xref: /spdk/lib/blob/blobstore.c (revision 795134891e85e6d445a25318ed4f03da9f3dc8f3)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "spdk/blob.h"
37 #include "spdk/crc32.h"
38 #include "spdk/env.h"
39 #include "spdk/queue.h"
40 #include "spdk/thread.h"
41 #include "spdk/bit_array.h"
42 #include "spdk/likely.h"
43 #include "spdk/util.h"
44 
45 #include "spdk_internal/assert.h"
46 #include "spdk_internal/log.h"
47 
48 #include "blobstore.h"
49 
50 #define BLOB_CRC32C_INITIAL    0xffffffffUL
51 
52 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs);
53 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs);
54 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
55 static void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
56 		uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg);
57 
58 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
59 				uint16_t value_len, bool internal);
60 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
61 				      const void **value, size_t *value_len, bool internal);
62 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal);
63 
64 static void
65 _spdk_blob_verify_md_op(struct spdk_blob *blob)
66 {
67 	assert(blob != NULL);
68 	assert(spdk_get_thread() == blob->bs->md_thread);
69 	assert(blob->state != SPDK_BLOB_STATE_LOADING);
70 }
71 
72 static void
73 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
74 {
75 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
76 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false);
77 	assert(bs->num_free_clusters > 0);
78 
79 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num);
80 
81 	spdk_bit_array_set(bs->used_clusters, cluster_num);
82 	bs->num_free_clusters--;
83 }
84 
85 static int
86 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
87 {
88 	uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
89 
90 	_spdk_blob_verify_md_op(blob);
91 
92 	if (*cluster_lba != 0) {
93 		return -EEXIST;
94 	}
95 
96 	*cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster);
97 	return 0;
98 }
99 
100 static int
101 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
102 			  uint64_t *lowest_free_cluster, bool update_map)
103 {
104 	pthread_mutex_lock(&blob->bs->used_clusters_mutex);
105 	*lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters,
106 			       *lowest_free_cluster);
107 	if (*lowest_free_cluster == UINT32_MAX) {
108 		/* No more free clusters. Cannot satisfy the request */
109 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
110 		return -ENOSPC;
111 	}
112 
113 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id);
114 	_spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster);
115 	pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
116 
117 	if (update_map) {
118 		_spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster);
119 	}
120 
121 	return 0;
122 }
123 
124 static void
125 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
126 {
127 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
128 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true);
129 	assert(bs->num_free_clusters < bs->total_clusters);
130 
131 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num);
132 
133 	pthread_mutex_lock(&bs->used_clusters_mutex);
134 	spdk_bit_array_clear(bs->used_clusters, cluster_num);
135 	bs->num_free_clusters++;
136 	pthread_mutex_unlock(&bs->used_clusters_mutex);
137 }
138 
139 static void
140 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
141 {
142 	xattrs->count = 0;
143 	xattrs->names = NULL;
144 	xattrs->ctx = NULL;
145 	xattrs->get_value = NULL;
146 }
147 
148 void
149 spdk_blob_opts_init(struct spdk_blob_opts *opts)
150 {
151 	opts->num_clusters = 0;
152 	opts->thin_provision = false;
153 	_spdk_blob_xattrs_init(&opts->xattrs);
154 }
155 
156 void
157 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts)
158 {
159 	opts->clear_method = BLOB_CLEAR_WITH_UNMAP;
160 }
161 
162 static struct spdk_blob *
163 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
164 {
165 	struct spdk_blob *blob;
166 
167 	blob = calloc(1, sizeof(*blob));
168 	if (!blob) {
169 		return NULL;
170 	}
171 
172 	blob->id = id;
173 	blob->bs = bs;
174 
175 	blob->parent_id = SPDK_BLOBID_INVALID;
176 
177 	blob->state = SPDK_BLOB_STATE_DIRTY;
178 	blob->active.num_pages = 1;
179 	blob->active.pages = calloc(1, sizeof(*blob->active.pages));
180 	if (!blob->active.pages) {
181 		free(blob);
182 		return NULL;
183 	}
184 
185 	blob->active.pages[0] = _spdk_bs_blobid_to_page(id);
186 
187 	TAILQ_INIT(&blob->xattrs);
188 	TAILQ_INIT(&blob->xattrs_internal);
189 
190 	return blob;
191 }
192 
193 static void
194 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs)
195 {
196 	struct spdk_xattr	*xattr, *xattr_tmp;
197 
198 	TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
199 		TAILQ_REMOVE(xattrs, xattr, link);
200 		free(xattr->name);
201 		free(xattr->value);
202 		free(xattr);
203 	}
204 }
205 
206 static void
207 _spdk_blob_free(struct spdk_blob *blob)
208 {
209 	assert(blob != NULL);
210 
211 	free(blob->active.clusters);
212 	free(blob->clean.clusters);
213 	free(blob->active.pages);
214 	free(blob->clean.pages);
215 
216 	_spdk_xattrs_free(&blob->xattrs);
217 	_spdk_xattrs_free(&blob->xattrs_internal);
218 
219 	if (blob->back_bs_dev) {
220 		blob->back_bs_dev->destroy(blob->back_bs_dev);
221 	}
222 
223 	free(blob);
224 }
225 
226 struct freeze_io_ctx {
227 	struct spdk_bs_cpl cpl;
228 	struct spdk_blob *blob;
229 };
230 
231 static void
232 _spdk_blob_io_sync(struct spdk_io_channel_iter *i)
233 {
234 	spdk_for_each_channel_continue(i, 0);
235 }
236 
237 static void
238 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i)
239 {
240 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
241 	struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
242 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
243 	struct spdk_bs_request_set	*set;
244 	struct spdk_bs_user_op_args	*args;
245 	spdk_bs_user_op_t *op, *tmp;
246 
247 	TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
248 		set = (struct spdk_bs_request_set *)op;
249 		args = &set->u.user_op;
250 
251 		if (args->blob == ctx->blob) {
252 			TAILQ_REMOVE(&ch->queued_io, op, link);
253 			spdk_bs_user_op_execute(op);
254 		}
255 	}
256 
257 	spdk_for_each_channel_continue(i, 0);
258 }
259 
260 static void
261 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status)
262 {
263 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
264 
265 	ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
266 
267 	free(ctx);
268 }
269 
270 static void
271 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
272 {
273 	struct freeze_io_ctx *ctx;
274 
275 	ctx = calloc(1, sizeof(*ctx));
276 	if (!ctx) {
277 		cb_fn(cb_arg, -ENOMEM);
278 		return;
279 	}
280 
281 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
282 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
283 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
284 	ctx->blob = blob;
285 
286 	/* Freeze I/O on blob */
287 	blob->frozen_refcnt++;
288 
289 	if (blob->frozen_refcnt == 1) {
290 		spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl);
291 	} else {
292 		cb_fn(cb_arg, 0);
293 		free(ctx);
294 	}
295 }
296 
297 static void
298 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
299 {
300 	struct freeze_io_ctx *ctx;
301 
302 	ctx = calloc(1, sizeof(*ctx));
303 	if (!ctx) {
304 		cb_fn(cb_arg, -ENOMEM);
305 		return;
306 	}
307 
308 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
309 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
310 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
311 	ctx->blob = blob;
312 
313 	assert(blob->frozen_refcnt > 0);
314 
315 	blob->frozen_refcnt--;
316 
317 	if (blob->frozen_refcnt == 0) {
318 		spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl);
319 	} else {
320 		cb_fn(cb_arg, 0);
321 		free(ctx);
322 	}
323 }
324 
325 static int
326 _spdk_blob_mark_clean(struct spdk_blob *blob)
327 {
328 	uint64_t *clusters = NULL;
329 	uint32_t *pages = NULL;
330 
331 	assert(blob != NULL);
332 
333 	if (blob->active.num_clusters) {
334 		assert(blob->active.clusters);
335 		clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
336 		if (!clusters) {
337 			return -ENOMEM;
338 		}
339 		memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*clusters));
340 	}
341 
342 	if (blob->active.num_pages) {
343 		assert(blob->active.pages);
344 		pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
345 		if (!pages) {
346 			free(clusters);
347 			return -ENOMEM;
348 		}
349 		memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*pages));
350 	}
351 
352 	free(blob->clean.clusters);
353 	free(blob->clean.pages);
354 
355 	blob->clean.num_clusters = blob->active.num_clusters;
356 	blob->clean.clusters = blob->active.clusters;
357 	blob->clean.num_pages = blob->active.num_pages;
358 	blob->clean.pages = blob->active.pages;
359 
360 	blob->active.clusters = clusters;
361 	blob->active.pages = pages;
362 
363 	/* If the metadata was dirtied again while the metadata was being written to disk,
364 	 *  we do not want to revert the DIRTY state back to CLEAN here.
365 	 */
366 	if (blob->state == SPDK_BLOB_STATE_LOADING) {
367 		blob->state = SPDK_BLOB_STATE_CLEAN;
368 	}
369 
370 	return 0;
371 }
372 
373 static int
374 _spdk_blob_deserialize_xattr(struct spdk_blob *blob,
375 			     struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
376 {
377 	struct spdk_xattr                       *xattr;
378 
379 	if (desc_xattr->length != sizeof(desc_xattr->name_length) +
380 	    sizeof(desc_xattr->value_length) +
381 	    desc_xattr->name_length + desc_xattr->value_length) {
382 		return -EINVAL;
383 	}
384 
385 	xattr = calloc(1, sizeof(*xattr));
386 	if (xattr == NULL) {
387 		return -ENOMEM;
388 	}
389 
390 	xattr->name = malloc(desc_xattr->name_length + 1);
391 	if (xattr->name == NULL) {
392 		free(xattr);
393 		return -ENOMEM;
394 	}
395 	memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
396 	xattr->name[desc_xattr->name_length] = '\0';
397 
398 	xattr->value = malloc(desc_xattr->value_length);
399 	if (xattr->value == NULL) {
400 		free(xattr->name);
401 		free(xattr);
402 		return -ENOMEM;
403 	}
404 	xattr->value_len = desc_xattr->value_length;
405 	memcpy(xattr->value,
406 	       (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
407 	       desc_xattr->value_length);
408 
409 	TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
410 
411 	return 0;
412 }
413 
414 
415 static int
416 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
417 {
418 	struct spdk_blob_md_descriptor *desc;
419 	size_t	cur_desc = 0;
420 	void *tmp;
421 
422 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
423 	while (cur_desc < sizeof(page->descriptors)) {
424 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
425 			if (desc->length == 0) {
426 				/* If padding and length are 0, this terminates the page */
427 				break;
428 			}
429 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
430 			struct spdk_blob_md_descriptor_flags	*desc_flags;
431 
432 			desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
433 
434 			if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
435 				return -EINVAL;
436 			}
437 
438 			if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
439 			    SPDK_BLOB_INVALID_FLAGS_MASK) {
440 				return -EINVAL;
441 			}
442 
443 			if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
444 			    SPDK_BLOB_DATA_RO_FLAGS_MASK) {
445 				blob->data_ro = true;
446 				blob->md_ro = true;
447 			}
448 
449 			if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
450 			    SPDK_BLOB_MD_RO_FLAGS_MASK) {
451 				blob->md_ro = true;
452 			}
453 
454 			if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
455 				blob->data_ro = true;
456 				blob->md_ro = true;
457 			}
458 
459 			blob->invalid_flags = desc_flags->invalid_flags;
460 			blob->data_ro_flags = desc_flags->data_ro_flags;
461 			blob->md_ro_flags = desc_flags->md_ro_flags;
462 
463 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
464 			struct spdk_blob_md_descriptor_extent	*desc_extent;
465 			unsigned int				i, j;
466 			unsigned int				cluster_count = blob->active.num_clusters;
467 
468 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
469 
470 			if (desc_extent->length == 0 ||
471 			    (desc_extent->length % sizeof(desc_extent->extents[0]) != 0)) {
472 				return -EINVAL;
473 			}
474 
475 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
476 				for (j = 0; j < desc_extent->extents[i].length; j++) {
477 					if (desc_extent->extents[i].cluster_idx != 0) {
478 						if (!spdk_bit_array_get(blob->bs->used_clusters,
479 									desc_extent->extents[i].cluster_idx + j)) {
480 							return -EINVAL;
481 						}
482 					}
483 					cluster_count++;
484 				}
485 			}
486 
487 			if (cluster_count == 0) {
488 				return -EINVAL;
489 			}
490 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t));
491 			if (tmp == NULL) {
492 				return -ENOMEM;
493 			}
494 			blob->active.clusters = tmp;
495 			blob->active.cluster_array_size = cluster_count;
496 
497 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
498 				for (j = 0; j < desc_extent->extents[i].length; j++) {
499 					if (desc_extent->extents[i].cluster_idx != 0) {
500 						blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
501 								desc_extent->extents[i].cluster_idx + j);
502 					} else if (spdk_blob_is_thin_provisioned(blob)) {
503 						blob->active.clusters[blob->active.num_clusters++] = 0;
504 					} else {
505 						return -EINVAL;
506 					}
507 				}
508 			}
509 
510 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
511 			int rc;
512 
513 			rc = _spdk_blob_deserialize_xattr(blob,
514 							  (struct spdk_blob_md_descriptor_xattr *) desc, false);
515 			if (rc != 0) {
516 				return rc;
517 			}
518 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
519 			int rc;
520 
521 			rc = _spdk_blob_deserialize_xattr(blob,
522 							  (struct spdk_blob_md_descriptor_xattr *) desc, true);
523 			if (rc != 0) {
524 				return rc;
525 			}
526 		} else {
527 			/* Unrecognized descriptor type.  Do not fail - just continue to the
528 			 *  next descriptor.  If this descriptor is associated with some feature
529 			 *  defined in a newer version of blobstore, that version of blobstore
530 			 *  should create and set an associated feature flag to specify if this
531 			 *  blob can be loaded or not.
532 			 */
533 		}
534 
535 		/* Advance to the next descriptor */
536 		cur_desc += sizeof(*desc) + desc->length;
537 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
538 			break;
539 		}
540 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
541 	}
542 
543 	return 0;
544 }
545 
546 static int
547 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
548 		 struct spdk_blob *blob)
549 {
550 	const struct spdk_blob_md_page *page;
551 	uint32_t i;
552 	int rc;
553 
554 	assert(page_count > 0);
555 	assert(pages[0].sequence_num == 0);
556 	assert(blob != NULL);
557 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
558 	assert(blob->active.clusters == NULL);
559 
560 	/* The blobid provided doesn't match what's in the MD, this can
561 	 * happen for example if a bogus blobid is passed in through open.
562 	 */
563 	if (blob->id != pages[0].id) {
564 		SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n",
565 			    blob->id, pages[0].id);
566 		return -ENOENT;
567 	}
568 
569 	for (i = 0; i < page_count; i++) {
570 		page = &pages[i];
571 
572 		assert(page->id == blob->id);
573 		assert(page->sequence_num == i);
574 
575 		rc = _spdk_blob_parse_page(page, blob);
576 		if (rc != 0) {
577 			return rc;
578 		}
579 	}
580 
581 	return 0;
582 }
583 
584 static int
585 _spdk_blob_serialize_add_page(const struct spdk_blob *blob,
586 			      struct spdk_blob_md_page **pages,
587 			      uint32_t *page_count,
588 			      struct spdk_blob_md_page **last_page)
589 {
590 	struct spdk_blob_md_page *page;
591 
592 	assert(pages != NULL);
593 	assert(page_count != NULL);
594 
595 	if (*page_count == 0) {
596 		assert(*pages == NULL);
597 		*page_count = 1;
598 		*pages = spdk_malloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
599 				     NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
600 	} else {
601 		assert(*pages != NULL);
602 		(*page_count)++;
603 		*pages = spdk_realloc(*pages,
604 				      SPDK_BS_PAGE_SIZE * (*page_count),
605 				      SPDK_BS_PAGE_SIZE);
606 	}
607 
608 	if (*pages == NULL) {
609 		*page_count = 0;
610 		*last_page = NULL;
611 		return -ENOMEM;
612 	}
613 
614 	page = &(*pages)[*page_count - 1];
615 	memset(page, 0, sizeof(*page));
616 	page->id = blob->id;
617 	page->sequence_num = *page_count - 1;
618 	page->next = SPDK_INVALID_MD_PAGE;
619 	*last_page = page;
620 
621 	return 0;
622 }
623 
624 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
625  * Update required_sz on both success and failure.
626  *
627  */
628 static int
629 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
630 			   uint8_t *buf, size_t buf_sz,
631 			   size_t *required_sz, bool internal)
632 {
633 	struct spdk_blob_md_descriptor_xattr	*desc;
634 
635 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
636 		       strlen(xattr->name) +
637 		       xattr->value_len;
638 
639 	if (buf_sz < *required_sz) {
640 		return -1;
641 	}
642 
643 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
644 
645 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
646 	desc->length = sizeof(desc->name_length) +
647 		       sizeof(desc->value_length) +
648 		       strlen(xattr->name) +
649 		       xattr->value_len;
650 	desc->name_length = strlen(xattr->name);
651 	desc->value_length = xattr->value_len;
652 
653 	memcpy(desc->name, xattr->name, desc->name_length);
654 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
655 	       xattr->value,
656 	       desc->value_length);
657 
658 	return 0;
659 }
660 
661 static void
662 _spdk_blob_serialize_extent(const struct spdk_blob *blob,
663 			    uint64_t start_cluster, uint64_t *next_cluster,
664 			    uint8_t *buf, size_t buf_sz)
665 {
666 	struct spdk_blob_md_descriptor_extent *desc;
667 	size_t cur_sz;
668 	uint64_t i, extent_idx;
669 	uint64_t lba, lba_per_cluster, lba_count;
670 
671 	/* The buffer must have room for at least one extent */
672 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]);
673 	if (buf_sz < cur_sz) {
674 		*next_cluster = start_cluster;
675 		return;
676 	}
677 
678 	desc = (struct spdk_blob_md_descriptor_extent *)buf;
679 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT;
680 
681 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
682 
683 	lba = blob->active.clusters[start_cluster];
684 	lba_count = lba_per_cluster;
685 	extent_idx = 0;
686 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
687 		if ((lba + lba_count) == blob->active.clusters[i]) {
688 			lba_count += lba_per_cluster;
689 			continue;
690 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
691 			lba_count += lba_per_cluster;
692 			continue;
693 		}
694 		desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
695 		desc->extents[extent_idx].length = lba_count / lba_per_cluster;
696 		extent_idx++;
697 
698 		cur_sz += sizeof(desc->extents[extent_idx]);
699 
700 		if (buf_sz < cur_sz) {
701 			/* If we ran out of buffer space, return */
702 			desc->length = sizeof(desc->extents[0]) * extent_idx;
703 			*next_cluster = i;
704 			return;
705 		}
706 
707 		lba = blob->active.clusters[i];
708 		lba_count = lba_per_cluster;
709 	}
710 
711 	desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
712 	desc->extents[extent_idx].length = lba_count / lba_per_cluster;
713 	extent_idx++;
714 
715 	desc->length = sizeof(desc->extents[0]) * extent_idx;
716 	*next_cluster = blob->active.num_clusters;
717 
718 	return;
719 }
720 
721 static void
722 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
723 			   uint8_t *buf, size_t *buf_sz)
724 {
725 	struct spdk_blob_md_descriptor_flags *desc;
726 
727 	/*
728 	 * Flags get serialized first, so we should always have room for the flags
729 	 *  descriptor.
730 	 */
731 	assert(*buf_sz >= sizeof(*desc));
732 
733 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
734 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
735 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
736 	desc->invalid_flags = blob->invalid_flags;
737 	desc->data_ro_flags = blob->data_ro_flags;
738 	desc->md_ro_flags = blob->md_ro_flags;
739 
740 	*buf_sz -= sizeof(*desc);
741 }
742 
743 static int
744 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
745 			    const struct spdk_xattr_tailq *xattrs, bool internal,
746 			    struct spdk_blob_md_page **pages,
747 			    struct spdk_blob_md_page *cur_page,
748 			    uint32_t *page_count, uint8_t **buf,
749 			    size_t *remaining_sz)
750 {
751 	const struct spdk_xattr	*xattr;
752 	int	rc;
753 
754 	TAILQ_FOREACH(xattr, xattrs, link) {
755 		size_t required_sz = 0;
756 
757 		rc = _spdk_blob_serialize_xattr(xattr,
758 						*buf, *remaining_sz,
759 						&required_sz, internal);
760 		if (rc < 0) {
761 			/* Need to add a new page to the chain */
762 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
763 							   &cur_page);
764 			if (rc < 0) {
765 				spdk_free(*pages);
766 				*pages = NULL;
767 				*page_count = 0;
768 				return rc;
769 			}
770 
771 			*buf = (uint8_t *)cur_page->descriptors;
772 			*remaining_sz = sizeof(cur_page->descriptors);
773 
774 			/* Try again */
775 			required_sz = 0;
776 			rc = _spdk_blob_serialize_xattr(xattr,
777 							*buf, *remaining_sz,
778 							&required_sz, internal);
779 
780 			if (rc < 0) {
781 				spdk_free(*pages);
782 				*pages = NULL;
783 				*page_count = 0;
784 				return rc;
785 			}
786 		}
787 
788 		*remaining_sz -= required_sz;
789 		*buf += required_sz;
790 	}
791 
792 	return 0;
793 }
794 
795 static int
796 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
797 		     uint32_t *page_count)
798 {
799 	struct spdk_blob_md_page		*cur_page;
800 	int					rc;
801 	uint8_t					*buf;
802 	size_t					remaining_sz;
803 	uint64_t				last_cluster;
804 
805 	assert(pages != NULL);
806 	assert(page_count != NULL);
807 	assert(blob != NULL);
808 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
809 
810 	*pages = NULL;
811 	*page_count = 0;
812 
813 	/* A blob always has at least 1 page, even if it has no descriptors */
814 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
815 	if (rc < 0) {
816 		return rc;
817 	}
818 
819 	buf = (uint8_t *)cur_page->descriptors;
820 	remaining_sz = sizeof(cur_page->descriptors);
821 
822 	/* Serialize flags */
823 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
824 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
825 
826 	/* Serialize xattrs */
827 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
828 					 pages, cur_page, page_count, &buf, &remaining_sz);
829 	if (rc < 0) {
830 		return rc;
831 	}
832 
833 	/* Serialize internal xattrs */
834 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
835 					 pages, cur_page, page_count, &buf, &remaining_sz);
836 	if (rc < 0) {
837 		return rc;
838 	}
839 
840 	/* Serialize extents */
841 	last_cluster = 0;
842 	while (last_cluster < blob->active.num_clusters) {
843 		_spdk_blob_serialize_extent(blob, last_cluster, &last_cluster,
844 					    buf, remaining_sz);
845 
846 		if (last_cluster == blob->active.num_clusters) {
847 			break;
848 		}
849 
850 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
851 						   &cur_page);
852 		if (rc < 0) {
853 			return rc;
854 		}
855 
856 		buf = (uint8_t *)cur_page->descriptors;
857 		remaining_sz = sizeof(cur_page->descriptors);
858 	}
859 
860 	return 0;
861 }
862 
863 struct spdk_blob_load_ctx {
864 	struct spdk_blob		*blob;
865 
866 	struct spdk_blob_md_page	*pages;
867 	uint32_t			num_pages;
868 	spdk_bs_sequence_t	        *seq;
869 
870 	spdk_bs_sequence_cpl		cb_fn;
871 	void				*cb_arg;
872 };
873 
874 static uint32_t
875 _spdk_blob_md_page_calc_crc(void *page)
876 {
877 	uint32_t		crc;
878 
879 	crc = BLOB_CRC32C_INITIAL;
880 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
881 	crc ^= BLOB_CRC32C_INITIAL;
882 
883 	return crc;
884 
885 }
886 
887 static void
888 _spdk_blob_load_final(void *cb_arg, int bserrno)
889 {
890 	struct spdk_blob_load_ctx	*ctx = cb_arg;
891 	struct spdk_blob		*blob = ctx->blob;
892 
893 	_spdk_blob_mark_clean(blob);
894 
895 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
896 
897 	/* Free the memory */
898 	spdk_free(ctx->pages);
899 	free(ctx);
900 }
901 
902 static void
903 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
904 {
905 	struct spdk_blob_load_ctx	*ctx = cb_arg;
906 	struct spdk_blob		*blob = ctx->blob;
907 
908 	if (bserrno != 0) {
909 		goto error;
910 	}
911 
912 	blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
913 
914 	if (blob->back_bs_dev == NULL) {
915 		bserrno = -ENOMEM;
916 		goto error;
917 	}
918 
919 	_spdk_blob_load_final(ctx, bserrno);
920 	return;
921 
922 error:
923 	SPDK_ERRLOG("Snapshot fail\n");
924 	_spdk_blob_free(blob);
925 	ctx->cb_fn(ctx->seq, NULL, bserrno);
926 	spdk_free(ctx->pages);
927 	free(ctx);
928 }
929 
930 static void
931 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
932 {
933 	struct spdk_blob_load_ctx	*ctx = cb_arg;
934 	struct spdk_blob		*blob = ctx->blob;
935 	struct spdk_blob_md_page	*page;
936 	const void			*value;
937 	size_t				len;
938 	int				rc;
939 	uint32_t			crc;
940 
941 	if (bserrno) {
942 		SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno);
943 		_spdk_blob_free(blob);
944 		ctx->cb_fn(seq, NULL, bserrno);
945 		spdk_free(ctx->pages);
946 		free(ctx);
947 		return;
948 	}
949 
950 	page = &ctx->pages[ctx->num_pages - 1];
951 	crc = _spdk_blob_md_page_calc_crc(page);
952 	if (crc != page->crc) {
953 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
954 		_spdk_blob_free(blob);
955 		ctx->cb_fn(seq, NULL, -EINVAL);
956 		spdk_free(ctx->pages);
957 		free(ctx);
958 		return;
959 	}
960 
961 	if (page->next != SPDK_INVALID_MD_PAGE) {
962 		uint32_t next_page = page->next;
963 		uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page);
964 
965 
966 		assert(next_lba < (blob->bs->md_start + blob->bs->md_len));
967 
968 		/* Read the next page */
969 		ctx->num_pages++;
970 		ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
971 					  sizeof(*page));
972 		if (ctx->pages == NULL) {
973 			ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM);
974 			free(ctx);
975 			return;
976 		}
977 
978 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
979 					  next_lba,
980 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
981 					  _spdk_blob_load_cpl, ctx);
982 		return;
983 	}
984 
985 	/* Parse the pages */
986 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
987 	if (rc) {
988 		_spdk_blob_free(blob);
989 		ctx->cb_fn(seq, NULL, rc);
990 		spdk_free(ctx->pages);
991 		free(ctx);
992 		return;
993 	}
994 	ctx->seq = seq;
995 
996 
997 	if (spdk_blob_is_thin_provisioned(blob)) {
998 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
999 		if (rc == 0) {
1000 			if (len != sizeof(spdk_blob_id)) {
1001 				_spdk_blob_free(blob);
1002 				ctx->cb_fn(seq, NULL, -EINVAL);
1003 				spdk_free(ctx->pages);
1004 				free(ctx);
1005 				return;
1006 			}
1007 			/* open snapshot blob and continue in the callback function */
1008 			blob->parent_id = *(spdk_blob_id *)value;
1009 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1010 					  _spdk_blob_load_snapshot_cpl, ctx);
1011 			return;
1012 		} else {
1013 			/* add zeroes_dev for thin provisioned blob */
1014 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1015 		}
1016 	} else {
1017 		/* standard blob */
1018 		blob->back_bs_dev = NULL;
1019 	}
1020 	_spdk_blob_load_final(ctx, bserrno);
1021 }
1022 
1023 /* Load a blob from disk given a blobid */
1024 static void
1025 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1026 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1027 {
1028 	struct spdk_blob_load_ctx *ctx;
1029 	struct spdk_blob_store *bs;
1030 	uint32_t page_num;
1031 	uint64_t lba;
1032 
1033 	_spdk_blob_verify_md_op(blob);
1034 
1035 	bs = blob->bs;
1036 
1037 	ctx = calloc(1, sizeof(*ctx));
1038 	if (!ctx) {
1039 		cb_fn(seq, cb_arg, -ENOMEM);
1040 		return;
1041 	}
1042 
1043 	ctx->blob = blob;
1044 	ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE);
1045 	if (!ctx->pages) {
1046 		free(ctx);
1047 		cb_fn(seq, cb_arg, -ENOMEM);
1048 		return;
1049 	}
1050 	ctx->num_pages = 1;
1051 	ctx->cb_fn = cb_fn;
1052 	ctx->cb_arg = cb_arg;
1053 
1054 	page_num = _spdk_bs_blobid_to_page(blob->id);
1055 	lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num);
1056 
1057 	blob->state = SPDK_BLOB_STATE_LOADING;
1058 
1059 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1060 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1061 				  _spdk_blob_load_cpl, ctx);
1062 }
1063 
1064 struct spdk_blob_persist_ctx {
1065 	struct spdk_blob		*blob;
1066 
1067 	struct spdk_bs_super_block	*super;
1068 
1069 	struct spdk_blob_md_page	*pages;
1070 
1071 	uint64_t			idx;
1072 
1073 	spdk_bs_sequence_t		*seq;
1074 	spdk_bs_sequence_cpl		cb_fn;
1075 	void				*cb_arg;
1076 };
1077 
1078 static void
1079 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1080 			uint32_t lba_count)
1081 {
1082 	if (ctx->blob->clear_method == BLOB_CLEAR_WITH_DEFAULT ||
1083 	    ctx->blob->clear_method == BLOB_CLEAR_WITH_UNMAP) {
1084 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1085 	} else if (ctx->blob->clear_method == BLOB_CLEAR_WITH_WRITE_ZEROES) {
1086 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1087 	}
1088 }
1089 
1090 static void
1091 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1092 {
1093 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1094 	struct spdk_blob		*blob = ctx->blob;
1095 
1096 	if (bserrno == 0) {
1097 		_spdk_blob_mark_clean(blob);
1098 	}
1099 
1100 	/* Call user callback */
1101 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1102 
1103 	/* Free the memory */
1104 	spdk_free(ctx->pages);
1105 	free(ctx);
1106 }
1107 
1108 static void
1109 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1110 {
1111 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1112 	struct spdk_blob		*blob = ctx->blob;
1113 	struct spdk_blob_store		*bs = blob->bs;
1114 	void				*tmp;
1115 	size_t				i;
1116 
1117 	/* Release all clusters that were truncated */
1118 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1119 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1120 
1121 		/* Nothing to release if it was not allocated */
1122 		if (blob->active.clusters[i] != 0) {
1123 			_spdk_bs_release_cluster(bs, cluster_num);
1124 		}
1125 	}
1126 
1127 	if (blob->active.num_clusters == 0) {
1128 		free(blob->active.clusters);
1129 		blob->active.clusters = NULL;
1130 		blob->active.cluster_array_size = 0;
1131 	} else if (blob->active.num_clusters != blob->active.cluster_array_size) {
1132 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters);
1133 		assert(tmp != NULL);
1134 		blob->active.clusters = tmp;
1135 		blob->active.cluster_array_size = blob->active.num_clusters;
1136 	}
1137 
1138 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1139 }
1140 
1141 static void
1142 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1143 {
1144 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1145 	struct spdk_blob		*blob = ctx->blob;
1146 	struct spdk_blob_store		*bs = blob->bs;
1147 	spdk_bs_batch_t			*batch;
1148 	size_t				i;
1149 	uint64_t			lba;
1150 	uint32_t			lba_count;
1151 
1152 	/* Clusters don't move around in blobs. The list shrinks or grows
1153 	 * at the end, but no changes ever occur in the middle of the list.
1154 	 */
1155 
1156 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx);
1157 
1158 	/* Clear all clusters that were truncated */
1159 	lba = 0;
1160 	lba_count = 0;
1161 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1162 		uint64_t next_lba = blob->active.clusters[i];
1163 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1164 
1165 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1166 			/* This cluster is contiguous with the previous one. */
1167 			lba_count += next_lba_count;
1168 			continue;
1169 		}
1170 
1171 		/* This cluster is not contiguous with the previous one. */
1172 
1173 		/* If a run of LBAs previously existing, clear them now */
1174 		if (lba_count > 0) {
1175 			spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1176 		}
1177 
1178 		/* Start building the next batch */
1179 		lba = next_lba;
1180 		if (next_lba > 0) {
1181 			lba_count = next_lba_count;
1182 		} else {
1183 			lba_count = 0;
1184 		}
1185 	}
1186 
1187 	/* If we ended with a contiguous set of LBAs, clear them now */
1188 	if (lba_count > 0) {
1189 		spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1190 	}
1191 
1192 	spdk_bs_batch_close(batch);
1193 }
1194 
1195 static void
1196 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1197 {
1198 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1199 	struct spdk_blob		*blob = ctx->blob;
1200 	struct spdk_blob_store		*bs = blob->bs;
1201 	size_t				i;
1202 
1203 	/* This loop starts at 1 because the first page is special and handled
1204 	 * below. The pages (except the first) are never written in place,
1205 	 * so any pages in the clean list must be zeroed.
1206 	 */
1207 	for (i = 1; i < blob->clean.num_pages; i++) {
1208 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1209 	}
1210 
1211 	if (blob->active.num_pages == 0) {
1212 		uint32_t page_num;
1213 
1214 		page_num = _spdk_bs_blobid_to_page(blob->id);
1215 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1216 	}
1217 
1218 	/* Move on to clearing clusters */
1219 	_spdk_blob_persist_clear_clusters(seq, ctx, 0);
1220 }
1221 
1222 static void
1223 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1224 {
1225 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1226 	struct spdk_blob		*blob = ctx->blob;
1227 	struct spdk_blob_store		*bs = blob->bs;
1228 	uint64_t			lba;
1229 	uint32_t			lba_count;
1230 	spdk_bs_batch_t			*batch;
1231 	size_t				i;
1232 
1233 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1234 
1235 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1236 
1237 	/* This loop starts at 1 because the first page is special and handled
1238 	 * below. The pages (except the first) are never written in place,
1239 	 * so any pages in the clean list must be zeroed.
1240 	 */
1241 	for (i = 1; i < blob->clean.num_pages; i++) {
1242 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);
1243 
1244 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1245 	}
1246 
1247 	/* The first page will only be zeroed if this is a delete. */
1248 	if (blob->active.num_pages == 0) {
1249 		uint32_t page_num;
1250 
1251 		/* The first page in the metadata goes where the blobid indicates */
1252 		page_num = _spdk_bs_blobid_to_page(blob->id);
1253 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);
1254 
1255 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1256 	}
1257 
1258 	spdk_bs_batch_close(batch);
1259 }
1260 
1261 static void
1262 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1263 {
1264 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1265 	struct spdk_blob		*blob = ctx->blob;
1266 	struct spdk_blob_store		*bs = blob->bs;
1267 	uint64_t			lba;
1268 	uint32_t			lba_count;
1269 	struct spdk_blob_md_page	*page;
1270 
1271 	if (blob->active.num_pages == 0) {
1272 		/* Move on to the next step */
1273 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1274 		return;
1275 	}
1276 
1277 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1278 
1279 	page = &ctx->pages[0];
1280 	/* The first page in the metadata goes where the blobid indicates */
1281 	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));
1282 
1283 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1284 				   _spdk_blob_persist_zero_pages, ctx);
1285 }
1286 
1287 static void
1288 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1289 {
1290 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1291 	struct spdk_blob		*blob = ctx->blob;
1292 	struct spdk_blob_store		*bs = blob->bs;
1293 	uint64_t			lba;
1294 	uint32_t			lba_count;
1295 	struct spdk_blob_md_page	*page;
1296 	spdk_bs_batch_t			*batch;
1297 	size_t				i;
1298 
1299 	/* Clusters don't move around in blobs. The list shrinks or grows
1300 	 * at the end, but no changes ever occur in the middle of the list.
1301 	 */
1302 
1303 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1304 
1305 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1306 
1307 	/* This starts at 1. The root page is not written until
1308 	 * all of the others are finished
1309 	 */
1310 	for (i = 1; i < blob->active.num_pages; i++) {
1311 		page = &ctx->pages[i];
1312 		assert(page->sequence_num == i);
1313 
1314 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]);
1315 
1316 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1317 	}
1318 
1319 	spdk_bs_batch_close(batch);
1320 }
1321 
1322 static int
1323 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1324 {
1325 	uint64_t	i;
1326 	uint64_t	*tmp;
1327 	uint64_t	lfc; /* lowest free cluster */
1328 	uint64_t	num_clusters;
1329 	struct spdk_blob_store *bs;
1330 
1331 	bs = blob->bs;
1332 
1333 	_spdk_blob_verify_md_op(blob);
1334 
1335 	if (blob->active.num_clusters == sz) {
1336 		return 0;
1337 	}
1338 
1339 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1340 		/* If this blob was resized to be larger, then smaller, then
1341 		 * larger without syncing, then the cluster array already
1342 		 * contains spare assigned clusters we can use.
1343 		 */
1344 		num_clusters = spdk_min(blob->active.cluster_array_size,
1345 					sz);
1346 	} else {
1347 		num_clusters = blob->active.num_clusters;
1348 	}
1349 
1350 	/* Do two passes - one to verify that we can obtain enough clusters
1351 	 * and another to actually claim them.
1352 	 */
1353 
1354 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1355 		lfc = 0;
1356 		for (i = num_clusters; i < sz; i++) {
1357 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1358 			if (lfc == UINT32_MAX) {
1359 				/* No more free clusters. Cannot satisfy the request */
1360 				return -ENOSPC;
1361 			}
1362 			lfc++;
1363 		}
1364 	}
1365 
1366 	if (sz > num_clusters) {
1367 		/* Expand the cluster array if necessary.
1368 		 * We only shrink the array when persisting.
1369 		 */
1370 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz);
1371 		if (sz > 0 && tmp == NULL) {
1372 			return -ENOMEM;
1373 		}
1374 		memset(tmp + blob->active.cluster_array_size, 0,
1375 		       sizeof(uint64_t) * (sz - blob->active.cluster_array_size));
1376 		blob->active.clusters = tmp;
1377 		blob->active.cluster_array_size = sz;
1378 	}
1379 
1380 	blob->state = SPDK_BLOB_STATE_DIRTY;
1381 
1382 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1383 		lfc = 0;
1384 		for (i = num_clusters; i < sz; i++) {
1385 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1386 			lfc++;
1387 		}
1388 	}
1389 
1390 	blob->active.num_clusters = sz;
1391 
1392 	return 0;
1393 }
1394 
1395 static void
1396 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1397 {
1398 	spdk_bs_sequence_t *seq = ctx->seq;
1399 	struct spdk_blob *blob = ctx->blob;
1400 	struct spdk_blob_store *bs = blob->bs;
1401 	uint64_t i;
1402 	uint32_t page_num;
1403 	void *tmp;
1404 	int rc;
1405 
1406 	if (blob->active.num_pages == 0) {
1407 		/* This is the signal that the blob should be deleted.
1408 		 * Immediately jump to the clean up routine. */
1409 		assert(blob->clean.num_pages > 0);
1410 		ctx->idx = blob->clean.num_pages - 1;
1411 		blob->state = SPDK_BLOB_STATE_CLEAN;
1412 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1413 		return;
1414 
1415 	}
1416 
1417 	/* Generate the new metadata */
1418 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1419 	if (rc < 0) {
1420 		_spdk_blob_persist_complete(seq, ctx, rc);
1421 		return;
1422 	}
1423 
1424 	assert(blob->active.num_pages >= 1);
1425 
1426 	/* Resize the cache of page indices */
1427 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1428 	if (!tmp) {
1429 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1430 		return;
1431 	}
1432 	blob->active.pages = tmp;
1433 
1434 	/* Assign this metadata to pages. This requires two passes -
1435 	 * one to verify that there are enough pages and a second
1436 	 * to actually claim them. */
1437 	page_num = 0;
1438 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1439 	for (i = 1; i < blob->active.num_pages; i++) {
1440 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1441 		if (page_num == UINT32_MAX) {
1442 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1443 			return;
1444 		}
1445 		page_num++;
1446 	}
1447 
1448 	page_num = 0;
1449 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1450 	for (i = 1; i < blob->active.num_pages; i++) {
1451 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1452 		ctx->pages[i - 1].next = page_num;
1453 		/* Now that previous metadata page is complete, calculate the crc for it. */
1454 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1455 		blob->active.pages[i] = page_num;
1456 		spdk_bit_array_set(bs->used_md_pages, page_num);
1457 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1458 		page_num++;
1459 	}
1460 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1461 	/* Start writing the metadata from last page to first */
1462 	ctx->idx = blob->active.num_pages - 1;
1463 	blob->state = SPDK_BLOB_STATE_CLEAN;
1464 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1465 }
1466 
1467 static void
1468 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1469 {
1470 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1471 
1472 	ctx->blob->bs->clean = 0;
1473 
1474 	spdk_free(ctx->super);
1475 
1476 	_spdk_blob_persist_start(ctx);
1477 }
1478 
1479 static void
1480 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1481 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1482 
1483 
1484 static void
1485 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1486 {
1487 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1488 
1489 	ctx->super->clean = 0;
1490 	if (ctx->super->size == 0) {
1491 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1492 	}
1493 
1494 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1495 }
1496 
1497 
1498 /* Write a blob to disk */
1499 static void
1500 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1501 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1502 {
1503 	struct spdk_blob_persist_ctx *ctx;
1504 
1505 	_spdk_blob_verify_md_op(blob);
1506 
1507 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1508 		cb_fn(seq, cb_arg, 0);
1509 		return;
1510 	}
1511 
1512 	ctx = calloc(1, sizeof(*ctx));
1513 	if (!ctx) {
1514 		cb_fn(seq, cb_arg, -ENOMEM);
1515 		return;
1516 	}
1517 	ctx->blob = blob;
1518 	ctx->seq = seq;
1519 	ctx->cb_fn = cb_fn;
1520 	ctx->cb_arg = cb_arg;
1521 
1522 	if (blob->bs->clean) {
1523 		ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
1524 					  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1525 		if (!ctx->super) {
1526 			cb_fn(seq, cb_arg, -ENOMEM);
1527 			free(ctx);
1528 			return;
1529 		}
1530 
1531 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1532 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1533 					  _spdk_blob_persist_dirty, ctx);
1534 	} else {
1535 		_spdk_blob_persist_start(ctx);
1536 	}
1537 }
1538 
1539 struct spdk_blob_copy_cluster_ctx {
1540 	struct spdk_blob *blob;
1541 	uint8_t *buf;
1542 	uint64_t page;
1543 	uint64_t new_cluster;
1544 	spdk_bs_sequence_t *seq;
1545 };
1546 
1547 static void
1548 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1549 {
1550 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1551 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1552 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1553 	spdk_bs_user_op_t *op;
1554 
1555 	TAILQ_INIT(&requests);
1556 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1557 
1558 	while (!TAILQ_EMPTY(&requests)) {
1559 		op = TAILQ_FIRST(&requests);
1560 		TAILQ_REMOVE(&requests, op, link);
1561 		if (bserrno == 0) {
1562 			spdk_bs_user_op_execute(op);
1563 		} else {
1564 			spdk_bs_user_op_abort(op);
1565 		}
1566 	}
1567 
1568 	spdk_free(ctx->buf);
1569 	free(ctx);
1570 }
1571 
1572 static void
1573 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1574 {
1575 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1576 
1577 	if (bserrno) {
1578 		uint32_t cluster_number;
1579 
1580 		if (bserrno == -EEXIST) {
1581 			/* The metadata insert failed because another thread
1582 			 * allocated the cluster first. Free our cluster
1583 			 * but continue without error. */
1584 			bserrno = 0;
1585 		}
1586 
1587 		cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1588 		_spdk_bs_release_cluster(ctx->blob->bs, cluster_number);
1589 	}
1590 
1591 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1592 }
1593 
1594 static void
1595 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1596 {
1597 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1598 	uint32_t cluster_number;
1599 
1600 	if (bserrno) {
1601 		/* The write failed, so jump to the final completion handler */
1602 		spdk_bs_sequence_finish(seq, bserrno);
1603 		return;
1604 	}
1605 
1606 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1607 
1608 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1609 					       _spdk_blob_insert_cluster_cpl, ctx);
1610 }
1611 
1612 static void
1613 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1614 {
1615 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1616 
1617 	if (bserrno != 0) {
1618 		/* The read failed, so jump to the final completion handler */
1619 		spdk_bs_sequence_finish(seq, bserrno);
1620 		return;
1621 	}
1622 
1623 	/* Write whole cluster */
1624 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1625 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1626 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1627 				   _spdk_blob_write_copy_cpl, ctx);
1628 }
1629 
1630 static void
1631 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1632 				   struct spdk_io_channel *_ch,
1633 				   uint64_t io_unit, spdk_bs_user_op_t *op)
1634 {
1635 	struct spdk_bs_cpl cpl;
1636 	struct spdk_bs_channel *ch;
1637 	struct spdk_blob_copy_cluster_ctx *ctx;
1638 	uint32_t cluster_start_page;
1639 	uint32_t cluster_number;
1640 	int rc;
1641 
1642 	ch = spdk_io_channel_get_ctx(_ch);
1643 
1644 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1645 		/* There are already operations pending. Queue this user op
1646 		 * and return because it will be re-executed when the outstanding
1647 		 * cluster allocation completes. */
1648 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1649 		return;
1650 	}
1651 
1652 	/* Round the io_unit offset down to the first page in the cluster */
1653 	cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit);
1654 
1655 	/* Calculate which index in the metadata cluster array the corresponding
1656 	 * cluster is supposed to be at. */
1657 	cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit);
1658 
1659 	ctx = calloc(1, sizeof(*ctx));
1660 	if (!ctx) {
1661 		spdk_bs_user_op_abort(op);
1662 		return;
1663 	}
1664 
1665 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1666 
1667 	ctx->blob = blob;
1668 	ctx->page = cluster_start_page;
1669 
1670 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1671 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
1672 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1673 		if (!ctx->buf) {
1674 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1675 				    blob->bs->cluster_sz);
1676 			free(ctx);
1677 			spdk_bs_user_op_abort(op);
1678 			return;
1679 		}
1680 	}
1681 
1682 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1683 	if (rc != 0) {
1684 		spdk_free(ctx->buf);
1685 		free(ctx);
1686 		spdk_bs_user_op_abort(op);
1687 		return;
1688 	}
1689 
1690 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1691 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1692 	cpl.u.blob_basic.cb_arg = ctx;
1693 
1694 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1695 	if (!ctx->seq) {
1696 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1697 		spdk_free(ctx->buf);
1698 		free(ctx);
1699 		spdk_bs_user_op_abort(op);
1700 		return;
1701 	}
1702 
1703 	/* Queue the user op to block other incoming operations */
1704 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1705 
1706 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1707 		/* Read cluster from backing device */
1708 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1709 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1710 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1711 					     _spdk_blob_write_copy, ctx);
1712 	} else {
1713 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1714 						       _spdk_blob_insert_cluster_cpl, ctx);
1715 	}
1716 }
1717 
1718 static void
1719 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
1720 				       uint64_t *lba,	uint32_t *lba_count)
1721 {
1722 	*lba_count = length;
1723 
1724 	if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) {
1725 		assert(blob->back_bs_dev != NULL);
1726 		*lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit);
1727 		*lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count);
1728 	} else {
1729 		*lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit);
1730 	}
1731 }
1732 
1733 struct op_split_ctx {
1734 	struct spdk_blob *blob;
1735 	struct spdk_io_channel *channel;
1736 	uint64_t io_unit_offset;
1737 	uint64_t io_units_remaining;
1738 	void *curr_payload;
1739 	enum spdk_blob_op_type op_type;
1740 	spdk_bs_sequence_t *seq;
1741 };
1742 
1743 static void
1744 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1745 {
1746 	struct op_split_ctx	*ctx = cb_arg;
1747 	struct spdk_blob	*blob = ctx->blob;
1748 	struct spdk_io_channel	*ch = ctx->channel;
1749 	enum spdk_blob_op_type	op_type = ctx->op_type;
1750 	uint8_t			*buf = ctx->curr_payload;
1751 	uint64_t		offset = ctx->io_unit_offset;
1752 	uint64_t		length = ctx->io_units_remaining;
1753 	uint64_t		op_length;
1754 
1755 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
1756 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1757 		free(ctx);
1758 		return;
1759 	}
1760 
1761 	op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob,
1762 			     offset));
1763 
1764 	/* Update length and payload for next operation */
1765 	ctx->io_units_remaining -= op_length;
1766 	ctx->io_unit_offset += op_length;
1767 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1768 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
1769 	}
1770 
1771 	switch (op_type) {
1772 	case SPDK_BLOB_READ:
1773 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1774 				  _spdk_blob_request_submit_op_split_next, ctx);
1775 		break;
1776 	case SPDK_BLOB_WRITE:
1777 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1778 				   _spdk_blob_request_submit_op_split_next, ctx);
1779 		break;
1780 	case SPDK_BLOB_UNMAP:
1781 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1782 				   _spdk_blob_request_submit_op_split_next, ctx);
1783 		break;
1784 	case SPDK_BLOB_WRITE_ZEROES:
1785 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1786 					  _spdk_blob_request_submit_op_split_next, ctx);
1787 		break;
1788 	case SPDK_BLOB_READV:
1789 	case SPDK_BLOB_WRITEV:
1790 		SPDK_ERRLOG("readv/write not valid for %s\n", __func__);
1791 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1792 		free(ctx);
1793 		break;
1794 	}
1795 }
1796 
1797 static void
1798 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1799 				   void *payload, uint64_t offset, uint64_t length,
1800 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1801 {
1802 	struct op_split_ctx *ctx;
1803 	spdk_bs_sequence_t *seq;
1804 	struct spdk_bs_cpl cpl;
1805 
1806 	assert(blob != NULL);
1807 
1808 	ctx = calloc(1, sizeof(struct op_split_ctx));
1809 	if (ctx == NULL) {
1810 		cb_fn(cb_arg, -ENOMEM);
1811 		return;
1812 	}
1813 
1814 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1815 	cpl.u.blob_basic.cb_fn = cb_fn;
1816 	cpl.u.blob_basic.cb_arg = cb_arg;
1817 
1818 	seq = spdk_bs_sequence_start(ch, &cpl);
1819 	if (!seq) {
1820 		free(ctx);
1821 		cb_fn(cb_arg, -ENOMEM);
1822 		return;
1823 	}
1824 
1825 	ctx->blob = blob;
1826 	ctx->channel = ch;
1827 	ctx->curr_payload = payload;
1828 	ctx->io_unit_offset = offset;
1829 	ctx->io_units_remaining = length;
1830 	ctx->op_type = op_type;
1831 	ctx->seq = seq;
1832 
1833 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1834 }
1835 
1836 static void
1837 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1838 				    void *payload, uint64_t offset, uint64_t length,
1839 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1840 {
1841 	struct spdk_bs_cpl cpl;
1842 	uint64_t lba;
1843 	uint32_t lba_count;
1844 
1845 	assert(blob != NULL);
1846 
1847 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1848 	cpl.u.blob_basic.cb_fn = cb_fn;
1849 	cpl.u.blob_basic.cb_arg = cb_arg;
1850 
1851 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1852 
1853 	if (blob->frozen_refcnt) {
1854 		/* This blob I/O is frozen */
1855 		spdk_bs_user_op_t *op;
1856 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1857 
1858 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1859 		if (!op) {
1860 			cb_fn(cb_arg, -ENOMEM);
1861 			return;
1862 		}
1863 
1864 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1865 
1866 		return;
1867 	}
1868 
1869 	switch (op_type) {
1870 	case SPDK_BLOB_READ: {
1871 		spdk_bs_batch_t *batch;
1872 
1873 		batch = spdk_bs_batch_open(_ch, &cpl);
1874 		if (!batch) {
1875 			cb_fn(cb_arg, -ENOMEM);
1876 			return;
1877 		}
1878 
1879 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1880 			/* Read from the blob */
1881 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1882 		} else {
1883 			/* Read from the backing block device */
1884 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1885 		}
1886 
1887 		spdk_bs_batch_close(batch);
1888 		break;
1889 	}
1890 	case SPDK_BLOB_WRITE:
1891 	case SPDK_BLOB_WRITE_ZEROES: {
1892 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1893 			/* Write to the blob */
1894 			spdk_bs_batch_t *batch;
1895 
1896 			if (lba_count == 0) {
1897 				cb_fn(cb_arg, 0);
1898 				return;
1899 			}
1900 
1901 			batch = spdk_bs_batch_open(_ch, &cpl);
1902 			if (!batch) {
1903 				cb_fn(cb_arg, -ENOMEM);
1904 				return;
1905 			}
1906 
1907 			if (op_type == SPDK_BLOB_WRITE) {
1908 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1909 			} else {
1910 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1911 			}
1912 
1913 			spdk_bs_batch_close(batch);
1914 		} else {
1915 			/* Queue this operation and allocate the cluster */
1916 			spdk_bs_user_op_t *op;
1917 
1918 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1919 			if (!op) {
1920 				cb_fn(cb_arg, -ENOMEM);
1921 				return;
1922 			}
1923 
1924 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1925 		}
1926 		break;
1927 	}
1928 	case SPDK_BLOB_UNMAP: {
1929 		spdk_bs_batch_t *batch;
1930 
1931 		batch = spdk_bs_batch_open(_ch, &cpl);
1932 		if (!batch) {
1933 			cb_fn(cb_arg, -ENOMEM);
1934 			return;
1935 		}
1936 
1937 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1938 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1939 		}
1940 
1941 		spdk_bs_batch_close(batch);
1942 		break;
1943 	}
1944 	case SPDK_BLOB_READV:
1945 	case SPDK_BLOB_WRITEV:
1946 		SPDK_ERRLOG("readv/write not valid\n");
1947 		cb_fn(cb_arg, -EINVAL);
1948 		break;
1949 	}
1950 }
1951 
1952 static void
1953 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1954 			     void *payload, uint64_t offset, uint64_t length,
1955 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1956 {
1957 	assert(blob != NULL);
1958 
1959 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1960 		cb_fn(cb_arg, -EPERM);
1961 		return;
1962 	}
1963 
1964 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
1965 		cb_fn(cb_arg, -EINVAL);
1966 		return;
1967 	}
1968 	if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) {
1969 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1970 						    cb_fn, cb_arg, op_type);
1971 	} else {
1972 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1973 						   cb_fn, cb_arg, op_type);
1974 	}
1975 }
1976 
1977 struct rw_iov_ctx {
1978 	struct spdk_blob *blob;
1979 	struct spdk_io_channel *channel;
1980 	spdk_blob_op_complete cb_fn;
1981 	void *cb_arg;
1982 	bool read;
1983 	int iovcnt;
1984 	struct iovec *orig_iov;
1985 	uint64_t io_unit_offset;
1986 	uint64_t io_units_remaining;
1987 	uint64_t io_units_done;
1988 	struct iovec iov[0];
1989 };
1990 
1991 static void
1992 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1993 {
1994 	assert(cb_arg == NULL);
1995 	spdk_bs_sequence_finish(seq, bserrno);
1996 }
1997 
1998 static void
1999 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
2000 {
2001 	struct rw_iov_ctx *ctx = cb_arg;
2002 	struct spdk_blob *blob = ctx->blob;
2003 	struct iovec *iov, *orig_iov;
2004 	int iovcnt;
2005 	size_t orig_iovoff;
2006 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2007 	uint64_t byte_count;
2008 
2009 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2010 		ctx->cb_fn(ctx->cb_arg, bserrno);
2011 		free(ctx);
2012 		return;
2013 	}
2014 
2015 	io_unit_offset = ctx->io_unit_offset;
2016 	io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2017 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2018 	/*
2019 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2020 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2021 	 *  point to the current position in the I/O sequence.
2022 	 */
2023 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2024 	orig_iov = &ctx->orig_iov[0];
2025 	orig_iovoff = 0;
2026 	while (byte_count > 0) {
2027 		if (byte_count >= orig_iov->iov_len) {
2028 			byte_count -= orig_iov->iov_len;
2029 			orig_iov++;
2030 		} else {
2031 			orig_iovoff = byte_count;
2032 			byte_count = 0;
2033 		}
2034 	}
2035 
2036 	/*
2037 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2038 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2039 	 */
2040 	byte_count = io_units_count * blob->bs->io_unit_size;
2041 	iov = &ctx->iov[0];
2042 	iovcnt = 0;
2043 	while (byte_count > 0) {
2044 		assert(iovcnt < ctx->iovcnt);
2045 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2046 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2047 		byte_count -= iov->iov_len;
2048 		orig_iovoff = 0;
2049 		orig_iov++;
2050 		iov++;
2051 		iovcnt++;
2052 	}
2053 
2054 	ctx->io_unit_offset += io_units_count;
2055 	ctx->io_units_remaining -= io_units_count;
2056 	ctx->io_units_done += io_units_count;
2057 	iov = &ctx->iov[0];
2058 
2059 	if (ctx->read) {
2060 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2061 				   io_units_count, _spdk_rw_iov_split_next, ctx);
2062 	} else {
2063 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2064 				    io_units_count, _spdk_rw_iov_split_next, ctx);
2065 	}
2066 }
2067 
2068 static void
2069 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2070 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2071 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2072 {
2073 	struct spdk_bs_cpl	cpl;
2074 
2075 	assert(blob != NULL);
2076 
2077 	if (!read && blob->data_ro) {
2078 		cb_fn(cb_arg, -EPERM);
2079 		return;
2080 	}
2081 
2082 	if (length == 0) {
2083 		cb_fn(cb_arg, 0);
2084 		return;
2085 	}
2086 
2087 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2088 		cb_fn(cb_arg, -EINVAL);
2089 		return;
2090 	}
2091 
2092 	/*
2093 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2094 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2095 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2096 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2097 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2098 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2099 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2100 	 *
2101 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2102 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2103 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2104 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2105 	 */
2106 	if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) {
2107 		uint32_t lba_count;
2108 		uint64_t lba;
2109 
2110 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2111 
2112 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2113 		cpl.u.blob_basic.cb_fn = cb_fn;
2114 		cpl.u.blob_basic.cb_arg = cb_arg;
2115 		if (blob->frozen_refcnt) {
2116 			/* This blob I/O is frozen */
2117 			spdk_bs_user_op_t *op;
2118 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2119 
2120 			op = spdk_bs_user_op_alloc(_channel, &cpl, read, blob, iov, iovcnt, offset, length);
2121 			if (!op) {
2122 				cb_fn(cb_arg, -ENOMEM);
2123 				return;
2124 			}
2125 
2126 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2127 
2128 			return;
2129 		}
2130 
2131 		if (read) {
2132 			spdk_bs_sequence_t *seq;
2133 
2134 			seq = spdk_bs_sequence_start(_channel, &cpl);
2135 			if (!seq) {
2136 				cb_fn(cb_arg, -ENOMEM);
2137 				return;
2138 			}
2139 
2140 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2141 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2142 			} else {
2143 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2144 							      _spdk_rw_iov_done, NULL);
2145 			}
2146 		} else {
2147 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2148 				spdk_bs_sequence_t *seq;
2149 
2150 				seq = spdk_bs_sequence_start(_channel, &cpl);
2151 				if (!seq) {
2152 					cb_fn(cb_arg, -ENOMEM);
2153 					return;
2154 				}
2155 
2156 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2157 			} else {
2158 				/* Queue this operation and allocate the cluster */
2159 				spdk_bs_user_op_t *op;
2160 
2161 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2162 							   length);
2163 				if (!op) {
2164 					cb_fn(cb_arg, -ENOMEM);
2165 					return;
2166 				}
2167 
2168 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2169 			}
2170 		}
2171 	} else {
2172 		struct rw_iov_ctx *ctx;
2173 
2174 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2175 		if (ctx == NULL) {
2176 			cb_fn(cb_arg, -ENOMEM);
2177 			return;
2178 		}
2179 
2180 		ctx->blob = blob;
2181 		ctx->channel = _channel;
2182 		ctx->cb_fn = cb_fn;
2183 		ctx->cb_arg = cb_arg;
2184 		ctx->read = read;
2185 		ctx->orig_iov = iov;
2186 		ctx->iovcnt = iovcnt;
2187 		ctx->io_unit_offset = offset;
2188 		ctx->io_units_remaining = length;
2189 		ctx->io_units_done = 0;
2190 
2191 		_spdk_rw_iov_split_next(ctx, 0);
2192 	}
2193 }
2194 
2195 static struct spdk_blob *
2196 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2197 {
2198 	struct spdk_blob *blob;
2199 
2200 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2201 		if (blob->id == blobid) {
2202 			return blob;
2203 		}
2204 	}
2205 
2206 	return NULL;
2207 }
2208 
2209 static void
2210 _spdk_blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
2211 		struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
2212 {
2213 	assert(blob != NULL);
2214 	*snapshot_entry = NULL;
2215 	*clone_entry = NULL;
2216 
2217 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
2218 		return;
2219 	}
2220 
2221 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
2222 		if ((*snapshot_entry)->id == blob->parent_id) {
2223 			break;
2224 		}
2225 	}
2226 
2227 	if (*snapshot_entry != NULL) {
2228 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
2229 			if ((*clone_entry)->id == blob->id) {
2230 				break;
2231 			}
2232 		}
2233 
2234 		assert(clone_entry != NULL);
2235 	}
2236 }
2237 
2238 static int
2239 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2240 {
2241 	struct spdk_blob_store		*bs = io_device;
2242 	struct spdk_bs_channel		*channel = ctx_buf;
2243 	struct spdk_bs_dev		*dev;
2244 	uint32_t			max_ops = bs->max_channel_ops;
2245 	uint32_t			i;
2246 
2247 	dev = bs->dev;
2248 
2249 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2250 	if (!channel->req_mem) {
2251 		return -1;
2252 	}
2253 
2254 	TAILQ_INIT(&channel->reqs);
2255 
2256 	for (i = 0; i < max_ops; i++) {
2257 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2258 	}
2259 
2260 	channel->bs = bs;
2261 	channel->dev = dev;
2262 	channel->dev_channel = dev->create_channel(dev);
2263 
2264 	if (!channel->dev_channel) {
2265 		SPDK_ERRLOG("Failed to create device channel.\n");
2266 		free(channel->req_mem);
2267 		return -1;
2268 	}
2269 
2270 	TAILQ_INIT(&channel->need_cluster_alloc);
2271 	TAILQ_INIT(&channel->queued_io);
2272 
2273 	return 0;
2274 }
2275 
2276 static void
2277 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2278 {
2279 	struct spdk_bs_channel *channel = ctx_buf;
2280 	spdk_bs_user_op_t *op;
2281 
2282 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2283 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2284 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2285 		spdk_bs_user_op_abort(op);
2286 	}
2287 
2288 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2289 		op = TAILQ_FIRST(&channel->queued_io);
2290 		TAILQ_REMOVE(&channel->queued_io, op, link);
2291 		spdk_bs_user_op_abort(op);
2292 	}
2293 
2294 	free(channel->req_mem);
2295 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2296 }
2297 
2298 static void
2299 _spdk_bs_dev_destroy(void *io_device)
2300 {
2301 	struct spdk_blob_store *bs = io_device;
2302 	struct spdk_blob	*blob, *blob_tmp;
2303 
2304 	bs->dev->destroy(bs->dev);
2305 
2306 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2307 		TAILQ_REMOVE(&bs->blobs, blob, link);
2308 		_spdk_blob_free(blob);
2309 	}
2310 
2311 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2312 
2313 	spdk_bit_array_free(&bs->used_blobids);
2314 	spdk_bit_array_free(&bs->used_md_pages);
2315 	spdk_bit_array_free(&bs->used_clusters);
2316 	/*
2317 	 * If this function is called for any reason except a successful unload,
2318 	 * the unload_cpl type will be NONE and this will be a nop.
2319 	 */
2320 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2321 
2322 	free(bs);
2323 }
2324 
2325 static int
2326 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2327 {
2328 	spdk_blob_id snapshot_id;
2329 	struct spdk_blob_list *snapshot_entry = NULL;
2330 	struct spdk_blob_list *clone_entry = NULL;
2331 
2332 	assert(blob != NULL);
2333 
2334 	snapshot_id = blob->parent_id;
2335 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2336 		return 0;
2337 	}
2338 
2339 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2340 		if (snapshot_entry->id == snapshot_id) {
2341 			break;
2342 		}
2343 	}
2344 
2345 	if (snapshot_entry == NULL) {
2346 		/* Snapshot not found */
2347 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2348 		if (snapshot_entry == NULL) {
2349 			return -ENOMEM;
2350 		}
2351 		snapshot_entry->id = snapshot_id;
2352 		TAILQ_INIT(&snapshot_entry->clones);
2353 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2354 	} else {
2355 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2356 			if (clone_entry->id == blob->id) {
2357 				break;
2358 			}
2359 		}
2360 	}
2361 
2362 	if (clone_entry == NULL) {
2363 		/* Clone not found */
2364 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2365 		if (clone_entry == NULL) {
2366 			return -ENOMEM;
2367 		}
2368 		clone_entry->id = blob->id;
2369 		TAILQ_INIT(&clone_entry->clones);
2370 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2371 		snapshot_entry->clone_count++;
2372 	}
2373 
2374 	return 0;
2375 }
2376 
2377 static void
2378 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2379 {
2380 	struct spdk_blob_list *snapshot_entry = NULL;
2381 	struct spdk_blob_list *clone_entry = NULL;
2382 
2383 	_spdk_blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
2384 
2385 	if (snapshot_entry == NULL) {
2386 		return;
2387 	}
2388 
2389 	blob->parent_id = SPDK_BLOBID_INVALID;
2390 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2391 	free(clone_entry);
2392 
2393 	snapshot_entry->clone_count--;
2394 }
2395 
2396 static int
2397 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2398 {
2399 	struct spdk_blob_list *snapshot_entry;
2400 	struct spdk_blob_list *snapshot_entry_tmp;
2401 	struct spdk_blob_list *clone_entry;
2402 	struct spdk_blob_list *clone_entry_tmp;
2403 
2404 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2405 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2406 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2407 			free(clone_entry);
2408 		}
2409 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2410 		free(snapshot_entry);
2411 	}
2412 
2413 	return 0;
2414 }
2415 
2416 static void
2417 _spdk_bs_free(struct spdk_blob_store *bs)
2418 {
2419 	_spdk_bs_blob_list_free(bs);
2420 
2421 	spdk_bs_unregister_md_thread(bs);
2422 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2423 }
2424 
2425 void
2426 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2427 {
2428 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2429 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2430 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2431 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2432 	opts->clear_method = BS_CLEAR_WITH_UNMAP;
2433 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2434 	opts->iter_cb_fn = NULL;
2435 	opts->iter_cb_arg = NULL;
2436 }
2437 
2438 static int
2439 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2440 {
2441 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2442 	    opts->max_channel_ops == 0) {
2443 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2444 		return -1;
2445 	}
2446 
2447 	return 0;
2448 }
2449 
2450 static int
2451 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2452 {
2453 	struct spdk_blob_store	*bs;
2454 	uint64_t dev_size;
2455 	int rc;
2456 
2457 	dev_size = dev->blocklen * dev->blockcnt;
2458 	if (dev_size < opts->cluster_sz) {
2459 		/* Device size cannot be smaller than cluster size of blobstore */
2460 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2461 			     dev_size, opts->cluster_sz);
2462 		return -ENOSPC;
2463 	}
2464 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2465 		/* Cluster size cannot be smaller than page size */
2466 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2467 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2468 		return -EINVAL;
2469 	}
2470 	bs = calloc(1, sizeof(struct spdk_blob_store));
2471 	if (!bs) {
2472 		return -ENOMEM;
2473 	}
2474 
2475 	TAILQ_INIT(&bs->blobs);
2476 	TAILQ_INIT(&bs->snapshots);
2477 	bs->dev = dev;
2478 	bs->md_thread = spdk_get_thread();
2479 	assert(bs->md_thread != NULL);
2480 
2481 	/*
2482 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2483 	 *  even multiple of the cluster size.
2484 	 */
2485 	bs->cluster_sz = opts->cluster_sz;
2486 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2487 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2488 	bs->num_free_clusters = bs->total_clusters;
2489 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2490 	bs->io_unit_size = dev->blocklen;
2491 	if (bs->used_clusters == NULL) {
2492 		free(bs);
2493 		return -ENOMEM;
2494 	}
2495 
2496 	bs->max_channel_ops = opts->max_channel_ops;
2497 	bs->super_blob = SPDK_BLOBID_INVALID;
2498 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2499 
2500 	/* The metadata is assumed to be at least 1 page */
2501 	bs->used_md_pages = spdk_bit_array_create(1);
2502 	bs->used_blobids = spdk_bit_array_create(0);
2503 
2504 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2505 
2506 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2507 				sizeof(struct spdk_bs_channel), "blobstore");
2508 	rc = spdk_bs_register_md_thread(bs);
2509 	if (rc == -1) {
2510 		spdk_io_device_unregister(bs, NULL);
2511 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2512 		spdk_bit_array_free(&bs->used_blobids);
2513 		spdk_bit_array_free(&bs->used_md_pages);
2514 		spdk_bit_array_free(&bs->used_clusters);
2515 		free(bs);
2516 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2517 		return -ENOMEM;
2518 	}
2519 
2520 	*_bs = bs;
2521 	return 0;
2522 }
2523 
2524 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2525 
2526 struct spdk_bs_load_ctx {
2527 	struct spdk_blob_store		*bs;
2528 	struct spdk_bs_super_block	*super;
2529 
2530 	struct spdk_bs_md_mask		*mask;
2531 	bool				in_page_chain;
2532 	uint32_t			page_index;
2533 	uint32_t			cur_page;
2534 	struct spdk_blob_md_page	*page;
2535 
2536 	spdk_bs_sequence_t			*seq;
2537 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2538 	void					*iter_cb_arg;
2539 };
2540 
2541 static void
2542 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2543 {
2544 	assert(bserrno != 0);
2545 
2546 	spdk_free(ctx->super);
2547 	spdk_bs_sequence_finish(seq, bserrno);
2548 	_spdk_bs_free(ctx->bs);
2549 	free(ctx);
2550 }
2551 
2552 static void
2553 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2554 {
2555 	uint32_t i = 0;
2556 
2557 	while (true) {
2558 		i = spdk_bit_array_find_first_set(array, i);
2559 		if (i >= mask->length) {
2560 			break;
2561 		}
2562 		mask->mask[i / 8] |= 1U << (i % 8);
2563 		i++;
2564 	}
2565 }
2566 
2567 static int
2568 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2569 {
2570 	struct spdk_bit_array *array;
2571 	uint32_t i;
2572 
2573 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2574 		return -ENOMEM;
2575 	}
2576 
2577 	array = *array_ptr;
2578 	for (i = 0; i < mask->length; i++) {
2579 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2580 			spdk_bit_array_set(array, i);
2581 		}
2582 	}
2583 
2584 	return 0;
2585 }
2586 
2587 static void
2588 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2589 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2590 {
2591 	/* Update the values in the super block */
2592 	super->super_blob = bs->super_blob;
2593 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2594 	super->crc = _spdk_blob_md_page_calc_crc(super);
2595 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2596 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2597 				   cb_fn, cb_arg);
2598 }
2599 
2600 static void
2601 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2602 {
2603 	struct spdk_bs_load_ctx	*ctx = arg;
2604 	uint64_t	mask_size, lba, lba_count;
2605 
2606 	/* Write out the used clusters mask */
2607 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2608 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2609 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2610 	if (!ctx->mask) {
2611 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2612 		return;
2613 	}
2614 
2615 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2616 	ctx->mask->length = ctx->bs->total_clusters;
2617 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2618 
2619 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2620 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2621 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2622 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2623 }
2624 
2625 static void
2626 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2627 {
2628 	struct spdk_bs_load_ctx	*ctx = arg;
2629 	uint64_t	mask_size, lba, lba_count;
2630 
2631 	if (seq->bserrno) {
2632 		_spdk_bs_load_ctx_fail(seq, ctx, seq->bserrno);
2633 		return;
2634 	}
2635 
2636 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2637 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2638 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2639 	if (!ctx->mask) {
2640 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2641 		return;
2642 	}
2643 
2644 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2645 	ctx->mask->length = ctx->super->md_len;
2646 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2647 
2648 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2649 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2650 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2651 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2652 }
2653 
2654 static void
2655 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2656 {
2657 	struct spdk_bs_load_ctx	*ctx = arg;
2658 	uint64_t	mask_size, lba, lba_count;
2659 
2660 	if (ctx->super->used_blobid_mask_len == 0) {
2661 		/*
2662 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2663 		 *  written to disk.
2664 		 */
2665 		cb_fn(seq, arg, 0);
2666 		return;
2667 	}
2668 
2669 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2670 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2671 				 SPDK_MALLOC_DMA);
2672 	if (!ctx->mask) {
2673 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2674 		return;
2675 	}
2676 
2677 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2678 	ctx->mask->length = ctx->super->md_len;
2679 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2680 
2681 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2682 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2683 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2684 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2685 }
2686 
2687 static void
2688 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2689 {
2690 	struct spdk_bs_load_ctx *ctx = arg;
2691 
2692 	if (bserrno == 0) {
2693 		if (ctx->iter_cb_fn) {
2694 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2695 		}
2696 		_spdk_bs_blob_list_add(blob);
2697 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2698 		return;
2699 	}
2700 
2701 	if (bserrno == -ENOENT) {
2702 		bserrno = 0;
2703 	} else {
2704 		/*
2705 		 * This case needs to be looked at further.  Same problem
2706 		 *  exists with applications that rely on explicit blob
2707 		 *  iteration.  We should just skip the blob that failed
2708 		 *  to load and continue on to the next one.
2709 		 */
2710 		SPDK_ERRLOG("Error in iterating blobs\n");
2711 	}
2712 
2713 	ctx->iter_cb_fn = NULL;
2714 
2715 	spdk_free(ctx->super);
2716 	spdk_free(ctx->mask);
2717 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2718 	free(ctx);
2719 }
2720 
2721 static void
2722 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2723 {
2724 	ctx->seq = seq;
2725 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2726 }
2727 
2728 static void
2729 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2730 {
2731 	struct spdk_bs_load_ctx *ctx = cb_arg;
2732 	int rc;
2733 
2734 	/* The type must be correct */
2735 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2736 
2737 	/* The length of the mask (in bits) must not be greater than
2738 	 * the length of the buffer (converted to bits) */
2739 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2740 
2741 	/* The length of the mask must be exactly equal to the size
2742 	 * (in pages) of the metadata region */
2743 	assert(ctx->mask->length == ctx->super->md_len);
2744 
2745 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2746 	if (rc < 0) {
2747 		spdk_free(ctx->mask);
2748 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2749 		return;
2750 	}
2751 
2752 	_spdk_bs_load_complete(seq, ctx, bserrno);
2753 }
2754 
2755 static void
2756 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2757 {
2758 	struct spdk_bs_load_ctx *ctx = cb_arg;
2759 	uint64_t		lba, lba_count, mask_size;
2760 	int			rc;
2761 
2762 	/* The type must be correct */
2763 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2764 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2765 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2766 					     struct spdk_blob_md_page) * 8));
2767 	/* The length of the mask must be exactly equal to the total number of clusters */
2768 	assert(ctx->mask->length == ctx->bs->total_clusters);
2769 
2770 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2771 	if (rc < 0) {
2772 		spdk_free(ctx->mask);
2773 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2774 		return;
2775 	}
2776 
2777 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2778 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2779 
2780 	spdk_free(ctx->mask);
2781 
2782 	/* Read the used blobids mask */
2783 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2784 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2785 				 SPDK_MALLOC_DMA);
2786 	if (!ctx->mask) {
2787 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2788 		return;
2789 	}
2790 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2791 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2792 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2793 				  _spdk_bs_load_used_blobids_cpl, ctx);
2794 }
2795 
2796 static void
2797 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2798 {
2799 	struct spdk_bs_load_ctx *ctx = cb_arg;
2800 	uint64_t		lba, lba_count, mask_size;
2801 	int			rc;
2802 
2803 	/* The type must be correct */
2804 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2805 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2806 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2807 				     8));
2808 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2809 	assert(ctx->mask->length == ctx->super->md_len);
2810 
2811 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2812 	if (rc < 0) {
2813 		spdk_free(ctx->mask);
2814 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2815 		return;
2816 	}
2817 
2818 	spdk_free(ctx->mask);
2819 
2820 	/* Read the used clusters mask */
2821 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2822 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2823 				 SPDK_MALLOC_DMA);
2824 	if (!ctx->mask) {
2825 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2826 		return;
2827 	}
2828 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2829 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2830 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2831 				  _spdk_bs_load_used_clusters_cpl, ctx);
2832 }
2833 
2834 static void
2835 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2836 {
2837 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2838 	uint64_t lba, lba_count, mask_size;
2839 
2840 	/* Read the used pages mask */
2841 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2842 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2843 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2844 	if (!ctx->mask) {
2845 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2846 		return;
2847 	}
2848 
2849 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2850 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2851 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2852 				  _spdk_bs_load_used_pages_cpl, ctx);
2853 }
2854 
2855 static int
2856 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2857 {
2858 	struct spdk_blob_md_descriptor *desc;
2859 	size_t	cur_desc = 0;
2860 
2861 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2862 	while (cur_desc < sizeof(page->descriptors)) {
2863 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2864 			if (desc->length == 0) {
2865 				/* If padding and length are 0, this terminates the page */
2866 				break;
2867 			}
2868 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2869 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2870 			unsigned int				i, j;
2871 			unsigned int				cluster_count = 0;
2872 			uint32_t				cluster_idx;
2873 
2874 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2875 
2876 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2877 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2878 					cluster_idx = desc_extent->extents[i].cluster_idx;
2879 					/*
2880 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2881 					 * in the used cluster map.
2882 					 */
2883 					if (cluster_idx != 0) {
2884 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2885 						if (bs->num_free_clusters == 0) {
2886 							return -ENOSPC;
2887 						}
2888 						bs->num_free_clusters--;
2889 					}
2890 					cluster_count++;
2891 				}
2892 			}
2893 			if (cluster_count == 0) {
2894 				return -EINVAL;
2895 			}
2896 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2897 			/* Skip this item */
2898 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2899 			/* Skip this item */
2900 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2901 			/* Skip this item */
2902 		} else {
2903 			/* Error */
2904 			return -EINVAL;
2905 		}
2906 		/* Advance to the next descriptor */
2907 		cur_desc += sizeof(*desc) + desc->length;
2908 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2909 			break;
2910 		}
2911 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2912 	}
2913 	return 0;
2914 }
2915 
2916 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2917 {
2918 	uint32_t crc;
2919 
2920 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2921 	if (crc != ctx->page->crc) {
2922 		return false;
2923 	}
2924 
2925 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2926 		return false;
2927 	}
2928 	return true;
2929 }
2930 
2931 static void
2932 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2933 
2934 static void
2935 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2936 {
2937 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2938 
2939 	_spdk_bs_load_complete(seq, ctx, bserrno);
2940 }
2941 
2942 static void
2943 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2944 {
2945 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2946 
2947 	spdk_free(ctx->mask);
2948 	ctx->mask = NULL;
2949 
2950 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2951 }
2952 
2953 static void
2954 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2955 {
2956 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2957 
2958 	spdk_free(ctx->mask);
2959 	ctx->mask = NULL;
2960 
2961 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2962 }
2963 
2964 static void
2965 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2966 {
2967 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2968 }
2969 
2970 static void
2971 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2972 {
2973 	struct spdk_bs_load_ctx *ctx = cb_arg;
2974 	uint64_t num_md_clusters;
2975 	uint64_t i;
2976 	uint32_t page_num;
2977 
2978 	if (bserrno != 0) {
2979 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2980 		return;
2981 	}
2982 
2983 	page_num = ctx->cur_page;
2984 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2985 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2986 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2987 			if (ctx->page->sequence_num == 0) {
2988 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2989 			}
2990 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2991 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
2992 				return;
2993 			}
2994 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
2995 				ctx->in_page_chain = true;
2996 				ctx->cur_page = ctx->page->next;
2997 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2998 				return;
2999 			}
3000 		}
3001 	}
3002 
3003 	ctx->in_page_chain = false;
3004 
3005 	do {
3006 		ctx->page_index++;
3007 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
3008 
3009 	if (ctx->page_index < ctx->super->md_len) {
3010 		ctx->cur_page = ctx->page_index;
3011 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3012 	} else {
3013 		/* Claim all of the clusters used by the metadata */
3014 		num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
3015 		for (i = 0; i < num_md_clusters; i++) {
3016 			_spdk_bs_claim_cluster(ctx->bs, i);
3017 		}
3018 		spdk_free(ctx->page);
3019 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
3020 	}
3021 }
3022 
3023 static void
3024 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3025 {
3026 	struct spdk_bs_load_ctx *ctx = cb_arg;
3027 	uint64_t lba;
3028 
3029 	assert(ctx->cur_page < ctx->super->md_len);
3030 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3031 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3032 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3033 				  _spdk_bs_load_replay_md_cpl, ctx);
3034 }
3035 
3036 static void
3037 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3038 {
3039 	struct spdk_bs_load_ctx *ctx = cb_arg;
3040 
3041 	ctx->page_index = 0;
3042 	ctx->cur_page = 0;
3043 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3044 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3045 	if (!ctx->page) {
3046 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3047 		return;
3048 	}
3049 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3050 }
3051 
3052 static void
3053 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3054 {
3055 	struct spdk_bs_load_ctx *ctx = cb_arg;
3056 	int		rc;
3057 
3058 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3059 	if (rc < 0) {
3060 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3061 		return;
3062 	}
3063 
3064 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3065 	if (rc < 0) {
3066 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3067 		return;
3068 	}
3069 
3070 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3071 	if (rc < 0) {
3072 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3073 		return;
3074 	}
3075 
3076 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3077 	_spdk_bs_load_replay_md(seq, cb_arg);
3078 }
3079 
3080 static void
3081 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3082 {
3083 	struct spdk_bs_load_ctx *ctx = cb_arg;
3084 	uint32_t	crc;
3085 	int		rc;
3086 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3087 
3088 	if (ctx->super->version > SPDK_BS_VERSION ||
3089 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3090 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3091 		return;
3092 	}
3093 
3094 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3095 		   sizeof(ctx->super->signature)) != 0) {
3096 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3097 		return;
3098 	}
3099 
3100 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3101 	if (crc != ctx->super->crc) {
3102 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3103 		return;
3104 	}
3105 
3106 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3107 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3108 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3109 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3110 	} else {
3111 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3112 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3113 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3114 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3115 		return;
3116 	}
3117 
3118 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3119 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3120 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3121 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3122 		return;
3123 	}
3124 
3125 	if (ctx->super->size == 0) {
3126 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
3127 	}
3128 
3129 	if (ctx->super->io_unit_size == 0) {
3130 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
3131 	}
3132 
3133 	/* Parse the super block */
3134 	ctx->bs->clean = 1;
3135 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3136 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
3137 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3138 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
3139 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3140 	if (rc < 0) {
3141 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3142 		return;
3143 	}
3144 	ctx->bs->md_start = ctx->super->md_start;
3145 	ctx->bs->md_len = ctx->super->md_len;
3146 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
3147 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3148 	ctx->bs->super_blob = ctx->super->super_blob;
3149 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3150 
3151 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3152 		_spdk_bs_recover(seq, ctx);
3153 	} else {
3154 		_spdk_bs_load_read_used_pages(seq, ctx);
3155 	}
3156 }
3157 
3158 void
3159 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3160 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3161 {
3162 	struct spdk_blob_store	*bs;
3163 	struct spdk_bs_cpl	cpl;
3164 	spdk_bs_sequence_t	*seq;
3165 	struct spdk_bs_load_ctx *ctx;
3166 	struct spdk_bs_opts	opts = {};
3167 	int err;
3168 
3169 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3170 
3171 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3172 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3173 		dev->destroy(dev);
3174 		cb_fn(cb_arg, NULL, -EINVAL);
3175 		return;
3176 	}
3177 
3178 	if (o) {
3179 		opts = *o;
3180 	} else {
3181 		spdk_bs_opts_init(&opts);
3182 	}
3183 
3184 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3185 		dev->destroy(dev);
3186 		cb_fn(cb_arg, NULL, -EINVAL);
3187 		return;
3188 	}
3189 
3190 	err = _spdk_bs_alloc(dev, &opts, &bs);
3191 	if (err) {
3192 		dev->destroy(dev);
3193 		cb_fn(cb_arg, NULL, err);
3194 		return;
3195 	}
3196 
3197 	ctx = calloc(1, sizeof(*ctx));
3198 	if (!ctx) {
3199 		_spdk_bs_free(bs);
3200 		cb_fn(cb_arg, NULL, -ENOMEM);
3201 		return;
3202 	}
3203 
3204 	ctx->bs = bs;
3205 	ctx->iter_cb_fn = opts.iter_cb_fn;
3206 	ctx->iter_cb_arg = opts.iter_cb_arg;
3207 
3208 	/* Allocate memory for the super block */
3209 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3210 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3211 	if (!ctx->super) {
3212 		free(ctx);
3213 		_spdk_bs_free(bs);
3214 		cb_fn(cb_arg, NULL, -ENOMEM);
3215 		return;
3216 	}
3217 
3218 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3219 	cpl.u.bs_handle.cb_fn = cb_fn;
3220 	cpl.u.bs_handle.cb_arg = cb_arg;
3221 	cpl.u.bs_handle.bs = bs;
3222 
3223 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3224 	if (!seq) {
3225 		spdk_free(ctx->super);
3226 		free(ctx);
3227 		_spdk_bs_free(bs);
3228 		cb_fn(cb_arg, NULL, -ENOMEM);
3229 		return;
3230 	}
3231 
3232 	/* Read the super block */
3233 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3234 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3235 				  _spdk_bs_load_super_cpl, ctx);
3236 }
3237 
3238 /* END spdk_bs_load */
3239 
3240 /* START spdk_bs_dump */
3241 
3242 struct spdk_bs_dump_ctx {
3243 	struct spdk_blob_store		*bs;
3244 	struct spdk_bs_super_block	*super;
3245 	uint32_t			cur_page;
3246 	struct spdk_blob_md_page	*page;
3247 	spdk_bs_sequence_t		*seq;
3248 	FILE				*fp;
3249 	spdk_bs_dump_print_xattr	print_xattr_fn;
3250 	char				xattr_name[4096];
3251 };
3252 
3253 static void
3254 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3255 {
3256 	spdk_free(ctx->super);
3257 
3258 	/*
3259 	 * We need to defer calling spdk_bs_call_cpl() until after
3260 	 * dev destruction, so tuck these away for later use.
3261 	 */
3262 	ctx->bs->unload_err = bserrno;
3263 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3264 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3265 
3266 	spdk_bs_sequence_finish(seq, 0);
3267 	_spdk_bs_free(ctx->bs);
3268 	free(ctx);
3269 }
3270 
3271 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3272 
3273 static void
3274 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3275 {
3276 	uint32_t page_idx = ctx->cur_page;
3277 	struct spdk_blob_md_page *page = ctx->page;
3278 	struct spdk_blob_md_descriptor *desc;
3279 	size_t cur_desc = 0;
3280 	uint32_t crc;
3281 
3282 	fprintf(ctx->fp, "=========\n");
3283 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3284 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3285 
3286 	crc = _spdk_blob_md_page_calc_crc(page);
3287 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3288 
3289 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3290 	while (cur_desc < sizeof(page->descriptors)) {
3291 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3292 			if (desc->length == 0) {
3293 				/* If padding and length are 0, this terminates the page */
3294 				break;
3295 			}
3296 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3297 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3298 			unsigned int				i;
3299 
3300 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3301 
3302 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3303 				if (desc_extent->extents[i].cluster_idx != 0) {
3304 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3305 						desc_extent->extents[i].cluster_idx);
3306 				} else {
3307 					fprintf(ctx->fp, "Unallocated Extent - ");
3308 				}
3309 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3310 				fprintf(ctx->fp, "\n");
3311 			}
3312 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3313 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3314 			uint32_t i;
3315 
3316 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3317 
3318 			if (desc_xattr->length !=
3319 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3320 			    desc_xattr->name_length + desc_xattr->value_length) {
3321 			}
3322 
3323 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3324 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3325 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3326 			fprintf(ctx->fp, "       value = \"");
3327 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3328 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3329 					    desc_xattr->value_length);
3330 			fprintf(ctx->fp, "\"\n");
3331 			for (i = 0; i < desc_xattr->value_length; i++) {
3332 				if (i % 16 == 0) {
3333 					fprintf(ctx->fp, "               ");
3334 				}
3335 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3336 				if ((i + 1) % 16 == 0) {
3337 					fprintf(ctx->fp, "\n");
3338 				}
3339 			}
3340 			if (i % 16 != 0) {
3341 				fprintf(ctx->fp, "\n");
3342 			}
3343 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3344 			/* TODO */
3345 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3346 			/* TODO */
3347 		} else {
3348 			/* Error */
3349 		}
3350 		/* Advance to the next descriptor */
3351 		cur_desc += sizeof(*desc) + desc->length;
3352 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3353 			break;
3354 		}
3355 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3356 	}
3357 }
3358 
3359 static void
3360 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3361 {
3362 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3363 
3364 	if (bserrno != 0) {
3365 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3366 		return;
3367 	}
3368 
3369 	if (ctx->page->id != 0) {
3370 		_spdk_bs_dump_print_md_page(ctx);
3371 	}
3372 
3373 	ctx->cur_page++;
3374 
3375 	if (ctx->cur_page < ctx->super->md_len) {
3376 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3377 	} else {
3378 		spdk_free(ctx->page);
3379 		_spdk_bs_dump_finish(seq, ctx, 0);
3380 	}
3381 }
3382 
3383 static void
3384 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3385 {
3386 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3387 	uint64_t lba;
3388 
3389 	assert(ctx->cur_page < ctx->super->md_len);
3390 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3391 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3392 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3393 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3394 }
3395 
3396 static void
3397 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3398 {
3399 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3400 
3401 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3402 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3403 		   sizeof(ctx->super->signature)) != 0) {
3404 		fprintf(ctx->fp, "(Mismatch)\n");
3405 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3406 		return;
3407 	} else {
3408 		fprintf(ctx->fp, "(OK)\n");
3409 	}
3410 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3411 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3412 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3413 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3414 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3415 	fprintf(ctx->fp, "Super Blob ID: ");
3416 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3417 		fprintf(ctx->fp, "(None)\n");
3418 	} else {
3419 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3420 	}
3421 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3422 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3423 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3424 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3425 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3426 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3427 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3428 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3429 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3430 
3431 	ctx->cur_page = 0;
3432 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3433 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3434 	if (!ctx->page) {
3435 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3436 		return;
3437 	}
3438 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3439 }
3440 
3441 void
3442 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3443 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3444 {
3445 	struct spdk_blob_store	*bs;
3446 	struct spdk_bs_cpl	cpl;
3447 	spdk_bs_sequence_t	*seq;
3448 	struct spdk_bs_dump_ctx *ctx;
3449 	struct spdk_bs_opts	opts = {};
3450 	int err;
3451 
3452 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3453 
3454 	spdk_bs_opts_init(&opts);
3455 
3456 	err = _spdk_bs_alloc(dev, &opts, &bs);
3457 	if (err) {
3458 		dev->destroy(dev);
3459 		cb_fn(cb_arg, err);
3460 		return;
3461 	}
3462 
3463 	ctx = calloc(1, sizeof(*ctx));
3464 	if (!ctx) {
3465 		_spdk_bs_free(bs);
3466 		cb_fn(cb_arg, -ENOMEM);
3467 		return;
3468 	}
3469 
3470 	ctx->bs = bs;
3471 	ctx->fp = fp;
3472 	ctx->print_xattr_fn = print_xattr_fn;
3473 
3474 	/* Allocate memory for the super block */
3475 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3476 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3477 	if (!ctx->super) {
3478 		free(ctx);
3479 		_spdk_bs_free(bs);
3480 		cb_fn(cb_arg, -ENOMEM);
3481 		return;
3482 	}
3483 
3484 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3485 	cpl.u.bs_basic.cb_fn = cb_fn;
3486 	cpl.u.bs_basic.cb_arg = cb_arg;
3487 
3488 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3489 	if (!seq) {
3490 		spdk_free(ctx->super);
3491 		free(ctx);
3492 		_spdk_bs_free(bs);
3493 		cb_fn(cb_arg, -ENOMEM);
3494 		return;
3495 	}
3496 
3497 	/* Read the super block */
3498 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3499 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3500 				  _spdk_bs_dump_super_cpl, ctx);
3501 }
3502 
3503 /* END spdk_bs_dump */
3504 
3505 /* START spdk_bs_init */
3506 
3507 struct spdk_bs_init_ctx {
3508 	struct spdk_blob_store		*bs;
3509 	struct spdk_bs_super_block	*super;
3510 };
3511 
3512 static void
3513 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3514 {
3515 	struct spdk_bs_init_ctx *ctx = cb_arg;
3516 
3517 	spdk_free(ctx->super);
3518 	free(ctx);
3519 
3520 	spdk_bs_sequence_finish(seq, bserrno);
3521 }
3522 
3523 static void
3524 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3525 {
3526 	struct spdk_bs_init_ctx *ctx = cb_arg;
3527 
3528 	/* Write super block */
3529 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3530 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3531 				   _spdk_bs_init_persist_super_cpl, ctx);
3532 }
3533 
3534 void
3535 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3536 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3537 {
3538 	struct spdk_bs_init_ctx *ctx;
3539 	struct spdk_blob_store	*bs;
3540 	struct spdk_bs_cpl	cpl;
3541 	spdk_bs_sequence_t	*seq;
3542 	spdk_bs_batch_t		*batch;
3543 	uint64_t		num_md_lba;
3544 	uint64_t		num_md_pages;
3545 	uint64_t		num_md_clusters;
3546 	uint32_t		i;
3547 	struct spdk_bs_opts	opts = {};
3548 	int			rc;
3549 
3550 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3551 
3552 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3553 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3554 			    dev->blocklen);
3555 		dev->destroy(dev);
3556 		cb_fn(cb_arg, NULL, -EINVAL);
3557 		return;
3558 	}
3559 
3560 	if (o) {
3561 		opts = *o;
3562 	} else {
3563 		spdk_bs_opts_init(&opts);
3564 	}
3565 
3566 	if (_spdk_bs_opts_verify(&opts) != 0) {
3567 		dev->destroy(dev);
3568 		cb_fn(cb_arg, NULL, -EINVAL);
3569 		return;
3570 	}
3571 
3572 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3573 	if (rc) {
3574 		dev->destroy(dev);
3575 		cb_fn(cb_arg, NULL, rc);
3576 		return;
3577 	}
3578 
3579 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3580 		/* By default, allocate 1 page per cluster.
3581 		 * Technically, this over-allocates metadata
3582 		 * because more metadata will reduce the number
3583 		 * of usable clusters. This can be addressed with
3584 		 * more complex math in the future.
3585 		 */
3586 		bs->md_len = bs->total_clusters;
3587 	} else {
3588 		bs->md_len = opts.num_md_pages;
3589 	}
3590 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3591 	if (rc < 0) {
3592 		_spdk_bs_free(bs);
3593 		cb_fn(cb_arg, NULL, -ENOMEM);
3594 		return;
3595 	}
3596 
3597 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3598 	if (rc < 0) {
3599 		_spdk_bs_free(bs);
3600 		cb_fn(cb_arg, NULL, -ENOMEM);
3601 		return;
3602 	}
3603 
3604 	ctx = calloc(1, sizeof(*ctx));
3605 	if (!ctx) {
3606 		_spdk_bs_free(bs);
3607 		cb_fn(cb_arg, NULL, -ENOMEM);
3608 		return;
3609 	}
3610 
3611 	ctx->bs = bs;
3612 
3613 	/* Allocate memory for the super block */
3614 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3615 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3616 	if (!ctx->super) {
3617 		free(ctx);
3618 		_spdk_bs_free(bs);
3619 		cb_fn(cb_arg, NULL, -ENOMEM);
3620 		return;
3621 	}
3622 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3623 	       sizeof(ctx->super->signature));
3624 	ctx->super->version = SPDK_BS_VERSION;
3625 	ctx->super->length = sizeof(*ctx->super);
3626 	ctx->super->super_blob = bs->super_blob;
3627 	ctx->super->clean = 0;
3628 	ctx->super->cluster_size = bs->cluster_sz;
3629 	ctx->super->io_unit_size = bs->io_unit_size;
3630 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3631 
3632 	/* Calculate how many pages the metadata consumes at the front
3633 	 * of the disk.
3634 	 */
3635 
3636 	/* The super block uses 1 page */
3637 	num_md_pages = 1;
3638 
3639 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3640 	 * up to the nearest page, plus a header.
3641 	 */
3642 	ctx->super->used_page_mask_start = num_md_pages;
3643 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3644 					 spdk_divide_round_up(bs->md_len, 8),
3645 					 SPDK_BS_PAGE_SIZE);
3646 	num_md_pages += ctx->super->used_page_mask_len;
3647 
3648 	/* The used_clusters mask requires 1 bit per cluster, rounded
3649 	 * up to the nearest page, plus a header.
3650 	 */
3651 	ctx->super->used_cluster_mask_start = num_md_pages;
3652 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3653 					    spdk_divide_round_up(bs->total_clusters, 8),
3654 					    SPDK_BS_PAGE_SIZE);
3655 	num_md_pages += ctx->super->used_cluster_mask_len;
3656 
3657 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3658 	 * up to the nearest page, plus a header.
3659 	 */
3660 	ctx->super->used_blobid_mask_start = num_md_pages;
3661 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3662 					   spdk_divide_round_up(bs->md_len, 8),
3663 					   SPDK_BS_PAGE_SIZE);
3664 	num_md_pages += ctx->super->used_blobid_mask_len;
3665 
3666 	/* The metadata region size was chosen above */
3667 	ctx->super->md_start = bs->md_start = num_md_pages;
3668 	ctx->super->md_len = bs->md_len;
3669 	num_md_pages += bs->md_len;
3670 
3671 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3672 
3673 	ctx->super->size = dev->blockcnt * dev->blocklen;
3674 
3675 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3676 
3677 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
3678 	if (num_md_clusters > bs->total_clusters) {
3679 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3680 			    "please decrease number of pages reserved for metadata "
3681 			    "or increase cluster size.\n");
3682 		spdk_free(ctx->super);
3683 		free(ctx);
3684 		_spdk_bs_free(bs);
3685 		cb_fn(cb_arg, NULL, -ENOMEM);
3686 		return;
3687 	}
3688 	/* Claim all of the clusters used by the metadata */
3689 	for (i = 0; i < num_md_clusters; i++) {
3690 		_spdk_bs_claim_cluster(bs, i);
3691 	}
3692 
3693 	bs->total_data_clusters = bs->num_free_clusters;
3694 
3695 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3696 	cpl.u.bs_handle.cb_fn = cb_fn;
3697 	cpl.u.bs_handle.cb_arg = cb_arg;
3698 	cpl.u.bs_handle.bs = bs;
3699 
3700 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3701 	if (!seq) {
3702 		spdk_free(ctx->super);
3703 		free(ctx);
3704 		_spdk_bs_free(bs);
3705 		cb_fn(cb_arg, NULL, -ENOMEM);
3706 		return;
3707 	}
3708 
3709 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3710 
3711 	/* Clear metadata space */
3712 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3713 
3714 	if (opts.clear_method == BS_CLEAR_WITH_UNMAP) {
3715 		/* Trim data clusters */
3716 		spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3717 	} else if (opts.clear_method == BS_CLEAR_WITH_WRITE_ZEROES) {
3718 		/* Write_zeroes to data clusters */
3719 		spdk_bs_batch_write_zeroes_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3720 	}
3721 
3722 	spdk_bs_batch_close(batch);
3723 }
3724 
3725 /* END spdk_bs_init */
3726 
3727 /* START spdk_bs_destroy */
3728 
3729 static void
3730 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3731 {
3732 	struct spdk_bs_init_ctx *ctx = cb_arg;
3733 	struct spdk_blob_store *bs = ctx->bs;
3734 
3735 	/*
3736 	 * We need to defer calling spdk_bs_call_cpl() until after
3737 	 * dev destruction, so tuck these away for later use.
3738 	 */
3739 	bs->unload_err = bserrno;
3740 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3741 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3742 
3743 	spdk_bs_sequence_finish(seq, bserrno);
3744 
3745 	_spdk_bs_free(bs);
3746 	free(ctx);
3747 }
3748 
3749 void
3750 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3751 		void *cb_arg)
3752 {
3753 	struct spdk_bs_cpl	cpl;
3754 	spdk_bs_sequence_t	*seq;
3755 	struct spdk_bs_init_ctx *ctx;
3756 
3757 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3758 
3759 	if (!TAILQ_EMPTY(&bs->blobs)) {
3760 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3761 		cb_fn(cb_arg, -EBUSY);
3762 		return;
3763 	}
3764 
3765 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3766 	cpl.u.bs_basic.cb_fn = cb_fn;
3767 	cpl.u.bs_basic.cb_arg = cb_arg;
3768 
3769 	ctx = calloc(1, sizeof(*ctx));
3770 	if (!ctx) {
3771 		cb_fn(cb_arg, -ENOMEM);
3772 		return;
3773 	}
3774 
3775 	ctx->bs = bs;
3776 
3777 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3778 	if (!seq) {
3779 		free(ctx);
3780 		cb_fn(cb_arg, -ENOMEM);
3781 		return;
3782 	}
3783 
3784 	/* Write zeroes to the super block */
3785 	spdk_bs_sequence_write_zeroes_dev(seq,
3786 					  _spdk_bs_page_to_lba(bs, 0),
3787 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3788 					  _spdk_bs_destroy_trim_cpl, ctx);
3789 }
3790 
3791 /* END spdk_bs_destroy */
3792 
3793 /* START spdk_bs_unload */
3794 
3795 static void
3796 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3797 {
3798 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3799 
3800 	spdk_free(ctx->super);
3801 
3802 	/*
3803 	 * We need to defer calling spdk_bs_call_cpl() until after
3804 	 * dev destruction, so tuck these away for later use.
3805 	 */
3806 	ctx->bs->unload_err = bserrno;
3807 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3808 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3809 
3810 	spdk_bs_sequence_finish(seq, bserrno);
3811 
3812 	_spdk_bs_free(ctx->bs);
3813 	free(ctx);
3814 }
3815 
3816 static void
3817 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3818 {
3819 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3820 
3821 	spdk_free(ctx->mask);
3822 	ctx->super->clean = 1;
3823 
3824 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3825 }
3826 
3827 static void
3828 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3829 {
3830 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3831 
3832 	spdk_free(ctx->mask);
3833 	ctx->mask = NULL;
3834 
3835 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3836 }
3837 
3838 static void
3839 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3840 {
3841 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3842 
3843 	spdk_free(ctx->mask);
3844 	ctx->mask = NULL;
3845 
3846 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3847 }
3848 
3849 static void
3850 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3851 {
3852 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3853 }
3854 
3855 void
3856 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3857 {
3858 	struct spdk_bs_cpl	cpl;
3859 	spdk_bs_sequence_t	*seq;
3860 	struct spdk_bs_load_ctx *ctx;
3861 
3862 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3863 
3864 	if (!TAILQ_EMPTY(&bs->blobs)) {
3865 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3866 		cb_fn(cb_arg, -EBUSY);
3867 		return;
3868 	}
3869 
3870 	ctx = calloc(1, sizeof(*ctx));
3871 	if (!ctx) {
3872 		cb_fn(cb_arg, -ENOMEM);
3873 		return;
3874 	}
3875 
3876 	ctx->bs = bs;
3877 
3878 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3879 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3880 	if (!ctx->super) {
3881 		free(ctx);
3882 		cb_fn(cb_arg, -ENOMEM);
3883 		return;
3884 	}
3885 
3886 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3887 	cpl.u.bs_basic.cb_fn = cb_fn;
3888 	cpl.u.bs_basic.cb_arg = cb_arg;
3889 
3890 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3891 	if (!seq) {
3892 		spdk_free(ctx->super);
3893 		free(ctx);
3894 		cb_fn(cb_arg, -ENOMEM);
3895 		return;
3896 	}
3897 
3898 	/* Read super block */
3899 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3900 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3901 				  _spdk_bs_unload_read_super_cpl, ctx);
3902 }
3903 
3904 /* END spdk_bs_unload */
3905 
3906 /* START spdk_bs_set_super */
3907 
3908 struct spdk_bs_set_super_ctx {
3909 	struct spdk_blob_store		*bs;
3910 	struct spdk_bs_super_block	*super;
3911 };
3912 
3913 static void
3914 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3915 {
3916 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3917 
3918 	if (bserrno != 0) {
3919 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3920 	}
3921 
3922 	spdk_free(ctx->super);
3923 
3924 	spdk_bs_sequence_finish(seq, bserrno);
3925 
3926 	free(ctx);
3927 }
3928 
3929 static void
3930 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3931 {
3932 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3933 
3934 	if (bserrno != 0) {
3935 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3936 		spdk_free(ctx->super);
3937 		spdk_bs_sequence_finish(seq, bserrno);
3938 		free(ctx);
3939 		return;
3940 	}
3941 
3942 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3943 }
3944 
3945 void
3946 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3947 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3948 {
3949 	struct spdk_bs_cpl		cpl;
3950 	spdk_bs_sequence_t		*seq;
3951 	struct spdk_bs_set_super_ctx	*ctx;
3952 
3953 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3954 
3955 	ctx = calloc(1, sizeof(*ctx));
3956 	if (!ctx) {
3957 		cb_fn(cb_arg, -ENOMEM);
3958 		return;
3959 	}
3960 
3961 	ctx->bs = bs;
3962 
3963 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3964 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3965 	if (!ctx->super) {
3966 		free(ctx);
3967 		cb_fn(cb_arg, -ENOMEM);
3968 		return;
3969 	}
3970 
3971 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3972 	cpl.u.bs_basic.cb_fn = cb_fn;
3973 	cpl.u.bs_basic.cb_arg = cb_arg;
3974 
3975 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3976 	if (!seq) {
3977 		spdk_free(ctx->super);
3978 		free(ctx);
3979 		cb_fn(cb_arg, -ENOMEM);
3980 		return;
3981 	}
3982 
3983 	bs->super_blob = blobid;
3984 
3985 	/* Read super block */
3986 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3987 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3988 				  _spdk_bs_set_super_read_cpl, ctx);
3989 }
3990 
3991 /* END spdk_bs_set_super */
3992 
3993 void
3994 spdk_bs_get_super(struct spdk_blob_store *bs,
3995 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
3996 {
3997 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
3998 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
3999 	} else {
4000 		cb_fn(cb_arg, bs->super_blob, 0);
4001 	}
4002 }
4003 
4004 uint64_t
4005 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
4006 {
4007 	return bs->cluster_sz;
4008 }
4009 
4010 uint64_t
4011 spdk_bs_get_page_size(struct spdk_blob_store *bs)
4012 {
4013 	return SPDK_BS_PAGE_SIZE;
4014 }
4015 
4016 uint64_t
4017 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
4018 {
4019 	return bs->io_unit_size;
4020 }
4021 
4022 uint64_t
4023 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
4024 {
4025 	return bs->num_free_clusters;
4026 }
4027 
4028 uint64_t
4029 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
4030 {
4031 	return bs->total_data_clusters;
4032 }
4033 
4034 static int
4035 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
4036 {
4037 	bs->md_channel = spdk_get_io_channel(bs);
4038 	if (!bs->md_channel) {
4039 		SPDK_ERRLOG("Failed to get IO channel.\n");
4040 		return -1;
4041 	}
4042 
4043 	return 0;
4044 }
4045 
4046 static int
4047 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
4048 {
4049 	spdk_put_io_channel(bs->md_channel);
4050 
4051 	return 0;
4052 }
4053 
4054 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4055 {
4056 	assert(blob != NULL);
4057 
4058 	return blob->id;
4059 }
4060 
4061 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4062 {
4063 	assert(blob != NULL);
4064 
4065 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4066 }
4067 
4068 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
4069 {
4070 	assert(blob != NULL);
4071 
4072 	return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs);
4073 }
4074 
4075 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4076 {
4077 	assert(blob != NULL);
4078 
4079 	return blob->active.num_clusters;
4080 }
4081 
4082 /* START spdk_bs_create_blob */
4083 
4084 static void
4085 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4086 {
4087 	struct spdk_blob *blob = cb_arg;
4088 
4089 	_spdk_blob_free(blob);
4090 
4091 	spdk_bs_sequence_finish(seq, bserrno);
4092 }
4093 
4094 static int
4095 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4096 		      bool internal)
4097 {
4098 	uint64_t i;
4099 	size_t value_len = 0;
4100 	int rc;
4101 	const void *value = NULL;
4102 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4103 		return -EINVAL;
4104 	}
4105 	for (i = 0; i < xattrs->count; i++) {
4106 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4107 		if (value == NULL || value_len == 0) {
4108 			return -EINVAL;
4109 		}
4110 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4111 		if (rc < 0) {
4112 			return rc;
4113 		}
4114 	}
4115 	return 0;
4116 }
4117 
4118 static void
4119 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4120 {
4121 	_spdk_blob_verify_md_op(blob);
4122 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4123 	blob->state = SPDK_BLOB_STATE_DIRTY;
4124 }
4125 
4126 static void
4127 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4128 		     const struct spdk_blob_opts *opts,
4129 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4130 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4131 {
4132 	struct spdk_blob	*blob;
4133 	uint32_t		page_idx;
4134 	struct spdk_bs_cpl	cpl;
4135 	struct spdk_blob_opts	opts_default;
4136 	struct spdk_blob_xattr_opts internal_xattrs_default;
4137 	spdk_bs_sequence_t	*seq;
4138 	spdk_blob_id		id;
4139 	int rc;
4140 
4141 	assert(spdk_get_thread() == bs->md_thread);
4142 
4143 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4144 	if (page_idx == UINT32_MAX) {
4145 		cb_fn(cb_arg, 0, -ENOMEM);
4146 		return;
4147 	}
4148 	spdk_bit_array_set(bs->used_blobids, page_idx);
4149 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4150 
4151 	id = _spdk_bs_page_to_blobid(page_idx);
4152 
4153 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4154 
4155 	blob = _spdk_blob_alloc(bs, id);
4156 	if (!blob) {
4157 		cb_fn(cb_arg, 0, -ENOMEM);
4158 		return;
4159 	}
4160 
4161 	if (!opts) {
4162 		spdk_blob_opts_init(&opts_default);
4163 		opts = &opts_default;
4164 	}
4165 	if (!internal_xattrs) {
4166 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4167 		internal_xattrs = &internal_xattrs_default;
4168 	}
4169 
4170 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4171 	if (rc < 0) {
4172 		_spdk_blob_free(blob);
4173 		cb_fn(cb_arg, 0, rc);
4174 		return;
4175 	}
4176 
4177 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4178 	if (rc < 0) {
4179 		_spdk_blob_free(blob);
4180 		cb_fn(cb_arg, 0, rc);
4181 		return;
4182 	}
4183 
4184 	if (opts->thin_provision) {
4185 		_spdk_blob_set_thin_provision(blob);
4186 	}
4187 
4188 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4189 	if (rc < 0) {
4190 		_spdk_blob_free(blob);
4191 		cb_fn(cb_arg, 0, rc);
4192 		return;
4193 	}
4194 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4195 	cpl.u.blobid.cb_fn = cb_fn;
4196 	cpl.u.blobid.cb_arg = cb_arg;
4197 	cpl.u.blobid.blobid = blob->id;
4198 
4199 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4200 	if (!seq) {
4201 		_spdk_blob_free(blob);
4202 		cb_fn(cb_arg, 0, -ENOMEM);
4203 		return;
4204 	}
4205 
4206 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4207 }
4208 
4209 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4210 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4211 {
4212 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4213 }
4214 
4215 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4216 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4217 {
4218 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4219 }
4220 
4221 /* END spdk_bs_create_blob */
4222 
4223 /* START blob_cleanup */
4224 
4225 struct spdk_clone_snapshot_ctx {
4226 	struct spdk_bs_cpl      cpl;
4227 	int bserrno;
4228 	bool frozen;
4229 
4230 	struct spdk_io_channel *channel;
4231 
4232 	/* Current cluster for inflate operation */
4233 	uint64_t cluster;
4234 
4235 	/* For inflation force allocation of all unallocated clusters and remove
4236 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4237 	bool allocate_all;
4238 
4239 	struct {
4240 		spdk_blob_id id;
4241 		struct spdk_blob *blob;
4242 	} original;
4243 	struct {
4244 		spdk_blob_id id;
4245 		struct spdk_blob *blob;
4246 	} new;
4247 
4248 	/* xattrs specified for snapshot/clones only. They have no impact on
4249 	 * the original blobs xattrs. */
4250 	const struct spdk_blob_xattr_opts *xattrs;
4251 };
4252 
4253 static void
4254 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4255 {
4256 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4257 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4258 
4259 	if (bserrno != 0) {
4260 		if (ctx->bserrno != 0) {
4261 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4262 		} else {
4263 			ctx->bserrno = bserrno;
4264 		}
4265 	}
4266 
4267 	switch (cpl->type) {
4268 	case SPDK_BS_CPL_TYPE_BLOBID:
4269 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4270 		break;
4271 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4272 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4273 		break;
4274 	default:
4275 		SPDK_UNREACHABLE();
4276 		break;
4277 	}
4278 
4279 	free(ctx);
4280 }
4281 
4282 static void
4283 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4284 {
4285 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4286 	struct spdk_blob *origblob = ctx->original.blob;
4287 
4288 	if (bserrno != 0) {
4289 		if (ctx->bserrno != 0) {
4290 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4291 		} else {
4292 			ctx->bserrno = bserrno;
4293 		}
4294 	}
4295 
4296 	ctx->original.id = origblob->id;
4297 	origblob->locked_operation_in_progress = false;
4298 
4299 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4300 }
4301 
4302 static void
4303 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4304 {
4305 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4306 	struct spdk_blob *origblob = ctx->original.blob;
4307 
4308 	if (bserrno != 0) {
4309 		if (ctx->bserrno != 0) {
4310 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4311 		} else {
4312 			ctx->bserrno = bserrno;
4313 		}
4314 	}
4315 
4316 	if (ctx->frozen) {
4317 		/* Unfreeze any outstanding I/O */
4318 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4319 	} else {
4320 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4321 	}
4322 
4323 }
4324 
4325 static void
4326 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4327 {
4328 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4329 	struct spdk_blob *newblob = ctx->new.blob;
4330 
4331 	if (bserrno != 0) {
4332 		if (ctx->bserrno != 0) {
4333 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4334 		} else {
4335 			ctx->bserrno = bserrno;
4336 		}
4337 	}
4338 
4339 	ctx->new.id = newblob->id;
4340 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4341 }
4342 
4343 /* END blob_cleanup */
4344 
4345 /* START spdk_bs_create_snapshot */
4346 
4347 static void
4348 _spdk_bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
4349 {
4350 	uint64_t *cluster_temp;
4351 
4352 	cluster_temp = blob1->active.clusters;
4353 	blob1->active.clusters = blob2->active.clusters;
4354 	blob2->active.clusters = cluster_temp;
4355 }
4356 
4357 static void
4358 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4359 {
4360 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4361 	struct spdk_blob *origblob = ctx->original.blob;
4362 	struct spdk_blob *newblob = ctx->new.blob;
4363 
4364 	if (bserrno != 0) {
4365 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4366 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4367 		return;
4368 	}
4369 
4370 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4371 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4372 	if (bserrno != 0) {
4373 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4374 		return;
4375 	}
4376 
4377 	_spdk_bs_blob_list_add(ctx->original.blob);
4378 
4379 	spdk_blob_set_read_only(newblob);
4380 
4381 	/* sync snapshot metadata */
4382 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4383 }
4384 
4385 static void
4386 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4387 {
4388 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4389 	struct spdk_blob *origblob = ctx->original.blob;
4390 	struct spdk_blob *newblob = ctx->new.blob;
4391 
4392 	if (bserrno != 0) {
4393 		/* return cluster map back to original */
4394 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4395 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4396 		return;
4397 	}
4398 
4399 	/* Set internal xattr for snapshot id */
4400 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4401 	if (bserrno != 0) {
4402 		/* return cluster map back to original */
4403 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4404 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4405 		return;
4406 	}
4407 
4408 	_spdk_bs_blob_list_remove(origblob);
4409 	origblob->parent_id = newblob->id;
4410 
4411 	/* Create new back_bs_dev for snapshot */
4412 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4413 	if (origblob->back_bs_dev == NULL) {
4414 		/* return cluster map back to original */
4415 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4416 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4417 		return;
4418 	}
4419 
4420 	/* set clone blob as thin provisioned */
4421 	_spdk_blob_set_thin_provision(origblob);
4422 
4423 	_spdk_bs_blob_list_add(newblob);
4424 
4425 	/* sync clone metadata */
4426 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4427 }
4428 
4429 static void
4430 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4431 {
4432 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4433 	struct spdk_blob *origblob = ctx->original.blob;
4434 	struct spdk_blob *newblob = ctx->new.blob;
4435 	int bserrno;
4436 
4437 	if (rc != 0) {
4438 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4439 		return;
4440 	}
4441 
4442 	ctx->frozen = true;
4443 
4444 	/* set new back_bs_dev for snapshot */
4445 	newblob->back_bs_dev = origblob->back_bs_dev;
4446 	/* Set invalid flags from origblob */
4447 	newblob->invalid_flags = origblob->invalid_flags;
4448 
4449 	/* inherit parent from original blob if set */
4450 	newblob->parent_id = origblob->parent_id;
4451 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4452 		/* Set internal xattr for snapshot id */
4453 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4454 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4455 		if (bserrno != 0) {
4456 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4457 			return;
4458 		}
4459 	}
4460 
4461 	/* swap cluster maps */
4462 	_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4463 
4464 	/* sync snapshot metadata */
4465 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4466 }
4467 
4468 static void
4469 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4470 {
4471 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4472 	struct spdk_blob *origblob = ctx->original.blob;
4473 	struct spdk_blob *newblob = _blob;
4474 
4475 	if (bserrno != 0) {
4476 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4477 		return;
4478 	}
4479 
4480 	ctx->new.blob = newblob;
4481 
4482 	/* Zero out newblob cluster map */
4483 	memset(newblob->active.clusters, 0,
4484 	       newblob->active.num_clusters * sizeof(newblob->active.clusters));
4485 
4486 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4487 }
4488 
4489 static void
4490 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4491 {
4492 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4493 	struct spdk_blob *origblob = ctx->original.blob;
4494 
4495 	if (bserrno != 0) {
4496 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4497 		return;
4498 	}
4499 
4500 	ctx->new.id = blobid;
4501 	ctx->cpl.u.blobid.blobid = blobid;
4502 
4503 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4504 }
4505 
4506 
4507 static void
4508 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4509 			const void **value, size_t *value_len)
4510 {
4511 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4512 
4513 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4514 	*value = &blob->id;
4515 	*value_len = sizeof(blob->id);
4516 }
4517 
4518 static void
4519 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4520 {
4521 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4522 	struct spdk_blob_opts opts;
4523 	struct spdk_blob_xattr_opts internal_xattrs;
4524 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4525 
4526 	if (bserrno != 0) {
4527 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4528 		return;
4529 	}
4530 
4531 	ctx->original.blob = _blob;
4532 
4533 	if (_blob->data_ro || _blob->md_ro) {
4534 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4535 			      _blob->id);
4536 		ctx->bserrno = -EINVAL;
4537 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4538 		return;
4539 	}
4540 
4541 	if (_blob->locked_operation_in_progress) {
4542 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot - another operation in progress\n");
4543 		ctx->bserrno = -EBUSY;
4544 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4545 		return;
4546 	}
4547 
4548 	_blob->locked_operation_in_progress = true;
4549 
4550 	spdk_blob_opts_init(&opts);
4551 	_spdk_blob_xattrs_init(&internal_xattrs);
4552 
4553 	/* Change the size of new blob to the same as in original blob,
4554 	 * but do not allocate clusters */
4555 	opts.thin_provision = true;
4556 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4557 
4558 	/* If there are any xattrs specified for snapshot, set them now */
4559 	if (ctx->xattrs) {
4560 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4561 	}
4562 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4563 	internal_xattrs.count = 1;
4564 	internal_xattrs.ctx = _blob;
4565 	internal_xattrs.names = xattrs_names;
4566 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4567 
4568 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4569 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4570 }
4571 
4572 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4573 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4574 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4575 {
4576 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4577 
4578 	if (!ctx) {
4579 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4580 		return;
4581 	}
4582 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4583 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4584 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4585 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4586 	ctx->bserrno = 0;
4587 	ctx->frozen = false;
4588 	ctx->original.id = blobid;
4589 	ctx->xattrs = snapshot_xattrs;
4590 
4591 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4592 }
4593 /* END spdk_bs_create_snapshot */
4594 
4595 /* START spdk_bs_create_clone */
4596 
4597 static void
4598 _spdk_bs_xattr_clone(void *arg, const char *name,
4599 		     const void **value, size_t *value_len)
4600 {
4601 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4602 
4603 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4604 	*value = &blob->id;
4605 	*value_len = sizeof(blob->id);
4606 }
4607 
4608 static void
4609 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4610 {
4611 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4612 	struct spdk_blob *clone = _blob;
4613 
4614 	ctx->new.blob = clone;
4615 	_spdk_bs_blob_list_add(clone);
4616 
4617 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4618 }
4619 
4620 static void
4621 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4622 {
4623 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4624 
4625 	ctx->cpl.u.blobid.blobid = blobid;
4626 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4627 }
4628 
4629 static void
4630 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4631 {
4632 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4633 	struct spdk_blob_opts		opts;
4634 	struct spdk_blob_xattr_opts internal_xattrs;
4635 	char *xattr_names[] = { BLOB_SNAPSHOT };
4636 
4637 	if (bserrno != 0) {
4638 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4639 		return;
4640 	}
4641 
4642 	ctx->original.blob = _blob;
4643 
4644 	if (!_blob->data_ro || !_blob->md_ro) {
4645 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4646 		ctx->bserrno = -EINVAL;
4647 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4648 		return;
4649 	}
4650 
4651 	if (_blob->locked_operation_in_progress) {
4652 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create clone - another operation in progress\n");
4653 		ctx->bserrno = -EBUSY;
4654 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4655 		return;
4656 	}
4657 
4658 	_blob->locked_operation_in_progress = true;
4659 
4660 	spdk_blob_opts_init(&opts);
4661 	_spdk_blob_xattrs_init(&internal_xattrs);
4662 
4663 	opts.thin_provision = true;
4664 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4665 	if (ctx->xattrs) {
4666 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4667 	}
4668 
4669 	/* Set internal xattr BLOB_SNAPSHOT */
4670 	internal_xattrs.count = 1;
4671 	internal_xattrs.ctx = _blob;
4672 	internal_xattrs.names = xattr_names;
4673 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4674 
4675 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4676 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4677 }
4678 
4679 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4680 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4681 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4682 {
4683 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4684 
4685 	if (!ctx) {
4686 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4687 		return;
4688 	}
4689 
4690 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4691 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4692 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4693 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4694 	ctx->bserrno = 0;
4695 	ctx->xattrs = clone_xattrs;
4696 	ctx->original.id = blobid;
4697 
4698 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4699 }
4700 
4701 /* END spdk_bs_create_clone */
4702 
4703 /* START spdk_bs_inflate_blob */
4704 
4705 static void
4706 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4707 {
4708 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4709 	struct spdk_blob *_blob = ctx->original.blob;
4710 
4711 	if (bserrno != 0) {
4712 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4713 		return;
4714 	}
4715 
4716 	assert(_parent != NULL);
4717 
4718 	_spdk_bs_blob_list_remove(_blob);
4719 	_blob->parent_id = _parent->id;
4720 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4721 			     sizeof(spdk_blob_id), true);
4722 
4723 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4724 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4725 	_spdk_bs_blob_list_add(_blob);
4726 
4727 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4728 }
4729 
4730 static void
4731 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4732 {
4733 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4734 	struct spdk_blob *_blob = ctx->original.blob;
4735 	struct spdk_blob *_parent;
4736 
4737 	if (bserrno != 0) {
4738 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4739 		return;
4740 	}
4741 
4742 	if (ctx->allocate_all) {
4743 		/* remove thin provisioning */
4744 		_spdk_bs_blob_list_remove(_blob);
4745 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4746 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4747 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4748 		_blob->back_bs_dev = NULL;
4749 		_blob->parent_id = SPDK_BLOBID_INVALID;
4750 	} else {
4751 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4752 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4753 			/* We must change the parent of the inflated blob */
4754 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4755 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4756 			return;
4757 		}
4758 
4759 		_spdk_bs_blob_list_remove(_blob);
4760 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4761 		_blob->parent_id = SPDK_BLOBID_INVALID;
4762 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4763 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4764 	}
4765 
4766 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4767 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4768 }
4769 
4770 /* Check if cluster needs allocation */
4771 static inline bool
4772 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4773 {
4774 	struct spdk_blob_bs_dev *b;
4775 
4776 	assert(blob != NULL);
4777 
4778 	if (blob->active.clusters[cluster] != 0) {
4779 		/* Cluster is already allocated */
4780 		return false;
4781 	}
4782 
4783 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4784 		/* Blob have no parent blob */
4785 		return allocate_all;
4786 	}
4787 
4788 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4789 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4790 }
4791 
4792 static void
4793 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4794 {
4795 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4796 	struct spdk_blob *_blob = ctx->original.blob;
4797 	uint64_t offset;
4798 
4799 	if (bserrno != 0) {
4800 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4801 		return;
4802 	}
4803 
4804 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4805 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4806 			break;
4807 		}
4808 	}
4809 
4810 	if (ctx->cluster < _blob->active.num_clusters) {
4811 		offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster);
4812 
4813 		/* We may safely increment a cluster before write */
4814 		ctx->cluster++;
4815 
4816 		/* Use zero length write to touch a cluster */
4817 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4818 				   _spdk_bs_inflate_blob_touch_next, ctx);
4819 	} else {
4820 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4821 	}
4822 }
4823 
4824 static void
4825 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4826 {
4827 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4828 	uint64_t lfc; /* lowest free cluster */
4829 	uint64_t i;
4830 
4831 	if (bserrno != 0) {
4832 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4833 		return;
4834 	}
4835 
4836 	ctx->original.blob = _blob;
4837 
4838 	if (_blob->locked_operation_in_progress) {
4839 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot inflate blob - another operation in progress\n");
4840 		ctx->bserrno = -EBUSY;
4841 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4842 		return;
4843 	}
4844 
4845 	_blob->locked_operation_in_progress = true;
4846 
4847 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4848 		/* This blob have no parent, so we cannot decouple it. */
4849 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4850 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4851 		return;
4852 	}
4853 
4854 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4855 		/* This is not thin provisioned blob. No need to inflate. */
4856 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4857 		return;
4858 	}
4859 
4860 	/* Do two passes - one to verify that we can obtain enough clusters
4861 	 * and another to actually claim them.
4862 	 */
4863 	lfc = 0;
4864 	for (i = 0; i < _blob->active.num_clusters; i++) {
4865 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4866 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4867 			if (lfc == UINT32_MAX) {
4868 				/* No more free clusters. Cannot satisfy the request */
4869 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4870 				return;
4871 			}
4872 			lfc++;
4873 		}
4874 	}
4875 
4876 	ctx->cluster = 0;
4877 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4878 }
4879 
4880 static void
4881 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4882 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4883 {
4884 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4885 
4886 	if (!ctx) {
4887 		cb_fn(cb_arg, -ENOMEM);
4888 		return;
4889 	}
4890 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4891 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4892 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4893 	ctx->bserrno = 0;
4894 	ctx->original.id = blobid;
4895 	ctx->channel = channel;
4896 	ctx->allocate_all = allocate_all;
4897 
4898 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4899 }
4900 
4901 void
4902 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4903 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4904 {
4905 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4906 }
4907 
4908 void
4909 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4910 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4911 {
4912 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
4913 }
4914 /* END spdk_bs_inflate_blob */
4915 
4916 /* START spdk_blob_resize */
4917 struct spdk_bs_resize_ctx {
4918 	spdk_blob_op_complete cb_fn;
4919 	void *cb_arg;
4920 	struct spdk_blob *blob;
4921 	uint64_t sz;
4922 	int rc;
4923 };
4924 
4925 static void
4926 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4927 {
4928 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4929 
4930 	if (rc != 0) {
4931 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4932 	}
4933 
4934 	if (ctx->rc != 0) {
4935 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4936 		rc = ctx->rc;
4937 	}
4938 
4939 	ctx->blob->locked_operation_in_progress = false;
4940 
4941 	ctx->cb_fn(ctx->cb_arg, rc);
4942 	free(ctx);
4943 }
4944 
4945 static void
4946 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4947 {
4948 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4949 
4950 	if (rc != 0) {
4951 		ctx->blob->locked_operation_in_progress = false;
4952 		ctx->cb_fn(ctx->cb_arg, rc);
4953 		free(ctx);
4954 		return;
4955 	}
4956 
4957 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4958 
4959 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4960 }
4961 
4962 void
4963 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4964 {
4965 	struct spdk_bs_resize_ctx *ctx;
4966 
4967 	_spdk_blob_verify_md_op(blob);
4968 
4969 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4970 
4971 	if (blob->md_ro) {
4972 		cb_fn(cb_arg, -EPERM);
4973 		return;
4974 	}
4975 
4976 	if (sz == blob->active.num_clusters) {
4977 		cb_fn(cb_arg, 0);
4978 		return;
4979 	}
4980 
4981 	if (blob->locked_operation_in_progress) {
4982 		cb_fn(cb_arg, -EBUSY);
4983 		return;
4984 	}
4985 
4986 	ctx = calloc(1, sizeof(*ctx));
4987 	if (!ctx) {
4988 		cb_fn(cb_arg, -ENOMEM);
4989 		return;
4990 	}
4991 
4992 	blob->locked_operation_in_progress = true;
4993 	ctx->cb_fn = cb_fn;
4994 	ctx->cb_arg = cb_arg;
4995 	ctx->blob = blob;
4996 	ctx->sz = sz;
4997 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
4998 }
4999 
5000 /* END spdk_blob_resize */
5001 
5002 
5003 /* START spdk_bs_delete_blob */
5004 
5005 static void
5006 _spdk_bs_delete_ebusy_close_cpl(void *cb_arg, int bserrno)
5007 {
5008 	spdk_bs_sequence_t *seq = cb_arg;
5009 
5010 	spdk_bs_sequence_finish(seq, -EBUSY);
5011 }
5012 
5013 static void
5014 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
5015 {
5016 	spdk_bs_sequence_t *seq = cb_arg;
5017 
5018 	spdk_bs_sequence_finish(seq, bserrno);
5019 }
5020 
5021 static void
5022 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5023 {
5024 	struct spdk_blob *blob = cb_arg;
5025 
5026 	if (bserrno != 0) {
5027 		/*
5028 		 * We already removed this blob from the blobstore tailq, so
5029 		 *  we need to free it here since this is the last reference
5030 		 *  to it.
5031 		 */
5032 		_spdk_blob_free(blob);
5033 		_spdk_bs_delete_close_cpl(seq, bserrno);
5034 		return;
5035 	}
5036 
5037 	/*
5038 	 * This will immediately decrement the ref_count and call
5039 	 *  the completion routine since the metadata state is clean.
5040 	 *  By calling spdk_blob_close, we reduce the number of call
5041 	 *  points into code that touches the blob->open_ref count
5042 	 *  and the blobstore's blob list.
5043 	 */
5044 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
5045 }
5046 
5047 static void
5048 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
5049 {
5050 	spdk_bs_sequence_t *seq = cb_arg;
5051 	struct spdk_blob_list *snapshot = NULL;
5052 	uint32_t page_num;
5053 
5054 	if (bserrno != 0) {
5055 		spdk_bs_sequence_finish(seq, bserrno);
5056 		return;
5057 	}
5058 
5059 	_spdk_blob_verify_md_op(blob);
5060 
5061 	if (blob->open_ref > 1) {
5062 		/*
5063 		 * Someone has this blob open (besides this delete context).
5064 		 *  Decrement the ref count directly and return -EBUSY.
5065 		 */
5066 		blob->open_ref--;
5067 		spdk_bs_sequence_finish(seq, -EBUSY);
5068 		return;
5069 	}
5070 
5071 	/* Check if this is a snapshot with clones */
5072 	TAILQ_FOREACH(snapshot, &blob->bs->snapshots, link) {
5073 		if (snapshot->id == blob->id) {
5074 			break;
5075 		}
5076 	}
5077 	if (snapshot != NULL) {
5078 		/* If snapshot have clones, we cannot remove it */
5079 		if (!TAILQ_EMPTY(&snapshot->clones)) {
5080 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
5081 			spdk_blob_close(blob, _spdk_bs_delete_ebusy_close_cpl, seq);
5082 			return;
5083 		}
5084 	}
5085 
5086 	_spdk_bs_blob_list_remove(blob);
5087 
5088 	if (blob->locked_operation_in_progress) {
5089 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress\n");
5090 		spdk_blob_close(blob, _spdk_bs_delete_ebusy_close_cpl, seq);
5091 		return;
5092 	}
5093 
5094 	blob->locked_operation_in_progress = true;
5095 
5096 	/*
5097 	 * Remove the blob from the blob_store list now, to ensure it does not
5098 	 *  get returned after this point by _spdk_blob_lookup().
5099 	 */
5100 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5101 
5102 	/* If blob is a snapshot then remove it from the list */
5103 	TAILQ_FOREACH(snapshot, &blob->bs->snapshots, link) {
5104 		if (snapshot->id == blob->id) {
5105 			TAILQ_REMOVE(&blob->bs->snapshots, snapshot, link);
5106 			free(snapshot);
5107 			break;
5108 		}
5109 	}
5110 
5111 	page_num = _spdk_bs_blobid_to_page(blob->id);
5112 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
5113 	blob->state = SPDK_BLOB_STATE_DIRTY;
5114 	blob->active.num_pages = 0;
5115 	_spdk_blob_resize(blob, 0);
5116 
5117 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
5118 }
5119 
5120 void
5121 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5122 		    spdk_blob_op_complete cb_fn, void *cb_arg)
5123 {
5124 	struct spdk_bs_cpl	cpl;
5125 	spdk_bs_sequence_t	*seq;
5126 
5127 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
5128 
5129 	assert(spdk_get_thread() == bs->md_thread);
5130 
5131 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5132 	cpl.u.blob_basic.cb_fn = cb_fn;
5133 	cpl.u.blob_basic.cb_arg = cb_arg;
5134 
5135 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5136 	if (!seq) {
5137 		cb_fn(cb_arg, -ENOMEM);
5138 		return;
5139 	}
5140 
5141 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5142 }
5143 
5144 /* END spdk_bs_delete_blob */
5145 
5146 /* START spdk_bs_open_blob */
5147 
5148 static void
5149 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5150 {
5151 	struct spdk_blob *blob = cb_arg;
5152 
5153 	/* If the blob have crc error, we just return NULL. */
5154 	if (blob == NULL) {
5155 		seq->cpl.u.blob_handle.blob = NULL;
5156 		spdk_bs_sequence_finish(seq, bserrno);
5157 		return;
5158 	}
5159 
5160 	blob->open_ref++;
5161 
5162 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5163 
5164 	spdk_bs_sequence_finish(seq, bserrno);
5165 }
5166 
5167 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5168 			       struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5169 {
5170 	struct spdk_blob		*blob;
5171 	struct spdk_bs_cpl		cpl;
5172 	struct spdk_blob_open_opts	opts_default;
5173 	spdk_bs_sequence_t		*seq;
5174 	uint32_t			page_num;
5175 
5176 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5177 	assert(spdk_get_thread() == bs->md_thread);
5178 
5179 	page_num = _spdk_bs_blobid_to_page(blobid);
5180 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5181 		/* Invalid blobid */
5182 		cb_fn(cb_arg, NULL, -ENOENT);
5183 		return;
5184 	}
5185 
5186 	blob = _spdk_blob_lookup(bs, blobid);
5187 	if (blob) {
5188 		blob->open_ref++;
5189 		cb_fn(cb_arg, blob, 0);
5190 		return;
5191 	}
5192 
5193 	blob = _spdk_blob_alloc(bs, blobid);
5194 	if (!blob) {
5195 		cb_fn(cb_arg, NULL, -ENOMEM);
5196 		return;
5197 	}
5198 
5199 	if (!opts) {
5200 		spdk_blob_open_opts_init(&opts_default);
5201 		opts = &opts_default;
5202 	}
5203 
5204 	blob->clear_method = opts->clear_method;
5205 
5206 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5207 	cpl.u.blob_handle.cb_fn = cb_fn;
5208 	cpl.u.blob_handle.cb_arg = cb_arg;
5209 	cpl.u.blob_handle.blob = blob;
5210 
5211 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5212 	if (!seq) {
5213 		_spdk_blob_free(blob);
5214 		cb_fn(cb_arg, NULL, -ENOMEM);
5215 		return;
5216 	}
5217 
5218 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5219 }
5220 
5221 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5222 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5223 {
5224 	_spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
5225 }
5226 
5227 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
5228 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5229 {
5230 	_spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
5231 }
5232 
5233 /* END spdk_bs_open_blob */
5234 
5235 /* START spdk_blob_set_read_only */
5236 int spdk_blob_set_read_only(struct spdk_blob *blob)
5237 {
5238 	_spdk_blob_verify_md_op(blob);
5239 
5240 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5241 
5242 	blob->state = SPDK_BLOB_STATE_DIRTY;
5243 	return 0;
5244 }
5245 /* END spdk_blob_set_read_only */
5246 
5247 /* START spdk_blob_sync_md */
5248 
5249 static void
5250 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5251 {
5252 	struct spdk_blob *blob = cb_arg;
5253 
5254 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5255 		blob->data_ro = true;
5256 		blob->md_ro = true;
5257 	}
5258 
5259 	spdk_bs_sequence_finish(seq, bserrno);
5260 }
5261 
5262 static void
5263 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5264 {
5265 	struct spdk_bs_cpl	cpl;
5266 	spdk_bs_sequence_t	*seq;
5267 
5268 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5269 	cpl.u.blob_basic.cb_fn = cb_fn;
5270 	cpl.u.blob_basic.cb_arg = cb_arg;
5271 
5272 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5273 	if (!seq) {
5274 		cb_fn(cb_arg, -ENOMEM);
5275 		return;
5276 	}
5277 
5278 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5279 }
5280 
5281 void
5282 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5283 {
5284 	_spdk_blob_verify_md_op(blob);
5285 
5286 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5287 
5288 	if (blob->md_ro) {
5289 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5290 		cb_fn(cb_arg, 0);
5291 		return;
5292 	}
5293 
5294 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5295 }
5296 
5297 /* END spdk_blob_sync_md */
5298 
5299 struct spdk_blob_insert_cluster_ctx {
5300 	struct spdk_thread	*thread;
5301 	struct spdk_blob	*blob;
5302 	uint32_t		cluster_num;	/* cluster index in blob */
5303 	uint32_t		cluster;	/* cluster on disk */
5304 	int			rc;
5305 	spdk_blob_op_complete	cb_fn;
5306 	void			*cb_arg;
5307 };
5308 
5309 static void
5310 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5311 {
5312 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5313 
5314 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5315 	free(ctx);
5316 }
5317 
5318 static void
5319 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5320 {
5321 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5322 
5323 	ctx->rc = bserrno;
5324 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5325 }
5326 
5327 static void
5328 _spdk_blob_insert_cluster_msg(void *arg)
5329 {
5330 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5331 
5332 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5333 	if (ctx->rc != 0) {
5334 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5335 		return;
5336 	}
5337 
5338 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5339 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5340 }
5341 
5342 static void
5343 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5344 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5345 {
5346 	struct spdk_blob_insert_cluster_ctx *ctx;
5347 
5348 	ctx = calloc(1, sizeof(*ctx));
5349 	if (ctx == NULL) {
5350 		cb_fn(cb_arg, -ENOMEM);
5351 		return;
5352 	}
5353 
5354 	ctx->thread = spdk_get_thread();
5355 	ctx->blob = blob;
5356 	ctx->cluster_num = cluster_num;
5357 	ctx->cluster = cluster;
5358 	ctx->cb_fn = cb_fn;
5359 	ctx->cb_arg = cb_arg;
5360 
5361 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5362 }
5363 
5364 /* START spdk_blob_close */
5365 
5366 static void
5367 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5368 {
5369 	struct spdk_blob *blob = cb_arg;
5370 
5371 	if (bserrno == 0) {
5372 		blob->open_ref--;
5373 		if (blob->open_ref == 0) {
5374 			/*
5375 			 * Blobs with active.num_pages == 0 are deleted blobs.
5376 			 *  these blobs are removed from the blob_store list
5377 			 *  when the deletion process starts - so don't try to
5378 			 *  remove them again.
5379 			 */
5380 			if (blob->active.num_pages > 0) {
5381 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5382 			}
5383 			_spdk_blob_free(blob);
5384 		}
5385 	}
5386 
5387 	spdk_bs_sequence_finish(seq, bserrno);
5388 }
5389 
5390 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5391 {
5392 	struct spdk_bs_cpl	cpl;
5393 	spdk_bs_sequence_t	*seq;
5394 
5395 	_spdk_blob_verify_md_op(blob);
5396 
5397 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5398 
5399 	if (blob->open_ref == 0) {
5400 		cb_fn(cb_arg, -EBADF);
5401 		return;
5402 	}
5403 
5404 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5405 	cpl.u.blob_basic.cb_fn = cb_fn;
5406 	cpl.u.blob_basic.cb_arg = cb_arg;
5407 
5408 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5409 	if (!seq) {
5410 		cb_fn(cb_arg, -ENOMEM);
5411 		return;
5412 	}
5413 
5414 	/* Sync metadata */
5415 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5416 }
5417 
5418 /* END spdk_blob_close */
5419 
5420 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5421 {
5422 	return spdk_get_io_channel(bs);
5423 }
5424 
5425 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5426 {
5427 	spdk_put_io_channel(channel);
5428 }
5429 
5430 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5431 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5432 {
5433 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5434 				     SPDK_BLOB_UNMAP);
5435 }
5436 
5437 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5438 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5439 {
5440 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5441 				     SPDK_BLOB_WRITE_ZEROES);
5442 }
5443 
5444 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5445 			void *payload, uint64_t offset, uint64_t length,
5446 			spdk_blob_op_complete cb_fn, void *cb_arg)
5447 {
5448 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5449 				     SPDK_BLOB_WRITE);
5450 }
5451 
5452 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5453 		       void *payload, uint64_t offset, uint64_t length,
5454 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5455 {
5456 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5457 				     SPDK_BLOB_READ);
5458 }
5459 
5460 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5461 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5462 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5463 {
5464 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5465 }
5466 
5467 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5468 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5469 			spdk_blob_op_complete cb_fn, void *cb_arg)
5470 {
5471 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5472 }
5473 
5474 struct spdk_bs_iter_ctx {
5475 	int64_t page_num;
5476 	struct spdk_blob_store *bs;
5477 
5478 	spdk_blob_op_with_handle_complete cb_fn;
5479 	void *cb_arg;
5480 };
5481 
5482 static void
5483 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5484 {
5485 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5486 	struct spdk_blob_store *bs = ctx->bs;
5487 	spdk_blob_id id;
5488 
5489 	if (bserrno == 0) {
5490 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5491 		free(ctx);
5492 		return;
5493 	}
5494 
5495 	ctx->page_num++;
5496 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5497 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5498 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5499 		free(ctx);
5500 		return;
5501 	}
5502 
5503 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5504 
5505 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5506 }
5507 
5508 void
5509 spdk_bs_iter_first(struct spdk_blob_store *bs,
5510 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5511 {
5512 	struct spdk_bs_iter_ctx *ctx;
5513 
5514 	ctx = calloc(1, sizeof(*ctx));
5515 	if (!ctx) {
5516 		cb_fn(cb_arg, NULL, -ENOMEM);
5517 		return;
5518 	}
5519 
5520 	ctx->page_num = -1;
5521 	ctx->bs = bs;
5522 	ctx->cb_fn = cb_fn;
5523 	ctx->cb_arg = cb_arg;
5524 
5525 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5526 }
5527 
5528 static void
5529 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5530 {
5531 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5532 
5533 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5534 }
5535 
5536 void
5537 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5538 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5539 {
5540 	struct spdk_bs_iter_ctx *ctx;
5541 
5542 	assert(blob != NULL);
5543 
5544 	ctx = calloc(1, sizeof(*ctx));
5545 	if (!ctx) {
5546 		cb_fn(cb_arg, NULL, -ENOMEM);
5547 		return;
5548 	}
5549 
5550 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5551 	ctx->bs = bs;
5552 	ctx->cb_fn = cb_fn;
5553 	ctx->cb_arg = cb_arg;
5554 
5555 	/* Close the existing blob */
5556 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5557 }
5558 
5559 static int
5560 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5561 		     uint16_t value_len, bool internal)
5562 {
5563 	struct spdk_xattr_tailq *xattrs;
5564 	struct spdk_xattr	*xattr;
5565 
5566 	_spdk_blob_verify_md_op(blob);
5567 
5568 	if (blob->md_ro) {
5569 		return -EPERM;
5570 	}
5571 
5572 	if (internal) {
5573 		xattrs = &blob->xattrs_internal;
5574 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5575 	} else {
5576 		xattrs = &blob->xattrs;
5577 	}
5578 
5579 	TAILQ_FOREACH(xattr, xattrs, link) {
5580 		if (!strcmp(name, xattr->name)) {
5581 			free(xattr->value);
5582 			xattr->value_len = value_len;
5583 			xattr->value = malloc(value_len);
5584 			memcpy(xattr->value, value, value_len);
5585 
5586 			blob->state = SPDK_BLOB_STATE_DIRTY;
5587 
5588 			return 0;
5589 		}
5590 	}
5591 
5592 	xattr = calloc(1, sizeof(*xattr));
5593 	if (!xattr) {
5594 		return -ENOMEM;
5595 	}
5596 	xattr->name = strdup(name);
5597 	xattr->value_len = value_len;
5598 	xattr->value = malloc(value_len);
5599 	memcpy(xattr->value, value, value_len);
5600 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5601 
5602 	blob->state = SPDK_BLOB_STATE_DIRTY;
5603 
5604 	return 0;
5605 }
5606 
5607 int
5608 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5609 		    uint16_t value_len)
5610 {
5611 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5612 }
5613 
5614 static int
5615 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5616 {
5617 	struct spdk_xattr_tailq *xattrs;
5618 	struct spdk_xattr	*xattr;
5619 
5620 	_spdk_blob_verify_md_op(blob);
5621 
5622 	if (blob->md_ro) {
5623 		return -EPERM;
5624 	}
5625 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5626 
5627 	TAILQ_FOREACH(xattr, xattrs, link) {
5628 		if (!strcmp(name, xattr->name)) {
5629 			TAILQ_REMOVE(xattrs, xattr, link);
5630 			free(xattr->value);
5631 			free(xattr->name);
5632 			free(xattr);
5633 
5634 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5635 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5636 			}
5637 			blob->state = SPDK_BLOB_STATE_DIRTY;
5638 
5639 			return 0;
5640 		}
5641 	}
5642 
5643 	return -ENOENT;
5644 }
5645 
5646 int
5647 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5648 {
5649 	return _spdk_blob_remove_xattr(blob, name, false);
5650 }
5651 
5652 static int
5653 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5654 			   const void **value, size_t *value_len, bool internal)
5655 {
5656 	struct spdk_xattr	*xattr;
5657 	struct spdk_xattr_tailq *xattrs;
5658 
5659 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5660 
5661 	TAILQ_FOREACH(xattr, xattrs, link) {
5662 		if (!strcmp(name, xattr->name)) {
5663 			*value = xattr->value;
5664 			*value_len = xattr->value_len;
5665 			return 0;
5666 		}
5667 	}
5668 	return -ENOENT;
5669 }
5670 
5671 int
5672 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5673 			  const void **value, size_t *value_len)
5674 {
5675 	_spdk_blob_verify_md_op(blob);
5676 
5677 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5678 }
5679 
5680 struct spdk_xattr_names {
5681 	uint32_t	count;
5682 	const char	*names[0];
5683 };
5684 
5685 static int
5686 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5687 {
5688 	struct spdk_xattr	*xattr;
5689 	int			count = 0;
5690 
5691 	TAILQ_FOREACH(xattr, xattrs, link) {
5692 		count++;
5693 	}
5694 
5695 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5696 	if (*names == NULL) {
5697 		return -ENOMEM;
5698 	}
5699 
5700 	TAILQ_FOREACH(xattr, xattrs, link) {
5701 		(*names)->names[(*names)->count++] = xattr->name;
5702 	}
5703 
5704 	return 0;
5705 }
5706 
5707 int
5708 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5709 {
5710 	_spdk_blob_verify_md_op(blob);
5711 
5712 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5713 }
5714 
5715 uint32_t
5716 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5717 {
5718 	assert(names != NULL);
5719 
5720 	return names->count;
5721 }
5722 
5723 const char *
5724 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5725 {
5726 	if (index >= names->count) {
5727 		return NULL;
5728 	}
5729 
5730 	return names->names[index];
5731 }
5732 
5733 void
5734 spdk_xattr_names_free(struct spdk_xattr_names *names)
5735 {
5736 	free(names);
5737 }
5738 
5739 struct spdk_bs_type
5740 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5741 {
5742 	return bs->bstype;
5743 }
5744 
5745 void
5746 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5747 {
5748 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5749 }
5750 
5751 bool
5752 spdk_blob_is_read_only(struct spdk_blob *blob)
5753 {
5754 	assert(blob != NULL);
5755 	return (blob->data_ro || blob->md_ro);
5756 }
5757 
5758 bool
5759 spdk_blob_is_snapshot(struct spdk_blob *blob)
5760 {
5761 	struct spdk_blob_list *snapshot_entry;
5762 
5763 	assert(blob != NULL);
5764 
5765 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
5766 		if (snapshot_entry->id == blob->id) {
5767 			break;
5768 		}
5769 	}
5770 
5771 	if (snapshot_entry == NULL) {
5772 		return false;
5773 	}
5774 
5775 	return true;
5776 }
5777 
5778 bool
5779 spdk_blob_is_clone(struct spdk_blob *blob)
5780 {
5781 	assert(blob != NULL);
5782 
5783 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5784 		assert(spdk_blob_is_thin_provisioned(blob));
5785 		return true;
5786 	}
5787 
5788 	return false;
5789 }
5790 
5791 bool
5792 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5793 {
5794 	assert(blob != NULL);
5795 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5796 }
5797 
5798 spdk_blob_id
5799 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5800 {
5801 	struct spdk_blob_list *snapshot_entry = NULL;
5802 	struct spdk_blob_list *clone_entry = NULL;
5803 
5804 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5805 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5806 			if (clone_entry->id == blob_id) {
5807 				return snapshot_entry->id;
5808 			}
5809 		}
5810 	}
5811 
5812 	return SPDK_BLOBID_INVALID;
5813 }
5814 
5815 int
5816 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5817 		     size_t *count)
5818 {
5819 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5820 	size_t n;
5821 
5822 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5823 		if (snapshot_entry->id == blobid) {
5824 			break;
5825 		}
5826 	}
5827 	if (snapshot_entry == NULL) {
5828 		*count = 0;
5829 		return 0;
5830 	}
5831 
5832 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5833 		*count = snapshot_entry->clone_count;
5834 		return -ENOMEM;
5835 	}
5836 	*count = snapshot_entry->clone_count;
5837 
5838 	n = 0;
5839 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5840 		ids[n++] = clone_entry->id;
5841 	}
5842 
5843 	return 0;
5844 }
5845 
5846 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5847