xref: /spdk/lib/bdev/bdev.c (revision bea2e2308fc7a98826f3cfe36c584cfb14dcd180)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (C) 2008-2012 Daisuke Aoyama <aoyama@peach.ne.jp>.
5  *   Copyright (c) Intel Corporation.
6  *   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/stdinc.h"
36 
37 #include "spdk/bdev.h"
38 
39 #include <rte_config.h>
40 #include <rte_lcore.h>
41 #include "spdk/env.h"
42 #include "spdk/io_channel.h"
43 #include "spdk/likely.h"
44 #include "spdk/queue.h"
45 #include "spdk/nvme_spec.h"
46 #include "spdk/scsi_spec.h"
47 
48 #include "spdk_internal/bdev.h"
49 #include "spdk_internal/event.h"
50 #include "spdk_internal/log.h"
51 
52 #define SPDK_BDEV_IO_POOL_SIZE	(64 * 1024)
53 #define BUF_SMALL_POOL_SIZE	8192
54 #define BUF_LARGE_POOL_SIZE	1024
55 
56 typedef TAILQ_HEAD(, spdk_bdev_io) need_buf_tailq_t;
57 
58 struct spdk_bdev_mgr {
59 	struct spdk_mempool *bdev_io_pool;
60 
61 	struct spdk_mempool *buf_small_pool;
62 	struct spdk_mempool *buf_large_pool;
63 
64 	need_buf_tailq_t need_buf_small[RTE_MAX_LCORE];
65 	need_buf_tailq_t need_buf_large[RTE_MAX_LCORE];
66 
67 	TAILQ_HEAD(, spdk_bdev_module_if) bdev_modules;
68 	TAILQ_HEAD(, spdk_bdev_module_if) vbdev_modules;
69 
70 	TAILQ_HEAD(, spdk_bdev) bdevs;
71 };
72 
73 static struct spdk_bdev_mgr g_bdev_mgr = {
74 	.bdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdev_modules),
75 	.vbdev_modules = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.vbdev_modules),
76 	.bdevs = TAILQ_HEAD_INITIALIZER(g_bdev_mgr.bdevs),
77 };
78 
79 struct spdk_bdev_channel {
80 	struct spdk_bdev	*bdev;
81 
82 	/* The channel for the underlying device */
83 	struct spdk_io_channel	*channel;
84 };
85 
86 struct spdk_bdev *
87 spdk_bdev_first(void)
88 {
89 	struct spdk_bdev *bdev;
90 
91 	bdev = TAILQ_FIRST(&g_bdev_mgr.bdevs);
92 	if (bdev) {
93 		SPDK_TRACELOG(SPDK_TRACE_DEBUG, "Starting bdev iteration at %s\n", bdev->name);
94 	}
95 
96 	return bdev;
97 }
98 
99 struct spdk_bdev *
100 spdk_bdev_next(struct spdk_bdev *prev)
101 {
102 	struct spdk_bdev *bdev;
103 
104 	bdev = TAILQ_NEXT(prev, link);
105 	if (bdev) {
106 		SPDK_TRACELOG(SPDK_TRACE_DEBUG, "Continuing bdev iteration at %s\n", bdev->name);
107 	}
108 
109 	return bdev;
110 }
111 
112 struct spdk_bdev *
113 spdk_bdev_get_by_name(const char *bdev_name)
114 {
115 	struct spdk_bdev *bdev = spdk_bdev_first();
116 
117 	while (bdev != NULL) {
118 		if (strncmp(bdev_name, bdev->name, sizeof(bdev->name)) == 0) {
119 			return bdev;
120 		}
121 		bdev = spdk_bdev_next(bdev);
122 	}
123 
124 	return NULL;
125 }
126 
127 static void
128 spdk_bdev_io_set_buf(struct spdk_bdev_io *bdev_io, void *buf)
129 {
130 	assert(bdev_io->get_buf_cb != NULL);
131 	assert(buf != NULL);
132 	assert(bdev_io->u.read.iovs != NULL);
133 
134 	bdev_io->buf = buf;
135 	bdev_io->u.read.iovs[0].iov_base = (void *)((unsigned long)((char *)buf + 512) & ~511UL);
136 	bdev_io->u.read.iovs[0].iov_len = bdev_io->u.read.len;
137 	bdev_io->get_buf_cb(bdev_io->ch->channel, bdev_io);
138 }
139 
140 static void
141 spdk_bdev_io_put_buf(struct spdk_bdev_io *bdev_io)
142 {
143 	struct spdk_mempool *pool;
144 	struct spdk_bdev_io *tmp;
145 	void *buf;
146 	need_buf_tailq_t *tailq;
147 	uint64_t length;
148 
149 	assert(bdev_io->u.read.iovcnt == 1);
150 
151 	length = bdev_io->u.read.len;
152 	buf = bdev_io->buf;
153 
154 	if (length <= SPDK_BDEV_SMALL_BUF_MAX_SIZE) {
155 		pool = g_bdev_mgr.buf_small_pool;
156 		tailq = &g_bdev_mgr.need_buf_small[rte_lcore_id()];
157 	} else {
158 		pool = g_bdev_mgr.buf_large_pool;
159 		tailq = &g_bdev_mgr.need_buf_large[rte_lcore_id()];
160 	}
161 
162 	if (TAILQ_EMPTY(tailq)) {
163 		spdk_mempool_put(pool, buf);
164 	} else {
165 		tmp = TAILQ_FIRST(tailq);
166 		TAILQ_REMOVE(tailq, tmp, buf_link);
167 		spdk_bdev_io_set_buf(tmp, buf);
168 	}
169 }
170 
171 void
172 spdk_bdev_io_get_buf(struct spdk_bdev_io *bdev_io, spdk_bdev_io_get_buf_cb cb)
173 {
174 	uint64_t len = bdev_io->u.read.len;
175 	struct spdk_mempool *pool;
176 	need_buf_tailq_t *tailq;
177 	void *buf = NULL;
178 
179 	assert(cb != NULL);
180 	assert(bdev_io->u.read.iovs != NULL);
181 
182 	if (spdk_unlikely(bdev_io->u.read.iovs[0].iov_base != NULL)) {
183 		/* Buffer already present */
184 		cb(bdev_io->ch->channel, bdev_io);
185 		return;
186 	}
187 
188 	bdev_io->get_buf_cb = cb;
189 	if (len <= SPDK_BDEV_SMALL_BUF_MAX_SIZE) {
190 		pool = g_bdev_mgr.buf_small_pool;
191 		tailq = &g_bdev_mgr.need_buf_small[rte_lcore_id()];
192 	} else {
193 		pool = g_bdev_mgr.buf_large_pool;
194 		tailq = &g_bdev_mgr.need_buf_large[rte_lcore_id()];
195 	}
196 
197 	buf = spdk_mempool_get(pool);
198 
199 	if (!buf) {
200 		TAILQ_INSERT_TAIL(tailq, bdev_io, buf_link);
201 	} else {
202 		spdk_bdev_io_set_buf(bdev_io, buf);
203 	}
204 }
205 
206 static int
207 spdk_bdev_module_get_max_ctx_size(void)
208 {
209 	struct spdk_bdev_module_if *bdev_module;
210 	int max_bdev_module_size = 0;
211 
212 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, tailq) {
213 		if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
214 			max_bdev_module_size = bdev_module->get_ctx_size();
215 		}
216 	}
217 
218 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.vbdev_modules, tailq) {
219 		if (bdev_module->get_ctx_size && bdev_module->get_ctx_size() > max_bdev_module_size) {
220 			max_bdev_module_size = bdev_module->get_ctx_size();
221 		}
222 	}
223 
224 	return max_bdev_module_size;
225 }
226 
227 static void
228 spdk_bdev_config_text(FILE *fp)
229 {
230 	struct spdk_bdev_module_if *bdev_module;
231 
232 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, tailq) {
233 		if (bdev_module->config_text) {
234 			bdev_module->config_text(fp);
235 		}
236 	}
237 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.vbdev_modules, tailq) {
238 		if (bdev_module->config_text) {
239 			bdev_module->config_text(fp);
240 		}
241 	}
242 }
243 
244 static int
245 spdk_bdev_initialize(void)
246 {
247 	int i, cache_size;
248 	struct spdk_bdev_module_if *bdev_module;
249 	int rc = 0;
250 
251 	g_bdev_mgr.bdev_io_pool = spdk_mempool_create("blockdev_io",
252 				  SPDK_BDEV_IO_POOL_SIZE,
253 				  sizeof(struct spdk_bdev_io) +
254 				  spdk_bdev_module_get_max_ctx_size(),
255 				  64,
256 				  SPDK_ENV_SOCKET_ID_ANY);
257 
258 	if (g_bdev_mgr.bdev_io_pool == NULL) {
259 		SPDK_ERRLOG("could not allocate spdk_bdev_io pool");
260 		return -1;
261 	}
262 
263 	for (i = 0; i < RTE_MAX_LCORE; i++) {
264 		TAILQ_INIT(&g_bdev_mgr.need_buf_small[i]);
265 		TAILQ_INIT(&g_bdev_mgr.need_buf_large[i]);
266 	}
267 
268 	/**
269 	 * Ensure no more than half of the total buffers end up local caches, by
270 	 *   using spdk_env_get_core_count() to determine how many local caches we need
271 	 *   to account for.
272 	 */
273 	cache_size = BUF_SMALL_POOL_SIZE / (2 * spdk_env_get_core_count());
274 	g_bdev_mgr.buf_small_pool = spdk_mempool_create("buf_small_pool",
275 				    BUF_SMALL_POOL_SIZE,
276 				    SPDK_BDEV_SMALL_BUF_MAX_SIZE + 512,
277 				    cache_size,
278 				    SPDK_ENV_SOCKET_ID_ANY);
279 	if (!g_bdev_mgr.buf_small_pool) {
280 		SPDK_ERRLOG("create rbuf small pool failed\n");
281 		return -1;
282 	}
283 
284 	cache_size = BUF_LARGE_POOL_SIZE / (2 * spdk_env_get_core_count());
285 	g_bdev_mgr.buf_large_pool = spdk_mempool_create("buf_large_pool",
286 				    BUF_LARGE_POOL_SIZE,
287 				    SPDK_BDEV_LARGE_BUF_MAX_SIZE + 512,
288 				    cache_size,
289 				    SPDK_ENV_SOCKET_ID_ANY);
290 	if (!g_bdev_mgr.buf_large_pool) {
291 		SPDK_ERRLOG("create rbuf large pool failed\n");
292 		return -1;
293 	}
294 
295 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, tailq) {
296 		rc = bdev_module->module_init();
297 		if (rc) {
298 			return rc;
299 		}
300 	}
301 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.vbdev_modules, tailq) {
302 		rc = bdev_module->module_init();
303 		if (rc) {
304 			return rc;
305 		}
306 	}
307 
308 	return 0;
309 }
310 
311 static int
312 spdk_bdev_finish(void)
313 {
314 	struct spdk_bdev_module_if *bdev_module;
315 
316 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.vbdev_modules, tailq) {
317 		if (bdev_module->module_fini) {
318 			bdev_module->module_fini();
319 		}
320 	}
321 
322 	TAILQ_FOREACH(bdev_module, &g_bdev_mgr.bdev_modules, tailq) {
323 		if (bdev_module->module_fini) {
324 			bdev_module->module_fini();
325 		}
326 	}
327 
328 	if (spdk_mempool_count(g_bdev_mgr.bdev_io_pool) != SPDK_BDEV_IO_POOL_SIZE) {
329 		SPDK_ERRLOG("bdev IO pool count is %zu but should be %u\n",
330 			    spdk_mempool_count(g_bdev_mgr.bdev_io_pool),
331 			    SPDK_BDEV_IO_POOL_SIZE);
332 	}
333 
334 	if (spdk_mempool_count(g_bdev_mgr.buf_small_pool) != BUF_SMALL_POOL_SIZE) {
335 		SPDK_ERRLOG("Small buffer pool count is %zu but should be %u\n",
336 			    spdk_mempool_count(g_bdev_mgr.buf_small_pool),
337 			    BUF_SMALL_POOL_SIZE);
338 		assert(false);
339 	}
340 
341 	if (spdk_mempool_count(g_bdev_mgr.buf_large_pool) != BUF_LARGE_POOL_SIZE) {
342 		SPDK_ERRLOG("Large buffer pool count is %zu but should be %u\n",
343 			    spdk_mempool_count(g_bdev_mgr.buf_large_pool),
344 			    BUF_LARGE_POOL_SIZE);
345 		assert(false);
346 	}
347 
348 	spdk_mempool_free(g_bdev_mgr.bdev_io_pool);
349 	spdk_mempool_free(g_bdev_mgr.buf_small_pool);
350 	spdk_mempool_free(g_bdev_mgr.buf_large_pool);
351 
352 	return 0;
353 }
354 
355 struct spdk_bdev_io *
356 spdk_bdev_get_io(void)
357 {
358 	struct spdk_bdev_io *bdev_io;
359 
360 	bdev_io = spdk_mempool_get(g_bdev_mgr.bdev_io_pool);
361 	if (!bdev_io) {
362 		SPDK_ERRLOG("Unable to get spdk_bdev_io\n");
363 		abort();
364 	}
365 
366 	memset(bdev_io, 0, sizeof(*bdev_io));
367 
368 	return bdev_io;
369 }
370 
371 static void
372 spdk_bdev_put_io(struct spdk_bdev_io *bdev_io)
373 {
374 	if (!bdev_io) {
375 		return;
376 	}
377 
378 	if (bdev_io->buf != NULL) {
379 		spdk_bdev_io_put_buf(bdev_io);
380 	}
381 
382 	spdk_mempool_put(g_bdev_mgr.bdev_io_pool, (void *)bdev_io);
383 }
384 
385 static void
386 spdk_bdev_cleanup_pending_buf_io(struct spdk_bdev *bdev)
387 {
388 	struct spdk_bdev_io *bdev_io, *tmp;
389 
390 	TAILQ_FOREACH_SAFE(bdev_io, &g_bdev_mgr.need_buf_small[rte_lcore_id()], buf_link, tmp) {
391 		if (bdev_io->bdev == bdev) {
392 			TAILQ_REMOVE(&g_bdev_mgr.need_buf_small[rte_lcore_id()], bdev_io, buf_link);
393 			bdev_io->status = SPDK_BDEV_IO_STATUS_FAILED;
394 		}
395 	}
396 
397 	TAILQ_FOREACH_SAFE(bdev_io, &g_bdev_mgr.need_buf_large[rte_lcore_id()], buf_link, tmp) {
398 		if (bdev_io->bdev == bdev) {
399 			TAILQ_REMOVE(&g_bdev_mgr.need_buf_large[rte_lcore_id()], bdev_io, buf_link);
400 			bdev_io->status = SPDK_BDEV_IO_STATUS_FAILED;
401 		}
402 	}
403 }
404 
405 static void
406 __submit_request(struct spdk_bdev *bdev, struct spdk_bdev_io *bdev_io)
407 {
408 	struct spdk_io_channel *ch;
409 
410 	assert(bdev_io->status == SPDK_BDEV_IO_STATUS_PENDING);
411 
412 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_RESET) {
413 		spdk_bdev_cleanup_pending_buf_io(bdev);
414 		ch = NULL;
415 	} else {
416 		ch = bdev_io->ch->channel;
417 	}
418 
419 	bdev_io->in_submit_request = true;
420 	bdev->fn_table->submit_request(ch, bdev_io);
421 	bdev_io->in_submit_request = false;
422 }
423 
424 static int
425 spdk_bdev_io_submit(struct spdk_bdev_io *bdev_io)
426 {
427 	struct spdk_bdev *bdev = bdev_io->bdev;
428 
429 	__submit_request(bdev, bdev_io);
430 	return 0;
431 }
432 
433 void
434 spdk_bdev_io_resubmit(struct spdk_bdev_io *bdev_io, struct spdk_bdev *new_bdev)
435 {
436 	assert(bdev_io->status == SPDK_BDEV_IO_STATUS_PENDING);
437 	bdev_io->bdev = new_bdev;
438 
439 	/*
440 	 * These fields are normally set during spdk_bdev_io_init(), but since bdev is
441 	 * being switched, they need to be reinitialized.
442 	 */
443 	bdev_io->gencnt = new_bdev->gencnt;
444 
445 	__submit_request(new_bdev, bdev_io);
446 }
447 
448 static void
449 spdk_bdev_io_init(struct spdk_bdev_io *bdev_io,
450 		  struct spdk_bdev *bdev, void *cb_arg,
451 		  spdk_bdev_io_completion_cb cb)
452 {
453 	bdev_io->bdev = bdev;
454 	bdev_io->caller_ctx = cb_arg;
455 	bdev_io->cb = cb;
456 	bdev_io->gencnt = bdev->gencnt;
457 	bdev_io->status = SPDK_BDEV_IO_STATUS_PENDING;
458 	bdev_io->in_submit_request = false;
459 	TAILQ_INIT(&bdev_io->child_io);
460 }
461 
462 struct spdk_bdev_io *
463 spdk_bdev_get_child_io(struct spdk_bdev_io *parent,
464 		       struct spdk_bdev *bdev,
465 		       spdk_bdev_io_completion_cb cb,
466 		       void *cb_arg)
467 {
468 	struct spdk_bdev_io *child;
469 
470 	child = spdk_bdev_get_io();
471 	if (!child) {
472 		SPDK_ERRLOG("Unable to get spdk_bdev_io\n");
473 		return NULL;
474 	}
475 
476 	if (cb_arg == NULL) {
477 		cb_arg = child;
478 	}
479 
480 	spdk_bdev_io_init(child, bdev, cb_arg, cb);
481 
482 	child->type = parent->type;
483 	memcpy(&child->u, &parent->u, sizeof(child->u));
484 	child->buf = NULL;
485 	child->get_buf_cb = NULL;
486 	child->parent = parent;
487 
488 	TAILQ_INSERT_TAIL(&parent->child_io, child, link);
489 
490 	return child;
491 }
492 
493 bool
494 spdk_bdev_io_type_supported(struct spdk_bdev *bdev, enum spdk_bdev_io_type io_type)
495 {
496 	return bdev->fn_table->io_type_supported(bdev->ctxt, io_type);
497 }
498 
499 int
500 spdk_bdev_dump_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w)
501 {
502 	if (bdev->fn_table->dump_config_json) {
503 		return bdev->fn_table->dump_config_json(bdev->ctxt, w);
504 	}
505 
506 	return 0;
507 }
508 
509 static int
510 spdk_bdev_channel_create(void *io_device, uint32_t priority, void *ctx_buf,
511 			 void *unique_ctx)
512 {
513 	struct spdk_bdev		*bdev = io_device;
514 	struct spdk_bdev_channel	*ch = ctx_buf;
515 
516 	ch->bdev = io_device;
517 	ch->channel = bdev->fn_table->get_io_channel(bdev->ctxt, priority);
518 
519 	return 0;
520 }
521 
522 static void
523 spdk_bdev_channel_destroy(void *io_device, void *ctx_buf)
524 {
525 	struct spdk_bdev_channel	*ch = ctx_buf;
526 
527 	spdk_put_io_channel(ch->channel);
528 }
529 
530 struct spdk_io_channel *
531 spdk_bdev_get_io_channel(struct spdk_bdev *bdev, uint32_t priority)
532 {
533 	return spdk_get_io_channel(bdev, priority, false, NULL);
534 }
535 
536 const char *
537 spdk_bdev_get_name(const struct spdk_bdev *bdev)
538 {
539 	return bdev->name;
540 }
541 
542 const char *
543 spdk_bdev_get_product_name(const struct spdk_bdev *bdev)
544 {
545 	return bdev->product_name;
546 }
547 
548 uint32_t
549 spdk_bdev_get_block_size(const struct spdk_bdev *bdev)
550 {
551 	return bdev->blocklen;
552 }
553 
554 uint64_t
555 spdk_bdev_get_num_blocks(const struct spdk_bdev *bdev)
556 {
557 	return bdev->blockcnt;
558 }
559 
560 uint32_t
561 spdk_bdev_get_max_unmap_descriptors(const struct spdk_bdev *bdev)
562 {
563 	return bdev->max_unmap_bdesc_count;
564 }
565 
566 size_t
567 spdk_bdev_get_buf_align(const struct spdk_bdev *bdev)
568 {
569 	/* TODO: push this logic down to the bdev modules */
570 	if (bdev->need_aligned_buffer) {
571 		return bdev->blocklen;
572 	}
573 
574 	return 1;
575 }
576 
577 bool
578 spdk_bdev_has_write_cache(const struct spdk_bdev *bdev)
579 {
580 	return bdev->write_cache;
581 }
582 
583 static int
584 spdk_bdev_io_valid(struct spdk_bdev *bdev, uint64_t offset, uint64_t nbytes)
585 {
586 	/* Return failure if nbytes is not a multiple of bdev->blocklen */
587 	if (nbytes % bdev->blocklen) {
588 		return -1;
589 	}
590 
591 	/* Return failure if offset + nbytes is less than offset; indicates there
592 	 * has been an overflow and hence the offset has been wrapped around */
593 	if (offset + nbytes < offset) {
594 		return -1;
595 	}
596 
597 	/* Return failure if offset + nbytes exceeds the size of the blockdev */
598 	if (offset + nbytes > bdev->blockcnt * bdev->blocklen) {
599 		return -1;
600 	}
601 
602 	return 0;
603 }
604 
605 struct spdk_bdev_io *
606 spdk_bdev_read(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
607 	       void *buf, uint64_t offset, uint64_t nbytes,
608 	       spdk_bdev_io_completion_cb cb, void *cb_arg)
609 {
610 	struct spdk_bdev_io *bdev_io;
611 	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
612 	int rc;
613 
614 	assert(bdev->status != SPDK_BDEV_STATUS_UNCLAIMED);
615 	if (spdk_bdev_io_valid(bdev, offset, nbytes) != 0) {
616 		return NULL;
617 	}
618 
619 	bdev_io = spdk_bdev_get_io();
620 	if (!bdev_io) {
621 		SPDK_ERRLOG("spdk_bdev_io memory allocation failed duing read\n");
622 		return NULL;
623 	}
624 
625 	bdev_io->ch = channel;
626 	bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
627 	bdev_io->u.read.iov.iov_base = buf;
628 	bdev_io->u.read.iov.iov_len = nbytes;
629 	bdev_io->u.read.iovs = &bdev_io->u.read.iov;
630 	bdev_io->u.read.iovcnt = 1;
631 	bdev_io->u.read.len = nbytes;
632 	bdev_io->u.read.offset = offset;
633 	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);
634 
635 	rc = spdk_bdev_io_submit(bdev_io);
636 	if (rc < 0) {
637 		spdk_bdev_put_io(bdev_io);
638 		return NULL;
639 	}
640 
641 	return bdev_io;
642 }
643 
644 struct spdk_bdev_io *
645 spdk_bdev_readv(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
646 		struct iovec *iov, int iovcnt,
647 		uint64_t offset, uint64_t nbytes,
648 		spdk_bdev_io_completion_cb cb, void *cb_arg)
649 {
650 	struct spdk_bdev_io *bdev_io;
651 	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
652 	int rc;
653 
654 	assert(bdev->status != SPDK_BDEV_STATUS_UNCLAIMED);
655 	if (spdk_bdev_io_valid(bdev, offset, nbytes) != 0) {
656 		return NULL;
657 	}
658 
659 	bdev_io = spdk_bdev_get_io();
660 	if (!bdev_io) {
661 		SPDK_ERRLOG("spdk_bdev_io memory allocation failed duing read\n");
662 		return NULL;
663 	}
664 
665 	bdev_io->ch = channel;
666 	bdev_io->type = SPDK_BDEV_IO_TYPE_READ;
667 	bdev_io->u.read.iovs = iov;
668 	bdev_io->u.read.iovcnt = iovcnt;
669 	bdev_io->u.read.len = nbytes;
670 	bdev_io->u.read.offset = offset;
671 	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);
672 
673 	rc = spdk_bdev_io_submit(bdev_io);
674 	if (rc < 0) {
675 		spdk_bdev_put_io(bdev_io);
676 		return NULL;
677 	}
678 
679 	return bdev_io;
680 }
681 
682 struct spdk_bdev_io *
683 spdk_bdev_write(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
684 		void *buf, uint64_t offset, uint64_t nbytes,
685 		spdk_bdev_io_completion_cb cb, void *cb_arg)
686 {
687 	struct spdk_bdev_io *bdev_io;
688 	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
689 	int rc;
690 
691 	assert(bdev->status != SPDK_BDEV_STATUS_UNCLAIMED);
692 	if (spdk_bdev_io_valid(bdev, offset, nbytes) != 0) {
693 		return NULL;
694 	}
695 
696 	bdev_io = spdk_bdev_get_io();
697 	if (!bdev_io) {
698 		SPDK_ERRLOG("blockdev_io memory allocation failed duing write\n");
699 		return NULL;
700 	}
701 
702 	bdev_io->ch = channel;
703 	bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
704 	bdev_io->u.write.iov.iov_base = buf;
705 	bdev_io->u.write.iov.iov_len = nbytes;
706 	bdev_io->u.write.iovs = &bdev_io->u.write.iov;
707 	bdev_io->u.write.iovcnt = 1;
708 	bdev_io->u.write.len = nbytes;
709 	bdev_io->u.write.offset = offset;
710 	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);
711 
712 	rc = spdk_bdev_io_submit(bdev_io);
713 	if (rc < 0) {
714 		spdk_bdev_put_io(bdev_io);
715 		return NULL;
716 	}
717 
718 	return bdev_io;
719 }
720 
721 struct spdk_bdev_io *
722 spdk_bdev_writev(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
723 		 struct iovec *iov, int iovcnt,
724 		 uint64_t offset, uint64_t len,
725 		 spdk_bdev_io_completion_cb cb, void *cb_arg)
726 {
727 	struct spdk_bdev_io *bdev_io;
728 	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
729 	int rc;
730 
731 	assert(bdev->status != SPDK_BDEV_STATUS_UNCLAIMED);
732 	if (spdk_bdev_io_valid(bdev, offset, len) != 0) {
733 		return NULL;
734 	}
735 
736 	bdev_io = spdk_bdev_get_io();
737 	if (!bdev_io) {
738 		SPDK_ERRLOG("bdev_io memory allocation failed duing writev\n");
739 		return NULL;
740 	}
741 
742 	bdev_io->ch = channel;
743 	bdev_io->type = SPDK_BDEV_IO_TYPE_WRITE;
744 	bdev_io->u.write.iovs = iov;
745 	bdev_io->u.write.iovcnt = iovcnt;
746 	bdev_io->u.write.len = len;
747 	bdev_io->u.write.offset = offset;
748 	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);
749 
750 	rc = spdk_bdev_io_submit(bdev_io);
751 	if (rc < 0) {
752 		spdk_bdev_put_io(bdev_io);
753 		return NULL;
754 	}
755 
756 	return bdev_io;
757 }
758 
759 struct spdk_bdev_io *
760 spdk_bdev_unmap(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
761 		struct spdk_scsi_unmap_bdesc *unmap_d,
762 		uint16_t bdesc_count,
763 		spdk_bdev_io_completion_cb cb, void *cb_arg)
764 {
765 	struct spdk_bdev_io *bdev_io;
766 	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
767 	int rc;
768 
769 	assert(bdev->status != SPDK_BDEV_STATUS_UNCLAIMED);
770 	if (bdesc_count == 0) {
771 		SPDK_ERRLOG("Invalid bdesc_count 0\n");
772 		return NULL;
773 	}
774 
775 	if (bdesc_count > bdev->max_unmap_bdesc_count) {
776 		SPDK_ERRLOG("Invalid bdesc_count %u > max_unmap_bdesc_count %u\n",
777 			    bdesc_count, bdev->max_unmap_bdesc_count);
778 		return NULL;
779 	}
780 
781 	bdev_io = spdk_bdev_get_io();
782 	if (!bdev_io) {
783 		SPDK_ERRLOG("bdev_io memory allocation failed duing unmap\n");
784 		return NULL;
785 	}
786 
787 	bdev_io->ch = channel;
788 	bdev_io->type = SPDK_BDEV_IO_TYPE_UNMAP;
789 	bdev_io->u.unmap.unmap_bdesc = unmap_d;
790 	bdev_io->u.unmap.bdesc_count = bdesc_count;
791 	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);
792 
793 	rc = spdk_bdev_io_submit(bdev_io);
794 	if (rc < 0) {
795 		spdk_bdev_put_io(bdev_io);
796 		return NULL;
797 	}
798 
799 	return bdev_io;
800 }
801 
802 struct spdk_bdev_io *
803 spdk_bdev_flush(struct spdk_bdev *bdev, struct spdk_io_channel *ch,
804 		uint64_t offset, uint64_t length,
805 		spdk_bdev_io_completion_cb cb, void *cb_arg)
806 {
807 	struct spdk_bdev_io *bdev_io;
808 	struct spdk_bdev_channel *channel = spdk_io_channel_get_ctx(ch);
809 	int rc;
810 
811 	assert(bdev->status != SPDK_BDEV_STATUS_UNCLAIMED);
812 	bdev_io = spdk_bdev_get_io();
813 	if (!bdev_io) {
814 		SPDK_ERRLOG("bdev_io memory allocation failed duing flush\n");
815 		return NULL;
816 	}
817 
818 	bdev_io->ch = channel;
819 	bdev_io->type = SPDK_BDEV_IO_TYPE_FLUSH;
820 	bdev_io->u.flush.offset = offset;
821 	bdev_io->u.flush.length = length;
822 	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);
823 
824 	rc = spdk_bdev_io_submit(bdev_io);
825 	if (rc < 0) {
826 		spdk_bdev_put_io(bdev_io);
827 		return NULL;
828 	}
829 
830 	return bdev_io;
831 }
832 
833 int
834 spdk_bdev_reset(struct spdk_bdev *bdev, enum spdk_bdev_reset_type reset_type,
835 		spdk_bdev_io_completion_cb cb, void *cb_arg)
836 {
837 	struct spdk_bdev_io *bdev_io;
838 	int rc;
839 
840 	assert(bdev->status != SPDK_BDEV_STATUS_UNCLAIMED);
841 	bdev_io = spdk_bdev_get_io();
842 	if (!bdev_io) {
843 		SPDK_ERRLOG("bdev_io memory allocation failed duing reset\n");
844 		return -1;
845 	}
846 
847 	bdev_io->type = SPDK_BDEV_IO_TYPE_RESET;
848 	bdev_io->u.reset.type = reset_type;
849 	spdk_bdev_io_init(bdev_io, bdev, cb_arg, cb);
850 
851 	rc = spdk_bdev_io_submit(bdev_io);
852 	if (rc < 0) {
853 		spdk_bdev_put_io(bdev_io);
854 		SPDK_ERRLOG("reset failed\n");
855 	}
856 
857 	return rc;
858 }
859 
860 int
861 spdk_bdev_free_io(struct spdk_bdev_io *bdev_io)
862 {
863 	struct spdk_bdev_io *child_io, *tmp;
864 
865 	if (!bdev_io) {
866 		SPDK_ERRLOG("bdev_io is NULL\n");
867 		return -1;
868 	}
869 
870 	if (bdev_io->status == SPDK_BDEV_IO_STATUS_PENDING) {
871 		SPDK_ERRLOG("bdev_io is in pending state\n");
872 		assert(false);
873 		return -1;
874 	}
875 
876 	TAILQ_FOREACH_SAFE(child_io, &bdev_io->child_io, link, tmp) {
877 		/*
878 		 * Make sure no references to the parent I/O remain, since it is being
879 		 * returned to the free pool.
880 		 */
881 		child_io->parent = NULL;
882 		TAILQ_REMOVE(&bdev_io->child_io, child_io, link);
883 
884 		/*
885 		 * Child I/O may have a buf that needs to be returned to a pool
886 		 *  on a different core, so free it through the request submission
887 		 *  process rather than calling put_io directly here.
888 		 */
889 		spdk_bdev_free_io(child_io);
890 	}
891 
892 	spdk_bdev_put_io(bdev_io);
893 
894 	return 0;
895 }
896 
897 static void
898 bdev_io_deferred_completion(void *arg1, void *arg2)
899 {
900 	struct spdk_bdev_io *bdev_io = arg1;
901 	enum spdk_bdev_io_status status = (enum spdk_bdev_io_status)arg2;
902 
903 	assert(bdev_io->in_submit_request == false);
904 
905 	spdk_bdev_io_complete(bdev_io, status);
906 }
907 
908 void
909 spdk_bdev_io_complete(struct spdk_bdev_io *bdev_io, enum spdk_bdev_io_status status)
910 {
911 	if (bdev_io->in_submit_request) {
912 		/*
913 		 * Defer completion via an event to avoid potential infinite recursion if the
914 		 * user's completion callback issues a new I/O.
915 		 */
916 		spdk_event_call(spdk_event_allocate(spdk_env_get_current_core(),
917 						    bdev_io_deferred_completion,
918 						    bdev_io,
919 						    (void *)status));
920 		return;
921 	}
922 
923 	if (bdev_io->type == SPDK_BDEV_IO_TYPE_RESET) {
924 		/* Successful reset */
925 		if (status == SPDK_BDEV_IO_STATUS_SUCCESS) {
926 			/* Increase the blockdev generation if it is a hard reset */
927 			if (bdev_io->u.reset.type == SPDK_BDEV_RESET_HARD) {
928 				bdev_io->bdev->gencnt++;
929 			}
930 		}
931 	} else {
932 		/*
933 		 * Check the gencnt, to see if this I/O was issued before the most
934 		 * recent reset. If the gencnt is not equal, then just free the I/O
935 		 * without calling the callback, since the caller will have already
936 		 * freed its context for this I/O.
937 		 */
938 		if (bdev_io->bdev->gencnt != bdev_io->gencnt) {
939 			spdk_bdev_put_io(bdev_io);
940 			return;
941 		}
942 	}
943 
944 	bdev_io->status = status;
945 
946 	assert(bdev_io->cb != NULL);
947 	bdev_io->cb(bdev_io, status, bdev_io->caller_ctx);
948 }
949 
950 void
951 spdk_bdev_io_complete_scsi_status(struct spdk_bdev_io *bdev_io, enum spdk_scsi_status sc,
952 				  enum spdk_scsi_sense sk, uint8_t asc, uint8_t ascq)
953 {
954 	if (sc == SPDK_SCSI_STATUS_GOOD) {
955 		bdev_io->status = SPDK_BDEV_IO_STATUS_SUCCESS;
956 	} else {
957 		bdev_io->status = SPDK_BDEV_IO_STATUS_SCSI_ERROR;
958 		bdev_io->error.scsi.sc = sc;
959 		bdev_io->error.scsi.sk = sk;
960 		bdev_io->error.scsi.asc = asc;
961 		bdev_io->error.scsi.ascq = ascq;
962 	}
963 
964 	spdk_bdev_io_complete(bdev_io, bdev_io->status);
965 }
966 
967 void
968 spdk_bdev_io_get_scsi_status(const struct spdk_bdev_io *bdev_io,
969 			     int *sc, int *sk, int *asc, int *ascq)
970 {
971 	assert(sc != NULL);
972 	assert(sk != NULL);
973 	assert(asc != NULL);
974 	assert(ascq != NULL);
975 
976 	switch (bdev_io->status) {
977 	case SPDK_BDEV_IO_STATUS_SUCCESS:
978 		*sc = SPDK_SCSI_STATUS_GOOD;
979 		*sk = SPDK_SCSI_SENSE_NO_SENSE;
980 		*asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
981 		*ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
982 		break;
983 	case SPDK_BDEV_IO_STATUS_NVME_ERROR:
984 		spdk_scsi_nvme_translate(bdev_io, sc, sk, asc, ascq);
985 		break;
986 	case SPDK_BDEV_IO_STATUS_SCSI_ERROR:
987 		*sc = bdev_io->error.scsi.sc;
988 		*sk = bdev_io->error.scsi.sk;
989 		*asc = bdev_io->error.scsi.asc;
990 		*ascq = bdev_io->error.scsi.ascq;
991 		break;
992 	default:
993 		*sc = SPDK_SCSI_STATUS_CHECK_CONDITION;
994 		*sk = SPDK_SCSI_SENSE_ABORTED_COMMAND;
995 		*asc = SPDK_SCSI_ASC_NO_ADDITIONAL_SENSE;
996 		*ascq = SPDK_SCSI_ASCQ_CAUSE_NOT_REPORTABLE;
997 		break;
998 	}
999 }
1000 
1001 void
1002 spdk_bdev_io_complete_nvme_status(struct spdk_bdev_io *bdev_io, int sct, int sc)
1003 {
1004 	if (sct == SPDK_NVME_SCT_GENERIC && sc == SPDK_NVME_SC_SUCCESS) {
1005 		bdev_io->status = SPDK_BDEV_IO_STATUS_SUCCESS;
1006 	} else {
1007 		bdev_io->error.nvme.sct = sct;
1008 		bdev_io->error.nvme.sc = sc;
1009 		bdev_io->status = SPDK_BDEV_IO_STATUS_NVME_ERROR;
1010 	}
1011 
1012 	spdk_bdev_io_complete(bdev_io, bdev_io->status);
1013 }
1014 
1015 void
1016 spdk_bdev_io_get_nvme_status(const struct spdk_bdev_io *bdev_io, int *sct, int *sc)
1017 {
1018 	assert(sct != NULL);
1019 	assert(sc != NULL);
1020 
1021 	if (bdev_io->status == SPDK_BDEV_IO_STATUS_NVME_ERROR) {
1022 		*sct = bdev_io->error.nvme.sct;
1023 		*sc = bdev_io->error.nvme.sc;
1024 	} else if (bdev_io->status == SPDK_BDEV_IO_STATUS_SUCCESS) {
1025 		*sct = SPDK_NVME_SCT_GENERIC;
1026 		*sc = SPDK_NVME_SC_SUCCESS;
1027 	} else {
1028 		*sct = SPDK_NVME_SCT_GENERIC;
1029 		*sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR;
1030 	}
1031 }
1032 
1033 void
1034 spdk_bdev_register(struct spdk_bdev *bdev)
1035 {
1036 	/* initialize the reset generation value to zero */
1037 	bdev->gencnt = 0;
1038 
1039 	spdk_io_device_register(bdev, spdk_bdev_channel_create, spdk_bdev_channel_destroy,
1040 				sizeof(struct spdk_bdev_channel));
1041 
1042 	pthread_mutex_init(&bdev->mutex, NULL);
1043 	bdev->status = SPDK_BDEV_STATUS_UNCLAIMED;
1044 	SPDK_TRACELOG(SPDK_TRACE_DEBUG, "Inserting bdev %s into list\n", bdev->name);
1045 	TAILQ_INSERT_TAIL(&g_bdev_mgr.bdevs, bdev, link);
1046 }
1047 
1048 void
1049 spdk_bdev_unregister(struct spdk_bdev *bdev)
1050 {
1051 	int			rc;
1052 
1053 	SPDK_TRACELOG(SPDK_TRACE_DEBUG, "Removing bdev %s from list\n", bdev->name);
1054 
1055 	pthread_mutex_lock(&bdev->mutex);
1056 	assert(bdev->status == SPDK_BDEV_STATUS_CLAIMED || bdev->status == SPDK_BDEV_STATUS_UNCLAIMED);
1057 	if (bdev->status == SPDK_BDEV_STATUS_CLAIMED) {
1058 		if (bdev->remove_cb) {
1059 			bdev->status = SPDK_BDEV_STATUS_REMOVING;
1060 			pthread_mutex_unlock(&bdev->mutex);
1061 			bdev->remove_cb(bdev->remove_ctx);
1062 			return;
1063 		} else {
1064 			bdev->status = SPDK_BDEV_STATUS_UNCLAIMED;
1065 		}
1066 	}
1067 
1068 	TAILQ_REMOVE(&g_bdev_mgr.bdevs, bdev, link);
1069 	pthread_mutex_unlock(&bdev->mutex);
1070 
1071 	pthread_mutex_destroy(&bdev->mutex);
1072 
1073 	spdk_io_device_unregister(bdev);
1074 
1075 	rc = bdev->fn_table->destruct(bdev->ctxt);
1076 	if (rc < 0) {
1077 		SPDK_ERRLOG("destruct failed\n");
1078 	}
1079 }
1080 
1081 bool
1082 spdk_bdev_claim(struct spdk_bdev *bdev, spdk_bdev_remove_cb_t remove_cb,
1083 		void *remove_ctx)
1084 {
1085 	bool success;
1086 
1087 	pthread_mutex_lock(&bdev->mutex);
1088 
1089 	if (bdev->status != SPDK_BDEV_STATUS_CLAIMED) {
1090 		/* Take ownership of bdev. */
1091 		bdev->remove_cb = remove_cb;
1092 		bdev->remove_ctx = remove_ctx;
1093 		bdev->status = SPDK_BDEV_STATUS_CLAIMED;
1094 		success = true;
1095 	} else {
1096 		/* bdev is already claimed. */
1097 		success = false;
1098 	}
1099 
1100 	pthread_mutex_unlock(&bdev->mutex);
1101 
1102 	return success;
1103 }
1104 
1105 void
1106 spdk_bdev_unclaim(struct spdk_bdev *bdev)
1107 {
1108 	bool do_unregister = false;
1109 
1110 	pthread_mutex_lock(&bdev->mutex);
1111 	assert(bdev->status == SPDK_BDEV_STATUS_CLAIMED || bdev->status == SPDK_BDEV_STATUS_REMOVING);
1112 	if (bdev->status == SPDK_BDEV_STATUS_REMOVING) {
1113 		do_unregister = true;
1114 	}
1115 	bdev->remove_cb = NULL;
1116 	bdev->remove_ctx = NULL;
1117 	bdev->status = SPDK_BDEV_STATUS_UNCLAIMED;
1118 	pthread_mutex_unlock(&bdev->mutex);
1119 
1120 	if (do_unregister == true) {
1121 		spdk_bdev_unregister(bdev);
1122 	}
1123 }
1124 
1125 void
1126 spdk_bdev_io_get_iovec(struct spdk_bdev_io *bdev_io, struct iovec **iovp, int *iovcntp)
1127 {
1128 	struct iovec *iovs;
1129 	int iovcnt;
1130 
1131 	if (bdev_io == NULL) {
1132 		return;
1133 	}
1134 
1135 	switch (bdev_io->type) {
1136 	case SPDK_BDEV_IO_TYPE_READ:
1137 		iovs = bdev_io->u.read.iovs;
1138 		iovcnt = bdev_io->u.read.iovcnt;
1139 		break;
1140 	case SPDK_BDEV_IO_TYPE_WRITE:
1141 		iovs = bdev_io->u.write.iovs;
1142 		iovcnt = bdev_io->u.write.iovcnt;
1143 		break;
1144 	default:
1145 		iovs = NULL;
1146 		iovcnt = 0;
1147 		break;
1148 	}
1149 
1150 	if (iovp) {
1151 		*iovp = iovs;
1152 	}
1153 	if (iovcntp) {
1154 		*iovcntp = iovcnt;
1155 	}
1156 }
1157 
1158 void
1159 spdk_bdev_module_list_add(struct spdk_bdev_module_if *bdev_module)
1160 {
1161 	TAILQ_INSERT_TAIL(&g_bdev_mgr.bdev_modules, bdev_module, tailq);
1162 }
1163 
1164 void
1165 spdk_vbdev_module_list_add(struct spdk_bdev_module_if *vbdev_module)
1166 {
1167 	TAILQ_INSERT_TAIL(&g_bdev_mgr.vbdev_modules, vbdev_module, tailq);
1168 }
1169 SPDK_SUBSYSTEM_REGISTER(bdev, spdk_bdev_initialize, spdk_bdev_finish, spdk_bdev_config_text)
1170 SPDK_SUBSYSTEM_DEPEND(bdev, copy)
1171