
655 	/*
656 	 * Open bdev5 read/write.  This should fail since bdev4 has been claimed
657 	 * by a vbdev module.
658 	 */
659 	rc = spdk_bdev_open_ext("bdev5", true, bdev_ut_event_cb, NULL, &desc[5]);
660 	CU_ASSERT(rc == -EPERM);
661 
662 	/* Open bdev4 read-only.  This should succeed. */
663 	rc = spdk_bdev_open_ext("bdev5", false, bdev_ut_event_cb, NULL, &desc[5]);
664 	CU_ASSERT(rc == 0);
665 	SPDK_CU_ASSERT_FATAL(desc[5] != NULL);
666 	CU_ASSERT(bdev[5] == spdk_bdev_desc_get_bdev(desc[5]));
667 	spdk_bdev_close(desc[5]);
668 
669 	free_vbdev(bdev[8]);
670 
671 	free_vbdev(bdev[5]);
672 	free_vbdev(bdev[6]);
673 	free_vbdev(bdev[7]);
674 
675 	free_vbdev(bdev[4]);
676 
677 	free_bdev(bdev[0]);
678 	free_bdev(bdev[1]);
679 	free_bdev(bdev[2]);
680 	free_bdev(bdev[3]);
681 }
682 
683 static void
684 bytes_to_blocks_test(void)
685 {
686 	struct spdk_bdev bdev;
687 	uint64_t offset_blocks, num_blocks;
688 
689 	memset(&bdev, 0, sizeof(bdev));
690 
691 	bdev.blocklen = 512;
692 
693 	/* All parameters valid */
694 	offset_blocks = 0;
695 	num_blocks = 0;
696 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 1024, &num_blocks) == 0);
697 	CU_ASSERT(offset_blocks == 1);
698 	CU_ASSERT(num_blocks == 2);
699 
700 	/* Offset not a block multiple */
701 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 512, &num_blocks) != 0);
702 
703 	/* Length not a block multiple */
704 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 3, &num_blocks) != 0);
705 
706 	/* In case blocklen not the power of two */
707 	bdev.blocklen = 100;
708 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 200, &num_blocks) == 0);
709 	CU_ASSERT(offset_blocks == 1);
710 	CU_ASSERT(num_blocks == 2);
711 
712 	/* Offset not a block multiple */
713 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 100, &num_blocks) != 0);
714 
715 	/* Length not a block multiple */
716 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 3, &num_blocks) != 0);
717 }
718 
719 static void
720 num_blocks_test(void)
721 {
722 	struct spdk_bdev bdev;
723 	struct spdk_bdev_desc *desc = NULL;
724 	int rc;
725 
726 	memset(&bdev, 0, sizeof(bdev));
727 	bdev.name = "num_blocks";
728 	bdev.fn_table = &fn_table;
729 	bdev.module = &bdev_ut_if;
730 	spdk_bdev_register(&bdev);
731 	spdk_bdev_notify_blockcnt_change(&bdev, 50);
732 
733 	/* Growing block number */
734 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 70) == 0);
735 	/* Shrinking block number */
736 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 30) == 0);
737 
738 	rc = spdk_bdev_open_ext("num_blocks", false, bdev_open_cb1, &desc, &desc);
739 	CU_ASSERT(rc == 0);
740 	SPDK_CU_ASSERT_FATAL(desc != NULL);
741 	CU_ASSERT(&bdev == spdk_bdev_desc_get_bdev(desc));
742 
743 	/* Growing block number */
744 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 80) == 0);
745 	/* Shrinking block number */
746 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 20) != 0);
747 
748 	g_event_type1 = 0xFF;
749 	/* Growing block number */
750 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 90) == 0);
751 
752 	poll_threads();
753 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_RESIZE);
754 
755 	g_event_type1 = 0xFF;
756 	/* Growing block number and closing */
757 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 100) == 0);
758 
759 	spdk_bdev_close(desc);
760 	spdk_bdev_unregister(&bdev, NULL, NULL);
761 
762 	poll_threads();
763 
764 	/* Callback is not called for closed device */
765 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
766 }
767 
768 static void
769 io_valid_test(void)
770 {
771 	struct spdk_bdev bdev;
772 
773 	memset(&bdev, 0, sizeof(bdev));
774 
775 	bdev.blocklen = 512;
776 	CU_ASSERT(pthread_mutex_init(&bdev.internal.mutex, NULL) == 0);
777 
778 	spdk_bdev_notify_blockcnt_change(&bdev, 100);
779 
780 	/* All parameters valid */
781 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 1, 2) == true);
782 
783 	/* Last valid block */
784 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 1) == true);
785 
786 	/* Offset past end of bdev */
787 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 100, 1) == false);
788 
789 	/* Offset + length past end of bdev */
790 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 2) == false);
791 
792 	/* Offset near end of uint64_t range (2^64 - 1) */
793 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 18446744073709551615ULL, 1) == false);
794 
795 	CU_ASSERT(pthread_mutex_destroy(&bdev.internal.mutex) == 0);
796 }
797 
798 static void
799 alias_add_del_test(void)
800 {
801 	struct spdk_bdev *bdev[3];
802 	int rc;
803 
804 	/* Creating and registering bdevs */
805 	bdev[0] = allocate_bdev("bdev0");
806 	SPDK_CU_ASSERT_FATAL(bdev[0] != 0);
807 
808 	bdev[1] = allocate_bdev("bdev1");
809 	SPDK_CU_ASSERT_FATAL(bdev[1] != 0);
810 
811 	bdev[2] = allocate_bdev("bdev2");
812 	SPDK_CU_ASSERT_FATAL(bdev[2] != 0);
813 
814 	poll_threads();
815 
816 	/*
817 	 * Trying adding an alias identical to name.
818 	 * Alias is identical to name, so it can not be added to aliases list
819 	 */
820 	rc = spdk_bdev_alias_add(bdev[0], bdev[0]->name);
821 	CU_ASSERT(rc == -EEXIST);
822 
823 	/*
824 	 * Trying to add empty alias,
825 	 * this one should fail
826 	 */
827 	rc = spdk_bdev_alias_add(bdev[0], NULL);
828 	CU_ASSERT(rc == -EINVAL);
829 
830 	/* Trying adding same alias to two different registered bdevs */
831 
832 	/* Alias is used first time, so this one should pass */
833 	rc = spdk_bdev_alias_add(bdev[0], "proper alias 0");
834 	CU_ASSERT(rc == 0);
835 
836 	/* Alias was added to another bdev, so this one should fail */
837 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 0");
838 	CU_ASSERT(rc == -EEXIST);
839 
840 	/* Alias is used first time, so this one should pass */
841 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 1");
842 	CU_ASSERT(rc == 0);
843 
844 	/* Trying removing an alias from registered bdevs */
845 
846 	/* Alias is not on a bdev aliases list, so this one should fail */
847 	rc = spdk_bdev_alias_del(bdev[0], "not existing");
848 	CU_ASSERT(rc == -ENOENT);
849 
850 	/* Alias is present on a bdev aliases list, so this one should pass */
851 	rc = spdk_bdev_alias_del(bdev[0], "proper alias 0");
852 	CU_ASSERT(rc == 0);
853 
854 	/* Alias is present on a bdev aliases list, so this one should pass */
855 	rc = spdk_bdev_alias_del(bdev[1], "proper alias 1");
856 	CU_ASSERT(rc == 0);
857 
858 	/* Trying to remove name instead of alias, so this one should fail, name cannot be changed or removed */
859 	rc = spdk_bdev_alias_del(bdev[0], bdev[0]->name);
860 	CU_ASSERT(rc != 0);
861 
862 	/* Trying to del all alias from empty alias list */
863 	spdk_bdev_alias_del_all(bdev[2]);
864 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev[2]->aliases));
865 
866 	/* Trying to del all alias from non-empty alias list */
867 	rc = spdk_bdev_alias_add(bdev[2], "alias0");
868 	CU_ASSERT(rc == 0);
869 	rc = spdk_bdev_alias_add(bdev[2], "alias1");
870 	CU_ASSERT(rc == 0);
871 	spdk_bdev_alias_del_all(bdev[2]);
872 	CU_ASSERT(TAILQ_EMPTY(&bdev[2]->aliases));
873 
874 	/* Unregister and free bdevs */
875 	spdk_bdev_unregister(bdev[0], NULL, NULL);
876 	spdk_bdev_unregister(bdev[1], NULL, NULL);
877 	spdk_bdev_unregister(bdev[2], NULL, NULL);
878 
879 	poll_threads();
880 
881 	free(bdev[0]);
882 	free(bdev[1]);
883 	free(bdev[2]);
884 }
885 
886 static void
887 io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
888 {
889 	g_io_done = true;
890 	g_io_status = bdev_io->internal.status;
891 	if ((bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) &&
892 	    (bdev_io->u.bdev.zcopy.start)) {
893 		g_zcopy_bdev_io = bdev_io;
894 	} else {
895 		spdk_bdev_free_io(bdev_io);
896 		g_zcopy_bdev_io = NULL;
897 	}
898 }
899 
900 static void
901 bdev_init_cb(void *arg, int rc)
902 {
903 	CU_ASSERT(rc == 0);
904 }
905 
906 static void
907 bdev_fini_cb(void *arg)
908 {
909 }
910 
911 struct bdev_ut_io_wait_entry {
912 	struct spdk_bdev_io_wait_entry	entry;
913 	struct spdk_io_channel		*io_ch;
914 	struct spdk_bdev_desc		*desc;
915 	bool				submitted;
916 };
917 
918 static void
919 io_wait_cb(void *arg)
920 {
921 	struct bdev_ut_io_wait_entry *entry = arg;
922 	int rc;
923 
924 	rc = spdk_bdev_read_blocks(entry->desc, entry->io_ch, NULL, 0, 1, io_done, NULL);
925 	CU_ASSERT(rc == 0);
926 	entry->submitted = true;
927 }
928 
929 static void
930 bdev_io_types_test(void)
931 {
932 	struct spdk_bdev *bdev;
933 	struct spdk_bdev_desc *desc = NULL;
934 	struct spdk_io_channel *io_ch;
935 	struct spdk_bdev_opts bdev_opts = {};
936 	int rc;
937 
938 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
939 	bdev_opts.bdev_io_pool_size = 4;
940 	bdev_opts.bdev_io_cache_size = 2;
941 
942 	rc = spdk_bdev_set_opts(&bdev_opts);
943 	CU_ASSERT(rc == 0);
944 	spdk_bdev_initialize(bdev_init_cb, NULL);
945 	poll_threads();
946 
947 	bdev = allocate_bdev("bdev0");
948 
949 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
950 	CU_ASSERT(rc == 0);
951 	poll_threads();
952 	SPDK_CU_ASSERT_FATAL(desc != NULL);
953 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
954 	io_ch = spdk_bdev_get_io_channel(desc);
955 	CU_ASSERT(io_ch != NULL);
956 
957 	/* WRITE and WRITE ZEROES are not supported */
958 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
959 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, false);
960 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 0, 128, io_done, NULL);
961 	CU_ASSERT(rc == -ENOTSUP);
962 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
963 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, true);
964 
965 	spdk_put_io_channel(io_ch);
966 	spdk_bdev_close(desc);
967 	free_bdev(bdev);
968 	spdk_bdev_finish(bdev_fini_cb, NULL);
969 	poll_threads();
970 }
971 
972 static void
973 bdev_io_wait_test(void)
974 {
975 	struct spdk_bdev *bdev;
976 	struct spdk_bdev_desc *desc = NULL;
977 	struct spdk_io_channel *io_ch;
978 	struct spdk_bdev_opts bdev_opts = {};
979 	struct bdev_ut_io_wait_entry io_wait_entry;
980 	struct bdev_ut_io_wait_entry io_wait_entry2;
981 	int rc;
982 
983 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
984 	bdev_opts.bdev_io_pool_size = 4;
985 	bdev_opts.bdev_io_cache_size = 2;
986 
987 	rc = spdk_bdev_set_opts(&bdev_opts);
988 	CU_ASSERT(rc == 0);
989 	spdk_bdev_initialize(bdev_init_cb, NULL);
990 	poll_threads();
991 
992 	bdev = allocate_bdev("bdev0");
993 
994 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
995 	CU_ASSERT(rc == 0);
996 	poll_threads();
997 	SPDK_CU_ASSERT_FATAL(desc != NULL);
998 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
999 	io_ch = spdk_bdev_get_io_channel(desc);
1000 	CU_ASSERT(io_ch != NULL);
1001 
1002 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1003 	CU_ASSERT(rc == 0);
1004 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1005 	CU_ASSERT(rc == 0);
1006 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1007 	CU_ASSERT(rc == 0);
1008 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1009 	CU_ASSERT(rc == 0);
1010 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1011 
1012 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1013 	CU_ASSERT(rc == -ENOMEM);
1014 
1015 	io_wait_entry.entry.bdev = bdev;
1016 	io_wait_entry.entry.cb_fn = io_wait_cb;
1017 	io_wait_entry.entry.cb_arg = &io_wait_entry;
1018 	io_wait_entry.io_ch = io_ch;
1019 	io_wait_entry.desc = desc;
1020 	io_wait_entry.submitted = false;
1021 	/* Cannot use the same io_wait_entry for two different calls. */
1022 	memcpy(&io_wait_entry2, &io_wait_entry, sizeof(io_wait_entry));
1023 	io_wait_entry2.entry.cb_arg = &io_wait_entry2;
1024 
1025 	/* Queue two I/O waits. */
1026 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry.entry);
1027 	CU_ASSERT(rc == 0);
1028 	CU_ASSERT(io_wait_entry.submitted == false);
1029 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry2.entry);
1030 	CU_ASSERT(rc == 0);
1031 	CU_ASSERT(io_wait_entry2.submitted == false);
1032 
1033 	stub_complete_io(1);
1034 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1035 	CU_ASSERT(io_wait_entry.submitted == true);
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_entry2.submitted == true);
1041 
1042 	stub_complete_io(4);
1043 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1044 
1045 	spdk_put_io_channel(io_ch);
1046 	spdk_bdev_close(desc);
1047 	free_bdev(bdev);
1048 	spdk_bdev_finish(bdev_fini_cb, NULL);
1049 	poll_threads();
1050 }
1051 
1052 static void
1053 bdev_io_spans_split_test(void)
1054 {
1055 	struct spdk_bdev bdev;
1056 	struct spdk_bdev_io bdev_io;
1057 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV];
1058 
1059 	memset(&bdev, 0, sizeof(bdev));
1060 	bdev_io.u.bdev.iovs = iov;
1061 
1062 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1063 	bdev.optimal_io_boundary = 0;
1064 	bdev.max_segment_size = 0;
1065 	bdev.max_num_segments = 0;
1066 	bdev_io.bdev = &bdev;
1067 
1068 	/* bdev has no optimal_io_boundary and max_size set - so this should return false. */
1069 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1070 
1071 	bdev.split_on_optimal_io_boundary = true;
1072 	bdev.optimal_io_boundary = 32;
1073 	bdev_io.type = SPDK_BDEV_IO_TYPE_RESET;
1074 
1075 	/* RESETs are not based on LBAs - so this should return false. */
1076 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1077 
1078 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1079 	bdev_io.u.bdev.offset_blocks = 0;
1080 	bdev_io.u.bdev.num_blocks = 32;
1081 
1082 	/* This I/O run right up to, but does not cross, the boundary - so this should return false. */
1083 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1084 
1085 	bdev_io.u.bdev.num_blocks = 33;
1086 
1087 	/* This I/O spans a boundary. */
1088 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1089 
1090 	bdev_io.u.bdev.num_blocks = 32;
1091 	bdev.max_segment_size = 512 * 32;
1092 	bdev.max_num_segments = 1;
1093 	bdev_io.u.bdev.iovcnt = 1;
1094 	iov[0].iov_len = 512;
1095 
1096 	/* Does not cross and exceed max_size or max_segs */
1097 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1098 
1099 	bdev.split_on_optimal_io_boundary = false;
1100 	bdev.max_segment_size = 512;
1101 	bdev.max_num_segments = 1;
1102 	bdev_io.u.bdev.iovcnt = 2;
1103 
1104 	/* Exceed max_segs */
1105 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1106 
1107 	bdev.max_num_segments = 2;
1108 	iov[0].iov_len = 513;
1109 	iov[1].iov_len = 512;
1110 
1111 	/* Exceed max_sizes */
1112 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1113 }
1114 
1115 static void
1116 bdev_io_boundary_split_test(void)
1117 {
1118 	struct spdk_bdev *bdev;
1119 	struct spdk_bdev_desc *desc = NULL;
1120 	struct spdk_io_channel *io_ch;
1121 	struct spdk_bdev_opts bdev_opts = {};
1122 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1123 	struct ut_expected_io *expected_io;
1124 	void *md_buf = (void *)0xFF000000;
1125 	uint64_t i;
1126 	int rc;
1127 
1128 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1129 	bdev_opts.bdev_io_pool_size = 512;
1130 	bdev_opts.bdev_io_cache_size = 64;
1131 
1132 	rc = spdk_bdev_set_opts(&bdev_opts);
1133 	CU_ASSERT(rc == 0);
1134 	spdk_bdev_initialize(bdev_init_cb, NULL);
1135 
1136 	bdev = allocate_bdev("bdev0");
1137 
1138 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1139 	CU_ASSERT(rc == 0);
1140 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1141 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1142 	io_ch = spdk_bdev_get_io_channel(desc);
1143 	CU_ASSERT(io_ch != NULL);
1144 
1145 	bdev->optimal_io_boundary = 16;
1146 	bdev->split_on_optimal_io_boundary = false;
1147 
1148 	g_io_done = false;
1149 
1150 	/* First test that the I/O does not get split if split_on_optimal_io_boundary == false. */
1151 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 8, 1);
1152 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 8 * 512);
1153 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1154 
1155 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
1156 	CU_ASSERT(rc == 0);
1157 	CU_ASSERT(g_io_done == false);
1158 
1159 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1160 	stub_complete_io(1);
1161 	CU_ASSERT(g_io_done == true);
1162 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1163 
1164 	bdev->split_on_optimal_io_boundary = true;
1165 	bdev->md_interleave = false;
1166 	bdev->md_len = 8;
1167 
1168 	/* Now test that a single-vector command is split correctly.
1169 	 * Offset 14, length 8, payload 0xF000
1170 	 *  Child - Offset 14, length 2, payload 0xF000
1171 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
1172 	 *
1173 	 * Set up the expected values before calling spdk_bdev_read_blocks
1174 	 */
1175 	g_io_done = false;
1176 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
1177 	expected_io->md_buf = md_buf;
1178 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
1179 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1180 
1181 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
1182 	expected_io->md_buf = md_buf + 2 * 8;
1183 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
1184 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1185 
1186 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1187 	rc = spdk_bdev_read_blocks_with_md(desc, io_ch, (void *)0xF000, md_buf,
1188 					   14, 8, io_done, NULL);
1189 	CU_ASSERT(rc == 0);
1190 	CU_ASSERT(g_io_done == false);
1191 
1192 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1193 	stub_complete_io(2);
1194 	CU_ASSERT(g_io_done == true);
1195 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1196 
1197 	/* Now set up a more complex, multi-vector command that needs to be split,
1198 	 *  including splitting iovecs.
1199 	 */
1200 	iov[0].iov_base = (void *)0x10000;
1201 	iov[0].iov_len = 512;
1202 	iov[1].iov_base = (void *)0x20000;
1203 	iov[1].iov_len = 20 * 512;
1204 	iov[2].iov_base = (void *)0x30000;
1205 	iov[2].iov_len = 11 * 512;
1206 
1207 	g_io_done = false;
1208 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
1209 	expected_io->md_buf = md_buf;
1210 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
1211 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
1212 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1213 
1214 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
1215 	expected_io->md_buf = md_buf + 2 * 8;
1216 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 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, 32, 14, 2);
1220 	expected_io->md_buf = md_buf + 18 * 8;
1221 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
1222 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
1223 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1224 
1225 	rc = spdk_bdev_writev_blocks_with_md(desc, io_ch, iov, 3, md_buf,
1226 					     14, 32, io_done, NULL);
1227 	CU_ASSERT(rc == 0);
1228 	CU_ASSERT(g_io_done == false);
1229 
1230 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
1231 	stub_complete_io(3);
1232 	CU_ASSERT(g_io_done == true);
1233 
1234 	/* Test multi vector command that needs to be split by strip and then needs to be
1235 	 * split further due to the capacity of child iovs.
1236 	 */
1237 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
1238 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1239 		iov[i].iov_len = 512;
1240 	}
1241 
1242 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1243 	g_io_done = false;
1244 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1245 					   BDEV_IO_NUM_CHILD_IOV);
1246 	expected_io->md_buf = md_buf;
1247 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1248 		ut_expected_io_set_iov(expected_io, i, (void *)((i + 1) * 0x10000), 512);
1249 	}
1250 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1251 
1252 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1253 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
1254 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1255 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1256 		ut_expected_io_set_iov(expected_io, i,
1257 				       (void *)((i + 1 + BDEV_IO_NUM_CHILD_IOV) * 0x10000), 512);
1258 	}
1259 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1260 
1261 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1262 					    0, BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1263 	CU_ASSERT(rc == 0);
1264 	CU_ASSERT(g_io_done == false);
1265 
1266 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1267 	stub_complete_io(1);
1268 	CU_ASSERT(g_io_done == false);
1269 
1270 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1271 	stub_complete_io(1);
1272 	CU_ASSERT(g_io_done == true);
1273 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1274 
1275 	/* Test multi vector command that needs to be split by strip and then needs to be
1276 	 * split further due to the capacity of child iovs. In this case, the length of
1277 	 * the rest of iovec array with an I/O boundary is the multiple of block size.
1278 	 */
1279 
1280 	/* Fill iovec array for exactly one boundary. The iovec cnt for this boundary
1281 	 * is BDEV_IO_NUM_CHILD_IOV + 1, which exceeds the capacity of child iovs.
1282 	 */
1283 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1284 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1285 		iov[i].iov_len = 512;
1286 	}
1287 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1288 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1289 		iov[i].iov_len = 256;
1290 	}
1291 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1292 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 512;
1293 
1294 	/* Add an extra iovec to trigger split */
1295 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1296 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1297 
1298 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1299 	g_io_done = false;
1300 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1301 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV);
1302 	expected_io->md_buf = md_buf;
1303 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1304 		ut_expected_io_set_iov(expected_io, i,
1305 				       (void *)((i + 1) * 0x10000), 512);
1306 	}
1307 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1308 		ut_expected_io_set_iov(expected_io, i,
1309 				       (void *)((i + 1) * 0x10000), 256);
1310 	}
1311 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1312 
1313 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1314 					   1, 1);
1315 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1316 	ut_expected_io_set_iov(expected_io, 0,
1317 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 512);
1318 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1319 
1320 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1321 					   1, 1);
1322 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1323 	ut_expected_io_set_iov(expected_io, 0,
1324 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1325 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1326 
1327 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, md_buf,
1328 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1329 	CU_ASSERT(rc == 0);
1330 	CU_ASSERT(g_io_done == false);
1331 
1332 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1333 	stub_complete_io(1);
1334 	CU_ASSERT(g_io_done == false);
1335 
1336 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1337 	stub_complete_io(2);
1338 	CU_ASSERT(g_io_done == true);
1339 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1340 
1341 	/* Test multi vector command that needs to be split by strip and then needs to be
1342 	 * split further due to the capacity of child iovs, the child request offset should
1343 	 * be rewind to last aligned offset and go success without error.
1344 	 */
1345 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1346 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1347 		iov[i].iov_len = 512;
1348 	}
1349 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1350 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1351 
1352 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1353 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1354 
1355 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1356 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1357 
1358 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1359 	g_io_done = false;
1360 	g_io_status = 0;
1361 	/* The first expected io should be start from offset 0 to BDEV_IO_NUM_CHILD_IOV - 1 */
1362 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1363 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV - 1);
1364 	expected_io->md_buf = md_buf;
1365 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1366 		ut_expected_io_set_iov(expected_io, i,
1367 				       (void *)((i + 1) * 0x10000), 512);
1368 	}
1369 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1370 	/* The second expected io should be start from offset BDEV_IO_NUM_CHILD_IOV - 1 to BDEV_IO_NUM_CHILD_IOV */
1371 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1372 					   1, 2);
1373 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1374 	ut_expected_io_set_iov(expected_io, 0,
1375 			       (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000), 256);
1376 	ut_expected_io_set_iov(expected_io, 1,
1377 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 256);
1378 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1379 	/* The third expected io should be start from offset BDEV_IO_NUM_CHILD_IOV to BDEV_IO_NUM_CHILD_IOV + 1 */
1380 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1381 					   1, 1);
1382 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1383 	ut_expected_io_set_iov(expected_io, 0,
1384 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1385 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1386 
1387 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1388 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1389 	CU_ASSERT(rc == 0);
1390 	CU_ASSERT(g_io_done == false);
1391 
1392 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1393 	stub_complete_io(1);
1394 	CU_ASSERT(g_io_done == false);
1395 
1396 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1397 	stub_complete_io(2);
1398 	CU_ASSERT(g_io_done == true);
1399 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1400 
1401 	/* Test multi vector command that needs to be split due to the IO boundary and
1402 	 * the capacity of child iovs. Especially test the case when the command is
1403 	 * split due to the capacity of child iovs, the tail address is not aligned with
1404 	 * block size and is rewinded to the aligned address.
1405 	 *
1406 	 * The iovecs used in read request is complex but is based on the data
1407 	 * collected in the real issue. We change the base addresses but keep the lengths
1408 	 * not to loose the credibility of the test.
1409 	 */
1410 	bdev->optimal_io_boundary = 128;
1411 	g_io_done = false;
1412 	g_io_status = 0;
1413 
1414 	for (i = 0; i < 31; i++) {
1415 		iov[i].iov_base = (void *)(0xFEED0000000 + (i << 20));
1416 		iov[i].iov_len = 1024;
1417 	}
1418 	iov[31].iov_base = (void *)0xFEED1F00000;
1419 	iov[31].iov_len = 32768;
1420 	iov[32].iov_base = (void *)0xFEED2000000;
1421 	iov[32].iov_len = 160;
1422 	iov[33].iov_base = (void *)0xFEED2100000;
1423 	iov[33].iov_len = 4096;
1424 	iov[34].iov_base = (void *)0xFEED2200000;
1425 	iov[34].iov_len = 4096;
1426 	iov[35].iov_base = (void *)0xFEED2300000;
1427 	iov[35].iov_len = 4096;
1428 	iov[36].iov_base = (void *)0xFEED2400000;
1429 	iov[36].iov_len = 4096;
1430 	iov[37].iov_base = (void *)0xFEED2500000;
1431 	iov[37].iov_len = 4096;
1432 	iov[38].iov_base = (void *)0xFEED2600000;
1433 	iov[38].iov_len = 4096;
1434 	iov[39].iov_base = (void *)0xFEED2700000;
1435 	iov[39].iov_len = 4096;
1436 	iov[40].iov_base = (void *)0xFEED2800000;
1437 	iov[40].iov_len = 4096;
1438 	iov[41].iov_base = (void *)0xFEED2900000;
1439 	iov[41].iov_len = 4096;
1440 	iov[42].iov_base = (void *)0xFEED2A00000;
1441 	iov[42].iov_len = 4096;
1442 	iov[43].iov_base = (void *)0xFEED2B00000;
1443 	iov[43].iov_len = 12288;
1444 	iov[44].iov_base = (void *)0xFEED2C00000;
1445 	iov[44].iov_len = 8192;
1446 	iov[45].iov_base = (void *)0xFEED2F00000;
1447 	iov[45].iov_len = 4096;
1448 	iov[46].iov_base = (void *)0xFEED3000000;
1449 	iov[46].iov_len = 4096;
1450 	iov[47].iov_base = (void *)0xFEED3100000;
1451 	iov[47].iov_len = 4096;
1452 	iov[48].iov_base = (void *)0xFEED3200000;
1453 	iov[48].iov_len = 24576;
1454 	iov[49].iov_base = (void *)0xFEED3300000;
1455 	iov[49].iov_len = 16384;
1456 	iov[50].iov_base = (void *)0xFEED3400000;
1457 	iov[50].iov_len = 12288;
1458 	iov[51].iov_base = (void *)0xFEED3500000;
1459 	iov[51].iov_len = 4096;
1460 	iov[52].iov_base = (void *)0xFEED3600000;
1461 	iov[52].iov_len = 4096;
1462 	iov[53].iov_base = (void *)0xFEED3700000;
1463 	iov[53].iov_len = 4096;
1464 	iov[54].iov_base = (void *)0xFEED3800000;
1465 	iov[54].iov_len = 28672;
1466 	iov[55].iov_base = (void *)0xFEED3900000;
1467 	iov[55].iov_len = 20480;
1468 	iov[56].iov_base = (void *)0xFEED3A00000;
1469 	iov[56].iov_len = 4096;
1470 	iov[57].iov_base = (void *)0xFEED3B00000;
1471 	iov[57].iov_len = 12288;
1472 	iov[58].iov_base = (void *)0xFEED3C00000;
1473 	iov[58].iov_len = 4096;
1474 	iov[59].iov_base = (void *)0xFEED3D00000;
1475 	iov[59].iov_len = 4096;
1476 	iov[60].iov_base = (void *)0xFEED3E00000;
1477 	iov[60].iov_len = 352;
1478 
1479 	/* The 1st child IO must be from iov[0] to iov[31] split by the capacity
1480 	 * of child iovs,
1481 	 */
1482 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 126, 32);
1483 	expected_io->md_buf = md_buf;
1484 	for (i = 0; i < 32; i++) {
1485 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1486 	}
1487 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1488 
1489 	/* The 2nd child IO must be from iov[32] to the first 864 bytes of iov[33]
1490 	 * split by the IO boundary requirement.
1491 	 */
1492 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 126, 2, 2);
1493 	expected_io->md_buf = md_buf + 126 * 8;
1494 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base, iov[32].iov_len);
1495 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 864);
1496 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1497 
1498 	/* The 3rd child IO must be from the remaining 3232 bytes of iov[33] to
1499 	 * the first 864 bytes of iov[46] split by the IO boundary requirement.
1500 	 */
1501 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 128, 128, 14);
1502 	expected_io->md_buf = md_buf + 128 * 8;
1503 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[33].iov_base + 864),
1504 			       iov[33].iov_len - 864);
1505 	ut_expected_io_set_iov(expected_io, 1, iov[34].iov_base, iov[34].iov_len);
1506 	ut_expected_io_set_iov(expected_io, 2, iov[35].iov_base, iov[35].iov_len);
1507 	ut_expected_io_set_iov(expected_io, 3, iov[36].iov_base, iov[36].iov_len);
1508 	ut_expected_io_set_iov(expected_io, 4, iov[37].iov_base, iov[37].iov_len);
1509 	ut_expected_io_set_iov(expected_io, 5, iov[38].iov_base, iov[38].iov_len);
1510 	ut_expected_io_set_iov(expected_io, 6, iov[39].iov_base, iov[39].iov_len);
1511 	ut_expected_io_set_iov(expected_io, 7, iov[40].iov_base, iov[40].iov_len);
1512 	ut_expected_io_set_iov(expected_io, 8, iov[41].iov_base, iov[41].iov_len);
1513 	ut_expected_io_set_iov(expected_io, 9, iov[42].iov_base, iov[42].iov_len);
1514 	ut_expected_io_set_iov(expected_io, 10, iov[43].iov_base, iov[43].iov_len);
1515 	ut_expected_io_set_iov(expected_io, 11, iov[44].iov_base, iov[44].iov_len);
1516 	ut_expected_io_set_iov(expected_io, 12, iov[45].iov_base, iov[45].iov_len);
1517 	ut_expected_io_set_iov(expected_io, 13, iov[46].iov_base, 864);
1518 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1519 
1520 	/* The 4th child IO must be from the remaining 3232 bytes of iov[46] to the
1521 	 * first 864 bytes of iov[52] split by the IO boundary requirement.
1522 	 */
1523 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 256, 128, 7);
1524 	expected_io->md_buf = md_buf + 256 * 8;
1525 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[46].iov_base + 864),
1526 			       iov[46].iov_len - 864);
1527 	ut_expected_io_set_iov(expected_io, 1, iov[47].iov_base, iov[47].iov_len);
1528 	ut_expected_io_set_iov(expected_io, 2, iov[48].iov_base, iov[48].iov_len);
1529 	ut_expected_io_set_iov(expected_io, 3, iov[49].iov_base, iov[49].iov_len);
1530 	ut_expected_io_set_iov(expected_io, 4, iov[50].iov_base, iov[50].iov_len);
1531 	ut_expected_io_set_iov(expected_io, 5, iov[51].iov_base, iov[51].iov_len);
1532 	ut_expected_io_set_iov(expected_io, 6, iov[52].iov_base, 864);
1533 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1534 
1535 	/* The 5th child IO must be from the remaining 3232 bytes of iov[52] to
1536 	 * the first 4096 bytes of iov[57] split by the IO boundary requirement.
1537 	 */
1538 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 384, 128, 6);
1539 	expected_io->md_buf = md_buf + 384 * 8;
1540 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[52].iov_base + 864),
1541 			       iov[52].iov_len - 864);
1542 	ut_expected_io_set_iov(expected_io, 1, iov[53].iov_base, iov[53].iov_len);
1543 	ut_expected_io_set_iov(expected_io, 2, iov[54].iov_base, iov[54].iov_len);
1544 	ut_expected_io_set_iov(expected_io, 3, iov[55].iov_base, iov[55].iov_len);
1545 	ut_expected_io_set_iov(expected_io, 4, iov[56].iov_base, iov[56].iov_len);
1546 	ut_expected_io_set_iov(expected_io, 5, iov[57].iov_base, 4960);
1547 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1548 
1549 	/* The 6th child IO must be from the remaining 7328 bytes of iov[57]
1550 	 * to the first 3936 bytes of iov[58] split by the capacity of child iovs.
1551 	 */
1552 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 512, 30, 3);
1553 	expected_io->md_buf = md_buf + 512 * 8;
1554 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[57].iov_base + 4960),
1555 			       iov[57].iov_len - 4960);
1556 	ut_expected_io_set_iov(expected_io, 1, iov[58].iov_base, iov[58].iov_len);
1557 	ut_expected_io_set_iov(expected_io, 2, iov[59].iov_base, 3936);
1558 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1559 
1560 	/* The 7th child IO is from the remaining 160 bytes of iov[59] and iov[60]. */
1561 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 542, 1, 2);
1562 	expected_io->md_buf = md_buf + 542 * 8;
1563 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[59].iov_base + 3936),
1564 			       iov[59].iov_len - 3936);
1565 	ut_expected_io_set_iov(expected_io, 1, iov[60].iov_base, iov[60].iov_len);
1566 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1567 
1568 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, 61, md_buf,
1569 					    0, 543, io_done, NULL);
1570 	CU_ASSERT(rc == 0);
1571 	CU_ASSERT(g_io_done == false);
1572 
1573 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1574 	stub_complete_io(1);
1575 	CU_ASSERT(g_io_done == false);
1576 
1577 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1578 	stub_complete_io(5);
1579 	CU_ASSERT(g_io_done == false);
1580 
1581 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1582 	stub_complete_io(1);
1583 	CU_ASSERT(g_io_done == true);
1584 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1585 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1586 
1587 	/* Test a WRITE_ZEROES that would span an I/O boundary.  WRITE_ZEROES should not be
1588 	 * split, so test that.
1589 	 */
1590 	bdev->optimal_io_boundary = 15;
1591 	g_io_done = false;
1592 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
1593 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1594 
1595 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
1596 	CU_ASSERT(rc == 0);
1597 	CU_ASSERT(g_io_done == false);
1598 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1599 	stub_complete_io(1);
1600 	CU_ASSERT(g_io_done == true);
1601 
1602 	/* Test an UNMAP.  This should also not be split. */
1603 	bdev->optimal_io_boundary = 16;
1604 	g_io_done = false;
1605 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 2, 0);
1606 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1607 
1608 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 2, io_done, NULL);
1609 	CU_ASSERT(rc == 0);
1610 	CU_ASSERT(g_io_done == false);
1611 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1612 	stub_complete_io(1);
1613 	CU_ASSERT(g_io_done == true);
1614 
1615 	/* Test a FLUSH.  This should also not be split. */
1616 	bdev->optimal_io_boundary = 16;
1617 	g_io_done = false;
1618 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 2, 0);
1619 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1620 
1621 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
1622 	CU_ASSERT(rc == 0);
1623 	CU_ASSERT(g_io_done == false);
1624 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1625 	stub_complete_io(1);
1626 	CU_ASSERT(g_io_done == true);
1627 
1628 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
1629 
1630 	/* Children requests return an error status */
1631 	bdev->optimal_io_boundary = 16;
1632 	iov[0].iov_base = (void *)0x10000;
1633 	iov[0].iov_len = 512 * 64;
1634 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1635 	g_io_done = false;
1636 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1637 
1638 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 1, 1, 64, io_done, NULL);
1639 	CU_ASSERT(rc == 0);
1640 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1641 	stub_complete_io(4);
1642 	CU_ASSERT(g_io_done == false);
1643 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1644 	stub_complete_io(1);
1645 	CU_ASSERT(g_io_done == true);
1646 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1647 
1648 	/* Test if a multi vector command terminated with failure before continueing
1649 	 * splitting process when one of child I/O failed.
1650 	 * The multi vector command is as same as the above that needs to be split by strip
1651 	 * and then needs to be split further due to the capacity of child iovs.
1652 	 */
1653 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1654 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1655 		iov[i].iov_len = 512;
1656 	}
1657 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1658 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1659 
1660 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1661 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1662 
1663 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1664 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1665 
1666 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1667 
1668 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1669 	g_io_done = false;
1670 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1671 
1672 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
1673 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1674 	CU_ASSERT(rc == 0);
1675 	CU_ASSERT(g_io_done == false);
1676 
1677 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1678 	stub_complete_io(1);
1679 	CU_ASSERT(g_io_done == true);
1680 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1681 
1682 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1683 
1684 	/* for this test we will create the following conditions to hit the code path where
1685 	 * we are trying to send and IO following a split that has no iovs because we had to
1686 	 * trim them for alignment reasons.
1687 	 *
1688 	 * - 16K boundary, our IO will start at offset 0 with a length of 0x4200
1689 	 * - Our IOVs are 0x212 in size so that we run into the 16K boundary at child IOV
1690 	 *   position 30 and overshoot by 0x2e.
1691 	 * - That means we'll send the IO and loop back to pick up the remaining bytes at
1692 	 *   child IOV index 31. When we do, we find that we have to shorten index 31 by 0x2e
1693 	 *   which eliniates that vector so we just send the first split IO with 30 vectors
1694 	 *   and let the completion pick up the last 2 vectors.
1695 	 */
1696 	bdev->optimal_io_boundary = 32;
1697 	bdev->split_on_optimal_io_boundary = true;
1698 	g_io_done = false;
1699 
1700 	/* Init all parent IOVs to 0x212 */
1701 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
1702 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1703 		iov[i].iov_len = 0x212;
1704 	}
1705 
1706 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1707 					   BDEV_IO_NUM_CHILD_IOV - 1);
1708 	/* expect 0-29 to be 1:1 with the parent iov */
1709 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1710 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1711 	}
1712 
1713 	/* expect index 30 to be shortened to 0x1e4 (0x212 - 0x1e) because of the alignment
1714 	 * where 0x1e is the amount we overshot the 16K boundary
1715 	 */
1716 	ut_expected_io_set_iov(expected_io, BDEV_IO_NUM_CHILD_IOV - 2,
1717 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base), 0x1e4);
1718 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1719 
1720 	/* 2nd child IO will have 2 remaining vectors, one to pick up from the one that was
1721 	 * shortened that take it to the next boundary and then a final one to get us to
1722 	 * 0x4200 bytes for the IO.
1723 	 */
1724 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1725 					   BDEV_IO_NUM_CHILD_IOV, 2);
1726 	/* position 30 picked up the remaining bytes to the next boundary */
1727 	ut_expected_io_set_iov(expected_io, 0,
1728 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base + 0x1e4), 0x2e);
1729 
1730 	/* position 31 picked the the rest of the trasnfer to get us to 0x4200 */
1731 	ut_expected_io_set_iov(expected_io, 1,
1732 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base), 0x1d2);
1733 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1734 
1735 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 1, 0,
1736 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1737 	CU_ASSERT(rc == 0);
1738 	CU_ASSERT(g_io_done == false);
1739 
1740 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1741 	stub_complete_io(1);
1742 	CU_ASSERT(g_io_done == false);
1743 
1744 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1745 	stub_complete_io(1);
1746 	CU_ASSERT(g_io_done == true);
1747 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1748 
1749 	spdk_put_io_channel(io_ch);
1750 	spdk_bdev_close(desc);
1751 	free_bdev(bdev);
1752 	spdk_bdev_finish(bdev_fini_cb, NULL);
1753 	poll_threads();
1754 }
1755 
1756 static void
1757 bdev_io_max_size_and_segment_split_test(void)
1758 {
1759 	struct spdk_bdev *bdev;
1760 	struct spdk_bdev_desc *desc = NULL;
1761 	struct spdk_io_channel *io_ch;
1762 	struct spdk_bdev_opts bdev_opts = {};
1763 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1764 	struct ut_expected_io *expected_io;
1765 	uint64_t i;
1766 	int rc;
1767 
1768 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1769 	bdev_opts.bdev_io_pool_size = 512;
1770 	bdev_opts.bdev_io_cache_size = 64;
1771 
1772 	bdev_opts.opts_size = sizeof(bdev_opts);
1773 	rc = spdk_bdev_set_opts(&bdev_opts);
1774 	CU_ASSERT(rc == 0);
1775 	spdk_bdev_initialize(bdev_init_cb, NULL);
1776 
1777 	bdev = allocate_bdev("bdev0");
1778 
1779 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
1780 	CU_ASSERT(rc == 0);
1781 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1782 	io_ch = spdk_bdev_get_io_channel(desc);
1783 	CU_ASSERT(io_ch != NULL);
1784 
1785 	bdev->split_on_optimal_io_boundary = false;
1786 	bdev->optimal_io_boundary = 0;
1787 
1788 	/* Case 0 max_num_segments == 0.
1789 	 * but segment size 2 * 512 > 512
1790 	 */
1791 	bdev->max_segment_size = 512;
1792 	bdev->max_num_segments = 0;
1793 	g_io_done = false;
1794 
1795 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
1796 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1797 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
1798 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1799 
1800 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1801 	CU_ASSERT(rc == 0);
1802 	CU_ASSERT(g_io_done == false);
1803 
1804 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1805 	stub_complete_io(1);
1806 	CU_ASSERT(g_io_done == true);
1807 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1808 
1809 	/* Case 1 max_segment_size == 0
1810 	 * but iov num 2 > 1.
1811 	 */
1812 	bdev->max_segment_size = 0;
1813 	bdev->max_num_segments = 1;
1814 	g_io_done = false;
1815 
1816 	iov[0].iov_base = (void *)0x10000;
1817 	iov[0].iov_len = 512;
1818 	iov[1].iov_base = (void *)0x20000;
1819 	iov[1].iov_len = 8 * 512;
1820 
1821 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1822 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, iov[0].iov_len);
1823 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1824 
1825 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 8, 1);
1826 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, iov[1].iov_len);
1827 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1828 
1829 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 14, 9, io_done, NULL);
1830 	CU_ASSERT(rc == 0);
1831 	CU_ASSERT(g_io_done == false);
1832 
1833 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1834 	stub_complete_io(2);
1835 	CU_ASSERT(g_io_done == true);
1836 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1837 
1838 	/* Test that a non-vector command is split correctly.
1839 	 * Set up the expected values before calling spdk_bdev_read_blocks
1840 	 */
1841 	bdev->max_segment_size = 512;
1842 	bdev->max_num_segments = 1;
1843 	g_io_done = false;
1844 
1845 	/* Child IO 0 */
1846 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1847 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1848 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1849 
1850 	/* Child IO 1 */
1851 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
1852 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 1 * 512), 512);
1853 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1854 
1855 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1856 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1857 	CU_ASSERT(rc == 0);
1858 	CU_ASSERT(g_io_done == false);
1859 
1860 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1861 	stub_complete_io(2);
1862 	CU_ASSERT(g_io_done == true);
1863 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1864 
1865 	/* Now set up a more complex, multi-vector command that needs to be split,
1866 	 * including splitting iovecs.
1867 	 */
1868 	bdev->max_segment_size = 2 * 512;
1869 	bdev->max_num_segments = 1;
1870 	g_io_done = false;
1871 
1872 	iov[0].iov_base = (void *)0x10000;
1873 	iov[0].iov_len = 2 * 512;
1874 	iov[1].iov_base = (void *)0x20000;
1875 	iov[1].iov_len = 4 * 512;
1876 	iov[2].iov_base = (void *)0x30000;
1877 	iov[2].iov_len = 6 * 512;
1878 
1879 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
1880 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 2);
1881 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1882 
1883 	/* Split iov[1].size to 2 iov entries then split the segments */
1884 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
1885 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, 512 * 2);
1886 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1887 
1888 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 18, 2, 1);
1889 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base + 512 * 2, 512 * 2);
1890 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1891 
1892 	/* Split iov[2].size to 3 iov entries then split the segments */
1893 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 20, 2, 1);
1894 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base, 512 * 2);
1895 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1896 
1897 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 22, 2, 1);
1898 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 2, 512 * 2);
1899 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1900 
1901 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 24, 2, 1);
1902 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 4, 512 * 2);
1903 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1904 
1905 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 12, io_done, NULL);
1906 	CU_ASSERT(rc == 0);
1907 	CU_ASSERT(g_io_done == false);
1908 
1909 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
1910 	stub_complete_io(6);
1911 	CU_ASSERT(g_io_done == true);
1912 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1913 
1914 	/* Test multi vector command that needs to be split by strip and then needs to be
1915 	 * split further due to the capacity of parent IO child iovs.
1916 	 */
1917 	bdev->max_segment_size = 512;
1918 	bdev->max_num_segments = 1;
1919 	g_io_done = false;
1920 
1921 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1922 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1923 		iov[i].iov_len = 512 * 2;
1924 	}
1925 
1926 	/* Each input iov.size is split into 2 iovs,
1927 	 * half of the input iov can fill all child iov entries of a single IO.
1928 	 */
1929 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV / 2; i++) {
1930 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i, 1, 1);
1931 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1932 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1933 
1934 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i + 1, 1, 1);
1935 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1936 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1937 	}
1938 
1939 	/* The remaining iov is split in the second round */
1940 	for (i = BDEV_IO_NUM_CHILD_IOV / 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1941 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2, 1, 1);
1942 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1943 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1944 
1945 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2 + 1, 1, 1);
1946 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1947 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1948 	}
1949 
1950 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
1951 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1952 	CU_ASSERT(rc == 0);
1953 	CU_ASSERT(g_io_done == false);
1954 
1955 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1956 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1957 	CU_ASSERT(g_io_done == false);
1958 
1959 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1960 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1961 	CU_ASSERT(g_io_done == true);
1962 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1963 
1964 	/* A wrong case, a child IO that is divided does
1965 	 * not meet the principle of multiples of block size,
1966 	 * and exits with error
1967 	 */
1968 	bdev->max_segment_size = 512;
1969 	bdev->max_num_segments = 1;
1970 	g_io_done = false;
1971 
1972 	iov[0].iov_base = (void *)0x10000;
1973 	iov[0].iov_len = 512 + 256;
1974 	iov[1].iov_base = (void *)0x20000;
1975 	iov[1].iov_len = 256;
1976 
1977 	/* iov[0] is split to 512 and 256.
1978 	 * 256 is less than a block size, and it is found
1979 	 * in the next round of split that it is the first child IO smaller than
1980 	 * the block size, so the error exit
1981 	 */
1982 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 1, 1);
1983 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512);
1984 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1985 
1986 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 0, 2, io_done, NULL);
1987 	CU_ASSERT(rc == 0);
1988 	CU_ASSERT(g_io_done == false);
1989 
1990 	/* First child IO is OK */
1991 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1992 	stub_complete_io(1);
1993 	CU_ASSERT(g_io_done == true);
1994 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1995 
1996 	/* error exit */
1997 	stub_complete_io(1);
1998 	CU_ASSERT(g_io_done == true);
1999 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
2000 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2001 
2002 	/* Test multi vector command that needs to be split by strip and then needs to be
2003 	 * split further due to the capacity of child iovs.
2004 	 *
2005 	 * In this case, the last two iovs need to be split, but it will exceed the capacity
2006 	 * of child iovs, so it needs to wait until the first batch completed.
2007 	 */
2008 	bdev->max_segment_size = 512;
2009 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2010 	g_io_done = false;
2011 
2012 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2013 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2014 		iov[i].iov_len = 512;
2015 	}
2016 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
2017 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2018 		iov[i].iov_len = 512 * 2;
2019 	}
2020 
2021 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2022 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
2023 	/* 0 ~ (BDEV_IO_NUM_CHILD_IOV - 2) Will not be split */
2024 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2025 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2026 	}
2027 	/* (BDEV_IO_NUM_CHILD_IOV - 2) is split */
2028 	ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, 512);
2029 	ut_expected_io_set_iov(expected_io, i + 1, iov[i].iov_base + 512, 512);
2030 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2031 
2032 	/* Child iov entries exceed the max num of parent IO so split it in next round */
2033 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV, 2, 2);
2034 	ut_expected_io_set_iov(expected_io, 0, iov[i + 1].iov_base, 512);
2035 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base + 512, 512);
2036 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2037 
2038 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
2039 				    BDEV_IO_NUM_CHILD_IOV + 2, io_done, NULL);
2040 	CU_ASSERT(rc == 0);
2041 	CU_ASSERT(g_io_done == false);
2042 
2043 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2044 	stub_complete_io(1);
2045 	CU_ASSERT(g_io_done == false);
2046 
2047 	/* Next round */
2048 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2049 	stub_complete_io(1);
2050 	CU_ASSERT(g_io_done == true);
2051 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2052 
2053 	/* This case is similar to the previous one, but the io composed of
2054 	 * the last few entries of child iov is not enough for a blocklen, so they
2055 	 * cannot be put into this IO, but wait until the next time.
2056 	 */
2057 	bdev->max_segment_size = 512;
2058 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2059 	g_io_done = false;
2060 
2061 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2062 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2063 		iov[i].iov_len = 512;
2064 	}
2065 
2066 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2067 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2068 		iov[i].iov_len = 128;
2069 	}
2070 
2071 	/* First child iovcnt is't BDEV_IO_NUM_CHILD_IOV but BDEV_IO_NUM_CHILD_IOV - 2.
2072 	 * Because the left 2 iov is not enough for a blocklen.
2073 	 */
2074 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2075 					   BDEV_IO_NUM_CHILD_IOV - 2, BDEV_IO_NUM_CHILD_IOV - 2);
2076 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2077 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2078 	}
2079 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2080 
2081 	/* The second child io waits until the end of the first child io before executing.
2082 	 * Because the iovcnt of the two IOs exceeds the child iovcnt of the parent IO.
2083 	 * BDEV_IO_NUM_CHILD_IOV - 2 to BDEV_IO_NUM_CHILD_IOV + 2
2084 	 */
2085 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 2,
2086 					   1, 4);
2087 	ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, iov[i].iov_len);
2088 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base, iov[i + 1].iov_len);
2089 	ut_expected_io_set_iov(expected_io, 2, iov[i + 2].iov_base, iov[i + 2].iov_len);
2090 	ut_expected_io_set_iov(expected_io, 3, iov[i + 3].iov_base, iov[i + 3].iov_len);
2091 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2092 
2093 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2094 				    BDEV_IO_NUM_CHILD_IOV - 1, io_done, NULL);
2095 	CU_ASSERT(rc == 0);
2096 	CU_ASSERT(g_io_done == false);
2097 
2098 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2099 	stub_complete_io(1);
2100 	CU_ASSERT(g_io_done == false);
2101 
2102 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2103 	stub_complete_io(1);
2104 	CU_ASSERT(g_io_done == true);
2105 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2106 
2107 	/* A very complicated case. Each sg entry exceeds max_segment_size and
2108 	 * needs to be split. At the same time, child io must be a multiple of blocklen.
2109 	 * At the same time, child iovcnt exceeds parent iovcnt.
2110 	 */
2111 	bdev->max_segment_size = 512 + 128;
2112 	bdev->max_num_segments = 3;
2113 	g_io_done = false;
2114 
2115 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2116 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2117 		iov[i].iov_len = 512 + 256;
2118 	}
2119 
2120 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2121 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2122 		iov[i].iov_len = 512 + 128;
2123 	}
2124 
2125 	/* Child IOs use 9 entries per for() round and 3 * 9 = 27 child iov entries.
2126 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2127 	 * Generate 9 child IOs.
2128 	 */
2129 	for (i = 0; i < 3; i++) {
2130 		uint32_t j = i * 4;
2131 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6, 2, 3);
2132 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2133 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2134 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2135 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2136 
2137 		/* Child io must be a multiple of blocklen
2138 		 * iov[j + 2] must be split. If the third entry is also added,
2139 		 * the multiple of blocklen cannot be guaranteed. But it still
2140 		 * occupies one iov entry of the parent child iov.
2141 		 */
2142 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 2, 2, 2);
2143 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2144 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2145 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2146 
2147 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 4, 2, 3);
2148 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2149 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2150 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2151 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2152 	}
2153 
2154 	/* Child iov position at 27, the 10th child IO
2155 	 * iov entry index is 3 * 4 and offset is 3 * 6
2156 	 */
2157 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 18, 2, 3);
2158 	ut_expected_io_set_iov(expected_io, 0, iov[12].iov_base, 640);
2159 	ut_expected_io_set_iov(expected_io, 1, iov[12].iov_base + 640, 128);
2160 	ut_expected_io_set_iov(expected_io, 2, iov[13].iov_base, 256);
2161 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2162 
2163 	/* Child iov position at 30, the 11th child IO */
2164 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 20, 2, 2);
2165 	ut_expected_io_set_iov(expected_io, 0, iov[13].iov_base + 256, 512);
2166 	ut_expected_io_set_iov(expected_io, 1, iov[14].iov_base, 512);
2167 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2168 
2169 	/* The 2nd split round and iovpos is 0, the 12th child IO */
2170 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 22, 2, 3);
2171 	ut_expected_io_set_iov(expected_io, 0, iov[14].iov_base + 512, 256);
2172 	ut_expected_io_set_iov(expected_io, 1, iov[15].iov_base, 640);
2173 	ut_expected_io_set_iov(expected_io, 2, iov[15].iov_base + 640, 128);
2174 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2175 
2176 	/* Consume 9 child IOs and 27 child iov entries.
2177 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2178 	 * Parent IO iov index start from 16 and block offset start from 24
2179 	 */
2180 	for (i = 0; i < 3; i++) {
2181 		uint32_t j = i * 4 + 16;
2182 		uint32_t offset = i * 6 + 24;
2183 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, 2, 3);
2184 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2185 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2186 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2187 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2188 
2189 		/* Child io must be a multiple of blocklen
2190 		 * iov[j + 2] must be split. If the third entry is also added,
2191 		 * the multiple of blocklen cannot be guaranteed. But it still
2192 		 * occupies one iov entry of the parent child iov.
2193 		 */
2194 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 2, 2, 2);
2195 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2196 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2197 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2198 
2199 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 4, 2, 3);
2200 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2201 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2202 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2203 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2204 	}
2205 
2206 	/* The 22th child IO, child iov position at 30 */
2207 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 42, 1, 1);
2208 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base, 512);
2209 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2210 
2211 	/* The third round */
2212 	/* Here is the 23nd child IO and child iovpos is 0 */
2213 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 43, 2, 3);
2214 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base + 512, 256);
2215 	ut_expected_io_set_iov(expected_io, 1, iov[29].iov_base, 640);
2216 	ut_expected_io_set_iov(expected_io, 2, iov[29].iov_base + 640, 128);
2217 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2218 
2219 	/* The 24th child IO */
2220 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 45, 3, 3);
2221 	ut_expected_io_set_iov(expected_io, 0, iov[30].iov_base, 640);
2222 	ut_expected_io_set_iov(expected_io, 1, iov[31].iov_base, 640);
2223 	ut_expected_io_set_iov(expected_io, 2, iov[32].iov_base, 256);
2224 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2225 
2226 	/* The 25th child IO */
2227 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 48, 2, 2);
2228 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base + 256, 384);
2229 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 640);
2230 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2231 
2232 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2233 				    50, io_done, NULL);
2234 	CU_ASSERT(rc == 0);
2235 	CU_ASSERT(g_io_done == false);
2236 
2237 	/* Parent IO supports up to 32 child iovs, so it is calculated that
2238 	 * a maximum of 11 IOs can be split at a time, and the
2239 	 * splitting will continue after the first batch is over.
2240 	 */
2241 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2242 	stub_complete_io(11);
2243 	CU_ASSERT(g_io_done == false);
2244 
2245 	/* The 2nd round */
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 last round */
2251 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2252 	stub_complete_io(3);
2253 	CU_ASSERT(g_io_done == true);
2254 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2255 
2256 	/* Test an WRITE_ZEROES.  This should also not be split. */
2257 	bdev->max_segment_size = 512;
2258 	bdev->max_num_segments = 1;
2259 	g_io_done = false;
2260 
2261 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
2262 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2263 
2264 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
2265 	CU_ASSERT(rc == 0);
2266 	CU_ASSERT(g_io_done == false);
2267 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2268 	stub_complete_io(1);
2269 	CU_ASSERT(g_io_done == true);
2270 
2271 	/* Test an UNMAP.  This should also not be split. */
2272 	g_io_done = false;
2273 
2274 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 4, 0);
2275 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2276 
2277 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 4, io_done, NULL);
2278 	CU_ASSERT(rc == 0);
2279 	CU_ASSERT(g_io_done == false);
2280 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2281 	stub_complete_io(1);
2282 	CU_ASSERT(g_io_done == true);
2283 
2284 	/* Test a FLUSH.  This should also not be split. */
2285 	g_io_done = false;
2286 
2287 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 4, 0);
2288 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2289 
2290 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
2291 	CU_ASSERT(rc == 0);
2292 	CU_ASSERT(g_io_done == false);
2293 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2294 	stub_complete_io(1);
2295 	CU_ASSERT(g_io_done == true);
2296 
2297 	spdk_put_io_channel(io_ch);
2298 	spdk_bdev_close(desc);
2299 	free_bdev(bdev);
2300 	spdk_bdev_finish(bdev_fini_cb, NULL);
2301 	poll_threads();
2302 }
2303 
2304 static void
2305 bdev_io_mix_split_test(void)
2306 {
2307 	struct spdk_bdev *bdev;
2308 	struct spdk_bdev_desc *desc = NULL;
2309 	struct spdk_io_channel *io_ch;
2310 	struct spdk_bdev_opts bdev_opts = {};
2311 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
2312 	struct ut_expected_io *expected_io;
2313 	uint64_t i;
2314 	int rc;
2315 
2316 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2317 	bdev_opts.bdev_io_pool_size = 512;
2318 	bdev_opts.bdev_io_cache_size = 64;
2319 
2320 	rc = spdk_bdev_set_opts(&bdev_opts);
2321 	CU_ASSERT(rc == 0);
2322 	spdk_bdev_initialize(bdev_init_cb, NULL);
2323 
2324 	bdev = allocate_bdev("bdev0");
2325 
2326 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
2327 	CU_ASSERT(rc == 0);
2328 	SPDK_CU_ASSERT_FATAL(desc != NULL);
2329 	io_ch = spdk_bdev_get_io_channel(desc);
2330 	CU_ASSERT(io_ch != NULL);
2331 
2332 	/* First case optimal_io_boundary == max_segment_size * max_num_segments */
2333 	bdev->split_on_optimal_io_boundary = true;
2334 	bdev->optimal_io_boundary = 16;
2335 
2336 	bdev->max_segment_size = 512;
2337 	bdev->max_num_segments = 16;
2338 	g_io_done = false;
2339 
2340 	/* IO crossing the IO boundary requires split
2341 	 * Total 2 child IOs.
2342 	 */
2343 
2344 	/* The 1st child IO split the segment_size to multiple segment entry */
2345 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
2346 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
2347 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
2348 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2349 
2350 	/* The 2nd child IO split the segment_size to multiple segment entry */
2351 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 2);
2352 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 512);
2353 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 3 * 512), 512);
2354 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2355 
2356 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 4, io_done, NULL);
2357 	CU_ASSERT(rc == 0);
2358 	CU_ASSERT(g_io_done == false);
2359 
2360 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2361 	stub_complete_io(2);
2362 	CU_ASSERT(g_io_done == true);
2363 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2364 
2365 	/* Second case optimal_io_boundary > max_segment_size * max_num_segments */
2366 	bdev->max_segment_size = 15 * 512;
2367 	bdev->max_num_segments = 1;
2368 	g_io_done = false;
2369 
2370 	/* IO crossing the IO boundary requires split.
2371 	 * The 1st child IO segment size exceeds the max_segment_size,
2372 	 * So 1st child IO will be splitted to multiple segment entry.
2373 	 * Then it split to 2 child IOs because of the max_num_segments.
2374 	 * Total 3 child IOs.
2375 	 */
2376 
2377 	/* The first 2 IOs are in an IO boundary.
2378 	 * Because the optimal_io_boundary > max_segment_size * max_num_segments
2379 	 * So it split to the first 2 IOs.
2380 	 */
2381 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 15, 1);
2382 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 15);
2383 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2384 
2385 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
2386 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 15), 512);
2387 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2388 
2389 	/* The 3rd Child IO is because of the io boundary */
2390 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2391 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2392 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2393 
2394 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2395 	CU_ASSERT(rc == 0);
2396 	CU_ASSERT(g_io_done == false);
2397 
2398 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2399 	stub_complete_io(3);
2400 	CU_ASSERT(g_io_done == true);
2401 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2402 
2403 	/* Third case optimal_io_boundary < max_segment_size * max_num_segments */
2404 	bdev->max_segment_size = 17 * 512;
2405 	bdev->max_num_segments = 1;
2406 	g_io_done = false;
2407 
2408 	/* IO crossing the IO boundary requires split.
2409 	 * Child IO does not split.
2410 	 * Total 2 child IOs.
2411 	 */
2412 
2413 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 16, 1);
2414 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 16);
2415 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2416 
2417 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2418 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2419 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2420 
2421 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2422 	CU_ASSERT(rc == 0);
2423 	CU_ASSERT(g_io_done == false);
2424 
2425 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2426 	stub_complete_io(2);
2427 	CU_ASSERT(g_io_done == true);
2428 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2429 
2430 	/* Now set up a more complex, multi-vector command that needs to be split,
2431 	 * including splitting iovecs.
2432 	 * optimal_io_boundary < max_segment_size * max_num_segments
2433 	 */
2434 	bdev->max_segment_size = 3 * 512;
2435 	bdev->max_num_segments = 6;
2436 	g_io_done = false;
2437 
2438 	iov[0].iov_base = (void *)0x10000;
2439 	iov[0].iov_len = 4 * 512;
2440 	iov[1].iov_base = (void *)0x20000;
2441 	iov[1].iov_len = 4 * 512;
2442 	iov[2].iov_base = (void *)0x30000;
2443 	iov[2].iov_len = 10 * 512;
2444 
2445 	/* IO crossing the IO boundary requires split.
2446 	 * The 1st child IO segment size exceeds the max_segment_size and after
2447 	 * splitting segment_size, the num_segments exceeds max_num_segments.
2448 	 * So 1st child IO will be splitted to 2 child IOs.
2449 	 * Total 3 child IOs.
2450 	 */
2451 
2452 	/* The first 2 IOs are in an IO boundary.
2453 	 * After splitting segmemt size the segment num exceeds.
2454 	 * So it splits to 2 child IOs.
2455 	 */
2456 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 14, 6);
2457 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 3);
2458 	ut_expected_io_set_iov(expected_io, 1, iov[0].iov_base + 512 * 3, 512);
2459 	ut_expected_io_set_iov(expected_io, 2, iov[1].iov_base, 512 * 3);
2460 	ut_expected_io_set_iov(expected_io, 3, iov[1].iov_base + 512 * 3, 512);
2461 	ut_expected_io_set_iov(expected_io, 4, iov[2].iov_base, 512 * 3);
2462 	ut_expected_io_set_iov(expected_io, 5, iov[2].iov_base + 512 * 3, 512 * 3);
2463 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2464 
2465 	/* The 2nd child IO has the left segment entry */
2466 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
2467 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 6, 512 * 2);
2468 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2469 
2470 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
2471 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 8, 512 * 2);
2472 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2473 
2474 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 0, 18, io_done, NULL);
2475 	CU_ASSERT(rc == 0);
2476 	CU_ASSERT(g_io_done == false);
2477 
2478 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2479 	stub_complete_io(3);
2480 	CU_ASSERT(g_io_done == true);
2481 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2482 
2483 	/* A very complicated case. Each sg entry exceeds max_segment_size
2484 	 * and split on io boundary.
2485 	 * optimal_io_boundary < max_segment_size * max_num_segments
2486 	 */
2487 	bdev->max_segment_size = 3 * 512;
2488 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2489 	g_io_done = false;
2490 
2491 	for (i = 0; i < 20; i++) {
2492 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2493 		iov[i].iov_len = 512 * 4;
2494 	}
2495 
2496 	/* IO crossing the IO boundary requires split.
2497 	 * 80 block length can split 5 child IOs base on offset and IO boundary.
2498 	 * Each iov entry needs to be splitted to 2 entries because of max_segment_size
2499 	 * Total 5 child IOs.
2500 	 */
2501 
2502 	/* 4 iov entries are in an IO boundary and each iov entry splits to 2.
2503 	 * So each child IO occupies 8 child iov entries.
2504 	 */
2505 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 16, 8);
2506 	for (i = 0; i < 4; i++) {
2507 		int iovcnt = i * 2;
2508 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2509 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2510 	}
2511 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2512 
2513 	/* 2nd child IO and total 16 child iov entries of parent IO */
2514 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 8);
2515 	for (i = 4; i < 8; i++) {
2516 		int iovcnt = (i - 4) * 2;
2517 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2518 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2519 	}
2520 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2521 
2522 	/* 3rd child IO and total 24 child iov entries of parent IO */
2523 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 16, 8);
2524 	for (i = 8; i < 12; i++) {
2525 		int iovcnt = (i - 8) * 2;
2526 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2527 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2528 	}
2529 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2530 
2531 	/* 4th child IO and total 32 child iov entries of parent IO */
2532 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 48, 16, 8);
2533 	for (i = 12; i < 16; i++) {
2534 		int iovcnt = (i - 12) * 2;
2535 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2536 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2537 	}
2538 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2539 
2540 	/* 5th child IO and because of the child iov entry it should be splitted
2541 	 * in next round.
2542 	 */
2543 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 64, 16, 8);
2544 	for (i = 16; i < 20; i++) {
2545 		int iovcnt = (i - 16) * 2;
2546 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2547 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2548 	}
2549 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2550 
2551 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 20, 0, 80, io_done, NULL);
2552 	CU_ASSERT(rc == 0);
2553 	CU_ASSERT(g_io_done == false);
2554 
2555 	/* First split round */
2556 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
2557 	stub_complete_io(4);
2558 	CU_ASSERT(g_io_done == false);
2559 
2560 	/* Second split round */
2561 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2562 	stub_complete_io(1);
2563 	CU_ASSERT(g_io_done == true);
2564 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2565 
2566 	spdk_put_io_channel(io_ch);
2567 	spdk_bdev_close(desc);
2568 	free_bdev(bdev);
2569 	spdk_bdev_finish(bdev_fini_cb, NULL);
2570 	poll_threads();
2571 }
2572 
2573 static void
2574 bdev_io_split_with_io_wait(void)
2575 {
2576 	struct spdk_bdev *bdev;
2577 	struct spdk_bdev_desc *desc = NULL;
2578 	struct spdk_io_channel *io_ch;
2579 	struct spdk_bdev_channel *channel;
2580 	struct spdk_bdev_mgmt_channel *mgmt_ch;
2581 	struct spdk_bdev_opts bdev_opts = {};
2582 	struct iovec iov[3];
2583 	struct ut_expected_io *expected_io;
2584 	int rc;
2585 
2586 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2587 	bdev_opts.bdev_io_pool_size = 2;
2588 	bdev_opts.bdev_io_cache_size = 1;
2589 
2590 	rc = spdk_bdev_set_opts(&bdev_opts);
2591 	CU_ASSERT(rc == 0);
2592 	spdk_bdev_initialize(bdev_init_cb, NULL);
2593 
2594 	bdev = allocate_bdev("bdev0");
2595 
2596 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2597 	CU_ASSERT(rc == 0);
2598 	CU_ASSERT(desc != NULL);
2599 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2600 	io_ch = spdk_bdev_get_io_channel(desc);
2601 	CU_ASSERT(io_ch != NULL);
2602 	channel = spdk_io_channel_get_ctx(io_ch);
2603 	mgmt_ch = channel->shared_resource->mgmt_ch;
2604 
2605 	bdev->optimal_io_boundary = 16;
2606 	bdev->split_on_optimal_io_boundary = true;
2607 
2608 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
2609 	CU_ASSERT(rc == 0);
2610 
2611 	/* Now test that a single-vector command is split correctly.
2612 	 * Offset 14, length 8, payload 0xF000
2613 	 *  Child - Offset 14, length 2, payload 0xF000
2614 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
2615 	 *
2616 	 * Set up the expected values before calling spdk_bdev_read_blocks
2617 	 */
2618 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
2619 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
2620 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2621 
2622 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
2623 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
2624 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2625 
2626 	/* The following children will be submitted sequentially due to the capacity of
2627 	 * spdk_bdev_io.
2628 	 */
2629 
2630 	/* The first child I/O will be queued to wait until an spdk_bdev_io becomes available */
2631 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
2632 	CU_ASSERT(rc == 0);
2633 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2634 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2635 
2636 	/* Completing the first read I/O will submit the first child */
2637 	stub_complete_io(1);
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 child will submit the second child */
2642 	stub_complete_io(1);
2643 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2644 
2645 	/* Complete the second child I/O.  This should result in our callback getting
2646 	 * invoked since the parent I/O is now complete.
2647 	 */
2648 	stub_complete_io(1);
2649 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2650 
2651 	/* Now set up a more complex, multi-vector command that needs to be split,
2652 	 *  including splitting iovecs.
2653 	 */
2654 	iov[0].iov_base = (void *)0x10000;
2655 	iov[0].iov_len = 512;
2656 	iov[1].iov_base = (void *)0x20000;
2657 	iov[1].iov_len = 20 * 512;
2658 	iov[2].iov_base = (void *)0x30000;
2659 	iov[2].iov_len = 11 * 512;
2660 
2661 	g_io_done = false;
2662 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
2663 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
2664 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
2665 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2666 
2667 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
2668 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
2669 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2670 
2671 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
2672 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
2673 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
2674 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2675 
2676 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 32, io_done, NULL);
2677 	CU_ASSERT(rc == 0);
2678 	CU_ASSERT(g_io_done == false);
2679 
2680 	/* The following children will be submitted sequentially due to the capacity of
2681 	 * spdk_bdev_io.
2682 	 */
2683 
2684 	/* Completing the first child will submit the second child */
2685 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2686 	stub_complete_io(1);
2687 	CU_ASSERT(g_io_done == false);
2688 
2689 	/* Completing the second child will submit the third 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 third child will result in our callback getting invoked
2695 	 * since the parent I/O is now complete.
2696 	 */
2697 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2698 	stub_complete_io(1);
2699 	CU_ASSERT(g_io_done == true);
2700 
2701 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
2702 
2703 	spdk_put_io_channel(io_ch);
2704 	spdk_bdev_close(desc);
2705 	free_bdev(bdev);
2706 	spdk_bdev_finish(bdev_fini_cb, NULL);
2707 	poll_threads();
2708 }
2709 
2710 static void
2711 bdev_io_alignment(void)
2712 {
2713 	struct spdk_bdev *bdev;
2714 	struct spdk_bdev_desc *desc = NULL;
2715 	struct spdk_io_channel *io_ch;
2716 	struct spdk_bdev_opts bdev_opts = {};
2717 	int rc;
2718 	void *buf = NULL;
2719 	struct iovec iovs[2];
2720 	int iovcnt;
2721 	uint64_t alignment;
2722 
2723 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2724 	bdev_opts.bdev_io_pool_size = 20;
2725 	bdev_opts.bdev_io_cache_size = 2;
2726 
2727 	rc = spdk_bdev_set_opts(&bdev_opts);
2728 	CU_ASSERT(rc == 0);
2729 	spdk_bdev_initialize(bdev_init_cb, NULL);
2730 
2731 	fn_table.submit_request = stub_submit_request_get_buf;
2732 	bdev = allocate_bdev("bdev0");
2733 
2734 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2735 	CU_ASSERT(rc == 0);
2736 	CU_ASSERT(desc != NULL);
2737 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2738 	io_ch = spdk_bdev_get_io_channel(desc);
2739 	CU_ASSERT(io_ch != NULL);
2740 
2741 	/* Create aligned buffer */
2742 	rc = posix_memalign(&buf, 4096, 8192);
2743 	SPDK_CU_ASSERT_FATAL(rc == 0);
2744 
2745 	/* Pass aligned single buffer with no alignment required */
2746 	alignment = 1;
2747 	bdev->required_alignment = spdk_u32log2(alignment);
2748 
2749 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2750 	CU_ASSERT(rc == 0);
2751 	stub_complete_io(1);
2752 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2753 				    alignment));
2754 
2755 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2756 	CU_ASSERT(rc == 0);
2757 	stub_complete_io(1);
2758 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2759 				    alignment));
2760 
2761 	/* Pass unaligned single buffer with no alignment required */
2762 	alignment = 1;
2763 	bdev->required_alignment = spdk_u32log2(alignment);
2764 
2765 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2766 	CU_ASSERT(rc == 0);
2767 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2768 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2769 	stub_complete_io(1);
2770 
2771 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2772 	CU_ASSERT(rc == 0);
2773 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2774 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2775 	stub_complete_io(1);
2776 
2777 	/* Pass unaligned single buffer with 512 alignment required */
2778 	alignment = 512;
2779 	bdev->required_alignment = spdk_u32log2(alignment);
2780 
2781 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2782 	CU_ASSERT(rc == 0);
2783 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2784 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2785 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2786 				    alignment));
2787 	stub_complete_io(1);
2788 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2789 
2790 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2791 	CU_ASSERT(rc == 0);
2792 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2793 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2794 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2795 				    alignment));
2796 	stub_complete_io(1);
2797 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2798 
2799 	/* Pass unaligned single buffer with 4096 alignment required */
2800 	alignment = 4096;
2801 	bdev->required_alignment = spdk_u32log2(alignment);
2802 
2803 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2804 	CU_ASSERT(rc == 0);
2805 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2806 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2807 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2808 				    alignment));
2809 	stub_complete_io(1);
2810 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2811 
2812 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2813 	CU_ASSERT(rc == 0);
2814 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2815 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2816 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2817 				    alignment));
2818 	stub_complete_io(1);
2819 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2820 
2821 	/* Pass aligned iovs with no alignment required */
2822 	alignment = 1;
2823 	bdev->required_alignment = spdk_u32log2(alignment);
2824 
2825 	iovcnt = 1;
2826 	iovs[0].iov_base = buf;
2827 	iovs[0].iov_len = 512;
2828 
2829 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2830 	CU_ASSERT(rc == 0);
2831 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2832 	stub_complete_io(1);
2833 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2834 
2835 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2836 	CU_ASSERT(rc == 0);
2837 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2838 	stub_complete_io(1);
2839 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2840 
2841 	/* Pass unaligned iovs with no alignment required */
2842 	alignment = 1;
2843 	bdev->required_alignment = spdk_u32log2(alignment);
2844 
2845 	iovcnt = 2;
2846 	iovs[0].iov_base = buf + 16;
2847 	iovs[0].iov_len = 256;
2848 	iovs[1].iov_base = buf + 16 + 256 + 32;
2849 	iovs[1].iov_len = 256;
2850 
2851 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2852 	CU_ASSERT(rc == 0);
2853 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2854 	stub_complete_io(1);
2855 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2856 
2857 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2858 	CU_ASSERT(rc == 0);
2859 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2860 	stub_complete_io(1);
2861 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2862 
2863 	/* Pass unaligned iov with 2048 alignment required */
2864 	alignment = 2048;
2865 	bdev->required_alignment = spdk_u32log2(alignment);
2866 
2867 	iovcnt = 2;
2868 	iovs[0].iov_base = buf + 16;
2869 	iovs[0].iov_len = 256;
2870 	iovs[1].iov_base = buf + 16 + 256 + 32;
2871 	iovs[1].iov_len = 256;
2872 
2873 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2874 	CU_ASSERT(rc == 0);
2875 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2876 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2877 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2878 				    alignment));
2879 	stub_complete_io(1);
2880 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2881 
2882 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2883 	CU_ASSERT(rc == 0);
2884 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2885 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2886 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2887 				    alignment));
2888 	stub_complete_io(1);
2889 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2890 
2891 	/* Pass iov without allocated buffer without alignment required */
2892 	alignment = 1;
2893 	bdev->required_alignment = spdk_u32log2(alignment);
2894 
2895 	iovcnt = 1;
2896 	iovs[0].iov_base = NULL;
2897 	iovs[0].iov_len = 0;
2898 
2899 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2900 	CU_ASSERT(rc == 0);
2901 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2902 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2903 				    alignment));
2904 	stub_complete_io(1);
2905 
2906 	/* Pass iov without allocated buffer with 1024 alignment required */
2907 	alignment = 1024;
2908 	bdev->required_alignment = spdk_u32log2(alignment);
2909 
2910 	iovcnt = 1;
2911 	iovs[0].iov_base = NULL;
2912 	iovs[0].iov_len = 0;
2913 
2914 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2915 	CU_ASSERT(rc == 0);
2916 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2917 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2918 				    alignment));
2919 	stub_complete_io(1);
2920 
2921 	spdk_put_io_channel(io_ch);
2922 	spdk_bdev_close(desc);
2923 	free_bdev(bdev);
2924 	fn_table.submit_request = stub_submit_request;
2925 	spdk_bdev_finish(bdev_fini_cb, NULL);
2926 	poll_threads();
2927 
2928 	free(buf);
2929 }
2930 
2931 static void
2932 bdev_io_alignment_with_boundary(void)
2933 {
2934 	struct spdk_bdev *bdev;
2935 	struct spdk_bdev_desc *desc = NULL;
2936 	struct spdk_io_channel *io_ch;
2937 	struct spdk_bdev_opts bdev_opts = {};
2938 	int rc;
2939 	void *buf = NULL;
2940 	struct iovec iovs[2];
2941 	int iovcnt;
2942 	uint64_t alignment;
2943 
2944 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2945 	bdev_opts.bdev_io_pool_size = 20;
2946 	bdev_opts.bdev_io_cache_size = 2;
2947 
2948 	bdev_opts.opts_size = sizeof(bdev_opts);
2949 	rc = spdk_bdev_set_opts(&bdev_opts);
2950 	CU_ASSERT(rc == 0);
2951 	spdk_bdev_initialize(bdev_init_cb, NULL);
2952 
2953 	fn_table.submit_request = stub_submit_request_get_buf;
2954 	bdev = allocate_bdev("bdev0");
2955 
2956 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2957 	CU_ASSERT(rc == 0);
2958 	CU_ASSERT(desc != NULL);
2959 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2960 	io_ch = spdk_bdev_get_io_channel(desc);
2961 	CU_ASSERT(io_ch != NULL);
2962 
2963 	/* Create aligned buffer */
2964 	rc = posix_memalign(&buf, 4096, 131072);
2965 	SPDK_CU_ASSERT_FATAL(rc == 0);
2966 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2967 
2968 	/* 512 * 3 with 2 IO boundary, allocate small data buffer from bdev layer */
2969 	alignment = 512;
2970 	bdev->required_alignment = spdk_u32log2(alignment);
2971 	bdev->optimal_io_boundary = 2;
2972 	bdev->split_on_optimal_io_boundary = true;
2973 
2974 	iovcnt = 1;
2975 	iovs[0].iov_base = NULL;
2976 	iovs[0].iov_len = 512 * 3;
2977 
2978 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
2979 	CU_ASSERT(rc == 0);
2980 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2981 	stub_complete_io(2);
2982 
2983 	/* 8KiB with 16 IO boundary, allocate large data buffer from bdev layer */
2984 	alignment = 512;
2985 	bdev->required_alignment = spdk_u32log2(alignment);
2986 	bdev->optimal_io_boundary = 16;
2987 	bdev->split_on_optimal_io_boundary = true;
2988 
2989 	iovcnt = 1;
2990 	iovs[0].iov_base = NULL;
2991 	iovs[0].iov_len = 512 * 16;
2992 
2993 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 16, io_done, NULL);
2994 	CU_ASSERT(rc == 0);
2995 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2996 	stub_complete_io(2);
2997 
2998 	/* 512 * 160 with 128 IO boundary, 63.5KiB + 16.5KiB for the two children requests */
2999 	alignment = 512;
3000 	bdev->required_alignment = spdk_u32log2(alignment);
3001 	bdev->optimal_io_boundary = 128;
3002 	bdev->split_on_optimal_io_boundary = true;
3003 
3004 	iovcnt = 1;
3005 	iovs[0].iov_base = buf + 16;
3006 	iovs[0].iov_len = 512 * 160;
3007 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3008 	CU_ASSERT(rc == 0);
3009 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3010 	stub_complete_io(2);
3011 
3012 	/* 512 * 3 with 2 IO boundary */
3013 	alignment = 512;
3014 	bdev->required_alignment = spdk_u32log2(alignment);
3015 	bdev->optimal_io_boundary = 2;
3016 	bdev->split_on_optimal_io_boundary = true;
3017 
3018 	iovcnt = 2;
3019 	iovs[0].iov_base = buf + 16;
3020 	iovs[0].iov_len = 512;
3021 	iovs[1].iov_base = buf + 16 + 512 + 32;
3022 	iovs[1].iov_len = 1024;
3023 
3024 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3025 	CU_ASSERT(rc == 0);
3026 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3027 	stub_complete_io(2);
3028 
3029 	rc = spdk_bdev_readv_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 	/* 512 * 64 with 32 IO boundary */
3035 	bdev->optimal_io_boundary = 32;
3036 	iovcnt = 2;
3037 	iovs[0].iov_base = buf + 16;
3038 	iovs[0].iov_len = 16384;
3039 	iovs[1].iov_base = buf + 16 + 16384 + 32;
3040 	iovs[1].iov_len = 16384;
3041 
3042 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3043 	CU_ASSERT(rc == 0);
3044 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3045 	stub_complete_io(3);
3046 
3047 	rc = spdk_bdev_readv_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 	/* 512 * 160 with 32 IO boundary */
3053 	iovcnt = 1;
3054 	iovs[0].iov_base = buf + 16;
3055 	iovs[0].iov_len = 16384 + 65536;
3056 
3057 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3058 	CU_ASSERT(rc == 0);
3059 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
3060 	stub_complete_io(6);
3061 
3062 	spdk_put_io_channel(io_ch);
3063 	spdk_bdev_close(desc);
3064 	free_bdev(bdev);
3065 	fn_table.submit_request = stub_submit_request;
3066 	spdk_bdev_finish(bdev_fini_cb, NULL);
3067 	poll_threads();
3068 
3069 	free(buf);
3070 }
3071 
3072 static void
3073 histogram_status_cb(void *cb_arg, int status)
3074 {
3075 	g_status = status;
3076 }
3077 
3078 static void
3079 histogram_data_cb(void *cb_arg, int status, struct spdk_histogram_data *histogram)
3080 {
3081 	g_status = status;
3082 	g_histogram = histogram;
3083 }
3084 
3085 static void
3086 histogram_io_count(void *ctx, uint64_t start, uint64_t end, uint64_t count,
3087 		   uint64_t total, uint64_t so_far)
3088 {
3089 	g_count += count;
3090 }
3091 
3092 static void
3093 bdev_histograms(void)
3094 {
3095 	struct spdk_bdev *bdev;
3096 	struct spdk_bdev_desc *desc = NULL;
3097 	struct spdk_io_channel *ch;
3098 	struct spdk_histogram_data *histogram;
3099 	uint8_t buf[4096];
3100 	int rc;
3101 
3102 	spdk_bdev_initialize(bdev_init_cb, NULL);
3103 
3104 	bdev = allocate_bdev("bdev");
3105 
3106 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3107 	CU_ASSERT(rc == 0);
3108 	CU_ASSERT(desc != NULL);
3109 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3110 
3111 	ch = spdk_bdev_get_io_channel(desc);
3112 	CU_ASSERT(ch != NULL);
3113 
3114 	/* Enable histogram */
3115 	g_status = -1;
3116 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, true);
3117 	poll_threads();
3118 	CU_ASSERT(g_status == 0);
3119 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3120 
3121 	/* Allocate histogram */
3122 	histogram = spdk_histogram_data_alloc();
3123 	SPDK_CU_ASSERT_FATAL(histogram != NULL);
3124 
3125 	/* Check if histogram is zeroed */
3126 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3127 	poll_threads();
3128 	CU_ASSERT(g_status == 0);
3129 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3130 
3131 	g_count = 0;
3132 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3133 
3134 	CU_ASSERT(g_count == 0);
3135 
3136 	rc = spdk_bdev_write_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3137 	CU_ASSERT(rc == 0);
3138 
3139 	spdk_delay_us(10);
3140 	stub_complete_io(1);
3141 	poll_threads();
3142 
3143 	rc = spdk_bdev_read_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3144 	CU_ASSERT(rc == 0);
3145 
3146 	spdk_delay_us(10);
3147 	stub_complete_io(1);
3148 	poll_threads();
3149 
3150 	/* Check if histogram gathered data from all I/O channels */
3151 	g_histogram = NULL;
3152 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3153 	poll_threads();
3154 	CU_ASSERT(g_status == 0);
3155 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3156 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3157 
3158 	g_count = 0;
3159 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3160 	CU_ASSERT(g_count == 2);
3161 
3162 	/* Disable histogram */
3163 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, false);
3164 	poll_threads();
3165 	CU_ASSERT(g_status == 0);
3166 	CU_ASSERT(bdev->internal.histogram_enabled == false);
3167 
3168 	/* Try to run histogram commands on disabled bdev */
3169 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3170 	poll_threads();
3171 	CU_ASSERT(g_status == -EFAULT);
3172 
3173 	spdk_histogram_data_free(histogram);
3174 	spdk_put_io_channel(ch);
3175 	spdk_bdev_close(desc);
3176 	free_bdev(bdev);
3177 	spdk_bdev_finish(bdev_fini_cb, NULL);
3178 	poll_threads();
3179 }
3180 
3181 static void
3182 _bdev_compare(bool emulated)
3183 {
3184 	struct spdk_bdev *bdev;
3185 	struct spdk_bdev_desc *desc = NULL;
3186 	struct spdk_io_channel *ioch;
3187 	struct ut_expected_io *expected_io;
3188 	uint64_t offset, num_blocks;
3189 	uint32_t num_completed;
3190 	char aa_buf[512];
3191 	char bb_buf[512];
3192 	struct iovec compare_iov;
3193 	uint8_t io_type;
3194 	int rc;
3195 
3196 	if (emulated) {
3197 		io_type = SPDK_BDEV_IO_TYPE_READ;
3198 	} else {
3199 		io_type = SPDK_BDEV_IO_TYPE_COMPARE;
3200 	}
3201 
3202 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3203 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3204 
3205 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = !emulated;
3206 
3207 	spdk_bdev_initialize(bdev_init_cb, NULL);
3208 	fn_table.submit_request = stub_submit_request_get_buf;
3209 	bdev = allocate_bdev("bdev");
3210 
3211 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3212 	CU_ASSERT_EQUAL(rc, 0);
3213 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3214 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3215 	ioch = spdk_bdev_get_io_channel(desc);
3216 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3217 
3218 	fn_table.submit_request = stub_submit_request_get_buf;
3219 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3220 
3221 	offset = 50;
3222 	num_blocks = 1;
3223 	compare_iov.iov_base = aa_buf;
3224 	compare_iov.iov_len = sizeof(aa_buf);
3225 
3226 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3227 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3228 
3229 	g_io_done = false;
3230 	g_compare_read_buf = aa_buf;
3231 	g_compare_read_buf_len = sizeof(aa_buf);
3232 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3233 	CU_ASSERT_EQUAL(rc, 0);
3234 	num_completed = stub_complete_io(1);
3235 	CU_ASSERT_EQUAL(num_completed, 1);
3236 	CU_ASSERT(g_io_done == true);
3237 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3238 
3239 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3240 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3241 
3242 	g_io_done = false;
3243 	g_compare_read_buf = bb_buf;
3244 	g_compare_read_buf_len = sizeof(bb_buf);
3245 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3246 	CU_ASSERT_EQUAL(rc, 0);
3247 	num_completed = stub_complete_io(1);
3248 	CU_ASSERT_EQUAL(num_completed, 1);
3249 	CU_ASSERT(g_io_done == true);
3250 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3251 
3252 	spdk_put_io_channel(ioch);
3253 	spdk_bdev_close(desc);
3254 	free_bdev(bdev);
3255 	fn_table.submit_request = stub_submit_request;
3256 	spdk_bdev_finish(bdev_fini_cb, NULL);
3257 	poll_threads();
3258 
3259 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3260 
3261 	g_compare_read_buf = NULL;
3262 }
3263 
3264 static void
3265 bdev_compare(void)
3266 {
3267 	_bdev_compare(true);
3268 	_bdev_compare(false);
3269 }
3270 
3271 static void
3272 bdev_compare_and_write(void)
3273 {
3274 	struct spdk_bdev *bdev;
3275 	struct spdk_bdev_desc *desc = NULL;
3276 	struct spdk_io_channel *ioch;
3277 	struct ut_expected_io *expected_io;
3278 	uint64_t offset, num_blocks;
3279 	uint32_t num_completed;
3280 	char aa_buf[512];
3281 	char bb_buf[512];
3282 	char cc_buf[512];
3283 	char write_buf[512];
3284 	struct iovec compare_iov;
3285 	struct iovec write_iov;
3286 	int rc;
3287 
3288 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3289 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3290 	memset(cc_buf, 0xcc, sizeof(cc_buf));
3291 
3292 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = false;
3293 
3294 	spdk_bdev_initialize(bdev_init_cb, NULL);
3295 	fn_table.submit_request = stub_submit_request_get_buf;
3296 	bdev = allocate_bdev("bdev");
3297 
3298 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3299 	CU_ASSERT_EQUAL(rc, 0);
3300 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3301 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3302 	ioch = spdk_bdev_get_io_channel(desc);
3303 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3304 
3305 	fn_table.submit_request = stub_submit_request_get_buf;
3306 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3307 
3308 	offset = 50;
3309 	num_blocks = 1;
3310 	compare_iov.iov_base = aa_buf;
3311 	compare_iov.iov_len = sizeof(aa_buf);
3312 	write_iov.iov_base = bb_buf;
3313 	write_iov.iov_len = sizeof(bb_buf);
3314 
3315 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3316 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3317 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, offset, num_blocks, 0);
3318 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3319 
3320 	g_io_done = false;
3321 	g_compare_read_buf = aa_buf;
3322 	g_compare_read_buf_len = sizeof(aa_buf);
3323 	memset(write_buf, 0, sizeof(write_buf));
3324 	g_compare_write_buf = write_buf;
3325 	g_compare_write_buf_len = sizeof(write_buf);
3326 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3327 			offset, num_blocks, io_done, NULL);
3328 	/* Trigger range locking */
3329 	poll_threads();
3330 	CU_ASSERT_EQUAL(rc, 0);
3331 	num_completed = stub_complete_io(1);
3332 	CU_ASSERT_EQUAL(num_completed, 1);
3333 	CU_ASSERT(g_io_done == false);
3334 	num_completed = stub_complete_io(1);
3335 	/* Trigger range unlocking */
3336 	poll_threads();
3337 	CU_ASSERT_EQUAL(num_completed, 1);
3338 	CU_ASSERT(g_io_done == true);
3339 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3340 	CU_ASSERT(memcmp(write_buf, bb_buf, sizeof(write_buf)) == 0);
3341 
3342 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3343 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3344 
3345 	g_io_done = false;
3346 	g_compare_read_buf = cc_buf;
3347 	g_compare_read_buf_len = sizeof(cc_buf);
3348 	memset(write_buf, 0, sizeof(write_buf));
3349 	g_compare_write_buf = write_buf;
3350 	g_compare_write_buf_len = sizeof(write_buf);
3351 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3352 			offset, num_blocks, io_done, NULL);
3353 	/* Trigger range locking */
3354 	poll_threads();
3355 	CU_ASSERT_EQUAL(rc, 0);
3356 	num_completed = stub_complete_io(1);
3357 	/* Trigger range unlocking earlier because we expect error here */
3358 	poll_threads();
3359 	CU_ASSERT_EQUAL(num_completed, 1);
3360 	CU_ASSERT(g_io_done == true);
3361 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3362 	num_completed = stub_complete_io(1);
3363 	CU_ASSERT_EQUAL(num_completed, 0);
3364 
3365 	spdk_put_io_channel(ioch);
3366 	spdk_bdev_close(desc);
3367 	free_bdev(bdev);
3368 	fn_table.submit_request = stub_submit_request;
3369 	spdk_bdev_finish(bdev_fini_cb, NULL);
3370 	poll_threads();
3371 
3372 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3373 
3374 	g_compare_read_buf = NULL;
3375 	g_compare_write_buf = NULL;
3376 }
3377 
3378 static void
3379 bdev_write_zeroes(void)
3380 {
3381 	struct spdk_bdev *bdev;
3382 	struct spdk_bdev_desc *desc = NULL;
3383 	struct spdk_io_channel *ioch;
3384 	struct ut_expected_io *expected_io;
3385 	uint64_t offset, num_io_blocks, num_blocks;
3386 	uint32_t num_completed, num_requests;
3387 	int rc;
3388 
3389 	spdk_bdev_initialize(bdev_init_cb, NULL);
3390 	bdev = allocate_bdev("bdev");
3391 
3392 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3393 	CU_ASSERT_EQUAL(rc, 0);
3394 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3395 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3396 	ioch = spdk_bdev_get_io_channel(desc);
3397 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3398 
3399 	fn_table.submit_request = stub_submit_request;
3400 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3401 
3402 	/* First test that if the bdev supports write_zeroes, the request won't be split */
3403 	bdev->md_len = 0;
3404 	bdev->blocklen = 4096;
3405 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3406 
3407 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
3408 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3409 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3410 	CU_ASSERT_EQUAL(rc, 0);
3411 	num_completed = stub_complete_io(1);
3412 	CU_ASSERT_EQUAL(num_completed, 1);
3413 
3414 	/* Check that if write zeroes is not supported it'll be replaced by regular writes */
3415 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
3416 	num_io_blocks = ZERO_BUFFER_SIZE / bdev->blocklen;
3417 	num_requests = 2;
3418 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * num_requests;
3419 
3420 	for (offset = 0; offset < num_requests; ++offset) {
3421 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3422 						   offset * num_io_blocks, num_io_blocks, 0);
3423 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3424 	}
3425 
3426 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3427 	CU_ASSERT_EQUAL(rc, 0);
3428 	num_completed = stub_complete_io(num_requests);
3429 	CU_ASSERT_EQUAL(num_completed, num_requests);
3430 
3431 	/* Check that the splitting is correct if bdev has interleaved metadata */
3432 	bdev->md_interleave = true;
3433 	bdev->md_len = 64;
3434 	bdev->blocklen = 4096 + 64;
3435 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3436 
3437 	num_requests = offset = 0;
3438 	while (offset < num_blocks) {
3439 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / bdev->blocklen, num_blocks - offset);
3440 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3441 						   offset, num_io_blocks, 0);
3442 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3443 		offset += num_io_blocks;
3444 		num_requests++;
3445 	}
3446 
3447 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3448 	CU_ASSERT_EQUAL(rc, 0);
3449 	num_completed = stub_complete_io(num_requests);
3450 	CU_ASSERT_EQUAL(num_completed, num_requests);
3451 	num_completed = stub_complete_io(num_requests);
3452 	assert(num_completed == 0);
3453 
3454 	/* Check the the same for separate metadata buffer */
3455 	bdev->md_interleave = false;
3456 	bdev->md_len = 64;
3457 	bdev->blocklen = 4096;
3458 
3459 	num_requests = offset = 0;
3460 	while (offset < num_blocks) {
3461 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / (bdev->blocklen + bdev->md_len), num_blocks);
3462 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3463 						   offset, num_io_blocks, 0);
3464 		expected_io->md_buf = (char *)g_bdev_mgr.zero_buffer + num_io_blocks * bdev->blocklen;
3465 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3466 		offset += num_io_blocks;
3467 		num_requests++;
3468 	}
3469 
3470 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3471 	CU_ASSERT_EQUAL(rc, 0);
3472 	num_completed = stub_complete_io(num_requests);
3473 	CU_ASSERT_EQUAL(num_completed, num_requests);
3474 
3475 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
3476 	spdk_put_io_channel(ioch);
3477 	spdk_bdev_close(desc);
3478 	free_bdev(bdev);
3479 	spdk_bdev_finish(bdev_fini_cb, NULL);
3480 	poll_threads();
3481 }
3482 
3483 static void
3484 bdev_zcopy_write(void)
3485 {
3486 	struct spdk_bdev *bdev;
3487 	struct spdk_bdev_desc *desc = NULL;
3488 	struct spdk_io_channel *ioch;
3489 	struct ut_expected_io *expected_io;
3490 	uint64_t offset, num_blocks;
3491 	uint32_t num_completed;
3492 	char aa_buf[512];
3493 	struct iovec iov;
3494 	int rc;
3495 	const bool populate = false;
3496 	const bool commit = true;
3497 
3498 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3499 
3500 	spdk_bdev_initialize(bdev_init_cb, NULL);
3501 	bdev = allocate_bdev("bdev");
3502 
3503 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3504 	CU_ASSERT_EQUAL(rc, 0);
3505 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3506 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3507 	ioch = spdk_bdev_get_io_channel(desc);
3508 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3509 
3510 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3511 
3512 	offset = 50;
3513 	num_blocks = 1;
3514 	iov.iov_base = NULL;
3515 	iov.iov_len = 0;
3516 
3517 	g_zcopy_read_buf = (void *) 0x1122334455667788UL;
3518 	g_zcopy_read_buf_len = (uint32_t) -1;
3519 	/* Do a zcopy start for a write (populate=false) */
3520 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3521 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3522 	g_io_done = false;
3523 	g_zcopy_write_buf = aa_buf;
3524 	g_zcopy_write_buf_len = sizeof(aa_buf);
3525 	g_zcopy_bdev_io = NULL;
3526 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3527 	CU_ASSERT_EQUAL(rc, 0);
3528 	num_completed = stub_complete_io(1);
3529 	CU_ASSERT_EQUAL(num_completed, 1);
3530 	CU_ASSERT(g_io_done == true);
3531 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3532 	/* Check that the iov has been set up */
3533 	CU_ASSERT(iov.iov_base == g_zcopy_write_buf);
3534 	CU_ASSERT(iov.iov_len == g_zcopy_write_buf_len);
3535 	/* Check that the bdev_io has been saved */
3536 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3537 	/* Now do the zcopy end for a write (commit=true) */
3538 	g_io_done = false;
3539 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3540 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3541 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3542 	CU_ASSERT_EQUAL(rc, 0);
3543 	num_completed = stub_complete_io(1);
3544 	CU_ASSERT_EQUAL(num_completed, 1);
3545 	CU_ASSERT(g_io_done == true);
3546 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3547 	/* Check the g_zcopy are reset by io_done */
3548 	CU_ASSERT(g_zcopy_write_buf == NULL);
3549 	CU_ASSERT(g_zcopy_write_buf_len == 0);
3550 	/* Check that io_done has freed the g_zcopy_bdev_io */
3551 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3552 
3553 	/* Check the zcopy read buffer has not been touched which
3554 	 * ensures that the correct buffers were used.
3555 	 */
3556 	CU_ASSERT(g_zcopy_read_buf == (void *) 0x1122334455667788UL);
3557 	CU_ASSERT(g_zcopy_read_buf_len == (uint32_t) -1);
3558 
3559 	spdk_put_io_channel(ioch);
3560 	spdk_bdev_close(desc);
3561 	free_bdev(bdev);
3562 	spdk_bdev_finish(bdev_fini_cb, NULL);
3563 	poll_threads();
3564 }
3565 
3566 static void
3567 bdev_zcopy_read(void)
3568 {
3569 	struct spdk_bdev *bdev;
3570 	struct spdk_bdev_desc *desc = NULL;
3571 	struct spdk_io_channel *ioch;
3572 	struct ut_expected_io *expected_io;
3573 	uint64_t offset, num_blocks;
3574 	uint32_t num_completed;
3575 	char aa_buf[512];
3576 	struct iovec iov;
3577 	int rc;
3578 	const bool populate = true;
3579 	const bool commit = false;
3580 
3581 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3582 
3583 	spdk_bdev_initialize(bdev_init_cb, NULL);
3584 	bdev = allocate_bdev("bdev");
3585 
3586 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3587 	CU_ASSERT_EQUAL(rc, 0);
3588 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3589 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3590 	ioch = spdk_bdev_get_io_channel(desc);
3591 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3592 
3593 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3594 
3595 	offset = 50;
3596 	num_blocks = 1;
3597 	iov.iov_base = NULL;
3598 	iov.iov_len = 0;
3599 
3600 	g_zcopy_write_buf = (void *) 0x1122334455667788UL;
3601 	g_zcopy_write_buf_len = (uint32_t) -1;
3602 
3603 	/* Do a zcopy start for a read (populate=true) */
3604 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3605 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3606 	g_io_done = false;
3607 	g_zcopy_read_buf = aa_buf;
3608 	g_zcopy_read_buf_len = sizeof(aa_buf);
3609 	g_zcopy_bdev_io = NULL;
3610 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3611 	CU_ASSERT_EQUAL(rc, 0);
3612 	num_completed = stub_complete_io(1);
3613 	CU_ASSERT_EQUAL(num_completed, 1);
3614 	CU_ASSERT(g_io_done == true);
3615 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3616 	/* Check that the iov has been set up */
3617 	CU_ASSERT(iov.iov_base == g_zcopy_read_buf);
3618 	CU_ASSERT(iov.iov_len == g_zcopy_read_buf_len);
3619 	/* Check that the bdev_io has been saved */
3620 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3621 
3622 	/* Now do the zcopy end for a read (commit=false) */
3623 	g_io_done = false;
3624 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3625 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3626 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3627 	CU_ASSERT_EQUAL(rc, 0);
3628 	num_completed = stub_complete_io(1);
3629 	CU_ASSERT_EQUAL(num_completed, 1);
3630 	CU_ASSERT(g_io_done == true);
3631 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3632 	/* Check the g_zcopy are reset by io_done */
3633 	CU_ASSERT(g_zcopy_read_buf == NULL);
3634 	CU_ASSERT(g_zcopy_read_buf_len == 0);
3635 	/* Check that io_done has freed the g_zcopy_bdev_io */
3636 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3637 
3638 	/* Check the zcopy write buffer has not been touched which
3639 	 * ensures that the correct buffers were used.
3640 	 */
3641 	CU_ASSERT(g_zcopy_write_buf == (void *) 0x1122334455667788UL);
3642 	CU_ASSERT(g_zcopy_write_buf_len == (uint32_t) -1);
3643 
3644 	spdk_put_io_channel(ioch);
3645 	spdk_bdev_close(desc);
3646 	free_bdev(bdev);
3647 	spdk_bdev_finish(bdev_fini_cb, NULL);
3648 	poll_threads();
3649 }
3650 
3651 static void
3652 bdev_open_while_hotremove(void)
3653 {
3654 	struct spdk_bdev *bdev;
3655 	struct spdk_bdev_desc *desc[2] = {};
3656 	int rc;
3657 
3658 	bdev = allocate_bdev("bdev");
3659 
3660 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[0]);
3661 	CU_ASSERT(rc == 0);
3662 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
3663 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc[0]));
3664 
3665 	spdk_bdev_unregister(bdev, NULL, NULL);
3666 
3667 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[1]);
3668 	CU_ASSERT(rc == -ENODEV);
3669 	SPDK_CU_ASSERT_FATAL(desc[1] == NULL);
3670 
3671 	spdk_bdev_close(desc[0]);
3672 	free_bdev(bdev);
3673 }
3674 
3675 static void
3676 bdev_close_while_hotremove(void)
3677 {
3678 	struct spdk_bdev *bdev;
3679 	struct spdk_bdev_desc *desc = NULL;
3680 	int rc = 0;
3681 
3682 	bdev = allocate_bdev("bdev");
3683 
3684 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc, &desc);
3685 	CU_ASSERT_EQUAL(rc, 0);
3686 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3687 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3688 
3689 	/* Simulate hot-unplug by unregistering bdev */
3690 	g_event_type1 = 0xFF;
3691 	g_unregister_arg = NULL;
3692 	g_unregister_rc = -1;
3693 	spdk_bdev_unregister(bdev, bdev_unregister_cb, (void *)0x12345678);
3694 	/* Close device while remove event is in flight */
3695 	spdk_bdev_close(desc);
3696 
3697 	/* Ensure that unregister callback is delayed */
3698 	CU_ASSERT_EQUAL(g_unregister_arg, NULL);
3699 	CU_ASSERT_EQUAL(g_unregister_rc, -1);
3700 
3701 	poll_threads();
3702 
3703 	/* Event callback shall not be issued because device was closed */
3704 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
3705 	/* Unregister callback is issued */
3706 	CU_ASSERT_EQUAL(g_unregister_arg, (void *)0x12345678);
3707 	CU_ASSERT_EQUAL(g_unregister_rc, 0);
3708 
3709 	free_bdev(bdev);
3710 }
3711 
3712 static void
3713 bdev_open_ext(void)
3714 {
3715 	struct spdk_bdev *bdev;
3716 	struct spdk_bdev_desc *desc1 = NULL;
3717 	struct spdk_bdev_desc *desc2 = NULL;
3718 	int rc = 0;
3719 
3720 	bdev = allocate_bdev("bdev");
3721 
3722 	rc = spdk_bdev_open_ext("bdev", true, NULL, NULL, &desc1);
3723 	CU_ASSERT_EQUAL(rc, -EINVAL);
3724 
3725 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc1, &desc1);
3726 	CU_ASSERT_EQUAL(rc, 0);
3727 
3728 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb2, &desc2, &desc2);
3729 	CU_ASSERT_EQUAL(rc, 0);
3730 
3731 	g_event_type1 = 0xFF;
3732 	g_event_type2 = 0xFF;
3733 
3734 	/* Simulate hot-unplug by unregistering bdev */
3735 	spdk_bdev_unregister(bdev, NULL, NULL);
3736 	poll_threads();
3737 
3738 	/* Check if correct events have been triggered in event callback fn */
3739 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_REMOVE);
3740 	CU_ASSERT_EQUAL(g_event_type2, SPDK_BDEV_EVENT_REMOVE);
3741 
3742 	free_bdev(bdev);
3743 	poll_threads();
3744 }
3745 
3746 struct timeout_io_cb_arg {
3747 	struct iovec iov;
3748 	uint8_t type;
3749 };
3750 
3751 static int
3752 bdev_channel_count_submitted_io(struct spdk_bdev_channel *ch)
3753 {
3754 	struct spdk_bdev_io *bdev_io;
3755 	int n = 0;
3756 
3757 	if (!ch) {
3758 		return -1;
3759 	}
3760 
3761 	TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
3762 		n++;
3763 	}
3764 
3765 	return n;
3766 }
3767 
3768 static void
3769 bdev_channel_io_timeout_cb(void *cb_arg, struct spdk_bdev_io *bdev_io)
3770 {
3771 	struct timeout_io_cb_arg *ctx = cb_arg;
3772 
3773 	ctx->type = bdev_io->type;
3774 	ctx->iov.iov_base = bdev_io->iov.iov_base;
3775 	ctx->iov.iov_len = bdev_io->iov.iov_len;
3776 }
3777 
3778 static void
3779 bdev_set_io_timeout(void)
3780 {
3781 	struct spdk_bdev *bdev;
3782 	struct spdk_bdev_desc *desc = NULL;
3783 	struct spdk_io_channel *io_ch = NULL;
3784 	struct spdk_bdev_channel *bdev_ch = NULL;
3785 	struct timeout_io_cb_arg cb_arg;
3786 
3787 	spdk_bdev_initialize(bdev_init_cb, NULL);
3788 
3789 	bdev = allocate_bdev("bdev");
3790 
3791 	CU_ASSERT(spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc) == 0);
3792 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3793 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3794 
3795 	io_ch = spdk_bdev_get_io_channel(desc);
3796 	CU_ASSERT(io_ch != NULL);
3797 
3798 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
3799 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
3800 
3801 	/* This is the part1.
3802 	 * We will check the bdev_ch->io_submitted list
3803 	 * TO make sure that it can link IOs and only the user submitted IOs
3804 	 */
3805 	CU_ASSERT(spdk_bdev_read(desc, io_ch, (void *)0x1000, 0, 4096, io_done, NULL) == 0);
3806 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3807 	CU_ASSERT(spdk_bdev_write(desc, io_ch, (void *)0x2000, 0, 4096, io_done, NULL) == 0);
3808 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3809 	stub_complete_io(1);
3810 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3811 	stub_complete_io(1);
3812 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3813 
3814 	/* Split IO */
3815 	bdev->optimal_io_boundary = 16;
3816 	bdev->split_on_optimal_io_boundary = true;
3817 
3818 	/* Now test that a single-vector command is split correctly.
3819 	 * Offset 14, length 8, payload 0xF000
3820 	 *  Child - Offset 14, length 2, payload 0xF000
3821 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
3822 	 *
3823 	 * Set up the expected values before calling spdk_bdev_read_blocks
3824 	 */
3825 	CU_ASSERT(spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3826 	/* We count all submitted IOs including IO that are generated by splitting. */
3827 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 3);
3828 	stub_complete_io(1);
3829 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3830 	stub_complete_io(1);
3831 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3832 
3833 	/* Also include the reset IO */
3834 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3835 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3836 	poll_threads();
3837 	stub_complete_io(1);
3838 	poll_threads();
3839 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3840 
3841 	/* This is part2
3842 	 * Test the desc timeout poller register
3843 	 */
3844 
3845 	/* Successfully set the timeout */
3846 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3847 	CU_ASSERT(desc->io_timeout_poller != NULL);
3848 	CU_ASSERT(desc->timeout_in_sec == 30);
3849 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3850 	CU_ASSERT(desc->cb_arg == &cb_arg);
3851 
3852 	/* Change the timeout limit */
3853 	CU_ASSERT(spdk_bdev_set_timeout(desc, 20, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3854 	CU_ASSERT(desc->io_timeout_poller != NULL);
3855 	CU_ASSERT(desc->timeout_in_sec == 20);
3856 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3857 	CU_ASSERT(desc->cb_arg == &cb_arg);
3858 
3859 	/* Disable the timeout */
3860 	CU_ASSERT(spdk_bdev_set_timeout(desc, 0, NULL, NULL) == 0);
3861 	CU_ASSERT(desc->io_timeout_poller == NULL);
3862 
3863 	/* This the part3
3864 	 * We will test to catch timeout IO and check whether the IO is
3865 	 * the submitted one.
3866 	 */
3867 	memset(&cb_arg, 0, sizeof(cb_arg));
3868 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3869 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0x1000, 0, 1, io_done, NULL) == 0);
3870 
3871 	/* Don't reach the limit */
3872 	spdk_delay_us(15 * spdk_get_ticks_hz());
3873 	poll_threads();
3874 	CU_ASSERT(cb_arg.type == 0);
3875 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3876 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3877 
3878 	/* 15 + 15 = 30 reach the limit */
3879 	spdk_delay_us(15 * spdk_get_ticks_hz());
3880 	poll_threads();
3881 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3882 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x1000);
3883 	CU_ASSERT(cb_arg.iov.iov_len == 1 * bdev->blocklen);
3884 	stub_complete_io(1);
3885 
3886 	/* Use the same split IO above and check the IO */
3887 	memset(&cb_arg, 0, sizeof(cb_arg));
3888 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3889 
3890 	/* The first child complete in time */
3891 	spdk_delay_us(15 * spdk_get_ticks_hz());
3892 	poll_threads();
3893 	stub_complete_io(1);
3894 	CU_ASSERT(cb_arg.type == 0);
3895 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3896 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3897 
3898 	/* The second child reach the limit */
3899 	spdk_delay_us(15 * spdk_get_ticks_hz());
3900 	poll_threads();
3901 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3902 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0xF000);
3903 	CU_ASSERT(cb_arg.iov.iov_len == 8 * bdev->blocklen);
3904 	stub_complete_io(1);
3905 
3906 	/* Also include the reset IO */
3907 	memset(&cb_arg, 0, sizeof(cb_arg));
3908 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3909 	spdk_delay_us(30 * spdk_get_ticks_hz());
3910 	poll_threads();
3911 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_RESET);
3912 	stub_complete_io(1);
3913 	poll_threads();
3914 
3915 	spdk_put_io_channel(io_ch);
3916 	spdk_bdev_close(desc);
3917 	free_bdev(bdev);
3918 	spdk_bdev_finish(bdev_fini_cb, NULL);
3919 	poll_threads();
3920 }
3921 
3922 static void
3923 lba_range_overlap(void)
3924 {
3925 	struct lba_range r1, r2;
3926 
3927 	r1.offset = 100;
3928 	r1.length = 50;
3929 
3930 	r2.offset = 0;
3931 	r2.length = 1;
3932 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3933 
3934 	r2.offset = 0;
3935 	r2.length = 100;
3936 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3937 
3938 	r2.offset = 0;
3939 	r2.length = 110;
3940 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3941 
3942 	r2.offset = 100;
3943 	r2.length = 10;
3944 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3945 
3946 	r2.offset = 110;
3947 	r2.length = 20;
3948 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3949 
3950 	r2.offset = 140;
3951 	r2.length = 150;
3952 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3953 
3954 	r2.offset = 130;
3955 	r2.length = 200;
3956 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3957 
3958 	r2.offset = 150;
3959 	r2.length = 100;
3960 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3961 
3962 	r2.offset = 110;
3963 	r2.length = 0;
3964 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3965 }
3966 
3967 static bool g_lock_lba_range_done;
3968 static bool g_unlock_lba_range_done;
3969 
3970 static void
3971 lock_lba_range_done(void *ctx, int status)
3972 {
3973 	g_lock_lba_range_done = true;
3974 }
3975 
3976 static void
3977 unlock_lba_range_done(void *ctx, int status)
3978 {
3979 	g_unlock_lba_range_done = true;
3980 }
3981 
3982 static void
3983 lock_lba_range_check_ranges(void)
3984 {
3985 	struct spdk_bdev *bdev;
3986 	struct spdk_bdev_desc *desc = NULL;
3987 	struct spdk_io_channel *io_ch;
3988 	struct spdk_bdev_channel *channel;
3989 	struct lba_range *range;
3990 	int ctx1;
3991 	int rc;
3992 
3993 	spdk_bdev_initialize(bdev_init_cb, NULL);
3994 
3995 	bdev = allocate_bdev("bdev0");
3996 
3997 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
3998 	CU_ASSERT(rc == 0);
3999 	CU_ASSERT(desc != NULL);
4000 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4001 	io_ch = spdk_bdev_get_io_channel(desc);
4002 	CU_ASSERT(io_ch != NULL);
4003 	channel = spdk_io_channel_get_ctx(io_ch);
4004 
4005 	g_lock_lba_range_done = false;
4006 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4007 	CU_ASSERT(rc == 0);
4008 	poll_threads();
4009 
4010 	CU_ASSERT(g_lock_lba_range_done == true);
4011 	range = TAILQ_FIRST(&channel->locked_ranges);
4012 	SPDK_CU_ASSERT_FATAL(range != NULL);
4013 	CU_ASSERT(range->offset == 20);
4014 	CU_ASSERT(range->length == 10);
4015 	CU_ASSERT(range->owner_ch == channel);
4016 
4017 	/* Unlocks must exactly match a lock. */
4018 	g_unlock_lba_range_done = false;
4019 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 1, unlock_lba_range_done, &ctx1);
4020 	CU_ASSERT(rc == -EINVAL);
4021 	CU_ASSERT(g_unlock_lba_range_done == false);
4022 
4023 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4024 	CU_ASSERT(rc == 0);
4025 	spdk_delay_us(100);
4026 	poll_threads();
4027 
4028 	CU_ASSERT(g_unlock_lba_range_done == true);
4029 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4030 
4031 	spdk_put_io_channel(io_ch);
4032 	spdk_bdev_close(desc);
4033 	free_bdev(bdev);
4034 	spdk_bdev_finish(bdev_fini_cb, NULL);
4035 	poll_threads();
4036 }
4037 
4038 static void
4039 lock_lba_range_with_io_outstanding(void)
4040 {
4041 	struct spdk_bdev *bdev;
4042 	struct spdk_bdev_desc *desc = NULL;
4043 	struct spdk_io_channel *io_ch;
4044 	struct spdk_bdev_channel *channel;
4045 	struct lba_range *range;
4046 	char buf[4096];
4047 	int ctx1;
4048 	int rc;
4049 
4050 	spdk_bdev_initialize(bdev_init_cb, NULL);
4051 
4052 	bdev = allocate_bdev("bdev0");
4053 
4054 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4055 	CU_ASSERT(rc == 0);
4056 	CU_ASSERT(desc != NULL);
4057 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4058 	io_ch = spdk_bdev_get_io_channel(desc);
4059 	CU_ASSERT(io_ch != NULL);
4060 	channel = spdk_io_channel_get_ctx(io_ch);
4061 
4062 	g_io_done = false;
4063 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4064 	CU_ASSERT(rc == 0);
4065 
4066 	g_lock_lba_range_done = false;
4067 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4068 	CU_ASSERT(rc == 0);
4069 	poll_threads();
4070 
4071 	/* The lock should immediately become valid, since there are no outstanding
4072 	 * write I/O.
4073 	 */
4074 	CU_ASSERT(g_io_done == false);
4075 	CU_ASSERT(g_lock_lba_range_done == true);
4076 	range = TAILQ_FIRST(&channel->locked_ranges);
4077 	SPDK_CU_ASSERT_FATAL(range != NULL);
4078 	CU_ASSERT(range->offset == 20);
4079 	CU_ASSERT(range->length == 10);
4080 	CU_ASSERT(range->owner_ch == channel);
4081 	CU_ASSERT(range->locked_ctx == &ctx1);
4082 
4083 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4084 	CU_ASSERT(rc == 0);
4085 	stub_complete_io(1);
4086 	spdk_delay_us(100);
4087 	poll_threads();
4088 
4089 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4090 
4091 	/* Now try again, but with a write I/O. */
4092 	g_io_done = false;
4093 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4094 	CU_ASSERT(rc == 0);
4095 
4096 	g_lock_lba_range_done = false;
4097 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4098 	CU_ASSERT(rc == 0);
4099 	poll_threads();
4100 
4101 	/* The lock should not be fully valid yet, since a write I/O is outstanding.
4102 	 * But note that the range should be on the channel's locked_list, to make sure no
4103 	 * new write I/O are started.
4104 	 */
4105 	CU_ASSERT(g_io_done == false);
4106 	CU_ASSERT(g_lock_lba_range_done == false);
4107 	range = TAILQ_FIRST(&channel->locked_ranges);
4108 	SPDK_CU_ASSERT_FATAL(range != NULL);
4109 	CU_ASSERT(range->offset == 20);
4110 	CU_ASSERT(range->length == 10);
4111 
4112 	/* Complete the write I/O.  This should make the lock valid (checked by confirming
4113 	 * our callback was invoked).
4114 	 */
4115 	stub_complete_io(1);
4116 	spdk_delay_us(100);
4117 	poll_threads();
4118 	CU_ASSERT(g_io_done == true);
4119 	CU_ASSERT(g_lock_lba_range_done == true);
4120 
4121 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4122 	CU_ASSERT(rc == 0);
4123 	poll_threads();
4124 
4125 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4126 
4127 	spdk_put_io_channel(io_ch);
4128 	spdk_bdev_close(desc);
4129 	free_bdev(bdev);
4130 	spdk_bdev_finish(bdev_fini_cb, NULL);
4131 	poll_threads();
4132 }
4133 
4134 static void
4135 lock_lba_range_overlapped(void)
4136 {
4137 	struct spdk_bdev *bdev;
4138 	struct spdk_bdev_desc *desc = NULL;
4139 	struct spdk_io_channel *io_ch;
4140 	struct spdk_bdev_channel *channel;
4141 	struct lba_range *range;
4142 	int ctx1;
4143 	int rc;
4144 
4145 	spdk_bdev_initialize(bdev_init_cb, NULL);
4146 
4147 	bdev = allocate_bdev("bdev0");
4148 
4149 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4150 	CU_ASSERT(rc == 0);
4151 	CU_ASSERT(desc != NULL);
4152 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4153 	io_ch = spdk_bdev_get_io_channel(desc);
4154 	CU_ASSERT(io_ch != NULL);
4155 	channel = spdk_io_channel_get_ctx(io_ch);
4156 
4157 	/* Lock range 20-29. */
4158 	g_lock_lba_range_done = false;
4159 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4160 	CU_ASSERT(rc == 0);
4161 	poll_threads();
4162 
4163 	CU_ASSERT(g_lock_lba_range_done == true);
4164 	range = TAILQ_FIRST(&channel->locked_ranges);
4165 	SPDK_CU_ASSERT_FATAL(range != NULL);
4166 	CU_ASSERT(range->offset == 20);
4167 	CU_ASSERT(range->length == 10);
4168 
4169 	/* Try to lock range 25-39.  It should not lock immediately, since it overlaps with
4170 	 * 20-29.
4171 	 */
4172 	g_lock_lba_range_done = false;
4173 	rc = bdev_lock_lba_range(desc, io_ch, 25, 15, lock_lba_range_done, &ctx1);
4174 	CU_ASSERT(rc == 0);
4175 	poll_threads();
4176 
4177 	CU_ASSERT(g_lock_lba_range_done == false);
4178 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4179 	SPDK_CU_ASSERT_FATAL(range != NULL);
4180 	CU_ASSERT(range->offset == 25);
4181 	CU_ASSERT(range->length == 15);
4182 
4183 	/* Unlock 20-29.  This should result in range 25-39 now getting locked since it
4184 	 * no longer overlaps with an active lock.
4185 	 */
4186 	g_unlock_lba_range_done = false;
4187 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4188 	CU_ASSERT(rc == 0);
4189 	poll_threads();
4190 
4191 	CU_ASSERT(g_unlock_lba_range_done == true);
4192 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4193 	range = TAILQ_FIRST(&channel->locked_ranges);
4194 	SPDK_CU_ASSERT_FATAL(range != NULL);
4195 	CU_ASSERT(range->offset == 25);
4196 	CU_ASSERT(range->length == 15);
4197 
4198 	/* Lock 40-59.  This should immediately lock since it does not overlap with the
4199 	 * currently active 25-39 lock.
4200 	 */
4201 	g_lock_lba_range_done = false;
4202 	rc = bdev_lock_lba_range(desc, io_ch, 40, 20, lock_lba_range_done, &ctx1);
4203 	CU_ASSERT(rc == 0);
4204 	poll_threads();
4205 
4206 	CU_ASSERT(g_lock_lba_range_done == true);
4207 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4208 	SPDK_CU_ASSERT_FATAL(range != NULL);
4209 	range = TAILQ_NEXT(range, tailq);
4210 	SPDK_CU_ASSERT_FATAL(range != NULL);
4211 	CU_ASSERT(range->offset == 40);
4212 	CU_ASSERT(range->length == 20);
4213 
4214 	/* Try to lock 35-44.  Note that this overlaps with both 25-39 and 40-59. */
4215 	g_lock_lba_range_done = false;
4216 	rc = bdev_lock_lba_range(desc, io_ch, 35, 10, lock_lba_range_done, &ctx1);
4217 	CU_ASSERT(rc == 0);
4218 	poll_threads();
4219 
4220 	CU_ASSERT(g_lock_lba_range_done == false);
4221 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4222 	SPDK_CU_ASSERT_FATAL(range != NULL);
4223 	CU_ASSERT(range->offset == 35);
4224 	CU_ASSERT(range->length == 10);
4225 
4226 	/* Unlock 25-39.  Make sure that 35-44 is still in the pending list, since
4227 	 * the 40-59 lock is still active.
4228 	 */
4229 	g_unlock_lba_range_done = false;
4230 	rc = bdev_unlock_lba_range(desc, io_ch, 25, 15, unlock_lba_range_done, &ctx1);
4231 	CU_ASSERT(rc == 0);
4232 	poll_threads();
4233 
4234 	CU_ASSERT(g_unlock_lba_range_done == true);
4235 	CU_ASSERT(g_lock_lba_range_done == false);
4236 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4237 	SPDK_CU_ASSERT_FATAL(range != NULL);
4238 	CU_ASSERT(range->offset == 35);
4239 	CU_ASSERT(range->length == 10);
4240 
4241 	/* Unlock 40-59.  This should result in 35-44 now getting locked, since there are
4242 	 * no longer any active overlapping locks.
4243 	 */
4244 	g_unlock_lba_range_done = false;
4245 	rc = bdev_unlock_lba_range(desc, io_ch, 40, 20, unlock_lba_range_done, &ctx1);
4246 	CU_ASSERT(rc == 0);
4247 	poll_threads();
4248 
4249 	CU_ASSERT(g_unlock_lba_range_done == true);
4250 	CU_ASSERT(g_lock_lba_range_done == true);
4251 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4252 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4253 	SPDK_CU_ASSERT_FATAL(range != NULL);
4254 	CU_ASSERT(range->offset == 35);
4255 	CU_ASSERT(range->length == 10);
4256 
4257 	/* Finally, unlock 35-44. */
4258 	g_unlock_lba_range_done = false;
4259 	rc = bdev_unlock_lba_range(desc, io_ch, 35, 10, unlock_lba_range_done, &ctx1);
4260 	CU_ASSERT(rc == 0);
4261 	poll_threads();
4262 
4263 	CU_ASSERT(g_unlock_lba_range_done == true);
4264 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.locked_ranges));
4265 
4266 	spdk_put_io_channel(io_ch);
4267 	spdk_bdev_close(desc);
4268 	free_bdev(bdev);
4269 	spdk_bdev_finish(bdev_fini_cb, NULL);
4270 	poll_threads();
4271 }
4272 
4273 static void
4274 abort_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
4275 {
4276 	g_abort_done = true;
4277 	g_abort_status = bdev_io->internal.status;
4278 	spdk_bdev_free_io(bdev_io);
4279 }
4280 
4281 static void
4282 bdev_io_abort(void)
4283 {
4284 	struct spdk_bdev *bdev;
4285 	struct spdk_bdev_desc *desc = NULL;
4286 	struct spdk_io_channel *io_ch;
4287 	struct spdk_bdev_channel *channel;
4288 	struct spdk_bdev_mgmt_channel *mgmt_ch;
4289 	struct spdk_bdev_opts bdev_opts = {};
4290 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
4291 	uint64_t io_ctx1 = 0, io_ctx2 = 0, i;
4292 	int rc;
4293 
4294 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4295 	bdev_opts.bdev_io_pool_size = 7;
4296 	bdev_opts.bdev_io_cache_size = 2;
4297 
4298 	rc = spdk_bdev_set_opts(&bdev_opts);
4299 	CU_ASSERT(rc == 0);
4300 	spdk_bdev_initialize(bdev_init_cb, NULL);
4301 
4302 	bdev = allocate_bdev("bdev0");
4303 
4304 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4305 	CU_ASSERT(rc == 0);
4306 	CU_ASSERT(desc != NULL);
4307 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4308 	io_ch = spdk_bdev_get_io_channel(desc);
4309 	CU_ASSERT(io_ch != NULL);
4310 	channel = spdk_io_channel_get_ctx(io_ch);
4311 	mgmt_ch = channel->shared_resource->mgmt_ch;
4312 
4313 	g_abort_done = false;
4314 
4315 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, false);
4316 
4317 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4318 	CU_ASSERT(rc == -ENOTSUP);
4319 
4320 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, true);
4321 
4322 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx2, abort_done, NULL);
4323 	CU_ASSERT(rc == 0);
4324 	CU_ASSERT(g_abort_done == true);
4325 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_FAILED);
4326 
4327 	/* Test the case that the target I/O was successfully aborted. */
4328 	g_io_done = false;
4329 
4330 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4331 	CU_ASSERT(rc == 0);
4332 	CU_ASSERT(g_io_done == false);
4333 
4334 	g_abort_done = false;
4335 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4336 
4337 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4338 	CU_ASSERT(rc == 0);
4339 	CU_ASSERT(g_io_done == true);
4340 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4341 	stub_complete_io(1);
4342 	CU_ASSERT(g_abort_done == true);
4343 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4344 
4345 	/* Test the case that the target I/O was not aborted because it completed
4346 	 * in the middle of execution of the abort.
4347 	 */
4348 	g_io_done = false;
4349 
4350 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4351 	CU_ASSERT(rc == 0);
4352 	CU_ASSERT(g_io_done == false);
4353 
4354 	g_abort_done = false;
4355 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4356 
4357 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4358 	CU_ASSERT(rc == 0);
4359 	CU_ASSERT(g_io_done == false);
4360 
4361 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4362 	stub_complete_io(1);
4363 	CU_ASSERT(g_io_done == true);
4364 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4365 
4366 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4367 	stub_complete_io(1);
4368 	CU_ASSERT(g_abort_done == true);
4369 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4370 
4371 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4372 
4373 	bdev->optimal_io_boundary = 16;
4374 	bdev->split_on_optimal_io_boundary = true;
4375 
4376 	/* Test that a single-vector command which is split is aborted correctly.
4377 	 * Offset 14, length 8, payload 0xF000
4378 	 *  Child - Offset 14, length 2, payload 0xF000
4379 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
4380 	 */
4381 	g_io_done = false;
4382 
4383 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, &io_ctx1);
4384 	CU_ASSERT(rc == 0);
4385 	CU_ASSERT(g_io_done == false);
4386 
4387 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4388 
4389 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4390 
4391 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4392 	CU_ASSERT(rc == 0);
4393 	CU_ASSERT(g_io_done == true);
4394 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4395 	stub_complete_io(2);
4396 	CU_ASSERT(g_abort_done == true);
4397 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4398 
4399 	/* Test that a multi-vector command that needs to be split by strip and then
4400 	 * needs to be split is aborted correctly. Abort is requested before the second
4401 	 * child I/O was submitted. The parent I/O should complete with failure without
4402 	 * submitting the second child I/O.
4403 	 */
4404 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
4405 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
4406 		iov[i].iov_len = 512;
4407 	}
4408 
4409 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
4410 	g_io_done = false;
4411 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
4412 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, &io_ctx1);
4413 	CU_ASSERT(rc == 0);
4414 	CU_ASSERT(g_io_done == false);
4415 
4416 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4417 
4418 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4419 
4420 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4421 	CU_ASSERT(rc == 0);
4422 	CU_ASSERT(g_io_done == true);
4423 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4424 	stub_complete_io(1);
4425 	CU_ASSERT(g_abort_done == true);
4426 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4427 
4428 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4429 
4430 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4431 
4432 	bdev->optimal_io_boundary = 16;
4433 	g_io_done = false;
4434 
4435 	/* Test that a ingle-vector command which is split is aborted correctly.
4436 	 * Differently from the above, the child abort request will be submitted
4437 	 * sequentially due to the capacity of spdk_bdev_io.
4438 	 */
4439 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 50, io_done, &io_ctx1);
4440 	CU_ASSERT(rc == 0);
4441 	CU_ASSERT(g_io_done == false);
4442 
4443 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4444 
4445 	g_abort_done = false;
4446 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4447 
4448 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4449 	CU_ASSERT(rc == 0);
4450 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
4451 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4452 
4453 	stub_complete_io(1);
4454 	CU_ASSERT(g_io_done == true);
4455 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4456 	stub_complete_io(3);
4457 	CU_ASSERT(g_abort_done == true);
4458 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4459 
4460 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4461 
4462 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4463 
4464 	spdk_put_io_channel(io_ch);
4465 	spdk_bdev_close(desc);
4466 	free_bdev(bdev);
4467 	spdk_bdev_finish(bdev_fini_cb, NULL);
4468 	poll_threads();
4469 }
4470 
4471 static void
4472 bdev_unmap(void)
4473 {
4474 	struct spdk_bdev *bdev;
4475 	struct spdk_bdev_desc *desc = NULL;
4476 	struct spdk_io_channel *ioch;
4477 	struct spdk_bdev_channel *bdev_ch;
4478 	struct ut_expected_io *expected_io;
4479 	struct spdk_bdev_opts bdev_opts = {};
4480 	uint32_t i, num_outstanding;
4481 	uint64_t offset, num_blocks, max_unmap_blocks, num_children;
4482 	int rc;
4483 
4484 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4485 	bdev_opts.bdev_io_pool_size = 512;
4486 	bdev_opts.bdev_io_cache_size = 64;
4487 	rc = spdk_bdev_set_opts(&bdev_opts);
4488 	CU_ASSERT(rc == 0);
4489 
4490 	spdk_bdev_initialize(bdev_init_cb, NULL);
4491 	bdev = allocate_bdev("bdev");
4492 
4493 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4494 	CU_ASSERT_EQUAL(rc, 0);
4495 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4496 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4497 	ioch = spdk_bdev_get_io_channel(desc);
4498 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4499 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4500 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4501 
4502 	fn_table.submit_request = stub_submit_request;
4503 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4504 
4505 	/* Case 1: First test the request won't be split */
4506 	num_blocks = 32;
4507 
4508 	g_io_done = false;
4509 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 0, num_blocks, 0);
4510 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4511 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4512 	CU_ASSERT_EQUAL(rc, 0);
4513 	CU_ASSERT(g_io_done == false);
4514 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4515 	stub_complete_io(1);
4516 	CU_ASSERT(g_io_done == true);
4517 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4518 
4519 	/* Case 2: Test the split with 2 children requests */
4520 	bdev->max_unmap = 8;
4521 	bdev->max_unmap_segments = 2;
4522 	max_unmap_blocks = bdev->max_unmap * bdev->max_unmap_segments;
4523 	num_blocks = max_unmap_blocks * 2;
4524 	offset = 0;
4525 
4526 	g_io_done = false;
4527 	for (i = 0; i < 2; i++) {
4528 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4529 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4530 		offset += max_unmap_blocks;
4531 	}
4532 
4533 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4534 	CU_ASSERT_EQUAL(rc, 0);
4535 	CU_ASSERT(g_io_done == false);
4536 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4537 	stub_complete_io(2);
4538 	CU_ASSERT(g_io_done == true);
4539 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4540 
4541 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4542 	num_children = 15;
4543 	num_blocks = max_unmap_blocks * num_children;
4544 	g_io_done = false;
4545 	offset = 0;
4546 	for (i = 0; i < num_children; i++) {
4547 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4548 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4549 		offset += max_unmap_blocks;
4550 	}
4551 
4552 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4553 	CU_ASSERT_EQUAL(rc, 0);
4554 	CU_ASSERT(g_io_done == false);
4555 
4556 	while (num_children > 0) {
4557 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4558 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4559 		stub_complete_io(num_outstanding);
4560 		num_children -= num_outstanding;
4561 	}
4562 	CU_ASSERT(g_io_done == true);
4563 
4564 	spdk_put_io_channel(ioch);
4565 	spdk_bdev_close(desc);
4566 	free_bdev(bdev);
4567 	spdk_bdev_finish(bdev_fini_cb, NULL);
4568 	poll_threads();
4569 }
4570 
4571 static void
4572 bdev_write_zeroes_split_test(void)
4573 {
4574 	struct spdk_bdev *bdev;
4575 	struct spdk_bdev_desc *desc = NULL;
4576 	struct spdk_io_channel *ioch;
4577 	struct spdk_bdev_channel *bdev_ch;
4578 	struct ut_expected_io *expected_io;
4579 	struct spdk_bdev_opts bdev_opts = {};
4580 	uint32_t i, num_outstanding;
4581 	uint64_t offset, num_blocks, max_write_zeroes_blocks, num_children;
4582 	int rc;
4583 
4584 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4585 	bdev_opts.bdev_io_pool_size = 512;
4586 	bdev_opts.bdev_io_cache_size = 64;
4587 	rc = spdk_bdev_set_opts(&bdev_opts);
4588 	CU_ASSERT(rc == 0);
4589 
4590 	spdk_bdev_initialize(bdev_init_cb, NULL);
4591 	bdev = allocate_bdev("bdev");
4592 
4593 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4594 	CU_ASSERT_EQUAL(rc, 0);
4595 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4596 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4597 	ioch = spdk_bdev_get_io_channel(desc);
4598 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4599 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4600 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4601 
4602 	fn_table.submit_request = stub_submit_request;
4603 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4604 
4605 	/* Case 1: First test the request won't be split */
4606 	num_blocks = 32;
4607 
4608 	g_io_done = false;
4609 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
4610 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4611 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4612 	CU_ASSERT_EQUAL(rc, 0);
4613 	CU_ASSERT(g_io_done == false);
4614 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4615 	stub_complete_io(1);
4616 	CU_ASSERT(g_io_done == true);
4617 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4618 
4619 	/* Case 2: Test the split with 2 children requests */
4620 	max_write_zeroes_blocks = 8;
4621 	bdev->max_write_zeroes = max_write_zeroes_blocks;
4622 	num_blocks = max_write_zeroes_blocks * 2;
4623 	offset = 0;
4624 
4625 	g_io_done = false;
4626 	for (i = 0; i < 2; i++) {
4627 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4628 						   0);
4629 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4630 		offset += max_write_zeroes_blocks;
4631 	}
4632 
4633 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4634 	CU_ASSERT_EQUAL(rc, 0);
4635 	CU_ASSERT(g_io_done == false);
4636 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4637 	stub_complete_io(2);
4638 	CU_ASSERT(g_io_done == true);
4639 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4640 
4641 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4642 	num_children = 15;
4643 	num_blocks = max_write_zeroes_blocks * num_children;
4644 	g_io_done = false;
4645 	offset = 0;
4646 	for (i = 0; i < num_children; i++) {
4647 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4648 						   0);
4649 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4650 		offset += max_write_zeroes_blocks;
4651 	}
4652 
4653 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4654 	CU_ASSERT_EQUAL(rc, 0);
4655 	CU_ASSERT(g_io_done == false);
4656 
4657 	while (num_children > 0) {
4658 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4659 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4660 		stub_complete_io(num_outstanding);
4661 		num_children -= num_outstanding;
4662 	}
4663 	CU_ASSERT(g_io_done == true);
4664 
4665 	spdk_put_io_channel(ioch);
4666 	spdk_bdev_close(desc);
4667 	free_bdev(bdev);
4668 	spdk_bdev_finish(bdev_fini_cb, NULL);
4669 	poll_threads();
4670 }
4671 
4672 static void
4673 bdev_set_options_test(void)
4674 {
4675 	struct spdk_bdev_opts bdev_opts = {};
4676 	int rc;
4677 
4678 	/* Case1: Do not set opts_size */
4679 	rc = spdk_bdev_set_opts(&bdev_opts);
4680 	CU_ASSERT(rc == -1);
4681 
4682 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4683 	bdev_opts.bdev_io_pool_size = 4;
4684 	bdev_opts.bdev_io_cache_size = 2;
4685 	bdev_opts.small_buf_pool_size = 4;
4686 
4687 	/* Case 2: Do not set valid small_buf_pool_size and large_buf_pool_size */
4688 	rc = spdk_bdev_set_opts(&bdev_opts);
4689 	CU_ASSERT(rc == -1);
4690 
4691 	/* Case 3: Do not set valid large_buf_pool_size */
4692 	bdev_opts.small_buf_pool_size = BUF_SMALL_POOL_SIZE;
4693 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE - 1;
4694 	rc = spdk_bdev_set_opts(&bdev_opts);
4695 	CU_ASSERT(rc == -1);
4696 
4697 	/* Case4: set valid large buf_pool_size */
4698 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE;
4699 	rc = spdk_bdev_set_opts(&bdev_opts);
4700 	CU_ASSERT(rc == 0);
4701 
4702 	/* Case5: Set different valid value for small and large buf pool */
4703 	bdev_opts.large_buf_pool_size = BUF_SMALL_POOL_SIZE + 3;
4704 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE + 3;
4705 	rc = spdk_bdev_set_opts(&bdev_opts);
4706 	CU_ASSERT(rc == 0);
4707 }
4708 
4709 static uint64_t
4710 get_ns_time(void)
4711 {
4712 	int rc;
4713 	struct timespec ts;
4714 
4715 	rc = clock_gettime(CLOCK_MONOTONIC, &ts);
4716 	CU_ASSERT(rc == 0);
4717 	return ts.tv_sec * 1000 * 1000 * 1000 + ts.tv_nsec;
4718 }
4719 
4720 static int
4721 rb_tree_get_height(struct spdk_bdev_name *bdev_name)
4722 {
4723 	int h1, h2;
4724 
4725 	if (bdev_name == NULL) {
4726 		return -1;
4727 	} else {
4728 		h1 = rb_tree_get_height(RB_LEFT(bdev_name, node));
4729 		h2 = rb_tree_get_height(RB_RIGHT(bdev_name, node));
4730 
4731 		return spdk_max(h1, h2) + 1;
4732 	}
4733 }
4734 
4735 static void
4736 bdev_multi_allocation(void)
4737 {
4738 	const int max_bdev_num = 1024 * 16;
4739 	char name[max_bdev_num][10];
4740 	char noexist_name[] = "invalid_bdev";
4741 	struct spdk_bdev *bdev[max_bdev_num];
4742 	int i, j;
4743 	uint64_t last_time;
4744 	int bdev_num;
4745 	int height;
4746 
4747 	for (j = 0; j < max_bdev_num; j++) {
4748 		snprintf(name[j], sizeof(name[j]), "bdev%d", j);
4749 	}
4750 
4751 	for (i = 0; i < 16; i++) {
4752 		last_time = get_ns_time();
4753 		bdev_num = 1024 * (i + 1);
4754 		for (j = 0; j < bdev_num; j++) {
4755 			bdev[j] = allocate_bdev(name[j]);
4756 			height = rb_tree_get_height(&bdev[j]->internal.bdev_name);
4757 			CU_ASSERT(height <= (int)(spdk_u32log2(j + 1)));
4758 		}
4759 		SPDK_NOTICELOG("alloc bdev num %d takes %" PRIu64 " ms\n", bdev_num,
4760 			       (get_ns_time() - last_time) / 1000 / 1000);
4761 		for (j = 0; j < bdev_num; j++) {
4762 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) != NULL);
4763 		}
4764 		CU_ASSERT(spdk_bdev_get_by_name(noexist_name) == NULL);
4765 
4766 		for (j = 0; j < bdev_num; j++) {
4767 			free_bdev(bdev[j]);
4768 		}
4769 		for (j = 0; j < bdev_num; j++) {
4770 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) == NULL);
4771 		}
4772 	}
4773 }
4774 
4775 static struct spdk_memory_domain *g_bdev_memory_domain = (struct spdk_memory_domain *) 0xf00df00d;
4776 
4777 static int
4778 test_bdev_get_supported_dma_device_types_op(void *ctx, struct spdk_memory_domain **domains,
4779 		int array_size)
4780 {
4781 	if (array_size > 0 && domains) {
4782 		domains[0] = g_bdev_memory_domain;
4783 	}
4784 
4785 	return 1;
4786 }
4787 
4788 static void
4789 bdev_get_memory_domains(void)
4790 {
4791 	struct spdk_bdev_fn_table fn_table = {
4792 		.get_memory_domains = test_bdev_get_supported_dma_device_types_op
4793 	};
4794 	struct spdk_bdev bdev = { .fn_table = &fn_table };
4795 	struct spdk_memory_domain *domains[2] = {};
4796 	int rc;
4797 
4798 	/* bdev is NULL */
4799 	rc = spdk_bdev_get_memory_domains(NULL, domains, 2);
4800 	CU_ASSERT(rc == -EINVAL);
4801 
4802 	/* domains is NULL */
4803 	rc = spdk_bdev_get_memory_domains(&bdev, NULL, 2);
4804 	CU_ASSERT(rc == 1);
4805 
4806 	/* array size is 0 */
4807 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 0);
4808 	CU_ASSERT(rc == 1);
4809 
4810 	/* get_supported_dma_device_types op is set */
4811 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4812 	CU_ASSERT(rc == 1);
4813 	CU_ASSERT(domains[0] == g_bdev_memory_domain);
4814 
4815 	/* get_supported_dma_device_types op is not set */
4816 	fn_table.get_memory_domains = NULL;
4817 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4818 	CU_ASSERT(rc == 0);
4819 }
4820 
4821 int
4822 main(int argc, char **argv)
4823 {
4824 	CU_pSuite		suite = NULL;
4825 	unsigned int		num_failures;
4826 
4827 	CU_set_error_action(CUEA_ABORT);
4828 	CU_initialize_registry();
4829 
4830 	suite = CU_add_suite("bdev", null_init, null_clean);
4831 
4832 	CU_ADD_TEST(suite, bytes_to_blocks_test);
4833 	CU_ADD_TEST(suite, num_blocks_test);
4834 	CU_ADD_TEST(suite, io_valid_test);
4835 	CU_ADD_TEST(suite, open_write_test);
4836 	CU_ADD_TEST(suite, alias_add_del_test);
4837 	CU_ADD_TEST(suite, get_device_stat_test);
4838 	CU_ADD_TEST(suite, bdev_io_types_test);
4839 	CU_ADD_TEST(suite, bdev_io_wait_test);
4840 	CU_ADD_TEST(suite, bdev_io_spans_split_test);
4841 	CU_ADD_TEST(suite, bdev_io_boundary_split_test);
4842 	CU_ADD_TEST(suite, bdev_io_max_size_and_segment_split_test);
4843 	CU_ADD_TEST(suite, bdev_io_mix_split_test);
4844 	CU_ADD_TEST(suite, bdev_io_split_with_io_wait);
4845 	CU_ADD_TEST(suite, bdev_io_alignment_with_boundary);
4846 	CU_ADD_TEST(suite, bdev_io_alignment);
4847 	CU_ADD_TEST(suite, bdev_histograms);
4848 	CU_ADD_TEST(suite, bdev_write_zeroes);
4849 	CU_ADD_TEST(suite, bdev_compare_and_write);
4850 	CU_ADD_TEST(suite, bdev_compare);
4851 	CU_ADD_TEST(suite, bdev_zcopy_write);
4852 	CU_ADD_TEST(suite, bdev_zcopy_read);
4853 	CU_ADD_TEST(suite, bdev_open_while_hotremove);
4854 	CU_ADD_TEST(suite, bdev_close_while_hotremove);
4855 	CU_ADD_TEST(suite, bdev_open_ext);
4856 	CU_ADD_TEST(suite, bdev_set_io_timeout);
4857 	CU_ADD_TEST(suite, lba_range_overlap);
4858 	CU_ADD_TEST(suite, lock_lba_range_check_ranges);
4859 	CU_ADD_TEST(suite, lock_lba_range_with_io_outstanding);
4860 	CU_ADD_TEST(suite, lock_lba_range_overlapped);
4861 	CU_ADD_TEST(suite, bdev_io_abort);
4862 	CU_ADD_TEST(suite, bdev_unmap);
4863 	CU_ADD_TEST(suite, bdev_write_zeroes_split_test);
4864 	CU_ADD_TEST(suite, bdev_set_options_test);
4865 	CU_ADD_TEST(suite, bdev_multi_allocation);
4866 	CU_ADD_TEST(suite, bdev_get_memory_domains);
4867 
4868 	allocate_cores(1);
4869 	allocate_threads(1);
4870 	set_thread(0);
4871 
4872 	CU_basic_set_mode(CU_BRM_VERBOSE);
4873 	CU_basic_run_tests();
4874 	num_failures = CU_get_number_of_failures();
4875 	CU_cleanup_registry();
4876 
4877 	free_threads();
4878 	free_cores();
4879 
4880 	return num_failures;
4881 }
4882