xref: /spdk/test/unit/lib/bdev/bdev.c/bdev_ut.c (revision 2f5c602574a98ede645991abe279a96e19c50196)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk_cunit.h"
35 
36 #include "common/lib/ut_multithread.c"
37 #include "unit/lib/json_mock.c"
38 
39 #include "spdk/config.h"
40 /* HACK: disable VTune integration so the unit test doesn't need VTune headers and libs to build */
41 #undef SPDK_CONFIG_VTUNE
42 
43 #include "bdev/bdev.c"
44 
45 DEFINE_STUB(spdk_notify_send, uint64_t, (const char *type, const char *ctx), 0);
46 DEFINE_STUB(spdk_notify_type_register, struct spdk_notify_type *, (const char *type), NULL);
47 
48 int g_status;
49 int g_count;
50 enum spdk_bdev_event_type g_event_type1;
51 enum spdk_bdev_event_type g_event_type2;
52 struct spdk_histogram_data *g_histogram;
53 void *g_unregister_arg;
54 int g_unregister_rc;
55 
56 void
57 spdk_scsi_nvme_translate(const struct spdk_bdev_io *bdev_io,
58 			 int *sc, int *sk, int *asc, int *ascq)
59 {
60 }
61 
62 static int
63 null_init(void)
64 {
65 	return 0;
66 }
67 
68 static int
69 null_clean(void)
70 {
71 	return 0;
72 }
73 
74 static int
75 stub_destruct(void *ctx)
76 {
77 	return 0;
78 }
79 
80 struct ut_expected_io {
81 	uint8_t				type;
82 	uint64_t			offset;
83 	uint64_t			length;
84 	int				iovcnt;
85 	struct iovec			iov[BDEV_IO_NUM_CHILD_IOV];
86 	void				*md_buf;
87 	TAILQ_ENTRY(ut_expected_io)	link;
88 };
89 
90 struct bdev_ut_channel {
91 	TAILQ_HEAD(, spdk_bdev_io)	outstanding_io;
92 	uint32_t			outstanding_io_count;
93 	TAILQ_HEAD(, ut_expected_io)	expected_io;
94 };
95 
96 static bool g_io_done;
97 static struct spdk_bdev_io *g_bdev_io;
98 static enum spdk_bdev_io_status g_io_status;
99 static enum spdk_bdev_io_status g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
100 static uint32_t g_bdev_ut_io_device;
101 static struct bdev_ut_channel *g_bdev_ut_channel;
102 static void *g_compare_read_buf;
103 static uint32_t g_compare_read_buf_len;
104 static void *g_compare_write_buf;
105 static uint32_t g_compare_write_buf_len;
106 static bool g_abort_done;
107 static enum spdk_bdev_io_status g_abort_status;
108 static void *g_zcopy_read_buf;
109 static uint32_t g_zcopy_read_buf_len;
110 static void *g_zcopy_write_buf;
111 static uint32_t g_zcopy_write_buf_len;
112 static struct spdk_bdev_io *g_zcopy_bdev_io;
113 
114 static struct ut_expected_io *
115 ut_alloc_expected_io(uint8_t type, uint64_t offset, uint64_t length, int iovcnt)
116 {
117 	struct ut_expected_io *expected_io;
118 
119 	expected_io = calloc(1, sizeof(*expected_io));
120 	SPDK_CU_ASSERT_FATAL(expected_io != NULL);
121 
122 	expected_io->type = type;
123 	expected_io->offset = offset;
124 	expected_io->length = length;
125 	expected_io->iovcnt = iovcnt;
126 
127 	return expected_io;
128 }
129 
130 static void
131 ut_expected_io_set_iov(struct ut_expected_io *expected_io, int pos, void *base, size_t len)
132 {
133 	expected_io->iov[pos].iov_base = base;
134 	expected_io->iov[pos].iov_len = len;
135 }
136 
137 static void
138 stub_submit_request(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
139 {
140 	struct bdev_ut_channel *ch = spdk_io_channel_get_ctx(_ch);
141 	struct ut_expected_io *expected_io;
142 	struct iovec *iov, *expected_iov;
143 	struct spdk_bdev_io *bio_to_abort;
144 	int i;
145 
146 	g_bdev_io = bdev_io;
147 
148 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
149 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
150 
151 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
152 		CU_ASSERT(g_compare_read_buf_len == len);
153 		memcpy(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len);
154 	}
155 
156 	if (g_compare_write_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
157 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
158 
159 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
160 		CU_ASSERT(g_compare_write_buf_len == len);
161 		memcpy(g_compare_write_buf, bdev_io->u.bdev.iovs[0].iov_base, len);
162 	}
163 
164 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_COMPARE) {
165 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
166 
167 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
168 		CU_ASSERT(g_compare_read_buf_len == len);
169 		if (memcmp(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len)) {
170 			g_io_exp_status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
171 		}
172 	}
173 
174 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) {
175 		if (g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS) {
176 			TAILQ_FOREACH(bio_to_abort, &ch->outstanding_io, module_link) {
177 				if (bio_to_abort == bdev_io->u.abort.bio_to_abort) {
178 					TAILQ_REMOVE(&ch->outstanding_io, bio_to_abort, module_link);
179 					ch->outstanding_io_count--;
180 					spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_FAILED);
181 					break;
182 				}
183 			}
184 		}
185 	}
186 
187 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) {
188 		if (bdev_io->u.bdev.zcopy.start) {
189 			g_zcopy_bdev_io = bdev_io;
190 			if (bdev_io->u.bdev.zcopy.populate) {
191 				/* Start of a read */
192 				CU_ASSERT(g_zcopy_read_buf != NULL);
193 				CU_ASSERT(g_zcopy_read_buf_len > 0);
194 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_read_buf;
195 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_read_buf_len;
196 				bdev_io->u.bdev.iovcnt = 1;
197 			} else {
198 				/* Start of a write */
199 				CU_ASSERT(g_zcopy_write_buf != NULL);
200 				CU_ASSERT(g_zcopy_write_buf_len > 0);
201 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_write_buf;
202 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_write_buf_len;
203 				bdev_io->u.bdev.iovcnt = 1;
204 			}
205 		} else {
206 			if (bdev_io->u.bdev.zcopy.commit) {
207 				/* End of write */
208 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_write_buf);
209 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_write_buf_len);
210 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
211 				g_zcopy_write_buf = NULL;
212 				g_zcopy_write_buf_len = 0;
213 			} else {
214 				/* End of read */
215 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_read_buf);
216 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_read_buf_len);
217 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
218 				g_zcopy_read_buf = NULL;
219 				g_zcopy_read_buf_len = 0;
220 			}
221 		}
222 	}
223 
224 	TAILQ_INSERT_TAIL(&ch->outstanding_io, bdev_io, module_link);
225 	ch->outstanding_io_count++;
226 
227 	expected_io = TAILQ_FIRST(&ch->expected_io);
228 	if (expected_io == NULL) {
229 		return;
230 	}
231 	TAILQ_REMOVE(&ch->expected_io, expected_io, link);
232 
233 	if (expected_io->type != SPDK_BDEV_IO_TYPE_INVALID) {
234 		CU_ASSERT(bdev_io->type == expected_io->type);
235 	}
236 
237 	if (expected_io->md_buf != NULL) {
238 		CU_ASSERT(expected_io->md_buf == bdev_io->u.bdev.md_buf);
239 	}
240 
241 	if (expected_io->length == 0) {
242 		free(expected_io);
243 		return;
244 	}
245 
246 	CU_ASSERT(expected_io->offset == bdev_io->u.bdev.offset_blocks);
247 	CU_ASSERT(expected_io->length = bdev_io->u.bdev.num_blocks);
248 
249 	if (expected_io->iovcnt == 0) {
250 		free(expected_io);
251 		/* UNMAP, WRITE_ZEROES and FLUSH don't have iovs, so we can just return now. */
252 		return;
253 	}
254 
255 	CU_ASSERT(expected_io->iovcnt == bdev_io->u.bdev.iovcnt);
256 	for (i = 0; i < expected_io->iovcnt; i++) {
257 		iov = &bdev_io->u.bdev.iovs[i];
258 		expected_iov = &expected_io->iov[i];
259 		CU_ASSERT(iov->iov_len == expected_iov->iov_len);
260 		CU_ASSERT(iov->iov_base == expected_iov->iov_base);
261 	}
262 
263 	free(expected_io);
264 }
265 
266 static void
267 stub_submit_request_get_buf_cb(struct spdk_io_channel *_ch,
268 			       struct spdk_bdev_io *bdev_io, bool success)
269 {
270 	CU_ASSERT(success == true);
271 
272 	stub_submit_request(_ch, bdev_io);
273 }
274 
275 static void
276 stub_submit_request_get_buf(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
277 {
278 	spdk_bdev_io_get_buf(bdev_io, stub_submit_request_get_buf_cb,
279 			     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
280 }
281 
282 static uint32_t
283 stub_complete_io(uint32_t num_to_complete)
284 {
285 	struct bdev_ut_channel *ch = g_bdev_ut_channel;
286 	struct spdk_bdev_io *bdev_io;
287 	static enum spdk_bdev_io_status io_status;
288 	uint32_t num_completed = 0;
289 
290 	while (num_completed < num_to_complete) {
291 		if (TAILQ_EMPTY(&ch->outstanding_io)) {
292 			break;
293 		}
294 		bdev_io = TAILQ_FIRST(&ch->outstanding_io);
295 		TAILQ_REMOVE(&ch->outstanding_io, bdev_io, module_link);
296 		ch->outstanding_io_count--;
297 		io_status = g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS ? SPDK_BDEV_IO_STATUS_SUCCESS :
298 			    g_io_exp_status;
299 		spdk_bdev_io_complete(bdev_io, io_status);
300 		num_completed++;
301 	}
302 
303 	return num_completed;
304 }
305 
306 static struct spdk_io_channel *
307 bdev_ut_get_io_channel(void *ctx)
308 {
309 	return spdk_get_io_channel(&g_bdev_ut_io_device);
310 }
311 
312 static bool g_io_types_supported[SPDK_BDEV_NUM_IO_TYPES] = {
313 	[SPDK_BDEV_IO_TYPE_READ]		= true,
314 	[SPDK_BDEV_IO_TYPE_WRITE]		= true,
315 	[SPDK_BDEV_IO_TYPE_COMPARE]		= true,
316 	[SPDK_BDEV_IO_TYPE_UNMAP]		= true,
317 	[SPDK_BDEV_IO_TYPE_FLUSH]		= true,
318 	[SPDK_BDEV_IO_TYPE_RESET]		= true,
319 	[SPDK_BDEV_IO_TYPE_NVME_ADMIN]		= true,
320 	[SPDK_BDEV_IO_TYPE_NVME_IO]		= true,
321 	[SPDK_BDEV_IO_TYPE_NVME_IO_MD]		= true,
322 	[SPDK_BDEV_IO_TYPE_WRITE_ZEROES]	= true,
323 	[SPDK_BDEV_IO_TYPE_ZCOPY]		= true,
324 	[SPDK_BDEV_IO_TYPE_ABORT]		= true,
325 };
326 
327 static void
328 ut_enable_io_type(enum spdk_bdev_io_type io_type, bool enable)
329 {
330 	g_io_types_supported[io_type] = enable;
331 }
332 
333 static bool
334 stub_io_type_supported(void *_bdev, enum spdk_bdev_io_type io_type)
335 {
336 	return g_io_types_supported[io_type];
337 }
338 
339 static struct spdk_bdev_fn_table fn_table = {
340 	.destruct = stub_destruct,
341 	.submit_request = stub_submit_request,
342 	.get_io_channel = bdev_ut_get_io_channel,
343 	.io_type_supported = stub_io_type_supported,
344 };
345 
346 static int
347 bdev_ut_create_ch(void *io_device, void *ctx_buf)
348 {
349 	struct bdev_ut_channel *ch = ctx_buf;
350 
351 	CU_ASSERT(g_bdev_ut_channel == NULL);
352 	g_bdev_ut_channel = ch;
353 
354 	TAILQ_INIT(&ch->outstanding_io);
355 	ch->outstanding_io_count = 0;
356 	TAILQ_INIT(&ch->expected_io);
357 	return 0;
358 }
359 
360 static void
361 bdev_ut_destroy_ch(void *io_device, void *ctx_buf)
362 {
363 	CU_ASSERT(g_bdev_ut_channel != NULL);
364 	g_bdev_ut_channel = NULL;
365 }
366 
367 struct spdk_bdev_module bdev_ut_if;
368 
369 static int
370 bdev_ut_module_init(void)
371 {
372 	spdk_io_device_register(&g_bdev_ut_io_device, bdev_ut_create_ch, bdev_ut_destroy_ch,
373 				sizeof(struct bdev_ut_channel), NULL);
374 	spdk_bdev_module_init_done(&bdev_ut_if);
375 	return 0;
376 }
377 
378 static void
379 bdev_ut_module_fini(void)
380 {
381 	spdk_io_device_unregister(&g_bdev_ut_io_device, NULL);
382 }
383 
384 struct spdk_bdev_module bdev_ut_if = {
385 	.name = "bdev_ut",
386 	.module_init = bdev_ut_module_init,
387 	.module_fini = bdev_ut_module_fini,
388 	.async_init = true,
389 };
390 
391 static void vbdev_ut_examine(struct spdk_bdev *bdev);
392 
393 static int
394 vbdev_ut_module_init(void)
395 {
396 	return 0;
397 }
398 
399 static void
400 vbdev_ut_module_fini(void)
401 {
402 }
403 
404 struct spdk_bdev_module vbdev_ut_if = {
405 	.name = "vbdev_ut",
406 	.module_init = vbdev_ut_module_init,
407 	.module_fini = vbdev_ut_module_fini,
408 	.examine_config = vbdev_ut_examine,
409 };
410 
411 SPDK_BDEV_MODULE_REGISTER(bdev_ut, &bdev_ut_if)
412 SPDK_BDEV_MODULE_REGISTER(vbdev_ut, &vbdev_ut_if)
413 
414 static void
415 vbdev_ut_examine(struct spdk_bdev *bdev)
416 {
417 	spdk_bdev_module_examine_done(&vbdev_ut_if);
418 }
419 
420 static struct spdk_bdev *
421 allocate_bdev(char *name)
422 {
423 	struct spdk_bdev *bdev;
424 	int rc;
425 
426 	bdev = calloc(1, sizeof(*bdev));
427 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
428 
429 	bdev->name = name;
430 	bdev->fn_table = &fn_table;
431 	bdev->module = &bdev_ut_if;
432 	bdev->blockcnt = 1024;
433 	bdev->blocklen = 512;
434 
435 	rc = spdk_bdev_register(bdev);
436 	CU_ASSERT(rc == 0);
437 
438 	return bdev;
439 }
440 
441 static struct spdk_bdev *
442 allocate_vbdev(char *name)
443 {
444 	struct spdk_bdev *bdev;
445 	int rc;
446 
447 	bdev = calloc(1, sizeof(*bdev));
448 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
449 
450 	bdev->name = name;
451 	bdev->fn_table = &fn_table;
452 	bdev->module = &vbdev_ut_if;
453 
454 	rc = spdk_bdev_register(bdev);
455 	CU_ASSERT(rc == 0);
456 
457 	return bdev;
458 }
459 
460 static void
461 free_bdev(struct spdk_bdev *bdev)
462 {
463 	spdk_bdev_unregister(bdev, NULL, NULL);
464 	poll_threads();
465 	memset(bdev, 0xFF, sizeof(*bdev));
466 	free(bdev);
467 }
468 
469 static void
470 free_vbdev(struct spdk_bdev *bdev)
471 {
472 	spdk_bdev_unregister(bdev, NULL, NULL);
473 	poll_threads();
474 	memset(bdev, 0xFF, sizeof(*bdev));
475 	free(bdev);
476 }
477 
478 static void
479 get_device_stat_cb(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat, void *cb_arg, int rc)
480 {
481 	const char *bdev_name;
482 
483 	CU_ASSERT(bdev != NULL);
484 	CU_ASSERT(rc == 0);
485 	bdev_name = spdk_bdev_get_name(bdev);
486 	CU_ASSERT_STRING_EQUAL(bdev_name, "bdev0");
487 
488 	free(stat);
489 
490 	*(bool *)cb_arg = true;
491 }
492 
493 static void
494 bdev_unregister_cb(void *cb_arg, int rc)
495 {
496 	g_unregister_arg = cb_arg;
497 	g_unregister_rc = rc;
498 }
499 
500 static void
501 bdev_ut_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
502 {
503 }
504 
505 static void
506 bdev_open_cb1(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
507 {
508 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
509 
510 	g_event_type1 = type;
511 	if (SPDK_BDEV_EVENT_REMOVE == type) {
512 		spdk_bdev_close(desc);
513 	}
514 }
515 
516 static void
517 bdev_open_cb2(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
518 {
519 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
520 
521 	g_event_type2 = type;
522 	if (SPDK_BDEV_EVENT_REMOVE == type) {
523 		spdk_bdev_close(desc);
524 	}
525 }
526 
527 static void
528 get_device_stat_test(void)
529 {
530 	struct spdk_bdev *bdev;
531 	struct spdk_bdev_io_stat *stat;
532 	bool done;
533 
534 	bdev = allocate_bdev("bdev0");
535 	stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
536 	if (stat == NULL) {
537 		free_bdev(bdev);
538 		return;
539 	}
540 
541 	done = false;
542 	spdk_bdev_get_device_stat(bdev, stat, get_device_stat_cb, &done);
543 	while (!done) { poll_threads(); }
544 
545 	free_bdev(bdev);
546 }
547 
548 static void
549 open_write_test(void)
550 {
551 	struct spdk_bdev *bdev[9];
552 	struct spdk_bdev_desc *desc[9] = {};
553 	int rc;
554 
555 	/*
556 	 * Create a tree of bdevs to test various open w/ write cases.
557 	 *
558 	 * bdev0 through bdev3 are physical block devices, such as NVMe
559 	 * namespaces or Ceph block devices.
560 	 *
561 	 * bdev4 is a virtual bdev with multiple base bdevs.  This models
562 	 * caching or RAID use cases.
563 	 *
564 	 * bdev5 through bdev7 are all virtual bdevs with the same base
565 	 * bdev (except bdev7). This models partitioning or logical volume
566 	 * use cases.
567 	 *
568 	 * bdev7 is a virtual bdev with multiple base bdevs. One of base bdevs
569 	 * (bdev2) is shared with other virtual bdevs: bdev5 and bdev6. This
570 	 * models caching, RAID, partitioning or logical volumes use cases.
571 	 *
572 	 * bdev8 is a virtual bdev with multiple base bdevs, but these
573 	 * base bdevs are themselves virtual bdevs.
574 	 *
575 	 *                bdev8
576 	 *                  |
577 	 *            +----------+
578 	 *            |          |
579 	 *          bdev4      bdev5   bdev6   bdev7
580 	 *            |          |       |       |
581 	 *        +---+---+      +---+   +   +---+---+
582 	 *        |       |           \  |  /         \
583 	 *      bdev0   bdev1          bdev2         bdev3
584 	 */
585 
586 	bdev[0] = allocate_bdev("bdev0");
587 	rc = spdk_bdev_module_claim_bdev(bdev[0], NULL, &bdev_ut_if);
588 	CU_ASSERT(rc == 0);
589 
590 	bdev[1] = allocate_bdev("bdev1");
591 	rc = spdk_bdev_module_claim_bdev(bdev[1], NULL, &bdev_ut_if);
592 	CU_ASSERT(rc == 0);
593 
594 	bdev[2] = allocate_bdev("bdev2");
595 	rc = spdk_bdev_module_claim_bdev(bdev[2], NULL, &bdev_ut_if);
596 	CU_ASSERT(rc == 0);
597 
598 	bdev[3] = allocate_bdev("bdev3");
599 	rc = spdk_bdev_module_claim_bdev(bdev[3], NULL, &bdev_ut_if);
600 	CU_ASSERT(rc == 0);
601 
602 	bdev[4] = allocate_vbdev("bdev4");
603 	rc = spdk_bdev_module_claim_bdev(bdev[4], NULL, &bdev_ut_if);
604 	CU_ASSERT(rc == 0);
605 
606 	bdev[5] = allocate_vbdev("bdev5");
607 	rc = spdk_bdev_module_claim_bdev(bdev[5], NULL, &bdev_ut_if);
608 	CU_ASSERT(rc == 0);
609 
610 	bdev[6] = allocate_vbdev("bdev6");
611 
612 	bdev[7] = allocate_vbdev("bdev7");
613 
614 	bdev[8] = allocate_vbdev("bdev8");
615 
616 	/* Open bdev0 read-only.  This should succeed. */
617 	rc = spdk_bdev_open_ext("bdev0", false, bdev_ut_event_cb, NULL, &desc[0]);
618 	CU_ASSERT(rc == 0);
619 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
620 	CU_ASSERT(bdev[0] == spdk_bdev_desc_get_bdev(desc[0]));
621 	spdk_bdev_close(desc[0]);
622 
623 	/*
624 	 * Open bdev1 read/write.  This should fail since bdev1 has been claimed
625 	 * by a vbdev module.
626 	 */
627 	rc = spdk_bdev_open_ext("bdev1", true, bdev_ut_event_cb, NULL, &desc[1]);
628 	CU_ASSERT(rc == -EPERM);
629 
630 	/*
631 	 * Open bdev4 read/write.  This should fail since bdev3 has been claimed
632 	 * by a vbdev module.
633 	 */
634 	rc = spdk_bdev_open_ext("bdev4", true, bdev_ut_event_cb, NULL, &desc[4]);
635 	CU_ASSERT(rc == -EPERM);
636 
637 	/* Open bdev4 read-only.  This should succeed. */
638 	rc = spdk_bdev_open_ext("bdev4", false, bdev_ut_event_cb, NULL, &desc[4]);
639 	CU_ASSERT(rc == 0);
640 	SPDK_CU_ASSERT_FATAL(desc[4] != NULL);
641 	CU_ASSERT(bdev[4] == spdk_bdev_desc_get_bdev(desc[4]));
642 	spdk_bdev_close(desc[4]);
643 
644 	/*
645 	 * Open bdev8 read/write.  This should succeed since it is a leaf
646 	 * bdev.
647 	 */
648 	rc = spdk_bdev_open_ext("bdev8", true, bdev_ut_event_cb, NULL, &desc[8]);
649 	CU_ASSERT(rc == 0);
650 	SPDK_CU_ASSERT_FATAL(desc[8] != NULL);
651 	CU_ASSERT(bdev[8] == spdk_bdev_desc_get_bdev(desc[8]));
652 	spdk_bdev_close(desc[8]);
653 
654 	/*
655 	 * Open bdev5 read/write.  This should fail since bdev4 has been claimed
656 	 * by a vbdev module.
657 	 */
658 	rc = spdk_bdev_open_ext("bdev5", true, bdev_ut_event_cb, NULL, &desc[5]);
659 	CU_ASSERT(rc == -EPERM);
660 
661 	/* Open bdev4 read-only.  This should succeed. */
662 	rc = spdk_bdev_open_ext("bdev5", false, bdev_ut_event_cb, NULL, &desc[5]);
663 	CU_ASSERT(rc == 0);
664 	SPDK_CU_ASSERT_FATAL(desc[5] != NULL);
665 	CU_ASSERT(bdev[5] == spdk_bdev_desc_get_bdev(desc[5]));
666 	spdk_bdev_close(desc[5]);
667 
668 	free_vbdev(bdev[8]);
669 
670 	free_vbdev(bdev[5]);
671 	free_vbdev(bdev[6]);
672 	free_vbdev(bdev[7]);
673 
674 	free_vbdev(bdev[4]);
675 
676 	free_bdev(bdev[0]);
677 	free_bdev(bdev[1]);
678 	free_bdev(bdev[2]);
679 	free_bdev(bdev[3]);
680 }
681 
682 static void
683 bytes_to_blocks_test(void)
684 {
685 	struct spdk_bdev bdev;
686 	uint64_t offset_blocks, num_blocks;
687 
688 	memset(&bdev, 0, sizeof(bdev));
689 
690 	bdev.blocklen = 512;
691 
692 	/* All parameters valid */
693 	offset_blocks = 0;
694 	num_blocks = 0;
695 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 1024, &num_blocks) == 0);
696 	CU_ASSERT(offset_blocks == 1);
697 	CU_ASSERT(num_blocks == 2);
698 
699 	/* Offset not a block multiple */
700 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 512, &num_blocks) != 0);
701 
702 	/* Length not a block multiple */
703 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 3, &num_blocks) != 0);
704 
705 	/* In case blocklen not the power of two */
706 	bdev.blocklen = 100;
707 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 200, &num_blocks) == 0);
708 	CU_ASSERT(offset_blocks == 1);
709 	CU_ASSERT(num_blocks == 2);
710 
711 	/* Offset not a block multiple */
712 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 100, &num_blocks) != 0);
713 
714 	/* Length not a block multiple */
715 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 3, &num_blocks) != 0);
716 }
717 
718 static void
719 num_blocks_test(void)
720 {
721 	struct spdk_bdev bdev;
722 	struct spdk_bdev_desc *desc = NULL;
723 	int rc;
724 
725 	memset(&bdev, 0, sizeof(bdev));
726 	bdev.name = "num_blocks";
727 	bdev.fn_table = &fn_table;
728 	bdev.module = &bdev_ut_if;
729 	spdk_bdev_register(&bdev);
730 	spdk_bdev_notify_blockcnt_change(&bdev, 50);
731 
732 	/* Growing block number */
733 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 70) == 0);
734 	/* Shrinking block number */
735 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 30) == 0);
736 
737 	rc = spdk_bdev_open_ext("num_blocks", false, bdev_open_cb1, &desc, &desc);
738 	CU_ASSERT(rc == 0);
739 	SPDK_CU_ASSERT_FATAL(desc != NULL);
740 	CU_ASSERT(&bdev == spdk_bdev_desc_get_bdev(desc));
741 
742 	/* Growing block number */
743 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 80) == 0);
744 	/* Shrinking block number */
745 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 20) != 0);
746 
747 	g_event_type1 = 0xFF;
748 	/* Growing block number */
749 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 90) == 0);
750 
751 	poll_threads();
752 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_RESIZE);
753 
754 	g_event_type1 = 0xFF;
755 	/* Growing block number and closing */
756 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 100) == 0);
757 
758 	spdk_bdev_close(desc);
759 	spdk_bdev_unregister(&bdev, NULL, NULL);
760 
761 	poll_threads();
762 
763 	/* Callback is not called for closed device */
764 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
765 }
766 
767 static void
768 io_valid_test(void)
769 {
770 	struct spdk_bdev bdev;
771 
772 	memset(&bdev, 0, sizeof(bdev));
773 
774 	bdev.blocklen = 512;
775 	CU_ASSERT(pthread_mutex_init(&bdev.internal.mutex, NULL) == 0);
776 
777 	spdk_bdev_notify_blockcnt_change(&bdev, 100);
778 
779 	/* All parameters valid */
780 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 1, 2) == true);
781 
782 	/* Last valid block */
783 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 1) == true);
784 
785 	/* Offset past end of bdev */
786 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 100, 1) == false);
787 
788 	/* Offset + length past end of bdev */
789 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 2) == false);
790 
791 	/* Offset near end of uint64_t range (2^64 - 1) */
792 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 18446744073709551615ULL, 1) == false);
793 
794 	CU_ASSERT(pthread_mutex_destroy(&bdev.internal.mutex) == 0);
795 }
796 
797 static void
798 alias_add_del_test(void)
799 {
800 	struct spdk_bdev *bdev[3];
801 	int rc;
802 
803 	/* Creating and registering bdevs */
804 	bdev[0] = allocate_bdev("bdev0");
805 	SPDK_CU_ASSERT_FATAL(bdev[0] != 0);
806 
807 	bdev[1] = allocate_bdev("bdev1");
808 	SPDK_CU_ASSERT_FATAL(bdev[1] != 0);
809 
810 	bdev[2] = allocate_bdev("bdev2");
811 	SPDK_CU_ASSERT_FATAL(bdev[2] != 0);
812 
813 	poll_threads();
814 
815 	/*
816 	 * Trying adding an alias identical to name.
817 	 * Alias is identical to name, so it can not be added to aliases list
818 	 */
819 	rc = spdk_bdev_alias_add(bdev[0], bdev[0]->name);
820 	CU_ASSERT(rc == -EEXIST);
821 
822 	/*
823 	 * Trying to add empty alias,
824 	 * this one should fail
825 	 */
826 	rc = spdk_bdev_alias_add(bdev[0], NULL);
827 	CU_ASSERT(rc == -EINVAL);
828 
829 	/* Trying adding same alias to two different registered bdevs */
830 
831 	/* Alias is used first time, so this one should pass */
832 	rc = spdk_bdev_alias_add(bdev[0], "proper alias 0");
833 	CU_ASSERT(rc == 0);
834 
835 	/* Alias was added to another bdev, so this one should fail */
836 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 0");
837 	CU_ASSERT(rc == -EEXIST);
838 
839 	/* Alias is used first time, so this one should pass */
840 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 1");
841 	CU_ASSERT(rc == 0);
842 
843 	/* Trying removing an alias from registered bdevs */
844 
845 	/* Alias is not on a bdev aliases list, so this one should fail */
846 	rc = spdk_bdev_alias_del(bdev[0], "not existing");
847 	CU_ASSERT(rc == -ENOENT);
848 
849 	/* Alias is present on a bdev aliases list, so this one should pass */
850 	rc = spdk_bdev_alias_del(bdev[0], "proper alias 0");
851 	CU_ASSERT(rc == 0);
852 
853 	/* Alias is present on a bdev aliases list, so this one should pass */
854 	rc = spdk_bdev_alias_del(bdev[1], "proper alias 1");
855 	CU_ASSERT(rc == 0);
856 
857 	/* Trying to remove name instead of alias, so this one should fail, name cannot be changed or removed */
858 	rc = spdk_bdev_alias_del(bdev[0], bdev[0]->name);
859 	CU_ASSERT(rc != 0);
860 
861 	/* Trying to del all alias from empty alias list */
862 	spdk_bdev_alias_del_all(bdev[2]);
863 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev[2]->aliases));
864 
865 	/* Trying to del all alias from non-empty alias list */
866 	rc = spdk_bdev_alias_add(bdev[2], "alias0");
867 	CU_ASSERT(rc == 0);
868 	rc = spdk_bdev_alias_add(bdev[2], "alias1");
869 	CU_ASSERT(rc == 0);
870 	spdk_bdev_alias_del_all(bdev[2]);
871 	CU_ASSERT(TAILQ_EMPTY(&bdev[2]->aliases));
872 
873 	/* Unregister and free bdevs */
874 	spdk_bdev_unregister(bdev[0], NULL, NULL);
875 	spdk_bdev_unregister(bdev[1], NULL, NULL);
876 	spdk_bdev_unregister(bdev[2], NULL, NULL);
877 
878 	poll_threads();
879 
880 	free(bdev[0]);
881 	free(bdev[1]);
882 	free(bdev[2]);
883 }
884 
885 static void
886 io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
887 {
888 	g_io_done = true;
889 	g_io_status = bdev_io->internal.status;
890 	if ((bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) &&
891 	    (bdev_io->u.bdev.zcopy.start)) {
892 		g_zcopy_bdev_io = bdev_io;
893 	} else {
894 		spdk_bdev_free_io(bdev_io);
895 		g_zcopy_bdev_io = NULL;
896 	}
897 }
898 
899 static void
900 bdev_init_cb(void *arg, int rc)
901 {
902 	CU_ASSERT(rc == 0);
903 }
904 
905 static void
906 bdev_fini_cb(void *arg)
907 {
908 }
909 
910 struct bdev_ut_io_wait_entry {
911 	struct spdk_bdev_io_wait_entry	entry;
912 	struct spdk_io_channel		*io_ch;
913 	struct spdk_bdev_desc		*desc;
914 	bool				submitted;
915 };
916 
917 static void
918 io_wait_cb(void *arg)
919 {
920 	struct bdev_ut_io_wait_entry *entry = arg;
921 	int rc;
922 
923 	rc = spdk_bdev_read_blocks(entry->desc, entry->io_ch, NULL, 0, 1, io_done, NULL);
924 	CU_ASSERT(rc == 0);
925 	entry->submitted = true;
926 }
927 
928 static void
929 bdev_io_types_test(void)
930 {
931 	struct spdk_bdev *bdev;
932 	struct spdk_bdev_desc *desc = NULL;
933 	struct spdk_io_channel *io_ch;
934 	struct spdk_bdev_opts bdev_opts = {};
935 	int rc;
936 
937 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
938 	bdev_opts.bdev_io_pool_size = 4;
939 	bdev_opts.bdev_io_cache_size = 2;
940 
941 	rc = spdk_bdev_set_opts(&bdev_opts);
942 	CU_ASSERT(rc == 0);
943 	spdk_bdev_initialize(bdev_init_cb, NULL);
944 	poll_threads();
945 
946 	bdev = allocate_bdev("bdev0");
947 
948 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
949 	CU_ASSERT(rc == 0);
950 	poll_threads();
951 	SPDK_CU_ASSERT_FATAL(desc != NULL);
952 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
953 	io_ch = spdk_bdev_get_io_channel(desc);
954 	CU_ASSERT(io_ch != NULL);
955 
956 	/* WRITE and WRITE ZEROES are not supported */
957 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
958 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, false);
959 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 0, 128, io_done, NULL);
960 	CU_ASSERT(rc == -ENOTSUP);
961 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
962 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, true);
963 
964 	spdk_put_io_channel(io_ch);
965 	spdk_bdev_close(desc);
966 	free_bdev(bdev);
967 	spdk_bdev_finish(bdev_fini_cb, NULL);
968 	poll_threads();
969 }
970 
971 static void
972 bdev_io_wait_test(void)
973 {
974 	struct spdk_bdev *bdev;
975 	struct spdk_bdev_desc *desc = NULL;
976 	struct spdk_io_channel *io_ch;
977 	struct spdk_bdev_opts bdev_opts = {};
978 	struct bdev_ut_io_wait_entry io_wait_entry;
979 	struct bdev_ut_io_wait_entry io_wait_entry2;
980 	int rc;
981 
982 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
983 	bdev_opts.bdev_io_pool_size = 4;
984 	bdev_opts.bdev_io_cache_size = 2;
985 
986 	rc = spdk_bdev_set_opts(&bdev_opts);
987 	CU_ASSERT(rc == 0);
988 	spdk_bdev_initialize(bdev_init_cb, NULL);
989 	poll_threads();
990 
991 	bdev = allocate_bdev("bdev0");
992 
993 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
994 	CU_ASSERT(rc == 0);
995 	poll_threads();
996 	SPDK_CU_ASSERT_FATAL(desc != NULL);
997 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
998 	io_ch = spdk_bdev_get_io_channel(desc);
999 	CU_ASSERT(io_ch != NULL);
1000 
1001 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1002 	CU_ASSERT(rc == 0);
1003 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1004 	CU_ASSERT(rc == 0);
1005 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1006 	CU_ASSERT(rc == 0);
1007 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1008 	CU_ASSERT(rc == 0);
1009 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1010 
1011 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1012 	CU_ASSERT(rc == -ENOMEM);
1013 
1014 	io_wait_entry.entry.bdev = bdev;
1015 	io_wait_entry.entry.cb_fn = io_wait_cb;
1016 	io_wait_entry.entry.cb_arg = &io_wait_entry;
1017 	io_wait_entry.io_ch = io_ch;
1018 	io_wait_entry.desc = desc;
1019 	io_wait_entry.submitted = false;
1020 	/* Cannot use the same io_wait_entry for two different calls. */
1021 	memcpy(&io_wait_entry2, &io_wait_entry, sizeof(io_wait_entry));
1022 	io_wait_entry2.entry.cb_arg = &io_wait_entry2;
1023 
1024 	/* Queue two I/O waits. */
1025 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry.entry);
1026 	CU_ASSERT(rc == 0);
1027 	CU_ASSERT(io_wait_entry.submitted == false);
1028 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry2.entry);
1029 	CU_ASSERT(rc == 0);
1030 	CU_ASSERT(io_wait_entry2.submitted == false);
1031 
1032 	stub_complete_io(1);
1033 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1034 	CU_ASSERT(io_wait_entry.submitted == true);
1035 	CU_ASSERT(io_wait_entry2.submitted == false);
1036 
1037 	stub_complete_io(1);
1038 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1039 	CU_ASSERT(io_wait_entry2.submitted == true);
1040 
1041 	stub_complete_io(4);
1042 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1043 
1044 	spdk_put_io_channel(io_ch);
1045 	spdk_bdev_close(desc);
1046 	free_bdev(bdev);
1047 	spdk_bdev_finish(bdev_fini_cb, NULL);
1048 	poll_threads();
1049 }
1050 
1051 static void
1052 bdev_io_spans_split_test(void)
1053 {
1054 	struct spdk_bdev bdev;
1055 	struct spdk_bdev_io bdev_io;
1056 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV];
1057 
1058 	memset(&bdev, 0, sizeof(bdev));
1059 	bdev_io.u.bdev.iovs = iov;
1060 
1061 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1062 	bdev.optimal_io_boundary = 0;
1063 	bdev.max_segment_size = 0;
1064 	bdev.max_num_segments = 0;
1065 	bdev_io.bdev = &bdev;
1066 
1067 	/* bdev has no optimal_io_boundary and max_size set - so this should return false. */
1068 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1069 
1070 	bdev.split_on_optimal_io_boundary = true;
1071 	bdev.optimal_io_boundary = 32;
1072 	bdev_io.type = SPDK_BDEV_IO_TYPE_RESET;
1073 
1074 	/* RESETs are not based on LBAs - so this should return false. */
1075 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1076 
1077 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1078 	bdev_io.u.bdev.offset_blocks = 0;
1079 	bdev_io.u.bdev.num_blocks = 32;
1080 
1081 	/* This I/O run right up to, but does not cross, the boundary - so this should return false. */
1082 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1083 
1084 	bdev_io.u.bdev.num_blocks = 33;
1085 
1086 	/* This I/O spans a boundary. */
1087 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1088 
1089 	bdev_io.u.bdev.num_blocks = 32;
1090 	bdev.max_segment_size = 512 * 32;
1091 	bdev.max_num_segments = 1;
1092 	bdev_io.u.bdev.iovcnt = 1;
1093 	iov[0].iov_len = 512;
1094 
1095 	/* Does not cross and exceed max_size or max_segs */
1096 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1097 
1098 	bdev.split_on_optimal_io_boundary = false;
1099 	bdev.max_segment_size = 512;
1100 	bdev.max_num_segments = 1;
1101 	bdev_io.u.bdev.iovcnt = 2;
1102 
1103 	/* Exceed max_segs */
1104 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1105 
1106 	bdev.max_num_segments = 2;
1107 	iov[0].iov_len = 513;
1108 	iov[1].iov_len = 512;
1109 
1110 	/* Exceed max_sizes */
1111 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1112 }
1113 
1114 static void
1115 bdev_io_boundary_split_test(void)
1116 {
1117 	struct spdk_bdev *bdev;
1118 	struct spdk_bdev_desc *desc = NULL;
1119 	struct spdk_io_channel *io_ch;
1120 	struct spdk_bdev_opts bdev_opts = {};
1121 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1122 	struct ut_expected_io *expected_io;
1123 	void *md_buf = (void *)0xFF000000;
1124 	uint64_t i;
1125 	int rc;
1126 
1127 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1128 	bdev_opts.bdev_io_pool_size = 512;
1129 	bdev_opts.bdev_io_cache_size = 64;
1130 
1131 	rc = spdk_bdev_set_opts(&bdev_opts);
1132 	CU_ASSERT(rc == 0);
1133 	spdk_bdev_initialize(bdev_init_cb, NULL);
1134 
1135 	bdev = allocate_bdev("bdev0");
1136 
1137 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1138 	CU_ASSERT(rc == 0);
1139 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1140 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1141 	io_ch = spdk_bdev_get_io_channel(desc);
1142 	CU_ASSERT(io_ch != NULL);
1143 
1144 	bdev->optimal_io_boundary = 16;
1145 	bdev->split_on_optimal_io_boundary = false;
1146 
1147 	g_io_done = false;
1148 
1149 	/* First test that the I/O does not get split if split_on_optimal_io_boundary == false. */
1150 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 8, 1);
1151 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 8 * 512);
1152 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1153 
1154 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
1155 	CU_ASSERT(rc == 0);
1156 	CU_ASSERT(g_io_done == false);
1157 
1158 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1159 	stub_complete_io(1);
1160 	CU_ASSERT(g_io_done == true);
1161 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1162 
1163 	bdev->split_on_optimal_io_boundary = true;
1164 	bdev->md_interleave = false;
1165 	bdev->md_len = 8;
1166 
1167 	/* Now test that a single-vector command is split correctly.
1168 	 * Offset 14, length 8, payload 0xF000
1169 	 *  Child - Offset 14, length 2, payload 0xF000
1170 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
1171 	 *
1172 	 * Set up the expected values before calling spdk_bdev_read_blocks
1173 	 */
1174 	g_io_done = false;
1175 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
1176 	expected_io->md_buf = md_buf;
1177 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
1178 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1179 
1180 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
1181 	expected_io->md_buf = md_buf + 2 * 8;
1182 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
1183 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1184 
1185 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1186 	rc = spdk_bdev_read_blocks_with_md(desc, io_ch, (void *)0xF000, md_buf,
1187 					   14, 8, io_done, NULL);
1188 	CU_ASSERT(rc == 0);
1189 	CU_ASSERT(g_io_done == false);
1190 
1191 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1192 	stub_complete_io(2);
1193 	CU_ASSERT(g_io_done == true);
1194 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1195 
1196 	/* Now set up a more complex, multi-vector command that needs to be split,
1197 	 *  including splitting iovecs.
1198 	 */
1199 	iov[0].iov_base = (void *)0x10000;
1200 	iov[0].iov_len = 512;
1201 	iov[1].iov_base = (void *)0x20000;
1202 	iov[1].iov_len = 20 * 512;
1203 	iov[2].iov_base = (void *)0x30000;
1204 	iov[2].iov_len = 11 * 512;
1205 
1206 	g_io_done = false;
1207 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
1208 	expected_io->md_buf = md_buf;
1209 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
1210 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
1211 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1212 
1213 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
1214 	expected_io->md_buf = md_buf + 2 * 8;
1215 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
1216 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1217 
1218 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
1219 	expected_io->md_buf = md_buf + 18 * 8;
1220 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
1221 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
1222 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1223 
1224 	rc = spdk_bdev_writev_blocks_with_md(desc, io_ch, iov, 3, md_buf,
1225 					     14, 32, io_done, NULL);
1226 	CU_ASSERT(rc == 0);
1227 	CU_ASSERT(g_io_done == false);
1228 
1229 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
1230 	stub_complete_io(3);
1231 	CU_ASSERT(g_io_done == true);
1232 
1233 	/* Test multi vector command that needs to be split by strip and then needs to be
1234 	 * split further due to the capacity of child iovs.
1235 	 */
1236 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
1237 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1238 		iov[i].iov_len = 512;
1239 	}
1240 
1241 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1242 	g_io_done = false;
1243 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1244 					   BDEV_IO_NUM_CHILD_IOV);
1245 	expected_io->md_buf = md_buf;
1246 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1247 		ut_expected_io_set_iov(expected_io, i, (void *)((i + 1) * 0x10000), 512);
1248 	}
1249 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1250 
1251 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1252 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
1253 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1254 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1255 		ut_expected_io_set_iov(expected_io, i,
1256 				       (void *)((i + 1 + BDEV_IO_NUM_CHILD_IOV) * 0x10000), 512);
1257 	}
1258 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1259 
1260 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1261 					    0, BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1262 	CU_ASSERT(rc == 0);
1263 	CU_ASSERT(g_io_done == false);
1264 
1265 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1266 	stub_complete_io(1);
1267 	CU_ASSERT(g_io_done == false);
1268 
1269 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1270 	stub_complete_io(1);
1271 	CU_ASSERT(g_io_done == true);
1272 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1273 
1274 	/* Test multi vector command that needs to be split by strip and then needs to be
1275 	 * split further due to the capacity of child iovs. In this case, the length of
1276 	 * the rest of iovec array with an I/O boundary is the multiple of block size.
1277 	 */
1278 
1279 	/* Fill iovec array for exactly one boundary. The iovec cnt for this boundary
1280 	 * is BDEV_IO_NUM_CHILD_IOV + 1, which exceeds the capacity of child iovs.
1281 	 */
1282 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1283 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1284 		iov[i].iov_len = 512;
1285 	}
1286 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1287 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1288 		iov[i].iov_len = 256;
1289 	}
1290 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1291 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 512;
1292 
1293 	/* Add an extra iovec to trigger split */
1294 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1295 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1296 
1297 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1298 	g_io_done = false;
1299 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1300 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV);
1301 	expected_io->md_buf = md_buf;
1302 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1303 		ut_expected_io_set_iov(expected_io, i,
1304 				       (void *)((i + 1) * 0x10000), 512);
1305 	}
1306 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1307 		ut_expected_io_set_iov(expected_io, i,
1308 				       (void *)((i + 1) * 0x10000), 256);
1309 	}
1310 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1311 
1312 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1313 					   1, 1);
1314 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1315 	ut_expected_io_set_iov(expected_io, 0,
1316 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 512);
1317 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1318 
1319 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1320 					   1, 1);
1321 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1322 	ut_expected_io_set_iov(expected_io, 0,
1323 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1324 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1325 
1326 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, md_buf,
1327 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1328 	CU_ASSERT(rc == 0);
1329 	CU_ASSERT(g_io_done == false);
1330 
1331 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1332 	stub_complete_io(1);
1333 	CU_ASSERT(g_io_done == false);
1334 
1335 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1336 	stub_complete_io(2);
1337 	CU_ASSERT(g_io_done == true);
1338 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1339 
1340 	/* Test multi vector command that needs to be split by strip and then needs to be
1341 	 * split further due to the capacity of child iovs, the child request offset should
1342 	 * be rewind to last aligned offset and go success without error.
1343 	 */
1344 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1345 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1346 		iov[i].iov_len = 512;
1347 	}
1348 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1349 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1350 
1351 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1352 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1353 
1354 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1355 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1356 
1357 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1358 	g_io_done = false;
1359 	g_io_status = 0;
1360 	/* The first expected io should be start from offset 0 to BDEV_IO_NUM_CHILD_IOV - 1 */
1361 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1362 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV - 1);
1363 	expected_io->md_buf = md_buf;
1364 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1365 		ut_expected_io_set_iov(expected_io, i,
1366 				       (void *)((i + 1) * 0x10000), 512);
1367 	}
1368 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1369 	/* The second expected io should be start from offset BDEV_IO_NUM_CHILD_IOV - 1 to BDEV_IO_NUM_CHILD_IOV */
1370 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1371 					   1, 2);
1372 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1373 	ut_expected_io_set_iov(expected_io, 0,
1374 			       (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000), 256);
1375 	ut_expected_io_set_iov(expected_io, 1,
1376 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 256);
1377 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1378 	/* The third expected io should be start from offset BDEV_IO_NUM_CHILD_IOV to BDEV_IO_NUM_CHILD_IOV + 1 */
1379 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1380 					   1, 1);
1381 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1382 	ut_expected_io_set_iov(expected_io, 0,
1383 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1384 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1385 
1386 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1387 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1388 	CU_ASSERT(rc == 0);
1389 	CU_ASSERT(g_io_done == false);
1390 
1391 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1392 	stub_complete_io(1);
1393 	CU_ASSERT(g_io_done == false);
1394 
1395 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1396 	stub_complete_io(2);
1397 	CU_ASSERT(g_io_done == true);
1398 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1399 
1400 	/* Test multi vector command that needs to be split due to the IO boundary and
1401 	 * the capacity of child iovs. Especially test the case when the command is
1402 	 * split due to the capacity of child iovs, the tail address is not aligned with
1403 	 * block size and is rewinded to the aligned address.
1404 	 *
1405 	 * The iovecs used in read request is complex but is based on the data
1406 	 * collected in the real issue. We change the base addresses but keep the lengths
1407 	 * not to loose the credibility of the test.
1408 	 */
1409 	bdev->optimal_io_boundary = 128;
1410 	g_io_done = false;
1411 	g_io_status = 0;
1412 
1413 	for (i = 0; i < 31; i++) {
1414 		iov[i].iov_base = (void *)(0xFEED0000000 + (i << 20));
1415 		iov[i].iov_len = 1024;
1416 	}
1417 	iov[31].iov_base = (void *)0xFEED1F00000;
1418 	iov[31].iov_len = 32768;
1419 	iov[32].iov_base = (void *)0xFEED2000000;
1420 	iov[32].iov_len = 160;
1421 	iov[33].iov_base = (void *)0xFEED2100000;
1422 	iov[33].iov_len = 4096;
1423 	iov[34].iov_base = (void *)0xFEED2200000;
1424 	iov[34].iov_len = 4096;
1425 	iov[35].iov_base = (void *)0xFEED2300000;
1426 	iov[35].iov_len = 4096;
1427 	iov[36].iov_base = (void *)0xFEED2400000;
1428 	iov[36].iov_len = 4096;
1429 	iov[37].iov_base = (void *)0xFEED2500000;
1430 	iov[37].iov_len = 4096;
1431 	iov[38].iov_base = (void *)0xFEED2600000;
1432 	iov[38].iov_len = 4096;
1433 	iov[39].iov_base = (void *)0xFEED2700000;
1434 	iov[39].iov_len = 4096;
1435 	iov[40].iov_base = (void *)0xFEED2800000;
1436 	iov[40].iov_len = 4096;
1437 	iov[41].iov_base = (void *)0xFEED2900000;
1438 	iov[41].iov_len = 4096;
1439 	iov[42].iov_base = (void *)0xFEED2A00000;
1440 	iov[42].iov_len = 4096;
1441 	iov[43].iov_base = (void *)0xFEED2B00000;
1442 	iov[43].iov_len = 12288;
1443 	iov[44].iov_base = (void *)0xFEED2C00000;
1444 	iov[44].iov_len = 8192;
1445 	iov[45].iov_base = (void *)0xFEED2F00000;
1446 	iov[45].iov_len = 4096;
1447 	iov[46].iov_base = (void *)0xFEED3000000;
1448 	iov[46].iov_len = 4096;
1449 	iov[47].iov_base = (void *)0xFEED3100000;
1450 	iov[47].iov_len = 4096;
1451 	iov[48].iov_base = (void *)0xFEED3200000;
1452 	iov[48].iov_len = 24576;
1453 	iov[49].iov_base = (void *)0xFEED3300000;
1454 	iov[49].iov_len = 16384;
1455 	iov[50].iov_base = (void *)0xFEED3400000;
1456 	iov[50].iov_len = 12288;
1457 	iov[51].iov_base = (void *)0xFEED3500000;
1458 	iov[51].iov_len = 4096;
1459 	iov[52].iov_base = (void *)0xFEED3600000;
1460 	iov[52].iov_len = 4096;
1461 	iov[53].iov_base = (void *)0xFEED3700000;
1462 	iov[53].iov_len = 4096;
1463 	iov[54].iov_base = (void *)0xFEED3800000;
1464 	iov[54].iov_len = 28672;
1465 	iov[55].iov_base = (void *)0xFEED3900000;
1466 	iov[55].iov_len = 20480;
1467 	iov[56].iov_base = (void *)0xFEED3A00000;
1468 	iov[56].iov_len = 4096;
1469 	iov[57].iov_base = (void *)0xFEED3B00000;
1470 	iov[57].iov_len = 12288;
1471 	iov[58].iov_base = (void *)0xFEED3C00000;
1472 	iov[58].iov_len = 4096;
1473 	iov[59].iov_base = (void *)0xFEED3D00000;
1474 	iov[59].iov_len = 4096;
1475 	iov[60].iov_base = (void *)0xFEED3E00000;
1476 	iov[60].iov_len = 352;
1477 
1478 	/* The 1st child IO must be from iov[0] to iov[31] split by the capacity
1479 	 * of child iovs,
1480 	 */
1481 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 126, 32);
1482 	expected_io->md_buf = md_buf;
1483 	for (i = 0; i < 32; i++) {
1484 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1485 	}
1486 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1487 
1488 	/* The 2nd child IO must be from iov[32] to the first 864 bytes of iov[33]
1489 	 * split by the IO boundary requirement.
1490 	 */
1491 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 126, 2, 2);
1492 	expected_io->md_buf = md_buf + 126 * 8;
1493 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base, iov[32].iov_len);
1494 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 864);
1495 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1496 
1497 	/* The 3rd child IO must be from the remaining 3232 bytes of iov[33] to
1498 	 * the first 864 bytes of iov[46] split by the IO boundary requirement.
1499 	 */
1500 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 128, 128, 14);
1501 	expected_io->md_buf = md_buf + 128 * 8;
1502 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[33].iov_base + 864),
1503 			       iov[33].iov_len - 864);
1504 	ut_expected_io_set_iov(expected_io, 1, iov[34].iov_base, iov[34].iov_len);
1505 	ut_expected_io_set_iov(expected_io, 2, iov[35].iov_base, iov[35].iov_len);
1506 	ut_expected_io_set_iov(expected_io, 3, iov[36].iov_base, iov[36].iov_len);
1507 	ut_expected_io_set_iov(expected_io, 4, iov[37].iov_base, iov[37].iov_len);
1508 	ut_expected_io_set_iov(expected_io, 5, iov[38].iov_base, iov[38].iov_len);
1509 	ut_expected_io_set_iov(expected_io, 6, iov[39].iov_base, iov[39].iov_len);
1510 	ut_expected_io_set_iov(expected_io, 7, iov[40].iov_base, iov[40].iov_len);
1511 	ut_expected_io_set_iov(expected_io, 8, iov[41].iov_base, iov[41].iov_len);
1512 	ut_expected_io_set_iov(expected_io, 9, iov[42].iov_base, iov[42].iov_len);
1513 	ut_expected_io_set_iov(expected_io, 10, iov[43].iov_base, iov[43].iov_len);
1514 	ut_expected_io_set_iov(expected_io, 11, iov[44].iov_base, iov[44].iov_len);
1515 	ut_expected_io_set_iov(expected_io, 12, iov[45].iov_base, iov[45].iov_len);
1516 	ut_expected_io_set_iov(expected_io, 13, iov[46].iov_base, 864);
1517 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1518 
1519 	/* The 4th child IO must be from the remaining 3232 bytes of iov[46] to the
1520 	 * first 864 bytes of iov[52] split by the IO boundary requirement.
1521 	 */
1522 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 256, 128, 7);
1523 	expected_io->md_buf = md_buf + 256 * 8;
1524 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[46].iov_base + 864),
1525 			       iov[46].iov_len - 864);
1526 	ut_expected_io_set_iov(expected_io, 1, iov[47].iov_base, iov[47].iov_len);
1527 	ut_expected_io_set_iov(expected_io, 2, iov[48].iov_base, iov[48].iov_len);
1528 	ut_expected_io_set_iov(expected_io, 3, iov[49].iov_base, iov[49].iov_len);
1529 	ut_expected_io_set_iov(expected_io, 4, iov[50].iov_base, iov[50].iov_len);
1530 	ut_expected_io_set_iov(expected_io, 5, iov[51].iov_base, iov[51].iov_len);
1531 	ut_expected_io_set_iov(expected_io, 6, iov[52].iov_base, 864);
1532 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1533 
1534 	/* The 5th child IO must be from the remaining 3232 bytes of iov[52] to
1535 	 * the first 4096 bytes of iov[57] split by the IO boundary requirement.
1536 	 */
1537 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 384, 128, 6);
1538 	expected_io->md_buf = md_buf + 384 * 8;
1539 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[52].iov_base + 864),
1540 			       iov[52].iov_len - 864);
1541 	ut_expected_io_set_iov(expected_io, 1, iov[53].iov_base, iov[53].iov_len);
1542 	ut_expected_io_set_iov(expected_io, 2, iov[54].iov_base, iov[54].iov_len);
1543 	ut_expected_io_set_iov(expected_io, 3, iov[55].iov_base, iov[55].iov_len);
1544 	ut_expected_io_set_iov(expected_io, 4, iov[56].iov_base, iov[56].iov_len);
1545 	ut_expected_io_set_iov(expected_io, 5, iov[57].iov_base, 4960);
1546 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1547 
1548 	/* The 6th child IO must be from the remaining 7328 bytes of iov[57]
1549 	 * to the first 3936 bytes of iov[58] split by the capacity of child iovs.
1550 	 */
1551 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 512, 30, 3);
1552 	expected_io->md_buf = md_buf + 512 * 8;
1553 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[57].iov_base + 4960),
1554 			       iov[57].iov_len - 4960);
1555 	ut_expected_io_set_iov(expected_io, 1, iov[58].iov_base, iov[58].iov_len);
1556 	ut_expected_io_set_iov(expected_io, 2, iov[59].iov_base, 3936);
1557 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1558 
1559 	/* The 7th child IO is from the remaining 160 bytes of iov[59] and iov[60]. */
1560 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 542, 1, 2);
1561 	expected_io->md_buf = md_buf + 542 * 8;
1562 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[59].iov_base + 3936),
1563 			       iov[59].iov_len - 3936);
1564 	ut_expected_io_set_iov(expected_io, 1, iov[60].iov_base, iov[60].iov_len);
1565 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1566 
1567 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, 61, md_buf,
1568 					    0, 543, io_done, NULL);
1569 	CU_ASSERT(rc == 0);
1570 	CU_ASSERT(g_io_done == false);
1571 
1572 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1573 	stub_complete_io(1);
1574 	CU_ASSERT(g_io_done == false);
1575 
1576 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1577 	stub_complete_io(5);
1578 	CU_ASSERT(g_io_done == false);
1579 
1580 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1581 	stub_complete_io(1);
1582 	CU_ASSERT(g_io_done == true);
1583 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1584 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1585 
1586 	/* Test a WRITE_ZEROES that would span an I/O boundary.  WRITE_ZEROES should not be
1587 	 * split, so test that.
1588 	 */
1589 	bdev->optimal_io_boundary = 15;
1590 	g_io_done = false;
1591 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
1592 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1593 
1594 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
1595 	CU_ASSERT(rc == 0);
1596 	CU_ASSERT(g_io_done == false);
1597 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1598 	stub_complete_io(1);
1599 	CU_ASSERT(g_io_done == true);
1600 
1601 	/* Test an UNMAP.  This should also not be split. */
1602 	bdev->optimal_io_boundary = 16;
1603 	g_io_done = false;
1604 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 2, 0);
1605 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1606 
1607 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 2, io_done, NULL);
1608 	CU_ASSERT(rc == 0);
1609 	CU_ASSERT(g_io_done == false);
1610 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1611 	stub_complete_io(1);
1612 	CU_ASSERT(g_io_done == true);
1613 
1614 	/* Test a FLUSH.  This should also not be split. */
1615 	bdev->optimal_io_boundary = 16;
1616 	g_io_done = false;
1617 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 2, 0);
1618 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1619 
1620 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
1621 	CU_ASSERT(rc == 0);
1622 	CU_ASSERT(g_io_done == false);
1623 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1624 	stub_complete_io(1);
1625 	CU_ASSERT(g_io_done == true);
1626 
1627 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
1628 
1629 	/* Children requests return an error status */
1630 	bdev->optimal_io_boundary = 16;
1631 	iov[0].iov_base = (void *)0x10000;
1632 	iov[0].iov_len = 512 * 64;
1633 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1634 	g_io_done = false;
1635 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1636 
1637 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 1, 1, 64, io_done, NULL);
1638 	CU_ASSERT(rc == 0);
1639 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1640 	stub_complete_io(4);
1641 	CU_ASSERT(g_io_done == false);
1642 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1643 	stub_complete_io(1);
1644 	CU_ASSERT(g_io_done == true);
1645 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1646 
1647 	/* Test if a multi vector command terminated with failure before continueing
1648 	 * splitting process when one of child I/O failed.
1649 	 * The multi vector command is as same as the above that needs to be split by strip
1650 	 * and then needs to be split further due to the capacity of child iovs.
1651 	 */
1652 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1653 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1654 		iov[i].iov_len = 512;
1655 	}
1656 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1657 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1658 
1659 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1660 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1661 
1662 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1663 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1664 
1665 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1666 
1667 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1668 	g_io_done = false;
1669 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1670 
1671 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
1672 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1673 	CU_ASSERT(rc == 0);
1674 	CU_ASSERT(g_io_done == false);
1675 
1676 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1677 	stub_complete_io(1);
1678 	CU_ASSERT(g_io_done == true);
1679 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1680 
1681 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1682 
1683 	/* for this test we will create the following conditions to hit the code path where
1684 	 * we are trying to send and IO following a split that has no iovs because we had to
1685 	 * trim them for alignment reasons.
1686 	 *
1687 	 * - 16K boundary, our IO will start at offset 0 with a length of 0x4200
1688 	 * - Our IOVs are 0x212 in size so that we run into the 16K boundary at child IOV
1689 	 *   position 30 and overshoot by 0x2e.
1690 	 * - That means we'll send the IO and loop back to pick up the remaining bytes at
1691 	 *   child IOV index 31. When we do, we find that we have to shorten index 31 by 0x2e
1692 	 *   which eliniates that vector so we just send the first split IO with 30 vectors
1693 	 *   and let the completion pick up the last 2 vectors.
1694 	 */
1695 	bdev->optimal_io_boundary = 32;
1696 	bdev->split_on_optimal_io_boundary = true;
1697 	g_io_done = false;
1698 
1699 	/* Init all parent IOVs to 0x212 */
1700 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
1701 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1702 		iov[i].iov_len = 0x212;
1703 	}
1704 
1705 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1706 					   BDEV_IO_NUM_CHILD_IOV - 1);
1707 	/* expect 0-29 to be 1:1 with the parent iov */
1708 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1709 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1710 	}
1711 
1712 	/* expect index 30 to be shortened to 0x1e4 (0x212 - 0x1e) because of the alignment
1713 	 * where 0x1e is the amount we overshot the 16K boundary
1714 	 */
1715 	ut_expected_io_set_iov(expected_io, BDEV_IO_NUM_CHILD_IOV - 2,
1716 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base), 0x1e4);
1717 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1718 
1719 	/* 2nd child IO will have 2 remaining vectors, one to pick up from the one that was
1720 	 * shortened that take it to the next boundary and then a final one to get us to
1721 	 * 0x4200 bytes for the IO.
1722 	 */
1723 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1724 					   BDEV_IO_NUM_CHILD_IOV, 2);
1725 	/* position 30 picked up the remaining bytes to the next boundary */
1726 	ut_expected_io_set_iov(expected_io, 0,
1727 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base + 0x1e4), 0x2e);
1728 
1729 	/* position 31 picked the the rest of the trasnfer to get us to 0x4200 */
1730 	ut_expected_io_set_iov(expected_io, 1,
1731 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base), 0x1d2);
1732 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1733 
1734 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 1, 0,
1735 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1736 	CU_ASSERT(rc == 0);
1737 	CU_ASSERT(g_io_done == false);
1738 
1739 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1740 	stub_complete_io(1);
1741 	CU_ASSERT(g_io_done == false);
1742 
1743 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1744 	stub_complete_io(1);
1745 	CU_ASSERT(g_io_done == true);
1746 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1747 
1748 	spdk_put_io_channel(io_ch);
1749 	spdk_bdev_close(desc);
1750 	free_bdev(bdev);
1751 	spdk_bdev_finish(bdev_fini_cb, NULL);
1752 	poll_threads();
1753 }
1754 
1755 static void
1756 bdev_io_max_size_and_segment_split_test(void)
1757 {
1758 	struct spdk_bdev *bdev;
1759 	struct spdk_bdev_desc *desc = NULL;
1760 	struct spdk_io_channel *io_ch;
1761 	struct spdk_bdev_opts bdev_opts = {};
1762 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1763 	struct ut_expected_io *expected_io;
1764 	uint64_t i;
1765 	int rc;
1766 
1767 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1768 	bdev_opts.bdev_io_pool_size = 512;
1769 	bdev_opts.bdev_io_cache_size = 64;
1770 
1771 	bdev_opts.opts_size = sizeof(bdev_opts);
1772 	rc = spdk_bdev_set_opts(&bdev_opts);
1773 	CU_ASSERT(rc == 0);
1774 	spdk_bdev_initialize(bdev_init_cb, NULL);
1775 
1776 	bdev = allocate_bdev("bdev0");
1777 
1778 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
1779 	CU_ASSERT(rc == 0);
1780 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1781 	io_ch = spdk_bdev_get_io_channel(desc);
1782 	CU_ASSERT(io_ch != NULL);
1783 
1784 	bdev->split_on_optimal_io_boundary = false;
1785 	bdev->optimal_io_boundary = 0;
1786 
1787 	/* Case 0 max_num_segments == 0.
1788 	 * but segment size 2 * 512 > 512
1789 	 */
1790 	bdev->max_segment_size = 512;
1791 	bdev->max_num_segments = 0;
1792 	g_io_done = false;
1793 
1794 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
1795 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1796 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
1797 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1798 
1799 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1800 	CU_ASSERT(rc == 0);
1801 	CU_ASSERT(g_io_done == false);
1802 
1803 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1804 	stub_complete_io(1);
1805 	CU_ASSERT(g_io_done == true);
1806 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1807 
1808 	/* Case 1 max_segment_size == 0
1809 	 * but iov num 2 > 1.
1810 	 */
1811 	bdev->max_segment_size = 0;
1812 	bdev->max_num_segments = 1;
1813 	g_io_done = false;
1814 
1815 	iov[0].iov_base = (void *)0x10000;
1816 	iov[0].iov_len = 512;
1817 	iov[1].iov_base = (void *)0x20000;
1818 	iov[1].iov_len = 8 * 512;
1819 
1820 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1821 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, iov[0].iov_len);
1822 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1823 
1824 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 8, 1);
1825 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, iov[1].iov_len);
1826 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1827 
1828 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 14, 9, io_done, NULL);
1829 	CU_ASSERT(rc == 0);
1830 	CU_ASSERT(g_io_done == false);
1831 
1832 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1833 	stub_complete_io(2);
1834 	CU_ASSERT(g_io_done == true);
1835 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1836 
1837 	/* Test that a non-vector command is split correctly.
1838 	 * Set up the expected values before calling spdk_bdev_read_blocks
1839 	 */
1840 	bdev->max_segment_size = 512;
1841 	bdev->max_num_segments = 1;
1842 	g_io_done = false;
1843 
1844 	/* Child IO 0 */
1845 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1846 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1847 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1848 
1849 	/* Child IO 1 */
1850 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
1851 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 1 * 512), 512);
1852 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1853 
1854 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1855 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1856 	CU_ASSERT(rc == 0);
1857 	CU_ASSERT(g_io_done == false);
1858 
1859 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1860 	stub_complete_io(2);
1861 	CU_ASSERT(g_io_done == true);
1862 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1863 
1864 	/* Now set up a more complex, multi-vector command that needs to be split,
1865 	 * including splitting iovecs.
1866 	 */
1867 	bdev->max_segment_size = 2 * 512;
1868 	bdev->max_num_segments = 1;
1869 	g_io_done = false;
1870 
1871 	iov[0].iov_base = (void *)0x10000;
1872 	iov[0].iov_len = 2 * 512;
1873 	iov[1].iov_base = (void *)0x20000;
1874 	iov[1].iov_len = 4 * 512;
1875 	iov[2].iov_base = (void *)0x30000;
1876 	iov[2].iov_len = 6 * 512;
1877 
1878 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
1879 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 2);
1880 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1881 
1882 	/* Split iov[1].size to 2 iov entries then split the segments */
1883 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
1884 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, 512 * 2);
1885 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1886 
1887 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 18, 2, 1);
1888 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base + 512 * 2, 512 * 2);
1889 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1890 
1891 	/* Split iov[2].size to 3 iov entries then split the segments */
1892 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 20, 2, 1);
1893 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base, 512 * 2);
1894 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1895 
1896 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 22, 2, 1);
1897 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 2, 512 * 2);
1898 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1899 
1900 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 24, 2, 1);
1901 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 4, 512 * 2);
1902 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1903 
1904 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 12, io_done, NULL);
1905 	CU_ASSERT(rc == 0);
1906 	CU_ASSERT(g_io_done == false);
1907 
1908 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
1909 	stub_complete_io(6);
1910 	CU_ASSERT(g_io_done == true);
1911 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1912 
1913 	/* Test multi vector command that needs to be split by strip and then needs to be
1914 	 * split further due to the capacity of parent IO child iovs.
1915 	 */
1916 	bdev->max_segment_size = 512;
1917 	bdev->max_num_segments = 1;
1918 	g_io_done = false;
1919 
1920 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1921 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1922 		iov[i].iov_len = 512 * 2;
1923 	}
1924 
1925 	/* Each input iov.size is split into 2 iovs,
1926 	 * half of the input iov can fill all child iov entries of a single IO.
1927 	 */
1928 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV / 2; i++) {
1929 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i, 1, 1);
1930 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1931 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1932 
1933 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i + 1, 1, 1);
1934 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1935 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1936 	}
1937 
1938 	/* The remaining iov is split in the second round */
1939 	for (i = BDEV_IO_NUM_CHILD_IOV / 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1940 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2, 1, 1);
1941 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1942 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1943 
1944 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2 + 1, 1, 1);
1945 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1946 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1947 	}
1948 
1949 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
1950 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1951 	CU_ASSERT(rc == 0);
1952 	CU_ASSERT(g_io_done == false);
1953 
1954 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1955 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1956 	CU_ASSERT(g_io_done == false);
1957 
1958 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1959 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1960 	CU_ASSERT(g_io_done == true);
1961 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1962 
1963 	/* A wrong case, a child IO that is divided does
1964 	 * not meet the principle of multiples of block size,
1965 	 * and exits with error
1966 	 */
1967 	bdev->max_segment_size = 512;
1968 	bdev->max_num_segments = 1;
1969 	g_io_done = false;
1970 
1971 	iov[0].iov_base = (void *)0x10000;
1972 	iov[0].iov_len = 512 + 256;
1973 	iov[1].iov_base = (void *)0x20000;
1974 	iov[1].iov_len = 256;
1975 
1976 	/* iov[0] is split to 512 and 256.
1977 	 * 256 is less than a block size, and it is found
1978 	 * in the next round of split that it is the first child IO smaller than
1979 	 * the block size, so the error exit
1980 	 */
1981 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 1, 1);
1982 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512);
1983 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1984 
1985 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 0, 2, io_done, NULL);
1986 	CU_ASSERT(rc == 0);
1987 	CU_ASSERT(g_io_done == false);
1988 
1989 	/* First child IO is OK */
1990 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1991 	stub_complete_io(1);
1992 	CU_ASSERT(g_io_done == true);
1993 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1994 
1995 	/* error exit */
1996 	stub_complete_io(1);
1997 	CU_ASSERT(g_io_done == true);
1998 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1999 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2000 
2001 	/* Test multi vector command that needs to be split by strip and then needs to be
2002 	 * split further due to the capacity of child iovs.
2003 	 *
2004 	 * In this case, the last two iovs need to be split, but it will exceed the capacity
2005 	 * of child iovs, so it needs to wait until the first batch completed.
2006 	 */
2007 	bdev->max_segment_size = 512;
2008 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2009 	g_io_done = false;
2010 
2011 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2012 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2013 		iov[i].iov_len = 512;
2014 	}
2015 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
2016 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2017 		iov[i].iov_len = 512 * 2;
2018 	}
2019 
2020 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2021 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
2022 	/* 0 ~ (BDEV_IO_NUM_CHILD_IOV - 2) Will not be split */
2023 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2024 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2025 	}
2026 	/* (BDEV_IO_NUM_CHILD_IOV - 2) is split */
2027 	ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, 512);
2028 	ut_expected_io_set_iov(expected_io, i + 1, iov[i].iov_base + 512, 512);
2029 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2030 
2031 	/* Child iov entries exceed the max num of parent IO so split it in next round */
2032 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV, 2, 2);
2033 	ut_expected_io_set_iov(expected_io, 0, iov[i + 1].iov_base, 512);
2034 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base + 512, 512);
2035 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2036 
2037 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
2038 				    BDEV_IO_NUM_CHILD_IOV + 2, io_done, NULL);
2039 	CU_ASSERT(rc == 0);
2040 	CU_ASSERT(g_io_done == false);
2041 
2042 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2043 	stub_complete_io(1);
2044 	CU_ASSERT(g_io_done == false);
2045 
2046 	/* Next round */
2047 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2048 	stub_complete_io(1);
2049 	CU_ASSERT(g_io_done == true);
2050 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2051 
2052 	/* This case is similar to the previous one, but the io composed of
2053 	 * the last few entries of child iov is not enough for a blocklen, so they
2054 	 * cannot be put into this IO, but wait until the next time.
2055 	 */
2056 	bdev->max_segment_size = 512;
2057 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2058 	g_io_done = false;
2059 
2060 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2061 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2062 		iov[i].iov_len = 512;
2063 	}
2064 
2065 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2066 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2067 		iov[i].iov_len = 128;
2068 	}
2069 
2070 	/* First child iovcnt is't BDEV_IO_NUM_CHILD_IOV but BDEV_IO_NUM_CHILD_IOV - 2.
2071 	 * Because the left 2 iov is not enough for a blocklen.
2072 	 */
2073 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2074 					   BDEV_IO_NUM_CHILD_IOV - 2, BDEV_IO_NUM_CHILD_IOV - 2);
2075 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2076 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2077 	}
2078 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2079 
2080 	/* The second child io waits until the end of the first child io before executing.
2081 	 * Because the iovcnt of the two IOs exceeds the child iovcnt of the parent IO.
2082 	 * BDEV_IO_NUM_CHILD_IOV - 2 to BDEV_IO_NUM_CHILD_IOV + 2
2083 	 */
2084 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 2,
2085 					   1, 4);
2086 	ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, iov[i].iov_len);
2087 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base, iov[i + 1].iov_len);
2088 	ut_expected_io_set_iov(expected_io, 2, iov[i + 2].iov_base, iov[i + 2].iov_len);
2089 	ut_expected_io_set_iov(expected_io, 3, iov[i + 3].iov_base, iov[i + 3].iov_len);
2090 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2091 
2092 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2093 				    BDEV_IO_NUM_CHILD_IOV - 1, io_done, NULL);
2094 	CU_ASSERT(rc == 0);
2095 	CU_ASSERT(g_io_done == false);
2096 
2097 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2098 	stub_complete_io(1);
2099 	CU_ASSERT(g_io_done == false);
2100 
2101 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2102 	stub_complete_io(1);
2103 	CU_ASSERT(g_io_done == true);
2104 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2105 
2106 	/* A very complicated case. Each sg entry exceeds max_segment_size and
2107 	 * needs to be split. At the same time, child io must be a multiple of blocklen.
2108 	 * At the same time, child iovcnt exceeds parent iovcnt.
2109 	 */
2110 	bdev->max_segment_size = 512 + 128;
2111 	bdev->max_num_segments = 3;
2112 	g_io_done = false;
2113 
2114 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2115 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2116 		iov[i].iov_len = 512 + 256;
2117 	}
2118 
2119 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2120 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2121 		iov[i].iov_len = 512 + 128;
2122 	}
2123 
2124 	/* Child IOs use 9 entries per for() round and 3 * 9 = 27 child iov entries.
2125 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2126 	 * Generate 9 child IOs.
2127 	 */
2128 	for (i = 0; i < 3; i++) {
2129 		uint32_t j = i * 4;
2130 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6, 2, 3);
2131 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2132 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2133 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2134 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2135 
2136 		/* Child io must be a multiple of blocklen
2137 		 * iov[j + 2] must be split. If the third entry is also added,
2138 		 * the multiple of blocklen cannot be guaranteed. But it still
2139 		 * occupies one iov entry of the parent child iov.
2140 		 */
2141 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 2, 2, 2);
2142 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2143 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2144 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2145 
2146 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 4, 2, 3);
2147 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2148 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2149 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2150 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2151 	}
2152 
2153 	/* Child iov position at 27, the 10th child IO
2154 	 * iov entry index is 3 * 4 and offset is 3 * 6
2155 	 */
2156 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 18, 2, 3);
2157 	ut_expected_io_set_iov(expected_io, 0, iov[12].iov_base, 640);
2158 	ut_expected_io_set_iov(expected_io, 1, iov[12].iov_base + 640, 128);
2159 	ut_expected_io_set_iov(expected_io, 2, iov[13].iov_base, 256);
2160 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2161 
2162 	/* Child iov position at 30, the 11th child IO */
2163 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 20, 2, 2);
2164 	ut_expected_io_set_iov(expected_io, 0, iov[13].iov_base + 256, 512);
2165 	ut_expected_io_set_iov(expected_io, 1, iov[14].iov_base, 512);
2166 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2167 
2168 	/* The 2nd split round and iovpos is 0, the 12th child IO */
2169 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 22, 2, 3);
2170 	ut_expected_io_set_iov(expected_io, 0, iov[14].iov_base + 512, 256);
2171 	ut_expected_io_set_iov(expected_io, 1, iov[15].iov_base, 640);
2172 	ut_expected_io_set_iov(expected_io, 2, iov[15].iov_base + 640, 128);
2173 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2174 
2175 	/* Consume 9 child IOs and 27 child iov entries.
2176 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2177 	 * Parent IO iov index start from 16 and block offset start from 24
2178 	 */
2179 	for (i = 0; i < 3; i++) {
2180 		uint32_t j = i * 4 + 16;
2181 		uint32_t offset = i * 6 + 24;
2182 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, 2, 3);
2183 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2184 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2185 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2186 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2187 
2188 		/* Child io must be a multiple of blocklen
2189 		 * iov[j + 2] must be split. If the third entry is also added,
2190 		 * the multiple of blocklen cannot be guaranteed. But it still
2191 		 * occupies one iov entry of the parent child iov.
2192 		 */
2193 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 2, 2, 2);
2194 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2195 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2196 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2197 
2198 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 4, 2, 3);
2199 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2200 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2201 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2202 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2203 	}
2204 
2205 	/* The 22th child IO, child iov position at 30 */
2206 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 42, 1, 1);
2207 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base, 512);
2208 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2209 
2210 	/* The third round */
2211 	/* Here is the 23nd child IO and child iovpos is 0 */
2212 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 43, 2, 3);
2213 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base + 512, 256);
2214 	ut_expected_io_set_iov(expected_io, 1, iov[29].iov_base, 640);
2215 	ut_expected_io_set_iov(expected_io, 2, iov[29].iov_base + 640, 128);
2216 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2217 
2218 	/* The 24th child IO */
2219 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 45, 3, 3);
2220 	ut_expected_io_set_iov(expected_io, 0, iov[30].iov_base, 640);
2221 	ut_expected_io_set_iov(expected_io, 1, iov[31].iov_base, 640);
2222 	ut_expected_io_set_iov(expected_io, 2, iov[32].iov_base, 256);
2223 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2224 
2225 	/* The 25th child IO */
2226 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 48, 2, 2);
2227 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base + 256, 384);
2228 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 640);
2229 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2230 
2231 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2232 				    50, io_done, NULL);
2233 	CU_ASSERT(rc == 0);
2234 	CU_ASSERT(g_io_done == false);
2235 
2236 	/* Parent IO supports up to 32 child iovs, so it is calculated that
2237 	 * a maximum of 11 IOs can be split at a time, and the
2238 	 * splitting will continue after the first batch is over.
2239 	 */
2240 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2241 	stub_complete_io(11);
2242 	CU_ASSERT(g_io_done == false);
2243 
2244 	/* The 2nd round */
2245 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2246 	stub_complete_io(11);
2247 	CU_ASSERT(g_io_done == false);
2248 
2249 	/* The last round */
2250 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2251 	stub_complete_io(3);
2252 	CU_ASSERT(g_io_done == true);
2253 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2254 
2255 	/* Test an WRITE_ZEROES.  This should also not be split. */
2256 	bdev->max_segment_size = 512;
2257 	bdev->max_num_segments = 1;
2258 	g_io_done = false;
2259 
2260 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
2261 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2262 
2263 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
2264 	CU_ASSERT(rc == 0);
2265 	CU_ASSERT(g_io_done == false);
2266 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2267 	stub_complete_io(1);
2268 	CU_ASSERT(g_io_done == true);
2269 
2270 	/* Test an UNMAP.  This should also not be split. */
2271 	g_io_done = false;
2272 
2273 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 4, 0);
2274 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2275 
2276 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 4, io_done, NULL);
2277 	CU_ASSERT(rc == 0);
2278 	CU_ASSERT(g_io_done == false);
2279 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2280 	stub_complete_io(1);
2281 	CU_ASSERT(g_io_done == true);
2282 
2283 	/* Test a FLUSH.  This should also not be split. */
2284 	g_io_done = false;
2285 
2286 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 4, 0);
2287 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2288 
2289 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
2290 	CU_ASSERT(rc == 0);
2291 	CU_ASSERT(g_io_done == false);
2292 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2293 	stub_complete_io(1);
2294 	CU_ASSERT(g_io_done == true);
2295 
2296 	spdk_put_io_channel(io_ch);
2297 	spdk_bdev_close(desc);
2298 	free_bdev(bdev);
2299 	spdk_bdev_finish(bdev_fini_cb, NULL);
2300 	poll_threads();
2301 }
2302 
2303 static void
2304 bdev_io_mix_split_test(void)
2305 {
2306 	struct spdk_bdev *bdev;
2307 	struct spdk_bdev_desc *desc = NULL;
2308 	struct spdk_io_channel *io_ch;
2309 	struct spdk_bdev_opts bdev_opts = {};
2310 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
2311 	struct ut_expected_io *expected_io;
2312 	uint64_t i;
2313 	int rc;
2314 
2315 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2316 	bdev_opts.bdev_io_pool_size = 512;
2317 	bdev_opts.bdev_io_cache_size = 64;
2318 
2319 	rc = spdk_bdev_set_opts(&bdev_opts);
2320 	CU_ASSERT(rc == 0);
2321 	spdk_bdev_initialize(bdev_init_cb, NULL);
2322 
2323 	bdev = allocate_bdev("bdev0");
2324 
2325 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
2326 	CU_ASSERT(rc == 0);
2327 	SPDK_CU_ASSERT_FATAL(desc != NULL);
2328 	io_ch = spdk_bdev_get_io_channel(desc);
2329 	CU_ASSERT(io_ch != NULL);
2330 
2331 	/* First case optimal_io_boundary == max_segment_size * max_num_segments */
2332 	bdev->split_on_optimal_io_boundary = true;
2333 	bdev->optimal_io_boundary = 16;
2334 
2335 	bdev->max_segment_size = 512;
2336 	bdev->max_num_segments = 16;
2337 	g_io_done = false;
2338 
2339 	/* IO crossing the IO boundary requires split
2340 	 * Total 2 child IOs.
2341 	 */
2342 
2343 	/* The 1st child IO split the segment_size to multiple segment entry */
2344 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
2345 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
2346 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
2347 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2348 
2349 	/* The 2nd child IO split the segment_size to multiple segment entry */
2350 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 2);
2351 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 512);
2352 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 3 * 512), 512);
2353 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2354 
2355 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 4, io_done, NULL);
2356 	CU_ASSERT(rc == 0);
2357 	CU_ASSERT(g_io_done == false);
2358 
2359 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2360 	stub_complete_io(2);
2361 	CU_ASSERT(g_io_done == true);
2362 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2363 
2364 	/* Second case optimal_io_boundary > max_segment_size * max_num_segments */
2365 	bdev->max_segment_size = 15 * 512;
2366 	bdev->max_num_segments = 1;
2367 	g_io_done = false;
2368 
2369 	/* IO crossing the IO boundary requires split.
2370 	 * The 1st child IO segment size exceeds the max_segment_size,
2371 	 * So 1st child IO will be splitted to multiple segment entry.
2372 	 * Then it split to 2 child IOs because of the max_num_segments.
2373 	 * Total 3 child IOs.
2374 	 */
2375 
2376 	/* The first 2 IOs are in an IO boundary.
2377 	 * Because the optimal_io_boundary > max_segment_size * max_num_segments
2378 	 * So it split to the first 2 IOs.
2379 	 */
2380 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 15, 1);
2381 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 15);
2382 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2383 
2384 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
2385 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 15), 512);
2386 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2387 
2388 	/* The 3rd Child IO is because of the io boundary */
2389 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2390 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2391 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2392 
2393 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2394 	CU_ASSERT(rc == 0);
2395 	CU_ASSERT(g_io_done == false);
2396 
2397 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2398 	stub_complete_io(3);
2399 	CU_ASSERT(g_io_done == true);
2400 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2401 
2402 	/* Third case optimal_io_boundary < max_segment_size * max_num_segments */
2403 	bdev->max_segment_size = 17 * 512;
2404 	bdev->max_num_segments = 1;
2405 	g_io_done = false;
2406 
2407 	/* IO crossing the IO boundary requires split.
2408 	 * Child IO does not split.
2409 	 * Total 2 child IOs.
2410 	 */
2411 
2412 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 16, 1);
2413 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 16);
2414 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2415 
2416 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2417 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2418 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2419 
2420 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2421 	CU_ASSERT(rc == 0);
2422 	CU_ASSERT(g_io_done == false);
2423 
2424 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2425 	stub_complete_io(2);
2426 	CU_ASSERT(g_io_done == true);
2427 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2428 
2429 	/* Now set up a more complex, multi-vector command that needs to be split,
2430 	 * including splitting iovecs.
2431 	 * optimal_io_boundary < max_segment_size * max_num_segments
2432 	 */
2433 	bdev->max_segment_size = 3 * 512;
2434 	bdev->max_num_segments = 6;
2435 	g_io_done = false;
2436 
2437 	iov[0].iov_base = (void *)0x10000;
2438 	iov[0].iov_len = 4 * 512;
2439 	iov[1].iov_base = (void *)0x20000;
2440 	iov[1].iov_len = 4 * 512;
2441 	iov[2].iov_base = (void *)0x30000;
2442 	iov[2].iov_len = 10 * 512;
2443 
2444 	/* IO crossing the IO boundary requires split.
2445 	 * The 1st child IO segment size exceeds the max_segment_size and after
2446 	 * splitting segment_size, the num_segments exceeds max_num_segments.
2447 	 * So 1st child IO will be splitted to 2 child IOs.
2448 	 * Total 3 child IOs.
2449 	 */
2450 
2451 	/* The first 2 IOs are in an IO boundary.
2452 	 * After splitting segmemt size the segment num exceeds.
2453 	 * So it splits to 2 child IOs.
2454 	 */
2455 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 14, 6);
2456 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 3);
2457 	ut_expected_io_set_iov(expected_io, 1, iov[0].iov_base + 512 * 3, 512);
2458 	ut_expected_io_set_iov(expected_io, 2, iov[1].iov_base, 512 * 3);
2459 	ut_expected_io_set_iov(expected_io, 3, iov[1].iov_base + 512 * 3, 512);
2460 	ut_expected_io_set_iov(expected_io, 4, iov[2].iov_base, 512 * 3);
2461 	ut_expected_io_set_iov(expected_io, 5, iov[2].iov_base + 512 * 3, 512 * 3);
2462 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2463 
2464 	/* The 2nd child IO has the left segment entry */
2465 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
2466 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 6, 512 * 2);
2467 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2468 
2469 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
2470 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 8, 512 * 2);
2471 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2472 
2473 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 0, 18, io_done, NULL);
2474 	CU_ASSERT(rc == 0);
2475 	CU_ASSERT(g_io_done == false);
2476 
2477 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2478 	stub_complete_io(3);
2479 	CU_ASSERT(g_io_done == true);
2480 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2481 
2482 	/* A very complicated case. Each sg entry exceeds max_segment_size
2483 	 * and split on io boundary.
2484 	 * optimal_io_boundary < max_segment_size * max_num_segments
2485 	 */
2486 	bdev->max_segment_size = 3 * 512;
2487 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2488 	g_io_done = false;
2489 
2490 	for (i = 0; i < 20; i++) {
2491 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2492 		iov[i].iov_len = 512 * 4;
2493 	}
2494 
2495 	/* IO crossing the IO boundary requires split.
2496 	 * 80 block length can split 5 child IOs base on offset and IO boundary.
2497 	 * Each iov entry needs to be splitted to 2 entries because of max_segment_size
2498 	 * Total 5 child IOs.
2499 	 */
2500 
2501 	/* 4 iov entries are in an IO boundary and each iov entry splits to 2.
2502 	 * So each child IO occupies 8 child iov entries.
2503 	 */
2504 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 16, 8);
2505 	for (i = 0; i < 4; i++) {
2506 		int iovcnt = i * 2;
2507 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2508 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2509 	}
2510 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2511 
2512 	/* 2nd child IO and total 16 child iov entries of parent IO */
2513 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 8);
2514 	for (i = 4; i < 8; i++) {
2515 		int iovcnt = (i - 4) * 2;
2516 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2517 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2518 	}
2519 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2520 
2521 	/* 3rd child IO and total 24 child iov entries of parent IO */
2522 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 16, 8);
2523 	for (i = 8; i < 12; i++) {
2524 		int iovcnt = (i - 8) * 2;
2525 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2526 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2527 	}
2528 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2529 
2530 	/* 4th child IO and total 32 child iov entries of parent IO */
2531 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 48, 16, 8);
2532 	for (i = 12; i < 16; i++) {
2533 		int iovcnt = (i - 12) * 2;
2534 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2535 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2536 	}
2537 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2538 
2539 	/* 5th child IO and because of the child iov entry it should be splitted
2540 	 * in next round.
2541 	 */
2542 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 64, 16, 8);
2543 	for (i = 16; i < 20; i++) {
2544 		int iovcnt = (i - 16) * 2;
2545 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2546 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2547 	}
2548 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2549 
2550 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 20, 0, 80, io_done, NULL);
2551 	CU_ASSERT(rc == 0);
2552 	CU_ASSERT(g_io_done == false);
2553 
2554 	/* First split round */
2555 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
2556 	stub_complete_io(4);
2557 	CU_ASSERT(g_io_done == false);
2558 
2559 	/* Second split round */
2560 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2561 	stub_complete_io(1);
2562 	CU_ASSERT(g_io_done == true);
2563 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2564 
2565 	spdk_put_io_channel(io_ch);
2566 	spdk_bdev_close(desc);
2567 	free_bdev(bdev);
2568 	spdk_bdev_finish(bdev_fini_cb, NULL);
2569 	poll_threads();
2570 }
2571 
2572 static void
2573 bdev_io_split_with_io_wait(void)
2574 {
2575 	struct spdk_bdev *bdev;
2576 	struct spdk_bdev_desc *desc = NULL;
2577 	struct spdk_io_channel *io_ch;
2578 	struct spdk_bdev_channel *channel;
2579 	struct spdk_bdev_mgmt_channel *mgmt_ch;
2580 	struct spdk_bdev_opts bdev_opts = {};
2581 	struct iovec iov[3];
2582 	struct ut_expected_io *expected_io;
2583 	int rc;
2584 
2585 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2586 	bdev_opts.bdev_io_pool_size = 2;
2587 	bdev_opts.bdev_io_cache_size = 1;
2588 
2589 	rc = spdk_bdev_set_opts(&bdev_opts);
2590 	CU_ASSERT(rc == 0);
2591 	spdk_bdev_initialize(bdev_init_cb, NULL);
2592 
2593 	bdev = allocate_bdev("bdev0");
2594 
2595 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2596 	CU_ASSERT(rc == 0);
2597 	CU_ASSERT(desc != NULL);
2598 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2599 	io_ch = spdk_bdev_get_io_channel(desc);
2600 	CU_ASSERT(io_ch != NULL);
2601 	channel = spdk_io_channel_get_ctx(io_ch);
2602 	mgmt_ch = channel->shared_resource->mgmt_ch;
2603 
2604 	bdev->optimal_io_boundary = 16;
2605 	bdev->split_on_optimal_io_boundary = true;
2606 
2607 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
2608 	CU_ASSERT(rc == 0);
2609 
2610 	/* Now test that a single-vector command is split correctly.
2611 	 * Offset 14, length 8, payload 0xF000
2612 	 *  Child - Offset 14, length 2, payload 0xF000
2613 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
2614 	 *
2615 	 * Set up the expected values before calling spdk_bdev_read_blocks
2616 	 */
2617 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
2618 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
2619 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2620 
2621 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
2622 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
2623 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2624 
2625 	/* The following children will be submitted sequentially due to the capacity of
2626 	 * spdk_bdev_io.
2627 	 */
2628 
2629 	/* The first child I/O will be queued to wait until an spdk_bdev_io becomes available */
2630 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
2631 	CU_ASSERT(rc == 0);
2632 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2633 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2634 
2635 	/* Completing the first read I/O will submit the first child */
2636 	stub_complete_io(1);
2637 	CU_ASSERT(TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2638 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2639 
2640 	/* Completing the first child will submit the second child */
2641 	stub_complete_io(1);
2642 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2643 
2644 	/* Complete the second child I/O.  This should result in our callback getting
2645 	 * invoked since the parent I/O is now complete.
2646 	 */
2647 	stub_complete_io(1);
2648 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2649 
2650 	/* Now set up a more complex, multi-vector command that needs to be split,
2651 	 *  including splitting iovecs.
2652 	 */
2653 	iov[0].iov_base = (void *)0x10000;
2654 	iov[0].iov_len = 512;
2655 	iov[1].iov_base = (void *)0x20000;
2656 	iov[1].iov_len = 20 * 512;
2657 	iov[2].iov_base = (void *)0x30000;
2658 	iov[2].iov_len = 11 * 512;
2659 
2660 	g_io_done = false;
2661 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
2662 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
2663 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
2664 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2665 
2666 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
2667 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
2668 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2669 
2670 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
2671 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
2672 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
2673 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2674 
2675 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 32, io_done, NULL);
2676 	CU_ASSERT(rc == 0);
2677 	CU_ASSERT(g_io_done == false);
2678 
2679 	/* The following children will be submitted sequentially due to the capacity of
2680 	 * spdk_bdev_io.
2681 	 */
2682 
2683 	/* Completing the first child will submit the second child */
2684 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2685 	stub_complete_io(1);
2686 	CU_ASSERT(g_io_done == false);
2687 
2688 	/* Completing the second child will submit the third child */
2689 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2690 	stub_complete_io(1);
2691 	CU_ASSERT(g_io_done == false);
2692 
2693 	/* Completing the third child will result in our callback getting invoked
2694 	 * since the parent I/O is now complete.
2695 	 */
2696 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2697 	stub_complete_io(1);
2698 	CU_ASSERT(g_io_done == true);
2699 
2700 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
2701 
2702 	spdk_put_io_channel(io_ch);
2703 	spdk_bdev_close(desc);
2704 	free_bdev(bdev);
2705 	spdk_bdev_finish(bdev_fini_cb, NULL);
2706 	poll_threads();
2707 }
2708 
2709 static void
2710 bdev_io_alignment(void)
2711 {
2712 	struct spdk_bdev *bdev;
2713 	struct spdk_bdev_desc *desc = NULL;
2714 	struct spdk_io_channel *io_ch;
2715 	struct spdk_bdev_opts bdev_opts = {};
2716 	int rc;
2717 	void *buf = NULL;
2718 	struct iovec iovs[2];
2719 	int iovcnt;
2720 	uint64_t alignment;
2721 
2722 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2723 	bdev_opts.bdev_io_pool_size = 20;
2724 	bdev_opts.bdev_io_cache_size = 2;
2725 
2726 	rc = spdk_bdev_set_opts(&bdev_opts);
2727 	CU_ASSERT(rc == 0);
2728 	spdk_bdev_initialize(bdev_init_cb, NULL);
2729 
2730 	fn_table.submit_request = stub_submit_request_get_buf;
2731 	bdev = allocate_bdev("bdev0");
2732 
2733 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2734 	CU_ASSERT(rc == 0);
2735 	CU_ASSERT(desc != NULL);
2736 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2737 	io_ch = spdk_bdev_get_io_channel(desc);
2738 	CU_ASSERT(io_ch != NULL);
2739 
2740 	/* Create aligned buffer */
2741 	rc = posix_memalign(&buf, 4096, 8192);
2742 	SPDK_CU_ASSERT_FATAL(rc == 0);
2743 
2744 	/* Pass aligned single buffer with no alignment required */
2745 	alignment = 1;
2746 	bdev->required_alignment = spdk_u32log2(alignment);
2747 
2748 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2749 	CU_ASSERT(rc == 0);
2750 	stub_complete_io(1);
2751 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2752 				    alignment));
2753 
2754 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2755 	CU_ASSERT(rc == 0);
2756 	stub_complete_io(1);
2757 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2758 				    alignment));
2759 
2760 	/* Pass unaligned single buffer with no alignment required */
2761 	alignment = 1;
2762 	bdev->required_alignment = spdk_u32log2(alignment);
2763 
2764 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2765 	CU_ASSERT(rc == 0);
2766 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2767 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2768 	stub_complete_io(1);
2769 
2770 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2771 	CU_ASSERT(rc == 0);
2772 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2773 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2774 	stub_complete_io(1);
2775 
2776 	/* Pass unaligned single buffer with 512 alignment required */
2777 	alignment = 512;
2778 	bdev->required_alignment = spdk_u32log2(alignment);
2779 
2780 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2781 	CU_ASSERT(rc == 0);
2782 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2783 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2784 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2785 				    alignment));
2786 	stub_complete_io(1);
2787 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2788 
2789 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2790 	CU_ASSERT(rc == 0);
2791 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2792 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2793 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2794 				    alignment));
2795 	stub_complete_io(1);
2796 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2797 
2798 	/* Pass unaligned single buffer with 4096 alignment required */
2799 	alignment = 4096;
2800 	bdev->required_alignment = spdk_u32log2(alignment);
2801 
2802 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2803 	CU_ASSERT(rc == 0);
2804 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2805 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2806 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2807 				    alignment));
2808 	stub_complete_io(1);
2809 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2810 
2811 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2812 	CU_ASSERT(rc == 0);
2813 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2814 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2815 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2816 				    alignment));
2817 	stub_complete_io(1);
2818 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2819 
2820 	/* Pass aligned iovs with no alignment required */
2821 	alignment = 1;
2822 	bdev->required_alignment = spdk_u32log2(alignment);
2823 
2824 	iovcnt = 1;
2825 	iovs[0].iov_base = buf;
2826 	iovs[0].iov_len = 512;
2827 
2828 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2829 	CU_ASSERT(rc == 0);
2830 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2831 	stub_complete_io(1);
2832 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2833 
2834 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2835 	CU_ASSERT(rc == 0);
2836 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2837 	stub_complete_io(1);
2838 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2839 
2840 	/* Pass unaligned iovs with no alignment required */
2841 	alignment = 1;
2842 	bdev->required_alignment = spdk_u32log2(alignment);
2843 
2844 	iovcnt = 2;
2845 	iovs[0].iov_base = buf + 16;
2846 	iovs[0].iov_len = 256;
2847 	iovs[1].iov_base = buf + 16 + 256 + 32;
2848 	iovs[1].iov_len = 256;
2849 
2850 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2851 	CU_ASSERT(rc == 0);
2852 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2853 	stub_complete_io(1);
2854 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2855 
2856 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2857 	CU_ASSERT(rc == 0);
2858 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2859 	stub_complete_io(1);
2860 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2861 
2862 	/* Pass unaligned iov with 2048 alignment required */
2863 	alignment = 2048;
2864 	bdev->required_alignment = spdk_u32log2(alignment);
2865 
2866 	iovcnt = 2;
2867 	iovs[0].iov_base = buf + 16;
2868 	iovs[0].iov_len = 256;
2869 	iovs[1].iov_base = buf + 16 + 256 + 32;
2870 	iovs[1].iov_len = 256;
2871 
2872 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2873 	CU_ASSERT(rc == 0);
2874 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2875 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2876 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2877 				    alignment));
2878 	stub_complete_io(1);
2879 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2880 
2881 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2882 	CU_ASSERT(rc == 0);
2883 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2884 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2885 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2886 				    alignment));
2887 	stub_complete_io(1);
2888 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2889 
2890 	/* Pass iov without allocated buffer without alignment required */
2891 	alignment = 1;
2892 	bdev->required_alignment = spdk_u32log2(alignment);
2893 
2894 	iovcnt = 1;
2895 	iovs[0].iov_base = NULL;
2896 	iovs[0].iov_len = 0;
2897 
2898 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2899 	CU_ASSERT(rc == 0);
2900 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2901 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2902 				    alignment));
2903 	stub_complete_io(1);
2904 
2905 	/* Pass iov without allocated buffer with 1024 alignment required */
2906 	alignment = 1024;
2907 	bdev->required_alignment = spdk_u32log2(alignment);
2908 
2909 	iovcnt = 1;
2910 	iovs[0].iov_base = NULL;
2911 	iovs[0].iov_len = 0;
2912 
2913 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2914 	CU_ASSERT(rc == 0);
2915 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2916 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2917 				    alignment));
2918 	stub_complete_io(1);
2919 
2920 	spdk_put_io_channel(io_ch);
2921 	spdk_bdev_close(desc);
2922 	free_bdev(bdev);
2923 	fn_table.submit_request = stub_submit_request;
2924 	spdk_bdev_finish(bdev_fini_cb, NULL);
2925 	poll_threads();
2926 
2927 	free(buf);
2928 }
2929 
2930 static void
2931 bdev_io_alignment_with_boundary(void)
2932 {
2933 	struct spdk_bdev *bdev;
2934 	struct spdk_bdev_desc *desc = NULL;
2935 	struct spdk_io_channel *io_ch;
2936 	struct spdk_bdev_opts bdev_opts = {};
2937 	int rc;
2938 	void *buf = NULL;
2939 	struct iovec iovs[2];
2940 	int iovcnt;
2941 	uint64_t alignment;
2942 
2943 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2944 	bdev_opts.bdev_io_pool_size = 20;
2945 	bdev_opts.bdev_io_cache_size = 2;
2946 
2947 	bdev_opts.opts_size = sizeof(bdev_opts);
2948 	rc = spdk_bdev_set_opts(&bdev_opts);
2949 	CU_ASSERT(rc == 0);
2950 	spdk_bdev_initialize(bdev_init_cb, NULL);
2951 
2952 	fn_table.submit_request = stub_submit_request_get_buf;
2953 	bdev = allocate_bdev("bdev0");
2954 
2955 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2956 	CU_ASSERT(rc == 0);
2957 	CU_ASSERT(desc != NULL);
2958 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2959 	io_ch = spdk_bdev_get_io_channel(desc);
2960 	CU_ASSERT(io_ch != NULL);
2961 
2962 	/* Create aligned buffer */
2963 	rc = posix_memalign(&buf, 4096, 131072);
2964 	SPDK_CU_ASSERT_FATAL(rc == 0);
2965 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2966 
2967 	/* 512 * 3 with 2 IO boundary, allocate small data buffer from bdev layer */
2968 	alignment = 512;
2969 	bdev->required_alignment = spdk_u32log2(alignment);
2970 	bdev->optimal_io_boundary = 2;
2971 	bdev->split_on_optimal_io_boundary = true;
2972 
2973 	iovcnt = 1;
2974 	iovs[0].iov_base = NULL;
2975 	iovs[0].iov_len = 512 * 3;
2976 
2977 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
2978 	CU_ASSERT(rc == 0);
2979 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2980 	stub_complete_io(2);
2981 
2982 	/* 8KiB with 16 IO boundary, allocate large data buffer from bdev layer */
2983 	alignment = 512;
2984 	bdev->required_alignment = spdk_u32log2(alignment);
2985 	bdev->optimal_io_boundary = 16;
2986 	bdev->split_on_optimal_io_boundary = true;
2987 
2988 	iovcnt = 1;
2989 	iovs[0].iov_base = NULL;
2990 	iovs[0].iov_len = 512 * 16;
2991 
2992 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 16, io_done, NULL);
2993 	CU_ASSERT(rc == 0);
2994 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2995 	stub_complete_io(2);
2996 
2997 	/* 512 * 160 with 128 IO boundary, 63.5KiB + 16.5KiB for the two children requests */
2998 	alignment = 512;
2999 	bdev->required_alignment = spdk_u32log2(alignment);
3000 	bdev->optimal_io_boundary = 128;
3001 	bdev->split_on_optimal_io_boundary = true;
3002 
3003 	iovcnt = 1;
3004 	iovs[0].iov_base = buf + 16;
3005 	iovs[0].iov_len = 512 * 160;
3006 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3007 	CU_ASSERT(rc == 0);
3008 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3009 	stub_complete_io(2);
3010 
3011 	/* 512 * 3 with 2 IO boundary */
3012 	alignment = 512;
3013 	bdev->required_alignment = spdk_u32log2(alignment);
3014 	bdev->optimal_io_boundary = 2;
3015 	bdev->split_on_optimal_io_boundary = true;
3016 
3017 	iovcnt = 2;
3018 	iovs[0].iov_base = buf + 16;
3019 	iovs[0].iov_len = 512;
3020 	iovs[1].iov_base = buf + 16 + 512 + 32;
3021 	iovs[1].iov_len = 1024;
3022 
3023 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3024 	CU_ASSERT(rc == 0);
3025 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3026 	stub_complete_io(2);
3027 
3028 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3029 	CU_ASSERT(rc == 0);
3030 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3031 	stub_complete_io(2);
3032 
3033 	/* 512 * 64 with 32 IO boundary */
3034 	bdev->optimal_io_boundary = 32;
3035 	iovcnt = 2;
3036 	iovs[0].iov_base = buf + 16;
3037 	iovs[0].iov_len = 16384;
3038 	iovs[1].iov_base = buf + 16 + 16384 + 32;
3039 	iovs[1].iov_len = 16384;
3040 
3041 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3042 	CU_ASSERT(rc == 0);
3043 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3044 	stub_complete_io(3);
3045 
3046 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3047 	CU_ASSERT(rc == 0);
3048 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3049 	stub_complete_io(3);
3050 
3051 	/* 512 * 160 with 32 IO boundary */
3052 	iovcnt = 1;
3053 	iovs[0].iov_base = buf + 16;
3054 	iovs[0].iov_len = 16384 + 65536;
3055 
3056 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3057 	CU_ASSERT(rc == 0);
3058 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
3059 	stub_complete_io(6);
3060 
3061 	spdk_put_io_channel(io_ch);
3062 	spdk_bdev_close(desc);
3063 	free_bdev(bdev);
3064 	fn_table.submit_request = stub_submit_request;
3065 	spdk_bdev_finish(bdev_fini_cb, NULL);
3066 	poll_threads();
3067 
3068 	free(buf);
3069 }
3070 
3071 static void
3072 histogram_status_cb(void *cb_arg, int status)
3073 {
3074 	g_status = status;
3075 }
3076 
3077 static void
3078 histogram_data_cb(void *cb_arg, int status, struct spdk_histogram_data *histogram)
3079 {
3080 	g_status = status;
3081 	g_histogram = histogram;
3082 }
3083 
3084 static void
3085 histogram_io_count(void *ctx, uint64_t start, uint64_t end, uint64_t count,
3086 		   uint64_t total, uint64_t so_far)
3087 {
3088 	g_count += count;
3089 }
3090 
3091 static void
3092 bdev_histograms(void)
3093 {
3094 	struct spdk_bdev *bdev;
3095 	struct spdk_bdev_desc *desc = NULL;
3096 	struct spdk_io_channel *ch;
3097 	struct spdk_histogram_data *histogram;
3098 	uint8_t buf[4096];
3099 	int rc;
3100 
3101 	spdk_bdev_initialize(bdev_init_cb, NULL);
3102 
3103 	bdev = allocate_bdev("bdev");
3104 
3105 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3106 	CU_ASSERT(rc == 0);
3107 	CU_ASSERT(desc != NULL);
3108 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3109 
3110 	ch = spdk_bdev_get_io_channel(desc);
3111 	CU_ASSERT(ch != NULL);
3112 
3113 	/* Enable histogram */
3114 	g_status = -1;
3115 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, true);
3116 	poll_threads();
3117 	CU_ASSERT(g_status == 0);
3118 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3119 
3120 	/* Allocate histogram */
3121 	histogram = spdk_histogram_data_alloc();
3122 	SPDK_CU_ASSERT_FATAL(histogram != NULL);
3123 
3124 	/* Check if histogram is zeroed */
3125 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3126 	poll_threads();
3127 	CU_ASSERT(g_status == 0);
3128 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3129 
3130 	g_count = 0;
3131 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3132 
3133 	CU_ASSERT(g_count == 0);
3134 
3135 	rc = spdk_bdev_write_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3136 	CU_ASSERT(rc == 0);
3137 
3138 	spdk_delay_us(10);
3139 	stub_complete_io(1);
3140 	poll_threads();
3141 
3142 	rc = spdk_bdev_read_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3143 	CU_ASSERT(rc == 0);
3144 
3145 	spdk_delay_us(10);
3146 	stub_complete_io(1);
3147 	poll_threads();
3148 
3149 	/* Check if histogram gathered data from all I/O channels */
3150 	g_histogram = NULL;
3151 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3152 	poll_threads();
3153 	CU_ASSERT(g_status == 0);
3154 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3155 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3156 
3157 	g_count = 0;
3158 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3159 	CU_ASSERT(g_count == 2);
3160 
3161 	/* Disable histogram */
3162 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, false);
3163 	poll_threads();
3164 	CU_ASSERT(g_status == 0);
3165 	CU_ASSERT(bdev->internal.histogram_enabled == false);
3166 
3167 	/* Try to run histogram commands on disabled bdev */
3168 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3169 	poll_threads();
3170 	CU_ASSERT(g_status == -EFAULT);
3171 
3172 	spdk_histogram_data_free(histogram);
3173 	spdk_put_io_channel(ch);
3174 	spdk_bdev_close(desc);
3175 	free_bdev(bdev);
3176 	spdk_bdev_finish(bdev_fini_cb, NULL);
3177 	poll_threads();
3178 }
3179 
3180 static void
3181 _bdev_compare(bool emulated)
3182 {
3183 	struct spdk_bdev *bdev;
3184 	struct spdk_bdev_desc *desc = NULL;
3185 	struct spdk_io_channel *ioch;
3186 	struct ut_expected_io *expected_io;
3187 	uint64_t offset, num_blocks;
3188 	uint32_t num_completed;
3189 	char aa_buf[512];
3190 	char bb_buf[512];
3191 	struct iovec compare_iov;
3192 	uint8_t io_type;
3193 	int rc;
3194 
3195 	if (emulated) {
3196 		io_type = SPDK_BDEV_IO_TYPE_READ;
3197 	} else {
3198 		io_type = SPDK_BDEV_IO_TYPE_COMPARE;
3199 	}
3200 
3201 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3202 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3203 
3204 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = !emulated;
3205 
3206 	spdk_bdev_initialize(bdev_init_cb, NULL);
3207 	fn_table.submit_request = stub_submit_request_get_buf;
3208 	bdev = allocate_bdev("bdev");
3209 
3210 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3211 	CU_ASSERT_EQUAL(rc, 0);
3212 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3213 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3214 	ioch = spdk_bdev_get_io_channel(desc);
3215 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3216 
3217 	fn_table.submit_request = stub_submit_request_get_buf;
3218 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3219 
3220 	offset = 50;
3221 	num_blocks = 1;
3222 	compare_iov.iov_base = aa_buf;
3223 	compare_iov.iov_len = sizeof(aa_buf);
3224 
3225 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3226 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3227 
3228 	g_io_done = false;
3229 	g_compare_read_buf = aa_buf;
3230 	g_compare_read_buf_len = sizeof(aa_buf);
3231 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3232 	CU_ASSERT_EQUAL(rc, 0);
3233 	num_completed = stub_complete_io(1);
3234 	CU_ASSERT_EQUAL(num_completed, 1);
3235 	CU_ASSERT(g_io_done == true);
3236 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3237 
3238 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3239 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3240 
3241 	g_io_done = false;
3242 	g_compare_read_buf = bb_buf;
3243 	g_compare_read_buf_len = sizeof(bb_buf);
3244 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3245 	CU_ASSERT_EQUAL(rc, 0);
3246 	num_completed = stub_complete_io(1);
3247 	CU_ASSERT_EQUAL(num_completed, 1);
3248 	CU_ASSERT(g_io_done == true);
3249 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3250 
3251 	spdk_put_io_channel(ioch);
3252 	spdk_bdev_close(desc);
3253 	free_bdev(bdev);
3254 	fn_table.submit_request = stub_submit_request;
3255 	spdk_bdev_finish(bdev_fini_cb, NULL);
3256 	poll_threads();
3257 
3258 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3259 
3260 	g_compare_read_buf = NULL;
3261 }
3262 
3263 static void
3264 bdev_compare(void)
3265 {
3266 	_bdev_compare(true);
3267 	_bdev_compare(false);
3268 }
3269 
3270 static void
3271 bdev_compare_and_write(void)
3272 {
3273 	struct spdk_bdev *bdev;
3274 	struct spdk_bdev_desc *desc = NULL;
3275 	struct spdk_io_channel *ioch;
3276 	struct ut_expected_io *expected_io;
3277 	uint64_t offset, num_blocks;
3278 	uint32_t num_completed;
3279 	char aa_buf[512];
3280 	char bb_buf[512];
3281 	char cc_buf[512];
3282 	char write_buf[512];
3283 	struct iovec compare_iov;
3284 	struct iovec write_iov;
3285 	int rc;
3286 
3287 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3288 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3289 	memset(cc_buf, 0xcc, sizeof(cc_buf));
3290 
3291 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = false;
3292 
3293 	spdk_bdev_initialize(bdev_init_cb, NULL);
3294 	fn_table.submit_request = stub_submit_request_get_buf;
3295 	bdev = allocate_bdev("bdev");
3296 
3297 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3298 	CU_ASSERT_EQUAL(rc, 0);
3299 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3300 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3301 	ioch = spdk_bdev_get_io_channel(desc);
3302 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3303 
3304 	fn_table.submit_request = stub_submit_request_get_buf;
3305 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3306 
3307 	offset = 50;
3308 	num_blocks = 1;
3309 	compare_iov.iov_base = aa_buf;
3310 	compare_iov.iov_len = sizeof(aa_buf);
3311 	write_iov.iov_base = bb_buf;
3312 	write_iov.iov_len = sizeof(bb_buf);
3313 
3314 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3315 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3316 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, offset, num_blocks, 0);
3317 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3318 
3319 	g_io_done = false;
3320 	g_compare_read_buf = aa_buf;
3321 	g_compare_read_buf_len = sizeof(aa_buf);
3322 	memset(write_buf, 0, sizeof(write_buf));
3323 	g_compare_write_buf = write_buf;
3324 	g_compare_write_buf_len = sizeof(write_buf);
3325 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3326 			offset, num_blocks, io_done, NULL);
3327 	/* Trigger range locking */
3328 	poll_threads();
3329 	CU_ASSERT_EQUAL(rc, 0);
3330 	num_completed = stub_complete_io(1);
3331 	CU_ASSERT_EQUAL(num_completed, 1);
3332 	CU_ASSERT(g_io_done == false);
3333 	num_completed = stub_complete_io(1);
3334 	/* Trigger range unlocking */
3335 	poll_threads();
3336 	CU_ASSERT_EQUAL(num_completed, 1);
3337 	CU_ASSERT(g_io_done == true);
3338 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3339 	CU_ASSERT(memcmp(write_buf, bb_buf, sizeof(write_buf)) == 0);
3340 
3341 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3342 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3343 
3344 	g_io_done = false;
3345 	g_compare_read_buf = cc_buf;
3346 	g_compare_read_buf_len = sizeof(cc_buf);
3347 	memset(write_buf, 0, sizeof(write_buf));
3348 	g_compare_write_buf = write_buf;
3349 	g_compare_write_buf_len = sizeof(write_buf);
3350 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3351 			offset, num_blocks, io_done, NULL);
3352 	/* Trigger range locking */
3353 	poll_threads();
3354 	CU_ASSERT_EQUAL(rc, 0);
3355 	num_completed = stub_complete_io(1);
3356 	/* Trigger range unlocking earlier because we expect error here */
3357 	poll_threads();
3358 	CU_ASSERT_EQUAL(num_completed, 1);
3359 	CU_ASSERT(g_io_done == true);
3360 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3361 	num_completed = stub_complete_io(1);
3362 	CU_ASSERT_EQUAL(num_completed, 0);
3363 
3364 	spdk_put_io_channel(ioch);
3365 	spdk_bdev_close(desc);
3366 	free_bdev(bdev);
3367 	fn_table.submit_request = stub_submit_request;
3368 	spdk_bdev_finish(bdev_fini_cb, NULL);
3369 	poll_threads();
3370 
3371 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3372 
3373 	g_compare_read_buf = NULL;
3374 	g_compare_write_buf = NULL;
3375 }
3376 
3377 static void
3378 bdev_write_zeroes(void)
3379 {
3380 	struct spdk_bdev *bdev;
3381 	struct spdk_bdev_desc *desc = NULL;
3382 	struct spdk_io_channel *ioch;
3383 	struct ut_expected_io *expected_io;
3384 	uint64_t offset, num_io_blocks, num_blocks;
3385 	uint32_t num_completed, num_requests;
3386 	int rc;
3387 
3388 	spdk_bdev_initialize(bdev_init_cb, NULL);
3389 	bdev = allocate_bdev("bdev");
3390 
3391 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3392 	CU_ASSERT_EQUAL(rc, 0);
3393 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3394 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3395 	ioch = spdk_bdev_get_io_channel(desc);
3396 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3397 
3398 	fn_table.submit_request = stub_submit_request;
3399 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3400 
3401 	/* First test that if the bdev supports write_zeroes, the request won't be split */
3402 	bdev->md_len = 0;
3403 	bdev->blocklen = 4096;
3404 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3405 
3406 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
3407 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3408 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3409 	CU_ASSERT_EQUAL(rc, 0);
3410 	num_completed = stub_complete_io(1);
3411 	CU_ASSERT_EQUAL(num_completed, 1);
3412 
3413 	/* Check that if write zeroes is not supported it'll be replaced by regular writes */
3414 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
3415 	num_io_blocks = ZERO_BUFFER_SIZE / bdev->blocklen;
3416 	num_requests = 2;
3417 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * num_requests;
3418 
3419 	for (offset = 0; offset < num_requests; ++offset) {
3420 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3421 						   offset * num_io_blocks, num_io_blocks, 0);
3422 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3423 	}
3424 
3425 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3426 	CU_ASSERT_EQUAL(rc, 0);
3427 	num_completed = stub_complete_io(num_requests);
3428 	CU_ASSERT_EQUAL(num_completed, num_requests);
3429 
3430 	/* Check that the splitting is correct if bdev has interleaved metadata */
3431 	bdev->md_interleave = true;
3432 	bdev->md_len = 64;
3433 	bdev->blocklen = 4096 + 64;
3434 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3435 
3436 	num_requests = offset = 0;
3437 	while (offset < num_blocks) {
3438 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / bdev->blocklen, num_blocks - offset);
3439 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3440 						   offset, num_io_blocks, 0);
3441 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3442 		offset += num_io_blocks;
3443 		num_requests++;
3444 	}
3445 
3446 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3447 	CU_ASSERT_EQUAL(rc, 0);
3448 	num_completed = stub_complete_io(num_requests);
3449 	CU_ASSERT_EQUAL(num_completed, num_requests);
3450 	num_completed = stub_complete_io(num_requests);
3451 	assert(num_completed == 0);
3452 
3453 	/* Check the the same for separate metadata buffer */
3454 	bdev->md_interleave = false;
3455 	bdev->md_len = 64;
3456 	bdev->blocklen = 4096;
3457 
3458 	num_requests = offset = 0;
3459 	while (offset < num_blocks) {
3460 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / (bdev->blocklen + bdev->md_len), num_blocks);
3461 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3462 						   offset, num_io_blocks, 0);
3463 		expected_io->md_buf = (char *)g_bdev_mgr.zero_buffer + num_io_blocks * bdev->blocklen;
3464 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3465 		offset += num_io_blocks;
3466 		num_requests++;
3467 	}
3468 
3469 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3470 	CU_ASSERT_EQUAL(rc, 0);
3471 	num_completed = stub_complete_io(num_requests);
3472 	CU_ASSERT_EQUAL(num_completed, num_requests);
3473 
3474 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
3475 	spdk_put_io_channel(ioch);
3476 	spdk_bdev_close(desc);
3477 	free_bdev(bdev);
3478 	spdk_bdev_finish(bdev_fini_cb, NULL);
3479 	poll_threads();
3480 }
3481 
3482 static void
3483 bdev_zcopy_write(void)
3484 {
3485 	struct spdk_bdev *bdev;
3486 	struct spdk_bdev_desc *desc = NULL;
3487 	struct spdk_io_channel *ioch;
3488 	struct ut_expected_io *expected_io;
3489 	uint64_t offset, num_blocks;
3490 	uint32_t num_completed;
3491 	char aa_buf[512];
3492 	struct iovec iov;
3493 	int rc;
3494 	const bool populate = false;
3495 	const bool commit = true;
3496 
3497 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3498 
3499 	spdk_bdev_initialize(bdev_init_cb, NULL);
3500 	bdev = allocate_bdev("bdev");
3501 
3502 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3503 	CU_ASSERT_EQUAL(rc, 0);
3504 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3505 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3506 	ioch = spdk_bdev_get_io_channel(desc);
3507 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3508 
3509 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3510 
3511 	offset = 50;
3512 	num_blocks = 1;
3513 	iov.iov_base = NULL;
3514 	iov.iov_len = 0;
3515 
3516 	g_zcopy_read_buf = (void *) 0x1122334455667788UL;
3517 	g_zcopy_read_buf_len = (uint32_t) -1;
3518 	/* Do a zcopy start for a write (populate=false) */
3519 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3520 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3521 	g_io_done = false;
3522 	g_zcopy_write_buf = aa_buf;
3523 	g_zcopy_write_buf_len = sizeof(aa_buf);
3524 	g_zcopy_bdev_io = NULL;
3525 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3526 	CU_ASSERT_EQUAL(rc, 0);
3527 	num_completed = stub_complete_io(1);
3528 	CU_ASSERT_EQUAL(num_completed, 1);
3529 	CU_ASSERT(g_io_done == true);
3530 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3531 	/* Check that the iov has been set up */
3532 	CU_ASSERT(iov.iov_base == g_zcopy_write_buf);
3533 	CU_ASSERT(iov.iov_len == g_zcopy_write_buf_len);
3534 	/* Check that the bdev_io has been saved */
3535 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3536 	/* Now do the zcopy end for a write (commit=true) */
3537 	g_io_done = false;
3538 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3539 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3540 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3541 	CU_ASSERT_EQUAL(rc, 0);
3542 	num_completed = stub_complete_io(1);
3543 	CU_ASSERT_EQUAL(num_completed, 1);
3544 	CU_ASSERT(g_io_done == true);
3545 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3546 	/* Check the g_zcopy are reset by io_done */
3547 	CU_ASSERT(g_zcopy_write_buf == NULL);
3548 	CU_ASSERT(g_zcopy_write_buf_len == 0);
3549 	/* Check that io_done has freed the g_zcopy_bdev_io */
3550 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3551 
3552 	/* Check the zcopy read buffer has not been touched which
3553 	 * ensures that the correct buffers were used.
3554 	 */
3555 	CU_ASSERT(g_zcopy_read_buf == (void *) 0x1122334455667788UL);
3556 	CU_ASSERT(g_zcopy_read_buf_len == (uint32_t) -1);
3557 
3558 	spdk_put_io_channel(ioch);
3559 	spdk_bdev_close(desc);
3560 	free_bdev(bdev);
3561 	spdk_bdev_finish(bdev_fini_cb, NULL);
3562 	poll_threads();
3563 }
3564 
3565 static void
3566 bdev_zcopy_read(void)
3567 {
3568 	struct spdk_bdev *bdev;
3569 	struct spdk_bdev_desc *desc = NULL;
3570 	struct spdk_io_channel *ioch;
3571 	struct ut_expected_io *expected_io;
3572 	uint64_t offset, num_blocks;
3573 	uint32_t num_completed;
3574 	char aa_buf[512];
3575 	struct iovec iov;
3576 	int rc;
3577 	const bool populate = true;
3578 	const bool commit = false;
3579 
3580 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3581 
3582 	spdk_bdev_initialize(bdev_init_cb, NULL);
3583 	bdev = allocate_bdev("bdev");
3584 
3585 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3586 	CU_ASSERT_EQUAL(rc, 0);
3587 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3588 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3589 	ioch = spdk_bdev_get_io_channel(desc);
3590 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3591 
3592 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3593 
3594 	offset = 50;
3595 	num_blocks = 1;
3596 	iov.iov_base = NULL;
3597 	iov.iov_len = 0;
3598 
3599 	g_zcopy_write_buf = (void *) 0x1122334455667788UL;
3600 	g_zcopy_write_buf_len = (uint32_t) -1;
3601 
3602 	/* Do a zcopy start for a read (populate=true) */
3603 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3604 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3605 	g_io_done = false;
3606 	g_zcopy_read_buf = aa_buf;
3607 	g_zcopy_read_buf_len = sizeof(aa_buf);
3608 	g_zcopy_bdev_io = NULL;
3609 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3610 	CU_ASSERT_EQUAL(rc, 0);
3611 	num_completed = stub_complete_io(1);
3612 	CU_ASSERT_EQUAL(num_completed, 1);
3613 	CU_ASSERT(g_io_done == true);
3614 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3615 	/* Check that the iov has been set up */
3616 	CU_ASSERT(iov.iov_base == g_zcopy_read_buf);
3617 	CU_ASSERT(iov.iov_len == g_zcopy_read_buf_len);
3618 	/* Check that the bdev_io has been saved */
3619 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3620 
3621 	/* Now do the zcopy end for a read (commit=false) */
3622 	g_io_done = false;
3623 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3624 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3625 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3626 	CU_ASSERT_EQUAL(rc, 0);
3627 	num_completed = stub_complete_io(1);
3628 	CU_ASSERT_EQUAL(num_completed, 1);
3629 	CU_ASSERT(g_io_done == true);
3630 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3631 	/* Check the g_zcopy are reset by io_done */
3632 	CU_ASSERT(g_zcopy_read_buf == NULL);
3633 	CU_ASSERT(g_zcopy_read_buf_len == 0);
3634 	/* Check that io_done has freed the g_zcopy_bdev_io */
3635 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3636 
3637 	/* Check the zcopy write buffer has not been touched which
3638 	 * ensures that the correct buffers were used.
3639 	 */
3640 	CU_ASSERT(g_zcopy_write_buf == (void *) 0x1122334455667788UL);
3641 	CU_ASSERT(g_zcopy_write_buf_len == (uint32_t) -1);
3642 
3643 	spdk_put_io_channel(ioch);
3644 	spdk_bdev_close(desc);
3645 	free_bdev(bdev);
3646 	spdk_bdev_finish(bdev_fini_cb, NULL);
3647 	poll_threads();
3648 }
3649 
3650 static void
3651 bdev_open_while_hotremove(void)
3652 {
3653 	struct spdk_bdev *bdev;
3654 	struct spdk_bdev_desc *desc[2] = {};
3655 	int rc;
3656 
3657 	bdev = allocate_bdev("bdev");
3658 
3659 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[0]);
3660 	CU_ASSERT(rc == 0);
3661 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
3662 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc[0]));
3663 
3664 	spdk_bdev_unregister(bdev, NULL, NULL);
3665 
3666 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[1]);
3667 	CU_ASSERT(rc == -ENODEV);
3668 	SPDK_CU_ASSERT_FATAL(desc[1] == NULL);
3669 
3670 	spdk_bdev_close(desc[0]);
3671 	free_bdev(bdev);
3672 }
3673 
3674 static void
3675 bdev_close_while_hotremove(void)
3676 {
3677 	struct spdk_bdev *bdev;
3678 	struct spdk_bdev_desc *desc = NULL;
3679 	int rc = 0;
3680 
3681 	bdev = allocate_bdev("bdev");
3682 
3683 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc, &desc);
3684 	CU_ASSERT_EQUAL(rc, 0);
3685 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3686 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3687 
3688 	/* Simulate hot-unplug by unregistering bdev */
3689 	g_event_type1 = 0xFF;
3690 	g_unregister_arg = NULL;
3691 	g_unregister_rc = -1;
3692 	spdk_bdev_unregister(bdev, bdev_unregister_cb, (void *)0x12345678);
3693 	/* Close device while remove event is in flight */
3694 	spdk_bdev_close(desc);
3695 
3696 	/* Ensure that unregister callback is delayed */
3697 	CU_ASSERT_EQUAL(g_unregister_arg, NULL);
3698 	CU_ASSERT_EQUAL(g_unregister_rc, -1);
3699 
3700 	poll_threads();
3701 
3702 	/* Event callback shall not be issued because device was closed */
3703 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
3704 	/* Unregister callback is issued */
3705 	CU_ASSERT_EQUAL(g_unregister_arg, (void *)0x12345678);
3706 	CU_ASSERT_EQUAL(g_unregister_rc, 0);
3707 
3708 	free_bdev(bdev);
3709 }
3710 
3711 static void
3712 bdev_open_ext(void)
3713 {
3714 	struct spdk_bdev *bdev;
3715 	struct spdk_bdev_desc *desc1 = NULL;
3716 	struct spdk_bdev_desc *desc2 = NULL;
3717 	int rc = 0;
3718 
3719 	bdev = allocate_bdev("bdev");
3720 
3721 	rc = spdk_bdev_open_ext("bdev", true, NULL, NULL, &desc1);
3722 	CU_ASSERT_EQUAL(rc, -EINVAL);
3723 
3724 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc1, &desc1);
3725 	CU_ASSERT_EQUAL(rc, 0);
3726 
3727 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb2, &desc2, &desc2);
3728 	CU_ASSERT_EQUAL(rc, 0);
3729 
3730 	g_event_type1 = 0xFF;
3731 	g_event_type2 = 0xFF;
3732 
3733 	/* Simulate hot-unplug by unregistering bdev */
3734 	spdk_bdev_unregister(bdev, NULL, NULL);
3735 	poll_threads();
3736 
3737 	/* Check if correct events have been triggered in event callback fn */
3738 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_REMOVE);
3739 	CU_ASSERT_EQUAL(g_event_type2, SPDK_BDEV_EVENT_REMOVE);
3740 
3741 	free_bdev(bdev);
3742 	poll_threads();
3743 }
3744 
3745 struct timeout_io_cb_arg {
3746 	struct iovec iov;
3747 	uint8_t type;
3748 };
3749 
3750 static int
3751 bdev_channel_count_submitted_io(struct spdk_bdev_channel *ch)
3752 {
3753 	struct spdk_bdev_io *bdev_io;
3754 	int n = 0;
3755 
3756 	if (!ch) {
3757 		return -1;
3758 	}
3759 
3760 	TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
3761 		n++;
3762 	}
3763 
3764 	return n;
3765 }
3766 
3767 static void
3768 bdev_channel_io_timeout_cb(void *cb_arg, struct spdk_bdev_io *bdev_io)
3769 {
3770 	struct timeout_io_cb_arg *ctx = cb_arg;
3771 
3772 	ctx->type = bdev_io->type;
3773 	ctx->iov.iov_base = bdev_io->iov.iov_base;
3774 	ctx->iov.iov_len = bdev_io->iov.iov_len;
3775 }
3776 
3777 static void
3778 bdev_set_io_timeout(void)
3779 {
3780 	struct spdk_bdev *bdev;
3781 	struct spdk_bdev_desc *desc = NULL;
3782 	struct spdk_io_channel *io_ch = NULL;
3783 	struct spdk_bdev_channel *bdev_ch = NULL;
3784 	struct timeout_io_cb_arg cb_arg;
3785 
3786 	spdk_bdev_initialize(bdev_init_cb, NULL);
3787 
3788 	bdev = allocate_bdev("bdev");
3789 
3790 	CU_ASSERT(spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc) == 0);
3791 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3792 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3793 
3794 	io_ch = spdk_bdev_get_io_channel(desc);
3795 	CU_ASSERT(io_ch != NULL);
3796 
3797 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
3798 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
3799 
3800 	/* This is the part1.
3801 	 * We will check the bdev_ch->io_submitted list
3802 	 * TO make sure that it can link IOs and only the user submitted IOs
3803 	 */
3804 	CU_ASSERT(spdk_bdev_read(desc, io_ch, (void *)0x1000, 0, 4096, io_done, NULL) == 0);
3805 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3806 	CU_ASSERT(spdk_bdev_write(desc, io_ch, (void *)0x2000, 0, 4096, io_done, NULL) == 0);
3807 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3808 	stub_complete_io(1);
3809 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3810 	stub_complete_io(1);
3811 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3812 
3813 	/* Split IO */
3814 	bdev->optimal_io_boundary = 16;
3815 	bdev->split_on_optimal_io_boundary = true;
3816 
3817 	/* Now test that a single-vector command is split correctly.
3818 	 * Offset 14, length 8, payload 0xF000
3819 	 *  Child - Offset 14, length 2, payload 0xF000
3820 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
3821 	 *
3822 	 * Set up the expected values before calling spdk_bdev_read_blocks
3823 	 */
3824 	CU_ASSERT(spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3825 	/* We count all submitted IOs including IO that are generated by splitting. */
3826 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 3);
3827 	stub_complete_io(1);
3828 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3829 	stub_complete_io(1);
3830 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3831 
3832 	/* Also include the reset IO */
3833 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3834 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3835 	poll_threads();
3836 	stub_complete_io(1);
3837 	poll_threads();
3838 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3839 
3840 	/* This is part2
3841 	 * Test the desc timeout poller register
3842 	 */
3843 
3844 	/* Successfully set the timeout */
3845 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3846 	CU_ASSERT(desc->io_timeout_poller != NULL);
3847 	CU_ASSERT(desc->timeout_in_sec == 30);
3848 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3849 	CU_ASSERT(desc->cb_arg == &cb_arg);
3850 
3851 	/* Change the timeout limit */
3852 	CU_ASSERT(spdk_bdev_set_timeout(desc, 20, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3853 	CU_ASSERT(desc->io_timeout_poller != NULL);
3854 	CU_ASSERT(desc->timeout_in_sec == 20);
3855 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3856 	CU_ASSERT(desc->cb_arg == &cb_arg);
3857 
3858 	/* Disable the timeout */
3859 	CU_ASSERT(spdk_bdev_set_timeout(desc, 0, NULL, NULL) == 0);
3860 	CU_ASSERT(desc->io_timeout_poller == NULL);
3861 
3862 	/* This the part3
3863 	 * We will test to catch timeout IO and check whether the IO is
3864 	 * the submitted one.
3865 	 */
3866 	memset(&cb_arg, 0, sizeof(cb_arg));
3867 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3868 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0x1000, 0, 1, io_done, NULL) == 0);
3869 
3870 	/* Don't reach the limit */
3871 	spdk_delay_us(15 * spdk_get_ticks_hz());
3872 	poll_threads();
3873 	CU_ASSERT(cb_arg.type == 0);
3874 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3875 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3876 
3877 	/* 15 + 15 = 30 reach the limit */
3878 	spdk_delay_us(15 * spdk_get_ticks_hz());
3879 	poll_threads();
3880 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3881 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x1000);
3882 	CU_ASSERT(cb_arg.iov.iov_len == 1 * bdev->blocklen);
3883 	stub_complete_io(1);
3884 
3885 	/* Use the same split IO above and check the IO */
3886 	memset(&cb_arg, 0, sizeof(cb_arg));
3887 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3888 
3889 	/* The first child complete in time */
3890 	spdk_delay_us(15 * spdk_get_ticks_hz());
3891 	poll_threads();
3892 	stub_complete_io(1);
3893 	CU_ASSERT(cb_arg.type == 0);
3894 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3895 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3896 
3897 	/* The second child reach the limit */
3898 	spdk_delay_us(15 * spdk_get_ticks_hz());
3899 	poll_threads();
3900 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3901 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0xF000);
3902 	CU_ASSERT(cb_arg.iov.iov_len == 8 * bdev->blocklen);
3903 	stub_complete_io(1);
3904 
3905 	/* Also include the reset IO */
3906 	memset(&cb_arg, 0, sizeof(cb_arg));
3907 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3908 	spdk_delay_us(30 * spdk_get_ticks_hz());
3909 	poll_threads();
3910 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_RESET);
3911 	stub_complete_io(1);
3912 	poll_threads();
3913 
3914 	spdk_put_io_channel(io_ch);
3915 	spdk_bdev_close(desc);
3916 	free_bdev(bdev);
3917 	spdk_bdev_finish(bdev_fini_cb, NULL);
3918 	poll_threads();
3919 }
3920 
3921 static void
3922 lba_range_overlap(void)
3923 {
3924 	struct lba_range r1, r2;
3925 
3926 	r1.offset = 100;
3927 	r1.length = 50;
3928 
3929 	r2.offset = 0;
3930 	r2.length = 1;
3931 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3932 
3933 	r2.offset = 0;
3934 	r2.length = 100;
3935 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3936 
3937 	r2.offset = 0;
3938 	r2.length = 110;
3939 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3940 
3941 	r2.offset = 100;
3942 	r2.length = 10;
3943 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3944 
3945 	r2.offset = 110;
3946 	r2.length = 20;
3947 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3948 
3949 	r2.offset = 140;
3950 	r2.length = 150;
3951 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3952 
3953 	r2.offset = 130;
3954 	r2.length = 200;
3955 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3956 
3957 	r2.offset = 150;
3958 	r2.length = 100;
3959 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3960 
3961 	r2.offset = 110;
3962 	r2.length = 0;
3963 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3964 }
3965 
3966 static bool g_lock_lba_range_done;
3967 static bool g_unlock_lba_range_done;
3968 
3969 static void
3970 lock_lba_range_done(void *ctx, int status)
3971 {
3972 	g_lock_lba_range_done = true;
3973 }
3974 
3975 static void
3976 unlock_lba_range_done(void *ctx, int status)
3977 {
3978 	g_unlock_lba_range_done = true;
3979 }
3980 
3981 static void
3982 lock_lba_range_check_ranges(void)
3983 {
3984 	struct spdk_bdev *bdev;
3985 	struct spdk_bdev_desc *desc = NULL;
3986 	struct spdk_io_channel *io_ch;
3987 	struct spdk_bdev_channel *channel;
3988 	struct lba_range *range;
3989 	int ctx1;
3990 	int rc;
3991 
3992 	spdk_bdev_initialize(bdev_init_cb, NULL);
3993 
3994 	bdev = allocate_bdev("bdev0");
3995 
3996 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
3997 	CU_ASSERT(rc == 0);
3998 	CU_ASSERT(desc != NULL);
3999 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4000 	io_ch = spdk_bdev_get_io_channel(desc);
4001 	CU_ASSERT(io_ch != NULL);
4002 	channel = spdk_io_channel_get_ctx(io_ch);
4003 
4004 	g_lock_lba_range_done = false;
4005 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4006 	CU_ASSERT(rc == 0);
4007 	poll_threads();
4008 
4009 	CU_ASSERT(g_lock_lba_range_done == true);
4010 	range = TAILQ_FIRST(&channel->locked_ranges);
4011 	SPDK_CU_ASSERT_FATAL(range != NULL);
4012 	CU_ASSERT(range->offset == 20);
4013 	CU_ASSERT(range->length == 10);
4014 	CU_ASSERT(range->owner_ch == channel);
4015 
4016 	/* Unlocks must exactly match a lock. */
4017 	g_unlock_lba_range_done = false;
4018 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 1, unlock_lba_range_done, &ctx1);
4019 	CU_ASSERT(rc == -EINVAL);
4020 	CU_ASSERT(g_unlock_lba_range_done == false);
4021 
4022 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4023 	CU_ASSERT(rc == 0);
4024 	spdk_delay_us(100);
4025 	poll_threads();
4026 
4027 	CU_ASSERT(g_unlock_lba_range_done == true);
4028 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4029 
4030 	spdk_put_io_channel(io_ch);
4031 	spdk_bdev_close(desc);
4032 	free_bdev(bdev);
4033 	spdk_bdev_finish(bdev_fini_cb, NULL);
4034 	poll_threads();
4035 }
4036 
4037 static void
4038 lock_lba_range_with_io_outstanding(void)
4039 {
4040 	struct spdk_bdev *bdev;
4041 	struct spdk_bdev_desc *desc = NULL;
4042 	struct spdk_io_channel *io_ch;
4043 	struct spdk_bdev_channel *channel;
4044 	struct lba_range *range;
4045 	char buf[4096];
4046 	int ctx1;
4047 	int rc;
4048 
4049 	spdk_bdev_initialize(bdev_init_cb, NULL);
4050 
4051 	bdev = allocate_bdev("bdev0");
4052 
4053 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4054 	CU_ASSERT(rc == 0);
4055 	CU_ASSERT(desc != NULL);
4056 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4057 	io_ch = spdk_bdev_get_io_channel(desc);
4058 	CU_ASSERT(io_ch != NULL);
4059 	channel = spdk_io_channel_get_ctx(io_ch);
4060 
4061 	g_io_done = false;
4062 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4063 	CU_ASSERT(rc == 0);
4064 
4065 	g_lock_lba_range_done = false;
4066 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4067 	CU_ASSERT(rc == 0);
4068 	poll_threads();
4069 
4070 	/* The lock should immediately become valid, since there are no outstanding
4071 	 * write I/O.
4072 	 */
4073 	CU_ASSERT(g_io_done == false);
4074 	CU_ASSERT(g_lock_lba_range_done == true);
4075 	range = TAILQ_FIRST(&channel->locked_ranges);
4076 	SPDK_CU_ASSERT_FATAL(range != NULL);
4077 	CU_ASSERT(range->offset == 20);
4078 	CU_ASSERT(range->length == 10);
4079 	CU_ASSERT(range->owner_ch == channel);
4080 	CU_ASSERT(range->locked_ctx == &ctx1);
4081 
4082 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4083 	CU_ASSERT(rc == 0);
4084 	stub_complete_io(1);
4085 	spdk_delay_us(100);
4086 	poll_threads();
4087 
4088 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4089 
4090 	/* Now try again, but with a write I/O. */
4091 	g_io_done = false;
4092 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4093 	CU_ASSERT(rc == 0);
4094 
4095 	g_lock_lba_range_done = false;
4096 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4097 	CU_ASSERT(rc == 0);
4098 	poll_threads();
4099 
4100 	/* The lock should not be fully valid yet, since a write I/O is outstanding.
4101 	 * But note that the range should be on the channel's locked_list, to make sure no
4102 	 * new write I/O are started.
4103 	 */
4104 	CU_ASSERT(g_io_done == false);
4105 	CU_ASSERT(g_lock_lba_range_done == false);
4106 	range = TAILQ_FIRST(&channel->locked_ranges);
4107 	SPDK_CU_ASSERT_FATAL(range != NULL);
4108 	CU_ASSERT(range->offset == 20);
4109 	CU_ASSERT(range->length == 10);
4110 
4111 	/* Complete the write I/O.  This should make the lock valid (checked by confirming
4112 	 * our callback was invoked).
4113 	 */
4114 	stub_complete_io(1);
4115 	spdk_delay_us(100);
4116 	poll_threads();
4117 	CU_ASSERT(g_io_done == true);
4118 	CU_ASSERT(g_lock_lba_range_done == true);
4119 
4120 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4121 	CU_ASSERT(rc == 0);
4122 	poll_threads();
4123 
4124 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4125 
4126 	spdk_put_io_channel(io_ch);
4127 	spdk_bdev_close(desc);
4128 	free_bdev(bdev);
4129 	spdk_bdev_finish(bdev_fini_cb, NULL);
4130 	poll_threads();
4131 }
4132 
4133 static void
4134 lock_lba_range_overlapped(void)
4135 {
4136 	struct spdk_bdev *bdev;
4137 	struct spdk_bdev_desc *desc = NULL;
4138 	struct spdk_io_channel *io_ch;
4139 	struct spdk_bdev_channel *channel;
4140 	struct lba_range *range;
4141 	int ctx1;
4142 	int rc;
4143 
4144 	spdk_bdev_initialize(bdev_init_cb, NULL);
4145 
4146 	bdev = allocate_bdev("bdev0");
4147 
4148 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4149 	CU_ASSERT(rc == 0);
4150 	CU_ASSERT(desc != NULL);
4151 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4152 	io_ch = spdk_bdev_get_io_channel(desc);
4153 	CU_ASSERT(io_ch != NULL);
4154 	channel = spdk_io_channel_get_ctx(io_ch);
4155 
4156 	/* Lock range 20-29. */
4157 	g_lock_lba_range_done = false;
4158 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4159 	CU_ASSERT(rc == 0);
4160 	poll_threads();
4161 
4162 	CU_ASSERT(g_lock_lba_range_done == true);
4163 	range = TAILQ_FIRST(&channel->locked_ranges);
4164 	SPDK_CU_ASSERT_FATAL(range != NULL);
4165 	CU_ASSERT(range->offset == 20);
4166 	CU_ASSERT(range->length == 10);
4167 
4168 	/* Try to lock range 25-39.  It should not lock immediately, since it overlaps with
4169 	 * 20-29.
4170 	 */
4171 	g_lock_lba_range_done = false;
4172 	rc = bdev_lock_lba_range(desc, io_ch, 25, 15, lock_lba_range_done, &ctx1);
4173 	CU_ASSERT(rc == 0);
4174 	poll_threads();
4175 
4176 	CU_ASSERT(g_lock_lba_range_done == false);
4177 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4178 	SPDK_CU_ASSERT_FATAL(range != NULL);
4179 	CU_ASSERT(range->offset == 25);
4180 	CU_ASSERT(range->length == 15);
4181 
4182 	/* Unlock 20-29.  This should result in range 25-39 now getting locked since it
4183 	 * no longer overlaps with an active lock.
4184 	 */
4185 	g_unlock_lba_range_done = false;
4186 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4187 	CU_ASSERT(rc == 0);
4188 	poll_threads();
4189 
4190 	CU_ASSERT(g_unlock_lba_range_done == true);
4191 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4192 	range = TAILQ_FIRST(&channel->locked_ranges);
4193 	SPDK_CU_ASSERT_FATAL(range != NULL);
4194 	CU_ASSERT(range->offset == 25);
4195 	CU_ASSERT(range->length == 15);
4196 
4197 	/* Lock 40-59.  This should immediately lock since it does not overlap with the
4198 	 * currently active 25-39 lock.
4199 	 */
4200 	g_lock_lba_range_done = false;
4201 	rc = bdev_lock_lba_range(desc, io_ch, 40, 20, lock_lba_range_done, &ctx1);
4202 	CU_ASSERT(rc == 0);
4203 	poll_threads();
4204 
4205 	CU_ASSERT(g_lock_lba_range_done == true);
4206 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4207 	SPDK_CU_ASSERT_FATAL(range != NULL);
4208 	range = TAILQ_NEXT(range, tailq);
4209 	SPDK_CU_ASSERT_FATAL(range != NULL);
4210 	CU_ASSERT(range->offset == 40);
4211 	CU_ASSERT(range->length == 20);
4212 
4213 	/* Try to lock 35-44.  Note that this overlaps with both 25-39 and 40-59. */
4214 	g_lock_lba_range_done = false;
4215 	rc = bdev_lock_lba_range(desc, io_ch, 35, 10, lock_lba_range_done, &ctx1);
4216 	CU_ASSERT(rc == 0);
4217 	poll_threads();
4218 
4219 	CU_ASSERT(g_lock_lba_range_done == false);
4220 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4221 	SPDK_CU_ASSERT_FATAL(range != NULL);
4222 	CU_ASSERT(range->offset == 35);
4223 	CU_ASSERT(range->length == 10);
4224 
4225 	/* Unlock 25-39.  Make sure that 35-44 is still in the pending list, since
4226 	 * the 40-59 lock is still active.
4227 	 */
4228 	g_unlock_lba_range_done = false;
4229 	rc = bdev_unlock_lba_range(desc, io_ch, 25, 15, unlock_lba_range_done, &ctx1);
4230 	CU_ASSERT(rc == 0);
4231 	poll_threads();
4232 
4233 	CU_ASSERT(g_unlock_lba_range_done == true);
4234 	CU_ASSERT(g_lock_lba_range_done == false);
4235 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4236 	SPDK_CU_ASSERT_FATAL(range != NULL);
4237 	CU_ASSERT(range->offset == 35);
4238 	CU_ASSERT(range->length == 10);
4239 
4240 	/* Unlock 40-59.  This should result in 35-44 now getting locked, since there are
4241 	 * no longer any active overlapping locks.
4242 	 */
4243 	g_unlock_lba_range_done = false;
4244 	rc = bdev_unlock_lba_range(desc, io_ch, 40, 20, unlock_lba_range_done, &ctx1);
4245 	CU_ASSERT(rc == 0);
4246 	poll_threads();
4247 
4248 	CU_ASSERT(g_unlock_lba_range_done == true);
4249 	CU_ASSERT(g_lock_lba_range_done == true);
4250 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4251 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4252 	SPDK_CU_ASSERT_FATAL(range != NULL);
4253 	CU_ASSERT(range->offset == 35);
4254 	CU_ASSERT(range->length == 10);
4255 
4256 	/* Finally, unlock 35-44. */
4257 	g_unlock_lba_range_done = false;
4258 	rc = bdev_unlock_lba_range(desc, io_ch, 35, 10, unlock_lba_range_done, &ctx1);
4259 	CU_ASSERT(rc == 0);
4260 	poll_threads();
4261 
4262 	CU_ASSERT(g_unlock_lba_range_done == true);
4263 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.locked_ranges));
4264 
4265 	spdk_put_io_channel(io_ch);
4266 	spdk_bdev_close(desc);
4267 	free_bdev(bdev);
4268 	spdk_bdev_finish(bdev_fini_cb, NULL);
4269 	poll_threads();
4270 }
4271 
4272 static void
4273 abort_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
4274 {
4275 	g_abort_done = true;
4276 	g_abort_status = bdev_io->internal.status;
4277 	spdk_bdev_free_io(bdev_io);
4278 }
4279 
4280 static void
4281 bdev_io_abort(void)
4282 {
4283 	struct spdk_bdev *bdev;
4284 	struct spdk_bdev_desc *desc = NULL;
4285 	struct spdk_io_channel *io_ch;
4286 	struct spdk_bdev_channel *channel;
4287 	struct spdk_bdev_mgmt_channel *mgmt_ch;
4288 	struct spdk_bdev_opts bdev_opts = {};
4289 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
4290 	uint64_t io_ctx1 = 0, io_ctx2 = 0, i;
4291 	int rc;
4292 
4293 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4294 	bdev_opts.bdev_io_pool_size = 7;
4295 	bdev_opts.bdev_io_cache_size = 2;
4296 
4297 	rc = spdk_bdev_set_opts(&bdev_opts);
4298 	CU_ASSERT(rc == 0);
4299 	spdk_bdev_initialize(bdev_init_cb, NULL);
4300 
4301 	bdev = allocate_bdev("bdev0");
4302 
4303 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4304 	CU_ASSERT(rc == 0);
4305 	CU_ASSERT(desc != NULL);
4306 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4307 	io_ch = spdk_bdev_get_io_channel(desc);
4308 	CU_ASSERT(io_ch != NULL);
4309 	channel = spdk_io_channel_get_ctx(io_ch);
4310 	mgmt_ch = channel->shared_resource->mgmt_ch;
4311 
4312 	g_abort_done = false;
4313 
4314 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, false);
4315 
4316 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4317 	CU_ASSERT(rc == -ENOTSUP);
4318 
4319 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, true);
4320 
4321 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx2, abort_done, NULL);
4322 	CU_ASSERT(rc == 0);
4323 	CU_ASSERT(g_abort_done == true);
4324 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_FAILED);
4325 
4326 	/* Test the case that the target I/O was successfully aborted. */
4327 	g_io_done = false;
4328 
4329 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4330 	CU_ASSERT(rc == 0);
4331 	CU_ASSERT(g_io_done == false);
4332 
4333 	g_abort_done = false;
4334 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4335 
4336 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4337 	CU_ASSERT(rc == 0);
4338 	CU_ASSERT(g_io_done == true);
4339 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4340 	stub_complete_io(1);
4341 	CU_ASSERT(g_abort_done == true);
4342 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4343 
4344 	/* Test the case that the target I/O was not aborted because it completed
4345 	 * in the middle of execution of the abort.
4346 	 */
4347 	g_io_done = false;
4348 
4349 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4350 	CU_ASSERT(rc == 0);
4351 	CU_ASSERT(g_io_done == false);
4352 
4353 	g_abort_done = false;
4354 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4355 
4356 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4357 	CU_ASSERT(rc == 0);
4358 	CU_ASSERT(g_io_done == false);
4359 
4360 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4361 	stub_complete_io(1);
4362 	CU_ASSERT(g_io_done == true);
4363 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4364 
4365 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4366 	stub_complete_io(1);
4367 	CU_ASSERT(g_abort_done == true);
4368 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4369 
4370 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4371 
4372 	bdev->optimal_io_boundary = 16;
4373 	bdev->split_on_optimal_io_boundary = true;
4374 
4375 	/* Test that a single-vector command which is split is aborted correctly.
4376 	 * Offset 14, length 8, payload 0xF000
4377 	 *  Child - Offset 14, length 2, payload 0xF000
4378 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
4379 	 */
4380 	g_io_done = false;
4381 
4382 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, &io_ctx1);
4383 	CU_ASSERT(rc == 0);
4384 	CU_ASSERT(g_io_done == false);
4385 
4386 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4387 
4388 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4389 
4390 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4391 	CU_ASSERT(rc == 0);
4392 	CU_ASSERT(g_io_done == true);
4393 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4394 	stub_complete_io(2);
4395 	CU_ASSERT(g_abort_done == true);
4396 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4397 
4398 	/* Test that a multi-vector command that needs to be split by strip and then
4399 	 * needs to be split is aborted correctly. Abort is requested before the second
4400 	 * child I/O was submitted. The parent I/O should complete with failure without
4401 	 * submitting the second child I/O.
4402 	 */
4403 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
4404 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
4405 		iov[i].iov_len = 512;
4406 	}
4407 
4408 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
4409 	g_io_done = false;
4410 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
4411 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, &io_ctx1);
4412 	CU_ASSERT(rc == 0);
4413 	CU_ASSERT(g_io_done == false);
4414 
4415 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4416 
4417 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4418 
4419 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4420 	CU_ASSERT(rc == 0);
4421 	CU_ASSERT(g_io_done == true);
4422 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4423 	stub_complete_io(1);
4424 	CU_ASSERT(g_abort_done == true);
4425 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4426 
4427 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4428 
4429 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4430 
4431 	bdev->optimal_io_boundary = 16;
4432 	g_io_done = false;
4433 
4434 	/* Test that a ingle-vector command which is split is aborted correctly.
4435 	 * Differently from the above, the child abort request will be submitted
4436 	 * sequentially due to the capacity of spdk_bdev_io.
4437 	 */
4438 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 50, io_done, &io_ctx1);
4439 	CU_ASSERT(rc == 0);
4440 	CU_ASSERT(g_io_done == false);
4441 
4442 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4443 
4444 	g_abort_done = false;
4445 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4446 
4447 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4448 	CU_ASSERT(rc == 0);
4449 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
4450 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4451 
4452 	stub_complete_io(1);
4453 	CU_ASSERT(g_io_done == true);
4454 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4455 	stub_complete_io(3);
4456 	CU_ASSERT(g_abort_done == true);
4457 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4458 
4459 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4460 
4461 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4462 
4463 	spdk_put_io_channel(io_ch);
4464 	spdk_bdev_close(desc);
4465 	free_bdev(bdev);
4466 	spdk_bdev_finish(bdev_fini_cb, NULL);
4467 	poll_threads();
4468 }
4469 
4470 static void
4471 bdev_unmap(void)
4472 {
4473 	struct spdk_bdev *bdev;
4474 	struct spdk_bdev_desc *desc = NULL;
4475 	struct spdk_io_channel *ioch;
4476 	struct spdk_bdev_channel *bdev_ch;
4477 	struct ut_expected_io *expected_io;
4478 	struct spdk_bdev_opts bdev_opts = {};
4479 	uint32_t i, num_outstanding;
4480 	uint64_t offset, num_blocks, max_unmap_blocks, num_children;
4481 	int rc;
4482 
4483 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4484 	bdev_opts.bdev_io_pool_size = 512;
4485 	bdev_opts.bdev_io_cache_size = 64;
4486 	rc = spdk_bdev_set_opts(&bdev_opts);
4487 	CU_ASSERT(rc == 0);
4488 
4489 	spdk_bdev_initialize(bdev_init_cb, NULL);
4490 	bdev = allocate_bdev("bdev");
4491 
4492 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4493 	CU_ASSERT_EQUAL(rc, 0);
4494 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4495 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4496 	ioch = spdk_bdev_get_io_channel(desc);
4497 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4498 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4499 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4500 
4501 	fn_table.submit_request = stub_submit_request;
4502 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4503 
4504 	/* Case 1: First test the request won't be split */
4505 	num_blocks = 32;
4506 
4507 	g_io_done = false;
4508 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 0, num_blocks, 0);
4509 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4510 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4511 	CU_ASSERT_EQUAL(rc, 0);
4512 	CU_ASSERT(g_io_done == false);
4513 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4514 	stub_complete_io(1);
4515 	CU_ASSERT(g_io_done == true);
4516 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4517 
4518 	/* Case 2: Test the split with 2 children requests */
4519 	bdev->max_unmap = 8;
4520 	bdev->max_unmap_segments = 2;
4521 	max_unmap_blocks = bdev->max_unmap * bdev->max_unmap_segments;
4522 	num_blocks = max_unmap_blocks * 2;
4523 	offset = 0;
4524 
4525 	g_io_done = false;
4526 	for (i = 0; i < 2; i++) {
4527 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4528 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4529 		offset += max_unmap_blocks;
4530 	}
4531 
4532 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4533 	CU_ASSERT_EQUAL(rc, 0);
4534 	CU_ASSERT(g_io_done == false);
4535 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4536 	stub_complete_io(2);
4537 	CU_ASSERT(g_io_done == true);
4538 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4539 
4540 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4541 	num_children = 15;
4542 	num_blocks = max_unmap_blocks * num_children;
4543 	g_io_done = false;
4544 	offset = 0;
4545 	for (i = 0; i < num_children; i++) {
4546 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4547 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4548 		offset += max_unmap_blocks;
4549 	}
4550 
4551 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4552 	CU_ASSERT_EQUAL(rc, 0);
4553 	CU_ASSERT(g_io_done == false);
4554 
4555 	while (num_children > 0) {
4556 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4557 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4558 		stub_complete_io(num_outstanding);
4559 		num_children -= num_outstanding;
4560 	}
4561 	CU_ASSERT(g_io_done == true);
4562 
4563 	spdk_put_io_channel(ioch);
4564 	spdk_bdev_close(desc);
4565 	free_bdev(bdev);
4566 	spdk_bdev_finish(bdev_fini_cb, NULL);
4567 	poll_threads();
4568 }
4569 
4570 static void
4571 bdev_write_zeroes_split_test(void)
4572 {
4573 	struct spdk_bdev *bdev;
4574 	struct spdk_bdev_desc *desc = NULL;
4575 	struct spdk_io_channel *ioch;
4576 	struct spdk_bdev_channel *bdev_ch;
4577 	struct ut_expected_io *expected_io;
4578 	struct spdk_bdev_opts bdev_opts = {};
4579 	uint32_t i, num_outstanding;
4580 	uint64_t offset, num_blocks, max_write_zeroes_blocks, num_children;
4581 	int rc;
4582 
4583 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4584 	bdev_opts.bdev_io_pool_size = 512;
4585 	bdev_opts.bdev_io_cache_size = 64;
4586 	rc = spdk_bdev_set_opts(&bdev_opts);
4587 	CU_ASSERT(rc == 0);
4588 
4589 	spdk_bdev_initialize(bdev_init_cb, NULL);
4590 	bdev = allocate_bdev("bdev");
4591 
4592 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4593 	CU_ASSERT_EQUAL(rc, 0);
4594 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4595 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4596 	ioch = spdk_bdev_get_io_channel(desc);
4597 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4598 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4599 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4600 
4601 	fn_table.submit_request = stub_submit_request;
4602 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4603 
4604 	/* Case 1: First test the request won't be split */
4605 	num_blocks = 32;
4606 
4607 	g_io_done = false;
4608 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
4609 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4610 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4611 	CU_ASSERT_EQUAL(rc, 0);
4612 	CU_ASSERT(g_io_done == false);
4613 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4614 	stub_complete_io(1);
4615 	CU_ASSERT(g_io_done == true);
4616 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4617 
4618 	/* Case 2: Test the split with 2 children requests */
4619 	max_write_zeroes_blocks = 8;
4620 	bdev->max_write_zeroes = max_write_zeroes_blocks;
4621 	num_blocks = max_write_zeroes_blocks * 2;
4622 	offset = 0;
4623 
4624 	g_io_done = false;
4625 	for (i = 0; i < 2; i++) {
4626 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4627 						   0);
4628 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4629 		offset += max_write_zeroes_blocks;
4630 	}
4631 
4632 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4633 	CU_ASSERT_EQUAL(rc, 0);
4634 	CU_ASSERT(g_io_done == false);
4635 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4636 	stub_complete_io(2);
4637 	CU_ASSERT(g_io_done == true);
4638 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4639 
4640 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4641 	num_children = 15;
4642 	num_blocks = max_write_zeroes_blocks * num_children;
4643 	g_io_done = false;
4644 	offset = 0;
4645 	for (i = 0; i < num_children; i++) {
4646 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4647 						   0);
4648 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4649 		offset += max_write_zeroes_blocks;
4650 	}
4651 
4652 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4653 	CU_ASSERT_EQUAL(rc, 0);
4654 	CU_ASSERT(g_io_done == false);
4655 
4656 	while (num_children > 0) {
4657 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4658 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4659 		stub_complete_io(num_outstanding);
4660 		num_children -= num_outstanding;
4661 	}
4662 	CU_ASSERT(g_io_done == true);
4663 
4664 	spdk_put_io_channel(ioch);
4665 	spdk_bdev_close(desc);
4666 	free_bdev(bdev);
4667 	spdk_bdev_finish(bdev_fini_cb, NULL);
4668 	poll_threads();
4669 }
4670 
4671 static void
4672 bdev_set_options_test(void)
4673 {
4674 	struct spdk_bdev_opts bdev_opts = {};
4675 	int rc;
4676 
4677 	/* Case1: Do not set opts_size */
4678 	rc = spdk_bdev_set_opts(&bdev_opts);
4679 	CU_ASSERT(rc == -1);
4680 
4681 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4682 	bdev_opts.bdev_io_pool_size = 4;
4683 	bdev_opts.bdev_io_cache_size = 2;
4684 	bdev_opts.small_buf_pool_size = 4;
4685 
4686 	/* Case 2: Do not set valid small_buf_pool_size and large_buf_pool_size */
4687 	rc = spdk_bdev_set_opts(&bdev_opts);
4688 	CU_ASSERT(rc == -1);
4689 
4690 	/* Case 3: Do not set valid large_buf_pool_size */
4691 	bdev_opts.small_buf_pool_size = BUF_SMALL_POOL_SIZE;
4692 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE - 1;
4693 	rc = spdk_bdev_set_opts(&bdev_opts);
4694 	CU_ASSERT(rc == -1);
4695 
4696 	/* Case4: set valid large buf_pool_size */
4697 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE;
4698 	rc = spdk_bdev_set_opts(&bdev_opts);
4699 	CU_ASSERT(rc == 0);
4700 
4701 	/* Case5: Set different valid value for small and large buf pool */
4702 	bdev_opts.large_buf_pool_size = BUF_SMALL_POOL_SIZE + 3;
4703 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE + 3;
4704 	rc = spdk_bdev_set_opts(&bdev_opts);
4705 	CU_ASSERT(rc == 0);
4706 }
4707 
4708 static uint64_t
4709 get_ns_time(void)
4710 {
4711 	int rc;
4712 	struct timespec ts;
4713 
4714 	rc = clock_gettime(CLOCK_MONOTONIC, &ts);
4715 	CU_ASSERT(rc == 0);
4716 	return ts.tv_sec * 1000 * 1000 * 1000 + ts.tv_nsec;
4717 }
4718 
4719 static int
4720 rb_tree_get_height(struct spdk_bdev_name *bdev_name)
4721 {
4722 	int h1, h2;
4723 
4724 	if (bdev_name == NULL) {
4725 		return -1;
4726 	} else {
4727 		h1 = rb_tree_get_height(RB_LEFT(bdev_name, node));
4728 		h2 = rb_tree_get_height(RB_RIGHT(bdev_name, node));
4729 
4730 		return spdk_max(h1, h2) + 1;
4731 	}
4732 }
4733 
4734 static void
4735 bdev_multi_allocation(void)
4736 {
4737 	const int max_bdev_num = 1024 * 16;
4738 	char name[max_bdev_num][10];
4739 	char noexist_name[] = "invalid_bdev";
4740 	struct spdk_bdev *bdev[max_bdev_num];
4741 	int i, j;
4742 	uint64_t last_time;
4743 	int bdev_num;
4744 	int height;
4745 
4746 	for (j = 0; j < max_bdev_num; j++) {
4747 		snprintf(name[j], sizeof(name[j]), "bdev%d", j);
4748 	}
4749 
4750 	for (i = 0; i < 16; i++) {
4751 		last_time = get_ns_time();
4752 		bdev_num = 1024 * (i + 1);
4753 		for (j = 0; j < bdev_num; j++) {
4754 			bdev[j] = allocate_bdev(name[j]);
4755 			height = rb_tree_get_height(&bdev[j]->internal.bdev_name);
4756 			CU_ASSERT(height <= (int)(spdk_u32log2(j + 1)));
4757 		}
4758 		SPDK_NOTICELOG("alloc bdev num %d takes %" PRIu64 " ms\n", bdev_num,
4759 			       (get_ns_time() - last_time) / 1000 / 1000);
4760 		for (j = 0; j < bdev_num; j++) {
4761 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) != NULL);
4762 		}
4763 		CU_ASSERT(spdk_bdev_get_by_name(noexist_name) == NULL);
4764 
4765 		for (j = 0; j < bdev_num; j++) {
4766 			free_bdev(bdev[j]);
4767 		}
4768 		for (j = 0; j < bdev_num; j++) {
4769 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) == NULL);
4770 		}
4771 	}
4772 }
4773 
4774 int
4775 main(int argc, char **argv)
4776 {
4777 	CU_pSuite		suite = NULL;
4778 	unsigned int		num_failures;
4779 
4780 	CU_set_error_action(CUEA_ABORT);
4781 	CU_initialize_registry();
4782 
4783 	suite = CU_add_suite("bdev", null_init, null_clean);
4784 
4785 	CU_ADD_TEST(suite, bytes_to_blocks_test);
4786 	CU_ADD_TEST(suite, num_blocks_test);
4787 	CU_ADD_TEST(suite, io_valid_test);
4788 	CU_ADD_TEST(suite, open_write_test);
4789 	CU_ADD_TEST(suite, alias_add_del_test);
4790 	CU_ADD_TEST(suite, get_device_stat_test);
4791 	CU_ADD_TEST(suite, bdev_io_types_test);
4792 	CU_ADD_TEST(suite, bdev_io_wait_test);
4793 	CU_ADD_TEST(suite, bdev_io_spans_split_test);
4794 	CU_ADD_TEST(suite, bdev_io_boundary_split_test);
4795 	CU_ADD_TEST(suite, bdev_io_max_size_and_segment_split_test);
4796 	CU_ADD_TEST(suite, bdev_io_mix_split_test);
4797 	CU_ADD_TEST(suite, bdev_io_split_with_io_wait);
4798 	CU_ADD_TEST(suite, bdev_io_alignment_with_boundary);
4799 	CU_ADD_TEST(suite, bdev_io_alignment);
4800 	CU_ADD_TEST(suite, bdev_histograms);
4801 	CU_ADD_TEST(suite, bdev_write_zeroes);
4802 	CU_ADD_TEST(suite, bdev_compare_and_write);
4803 	CU_ADD_TEST(suite, bdev_compare);
4804 	CU_ADD_TEST(suite, bdev_zcopy_write);
4805 	CU_ADD_TEST(suite, bdev_zcopy_read);
4806 	CU_ADD_TEST(suite, bdev_open_while_hotremove);
4807 	CU_ADD_TEST(suite, bdev_close_while_hotremove);
4808 	CU_ADD_TEST(suite, bdev_open_ext);
4809 	CU_ADD_TEST(suite, bdev_set_io_timeout);
4810 	CU_ADD_TEST(suite, lba_range_overlap);
4811 	CU_ADD_TEST(suite, lock_lba_range_check_ranges);
4812 	CU_ADD_TEST(suite, lock_lba_range_with_io_outstanding);
4813 	CU_ADD_TEST(suite, lock_lba_range_overlapped);
4814 	CU_ADD_TEST(suite, bdev_io_abort);
4815 	CU_ADD_TEST(suite, bdev_unmap);
4816 	CU_ADD_TEST(suite, bdev_write_zeroes_split_test);
4817 	CU_ADD_TEST(suite, bdev_set_options_test);
4818 	CU_ADD_TEST(suite, bdev_multi_allocation);
4819 
4820 	allocate_cores(1);
4821 	allocate_threads(1);
4822 	set_thread(0);
4823 
4824 	CU_basic_set_mode(CU_BRM_VERBOSE);
4825 	CU_basic_run_tests();
4826 	num_failures = CU_get_number_of_failures();
4827 	CU_cleanup_registry();
4828 
4829 	free_threads();
4830 	free_cores();
4831 
4832 	return num_failures;
4833 }
4834