xref: /spdk/test/unit/lib/bdev/bdev.c/bdev_ut.c (revision c20dd8afeedf9b956451d9fe0f66de6c022137bd)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
6  *   Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "spdk_cunit.h"
36 
37 #include "common/lib/ut_multithread.c"
38 #include "unit/lib/json_mock.c"
39 
40 #include "spdk/config.h"
41 /* HACK: disable VTune integration so the unit test doesn't need VTune headers and libs to build */
42 #undef SPDK_CONFIG_VTUNE
43 
44 #include "bdev/bdev.c"
45 
46 DEFINE_STUB(spdk_notify_send, uint64_t, (const char *type, const char *ctx), 0);
47 DEFINE_STUB(spdk_notify_type_register, struct spdk_notify_type *, (const char *type), NULL);
48 
49 int g_status;
50 int g_count;
51 enum spdk_bdev_event_type g_event_type1;
52 enum spdk_bdev_event_type g_event_type2;
53 struct spdk_histogram_data *g_histogram;
54 void *g_unregister_arg;
55 int g_unregister_rc;
56 
57 void
58 spdk_scsi_nvme_translate(const struct spdk_bdev_io *bdev_io,
59 			 int *sc, int *sk, int *asc, int *ascq)
60 {
61 }
62 
63 static int
64 null_init(void)
65 {
66 	return 0;
67 }
68 
69 static int
70 null_clean(void)
71 {
72 	return 0;
73 }
74 
75 static int
76 stub_destruct(void *ctx)
77 {
78 	return 0;
79 }
80 
81 struct ut_expected_io {
82 	uint8_t				type;
83 	uint64_t			offset;
84 	uint64_t			length;
85 	int				iovcnt;
86 	struct iovec			iov[BDEV_IO_NUM_CHILD_IOV];
87 	void				*md_buf;
88 	struct spdk_bdev_ext_io_opts	*ext_io_opts;
89 	TAILQ_ENTRY(ut_expected_io)	link;
90 };
91 
92 struct bdev_ut_channel {
93 	TAILQ_HEAD(, spdk_bdev_io)	outstanding_io;
94 	uint32_t			outstanding_io_count;
95 	TAILQ_HEAD(, ut_expected_io)	expected_io;
96 };
97 
98 static bool g_io_done;
99 static struct spdk_bdev_io *g_bdev_io;
100 static enum spdk_bdev_io_status g_io_status;
101 static enum spdk_bdev_io_status g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
102 static uint32_t g_bdev_ut_io_device;
103 static struct bdev_ut_channel *g_bdev_ut_channel;
104 static void *g_compare_read_buf;
105 static uint32_t g_compare_read_buf_len;
106 static void *g_compare_write_buf;
107 static uint32_t g_compare_write_buf_len;
108 static bool g_abort_done;
109 static enum spdk_bdev_io_status g_abort_status;
110 static void *g_zcopy_read_buf;
111 static uint32_t g_zcopy_read_buf_len;
112 static void *g_zcopy_write_buf;
113 static uint32_t g_zcopy_write_buf_len;
114 static struct spdk_bdev_io *g_zcopy_bdev_io;
115 
116 static struct ut_expected_io *
117 ut_alloc_expected_io(uint8_t type, uint64_t offset, uint64_t length, int iovcnt)
118 {
119 	struct ut_expected_io *expected_io;
120 
121 	expected_io = calloc(1, sizeof(*expected_io));
122 	SPDK_CU_ASSERT_FATAL(expected_io != NULL);
123 
124 	expected_io->type = type;
125 	expected_io->offset = offset;
126 	expected_io->length = length;
127 	expected_io->iovcnt = iovcnt;
128 
129 	return expected_io;
130 }
131 
132 static void
133 ut_expected_io_set_iov(struct ut_expected_io *expected_io, int pos, void *base, size_t len)
134 {
135 	expected_io->iov[pos].iov_base = base;
136 	expected_io->iov[pos].iov_len = len;
137 }
138 
139 static void
140 stub_submit_request(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
141 {
142 	struct bdev_ut_channel *ch = spdk_io_channel_get_ctx(_ch);
143 	struct ut_expected_io *expected_io;
144 	struct iovec *iov, *expected_iov;
145 	struct spdk_bdev_io *bio_to_abort;
146 	int i;
147 
148 	g_bdev_io = bdev_io;
149 
150 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
151 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
152 
153 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
154 		CU_ASSERT(g_compare_read_buf_len == len);
155 		memcpy(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len);
156 	}
157 
158 	if (g_compare_write_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
159 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
160 
161 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
162 		CU_ASSERT(g_compare_write_buf_len == len);
163 		memcpy(g_compare_write_buf, bdev_io->u.bdev.iovs[0].iov_base, len);
164 	}
165 
166 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_COMPARE) {
167 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
168 
169 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
170 		CU_ASSERT(g_compare_read_buf_len == len);
171 		if (memcmp(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len)) {
172 			g_io_exp_status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
173 		}
174 	}
175 
176 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) {
177 		if (g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS) {
178 			TAILQ_FOREACH(bio_to_abort, &ch->outstanding_io, module_link) {
179 				if (bio_to_abort == bdev_io->u.abort.bio_to_abort) {
180 					TAILQ_REMOVE(&ch->outstanding_io, bio_to_abort, module_link);
181 					ch->outstanding_io_count--;
182 					spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_FAILED);
183 					break;
184 				}
185 			}
186 		}
187 	}
188 
189 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) {
190 		if (bdev_io->u.bdev.zcopy.start) {
191 			g_zcopy_bdev_io = bdev_io;
192 			if (bdev_io->u.bdev.zcopy.populate) {
193 				/* Start of a read */
194 				CU_ASSERT(g_zcopy_read_buf != NULL);
195 				CU_ASSERT(g_zcopy_read_buf_len > 0);
196 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_read_buf;
197 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_read_buf_len;
198 				bdev_io->u.bdev.iovcnt = 1;
199 			} else {
200 				/* Start of a write */
201 				CU_ASSERT(g_zcopy_write_buf != NULL);
202 				CU_ASSERT(g_zcopy_write_buf_len > 0);
203 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_write_buf;
204 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_write_buf_len;
205 				bdev_io->u.bdev.iovcnt = 1;
206 			}
207 		} else {
208 			if (bdev_io->u.bdev.zcopy.commit) {
209 				/* End of write */
210 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_write_buf);
211 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_write_buf_len);
212 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
213 				g_zcopy_write_buf = NULL;
214 				g_zcopy_write_buf_len = 0;
215 			} else {
216 				/* End of read */
217 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_read_buf);
218 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_read_buf_len);
219 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
220 				g_zcopy_read_buf = NULL;
221 				g_zcopy_read_buf_len = 0;
222 			}
223 		}
224 	}
225 
226 	TAILQ_INSERT_TAIL(&ch->outstanding_io, bdev_io, module_link);
227 	ch->outstanding_io_count++;
228 
229 	expected_io = TAILQ_FIRST(&ch->expected_io);
230 	if (expected_io == NULL) {
231 		return;
232 	}
233 	TAILQ_REMOVE(&ch->expected_io, expected_io, link);
234 
235 	if (expected_io->type != SPDK_BDEV_IO_TYPE_INVALID) {
236 		CU_ASSERT(bdev_io->type == expected_io->type);
237 	}
238 
239 	if (expected_io->md_buf != NULL) {
240 		CU_ASSERT(expected_io->md_buf == bdev_io->u.bdev.md_buf);
241 	}
242 
243 	if (expected_io->length == 0) {
244 		free(expected_io);
245 		return;
246 	}
247 
248 	CU_ASSERT(expected_io->offset == bdev_io->u.bdev.offset_blocks);
249 	CU_ASSERT(expected_io->length = bdev_io->u.bdev.num_blocks);
250 
251 	if (expected_io->iovcnt == 0) {
252 		free(expected_io);
253 		/* UNMAP, WRITE_ZEROES and FLUSH don't have iovs, so we can just return now. */
254 		return;
255 	}
256 
257 	CU_ASSERT(expected_io->iovcnt == bdev_io->u.bdev.iovcnt);
258 	for (i = 0; i < expected_io->iovcnt; i++) {
259 		expected_iov = &expected_io->iov[i];
260 		if (bdev_io->internal.orig_iovcnt == 0) {
261 			iov = &bdev_io->u.bdev.iovs[i];
262 		} else {
263 			iov = bdev_io->internal.orig_iovs;
264 		}
265 		CU_ASSERT(iov->iov_len == expected_iov->iov_len);
266 		CU_ASSERT(iov->iov_base == expected_iov->iov_base);
267 	}
268 
269 	if (expected_io->ext_io_opts) {
270 		CU_ASSERT(expected_io->ext_io_opts == bdev_io->internal.ext_opts)
271 	}
272 
273 	free(expected_io);
274 }
275 
276 static void
277 stub_submit_request_get_buf_cb(struct spdk_io_channel *_ch,
278 			       struct spdk_bdev_io *bdev_io, bool success)
279 {
280 	CU_ASSERT(success == true);
281 
282 	stub_submit_request(_ch, bdev_io);
283 }
284 
285 static void
286 stub_submit_request_get_buf(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
287 {
288 	spdk_bdev_io_get_buf(bdev_io, stub_submit_request_get_buf_cb,
289 			     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
290 }
291 
292 static uint32_t
293 stub_complete_io(uint32_t num_to_complete)
294 {
295 	struct bdev_ut_channel *ch = g_bdev_ut_channel;
296 	struct spdk_bdev_io *bdev_io;
297 	static enum spdk_bdev_io_status io_status;
298 	uint32_t num_completed = 0;
299 
300 	while (num_completed < num_to_complete) {
301 		if (TAILQ_EMPTY(&ch->outstanding_io)) {
302 			break;
303 		}
304 		bdev_io = TAILQ_FIRST(&ch->outstanding_io);
305 		TAILQ_REMOVE(&ch->outstanding_io, bdev_io, module_link);
306 		ch->outstanding_io_count--;
307 		io_status = g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS ? SPDK_BDEV_IO_STATUS_SUCCESS :
308 			    g_io_exp_status;
309 		spdk_bdev_io_complete(bdev_io, io_status);
310 		num_completed++;
311 	}
312 
313 	return num_completed;
314 }
315 
316 static struct spdk_io_channel *
317 bdev_ut_get_io_channel(void *ctx)
318 {
319 	return spdk_get_io_channel(&g_bdev_ut_io_device);
320 }
321 
322 static bool g_io_types_supported[SPDK_BDEV_NUM_IO_TYPES] = {
323 	[SPDK_BDEV_IO_TYPE_READ]		= true,
324 	[SPDK_BDEV_IO_TYPE_WRITE]		= true,
325 	[SPDK_BDEV_IO_TYPE_COMPARE]		= true,
326 	[SPDK_BDEV_IO_TYPE_UNMAP]		= true,
327 	[SPDK_BDEV_IO_TYPE_FLUSH]		= true,
328 	[SPDK_BDEV_IO_TYPE_RESET]		= true,
329 	[SPDK_BDEV_IO_TYPE_NVME_ADMIN]		= true,
330 	[SPDK_BDEV_IO_TYPE_NVME_IO]		= true,
331 	[SPDK_BDEV_IO_TYPE_NVME_IO_MD]		= true,
332 	[SPDK_BDEV_IO_TYPE_WRITE_ZEROES]	= true,
333 	[SPDK_BDEV_IO_TYPE_ZCOPY]		= true,
334 	[SPDK_BDEV_IO_TYPE_ABORT]		= true,
335 };
336 
337 static void
338 ut_enable_io_type(enum spdk_bdev_io_type io_type, bool enable)
339 {
340 	g_io_types_supported[io_type] = enable;
341 }
342 
343 static bool
344 stub_io_type_supported(void *_bdev, enum spdk_bdev_io_type io_type)
345 {
346 	return g_io_types_supported[io_type];
347 }
348 
349 static struct spdk_bdev_fn_table fn_table = {
350 	.destruct = stub_destruct,
351 	.submit_request = stub_submit_request,
352 	.get_io_channel = bdev_ut_get_io_channel,
353 	.io_type_supported = stub_io_type_supported,
354 };
355 
356 static int
357 bdev_ut_create_ch(void *io_device, void *ctx_buf)
358 {
359 	struct bdev_ut_channel *ch = ctx_buf;
360 
361 	CU_ASSERT(g_bdev_ut_channel == NULL);
362 	g_bdev_ut_channel = ch;
363 
364 	TAILQ_INIT(&ch->outstanding_io);
365 	ch->outstanding_io_count = 0;
366 	TAILQ_INIT(&ch->expected_io);
367 	return 0;
368 }
369 
370 static void
371 bdev_ut_destroy_ch(void *io_device, void *ctx_buf)
372 {
373 	CU_ASSERT(g_bdev_ut_channel != NULL);
374 	g_bdev_ut_channel = NULL;
375 }
376 
377 struct spdk_bdev_module bdev_ut_if;
378 
379 static int
380 bdev_ut_module_init(void)
381 {
382 	spdk_io_device_register(&g_bdev_ut_io_device, bdev_ut_create_ch, bdev_ut_destroy_ch,
383 				sizeof(struct bdev_ut_channel), NULL);
384 	spdk_bdev_module_init_done(&bdev_ut_if);
385 	return 0;
386 }
387 
388 static void
389 bdev_ut_module_fini(void)
390 {
391 	spdk_io_device_unregister(&g_bdev_ut_io_device, NULL);
392 }
393 
394 struct spdk_bdev_module bdev_ut_if = {
395 	.name = "bdev_ut",
396 	.module_init = bdev_ut_module_init,
397 	.module_fini = bdev_ut_module_fini,
398 	.async_init = true,
399 };
400 
401 static void vbdev_ut_examine(struct spdk_bdev *bdev);
402 
403 static int
404 vbdev_ut_module_init(void)
405 {
406 	return 0;
407 }
408 
409 static void
410 vbdev_ut_module_fini(void)
411 {
412 }
413 
414 struct spdk_bdev_module vbdev_ut_if = {
415 	.name = "vbdev_ut",
416 	.module_init = vbdev_ut_module_init,
417 	.module_fini = vbdev_ut_module_fini,
418 	.examine_config = vbdev_ut_examine,
419 };
420 
421 SPDK_BDEV_MODULE_REGISTER(bdev_ut, &bdev_ut_if)
422 SPDK_BDEV_MODULE_REGISTER(vbdev_ut, &vbdev_ut_if)
423 
424 static void
425 vbdev_ut_examine(struct spdk_bdev *bdev)
426 {
427 	spdk_bdev_module_examine_done(&vbdev_ut_if);
428 }
429 
430 static struct spdk_bdev *
431 allocate_bdev(char *name)
432 {
433 	struct spdk_bdev *bdev;
434 	int rc;
435 
436 	bdev = calloc(1, sizeof(*bdev));
437 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
438 
439 	bdev->name = name;
440 	bdev->fn_table = &fn_table;
441 	bdev->module = &bdev_ut_if;
442 	bdev->blockcnt = 1024;
443 	bdev->blocklen = 512;
444 
445 	rc = spdk_bdev_register(bdev);
446 	CU_ASSERT(rc == 0);
447 
448 	return bdev;
449 }
450 
451 static struct spdk_bdev *
452 allocate_vbdev(char *name)
453 {
454 	struct spdk_bdev *bdev;
455 	int rc;
456 
457 	bdev = calloc(1, sizeof(*bdev));
458 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
459 
460 	bdev->name = name;
461 	bdev->fn_table = &fn_table;
462 	bdev->module = &vbdev_ut_if;
463 
464 	rc = spdk_bdev_register(bdev);
465 	CU_ASSERT(rc == 0);
466 
467 	return bdev;
468 }
469 
470 static void
471 free_bdev(struct spdk_bdev *bdev)
472 {
473 	spdk_bdev_unregister(bdev, NULL, NULL);
474 	poll_threads();
475 	memset(bdev, 0xFF, sizeof(*bdev));
476 	free(bdev);
477 }
478 
479 static void
480 free_vbdev(struct spdk_bdev *bdev)
481 {
482 	spdk_bdev_unregister(bdev, NULL, NULL);
483 	poll_threads();
484 	memset(bdev, 0xFF, sizeof(*bdev));
485 	free(bdev);
486 }
487 
488 static void
489 get_device_stat_cb(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat, void *cb_arg, int rc)
490 {
491 	const char *bdev_name;
492 
493 	CU_ASSERT(bdev != NULL);
494 	CU_ASSERT(rc == 0);
495 	bdev_name = spdk_bdev_get_name(bdev);
496 	CU_ASSERT_STRING_EQUAL(bdev_name, "bdev0");
497 
498 	free(stat);
499 
500 	*(bool *)cb_arg = true;
501 }
502 
503 static void
504 bdev_unregister_cb(void *cb_arg, int rc)
505 {
506 	g_unregister_arg = cb_arg;
507 	g_unregister_rc = rc;
508 }
509 
510 static void
511 bdev_ut_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
512 {
513 }
514 
515 static void
516 bdev_open_cb1(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
517 {
518 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
519 
520 	g_event_type1 = type;
521 	if (SPDK_BDEV_EVENT_REMOVE == type) {
522 		spdk_bdev_close(desc);
523 	}
524 }
525 
526 static void
527 bdev_open_cb2(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
528 {
529 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
530 
531 	g_event_type2 = type;
532 	if (SPDK_BDEV_EVENT_REMOVE == type) {
533 		spdk_bdev_close(desc);
534 	}
535 }
536 
537 static void
538 get_device_stat_test(void)
539 {
540 	struct spdk_bdev *bdev;
541 	struct spdk_bdev_io_stat *stat;
542 	bool done;
543 
544 	bdev = allocate_bdev("bdev0");
545 	stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
546 	if (stat == NULL) {
547 		free_bdev(bdev);
548 		return;
549 	}
550 
551 	done = false;
552 	spdk_bdev_get_device_stat(bdev, stat, get_device_stat_cb, &done);
553 	while (!done) { poll_threads(); }
554 
555 	free_bdev(bdev);
556 }
557 
558 static void
559 open_write_test(void)
560 {
561 	struct spdk_bdev *bdev[9];
562 	struct spdk_bdev_desc *desc[9] = {};
563 	int rc;
564 
565 	/*
566 	 * Create a tree of bdevs to test various open w/ write cases.
567 	 *
568 	 * bdev0 through bdev3 are physical block devices, such as NVMe
569 	 * namespaces or Ceph block devices.
570 	 *
571 	 * bdev4 is a virtual bdev with multiple base bdevs.  This models
572 	 * caching or RAID use cases.
573 	 *
574 	 * bdev5 through bdev7 are all virtual bdevs with the same base
575 	 * bdev (except bdev7). This models partitioning or logical volume
576 	 * use cases.
577 	 *
578 	 * bdev7 is a virtual bdev with multiple base bdevs. One of base bdevs
579 	 * (bdev2) is shared with other virtual bdevs: bdev5 and bdev6. This
580 	 * models caching, RAID, partitioning or logical volumes use cases.
581 	 *
582 	 * bdev8 is a virtual bdev with multiple base bdevs, but these
583 	 * base bdevs are themselves virtual bdevs.
584 	 *
585 	 *                bdev8
586 	 *                  |
587 	 *            +----------+
588 	 *            |          |
589 	 *          bdev4      bdev5   bdev6   bdev7
590 	 *            |          |       |       |
591 	 *        +---+---+      +---+   +   +---+---+
592 	 *        |       |           \  |  /         \
593 	 *      bdev0   bdev1          bdev2         bdev3
594 	 */
595 
596 	bdev[0] = allocate_bdev("bdev0");
597 	rc = spdk_bdev_module_claim_bdev(bdev[0], NULL, &bdev_ut_if);
598 	CU_ASSERT(rc == 0);
599 
600 	bdev[1] = allocate_bdev("bdev1");
601 	rc = spdk_bdev_module_claim_bdev(bdev[1], NULL, &bdev_ut_if);
602 	CU_ASSERT(rc == 0);
603 
604 	bdev[2] = allocate_bdev("bdev2");
605 	rc = spdk_bdev_module_claim_bdev(bdev[2], NULL, &bdev_ut_if);
606 	CU_ASSERT(rc == 0);
607 
608 	bdev[3] = allocate_bdev("bdev3");
609 	rc = spdk_bdev_module_claim_bdev(bdev[3], NULL, &bdev_ut_if);
610 	CU_ASSERT(rc == 0);
611 
612 	bdev[4] = allocate_vbdev("bdev4");
613 	rc = spdk_bdev_module_claim_bdev(bdev[4], NULL, &bdev_ut_if);
614 	CU_ASSERT(rc == 0);
615 
616 	bdev[5] = allocate_vbdev("bdev5");
617 	rc = spdk_bdev_module_claim_bdev(bdev[5], NULL, &bdev_ut_if);
618 	CU_ASSERT(rc == 0);
619 
620 	bdev[6] = allocate_vbdev("bdev6");
621 
622 	bdev[7] = allocate_vbdev("bdev7");
623 
624 	bdev[8] = allocate_vbdev("bdev8");
625 
626 	/* Open bdev0 read-only.  This should succeed. */
627 	rc = spdk_bdev_open_ext("bdev0", false, bdev_ut_event_cb, NULL, &desc[0]);
628 	CU_ASSERT(rc == 0);
629 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
630 	CU_ASSERT(bdev[0] == spdk_bdev_desc_get_bdev(desc[0]));
631 	spdk_bdev_close(desc[0]);
632 
633 	/*
634 	 * Open bdev1 read/write.  This should fail since bdev1 has been claimed
635 	 * by a vbdev module.
636 	 */
637 	rc = spdk_bdev_open_ext("bdev1", true, bdev_ut_event_cb, NULL, &desc[1]);
638 	CU_ASSERT(rc == -EPERM);
639 
640 	/*
641 	 * Open bdev4 read/write.  This should fail since bdev3 has been claimed
642 	 * by a vbdev module.
643 	 */
644 	rc = spdk_bdev_open_ext("bdev4", true, bdev_ut_event_cb, NULL, &desc[4]);
645 	CU_ASSERT(rc == -EPERM);
646 
647 	/* Open bdev4 read-only.  This should succeed. */
648 	rc = spdk_bdev_open_ext("bdev4", false, bdev_ut_event_cb, NULL, &desc[4]);
649 	CU_ASSERT(rc == 0);
650 	SPDK_CU_ASSERT_FATAL(desc[4] != NULL);
651 	CU_ASSERT(bdev[4] == spdk_bdev_desc_get_bdev(desc[4]));
652 	spdk_bdev_close(desc[4]);
653 
654 	/*
655 	 * Open bdev8 read/write.  This should succeed since it is a leaf
656 	 * bdev.
657 	 */
658 	rc = spdk_bdev_open_ext("bdev8", true, bdev_ut_event_cb, NULL, &desc[8]);
659 	CU_ASSERT(rc == 0);
660 	SPDK_CU_ASSERT_FATAL(desc[8] != NULL);
661 	CU_ASSERT(bdev[8] == spdk_bdev_desc_get_bdev(desc[8]));
662 	spdk_bdev_close(desc[8]);
663 
664 	/*
665 	 * Open bdev5 read/write.  This should fail since bdev4 has been claimed
666 	 * by a vbdev module.
667 	 */
668 	rc = spdk_bdev_open_ext("bdev5", true, bdev_ut_event_cb, NULL, &desc[5]);
669 	CU_ASSERT(rc == -EPERM);
670 
671 	/* Open bdev4 read-only.  This should succeed. */
672 	rc = spdk_bdev_open_ext("bdev5", false, bdev_ut_event_cb, NULL, &desc[5]);
673 	CU_ASSERT(rc == 0);
674 	SPDK_CU_ASSERT_FATAL(desc[5] != NULL);
675 	CU_ASSERT(bdev[5] == spdk_bdev_desc_get_bdev(desc[5]));
676 	spdk_bdev_close(desc[5]);
677 
678 	free_vbdev(bdev[8]);
679 
680 	free_vbdev(bdev[5]);
681 	free_vbdev(bdev[6]);
682 	free_vbdev(bdev[7]);
683 
684 	free_vbdev(bdev[4]);
685 
686 	free_bdev(bdev[0]);
687 	free_bdev(bdev[1]);
688 	free_bdev(bdev[2]);
689 	free_bdev(bdev[3]);
690 }
691 
692 static void
693 claim_test(void)
694 {
695 	struct spdk_bdev *bdev;
696 	struct spdk_bdev_desc *desc, *open_desc;
697 	int rc;
698 	uint32_t count;
699 
700 	/*
701 	 * A vbdev that uses a read-only bdev may need it to remain read-only.
702 	 * To do so, it opens the bdev read-only, then claims it without
703 	 * passing a spdk_bdev_desc.
704 	 */
705 	bdev = allocate_bdev("bdev0");
706 	rc = spdk_bdev_open_ext("bdev0", false, bdev_ut_event_cb, NULL, &desc);
707 	CU_ASSERT(rc == 0);
708 	CU_ASSERT(desc->write == false);
709 
710 	rc = spdk_bdev_module_claim_bdev(bdev, NULL, &bdev_ut_if);
711 	CU_ASSERT(rc == 0);
712 	CU_ASSERT(bdev->internal.claim_module == &bdev_ut_if);
713 
714 	/* There should be only one open descriptor and it should still be ro */
715 	count = 0;
716 	TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
717 		CU_ASSERT(open_desc == desc);
718 		CU_ASSERT(!open_desc->write);
719 		count++;
720 	}
721 	CU_ASSERT(count == 1);
722 
723 	/* A read-only bdev is upgraded to read-write if desc is passed. */
724 	spdk_bdev_module_release_bdev(bdev);
725 	rc = spdk_bdev_module_claim_bdev(bdev, desc, &bdev_ut_if);
726 	CU_ASSERT(rc == 0);
727 	CU_ASSERT(bdev->internal.claim_module == &bdev_ut_if);
728 
729 	/* There should be only one open descriptor and it should be rw */
730 	count = 0;
731 	TAILQ_FOREACH(open_desc, &bdev->internal.open_descs, link) {
732 		CU_ASSERT(open_desc == desc);
733 		CU_ASSERT(open_desc->write);
734 		count++;
735 	}
736 	CU_ASSERT(count == 1);
737 
738 	spdk_bdev_close(desc);
739 	free_bdev(bdev);
740 }
741 
742 static void
743 bytes_to_blocks_test(void)
744 {
745 	struct spdk_bdev bdev;
746 	uint64_t offset_blocks, num_blocks;
747 
748 	memset(&bdev, 0, sizeof(bdev));
749 
750 	bdev.blocklen = 512;
751 
752 	/* All parameters valid */
753 	offset_blocks = 0;
754 	num_blocks = 0;
755 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 1024, &num_blocks) == 0);
756 	CU_ASSERT(offset_blocks == 1);
757 	CU_ASSERT(num_blocks == 2);
758 
759 	/* Offset not a block multiple */
760 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 512, &num_blocks) != 0);
761 
762 	/* Length not a block multiple */
763 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 3, &num_blocks) != 0);
764 
765 	/* In case blocklen not the power of two */
766 	bdev.blocklen = 100;
767 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 200, &num_blocks) == 0);
768 	CU_ASSERT(offset_blocks == 1);
769 	CU_ASSERT(num_blocks == 2);
770 
771 	/* Offset not a block multiple */
772 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 100, &num_blocks) != 0);
773 
774 	/* Length not a block multiple */
775 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 3, &num_blocks) != 0);
776 }
777 
778 static void
779 num_blocks_test(void)
780 {
781 	struct spdk_bdev bdev;
782 	struct spdk_bdev_desc *desc = NULL;
783 	int rc;
784 
785 	memset(&bdev, 0, sizeof(bdev));
786 	bdev.name = "num_blocks";
787 	bdev.fn_table = &fn_table;
788 	bdev.module = &bdev_ut_if;
789 	spdk_bdev_register(&bdev);
790 	spdk_bdev_notify_blockcnt_change(&bdev, 50);
791 
792 	/* Growing block number */
793 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 70) == 0);
794 	/* Shrinking block number */
795 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 30) == 0);
796 
797 	rc = spdk_bdev_open_ext("num_blocks", false, bdev_open_cb1, &desc, &desc);
798 	CU_ASSERT(rc == 0);
799 	SPDK_CU_ASSERT_FATAL(desc != NULL);
800 	CU_ASSERT(&bdev == spdk_bdev_desc_get_bdev(desc));
801 
802 	/* Growing block number */
803 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 80) == 0);
804 	/* Shrinking block number */
805 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 20) != 0);
806 
807 	g_event_type1 = 0xFF;
808 	/* Growing block number */
809 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 90) == 0);
810 
811 	poll_threads();
812 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_RESIZE);
813 
814 	g_event_type1 = 0xFF;
815 	/* Growing block number and closing */
816 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 100) == 0);
817 
818 	spdk_bdev_close(desc);
819 	spdk_bdev_unregister(&bdev, NULL, NULL);
820 
821 	poll_threads();
822 
823 	/* Callback is not called for closed device */
824 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
825 }
826 
827 static void
828 io_valid_test(void)
829 {
830 	struct spdk_bdev bdev;
831 
832 	memset(&bdev, 0, sizeof(bdev));
833 
834 	bdev.blocklen = 512;
835 	CU_ASSERT(pthread_mutex_init(&bdev.internal.mutex, NULL) == 0);
836 
837 	spdk_bdev_notify_blockcnt_change(&bdev, 100);
838 
839 	/* All parameters valid */
840 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 1, 2) == true);
841 
842 	/* Last valid block */
843 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 1) == true);
844 
845 	/* Offset past end of bdev */
846 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 100, 1) == false);
847 
848 	/* Offset + length past end of bdev */
849 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 2) == false);
850 
851 	/* Offset near end of uint64_t range (2^64 - 1) */
852 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 18446744073709551615ULL, 1) == false);
853 
854 	CU_ASSERT(pthread_mutex_destroy(&bdev.internal.mutex) == 0);
855 }
856 
857 static void
858 alias_add_del_test(void)
859 {
860 	struct spdk_bdev *bdev[3];
861 	int rc;
862 
863 	/* Creating and registering bdevs */
864 	bdev[0] = allocate_bdev("bdev0");
865 	SPDK_CU_ASSERT_FATAL(bdev[0] != 0);
866 
867 	bdev[1] = allocate_bdev("bdev1");
868 	SPDK_CU_ASSERT_FATAL(bdev[1] != 0);
869 
870 	bdev[2] = allocate_bdev("bdev2");
871 	SPDK_CU_ASSERT_FATAL(bdev[2] != 0);
872 
873 	poll_threads();
874 
875 	/*
876 	 * Trying adding an alias identical to name.
877 	 * Alias is identical to name, so it can not be added to aliases list
878 	 */
879 	rc = spdk_bdev_alias_add(bdev[0], bdev[0]->name);
880 	CU_ASSERT(rc == -EEXIST);
881 
882 	/*
883 	 * Trying to add empty alias,
884 	 * this one should fail
885 	 */
886 	rc = spdk_bdev_alias_add(bdev[0], NULL);
887 	CU_ASSERT(rc == -EINVAL);
888 
889 	/* Trying adding same alias to two different registered bdevs */
890 
891 	/* Alias is used first time, so this one should pass */
892 	rc = spdk_bdev_alias_add(bdev[0], "proper alias 0");
893 	CU_ASSERT(rc == 0);
894 
895 	/* Alias was added to another bdev, so this one should fail */
896 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 0");
897 	CU_ASSERT(rc == -EEXIST);
898 
899 	/* Alias is used first time, so this one should pass */
900 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 1");
901 	CU_ASSERT(rc == 0);
902 
903 	/* Trying removing an alias from registered bdevs */
904 
905 	/* Alias is not on a bdev aliases list, so this one should fail */
906 	rc = spdk_bdev_alias_del(bdev[0], "not existing");
907 	CU_ASSERT(rc == -ENOENT);
908 
909 	/* Alias is present on a bdev aliases list, so this one should pass */
910 	rc = spdk_bdev_alias_del(bdev[0], "proper alias 0");
911 	CU_ASSERT(rc == 0);
912 
913 	/* Alias is present on a bdev aliases list, so this one should pass */
914 	rc = spdk_bdev_alias_del(bdev[1], "proper alias 1");
915 	CU_ASSERT(rc == 0);
916 
917 	/* Trying to remove name instead of alias, so this one should fail, name cannot be changed or removed */
918 	rc = spdk_bdev_alias_del(bdev[0], bdev[0]->name);
919 	CU_ASSERT(rc != 0);
920 
921 	/* Trying to del all alias from empty alias list */
922 	spdk_bdev_alias_del_all(bdev[2]);
923 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev[2]->aliases));
924 
925 	/* Trying to del all alias from non-empty alias list */
926 	rc = spdk_bdev_alias_add(bdev[2], "alias0");
927 	CU_ASSERT(rc == 0);
928 	rc = spdk_bdev_alias_add(bdev[2], "alias1");
929 	CU_ASSERT(rc == 0);
930 	spdk_bdev_alias_del_all(bdev[2]);
931 	CU_ASSERT(TAILQ_EMPTY(&bdev[2]->aliases));
932 
933 	/* Unregister and free bdevs */
934 	spdk_bdev_unregister(bdev[0], NULL, NULL);
935 	spdk_bdev_unregister(bdev[1], NULL, NULL);
936 	spdk_bdev_unregister(bdev[2], NULL, NULL);
937 
938 	poll_threads();
939 
940 	free(bdev[0]);
941 	free(bdev[1]);
942 	free(bdev[2]);
943 }
944 
945 static void
946 io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
947 {
948 	g_io_done = true;
949 	g_io_status = bdev_io->internal.status;
950 	if ((bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) &&
951 	    (bdev_io->u.bdev.zcopy.start)) {
952 		g_zcopy_bdev_io = bdev_io;
953 	} else {
954 		spdk_bdev_free_io(bdev_io);
955 		g_zcopy_bdev_io = NULL;
956 	}
957 }
958 
959 static void
960 bdev_init_cb(void *arg, int rc)
961 {
962 	CU_ASSERT(rc == 0);
963 }
964 
965 static void
966 bdev_fini_cb(void *arg)
967 {
968 }
969 
970 struct bdev_ut_io_wait_entry {
971 	struct spdk_bdev_io_wait_entry	entry;
972 	struct spdk_io_channel		*io_ch;
973 	struct spdk_bdev_desc		*desc;
974 	bool				submitted;
975 };
976 
977 static void
978 io_wait_cb(void *arg)
979 {
980 	struct bdev_ut_io_wait_entry *entry = arg;
981 	int rc;
982 
983 	rc = spdk_bdev_read_blocks(entry->desc, entry->io_ch, NULL, 0, 1, io_done, NULL);
984 	CU_ASSERT(rc == 0);
985 	entry->submitted = true;
986 }
987 
988 static void
989 bdev_io_types_test(void)
990 {
991 	struct spdk_bdev *bdev;
992 	struct spdk_bdev_desc *desc = NULL;
993 	struct spdk_io_channel *io_ch;
994 	struct spdk_bdev_opts bdev_opts = {};
995 	int rc;
996 
997 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
998 	bdev_opts.bdev_io_pool_size = 4;
999 	bdev_opts.bdev_io_cache_size = 2;
1000 
1001 	rc = spdk_bdev_set_opts(&bdev_opts);
1002 	CU_ASSERT(rc == 0);
1003 	spdk_bdev_initialize(bdev_init_cb, NULL);
1004 	poll_threads();
1005 
1006 	bdev = allocate_bdev("bdev0");
1007 
1008 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1009 	CU_ASSERT(rc == 0);
1010 	poll_threads();
1011 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1012 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1013 	io_ch = spdk_bdev_get_io_channel(desc);
1014 	CU_ASSERT(io_ch != NULL);
1015 
1016 	/* WRITE and WRITE ZEROES are not supported */
1017 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
1018 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, false);
1019 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 0, 128, io_done, NULL);
1020 	CU_ASSERT(rc == -ENOTSUP);
1021 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
1022 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, true);
1023 
1024 	/* NVME_IO, NVME_IO_MD and NVME_ADMIN are not supported */
1025 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO, false);
1026 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO_MD, false);
1027 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_ADMIN, false);
1028 	rc = spdk_bdev_nvme_io_passthru(desc, io_ch, NULL, NULL, 0, NULL, NULL);
1029 	CU_ASSERT(rc == -ENOTSUP);
1030 	rc = spdk_bdev_nvme_io_passthru_md(desc, io_ch, NULL, NULL, 0, NULL, 0, NULL, NULL);
1031 	CU_ASSERT(rc == -ENOTSUP);
1032 	rc = spdk_bdev_nvme_admin_passthru(desc, io_ch, NULL, NULL, 0, NULL, NULL);
1033 	CU_ASSERT(rc == -ENOTSUP);
1034 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO, true);
1035 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO_MD, true);
1036 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_ADMIN, true);
1037 
1038 	spdk_put_io_channel(io_ch);
1039 	spdk_bdev_close(desc);
1040 	free_bdev(bdev);
1041 	spdk_bdev_finish(bdev_fini_cb, NULL);
1042 	poll_threads();
1043 }
1044 
1045 static void
1046 bdev_io_wait_test(void)
1047 {
1048 	struct spdk_bdev *bdev;
1049 	struct spdk_bdev_desc *desc = NULL;
1050 	struct spdk_io_channel *io_ch;
1051 	struct spdk_bdev_opts bdev_opts = {};
1052 	struct bdev_ut_io_wait_entry io_wait_entry;
1053 	struct bdev_ut_io_wait_entry io_wait_entry2;
1054 	int rc;
1055 
1056 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1057 	bdev_opts.bdev_io_pool_size = 4;
1058 	bdev_opts.bdev_io_cache_size = 2;
1059 
1060 	rc = spdk_bdev_set_opts(&bdev_opts);
1061 	CU_ASSERT(rc == 0);
1062 	spdk_bdev_initialize(bdev_init_cb, NULL);
1063 	poll_threads();
1064 
1065 	bdev = allocate_bdev("bdev0");
1066 
1067 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1068 	CU_ASSERT(rc == 0);
1069 	poll_threads();
1070 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1071 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1072 	io_ch = spdk_bdev_get_io_channel(desc);
1073 	CU_ASSERT(io_ch != NULL);
1074 
1075 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1076 	CU_ASSERT(rc == 0);
1077 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1078 	CU_ASSERT(rc == 0);
1079 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1080 	CU_ASSERT(rc == 0);
1081 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1082 	CU_ASSERT(rc == 0);
1083 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1084 
1085 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1086 	CU_ASSERT(rc == -ENOMEM);
1087 
1088 	io_wait_entry.entry.bdev = bdev;
1089 	io_wait_entry.entry.cb_fn = io_wait_cb;
1090 	io_wait_entry.entry.cb_arg = &io_wait_entry;
1091 	io_wait_entry.io_ch = io_ch;
1092 	io_wait_entry.desc = desc;
1093 	io_wait_entry.submitted = false;
1094 	/* Cannot use the same io_wait_entry for two different calls. */
1095 	memcpy(&io_wait_entry2, &io_wait_entry, sizeof(io_wait_entry));
1096 	io_wait_entry2.entry.cb_arg = &io_wait_entry2;
1097 
1098 	/* Queue two I/O waits. */
1099 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry.entry);
1100 	CU_ASSERT(rc == 0);
1101 	CU_ASSERT(io_wait_entry.submitted == false);
1102 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry2.entry);
1103 	CU_ASSERT(rc == 0);
1104 	CU_ASSERT(io_wait_entry2.submitted == false);
1105 
1106 	stub_complete_io(1);
1107 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1108 	CU_ASSERT(io_wait_entry.submitted == true);
1109 	CU_ASSERT(io_wait_entry2.submitted == false);
1110 
1111 	stub_complete_io(1);
1112 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1113 	CU_ASSERT(io_wait_entry2.submitted == true);
1114 
1115 	stub_complete_io(4);
1116 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1117 
1118 	spdk_put_io_channel(io_ch);
1119 	spdk_bdev_close(desc);
1120 	free_bdev(bdev);
1121 	spdk_bdev_finish(bdev_fini_cb, NULL);
1122 	poll_threads();
1123 }
1124 
1125 static void
1126 bdev_io_spans_split_test(void)
1127 {
1128 	struct spdk_bdev bdev;
1129 	struct spdk_bdev_io bdev_io;
1130 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV];
1131 
1132 	memset(&bdev, 0, sizeof(bdev));
1133 	bdev_io.u.bdev.iovs = iov;
1134 
1135 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1136 	bdev.optimal_io_boundary = 0;
1137 	bdev.max_segment_size = 0;
1138 	bdev.max_num_segments = 0;
1139 	bdev_io.bdev = &bdev;
1140 
1141 	/* bdev has no optimal_io_boundary and max_size set - so this should return false. */
1142 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1143 
1144 	bdev.split_on_optimal_io_boundary = true;
1145 	bdev.optimal_io_boundary = 32;
1146 	bdev_io.type = SPDK_BDEV_IO_TYPE_RESET;
1147 
1148 	/* RESETs are not based on LBAs - so this should return false. */
1149 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1150 
1151 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1152 	bdev_io.u.bdev.offset_blocks = 0;
1153 	bdev_io.u.bdev.num_blocks = 32;
1154 
1155 	/* This I/O run right up to, but does not cross, the boundary - so this should return false. */
1156 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1157 
1158 	bdev_io.u.bdev.num_blocks = 33;
1159 
1160 	/* This I/O spans a boundary. */
1161 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1162 
1163 	bdev_io.u.bdev.num_blocks = 32;
1164 	bdev.max_segment_size = 512 * 32;
1165 	bdev.max_num_segments = 1;
1166 	bdev_io.u.bdev.iovcnt = 1;
1167 	iov[0].iov_len = 512;
1168 
1169 	/* Does not cross and exceed max_size or max_segs */
1170 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1171 
1172 	bdev.split_on_optimal_io_boundary = false;
1173 	bdev.max_segment_size = 512;
1174 	bdev.max_num_segments = 1;
1175 	bdev_io.u.bdev.iovcnt = 2;
1176 
1177 	/* Exceed max_segs */
1178 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1179 
1180 	bdev.max_num_segments = 2;
1181 	iov[0].iov_len = 513;
1182 	iov[1].iov_len = 512;
1183 
1184 	/* Exceed max_sizes */
1185 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1186 }
1187 
1188 static void
1189 bdev_io_boundary_split_test(void)
1190 {
1191 	struct spdk_bdev *bdev;
1192 	struct spdk_bdev_desc *desc = NULL;
1193 	struct spdk_io_channel *io_ch;
1194 	struct spdk_bdev_opts bdev_opts = {};
1195 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1196 	struct ut_expected_io *expected_io;
1197 	void *md_buf = (void *)0xFF000000;
1198 	uint64_t i;
1199 	int rc;
1200 
1201 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1202 	bdev_opts.bdev_io_pool_size = 512;
1203 	bdev_opts.bdev_io_cache_size = 64;
1204 
1205 	rc = spdk_bdev_set_opts(&bdev_opts);
1206 	CU_ASSERT(rc == 0);
1207 	spdk_bdev_initialize(bdev_init_cb, NULL);
1208 
1209 	bdev = allocate_bdev("bdev0");
1210 
1211 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1212 	CU_ASSERT(rc == 0);
1213 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1214 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1215 	io_ch = spdk_bdev_get_io_channel(desc);
1216 	CU_ASSERT(io_ch != NULL);
1217 
1218 	bdev->optimal_io_boundary = 16;
1219 	bdev->split_on_optimal_io_boundary = false;
1220 
1221 	g_io_done = false;
1222 
1223 	/* First test that the I/O does not get split if split_on_optimal_io_boundary == false. */
1224 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 8, 1);
1225 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 8 * 512);
1226 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1227 
1228 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
1229 	CU_ASSERT(rc == 0);
1230 	CU_ASSERT(g_io_done == false);
1231 
1232 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1233 	stub_complete_io(1);
1234 	CU_ASSERT(g_io_done == true);
1235 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1236 
1237 	bdev->split_on_optimal_io_boundary = true;
1238 	bdev->md_interleave = false;
1239 	bdev->md_len = 8;
1240 
1241 	/* Now test that a single-vector command is split correctly.
1242 	 * Offset 14, length 8, payload 0xF000
1243 	 *  Child - Offset 14, length 2, payload 0xF000
1244 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
1245 	 *
1246 	 * Set up the expected values before calling spdk_bdev_read_blocks
1247 	 */
1248 	g_io_done = false;
1249 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
1250 	expected_io->md_buf = md_buf;
1251 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
1252 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1253 
1254 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
1255 	expected_io->md_buf = md_buf + 2 * 8;
1256 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
1257 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1258 
1259 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1260 	rc = spdk_bdev_read_blocks_with_md(desc, io_ch, (void *)0xF000, md_buf,
1261 					   14, 8, 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 == 2);
1266 	stub_complete_io(2);
1267 	CU_ASSERT(g_io_done == true);
1268 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1269 
1270 	/* Now set up a more complex, multi-vector command that needs to be split,
1271 	 *  including splitting iovecs.
1272 	 */
1273 	iov[0].iov_base = (void *)0x10000;
1274 	iov[0].iov_len = 512;
1275 	iov[1].iov_base = (void *)0x20000;
1276 	iov[1].iov_len = 20 * 512;
1277 	iov[2].iov_base = (void *)0x30000;
1278 	iov[2].iov_len = 11 * 512;
1279 
1280 	g_io_done = false;
1281 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
1282 	expected_io->md_buf = md_buf;
1283 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
1284 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
1285 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1286 
1287 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
1288 	expected_io->md_buf = md_buf + 2 * 8;
1289 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
1290 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1291 
1292 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
1293 	expected_io->md_buf = md_buf + 18 * 8;
1294 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
1295 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
1296 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1297 
1298 	rc = spdk_bdev_writev_blocks_with_md(desc, io_ch, iov, 3, md_buf,
1299 					     14, 32, io_done, NULL);
1300 	CU_ASSERT(rc == 0);
1301 	CU_ASSERT(g_io_done == false);
1302 
1303 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
1304 	stub_complete_io(3);
1305 	CU_ASSERT(g_io_done == true);
1306 
1307 	/* Test multi vector command that needs to be split by strip and then needs to be
1308 	 * split further due to the capacity of child iovs.
1309 	 */
1310 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
1311 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1312 		iov[i].iov_len = 512;
1313 	}
1314 
1315 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1316 	g_io_done = false;
1317 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1318 					   BDEV_IO_NUM_CHILD_IOV);
1319 	expected_io->md_buf = md_buf;
1320 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1321 		ut_expected_io_set_iov(expected_io, i, (void *)((i + 1) * 0x10000), 512);
1322 	}
1323 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1324 
1325 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1326 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
1327 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1328 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1329 		ut_expected_io_set_iov(expected_io, i,
1330 				       (void *)((i + 1 + BDEV_IO_NUM_CHILD_IOV) * 0x10000), 512);
1331 	}
1332 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1333 
1334 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1335 					    0, BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1336 	CU_ASSERT(rc == 0);
1337 	CU_ASSERT(g_io_done == false);
1338 
1339 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1340 	stub_complete_io(1);
1341 	CU_ASSERT(g_io_done == false);
1342 
1343 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1344 	stub_complete_io(1);
1345 	CU_ASSERT(g_io_done == true);
1346 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1347 
1348 	/* Test multi vector command that needs to be split by strip and then needs to be
1349 	 * split further due to the capacity of child iovs. In this case, the length of
1350 	 * the rest of iovec array with an I/O boundary is the multiple of block size.
1351 	 */
1352 
1353 	/* Fill iovec array for exactly one boundary. The iovec cnt for this boundary
1354 	 * is BDEV_IO_NUM_CHILD_IOV + 1, which exceeds the capacity of child iovs.
1355 	 */
1356 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1357 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1358 		iov[i].iov_len = 512;
1359 	}
1360 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1361 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1362 		iov[i].iov_len = 256;
1363 	}
1364 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1365 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 512;
1366 
1367 	/* Add an extra iovec to trigger split */
1368 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1369 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1370 
1371 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1372 	g_io_done = false;
1373 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1374 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV);
1375 	expected_io->md_buf = md_buf;
1376 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1377 		ut_expected_io_set_iov(expected_io, i,
1378 				       (void *)((i + 1) * 0x10000), 512);
1379 	}
1380 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1381 		ut_expected_io_set_iov(expected_io, i,
1382 				       (void *)((i + 1) * 0x10000), 256);
1383 	}
1384 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1385 
1386 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1387 					   1, 1);
1388 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1389 	ut_expected_io_set_iov(expected_io, 0,
1390 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 512);
1391 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1392 
1393 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1394 					   1, 1);
1395 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1396 	ut_expected_io_set_iov(expected_io, 0,
1397 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1398 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1399 
1400 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, md_buf,
1401 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1402 	CU_ASSERT(rc == 0);
1403 	CU_ASSERT(g_io_done == false);
1404 
1405 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1406 	stub_complete_io(1);
1407 	CU_ASSERT(g_io_done == false);
1408 
1409 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1410 	stub_complete_io(2);
1411 	CU_ASSERT(g_io_done == true);
1412 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1413 
1414 	/* Test multi vector command that needs to be split by strip and then needs to be
1415 	 * split further due to the capacity of child iovs, the child request offset should
1416 	 * be rewind to last aligned offset and go success without error.
1417 	 */
1418 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1419 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1420 		iov[i].iov_len = 512;
1421 	}
1422 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1423 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1424 
1425 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1426 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1427 
1428 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1429 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1430 
1431 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1432 	g_io_done = false;
1433 	g_io_status = 0;
1434 	/* The first expected io should be start from offset 0 to BDEV_IO_NUM_CHILD_IOV - 1 */
1435 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1436 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV - 1);
1437 	expected_io->md_buf = md_buf;
1438 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1439 		ut_expected_io_set_iov(expected_io, i,
1440 				       (void *)((i + 1) * 0x10000), 512);
1441 	}
1442 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1443 	/* The second expected io should be start from offset BDEV_IO_NUM_CHILD_IOV - 1 to BDEV_IO_NUM_CHILD_IOV */
1444 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1445 					   1, 2);
1446 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1447 	ut_expected_io_set_iov(expected_io, 0,
1448 			       (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000), 256);
1449 	ut_expected_io_set_iov(expected_io, 1,
1450 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 256);
1451 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1452 	/* The third expected io should be start from offset BDEV_IO_NUM_CHILD_IOV to BDEV_IO_NUM_CHILD_IOV + 1 */
1453 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1454 					   1, 1);
1455 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1456 	ut_expected_io_set_iov(expected_io, 0,
1457 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1458 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1459 
1460 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1461 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1462 	CU_ASSERT(rc == 0);
1463 	CU_ASSERT(g_io_done == false);
1464 
1465 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1466 	stub_complete_io(1);
1467 	CU_ASSERT(g_io_done == false);
1468 
1469 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1470 	stub_complete_io(2);
1471 	CU_ASSERT(g_io_done == true);
1472 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1473 
1474 	/* Test multi vector command that needs to be split due to the IO boundary and
1475 	 * the capacity of child iovs. Especially test the case when the command is
1476 	 * split due to the capacity of child iovs, the tail address is not aligned with
1477 	 * block size and is rewinded to the aligned address.
1478 	 *
1479 	 * The iovecs used in read request is complex but is based on the data
1480 	 * collected in the real issue. We change the base addresses but keep the lengths
1481 	 * not to loose the credibility of the test.
1482 	 */
1483 	bdev->optimal_io_boundary = 128;
1484 	g_io_done = false;
1485 	g_io_status = 0;
1486 
1487 	for (i = 0; i < 31; i++) {
1488 		iov[i].iov_base = (void *)(0xFEED0000000 + (i << 20));
1489 		iov[i].iov_len = 1024;
1490 	}
1491 	iov[31].iov_base = (void *)0xFEED1F00000;
1492 	iov[31].iov_len = 32768;
1493 	iov[32].iov_base = (void *)0xFEED2000000;
1494 	iov[32].iov_len = 160;
1495 	iov[33].iov_base = (void *)0xFEED2100000;
1496 	iov[33].iov_len = 4096;
1497 	iov[34].iov_base = (void *)0xFEED2200000;
1498 	iov[34].iov_len = 4096;
1499 	iov[35].iov_base = (void *)0xFEED2300000;
1500 	iov[35].iov_len = 4096;
1501 	iov[36].iov_base = (void *)0xFEED2400000;
1502 	iov[36].iov_len = 4096;
1503 	iov[37].iov_base = (void *)0xFEED2500000;
1504 	iov[37].iov_len = 4096;
1505 	iov[38].iov_base = (void *)0xFEED2600000;
1506 	iov[38].iov_len = 4096;
1507 	iov[39].iov_base = (void *)0xFEED2700000;
1508 	iov[39].iov_len = 4096;
1509 	iov[40].iov_base = (void *)0xFEED2800000;
1510 	iov[40].iov_len = 4096;
1511 	iov[41].iov_base = (void *)0xFEED2900000;
1512 	iov[41].iov_len = 4096;
1513 	iov[42].iov_base = (void *)0xFEED2A00000;
1514 	iov[42].iov_len = 4096;
1515 	iov[43].iov_base = (void *)0xFEED2B00000;
1516 	iov[43].iov_len = 12288;
1517 	iov[44].iov_base = (void *)0xFEED2C00000;
1518 	iov[44].iov_len = 8192;
1519 	iov[45].iov_base = (void *)0xFEED2F00000;
1520 	iov[45].iov_len = 4096;
1521 	iov[46].iov_base = (void *)0xFEED3000000;
1522 	iov[46].iov_len = 4096;
1523 	iov[47].iov_base = (void *)0xFEED3100000;
1524 	iov[47].iov_len = 4096;
1525 	iov[48].iov_base = (void *)0xFEED3200000;
1526 	iov[48].iov_len = 24576;
1527 	iov[49].iov_base = (void *)0xFEED3300000;
1528 	iov[49].iov_len = 16384;
1529 	iov[50].iov_base = (void *)0xFEED3400000;
1530 	iov[50].iov_len = 12288;
1531 	iov[51].iov_base = (void *)0xFEED3500000;
1532 	iov[51].iov_len = 4096;
1533 	iov[52].iov_base = (void *)0xFEED3600000;
1534 	iov[52].iov_len = 4096;
1535 	iov[53].iov_base = (void *)0xFEED3700000;
1536 	iov[53].iov_len = 4096;
1537 	iov[54].iov_base = (void *)0xFEED3800000;
1538 	iov[54].iov_len = 28672;
1539 	iov[55].iov_base = (void *)0xFEED3900000;
1540 	iov[55].iov_len = 20480;
1541 	iov[56].iov_base = (void *)0xFEED3A00000;
1542 	iov[56].iov_len = 4096;
1543 	iov[57].iov_base = (void *)0xFEED3B00000;
1544 	iov[57].iov_len = 12288;
1545 	iov[58].iov_base = (void *)0xFEED3C00000;
1546 	iov[58].iov_len = 4096;
1547 	iov[59].iov_base = (void *)0xFEED3D00000;
1548 	iov[59].iov_len = 4096;
1549 	iov[60].iov_base = (void *)0xFEED3E00000;
1550 	iov[60].iov_len = 352;
1551 
1552 	/* The 1st child IO must be from iov[0] to iov[31] split by the capacity
1553 	 * of child iovs,
1554 	 */
1555 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 126, 32);
1556 	expected_io->md_buf = md_buf;
1557 	for (i = 0; i < 32; i++) {
1558 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1559 	}
1560 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1561 
1562 	/* The 2nd child IO must be from iov[32] to the first 864 bytes of iov[33]
1563 	 * split by the IO boundary requirement.
1564 	 */
1565 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 126, 2, 2);
1566 	expected_io->md_buf = md_buf + 126 * 8;
1567 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base, iov[32].iov_len);
1568 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 864);
1569 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1570 
1571 	/* The 3rd child IO must be from the remaining 3232 bytes of iov[33] to
1572 	 * the first 864 bytes of iov[46] split by the IO boundary requirement.
1573 	 */
1574 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 128, 128, 14);
1575 	expected_io->md_buf = md_buf + 128 * 8;
1576 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[33].iov_base + 864),
1577 			       iov[33].iov_len - 864);
1578 	ut_expected_io_set_iov(expected_io, 1, iov[34].iov_base, iov[34].iov_len);
1579 	ut_expected_io_set_iov(expected_io, 2, iov[35].iov_base, iov[35].iov_len);
1580 	ut_expected_io_set_iov(expected_io, 3, iov[36].iov_base, iov[36].iov_len);
1581 	ut_expected_io_set_iov(expected_io, 4, iov[37].iov_base, iov[37].iov_len);
1582 	ut_expected_io_set_iov(expected_io, 5, iov[38].iov_base, iov[38].iov_len);
1583 	ut_expected_io_set_iov(expected_io, 6, iov[39].iov_base, iov[39].iov_len);
1584 	ut_expected_io_set_iov(expected_io, 7, iov[40].iov_base, iov[40].iov_len);
1585 	ut_expected_io_set_iov(expected_io, 8, iov[41].iov_base, iov[41].iov_len);
1586 	ut_expected_io_set_iov(expected_io, 9, iov[42].iov_base, iov[42].iov_len);
1587 	ut_expected_io_set_iov(expected_io, 10, iov[43].iov_base, iov[43].iov_len);
1588 	ut_expected_io_set_iov(expected_io, 11, iov[44].iov_base, iov[44].iov_len);
1589 	ut_expected_io_set_iov(expected_io, 12, iov[45].iov_base, iov[45].iov_len);
1590 	ut_expected_io_set_iov(expected_io, 13, iov[46].iov_base, 864);
1591 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1592 
1593 	/* The 4th child IO must be from the remaining 3232 bytes of iov[46] to the
1594 	 * first 864 bytes of iov[52] split by the IO boundary requirement.
1595 	 */
1596 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 256, 128, 7);
1597 	expected_io->md_buf = md_buf + 256 * 8;
1598 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[46].iov_base + 864),
1599 			       iov[46].iov_len - 864);
1600 	ut_expected_io_set_iov(expected_io, 1, iov[47].iov_base, iov[47].iov_len);
1601 	ut_expected_io_set_iov(expected_io, 2, iov[48].iov_base, iov[48].iov_len);
1602 	ut_expected_io_set_iov(expected_io, 3, iov[49].iov_base, iov[49].iov_len);
1603 	ut_expected_io_set_iov(expected_io, 4, iov[50].iov_base, iov[50].iov_len);
1604 	ut_expected_io_set_iov(expected_io, 5, iov[51].iov_base, iov[51].iov_len);
1605 	ut_expected_io_set_iov(expected_io, 6, iov[52].iov_base, 864);
1606 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1607 
1608 	/* The 5th child IO must be from the remaining 3232 bytes of iov[52] to
1609 	 * the first 4096 bytes of iov[57] split by the IO boundary requirement.
1610 	 */
1611 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 384, 128, 6);
1612 	expected_io->md_buf = md_buf + 384 * 8;
1613 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[52].iov_base + 864),
1614 			       iov[52].iov_len - 864);
1615 	ut_expected_io_set_iov(expected_io, 1, iov[53].iov_base, iov[53].iov_len);
1616 	ut_expected_io_set_iov(expected_io, 2, iov[54].iov_base, iov[54].iov_len);
1617 	ut_expected_io_set_iov(expected_io, 3, iov[55].iov_base, iov[55].iov_len);
1618 	ut_expected_io_set_iov(expected_io, 4, iov[56].iov_base, iov[56].iov_len);
1619 	ut_expected_io_set_iov(expected_io, 5, iov[57].iov_base, 4960);
1620 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1621 
1622 	/* The 6th child IO must be from the remaining 7328 bytes of iov[57]
1623 	 * to the first 3936 bytes of iov[58] split by the capacity of child iovs.
1624 	 */
1625 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 512, 30, 3);
1626 	expected_io->md_buf = md_buf + 512 * 8;
1627 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[57].iov_base + 4960),
1628 			       iov[57].iov_len - 4960);
1629 	ut_expected_io_set_iov(expected_io, 1, iov[58].iov_base, iov[58].iov_len);
1630 	ut_expected_io_set_iov(expected_io, 2, iov[59].iov_base, 3936);
1631 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1632 
1633 	/* The 7th child IO is from the remaining 160 bytes of iov[59] and iov[60]. */
1634 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 542, 1, 2);
1635 	expected_io->md_buf = md_buf + 542 * 8;
1636 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[59].iov_base + 3936),
1637 			       iov[59].iov_len - 3936);
1638 	ut_expected_io_set_iov(expected_io, 1, iov[60].iov_base, iov[60].iov_len);
1639 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1640 
1641 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, 61, md_buf,
1642 					    0, 543, io_done, NULL);
1643 	CU_ASSERT(rc == 0);
1644 	CU_ASSERT(g_io_done == false);
1645 
1646 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1647 	stub_complete_io(1);
1648 	CU_ASSERT(g_io_done == false);
1649 
1650 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1651 	stub_complete_io(5);
1652 	CU_ASSERT(g_io_done == false);
1653 
1654 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1655 	stub_complete_io(1);
1656 	CU_ASSERT(g_io_done == true);
1657 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1658 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1659 
1660 	/* Test a WRITE_ZEROES that would span an I/O boundary.  WRITE_ZEROES should not be
1661 	 * split, so test that.
1662 	 */
1663 	bdev->optimal_io_boundary = 15;
1664 	g_io_done = false;
1665 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
1666 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1667 
1668 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
1669 	CU_ASSERT(rc == 0);
1670 	CU_ASSERT(g_io_done == false);
1671 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1672 	stub_complete_io(1);
1673 	CU_ASSERT(g_io_done == true);
1674 
1675 	/* Test an UNMAP.  This should also not be split. */
1676 	bdev->optimal_io_boundary = 16;
1677 	g_io_done = false;
1678 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 2, 0);
1679 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1680 
1681 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 2, io_done, NULL);
1682 	CU_ASSERT(rc == 0);
1683 	CU_ASSERT(g_io_done == false);
1684 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1685 	stub_complete_io(1);
1686 	CU_ASSERT(g_io_done == true);
1687 
1688 	/* Test a FLUSH.  This should also not be split. */
1689 	bdev->optimal_io_boundary = 16;
1690 	g_io_done = false;
1691 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 2, 0);
1692 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1693 
1694 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
1695 	CU_ASSERT(rc == 0);
1696 	CU_ASSERT(g_io_done == false);
1697 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1698 	stub_complete_io(1);
1699 	CU_ASSERT(g_io_done == true);
1700 
1701 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
1702 
1703 	/* Children requests return an error status */
1704 	bdev->optimal_io_boundary = 16;
1705 	iov[0].iov_base = (void *)0x10000;
1706 	iov[0].iov_len = 512 * 64;
1707 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1708 	g_io_done = false;
1709 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1710 
1711 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 1, 1, 64, io_done, NULL);
1712 	CU_ASSERT(rc == 0);
1713 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1714 	stub_complete_io(4);
1715 	CU_ASSERT(g_io_done == false);
1716 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1717 	stub_complete_io(1);
1718 	CU_ASSERT(g_io_done == true);
1719 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1720 
1721 	/* Test if a multi vector command terminated with failure before continuing
1722 	 * splitting process when one of child I/O failed.
1723 	 * The multi vector command is as same as the above that needs to be split by strip
1724 	 * and then needs to be split further due to the capacity of child iovs.
1725 	 */
1726 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1727 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1728 		iov[i].iov_len = 512;
1729 	}
1730 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1731 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1732 
1733 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1734 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1735 
1736 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1737 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1738 
1739 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1740 
1741 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1742 	g_io_done = false;
1743 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1744 
1745 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
1746 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1747 	CU_ASSERT(rc == 0);
1748 	CU_ASSERT(g_io_done == false);
1749 
1750 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1751 	stub_complete_io(1);
1752 	CU_ASSERT(g_io_done == true);
1753 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1754 
1755 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1756 
1757 	/* for this test we will create the following conditions to hit the code path where
1758 	 * we are trying to send and IO following a split that has no iovs because we had to
1759 	 * trim them for alignment reasons.
1760 	 *
1761 	 * - 16K boundary, our IO will start at offset 0 with a length of 0x4200
1762 	 * - Our IOVs are 0x212 in size so that we run into the 16K boundary at child IOV
1763 	 *   position 30 and overshoot by 0x2e.
1764 	 * - That means we'll send the IO and loop back to pick up the remaining bytes at
1765 	 *   child IOV index 31. When we do, we find that we have to shorten index 31 by 0x2e
1766 	 *   which eliniates that vector so we just send the first split IO with 30 vectors
1767 	 *   and let the completion pick up the last 2 vectors.
1768 	 */
1769 	bdev->optimal_io_boundary = 32;
1770 	bdev->split_on_optimal_io_boundary = true;
1771 	g_io_done = false;
1772 
1773 	/* Init all parent IOVs to 0x212 */
1774 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
1775 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1776 		iov[i].iov_len = 0x212;
1777 	}
1778 
1779 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1780 					   BDEV_IO_NUM_CHILD_IOV - 1);
1781 	/* expect 0-29 to be 1:1 with the parent iov */
1782 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1783 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1784 	}
1785 
1786 	/* expect index 30 to be shortened to 0x1e4 (0x212 - 0x1e) because of the alignment
1787 	 * where 0x1e is the amount we overshot the 16K boundary
1788 	 */
1789 	ut_expected_io_set_iov(expected_io, BDEV_IO_NUM_CHILD_IOV - 2,
1790 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base), 0x1e4);
1791 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1792 
1793 	/* 2nd child IO will have 2 remaining vectors, one to pick up from the one that was
1794 	 * shortened that take it to the next boundary and then a final one to get us to
1795 	 * 0x4200 bytes for the IO.
1796 	 */
1797 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1798 					   BDEV_IO_NUM_CHILD_IOV, 2);
1799 	/* position 30 picked up the remaining bytes to the next boundary */
1800 	ut_expected_io_set_iov(expected_io, 0,
1801 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base + 0x1e4), 0x2e);
1802 
1803 	/* position 31 picked the the rest of the transfer to get us to 0x4200 */
1804 	ut_expected_io_set_iov(expected_io, 1,
1805 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base), 0x1d2);
1806 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1807 
1808 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 1, 0,
1809 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1810 	CU_ASSERT(rc == 0);
1811 	CU_ASSERT(g_io_done == false);
1812 
1813 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1814 	stub_complete_io(1);
1815 	CU_ASSERT(g_io_done == false);
1816 
1817 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1818 	stub_complete_io(1);
1819 	CU_ASSERT(g_io_done == true);
1820 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1821 
1822 	spdk_put_io_channel(io_ch);
1823 	spdk_bdev_close(desc);
1824 	free_bdev(bdev);
1825 	spdk_bdev_finish(bdev_fini_cb, NULL);
1826 	poll_threads();
1827 }
1828 
1829 static void
1830 bdev_io_max_size_and_segment_split_test(void)
1831 {
1832 	struct spdk_bdev *bdev;
1833 	struct spdk_bdev_desc *desc = NULL;
1834 	struct spdk_io_channel *io_ch;
1835 	struct spdk_bdev_opts bdev_opts = {};
1836 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1837 	struct ut_expected_io *expected_io;
1838 	uint64_t i;
1839 	int rc;
1840 
1841 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1842 	bdev_opts.bdev_io_pool_size = 512;
1843 	bdev_opts.bdev_io_cache_size = 64;
1844 
1845 	bdev_opts.opts_size = sizeof(bdev_opts);
1846 	rc = spdk_bdev_set_opts(&bdev_opts);
1847 	CU_ASSERT(rc == 0);
1848 	spdk_bdev_initialize(bdev_init_cb, NULL);
1849 
1850 	bdev = allocate_bdev("bdev0");
1851 
1852 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
1853 	CU_ASSERT(rc == 0);
1854 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1855 	io_ch = spdk_bdev_get_io_channel(desc);
1856 	CU_ASSERT(io_ch != NULL);
1857 
1858 	bdev->split_on_optimal_io_boundary = false;
1859 	bdev->optimal_io_boundary = 0;
1860 
1861 	/* Case 0 max_num_segments == 0.
1862 	 * but segment size 2 * 512 > 512
1863 	 */
1864 	bdev->max_segment_size = 512;
1865 	bdev->max_num_segments = 0;
1866 	g_io_done = false;
1867 
1868 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
1869 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1870 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
1871 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1872 
1873 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1874 	CU_ASSERT(rc == 0);
1875 	CU_ASSERT(g_io_done == false);
1876 
1877 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1878 	stub_complete_io(1);
1879 	CU_ASSERT(g_io_done == true);
1880 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1881 
1882 	/* Case 1 max_segment_size == 0
1883 	 * but iov num 2 > 1.
1884 	 */
1885 	bdev->max_segment_size = 0;
1886 	bdev->max_num_segments = 1;
1887 	g_io_done = false;
1888 
1889 	iov[0].iov_base = (void *)0x10000;
1890 	iov[0].iov_len = 512;
1891 	iov[1].iov_base = (void *)0x20000;
1892 	iov[1].iov_len = 8 * 512;
1893 
1894 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1895 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, iov[0].iov_len);
1896 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1897 
1898 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 8, 1);
1899 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, iov[1].iov_len);
1900 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1901 
1902 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 14, 9, io_done, NULL);
1903 	CU_ASSERT(rc == 0);
1904 	CU_ASSERT(g_io_done == false);
1905 
1906 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1907 	stub_complete_io(2);
1908 	CU_ASSERT(g_io_done == true);
1909 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1910 
1911 	/* Test that a non-vector command is split correctly.
1912 	 * Set up the expected values before calling spdk_bdev_read_blocks
1913 	 */
1914 	bdev->max_segment_size = 512;
1915 	bdev->max_num_segments = 1;
1916 	g_io_done = false;
1917 
1918 	/* Child IO 0 */
1919 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1920 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1921 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1922 
1923 	/* Child IO 1 */
1924 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
1925 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 1 * 512), 512);
1926 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1927 
1928 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1929 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1930 	CU_ASSERT(rc == 0);
1931 	CU_ASSERT(g_io_done == false);
1932 
1933 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1934 	stub_complete_io(2);
1935 	CU_ASSERT(g_io_done == true);
1936 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1937 
1938 	/* Now set up a more complex, multi-vector command that needs to be split,
1939 	 * including splitting iovecs.
1940 	 */
1941 	bdev->max_segment_size = 2 * 512;
1942 	bdev->max_num_segments = 1;
1943 	g_io_done = false;
1944 
1945 	iov[0].iov_base = (void *)0x10000;
1946 	iov[0].iov_len = 2 * 512;
1947 	iov[1].iov_base = (void *)0x20000;
1948 	iov[1].iov_len = 4 * 512;
1949 	iov[2].iov_base = (void *)0x30000;
1950 	iov[2].iov_len = 6 * 512;
1951 
1952 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
1953 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 2);
1954 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1955 
1956 	/* Split iov[1].size to 2 iov entries then split the segments */
1957 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
1958 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, 512 * 2);
1959 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1960 
1961 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 18, 2, 1);
1962 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base + 512 * 2, 512 * 2);
1963 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1964 
1965 	/* Split iov[2].size to 3 iov entries then split the segments */
1966 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 20, 2, 1);
1967 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base, 512 * 2);
1968 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1969 
1970 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 22, 2, 1);
1971 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 2, 512 * 2);
1972 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1973 
1974 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 24, 2, 1);
1975 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 4, 512 * 2);
1976 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1977 
1978 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 12, io_done, NULL);
1979 	CU_ASSERT(rc == 0);
1980 	CU_ASSERT(g_io_done == false);
1981 
1982 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
1983 	stub_complete_io(6);
1984 	CU_ASSERT(g_io_done == true);
1985 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1986 
1987 	/* Test multi vector command that needs to be split by strip and then needs to be
1988 	 * split further due to the capacity of parent IO child iovs.
1989 	 */
1990 	bdev->max_segment_size = 512;
1991 	bdev->max_num_segments = 1;
1992 	g_io_done = false;
1993 
1994 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1995 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1996 		iov[i].iov_len = 512 * 2;
1997 	}
1998 
1999 	/* Each input iov.size is split into 2 iovs,
2000 	 * half of the input iov can fill all child iov entries of a single IO.
2001 	 */
2002 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV / 2; i++) {
2003 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i, 1, 1);
2004 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
2005 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2006 
2007 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i + 1, 1, 1);
2008 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
2009 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2010 	}
2011 
2012 	/* The remaining iov is split in the second round */
2013 	for (i = BDEV_IO_NUM_CHILD_IOV / 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
2014 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2, 1, 1);
2015 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
2016 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2017 
2018 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2 + 1, 1, 1);
2019 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
2020 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2021 	}
2022 
2023 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
2024 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
2025 	CU_ASSERT(rc == 0);
2026 	CU_ASSERT(g_io_done == false);
2027 
2028 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
2029 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
2030 	CU_ASSERT(g_io_done == false);
2031 
2032 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
2033 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
2034 	CU_ASSERT(g_io_done == true);
2035 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2036 
2037 	/* A wrong case, a child IO that is divided does
2038 	 * not meet the principle of multiples of block size,
2039 	 * and exits with error
2040 	 */
2041 	bdev->max_segment_size = 512;
2042 	bdev->max_num_segments = 1;
2043 	g_io_done = false;
2044 
2045 	iov[0].iov_base = (void *)0x10000;
2046 	iov[0].iov_len = 512 + 256;
2047 	iov[1].iov_base = (void *)0x20000;
2048 	iov[1].iov_len = 256;
2049 
2050 	/* iov[0] is split to 512 and 256.
2051 	 * 256 is less than a block size, and it is found
2052 	 * in the next round of split that it is the first child IO smaller than
2053 	 * the block size, so the error exit
2054 	 */
2055 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 1, 1);
2056 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512);
2057 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2058 
2059 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 0, 2, io_done, NULL);
2060 	CU_ASSERT(rc == 0);
2061 	CU_ASSERT(g_io_done == false);
2062 
2063 	/* First child IO is OK */
2064 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2065 	stub_complete_io(1);
2066 	CU_ASSERT(g_io_done == true);
2067 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2068 
2069 	/* error exit */
2070 	stub_complete_io(1);
2071 	CU_ASSERT(g_io_done == true);
2072 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
2073 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2074 
2075 	/* Test multi vector command that needs to be split by strip and then needs to be
2076 	 * split further due to the capacity of child iovs.
2077 	 *
2078 	 * In this case, the last two iovs need to be split, but it will exceed the capacity
2079 	 * of child iovs, so it needs to wait until the first batch completed.
2080 	 */
2081 	bdev->max_segment_size = 512;
2082 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2083 	g_io_done = false;
2084 
2085 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2086 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2087 		iov[i].iov_len = 512;
2088 	}
2089 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
2090 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2091 		iov[i].iov_len = 512 * 2;
2092 	}
2093 
2094 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2095 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
2096 	/* 0 ~ (BDEV_IO_NUM_CHILD_IOV - 2) Will not be split */
2097 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2098 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2099 	}
2100 	/* (BDEV_IO_NUM_CHILD_IOV - 2) is split */
2101 	ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, 512);
2102 	ut_expected_io_set_iov(expected_io, i + 1, iov[i].iov_base + 512, 512);
2103 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2104 
2105 	/* Child iov entries exceed the max num of parent IO so split it in next round */
2106 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV, 2, 2);
2107 	ut_expected_io_set_iov(expected_io, 0, iov[i + 1].iov_base, 512);
2108 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base + 512, 512);
2109 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2110 
2111 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
2112 				    BDEV_IO_NUM_CHILD_IOV + 2, io_done, NULL);
2113 	CU_ASSERT(rc == 0);
2114 	CU_ASSERT(g_io_done == false);
2115 
2116 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2117 	stub_complete_io(1);
2118 	CU_ASSERT(g_io_done == false);
2119 
2120 	/* Next round */
2121 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2122 	stub_complete_io(1);
2123 	CU_ASSERT(g_io_done == true);
2124 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2125 
2126 	/* This case is similar to the previous one, but the io composed of
2127 	 * the last few entries of child iov is not enough for a blocklen, so they
2128 	 * cannot be put into this IO, but wait until the next time.
2129 	 */
2130 	bdev->max_segment_size = 512;
2131 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2132 	g_io_done = false;
2133 
2134 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2135 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2136 		iov[i].iov_len = 512;
2137 	}
2138 
2139 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2140 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2141 		iov[i].iov_len = 128;
2142 	}
2143 
2144 	/* First child iovcnt is't BDEV_IO_NUM_CHILD_IOV but BDEV_IO_NUM_CHILD_IOV - 2.
2145 	 * Because the left 2 iov is not enough for a blocklen.
2146 	 */
2147 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2148 					   BDEV_IO_NUM_CHILD_IOV - 2, BDEV_IO_NUM_CHILD_IOV - 2);
2149 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2150 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2151 	}
2152 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2153 
2154 	/* The second child io waits until the end of the first child io before executing.
2155 	 * Because the iovcnt of the two IOs exceeds the child iovcnt of the parent IO.
2156 	 * BDEV_IO_NUM_CHILD_IOV - 2 to BDEV_IO_NUM_CHILD_IOV + 2
2157 	 */
2158 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 2,
2159 					   1, 4);
2160 	ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, iov[i].iov_len);
2161 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base, iov[i + 1].iov_len);
2162 	ut_expected_io_set_iov(expected_io, 2, iov[i + 2].iov_base, iov[i + 2].iov_len);
2163 	ut_expected_io_set_iov(expected_io, 3, iov[i + 3].iov_base, iov[i + 3].iov_len);
2164 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2165 
2166 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2167 				    BDEV_IO_NUM_CHILD_IOV - 1, io_done, NULL);
2168 	CU_ASSERT(rc == 0);
2169 	CU_ASSERT(g_io_done == false);
2170 
2171 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2172 	stub_complete_io(1);
2173 	CU_ASSERT(g_io_done == false);
2174 
2175 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2176 	stub_complete_io(1);
2177 	CU_ASSERT(g_io_done == true);
2178 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2179 
2180 	/* A very complicated case. Each sg entry exceeds max_segment_size and
2181 	 * needs to be split. At the same time, child io must be a multiple of blocklen.
2182 	 * At the same time, child iovcnt exceeds parent iovcnt.
2183 	 */
2184 	bdev->max_segment_size = 512 + 128;
2185 	bdev->max_num_segments = 3;
2186 	g_io_done = false;
2187 
2188 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2189 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2190 		iov[i].iov_len = 512 + 256;
2191 	}
2192 
2193 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2194 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2195 		iov[i].iov_len = 512 + 128;
2196 	}
2197 
2198 	/* Child IOs use 9 entries per for() round and 3 * 9 = 27 child iov entries.
2199 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2200 	 * Generate 9 child IOs.
2201 	 */
2202 	for (i = 0; i < 3; i++) {
2203 		uint32_t j = i * 4;
2204 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6, 2, 3);
2205 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2206 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2207 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2208 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2209 
2210 		/* Child io must be a multiple of blocklen
2211 		 * iov[j + 2] must be split. If the third entry is also added,
2212 		 * the multiple of blocklen cannot be guaranteed. But it still
2213 		 * occupies one iov entry of the parent child iov.
2214 		 */
2215 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 2, 2, 2);
2216 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2217 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2218 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2219 
2220 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 4, 2, 3);
2221 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2222 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2223 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2224 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2225 	}
2226 
2227 	/* Child iov position at 27, the 10th child IO
2228 	 * iov entry index is 3 * 4 and offset is 3 * 6
2229 	 */
2230 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 18, 2, 3);
2231 	ut_expected_io_set_iov(expected_io, 0, iov[12].iov_base, 640);
2232 	ut_expected_io_set_iov(expected_io, 1, iov[12].iov_base + 640, 128);
2233 	ut_expected_io_set_iov(expected_io, 2, iov[13].iov_base, 256);
2234 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2235 
2236 	/* Child iov position at 30, the 11th child IO */
2237 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 20, 2, 2);
2238 	ut_expected_io_set_iov(expected_io, 0, iov[13].iov_base + 256, 512);
2239 	ut_expected_io_set_iov(expected_io, 1, iov[14].iov_base, 512);
2240 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2241 
2242 	/* The 2nd split round and iovpos is 0, the 12th child IO */
2243 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 22, 2, 3);
2244 	ut_expected_io_set_iov(expected_io, 0, iov[14].iov_base + 512, 256);
2245 	ut_expected_io_set_iov(expected_io, 1, iov[15].iov_base, 640);
2246 	ut_expected_io_set_iov(expected_io, 2, iov[15].iov_base + 640, 128);
2247 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2248 
2249 	/* Consume 9 child IOs and 27 child iov entries.
2250 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2251 	 * Parent IO iov index start from 16 and block offset start from 24
2252 	 */
2253 	for (i = 0; i < 3; i++) {
2254 		uint32_t j = i * 4 + 16;
2255 		uint32_t offset = i * 6 + 24;
2256 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, 2, 3);
2257 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2258 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2259 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2260 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2261 
2262 		/* Child io must be a multiple of blocklen
2263 		 * iov[j + 2] must be split. If the third entry is also added,
2264 		 * the multiple of blocklen cannot be guaranteed. But it still
2265 		 * occupies one iov entry of the parent child iov.
2266 		 */
2267 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 2, 2, 2);
2268 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2269 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2270 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2271 
2272 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 4, 2, 3);
2273 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2274 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2275 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2276 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2277 	}
2278 
2279 	/* The 22th child IO, child iov position at 30 */
2280 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 42, 1, 1);
2281 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base, 512);
2282 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2283 
2284 	/* The third round */
2285 	/* Here is the 23nd child IO and child iovpos is 0 */
2286 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 43, 2, 3);
2287 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base + 512, 256);
2288 	ut_expected_io_set_iov(expected_io, 1, iov[29].iov_base, 640);
2289 	ut_expected_io_set_iov(expected_io, 2, iov[29].iov_base + 640, 128);
2290 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2291 
2292 	/* The 24th child IO */
2293 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 45, 3, 3);
2294 	ut_expected_io_set_iov(expected_io, 0, iov[30].iov_base, 640);
2295 	ut_expected_io_set_iov(expected_io, 1, iov[31].iov_base, 640);
2296 	ut_expected_io_set_iov(expected_io, 2, iov[32].iov_base, 256);
2297 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2298 
2299 	/* The 25th child IO */
2300 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 48, 2, 2);
2301 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base + 256, 384);
2302 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 640);
2303 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2304 
2305 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2306 				    50, io_done, NULL);
2307 	CU_ASSERT(rc == 0);
2308 	CU_ASSERT(g_io_done == false);
2309 
2310 	/* Parent IO supports up to 32 child iovs, so it is calculated that
2311 	 * a maximum of 11 IOs can be split at a time, and the
2312 	 * splitting will continue after the first batch is over.
2313 	 */
2314 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2315 	stub_complete_io(11);
2316 	CU_ASSERT(g_io_done == false);
2317 
2318 	/* The 2nd round */
2319 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2320 	stub_complete_io(11);
2321 	CU_ASSERT(g_io_done == false);
2322 
2323 	/* The last round */
2324 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2325 	stub_complete_io(3);
2326 	CU_ASSERT(g_io_done == true);
2327 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2328 
2329 	/* Test an WRITE_ZEROES.  This should also not be split. */
2330 	bdev->max_segment_size = 512;
2331 	bdev->max_num_segments = 1;
2332 	g_io_done = false;
2333 
2334 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
2335 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2336 
2337 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
2338 	CU_ASSERT(rc == 0);
2339 	CU_ASSERT(g_io_done == false);
2340 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2341 	stub_complete_io(1);
2342 	CU_ASSERT(g_io_done == true);
2343 
2344 	/* Test an UNMAP.  This should also not be split. */
2345 	g_io_done = false;
2346 
2347 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 4, 0);
2348 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2349 
2350 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 4, io_done, NULL);
2351 	CU_ASSERT(rc == 0);
2352 	CU_ASSERT(g_io_done == false);
2353 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2354 	stub_complete_io(1);
2355 	CU_ASSERT(g_io_done == true);
2356 
2357 	/* Test a FLUSH.  This should also not be split. */
2358 	g_io_done = false;
2359 
2360 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 4, 0);
2361 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2362 
2363 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
2364 	CU_ASSERT(rc == 0);
2365 	CU_ASSERT(g_io_done == false);
2366 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2367 	stub_complete_io(1);
2368 	CU_ASSERT(g_io_done == true);
2369 
2370 	spdk_put_io_channel(io_ch);
2371 	spdk_bdev_close(desc);
2372 	free_bdev(bdev);
2373 	spdk_bdev_finish(bdev_fini_cb, NULL);
2374 	poll_threads();
2375 }
2376 
2377 static void
2378 bdev_io_mix_split_test(void)
2379 {
2380 	struct spdk_bdev *bdev;
2381 	struct spdk_bdev_desc *desc = NULL;
2382 	struct spdk_io_channel *io_ch;
2383 	struct spdk_bdev_opts bdev_opts = {};
2384 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
2385 	struct ut_expected_io *expected_io;
2386 	uint64_t i;
2387 	int rc;
2388 
2389 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2390 	bdev_opts.bdev_io_pool_size = 512;
2391 	bdev_opts.bdev_io_cache_size = 64;
2392 
2393 	rc = spdk_bdev_set_opts(&bdev_opts);
2394 	CU_ASSERT(rc == 0);
2395 	spdk_bdev_initialize(bdev_init_cb, NULL);
2396 
2397 	bdev = allocate_bdev("bdev0");
2398 
2399 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
2400 	CU_ASSERT(rc == 0);
2401 	SPDK_CU_ASSERT_FATAL(desc != NULL);
2402 	io_ch = spdk_bdev_get_io_channel(desc);
2403 	CU_ASSERT(io_ch != NULL);
2404 
2405 	/* First case optimal_io_boundary == max_segment_size * max_num_segments */
2406 	bdev->split_on_optimal_io_boundary = true;
2407 	bdev->optimal_io_boundary = 16;
2408 
2409 	bdev->max_segment_size = 512;
2410 	bdev->max_num_segments = 16;
2411 	g_io_done = false;
2412 
2413 	/* IO crossing the IO boundary requires split
2414 	 * Total 2 child IOs.
2415 	 */
2416 
2417 	/* The 1st child IO split the segment_size to multiple segment entry */
2418 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
2419 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
2420 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
2421 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2422 
2423 	/* The 2nd child IO split the segment_size to multiple segment entry */
2424 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 2);
2425 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 512);
2426 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 3 * 512), 512);
2427 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2428 
2429 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 4, io_done, NULL);
2430 	CU_ASSERT(rc == 0);
2431 	CU_ASSERT(g_io_done == false);
2432 
2433 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2434 	stub_complete_io(2);
2435 	CU_ASSERT(g_io_done == true);
2436 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2437 
2438 	/* Second case optimal_io_boundary > max_segment_size * max_num_segments */
2439 	bdev->max_segment_size = 15 * 512;
2440 	bdev->max_num_segments = 1;
2441 	g_io_done = false;
2442 
2443 	/* IO crossing the IO boundary requires split.
2444 	 * The 1st child IO segment size exceeds the max_segment_size,
2445 	 * So 1st child IO will be splitted to multiple segment entry.
2446 	 * Then it split to 2 child IOs because of the max_num_segments.
2447 	 * Total 3 child IOs.
2448 	 */
2449 
2450 	/* The first 2 IOs are in an IO boundary.
2451 	 * Because the optimal_io_boundary > max_segment_size * max_num_segments
2452 	 * So it split to the first 2 IOs.
2453 	 */
2454 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 15, 1);
2455 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 15);
2456 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2457 
2458 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
2459 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 15), 512);
2460 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2461 
2462 	/* The 3rd Child IO is because of the io boundary */
2463 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2464 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2465 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2466 
2467 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2468 	CU_ASSERT(rc == 0);
2469 	CU_ASSERT(g_io_done == false);
2470 
2471 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2472 	stub_complete_io(3);
2473 	CU_ASSERT(g_io_done == true);
2474 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2475 
2476 	/* Third case optimal_io_boundary < max_segment_size * max_num_segments */
2477 	bdev->max_segment_size = 17 * 512;
2478 	bdev->max_num_segments = 1;
2479 	g_io_done = false;
2480 
2481 	/* IO crossing the IO boundary requires split.
2482 	 * Child IO does not split.
2483 	 * Total 2 child IOs.
2484 	 */
2485 
2486 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 16, 1);
2487 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 16);
2488 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2489 
2490 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2491 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2492 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2493 
2494 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2495 	CU_ASSERT(rc == 0);
2496 	CU_ASSERT(g_io_done == false);
2497 
2498 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2499 	stub_complete_io(2);
2500 	CU_ASSERT(g_io_done == true);
2501 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2502 
2503 	/* Now set up a more complex, multi-vector command that needs to be split,
2504 	 * including splitting iovecs.
2505 	 * optimal_io_boundary < max_segment_size * max_num_segments
2506 	 */
2507 	bdev->max_segment_size = 3 * 512;
2508 	bdev->max_num_segments = 6;
2509 	g_io_done = false;
2510 
2511 	iov[0].iov_base = (void *)0x10000;
2512 	iov[0].iov_len = 4 * 512;
2513 	iov[1].iov_base = (void *)0x20000;
2514 	iov[1].iov_len = 4 * 512;
2515 	iov[2].iov_base = (void *)0x30000;
2516 	iov[2].iov_len = 10 * 512;
2517 
2518 	/* IO crossing the IO boundary requires split.
2519 	 * The 1st child IO segment size exceeds the max_segment_size and after
2520 	 * splitting segment_size, the num_segments exceeds max_num_segments.
2521 	 * So 1st child IO will be splitted to 2 child IOs.
2522 	 * Total 3 child IOs.
2523 	 */
2524 
2525 	/* The first 2 IOs are in an IO boundary.
2526 	 * After splitting segment size the segment num exceeds.
2527 	 * So it splits to 2 child IOs.
2528 	 */
2529 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 14, 6);
2530 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 3);
2531 	ut_expected_io_set_iov(expected_io, 1, iov[0].iov_base + 512 * 3, 512);
2532 	ut_expected_io_set_iov(expected_io, 2, iov[1].iov_base, 512 * 3);
2533 	ut_expected_io_set_iov(expected_io, 3, iov[1].iov_base + 512 * 3, 512);
2534 	ut_expected_io_set_iov(expected_io, 4, iov[2].iov_base, 512 * 3);
2535 	ut_expected_io_set_iov(expected_io, 5, iov[2].iov_base + 512 * 3, 512 * 3);
2536 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2537 
2538 	/* The 2nd child IO has the left segment entry */
2539 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
2540 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 6, 512 * 2);
2541 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2542 
2543 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
2544 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 8, 512 * 2);
2545 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2546 
2547 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 0, 18, io_done, NULL);
2548 	CU_ASSERT(rc == 0);
2549 	CU_ASSERT(g_io_done == false);
2550 
2551 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2552 	stub_complete_io(3);
2553 	CU_ASSERT(g_io_done == true);
2554 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2555 
2556 	/* A very complicated case. Each sg entry exceeds max_segment_size
2557 	 * and split on io boundary.
2558 	 * optimal_io_boundary < max_segment_size * max_num_segments
2559 	 */
2560 	bdev->max_segment_size = 3 * 512;
2561 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2562 	g_io_done = false;
2563 
2564 	for (i = 0; i < 20; i++) {
2565 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2566 		iov[i].iov_len = 512 * 4;
2567 	}
2568 
2569 	/* IO crossing the IO boundary requires split.
2570 	 * 80 block length can split 5 child IOs base on offset and IO boundary.
2571 	 * Each iov entry needs to be splitted to 2 entries because of max_segment_size
2572 	 * Total 5 child IOs.
2573 	 */
2574 
2575 	/* 4 iov entries are in an IO boundary and each iov entry splits to 2.
2576 	 * So each child IO occupies 8 child iov entries.
2577 	 */
2578 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 16, 8);
2579 	for (i = 0; i < 4; i++) {
2580 		int iovcnt = i * 2;
2581 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2582 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2583 	}
2584 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2585 
2586 	/* 2nd child IO and total 16 child iov entries of parent IO */
2587 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 8);
2588 	for (i = 4; i < 8; i++) {
2589 		int iovcnt = (i - 4) * 2;
2590 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2591 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2592 	}
2593 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2594 
2595 	/* 3rd child IO and total 24 child iov entries of parent IO */
2596 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 16, 8);
2597 	for (i = 8; i < 12; i++) {
2598 		int iovcnt = (i - 8) * 2;
2599 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2600 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2601 	}
2602 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2603 
2604 	/* 4th child IO and total 32 child iov entries of parent IO */
2605 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 48, 16, 8);
2606 	for (i = 12; i < 16; i++) {
2607 		int iovcnt = (i - 12) * 2;
2608 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2609 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2610 	}
2611 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2612 
2613 	/* 5th child IO and because of the child iov entry it should be splitted
2614 	 * in next round.
2615 	 */
2616 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 64, 16, 8);
2617 	for (i = 16; i < 20; i++) {
2618 		int iovcnt = (i - 16) * 2;
2619 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2620 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2621 	}
2622 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2623 
2624 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 20, 0, 80, io_done, NULL);
2625 	CU_ASSERT(rc == 0);
2626 	CU_ASSERT(g_io_done == false);
2627 
2628 	/* First split round */
2629 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
2630 	stub_complete_io(4);
2631 	CU_ASSERT(g_io_done == false);
2632 
2633 	/* Second split round */
2634 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2635 	stub_complete_io(1);
2636 	CU_ASSERT(g_io_done == true);
2637 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2638 
2639 	spdk_put_io_channel(io_ch);
2640 	spdk_bdev_close(desc);
2641 	free_bdev(bdev);
2642 	spdk_bdev_finish(bdev_fini_cb, NULL);
2643 	poll_threads();
2644 }
2645 
2646 static void
2647 bdev_io_split_with_io_wait(void)
2648 {
2649 	struct spdk_bdev *bdev;
2650 	struct spdk_bdev_desc *desc = NULL;
2651 	struct spdk_io_channel *io_ch;
2652 	struct spdk_bdev_channel *channel;
2653 	struct spdk_bdev_mgmt_channel *mgmt_ch;
2654 	struct spdk_bdev_opts bdev_opts = {};
2655 	struct iovec iov[3];
2656 	struct ut_expected_io *expected_io;
2657 	int rc;
2658 
2659 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2660 	bdev_opts.bdev_io_pool_size = 2;
2661 	bdev_opts.bdev_io_cache_size = 1;
2662 
2663 	rc = spdk_bdev_set_opts(&bdev_opts);
2664 	CU_ASSERT(rc == 0);
2665 	spdk_bdev_initialize(bdev_init_cb, NULL);
2666 
2667 	bdev = allocate_bdev("bdev0");
2668 
2669 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2670 	CU_ASSERT(rc == 0);
2671 	CU_ASSERT(desc != NULL);
2672 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2673 	io_ch = spdk_bdev_get_io_channel(desc);
2674 	CU_ASSERT(io_ch != NULL);
2675 	channel = spdk_io_channel_get_ctx(io_ch);
2676 	mgmt_ch = channel->shared_resource->mgmt_ch;
2677 
2678 	bdev->optimal_io_boundary = 16;
2679 	bdev->split_on_optimal_io_boundary = true;
2680 
2681 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
2682 	CU_ASSERT(rc == 0);
2683 
2684 	/* Now test that a single-vector command is split correctly.
2685 	 * Offset 14, length 8, payload 0xF000
2686 	 *  Child - Offset 14, length 2, payload 0xF000
2687 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
2688 	 *
2689 	 * Set up the expected values before calling spdk_bdev_read_blocks
2690 	 */
2691 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
2692 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
2693 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2694 
2695 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
2696 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
2697 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2698 
2699 	/* The following children will be submitted sequentially due to the capacity of
2700 	 * spdk_bdev_io.
2701 	 */
2702 
2703 	/* The first child I/O will be queued to wait until an spdk_bdev_io becomes available */
2704 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
2705 	CU_ASSERT(rc == 0);
2706 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2707 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2708 
2709 	/* Completing the first read I/O will submit the first child */
2710 	stub_complete_io(1);
2711 	CU_ASSERT(TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2712 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2713 
2714 	/* Completing the first child will submit the second child */
2715 	stub_complete_io(1);
2716 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2717 
2718 	/* Complete the second child I/O.  This should result in our callback getting
2719 	 * invoked since the parent I/O is now complete.
2720 	 */
2721 	stub_complete_io(1);
2722 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2723 
2724 	/* Now set up a more complex, multi-vector command that needs to be split,
2725 	 *  including splitting iovecs.
2726 	 */
2727 	iov[0].iov_base = (void *)0x10000;
2728 	iov[0].iov_len = 512;
2729 	iov[1].iov_base = (void *)0x20000;
2730 	iov[1].iov_len = 20 * 512;
2731 	iov[2].iov_base = (void *)0x30000;
2732 	iov[2].iov_len = 11 * 512;
2733 
2734 	g_io_done = false;
2735 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
2736 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
2737 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
2738 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2739 
2740 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
2741 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
2742 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2743 
2744 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
2745 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
2746 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
2747 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2748 
2749 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 32, io_done, NULL);
2750 	CU_ASSERT(rc == 0);
2751 	CU_ASSERT(g_io_done == false);
2752 
2753 	/* The following children will be submitted sequentially due to the capacity of
2754 	 * spdk_bdev_io.
2755 	 */
2756 
2757 	/* Completing the first child will submit the second child */
2758 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2759 	stub_complete_io(1);
2760 	CU_ASSERT(g_io_done == false);
2761 
2762 	/* Completing the second child will submit the third child */
2763 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2764 	stub_complete_io(1);
2765 	CU_ASSERT(g_io_done == false);
2766 
2767 	/* Completing the third child will result in our callback getting invoked
2768 	 * since the parent I/O is now complete.
2769 	 */
2770 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2771 	stub_complete_io(1);
2772 	CU_ASSERT(g_io_done == true);
2773 
2774 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
2775 
2776 	spdk_put_io_channel(io_ch);
2777 	spdk_bdev_close(desc);
2778 	free_bdev(bdev);
2779 	spdk_bdev_finish(bdev_fini_cb, NULL);
2780 	poll_threads();
2781 }
2782 
2783 static void
2784 bdev_io_alignment(void)
2785 {
2786 	struct spdk_bdev *bdev;
2787 	struct spdk_bdev_desc *desc = NULL;
2788 	struct spdk_io_channel *io_ch;
2789 	struct spdk_bdev_opts bdev_opts = {};
2790 	int rc;
2791 	void *buf = NULL;
2792 	struct iovec iovs[2];
2793 	int iovcnt;
2794 	uint64_t alignment;
2795 
2796 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2797 	bdev_opts.bdev_io_pool_size = 20;
2798 	bdev_opts.bdev_io_cache_size = 2;
2799 
2800 	rc = spdk_bdev_set_opts(&bdev_opts);
2801 	CU_ASSERT(rc == 0);
2802 	spdk_bdev_initialize(bdev_init_cb, NULL);
2803 
2804 	fn_table.submit_request = stub_submit_request_get_buf;
2805 	bdev = allocate_bdev("bdev0");
2806 
2807 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2808 	CU_ASSERT(rc == 0);
2809 	CU_ASSERT(desc != NULL);
2810 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2811 	io_ch = spdk_bdev_get_io_channel(desc);
2812 	CU_ASSERT(io_ch != NULL);
2813 
2814 	/* Create aligned buffer */
2815 	rc = posix_memalign(&buf, 4096, 8192);
2816 	SPDK_CU_ASSERT_FATAL(rc == 0);
2817 
2818 	/* Pass aligned single buffer with no alignment required */
2819 	alignment = 1;
2820 	bdev->required_alignment = spdk_u32log2(alignment);
2821 
2822 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2823 	CU_ASSERT(rc == 0);
2824 	stub_complete_io(1);
2825 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2826 				    alignment));
2827 
2828 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2829 	CU_ASSERT(rc == 0);
2830 	stub_complete_io(1);
2831 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2832 				    alignment));
2833 
2834 	/* Pass unaligned single buffer with no alignment required */
2835 	alignment = 1;
2836 	bdev->required_alignment = spdk_u32log2(alignment);
2837 
2838 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2839 	CU_ASSERT(rc == 0);
2840 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2841 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2842 	stub_complete_io(1);
2843 
2844 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2845 	CU_ASSERT(rc == 0);
2846 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2847 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2848 	stub_complete_io(1);
2849 
2850 	/* Pass unaligned single buffer with 512 alignment required */
2851 	alignment = 512;
2852 	bdev->required_alignment = spdk_u32log2(alignment);
2853 
2854 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2855 	CU_ASSERT(rc == 0);
2856 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2857 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2858 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2859 				    alignment));
2860 	stub_complete_io(1);
2861 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2862 
2863 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2864 	CU_ASSERT(rc == 0);
2865 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2866 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2867 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2868 				    alignment));
2869 	stub_complete_io(1);
2870 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2871 
2872 	/* Pass unaligned single buffer with 4096 alignment required */
2873 	alignment = 4096;
2874 	bdev->required_alignment = spdk_u32log2(alignment);
2875 
2876 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2877 	CU_ASSERT(rc == 0);
2878 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2879 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2880 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2881 				    alignment));
2882 	stub_complete_io(1);
2883 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2884 
2885 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2886 	CU_ASSERT(rc == 0);
2887 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2888 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2889 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2890 				    alignment));
2891 	stub_complete_io(1);
2892 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2893 
2894 	/* Pass aligned iovs with no alignment required */
2895 	alignment = 1;
2896 	bdev->required_alignment = spdk_u32log2(alignment);
2897 
2898 	iovcnt = 1;
2899 	iovs[0].iov_base = buf;
2900 	iovs[0].iov_len = 512;
2901 
2902 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2903 	CU_ASSERT(rc == 0);
2904 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2905 	stub_complete_io(1);
2906 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2907 
2908 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2909 	CU_ASSERT(rc == 0);
2910 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2911 	stub_complete_io(1);
2912 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2913 
2914 	/* Pass unaligned iovs with no alignment required */
2915 	alignment = 1;
2916 	bdev->required_alignment = spdk_u32log2(alignment);
2917 
2918 	iovcnt = 2;
2919 	iovs[0].iov_base = buf + 16;
2920 	iovs[0].iov_len = 256;
2921 	iovs[1].iov_base = buf + 16 + 256 + 32;
2922 	iovs[1].iov_len = 256;
2923 
2924 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2925 	CU_ASSERT(rc == 0);
2926 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2927 	stub_complete_io(1);
2928 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2929 
2930 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2931 	CU_ASSERT(rc == 0);
2932 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2933 	stub_complete_io(1);
2934 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2935 
2936 	/* Pass unaligned iov with 2048 alignment required */
2937 	alignment = 2048;
2938 	bdev->required_alignment = spdk_u32log2(alignment);
2939 
2940 	iovcnt = 2;
2941 	iovs[0].iov_base = buf + 16;
2942 	iovs[0].iov_len = 256;
2943 	iovs[1].iov_base = buf + 16 + 256 + 32;
2944 	iovs[1].iov_len = 256;
2945 
2946 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2947 	CU_ASSERT(rc == 0);
2948 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2949 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2950 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2951 				    alignment));
2952 	stub_complete_io(1);
2953 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2954 
2955 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2956 	CU_ASSERT(rc == 0);
2957 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2958 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2959 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2960 				    alignment));
2961 	stub_complete_io(1);
2962 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2963 
2964 	/* Pass iov without allocated buffer without alignment required */
2965 	alignment = 1;
2966 	bdev->required_alignment = spdk_u32log2(alignment);
2967 
2968 	iovcnt = 1;
2969 	iovs[0].iov_base = NULL;
2970 	iovs[0].iov_len = 0;
2971 
2972 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2973 	CU_ASSERT(rc == 0);
2974 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2975 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2976 				    alignment));
2977 	stub_complete_io(1);
2978 
2979 	/* Pass iov without allocated buffer with 1024 alignment required */
2980 	alignment = 1024;
2981 	bdev->required_alignment = spdk_u32log2(alignment);
2982 
2983 	iovcnt = 1;
2984 	iovs[0].iov_base = NULL;
2985 	iovs[0].iov_len = 0;
2986 
2987 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2988 	CU_ASSERT(rc == 0);
2989 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2990 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2991 				    alignment));
2992 	stub_complete_io(1);
2993 
2994 	spdk_put_io_channel(io_ch);
2995 	spdk_bdev_close(desc);
2996 	free_bdev(bdev);
2997 	fn_table.submit_request = stub_submit_request;
2998 	spdk_bdev_finish(bdev_fini_cb, NULL);
2999 	poll_threads();
3000 
3001 	free(buf);
3002 }
3003 
3004 static void
3005 bdev_io_alignment_with_boundary(void)
3006 {
3007 	struct spdk_bdev *bdev;
3008 	struct spdk_bdev_desc *desc = NULL;
3009 	struct spdk_io_channel *io_ch;
3010 	struct spdk_bdev_opts bdev_opts = {};
3011 	int rc;
3012 	void *buf = NULL;
3013 	struct iovec iovs[2];
3014 	int iovcnt;
3015 	uint64_t alignment;
3016 
3017 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
3018 	bdev_opts.bdev_io_pool_size = 20;
3019 	bdev_opts.bdev_io_cache_size = 2;
3020 
3021 	bdev_opts.opts_size = sizeof(bdev_opts);
3022 	rc = spdk_bdev_set_opts(&bdev_opts);
3023 	CU_ASSERT(rc == 0);
3024 	spdk_bdev_initialize(bdev_init_cb, NULL);
3025 
3026 	fn_table.submit_request = stub_submit_request_get_buf;
3027 	bdev = allocate_bdev("bdev0");
3028 
3029 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
3030 	CU_ASSERT(rc == 0);
3031 	CU_ASSERT(desc != NULL);
3032 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3033 	io_ch = spdk_bdev_get_io_channel(desc);
3034 	CU_ASSERT(io_ch != NULL);
3035 
3036 	/* Create aligned buffer */
3037 	rc = posix_memalign(&buf, 4096, 131072);
3038 	SPDK_CU_ASSERT_FATAL(rc == 0);
3039 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3040 
3041 	/* 512 * 3 with 2 IO boundary, allocate small data buffer from bdev layer */
3042 	alignment = 512;
3043 	bdev->required_alignment = spdk_u32log2(alignment);
3044 	bdev->optimal_io_boundary = 2;
3045 	bdev->split_on_optimal_io_boundary = true;
3046 
3047 	iovcnt = 1;
3048 	iovs[0].iov_base = NULL;
3049 	iovs[0].iov_len = 512 * 3;
3050 
3051 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3052 	CU_ASSERT(rc == 0);
3053 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3054 	stub_complete_io(2);
3055 
3056 	/* 8KiB with 16 IO boundary, allocate large data buffer from bdev layer */
3057 	alignment = 512;
3058 	bdev->required_alignment = spdk_u32log2(alignment);
3059 	bdev->optimal_io_boundary = 16;
3060 	bdev->split_on_optimal_io_boundary = true;
3061 
3062 	iovcnt = 1;
3063 	iovs[0].iov_base = NULL;
3064 	iovs[0].iov_len = 512 * 16;
3065 
3066 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 16, io_done, NULL);
3067 	CU_ASSERT(rc == 0);
3068 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3069 	stub_complete_io(2);
3070 
3071 	/* 512 * 160 with 128 IO boundary, 63.5KiB + 16.5KiB for the two children requests */
3072 	alignment = 512;
3073 	bdev->required_alignment = spdk_u32log2(alignment);
3074 	bdev->optimal_io_boundary = 128;
3075 	bdev->split_on_optimal_io_boundary = true;
3076 
3077 	iovcnt = 1;
3078 	iovs[0].iov_base = buf + 16;
3079 	iovs[0].iov_len = 512 * 160;
3080 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3081 	CU_ASSERT(rc == 0);
3082 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3083 	stub_complete_io(2);
3084 
3085 	/* 512 * 3 with 2 IO boundary */
3086 	alignment = 512;
3087 	bdev->required_alignment = spdk_u32log2(alignment);
3088 	bdev->optimal_io_boundary = 2;
3089 	bdev->split_on_optimal_io_boundary = true;
3090 
3091 	iovcnt = 2;
3092 	iovs[0].iov_base = buf + 16;
3093 	iovs[0].iov_len = 512;
3094 	iovs[1].iov_base = buf + 16 + 512 + 32;
3095 	iovs[1].iov_len = 1024;
3096 
3097 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3098 	CU_ASSERT(rc == 0);
3099 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3100 	stub_complete_io(2);
3101 
3102 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3103 	CU_ASSERT(rc == 0);
3104 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3105 	stub_complete_io(2);
3106 
3107 	/* 512 * 64 with 32 IO boundary */
3108 	bdev->optimal_io_boundary = 32;
3109 	iovcnt = 2;
3110 	iovs[0].iov_base = buf + 16;
3111 	iovs[0].iov_len = 16384;
3112 	iovs[1].iov_base = buf + 16 + 16384 + 32;
3113 	iovs[1].iov_len = 16384;
3114 
3115 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3116 	CU_ASSERT(rc == 0);
3117 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3118 	stub_complete_io(3);
3119 
3120 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3121 	CU_ASSERT(rc == 0);
3122 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3123 	stub_complete_io(3);
3124 
3125 	/* 512 * 160 with 32 IO boundary */
3126 	iovcnt = 1;
3127 	iovs[0].iov_base = buf + 16;
3128 	iovs[0].iov_len = 16384 + 65536;
3129 
3130 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3131 	CU_ASSERT(rc == 0);
3132 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
3133 	stub_complete_io(6);
3134 
3135 	spdk_put_io_channel(io_ch);
3136 	spdk_bdev_close(desc);
3137 	free_bdev(bdev);
3138 	fn_table.submit_request = stub_submit_request;
3139 	spdk_bdev_finish(bdev_fini_cb, NULL);
3140 	poll_threads();
3141 
3142 	free(buf);
3143 }
3144 
3145 static void
3146 histogram_status_cb(void *cb_arg, int status)
3147 {
3148 	g_status = status;
3149 }
3150 
3151 static void
3152 histogram_data_cb(void *cb_arg, int status, struct spdk_histogram_data *histogram)
3153 {
3154 	g_status = status;
3155 	g_histogram = histogram;
3156 }
3157 
3158 static void
3159 histogram_io_count(void *ctx, uint64_t start, uint64_t end, uint64_t count,
3160 		   uint64_t total, uint64_t so_far)
3161 {
3162 	g_count += count;
3163 }
3164 
3165 static void
3166 bdev_histograms(void)
3167 {
3168 	struct spdk_bdev *bdev;
3169 	struct spdk_bdev_desc *desc = NULL;
3170 	struct spdk_io_channel *ch;
3171 	struct spdk_histogram_data *histogram;
3172 	uint8_t buf[4096];
3173 	int rc;
3174 
3175 	spdk_bdev_initialize(bdev_init_cb, NULL);
3176 
3177 	bdev = allocate_bdev("bdev");
3178 
3179 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3180 	CU_ASSERT(rc == 0);
3181 	CU_ASSERT(desc != NULL);
3182 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3183 
3184 	ch = spdk_bdev_get_io_channel(desc);
3185 	CU_ASSERT(ch != NULL);
3186 
3187 	/* Enable histogram */
3188 	g_status = -1;
3189 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, true);
3190 	poll_threads();
3191 	CU_ASSERT(g_status == 0);
3192 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3193 
3194 	/* Allocate histogram */
3195 	histogram = spdk_histogram_data_alloc();
3196 	SPDK_CU_ASSERT_FATAL(histogram != NULL);
3197 
3198 	/* Check if histogram is zeroed */
3199 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3200 	poll_threads();
3201 	CU_ASSERT(g_status == 0);
3202 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3203 
3204 	g_count = 0;
3205 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3206 
3207 	CU_ASSERT(g_count == 0);
3208 
3209 	rc = spdk_bdev_write_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3210 	CU_ASSERT(rc == 0);
3211 
3212 	spdk_delay_us(10);
3213 	stub_complete_io(1);
3214 	poll_threads();
3215 
3216 	rc = spdk_bdev_read_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3217 	CU_ASSERT(rc == 0);
3218 
3219 	spdk_delay_us(10);
3220 	stub_complete_io(1);
3221 	poll_threads();
3222 
3223 	/* Check if histogram gathered data from all I/O channels */
3224 	g_histogram = NULL;
3225 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3226 	poll_threads();
3227 	CU_ASSERT(g_status == 0);
3228 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3229 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3230 
3231 	g_count = 0;
3232 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3233 	CU_ASSERT(g_count == 2);
3234 
3235 	/* Disable histogram */
3236 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, false);
3237 	poll_threads();
3238 	CU_ASSERT(g_status == 0);
3239 	CU_ASSERT(bdev->internal.histogram_enabled == false);
3240 
3241 	/* Try to run histogram commands on disabled bdev */
3242 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3243 	poll_threads();
3244 	CU_ASSERT(g_status == -EFAULT);
3245 
3246 	spdk_histogram_data_free(histogram);
3247 	spdk_put_io_channel(ch);
3248 	spdk_bdev_close(desc);
3249 	free_bdev(bdev);
3250 	spdk_bdev_finish(bdev_fini_cb, NULL);
3251 	poll_threads();
3252 }
3253 
3254 static void
3255 _bdev_compare(bool emulated)
3256 {
3257 	struct spdk_bdev *bdev;
3258 	struct spdk_bdev_desc *desc = NULL;
3259 	struct spdk_io_channel *ioch;
3260 	struct ut_expected_io *expected_io;
3261 	uint64_t offset, num_blocks;
3262 	uint32_t num_completed;
3263 	char aa_buf[512];
3264 	char bb_buf[512];
3265 	struct iovec compare_iov;
3266 	uint8_t io_type;
3267 	int rc;
3268 
3269 	if (emulated) {
3270 		io_type = SPDK_BDEV_IO_TYPE_READ;
3271 	} else {
3272 		io_type = SPDK_BDEV_IO_TYPE_COMPARE;
3273 	}
3274 
3275 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3276 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3277 
3278 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = !emulated;
3279 
3280 	spdk_bdev_initialize(bdev_init_cb, NULL);
3281 	fn_table.submit_request = stub_submit_request_get_buf;
3282 	bdev = allocate_bdev("bdev");
3283 
3284 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3285 	CU_ASSERT_EQUAL(rc, 0);
3286 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3287 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3288 	ioch = spdk_bdev_get_io_channel(desc);
3289 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3290 
3291 	fn_table.submit_request = stub_submit_request_get_buf;
3292 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3293 
3294 	offset = 50;
3295 	num_blocks = 1;
3296 	compare_iov.iov_base = aa_buf;
3297 	compare_iov.iov_len = sizeof(aa_buf);
3298 
3299 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3300 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3301 
3302 	g_io_done = false;
3303 	g_compare_read_buf = aa_buf;
3304 	g_compare_read_buf_len = sizeof(aa_buf);
3305 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3306 	CU_ASSERT_EQUAL(rc, 0);
3307 	num_completed = stub_complete_io(1);
3308 	CU_ASSERT_EQUAL(num_completed, 1);
3309 	CU_ASSERT(g_io_done == true);
3310 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3311 
3312 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3313 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3314 
3315 	g_io_done = false;
3316 	g_compare_read_buf = bb_buf;
3317 	g_compare_read_buf_len = sizeof(bb_buf);
3318 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3319 	CU_ASSERT_EQUAL(rc, 0);
3320 	num_completed = stub_complete_io(1);
3321 	CU_ASSERT_EQUAL(num_completed, 1);
3322 	CU_ASSERT(g_io_done == true);
3323 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3324 
3325 	spdk_put_io_channel(ioch);
3326 	spdk_bdev_close(desc);
3327 	free_bdev(bdev);
3328 	fn_table.submit_request = stub_submit_request;
3329 	spdk_bdev_finish(bdev_fini_cb, NULL);
3330 	poll_threads();
3331 
3332 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3333 
3334 	g_compare_read_buf = NULL;
3335 }
3336 
3337 static void
3338 bdev_compare(void)
3339 {
3340 	_bdev_compare(true);
3341 	_bdev_compare(false);
3342 }
3343 
3344 static void
3345 bdev_compare_and_write(void)
3346 {
3347 	struct spdk_bdev *bdev;
3348 	struct spdk_bdev_desc *desc = NULL;
3349 	struct spdk_io_channel *ioch;
3350 	struct ut_expected_io *expected_io;
3351 	uint64_t offset, num_blocks;
3352 	uint32_t num_completed;
3353 	char aa_buf[512];
3354 	char bb_buf[512];
3355 	char cc_buf[512];
3356 	char write_buf[512];
3357 	struct iovec compare_iov;
3358 	struct iovec write_iov;
3359 	int rc;
3360 
3361 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3362 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3363 	memset(cc_buf, 0xcc, sizeof(cc_buf));
3364 
3365 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = false;
3366 
3367 	spdk_bdev_initialize(bdev_init_cb, NULL);
3368 	fn_table.submit_request = stub_submit_request_get_buf;
3369 	bdev = allocate_bdev("bdev");
3370 
3371 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3372 	CU_ASSERT_EQUAL(rc, 0);
3373 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3374 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3375 	ioch = spdk_bdev_get_io_channel(desc);
3376 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3377 
3378 	fn_table.submit_request = stub_submit_request_get_buf;
3379 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3380 
3381 	offset = 50;
3382 	num_blocks = 1;
3383 	compare_iov.iov_base = aa_buf;
3384 	compare_iov.iov_len = sizeof(aa_buf);
3385 	write_iov.iov_base = bb_buf;
3386 	write_iov.iov_len = sizeof(bb_buf);
3387 
3388 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3389 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3390 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, offset, num_blocks, 0);
3391 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3392 
3393 	g_io_done = false;
3394 	g_compare_read_buf = aa_buf;
3395 	g_compare_read_buf_len = sizeof(aa_buf);
3396 	memset(write_buf, 0, sizeof(write_buf));
3397 	g_compare_write_buf = write_buf;
3398 	g_compare_write_buf_len = sizeof(write_buf);
3399 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3400 			offset, num_blocks, io_done, NULL);
3401 	/* Trigger range locking */
3402 	poll_threads();
3403 	CU_ASSERT_EQUAL(rc, 0);
3404 	num_completed = stub_complete_io(1);
3405 	CU_ASSERT_EQUAL(num_completed, 1);
3406 	CU_ASSERT(g_io_done == false);
3407 	num_completed = stub_complete_io(1);
3408 	/* Trigger range unlocking */
3409 	poll_threads();
3410 	CU_ASSERT_EQUAL(num_completed, 1);
3411 	CU_ASSERT(g_io_done == true);
3412 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3413 	CU_ASSERT(memcmp(write_buf, bb_buf, sizeof(write_buf)) == 0);
3414 
3415 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3416 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3417 
3418 	g_io_done = false;
3419 	g_compare_read_buf = cc_buf;
3420 	g_compare_read_buf_len = sizeof(cc_buf);
3421 	memset(write_buf, 0, sizeof(write_buf));
3422 	g_compare_write_buf = write_buf;
3423 	g_compare_write_buf_len = sizeof(write_buf);
3424 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3425 			offset, num_blocks, io_done, NULL);
3426 	/* Trigger range locking */
3427 	poll_threads();
3428 	CU_ASSERT_EQUAL(rc, 0);
3429 	num_completed = stub_complete_io(1);
3430 	/* Trigger range unlocking earlier because we expect error here */
3431 	poll_threads();
3432 	CU_ASSERT_EQUAL(num_completed, 1);
3433 	CU_ASSERT(g_io_done == true);
3434 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3435 	num_completed = stub_complete_io(1);
3436 	CU_ASSERT_EQUAL(num_completed, 0);
3437 
3438 	spdk_put_io_channel(ioch);
3439 	spdk_bdev_close(desc);
3440 	free_bdev(bdev);
3441 	fn_table.submit_request = stub_submit_request;
3442 	spdk_bdev_finish(bdev_fini_cb, NULL);
3443 	poll_threads();
3444 
3445 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3446 
3447 	g_compare_read_buf = NULL;
3448 	g_compare_write_buf = NULL;
3449 }
3450 
3451 static void
3452 bdev_write_zeroes(void)
3453 {
3454 	struct spdk_bdev *bdev;
3455 	struct spdk_bdev_desc *desc = NULL;
3456 	struct spdk_io_channel *ioch;
3457 	struct ut_expected_io *expected_io;
3458 	uint64_t offset, num_io_blocks, num_blocks;
3459 	uint32_t num_completed, num_requests;
3460 	int rc;
3461 
3462 	spdk_bdev_initialize(bdev_init_cb, NULL);
3463 	bdev = allocate_bdev("bdev");
3464 
3465 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3466 	CU_ASSERT_EQUAL(rc, 0);
3467 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3468 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3469 	ioch = spdk_bdev_get_io_channel(desc);
3470 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3471 
3472 	fn_table.submit_request = stub_submit_request;
3473 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3474 
3475 	/* First test that if the bdev supports write_zeroes, the request won't be split */
3476 	bdev->md_len = 0;
3477 	bdev->blocklen = 4096;
3478 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3479 
3480 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
3481 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3482 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3483 	CU_ASSERT_EQUAL(rc, 0);
3484 	num_completed = stub_complete_io(1);
3485 	CU_ASSERT_EQUAL(num_completed, 1);
3486 
3487 	/* Check that if write zeroes is not supported it'll be replaced by regular writes */
3488 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
3489 	num_io_blocks = ZERO_BUFFER_SIZE / bdev->blocklen;
3490 	num_requests = 2;
3491 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * num_requests;
3492 
3493 	for (offset = 0; offset < num_requests; ++offset) {
3494 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3495 						   offset * num_io_blocks, num_io_blocks, 0);
3496 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3497 	}
3498 
3499 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3500 	CU_ASSERT_EQUAL(rc, 0);
3501 	num_completed = stub_complete_io(num_requests);
3502 	CU_ASSERT_EQUAL(num_completed, num_requests);
3503 
3504 	/* Check that the splitting is correct if bdev has interleaved metadata */
3505 	bdev->md_interleave = true;
3506 	bdev->md_len = 64;
3507 	bdev->blocklen = 4096 + 64;
3508 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3509 
3510 	num_requests = offset = 0;
3511 	while (offset < num_blocks) {
3512 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / bdev->blocklen, num_blocks - offset);
3513 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3514 						   offset, num_io_blocks, 0);
3515 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3516 		offset += num_io_blocks;
3517 		num_requests++;
3518 	}
3519 
3520 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3521 	CU_ASSERT_EQUAL(rc, 0);
3522 	num_completed = stub_complete_io(num_requests);
3523 	CU_ASSERT_EQUAL(num_completed, num_requests);
3524 	num_completed = stub_complete_io(num_requests);
3525 	assert(num_completed == 0);
3526 
3527 	/* Check the the same for separate metadata buffer */
3528 	bdev->md_interleave = false;
3529 	bdev->md_len = 64;
3530 	bdev->blocklen = 4096;
3531 
3532 	num_requests = offset = 0;
3533 	while (offset < num_blocks) {
3534 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / (bdev->blocklen + bdev->md_len), num_blocks);
3535 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3536 						   offset, num_io_blocks, 0);
3537 		expected_io->md_buf = (char *)g_bdev_mgr.zero_buffer + num_io_blocks * bdev->blocklen;
3538 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3539 		offset += num_io_blocks;
3540 		num_requests++;
3541 	}
3542 
3543 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3544 	CU_ASSERT_EQUAL(rc, 0);
3545 	num_completed = stub_complete_io(num_requests);
3546 	CU_ASSERT_EQUAL(num_completed, num_requests);
3547 
3548 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
3549 	spdk_put_io_channel(ioch);
3550 	spdk_bdev_close(desc);
3551 	free_bdev(bdev);
3552 	spdk_bdev_finish(bdev_fini_cb, NULL);
3553 	poll_threads();
3554 }
3555 
3556 static void
3557 bdev_zcopy_write(void)
3558 {
3559 	struct spdk_bdev *bdev;
3560 	struct spdk_bdev_desc *desc = NULL;
3561 	struct spdk_io_channel *ioch;
3562 	struct ut_expected_io *expected_io;
3563 	uint64_t offset, num_blocks;
3564 	uint32_t num_completed;
3565 	char aa_buf[512];
3566 	struct iovec iov;
3567 	int rc;
3568 	const bool populate = false;
3569 	const bool commit = true;
3570 
3571 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3572 
3573 	spdk_bdev_initialize(bdev_init_cb, NULL);
3574 	bdev = allocate_bdev("bdev");
3575 
3576 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3577 	CU_ASSERT_EQUAL(rc, 0);
3578 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3579 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3580 	ioch = spdk_bdev_get_io_channel(desc);
3581 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3582 
3583 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3584 
3585 	offset = 50;
3586 	num_blocks = 1;
3587 	iov.iov_base = NULL;
3588 	iov.iov_len = 0;
3589 
3590 	g_zcopy_read_buf = (void *) 0x1122334455667788UL;
3591 	g_zcopy_read_buf_len = (uint32_t) -1;
3592 	/* Do a zcopy start for a write (populate=false) */
3593 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3594 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3595 	g_io_done = false;
3596 	g_zcopy_write_buf = aa_buf;
3597 	g_zcopy_write_buf_len = sizeof(aa_buf);
3598 	g_zcopy_bdev_io = NULL;
3599 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3600 	CU_ASSERT_EQUAL(rc, 0);
3601 	num_completed = stub_complete_io(1);
3602 	CU_ASSERT_EQUAL(num_completed, 1);
3603 	CU_ASSERT(g_io_done == true);
3604 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3605 	/* Check that the iov has been set up */
3606 	CU_ASSERT(iov.iov_base == g_zcopy_write_buf);
3607 	CU_ASSERT(iov.iov_len == g_zcopy_write_buf_len);
3608 	/* Check that the bdev_io has been saved */
3609 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3610 	/* Now do the zcopy end for a write (commit=true) */
3611 	g_io_done = false;
3612 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3613 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3614 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3615 	CU_ASSERT_EQUAL(rc, 0);
3616 	num_completed = stub_complete_io(1);
3617 	CU_ASSERT_EQUAL(num_completed, 1);
3618 	CU_ASSERT(g_io_done == true);
3619 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3620 	/* Check the g_zcopy are reset by io_done */
3621 	CU_ASSERT(g_zcopy_write_buf == NULL);
3622 	CU_ASSERT(g_zcopy_write_buf_len == 0);
3623 	/* Check that io_done has freed the g_zcopy_bdev_io */
3624 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3625 
3626 	/* Check the zcopy read buffer has not been touched which
3627 	 * ensures that the correct buffers were used.
3628 	 */
3629 	CU_ASSERT(g_zcopy_read_buf == (void *) 0x1122334455667788UL);
3630 	CU_ASSERT(g_zcopy_read_buf_len == (uint32_t) -1);
3631 
3632 	spdk_put_io_channel(ioch);
3633 	spdk_bdev_close(desc);
3634 	free_bdev(bdev);
3635 	spdk_bdev_finish(bdev_fini_cb, NULL);
3636 	poll_threads();
3637 }
3638 
3639 static void
3640 bdev_zcopy_read(void)
3641 {
3642 	struct spdk_bdev *bdev;
3643 	struct spdk_bdev_desc *desc = NULL;
3644 	struct spdk_io_channel *ioch;
3645 	struct ut_expected_io *expected_io;
3646 	uint64_t offset, num_blocks;
3647 	uint32_t num_completed;
3648 	char aa_buf[512];
3649 	struct iovec iov;
3650 	int rc;
3651 	const bool populate = true;
3652 	const bool commit = false;
3653 
3654 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3655 
3656 	spdk_bdev_initialize(bdev_init_cb, NULL);
3657 	bdev = allocate_bdev("bdev");
3658 
3659 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3660 	CU_ASSERT_EQUAL(rc, 0);
3661 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3662 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3663 	ioch = spdk_bdev_get_io_channel(desc);
3664 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3665 
3666 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3667 
3668 	offset = 50;
3669 	num_blocks = 1;
3670 	iov.iov_base = NULL;
3671 	iov.iov_len = 0;
3672 
3673 	g_zcopy_write_buf = (void *) 0x1122334455667788UL;
3674 	g_zcopy_write_buf_len = (uint32_t) -1;
3675 
3676 	/* Do a zcopy start for a read (populate=true) */
3677 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3678 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3679 	g_io_done = false;
3680 	g_zcopy_read_buf = aa_buf;
3681 	g_zcopy_read_buf_len = sizeof(aa_buf);
3682 	g_zcopy_bdev_io = NULL;
3683 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3684 	CU_ASSERT_EQUAL(rc, 0);
3685 	num_completed = stub_complete_io(1);
3686 	CU_ASSERT_EQUAL(num_completed, 1);
3687 	CU_ASSERT(g_io_done == true);
3688 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3689 	/* Check that the iov has been set up */
3690 	CU_ASSERT(iov.iov_base == g_zcopy_read_buf);
3691 	CU_ASSERT(iov.iov_len == g_zcopy_read_buf_len);
3692 	/* Check that the bdev_io has been saved */
3693 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3694 
3695 	/* Now do the zcopy end for a read (commit=false) */
3696 	g_io_done = false;
3697 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3698 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3699 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3700 	CU_ASSERT_EQUAL(rc, 0);
3701 	num_completed = stub_complete_io(1);
3702 	CU_ASSERT_EQUAL(num_completed, 1);
3703 	CU_ASSERT(g_io_done == true);
3704 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3705 	/* Check the g_zcopy are reset by io_done */
3706 	CU_ASSERT(g_zcopy_read_buf == NULL);
3707 	CU_ASSERT(g_zcopy_read_buf_len == 0);
3708 	/* Check that io_done has freed the g_zcopy_bdev_io */
3709 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3710 
3711 	/* Check the zcopy write buffer has not been touched which
3712 	 * ensures that the correct buffers were used.
3713 	 */
3714 	CU_ASSERT(g_zcopy_write_buf == (void *) 0x1122334455667788UL);
3715 	CU_ASSERT(g_zcopy_write_buf_len == (uint32_t) -1);
3716 
3717 	spdk_put_io_channel(ioch);
3718 	spdk_bdev_close(desc);
3719 	free_bdev(bdev);
3720 	spdk_bdev_finish(bdev_fini_cb, NULL);
3721 	poll_threads();
3722 }
3723 
3724 static void
3725 bdev_open_while_hotremove(void)
3726 {
3727 	struct spdk_bdev *bdev;
3728 	struct spdk_bdev_desc *desc[2] = {};
3729 	int rc;
3730 
3731 	bdev = allocate_bdev("bdev");
3732 
3733 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[0]);
3734 	CU_ASSERT(rc == 0);
3735 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
3736 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc[0]));
3737 
3738 	spdk_bdev_unregister(bdev, NULL, NULL);
3739 
3740 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[1]);
3741 	CU_ASSERT(rc == -ENODEV);
3742 	SPDK_CU_ASSERT_FATAL(desc[1] == NULL);
3743 
3744 	spdk_bdev_close(desc[0]);
3745 	free_bdev(bdev);
3746 }
3747 
3748 static void
3749 bdev_close_while_hotremove(void)
3750 {
3751 	struct spdk_bdev *bdev;
3752 	struct spdk_bdev_desc *desc = NULL;
3753 	int rc = 0;
3754 
3755 	bdev = allocate_bdev("bdev");
3756 
3757 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc, &desc);
3758 	CU_ASSERT_EQUAL(rc, 0);
3759 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3760 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3761 
3762 	/* Simulate hot-unplug by unregistering bdev */
3763 	g_event_type1 = 0xFF;
3764 	g_unregister_arg = NULL;
3765 	g_unregister_rc = -1;
3766 	spdk_bdev_unregister(bdev, bdev_unregister_cb, (void *)0x12345678);
3767 	/* Close device while remove event is in flight */
3768 	spdk_bdev_close(desc);
3769 
3770 	/* Ensure that unregister callback is delayed */
3771 	CU_ASSERT_EQUAL(g_unregister_arg, NULL);
3772 	CU_ASSERT_EQUAL(g_unregister_rc, -1);
3773 
3774 	poll_threads();
3775 
3776 	/* Event callback shall not be issued because device was closed */
3777 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
3778 	/* Unregister callback is issued */
3779 	CU_ASSERT_EQUAL(g_unregister_arg, (void *)0x12345678);
3780 	CU_ASSERT_EQUAL(g_unregister_rc, 0);
3781 
3782 	free_bdev(bdev);
3783 }
3784 
3785 static void
3786 bdev_open_ext(void)
3787 {
3788 	struct spdk_bdev *bdev;
3789 	struct spdk_bdev_desc *desc1 = NULL;
3790 	struct spdk_bdev_desc *desc2 = NULL;
3791 	int rc = 0;
3792 
3793 	bdev = allocate_bdev("bdev");
3794 
3795 	rc = spdk_bdev_open_ext("bdev", true, NULL, NULL, &desc1);
3796 	CU_ASSERT_EQUAL(rc, -EINVAL);
3797 
3798 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc1, &desc1);
3799 	CU_ASSERT_EQUAL(rc, 0);
3800 
3801 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb2, &desc2, &desc2);
3802 	CU_ASSERT_EQUAL(rc, 0);
3803 
3804 	g_event_type1 = 0xFF;
3805 	g_event_type2 = 0xFF;
3806 
3807 	/* Simulate hot-unplug by unregistering bdev */
3808 	spdk_bdev_unregister(bdev, NULL, NULL);
3809 	poll_threads();
3810 
3811 	/* Check if correct events have been triggered in event callback fn */
3812 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_REMOVE);
3813 	CU_ASSERT_EQUAL(g_event_type2, SPDK_BDEV_EVENT_REMOVE);
3814 
3815 	free_bdev(bdev);
3816 	poll_threads();
3817 }
3818 
3819 struct timeout_io_cb_arg {
3820 	struct iovec iov;
3821 	uint8_t type;
3822 };
3823 
3824 static int
3825 bdev_channel_count_submitted_io(struct spdk_bdev_channel *ch)
3826 {
3827 	struct spdk_bdev_io *bdev_io;
3828 	int n = 0;
3829 
3830 	if (!ch) {
3831 		return -1;
3832 	}
3833 
3834 	TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
3835 		n++;
3836 	}
3837 
3838 	return n;
3839 }
3840 
3841 static void
3842 bdev_channel_io_timeout_cb(void *cb_arg, struct spdk_bdev_io *bdev_io)
3843 {
3844 	struct timeout_io_cb_arg *ctx = cb_arg;
3845 
3846 	ctx->type = bdev_io->type;
3847 	ctx->iov.iov_base = bdev_io->iov.iov_base;
3848 	ctx->iov.iov_len = bdev_io->iov.iov_len;
3849 }
3850 
3851 static void
3852 bdev_set_io_timeout(void)
3853 {
3854 	struct spdk_bdev *bdev;
3855 	struct spdk_bdev_desc *desc = NULL;
3856 	struct spdk_io_channel *io_ch = NULL;
3857 	struct spdk_bdev_channel *bdev_ch = NULL;
3858 	struct timeout_io_cb_arg cb_arg;
3859 
3860 	spdk_bdev_initialize(bdev_init_cb, NULL);
3861 
3862 	bdev = allocate_bdev("bdev");
3863 
3864 	CU_ASSERT(spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc) == 0);
3865 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3866 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3867 
3868 	io_ch = spdk_bdev_get_io_channel(desc);
3869 	CU_ASSERT(io_ch != NULL);
3870 
3871 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
3872 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
3873 
3874 	/* This is the part1.
3875 	 * We will check the bdev_ch->io_submitted list
3876 	 * TO make sure that it can link IOs and only the user submitted IOs
3877 	 */
3878 	CU_ASSERT(spdk_bdev_read(desc, io_ch, (void *)0x1000, 0, 4096, io_done, NULL) == 0);
3879 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3880 	CU_ASSERT(spdk_bdev_write(desc, io_ch, (void *)0x2000, 0, 4096, io_done, NULL) == 0);
3881 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3882 	stub_complete_io(1);
3883 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3884 	stub_complete_io(1);
3885 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3886 
3887 	/* Split IO */
3888 	bdev->optimal_io_boundary = 16;
3889 	bdev->split_on_optimal_io_boundary = true;
3890 
3891 	/* Now test that a single-vector command is split correctly.
3892 	 * Offset 14, length 8, payload 0xF000
3893 	 *  Child - Offset 14, length 2, payload 0xF000
3894 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
3895 	 *
3896 	 * Set up the expected values before calling spdk_bdev_read_blocks
3897 	 */
3898 	CU_ASSERT(spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3899 	/* We count all submitted IOs including IO that are generated by splitting. */
3900 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 3);
3901 	stub_complete_io(1);
3902 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3903 	stub_complete_io(1);
3904 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3905 
3906 	/* Also include the reset IO */
3907 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3908 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3909 	poll_threads();
3910 	stub_complete_io(1);
3911 	poll_threads();
3912 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3913 
3914 	/* This is part2
3915 	 * Test the desc timeout poller register
3916 	 */
3917 
3918 	/* Successfully set the timeout */
3919 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3920 	CU_ASSERT(desc->io_timeout_poller != NULL);
3921 	CU_ASSERT(desc->timeout_in_sec == 30);
3922 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3923 	CU_ASSERT(desc->cb_arg == &cb_arg);
3924 
3925 	/* Change the timeout limit */
3926 	CU_ASSERT(spdk_bdev_set_timeout(desc, 20, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3927 	CU_ASSERT(desc->io_timeout_poller != NULL);
3928 	CU_ASSERT(desc->timeout_in_sec == 20);
3929 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3930 	CU_ASSERT(desc->cb_arg == &cb_arg);
3931 
3932 	/* Disable the timeout */
3933 	CU_ASSERT(spdk_bdev_set_timeout(desc, 0, NULL, NULL) == 0);
3934 	CU_ASSERT(desc->io_timeout_poller == NULL);
3935 
3936 	/* This the part3
3937 	 * We will test to catch timeout IO and check whether the IO is
3938 	 * the submitted one.
3939 	 */
3940 	memset(&cb_arg, 0, sizeof(cb_arg));
3941 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3942 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0x1000, 0, 1, io_done, NULL) == 0);
3943 
3944 	/* Don't reach the limit */
3945 	spdk_delay_us(15 * spdk_get_ticks_hz());
3946 	poll_threads();
3947 	CU_ASSERT(cb_arg.type == 0);
3948 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3949 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3950 
3951 	/* 15 + 15 = 30 reach the limit */
3952 	spdk_delay_us(15 * spdk_get_ticks_hz());
3953 	poll_threads();
3954 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3955 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x1000);
3956 	CU_ASSERT(cb_arg.iov.iov_len == 1 * bdev->blocklen);
3957 	stub_complete_io(1);
3958 
3959 	/* Use the same split IO above and check the IO */
3960 	memset(&cb_arg, 0, sizeof(cb_arg));
3961 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3962 
3963 	/* The first child complete in time */
3964 	spdk_delay_us(15 * spdk_get_ticks_hz());
3965 	poll_threads();
3966 	stub_complete_io(1);
3967 	CU_ASSERT(cb_arg.type == 0);
3968 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3969 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3970 
3971 	/* The second child reach the limit */
3972 	spdk_delay_us(15 * spdk_get_ticks_hz());
3973 	poll_threads();
3974 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3975 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0xF000);
3976 	CU_ASSERT(cb_arg.iov.iov_len == 8 * bdev->blocklen);
3977 	stub_complete_io(1);
3978 
3979 	/* Also include the reset IO */
3980 	memset(&cb_arg, 0, sizeof(cb_arg));
3981 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3982 	spdk_delay_us(30 * spdk_get_ticks_hz());
3983 	poll_threads();
3984 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_RESET);
3985 	stub_complete_io(1);
3986 	poll_threads();
3987 
3988 	spdk_put_io_channel(io_ch);
3989 	spdk_bdev_close(desc);
3990 	free_bdev(bdev);
3991 	spdk_bdev_finish(bdev_fini_cb, NULL);
3992 	poll_threads();
3993 }
3994 
3995 static void
3996 lba_range_overlap(void)
3997 {
3998 	struct lba_range r1, r2;
3999 
4000 	r1.offset = 100;
4001 	r1.length = 50;
4002 
4003 	r2.offset = 0;
4004 	r2.length = 1;
4005 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
4006 
4007 	r2.offset = 0;
4008 	r2.length = 100;
4009 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
4010 
4011 	r2.offset = 0;
4012 	r2.length = 110;
4013 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
4014 
4015 	r2.offset = 100;
4016 	r2.length = 10;
4017 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
4018 
4019 	r2.offset = 110;
4020 	r2.length = 20;
4021 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
4022 
4023 	r2.offset = 140;
4024 	r2.length = 150;
4025 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
4026 
4027 	r2.offset = 130;
4028 	r2.length = 200;
4029 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
4030 
4031 	r2.offset = 150;
4032 	r2.length = 100;
4033 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
4034 
4035 	r2.offset = 110;
4036 	r2.length = 0;
4037 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
4038 }
4039 
4040 static bool g_lock_lba_range_done;
4041 static bool g_unlock_lba_range_done;
4042 
4043 static void
4044 lock_lba_range_done(void *ctx, int status)
4045 {
4046 	g_lock_lba_range_done = true;
4047 }
4048 
4049 static void
4050 unlock_lba_range_done(void *ctx, int status)
4051 {
4052 	g_unlock_lba_range_done = true;
4053 }
4054 
4055 static void
4056 lock_lba_range_check_ranges(void)
4057 {
4058 	struct spdk_bdev *bdev;
4059 	struct spdk_bdev_desc *desc = NULL;
4060 	struct spdk_io_channel *io_ch;
4061 	struct spdk_bdev_channel *channel;
4062 	struct lba_range *range;
4063 	int ctx1;
4064 	int rc;
4065 
4066 	spdk_bdev_initialize(bdev_init_cb, NULL);
4067 
4068 	bdev = allocate_bdev("bdev0");
4069 
4070 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4071 	CU_ASSERT(rc == 0);
4072 	CU_ASSERT(desc != NULL);
4073 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4074 	io_ch = spdk_bdev_get_io_channel(desc);
4075 	CU_ASSERT(io_ch != NULL);
4076 	channel = spdk_io_channel_get_ctx(io_ch);
4077 
4078 	g_lock_lba_range_done = false;
4079 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4080 	CU_ASSERT(rc == 0);
4081 	poll_threads();
4082 
4083 	CU_ASSERT(g_lock_lba_range_done == true);
4084 	range = TAILQ_FIRST(&channel->locked_ranges);
4085 	SPDK_CU_ASSERT_FATAL(range != NULL);
4086 	CU_ASSERT(range->offset == 20);
4087 	CU_ASSERT(range->length == 10);
4088 	CU_ASSERT(range->owner_ch == channel);
4089 
4090 	/* Unlocks must exactly match a lock. */
4091 	g_unlock_lba_range_done = false;
4092 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 1, unlock_lba_range_done, &ctx1);
4093 	CU_ASSERT(rc == -EINVAL);
4094 	CU_ASSERT(g_unlock_lba_range_done == false);
4095 
4096 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4097 	CU_ASSERT(rc == 0);
4098 	spdk_delay_us(100);
4099 	poll_threads();
4100 
4101 	CU_ASSERT(g_unlock_lba_range_done == true);
4102 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4103 
4104 	spdk_put_io_channel(io_ch);
4105 	spdk_bdev_close(desc);
4106 	free_bdev(bdev);
4107 	spdk_bdev_finish(bdev_fini_cb, NULL);
4108 	poll_threads();
4109 }
4110 
4111 static void
4112 lock_lba_range_with_io_outstanding(void)
4113 {
4114 	struct spdk_bdev *bdev;
4115 	struct spdk_bdev_desc *desc = NULL;
4116 	struct spdk_io_channel *io_ch;
4117 	struct spdk_bdev_channel *channel;
4118 	struct lba_range *range;
4119 	char buf[4096];
4120 	int ctx1;
4121 	int rc;
4122 
4123 	spdk_bdev_initialize(bdev_init_cb, NULL);
4124 
4125 	bdev = allocate_bdev("bdev0");
4126 
4127 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4128 	CU_ASSERT(rc == 0);
4129 	CU_ASSERT(desc != NULL);
4130 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4131 	io_ch = spdk_bdev_get_io_channel(desc);
4132 	CU_ASSERT(io_ch != NULL);
4133 	channel = spdk_io_channel_get_ctx(io_ch);
4134 
4135 	g_io_done = false;
4136 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4137 	CU_ASSERT(rc == 0);
4138 
4139 	g_lock_lba_range_done = false;
4140 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4141 	CU_ASSERT(rc == 0);
4142 	poll_threads();
4143 
4144 	/* The lock should immediately become valid, since there are no outstanding
4145 	 * write I/O.
4146 	 */
4147 	CU_ASSERT(g_io_done == false);
4148 	CU_ASSERT(g_lock_lba_range_done == true);
4149 	range = TAILQ_FIRST(&channel->locked_ranges);
4150 	SPDK_CU_ASSERT_FATAL(range != NULL);
4151 	CU_ASSERT(range->offset == 20);
4152 	CU_ASSERT(range->length == 10);
4153 	CU_ASSERT(range->owner_ch == channel);
4154 	CU_ASSERT(range->locked_ctx == &ctx1);
4155 
4156 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4157 	CU_ASSERT(rc == 0);
4158 	stub_complete_io(1);
4159 	spdk_delay_us(100);
4160 	poll_threads();
4161 
4162 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4163 
4164 	/* Now try again, but with a write I/O. */
4165 	g_io_done = false;
4166 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4167 	CU_ASSERT(rc == 0);
4168 
4169 	g_lock_lba_range_done = false;
4170 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4171 	CU_ASSERT(rc == 0);
4172 	poll_threads();
4173 
4174 	/* The lock should not be fully valid yet, since a write I/O is outstanding.
4175 	 * But note that the range should be on the channel's locked_list, to make sure no
4176 	 * new write I/O are started.
4177 	 */
4178 	CU_ASSERT(g_io_done == false);
4179 	CU_ASSERT(g_lock_lba_range_done == false);
4180 	range = TAILQ_FIRST(&channel->locked_ranges);
4181 	SPDK_CU_ASSERT_FATAL(range != NULL);
4182 	CU_ASSERT(range->offset == 20);
4183 	CU_ASSERT(range->length == 10);
4184 
4185 	/* Complete the write I/O.  This should make the lock valid (checked by confirming
4186 	 * our callback was invoked).
4187 	 */
4188 	stub_complete_io(1);
4189 	spdk_delay_us(100);
4190 	poll_threads();
4191 	CU_ASSERT(g_io_done == true);
4192 	CU_ASSERT(g_lock_lba_range_done == true);
4193 
4194 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4195 	CU_ASSERT(rc == 0);
4196 	poll_threads();
4197 
4198 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4199 
4200 	spdk_put_io_channel(io_ch);
4201 	spdk_bdev_close(desc);
4202 	free_bdev(bdev);
4203 	spdk_bdev_finish(bdev_fini_cb, NULL);
4204 	poll_threads();
4205 }
4206 
4207 static void
4208 lock_lba_range_overlapped(void)
4209 {
4210 	struct spdk_bdev *bdev;
4211 	struct spdk_bdev_desc *desc = NULL;
4212 	struct spdk_io_channel *io_ch;
4213 	struct spdk_bdev_channel *channel;
4214 	struct lba_range *range;
4215 	int ctx1;
4216 	int rc;
4217 
4218 	spdk_bdev_initialize(bdev_init_cb, NULL);
4219 
4220 	bdev = allocate_bdev("bdev0");
4221 
4222 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4223 	CU_ASSERT(rc == 0);
4224 	CU_ASSERT(desc != NULL);
4225 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4226 	io_ch = spdk_bdev_get_io_channel(desc);
4227 	CU_ASSERT(io_ch != NULL);
4228 	channel = spdk_io_channel_get_ctx(io_ch);
4229 
4230 	/* Lock range 20-29. */
4231 	g_lock_lba_range_done = false;
4232 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4233 	CU_ASSERT(rc == 0);
4234 	poll_threads();
4235 
4236 	CU_ASSERT(g_lock_lba_range_done == true);
4237 	range = TAILQ_FIRST(&channel->locked_ranges);
4238 	SPDK_CU_ASSERT_FATAL(range != NULL);
4239 	CU_ASSERT(range->offset == 20);
4240 	CU_ASSERT(range->length == 10);
4241 
4242 	/* Try to lock range 25-39.  It should not lock immediately, since it overlaps with
4243 	 * 20-29.
4244 	 */
4245 	g_lock_lba_range_done = false;
4246 	rc = bdev_lock_lba_range(desc, io_ch, 25, 15, lock_lba_range_done, &ctx1);
4247 	CU_ASSERT(rc == 0);
4248 	poll_threads();
4249 
4250 	CU_ASSERT(g_lock_lba_range_done == false);
4251 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4252 	SPDK_CU_ASSERT_FATAL(range != NULL);
4253 	CU_ASSERT(range->offset == 25);
4254 	CU_ASSERT(range->length == 15);
4255 
4256 	/* Unlock 20-29.  This should result in range 25-39 now getting locked since it
4257 	 * no longer overlaps with an active lock.
4258 	 */
4259 	g_unlock_lba_range_done = false;
4260 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4261 	CU_ASSERT(rc == 0);
4262 	poll_threads();
4263 
4264 	CU_ASSERT(g_unlock_lba_range_done == true);
4265 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4266 	range = TAILQ_FIRST(&channel->locked_ranges);
4267 	SPDK_CU_ASSERT_FATAL(range != NULL);
4268 	CU_ASSERT(range->offset == 25);
4269 	CU_ASSERT(range->length == 15);
4270 
4271 	/* Lock 40-59.  This should immediately lock since it does not overlap with the
4272 	 * currently active 25-39 lock.
4273 	 */
4274 	g_lock_lba_range_done = false;
4275 	rc = bdev_lock_lba_range(desc, io_ch, 40, 20, lock_lba_range_done, &ctx1);
4276 	CU_ASSERT(rc == 0);
4277 	poll_threads();
4278 
4279 	CU_ASSERT(g_lock_lba_range_done == true);
4280 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4281 	SPDK_CU_ASSERT_FATAL(range != NULL);
4282 	range = TAILQ_NEXT(range, tailq);
4283 	SPDK_CU_ASSERT_FATAL(range != NULL);
4284 	CU_ASSERT(range->offset == 40);
4285 	CU_ASSERT(range->length == 20);
4286 
4287 	/* Try to lock 35-44.  Note that this overlaps with both 25-39 and 40-59. */
4288 	g_lock_lba_range_done = false;
4289 	rc = bdev_lock_lba_range(desc, io_ch, 35, 10, lock_lba_range_done, &ctx1);
4290 	CU_ASSERT(rc == 0);
4291 	poll_threads();
4292 
4293 	CU_ASSERT(g_lock_lba_range_done == false);
4294 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4295 	SPDK_CU_ASSERT_FATAL(range != NULL);
4296 	CU_ASSERT(range->offset == 35);
4297 	CU_ASSERT(range->length == 10);
4298 
4299 	/* Unlock 25-39.  Make sure that 35-44 is still in the pending list, since
4300 	 * the 40-59 lock is still active.
4301 	 */
4302 	g_unlock_lba_range_done = false;
4303 	rc = bdev_unlock_lba_range(desc, io_ch, 25, 15, unlock_lba_range_done, &ctx1);
4304 	CU_ASSERT(rc == 0);
4305 	poll_threads();
4306 
4307 	CU_ASSERT(g_unlock_lba_range_done == true);
4308 	CU_ASSERT(g_lock_lba_range_done == false);
4309 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4310 	SPDK_CU_ASSERT_FATAL(range != NULL);
4311 	CU_ASSERT(range->offset == 35);
4312 	CU_ASSERT(range->length == 10);
4313 
4314 	/* Unlock 40-59.  This should result in 35-44 now getting locked, since there are
4315 	 * no longer any active overlapping locks.
4316 	 */
4317 	g_unlock_lba_range_done = false;
4318 	rc = bdev_unlock_lba_range(desc, io_ch, 40, 20, unlock_lba_range_done, &ctx1);
4319 	CU_ASSERT(rc == 0);
4320 	poll_threads();
4321 
4322 	CU_ASSERT(g_unlock_lba_range_done == true);
4323 	CU_ASSERT(g_lock_lba_range_done == true);
4324 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4325 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4326 	SPDK_CU_ASSERT_FATAL(range != NULL);
4327 	CU_ASSERT(range->offset == 35);
4328 	CU_ASSERT(range->length == 10);
4329 
4330 	/* Finally, unlock 35-44. */
4331 	g_unlock_lba_range_done = false;
4332 	rc = bdev_unlock_lba_range(desc, io_ch, 35, 10, unlock_lba_range_done, &ctx1);
4333 	CU_ASSERT(rc == 0);
4334 	poll_threads();
4335 
4336 	CU_ASSERT(g_unlock_lba_range_done == true);
4337 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.locked_ranges));
4338 
4339 	spdk_put_io_channel(io_ch);
4340 	spdk_bdev_close(desc);
4341 	free_bdev(bdev);
4342 	spdk_bdev_finish(bdev_fini_cb, NULL);
4343 	poll_threads();
4344 }
4345 
4346 static void
4347 abort_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
4348 {
4349 	g_abort_done = true;
4350 	g_abort_status = bdev_io->internal.status;
4351 	spdk_bdev_free_io(bdev_io);
4352 }
4353 
4354 static void
4355 bdev_io_abort(void)
4356 {
4357 	struct spdk_bdev *bdev;
4358 	struct spdk_bdev_desc *desc = NULL;
4359 	struct spdk_io_channel *io_ch;
4360 	struct spdk_bdev_channel *channel;
4361 	struct spdk_bdev_mgmt_channel *mgmt_ch;
4362 	struct spdk_bdev_opts bdev_opts = {};
4363 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
4364 	uint64_t io_ctx1 = 0, io_ctx2 = 0, i;
4365 	int rc;
4366 
4367 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4368 	bdev_opts.bdev_io_pool_size = 7;
4369 	bdev_opts.bdev_io_cache_size = 2;
4370 
4371 	rc = spdk_bdev_set_opts(&bdev_opts);
4372 	CU_ASSERT(rc == 0);
4373 	spdk_bdev_initialize(bdev_init_cb, NULL);
4374 
4375 	bdev = allocate_bdev("bdev0");
4376 
4377 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4378 	CU_ASSERT(rc == 0);
4379 	CU_ASSERT(desc != NULL);
4380 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4381 	io_ch = spdk_bdev_get_io_channel(desc);
4382 	CU_ASSERT(io_ch != NULL);
4383 	channel = spdk_io_channel_get_ctx(io_ch);
4384 	mgmt_ch = channel->shared_resource->mgmt_ch;
4385 
4386 	g_abort_done = false;
4387 
4388 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, false);
4389 
4390 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4391 	CU_ASSERT(rc == -ENOTSUP);
4392 
4393 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, true);
4394 
4395 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx2, abort_done, NULL);
4396 	CU_ASSERT(rc == 0);
4397 	CU_ASSERT(g_abort_done == true);
4398 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_FAILED);
4399 
4400 	/* Test the case that the target I/O was successfully aborted. */
4401 	g_io_done = false;
4402 
4403 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4404 	CU_ASSERT(rc == 0);
4405 	CU_ASSERT(g_io_done == false);
4406 
4407 	g_abort_done = false;
4408 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4409 
4410 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4411 	CU_ASSERT(rc == 0);
4412 	CU_ASSERT(g_io_done == true);
4413 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4414 	stub_complete_io(1);
4415 	CU_ASSERT(g_abort_done == true);
4416 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4417 
4418 	/* Test the case that the target I/O was not aborted because it completed
4419 	 * in the middle of execution of the abort.
4420 	 */
4421 	g_io_done = false;
4422 
4423 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4424 	CU_ASSERT(rc == 0);
4425 	CU_ASSERT(g_io_done == false);
4426 
4427 	g_abort_done = false;
4428 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4429 
4430 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4431 	CU_ASSERT(rc == 0);
4432 	CU_ASSERT(g_io_done == false);
4433 
4434 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4435 	stub_complete_io(1);
4436 	CU_ASSERT(g_io_done == true);
4437 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4438 
4439 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4440 	stub_complete_io(1);
4441 	CU_ASSERT(g_abort_done == true);
4442 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4443 
4444 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4445 
4446 	bdev->optimal_io_boundary = 16;
4447 	bdev->split_on_optimal_io_boundary = true;
4448 
4449 	/* Test that a single-vector command which is split is aborted correctly.
4450 	 * Offset 14, length 8, payload 0xF000
4451 	 *  Child - Offset 14, length 2, payload 0xF000
4452 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
4453 	 */
4454 	g_io_done = false;
4455 
4456 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, &io_ctx1);
4457 	CU_ASSERT(rc == 0);
4458 	CU_ASSERT(g_io_done == false);
4459 
4460 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4461 
4462 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4463 
4464 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4465 	CU_ASSERT(rc == 0);
4466 	CU_ASSERT(g_io_done == true);
4467 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4468 	stub_complete_io(2);
4469 	CU_ASSERT(g_abort_done == true);
4470 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4471 
4472 	/* Test that a multi-vector command that needs to be split by strip and then
4473 	 * needs to be split is aborted correctly. Abort is requested before the second
4474 	 * child I/O was submitted. The parent I/O should complete with failure without
4475 	 * submitting the second child I/O.
4476 	 */
4477 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
4478 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
4479 		iov[i].iov_len = 512;
4480 	}
4481 
4482 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
4483 	g_io_done = false;
4484 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
4485 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, &io_ctx1);
4486 	CU_ASSERT(rc == 0);
4487 	CU_ASSERT(g_io_done == false);
4488 
4489 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4490 
4491 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4492 
4493 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4494 	CU_ASSERT(rc == 0);
4495 	CU_ASSERT(g_io_done == true);
4496 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4497 	stub_complete_io(1);
4498 	CU_ASSERT(g_abort_done == true);
4499 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4500 
4501 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4502 
4503 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4504 
4505 	bdev->optimal_io_boundary = 16;
4506 	g_io_done = false;
4507 
4508 	/* Test that a ingle-vector command which is split is aborted correctly.
4509 	 * Differently from the above, the child abort request will be submitted
4510 	 * sequentially due to the capacity of spdk_bdev_io.
4511 	 */
4512 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 50, io_done, &io_ctx1);
4513 	CU_ASSERT(rc == 0);
4514 	CU_ASSERT(g_io_done == false);
4515 
4516 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4517 
4518 	g_abort_done = false;
4519 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4520 
4521 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4522 	CU_ASSERT(rc == 0);
4523 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
4524 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4525 
4526 	stub_complete_io(1);
4527 	CU_ASSERT(g_io_done == true);
4528 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4529 	stub_complete_io(3);
4530 	CU_ASSERT(g_abort_done == true);
4531 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4532 
4533 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4534 
4535 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4536 
4537 	spdk_put_io_channel(io_ch);
4538 	spdk_bdev_close(desc);
4539 	free_bdev(bdev);
4540 	spdk_bdev_finish(bdev_fini_cb, NULL);
4541 	poll_threads();
4542 }
4543 
4544 static void
4545 bdev_unmap(void)
4546 {
4547 	struct spdk_bdev *bdev;
4548 	struct spdk_bdev_desc *desc = NULL;
4549 	struct spdk_io_channel *ioch;
4550 	struct spdk_bdev_channel *bdev_ch;
4551 	struct ut_expected_io *expected_io;
4552 	struct spdk_bdev_opts bdev_opts = {};
4553 	uint32_t i, num_outstanding;
4554 	uint64_t offset, num_blocks, max_unmap_blocks, num_children;
4555 	int rc;
4556 
4557 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4558 	bdev_opts.bdev_io_pool_size = 512;
4559 	bdev_opts.bdev_io_cache_size = 64;
4560 	rc = spdk_bdev_set_opts(&bdev_opts);
4561 	CU_ASSERT(rc == 0);
4562 
4563 	spdk_bdev_initialize(bdev_init_cb, NULL);
4564 	bdev = allocate_bdev("bdev");
4565 
4566 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4567 	CU_ASSERT_EQUAL(rc, 0);
4568 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4569 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4570 	ioch = spdk_bdev_get_io_channel(desc);
4571 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4572 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4573 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4574 
4575 	fn_table.submit_request = stub_submit_request;
4576 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4577 
4578 	/* Case 1: First test the request won't be split */
4579 	num_blocks = 32;
4580 
4581 	g_io_done = false;
4582 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 0, num_blocks, 0);
4583 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4584 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4585 	CU_ASSERT_EQUAL(rc, 0);
4586 	CU_ASSERT(g_io_done == false);
4587 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4588 	stub_complete_io(1);
4589 	CU_ASSERT(g_io_done == true);
4590 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4591 
4592 	/* Case 2: Test the split with 2 children requests */
4593 	bdev->max_unmap = 8;
4594 	bdev->max_unmap_segments = 2;
4595 	max_unmap_blocks = bdev->max_unmap * bdev->max_unmap_segments;
4596 	num_blocks = max_unmap_blocks * 2;
4597 	offset = 0;
4598 
4599 	g_io_done = false;
4600 	for (i = 0; i < 2; i++) {
4601 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4602 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4603 		offset += max_unmap_blocks;
4604 	}
4605 
4606 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4607 	CU_ASSERT_EQUAL(rc, 0);
4608 	CU_ASSERT(g_io_done == false);
4609 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4610 	stub_complete_io(2);
4611 	CU_ASSERT(g_io_done == true);
4612 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4613 
4614 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4615 	num_children = 15;
4616 	num_blocks = max_unmap_blocks * num_children;
4617 	g_io_done = false;
4618 	offset = 0;
4619 	for (i = 0; i < num_children; i++) {
4620 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4621 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4622 		offset += max_unmap_blocks;
4623 	}
4624 
4625 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4626 	CU_ASSERT_EQUAL(rc, 0);
4627 	CU_ASSERT(g_io_done == false);
4628 
4629 	while (num_children > 0) {
4630 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4631 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4632 		stub_complete_io(num_outstanding);
4633 		num_children -= num_outstanding;
4634 	}
4635 	CU_ASSERT(g_io_done == true);
4636 
4637 	spdk_put_io_channel(ioch);
4638 	spdk_bdev_close(desc);
4639 	free_bdev(bdev);
4640 	spdk_bdev_finish(bdev_fini_cb, NULL);
4641 	poll_threads();
4642 }
4643 
4644 static void
4645 bdev_write_zeroes_split_test(void)
4646 {
4647 	struct spdk_bdev *bdev;
4648 	struct spdk_bdev_desc *desc = NULL;
4649 	struct spdk_io_channel *ioch;
4650 	struct spdk_bdev_channel *bdev_ch;
4651 	struct ut_expected_io *expected_io;
4652 	struct spdk_bdev_opts bdev_opts = {};
4653 	uint32_t i, num_outstanding;
4654 	uint64_t offset, num_blocks, max_write_zeroes_blocks, num_children;
4655 	int rc;
4656 
4657 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4658 	bdev_opts.bdev_io_pool_size = 512;
4659 	bdev_opts.bdev_io_cache_size = 64;
4660 	rc = spdk_bdev_set_opts(&bdev_opts);
4661 	CU_ASSERT(rc == 0);
4662 
4663 	spdk_bdev_initialize(bdev_init_cb, NULL);
4664 	bdev = allocate_bdev("bdev");
4665 
4666 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4667 	CU_ASSERT_EQUAL(rc, 0);
4668 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4669 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4670 	ioch = spdk_bdev_get_io_channel(desc);
4671 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4672 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4673 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4674 
4675 	fn_table.submit_request = stub_submit_request;
4676 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4677 
4678 	/* Case 1: First test the request won't be split */
4679 	num_blocks = 32;
4680 
4681 	g_io_done = false;
4682 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
4683 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4684 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4685 	CU_ASSERT_EQUAL(rc, 0);
4686 	CU_ASSERT(g_io_done == false);
4687 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4688 	stub_complete_io(1);
4689 	CU_ASSERT(g_io_done == true);
4690 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4691 
4692 	/* Case 2: Test the split with 2 children requests */
4693 	max_write_zeroes_blocks = 8;
4694 	bdev->max_write_zeroes = max_write_zeroes_blocks;
4695 	num_blocks = max_write_zeroes_blocks * 2;
4696 	offset = 0;
4697 
4698 	g_io_done = false;
4699 	for (i = 0; i < 2; i++) {
4700 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4701 						   0);
4702 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4703 		offset += max_write_zeroes_blocks;
4704 	}
4705 
4706 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4707 	CU_ASSERT_EQUAL(rc, 0);
4708 	CU_ASSERT(g_io_done == false);
4709 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4710 	stub_complete_io(2);
4711 	CU_ASSERT(g_io_done == true);
4712 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4713 
4714 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4715 	num_children = 15;
4716 	num_blocks = max_write_zeroes_blocks * num_children;
4717 	g_io_done = false;
4718 	offset = 0;
4719 	for (i = 0; i < num_children; i++) {
4720 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4721 						   0);
4722 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4723 		offset += max_write_zeroes_blocks;
4724 	}
4725 
4726 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4727 	CU_ASSERT_EQUAL(rc, 0);
4728 	CU_ASSERT(g_io_done == false);
4729 
4730 	while (num_children > 0) {
4731 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4732 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4733 		stub_complete_io(num_outstanding);
4734 		num_children -= num_outstanding;
4735 	}
4736 	CU_ASSERT(g_io_done == true);
4737 
4738 	spdk_put_io_channel(ioch);
4739 	spdk_bdev_close(desc);
4740 	free_bdev(bdev);
4741 	spdk_bdev_finish(bdev_fini_cb, NULL);
4742 	poll_threads();
4743 }
4744 
4745 static void
4746 bdev_set_options_test(void)
4747 {
4748 	struct spdk_bdev_opts bdev_opts = {};
4749 	int rc;
4750 
4751 	/* Case1: Do not set opts_size */
4752 	rc = spdk_bdev_set_opts(&bdev_opts);
4753 	CU_ASSERT(rc == -1);
4754 
4755 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4756 	bdev_opts.bdev_io_pool_size = 4;
4757 	bdev_opts.bdev_io_cache_size = 2;
4758 	bdev_opts.small_buf_pool_size = 4;
4759 
4760 	/* Case 2: Do not set valid small_buf_pool_size and large_buf_pool_size */
4761 	rc = spdk_bdev_set_opts(&bdev_opts);
4762 	CU_ASSERT(rc == -1);
4763 
4764 	/* Case 3: Do not set valid large_buf_pool_size */
4765 	bdev_opts.small_buf_pool_size = BUF_SMALL_POOL_SIZE;
4766 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE - 1;
4767 	rc = spdk_bdev_set_opts(&bdev_opts);
4768 	CU_ASSERT(rc == -1);
4769 
4770 	/* Case4: set valid large buf_pool_size */
4771 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE;
4772 	rc = spdk_bdev_set_opts(&bdev_opts);
4773 	CU_ASSERT(rc == 0);
4774 
4775 	/* Case5: Set different valid value for small and large buf pool */
4776 	bdev_opts.large_buf_pool_size = BUF_SMALL_POOL_SIZE + 3;
4777 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE + 3;
4778 	rc = spdk_bdev_set_opts(&bdev_opts);
4779 	CU_ASSERT(rc == 0);
4780 }
4781 
4782 static uint64_t
4783 get_ns_time(void)
4784 {
4785 	int rc;
4786 	struct timespec ts;
4787 
4788 	rc = clock_gettime(CLOCK_MONOTONIC, &ts);
4789 	CU_ASSERT(rc == 0);
4790 	return ts.tv_sec * 1000 * 1000 * 1000 + ts.tv_nsec;
4791 }
4792 
4793 static int
4794 rb_tree_get_height(struct spdk_bdev_name *bdev_name)
4795 {
4796 	int h1, h2;
4797 
4798 	if (bdev_name == NULL) {
4799 		return -1;
4800 	} else {
4801 		h1 = rb_tree_get_height(RB_LEFT(bdev_name, node));
4802 		h2 = rb_tree_get_height(RB_RIGHT(bdev_name, node));
4803 
4804 		return spdk_max(h1, h2) + 1;
4805 	}
4806 }
4807 
4808 static void
4809 bdev_multi_allocation(void)
4810 {
4811 	const int max_bdev_num = 1024 * 16;
4812 	char name[max_bdev_num][16];
4813 	char noexist_name[] = "invalid_bdev";
4814 	struct spdk_bdev *bdev[max_bdev_num];
4815 	int i, j;
4816 	uint64_t last_time;
4817 	int bdev_num;
4818 	int height;
4819 
4820 	for (j = 0; j < max_bdev_num; j++) {
4821 		snprintf(name[j], sizeof(name[j]), "bdev%d", j);
4822 	}
4823 
4824 	for (i = 0; i < 16; i++) {
4825 		last_time = get_ns_time();
4826 		bdev_num = 1024 * (i + 1);
4827 		for (j = 0; j < bdev_num; j++) {
4828 			bdev[j] = allocate_bdev(name[j]);
4829 			height = rb_tree_get_height(&bdev[j]->internal.bdev_name);
4830 			CU_ASSERT(height <= (int)(spdk_u32log2(2 * j + 2)));
4831 		}
4832 		SPDK_NOTICELOG("alloc bdev num %d takes %" PRIu64 " ms\n", bdev_num,
4833 			       (get_ns_time() - last_time) / 1000 / 1000);
4834 		for (j = 0; j < bdev_num; j++) {
4835 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) != NULL);
4836 		}
4837 		CU_ASSERT(spdk_bdev_get_by_name(noexist_name) == NULL);
4838 
4839 		for (j = 0; j < bdev_num; j++) {
4840 			free_bdev(bdev[j]);
4841 		}
4842 		for (j = 0; j < bdev_num; j++) {
4843 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) == NULL);
4844 		}
4845 	}
4846 }
4847 
4848 static struct spdk_memory_domain *g_bdev_memory_domain = (struct spdk_memory_domain *) 0xf00df00d;
4849 
4850 static int
4851 test_bdev_get_supported_dma_device_types_op(void *ctx, struct spdk_memory_domain **domains,
4852 		int array_size)
4853 {
4854 	if (array_size > 0 && domains) {
4855 		domains[0] = g_bdev_memory_domain;
4856 	}
4857 
4858 	return 1;
4859 }
4860 
4861 static void
4862 bdev_get_memory_domains(void)
4863 {
4864 	struct spdk_bdev_fn_table fn_table = {
4865 		.get_memory_domains = test_bdev_get_supported_dma_device_types_op
4866 	};
4867 	struct spdk_bdev bdev = { .fn_table = &fn_table };
4868 	struct spdk_memory_domain *domains[2] = {};
4869 	int rc;
4870 
4871 	/* bdev is NULL */
4872 	rc = spdk_bdev_get_memory_domains(NULL, domains, 2);
4873 	CU_ASSERT(rc == -EINVAL);
4874 
4875 	/* domains is NULL */
4876 	rc = spdk_bdev_get_memory_domains(&bdev, NULL, 2);
4877 	CU_ASSERT(rc == 1);
4878 
4879 	/* array size is 0 */
4880 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 0);
4881 	CU_ASSERT(rc == 1);
4882 
4883 	/* get_supported_dma_device_types op is set */
4884 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4885 	CU_ASSERT(rc == 1);
4886 	CU_ASSERT(domains[0] == g_bdev_memory_domain);
4887 
4888 	/* get_supported_dma_device_types op is not set */
4889 	fn_table.get_memory_domains = NULL;
4890 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4891 	CU_ASSERT(rc == 0);
4892 }
4893 
4894 static void
4895 bdev_writev_readv_ext(void)
4896 {
4897 	struct spdk_bdev *bdev;
4898 	struct spdk_bdev_desc *desc = NULL;
4899 	struct spdk_io_channel *io_ch;
4900 	struct iovec iov = { .iov_base = (void *)0xbaaddead, .iov_len = 0x1000 };
4901 	struct ut_expected_io *expected_io;
4902 	struct spdk_bdev_ext_io_opts ext_io_opts = {
4903 		.metadata = (void *)0xFF000000
4904 	};
4905 	int rc;
4906 
4907 	spdk_bdev_initialize(bdev_init_cb, NULL);
4908 
4909 	bdev = allocate_bdev("bdev0");
4910 	bdev->md_interleave = false;
4911 	bdev->md_len = 8;
4912 
4913 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4914 	CU_ASSERT(rc == 0);
4915 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4916 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4917 	io_ch = spdk_bdev_get_io_channel(desc);
4918 	CU_ASSERT(io_ch != NULL);
4919 
4920 	g_io_done = false;
4921 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 32, 14, 1);
4922 	expected_io->md_buf = ext_io_opts.metadata;
4923 	expected_io->ext_io_opts = &ext_io_opts;
4924 	ut_expected_io_set_iov(expected_io, 0, iov.iov_base, iov.iov_len);
4925 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4926 
4927 	rc = spdk_bdev_readv_blocks_ext(desc, io_ch, &iov, 1, 32, 14, io_done, NULL, &ext_io_opts);
4928 
4929 	CU_ASSERT(rc == 0);
4930 	CU_ASSERT(g_io_done == false);
4931 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4932 	stub_complete_io(1);
4933 	CU_ASSERT(g_io_done == true);
4934 
4935 	g_io_done = false;
4936 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 1);
4937 	expected_io->md_buf = ext_io_opts.metadata;
4938 	expected_io->ext_io_opts = &ext_io_opts;
4939 	ut_expected_io_set_iov(expected_io, 0, iov.iov_base, iov.iov_len);
4940 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4941 
4942 	rc = spdk_bdev_writev_blocks_ext(desc, io_ch, &iov, 1, 32, 14, io_done, NULL, &ext_io_opts);
4943 
4944 	CU_ASSERT(rc == 0);
4945 	CU_ASSERT(g_io_done == false);
4946 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4947 	stub_complete_io(1);
4948 	CU_ASSERT(g_io_done == true);
4949 
4950 	spdk_put_io_channel(io_ch);
4951 	spdk_bdev_close(desc);
4952 	free_bdev(bdev);
4953 	spdk_bdev_finish(bdev_fini_cb, NULL);
4954 	poll_threads();
4955 }
4956 
4957 static void
4958 bdev_register_uuid_alias(void)
4959 {
4960 	struct spdk_bdev *bdev, *second;
4961 	char uuid[SPDK_UUID_STRING_LEN];
4962 	int rc;
4963 
4964 	spdk_bdev_initialize(bdev_init_cb, NULL);
4965 	bdev = allocate_bdev("bdev0");
4966 
4967 	/* Make sure an UUID was generated  */
4968 	CU_ASSERT_FALSE(spdk_mem_all_zero(&bdev->uuid, sizeof(bdev->uuid)));
4969 
4970 	/* Check that an UUID alias was registered */
4971 	spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
4972 	CU_ASSERT_EQUAL(spdk_bdev_get_by_name(uuid), bdev);
4973 
4974 	/* Unregister the bdev */
4975 	spdk_bdev_unregister(bdev, NULL, NULL);
4976 	poll_threads();
4977 	CU_ASSERT_PTR_NULL(spdk_bdev_get_by_name(uuid));
4978 
4979 	/* Check the same, but this time register the bdev with non-zero UUID */
4980 	rc = spdk_bdev_register(bdev);
4981 	CU_ASSERT_EQUAL(rc, 0);
4982 	CU_ASSERT_EQUAL(spdk_bdev_get_by_name(uuid), bdev);
4983 
4984 	/* Unregister the bdev */
4985 	spdk_bdev_unregister(bdev, NULL, NULL);
4986 	poll_threads();
4987 	CU_ASSERT_PTR_NULL(spdk_bdev_get_by_name(uuid));
4988 
4989 	/* Regiser the bdev using UUID as the name */
4990 	bdev->name = uuid;
4991 	rc = spdk_bdev_register(bdev);
4992 	CU_ASSERT_EQUAL(rc, 0);
4993 	CU_ASSERT_EQUAL(spdk_bdev_get_by_name(uuid), bdev);
4994 
4995 	/* Unregister the bdev */
4996 	spdk_bdev_unregister(bdev, NULL, NULL);
4997 	poll_threads();
4998 	CU_ASSERT_PTR_NULL(spdk_bdev_get_by_name(uuid));
4999 
5000 	/* Check that it's not possible to register two bdevs with the same UUIDs */
5001 	bdev->name = "bdev0";
5002 	second = allocate_bdev("bdev1");
5003 	spdk_uuid_copy(&bdev->uuid, &second->uuid);
5004 	rc = spdk_bdev_register(bdev);
5005 	CU_ASSERT_EQUAL(rc, -EEXIST);
5006 
5007 	/* Regenerate the UUID and re-check */
5008 	spdk_uuid_generate(&bdev->uuid);
5009 	rc = spdk_bdev_register(bdev);
5010 	CU_ASSERT_EQUAL(rc, 0);
5011 
5012 	/* And check that both bdevs can be retrieved through their UUIDs */
5013 	spdk_uuid_fmt_lower(uuid, sizeof(uuid), &bdev->uuid);
5014 	CU_ASSERT_EQUAL(spdk_bdev_get_by_name(uuid), bdev);
5015 	spdk_uuid_fmt_lower(uuid, sizeof(uuid), &second->uuid);
5016 	CU_ASSERT_EQUAL(spdk_bdev_get_by_name(uuid), second);
5017 
5018 	free_bdev(second);
5019 	free_bdev(bdev);
5020 	spdk_bdev_finish(bdev_fini_cb, NULL);
5021 	poll_threads();
5022 }
5023 
5024 int
5025 main(int argc, char **argv)
5026 {
5027 	CU_pSuite		suite = NULL;
5028 	unsigned int		num_failures;
5029 
5030 	CU_set_error_action(CUEA_ABORT);
5031 	CU_initialize_registry();
5032 
5033 	suite = CU_add_suite("bdev", null_init, null_clean);
5034 
5035 	CU_ADD_TEST(suite, bytes_to_blocks_test);
5036 	CU_ADD_TEST(suite, num_blocks_test);
5037 	CU_ADD_TEST(suite, io_valid_test);
5038 	CU_ADD_TEST(suite, open_write_test);
5039 	CU_ADD_TEST(suite, claim_test);
5040 	CU_ADD_TEST(suite, alias_add_del_test);
5041 	CU_ADD_TEST(suite, get_device_stat_test);
5042 	CU_ADD_TEST(suite, bdev_io_types_test);
5043 	CU_ADD_TEST(suite, bdev_io_wait_test);
5044 	CU_ADD_TEST(suite, bdev_io_spans_split_test);
5045 	CU_ADD_TEST(suite, bdev_io_boundary_split_test);
5046 	CU_ADD_TEST(suite, bdev_io_max_size_and_segment_split_test);
5047 	CU_ADD_TEST(suite, bdev_io_mix_split_test);
5048 	CU_ADD_TEST(suite, bdev_io_split_with_io_wait);
5049 	CU_ADD_TEST(suite, bdev_io_alignment_with_boundary);
5050 	CU_ADD_TEST(suite, bdev_io_alignment);
5051 	CU_ADD_TEST(suite, bdev_histograms);
5052 	CU_ADD_TEST(suite, bdev_write_zeroes);
5053 	CU_ADD_TEST(suite, bdev_compare_and_write);
5054 	CU_ADD_TEST(suite, bdev_compare);
5055 	CU_ADD_TEST(suite, bdev_zcopy_write);
5056 	CU_ADD_TEST(suite, bdev_zcopy_read);
5057 	CU_ADD_TEST(suite, bdev_open_while_hotremove);
5058 	CU_ADD_TEST(suite, bdev_close_while_hotremove);
5059 	CU_ADD_TEST(suite, bdev_open_ext);
5060 	CU_ADD_TEST(suite, bdev_set_io_timeout);
5061 	CU_ADD_TEST(suite, lba_range_overlap);
5062 	CU_ADD_TEST(suite, lock_lba_range_check_ranges);
5063 	CU_ADD_TEST(suite, lock_lba_range_with_io_outstanding);
5064 	CU_ADD_TEST(suite, lock_lba_range_overlapped);
5065 	CU_ADD_TEST(suite, bdev_io_abort);
5066 	CU_ADD_TEST(suite, bdev_unmap);
5067 	CU_ADD_TEST(suite, bdev_write_zeroes_split_test);
5068 	CU_ADD_TEST(suite, bdev_set_options_test);
5069 	CU_ADD_TEST(suite, bdev_multi_allocation);
5070 	CU_ADD_TEST(suite, bdev_get_memory_domains);
5071 	CU_ADD_TEST(suite, bdev_writev_readv_ext);
5072 	CU_ADD_TEST(suite, bdev_register_uuid_alias);
5073 
5074 	allocate_cores(1);
5075 	allocate_threads(1);
5076 	set_thread(0);
5077 
5078 	CU_basic_set_mode(CU_BRM_VERBOSE);
5079 	CU_basic_run_tests();
5080 	num_failures = CU_get_number_of_failures();
5081 	CU_cleanup_registry();
5082 
5083 	free_threads();
5084 	free_cores();
5085 
5086 	return num_failures;
5087 }
5088