xref: /spdk/test/unit/lib/bdev/bdev.c/bdev_ut.c (revision 8afa746b4d4dd44195380dd8b5e3265c514c1ca5)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation. All rights reserved.
5  *   Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved.
6  *   Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "spdk_cunit.h"
36 
37 #include "common/lib/ut_multithread.c"
38 #include "unit/lib/json_mock.c"
39 
40 #include "spdk/config.h"
41 /* HACK: disable VTune integration so the unit test doesn't need VTune headers and libs to build */
42 #undef SPDK_CONFIG_VTUNE
43 
44 #include "bdev/bdev.c"
45 
46 DEFINE_STUB(spdk_notify_send, uint64_t, (const char *type, const char *ctx), 0);
47 DEFINE_STUB(spdk_notify_type_register, struct spdk_notify_type *, (const char *type), NULL);
48 
49 int g_status;
50 int g_count;
51 enum spdk_bdev_event_type g_event_type1;
52 enum spdk_bdev_event_type g_event_type2;
53 struct spdk_histogram_data *g_histogram;
54 void *g_unregister_arg;
55 int g_unregister_rc;
56 
57 void
58 spdk_scsi_nvme_translate(const struct spdk_bdev_io *bdev_io,
59 			 int *sc, int *sk, int *asc, int *ascq)
60 {
61 }
62 
63 static int
64 null_init(void)
65 {
66 	return 0;
67 }
68 
69 static int
70 null_clean(void)
71 {
72 	return 0;
73 }
74 
75 static int
76 stub_destruct(void *ctx)
77 {
78 	return 0;
79 }
80 
81 struct ut_expected_io {
82 	uint8_t				type;
83 	uint64_t			offset;
84 	uint64_t			length;
85 	int				iovcnt;
86 	struct iovec			iov[BDEV_IO_NUM_CHILD_IOV];
87 	void				*md_buf;
88 	struct spdk_bdev_ext_io_opts	*ext_io_opts;
89 	TAILQ_ENTRY(ut_expected_io)	link;
90 };
91 
92 struct bdev_ut_channel {
93 	TAILQ_HEAD(, spdk_bdev_io)	outstanding_io;
94 	uint32_t			outstanding_io_count;
95 	TAILQ_HEAD(, ut_expected_io)	expected_io;
96 };
97 
98 static bool g_io_done;
99 static struct spdk_bdev_io *g_bdev_io;
100 static enum spdk_bdev_io_status g_io_status;
101 static enum spdk_bdev_io_status g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
102 static uint32_t g_bdev_ut_io_device;
103 static struct bdev_ut_channel *g_bdev_ut_channel;
104 static void *g_compare_read_buf;
105 static uint32_t g_compare_read_buf_len;
106 static void *g_compare_write_buf;
107 static uint32_t g_compare_write_buf_len;
108 static bool g_abort_done;
109 static enum spdk_bdev_io_status g_abort_status;
110 static void *g_zcopy_read_buf;
111 static uint32_t g_zcopy_read_buf_len;
112 static void *g_zcopy_write_buf;
113 static uint32_t g_zcopy_write_buf_len;
114 static struct spdk_bdev_io *g_zcopy_bdev_io;
115 
116 static struct ut_expected_io *
117 ut_alloc_expected_io(uint8_t type, uint64_t offset, uint64_t length, int iovcnt)
118 {
119 	struct ut_expected_io *expected_io;
120 
121 	expected_io = calloc(1, sizeof(*expected_io));
122 	SPDK_CU_ASSERT_FATAL(expected_io != NULL);
123 
124 	expected_io->type = type;
125 	expected_io->offset = offset;
126 	expected_io->length = length;
127 	expected_io->iovcnt = iovcnt;
128 
129 	return expected_io;
130 }
131 
132 static void
133 ut_expected_io_set_iov(struct ut_expected_io *expected_io, int pos, void *base, size_t len)
134 {
135 	expected_io->iov[pos].iov_base = base;
136 	expected_io->iov[pos].iov_len = len;
137 }
138 
139 static void
140 stub_submit_request(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
141 {
142 	struct bdev_ut_channel *ch = spdk_io_channel_get_ctx(_ch);
143 	struct ut_expected_io *expected_io;
144 	struct iovec *iov, *expected_iov;
145 	struct spdk_bdev_io *bio_to_abort;
146 	int i;
147 
148 	g_bdev_io = bdev_io;
149 
150 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_READ) {
151 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
152 
153 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
154 		CU_ASSERT(g_compare_read_buf_len == len);
155 		memcpy(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len);
156 	}
157 
158 	if (g_compare_write_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
159 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
160 
161 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
162 		CU_ASSERT(g_compare_write_buf_len == len);
163 		memcpy(g_compare_write_buf, bdev_io->u.bdev.iovs[0].iov_base, len);
164 	}
165 
166 	if (g_compare_read_buf && bdev_io->type == SPDK_BDEV_IO_TYPE_COMPARE) {
167 		uint32_t len = bdev_io->u.bdev.iovs[0].iov_len;
168 
169 		CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
170 		CU_ASSERT(g_compare_read_buf_len == len);
171 		if (memcmp(bdev_io->u.bdev.iovs[0].iov_base, g_compare_read_buf, len)) {
172 			g_io_exp_status = SPDK_BDEV_IO_STATUS_MISCOMPARE;
173 		}
174 	}
175 
176 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ABORT) {
177 		if (g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS) {
178 			TAILQ_FOREACH(bio_to_abort, &ch->outstanding_io, module_link) {
179 				if (bio_to_abort == bdev_io->u.abort.bio_to_abort) {
180 					TAILQ_REMOVE(&ch->outstanding_io, bio_to_abort, module_link);
181 					ch->outstanding_io_count--;
182 					spdk_bdev_io_complete(bio_to_abort, SPDK_BDEV_IO_STATUS_FAILED);
183 					break;
184 				}
185 			}
186 		}
187 	}
188 
189 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) {
190 		if (bdev_io->u.bdev.zcopy.start) {
191 			g_zcopy_bdev_io = bdev_io;
192 			if (bdev_io->u.bdev.zcopy.populate) {
193 				/* Start of a read */
194 				CU_ASSERT(g_zcopy_read_buf != NULL);
195 				CU_ASSERT(g_zcopy_read_buf_len > 0);
196 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_read_buf;
197 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_read_buf_len;
198 				bdev_io->u.bdev.iovcnt = 1;
199 			} else {
200 				/* Start of a write */
201 				CU_ASSERT(g_zcopy_write_buf != NULL);
202 				CU_ASSERT(g_zcopy_write_buf_len > 0);
203 				bdev_io->u.bdev.iovs[0].iov_base = g_zcopy_write_buf;
204 				bdev_io->u.bdev.iovs[0].iov_len = g_zcopy_write_buf_len;
205 				bdev_io->u.bdev.iovcnt = 1;
206 			}
207 		} else {
208 			if (bdev_io->u.bdev.zcopy.commit) {
209 				/* End of write */
210 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_write_buf);
211 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_write_buf_len);
212 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
213 				g_zcopy_write_buf = NULL;
214 				g_zcopy_write_buf_len = 0;
215 			} else {
216 				/* End of read */
217 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_base == g_zcopy_read_buf);
218 				CU_ASSERT(bdev_io->u.bdev.iovs[0].iov_len == g_zcopy_read_buf_len);
219 				CU_ASSERT(bdev_io->u.bdev.iovcnt == 1);
220 				g_zcopy_read_buf = NULL;
221 				g_zcopy_read_buf_len = 0;
222 			}
223 		}
224 	}
225 
226 	TAILQ_INSERT_TAIL(&ch->outstanding_io, bdev_io, module_link);
227 	ch->outstanding_io_count++;
228 
229 	expected_io = TAILQ_FIRST(&ch->expected_io);
230 	if (expected_io == NULL) {
231 		return;
232 	}
233 	TAILQ_REMOVE(&ch->expected_io, expected_io, link);
234 
235 	if (expected_io->type != SPDK_BDEV_IO_TYPE_INVALID) {
236 		CU_ASSERT(bdev_io->type == expected_io->type);
237 	}
238 
239 	if (expected_io->md_buf != NULL) {
240 		CU_ASSERT(expected_io->md_buf == bdev_io->u.bdev.md_buf);
241 	}
242 
243 	if (expected_io->length == 0) {
244 		free(expected_io);
245 		return;
246 	}
247 
248 	CU_ASSERT(expected_io->offset == bdev_io->u.bdev.offset_blocks);
249 	CU_ASSERT(expected_io->length = bdev_io->u.bdev.num_blocks);
250 
251 	if (expected_io->iovcnt == 0) {
252 		free(expected_io);
253 		/* UNMAP, WRITE_ZEROES and FLUSH don't have iovs, so we can just return now. */
254 		return;
255 	}
256 
257 	CU_ASSERT(expected_io->iovcnt == bdev_io->u.bdev.iovcnt);
258 	for (i = 0; i < expected_io->iovcnt; i++) {
259 		iov = &bdev_io->u.bdev.iovs[i];
260 		expected_iov = &expected_io->iov[i];
261 		CU_ASSERT(iov->iov_len == expected_iov->iov_len);
262 		CU_ASSERT(iov->iov_base == expected_iov->iov_base);
263 	}
264 
265 	if (expected_io->ext_io_opts) {
266 		CU_ASSERT(expected_io->ext_io_opts == bdev_io->internal.ext_opts)
267 	}
268 
269 	free(expected_io);
270 }
271 
272 static void
273 stub_submit_request_get_buf_cb(struct spdk_io_channel *_ch,
274 			       struct spdk_bdev_io *bdev_io, bool success)
275 {
276 	CU_ASSERT(success == true);
277 
278 	stub_submit_request(_ch, bdev_io);
279 }
280 
281 static void
282 stub_submit_request_get_buf(struct spdk_io_channel *_ch, struct spdk_bdev_io *bdev_io)
283 {
284 	spdk_bdev_io_get_buf(bdev_io, stub_submit_request_get_buf_cb,
285 			     bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen);
286 }
287 
288 static uint32_t
289 stub_complete_io(uint32_t num_to_complete)
290 {
291 	struct bdev_ut_channel *ch = g_bdev_ut_channel;
292 	struct spdk_bdev_io *bdev_io;
293 	static enum spdk_bdev_io_status io_status;
294 	uint32_t num_completed = 0;
295 
296 	while (num_completed < num_to_complete) {
297 		if (TAILQ_EMPTY(&ch->outstanding_io)) {
298 			break;
299 		}
300 		bdev_io = TAILQ_FIRST(&ch->outstanding_io);
301 		TAILQ_REMOVE(&ch->outstanding_io, bdev_io, module_link);
302 		ch->outstanding_io_count--;
303 		io_status = g_io_exp_status == SPDK_BDEV_IO_STATUS_SUCCESS ? SPDK_BDEV_IO_STATUS_SUCCESS :
304 			    g_io_exp_status;
305 		spdk_bdev_io_complete(bdev_io, io_status);
306 		num_completed++;
307 	}
308 
309 	return num_completed;
310 }
311 
312 static struct spdk_io_channel *
313 bdev_ut_get_io_channel(void *ctx)
314 {
315 	return spdk_get_io_channel(&g_bdev_ut_io_device);
316 }
317 
318 static bool g_io_types_supported[SPDK_BDEV_NUM_IO_TYPES] = {
319 	[SPDK_BDEV_IO_TYPE_READ]		= true,
320 	[SPDK_BDEV_IO_TYPE_WRITE]		= true,
321 	[SPDK_BDEV_IO_TYPE_COMPARE]		= true,
322 	[SPDK_BDEV_IO_TYPE_UNMAP]		= true,
323 	[SPDK_BDEV_IO_TYPE_FLUSH]		= true,
324 	[SPDK_BDEV_IO_TYPE_RESET]		= true,
325 	[SPDK_BDEV_IO_TYPE_NVME_ADMIN]		= true,
326 	[SPDK_BDEV_IO_TYPE_NVME_IO]		= true,
327 	[SPDK_BDEV_IO_TYPE_NVME_IO_MD]		= true,
328 	[SPDK_BDEV_IO_TYPE_WRITE_ZEROES]	= true,
329 	[SPDK_BDEV_IO_TYPE_ZCOPY]		= true,
330 	[SPDK_BDEV_IO_TYPE_ABORT]		= true,
331 };
332 
333 static void
334 ut_enable_io_type(enum spdk_bdev_io_type io_type, bool enable)
335 {
336 	g_io_types_supported[io_type] = enable;
337 }
338 
339 static bool
340 stub_io_type_supported(void *_bdev, enum spdk_bdev_io_type io_type)
341 {
342 	return g_io_types_supported[io_type];
343 }
344 
345 static struct spdk_bdev_fn_table fn_table = {
346 	.destruct = stub_destruct,
347 	.submit_request = stub_submit_request,
348 	.get_io_channel = bdev_ut_get_io_channel,
349 	.io_type_supported = stub_io_type_supported,
350 };
351 
352 static int
353 bdev_ut_create_ch(void *io_device, void *ctx_buf)
354 {
355 	struct bdev_ut_channel *ch = ctx_buf;
356 
357 	CU_ASSERT(g_bdev_ut_channel == NULL);
358 	g_bdev_ut_channel = ch;
359 
360 	TAILQ_INIT(&ch->outstanding_io);
361 	ch->outstanding_io_count = 0;
362 	TAILQ_INIT(&ch->expected_io);
363 	return 0;
364 }
365 
366 static void
367 bdev_ut_destroy_ch(void *io_device, void *ctx_buf)
368 {
369 	CU_ASSERT(g_bdev_ut_channel != NULL);
370 	g_bdev_ut_channel = NULL;
371 }
372 
373 struct spdk_bdev_module bdev_ut_if;
374 
375 static int
376 bdev_ut_module_init(void)
377 {
378 	spdk_io_device_register(&g_bdev_ut_io_device, bdev_ut_create_ch, bdev_ut_destroy_ch,
379 				sizeof(struct bdev_ut_channel), NULL);
380 	spdk_bdev_module_init_done(&bdev_ut_if);
381 	return 0;
382 }
383 
384 static void
385 bdev_ut_module_fini(void)
386 {
387 	spdk_io_device_unregister(&g_bdev_ut_io_device, NULL);
388 }
389 
390 struct spdk_bdev_module bdev_ut_if = {
391 	.name = "bdev_ut",
392 	.module_init = bdev_ut_module_init,
393 	.module_fini = bdev_ut_module_fini,
394 	.async_init = true,
395 };
396 
397 static void vbdev_ut_examine(struct spdk_bdev *bdev);
398 
399 static int
400 vbdev_ut_module_init(void)
401 {
402 	return 0;
403 }
404 
405 static void
406 vbdev_ut_module_fini(void)
407 {
408 }
409 
410 struct spdk_bdev_module vbdev_ut_if = {
411 	.name = "vbdev_ut",
412 	.module_init = vbdev_ut_module_init,
413 	.module_fini = vbdev_ut_module_fini,
414 	.examine_config = vbdev_ut_examine,
415 };
416 
417 SPDK_BDEV_MODULE_REGISTER(bdev_ut, &bdev_ut_if)
418 SPDK_BDEV_MODULE_REGISTER(vbdev_ut, &vbdev_ut_if)
419 
420 static void
421 vbdev_ut_examine(struct spdk_bdev *bdev)
422 {
423 	spdk_bdev_module_examine_done(&vbdev_ut_if);
424 }
425 
426 static struct spdk_bdev *
427 allocate_bdev(char *name)
428 {
429 	struct spdk_bdev *bdev;
430 	int rc;
431 
432 	bdev = calloc(1, sizeof(*bdev));
433 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
434 
435 	bdev->name = name;
436 	bdev->fn_table = &fn_table;
437 	bdev->module = &bdev_ut_if;
438 	bdev->blockcnt = 1024;
439 	bdev->blocklen = 512;
440 
441 	rc = spdk_bdev_register(bdev);
442 	CU_ASSERT(rc == 0);
443 
444 	return bdev;
445 }
446 
447 static struct spdk_bdev *
448 allocate_vbdev(char *name)
449 {
450 	struct spdk_bdev *bdev;
451 	int rc;
452 
453 	bdev = calloc(1, sizeof(*bdev));
454 	SPDK_CU_ASSERT_FATAL(bdev != NULL);
455 
456 	bdev->name = name;
457 	bdev->fn_table = &fn_table;
458 	bdev->module = &vbdev_ut_if;
459 
460 	rc = spdk_bdev_register(bdev);
461 	CU_ASSERT(rc == 0);
462 
463 	return bdev;
464 }
465 
466 static void
467 free_bdev(struct spdk_bdev *bdev)
468 {
469 	spdk_bdev_unregister(bdev, NULL, NULL);
470 	poll_threads();
471 	memset(bdev, 0xFF, sizeof(*bdev));
472 	free(bdev);
473 }
474 
475 static void
476 free_vbdev(struct spdk_bdev *bdev)
477 {
478 	spdk_bdev_unregister(bdev, NULL, NULL);
479 	poll_threads();
480 	memset(bdev, 0xFF, sizeof(*bdev));
481 	free(bdev);
482 }
483 
484 static void
485 get_device_stat_cb(struct spdk_bdev *bdev, struct spdk_bdev_io_stat *stat, void *cb_arg, int rc)
486 {
487 	const char *bdev_name;
488 
489 	CU_ASSERT(bdev != NULL);
490 	CU_ASSERT(rc == 0);
491 	bdev_name = spdk_bdev_get_name(bdev);
492 	CU_ASSERT_STRING_EQUAL(bdev_name, "bdev0");
493 
494 	free(stat);
495 
496 	*(bool *)cb_arg = true;
497 }
498 
499 static void
500 bdev_unregister_cb(void *cb_arg, int rc)
501 {
502 	g_unregister_arg = cb_arg;
503 	g_unregister_rc = rc;
504 }
505 
506 static void
507 bdev_ut_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
508 {
509 }
510 
511 static void
512 bdev_open_cb1(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
513 {
514 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
515 
516 	g_event_type1 = type;
517 	if (SPDK_BDEV_EVENT_REMOVE == type) {
518 		spdk_bdev_close(desc);
519 	}
520 }
521 
522 static void
523 bdev_open_cb2(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, void *event_ctx)
524 {
525 	struct spdk_bdev_desc *desc = *(struct spdk_bdev_desc **)event_ctx;
526 
527 	g_event_type2 = type;
528 	if (SPDK_BDEV_EVENT_REMOVE == type) {
529 		spdk_bdev_close(desc);
530 	}
531 }
532 
533 static void
534 get_device_stat_test(void)
535 {
536 	struct spdk_bdev *bdev;
537 	struct spdk_bdev_io_stat *stat;
538 	bool done;
539 
540 	bdev = allocate_bdev("bdev0");
541 	stat = calloc(1, sizeof(struct spdk_bdev_io_stat));
542 	if (stat == NULL) {
543 		free_bdev(bdev);
544 		return;
545 	}
546 
547 	done = false;
548 	spdk_bdev_get_device_stat(bdev, stat, get_device_stat_cb, &done);
549 	while (!done) { poll_threads(); }
550 
551 	free_bdev(bdev);
552 }
553 
554 static void
555 open_write_test(void)
556 {
557 	struct spdk_bdev *bdev[9];
558 	struct spdk_bdev_desc *desc[9] = {};
559 	int rc;
560 
561 	/*
562 	 * Create a tree of bdevs to test various open w/ write cases.
563 	 *
564 	 * bdev0 through bdev3 are physical block devices, such as NVMe
565 	 * namespaces or Ceph block devices.
566 	 *
567 	 * bdev4 is a virtual bdev with multiple base bdevs.  This models
568 	 * caching or RAID use cases.
569 	 *
570 	 * bdev5 through bdev7 are all virtual bdevs with the same base
571 	 * bdev (except bdev7). This models partitioning or logical volume
572 	 * use cases.
573 	 *
574 	 * bdev7 is a virtual bdev with multiple base bdevs. One of base bdevs
575 	 * (bdev2) is shared with other virtual bdevs: bdev5 and bdev6. This
576 	 * models caching, RAID, partitioning or logical volumes use cases.
577 	 *
578 	 * bdev8 is a virtual bdev with multiple base bdevs, but these
579 	 * base bdevs are themselves virtual bdevs.
580 	 *
581 	 *                bdev8
582 	 *                  |
583 	 *            +----------+
584 	 *            |          |
585 	 *          bdev4      bdev5   bdev6   bdev7
586 	 *            |          |       |       |
587 	 *        +---+---+      +---+   +   +---+---+
588 	 *        |       |           \  |  /         \
589 	 *      bdev0   bdev1          bdev2         bdev3
590 	 */
591 
592 	bdev[0] = allocate_bdev("bdev0");
593 	rc = spdk_bdev_module_claim_bdev(bdev[0], NULL, &bdev_ut_if);
594 	CU_ASSERT(rc == 0);
595 
596 	bdev[1] = allocate_bdev("bdev1");
597 	rc = spdk_bdev_module_claim_bdev(bdev[1], NULL, &bdev_ut_if);
598 	CU_ASSERT(rc == 0);
599 
600 	bdev[2] = allocate_bdev("bdev2");
601 	rc = spdk_bdev_module_claim_bdev(bdev[2], NULL, &bdev_ut_if);
602 	CU_ASSERT(rc == 0);
603 
604 	bdev[3] = allocate_bdev("bdev3");
605 	rc = spdk_bdev_module_claim_bdev(bdev[3], NULL, &bdev_ut_if);
606 	CU_ASSERT(rc == 0);
607 
608 	bdev[4] = allocate_vbdev("bdev4");
609 	rc = spdk_bdev_module_claim_bdev(bdev[4], NULL, &bdev_ut_if);
610 	CU_ASSERT(rc == 0);
611 
612 	bdev[5] = allocate_vbdev("bdev5");
613 	rc = spdk_bdev_module_claim_bdev(bdev[5], NULL, &bdev_ut_if);
614 	CU_ASSERT(rc == 0);
615 
616 	bdev[6] = allocate_vbdev("bdev6");
617 
618 	bdev[7] = allocate_vbdev("bdev7");
619 
620 	bdev[8] = allocate_vbdev("bdev8");
621 
622 	/* Open bdev0 read-only.  This should succeed. */
623 	rc = spdk_bdev_open_ext("bdev0", false, bdev_ut_event_cb, NULL, &desc[0]);
624 	CU_ASSERT(rc == 0);
625 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
626 	CU_ASSERT(bdev[0] == spdk_bdev_desc_get_bdev(desc[0]));
627 	spdk_bdev_close(desc[0]);
628 
629 	/*
630 	 * Open bdev1 read/write.  This should fail since bdev1 has been claimed
631 	 * by a vbdev module.
632 	 */
633 	rc = spdk_bdev_open_ext("bdev1", true, bdev_ut_event_cb, NULL, &desc[1]);
634 	CU_ASSERT(rc == -EPERM);
635 
636 	/*
637 	 * Open bdev4 read/write.  This should fail since bdev3 has been claimed
638 	 * by a vbdev module.
639 	 */
640 	rc = spdk_bdev_open_ext("bdev4", true, bdev_ut_event_cb, NULL, &desc[4]);
641 	CU_ASSERT(rc == -EPERM);
642 
643 	/* Open bdev4 read-only.  This should succeed. */
644 	rc = spdk_bdev_open_ext("bdev4", false, bdev_ut_event_cb, NULL, &desc[4]);
645 	CU_ASSERT(rc == 0);
646 	SPDK_CU_ASSERT_FATAL(desc[4] != NULL);
647 	CU_ASSERT(bdev[4] == spdk_bdev_desc_get_bdev(desc[4]));
648 	spdk_bdev_close(desc[4]);
649 
650 	/*
651 	 * Open bdev8 read/write.  This should succeed since it is a leaf
652 	 * bdev.
653 	 */
654 	rc = spdk_bdev_open_ext("bdev8", true, bdev_ut_event_cb, NULL, &desc[8]);
655 	CU_ASSERT(rc == 0);
656 	SPDK_CU_ASSERT_FATAL(desc[8] != NULL);
657 	CU_ASSERT(bdev[8] == spdk_bdev_desc_get_bdev(desc[8]));
658 	spdk_bdev_close(desc[8]);
659 
660 	/*
661 	 * Open bdev5 read/write.  This should fail since bdev4 has been claimed
662 	 * by a vbdev module.
663 	 */
664 	rc = spdk_bdev_open_ext("bdev5", true, bdev_ut_event_cb, NULL, &desc[5]);
665 	CU_ASSERT(rc == -EPERM);
666 
667 	/* Open bdev4 read-only.  This should succeed. */
668 	rc = spdk_bdev_open_ext("bdev5", false, bdev_ut_event_cb, NULL, &desc[5]);
669 	CU_ASSERT(rc == 0);
670 	SPDK_CU_ASSERT_FATAL(desc[5] != NULL);
671 	CU_ASSERT(bdev[5] == spdk_bdev_desc_get_bdev(desc[5]));
672 	spdk_bdev_close(desc[5]);
673 
674 	free_vbdev(bdev[8]);
675 
676 	free_vbdev(bdev[5]);
677 	free_vbdev(bdev[6]);
678 	free_vbdev(bdev[7]);
679 
680 	free_vbdev(bdev[4]);
681 
682 	free_bdev(bdev[0]);
683 	free_bdev(bdev[1]);
684 	free_bdev(bdev[2]);
685 	free_bdev(bdev[3]);
686 }
687 
688 static void
689 bytes_to_blocks_test(void)
690 {
691 	struct spdk_bdev bdev;
692 	uint64_t offset_blocks, num_blocks;
693 
694 	memset(&bdev, 0, sizeof(bdev));
695 
696 	bdev.blocklen = 512;
697 
698 	/* All parameters valid */
699 	offset_blocks = 0;
700 	num_blocks = 0;
701 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 1024, &num_blocks) == 0);
702 	CU_ASSERT(offset_blocks == 1);
703 	CU_ASSERT(num_blocks == 2);
704 
705 	/* Offset not a block multiple */
706 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 512, &num_blocks) != 0);
707 
708 	/* Length not a block multiple */
709 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 512, &offset_blocks, 3, &num_blocks) != 0);
710 
711 	/* In case blocklen not the power of two */
712 	bdev.blocklen = 100;
713 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 200, &num_blocks) == 0);
714 	CU_ASSERT(offset_blocks == 1);
715 	CU_ASSERT(num_blocks == 2);
716 
717 	/* Offset not a block multiple */
718 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 3, &offset_blocks, 100, &num_blocks) != 0);
719 
720 	/* Length not a block multiple */
721 	CU_ASSERT(bdev_bytes_to_blocks(&bdev, 100, &offset_blocks, 3, &num_blocks) != 0);
722 }
723 
724 static void
725 num_blocks_test(void)
726 {
727 	struct spdk_bdev bdev;
728 	struct spdk_bdev_desc *desc = NULL;
729 	int rc;
730 
731 	memset(&bdev, 0, sizeof(bdev));
732 	bdev.name = "num_blocks";
733 	bdev.fn_table = &fn_table;
734 	bdev.module = &bdev_ut_if;
735 	spdk_bdev_register(&bdev);
736 	spdk_bdev_notify_blockcnt_change(&bdev, 50);
737 
738 	/* Growing block number */
739 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 70) == 0);
740 	/* Shrinking block number */
741 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 30) == 0);
742 
743 	rc = spdk_bdev_open_ext("num_blocks", false, bdev_open_cb1, &desc, &desc);
744 	CU_ASSERT(rc == 0);
745 	SPDK_CU_ASSERT_FATAL(desc != NULL);
746 	CU_ASSERT(&bdev == spdk_bdev_desc_get_bdev(desc));
747 
748 	/* Growing block number */
749 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 80) == 0);
750 	/* Shrinking block number */
751 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 20) != 0);
752 
753 	g_event_type1 = 0xFF;
754 	/* Growing block number */
755 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 90) == 0);
756 
757 	poll_threads();
758 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_RESIZE);
759 
760 	g_event_type1 = 0xFF;
761 	/* Growing block number and closing */
762 	CU_ASSERT(spdk_bdev_notify_blockcnt_change(&bdev, 100) == 0);
763 
764 	spdk_bdev_close(desc);
765 	spdk_bdev_unregister(&bdev, NULL, NULL);
766 
767 	poll_threads();
768 
769 	/* Callback is not called for closed device */
770 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
771 }
772 
773 static void
774 io_valid_test(void)
775 {
776 	struct spdk_bdev bdev;
777 
778 	memset(&bdev, 0, sizeof(bdev));
779 
780 	bdev.blocklen = 512;
781 	CU_ASSERT(pthread_mutex_init(&bdev.internal.mutex, NULL) == 0);
782 
783 	spdk_bdev_notify_blockcnt_change(&bdev, 100);
784 
785 	/* All parameters valid */
786 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 1, 2) == true);
787 
788 	/* Last valid block */
789 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 1) == true);
790 
791 	/* Offset past end of bdev */
792 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 100, 1) == false);
793 
794 	/* Offset + length past end of bdev */
795 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 99, 2) == false);
796 
797 	/* Offset near end of uint64_t range (2^64 - 1) */
798 	CU_ASSERT(bdev_io_valid_blocks(&bdev, 18446744073709551615ULL, 1) == false);
799 
800 	CU_ASSERT(pthread_mutex_destroy(&bdev.internal.mutex) == 0);
801 }
802 
803 static void
804 alias_add_del_test(void)
805 {
806 	struct spdk_bdev *bdev[3];
807 	int rc;
808 
809 	/* Creating and registering bdevs */
810 	bdev[0] = allocate_bdev("bdev0");
811 	SPDK_CU_ASSERT_FATAL(bdev[0] != 0);
812 
813 	bdev[1] = allocate_bdev("bdev1");
814 	SPDK_CU_ASSERT_FATAL(bdev[1] != 0);
815 
816 	bdev[2] = allocate_bdev("bdev2");
817 	SPDK_CU_ASSERT_FATAL(bdev[2] != 0);
818 
819 	poll_threads();
820 
821 	/*
822 	 * Trying adding an alias identical to name.
823 	 * Alias is identical to name, so it can not be added to aliases list
824 	 */
825 	rc = spdk_bdev_alias_add(bdev[0], bdev[0]->name);
826 	CU_ASSERT(rc == -EEXIST);
827 
828 	/*
829 	 * Trying to add empty alias,
830 	 * this one should fail
831 	 */
832 	rc = spdk_bdev_alias_add(bdev[0], NULL);
833 	CU_ASSERT(rc == -EINVAL);
834 
835 	/* Trying adding same alias to two different registered bdevs */
836 
837 	/* Alias is used first time, so this one should pass */
838 	rc = spdk_bdev_alias_add(bdev[0], "proper alias 0");
839 	CU_ASSERT(rc == 0);
840 
841 	/* Alias was added to another bdev, so this one should fail */
842 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 0");
843 	CU_ASSERT(rc == -EEXIST);
844 
845 	/* Alias is used first time, so this one should pass */
846 	rc = spdk_bdev_alias_add(bdev[1], "proper alias 1");
847 	CU_ASSERT(rc == 0);
848 
849 	/* Trying removing an alias from registered bdevs */
850 
851 	/* Alias is not on a bdev aliases list, so this one should fail */
852 	rc = spdk_bdev_alias_del(bdev[0], "not existing");
853 	CU_ASSERT(rc == -ENOENT);
854 
855 	/* Alias is present on a bdev aliases list, so this one should pass */
856 	rc = spdk_bdev_alias_del(bdev[0], "proper alias 0");
857 	CU_ASSERT(rc == 0);
858 
859 	/* Alias is present on a bdev aliases list, so this one should pass */
860 	rc = spdk_bdev_alias_del(bdev[1], "proper alias 1");
861 	CU_ASSERT(rc == 0);
862 
863 	/* Trying to remove name instead of alias, so this one should fail, name cannot be changed or removed */
864 	rc = spdk_bdev_alias_del(bdev[0], bdev[0]->name);
865 	CU_ASSERT(rc != 0);
866 
867 	/* Trying to del all alias from empty alias list */
868 	spdk_bdev_alias_del_all(bdev[2]);
869 	SPDK_CU_ASSERT_FATAL(TAILQ_EMPTY(&bdev[2]->aliases));
870 
871 	/* Trying to del all alias from non-empty alias list */
872 	rc = spdk_bdev_alias_add(bdev[2], "alias0");
873 	CU_ASSERT(rc == 0);
874 	rc = spdk_bdev_alias_add(bdev[2], "alias1");
875 	CU_ASSERT(rc == 0);
876 	spdk_bdev_alias_del_all(bdev[2]);
877 	CU_ASSERT(TAILQ_EMPTY(&bdev[2]->aliases));
878 
879 	/* Unregister and free bdevs */
880 	spdk_bdev_unregister(bdev[0], NULL, NULL);
881 	spdk_bdev_unregister(bdev[1], NULL, NULL);
882 	spdk_bdev_unregister(bdev[2], NULL, NULL);
883 
884 	poll_threads();
885 
886 	free(bdev[0]);
887 	free(bdev[1]);
888 	free(bdev[2]);
889 }
890 
891 static void
892 io_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
893 {
894 	g_io_done = true;
895 	g_io_status = bdev_io->internal.status;
896 	if ((bdev_io->type == SPDK_BDEV_IO_TYPE_ZCOPY) &&
897 	    (bdev_io->u.bdev.zcopy.start)) {
898 		g_zcopy_bdev_io = bdev_io;
899 	} else {
900 		spdk_bdev_free_io(bdev_io);
901 		g_zcopy_bdev_io = NULL;
902 	}
903 }
904 
905 static void
906 bdev_init_cb(void *arg, int rc)
907 {
908 	CU_ASSERT(rc == 0);
909 }
910 
911 static void
912 bdev_fini_cb(void *arg)
913 {
914 }
915 
916 struct bdev_ut_io_wait_entry {
917 	struct spdk_bdev_io_wait_entry	entry;
918 	struct spdk_io_channel		*io_ch;
919 	struct spdk_bdev_desc		*desc;
920 	bool				submitted;
921 };
922 
923 static void
924 io_wait_cb(void *arg)
925 {
926 	struct bdev_ut_io_wait_entry *entry = arg;
927 	int rc;
928 
929 	rc = spdk_bdev_read_blocks(entry->desc, entry->io_ch, NULL, 0, 1, io_done, NULL);
930 	CU_ASSERT(rc == 0);
931 	entry->submitted = true;
932 }
933 
934 static void
935 bdev_io_types_test(void)
936 {
937 	struct spdk_bdev *bdev;
938 	struct spdk_bdev_desc *desc = NULL;
939 	struct spdk_io_channel *io_ch;
940 	struct spdk_bdev_opts bdev_opts = {};
941 	int rc;
942 
943 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
944 	bdev_opts.bdev_io_pool_size = 4;
945 	bdev_opts.bdev_io_cache_size = 2;
946 
947 	rc = spdk_bdev_set_opts(&bdev_opts);
948 	CU_ASSERT(rc == 0);
949 	spdk_bdev_initialize(bdev_init_cb, NULL);
950 	poll_threads();
951 
952 	bdev = allocate_bdev("bdev0");
953 
954 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
955 	CU_ASSERT(rc == 0);
956 	poll_threads();
957 	SPDK_CU_ASSERT_FATAL(desc != NULL);
958 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
959 	io_ch = spdk_bdev_get_io_channel(desc);
960 	CU_ASSERT(io_ch != NULL);
961 
962 	/* WRITE and WRITE ZEROES are not supported */
963 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
964 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, false);
965 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 0, 128, io_done, NULL);
966 	CU_ASSERT(rc == -ENOTSUP);
967 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
968 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE, true);
969 
970 	/* NVME_IO, NVME_IO_MD and NVME_ADMIN are not supported */
971 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO, false);
972 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO_MD, false);
973 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_ADMIN, false);
974 	rc = spdk_bdev_nvme_io_passthru(desc, io_ch, NULL, NULL, 0, NULL, NULL);
975 	CU_ASSERT(rc == -ENOTSUP);
976 	rc = spdk_bdev_nvme_io_passthru_md(desc, io_ch, NULL, NULL, 0, NULL, 0, NULL, NULL);
977 	CU_ASSERT(rc == -ENOTSUP);
978 	rc = spdk_bdev_nvme_admin_passthru(desc, io_ch, NULL, NULL, 0, NULL, NULL);
979 	CU_ASSERT(rc == -ENOTSUP);
980 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO, true);
981 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_IO_MD, true);
982 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_NVME_ADMIN, true);
983 
984 	spdk_put_io_channel(io_ch);
985 	spdk_bdev_close(desc);
986 	free_bdev(bdev);
987 	spdk_bdev_finish(bdev_fini_cb, NULL);
988 	poll_threads();
989 }
990 
991 static void
992 bdev_io_wait_test(void)
993 {
994 	struct spdk_bdev *bdev;
995 	struct spdk_bdev_desc *desc = NULL;
996 	struct spdk_io_channel *io_ch;
997 	struct spdk_bdev_opts bdev_opts = {};
998 	struct bdev_ut_io_wait_entry io_wait_entry;
999 	struct bdev_ut_io_wait_entry io_wait_entry2;
1000 	int rc;
1001 
1002 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1003 	bdev_opts.bdev_io_pool_size = 4;
1004 	bdev_opts.bdev_io_cache_size = 2;
1005 
1006 	rc = spdk_bdev_set_opts(&bdev_opts);
1007 	CU_ASSERT(rc == 0);
1008 	spdk_bdev_initialize(bdev_init_cb, NULL);
1009 	poll_threads();
1010 
1011 	bdev = allocate_bdev("bdev0");
1012 
1013 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1014 	CU_ASSERT(rc == 0);
1015 	poll_threads();
1016 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1017 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1018 	io_ch = spdk_bdev_get_io_channel(desc);
1019 	CU_ASSERT(io_ch != NULL);
1020 
1021 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1022 	CU_ASSERT(rc == 0);
1023 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1024 	CU_ASSERT(rc == 0);
1025 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1026 	CU_ASSERT(rc == 0);
1027 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1028 	CU_ASSERT(rc == 0);
1029 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1030 
1031 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
1032 	CU_ASSERT(rc == -ENOMEM);
1033 
1034 	io_wait_entry.entry.bdev = bdev;
1035 	io_wait_entry.entry.cb_fn = io_wait_cb;
1036 	io_wait_entry.entry.cb_arg = &io_wait_entry;
1037 	io_wait_entry.io_ch = io_ch;
1038 	io_wait_entry.desc = desc;
1039 	io_wait_entry.submitted = false;
1040 	/* Cannot use the same io_wait_entry for two different calls. */
1041 	memcpy(&io_wait_entry2, &io_wait_entry, sizeof(io_wait_entry));
1042 	io_wait_entry2.entry.cb_arg = &io_wait_entry2;
1043 
1044 	/* Queue two I/O waits. */
1045 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry.entry);
1046 	CU_ASSERT(rc == 0);
1047 	CU_ASSERT(io_wait_entry.submitted == false);
1048 	rc = spdk_bdev_queue_io_wait(bdev, io_ch, &io_wait_entry2.entry);
1049 	CU_ASSERT(rc == 0);
1050 	CU_ASSERT(io_wait_entry2.submitted == false);
1051 
1052 	stub_complete_io(1);
1053 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1054 	CU_ASSERT(io_wait_entry.submitted == true);
1055 	CU_ASSERT(io_wait_entry2.submitted == false);
1056 
1057 	stub_complete_io(1);
1058 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
1059 	CU_ASSERT(io_wait_entry2.submitted == true);
1060 
1061 	stub_complete_io(4);
1062 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1063 
1064 	spdk_put_io_channel(io_ch);
1065 	spdk_bdev_close(desc);
1066 	free_bdev(bdev);
1067 	spdk_bdev_finish(bdev_fini_cb, NULL);
1068 	poll_threads();
1069 }
1070 
1071 static void
1072 bdev_io_spans_split_test(void)
1073 {
1074 	struct spdk_bdev bdev;
1075 	struct spdk_bdev_io bdev_io;
1076 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV];
1077 
1078 	memset(&bdev, 0, sizeof(bdev));
1079 	bdev_io.u.bdev.iovs = iov;
1080 
1081 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1082 	bdev.optimal_io_boundary = 0;
1083 	bdev.max_segment_size = 0;
1084 	bdev.max_num_segments = 0;
1085 	bdev_io.bdev = &bdev;
1086 
1087 	/* bdev has no optimal_io_boundary and max_size set - so this should return false. */
1088 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1089 
1090 	bdev.split_on_optimal_io_boundary = true;
1091 	bdev.optimal_io_boundary = 32;
1092 	bdev_io.type = SPDK_BDEV_IO_TYPE_RESET;
1093 
1094 	/* RESETs are not based on LBAs - so this should return false. */
1095 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1096 
1097 	bdev_io.type = SPDK_BDEV_IO_TYPE_READ;
1098 	bdev_io.u.bdev.offset_blocks = 0;
1099 	bdev_io.u.bdev.num_blocks = 32;
1100 
1101 	/* This I/O run right up to, but does not cross, the boundary - so this should return false. */
1102 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1103 
1104 	bdev_io.u.bdev.num_blocks = 33;
1105 
1106 	/* This I/O spans a boundary. */
1107 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1108 
1109 	bdev_io.u.bdev.num_blocks = 32;
1110 	bdev.max_segment_size = 512 * 32;
1111 	bdev.max_num_segments = 1;
1112 	bdev_io.u.bdev.iovcnt = 1;
1113 	iov[0].iov_len = 512;
1114 
1115 	/* Does not cross and exceed max_size or max_segs */
1116 	CU_ASSERT(bdev_io_should_split(&bdev_io) == false);
1117 
1118 	bdev.split_on_optimal_io_boundary = false;
1119 	bdev.max_segment_size = 512;
1120 	bdev.max_num_segments = 1;
1121 	bdev_io.u.bdev.iovcnt = 2;
1122 
1123 	/* Exceed max_segs */
1124 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1125 
1126 	bdev.max_num_segments = 2;
1127 	iov[0].iov_len = 513;
1128 	iov[1].iov_len = 512;
1129 
1130 	/* Exceed max_sizes */
1131 	CU_ASSERT(bdev_io_should_split(&bdev_io) == true);
1132 }
1133 
1134 static void
1135 bdev_io_boundary_split_test(void)
1136 {
1137 	struct spdk_bdev *bdev;
1138 	struct spdk_bdev_desc *desc = NULL;
1139 	struct spdk_io_channel *io_ch;
1140 	struct spdk_bdev_opts bdev_opts = {};
1141 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1142 	struct ut_expected_io *expected_io;
1143 	void *md_buf = (void *)0xFF000000;
1144 	uint64_t i;
1145 	int rc;
1146 
1147 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1148 	bdev_opts.bdev_io_pool_size = 512;
1149 	bdev_opts.bdev_io_cache_size = 64;
1150 
1151 	rc = spdk_bdev_set_opts(&bdev_opts);
1152 	CU_ASSERT(rc == 0);
1153 	spdk_bdev_initialize(bdev_init_cb, NULL);
1154 
1155 	bdev = allocate_bdev("bdev0");
1156 
1157 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
1158 	CU_ASSERT(rc == 0);
1159 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1160 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
1161 	io_ch = spdk_bdev_get_io_channel(desc);
1162 	CU_ASSERT(io_ch != NULL);
1163 
1164 	bdev->optimal_io_boundary = 16;
1165 	bdev->split_on_optimal_io_boundary = false;
1166 
1167 	g_io_done = false;
1168 
1169 	/* First test that the I/O does not get split if split_on_optimal_io_boundary == false. */
1170 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 8, 1);
1171 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 8 * 512);
1172 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1173 
1174 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
1175 	CU_ASSERT(rc == 0);
1176 	CU_ASSERT(g_io_done == false);
1177 
1178 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1179 	stub_complete_io(1);
1180 	CU_ASSERT(g_io_done == true);
1181 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1182 
1183 	bdev->split_on_optimal_io_boundary = true;
1184 	bdev->md_interleave = false;
1185 	bdev->md_len = 8;
1186 
1187 	/* Now test that a single-vector command is split correctly.
1188 	 * Offset 14, length 8, payload 0xF000
1189 	 *  Child - Offset 14, length 2, payload 0xF000
1190 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
1191 	 *
1192 	 * Set up the expected values before calling spdk_bdev_read_blocks
1193 	 */
1194 	g_io_done = false;
1195 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
1196 	expected_io->md_buf = md_buf;
1197 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
1198 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1199 
1200 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
1201 	expected_io->md_buf = md_buf + 2 * 8;
1202 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
1203 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1204 
1205 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1206 	rc = spdk_bdev_read_blocks_with_md(desc, io_ch, (void *)0xF000, md_buf,
1207 					   14, 8, io_done, NULL);
1208 	CU_ASSERT(rc == 0);
1209 	CU_ASSERT(g_io_done == false);
1210 
1211 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1212 	stub_complete_io(2);
1213 	CU_ASSERT(g_io_done == true);
1214 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1215 
1216 	/* Now set up a more complex, multi-vector command that needs to be split,
1217 	 *  including splitting iovecs.
1218 	 */
1219 	iov[0].iov_base = (void *)0x10000;
1220 	iov[0].iov_len = 512;
1221 	iov[1].iov_base = (void *)0x20000;
1222 	iov[1].iov_len = 20 * 512;
1223 	iov[2].iov_base = (void *)0x30000;
1224 	iov[2].iov_len = 11 * 512;
1225 
1226 	g_io_done = false;
1227 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
1228 	expected_io->md_buf = md_buf;
1229 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
1230 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
1231 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1232 
1233 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
1234 	expected_io->md_buf = md_buf + 2 * 8;
1235 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
1236 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1237 
1238 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
1239 	expected_io->md_buf = md_buf + 18 * 8;
1240 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
1241 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
1242 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1243 
1244 	rc = spdk_bdev_writev_blocks_with_md(desc, io_ch, iov, 3, md_buf,
1245 					     14, 32, io_done, NULL);
1246 	CU_ASSERT(rc == 0);
1247 	CU_ASSERT(g_io_done == false);
1248 
1249 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
1250 	stub_complete_io(3);
1251 	CU_ASSERT(g_io_done == true);
1252 
1253 	/* Test multi vector command that needs to be split by strip and then needs to be
1254 	 * split further due to the capacity of child iovs.
1255 	 */
1256 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
1257 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1258 		iov[i].iov_len = 512;
1259 	}
1260 
1261 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1262 	g_io_done = false;
1263 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1264 					   BDEV_IO_NUM_CHILD_IOV);
1265 	expected_io->md_buf = md_buf;
1266 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1267 		ut_expected_io_set_iov(expected_io, i, (void *)((i + 1) * 0x10000), 512);
1268 	}
1269 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1270 
1271 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1272 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
1273 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1274 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1275 		ut_expected_io_set_iov(expected_io, i,
1276 				       (void *)((i + 1 + BDEV_IO_NUM_CHILD_IOV) * 0x10000), 512);
1277 	}
1278 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1279 
1280 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1281 					    0, BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1282 	CU_ASSERT(rc == 0);
1283 	CU_ASSERT(g_io_done == false);
1284 
1285 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1286 	stub_complete_io(1);
1287 	CU_ASSERT(g_io_done == false);
1288 
1289 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1290 	stub_complete_io(1);
1291 	CU_ASSERT(g_io_done == true);
1292 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1293 
1294 	/* Test multi vector command that needs to be split by strip and then needs to be
1295 	 * split further due to the capacity of child iovs. In this case, the length of
1296 	 * the rest of iovec array with an I/O boundary is the multiple of block size.
1297 	 */
1298 
1299 	/* Fill iovec array for exactly one boundary. The iovec cnt for this boundary
1300 	 * is BDEV_IO_NUM_CHILD_IOV + 1, which exceeds the capacity of child iovs.
1301 	 */
1302 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1303 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1304 		iov[i].iov_len = 512;
1305 	}
1306 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1307 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1308 		iov[i].iov_len = 256;
1309 	}
1310 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1311 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 512;
1312 
1313 	/* Add an extra iovec to trigger split */
1314 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1315 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1316 
1317 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1318 	g_io_done = false;
1319 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1320 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV);
1321 	expected_io->md_buf = md_buf;
1322 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1323 		ut_expected_io_set_iov(expected_io, i,
1324 				       (void *)((i + 1) * 0x10000), 512);
1325 	}
1326 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1327 		ut_expected_io_set_iov(expected_io, i,
1328 				       (void *)((i + 1) * 0x10000), 256);
1329 	}
1330 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1331 
1332 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1333 					   1, 1);
1334 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1335 	ut_expected_io_set_iov(expected_io, 0,
1336 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 512);
1337 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1338 
1339 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1340 					   1, 1);
1341 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1342 	ut_expected_io_set_iov(expected_io, 0,
1343 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1344 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1345 
1346 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, md_buf,
1347 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1348 	CU_ASSERT(rc == 0);
1349 	CU_ASSERT(g_io_done == false);
1350 
1351 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1352 	stub_complete_io(1);
1353 	CU_ASSERT(g_io_done == false);
1354 
1355 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1356 	stub_complete_io(2);
1357 	CU_ASSERT(g_io_done == true);
1358 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1359 
1360 	/* Test multi vector command that needs to be split by strip and then needs to be
1361 	 * split further due to the capacity of child iovs, the child request offset should
1362 	 * be rewind to last aligned offset and go success without error.
1363 	 */
1364 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1365 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1366 		iov[i].iov_len = 512;
1367 	}
1368 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1369 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1370 
1371 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1372 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1373 
1374 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1375 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1376 
1377 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1378 	g_io_done = false;
1379 	g_io_status = 0;
1380 	/* The first expected io should be start from offset 0 to BDEV_IO_NUM_CHILD_IOV - 1 */
1381 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
1382 					   BDEV_IO_NUM_CHILD_IOV - 1, BDEV_IO_NUM_CHILD_IOV - 1);
1383 	expected_io->md_buf = md_buf;
1384 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1385 		ut_expected_io_set_iov(expected_io, i,
1386 				       (void *)((i + 1) * 0x10000), 512);
1387 	}
1388 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1389 	/* The second expected io should be start from offset BDEV_IO_NUM_CHILD_IOV - 1 to BDEV_IO_NUM_CHILD_IOV */
1390 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 1,
1391 					   1, 2);
1392 	expected_io->md_buf = md_buf + (BDEV_IO_NUM_CHILD_IOV - 1) * 8;
1393 	ut_expected_io_set_iov(expected_io, 0,
1394 			       (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000), 256);
1395 	ut_expected_io_set_iov(expected_io, 1,
1396 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000), 256);
1397 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1398 	/* The third expected io should be start from offset BDEV_IO_NUM_CHILD_IOV to BDEV_IO_NUM_CHILD_IOV + 1 */
1399 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1400 					   1, 1);
1401 	expected_io->md_buf = md_buf + BDEV_IO_NUM_CHILD_IOV * 8;
1402 	ut_expected_io_set_iov(expected_io, 0,
1403 			       (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000), 512);
1404 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1405 
1406 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, md_buf,
1407 					    0, BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1408 	CU_ASSERT(rc == 0);
1409 	CU_ASSERT(g_io_done == false);
1410 
1411 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1412 	stub_complete_io(1);
1413 	CU_ASSERT(g_io_done == false);
1414 
1415 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1416 	stub_complete_io(2);
1417 	CU_ASSERT(g_io_done == true);
1418 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1419 
1420 	/* Test multi vector command that needs to be split due to the IO boundary and
1421 	 * the capacity of child iovs. Especially test the case when the command is
1422 	 * split due to the capacity of child iovs, the tail address is not aligned with
1423 	 * block size and is rewinded to the aligned address.
1424 	 *
1425 	 * The iovecs used in read request is complex but is based on the data
1426 	 * collected in the real issue. We change the base addresses but keep the lengths
1427 	 * not to loose the credibility of the test.
1428 	 */
1429 	bdev->optimal_io_boundary = 128;
1430 	g_io_done = false;
1431 	g_io_status = 0;
1432 
1433 	for (i = 0; i < 31; i++) {
1434 		iov[i].iov_base = (void *)(0xFEED0000000 + (i << 20));
1435 		iov[i].iov_len = 1024;
1436 	}
1437 	iov[31].iov_base = (void *)0xFEED1F00000;
1438 	iov[31].iov_len = 32768;
1439 	iov[32].iov_base = (void *)0xFEED2000000;
1440 	iov[32].iov_len = 160;
1441 	iov[33].iov_base = (void *)0xFEED2100000;
1442 	iov[33].iov_len = 4096;
1443 	iov[34].iov_base = (void *)0xFEED2200000;
1444 	iov[34].iov_len = 4096;
1445 	iov[35].iov_base = (void *)0xFEED2300000;
1446 	iov[35].iov_len = 4096;
1447 	iov[36].iov_base = (void *)0xFEED2400000;
1448 	iov[36].iov_len = 4096;
1449 	iov[37].iov_base = (void *)0xFEED2500000;
1450 	iov[37].iov_len = 4096;
1451 	iov[38].iov_base = (void *)0xFEED2600000;
1452 	iov[38].iov_len = 4096;
1453 	iov[39].iov_base = (void *)0xFEED2700000;
1454 	iov[39].iov_len = 4096;
1455 	iov[40].iov_base = (void *)0xFEED2800000;
1456 	iov[40].iov_len = 4096;
1457 	iov[41].iov_base = (void *)0xFEED2900000;
1458 	iov[41].iov_len = 4096;
1459 	iov[42].iov_base = (void *)0xFEED2A00000;
1460 	iov[42].iov_len = 4096;
1461 	iov[43].iov_base = (void *)0xFEED2B00000;
1462 	iov[43].iov_len = 12288;
1463 	iov[44].iov_base = (void *)0xFEED2C00000;
1464 	iov[44].iov_len = 8192;
1465 	iov[45].iov_base = (void *)0xFEED2F00000;
1466 	iov[45].iov_len = 4096;
1467 	iov[46].iov_base = (void *)0xFEED3000000;
1468 	iov[46].iov_len = 4096;
1469 	iov[47].iov_base = (void *)0xFEED3100000;
1470 	iov[47].iov_len = 4096;
1471 	iov[48].iov_base = (void *)0xFEED3200000;
1472 	iov[48].iov_len = 24576;
1473 	iov[49].iov_base = (void *)0xFEED3300000;
1474 	iov[49].iov_len = 16384;
1475 	iov[50].iov_base = (void *)0xFEED3400000;
1476 	iov[50].iov_len = 12288;
1477 	iov[51].iov_base = (void *)0xFEED3500000;
1478 	iov[51].iov_len = 4096;
1479 	iov[52].iov_base = (void *)0xFEED3600000;
1480 	iov[52].iov_len = 4096;
1481 	iov[53].iov_base = (void *)0xFEED3700000;
1482 	iov[53].iov_len = 4096;
1483 	iov[54].iov_base = (void *)0xFEED3800000;
1484 	iov[54].iov_len = 28672;
1485 	iov[55].iov_base = (void *)0xFEED3900000;
1486 	iov[55].iov_len = 20480;
1487 	iov[56].iov_base = (void *)0xFEED3A00000;
1488 	iov[56].iov_len = 4096;
1489 	iov[57].iov_base = (void *)0xFEED3B00000;
1490 	iov[57].iov_len = 12288;
1491 	iov[58].iov_base = (void *)0xFEED3C00000;
1492 	iov[58].iov_len = 4096;
1493 	iov[59].iov_base = (void *)0xFEED3D00000;
1494 	iov[59].iov_len = 4096;
1495 	iov[60].iov_base = (void *)0xFEED3E00000;
1496 	iov[60].iov_len = 352;
1497 
1498 	/* The 1st child IO must be from iov[0] to iov[31] split by the capacity
1499 	 * of child iovs,
1500 	 */
1501 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 126, 32);
1502 	expected_io->md_buf = md_buf;
1503 	for (i = 0; i < 32; i++) {
1504 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1505 	}
1506 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1507 
1508 	/* The 2nd child IO must be from iov[32] to the first 864 bytes of iov[33]
1509 	 * split by the IO boundary requirement.
1510 	 */
1511 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 126, 2, 2);
1512 	expected_io->md_buf = md_buf + 126 * 8;
1513 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base, iov[32].iov_len);
1514 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 864);
1515 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1516 
1517 	/* The 3rd child IO must be from the remaining 3232 bytes of iov[33] to
1518 	 * the first 864 bytes of iov[46] split by the IO boundary requirement.
1519 	 */
1520 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 128, 128, 14);
1521 	expected_io->md_buf = md_buf + 128 * 8;
1522 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[33].iov_base + 864),
1523 			       iov[33].iov_len - 864);
1524 	ut_expected_io_set_iov(expected_io, 1, iov[34].iov_base, iov[34].iov_len);
1525 	ut_expected_io_set_iov(expected_io, 2, iov[35].iov_base, iov[35].iov_len);
1526 	ut_expected_io_set_iov(expected_io, 3, iov[36].iov_base, iov[36].iov_len);
1527 	ut_expected_io_set_iov(expected_io, 4, iov[37].iov_base, iov[37].iov_len);
1528 	ut_expected_io_set_iov(expected_io, 5, iov[38].iov_base, iov[38].iov_len);
1529 	ut_expected_io_set_iov(expected_io, 6, iov[39].iov_base, iov[39].iov_len);
1530 	ut_expected_io_set_iov(expected_io, 7, iov[40].iov_base, iov[40].iov_len);
1531 	ut_expected_io_set_iov(expected_io, 8, iov[41].iov_base, iov[41].iov_len);
1532 	ut_expected_io_set_iov(expected_io, 9, iov[42].iov_base, iov[42].iov_len);
1533 	ut_expected_io_set_iov(expected_io, 10, iov[43].iov_base, iov[43].iov_len);
1534 	ut_expected_io_set_iov(expected_io, 11, iov[44].iov_base, iov[44].iov_len);
1535 	ut_expected_io_set_iov(expected_io, 12, iov[45].iov_base, iov[45].iov_len);
1536 	ut_expected_io_set_iov(expected_io, 13, iov[46].iov_base, 864);
1537 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1538 
1539 	/* The 4th child IO must be from the remaining 3232 bytes of iov[46] to the
1540 	 * first 864 bytes of iov[52] split by the IO boundary requirement.
1541 	 */
1542 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 256, 128, 7);
1543 	expected_io->md_buf = md_buf + 256 * 8;
1544 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[46].iov_base + 864),
1545 			       iov[46].iov_len - 864);
1546 	ut_expected_io_set_iov(expected_io, 1, iov[47].iov_base, iov[47].iov_len);
1547 	ut_expected_io_set_iov(expected_io, 2, iov[48].iov_base, iov[48].iov_len);
1548 	ut_expected_io_set_iov(expected_io, 3, iov[49].iov_base, iov[49].iov_len);
1549 	ut_expected_io_set_iov(expected_io, 4, iov[50].iov_base, iov[50].iov_len);
1550 	ut_expected_io_set_iov(expected_io, 5, iov[51].iov_base, iov[51].iov_len);
1551 	ut_expected_io_set_iov(expected_io, 6, iov[52].iov_base, 864);
1552 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1553 
1554 	/* The 5th child IO must be from the remaining 3232 bytes of iov[52] to
1555 	 * the first 4096 bytes of iov[57] split by the IO boundary requirement.
1556 	 */
1557 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 384, 128, 6);
1558 	expected_io->md_buf = md_buf + 384 * 8;
1559 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[52].iov_base + 864),
1560 			       iov[52].iov_len - 864);
1561 	ut_expected_io_set_iov(expected_io, 1, iov[53].iov_base, iov[53].iov_len);
1562 	ut_expected_io_set_iov(expected_io, 2, iov[54].iov_base, iov[54].iov_len);
1563 	ut_expected_io_set_iov(expected_io, 3, iov[55].iov_base, iov[55].iov_len);
1564 	ut_expected_io_set_iov(expected_io, 4, iov[56].iov_base, iov[56].iov_len);
1565 	ut_expected_io_set_iov(expected_io, 5, iov[57].iov_base, 4960);
1566 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1567 
1568 	/* The 6th child IO must be from the remaining 7328 bytes of iov[57]
1569 	 * to the first 3936 bytes of iov[58] split by the capacity of child iovs.
1570 	 */
1571 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 512, 30, 3);
1572 	expected_io->md_buf = md_buf + 512 * 8;
1573 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[57].iov_base + 4960),
1574 			       iov[57].iov_len - 4960);
1575 	ut_expected_io_set_iov(expected_io, 1, iov[58].iov_base, iov[58].iov_len);
1576 	ut_expected_io_set_iov(expected_io, 2, iov[59].iov_base, 3936);
1577 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1578 
1579 	/* The 7th child IO is from the remaining 160 bytes of iov[59] and iov[60]. */
1580 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 542, 1, 2);
1581 	expected_io->md_buf = md_buf + 542 * 8;
1582 	ut_expected_io_set_iov(expected_io, 0, (void *)((uintptr_t)iov[59].iov_base + 3936),
1583 			       iov[59].iov_len - 3936);
1584 	ut_expected_io_set_iov(expected_io, 1, iov[60].iov_base, iov[60].iov_len);
1585 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1586 
1587 	rc = spdk_bdev_readv_blocks_with_md(desc, io_ch, iov, 61, md_buf,
1588 					    0, 543, io_done, NULL);
1589 	CU_ASSERT(rc == 0);
1590 	CU_ASSERT(g_io_done == false);
1591 
1592 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1593 	stub_complete_io(1);
1594 	CU_ASSERT(g_io_done == false);
1595 
1596 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1597 	stub_complete_io(5);
1598 	CU_ASSERT(g_io_done == false);
1599 
1600 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1601 	stub_complete_io(1);
1602 	CU_ASSERT(g_io_done == true);
1603 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1604 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1605 
1606 	/* Test a WRITE_ZEROES that would span an I/O boundary.  WRITE_ZEROES should not be
1607 	 * split, so test that.
1608 	 */
1609 	bdev->optimal_io_boundary = 15;
1610 	g_io_done = false;
1611 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
1612 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1613 
1614 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
1615 	CU_ASSERT(rc == 0);
1616 	CU_ASSERT(g_io_done == false);
1617 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1618 	stub_complete_io(1);
1619 	CU_ASSERT(g_io_done == true);
1620 
1621 	/* Test an UNMAP.  This should also not be split. */
1622 	bdev->optimal_io_boundary = 16;
1623 	g_io_done = false;
1624 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 2, 0);
1625 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1626 
1627 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 2, io_done, NULL);
1628 	CU_ASSERT(rc == 0);
1629 	CU_ASSERT(g_io_done == false);
1630 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1631 	stub_complete_io(1);
1632 	CU_ASSERT(g_io_done == true);
1633 
1634 	/* Test a FLUSH.  This should also not be split. */
1635 	bdev->optimal_io_boundary = 16;
1636 	g_io_done = false;
1637 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 2, 0);
1638 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1639 
1640 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
1641 	CU_ASSERT(rc == 0);
1642 	CU_ASSERT(g_io_done == false);
1643 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1644 	stub_complete_io(1);
1645 	CU_ASSERT(g_io_done == true);
1646 
1647 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
1648 
1649 	/* Children requests return an error status */
1650 	bdev->optimal_io_boundary = 16;
1651 	iov[0].iov_base = (void *)0x10000;
1652 	iov[0].iov_len = 512 * 64;
1653 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1654 	g_io_done = false;
1655 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1656 
1657 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 1, 1, 64, io_done, NULL);
1658 	CU_ASSERT(rc == 0);
1659 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 5);
1660 	stub_complete_io(4);
1661 	CU_ASSERT(g_io_done == false);
1662 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
1663 	stub_complete_io(1);
1664 	CU_ASSERT(g_io_done == true);
1665 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1666 
1667 	/* Test if a multi vector command terminated with failure before continuing
1668 	 * splitting process when one of child I/O failed.
1669 	 * The multi vector command is as same as the above that needs to be split by strip
1670 	 * and then needs to be split further due to the capacity of child iovs.
1671 	 */
1672 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 1; i++) {
1673 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1674 		iov[i].iov_len = 512;
1675 	}
1676 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base = (void *)(BDEV_IO_NUM_CHILD_IOV * 0x10000);
1677 	iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_len = 256;
1678 
1679 	iov[BDEV_IO_NUM_CHILD_IOV].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 1) * 0x10000);
1680 	iov[BDEV_IO_NUM_CHILD_IOV].iov_len = 256;
1681 
1682 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_base = (void *)((BDEV_IO_NUM_CHILD_IOV + 2) * 0x10000);
1683 	iov[BDEV_IO_NUM_CHILD_IOV + 1].iov_len = 512;
1684 
1685 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
1686 
1687 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
1688 	g_io_done = false;
1689 	g_io_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1690 
1691 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
1692 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1693 	CU_ASSERT(rc == 0);
1694 	CU_ASSERT(g_io_done == false);
1695 
1696 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1697 	stub_complete_io(1);
1698 	CU_ASSERT(g_io_done == true);
1699 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
1700 
1701 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
1702 
1703 	/* for this test we will create the following conditions to hit the code path where
1704 	 * we are trying to send and IO following a split that has no iovs because we had to
1705 	 * trim them for alignment reasons.
1706 	 *
1707 	 * - 16K boundary, our IO will start at offset 0 with a length of 0x4200
1708 	 * - Our IOVs are 0x212 in size so that we run into the 16K boundary at child IOV
1709 	 *   position 30 and overshoot by 0x2e.
1710 	 * - That means we'll send the IO and loop back to pick up the remaining bytes at
1711 	 *   child IOV index 31. When we do, we find that we have to shorten index 31 by 0x2e
1712 	 *   which eliniates that vector so we just send the first split IO with 30 vectors
1713 	 *   and let the completion pick up the last 2 vectors.
1714 	 */
1715 	bdev->optimal_io_boundary = 32;
1716 	bdev->split_on_optimal_io_boundary = true;
1717 	g_io_done = false;
1718 
1719 	/* Init all parent IOVs to 0x212 */
1720 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
1721 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1722 		iov[i].iov_len = 0x212;
1723 	}
1724 
1725 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, BDEV_IO_NUM_CHILD_IOV,
1726 					   BDEV_IO_NUM_CHILD_IOV - 1);
1727 	/* expect 0-29 to be 1:1 with the parent iov */
1728 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
1729 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
1730 	}
1731 
1732 	/* expect index 30 to be shortened to 0x1e4 (0x212 - 0x1e) because of the alignment
1733 	 * where 0x1e is the amount we overshot the 16K boundary
1734 	 */
1735 	ut_expected_io_set_iov(expected_io, BDEV_IO_NUM_CHILD_IOV - 2,
1736 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base), 0x1e4);
1737 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1738 
1739 	/* 2nd child IO will have 2 remaining vectors, one to pick up from the one that was
1740 	 * shortened that take it to the next boundary and then a final one to get us to
1741 	 * 0x4200 bytes for the IO.
1742 	 */
1743 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV,
1744 					   BDEV_IO_NUM_CHILD_IOV, 2);
1745 	/* position 30 picked up the remaining bytes to the next boundary */
1746 	ut_expected_io_set_iov(expected_io, 0,
1747 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 2].iov_base + 0x1e4), 0x2e);
1748 
1749 	/* position 31 picked the the rest of the transfer to get us to 0x4200 */
1750 	ut_expected_io_set_iov(expected_io, 1,
1751 			       (void *)(iov[BDEV_IO_NUM_CHILD_IOV - 1].iov_base), 0x1d2);
1752 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1753 
1754 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 1, 0,
1755 				    BDEV_IO_NUM_CHILD_IOV + 1, io_done, NULL);
1756 	CU_ASSERT(rc == 0);
1757 	CU_ASSERT(g_io_done == false);
1758 
1759 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1760 	stub_complete_io(1);
1761 	CU_ASSERT(g_io_done == false);
1762 
1763 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1764 	stub_complete_io(1);
1765 	CU_ASSERT(g_io_done == true);
1766 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1767 
1768 	spdk_put_io_channel(io_ch);
1769 	spdk_bdev_close(desc);
1770 	free_bdev(bdev);
1771 	spdk_bdev_finish(bdev_fini_cb, NULL);
1772 	poll_threads();
1773 }
1774 
1775 static void
1776 bdev_io_max_size_and_segment_split_test(void)
1777 {
1778 	struct spdk_bdev *bdev;
1779 	struct spdk_bdev_desc *desc = NULL;
1780 	struct spdk_io_channel *io_ch;
1781 	struct spdk_bdev_opts bdev_opts = {};
1782 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
1783 	struct ut_expected_io *expected_io;
1784 	uint64_t i;
1785 	int rc;
1786 
1787 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
1788 	bdev_opts.bdev_io_pool_size = 512;
1789 	bdev_opts.bdev_io_cache_size = 64;
1790 
1791 	bdev_opts.opts_size = sizeof(bdev_opts);
1792 	rc = spdk_bdev_set_opts(&bdev_opts);
1793 	CU_ASSERT(rc == 0);
1794 	spdk_bdev_initialize(bdev_init_cb, NULL);
1795 
1796 	bdev = allocate_bdev("bdev0");
1797 
1798 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
1799 	CU_ASSERT(rc == 0);
1800 	SPDK_CU_ASSERT_FATAL(desc != NULL);
1801 	io_ch = spdk_bdev_get_io_channel(desc);
1802 	CU_ASSERT(io_ch != NULL);
1803 
1804 	bdev->split_on_optimal_io_boundary = false;
1805 	bdev->optimal_io_boundary = 0;
1806 
1807 	/* Case 0 max_num_segments == 0.
1808 	 * but segment size 2 * 512 > 512
1809 	 */
1810 	bdev->max_segment_size = 512;
1811 	bdev->max_num_segments = 0;
1812 	g_io_done = false;
1813 
1814 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
1815 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1816 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
1817 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1818 
1819 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1820 	CU_ASSERT(rc == 0);
1821 	CU_ASSERT(g_io_done == false);
1822 
1823 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
1824 	stub_complete_io(1);
1825 	CU_ASSERT(g_io_done == true);
1826 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1827 
1828 	/* Case 1 max_segment_size == 0
1829 	 * but iov num 2 > 1.
1830 	 */
1831 	bdev->max_segment_size = 0;
1832 	bdev->max_num_segments = 1;
1833 	g_io_done = false;
1834 
1835 	iov[0].iov_base = (void *)0x10000;
1836 	iov[0].iov_len = 512;
1837 	iov[1].iov_base = (void *)0x20000;
1838 	iov[1].iov_len = 8 * 512;
1839 
1840 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1841 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, iov[0].iov_len);
1842 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1843 
1844 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 8, 1);
1845 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, iov[1].iov_len);
1846 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1847 
1848 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 14, 9, io_done, NULL);
1849 	CU_ASSERT(rc == 0);
1850 	CU_ASSERT(g_io_done == false);
1851 
1852 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1853 	stub_complete_io(2);
1854 	CU_ASSERT(g_io_done == true);
1855 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1856 
1857 	/* Test that a non-vector command is split correctly.
1858 	 * Set up the expected values before calling spdk_bdev_read_blocks
1859 	 */
1860 	bdev->max_segment_size = 512;
1861 	bdev->max_num_segments = 1;
1862 	g_io_done = false;
1863 
1864 	/* Child IO 0 */
1865 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 1, 1);
1866 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
1867 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1868 
1869 	/* Child IO 1 */
1870 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
1871 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 1 * 512), 512);
1872 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1873 
1874 	/* spdk_bdev_read_blocks will submit the first child immediately. */
1875 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 2, io_done, NULL);
1876 	CU_ASSERT(rc == 0);
1877 	CU_ASSERT(g_io_done == false);
1878 
1879 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
1880 	stub_complete_io(2);
1881 	CU_ASSERT(g_io_done == true);
1882 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1883 
1884 	/* Now set up a more complex, multi-vector command that needs to be split,
1885 	 * including splitting iovecs.
1886 	 */
1887 	bdev->max_segment_size = 2 * 512;
1888 	bdev->max_num_segments = 1;
1889 	g_io_done = false;
1890 
1891 	iov[0].iov_base = (void *)0x10000;
1892 	iov[0].iov_len = 2 * 512;
1893 	iov[1].iov_base = (void *)0x20000;
1894 	iov[1].iov_len = 4 * 512;
1895 	iov[2].iov_base = (void *)0x30000;
1896 	iov[2].iov_len = 6 * 512;
1897 
1898 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
1899 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 2);
1900 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1901 
1902 	/* Split iov[1].size to 2 iov entries then split the segments */
1903 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
1904 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base, 512 * 2);
1905 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1906 
1907 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 18, 2, 1);
1908 	ut_expected_io_set_iov(expected_io, 0, iov[1].iov_base + 512 * 2, 512 * 2);
1909 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1910 
1911 	/* Split iov[2].size to 3 iov entries then split the segments */
1912 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 20, 2, 1);
1913 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base, 512 * 2);
1914 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1915 
1916 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 22, 2, 1);
1917 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 2, 512 * 2);
1918 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1919 
1920 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 24, 2, 1);
1921 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 4, 512 * 2);
1922 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1923 
1924 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 12, io_done, NULL);
1925 	CU_ASSERT(rc == 0);
1926 	CU_ASSERT(g_io_done == false);
1927 
1928 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
1929 	stub_complete_io(6);
1930 	CU_ASSERT(g_io_done == true);
1931 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1932 
1933 	/* Test multi vector command that needs to be split by strip and then needs to be
1934 	 * split further due to the capacity of parent IO child iovs.
1935 	 */
1936 	bdev->max_segment_size = 512;
1937 	bdev->max_num_segments = 1;
1938 	g_io_done = false;
1939 
1940 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1941 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
1942 		iov[i].iov_len = 512 * 2;
1943 	}
1944 
1945 	/* Each input iov.size is split into 2 iovs,
1946 	 * half of the input iov can fill all child iov entries of a single IO.
1947 	 */
1948 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV / 2; i++) {
1949 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i, 1, 1);
1950 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1951 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1952 
1953 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 2 * i + 1, 1, 1);
1954 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1955 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1956 	}
1957 
1958 	/* The remaining iov is split in the second round */
1959 	for (i = BDEV_IO_NUM_CHILD_IOV / 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
1960 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2, 1, 1);
1961 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, 512);
1962 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1963 
1964 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 2 + 1, 1, 1);
1965 		ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base + 512, 512);
1966 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
1967 	}
1968 
1969 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
1970 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, NULL);
1971 	CU_ASSERT(rc == 0);
1972 	CU_ASSERT(g_io_done == false);
1973 
1974 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1975 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1976 	CU_ASSERT(g_io_done == false);
1977 
1978 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == BDEV_IO_NUM_CHILD_IOV);
1979 	stub_complete_io(BDEV_IO_NUM_CHILD_IOV);
1980 	CU_ASSERT(g_io_done == true);
1981 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
1982 
1983 	/* A wrong case, a child IO that is divided does
1984 	 * not meet the principle of multiples of block size,
1985 	 * and exits with error
1986 	 */
1987 	bdev->max_segment_size = 512;
1988 	bdev->max_num_segments = 1;
1989 	g_io_done = false;
1990 
1991 	iov[0].iov_base = (void *)0x10000;
1992 	iov[0].iov_len = 512 + 256;
1993 	iov[1].iov_base = (void *)0x20000;
1994 	iov[1].iov_len = 256;
1995 
1996 	/* iov[0] is split to 512 and 256.
1997 	 * 256 is less than a block size, and it is found
1998 	 * in the next round of split that it is the first child IO smaller than
1999 	 * the block size, so the error exit
2000 	 */
2001 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 1, 1);
2002 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512);
2003 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2004 
2005 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, 2, 0, 2, io_done, NULL);
2006 	CU_ASSERT(rc == 0);
2007 	CU_ASSERT(g_io_done == false);
2008 
2009 	/* First child IO is OK */
2010 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2011 	stub_complete_io(1);
2012 	CU_ASSERT(g_io_done == true);
2013 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2014 
2015 	/* error exit */
2016 	stub_complete_io(1);
2017 	CU_ASSERT(g_io_done == true);
2018 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
2019 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2020 
2021 	/* Test multi vector command that needs to be split by strip and then needs to be
2022 	 * split further due to the capacity of child iovs.
2023 	 *
2024 	 * In this case, the last two iovs need to be split, but it will exceed the capacity
2025 	 * of child iovs, so it needs to wait until the first batch completed.
2026 	 */
2027 	bdev->max_segment_size = 512;
2028 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2029 	g_io_done = false;
2030 
2031 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2032 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2033 		iov[i].iov_len = 512;
2034 	}
2035 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV; i++) {
2036 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2037 		iov[i].iov_len = 512 * 2;
2038 	}
2039 
2040 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2041 					   BDEV_IO_NUM_CHILD_IOV, BDEV_IO_NUM_CHILD_IOV);
2042 	/* 0 ~ (BDEV_IO_NUM_CHILD_IOV - 2) Will not be split */
2043 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2044 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2045 	}
2046 	/* (BDEV_IO_NUM_CHILD_IOV - 2) is split */
2047 	ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, 512);
2048 	ut_expected_io_set_iov(expected_io, i + 1, iov[i].iov_base + 512, 512);
2049 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2050 
2051 	/* Child iov entries exceed the max num of parent IO so split it in next round */
2052 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV, 2, 2);
2053 	ut_expected_io_set_iov(expected_io, 0, iov[i + 1].iov_base, 512);
2054 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base + 512, 512);
2055 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2056 
2057 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV, 0,
2058 				    BDEV_IO_NUM_CHILD_IOV + 2, io_done, NULL);
2059 	CU_ASSERT(rc == 0);
2060 	CU_ASSERT(g_io_done == false);
2061 
2062 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2063 	stub_complete_io(1);
2064 	CU_ASSERT(g_io_done == false);
2065 
2066 	/* Next round */
2067 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2068 	stub_complete_io(1);
2069 	CU_ASSERT(g_io_done == true);
2070 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2071 
2072 	/* This case is similar to the previous one, but the io composed of
2073 	 * the last few entries of child iov is not enough for a blocklen, so they
2074 	 * cannot be put into this IO, but wait until the next time.
2075 	 */
2076 	bdev->max_segment_size = 512;
2077 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2078 	g_io_done = false;
2079 
2080 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2081 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2082 		iov[i].iov_len = 512;
2083 	}
2084 
2085 	for (i = BDEV_IO_NUM_CHILD_IOV - 2; i < BDEV_IO_NUM_CHILD_IOV + 2; i++) {
2086 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2087 		iov[i].iov_len = 128;
2088 	}
2089 
2090 	/* First child iovcnt is't BDEV_IO_NUM_CHILD_IOV but BDEV_IO_NUM_CHILD_IOV - 2.
2091 	 * Because the left 2 iov is not enough for a blocklen.
2092 	 */
2093 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0,
2094 					   BDEV_IO_NUM_CHILD_IOV - 2, BDEV_IO_NUM_CHILD_IOV - 2);
2095 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV - 2; i++) {
2096 		ut_expected_io_set_iov(expected_io, i, iov[i].iov_base, iov[i].iov_len);
2097 	}
2098 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2099 
2100 	/* The second child io waits until the end of the first child io before executing.
2101 	 * Because the iovcnt of the two IOs exceeds the child iovcnt of the parent IO.
2102 	 * BDEV_IO_NUM_CHILD_IOV - 2 to BDEV_IO_NUM_CHILD_IOV + 2
2103 	 */
2104 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, BDEV_IO_NUM_CHILD_IOV - 2,
2105 					   1, 4);
2106 	ut_expected_io_set_iov(expected_io, 0, iov[i].iov_base, iov[i].iov_len);
2107 	ut_expected_io_set_iov(expected_io, 1, iov[i + 1].iov_base, iov[i + 1].iov_len);
2108 	ut_expected_io_set_iov(expected_io, 2, iov[i + 2].iov_base, iov[i + 2].iov_len);
2109 	ut_expected_io_set_iov(expected_io, 3, iov[i + 3].iov_base, iov[i + 3].iov_len);
2110 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2111 
2112 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2113 				    BDEV_IO_NUM_CHILD_IOV - 1, io_done, NULL);
2114 	CU_ASSERT(rc == 0);
2115 	CU_ASSERT(g_io_done == false);
2116 
2117 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2118 	stub_complete_io(1);
2119 	CU_ASSERT(g_io_done == false);
2120 
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 	/* A very complicated case. Each sg entry exceeds max_segment_size and
2127 	 * needs to be split. At the same time, child io must be a multiple of blocklen.
2128 	 * At the same time, child iovcnt exceeds parent iovcnt.
2129 	 */
2130 	bdev->max_segment_size = 512 + 128;
2131 	bdev->max_num_segments = 3;
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 + 256;
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 = 512 + 128;
2142 	}
2143 
2144 	/* Child IOs use 9 entries per for() round and 3 * 9 = 27 child iov entries.
2145 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2146 	 * Generate 9 child IOs.
2147 	 */
2148 	for (i = 0; i < 3; i++) {
2149 		uint32_t j = i * 4;
2150 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6, 2, 3);
2151 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2152 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2153 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2154 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2155 
2156 		/* Child io must be a multiple of blocklen
2157 		 * iov[j + 2] must be split. If the third entry is also added,
2158 		 * the multiple of blocklen cannot be guaranteed. But it still
2159 		 * occupies one iov entry of the parent child iov.
2160 		 */
2161 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 2, 2, 2);
2162 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2163 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2164 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2165 
2166 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, i * 6 + 4, 2, 3);
2167 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2168 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2169 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2170 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2171 	}
2172 
2173 	/* Child iov position at 27, the 10th child IO
2174 	 * iov entry index is 3 * 4 and offset is 3 * 6
2175 	 */
2176 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 18, 2, 3);
2177 	ut_expected_io_set_iov(expected_io, 0, iov[12].iov_base, 640);
2178 	ut_expected_io_set_iov(expected_io, 1, iov[12].iov_base + 640, 128);
2179 	ut_expected_io_set_iov(expected_io, 2, iov[13].iov_base, 256);
2180 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2181 
2182 	/* Child iov position at 30, the 11th child IO */
2183 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 20, 2, 2);
2184 	ut_expected_io_set_iov(expected_io, 0, iov[13].iov_base + 256, 512);
2185 	ut_expected_io_set_iov(expected_io, 1, iov[14].iov_base, 512);
2186 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2187 
2188 	/* The 2nd split round and iovpos is 0, the 12th child IO */
2189 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 22, 2, 3);
2190 	ut_expected_io_set_iov(expected_io, 0, iov[14].iov_base + 512, 256);
2191 	ut_expected_io_set_iov(expected_io, 1, iov[15].iov_base, 640);
2192 	ut_expected_io_set_iov(expected_io, 2, iov[15].iov_base + 640, 128);
2193 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2194 
2195 	/* Consume 9 child IOs and 27 child iov entries.
2196 	 * Consume 4 parent IO iov entries per for() round and 6 block size.
2197 	 * Parent IO iov index start from 16 and block offset start from 24
2198 	 */
2199 	for (i = 0; i < 3; i++) {
2200 		uint32_t j = i * 4 + 16;
2201 		uint32_t offset = i * 6 + 24;
2202 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, 2, 3);
2203 		ut_expected_io_set_iov(expected_io, 0, iov[j].iov_base, 640);
2204 		ut_expected_io_set_iov(expected_io, 1, iov[j].iov_base + 640, 128);
2205 		ut_expected_io_set_iov(expected_io, 2, iov[j + 1].iov_base, 256);
2206 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2207 
2208 		/* Child io must be a multiple of blocklen
2209 		 * iov[j + 2] must be split. If the third entry is also added,
2210 		 * the multiple of blocklen cannot be guaranteed. But it still
2211 		 * occupies one iov entry of the parent child iov.
2212 		 */
2213 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 2, 2, 2);
2214 		ut_expected_io_set_iov(expected_io, 0, iov[j + 1].iov_base + 256, 512);
2215 		ut_expected_io_set_iov(expected_io, 1, iov[j + 2].iov_base, 512);
2216 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2217 
2218 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset + 4, 2, 3);
2219 		ut_expected_io_set_iov(expected_io, 0, iov[j + 2].iov_base + 512, 256);
2220 		ut_expected_io_set_iov(expected_io, 1, iov[j + 3].iov_base, 640);
2221 		ut_expected_io_set_iov(expected_io, 2, iov[j + 3].iov_base + 640, 128);
2222 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2223 	}
2224 
2225 	/* The 22th child IO, child iov position at 30 */
2226 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 42, 1, 1);
2227 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base, 512);
2228 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2229 
2230 	/* The third round */
2231 	/* Here is the 23nd child IO and child iovpos is 0 */
2232 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 43, 2, 3);
2233 	ut_expected_io_set_iov(expected_io, 0, iov[28].iov_base + 512, 256);
2234 	ut_expected_io_set_iov(expected_io, 1, iov[29].iov_base, 640);
2235 	ut_expected_io_set_iov(expected_io, 2, iov[29].iov_base + 640, 128);
2236 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2237 
2238 	/* The 24th child IO */
2239 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 45, 3, 3);
2240 	ut_expected_io_set_iov(expected_io, 0, iov[30].iov_base, 640);
2241 	ut_expected_io_set_iov(expected_io, 1, iov[31].iov_base, 640);
2242 	ut_expected_io_set_iov(expected_io, 2, iov[32].iov_base, 256);
2243 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2244 
2245 	/* The 25th child IO */
2246 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 48, 2, 2);
2247 	ut_expected_io_set_iov(expected_io, 0, iov[32].iov_base + 256, 384);
2248 	ut_expected_io_set_iov(expected_io, 1, iov[33].iov_base, 640);
2249 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2250 
2251 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV + 2, 0,
2252 				    50, io_done, NULL);
2253 	CU_ASSERT(rc == 0);
2254 	CU_ASSERT(g_io_done == false);
2255 
2256 	/* Parent IO supports up to 32 child iovs, so it is calculated that
2257 	 * a maximum of 11 IOs can be split at a time, and the
2258 	 * splitting will continue after the first batch is over.
2259 	 */
2260 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2261 	stub_complete_io(11);
2262 	CU_ASSERT(g_io_done == false);
2263 
2264 	/* The 2nd round */
2265 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 11);
2266 	stub_complete_io(11);
2267 	CU_ASSERT(g_io_done == false);
2268 
2269 	/* The last round */
2270 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2271 	stub_complete_io(3);
2272 	CU_ASSERT(g_io_done == true);
2273 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2274 
2275 	/* Test an WRITE_ZEROES.  This should also not be split. */
2276 	bdev->max_segment_size = 512;
2277 	bdev->max_num_segments = 1;
2278 	g_io_done = false;
2279 
2280 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 9, 36, 0);
2281 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2282 
2283 	rc = spdk_bdev_write_zeroes_blocks(desc, io_ch, 9, 36, io_done, NULL);
2284 	CU_ASSERT(rc == 0);
2285 	CU_ASSERT(g_io_done == false);
2286 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2287 	stub_complete_io(1);
2288 	CU_ASSERT(g_io_done == true);
2289 
2290 	/* Test an UNMAP.  This should also not be split. */
2291 	g_io_done = false;
2292 
2293 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 15, 4, 0);
2294 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2295 
2296 	rc = spdk_bdev_unmap_blocks(desc, io_ch, 15, 4, io_done, NULL);
2297 	CU_ASSERT(rc == 0);
2298 	CU_ASSERT(g_io_done == false);
2299 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2300 	stub_complete_io(1);
2301 	CU_ASSERT(g_io_done == true);
2302 
2303 	/* Test a FLUSH.  This should also not be split. */
2304 	g_io_done = false;
2305 
2306 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_FLUSH, 15, 4, 0);
2307 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2308 
2309 	rc = spdk_bdev_flush_blocks(desc, io_ch, 15, 2, io_done, NULL);
2310 	CU_ASSERT(rc == 0);
2311 	CU_ASSERT(g_io_done == false);
2312 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2313 	stub_complete_io(1);
2314 	CU_ASSERT(g_io_done == true);
2315 
2316 	spdk_put_io_channel(io_ch);
2317 	spdk_bdev_close(desc);
2318 	free_bdev(bdev);
2319 	spdk_bdev_finish(bdev_fini_cb, NULL);
2320 	poll_threads();
2321 }
2322 
2323 static void
2324 bdev_io_mix_split_test(void)
2325 {
2326 	struct spdk_bdev *bdev;
2327 	struct spdk_bdev_desc *desc = NULL;
2328 	struct spdk_io_channel *io_ch;
2329 	struct spdk_bdev_opts bdev_opts = {};
2330 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
2331 	struct ut_expected_io *expected_io;
2332 	uint64_t i;
2333 	int rc;
2334 
2335 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2336 	bdev_opts.bdev_io_pool_size = 512;
2337 	bdev_opts.bdev_io_cache_size = 64;
2338 
2339 	rc = spdk_bdev_set_opts(&bdev_opts);
2340 	CU_ASSERT(rc == 0);
2341 	spdk_bdev_initialize(bdev_init_cb, NULL);
2342 
2343 	bdev = allocate_bdev("bdev0");
2344 
2345 	rc = spdk_bdev_open_ext(bdev->name, true, bdev_ut_event_cb, NULL, &desc);
2346 	CU_ASSERT(rc == 0);
2347 	SPDK_CU_ASSERT_FATAL(desc != NULL);
2348 	io_ch = spdk_bdev_get_io_channel(desc);
2349 	CU_ASSERT(io_ch != NULL);
2350 
2351 	/* First case optimal_io_boundary == max_segment_size * max_num_segments */
2352 	bdev->split_on_optimal_io_boundary = true;
2353 	bdev->optimal_io_boundary = 16;
2354 
2355 	bdev->max_segment_size = 512;
2356 	bdev->max_num_segments = 16;
2357 	g_io_done = false;
2358 
2359 	/* IO crossing the IO boundary requires split
2360 	 * Total 2 child IOs.
2361 	 */
2362 
2363 	/* The 1st child IO split the segment_size to multiple segment entry */
2364 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 2);
2365 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512);
2366 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 512), 512);
2367 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2368 
2369 	/* The 2nd child IO split the segment_size to multiple segment entry */
2370 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 2);
2371 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 512);
2372 	ut_expected_io_set_iov(expected_io, 1, (void *)(0xF000 + 3 * 512), 512);
2373 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2374 
2375 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 4, io_done, NULL);
2376 	CU_ASSERT(rc == 0);
2377 	CU_ASSERT(g_io_done == false);
2378 
2379 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2380 	stub_complete_io(2);
2381 	CU_ASSERT(g_io_done == true);
2382 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2383 
2384 	/* Second case optimal_io_boundary > max_segment_size * max_num_segments */
2385 	bdev->max_segment_size = 15 * 512;
2386 	bdev->max_num_segments = 1;
2387 	g_io_done = false;
2388 
2389 	/* IO crossing the IO boundary requires split.
2390 	 * The 1st child IO segment size exceeds the max_segment_size,
2391 	 * So 1st child IO will be splitted to multiple segment entry.
2392 	 * Then it split to 2 child IOs because of the max_num_segments.
2393 	 * Total 3 child IOs.
2394 	 */
2395 
2396 	/* The first 2 IOs are in an IO boundary.
2397 	 * Because the optimal_io_boundary > max_segment_size * max_num_segments
2398 	 * So it split to the first 2 IOs.
2399 	 */
2400 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 15, 1);
2401 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 15);
2402 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2403 
2404 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 15, 1, 1);
2405 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 15), 512);
2406 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2407 
2408 	/* The 3rd Child IO is because of the io boundary */
2409 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2410 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2411 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2412 
2413 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2414 	CU_ASSERT(rc == 0);
2415 	CU_ASSERT(g_io_done == false);
2416 
2417 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2418 	stub_complete_io(3);
2419 	CU_ASSERT(g_io_done == true);
2420 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2421 
2422 	/* Third case optimal_io_boundary < max_segment_size * max_num_segments */
2423 	bdev->max_segment_size = 17 * 512;
2424 	bdev->max_num_segments = 1;
2425 	g_io_done = false;
2426 
2427 	/* IO crossing the IO boundary requires split.
2428 	 * Child IO does not split.
2429 	 * Total 2 child IOs.
2430 	 */
2431 
2432 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 0, 16, 1);
2433 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 512 * 16);
2434 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2435 
2436 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 2, 1);
2437 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 512 * 16), 512 * 2);
2438 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2439 
2440 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 0, 18, io_done, NULL);
2441 	CU_ASSERT(rc == 0);
2442 	CU_ASSERT(g_io_done == false);
2443 
2444 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
2445 	stub_complete_io(2);
2446 	CU_ASSERT(g_io_done == true);
2447 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2448 
2449 	/* Now set up a more complex, multi-vector command that needs to be split,
2450 	 * including splitting iovecs.
2451 	 * optimal_io_boundary < max_segment_size * max_num_segments
2452 	 */
2453 	bdev->max_segment_size = 3 * 512;
2454 	bdev->max_num_segments = 6;
2455 	g_io_done = false;
2456 
2457 	iov[0].iov_base = (void *)0x10000;
2458 	iov[0].iov_len = 4 * 512;
2459 	iov[1].iov_base = (void *)0x20000;
2460 	iov[1].iov_len = 4 * 512;
2461 	iov[2].iov_base = (void *)0x30000;
2462 	iov[2].iov_len = 10 * 512;
2463 
2464 	/* IO crossing the IO boundary requires split.
2465 	 * The 1st child IO segment size exceeds the max_segment_size and after
2466 	 * splitting segment_size, the num_segments exceeds max_num_segments.
2467 	 * So 1st child IO will be splitted to 2 child IOs.
2468 	 * Total 3 child IOs.
2469 	 */
2470 
2471 	/* The first 2 IOs are in an IO boundary.
2472 	 * After splitting segment size the segment num exceeds.
2473 	 * So it splits to 2 child IOs.
2474 	 */
2475 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 14, 6);
2476 	ut_expected_io_set_iov(expected_io, 0, iov[0].iov_base, 512 * 3);
2477 	ut_expected_io_set_iov(expected_io, 1, iov[0].iov_base + 512 * 3, 512);
2478 	ut_expected_io_set_iov(expected_io, 2, iov[1].iov_base, 512 * 3);
2479 	ut_expected_io_set_iov(expected_io, 3, iov[1].iov_base + 512 * 3, 512);
2480 	ut_expected_io_set_iov(expected_io, 4, iov[2].iov_base, 512 * 3);
2481 	ut_expected_io_set_iov(expected_io, 5, iov[2].iov_base + 512 * 3, 512 * 3);
2482 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2483 
2484 	/* The 2nd child IO has the left segment entry */
2485 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 1);
2486 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 6, 512 * 2);
2487 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2488 
2489 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 2, 1);
2490 	ut_expected_io_set_iov(expected_io, 0, iov[2].iov_base + 512 * 8, 512 * 2);
2491 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2492 
2493 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 0, 18, io_done, NULL);
2494 	CU_ASSERT(rc == 0);
2495 	CU_ASSERT(g_io_done == false);
2496 
2497 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
2498 	stub_complete_io(3);
2499 	CU_ASSERT(g_io_done == true);
2500 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2501 
2502 	/* A very complicated case. Each sg entry exceeds max_segment_size
2503 	 * and split on io boundary.
2504 	 * optimal_io_boundary < max_segment_size * max_num_segments
2505 	 */
2506 	bdev->max_segment_size = 3 * 512;
2507 	bdev->max_num_segments = BDEV_IO_NUM_CHILD_IOV;
2508 	g_io_done = false;
2509 
2510 	for (i = 0; i < 20; i++) {
2511 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
2512 		iov[i].iov_len = 512 * 4;
2513 	}
2514 
2515 	/* IO crossing the IO boundary requires split.
2516 	 * 80 block length can split 5 child IOs base on offset and IO boundary.
2517 	 * Each iov entry needs to be splitted to 2 entries because of max_segment_size
2518 	 * Total 5 child IOs.
2519 	 */
2520 
2521 	/* 4 iov entries are in an IO boundary and each iov entry splits to 2.
2522 	 * So each child IO occupies 8 child iov entries.
2523 	 */
2524 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 0, 16, 8);
2525 	for (i = 0; i < 4; i++) {
2526 		int iovcnt = i * 2;
2527 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2528 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2529 	}
2530 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2531 
2532 	/* 2nd child IO and total 16 child iov entries of parent IO */
2533 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 8);
2534 	for (i = 4; i < 8; i++) {
2535 		int iovcnt = (i - 4) * 2;
2536 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2537 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2538 	}
2539 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2540 
2541 	/* 3rd child IO and total 24 child iov entries of parent IO */
2542 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 16, 8);
2543 	for (i = 8; i < 12; i++) {
2544 		int iovcnt = (i - 8) * 2;
2545 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2546 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2547 	}
2548 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2549 
2550 	/* 4th child IO and total 32 child iov entries of parent IO */
2551 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 48, 16, 8);
2552 	for (i = 12; i < 16; i++) {
2553 		int iovcnt = (i - 12) * 2;
2554 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2555 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2556 	}
2557 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2558 
2559 	/* 5th child IO and because of the child iov entry it should be splitted
2560 	 * in next round.
2561 	 */
2562 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 64, 16, 8);
2563 	for (i = 16; i < 20; i++) {
2564 		int iovcnt = (i - 16) * 2;
2565 		ut_expected_io_set_iov(expected_io, iovcnt, iov[i].iov_base, 512 * 3);
2566 		ut_expected_io_set_iov(expected_io, iovcnt + 1, iov[i].iov_base + 512 * 3, 512);
2567 	}
2568 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2569 
2570 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 20, 0, 80, io_done, NULL);
2571 	CU_ASSERT(rc == 0);
2572 	CU_ASSERT(g_io_done == false);
2573 
2574 	/* First split round */
2575 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
2576 	stub_complete_io(4);
2577 	CU_ASSERT(g_io_done == false);
2578 
2579 	/* Second split round */
2580 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2581 	stub_complete_io(1);
2582 	CU_ASSERT(g_io_done == true);
2583 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2584 
2585 	spdk_put_io_channel(io_ch);
2586 	spdk_bdev_close(desc);
2587 	free_bdev(bdev);
2588 	spdk_bdev_finish(bdev_fini_cb, NULL);
2589 	poll_threads();
2590 }
2591 
2592 static void
2593 bdev_io_split_with_io_wait(void)
2594 {
2595 	struct spdk_bdev *bdev;
2596 	struct spdk_bdev_desc *desc = NULL;
2597 	struct spdk_io_channel *io_ch;
2598 	struct spdk_bdev_channel *channel;
2599 	struct spdk_bdev_mgmt_channel *mgmt_ch;
2600 	struct spdk_bdev_opts bdev_opts = {};
2601 	struct iovec iov[3];
2602 	struct ut_expected_io *expected_io;
2603 	int rc;
2604 
2605 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2606 	bdev_opts.bdev_io_pool_size = 2;
2607 	bdev_opts.bdev_io_cache_size = 1;
2608 
2609 	rc = spdk_bdev_set_opts(&bdev_opts);
2610 	CU_ASSERT(rc == 0);
2611 	spdk_bdev_initialize(bdev_init_cb, NULL);
2612 
2613 	bdev = allocate_bdev("bdev0");
2614 
2615 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2616 	CU_ASSERT(rc == 0);
2617 	CU_ASSERT(desc != NULL);
2618 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2619 	io_ch = spdk_bdev_get_io_channel(desc);
2620 	CU_ASSERT(io_ch != NULL);
2621 	channel = spdk_io_channel_get_ctx(io_ch);
2622 	mgmt_ch = channel->shared_resource->mgmt_ch;
2623 
2624 	bdev->optimal_io_boundary = 16;
2625 	bdev->split_on_optimal_io_boundary = true;
2626 
2627 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, NULL);
2628 	CU_ASSERT(rc == 0);
2629 
2630 	/* Now test that a single-vector command is split correctly.
2631 	 * Offset 14, length 8, payload 0xF000
2632 	 *  Child - Offset 14, length 2, payload 0xF000
2633 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
2634 	 *
2635 	 * Set up the expected values before calling spdk_bdev_read_blocks
2636 	 */
2637 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 14, 2, 1);
2638 	ut_expected_io_set_iov(expected_io, 0, (void *)0xF000, 2 * 512);
2639 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2640 
2641 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 16, 6, 1);
2642 	ut_expected_io_set_iov(expected_io, 0, (void *)(0xF000 + 2 * 512), 6 * 512);
2643 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2644 
2645 	/* The following children will be submitted sequentially due to the capacity of
2646 	 * spdk_bdev_io.
2647 	 */
2648 
2649 	/* The first child I/O will be queued to wait until an spdk_bdev_io becomes available */
2650 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL);
2651 	CU_ASSERT(rc == 0);
2652 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2653 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2654 
2655 	/* Completing the first read I/O will submit the first child */
2656 	stub_complete_io(1);
2657 	CU_ASSERT(TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
2658 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2659 
2660 	/* Completing the first child will submit the second child */
2661 	stub_complete_io(1);
2662 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2663 
2664 	/* Complete the second child I/O.  This should result in our callback getting
2665 	 * invoked since the parent I/O is now complete.
2666 	 */
2667 	stub_complete_io(1);
2668 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
2669 
2670 	/* Now set up a more complex, multi-vector command that needs to be split,
2671 	 *  including splitting iovecs.
2672 	 */
2673 	iov[0].iov_base = (void *)0x10000;
2674 	iov[0].iov_len = 512;
2675 	iov[1].iov_base = (void *)0x20000;
2676 	iov[1].iov_len = 20 * 512;
2677 	iov[2].iov_base = (void *)0x30000;
2678 	iov[2].iov_len = 11 * 512;
2679 
2680 	g_io_done = false;
2681 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 14, 2, 2);
2682 	ut_expected_io_set_iov(expected_io, 0, (void *)0x10000, 512);
2683 	ut_expected_io_set_iov(expected_io, 1, (void *)0x20000, 512);
2684 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2685 
2686 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 16, 16, 1);
2687 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 512), 16 * 512);
2688 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2689 
2690 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 2);
2691 	ut_expected_io_set_iov(expected_io, 0, (void *)(0x20000 + 17 * 512), 3 * 512);
2692 	ut_expected_io_set_iov(expected_io, 1, (void *)0x30000, 11 * 512);
2693 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
2694 
2695 	rc = spdk_bdev_writev_blocks(desc, io_ch, iov, 3, 14, 32, io_done, NULL);
2696 	CU_ASSERT(rc == 0);
2697 	CU_ASSERT(g_io_done == false);
2698 
2699 	/* The following children will be submitted sequentially due to the capacity of
2700 	 * spdk_bdev_io.
2701 	 */
2702 
2703 	/* Completing the first child will submit the second child */
2704 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2705 	stub_complete_io(1);
2706 	CU_ASSERT(g_io_done == false);
2707 
2708 	/* Completing the second child will submit the third child */
2709 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2710 	stub_complete_io(1);
2711 	CU_ASSERT(g_io_done == false);
2712 
2713 	/* Completing the third child will result in our callback getting invoked
2714 	 * since the parent I/O is now complete.
2715 	 */
2716 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
2717 	stub_complete_io(1);
2718 	CU_ASSERT(g_io_done == true);
2719 
2720 	CU_ASSERT(TAILQ_EMPTY(&g_bdev_ut_channel->expected_io));
2721 
2722 	spdk_put_io_channel(io_ch);
2723 	spdk_bdev_close(desc);
2724 	free_bdev(bdev);
2725 	spdk_bdev_finish(bdev_fini_cb, NULL);
2726 	poll_threads();
2727 }
2728 
2729 static void
2730 bdev_io_alignment(void)
2731 {
2732 	struct spdk_bdev *bdev;
2733 	struct spdk_bdev_desc *desc = NULL;
2734 	struct spdk_io_channel *io_ch;
2735 	struct spdk_bdev_opts bdev_opts = {};
2736 	int rc;
2737 	void *buf = NULL;
2738 	struct iovec iovs[2];
2739 	int iovcnt;
2740 	uint64_t alignment;
2741 
2742 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2743 	bdev_opts.bdev_io_pool_size = 20;
2744 	bdev_opts.bdev_io_cache_size = 2;
2745 
2746 	rc = spdk_bdev_set_opts(&bdev_opts);
2747 	CU_ASSERT(rc == 0);
2748 	spdk_bdev_initialize(bdev_init_cb, NULL);
2749 
2750 	fn_table.submit_request = stub_submit_request_get_buf;
2751 	bdev = allocate_bdev("bdev0");
2752 
2753 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2754 	CU_ASSERT(rc == 0);
2755 	CU_ASSERT(desc != NULL);
2756 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2757 	io_ch = spdk_bdev_get_io_channel(desc);
2758 	CU_ASSERT(io_ch != NULL);
2759 
2760 	/* Create aligned buffer */
2761 	rc = posix_memalign(&buf, 4096, 8192);
2762 	SPDK_CU_ASSERT_FATAL(rc == 0);
2763 
2764 	/* Pass aligned single buffer with no alignment required */
2765 	alignment = 1;
2766 	bdev->required_alignment = spdk_u32log2(alignment);
2767 
2768 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2769 	CU_ASSERT(rc == 0);
2770 	stub_complete_io(1);
2771 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2772 				    alignment));
2773 
2774 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 0, 1, io_done, NULL);
2775 	CU_ASSERT(rc == 0);
2776 	stub_complete_io(1);
2777 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2778 				    alignment));
2779 
2780 	/* Pass unaligned single buffer with no alignment required */
2781 	alignment = 1;
2782 	bdev->required_alignment = spdk_u32log2(alignment);
2783 
2784 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2785 	CU_ASSERT(rc == 0);
2786 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2787 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2788 	stub_complete_io(1);
2789 
2790 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2791 	CU_ASSERT(rc == 0);
2792 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2793 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == buf + 4);
2794 	stub_complete_io(1);
2795 
2796 	/* Pass unaligned single buffer with 512 alignment required */
2797 	alignment = 512;
2798 	bdev->required_alignment = spdk_u32log2(alignment);
2799 
2800 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2801 	CU_ASSERT(rc == 0);
2802 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2803 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2804 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2805 				    alignment));
2806 	stub_complete_io(1);
2807 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2808 
2809 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 4, 0, 1, io_done, NULL);
2810 	CU_ASSERT(rc == 0);
2811 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2812 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2813 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2814 				    alignment));
2815 	stub_complete_io(1);
2816 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2817 
2818 	/* Pass unaligned single buffer with 4096 alignment required */
2819 	alignment = 4096;
2820 	bdev->required_alignment = spdk_u32log2(alignment);
2821 
2822 	rc = spdk_bdev_write_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2823 	CU_ASSERT(rc == 0);
2824 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2825 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2826 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2827 				    alignment));
2828 	stub_complete_io(1);
2829 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2830 
2831 	rc = spdk_bdev_read_blocks(desc, io_ch, buf + 8, 0, 1, io_done, NULL);
2832 	CU_ASSERT(rc == 0);
2833 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 1);
2834 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2835 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2836 				    alignment));
2837 	stub_complete_io(1);
2838 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2839 
2840 	/* Pass aligned iovs with no alignment required */
2841 	alignment = 1;
2842 	bdev->required_alignment = spdk_u32log2(alignment);
2843 
2844 	iovcnt = 1;
2845 	iovs[0].iov_base = buf;
2846 	iovs[0].iov_len = 512;
2847 
2848 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2849 	CU_ASSERT(rc == 0);
2850 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2851 	stub_complete_io(1);
2852 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2853 
2854 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2855 	CU_ASSERT(rc == 0);
2856 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2857 	stub_complete_io(1);
2858 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2859 
2860 	/* Pass unaligned iovs with no alignment required */
2861 	alignment = 1;
2862 	bdev->required_alignment = spdk_u32log2(alignment);
2863 
2864 	iovcnt = 2;
2865 	iovs[0].iov_base = buf + 16;
2866 	iovs[0].iov_len = 256;
2867 	iovs[1].iov_base = buf + 16 + 256 + 32;
2868 	iovs[1].iov_len = 256;
2869 
2870 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2871 	CU_ASSERT(rc == 0);
2872 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2873 	stub_complete_io(1);
2874 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2875 
2876 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2877 	CU_ASSERT(rc == 0);
2878 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2879 	stub_complete_io(1);
2880 	CU_ASSERT(g_bdev_io->u.bdev.iovs[0].iov_base == iovs[0].iov_base);
2881 
2882 	/* Pass unaligned iov with 2048 alignment required */
2883 	alignment = 2048;
2884 	bdev->required_alignment = spdk_u32log2(alignment);
2885 
2886 	iovcnt = 2;
2887 	iovs[0].iov_base = buf + 16;
2888 	iovs[0].iov_len = 256;
2889 	iovs[1].iov_base = buf + 16 + 256 + 32;
2890 	iovs[1].iov_len = 256;
2891 
2892 	rc = spdk_bdev_writev(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2893 	CU_ASSERT(rc == 0);
2894 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2895 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2896 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2897 				    alignment));
2898 	stub_complete_io(1);
2899 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2900 
2901 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2902 	CU_ASSERT(rc == 0);
2903 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == iovcnt);
2904 	CU_ASSERT(g_bdev_io->u.bdev.iovs == &g_bdev_io->internal.bounce_iov);
2905 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2906 				    alignment));
2907 	stub_complete_io(1);
2908 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2909 
2910 	/* Pass iov without allocated buffer without alignment required */
2911 	alignment = 1;
2912 	bdev->required_alignment = spdk_u32log2(alignment);
2913 
2914 	iovcnt = 1;
2915 	iovs[0].iov_base = NULL;
2916 	iovs[0].iov_len = 0;
2917 
2918 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2919 	CU_ASSERT(rc == 0);
2920 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2921 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2922 				    alignment));
2923 	stub_complete_io(1);
2924 
2925 	/* Pass iov without allocated buffer with 1024 alignment required */
2926 	alignment = 1024;
2927 	bdev->required_alignment = spdk_u32log2(alignment);
2928 
2929 	iovcnt = 1;
2930 	iovs[0].iov_base = NULL;
2931 	iovs[0].iov_len = 0;
2932 
2933 	rc = spdk_bdev_readv(desc, io_ch, iovs, iovcnt, 0, 512, io_done, NULL);
2934 	CU_ASSERT(rc == 0);
2935 	CU_ASSERT(g_bdev_io->internal.orig_iovcnt == 0);
2936 	CU_ASSERT(_are_iovs_aligned(g_bdev_io->u.bdev.iovs, g_bdev_io->u.bdev.iovcnt,
2937 				    alignment));
2938 	stub_complete_io(1);
2939 
2940 	spdk_put_io_channel(io_ch);
2941 	spdk_bdev_close(desc);
2942 	free_bdev(bdev);
2943 	fn_table.submit_request = stub_submit_request;
2944 	spdk_bdev_finish(bdev_fini_cb, NULL);
2945 	poll_threads();
2946 
2947 	free(buf);
2948 }
2949 
2950 static void
2951 bdev_io_alignment_with_boundary(void)
2952 {
2953 	struct spdk_bdev *bdev;
2954 	struct spdk_bdev_desc *desc = NULL;
2955 	struct spdk_io_channel *io_ch;
2956 	struct spdk_bdev_opts bdev_opts = {};
2957 	int rc;
2958 	void *buf = NULL;
2959 	struct iovec iovs[2];
2960 	int iovcnt;
2961 	uint64_t alignment;
2962 
2963 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
2964 	bdev_opts.bdev_io_pool_size = 20;
2965 	bdev_opts.bdev_io_cache_size = 2;
2966 
2967 	bdev_opts.opts_size = sizeof(bdev_opts);
2968 	rc = spdk_bdev_set_opts(&bdev_opts);
2969 	CU_ASSERT(rc == 0);
2970 	spdk_bdev_initialize(bdev_init_cb, NULL);
2971 
2972 	fn_table.submit_request = stub_submit_request_get_buf;
2973 	bdev = allocate_bdev("bdev0");
2974 
2975 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
2976 	CU_ASSERT(rc == 0);
2977 	CU_ASSERT(desc != NULL);
2978 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
2979 	io_ch = spdk_bdev_get_io_channel(desc);
2980 	CU_ASSERT(io_ch != NULL);
2981 
2982 	/* Create aligned buffer */
2983 	rc = posix_memalign(&buf, 4096, 131072);
2984 	SPDK_CU_ASSERT_FATAL(rc == 0);
2985 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
2986 
2987 	/* 512 * 3 with 2 IO boundary, allocate small data buffer from bdev layer */
2988 	alignment = 512;
2989 	bdev->required_alignment = spdk_u32log2(alignment);
2990 	bdev->optimal_io_boundary = 2;
2991 	bdev->split_on_optimal_io_boundary = true;
2992 
2993 	iovcnt = 1;
2994 	iovs[0].iov_base = NULL;
2995 	iovs[0].iov_len = 512 * 3;
2996 
2997 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
2998 	CU_ASSERT(rc == 0);
2999 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3000 	stub_complete_io(2);
3001 
3002 	/* 8KiB with 16 IO boundary, allocate large data buffer from bdev layer */
3003 	alignment = 512;
3004 	bdev->required_alignment = spdk_u32log2(alignment);
3005 	bdev->optimal_io_boundary = 16;
3006 	bdev->split_on_optimal_io_boundary = true;
3007 
3008 	iovcnt = 1;
3009 	iovs[0].iov_base = NULL;
3010 	iovs[0].iov_len = 512 * 16;
3011 
3012 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 16, io_done, NULL);
3013 	CU_ASSERT(rc == 0);
3014 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3015 	stub_complete_io(2);
3016 
3017 	/* 512 * 160 with 128 IO boundary, 63.5KiB + 16.5KiB for the two children requests */
3018 	alignment = 512;
3019 	bdev->required_alignment = spdk_u32log2(alignment);
3020 	bdev->optimal_io_boundary = 128;
3021 	bdev->split_on_optimal_io_boundary = true;
3022 
3023 	iovcnt = 1;
3024 	iovs[0].iov_base = buf + 16;
3025 	iovs[0].iov_len = 512 * 160;
3026 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3027 	CU_ASSERT(rc == 0);
3028 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3029 	stub_complete_io(2);
3030 
3031 	/* 512 * 3 with 2 IO boundary */
3032 	alignment = 512;
3033 	bdev->required_alignment = spdk_u32log2(alignment);
3034 	bdev->optimal_io_boundary = 2;
3035 	bdev->split_on_optimal_io_boundary = true;
3036 
3037 	iovcnt = 2;
3038 	iovs[0].iov_base = buf + 16;
3039 	iovs[0].iov_len = 512;
3040 	iovs[1].iov_base = buf + 16 + 512 + 32;
3041 	iovs[1].iov_len = 1024;
3042 
3043 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3044 	CU_ASSERT(rc == 0);
3045 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3046 	stub_complete_io(2);
3047 
3048 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 3, io_done, NULL);
3049 	CU_ASSERT(rc == 0);
3050 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
3051 	stub_complete_io(2);
3052 
3053 	/* 512 * 64 with 32 IO boundary */
3054 	bdev->optimal_io_boundary = 32;
3055 	iovcnt = 2;
3056 	iovs[0].iov_base = buf + 16;
3057 	iovs[0].iov_len = 16384;
3058 	iovs[1].iov_base = buf + 16 + 16384 + 32;
3059 	iovs[1].iov_len = 16384;
3060 
3061 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3062 	CU_ASSERT(rc == 0);
3063 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3064 	stub_complete_io(3);
3065 
3066 	rc = spdk_bdev_readv_blocks(desc, io_ch, iovs, iovcnt, 1, 64, io_done, NULL);
3067 	CU_ASSERT(rc == 0);
3068 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 3);
3069 	stub_complete_io(3);
3070 
3071 	/* 512 * 160 with 32 IO boundary */
3072 	iovcnt = 1;
3073 	iovs[0].iov_base = buf + 16;
3074 	iovs[0].iov_len = 16384 + 65536;
3075 
3076 	rc = spdk_bdev_writev_blocks(desc, io_ch, iovs, iovcnt, 1, 160, io_done, NULL);
3077 	CU_ASSERT(rc == 0);
3078 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 6);
3079 	stub_complete_io(6);
3080 
3081 	spdk_put_io_channel(io_ch);
3082 	spdk_bdev_close(desc);
3083 	free_bdev(bdev);
3084 	fn_table.submit_request = stub_submit_request;
3085 	spdk_bdev_finish(bdev_fini_cb, NULL);
3086 	poll_threads();
3087 
3088 	free(buf);
3089 }
3090 
3091 static void
3092 histogram_status_cb(void *cb_arg, int status)
3093 {
3094 	g_status = status;
3095 }
3096 
3097 static void
3098 histogram_data_cb(void *cb_arg, int status, struct spdk_histogram_data *histogram)
3099 {
3100 	g_status = status;
3101 	g_histogram = histogram;
3102 }
3103 
3104 static void
3105 histogram_io_count(void *ctx, uint64_t start, uint64_t end, uint64_t count,
3106 		   uint64_t total, uint64_t so_far)
3107 {
3108 	g_count += count;
3109 }
3110 
3111 static void
3112 bdev_histograms(void)
3113 {
3114 	struct spdk_bdev *bdev;
3115 	struct spdk_bdev_desc *desc = NULL;
3116 	struct spdk_io_channel *ch;
3117 	struct spdk_histogram_data *histogram;
3118 	uint8_t buf[4096];
3119 	int rc;
3120 
3121 	spdk_bdev_initialize(bdev_init_cb, NULL);
3122 
3123 	bdev = allocate_bdev("bdev");
3124 
3125 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3126 	CU_ASSERT(rc == 0);
3127 	CU_ASSERT(desc != NULL);
3128 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3129 
3130 	ch = spdk_bdev_get_io_channel(desc);
3131 	CU_ASSERT(ch != NULL);
3132 
3133 	/* Enable histogram */
3134 	g_status = -1;
3135 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, true);
3136 	poll_threads();
3137 	CU_ASSERT(g_status == 0);
3138 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3139 
3140 	/* Allocate histogram */
3141 	histogram = spdk_histogram_data_alloc();
3142 	SPDK_CU_ASSERT_FATAL(histogram != NULL);
3143 
3144 	/* Check if histogram is zeroed */
3145 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3146 	poll_threads();
3147 	CU_ASSERT(g_status == 0);
3148 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3149 
3150 	g_count = 0;
3151 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3152 
3153 	CU_ASSERT(g_count == 0);
3154 
3155 	rc = spdk_bdev_write_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3156 	CU_ASSERT(rc == 0);
3157 
3158 	spdk_delay_us(10);
3159 	stub_complete_io(1);
3160 	poll_threads();
3161 
3162 	rc = spdk_bdev_read_blocks(desc, ch, buf, 0, 1, io_done, NULL);
3163 	CU_ASSERT(rc == 0);
3164 
3165 	spdk_delay_us(10);
3166 	stub_complete_io(1);
3167 	poll_threads();
3168 
3169 	/* Check if histogram gathered data from all I/O channels */
3170 	g_histogram = NULL;
3171 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3172 	poll_threads();
3173 	CU_ASSERT(g_status == 0);
3174 	CU_ASSERT(bdev->internal.histogram_enabled == true);
3175 	SPDK_CU_ASSERT_FATAL(g_histogram != NULL);
3176 
3177 	g_count = 0;
3178 	spdk_histogram_data_iterate(g_histogram, histogram_io_count, NULL);
3179 	CU_ASSERT(g_count == 2);
3180 
3181 	/* Disable histogram */
3182 	spdk_bdev_histogram_enable(bdev, histogram_status_cb, NULL, false);
3183 	poll_threads();
3184 	CU_ASSERT(g_status == 0);
3185 	CU_ASSERT(bdev->internal.histogram_enabled == false);
3186 
3187 	/* Try to run histogram commands on disabled bdev */
3188 	spdk_bdev_histogram_get(bdev, histogram, histogram_data_cb, NULL);
3189 	poll_threads();
3190 	CU_ASSERT(g_status == -EFAULT);
3191 
3192 	spdk_histogram_data_free(histogram);
3193 	spdk_put_io_channel(ch);
3194 	spdk_bdev_close(desc);
3195 	free_bdev(bdev);
3196 	spdk_bdev_finish(bdev_fini_cb, NULL);
3197 	poll_threads();
3198 }
3199 
3200 static void
3201 _bdev_compare(bool emulated)
3202 {
3203 	struct spdk_bdev *bdev;
3204 	struct spdk_bdev_desc *desc = NULL;
3205 	struct spdk_io_channel *ioch;
3206 	struct ut_expected_io *expected_io;
3207 	uint64_t offset, num_blocks;
3208 	uint32_t num_completed;
3209 	char aa_buf[512];
3210 	char bb_buf[512];
3211 	struct iovec compare_iov;
3212 	uint8_t io_type;
3213 	int rc;
3214 
3215 	if (emulated) {
3216 		io_type = SPDK_BDEV_IO_TYPE_READ;
3217 	} else {
3218 		io_type = SPDK_BDEV_IO_TYPE_COMPARE;
3219 	}
3220 
3221 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3222 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3223 
3224 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = !emulated;
3225 
3226 	spdk_bdev_initialize(bdev_init_cb, NULL);
3227 	fn_table.submit_request = stub_submit_request_get_buf;
3228 	bdev = allocate_bdev("bdev");
3229 
3230 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3231 	CU_ASSERT_EQUAL(rc, 0);
3232 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3233 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3234 	ioch = spdk_bdev_get_io_channel(desc);
3235 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3236 
3237 	fn_table.submit_request = stub_submit_request_get_buf;
3238 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3239 
3240 	offset = 50;
3241 	num_blocks = 1;
3242 	compare_iov.iov_base = aa_buf;
3243 	compare_iov.iov_len = sizeof(aa_buf);
3244 
3245 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3246 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3247 
3248 	g_io_done = false;
3249 	g_compare_read_buf = aa_buf;
3250 	g_compare_read_buf_len = sizeof(aa_buf);
3251 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3252 	CU_ASSERT_EQUAL(rc, 0);
3253 	num_completed = stub_complete_io(1);
3254 	CU_ASSERT_EQUAL(num_completed, 1);
3255 	CU_ASSERT(g_io_done == true);
3256 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3257 
3258 	expected_io = ut_alloc_expected_io(io_type, offset, num_blocks, 0);
3259 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3260 
3261 	g_io_done = false;
3262 	g_compare_read_buf = bb_buf;
3263 	g_compare_read_buf_len = sizeof(bb_buf);
3264 	rc = spdk_bdev_comparev_blocks(desc, ioch, &compare_iov, 1, offset, num_blocks, io_done, NULL);
3265 	CU_ASSERT_EQUAL(rc, 0);
3266 	num_completed = stub_complete_io(1);
3267 	CU_ASSERT_EQUAL(num_completed, 1);
3268 	CU_ASSERT(g_io_done == true);
3269 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3270 
3271 	spdk_put_io_channel(ioch);
3272 	spdk_bdev_close(desc);
3273 	free_bdev(bdev);
3274 	fn_table.submit_request = stub_submit_request;
3275 	spdk_bdev_finish(bdev_fini_cb, NULL);
3276 	poll_threads();
3277 
3278 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3279 
3280 	g_compare_read_buf = NULL;
3281 }
3282 
3283 static void
3284 bdev_compare(void)
3285 {
3286 	_bdev_compare(true);
3287 	_bdev_compare(false);
3288 }
3289 
3290 static void
3291 bdev_compare_and_write(void)
3292 {
3293 	struct spdk_bdev *bdev;
3294 	struct spdk_bdev_desc *desc = NULL;
3295 	struct spdk_io_channel *ioch;
3296 	struct ut_expected_io *expected_io;
3297 	uint64_t offset, num_blocks;
3298 	uint32_t num_completed;
3299 	char aa_buf[512];
3300 	char bb_buf[512];
3301 	char cc_buf[512];
3302 	char write_buf[512];
3303 	struct iovec compare_iov;
3304 	struct iovec write_iov;
3305 	int rc;
3306 
3307 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3308 	memset(bb_buf, 0xbb, sizeof(bb_buf));
3309 	memset(cc_buf, 0xcc, sizeof(cc_buf));
3310 
3311 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = false;
3312 
3313 	spdk_bdev_initialize(bdev_init_cb, NULL);
3314 	fn_table.submit_request = stub_submit_request_get_buf;
3315 	bdev = allocate_bdev("bdev");
3316 
3317 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3318 	CU_ASSERT_EQUAL(rc, 0);
3319 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3320 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3321 	ioch = spdk_bdev_get_io_channel(desc);
3322 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3323 
3324 	fn_table.submit_request = stub_submit_request_get_buf;
3325 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3326 
3327 	offset = 50;
3328 	num_blocks = 1;
3329 	compare_iov.iov_base = aa_buf;
3330 	compare_iov.iov_len = sizeof(aa_buf);
3331 	write_iov.iov_base = bb_buf;
3332 	write_iov.iov_len = sizeof(bb_buf);
3333 
3334 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3335 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3336 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, offset, num_blocks, 0);
3337 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3338 
3339 	g_io_done = false;
3340 	g_compare_read_buf = aa_buf;
3341 	g_compare_read_buf_len = sizeof(aa_buf);
3342 	memset(write_buf, 0, sizeof(write_buf));
3343 	g_compare_write_buf = write_buf;
3344 	g_compare_write_buf_len = sizeof(write_buf);
3345 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3346 			offset, num_blocks, io_done, NULL);
3347 	/* Trigger range locking */
3348 	poll_threads();
3349 	CU_ASSERT_EQUAL(rc, 0);
3350 	num_completed = stub_complete_io(1);
3351 	CU_ASSERT_EQUAL(num_completed, 1);
3352 	CU_ASSERT(g_io_done == false);
3353 	num_completed = stub_complete_io(1);
3354 	/* Trigger range unlocking */
3355 	poll_threads();
3356 	CU_ASSERT_EQUAL(num_completed, 1);
3357 	CU_ASSERT(g_io_done == true);
3358 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3359 	CU_ASSERT(memcmp(write_buf, bb_buf, sizeof(write_buf)) == 0);
3360 
3361 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, offset, num_blocks, 0);
3362 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3363 
3364 	g_io_done = false;
3365 	g_compare_read_buf = cc_buf;
3366 	g_compare_read_buf_len = sizeof(cc_buf);
3367 	memset(write_buf, 0, sizeof(write_buf));
3368 	g_compare_write_buf = write_buf;
3369 	g_compare_write_buf_len = sizeof(write_buf);
3370 	rc = spdk_bdev_comparev_and_writev_blocks(desc, ioch, &compare_iov, 1, &write_iov, 1,
3371 			offset, num_blocks, io_done, NULL);
3372 	/* Trigger range locking */
3373 	poll_threads();
3374 	CU_ASSERT_EQUAL(rc, 0);
3375 	num_completed = stub_complete_io(1);
3376 	/* Trigger range unlocking earlier because we expect error here */
3377 	poll_threads();
3378 	CU_ASSERT_EQUAL(num_completed, 1);
3379 	CU_ASSERT(g_io_done == true);
3380 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_MISCOMPARE);
3381 	num_completed = stub_complete_io(1);
3382 	CU_ASSERT_EQUAL(num_completed, 0);
3383 
3384 	spdk_put_io_channel(ioch);
3385 	spdk_bdev_close(desc);
3386 	free_bdev(bdev);
3387 	fn_table.submit_request = stub_submit_request;
3388 	spdk_bdev_finish(bdev_fini_cb, NULL);
3389 	poll_threads();
3390 
3391 	g_io_types_supported[SPDK_BDEV_IO_TYPE_COMPARE] = true;
3392 
3393 	g_compare_read_buf = NULL;
3394 	g_compare_write_buf = NULL;
3395 }
3396 
3397 static void
3398 bdev_write_zeroes(void)
3399 {
3400 	struct spdk_bdev *bdev;
3401 	struct spdk_bdev_desc *desc = NULL;
3402 	struct spdk_io_channel *ioch;
3403 	struct ut_expected_io *expected_io;
3404 	uint64_t offset, num_io_blocks, num_blocks;
3405 	uint32_t num_completed, num_requests;
3406 	int rc;
3407 
3408 	spdk_bdev_initialize(bdev_init_cb, NULL);
3409 	bdev = allocate_bdev("bdev");
3410 
3411 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3412 	CU_ASSERT_EQUAL(rc, 0);
3413 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3414 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3415 	ioch = spdk_bdev_get_io_channel(desc);
3416 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3417 
3418 	fn_table.submit_request = stub_submit_request;
3419 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3420 
3421 	/* First test that if the bdev supports write_zeroes, the request won't be split */
3422 	bdev->md_len = 0;
3423 	bdev->blocklen = 4096;
3424 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3425 
3426 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
3427 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3428 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3429 	CU_ASSERT_EQUAL(rc, 0);
3430 	num_completed = stub_complete_io(1);
3431 	CU_ASSERT_EQUAL(num_completed, 1);
3432 
3433 	/* Check that if write zeroes is not supported it'll be replaced by regular writes */
3434 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, false);
3435 	num_io_blocks = ZERO_BUFFER_SIZE / bdev->blocklen;
3436 	num_requests = 2;
3437 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * num_requests;
3438 
3439 	for (offset = 0; offset < num_requests; ++offset) {
3440 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3441 						   offset * num_io_blocks, num_io_blocks, 0);
3442 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3443 	}
3444 
3445 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3446 	CU_ASSERT_EQUAL(rc, 0);
3447 	num_completed = stub_complete_io(num_requests);
3448 	CU_ASSERT_EQUAL(num_completed, num_requests);
3449 
3450 	/* Check that the splitting is correct if bdev has interleaved metadata */
3451 	bdev->md_interleave = true;
3452 	bdev->md_len = 64;
3453 	bdev->blocklen = 4096 + 64;
3454 	num_blocks = (ZERO_BUFFER_SIZE / bdev->blocklen) * 2;
3455 
3456 	num_requests = offset = 0;
3457 	while (offset < num_blocks) {
3458 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / bdev->blocklen, num_blocks - offset);
3459 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3460 						   offset, num_io_blocks, 0);
3461 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3462 		offset += num_io_blocks;
3463 		num_requests++;
3464 	}
3465 
3466 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3467 	CU_ASSERT_EQUAL(rc, 0);
3468 	num_completed = stub_complete_io(num_requests);
3469 	CU_ASSERT_EQUAL(num_completed, num_requests);
3470 	num_completed = stub_complete_io(num_requests);
3471 	assert(num_completed == 0);
3472 
3473 	/* Check the the same for separate metadata buffer */
3474 	bdev->md_interleave = false;
3475 	bdev->md_len = 64;
3476 	bdev->blocklen = 4096;
3477 
3478 	num_requests = offset = 0;
3479 	while (offset < num_blocks) {
3480 		num_io_blocks = spdk_min(ZERO_BUFFER_SIZE / (bdev->blocklen + bdev->md_len), num_blocks);
3481 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE,
3482 						   offset, num_io_blocks, 0);
3483 		expected_io->md_buf = (char *)g_bdev_mgr.zero_buffer + num_io_blocks * bdev->blocklen;
3484 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3485 		offset += num_io_blocks;
3486 		num_requests++;
3487 	}
3488 
3489 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
3490 	CU_ASSERT_EQUAL(rc, 0);
3491 	num_completed = stub_complete_io(num_requests);
3492 	CU_ASSERT_EQUAL(num_completed, num_requests);
3493 
3494 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, true);
3495 	spdk_put_io_channel(ioch);
3496 	spdk_bdev_close(desc);
3497 	free_bdev(bdev);
3498 	spdk_bdev_finish(bdev_fini_cb, NULL);
3499 	poll_threads();
3500 }
3501 
3502 static void
3503 bdev_zcopy_write(void)
3504 {
3505 	struct spdk_bdev *bdev;
3506 	struct spdk_bdev_desc *desc = NULL;
3507 	struct spdk_io_channel *ioch;
3508 	struct ut_expected_io *expected_io;
3509 	uint64_t offset, num_blocks;
3510 	uint32_t num_completed;
3511 	char aa_buf[512];
3512 	struct iovec iov;
3513 	int rc;
3514 	const bool populate = false;
3515 	const bool commit = true;
3516 
3517 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3518 
3519 	spdk_bdev_initialize(bdev_init_cb, NULL);
3520 	bdev = allocate_bdev("bdev");
3521 
3522 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3523 	CU_ASSERT_EQUAL(rc, 0);
3524 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3525 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3526 	ioch = spdk_bdev_get_io_channel(desc);
3527 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3528 
3529 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3530 
3531 	offset = 50;
3532 	num_blocks = 1;
3533 	iov.iov_base = NULL;
3534 	iov.iov_len = 0;
3535 
3536 	g_zcopy_read_buf = (void *) 0x1122334455667788UL;
3537 	g_zcopy_read_buf_len = (uint32_t) -1;
3538 	/* Do a zcopy start for a write (populate=false) */
3539 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3540 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3541 	g_io_done = false;
3542 	g_zcopy_write_buf = aa_buf;
3543 	g_zcopy_write_buf_len = sizeof(aa_buf);
3544 	g_zcopy_bdev_io = NULL;
3545 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3546 	CU_ASSERT_EQUAL(rc, 0);
3547 	num_completed = stub_complete_io(1);
3548 	CU_ASSERT_EQUAL(num_completed, 1);
3549 	CU_ASSERT(g_io_done == true);
3550 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3551 	/* Check that the iov has been set up */
3552 	CU_ASSERT(iov.iov_base == g_zcopy_write_buf);
3553 	CU_ASSERT(iov.iov_len == g_zcopy_write_buf_len);
3554 	/* Check that the bdev_io has been saved */
3555 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3556 	/* Now do the zcopy end for a write (commit=true) */
3557 	g_io_done = false;
3558 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3559 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3560 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3561 	CU_ASSERT_EQUAL(rc, 0);
3562 	num_completed = stub_complete_io(1);
3563 	CU_ASSERT_EQUAL(num_completed, 1);
3564 	CU_ASSERT(g_io_done == true);
3565 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3566 	/* Check the g_zcopy are reset by io_done */
3567 	CU_ASSERT(g_zcopy_write_buf == NULL);
3568 	CU_ASSERT(g_zcopy_write_buf_len == 0);
3569 	/* Check that io_done has freed the g_zcopy_bdev_io */
3570 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3571 
3572 	/* Check the zcopy read buffer has not been touched which
3573 	 * ensures that the correct buffers were used.
3574 	 */
3575 	CU_ASSERT(g_zcopy_read_buf == (void *) 0x1122334455667788UL);
3576 	CU_ASSERT(g_zcopy_read_buf_len == (uint32_t) -1);
3577 
3578 	spdk_put_io_channel(ioch);
3579 	spdk_bdev_close(desc);
3580 	free_bdev(bdev);
3581 	spdk_bdev_finish(bdev_fini_cb, NULL);
3582 	poll_threads();
3583 }
3584 
3585 static void
3586 bdev_zcopy_read(void)
3587 {
3588 	struct spdk_bdev *bdev;
3589 	struct spdk_bdev_desc *desc = NULL;
3590 	struct spdk_io_channel *ioch;
3591 	struct ut_expected_io *expected_io;
3592 	uint64_t offset, num_blocks;
3593 	uint32_t num_completed;
3594 	char aa_buf[512];
3595 	struct iovec iov;
3596 	int rc;
3597 	const bool populate = true;
3598 	const bool commit = false;
3599 
3600 	memset(aa_buf, 0xaa, sizeof(aa_buf));
3601 
3602 	spdk_bdev_initialize(bdev_init_cb, NULL);
3603 	bdev = allocate_bdev("bdev");
3604 
3605 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
3606 	CU_ASSERT_EQUAL(rc, 0);
3607 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3608 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3609 	ioch = spdk_bdev_get_io_channel(desc);
3610 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
3611 
3612 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
3613 
3614 	offset = 50;
3615 	num_blocks = 1;
3616 	iov.iov_base = NULL;
3617 	iov.iov_len = 0;
3618 
3619 	g_zcopy_write_buf = (void *) 0x1122334455667788UL;
3620 	g_zcopy_write_buf_len = (uint32_t) -1;
3621 
3622 	/* Do a zcopy start for a read (populate=true) */
3623 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3624 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3625 	g_io_done = false;
3626 	g_zcopy_read_buf = aa_buf;
3627 	g_zcopy_read_buf_len = sizeof(aa_buf);
3628 	g_zcopy_bdev_io = NULL;
3629 	rc = spdk_bdev_zcopy_start(desc, ioch, &iov, 1, offset, num_blocks, populate, io_done, NULL);
3630 	CU_ASSERT_EQUAL(rc, 0);
3631 	num_completed = stub_complete_io(1);
3632 	CU_ASSERT_EQUAL(num_completed, 1);
3633 	CU_ASSERT(g_io_done == true);
3634 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3635 	/* Check that the iov has been set up */
3636 	CU_ASSERT(iov.iov_base == g_zcopy_read_buf);
3637 	CU_ASSERT(iov.iov_len == g_zcopy_read_buf_len);
3638 	/* Check that the bdev_io has been saved */
3639 	CU_ASSERT(g_zcopy_bdev_io != NULL);
3640 
3641 	/* Now do the zcopy end for a read (commit=false) */
3642 	g_io_done = false;
3643 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_ZCOPY, offset, num_blocks, 0);
3644 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
3645 	rc = spdk_bdev_zcopy_end(g_zcopy_bdev_io, commit, io_done, NULL);
3646 	CU_ASSERT_EQUAL(rc, 0);
3647 	num_completed = stub_complete_io(1);
3648 	CU_ASSERT_EQUAL(num_completed, 1);
3649 	CU_ASSERT(g_io_done == true);
3650 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
3651 	/* Check the g_zcopy are reset by io_done */
3652 	CU_ASSERT(g_zcopy_read_buf == NULL);
3653 	CU_ASSERT(g_zcopy_read_buf_len == 0);
3654 	/* Check that io_done has freed the g_zcopy_bdev_io */
3655 	CU_ASSERT(g_zcopy_bdev_io == NULL);
3656 
3657 	/* Check the zcopy write buffer has not been touched which
3658 	 * ensures that the correct buffers were used.
3659 	 */
3660 	CU_ASSERT(g_zcopy_write_buf == (void *) 0x1122334455667788UL);
3661 	CU_ASSERT(g_zcopy_write_buf_len == (uint32_t) -1);
3662 
3663 	spdk_put_io_channel(ioch);
3664 	spdk_bdev_close(desc);
3665 	free_bdev(bdev);
3666 	spdk_bdev_finish(bdev_fini_cb, NULL);
3667 	poll_threads();
3668 }
3669 
3670 static void
3671 bdev_open_while_hotremove(void)
3672 {
3673 	struct spdk_bdev *bdev;
3674 	struct spdk_bdev_desc *desc[2] = {};
3675 	int rc;
3676 
3677 	bdev = allocate_bdev("bdev");
3678 
3679 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[0]);
3680 	CU_ASSERT(rc == 0);
3681 	SPDK_CU_ASSERT_FATAL(desc[0] != NULL);
3682 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc[0]));
3683 
3684 	spdk_bdev_unregister(bdev, NULL, NULL);
3685 
3686 	rc = spdk_bdev_open_ext("bdev", false, bdev_ut_event_cb, NULL, &desc[1]);
3687 	CU_ASSERT(rc == -ENODEV);
3688 	SPDK_CU_ASSERT_FATAL(desc[1] == NULL);
3689 
3690 	spdk_bdev_close(desc[0]);
3691 	free_bdev(bdev);
3692 }
3693 
3694 static void
3695 bdev_close_while_hotremove(void)
3696 {
3697 	struct spdk_bdev *bdev;
3698 	struct spdk_bdev_desc *desc = NULL;
3699 	int rc = 0;
3700 
3701 	bdev = allocate_bdev("bdev");
3702 
3703 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc, &desc);
3704 	CU_ASSERT_EQUAL(rc, 0);
3705 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3706 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3707 
3708 	/* Simulate hot-unplug by unregistering bdev */
3709 	g_event_type1 = 0xFF;
3710 	g_unregister_arg = NULL;
3711 	g_unregister_rc = -1;
3712 	spdk_bdev_unregister(bdev, bdev_unregister_cb, (void *)0x12345678);
3713 	/* Close device while remove event is in flight */
3714 	spdk_bdev_close(desc);
3715 
3716 	/* Ensure that unregister callback is delayed */
3717 	CU_ASSERT_EQUAL(g_unregister_arg, NULL);
3718 	CU_ASSERT_EQUAL(g_unregister_rc, -1);
3719 
3720 	poll_threads();
3721 
3722 	/* Event callback shall not be issued because device was closed */
3723 	CU_ASSERT_EQUAL(g_event_type1, 0xFF);
3724 	/* Unregister callback is issued */
3725 	CU_ASSERT_EQUAL(g_unregister_arg, (void *)0x12345678);
3726 	CU_ASSERT_EQUAL(g_unregister_rc, 0);
3727 
3728 	free_bdev(bdev);
3729 }
3730 
3731 static void
3732 bdev_open_ext(void)
3733 {
3734 	struct spdk_bdev *bdev;
3735 	struct spdk_bdev_desc *desc1 = NULL;
3736 	struct spdk_bdev_desc *desc2 = NULL;
3737 	int rc = 0;
3738 
3739 	bdev = allocate_bdev("bdev");
3740 
3741 	rc = spdk_bdev_open_ext("bdev", true, NULL, NULL, &desc1);
3742 	CU_ASSERT_EQUAL(rc, -EINVAL);
3743 
3744 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb1, &desc1, &desc1);
3745 	CU_ASSERT_EQUAL(rc, 0);
3746 
3747 	rc = spdk_bdev_open_ext("bdev", true, bdev_open_cb2, &desc2, &desc2);
3748 	CU_ASSERT_EQUAL(rc, 0);
3749 
3750 	g_event_type1 = 0xFF;
3751 	g_event_type2 = 0xFF;
3752 
3753 	/* Simulate hot-unplug by unregistering bdev */
3754 	spdk_bdev_unregister(bdev, NULL, NULL);
3755 	poll_threads();
3756 
3757 	/* Check if correct events have been triggered in event callback fn */
3758 	CU_ASSERT_EQUAL(g_event_type1, SPDK_BDEV_EVENT_REMOVE);
3759 	CU_ASSERT_EQUAL(g_event_type2, SPDK_BDEV_EVENT_REMOVE);
3760 
3761 	free_bdev(bdev);
3762 	poll_threads();
3763 }
3764 
3765 struct timeout_io_cb_arg {
3766 	struct iovec iov;
3767 	uint8_t type;
3768 };
3769 
3770 static int
3771 bdev_channel_count_submitted_io(struct spdk_bdev_channel *ch)
3772 {
3773 	struct spdk_bdev_io *bdev_io;
3774 	int n = 0;
3775 
3776 	if (!ch) {
3777 		return -1;
3778 	}
3779 
3780 	TAILQ_FOREACH(bdev_io, &ch->io_submitted, internal.ch_link) {
3781 		n++;
3782 	}
3783 
3784 	return n;
3785 }
3786 
3787 static void
3788 bdev_channel_io_timeout_cb(void *cb_arg, struct spdk_bdev_io *bdev_io)
3789 {
3790 	struct timeout_io_cb_arg *ctx = cb_arg;
3791 
3792 	ctx->type = bdev_io->type;
3793 	ctx->iov.iov_base = bdev_io->iov.iov_base;
3794 	ctx->iov.iov_len = bdev_io->iov.iov_len;
3795 }
3796 
3797 static void
3798 bdev_set_io_timeout(void)
3799 {
3800 	struct spdk_bdev *bdev;
3801 	struct spdk_bdev_desc *desc = NULL;
3802 	struct spdk_io_channel *io_ch = NULL;
3803 	struct spdk_bdev_channel *bdev_ch = NULL;
3804 	struct timeout_io_cb_arg cb_arg;
3805 
3806 	spdk_bdev_initialize(bdev_init_cb, NULL);
3807 
3808 	bdev = allocate_bdev("bdev");
3809 
3810 	CU_ASSERT(spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc) == 0);
3811 	SPDK_CU_ASSERT_FATAL(desc != NULL);
3812 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
3813 
3814 	io_ch = spdk_bdev_get_io_channel(desc);
3815 	CU_ASSERT(io_ch != NULL);
3816 
3817 	bdev_ch = spdk_io_channel_get_ctx(io_ch);
3818 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
3819 
3820 	/* This is the part1.
3821 	 * We will check the bdev_ch->io_submitted list
3822 	 * TO make sure that it can link IOs and only the user submitted IOs
3823 	 */
3824 	CU_ASSERT(spdk_bdev_read(desc, io_ch, (void *)0x1000, 0, 4096, io_done, NULL) == 0);
3825 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3826 	CU_ASSERT(spdk_bdev_write(desc, io_ch, (void *)0x2000, 0, 4096, io_done, NULL) == 0);
3827 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3828 	stub_complete_io(1);
3829 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3830 	stub_complete_io(1);
3831 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3832 
3833 	/* Split IO */
3834 	bdev->optimal_io_boundary = 16;
3835 	bdev->split_on_optimal_io_boundary = true;
3836 
3837 	/* Now test that a single-vector command is split correctly.
3838 	 * Offset 14, length 8, payload 0xF000
3839 	 *  Child - Offset 14, length 2, payload 0xF000
3840 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
3841 	 *
3842 	 * Set up the expected values before calling spdk_bdev_read_blocks
3843 	 */
3844 	CU_ASSERT(spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3845 	/* We count all submitted IOs including IO that are generated by splitting. */
3846 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 3);
3847 	stub_complete_io(1);
3848 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 2);
3849 	stub_complete_io(1);
3850 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3851 
3852 	/* Also include the reset IO */
3853 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3854 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 1);
3855 	poll_threads();
3856 	stub_complete_io(1);
3857 	poll_threads();
3858 	CU_ASSERT(bdev_channel_count_submitted_io(bdev_ch) == 0);
3859 
3860 	/* This is part2
3861 	 * Test the desc timeout poller register
3862 	 */
3863 
3864 	/* Successfully set the timeout */
3865 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3866 	CU_ASSERT(desc->io_timeout_poller != NULL);
3867 	CU_ASSERT(desc->timeout_in_sec == 30);
3868 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3869 	CU_ASSERT(desc->cb_arg == &cb_arg);
3870 
3871 	/* Change the timeout limit */
3872 	CU_ASSERT(spdk_bdev_set_timeout(desc, 20, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3873 	CU_ASSERT(desc->io_timeout_poller != NULL);
3874 	CU_ASSERT(desc->timeout_in_sec == 20);
3875 	CU_ASSERT(desc->cb_fn == bdev_channel_io_timeout_cb);
3876 	CU_ASSERT(desc->cb_arg == &cb_arg);
3877 
3878 	/* Disable the timeout */
3879 	CU_ASSERT(spdk_bdev_set_timeout(desc, 0, NULL, NULL) == 0);
3880 	CU_ASSERT(desc->io_timeout_poller == NULL);
3881 
3882 	/* This the part3
3883 	 * We will test to catch timeout IO and check whether the IO is
3884 	 * the submitted one.
3885 	 */
3886 	memset(&cb_arg, 0, sizeof(cb_arg));
3887 	CU_ASSERT(spdk_bdev_set_timeout(desc, 30, bdev_channel_io_timeout_cb, &cb_arg) == 0);
3888 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0x1000, 0, 1, io_done, NULL) == 0);
3889 
3890 	/* Don't reach the limit */
3891 	spdk_delay_us(15 * spdk_get_ticks_hz());
3892 	poll_threads();
3893 	CU_ASSERT(cb_arg.type == 0);
3894 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3895 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3896 
3897 	/* 15 + 15 = 30 reach the limit */
3898 	spdk_delay_us(15 * spdk_get_ticks_hz());
3899 	poll_threads();
3900 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3901 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x1000);
3902 	CU_ASSERT(cb_arg.iov.iov_len == 1 * bdev->blocklen);
3903 	stub_complete_io(1);
3904 
3905 	/* Use the same split IO above and check the IO */
3906 	memset(&cb_arg, 0, sizeof(cb_arg));
3907 	CU_ASSERT(spdk_bdev_write_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, NULL) == 0);
3908 
3909 	/* The first child complete in time */
3910 	spdk_delay_us(15 * spdk_get_ticks_hz());
3911 	poll_threads();
3912 	stub_complete_io(1);
3913 	CU_ASSERT(cb_arg.type == 0);
3914 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0x0);
3915 	CU_ASSERT(cb_arg.iov.iov_len == 0);
3916 
3917 	/* The second child reach the limit */
3918 	spdk_delay_us(15 * spdk_get_ticks_hz());
3919 	poll_threads();
3920 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_WRITE);
3921 	CU_ASSERT(cb_arg.iov.iov_base == (void *)0xF000);
3922 	CU_ASSERT(cb_arg.iov.iov_len == 8 * bdev->blocklen);
3923 	stub_complete_io(1);
3924 
3925 	/* Also include the reset IO */
3926 	memset(&cb_arg, 0, sizeof(cb_arg));
3927 	CU_ASSERT(spdk_bdev_reset(desc, io_ch, io_done, NULL) == 0);
3928 	spdk_delay_us(30 * spdk_get_ticks_hz());
3929 	poll_threads();
3930 	CU_ASSERT(cb_arg.type == SPDK_BDEV_IO_TYPE_RESET);
3931 	stub_complete_io(1);
3932 	poll_threads();
3933 
3934 	spdk_put_io_channel(io_ch);
3935 	spdk_bdev_close(desc);
3936 	free_bdev(bdev);
3937 	spdk_bdev_finish(bdev_fini_cb, NULL);
3938 	poll_threads();
3939 }
3940 
3941 static void
3942 lba_range_overlap(void)
3943 {
3944 	struct lba_range r1, r2;
3945 
3946 	r1.offset = 100;
3947 	r1.length = 50;
3948 
3949 	r2.offset = 0;
3950 	r2.length = 1;
3951 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3952 
3953 	r2.offset = 0;
3954 	r2.length = 100;
3955 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3956 
3957 	r2.offset = 0;
3958 	r2.length = 110;
3959 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3960 
3961 	r2.offset = 100;
3962 	r2.length = 10;
3963 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3964 
3965 	r2.offset = 110;
3966 	r2.length = 20;
3967 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3968 
3969 	r2.offset = 140;
3970 	r2.length = 150;
3971 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3972 
3973 	r2.offset = 130;
3974 	r2.length = 200;
3975 	CU_ASSERT(bdev_lba_range_overlapped(&r1, &r2));
3976 
3977 	r2.offset = 150;
3978 	r2.length = 100;
3979 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3980 
3981 	r2.offset = 110;
3982 	r2.length = 0;
3983 	CU_ASSERT(!bdev_lba_range_overlapped(&r1, &r2));
3984 }
3985 
3986 static bool g_lock_lba_range_done;
3987 static bool g_unlock_lba_range_done;
3988 
3989 static void
3990 lock_lba_range_done(void *ctx, int status)
3991 {
3992 	g_lock_lba_range_done = true;
3993 }
3994 
3995 static void
3996 unlock_lba_range_done(void *ctx, int status)
3997 {
3998 	g_unlock_lba_range_done = true;
3999 }
4000 
4001 static void
4002 lock_lba_range_check_ranges(void)
4003 {
4004 	struct spdk_bdev *bdev;
4005 	struct spdk_bdev_desc *desc = NULL;
4006 	struct spdk_io_channel *io_ch;
4007 	struct spdk_bdev_channel *channel;
4008 	struct lba_range *range;
4009 	int ctx1;
4010 	int rc;
4011 
4012 	spdk_bdev_initialize(bdev_init_cb, NULL);
4013 
4014 	bdev = allocate_bdev("bdev0");
4015 
4016 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4017 	CU_ASSERT(rc == 0);
4018 	CU_ASSERT(desc != NULL);
4019 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4020 	io_ch = spdk_bdev_get_io_channel(desc);
4021 	CU_ASSERT(io_ch != NULL);
4022 	channel = spdk_io_channel_get_ctx(io_ch);
4023 
4024 	g_lock_lba_range_done = false;
4025 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4026 	CU_ASSERT(rc == 0);
4027 	poll_threads();
4028 
4029 	CU_ASSERT(g_lock_lba_range_done == true);
4030 	range = TAILQ_FIRST(&channel->locked_ranges);
4031 	SPDK_CU_ASSERT_FATAL(range != NULL);
4032 	CU_ASSERT(range->offset == 20);
4033 	CU_ASSERT(range->length == 10);
4034 	CU_ASSERT(range->owner_ch == channel);
4035 
4036 	/* Unlocks must exactly match a lock. */
4037 	g_unlock_lba_range_done = false;
4038 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 1, unlock_lba_range_done, &ctx1);
4039 	CU_ASSERT(rc == -EINVAL);
4040 	CU_ASSERT(g_unlock_lba_range_done == false);
4041 
4042 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4043 	CU_ASSERT(rc == 0);
4044 	spdk_delay_us(100);
4045 	poll_threads();
4046 
4047 	CU_ASSERT(g_unlock_lba_range_done == true);
4048 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4049 
4050 	spdk_put_io_channel(io_ch);
4051 	spdk_bdev_close(desc);
4052 	free_bdev(bdev);
4053 	spdk_bdev_finish(bdev_fini_cb, NULL);
4054 	poll_threads();
4055 }
4056 
4057 static void
4058 lock_lba_range_with_io_outstanding(void)
4059 {
4060 	struct spdk_bdev *bdev;
4061 	struct spdk_bdev_desc *desc = NULL;
4062 	struct spdk_io_channel *io_ch;
4063 	struct spdk_bdev_channel *channel;
4064 	struct lba_range *range;
4065 	char buf[4096];
4066 	int ctx1;
4067 	int rc;
4068 
4069 	spdk_bdev_initialize(bdev_init_cb, NULL);
4070 
4071 	bdev = allocate_bdev("bdev0");
4072 
4073 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4074 	CU_ASSERT(rc == 0);
4075 	CU_ASSERT(desc != NULL);
4076 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4077 	io_ch = spdk_bdev_get_io_channel(desc);
4078 	CU_ASSERT(io_ch != NULL);
4079 	channel = spdk_io_channel_get_ctx(io_ch);
4080 
4081 	g_io_done = false;
4082 	rc = spdk_bdev_read_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4083 	CU_ASSERT(rc == 0);
4084 
4085 	g_lock_lba_range_done = false;
4086 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4087 	CU_ASSERT(rc == 0);
4088 	poll_threads();
4089 
4090 	/* The lock should immediately become valid, since there are no outstanding
4091 	 * write I/O.
4092 	 */
4093 	CU_ASSERT(g_io_done == false);
4094 	CU_ASSERT(g_lock_lba_range_done == true);
4095 	range = TAILQ_FIRST(&channel->locked_ranges);
4096 	SPDK_CU_ASSERT_FATAL(range != NULL);
4097 	CU_ASSERT(range->offset == 20);
4098 	CU_ASSERT(range->length == 10);
4099 	CU_ASSERT(range->owner_ch == channel);
4100 	CU_ASSERT(range->locked_ctx == &ctx1);
4101 
4102 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4103 	CU_ASSERT(rc == 0);
4104 	stub_complete_io(1);
4105 	spdk_delay_us(100);
4106 	poll_threads();
4107 
4108 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4109 
4110 	/* Now try again, but with a write I/O. */
4111 	g_io_done = false;
4112 	rc = spdk_bdev_write_blocks(desc, io_ch, buf, 20, 1, io_done, &ctx1);
4113 	CU_ASSERT(rc == 0);
4114 
4115 	g_lock_lba_range_done = false;
4116 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4117 	CU_ASSERT(rc == 0);
4118 	poll_threads();
4119 
4120 	/* The lock should not be fully valid yet, since a write I/O is outstanding.
4121 	 * But note that the range should be on the channel's locked_list, to make sure no
4122 	 * new write I/O are started.
4123 	 */
4124 	CU_ASSERT(g_io_done == false);
4125 	CU_ASSERT(g_lock_lba_range_done == false);
4126 	range = TAILQ_FIRST(&channel->locked_ranges);
4127 	SPDK_CU_ASSERT_FATAL(range != NULL);
4128 	CU_ASSERT(range->offset == 20);
4129 	CU_ASSERT(range->length == 10);
4130 
4131 	/* Complete the write I/O.  This should make the lock valid (checked by confirming
4132 	 * our callback was invoked).
4133 	 */
4134 	stub_complete_io(1);
4135 	spdk_delay_us(100);
4136 	poll_threads();
4137 	CU_ASSERT(g_io_done == true);
4138 	CU_ASSERT(g_lock_lba_range_done == true);
4139 
4140 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4141 	CU_ASSERT(rc == 0);
4142 	poll_threads();
4143 
4144 	CU_ASSERT(TAILQ_EMPTY(&channel->locked_ranges));
4145 
4146 	spdk_put_io_channel(io_ch);
4147 	spdk_bdev_close(desc);
4148 	free_bdev(bdev);
4149 	spdk_bdev_finish(bdev_fini_cb, NULL);
4150 	poll_threads();
4151 }
4152 
4153 static void
4154 lock_lba_range_overlapped(void)
4155 {
4156 	struct spdk_bdev *bdev;
4157 	struct spdk_bdev_desc *desc = NULL;
4158 	struct spdk_io_channel *io_ch;
4159 	struct spdk_bdev_channel *channel;
4160 	struct lba_range *range;
4161 	int ctx1;
4162 	int rc;
4163 
4164 	spdk_bdev_initialize(bdev_init_cb, NULL);
4165 
4166 	bdev = allocate_bdev("bdev0");
4167 
4168 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4169 	CU_ASSERT(rc == 0);
4170 	CU_ASSERT(desc != NULL);
4171 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4172 	io_ch = spdk_bdev_get_io_channel(desc);
4173 	CU_ASSERT(io_ch != NULL);
4174 	channel = spdk_io_channel_get_ctx(io_ch);
4175 
4176 	/* Lock range 20-29. */
4177 	g_lock_lba_range_done = false;
4178 	rc = bdev_lock_lba_range(desc, io_ch, 20, 10, lock_lba_range_done, &ctx1);
4179 	CU_ASSERT(rc == 0);
4180 	poll_threads();
4181 
4182 	CU_ASSERT(g_lock_lba_range_done == true);
4183 	range = TAILQ_FIRST(&channel->locked_ranges);
4184 	SPDK_CU_ASSERT_FATAL(range != NULL);
4185 	CU_ASSERT(range->offset == 20);
4186 	CU_ASSERT(range->length == 10);
4187 
4188 	/* Try to lock range 25-39.  It should not lock immediately, since it overlaps with
4189 	 * 20-29.
4190 	 */
4191 	g_lock_lba_range_done = false;
4192 	rc = bdev_lock_lba_range(desc, io_ch, 25, 15, lock_lba_range_done, &ctx1);
4193 	CU_ASSERT(rc == 0);
4194 	poll_threads();
4195 
4196 	CU_ASSERT(g_lock_lba_range_done == false);
4197 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4198 	SPDK_CU_ASSERT_FATAL(range != NULL);
4199 	CU_ASSERT(range->offset == 25);
4200 	CU_ASSERT(range->length == 15);
4201 
4202 	/* Unlock 20-29.  This should result in range 25-39 now getting locked since it
4203 	 * no longer overlaps with an active lock.
4204 	 */
4205 	g_unlock_lba_range_done = false;
4206 	rc = bdev_unlock_lba_range(desc, io_ch, 20, 10, unlock_lba_range_done, &ctx1);
4207 	CU_ASSERT(rc == 0);
4208 	poll_threads();
4209 
4210 	CU_ASSERT(g_unlock_lba_range_done == true);
4211 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4212 	range = TAILQ_FIRST(&channel->locked_ranges);
4213 	SPDK_CU_ASSERT_FATAL(range != NULL);
4214 	CU_ASSERT(range->offset == 25);
4215 	CU_ASSERT(range->length == 15);
4216 
4217 	/* Lock 40-59.  This should immediately lock since it does not overlap with the
4218 	 * currently active 25-39 lock.
4219 	 */
4220 	g_lock_lba_range_done = false;
4221 	rc = bdev_lock_lba_range(desc, io_ch, 40, 20, lock_lba_range_done, &ctx1);
4222 	CU_ASSERT(rc == 0);
4223 	poll_threads();
4224 
4225 	CU_ASSERT(g_lock_lba_range_done == true);
4226 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4227 	SPDK_CU_ASSERT_FATAL(range != NULL);
4228 	range = TAILQ_NEXT(range, tailq);
4229 	SPDK_CU_ASSERT_FATAL(range != NULL);
4230 	CU_ASSERT(range->offset == 40);
4231 	CU_ASSERT(range->length == 20);
4232 
4233 	/* Try to lock 35-44.  Note that this overlaps with both 25-39 and 40-59. */
4234 	g_lock_lba_range_done = false;
4235 	rc = bdev_lock_lba_range(desc, io_ch, 35, 10, lock_lba_range_done, &ctx1);
4236 	CU_ASSERT(rc == 0);
4237 	poll_threads();
4238 
4239 	CU_ASSERT(g_lock_lba_range_done == false);
4240 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4241 	SPDK_CU_ASSERT_FATAL(range != NULL);
4242 	CU_ASSERT(range->offset == 35);
4243 	CU_ASSERT(range->length == 10);
4244 
4245 	/* Unlock 25-39.  Make sure that 35-44 is still in the pending list, since
4246 	 * the 40-59 lock is still active.
4247 	 */
4248 	g_unlock_lba_range_done = false;
4249 	rc = bdev_unlock_lba_range(desc, io_ch, 25, 15, unlock_lba_range_done, &ctx1);
4250 	CU_ASSERT(rc == 0);
4251 	poll_threads();
4252 
4253 	CU_ASSERT(g_unlock_lba_range_done == true);
4254 	CU_ASSERT(g_lock_lba_range_done == false);
4255 	range = TAILQ_FIRST(&bdev->internal.pending_locked_ranges);
4256 	SPDK_CU_ASSERT_FATAL(range != NULL);
4257 	CU_ASSERT(range->offset == 35);
4258 	CU_ASSERT(range->length == 10);
4259 
4260 	/* Unlock 40-59.  This should result in 35-44 now getting locked, since there are
4261 	 * no longer any active overlapping locks.
4262 	 */
4263 	g_unlock_lba_range_done = false;
4264 	rc = bdev_unlock_lba_range(desc, io_ch, 40, 20, unlock_lba_range_done, &ctx1);
4265 	CU_ASSERT(rc == 0);
4266 	poll_threads();
4267 
4268 	CU_ASSERT(g_unlock_lba_range_done == true);
4269 	CU_ASSERT(g_lock_lba_range_done == true);
4270 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.pending_locked_ranges));
4271 	range = TAILQ_FIRST(&bdev->internal.locked_ranges);
4272 	SPDK_CU_ASSERT_FATAL(range != NULL);
4273 	CU_ASSERT(range->offset == 35);
4274 	CU_ASSERT(range->length == 10);
4275 
4276 	/* Finally, unlock 35-44. */
4277 	g_unlock_lba_range_done = false;
4278 	rc = bdev_unlock_lba_range(desc, io_ch, 35, 10, unlock_lba_range_done, &ctx1);
4279 	CU_ASSERT(rc == 0);
4280 	poll_threads();
4281 
4282 	CU_ASSERT(g_unlock_lba_range_done == true);
4283 	CU_ASSERT(TAILQ_EMPTY(&bdev->internal.locked_ranges));
4284 
4285 	spdk_put_io_channel(io_ch);
4286 	spdk_bdev_close(desc);
4287 	free_bdev(bdev);
4288 	spdk_bdev_finish(bdev_fini_cb, NULL);
4289 	poll_threads();
4290 }
4291 
4292 static void
4293 abort_done(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
4294 {
4295 	g_abort_done = true;
4296 	g_abort_status = bdev_io->internal.status;
4297 	spdk_bdev_free_io(bdev_io);
4298 }
4299 
4300 static void
4301 bdev_io_abort(void)
4302 {
4303 	struct spdk_bdev *bdev;
4304 	struct spdk_bdev_desc *desc = NULL;
4305 	struct spdk_io_channel *io_ch;
4306 	struct spdk_bdev_channel *channel;
4307 	struct spdk_bdev_mgmt_channel *mgmt_ch;
4308 	struct spdk_bdev_opts bdev_opts = {};
4309 	struct iovec iov[BDEV_IO_NUM_CHILD_IOV * 2];
4310 	uint64_t io_ctx1 = 0, io_ctx2 = 0, i;
4311 	int rc;
4312 
4313 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4314 	bdev_opts.bdev_io_pool_size = 7;
4315 	bdev_opts.bdev_io_cache_size = 2;
4316 
4317 	rc = spdk_bdev_set_opts(&bdev_opts);
4318 	CU_ASSERT(rc == 0);
4319 	spdk_bdev_initialize(bdev_init_cb, NULL);
4320 
4321 	bdev = allocate_bdev("bdev0");
4322 
4323 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4324 	CU_ASSERT(rc == 0);
4325 	CU_ASSERT(desc != NULL);
4326 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4327 	io_ch = spdk_bdev_get_io_channel(desc);
4328 	CU_ASSERT(io_ch != NULL);
4329 	channel = spdk_io_channel_get_ctx(io_ch);
4330 	mgmt_ch = channel->shared_resource->mgmt_ch;
4331 
4332 	g_abort_done = false;
4333 
4334 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, false);
4335 
4336 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4337 	CU_ASSERT(rc == -ENOTSUP);
4338 
4339 	ut_enable_io_type(SPDK_BDEV_IO_TYPE_ABORT, true);
4340 
4341 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx2, abort_done, NULL);
4342 	CU_ASSERT(rc == 0);
4343 	CU_ASSERT(g_abort_done == true);
4344 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_FAILED);
4345 
4346 	/* Test the case that the target I/O was successfully aborted. */
4347 	g_io_done = false;
4348 
4349 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4350 	CU_ASSERT(rc == 0);
4351 	CU_ASSERT(g_io_done == false);
4352 
4353 	g_abort_done = false;
4354 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4355 
4356 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4357 	CU_ASSERT(rc == 0);
4358 	CU_ASSERT(g_io_done == true);
4359 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4360 	stub_complete_io(1);
4361 	CU_ASSERT(g_abort_done == true);
4362 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4363 
4364 	/* Test the case that the target I/O was not aborted because it completed
4365 	 * in the middle of execution of the abort.
4366 	 */
4367 	g_io_done = false;
4368 
4369 	rc = spdk_bdev_read_blocks(desc, io_ch, NULL, 0, 1, io_done, &io_ctx1);
4370 	CU_ASSERT(rc == 0);
4371 	CU_ASSERT(g_io_done == false);
4372 
4373 	g_abort_done = false;
4374 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4375 
4376 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4377 	CU_ASSERT(rc == 0);
4378 	CU_ASSERT(g_io_done == false);
4379 
4380 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4381 	stub_complete_io(1);
4382 	CU_ASSERT(g_io_done == true);
4383 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4384 
4385 	g_io_exp_status = SPDK_BDEV_IO_STATUS_FAILED;
4386 	stub_complete_io(1);
4387 	CU_ASSERT(g_abort_done == true);
4388 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4389 
4390 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4391 
4392 	bdev->optimal_io_boundary = 16;
4393 	bdev->split_on_optimal_io_boundary = true;
4394 
4395 	/* Test that a single-vector command which is split is aborted correctly.
4396 	 * Offset 14, length 8, payload 0xF000
4397 	 *  Child - Offset 14, length 2, payload 0xF000
4398 	 *  Child - Offset 16, length 6, payload 0xF000 + 2 * 512
4399 	 */
4400 	g_io_done = false;
4401 
4402 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 8, io_done, &io_ctx1);
4403 	CU_ASSERT(rc == 0);
4404 	CU_ASSERT(g_io_done == false);
4405 
4406 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4407 
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(2);
4415 	CU_ASSERT(g_abort_done == true);
4416 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4417 
4418 	/* Test that a multi-vector command that needs to be split by strip and then
4419 	 * needs to be split is aborted correctly. Abort is requested before the second
4420 	 * child I/O was submitted. The parent I/O should complete with failure without
4421 	 * submitting the second child I/O.
4422 	 */
4423 	for (i = 0; i < BDEV_IO_NUM_CHILD_IOV * 2; i++) {
4424 		iov[i].iov_base = (void *)((i + 1) * 0x10000);
4425 		iov[i].iov_len = 512;
4426 	}
4427 
4428 	bdev->optimal_io_boundary = BDEV_IO_NUM_CHILD_IOV;
4429 	g_io_done = false;
4430 	rc = spdk_bdev_readv_blocks(desc, io_ch, iov, BDEV_IO_NUM_CHILD_IOV * 2, 0,
4431 				    BDEV_IO_NUM_CHILD_IOV * 2, io_done, &io_ctx1);
4432 	CU_ASSERT(rc == 0);
4433 	CU_ASSERT(g_io_done == false);
4434 
4435 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4436 
4437 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4438 
4439 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4440 	CU_ASSERT(rc == 0);
4441 	CU_ASSERT(g_io_done == true);
4442 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4443 	stub_complete_io(1);
4444 	CU_ASSERT(g_abort_done == true);
4445 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4446 
4447 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4448 
4449 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4450 
4451 	bdev->optimal_io_boundary = 16;
4452 	g_io_done = false;
4453 
4454 	/* Test that a ingle-vector command which is split is aborted correctly.
4455 	 * Differently from the above, the child abort request will be submitted
4456 	 * sequentially due to the capacity of spdk_bdev_io.
4457 	 */
4458 	rc = spdk_bdev_read_blocks(desc, io_ch, (void *)0xF000, 14, 50, io_done, &io_ctx1);
4459 	CU_ASSERT(rc == 0);
4460 	CU_ASSERT(g_io_done == false);
4461 
4462 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4463 
4464 	g_abort_done = false;
4465 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4466 
4467 	rc = spdk_bdev_abort(desc, io_ch, &io_ctx1, abort_done, NULL);
4468 	CU_ASSERT(rc == 0);
4469 	CU_ASSERT(!TAILQ_EMPTY(&mgmt_ch->io_wait_queue));
4470 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 4);
4471 
4472 	stub_complete_io(1);
4473 	CU_ASSERT(g_io_done == true);
4474 	CU_ASSERT(g_io_status == SPDK_BDEV_IO_STATUS_FAILED);
4475 	stub_complete_io(3);
4476 	CU_ASSERT(g_abort_done == true);
4477 	CU_ASSERT(g_abort_status == SPDK_BDEV_IO_STATUS_SUCCESS);
4478 
4479 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4480 
4481 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4482 
4483 	spdk_put_io_channel(io_ch);
4484 	spdk_bdev_close(desc);
4485 	free_bdev(bdev);
4486 	spdk_bdev_finish(bdev_fini_cb, NULL);
4487 	poll_threads();
4488 }
4489 
4490 static void
4491 bdev_unmap(void)
4492 {
4493 	struct spdk_bdev *bdev;
4494 	struct spdk_bdev_desc *desc = NULL;
4495 	struct spdk_io_channel *ioch;
4496 	struct spdk_bdev_channel *bdev_ch;
4497 	struct ut_expected_io *expected_io;
4498 	struct spdk_bdev_opts bdev_opts = {};
4499 	uint32_t i, num_outstanding;
4500 	uint64_t offset, num_blocks, max_unmap_blocks, num_children;
4501 	int rc;
4502 
4503 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4504 	bdev_opts.bdev_io_pool_size = 512;
4505 	bdev_opts.bdev_io_cache_size = 64;
4506 	rc = spdk_bdev_set_opts(&bdev_opts);
4507 	CU_ASSERT(rc == 0);
4508 
4509 	spdk_bdev_initialize(bdev_init_cb, NULL);
4510 	bdev = allocate_bdev("bdev");
4511 
4512 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4513 	CU_ASSERT_EQUAL(rc, 0);
4514 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4515 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4516 	ioch = spdk_bdev_get_io_channel(desc);
4517 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4518 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4519 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4520 
4521 	fn_table.submit_request = stub_submit_request;
4522 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4523 
4524 	/* Case 1: First test the request won't be split */
4525 	num_blocks = 32;
4526 
4527 	g_io_done = false;
4528 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, 0, num_blocks, 0);
4529 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4530 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4531 	CU_ASSERT_EQUAL(rc, 0);
4532 	CU_ASSERT(g_io_done == false);
4533 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4534 	stub_complete_io(1);
4535 	CU_ASSERT(g_io_done == true);
4536 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4537 
4538 	/* Case 2: Test the split with 2 children requests */
4539 	bdev->max_unmap = 8;
4540 	bdev->max_unmap_segments = 2;
4541 	max_unmap_blocks = bdev->max_unmap * bdev->max_unmap_segments;
4542 	num_blocks = max_unmap_blocks * 2;
4543 	offset = 0;
4544 
4545 	g_io_done = false;
4546 	for (i = 0; i < 2; i++) {
4547 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4548 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4549 		offset += max_unmap_blocks;
4550 	}
4551 
4552 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4553 	CU_ASSERT_EQUAL(rc, 0);
4554 	CU_ASSERT(g_io_done == false);
4555 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4556 	stub_complete_io(2);
4557 	CU_ASSERT(g_io_done == true);
4558 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4559 
4560 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4561 	num_children = 15;
4562 	num_blocks = max_unmap_blocks * num_children;
4563 	g_io_done = false;
4564 	offset = 0;
4565 	for (i = 0; i < num_children; i++) {
4566 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_UNMAP, offset, max_unmap_blocks, 0);
4567 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4568 		offset += max_unmap_blocks;
4569 	}
4570 
4571 	rc = spdk_bdev_unmap_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4572 	CU_ASSERT_EQUAL(rc, 0);
4573 	CU_ASSERT(g_io_done == false);
4574 
4575 	while (num_children > 0) {
4576 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4577 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4578 		stub_complete_io(num_outstanding);
4579 		num_children -= num_outstanding;
4580 	}
4581 	CU_ASSERT(g_io_done == true);
4582 
4583 	spdk_put_io_channel(ioch);
4584 	spdk_bdev_close(desc);
4585 	free_bdev(bdev);
4586 	spdk_bdev_finish(bdev_fini_cb, NULL);
4587 	poll_threads();
4588 }
4589 
4590 static void
4591 bdev_write_zeroes_split_test(void)
4592 {
4593 	struct spdk_bdev *bdev;
4594 	struct spdk_bdev_desc *desc = NULL;
4595 	struct spdk_io_channel *ioch;
4596 	struct spdk_bdev_channel *bdev_ch;
4597 	struct ut_expected_io *expected_io;
4598 	struct spdk_bdev_opts bdev_opts = {};
4599 	uint32_t i, num_outstanding;
4600 	uint64_t offset, num_blocks, max_write_zeroes_blocks, num_children;
4601 	int rc;
4602 
4603 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4604 	bdev_opts.bdev_io_pool_size = 512;
4605 	bdev_opts.bdev_io_cache_size = 64;
4606 	rc = spdk_bdev_set_opts(&bdev_opts);
4607 	CU_ASSERT(rc == 0);
4608 
4609 	spdk_bdev_initialize(bdev_init_cb, NULL);
4610 	bdev = allocate_bdev("bdev");
4611 
4612 	rc = spdk_bdev_open_ext("bdev", true, bdev_ut_event_cb, NULL, &desc);
4613 	CU_ASSERT_EQUAL(rc, 0);
4614 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4615 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4616 	ioch = spdk_bdev_get_io_channel(desc);
4617 	SPDK_CU_ASSERT_FATAL(ioch != NULL);
4618 	bdev_ch = spdk_io_channel_get_ctx(ioch);
4619 	CU_ASSERT(TAILQ_EMPTY(&bdev_ch->io_submitted));
4620 
4621 	fn_table.submit_request = stub_submit_request;
4622 	g_io_exp_status = SPDK_BDEV_IO_STATUS_SUCCESS;
4623 
4624 	/* Case 1: First test the request won't be split */
4625 	num_blocks = 32;
4626 
4627 	g_io_done = false;
4628 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, 0, num_blocks, 0);
4629 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4630 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4631 	CU_ASSERT_EQUAL(rc, 0);
4632 	CU_ASSERT(g_io_done == false);
4633 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4634 	stub_complete_io(1);
4635 	CU_ASSERT(g_io_done == true);
4636 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4637 
4638 	/* Case 2: Test the split with 2 children requests */
4639 	max_write_zeroes_blocks = 8;
4640 	bdev->max_write_zeroes = max_write_zeroes_blocks;
4641 	num_blocks = max_write_zeroes_blocks * 2;
4642 	offset = 0;
4643 
4644 	g_io_done = false;
4645 	for (i = 0; i < 2; i++) {
4646 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4647 						   0);
4648 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4649 		offset += max_write_zeroes_blocks;
4650 	}
4651 
4652 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4653 	CU_ASSERT_EQUAL(rc, 0);
4654 	CU_ASSERT(g_io_done == false);
4655 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 2);
4656 	stub_complete_io(2);
4657 	CU_ASSERT(g_io_done == true);
4658 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 0);
4659 
4660 	/* Case 3: Test the split with 15 children requests, will finish 8 requests first */
4661 	num_children = 15;
4662 	num_blocks = max_write_zeroes_blocks * num_children;
4663 	g_io_done = false;
4664 	offset = 0;
4665 	for (i = 0; i < num_children; i++) {
4666 		expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE_ZEROES, offset, max_write_zeroes_blocks,
4667 						   0);
4668 		TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4669 		offset += max_write_zeroes_blocks;
4670 	}
4671 
4672 	rc = spdk_bdev_write_zeroes_blocks(desc, ioch, 0, num_blocks, io_done, NULL);
4673 	CU_ASSERT_EQUAL(rc, 0);
4674 	CU_ASSERT(g_io_done == false);
4675 
4676 	while (num_children > 0) {
4677 		num_outstanding = spdk_min(num_children, SPDK_BDEV_MAX_CHILDREN_UNMAP_WRITE_ZEROES_REQS);
4678 		CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == num_outstanding);
4679 		stub_complete_io(num_outstanding);
4680 		num_children -= num_outstanding;
4681 	}
4682 	CU_ASSERT(g_io_done == true);
4683 
4684 	spdk_put_io_channel(ioch);
4685 	spdk_bdev_close(desc);
4686 	free_bdev(bdev);
4687 	spdk_bdev_finish(bdev_fini_cb, NULL);
4688 	poll_threads();
4689 }
4690 
4691 static void
4692 bdev_set_options_test(void)
4693 {
4694 	struct spdk_bdev_opts bdev_opts = {};
4695 	int rc;
4696 
4697 	/* Case1: Do not set opts_size */
4698 	rc = spdk_bdev_set_opts(&bdev_opts);
4699 	CU_ASSERT(rc == -1);
4700 
4701 	spdk_bdev_get_opts(&bdev_opts, sizeof(bdev_opts));
4702 	bdev_opts.bdev_io_pool_size = 4;
4703 	bdev_opts.bdev_io_cache_size = 2;
4704 	bdev_opts.small_buf_pool_size = 4;
4705 
4706 	/* Case 2: Do not set valid small_buf_pool_size and large_buf_pool_size */
4707 	rc = spdk_bdev_set_opts(&bdev_opts);
4708 	CU_ASSERT(rc == -1);
4709 
4710 	/* Case 3: Do not set valid large_buf_pool_size */
4711 	bdev_opts.small_buf_pool_size = BUF_SMALL_POOL_SIZE;
4712 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE - 1;
4713 	rc = spdk_bdev_set_opts(&bdev_opts);
4714 	CU_ASSERT(rc == -1);
4715 
4716 	/* Case4: set valid large buf_pool_size */
4717 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE;
4718 	rc = spdk_bdev_set_opts(&bdev_opts);
4719 	CU_ASSERT(rc == 0);
4720 
4721 	/* Case5: Set different valid value for small and large buf pool */
4722 	bdev_opts.large_buf_pool_size = BUF_SMALL_POOL_SIZE + 3;
4723 	bdev_opts.large_buf_pool_size = BUF_LARGE_POOL_SIZE + 3;
4724 	rc = spdk_bdev_set_opts(&bdev_opts);
4725 	CU_ASSERT(rc == 0);
4726 }
4727 
4728 static uint64_t
4729 get_ns_time(void)
4730 {
4731 	int rc;
4732 	struct timespec ts;
4733 
4734 	rc = clock_gettime(CLOCK_MONOTONIC, &ts);
4735 	CU_ASSERT(rc == 0);
4736 	return ts.tv_sec * 1000 * 1000 * 1000 + ts.tv_nsec;
4737 }
4738 
4739 static int
4740 rb_tree_get_height(struct spdk_bdev_name *bdev_name)
4741 {
4742 	int h1, h2;
4743 
4744 	if (bdev_name == NULL) {
4745 		return -1;
4746 	} else {
4747 		h1 = rb_tree_get_height(RB_LEFT(bdev_name, node));
4748 		h2 = rb_tree_get_height(RB_RIGHT(bdev_name, node));
4749 
4750 		return spdk_max(h1, h2) + 1;
4751 	}
4752 }
4753 
4754 static void
4755 bdev_multi_allocation(void)
4756 {
4757 	const int max_bdev_num = 1024 * 16;
4758 	char name[max_bdev_num][10];
4759 	char noexist_name[] = "invalid_bdev";
4760 	struct spdk_bdev *bdev[max_bdev_num];
4761 	int i, j;
4762 	uint64_t last_time;
4763 	int bdev_num;
4764 	int height;
4765 
4766 	for (j = 0; j < max_bdev_num; j++) {
4767 		snprintf(name[j], sizeof(name[j]), "bdev%d", j);
4768 	}
4769 
4770 	for (i = 0; i < 16; i++) {
4771 		last_time = get_ns_time();
4772 		bdev_num = 1024 * (i + 1);
4773 		for (j = 0; j < bdev_num; j++) {
4774 			bdev[j] = allocate_bdev(name[j]);
4775 			height = rb_tree_get_height(&bdev[j]->internal.bdev_name);
4776 			CU_ASSERT(height <= (int)(spdk_u32log2(j + 1)));
4777 		}
4778 		SPDK_NOTICELOG("alloc bdev num %d takes %" PRIu64 " ms\n", bdev_num,
4779 			       (get_ns_time() - last_time) / 1000 / 1000);
4780 		for (j = 0; j < bdev_num; j++) {
4781 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) != NULL);
4782 		}
4783 		CU_ASSERT(spdk_bdev_get_by_name(noexist_name) == NULL);
4784 
4785 		for (j = 0; j < bdev_num; j++) {
4786 			free_bdev(bdev[j]);
4787 		}
4788 		for (j = 0; j < bdev_num; j++) {
4789 			CU_ASSERT(spdk_bdev_get_by_name(name[j]) == NULL);
4790 		}
4791 	}
4792 }
4793 
4794 static struct spdk_memory_domain *g_bdev_memory_domain = (struct spdk_memory_domain *) 0xf00df00d;
4795 
4796 static int
4797 test_bdev_get_supported_dma_device_types_op(void *ctx, struct spdk_memory_domain **domains,
4798 		int array_size)
4799 {
4800 	if (array_size > 0 && domains) {
4801 		domains[0] = g_bdev_memory_domain;
4802 	}
4803 
4804 	return 1;
4805 }
4806 
4807 static void
4808 bdev_get_memory_domains(void)
4809 {
4810 	struct spdk_bdev_fn_table fn_table = {
4811 		.get_memory_domains = test_bdev_get_supported_dma_device_types_op
4812 	};
4813 	struct spdk_bdev bdev = { .fn_table = &fn_table };
4814 	struct spdk_memory_domain *domains[2] = {};
4815 	int rc;
4816 
4817 	/* bdev is NULL */
4818 	rc = spdk_bdev_get_memory_domains(NULL, domains, 2);
4819 	CU_ASSERT(rc == -EINVAL);
4820 
4821 	/* domains is NULL */
4822 	rc = spdk_bdev_get_memory_domains(&bdev, NULL, 2);
4823 	CU_ASSERT(rc == 1);
4824 
4825 	/* array size is 0 */
4826 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 0);
4827 	CU_ASSERT(rc == 1);
4828 
4829 	/* get_supported_dma_device_types op is set */
4830 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4831 	CU_ASSERT(rc == 1);
4832 	CU_ASSERT(domains[0] == g_bdev_memory_domain);
4833 
4834 	/* get_supported_dma_device_types op is not set */
4835 	fn_table.get_memory_domains = NULL;
4836 	rc = spdk_bdev_get_memory_domains(&bdev, domains, 2);
4837 	CU_ASSERT(rc == 0);
4838 }
4839 
4840 static void
4841 bdev_writev_readv_ext(void)
4842 {
4843 	struct spdk_bdev *bdev;
4844 	struct spdk_bdev_desc *desc = NULL;
4845 	struct spdk_io_channel *io_ch;
4846 	struct iovec iov = { .iov_base = (void *)0xbaaddead, .iov_len = 0x1000 };
4847 	struct ut_expected_io *expected_io;
4848 	struct spdk_bdev_ext_io_opts ext_io_opts = {
4849 		.metadata = (void *)0xFF000000
4850 	};
4851 	int rc;
4852 
4853 	spdk_bdev_initialize(bdev_init_cb, NULL);
4854 
4855 	bdev = allocate_bdev("bdev0");
4856 	bdev->md_interleave = false;
4857 	bdev->md_len = 8;
4858 
4859 	rc = spdk_bdev_open_ext("bdev0", true, bdev_ut_event_cb, NULL, &desc);
4860 	CU_ASSERT(rc == 0);
4861 	SPDK_CU_ASSERT_FATAL(desc != NULL);
4862 	CU_ASSERT(bdev == spdk_bdev_desc_get_bdev(desc));
4863 	io_ch = spdk_bdev_get_io_channel(desc);
4864 	CU_ASSERT(io_ch != NULL);
4865 
4866 	g_io_done = false;
4867 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_READ, 32, 14, 1);
4868 	expected_io->md_buf = ext_io_opts.metadata;
4869 	expected_io->ext_io_opts = &ext_io_opts;
4870 	ut_expected_io_set_iov(expected_io, 0, iov.iov_base, iov.iov_len);
4871 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4872 
4873 	rc = spdk_bdev_readv_blocks_ext(desc, io_ch, &iov, 1, 32, 14, io_done, NULL, &ext_io_opts);
4874 
4875 	CU_ASSERT(rc == 0);
4876 	CU_ASSERT(g_io_done == false);
4877 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4878 	stub_complete_io(1);
4879 	CU_ASSERT(g_io_done == true);
4880 
4881 	g_io_done = false;
4882 	expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 1);
4883 	expected_io->md_buf = ext_io_opts.metadata;
4884 	expected_io->ext_io_opts = &ext_io_opts;
4885 	ut_expected_io_set_iov(expected_io, 0, iov.iov_base, iov.iov_len);
4886 	TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
4887 
4888 	rc = spdk_bdev_writev_blocks_ext(desc, io_ch, &iov, 1, 32, 14, io_done, NULL, &ext_io_opts);
4889 
4890 	CU_ASSERT(rc == 0);
4891 	CU_ASSERT(g_io_done == false);
4892 	CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
4893 	stub_complete_io(1);
4894 	CU_ASSERT(g_io_done == true);
4895 
4896 	spdk_put_io_channel(io_ch);
4897 	spdk_bdev_close(desc);
4898 	free_bdev(bdev);
4899 	spdk_bdev_finish(bdev_fini_cb, NULL);
4900 	poll_threads();
4901 }
4902 
4903 int
4904 main(int argc, char **argv)
4905 {
4906 	CU_pSuite		suite = NULL;
4907 	unsigned int		num_failures;
4908 
4909 	CU_set_error_action(CUEA_ABORT);
4910 	CU_initialize_registry();
4911 
4912 	suite = CU_add_suite("bdev", null_init, null_clean);
4913 
4914 	CU_ADD_TEST(suite, bytes_to_blocks_test);
4915 	CU_ADD_TEST(suite, num_blocks_test);
4916 	CU_ADD_TEST(suite, io_valid_test);
4917 	CU_ADD_TEST(suite, open_write_test);
4918 	CU_ADD_TEST(suite, alias_add_del_test);
4919 	CU_ADD_TEST(suite, get_device_stat_test);
4920 	CU_ADD_TEST(suite, bdev_io_types_test);
4921 	CU_ADD_TEST(suite, bdev_io_wait_test);
4922 	CU_ADD_TEST(suite, bdev_io_spans_split_test);
4923 	CU_ADD_TEST(suite, bdev_io_boundary_split_test);
4924 	CU_ADD_TEST(suite, bdev_io_max_size_and_segment_split_test);
4925 	CU_ADD_TEST(suite, bdev_io_mix_split_test);
4926 	CU_ADD_TEST(suite, bdev_io_split_with_io_wait);
4927 	CU_ADD_TEST(suite, bdev_io_alignment_with_boundary);
4928 	CU_ADD_TEST(suite, bdev_io_alignment);
4929 	CU_ADD_TEST(suite, bdev_histograms);
4930 	CU_ADD_TEST(suite, bdev_write_zeroes);
4931 	CU_ADD_TEST(suite, bdev_compare_and_write);
4932 	CU_ADD_TEST(suite, bdev_compare);
4933 	CU_ADD_TEST(suite, bdev_zcopy_write);
4934 	CU_ADD_TEST(suite, bdev_zcopy_read);
4935 	CU_ADD_TEST(suite, bdev_open_while_hotremove);
4936 	CU_ADD_TEST(suite, bdev_close_while_hotremove);
4937 	CU_ADD_TEST(suite, bdev_open_ext);
4938 	CU_ADD_TEST(suite, bdev_set_io_timeout);
4939 	CU_ADD_TEST(suite, lba_range_overlap);
4940 	CU_ADD_TEST(suite, lock_lba_range_check_ranges);
4941 	CU_ADD_TEST(suite, lock_lba_range_with_io_outstanding);
4942 	CU_ADD_TEST(suite, lock_lba_range_overlapped);
4943 	CU_ADD_TEST(suite, bdev_io_abort);
4944 	CU_ADD_TEST(suite, bdev_unmap);
4945 	CU_ADD_TEST(suite, bdev_write_zeroes_split_test);
4946 	CU_ADD_TEST(suite, bdev_set_options_test);
4947 	CU_ADD_TEST(suite, bdev_multi_allocation);
4948 	CU_ADD_TEST(suite, bdev_get_memory_domains);
4949 	CU_ADD_TEST(suite, bdev_writev_readv_ext);
4950 
4951 	allocate_cores(1);
4952 	allocate_threads(1);
4953 	set_thread(0);
4954 
4955 	CU_basic_set_mode(CU_BRM_VERBOSE);
4956 	CU_basic_run_tests();
4957 	num_failures = CU_get_number_of_failures();
4958 	CU_cleanup_registry();
4959 
4960 	free_threads();
4961 	free_cores();
4962 
4963 	return num_failures;
4964 }
4965