xref: /spdk/lib/blob/blobstore.c (revision 552e21cce6cccbf833ed9109827e08337377d7ce)
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_dma_malloc(SPDK_BS_PAGE_SIZE,
599 					 SPDK_BS_PAGE_SIZE,
600 					 NULL);
601 	} else {
602 		assert(*pages != NULL);
603 		(*page_count)++;
604 		*pages = spdk_dma_realloc(*pages,
605 					  SPDK_BS_PAGE_SIZE * (*page_count),
606 					  SPDK_BS_PAGE_SIZE,
607 					  NULL);
608 	}
609 
610 	if (*pages == NULL) {
611 		*page_count = 0;
612 		*last_page = NULL;
613 		return -ENOMEM;
614 	}
615 
616 	page = &(*pages)[*page_count - 1];
617 	memset(page, 0, sizeof(*page));
618 	page->id = blob->id;
619 	page->sequence_num = *page_count - 1;
620 	page->next = SPDK_INVALID_MD_PAGE;
621 	*last_page = page;
622 
623 	return 0;
624 }
625 
626 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
627  * Update required_sz on both success and failure.
628  *
629  */
630 static int
631 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
632 			   uint8_t *buf, size_t buf_sz,
633 			   size_t *required_sz, bool internal)
634 {
635 	struct spdk_blob_md_descriptor_xattr	*desc;
636 
637 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
638 		       strlen(xattr->name) +
639 		       xattr->value_len;
640 
641 	if (buf_sz < *required_sz) {
642 		return -1;
643 	}
644 
645 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
646 
647 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
648 	desc->length = sizeof(desc->name_length) +
649 		       sizeof(desc->value_length) +
650 		       strlen(xattr->name) +
651 		       xattr->value_len;
652 	desc->name_length = strlen(xattr->name);
653 	desc->value_length = xattr->value_len;
654 
655 	memcpy(desc->name, xattr->name, desc->name_length);
656 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
657 	       xattr->value,
658 	       desc->value_length);
659 
660 	return 0;
661 }
662 
663 static void
664 _spdk_blob_serialize_extent(const struct spdk_blob *blob,
665 			    uint64_t start_cluster, uint64_t *next_cluster,
666 			    uint8_t *buf, size_t buf_sz)
667 {
668 	struct spdk_blob_md_descriptor_extent *desc;
669 	size_t cur_sz;
670 	uint64_t i, extent_idx;
671 	uint64_t lba, lba_per_cluster, lba_count;
672 
673 	/* The buffer must have room for at least one extent */
674 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]);
675 	if (buf_sz < cur_sz) {
676 		*next_cluster = start_cluster;
677 		return;
678 	}
679 
680 	desc = (struct spdk_blob_md_descriptor_extent *)buf;
681 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT;
682 
683 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
684 
685 	lba = blob->active.clusters[start_cluster];
686 	lba_count = lba_per_cluster;
687 	extent_idx = 0;
688 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
689 		if ((lba + lba_count) == blob->active.clusters[i]) {
690 			lba_count += lba_per_cluster;
691 			continue;
692 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
693 			lba_count += lba_per_cluster;
694 			continue;
695 		}
696 		desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
697 		desc->extents[extent_idx].length = lba_count / lba_per_cluster;
698 		extent_idx++;
699 
700 		cur_sz += sizeof(desc->extents[extent_idx]);
701 
702 		if (buf_sz < cur_sz) {
703 			/* If we ran out of buffer space, return */
704 			desc->length = sizeof(desc->extents[0]) * extent_idx;
705 			*next_cluster = i;
706 			return;
707 		}
708 
709 		lba = blob->active.clusters[i];
710 		lba_count = lba_per_cluster;
711 	}
712 
713 	desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
714 	desc->extents[extent_idx].length = lba_count / lba_per_cluster;
715 	extent_idx++;
716 
717 	desc->length = sizeof(desc->extents[0]) * extent_idx;
718 	*next_cluster = blob->active.num_clusters;
719 
720 	return;
721 }
722 
723 static void
724 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
725 			   uint8_t *buf, size_t *buf_sz)
726 {
727 	struct spdk_blob_md_descriptor_flags *desc;
728 
729 	/*
730 	 * Flags get serialized first, so we should always have room for the flags
731 	 *  descriptor.
732 	 */
733 	assert(*buf_sz >= sizeof(*desc));
734 
735 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
736 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
737 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
738 	desc->invalid_flags = blob->invalid_flags;
739 	desc->data_ro_flags = blob->data_ro_flags;
740 	desc->md_ro_flags = blob->md_ro_flags;
741 
742 	*buf_sz -= sizeof(*desc);
743 }
744 
745 static int
746 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
747 			    const struct spdk_xattr_tailq *xattrs, bool internal,
748 			    struct spdk_blob_md_page **pages,
749 			    struct spdk_blob_md_page *cur_page,
750 			    uint32_t *page_count, uint8_t **buf,
751 			    size_t *remaining_sz)
752 {
753 	const struct spdk_xattr	*xattr;
754 	int	rc;
755 
756 	TAILQ_FOREACH(xattr, xattrs, link) {
757 		size_t required_sz = 0;
758 
759 		rc = _spdk_blob_serialize_xattr(xattr,
760 						*buf, *remaining_sz,
761 						&required_sz, internal);
762 		if (rc < 0) {
763 			/* Need to add a new page to the chain */
764 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
765 							   &cur_page);
766 			if (rc < 0) {
767 				spdk_dma_free(*pages);
768 				*pages = NULL;
769 				*page_count = 0;
770 				return rc;
771 			}
772 
773 			*buf = (uint8_t *)cur_page->descriptors;
774 			*remaining_sz = sizeof(cur_page->descriptors);
775 
776 			/* Try again */
777 			required_sz = 0;
778 			rc = _spdk_blob_serialize_xattr(xattr,
779 							*buf, *remaining_sz,
780 							&required_sz, internal);
781 
782 			if (rc < 0) {
783 				spdk_dma_free(*pages);
784 				*pages = NULL;
785 				*page_count = 0;
786 				return rc;
787 			}
788 		}
789 
790 		*remaining_sz -= required_sz;
791 		*buf += required_sz;
792 	}
793 
794 	return 0;
795 }
796 
797 static int
798 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
799 		     uint32_t *page_count)
800 {
801 	struct spdk_blob_md_page		*cur_page;
802 	int					rc;
803 	uint8_t					*buf;
804 	size_t					remaining_sz;
805 	uint64_t				last_cluster;
806 
807 	assert(pages != NULL);
808 	assert(page_count != NULL);
809 	assert(blob != NULL);
810 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
811 
812 	*pages = NULL;
813 	*page_count = 0;
814 
815 	/* A blob always has at least 1 page, even if it has no descriptors */
816 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
817 	if (rc < 0) {
818 		return rc;
819 	}
820 
821 	buf = (uint8_t *)cur_page->descriptors;
822 	remaining_sz = sizeof(cur_page->descriptors);
823 
824 	/* Serialize flags */
825 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
826 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
827 
828 	/* Serialize xattrs */
829 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
830 					 pages, cur_page, page_count, &buf, &remaining_sz);
831 	if (rc < 0) {
832 		return rc;
833 	}
834 
835 	/* Serialize internal xattrs */
836 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
837 					 pages, cur_page, page_count, &buf, &remaining_sz);
838 	if (rc < 0) {
839 		return rc;
840 	}
841 
842 	/* Serialize extents */
843 	last_cluster = 0;
844 	while (last_cluster < blob->active.num_clusters) {
845 		_spdk_blob_serialize_extent(blob, last_cluster, &last_cluster,
846 					    buf, remaining_sz);
847 
848 		if (last_cluster == blob->active.num_clusters) {
849 			break;
850 		}
851 
852 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
853 						   &cur_page);
854 		if (rc < 0) {
855 			return rc;
856 		}
857 
858 		buf = (uint8_t *)cur_page->descriptors;
859 		remaining_sz = sizeof(cur_page->descriptors);
860 	}
861 
862 	return 0;
863 }
864 
865 struct spdk_blob_load_ctx {
866 	struct spdk_blob		*blob;
867 
868 	struct spdk_blob_md_page	*pages;
869 	uint32_t			num_pages;
870 	spdk_bs_sequence_t	        *seq;
871 
872 	spdk_bs_sequence_cpl		cb_fn;
873 	void				*cb_arg;
874 };
875 
876 static uint32_t
877 _spdk_blob_md_page_calc_crc(void *page)
878 {
879 	uint32_t		crc;
880 
881 	crc = BLOB_CRC32C_INITIAL;
882 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
883 	crc ^= BLOB_CRC32C_INITIAL;
884 
885 	return crc;
886 
887 }
888 
889 static void
890 _spdk_blob_load_final(void *cb_arg, int bserrno)
891 {
892 	struct spdk_blob_load_ctx	*ctx = cb_arg;
893 	struct spdk_blob		*blob = ctx->blob;
894 
895 	_spdk_blob_mark_clean(blob);
896 
897 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
898 
899 	/* Free the memory */
900 	spdk_dma_free(ctx->pages);
901 	free(ctx);
902 }
903 
904 static void
905 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
906 {
907 	struct spdk_blob_load_ctx	*ctx = cb_arg;
908 	struct spdk_blob		*blob = ctx->blob;
909 
910 	if (bserrno != 0) {
911 		goto error;
912 	}
913 
914 	blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
915 
916 	if (blob->back_bs_dev == NULL) {
917 		bserrno = -ENOMEM;
918 		goto error;
919 	}
920 
921 	_spdk_blob_load_final(ctx, bserrno);
922 	return;
923 
924 error:
925 	SPDK_ERRLOG("Snapshot fail\n");
926 	_spdk_blob_free(blob);
927 	ctx->cb_fn(ctx->seq, NULL, bserrno);
928 	spdk_dma_free(ctx->pages);
929 	free(ctx);
930 }
931 
932 static void
933 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
934 {
935 	struct spdk_blob_load_ctx	*ctx = cb_arg;
936 	struct spdk_blob		*blob = ctx->blob;
937 	struct spdk_blob_md_page	*page;
938 	const void			*value;
939 	size_t				len;
940 	int				rc;
941 	uint32_t			crc;
942 
943 	if (bserrno) {
944 		SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno);
945 		_spdk_blob_free(blob);
946 		ctx->cb_fn(seq, NULL, bserrno);
947 		spdk_dma_free(ctx->pages);
948 		free(ctx);
949 		return;
950 	}
951 
952 	page = &ctx->pages[ctx->num_pages - 1];
953 	crc = _spdk_blob_md_page_calc_crc(page);
954 	if (crc != page->crc) {
955 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
956 		_spdk_blob_free(blob);
957 		ctx->cb_fn(seq, NULL, -EINVAL);
958 		spdk_dma_free(ctx->pages);
959 		free(ctx);
960 		return;
961 	}
962 
963 	if (page->next != SPDK_INVALID_MD_PAGE) {
964 		uint32_t next_page = page->next;
965 		uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page);
966 
967 
968 		assert(next_lba < (blob->bs->md_start + blob->bs->md_len));
969 
970 		/* Read the next page */
971 		ctx->num_pages++;
972 		ctx->pages = spdk_dma_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
973 					      sizeof(*page), NULL);
974 		if (ctx->pages == NULL) {
975 			ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM);
976 			free(ctx);
977 			return;
978 		}
979 
980 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
981 					  next_lba,
982 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
983 					  _spdk_blob_load_cpl, ctx);
984 		return;
985 	}
986 
987 	/* Parse the pages */
988 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
989 	if (rc) {
990 		_spdk_blob_free(blob);
991 		ctx->cb_fn(seq, NULL, rc);
992 		spdk_dma_free(ctx->pages);
993 		free(ctx);
994 		return;
995 	}
996 	ctx->seq = seq;
997 
998 
999 	if (spdk_blob_is_thin_provisioned(blob)) {
1000 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
1001 		if (rc == 0) {
1002 			if (len != sizeof(spdk_blob_id)) {
1003 				_spdk_blob_free(blob);
1004 				ctx->cb_fn(seq, NULL, -EINVAL);
1005 				spdk_dma_free(ctx->pages);
1006 				free(ctx);
1007 				return;
1008 			}
1009 			/* open snapshot blob and continue in the callback function */
1010 			blob->parent_id = *(spdk_blob_id *)value;
1011 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1012 					  _spdk_blob_load_snapshot_cpl, ctx);
1013 			return;
1014 		} else {
1015 			/* add zeroes_dev for thin provisioned blob */
1016 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1017 		}
1018 	} else {
1019 		/* standard blob */
1020 		blob->back_bs_dev = NULL;
1021 	}
1022 	_spdk_blob_load_final(ctx, bserrno);
1023 }
1024 
1025 /* Load a blob from disk given a blobid */
1026 static void
1027 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1028 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1029 {
1030 	struct spdk_blob_load_ctx *ctx;
1031 	struct spdk_blob_store *bs;
1032 	uint32_t page_num;
1033 	uint64_t lba;
1034 
1035 	_spdk_blob_verify_md_op(blob);
1036 
1037 	bs = blob->bs;
1038 
1039 	ctx = calloc(1, sizeof(*ctx));
1040 	if (!ctx) {
1041 		cb_fn(seq, cb_arg, -ENOMEM);
1042 		return;
1043 	}
1044 
1045 	ctx->blob = blob;
1046 	ctx->pages = spdk_dma_realloc(ctx->pages, SPDK_BS_PAGE_SIZE,
1047 				      SPDK_BS_PAGE_SIZE, NULL);
1048 	if (!ctx->pages) {
1049 		free(ctx);
1050 		cb_fn(seq, cb_arg, -ENOMEM);
1051 		return;
1052 	}
1053 	ctx->num_pages = 1;
1054 	ctx->cb_fn = cb_fn;
1055 	ctx->cb_arg = cb_arg;
1056 
1057 	page_num = _spdk_bs_blobid_to_page(blob->id);
1058 	lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num);
1059 
1060 	blob->state = SPDK_BLOB_STATE_LOADING;
1061 
1062 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1063 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1064 				  _spdk_blob_load_cpl, ctx);
1065 }
1066 
1067 struct spdk_blob_persist_ctx {
1068 	struct spdk_blob		*blob;
1069 
1070 	struct spdk_bs_super_block	*super;
1071 
1072 	struct spdk_blob_md_page	*pages;
1073 
1074 	uint64_t			idx;
1075 
1076 	spdk_bs_sequence_t		*seq;
1077 	spdk_bs_sequence_cpl		cb_fn;
1078 	void				*cb_arg;
1079 };
1080 
1081 static void
1082 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1083 			uint32_t lba_count)
1084 {
1085 	if (ctx->blob->clear_method == BLOB_CLEAR_WITH_DEFAULT ||
1086 	    ctx->blob->clear_method == BLOB_CLEAR_WITH_UNMAP) {
1087 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1088 	} else if (ctx->blob->clear_method == BLOB_CLEAR_WITH_WRITE_ZEROES) {
1089 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1090 	}
1091 }
1092 
1093 static void
1094 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1095 {
1096 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1097 	struct spdk_blob		*blob = ctx->blob;
1098 
1099 	if (bserrno == 0) {
1100 		_spdk_blob_mark_clean(blob);
1101 	}
1102 
1103 	/* Call user callback */
1104 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1105 
1106 	/* Free the memory */
1107 	spdk_dma_free(ctx->pages);
1108 	free(ctx);
1109 }
1110 
1111 static void
1112 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1113 {
1114 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1115 	struct spdk_blob		*blob = ctx->blob;
1116 	struct spdk_blob_store		*bs = blob->bs;
1117 	void				*tmp;
1118 	size_t				i;
1119 
1120 	/* Release all clusters that were truncated */
1121 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1122 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1123 
1124 		/* Nothing to release if it was not allocated */
1125 		if (blob->active.clusters[i] != 0) {
1126 			_spdk_bs_release_cluster(bs, cluster_num);
1127 		}
1128 	}
1129 
1130 	if (blob->active.num_clusters == 0) {
1131 		free(blob->active.clusters);
1132 		blob->active.clusters = NULL;
1133 		blob->active.cluster_array_size = 0;
1134 	} else {
1135 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters);
1136 		assert(tmp != NULL);
1137 		blob->active.clusters = tmp;
1138 		blob->active.cluster_array_size = blob->active.num_clusters;
1139 	}
1140 
1141 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1142 }
1143 
1144 static void
1145 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1146 {
1147 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1148 	struct spdk_blob		*blob = ctx->blob;
1149 	struct spdk_blob_store		*bs = blob->bs;
1150 	spdk_bs_batch_t			*batch;
1151 	size_t				i;
1152 	uint64_t			lba;
1153 	uint32_t			lba_count;
1154 
1155 	/* Clusters don't move around in blobs. The list shrinks or grows
1156 	 * at the end, but no changes ever occur in the middle of the list.
1157 	 */
1158 
1159 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx);
1160 
1161 	/* Clear all clusters that were truncated */
1162 	lba = 0;
1163 	lba_count = 0;
1164 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1165 		uint64_t next_lba = blob->active.clusters[i];
1166 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1167 
1168 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1169 			/* This cluster is contiguous with the previous one. */
1170 			lba_count += next_lba_count;
1171 			continue;
1172 		}
1173 
1174 		/* This cluster is not contiguous with the previous one. */
1175 
1176 		/* If a run of LBAs previously existing, clear them now */
1177 		if (lba_count > 0) {
1178 			spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1179 		}
1180 
1181 		/* Start building the next batch */
1182 		lba = next_lba;
1183 		if (next_lba > 0) {
1184 			lba_count = next_lba_count;
1185 		} else {
1186 			lba_count = 0;
1187 		}
1188 	}
1189 
1190 	/* If we ended with a contiguous set of LBAs, clear them now */
1191 	if (lba_count > 0) {
1192 		spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1193 	}
1194 
1195 	spdk_bs_batch_close(batch);
1196 }
1197 
1198 static void
1199 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1200 {
1201 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1202 	struct spdk_blob		*blob = ctx->blob;
1203 	struct spdk_blob_store		*bs = blob->bs;
1204 	size_t				i;
1205 
1206 	/* This loop starts at 1 because the first page is special and handled
1207 	 * below. The pages (except the first) are never written in place,
1208 	 * so any pages in the clean list must be zeroed.
1209 	 */
1210 	for (i = 1; i < blob->clean.num_pages; i++) {
1211 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1212 	}
1213 
1214 	if (blob->active.num_pages == 0) {
1215 		uint32_t page_num;
1216 
1217 		page_num = _spdk_bs_blobid_to_page(blob->id);
1218 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1219 	}
1220 
1221 	/* Move on to clearing clusters */
1222 	_spdk_blob_persist_clear_clusters(seq, ctx, 0);
1223 }
1224 
1225 static void
1226 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1227 {
1228 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1229 	struct spdk_blob		*blob = ctx->blob;
1230 	struct spdk_blob_store		*bs = blob->bs;
1231 	uint64_t			lba;
1232 	uint32_t			lba_count;
1233 	spdk_bs_batch_t			*batch;
1234 	size_t				i;
1235 
1236 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1237 
1238 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1239 
1240 	/* This loop starts at 1 because the first page is special and handled
1241 	 * below. The pages (except the first) are never written in place,
1242 	 * so any pages in the clean list must be zeroed.
1243 	 */
1244 	for (i = 1; i < blob->clean.num_pages; i++) {
1245 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);
1246 
1247 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1248 	}
1249 
1250 	/* The first page will only be zeroed if this is a delete. */
1251 	if (blob->active.num_pages == 0) {
1252 		uint32_t page_num;
1253 
1254 		/* The first page in the metadata goes where the blobid indicates */
1255 		page_num = _spdk_bs_blobid_to_page(blob->id);
1256 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);
1257 
1258 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1259 	}
1260 
1261 	spdk_bs_batch_close(batch);
1262 }
1263 
1264 static void
1265 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1266 {
1267 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1268 	struct spdk_blob		*blob = ctx->blob;
1269 	struct spdk_blob_store		*bs = blob->bs;
1270 	uint64_t			lba;
1271 	uint32_t			lba_count;
1272 	struct spdk_blob_md_page	*page;
1273 
1274 	if (blob->active.num_pages == 0) {
1275 		/* Move on to the next step */
1276 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1277 		return;
1278 	}
1279 
1280 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1281 
1282 	page = &ctx->pages[0];
1283 	/* The first page in the metadata goes where the blobid indicates */
1284 	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));
1285 
1286 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1287 				   _spdk_blob_persist_zero_pages, ctx);
1288 }
1289 
1290 static void
1291 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1292 {
1293 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1294 	struct spdk_blob		*blob = ctx->blob;
1295 	struct spdk_blob_store		*bs = blob->bs;
1296 	uint64_t			lba;
1297 	uint32_t			lba_count;
1298 	struct spdk_blob_md_page	*page;
1299 	spdk_bs_batch_t			*batch;
1300 	size_t				i;
1301 
1302 	/* Clusters don't move around in blobs. The list shrinks or grows
1303 	 * at the end, but no changes ever occur in the middle of the list.
1304 	 */
1305 
1306 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1307 
1308 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1309 
1310 	/* This starts at 1. The root page is not written until
1311 	 * all of the others are finished
1312 	 */
1313 	for (i = 1; i < blob->active.num_pages; i++) {
1314 		page = &ctx->pages[i];
1315 		assert(page->sequence_num == i);
1316 
1317 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]);
1318 
1319 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1320 	}
1321 
1322 	spdk_bs_batch_close(batch);
1323 }
1324 
1325 static int
1326 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1327 {
1328 	uint64_t	i;
1329 	uint64_t	*tmp;
1330 	uint64_t	lfc; /* lowest free cluster */
1331 	uint64_t	num_clusters;
1332 	struct spdk_blob_store *bs;
1333 
1334 	bs = blob->bs;
1335 
1336 	_spdk_blob_verify_md_op(blob);
1337 
1338 	if (blob->active.num_clusters == sz) {
1339 		return 0;
1340 	}
1341 
1342 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1343 		/* If this blob was resized to be larger, then smaller, then
1344 		 * larger without syncing, then the cluster array already
1345 		 * contains spare assigned clusters we can use.
1346 		 */
1347 		num_clusters = spdk_min(blob->active.cluster_array_size,
1348 					sz);
1349 	} else {
1350 		num_clusters = blob->active.num_clusters;
1351 	}
1352 
1353 	/* Do two passes - one to verify that we can obtain enough clusters
1354 	 * and another to actually claim them.
1355 	 */
1356 
1357 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1358 		lfc = 0;
1359 		for (i = num_clusters; i < sz; i++) {
1360 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1361 			if (lfc == UINT32_MAX) {
1362 				/* No more free clusters. Cannot satisfy the request */
1363 				return -ENOSPC;
1364 			}
1365 			lfc++;
1366 		}
1367 	}
1368 
1369 	if (sz > num_clusters) {
1370 		/* Expand the cluster array if necessary.
1371 		 * We only shrink the array when persisting.
1372 		 */
1373 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz);
1374 		if (sz > 0 && tmp == NULL) {
1375 			return -ENOMEM;
1376 		}
1377 		memset(tmp + blob->active.cluster_array_size, 0,
1378 		       sizeof(uint64_t) * (sz - blob->active.cluster_array_size));
1379 		blob->active.clusters = tmp;
1380 		blob->active.cluster_array_size = sz;
1381 	}
1382 
1383 	blob->state = SPDK_BLOB_STATE_DIRTY;
1384 
1385 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1386 		lfc = 0;
1387 		for (i = num_clusters; i < sz; i++) {
1388 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1389 			lfc++;
1390 		}
1391 	}
1392 
1393 	blob->active.num_clusters = sz;
1394 
1395 	return 0;
1396 }
1397 
1398 static void
1399 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1400 {
1401 	spdk_bs_sequence_t *seq = ctx->seq;
1402 	struct spdk_blob *blob = ctx->blob;
1403 	struct spdk_blob_store *bs = blob->bs;
1404 	uint64_t i;
1405 	uint32_t page_num;
1406 	void *tmp;
1407 	int rc;
1408 
1409 	if (blob->active.num_pages == 0) {
1410 		/* This is the signal that the blob should be deleted.
1411 		 * Immediately jump to the clean up routine. */
1412 		assert(blob->clean.num_pages > 0);
1413 		ctx->idx = blob->clean.num_pages - 1;
1414 		blob->state = SPDK_BLOB_STATE_CLEAN;
1415 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1416 		return;
1417 
1418 	}
1419 
1420 	/* Generate the new metadata */
1421 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1422 	if (rc < 0) {
1423 		_spdk_blob_persist_complete(seq, ctx, rc);
1424 		return;
1425 	}
1426 
1427 	assert(blob->active.num_pages >= 1);
1428 
1429 	/* Resize the cache of page indices */
1430 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1431 	if (!tmp) {
1432 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1433 		return;
1434 	}
1435 	blob->active.pages = tmp;
1436 
1437 	/* Assign this metadata to pages. This requires two passes -
1438 	 * one to verify that there are enough pages and a second
1439 	 * to actually claim them. */
1440 	page_num = 0;
1441 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1442 	for (i = 1; i < blob->active.num_pages; i++) {
1443 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1444 		if (page_num == UINT32_MAX) {
1445 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1446 			return;
1447 		}
1448 		page_num++;
1449 	}
1450 
1451 	page_num = 0;
1452 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1453 	for (i = 1; i < blob->active.num_pages; i++) {
1454 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1455 		ctx->pages[i - 1].next = page_num;
1456 		/* Now that previous metadata page is complete, calculate the crc for it. */
1457 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1458 		blob->active.pages[i] = page_num;
1459 		spdk_bit_array_set(bs->used_md_pages, page_num);
1460 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1461 		page_num++;
1462 	}
1463 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1464 	/* Start writing the metadata from last page to first */
1465 	ctx->idx = blob->active.num_pages - 1;
1466 	blob->state = SPDK_BLOB_STATE_CLEAN;
1467 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1468 }
1469 
1470 static void
1471 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1472 {
1473 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1474 
1475 	ctx->blob->bs->clean = 0;
1476 
1477 	spdk_dma_free(ctx->super);
1478 
1479 	_spdk_blob_persist_start(ctx);
1480 }
1481 
1482 static void
1483 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1484 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1485 
1486 
1487 static void
1488 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1489 {
1490 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1491 
1492 	ctx->super->clean = 0;
1493 	if (ctx->super->size == 0) {
1494 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1495 	}
1496 
1497 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1498 }
1499 
1500 
1501 /* Write a blob to disk */
1502 static void
1503 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1504 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1505 {
1506 	struct spdk_blob_persist_ctx *ctx;
1507 
1508 	_spdk_blob_verify_md_op(blob);
1509 
1510 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1511 		cb_fn(seq, cb_arg, 0);
1512 		return;
1513 	}
1514 
1515 	ctx = calloc(1, sizeof(*ctx));
1516 	if (!ctx) {
1517 		cb_fn(seq, cb_arg, -ENOMEM);
1518 		return;
1519 	}
1520 	ctx->blob = blob;
1521 	ctx->seq = seq;
1522 	ctx->cb_fn = cb_fn;
1523 	ctx->cb_arg = cb_arg;
1524 
1525 	if (blob->bs->clean) {
1526 		ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
1527 		if (!ctx->super) {
1528 			cb_fn(seq, cb_arg, -ENOMEM);
1529 			free(ctx);
1530 			return;
1531 		}
1532 
1533 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1534 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1535 					  _spdk_blob_persist_dirty, ctx);
1536 	} else {
1537 		_spdk_blob_persist_start(ctx);
1538 	}
1539 }
1540 
1541 struct spdk_blob_copy_cluster_ctx {
1542 	struct spdk_blob *blob;
1543 	uint8_t *buf;
1544 	uint64_t page;
1545 	uint64_t new_cluster;
1546 	spdk_bs_sequence_t *seq;
1547 };
1548 
1549 static void
1550 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1551 {
1552 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1553 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1554 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1555 	spdk_bs_user_op_t *op;
1556 
1557 	TAILQ_INIT(&requests);
1558 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1559 
1560 	while (!TAILQ_EMPTY(&requests)) {
1561 		op = TAILQ_FIRST(&requests);
1562 		TAILQ_REMOVE(&requests, op, link);
1563 		if (bserrno == 0) {
1564 			spdk_bs_user_op_execute(op);
1565 		} else {
1566 			spdk_bs_user_op_abort(op);
1567 		}
1568 	}
1569 
1570 	spdk_dma_free(ctx->buf);
1571 	free(ctx);
1572 }
1573 
1574 static void
1575 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1576 {
1577 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1578 
1579 	if (bserrno) {
1580 		uint32_t cluster_number;
1581 
1582 		if (bserrno == -EEXIST) {
1583 			/* The metadata insert failed because another thread
1584 			 * allocated the cluster first. Free our cluster
1585 			 * but continue without error. */
1586 			bserrno = 0;
1587 		}
1588 
1589 		cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1590 		_spdk_bs_release_cluster(ctx->blob->bs, cluster_number);
1591 	}
1592 
1593 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1594 }
1595 
1596 static void
1597 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1598 {
1599 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1600 	uint32_t cluster_number;
1601 
1602 	if (bserrno) {
1603 		/* The write failed, so jump to the final completion handler */
1604 		spdk_bs_sequence_finish(seq, bserrno);
1605 		return;
1606 	}
1607 
1608 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1609 
1610 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1611 					       _spdk_blob_insert_cluster_cpl, ctx);
1612 }
1613 
1614 static void
1615 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1616 {
1617 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1618 
1619 	if (bserrno != 0) {
1620 		/* The read failed, so jump to the final completion handler */
1621 		spdk_bs_sequence_finish(seq, bserrno);
1622 		return;
1623 	}
1624 
1625 	/* Write whole cluster */
1626 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1627 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1628 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1629 				   _spdk_blob_write_copy_cpl, ctx);
1630 }
1631 
1632 static void
1633 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1634 				   struct spdk_io_channel *_ch,
1635 				   uint64_t io_unit, spdk_bs_user_op_t *op)
1636 {
1637 	struct spdk_bs_cpl cpl;
1638 	struct spdk_bs_channel *ch;
1639 	struct spdk_blob_copy_cluster_ctx *ctx;
1640 	uint32_t cluster_start_page;
1641 	uint32_t cluster_number;
1642 	int rc;
1643 
1644 	ch = spdk_io_channel_get_ctx(_ch);
1645 
1646 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1647 		/* There are already operations pending. Queue this user op
1648 		 * and return because it will be re-executed when the outstanding
1649 		 * cluster allocation completes. */
1650 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1651 		return;
1652 	}
1653 
1654 	/* Round the io_unit offset down to the first page in the cluster */
1655 	cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit);
1656 
1657 	/* Calculate which index in the metadata cluster array the corresponding
1658 	 * cluster is supposed to be at. */
1659 	cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit);
1660 
1661 	ctx = calloc(1, sizeof(*ctx));
1662 	if (!ctx) {
1663 		spdk_bs_user_op_abort(op);
1664 		return;
1665 	}
1666 
1667 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1668 
1669 	ctx->blob = blob;
1670 	ctx->page = cluster_start_page;
1671 
1672 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1673 		ctx->buf = spdk_dma_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, NULL);
1674 		if (!ctx->buf) {
1675 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1676 				    blob->bs->cluster_sz);
1677 			free(ctx);
1678 			spdk_bs_user_op_abort(op);
1679 			return;
1680 		}
1681 	}
1682 
1683 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1684 	if (rc != 0) {
1685 		spdk_dma_free(ctx->buf);
1686 		free(ctx);
1687 		spdk_bs_user_op_abort(op);
1688 		return;
1689 	}
1690 
1691 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1692 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1693 	cpl.u.blob_basic.cb_arg = ctx;
1694 
1695 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1696 	if (!ctx->seq) {
1697 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1698 		spdk_dma_free(ctx->buf);
1699 		free(ctx);
1700 		spdk_bs_user_op_abort(op);
1701 		return;
1702 	}
1703 
1704 	/* Queue the user op to block other incoming operations */
1705 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1706 
1707 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1708 		/* Read cluster from backing device */
1709 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1710 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1711 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1712 					     _spdk_blob_write_copy, ctx);
1713 	} else {
1714 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1715 						       _spdk_blob_insert_cluster_cpl, ctx);
1716 	}
1717 }
1718 
1719 static void
1720 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
1721 				       uint64_t *lba,	uint32_t *lba_count)
1722 {
1723 	*lba_count = length;
1724 
1725 	if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) {
1726 		assert(blob->back_bs_dev != NULL);
1727 		*lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit);
1728 		*lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count);
1729 	} else {
1730 		*lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit);
1731 	}
1732 }
1733 
1734 struct op_split_ctx {
1735 	struct spdk_blob *blob;
1736 	struct spdk_io_channel *channel;
1737 	uint64_t io_unit_offset;
1738 	uint64_t io_units_remaining;
1739 	void *curr_payload;
1740 	enum spdk_blob_op_type op_type;
1741 	spdk_bs_sequence_t *seq;
1742 };
1743 
1744 static void
1745 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1746 {
1747 	struct op_split_ctx	*ctx = cb_arg;
1748 	struct spdk_blob	*blob = ctx->blob;
1749 	struct spdk_io_channel	*ch = ctx->channel;
1750 	enum spdk_blob_op_type	op_type = ctx->op_type;
1751 	uint8_t			*buf = ctx->curr_payload;
1752 	uint64_t		offset = ctx->io_unit_offset;
1753 	uint64_t		length = ctx->io_units_remaining;
1754 	uint64_t		op_length;
1755 
1756 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
1757 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1758 		free(ctx);
1759 		return;
1760 	}
1761 
1762 	op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob,
1763 			     offset));
1764 
1765 	/* Update length and payload for next operation */
1766 	ctx->io_units_remaining -= op_length;
1767 	ctx->io_unit_offset += op_length;
1768 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1769 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
1770 	}
1771 
1772 	switch (op_type) {
1773 	case SPDK_BLOB_READ:
1774 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1775 				  _spdk_blob_request_submit_op_split_next, ctx);
1776 		break;
1777 	case SPDK_BLOB_WRITE:
1778 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1779 				   _spdk_blob_request_submit_op_split_next, ctx);
1780 		break;
1781 	case SPDK_BLOB_UNMAP:
1782 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1783 				   _spdk_blob_request_submit_op_split_next, ctx);
1784 		break;
1785 	case SPDK_BLOB_WRITE_ZEROES:
1786 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1787 					  _spdk_blob_request_submit_op_split_next, ctx);
1788 		break;
1789 	case SPDK_BLOB_READV:
1790 	case SPDK_BLOB_WRITEV:
1791 		SPDK_ERRLOG("readv/write not valid for %s\n", __func__);
1792 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1793 		free(ctx);
1794 		break;
1795 	}
1796 }
1797 
1798 static void
1799 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1800 				   void *payload, uint64_t offset, uint64_t length,
1801 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1802 {
1803 	struct op_split_ctx *ctx;
1804 	spdk_bs_sequence_t *seq;
1805 	struct spdk_bs_cpl cpl;
1806 
1807 	assert(blob != NULL);
1808 
1809 	ctx = calloc(1, sizeof(struct op_split_ctx));
1810 	if (ctx == NULL) {
1811 		cb_fn(cb_arg, -ENOMEM);
1812 		return;
1813 	}
1814 
1815 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1816 	cpl.u.blob_basic.cb_fn = cb_fn;
1817 	cpl.u.blob_basic.cb_arg = cb_arg;
1818 
1819 	seq = spdk_bs_sequence_start(ch, &cpl);
1820 	if (!seq) {
1821 		free(ctx);
1822 		cb_fn(cb_arg, -ENOMEM);
1823 		return;
1824 	}
1825 
1826 	ctx->blob = blob;
1827 	ctx->channel = ch;
1828 	ctx->curr_payload = payload;
1829 	ctx->io_unit_offset = offset;
1830 	ctx->io_units_remaining = length;
1831 	ctx->op_type = op_type;
1832 	ctx->seq = seq;
1833 
1834 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1835 }
1836 
1837 static void
1838 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1839 				    void *payload, uint64_t offset, uint64_t length,
1840 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1841 {
1842 	struct spdk_bs_cpl cpl;
1843 	uint64_t lba;
1844 	uint32_t lba_count;
1845 
1846 	assert(blob != NULL);
1847 
1848 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1849 	cpl.u.blob_basic.cb_fn = cb_fn;
1850 	cpl.u.blob_basic.cb_arg = cb_arg;
1851 
1852 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1853 
1854 	if (blob->frozen_refcnt) {
1855 		/* This blob I/O is frozen */
1856 		spdk_bs_user_op_t *op;
1857 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1858 
1859 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1860 		if (!op) {
1861 			cb_fn(cb_arg, -ENOMEM);
1862 			return;
1863 		}
1864 
1865 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1866 
1867 		return;
1868 	}
1869 
1870 	switch (op_type) {
1871 	case SPDK_BLOB_READ: {
1872 		spdk_bs_batch_t *batch;
1873 
1874 		batch = spdk_bs_batch_open(_ch, &cpl);
1875 		if (!batch) {
1876 			cb_fn(cb_arg, -ENOMEM);
1877 			return;
1878 		}
1879 
1880 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1881 			/* Read from the blob */
1882 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1883 		} else {
1884 			/* Read from the backing block device */
1885 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1886 		}
1887 
1888 		spdk_bs_batch_close(batch);
1889 		break;
1890 	}
1891 	case SPDK_BLOB_WRITE:
1892 	case SPDK_BLOB_WRITE_ZEROES: {
1893 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1894 			/* Write to the blob */
1895 			spdk_bs_batch_t *batch;
1896 
1897 			if (lba_count == 0) {
1898 				cb_fn(cb_arg, 0);
1899 				return;
1900 			}
1901 
1902 			batch = spdk_bs_batch_open(_ch, &cpl);
1903 			if (!batch) {
1904 				cb_fn(cb_arg, -ENOMEM);
1905 				return;
1906 			}
1907 
1908 			if (op_type == SPDK_BLOB_WRITE) {
1909 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1910 			} else {
1911 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1912 			}
1913 
1914 			spdk_bs_batch_close(batch);
1915 		} else {
1916 			/* Queue this operation and allocate the cluster */
1917 			spdk_bs_user_op_t *op;
1918 
1919 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1920 			if (!op) {
1921 				cb_fn(cb_arg, -ENOMEM);
1922 				return;
1923 			}
1924 
1925 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1926 		}
1927 		break;
1928 	}
1929 	case SPDK_BLOB_UNMAP: {
1930 		spdk_bs_batch_t *batch;
1931 
1932 		batch = spdk_bs_batch_open(_ch, &cpl);
1933 		if (!batch) {
1934 			cb_fn(cb_arg, -ENOMEM);
1935 			return;
1936 		}
1937 
1938 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1939 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1940 		}
1941 
1942 		spdk_bs_batch_close(batch);
1943 		break;
1944 	}
1945 	case SPDK_BLOB_READV:
1946 	case SPDK_BLOB_WRITEV:
1947 		SPDK_ERRLOG("readv/write not valid\n");
1948 		cb_fn(cb_arg, -EINVAL);
1949 		break;
1950 	}
1951 }
1952 
1953 static void
1954 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1955 			     void *payload, uint64_t offset, uint64_t length,
1956 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1957 {
1958 	assert(blob != NULL);
1959 
1960 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1961 		cb_fn(cb_arg, -EPERM);
1962 		return;
1963 	}
1964 
1965 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
1966 		cb_fn(cb_arg, -EINVAL);
1967 		return;
1968 	}
1969 	if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) {
1970 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1971 						    cb_fn, cb_arg, op_type);
1972 	} else {
1973 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1974 						   cb_fn, cb_arg, op_type);
1975 	}
1976 }
1977 
1978 struct rw_iov_ctx {
1979 	struct spdk_blob *blob;
1980 	struct spdk_io_channel *channel;
1981 	spdk_blob_op_complete cb_fn;
1982 	void *cb_arg;
1983 	bool read;
1984 	int iovcnt;
1985 	struct iovec *orig_iov;
1986 	uint64_t io_unit_offset;
1987 	uint64_t io_units_remaining;
1988 	uint64_t io_units_done;
1989 	struct iovec iov[0];
1990 };
1991 
1992 static void
1993 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1994 {
1995 	assert(cb_arg == NULL);
1996 	spdk_bs_sequence_finish(seq, bserrno);
1997 }
1998 
1999 static void
2000 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
2001 {
2002 	struct rw_iov_ctx *ctx = cb_arg;
2003 	struct spdk_blob *blob = ctx->blob;
2004 	struct iovec *iov, *orig_iov;
2005 	int iovcnt;
2006 	size_t orig_iovoff;
2007 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2008 	uint64_t byte_count;
2009 
2010 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2011 		ctx->cb_fn(ctx->cb_arg, bserrno);
2012 		free(ctx);
2013 		return;
2014 	}
2015 
2016 	io_unit_offset = ctx->io_unit_offset;
2017 	io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2018 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2019 	/*
2020 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2021 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2022 	 *  point to the current position in the I/O sequence.
2023 	 */
2024 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2025 	orig_iov = &ctx->orig_iov[0];
2026 	orig_iovoff = 0;
2027 	while (byte_count > 0) {
2028 		if (byte_count >= orig_iov->iov_len) {
2029 			byte_count -= orig_iov->iov_len;
2030 			orig_iov++;
2031 		} else {
2032 			orig_iovoff = byte_count;
2033 			byte_count = 0;
2034 		}
2035 	}
2036 
2037 	/*
2038 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2039 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2040 	 */
2041 	byte_count = io_units_count * blob->bs->io_unit_size;
2042 	iov = &ctx->iov[0];
2043 	iovcnt = 0;
2044 	while (byte_count > 0) {
2045 		assert(iovcnt < ctx->iovcnt);
2046 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2047 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2048 		byte_count -= iov->iov_len;
2049 		orig_iovoff = 0;
2050 		orig_iov++;
2051 		iov++;
2052 		iovcnt++;
2053 	}
2054 
2055 	ctx->io_unit_offset += io_units_count;
2056 	ctx->io_units_remaining -= io_units_count;
2057 	ctx->io_units_done += io_units_count;
2058 	iov = &ctx->iov[0];
2059 
2060 	if (ctx->read) {
2061 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2062 				   io_units_count, _spdk_rw_iov_split_next, ctx);
2063 	} else {
2064 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2065 				    io_units_count, _spdk_rw_iov_split_next, ctx);
2066 	}
2067 }
2068 
2069 static void
2070 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2071 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2072 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2073 {
2074 	struct spdk_bs_cpl	cpl;
2075 
2076 	assert(blob != NULL);
2077 
2078 	if (!read && blob->data_ro) {
2079 		cb_fn(cb_arg, -EPERM);
2080 		return;
2081 	}
2082 
2083 	if (length == 0) {
2084 		cb_fn(cb_arg, 0);
2085 		return;
2086 	}
2087 
2088 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2089 		cb_fn(cb_arg, -EINVAL);
2090 		return;
2091 	}
2092 
2093 	/*
2094 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2095 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2096 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2097 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2098 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2099 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2100 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2101 	 *
2102 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2103 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2104 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2105 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2106 	 */
2107 	if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) {
2108 		uint32_t lba_count;
2109 		uint64_t lba;
2110 
2111 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2112 
2113 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2114 		cpl.u.blob_basic.cb_fn = cb_fn;
2115 		cpl.u.blob_basic.cb_arg = cb_arg;
2116 		if (blob->frozen_refcnt) {
2117 			/* This blob I/O is frozen */
2118 			spdk_bs_user_op_t *op;
2119 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2120 
2121 			op = spdk_bs_user_op_alloc(_channel, &cpl, read, blob, iov, iovcnt, offset, length);
2122 			if (!op) {
2123 				cb_fn(cb_arg, -ENOMEM);
2124 				return;
2125 			}
2126 
2127 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2128 
2129 			return;
2130 		}
2131 
2132 		if (read) {
2133 			spdk_bs_sequence_t *seq;
2134 
2135 			seq = spdk_bs_sequence_start(_channel, &cpl);
2136 			if (!seq) {
2137 				cb_fn(cb_arg, -ENOMEM);
2138 				return;
2139 			}
2140 
2141 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2142 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2143 			} else {
2144 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2145 							      _spdk_rw_iov_done, NULL);
2146 			}
2147 		} else {
2148 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2149 				spdk_bs_sequence_t *seq;
2150 
2151 				seq = spdk_bs_sequence_start(_channel, &cpl);
2152 				if (!seq) {
2153 					cb_fn(cb_arg, -ENOMEM);
2154 					return;
2155 				}
2156 
2157 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2158 			} else {
2159 				/* Queue this operation and allocate the cluster */
2160 				spdk_bs_user_op_t *op;
2161 
2162 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2163 							   length);
2164 				if (!op) {
2165 					cb_fn(cb_arg, -ENOMEM);
2166 					return;
2167 				}
2168 
2169 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2170 			}
2171 		}
2172 	} else {
2173 		struct rw_iov_ctx *ctx;
2174 
2175 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2176 		if (ctx == NULL) {
2177 			cb_fn(cb_arg, -ENOMEM);
2178 			return;
2179 		}
2180 
2181 		ctx->blob = blob;
2182 		ctx->channel = _channel;
2183 		ctx->cb_fn = cb_fn;
2184 		ctx->cb_arg = cb_arg;
2185 		ctx->read = read;
2186 		ctx->orig_iov = iov;
2187 		ctx->iovcnt = iovcnt;
2188 		ctx->io_unit_offset = offset;
2189 		ctx->io_units_remaining = length;
2190 		ctx->io_units_done = 0;
2191 
2192 		_spdk_rw_iov_split_next(ctx, 0);
2193 	}
2194 }
2195 
2196 static struct spdk_blob *
2197 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2198 {
2199 	struct spdk_blob *blob;
2200 
2201 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2202 		if (blob->id == blobid) {
2203 			return blob;
2204 		}
2205 	}
2206 
2207 	return NULL;
2208 }
2209 
2210 static int
2211 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2212 {
2213 	struct spdk_blob_store		*bs = io_device;
2214 	struct spdk_bs_channel		*channel = ctx_buf;
2215 	struct spdk_bs_dev		*dev;
2216 	uint32_t			max_ops = bs->max_channel_ops;
2217 	uint32_t			i;
2218 
2219 	dev = bs->dev;
2220 
2221 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2222 	if (!channel->req_mem) {
2223 		return -1;
2224 	}
2225 
2226 	TAILQ_INIT(&channel->reqs);
2227 
2228 	for (i = 0; i < max_ops; i++) {
2229 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2230 	}
2231 
2232 	channel->bs = bs;
2233 	channel->dev = dev;
2234 	channel->dev_channel = dev->create_channel(dev);
2235 
2236 	if (!channel->dev_channel) {
2237 		SPDK_ERRLOG("Failed to create device channel.\n");
2238 		free(channel->req_mem);
2239 		return -1;
2240 	}
2241 
2242 	TAILQ_INIT(&channel->need_cluster_alloc);
2243 	TAILQ_INIT(&channel->queued_io);
2244 
2245 	return 0;
2246 }
2247 
2248 static void
2249 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2250 {
2251 	struct spdk_bs_channel *channel = ctx_buf;
2252 	spdk_bs_user_op_t *op;
2253 
2254 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2255 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2256 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2257 		spdk_bs_user_op_abort(op);
2258 	}
2259 
2260 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2261 		op = TAILQ_FIRST(&channel->queued_io);
2262 		TAILQ_REMOVE(&channel->queued_io, op, link);
2263 		spdk_bs_user_op_abort(op);
2264 	}
2265 
2266 	free(channel->req_mem);
2267 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2268 }
2269 
2270 static void
2271 _spdk_bs_dev_destroy(void *io_device)
2272 {
2273 	struct spdk_blob_store *bs = io_device;
2274 	struct spdk_blob	*blob, *blob_tmp;
2275 
2276 	bs->dev->destroy(bs->dev);
2277 
2278 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2279 		TAILQ_REMOVE(&bs->blobs, blob, link);
2280 		_spdk_blob_free(blob);
2281 	}
2282 
2283 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2284 
2285 	spdk_bit_array_free(&bs->used_blobids);
2286 	spdk_bit_array_free(&bs->used_md_pages);
2287 	spdk_bit_array_free(&bs->used_clusters);
2288 	/*
2289 	 * If this function is called for any reason except a successful unload,
2290 	 * the unload_cpl type will be NONE and this will be a nop.
2291 	 */
2292 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2293 
2294 	free(bs);
2295 }
2296 
2297 static int
2298 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2299 {
2300 	spdk_blob_id snapshot_id;
2301 	struct spdk_blob_list *snapshot_entry = NULL;
2302 	struct spdk_blob_list *clone_entry = NULL;
2303 
2304 	assert(blob != NULL);
2305 
2306 	snapshot_id = blob->parent_id;
2307 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2308 		return 0;
2309 	}
2310 
2311 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2312 		if (snapshot_entry->id == snapshot_id) {
2313 			break;
2314 		}
2315 	}
2316 
2317 	if (snapshot_entry == NULL) {
2318 		/* Snapshot not found */
2319 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2320 		if (snapshot_entry == NULL) {
2321 			return -ENOMEM;
2322 		}
2323 		snapshot_entry->id = snapshot_id;
2324 		TAILQ_INIT(&snapshot_entry->clones);
2325 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2326 	} else {
2327 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2328 			if (clone_entry->id == blob->id) {
2329 				break;
2330 			}
2331 		}
2332 	}
2333 
2334 	if (clone_entry == NULL) {
2335 		/* Clone not found */
2336 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2337 		if (clone_entry == NULL) {
2338 			return -ENOMEM;
2339 		}
2340 		clone_entry->id = blob->id;
2341 		TAILQ_INIT(&clone_entry->clones);
2342 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2343 		snapshot_entry->clone_count++;
2344 	}
2345 
2346 	return 0;
2347 }
2348 
2349 static int
2350 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2351 {
2352 	struct spdk_blob_list *snapshot_entry = NULL;
2353 	struct spdk_blob_list *clone_entry = NULL;
2354 	spdk_blob_id snapshot_id;
2355 
2356 	assert(blob != NULL);
2357 
2358 	snapshot_id = blob->parent_id;
2359 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2360 		return 0;
2361 	}
2362 
2363 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2364 		if (snapshot_entry->id == snapshot_id) {
2365 			break;
2366 		}
2367 	}
2368 
2369 	assert(snapshot_entry != NULL);
2370 
2371 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2372 		if (clone_entry->id == blob->id) {
2373 			break;
2374 		}
2375 	}
2376 
2377 	assert(clone_entry != NULL);
2378 
2379 	blob->parent_id = SPDK_BLOBID_INVALID;
2380 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2381 	free(clone_entry);
2382 
2383 	snapshot_entry->clone_count--;
2384 
2385 	return 0;
2386 }
2387 
2388 static int
2389 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2390 {
2391 	struct spdk_blob_list *snapshot_entry;
2392 	struct spdk_blob_list *snapshot_entry_tmp;
2393 	struct spdk_blob_list *clone_entry;
2394 	struct spdk_blob_list *clone_entry_tmp;
2395 
2396 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2397 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2398 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2399 			free(clone_entry);
2400 		}
2401 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2402 		free(snapshot_entry);
2403 	}
2404 
2405 	return 0;
2406 }
2407 
2408 static void
2409 _spdk_bs_free(struct spdk_blob_store *bs)
2410 {
2411 	_spdk_bs_blob_list_free(bs);
2412 
2413 	spdk_bs_unregister_md_thread(bs);
2414 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2415 }
2416 
2417 void
2418 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2419 {
2420 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2421 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2422 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2423 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2424 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2425 	opts->iter_cb_fn = NULL;
2426 	opts->iter_cb_arg = NULL;
2427 }
2428 
2429 static int
2430 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2431 {
2432 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2433 	    opts->max_channel_ops == 0) {
2434 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2435 		return -1;
2436 	}
2437 
2438 	return 0;
2439 }
2440 
2441 static int
2442 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2443 {
2444 	struct spdk_blob_store	*bs;
2445 	uint64_t dev_size;
2446 	int rc;
2447 
2448 	dev_size = dev->blocklen * dev->blockcnt;
2449 	if (dev_size < opts->cluster_sz) {
2450 		/* Device size cannot be smaller than cluster size of blobstore */
2451 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2452 			     dev_size, opts->cluster_sz);
2453 		return -ENOSPC;
2454 	}
2455 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2456 		/* Cluster size cannot be smaller than page size */
2457 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2458 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2459 		return -EINVAL;
2460 	}
2461 	bs = calloc(1, sizeof(struct spdk_blob_store));
2462 	if (!bs) {
2463 		return -ENOMEM;
2464 	}
2465 
2466 	TAILQ_INIT(&bs->blobs);
2467 	TAILQ_INIT(&bs->snapshots);
2468 	bs->dev = dev;
2469 	bs->md_thread = spdk_get_thread();
2470 	assert(bs->md_thread != NULL);
2471 
2472 	/*
2473 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2474 	 *  even multiple of the cluster size.
2475 	 */
2476 	bs->cluster_sz = opts->cluster_sz;
2477 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2478 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2479 	bs->num_free_clusters = bs->total_clusters;
2480 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2481 	bs->io_unit_size = dev->blocklen;
2482 	if (bs->used_clusters == NULL) {
2483 		free(bs);
2484 		return -ENOMEM;
2485 	}
2486 
2487 	bs->max_channel_ops = opts->max_channel_ops;
2488 	bs->super_blob = SPDK_BLOBID_INVALID;
2489 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2490 
2491 	/* The metadata is assumed to be at least 1 page */
2492 	bs->used_md_pages = spdk_bit_array_create(1);
2493 	bs->used_blobids = spdk_bit_array_create(0);
2494 
2495 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2496 
2497 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2498 				sizeof(struct spdk_bs_channel), "blobstore");
2499 	rc = spdk_bs_register_md_thread(bs);
2500 	if (rc == -1) {
2501 		spdk_io_device_unregister(bs, NULL);
2502 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2503 		spdk_bit_array_free(&bs->used_blobids);
2504 		spdk_bit_array_free(&bs->used_md_pages);
2505 		spdk_bit_array_free(&bs->used_clusters);
2506 		free(bs);
2507 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2508 		return -ENOMEM;
2509 	}
2510 
2511 	*_bs = bs;
2512 	return 0;
2513 }
2514 
2515 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2516 
2517 struct spdk_bs_load_ctx {
2518 	struct spdk_blob_store		*bs;
2519 	struct spdk_bs_super_block	*super;
2520 
2521 	struct spdk_bs_md_mask		*mask;
2522 	bool				in_page_chain;
2523 	uint32_t			page_index;
2524 	uint32_t			cur_page;
2525 	struct spdk_blob_md_page	*page;
2526 
2527 	spdk_bs_sequence_t			*seq;
2528 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2529 	void					*iter_cb_arg;
2530 };
2531 
2532 static void
2533 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2534 {
2535 	assert(bserrno != 0);
2536 
2537 	spdk_dma_free(ctx->super);
2538 	spdk_bs_sequence_finish(seq, bserrno);
2539 	_spdk_bs_free(ctx->bs);
2540 	free(ctx);
2541 }
2542 
2543 static void
2544 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2545 {
2546 	uint32_t i = 0;
2547 
2548 	while (true) {
2549 		i = spdk_bit_array_find_first_set(array, i);
2550 		if (i >= mask->length) {
2551 			break;
2552 		}
2553 		mask->mask[i / 8] |= 1U << (i % 8);
2554 		i++;
2555 	}
2556 }
2557 
2558 static int
2559 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2560 {
2561 	struct spdk_bit_array *array;
2562 	uint32_t i;
2563 
2564 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2565 		return -ENOMEM;
2566 	}
2567 
2568 	array = *array_ptr;
2569 	for (i = 0; i < mask->length; i++) {
2570 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2571 			spdk_bit_array_set(array, i);
2572 		}
2573 	}
2574 
2575 	return 0;
2576 }
2577 
2578 static void
2579 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2580 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2581 {
2582 	/* Update the values in the super block */
2583 	super->super_blob = bs->super_blob;
2584 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2585 	super->crc = _spdk_blob_md_page_calc_crc(super);
2586 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2587 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2588 				   cb_fn, cb_arg);
2589 }
2590 
2591 static void
2592 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2593 {
2594 	struct spdk_bs_load_ctx	*ctx = arg;
2595 	uint64_t	mask_size, lba, lba_count;
2596 
2597 	/* Write out the used clusters mask */
2598 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2599 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2600 	if (!ctx->mask) {
2601 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2602 		return;
2603 	}
2604 
2605 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2606 	ctx->mask->length = ctx->bs->total_clusters;
2607 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2608 
2609 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2610 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2611 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2612 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2613 }
2614 
2615 static void
2616 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2617 {
2618 	struct spdk_bs_load_ctx	*ctx = arg;
2619 	uint64_t	mask_size, lba, lba_count;
2620 
2621 	if (seq->bserrno) {
2622 		_spdk_bs_load_ctx_fail(seq, ctx, seq->bserrno);
2623 		return;
2624 	}
2625 
2626 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2627 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2628 	if (!ctx->mask) {
2629 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2630 		return;
2631 	}
2632 
2633 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2634 	ctx->mask->length = ctx->super->md_len;
2635 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2636 
2637 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2638 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2639 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2640 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2641 }
2642 
2643 static void
2644 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2645 {
2646 	struct spdk_bs_load_ctx	*ctx = arg;
2647 	uint64_t	mask_size, lba, lba_count;
2648 
2649 	if (ctx->super->used_blobid_mask_len == 0) {
2650 		/*
2651 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2652 		 *  written to disk.
2653 		 */
2654 		cb_fn(seq, arg, 0);
2655 		return;
2656 	}
2657 
2658 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2659 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2660 	if (!ctx->mask) {
2661 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2662 		return;
2663 	}
2664 
2665 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2666 	ctx->mask->length = ctx->super->md_len;
2667 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2668 
2669 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2670 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2671 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2672 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2673 }
2674 
2675 static void
2676 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2677 {
2678 	struct spdk_bs_load_ctx *ctx = arg;
2679 
2680 	if (bserrno == 0) {
2681 		if (ctx->iter_cb_fn) {
2682 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2683 		}
2684 		_spdk_bs_blob_list_add(blob);
2685 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2686 		return;
2687 	}
2688 
2689 	if (bserrno == -ENOENT) {
2690 		bserrno = 0;
2691 	} else {
2692 		/*
2693 		 * This case needs to be looked at further.  Same problem
2694 		 *  exists with applications that rely on explicit blob
2695 		 *  iteration.  We should just skip the blob that failed
2696 		 *  to load and continue on to the next one.
2697 		 */
2698 		SPDK_ERRLOG("Error in iterating blobs\n");
2699 	}
2700 
2701 	ctx->iter_cb_fn = NULL;
2702 
2703 	spdk_dma_free(ctx->super);
2704 	spdk_dma_free(ctx->mask);
2705 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2706 	free(ctx);
2707 }
2708 
2709 static void
2710 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2711 {
2712 	ctx->seq = seq;
2713 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2714 }
2715 
2716 static void
2717 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2718 {
2719 	struct spdk_bs_load_ctx *ctx = cb_arg;
2720 	int rc;
2721 
2722 	/* The type must be correct */
2723 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2724 
2725 	/* The length of the mask (in bits) must not be greater than
2726 	 * the length of the buffer (converted to bits) */
2727 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2728 
2729 	/* The length of the mask must be exactly equal to the size
2730 	 * (in pages) of the metadata region */
2731 	assert(ctx->mask->length == ctx->super->md_len);
2732 
2733 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2734 	if (rc < 0) {
2735 		spdk_dma_free(ctx->mask);
2736 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2737 		return;
2738 	}
2739 
2740 	_spdk_bs_load_complete(seq, ctx, bserrno);
2741 }
2742 
2743 static void
2744 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2745 {
2746 	struct spdk_bs_load_ctx *ctx = cb_arg;
2747 	uint64_t		lba, lba_count, mask_size;
2748 	int			rc;
2749 
2750 	/* The type must be correct */
2751 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2752 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2753 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2754 					     struct spdk_blob_md_page) * 8));
2755 	/* The length of the mask must be exactly equal to the total number of clusters */
2756 	assert(ctx->mask->length == ctx->bs->total_clusters);
2757 
2758 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2759 	if (rc < 0) {
2760 		spdk_dma_free(ctx->mask);
2761 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2762 		return;
2763 	}
2764 
2765 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2766 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2767 
2768 	spdk_dma_free(ctx->mask);
2769 
2770 	/* Read the used blobids mask */
2771 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2772 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2773 	if (!ctx->mask) {
2774 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2775 		return;
2776 	}
2777 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2778 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2779 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2780 				  _spdk_bs_load_used_blobids_cpl, ctx);
2781 }
2782 
2783 static void
2784 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2785 {
2786 	struct spdk_bs_load_ctx *ctx = cb_arg;
2787 	uint64_t		lba, lba_count, mask_size;
2788 	int			rc;
2789 
2790 	/* The type must be correct */
2791 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2792 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2793 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2794 				     8));
2795 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2796 	assert(ctx->mask->length == ctx->super->md_len);
2797 
2798 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2799 	if (rc < 0) {
2800 		spdk_dma_free(ctx->mask);
2801 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2802 		return;
2803 	}
2804 
2805 	spdk_dma_free(ctx->mask);
2806 
2807 	/* Read the used clusters mask */
2808 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2809 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2810 	if (!ctx->mask) {
2811 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2812 		return;
2813 	}
2814 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2815 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2816 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2817 				  _spdk_bs_load_used_clusters_cpl, ctx);
2818 }
2819 
2820 static void
2821 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2822 {
2823 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2824 	uint64_t lba, lba_count, mask_size;
2825 
2826 	/* Read the used pages mask */
2827 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2828 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2829 	if (!ctx->mask) {
2830 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2831 		return;
2832 	}
2833 
2834 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2835 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2836 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2837 				  _spdk_bs_load_used_pages_cpl, ctx);
2838 }
2839 
2840 static int
2841 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2842 {
2843 	struct spdk_blob_md_descriptor *desc;
2844 	size_t	cur_desc = 0;
2845 
2846 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2847 	while (cur_desc < sizeof(page->descriptors)) {
2848 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2849 			if (desc->length == 0) {
2850 				/* If padding and length are 0, this terminates the page */
2851 				break;
2852 			}
2853 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2854 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2855 			unsigned int				i, j;
2856 			unsigned int				cluster_count = 0;
2857 			uint32_t				cluster_idx;
2858 
2859 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2860 
2861 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2862 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2863 					cluster_idx = desc_extent->extents[i].cluster_idx;
2864 					/*
2865 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2866 					 * in the used cluster map.
2867 					 */
2868 					if (cluster_idx != 0) {
2869 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2870 						if (bs->num_free_clusters == 0) {
2871 							return -ENOSPC;
2872 						}
2873 						bs->num_free_clusters--;
2874 					}
2875 					cluster_count++;
2876 				}
2877 			}
2878 			if (cluster_count == 0) {
2879 				return -EINVAL;
2880 			}
2881 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2882 			/* Skip this item */
2883 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2884 			/* Skip this item */
2885 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2886 			/* Skip this item */
2887 		} else {
2888 			/* Error */
2889 			return -EINVAL;
2890 		}
2891 		/* Advance to the next descriptor */
2892 		cur_desc += sizeof(*desc) + desc->length;
2893 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2894 			break;
2895 		}
2896 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2897 	}
2898 	return 0;
2899 }
2900 
2901 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2902 {
2903 	uint32_t crc;
2904 
2905 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2906 	if (crc != ctx->page->crc) {
2907 		return false;
2908 	}
2909 
2910 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2911 		return false;
2912 	}
2913 	return true;
2914 }
2915 
2916 static void
2917 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2918 
2919 static void
2920 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2921 {
2922 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2923 
2924 	_spdk_bs_load_complete(seq, ctx, bserrno);
2925 }
2926 
2927 static void
2928 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2929 {
2930 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2931 
2932 	spdk_dma_free(ctx->mask);
2933 	ctx->mask = NULL;
2934 
2935 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2936 }
2937 
2938 static void
2939 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2940 {
2941 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2942 
2943 	spdk_dma_free(ctx->mask);
2944 	ctx->mask = NULL;
2945 
2946 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2947 }
2948 
2949 static void
2950 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2951 {
2952 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2953 }
2954 
2955 static void
2956 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2957 {
2958 	struct spdk_bs_load_ctx *ctx = cb_arg;
2959 	uint64_t num_md_clusters;
2960 	uint64_t i;
2961 	uint32_t page_num;
2962 
2963 	if (bserrno != 0) {
2964 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2965 		return;
2966 	}
2967 
2968 	page_num = ctx->cur_page;
2969 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2970 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2971 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2972 			if (ctx->page->sequence_num == 0) {
2973 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2974 			}
2975 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2976 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
2977 				return;
2978 			}
2979 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
2980 				ctx->in_page_chain = true;
2981 				ctx->cur_page = ctx->page->next;
2982 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2983 				return;
2984 			}
2985 		}
2986 	}
2987 
2988 	ctx->in_page_chain = false;
2989 
2990 	do {
2991 		ctx->page_index++;
2992 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
2993 
2994 	if (ctx->page_index < ctx->super->md_len) {
2995 		ctx->cur_page = ctx->page_index;
2996 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2997 	} else {
2998 		/* Claim all of the clusters used by the metadata */
2999 		num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
3000 		for (i = 0; i < num_md_clusters; i++) {
3001 			_spdk_bs_claim_cluster(ctx->bs, i);
3002 		}
3003 		spdk_dma_free(ctx->page);
3004 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
3005 	}
3006 }
3007 
3008 static void
3009 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3010 {
3011 	struct spdk_bs_load_ctx *ctx = cb_arg;
3012 	uint64_t lba;
3013 
3014 	assert(ctx->cur_page < ctx->super->md_len);
3015 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3016 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3017 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3018 				  _spdk_bs_load_replay_md_cpl, ctx);
3019 }
3020 
3021 static void
3022 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3023 {
3024 	struct spdk_bs_load_ctx *ctx = cb_arg;
3025 
3026 	ctx->page_index = 0;
3027 	ctx->cur_page = 0;
3028 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3029 				     SPDK_BS_PAGE_SIZE,
3030 				     NULL);
3031 	if (!ctx->page) {
3032 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3033 		return;
3034 	}
3035 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3036 }
3037 
3038 static void
3039 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3040 {
3041 	struct spdk_bs_load_ctx *ctx = cb_arg;
3042 	int		rc;
3043 
3044 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3045 	if (rc < 0) {
3046 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3047 		return;
3048 	}
3049 
3050 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3051 	if (rc < 0) {
3052 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3053 		return;
3054 	}
3055 
3056 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3057 	if (rc < 0) {
3058 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3059 		return;
3060 	}
3061 
3062 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3063 	_spdk_bs_load_replay_md(seq, cb_arg);
3064 }
3065 
3066 static void
3067 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3068 {
3069 	struct spdk_bs_load_ctx *ctx = cb_arg;
3070 	uint32_t	crc;
3071 	int		rc;
3072 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3073 
3074 	if (ctx->super->version > SPDK_BS_VERSION ||
3075 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3076 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3077 		return;
3078 	}
3079 
3080 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3081 		   sizeof(ctx->super->signature)) != 0) {
3082 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3083 		return;
3084 	}
3085 
3086 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3087 	if (crc != ctx->super->crc) {
3088 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3089 		return;
3090 	}
3091 
3092 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3093 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3094 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3095 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3096 	} else {
3097 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3098 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3099 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3100 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3101 		return;
3102 	}
3103 
3104 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3105 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3106 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3107 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3108 		return;
3109 	}
3110 
3111 	if (ctx->super->size == 0) {
3112 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
3113 	}
3114 
3115 	if (ctx->super->io_unit_size == 0) {
3116 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
3117 	}
3118 
3119 	/* Parse the super block */
3120 	ctx->bs->clean = 1;
3121 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3122 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
3123 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3124 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
3125 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3126 	if (rc < 0) {
3127 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3128 		return;
3129 	}
3130 	ctx->bs->md_start = ctx->super->md_start;
3131 	ctx->bs->md_len = ctx->super->md_len;
3132 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
3133 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3134 	ctx->bs->super_blob = ctx->super->super_blob;
3135 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3136 
3137 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3138 		_spdk_bs_recover(seq, ctx);
3139 	} else {
3140 		_spdk_bs_load_read_used_pages(seq, ctx);
3141 	}
3142 }
3143 
3144 void
3145 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3146 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3147 {
3148 	struct spdk_blob_store	*bs;
3149 	struct spdk_bs_cpl	cpl;
3150 	spdk_bs_sequence_t	*seq;
3151 	struct spdk_bs_load_ctx *ctx;
3152 	struct spdk_bs_opts	opts = {};
3153 	int err;
3154 
3155 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3156 
3157 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3158 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3159 		dev->destroy(dev);
3160 		cb_fn(cb_arg, NULL, -EINVAL);
3161 		return;
3162 	}
3163 
3164 	if (o) {
3165 		opts = *o;
3166 	} else {
3167 		spdk_bs_opts_init(&opts);
3168 	}
3169 
3170 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3171 		dev->destroy(dev);
3172 		cb_fn(cb_arg, NULL, -EINVAL);
3173 		return;
3174 	}
3175 
3176 	err = _spdk_bs_alloc(dev, &opts, &bs);
3177 	if (err) {
3178 		dev->destroy(dev);
3179 		cb_fn(cb_arg, NULL, err);
3180 		return;
3181 	}
3182 
3183 	ctx = calloc(1, sizeof(*ctx));
3184 	if (!ctx) {
3185 		_spdk_bs_free(bs);
3186 		cb_fn(cb_arg, NULL, -ENOMEM);
3187 		return;
3188 	}
3189 
3190 	ctx->bs = bs;
3191 	ctx->iter_cb_fn = opts.iter_cb_fn;
3192 	ctx->iter_cb_arg = opts.iter_cb_arg;
3193 
3194 	/* Allocate memory for the super block */
3195 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3196 	if (!ctx->super) {
3197 		free(ctx);
3198 		_spdk_bs_free(bs);
3199 		cb_fn(cb_arg, NULL, -ENOMEM);
3200 		return;
3201 	}
3202 
3203 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3204 	cpl.u.bs_handle.cb_fn = cb_fn;
3205 	cpl.u.bs_handle.cb_arg = cb_arg;
3206 	cpl.u.bs_handle.bs = bs;
3207 
3208 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3209 	if (!seq) {
3210 		spdk_dma_free(ctx->super);
3211 		free(ctx);
3212 		_spdk_bs_free(bs);
3213 		cb_fn(cb_arg, NULL, -ENOMEM);
3214 		return;
3215 	}
3216 
3217 	/* Read the super block */
3218 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3219 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3220 				  _spdk_bs_load_super_cpl, ctx);
3221 }
3222 
3223 /* END spdk_bs_load */
3224 
3225 /* START spdk_bs_dump */
3226 
3227 struct spdk_bs_dump_ctx {
3228 	struct spdk_blob_store		*bs;
3229 	struct spdk_bs_super_block	*super;
3230 	uint32_t			cur_page;
3231 	struct spdk_blob_md_page	*page;
3232 	spdk_bs_sequence_t		*seq;
3233 	FILE				*fp;
3234 	spdk_bs_dump_print_xattr	print_xattr_fn;
3235 	char				xattr_name[4096];
3236 };
3237 
3238 static void
3239 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3240 {
3241 	spdk_dma_free(ctx->super);
3242 
3243 	/*
3244 	 * We need to defer calling spdk_bs_call_cpl() until after
3245 	 * dev destruction, so tuck these away for later use.
3246 	 */
3247 	ctx->bs->unload_err = bserrno;
3248 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3249 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3250 
3251 	spdk_bs_sequence_finish(seq, 0);
3252 	_spdk_bs_free(ctx->bs);
3253 	free(ctx);
3254 }
3255 
3256 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3257 
3258 static void
3259 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3260 {
3261 	uint32_t page_idx = ctx->cur_page;
3262 	struct spdk_blob_md_page *page = ctx->page;
3263 	struct spdk_blob_md_descriptor *desc;
3264 	size_t cur_desc = 0;
3265 	uint32_t crc;
3266 
3267 	fprintf(ctx->fp, "=========\n");
3268 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3269 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3270 
3271 	crc = _spdk_blob_md_page_calc_crc(page);
3272 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3273 
3274 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3275 	while (cur_desc < sizeof(page->descriptors)) {
3276 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3277 			if (desc->length == 0) {
3278 				/* If padding and length are 0, this terminates the page */
3279 				break;
3280 			}
3281 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3282 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3283 			unsigned int				i;
3284 
3285 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3286 
3287 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3288 				if (desc_extent->extents[i].cluster_idx != 0) {
3289 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3290 						desc_extent->extents[i].cluster_idx);
3291 				} else {
3292 					fprintf(ctx->fp, "Unallocated Extent - ");
3293 				}
3294 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3295 				fprintf(ctx->fp, "\n");
3296 			}
3297 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3298 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3299 			uint32_t i;
3300 
3301 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3302 
3303 			if (desc_xattr->length !=
3304 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3305 			    desc_xattr->name_length + desc_xattr->value_length) {
3306 			}
3307 
3308 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3309 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3310 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3311 			fprintf(ctx->fp, "       value = \"");
3312 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3313 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3314 					    desc_xattr->value_length);
3315 			fprintf(ctx->fp, "\"\n");
3316 			for (i = 0; i < desc_xattr->value_length; i++) {
3317 				if (i % 16 == 0) {
3318 					fprintf(ctx->fp, "               ");
3319 				}
3320 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3321 				if ((i + 1) % 16 == 0) {
3322 					fprintf(ctx->fp, "\n");
3323 				}
3324 			}
3325 			if (i % 16 != 0) {
3326 				fprintf(ctx->fp, "\n");
3327 			}
3328 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3329 			/* TODO */
3330 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3331 			/* TODO */
3332 		} else {
3333 			/* Error */
3334 		}
3335 		/* Advance to the next descriptor */
3336 		cur_desc += sizeof(*desc) + desc->length;
3337 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3338 			break;
3339 		}
3340 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3341 	}
3342 }
3343 
3344 static void
3345 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3346 {
3347 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3348 
3349 	if (bserrno != 0) {
3350 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3351 		return;
3352 	}
3353 
3354 	if (ctx->page->id != 0) {
3355 		_spdk_bs_dump_print_md_page(ctx);
3356 	}
3357 
3358 	ctx->cur_page++;
3359 
3360 	if (ctx->cur_page < ctx->super->md_len) {
3361 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3362 	} else {
3363 		spdk_dma_free(ctx->page);
3364 		_spdk_bs_dump_finish(seq, ctx, 0);
3365 	}
3366 }
3367 
3368 static void
3369 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3370 {
3371 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3372 	uint64_t lba;
3373 
3374 	assert(ctx->cur_page < ctx->super->md_len);
3375 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3376 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3377 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3378 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3379 }
3380 
3381 static void
3382 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3383 {
3384 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3385 
3386 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3387 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3388 		   sizeof(ctx->super->signature)) != 0) {
3389 		fprintf(ctx->fp, "(Mismatch)\n");
3390 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3391 		return;
3392 	} else {
3393 		fprintf(ctx->fp, "(OK)\n");
3394 	}
3395 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3396 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3397 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3398 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3399 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3400 	fprintf(ctx->fp, "Super Blob ID: ");
3401 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3402 		fprintf(ctx->fp, "(None)\n");
3403 	} else {
3404 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3405 	}
3406 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3407 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3408 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3409 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3410 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3411 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3412 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3413 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3414 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3415 
3416 	ctx->cur_page = 0;
3417 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3418 				     SPDK_BS_PAGE_SIZE,
3419 				     NULL);
3420 	if (!ctx->page) {
3421 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3422 		return;
3423 	}
3424 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3425 }
3426 
3427 void
3428 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3429 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3430 {
3431 	struct spdk_blob_store	*bs;
3432 	struct spdk_bs_cpl	cpl;
3433 	spdk_bs_sequence_t	*seq;
3434 	struct spdk_bs_dump_ctx *ctx;
3435 	struct spdk_bs_opts	opts = {};
3436 	int err;
3437 
3438 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3439 
3440 	spdk_bs_opts_init(&opts);
3441 
3442 	err = _spdk_bs_alloc(dev, &opts, &bs);
3443 	if (err) {
3444 		dev->destroy(dev);
3445 		cb_fn(cb_arg, err);
3446 		return;
3447 	}
3448 
3449 	ctx = calloc(1, sizeof(*ctx));
3450 	if (!ctx) {
3451 		_spdk_bs_free(bs);
3452 		cb_fn(cb_arg, -ENOMEM);
3453 		return;
3454 	}
3455 
3456 	ctx->bs = bs;
3457 	ctx->fp = fp;
3458 	ctx->print_xattr_fn = print_xattr_fn;
3459 
3460 	/* Allocate memory for the super block */
3461 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3462 	if (!ctx->super) {
3463 		free(ctx);
3464 		_spdk_bs_free(bs);
3465 		cb_fn(cb_arg, -ENOMEM);
3466 		return;
3467 	}
3468 
3469 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3470 	cpl.u.bs_basic.cb_fn = cb_fn;
3471 	cpl.u.bs_basic.cb_arg = cb_arg;
3472 
3473 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3474 	if (!seq) {
3475 		spdk_dma_free(ctx->super);
3476 		free(ctx);
3477 		_spdk_bs_free(bs);
3478 		cb_fn(cb_arg, -ENOMEM);
3479 		return;
3480 	}
3481 
3482 	/* Read the super block */
3483 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3484 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3485 				  _spdk_bs_dump_super_cpl, ctx);
3486 }
3487 
3488 /* END spdk_bs_dump */
3489 
3490 /* START spdk_bs_init */
3491 
3492 struct spdk_bs_init_ctx {
3493 	struct spdk_blob_store		*bs;
3494 	struct spdk_bs_super_block	*super;
3495 };
3496 
3497 static void
3498 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3499 {
3500 	struct spdk_bs_init_ctx *ctx = cb_arg;
3501 
3502 	spdk_dma_free(ctx->super);
3503 	free(ctx);
3504 
3505 	spdk_bs_sequence_finish(seq, bserrno);
3506 }
3507 
3508 static void
3509 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3510 {
3511 	struct spdk_bs_init_ctx *ctx = cb_arg;
3512 
3513 	/* Write super block */
3514 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3515 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3516 				   _spdk_bs_init_persist_super_cpl, ctx);
3517 }
3518 
3519 void
3520 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3521 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3522 {
3523 	struct spdk_bs_init_ctx *ctx;
3524 	struct spdk_blob_store	*bs;
3525 	struct spdk_bs_cpl	cpl;
3526 	spdk_bs_sequence_t	*seq;
3527 	spdk_bs_batch_t		*batch;
3528 	uint64_t		num_md_lba;
3529 	uint64_t		num_md_pages;
3530 	uint64_t		num_md_clusters;
3531 	uint32_t		i;
3532 	struct spdk_bs_opts	opts = {};
3533 	int			rc;
3534 
3535 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3536 
3537 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3538 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3539 			    dev->blocklen);
3540 		dev->destroy(dev);
3541 		cb_fn(cb_arg, NULL, -EINVAL);
3542 		return;
3543 	}
3544 
3545 	if (o) {
3546 		opts = *o;
3547 	} else {
3548 		spdk_bs_opts_init(&opts);
3549 	}
3550 
3551 	if (_spdk_bs_opts_verify(&opts) != 0) {
3552 		dev->destroy(dev);
3553 		cb_fn(cb_arg, NULL, -EINVAL);
3554 		return;
3555 	}
3556 
3557 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3558 	if (rc) {
3559 		dev->destroy(dev);
3560 		cb_fn(cb_arg, NULL, rc);
3561 		return;
3562 	}
3563 
3564 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3565 		/* By default, allocate 1 page per cluster.
3566 		 * Technically, this over-allocates metadata
3567 		 * because more metadata will reduce the number
3568 		 * of usable clusters. This can be addressed with
3569 		 * more complex math in the future.
3570 		 */
3571 		bs->md_len = bs->total_clusters;
3572 	} else {
3573 		bs->md_len = opts.num_md_pages;
3574 	}
3575 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3576 	if (rc < 0) {
3577 		_spdk_bs_free(bs);
3578 		cb_fn(cb_arg, NULL, -ENOMEM);
3579 		return;
3580 	}
3581 
3582 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3583 	if (rc < 0) {
3584 		_spdk_bs_free(bs);
3585 		cb_fn(cb_arg, NULL, -ENOMEM);
3586 		return;
3587 	}
3588 
3589 	ctx = calloc(1, sizeof(*ctx));
3590 	if (!ctx) {
3591 		_spdk_bs_free(bs);
3592 		cb_fn(cb_arg, NULL, -ENOMEM);
3593 		return;
3594 	}
3595 
3596 	ctx->bs = bs;
3597 
3598 	/* Allocate memory for the super block */
3599 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3600 	if (!ctx->super) {
3601 		free(ctx);
3602 		_spdk_bs_free(bs);
3603 		cb_fn(cb_arg, NULL, -ENOMEM);
3604 		return;
3605 	}
3606 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3607 	       sizeof(ctx->super->signature));
3608 	ctx->super->version = SPDK_BS_VERSION;
3609 	ctx->super->length = sizeof(*ctx->super);
3610 	ctx->super->super_blob = bs->super_blob;
3611 	ctx->super->clean = 0;
3612 	ctx->super->cluster_size = bs->cluster_sz;
3613 	ctx->super->io_unit_size = bs->io_unit_size;
3614 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3615 
3616 	/* Calculate how many pages the metadata consumes at the front
3617 	 * of the disk.
3618 	 */
3619 
3620 	/* The super block uses 1 page */
3621 	num_md_pages = 1;
3622 
3623 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3624 	 * up to the nearest page, plus a header.
3625 	 */
3626 	ctx->super->used_page_mask_start = num_md_pages;
3627 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3628 					 spdk_divide_round_up(bs->md_len, 8),
3629 					 SPDK_BS_PAGE_SIZE);
3630 	num_md_pages += ctx->super->used_page_mask_len;
3631 
3632 	/* The used_clusters mask requires 1 bit per cluster, rounded
3633 	 * up to the nearest page, plus a header.
3634 	 */
3635 	ctx->super->used_cluster_mask_start = num_md_pages;
3636 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3637 					    spdk_divide_round_up(bs->total_clusters, 8),
3638 					    SPDK_BS_PAGE_SIZE);
3639 	num_md_pages += ctx->super->used_cluster_mask_len;
3640 
3641 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3642 	 * up to the nearest page, plus a header.
3643 	 */
3644 	ctx->super->used_blobid_mask_start = num_md_pages;
3645 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3646 					   spdk_divide_round_up(bs->md_len, 8),
3647 					   SPDK_BS_PAGE_SIZE);
3648 	num_md_pages += ctx->super->used_blobid_mask_len;
3649 
3650 	/* The metadata region size was chosen above */
3651 	ctx->super->md_start = bs->md_start = num_md_pages;
3652 	ctx->super->md_len = bs->md_len;
3653 	num_md_pages += bs->md_len;
3654 
3655 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3656 
3657 	ctx->super->size = dev->blockcnt * dev->blocklen;
3658 
3659 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3660 
3661 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
3662 	if (num_md_clusters > bs->total_clusters) {
3663 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3664 			    "please decrease number of pages reserved for metadata "
3665 			    "or increase cluster size.\n");
3666 		spdk_dma_free(ctx->super);
3667 		free(ctx);
3668 		_spdk_bs_free(bs);
3669 		cb_fn(cb_arg, NULL, -ENOMEM);
3670 		return;
3671 	}
3672 	/* Claim all of the clusters used by the metadata */
3673 	for (i = 0; i < num_md_clusters; i++) {
3674 		_spdk_bs_claim_cluster(bs, i);
3675 	}
3676 
3677 	bs->total_data_clusters = bs->num_free_clusters;
3678 
3679 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3680 	cpl.u.bs_handle.cb_fn = cb_fn;
3681 	cpl.u.bs_handle.cb_arg = cb_arg;
3682 	cpl.u.bs_handle.bs = bs;
3683 
3684 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3685 	if (!seq) {
3686 		spdk_dma_free(ctx->super);
3687 		free(ctx);
3688 		_spdk_bs_free(bs);
3689 		cb_fn(cb_arg, NULL, -ENOMEM);
3690 		return;
3691 	}
3692 
3693 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3694 
3695 	/* Clear metadata space */
3696 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3697 	/* Trim data clusters */
3698 	spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3699 
3700 	spdk_bs_batch_close(batch);
3701 }
3702 
3703 /* END spdk_bs_init */
3704 
3705 /* START spdk_bs_destroy */
3706 
3707 static void
3708 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3709 {
3710 	struct spdk_bs_init_ctx *ctx = cb_arg;
3711 	struct spdk_blob_store *bs = ctx->bs;
3712 
3713 	/*
3714 	 * We need to defer calling spdk_bs_call_cpl() until after
3715 	 * dev destruction, so tuck these away for later use.
3716 	 */
3717 	bs->unload_err = bserrno;
3718 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3719 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3720 
3721 	spdk_bs_sequence_finish(seq, bserrno);
3722 
3723 	_spdk_bs_free(bs);
3724 	free(ctx);
3725 }
3726 
3727 void
3728 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3729 		void *cb_arg)
3730 {
3731 	struct spdk_bs_cpl	cpl;
3732 	spdk_bs_sequence_t	*seq;
3733 	struct spdk_bs_init_ctx *ctx;
3734 
3735 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3736 
3737 	if (!TAILQ_EMPTY(&bs->blobs)) {
3738 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3739 		cb_fn(cb_arg, -EBUSY);
3740 		return;
3741 	}
3742 
3743 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3744 	cpl.u.bs_basic.cb_fn = cb_fn;
3745 	cpl.u.bs_basic.cb_arg = cb_arg;
3746 
3747 	ctx = calloc(1, sizeof(*ctx));
3748 	if (!ctx) {
3749 		cb_fn(cb_arg, -ENOMEM);
3750 		return;
3751 	}
3752 
3753 	ctx->bs = bs;
3754 
3755 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3756 	if (!seq) {
3757 		free(ctx);
3758 		cb_fn(cb_arg, -ENOMEM);
3759 		return;
3760 	}
3761 
3762 	/* Write zeroes to the super block */
3763 	spdk_bs_sequence_write_zeroes_dev(seq,
3764 					  _spdk_bs_page_to_lba(bs, 0),
3765 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3766 					  _spdk_bs_destroy_trim_cpl, ctx);
3767 }
3768 
3769 /* END spdk_bs_destroy */
3770 
3771 /* START spdk_bs_unload */
3772 
3773 static void
3774 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3775 {
3776 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3777 
3778 	spdk_dma_free(ctx->super);
3779 
3780 	/*
3781 	 * We need to defer calling spdk_bs_call_cpl() until after
3782 	 * dev destruction, so tuck these away for later use.
3783 	 */
3784 	ctx->bs->unload_err = bserrno;
3785 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3786 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3787 
3788 	spdk_bs_sequence_finish(seq, bserrno);
3789 
3790 	_spdk_bs_free(ctx->bs);
3791 	free(ctx);
3792 }
3793 
3794 static void
3795 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3796 {
3797 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3798 
3799 	spdk_dma_free(ctx->mask);
3800 	ctx->super->clean = 1;
3801 
3802 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3803 }
3804 
3805 static void
3806 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3807 {
3808 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3809 
3810 	spdk_dma_free(ctx->mask);
3811 	ctx->mask = NULL;
3812 
3813 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3814 }
3815 
3816 static void
3817 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3818 {
3819 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3820 
3821 	spdk_dma_free(ctx->mask);
3822 	ctx->mask = NULL;
3823 
3824 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3825 }
3826 
3827 static void
3828 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3829 {
3830 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3831 }
3832 
3833 void
3834 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3835 {
3836 	struct spdk_bs_cpl	cpl;
3837 	spdk_bs_sequence_t	*seq;
3838 	struct spdk_bs_load_ctx *ctx;
3839 
3840 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3841 
3842 	if (!TAILQ_EMPTY(&bs->blobs)) {
3843 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3844 		cb_fn(cb_arg, -EBUSY);
3845 		return;
3846 	}
3847 
3848 	ctx = calloc(1, sizeof(*ctx));
3849 	if (!ctx) {
3850 		cb_fn(cb_arg, -ENOMEM);
3851 		return;
3852 	}
3853 
3854 	ctx->bs = bs;
3855 
3856 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3857 	if (!ctx->super) {
3858 		free(ctx);
3859 		cb_fn(cb_arg, -ENOMEM);
3860 		return;
3861 	}
3862 
3863 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3864 	cpl.u.bs_basic.cb_fn = cb_fn;
3865 	cpl.u.bs_basic.cb_arg = cb_arg;
3866 
3867 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3868 	if (!seq) {
3869 		spdk_dma_free(ctx->super);
3870 		free(ctx);
3871 		cb_fn(cb_arg, -ENOMEM);
3872 		return;
3873 	}
3874 
3875 	/* Read super block */
3876 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3877 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3878 				  _spdk_bs_unload_read_super_cpl, ctx);
3879 }
3880 
3881 /* END spdk_bs_unload */
3882 
3883 /* START spdk_bs_set_super */
3884 
3885 struct spdk_bs_set_super_ctx {
3886 	struct spdk_blob_store		*bs;
3887 	struct spdk_bs_super_block	*super;
3888 };
3889 
3890 static void
3891 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3892 {
3893 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3894 
3895 	if (bserrno != 0) {
3896 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3897 	}
3898 
3899 	spdk_dma_free(ctx->super);
3900 
3901 	spdk_bs_sequence_finish(seq, bserrno);
3902 
3903 	free(ctx);
3904 }
3905 
3906 static void
3907 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3908 {
3909 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3910 
3911 	if (bserrno != 0) {
3912 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3913 		spdk_dma_free(ctx->super);
3914 		spdk_bs_sequence_finish(seq, bserrno);
3915 		free(ctx);
3916 		return;
3917 	}
3918 
3919 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3920 }
3921 
3922 void
3923 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3924 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3925 {
3926 	struct spdk_bs_cpl		cpl;
3927 	spdk_bs_sequence_t		*seq;
3928 	struct spdk_bs_set_super_ctx	*ctx;
3929 
3930 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3931 
3932 	ctx = calloc(1, sizeof(*ctx));
3933 	if (!ctx) {
3934 		cb_fn(cb_arg, -ENOMEM);
3935 		return;
3936 	}
3937 
3938 	ctx->bs = bs;
3939 
3940 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3941 	if (!ctx->super) {
3942 		free(ctx);
3943 		cb_fn(cb_arg, -ENOMEM);
3944 		return;
3945 	}
3946 
3947 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3948 	cpl.u.bs_basic.cb_fn = cb_fn;
3949 	cpl.u.bs_basic.cb_arg = cb_arg;
3950 
3951 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3952 	if (!seq) {
3953 		spdk_dma_free(ctx->super);
3954 		free(ctx);
3955 		cb_fn(cb_arg, -ENOMEM);
3956 		return;
3957 	}
3958 
3959 	bs->super_blob = blobid;
3960 
3961 	/* Read super block */
3962 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3963 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3964 				  _spdk_bs_set_super_read_cpl, ctx);
3965 }
3966 
3967 /* END spdk_bs_set_super */
3968 
3969 void
3970 spdk_bs_get_super(struct spdk_blob_store *bs,
3971 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
3972 {
3973 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
3974 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
3975 	} else {
3976 		cb_fn(cb_arg, bs->super_blob, 0);
3977 	}
3978 }
3979 
3980 uint64_t
3981 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
3982 {
3983 	return bs->cluster_sz;
3984 }
3985 
3986 uint64_t
3987 spdk_bs_get_page_size(struct spdk_blob_store *bs)
3988 {
3989 	return SPDK_BS_PAGE_SIZE;
3990 }
3991 
3992 uint64_t
3993 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
3994 {
3995 	return bs->io_unit_size;
3996 }
3997 
3998 uint64_t
3999 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
4000 {
4001 	return bs->num_free_clusters;
4002 }
4003 
4004 uint64_t
4005 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
4006 {
4007 	return bs->total_data_clusters;
4008 }
4009 
4010 static int
4011 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
4012 {
4013 	bs->md_channel = spdk_get_io_channel(bs);
4014 	if (!bs->md_channel) {
4015 		SPDK_ERRLOG("Failed to get IO channel.\n");
4016 		return -1;
4017 	}
4018 
4019 	return 0;
4020 }
4021 
4022 static int
4023 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
4024 {
4025 	spdk_put_io_channel(bs->md_channel);
4026 
4027 	return 0;
4028 }
4029 
4030 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4031 {
4032 	assert(blob != NULL);
4033 
4034 	return blob->id;
4035 }
4036 
4037 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4038 {
4039 	assert(blob != NULL);
4040 
4041 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4042 }
4043 
4044 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
4045 {
4046 	assert(blob != NULL);
4047 
4048 	return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs);
4049 }
4050 
4051 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4052 {
4053 	assert(blob != NULL);
4054 
4055 	return blob->active.num_clusters;
4056 }
4057 
4058 /* START spdk_bs_create_blob */
4059 
4060 static void
4061 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4062 {
4063 	struct spdk_blob *blob = cb_arg;
4064 
4065 	_spdk_blob_free(blob);
4066 
4067 	spdk_bs_sequence_finish(seq, bserrno);
4068 }
4069 
4070 static int
4071 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4072 		      bool internal)
4073 {
4074 	uint64_t i;
4075 	size_t value_len = 0;
4076 	int rc;
4077 	const void *value = NULL;
4078 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4079 		return -EINVAL;
4080 	}
4081 	for (i = 0; i < xattrs->count; i++) {
4082 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4083 		if (value == NULL || value_len == 0) {
4084 			return -EINVAL;
4085 		}
4086 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4087 		if (rc < 0) {
4088 			return rc;
4089 		}
4090 	}
4091 	return 0;
4092 }
4093 
4094 static void
4095 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4096 {
4097 	_spdk_blob_verify_md_op(blob);
4098 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4099 	blob->state = SPDK_BLOB_STATE_DIRTY;
4100 }
4101 
4102 static void
4103 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4104 		     const struct spdk_blob_opts *opts,
4105 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4106 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4107 {
4108 	struct spdk_blob	*blob;
4109 	uint32_t		page_idx;
4110 	struct spdk_bs_cpl	cpl;
4111 	struct spdk_blob_opts	opts_default;
4112 	struct spdk_blob_xattr_opts internal_xattrs_default;
4113 	spdk_bs_sequence_t	*seq;
4114 	spdk_blob_id		id;
4115 	int rc;
4116 
4117 	assert(spdk_get_thread() == bs->md_thread);
4118 
4119 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4120 	if (page_idx == UINT32_MAX) {
4121 		cb_fn(cb_arg, 0, -ENOMEM);
4122 		return;
4123 	}
4124 	spdk_bit_array_set(bs->used_blobids, page_idx);
4125 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4126 
4127 	id = _spdk_bs_page_to_blobid(page_idx);
4128 
4129 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4130 
4131 	blob = _spdk_blob_alloc(bs, id);
4132 	if (!blob) {
4133 		cb_fn(cb_arg, 0, -ENOMEM);
4134 		return;
4135 	}
4136 
4137 	if (!opts) {
4138 		spdk_blob_opts_init(&opts_default);
4139 		opts = &opts_default;
4140 	}
4141 	if (!internal_xattrs) {
4142 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4143 		internal_xattrs = &internal_xattrs_default;
4144 	}
4145 
4146 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4147 	if (rc < 0) {
4148 		_spdk_blob_free(blob);
4149 		cb_fn(cb_arg, 0, rc);
4150 		return;
4151 	}
4152 
4153 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4154 	if (rc < 0) {
4155 		_spdk_blob_free(blob);
4156 		cb_fn(cb_arg, 0, rc);
4157 		return;
4158 	}
4159 
4160 	if (opts->thin_provision) {
4161 		_spdk_blob_set_thin_provision(blob);
4162 	}
4163 
4164 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4165 	if (rc < 0) {
4166 		_spdk_blob_free(blob);
4167 		cb_fn(cb_arg, 0, rc);
4168 		return;
4169 	}
4170 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4171 	cpl.u.blobid.cb_fn = cb_fn;
4172 	cpl.u.blobid.cb_arg = cb_arg;
4173 	cpl.u.blobid.blobid = blob->id;
4174 
4175 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4176 	if (!seq) {
4177 		_spdk_blob_free(blob);
4178 		cb_fn(cb_arg, 0, -ENOMEM);
4179 		return;
4180 	}
4181 
4182 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4183 }
4184 
4185 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4186 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4187 {
4188 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4189 }
4190 
4191 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4192 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4193 {
4194 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4195 }
4196 
4197 /* END spdk_bs_create_blob */
4198 
4199 /* START blob_cleanup */
4200 
4201 struct spdk_clone_snapshot_ctx {
4202 	struct spdk_bs_cpl      cpl;
4203 	int bserrno;
4204 	bool frozen;
4205 
4206 	struct spdk_io_channel *channel;
4207 
4208 	/* Current cluster for inflate operation */
4209 	uint64_t cluster;
4210 
4211 	/* For inflation force allocation of all unallocated clusters and remove
4212 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4213 	bool allocate_all;
4214 
4215 	struct {
4216 		spdk_blob_id id;
4217 		struct spdk_blob *blob;
4218 	} original;
4219 	struct {
4220 		spdk_blob_id id;
4221 		struct spdk_blob *blob;
4222 	} new;
4223 
4224 	/* xattrs specified for snapshot/clones only. They have no impact on
4225 	 * the original blobs xattrs. */
4226 	const struct spdk_blob_xattr_opts *xattrs;
4227 };
4228 
4229 static void
4230 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4231 {
4232 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4233 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4234 
4235 	if (bserrno != 0) {
4236 		if (ctx->bserrno != 0) {
4237 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4238 		} else {
4239 			ctx->bserrno = bserrno;
4240 		}
4241 	}
4242 
4243 	switch (cpl->type) {
4244 	case SPDK_BS_CPL_TYPE_BLOBID:
4245 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4246 		break;
4247 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4248 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4249 		break;
4250 	default:
4251 		SPDK_UNREACHABLE();
4252 		break;
4253 	}
4254 
4255 	free(ctx);
4256 }
4257 
4258 static void
4259 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4260 {
4261 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4262 	struct spdk_blob *origblob = ctx->original.blob;
4263 
4264 	if (bserrno != 0) {
4265 		if (ctx->bserrno != 0) {
4266 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4267 		} else {
4268 			ctx->bserrno = bserrno;
4269 		}
4270 	}
4271 
4272 	ctx->original.id = origblob->id;
4273 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4274 }
4275 
4276 static void
4277 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4278 {
4279 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4280 	struct spdk_blob *origblob = ctx->original.blob;
4281 
4282 	if (bserrno != 0) {
4283 		if (ctx->bserrno != 0) {
4284 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4285 		} else {
4286 			ctx->bserrno = bserrno;
4287 		}
4288 	}
4289 
4290 	if (ctx->frozen) {
4291 		/* Unfreeze any outstanding I/O */
4292 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4293 	} else {
4294 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4295 	}
4296 
4297 }
4298 
4299 static void
4300 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4301 {
4302 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4303 	struct spdk_blob *newblob = ctx->new.blob;
4304 
4305 	if (bserrno != 0) {
4306 		if (ctx->bserrno != 0) {
4307 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4308 		} else {
4309 			ctx->bserrno = bserrno;
4310 		}
4311 	}
4312 
4313 	ctx->new.id = newblob->id;
4314 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4315 }
4316 
4317 /* END blob_cleanup */
4318 
4319 /* START spdk_bs_create_snapshot */
4320 
4321 static void
4322 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4323 {
4324 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4325 	struct spdk_blob *newblob = ctx->new.blob;
4326 
4327 	if (bserrno != 0) {
4328 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4329 		return;
4330 	}
4331 
4332 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4333 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4334 	if (bserrno != 0) {
4335 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4336 		return;
4337 	}
4338 
4339 	_spdk_bs_blob_list_add(ctx->original.blob);
4340 
4341 	spdk_blob_set_read_only(newblob);
4342 
4343 	/* sync snapshot metadata */
4344 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4345 }
4346 
4347 static void
4348 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4349 {
4350 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4351 	struct spdk_blob *origblob = ctx->original.blob;
4352 	struct spdk_blob *newblob = ctx->new.blob;
4353 
4354 	if (bserrno != 0) {
4355 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4356 		return;
4357 	}
4358 
4359 	/* Set internal xattr for snapshot id */
4360 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4361 	if (bserrno != 0) {
4362 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4363 		return;
4364 	}
4365 
4366 	_spdk_bs_blob_list_remove(origblob);
4367 	origblob->parent_id = newblob->id;
4368 
4369 	/* Create new back_bs_dev for snapshot */
4370 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4371 	if (origblob->back_bs_dev == NULL) {
4372 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4373 		return;
4374 	}
4375 
4376 	/* set clone blob as thin provisioned */
4377 	_spdk_blob_set_thin_provision(origblob);
4378 
4379 	_spdk_bs_blob_list_add(newblob);
4380 
4381 	/* Zero out origblob cluster map */
4382 	memset(origblob->active.clusters, 0,
4383 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4384 
4385 	/* sync clone metadata */
4386 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4387 }
4388 
4389 static void
4390 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4391 {
4392 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4393 	struct spdk_blob *origblob = ctx->original.blob;
4394 	struct spdk_blob *newblob = ctx->new.blob;
4395 	int bserrno;
4396 
4397 	if (rc != 0) {
4398 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4399 		return;
4400 	}
4401 
4402 	ctx->frozen = true;
4403 
4404 	/* set new back_bs_dev for snapshot */
4405 	newblob->back_bs_dev = origblob->back_bs_dev;
4406 	/* Set invalid flags from origblob */
4407 	newblob->invalid_flags = origblob->invalid_flags;
4408 
4409 	/* inherit parent from original blob if set */
4410 	newblob->parent_id = origblob->parent_id;
4411 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4412 		/* Set internal xattr for snapshot id */
4413 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4414 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4415 		if (bserrno != 0) {
4416 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4417 			return;
4418 		}
4419 	}
4420 
4421 	/* Copy cluster map to snapshot */
4422 	memcpy(newblob->active.clusters, origblob->active.clusters,
4423 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4424 
4425 	/* sync snapshot metadata */
4426 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4427 }
4428 
4429 static void
4430 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
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 = _blob;
4435 
4436 	if (bserrno != 0) {
4437 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4438 		return;
4439 	}
4440 
4441 	ctx->new.blob = newblob;
4442 
4443 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4444 }
4445 
4446 static void
4447 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4448 {
4449 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4450 	struct spdk_blob *origblob = ctx->original.blob;
4451 
4452 	if (bserrno != 0) {
4453 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4454 		return;
4455 	}
4456 
4457 	ctx->new.id = blobid;
4458 	ctx->cpl.u.blobid.blobid = blobid;
4459 
4460 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4461 }
4462 
4463 
4464 static void
4465 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4466 			const void **value, size_t *value_len)
4467 {
4468 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4469 
4470 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4471 	*value = &blob->id;
4472 	*value_len = sizeof(blob->id);
4473 }
4474 
4475 static void
4476 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4477 {
4478 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4479 	struct spdk_blob_opts opts;
4480 	struct spdk_blob_xattr_opts internal_xattrs;
4481 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4482 
4483 	if (bserrno != 0) {
4484 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4485 		return;
4486 	}
4487 
4488 	ctx->original.blob = _blob;
4489 
4490 	if (_blob->data_ro || _blob->md_ro) {
4491 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4492 			      _blob->id);
4493 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4494 		return;
4495 	}
4496 
4497 	spdk_blob_opts_init(&opts);
4498 	_spdk_blob_xattrs_init(&internal_xattrs);
4499 
4500 	/* Change the size of new blob to the same as in original blob,
4501 	 * but do not allocate clusters */
4502 	opts.thin_provision = true;
4503 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4504 
4505 	/* If there are any xattrs specified for snapshot, set them now */
4506 	if (ctx->xattrs) {
4507 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4508 	}
4509 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4510 	internal_xattrs.count = 1;
4511 	internal_xattrs.ctx = _blob;
4512 	internal_xattrs.names = xattrs_names;
4513 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4514 
4515 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4516 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4517 }
4518 
4519 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4520 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4521 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4522 {
4523 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4524 
4525 	if (!ctx) {
4526 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4527 		return;
4528 	}
4529 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4530 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4531 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4532 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4533 	ctx->bserrno = 0;
4534 	ctx->frozen = false;
4535 	ctx->original.id = blobid;
4536 	ctx->xattrs = snapshot_xattrs;
4537 
4538 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4539 }
4540 /* END spdk_bs_create_snapshot */
4541 
4542 /* START spdk_bs_create_clone */
4543 
4544 static void
4545 _spdk_bs_xattr_clone(void *arg, const char *name,
4546 		     const void **value, size_t *value_len)
4547 {
4548 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4549 
4550 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4551 	*value = &blob->id;
4552 	*value_len = sizeof(blob->id);
4553 }
4554 
4555 static void
4556 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4557 {
4558 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4559 	struct spdk_blob *clone = _blob;
4560 
4561 	ctx->new.blob = clone;
4562 	_spdk_bs_blob_list_add(clone);
4563 
4564 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4565 }
4566 
4567 static void
4568 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4569 {
4570 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4571 
4572 	ctx->cpl.u.blobid.blobid = blobid;
4573 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4574 }
4575 
4576 static void
4577 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4578 {
4579 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4580 	struct spdk_blob_opts		opts;
4581 	struct spdk_blob_xattr_opts internal_xattrs;
4582 	char *xattr_names[] = { BLOB_SNAPSHOT };
4583 
4584 	if (bserrno != 0) {
4585 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4586 		return;
4587 	}
4588 
4589 	ctx->original.blob = _blob;
4590 
4591 	if (!_blob->data_ro || !_blob->md_ro) {
4592 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4593 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4594 		return;
4595 	}
4596 
4597 	spdk_blob_opts_init(&opts);
4598 	_spdk_blob_xattrs_init(&internal_xattrs);
4599 
4600 	opts.thin_provision = true;
4601 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4602 	if (ctx->xattrs) {
4603 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4604 	}
4605 
4606 	/* Set internal xattr BLOB_SNAPSHOT */
4607 	internal_xattrs.count = 1;
4608 	internal_xattrs.ctx = _blob;
4609 	internal_xattrs.names = xattr_names;
4610 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4611 
4612 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4613 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4614 }
4615 
4616 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4617 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4618 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4619 {
4620 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4621 
4622 	if (!ctx) {
4623 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4624 		return;
4625 	}
4626 
4627 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4628 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4629 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4630 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4631 	ctx->bserrno = 0;
4632 	ctx->xattrs = clone_xattrs;
4633 	ctx->original.id = blobid;
4634 
4635 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4636 }
4637 
4638 /* END spdk_bs_create_clone */
4639 
4640 /* START spdk_bs_inflate_blob */
4641 
4642 static void
4643 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4644 {
4645 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4646 	struct spdk_blob *_blob = ctx->original.blob;
4647 
4648 	if (bserrno != 0) {
4649 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4650 		return;
4651 	}
4652 
4653 	assert(_parent != NULL);
4654 
4655 	_spdk_bs_blob_list_remove(_blob);
4656 	_blob->parent_id = _parent->id;
4657 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4658 			     sizeof(spdk_blob_id), true);
4659 
4660 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4661 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4662 	_spdk_bs_blob_list_add(_blob);
4663 
4664 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4665 }
4666 
4667 static void
4668 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4669 {
4670 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4671 	struct spdk_blob *_blob = ctx->original.blob;
4672 	struct spdk_blob *_parent;
4673 
4674 	if (bserrno != 0) {
4675 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4676 		return;
4677 	}
4678 
4679 	if (ctx->allocate_all) {
4680 		/* remove thin provisioning */
4681 		_spdk_bs_blob_list_remove(_blob);
4682 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4683 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4684 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4685 		_blob->back_bs_dev = NULL;
4686 		_blob->parent_id = SPDK_BLOBID_INVALID;
4687 	} else {
4688 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4689 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4690 			/* We must change the parent of the inflated blob */
4691 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4692 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4693 			return;
4694 		}
4695 
4696 		_spdk_bs_blob_list_remove(_blob);
4697 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4698 		_blob->parent_id = SPDK_BLOBID_INVALID;
4699 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4700 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4701 	}
4702 
4703 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4704 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4705 }
4706 
4707 /* Check if cluster needs allocation */
4708 static inline bool
4709 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4710 {
4711 	struct spdk_blob_bs_dev *b;
4712 
4713 	assert(blob != NULL);
4714 
4715 	if (blob->active.clusters[cluster] != 0) {
4716 		/* Cluster is already allocated */
4717 		return false;
4718 	}
4719 
4720 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4721 		/* Blob have no parent blob */
4722 		return allocate_all;
4723 	}
4724 
4725 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4726 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4727 }
4728 
4729 static void
4730 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4731 {
4732 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4733 	struct spdk_blob *_blob = ctx->original.blob;
4734 	uint64_t offset;
4735 
4736 	if (bserrno != 0) {
4737 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4738 		return;
4739 	}
4740 
4741 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4742 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4743 			break;
4744 		}
4745 	}
4746 
4747 	if (ctx->cluster < _blob->active.num_clusters) {
4748 		offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster);
4749 
4750 		/* We may safely increment a cluster before write */
4751 		ctx->cluster++;
4752 
4753 		/* Use zero length write to touch a cluster */
4754 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4755 				   _spdk_bs_inflate_blob_touch_next, ctx);
4756 	} else {
4757 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4758 	}
4759 }
4760 
4761 static void
4762 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4763 {
4764 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4765 	uint64_t lfc; /* lowest free cluster */
4766 	uint64_t i;
4767 
4768 	if (bserrno != 0) {
4769 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4770 		return;
4771 	}
4772 	ctx->original.blob = _blob;
4773 
4774 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4775 		/* This blob have no parent, so we cannot decouple it. */
4776 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4777 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4778 		return;
4779 	}
4780 
4781 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4782 		/* This is not thin provisioned blob. No need to inflate. */
4783 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4784 		return;
4785 	}
4786 
4787 	/* Do two passes - one to verify that we can obtain enough clusters
4788 	 * and another to actually claim them.
4789 	 */
4790 	lfc = 0;
4791 	for (i = 0; i < _blob->active.num_clusters; i++) {
4792 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4793 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4794 			if (lfc == UINT32_MAX) {
4795 				/* No more free clusters. Cannot satisfy the request */
4796 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4797 				return;
4798 			}
4799 			lfc++;
4800 		}
4801 	}
4802 
4803 	ctx->cluster = 0;
4804 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4805 }
4806 
4807 static void
4808 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4809 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4810 {
4811 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4812 
4813 	if (!ctx) {
4814 		cb_fn(cb_arg, -ENOMEM);
4815 		return;
4816 	}
4817 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4818 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4819 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4820 	ctx->bserrno = 0;
4821 	ctx->original.id = blobid;
4822 	ctx->channel = channel;
4823 	ctx->allocate_all = allocate_all;
4824 
4825 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4826 }
4827 
4828 void
4829 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4830 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4831 {
4832 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4833 }
4834 
4835 void
4836 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4837 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4838 {
4839 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
4840 }
4841 /* END spdk_bs_inflate_blob */
4842 
4843 /* START spdk_blob_resize */
4844 struct spdk_bs_resize_ctx {
4845 	spdk_blob_op_complete cb_fn;
4846 	void *cb_arg;
4847 	struct spdk_blob *blob;
4848 	uint64_t sz;
4849 	int rc;
4850 };
4851 
4852 static void
4853 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4854 {
4855 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4856 
4857 	if (rc != 0) {
4858 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4859 	}
4860 
4861 	if (ctx->rc != 0) {
4862 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4863 		rc = ctx->rc;
4864 	}
4865 
4866 	ctx->blob->resize_in_progress = false;
4867 
4868 	ctx->cb_fn(ctx->cb_arg, rc);
4869 	free(ctx);
4870 }
4871 
4872 static void
4873 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4874 {
4875 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4876 
4877 	if (rc != 0) {
4878 		ctx->blob->resize_in_progress = false;
4879 		ctx->cb_fn(ctx->cb_arg, rc);
4880 		free(ctx);
4881 		return;
4882 	}
4883 
4884 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4885 
4886 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4887 }
4888 
4889 void
4890 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4891 {
4892 	struct spdk_bs_resize_ctx *ctx;
4893 
4894 	_spdk_blob_verify_md_op(blob);
4895 
4896 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4897 
4898 	if (blob->md_ro) {
4899 		cb_fn(cb_arg, -EPERM);
4900 		return;
4901 	}
4902 
4903 	if (sz == blob->active.num_clusters) {
4904 		cb_fn(cb_arg, 0);
4905 		return;
4906 	}
4907 
4908 	if (blob->resize_in_progress) {
4909 		cb_fn(cb_arg, -EBUSY);
4910 		return;
4911 	}
4912 
4913 	ctx = calloc(1, sizeof(*ctx));
4914 	if (!ctx) {
4915 		cb_fn(cb_arg, -ENOMEM);
4916 		return;
4917 	}
4918 
4919 	blob->resize_in_progress = true;
4920 	ctx->cb_fn = cb_fn;
4921 	ctx->cb_arg = cb_arg;
4922 	ctx->blob = blob;
4923 	ctx->sz = sz;
4924 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
4925 }
4926 
4927 /* END spdk_blob_resize */
4928 
4929 
4930 /* START spdk_bs_delete_blob */
4931 
4932 static void
4933 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
4934 {
4935 	spdk_bs_sequence_t *seq = cb_arg;
4936 
4937 	spdk_bs_sequence_finish(seq, bserrno);
4938 }
4939 
4940 static void
4941 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4942 {
4943 	struct spdk_blob *blob = cb_arg;
4944 
4945 	if (bserrno != 0) {
4946 		/*
4947 		 * We already removed this blob from the blobstore tailq, so
4948 		 *  we need to free it here since this is the last reference
4949 		 *  to it.
4950 		 */
4951 		_spdk_blob_free(blob);
4952 		_spdk_bs_delete_close_cpl(seq, bserrno);
4953 		return;
4954 	}
4955 
4956 	/*
4957 	 * This will immediately decrement the ref_count and call
4958 	 *  the completion routine since the metadata state is clean.
4959 	 *  By calling spdk_blob_close, we reduce the number of call
4960 	 *  points into code that touches the blob->open_ref count
4961 	 *  and the blobstore's blob list.
4962 	 */
4963 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
4964 }
4965 
4966 static void
4967 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
4968 {
4969 	spdk_bs_sequence_t *seq = cb_arg;
4970 	struct spdk_blob_list *snapshot = NULL;
4971 	uint32_t page_num;
4972 
4973 	if (bserrno != 0) {
4974 		spdk_bs_sequence_finish(seq, bserrno);
4975 		return;
4976 	}
4977 
4978 	_spdk_blob_verify_md_op(blob);
4979 
4980 	if (blob->open_ref > 1) {
4981 		/*
4982 		 * Someone has this blob open (besides this delete context).
4983 		 *  Decrement the ref count directly and return -EBUSY.
4984 		 */
4985 		blob->open_ref--;
4986 		spdk_bs_sequence_finish(seq, -EBUSY);
4987 		return;
4988 	}
4989 
4990 	bserrno = _spdk_bs_blob_list_remove(blob);
4991 	if (bserrno != 0) {
4992 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Remove blob #%" PRIu64 " from a list\n", blob->id);
4993 		spdk_bs_sequence_finish(seq, bserrno);
4994 		return;
4995 	}
4996 
4997 	/*
4998 	 * Remove the blob from the blob_store list now, to ensure it does not
4999 	 *  get returned after this point by _spdk_blob_lookup().
5000 	 */
5001 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5002 
5003 	/* If blob is a snapshot then remove it from the list */
5004 	TAILQ_FOREACH(snapshot, &blob->bs->snapshots, link) {
5005 		if (snapshot->id == blob->id) {
5006 			TAILQ_REMOVE(&blob->bs->snapshots, snapshot, link);
5007 			free(snapshot);
5008 			break;
5009 		}
5010 	}
5011 
5012 	page_num = _spdk_bs_blobid_to_page(blob->id);
5013 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
5014 	blob->state = SPDK_BLOB_STATE_DIRTY;
5015 	blob->active.num_pages = 0;
5016 	_spdk_blob_resize(blob, 0);
5017 
5018 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
5019 }
5020 
5021 void
5022 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5023 		    spdk_blob_op_complete cb_fn, void *cb_arg)
5024 {
5025 	struct spdk_bs_cpl	cpl;
5026 	spdk_bs_sequence_t	*seq;
5027 	struct spdk_blob_list	*snapshot_entry = NULL;
5028 
5029 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
5030 
5031 	assert(spdk_get_thread() == bs->md_thread);
5032 
5033 	/* Check if this is a snapshot with clones */
5034 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5035 		if (snapshot_entry->id == blobid) {
5036 			break;
5037 		}
5038 	}
5039 	if (snapshot_entry != NULL) {
5040 		/* If snapshot have clones, we cannot remove it */
5041 		if (!TAILQ_EMPTY(&snapshot_entry->clones)) {
5042 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
5043 			cb_fn(cb_arg, -EBUSY);
5044 			return;
5045 		}
5046 	}
5047 
5048 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5049 	cpl.u.blob_basic.cb_fn = cb_fn;
5050 	cpl.u.blob_basic.cb_arg = cb_arg;
5051 
5052 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5053 	if (!seq) {
5054 		cb_fn(cb_arg, -ENOMEM);
5055 		return;
5056 	}
5057 
5058 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5059 }
5060 
5061 /* END spdk_bs_delete_blob */
5062 
5063 /* START spdk_bs_open_blob */
5064 
5065 static void
5066 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5067 {
5068 	struct spdk_blob *blob = cb_arg;
5069 
5070 	/* If the blob have crc error, we just return NULL. */
5071 	if (blob == NULL) {
5072 		seq->cpl.u.blob_handle.blob = NULL;
5073 		spdk_bs_sequence_finish(seq, bserrno);
5074 		return;
5075 	}
5076 
5077 	blob->open_ref++;
5078 
5079 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5080 
5081 	spdk_bs_sequence_finish(seq, bserrno);
5082 }
5083 
5084 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5085 			       struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5086 {
5087 	struct spdk_blob		*blob;
5088 	struct spdk_bs_cpl		cpl;
5089 	struct spdk_blob_open_opts	opts_default;
5090 	spdk_bs_sequence_t		*seq;
5091 	uint32_t			page_num;
5092 
5093 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5094 	assert(spdk_get_thread() == bs->md_thread);
5095 
5096 	page_num = _spdk_bs_blobid_to_page(blobid);
5097 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5098 		/* Invalid blobid */
5099 		cb_fn(cb_arg, NULL, -ENOENT);
5100 		return;
5101 	}
5102 
5103 	blob = _spdk_blob_lookup(bs, blobid);
5104 	if (blob) {
5105 		blob->open_ref++;
5106 		cb_fn(cb_arg, blob, 0);
5107 		return;
5108 	}
5109 
5110 	blob = _spdk_blob_alloc(bs, blobid);
5111 	if (!blob) {
5112 		cb_fn(cb_arg, NULL, -ENOMEM);
5113 		return;
5114 	}
5115 
5116 	if (!opts) {
5117 		spdk_blob_open_opts_init(&opts_default);
5118 		opts = &opts_default;
5119 	}
5120 
5121 	blob->clear_method = opts->clear_method;
5122 
5123 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5124 	cpl.u.blob_handle.cb_fn = cb_fn;
5125 	cpl.u.blob_handle.cb_arg = cb_arg;
5126 	cpl.u.blob_handle.blob = blob;
5127 
5128 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5129 	if (!seq) {
5130 		_spdk_blob_free(blob);
5131 		cb_fn(cb_arg, NULL, -ENOMEM);
5132 		return;
5133 	}
5134 
5135 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5136 }
5137 
5138 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5139 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5140 {
5141 	_spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
5142 }
5143 
5144 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
5145 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5146 {
5147 	_spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
5148 }
5149 
5150 /* END spdk_bs_open_blob */
5151 
5152 /* START spdk_blob_set_read_only */
5153 int spdk_blob_set_read_only(struct spdk_blob *blob)
5154 {
5155 	_spdk_blob_verify_md_op(blob);
5156 
5157 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5158 
5159 	blob->state = SPDK_BLOB_STATE_DIRTY;
5160 	return 0;
5161 }
5162 /* END spdk_blob_set_read_only */
5163 
5164 /* START spdk_blob_sync_md */
5165 
5166 static void
5167 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5168 {
5169 	struct spdk_blob *blob = cb_arg;
5170 
5171 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5172 		blob->data_ro = true;
5173 		blob->md_ro = true;
5174 	}
5175 
5176 	spdk_bs_sequence_finish(seq, bserrno);
5177 }
5178 
5179 static void
5180 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5181 {
5182 	struct spdk_bs_cpl	cpl;
5183 	spdk_bs_sequence_t	*seq;
5184 
5185 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5186 	cpl.u.blob_basic.cb_fn = cb_fn;
5187 	cpl.u.blob_basic.cb_arg = cb_arg;
5188 
5189 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5190 	if (!seq) {
5191 		cb_fn(cb_arg, -ENOMEM);
5192 		return;
5193 	}
5194 
5195 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5196 }
5197 
5198 void
5199 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5200 {
5201 	_spdk_blob_verify_md_op(blob);
5202 
5203 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5204 
5205 	if (blob->md_ro) {
5206 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5207 		cb_fn(cb_arg, 0);
5208 		return;
5209 	}
5210 
5211 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5212 }
5213 
5214 /* END spdk_blob_sync_md */
5215 
5216 struct spdk_blob_insert_cluster_ctx {
5217 	struct spdk_thread	*thread;
5218 	struct spdk_blob	*blob;
5219 	uint32_t		cluster_num;	/* cluster index in blob */
5220 	uint32_t		cluster;	/* cluster on disk */
5221 	int			rc;
5222 	spdk_blob_op_complete	cb_fn;
5223 	void			*cb_arg;
5224 };
5225 
5226 static void
5227 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5228 {
5229 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5230 
5231 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5232 	free(ctx);
5233 }
5234 
5235 static void
5236 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5237 {
5238 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5239 
5240 	ctx->rc = bserrno;
5241 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5242 }
5243 
5244 static void
5245 _spdk_blob_insert_cluster_msg(void *arg)
5246 {
5247 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5248 
5249 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5250 	if (ctx->rc != 0) {
5251 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5252 		return;
5253 	}
5254 
5255 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5256 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5257 }
5258 
5259 static void
5260 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5261 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5262 {
5263 	struct spdk_blob_insert_cluster_ctx *ctx;
5264 
5265 	ctx = calloc(1, sizeof(*ctx));
5266 	if (ctx == NULL) {
5267 		cb_fn(cb_arg, -ENOMEM);
5268 		return;
5269 	}
5270 
5271 	ctx->thread = spdk_get_thread();
5272 	ctx->blob = blob;
5273 	ctx->cluster_num = cluster_num;
5274 	ctx->cluster = cluster;
5275 	ctx->cb_fn = cb_fn;
5276 	ctx->cb_arg = cb_arg;
5277 
5278 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5279 }
5280 
5281 /* START spdk_blob_close */
5282 
5283 static void
5284 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5285 {
5286 	struct spdk_blob *blob = cb_arg;
5287 
5288 	if (bserrno == 0) {
5289 		blob->open_ref--;
5290 		if (blob->open_ref == 0) {
5291 			/*
5292 			 * Blobs with active.num_pages == 0 are deleted blobs.
5293 			 *  these blobs are removed from the blob_store list
5294 			 *  when the deletion process starts - so don't try to
5295 			 *  remove them again.
5296 			 */
5297 			if (blob->active.num_pages > 0) {
5298 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5299 			}
5300 			_spdk_blob_free(blob);
5301 		}
5302 	}
5303 
5304 	spdk_bs_sequence_finish(seq, bserrno);
5305 }
5306 
5307 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5308 {
5309 	struct spdk_bs_cpl	cpl;
5310 	spdk_bs_sequence_t	*seq;
5311 
5312 	_spdk_blob_verify_md_op(blob);
5313 
5314 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5315 
5316 	if (blob->open_ref == 0) {
5317 		cb_fn(cb_arg, -EBADF);
5318 		return;
5319 	}
5320 
5321 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5322 	cpl.u.blob_basic.cb_fn = cb_fn;
5323 	cpl.u.blob_basic.cb_arg = cb_arg;
5324 
5325 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5326 	if (!seq) {
5327 		cb_fn(cb_arg, -ENOMEM);
5328 		return;
5329 	}
5330 
5331 	/* Sync metadata */
5332 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5333 }
5334 
5335 /* END spdk_blob_close */
5336 
5337 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5338 {
5339 	return spdk_get_io_channel(bs);
5340 }
5341 
5342 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5343 {
5344 	spdk_put_io_channel(channel);
5345 }
5346 
5347 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5348 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5349 {
5350 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5351 				     SPDK_BLOB_UNMAP);
5352 }
5353 
5354 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5355 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5356 {
5357 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5358 				     SPDK_BLOB_WRITE_ZEROES);
5359 }
5360 
5361 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5362 			void *payload, uint64_t offset, uint64_t length,
5363 			spdk_blob_op_complete cb_fn, void *cb_arg)
5364 {
5365 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5366 				     SPDK_BLOB_WRITE);
5367 }
5368 
5369 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5370 		       void *payload, uint64_t offset, uint64_t length,
5371 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5372 {
5373 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5374 				     SPDK_BLOB_READ);
5375 }
5376 
5377 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5378 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5379 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5380 {
5381 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5382 }
5383 
5384 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5385 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5386 			spdk_blob_op_complete cb_fn, void *cb_arg)
5387 {
5388 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5389 }
5390 
5391 struct spdk_bs_iter_ctx {
5392 	int64_t page_num;
5393 	struct spdk_blob_store *bs;
5394 
5395 	spdk_blob_op_with_handle_complete cb_fn;
5396 	void *cb_arg;
5397 };
5398 
5399 static void
5400 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5401 {
5402 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5403 	struct spdk_blob_store *bs = ctx->bs;
5404 	spdk_blob_id id;
5405 
5406 	if (bserrno == 0) {
5407 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5408 		free(ctx);
5409 		return;
5410 	}
5411 
5412 	ctx->page_num++;
5413 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5414 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5415 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5416 		free(ctx);
5417 		return;
5418 	}
5419 
5420 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5421 
5422 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5423 }
5424 
5425 void
5426 spdk_bs_iter_first(struct spdk_blob_store *bs,
5427 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5428 {
5429 	struct spdk_bs_iter_ctx *ctx;
5430 
5431 	ctx = calloc(1, sizeof(*ctx));
5432 	if (!ctx) {
5433 		cb_fn(cb_arg, NULL, -ENOMEM);
5434 		return;
5435 	}
5436 
5437 	ctx->page_num = -1;
5438 	ctx->bs = bs;
5439 	ctx->cb_fn = cb_fn;
5440 	ctx->cb_arg = cb_arg;
5441 
5442 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5443 }
5444 
5445 static void
5446 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5447 {
5448 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5449 
5450 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5451 }
5452 
5453 void
5454 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5455 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5456 {
5457 	struct spdk_bs_iter_ctx *ctx;
5458 
5459 	assert(blob != NULL);
5460 
5461 	ctx = calloc(1, sizeof(*ctx));
5462 	if (!ctx) {
5463 		cb_fn(cb_arg, NULL, -ENOMEM);
5464 		return;
5465 	}
5466 
5467 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5468 	ctx->bs = bs;
5469 	ctx->cb_fn = cb_fn;
5470 	ctx->cb_arg = cb_arg;
5471 
5472 	/* Close the existing blob */
5473 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5474 }
5475 
5476 static int
5477 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5478 		     uint16_t value_len, bool internal)
5479 {
5480 	struct spdk_xattr_tailq *xattrs;
5481 	struct spdk_xattr	*xattr;
5482 
5483 	_spdk_blob_verify_md_op(blob);
5484 
5485 	if (blob->md_ro) {
5486 		return -EPERM;
5487 	}
5488 
5489 	if (internal) {
5490 		xattrs = &blob->xattrs_internal;
5491 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5492 	} else {
5493 		xattrs = &blob->xattrs;
5494 	}
5495 
5496 	TAILQ_FOREACH(xattr, xattrs, link) {
5497 		if (!strcmp(name, xattr->name)) {
5498 			free(xattr->value);
5499 			xattr->value_len = value_len;
5500 			xattr->value = malloc(value_len);
5501 			memcpy(xattr->value, value, value_len);
5502 
5503 			blob->state = SPDK_BLOB_STATE_DIRTY;
5504 
5505 			return 0;
5506 		}
5507 	}
5508 
5509 	xattr = calloc(1, sizeof(*xattr));
5510 	if (!xattr) {
5511 		return -ENOMEM;
5512 	}
5513 	xattr->name = strdup(name);
5514 	xattr->value_len = value_len;
5515 	xattr->value = malloc(value_len);
5516 	memcpy(xattr->value, value, value_len);
5517 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5518 
5519 	blob->state = SPDK_BLOB_STATE_DIRTY;
5520 
5521 	return 0;
5522 }
5523 
5524 int
5525 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5526 		    uint16_t value_len)
5527 {
5528 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5529 }
5530 
5531 static int
5532 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5533 {
5534 	struct spdk_xattr_tailq *xattrs;
5535 	struct spdk_xattr	*xattr;
5536 
5537 	_spdk_blob_verify_md_op(blob);
5538 
5539 	if (blob->md_ro) {
5540 		return -EPERM;
5541 	}
5542 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5543 
5544 	TAILQ_FOREACH(xattr, xattrs, link) {
5545 		if (!strcmp(name, xattr->name)) {
5546 			TAILQ_REMOVE(xattrs, xattr, link);
5547 			free(xattr->value);
5548 			free(xattr->name);
5549 			free(xattr);
5550 
5551 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5552 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5553 			}
5554 			blob->state = SPDK_BLOB_STATE_DIRTY;
5555 
5556 			return 0;
5557 		}
5558 	}
5559 
5560 	return -ENOENT;
5561 }
5562 
5563 int
5564 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5565 {
5566 	return _spdk_blob_remove_xattr(blob, name, false);
5567 }
5568 
5569 static int
5570 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5571 			   const void **value, size_t *value_len, bool internal)
5572 {
5573 	struct spdk_xattr	*xattr;
5574 	struct spdk_xattr_tailq *xattrs;
5575 
5576 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5577 
5578 	TAILQ_FOREACH(xattr, xattrs, link) {
5579 		if (!strcmp(name, xattr->name)) {
5580 			*value = xattr->value;
5581 			*value_len = xattr->value_len;
5582 			return 0;
5583 		}
5584 	}
5585 	return -ENOENT;
5586 }
5587 
5588 int
5589 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5590 			  const void **value, size_t *value_len)
5591 {
5592 	_spdk_blob_verify_md_op(blob);
5593 
5594 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5595 }
5596 
5597 struct spdk_xattr_names {
5598 	uint32_t	count;
5599 	const char	*names[0];
5600 };
5601 
5602 static int
5603 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5604 {
5605 	struct spdk_xattr	*xattr;
5606 	int			count = 0;
5607 
5608 	TAILQ_FOREACH(xattr, xattrs, link) {
5609 		count++;
5610 	}
5611 
5612 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5613 	if (*names == NULL) {
5614 		return -ENOMEM;
5615 	}
5616 
5617 	TAILQ_FOREACH(xattr, xattrs, link) {
5618 		(*names)->names[(*names)->count++] = xattr->name;
5619 	}
5620 
5621 	return 0;
5622 }
5623 
5624 int
5625 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5626 {
5627 	_spdk_blob_verify_md_op(blob);
5628 
5629 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5630 }
5631 
5632 uint32_t
5633 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5634 {
5635 	assert(names != NULL);
5636 
5637 	return names->count;
5638 }
5639 
5640 const char *
5641 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5642 {
5643 	if (index >= names->count) {
5644 		return NULL;
5645 	}
5646 
5647 	return names->names[index];
5648 }
5649 
5650 void
5651 spdk_xattr_names_free(struct spdk_xattr_names *names)
5652 {
5653 	free(names);
5654 }
5655 
5656 struct spdk_bs_type
5657 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5658 {
5659 	return bs->bstype;
5660 }
5661 
5662 void
5663 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5664 {
5665 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5666 }
5667 
5668 bool
5669 spdk_blob_is_read_only(struct spdk_blob *blob)
5670 {
5671 	assert(blob != NULL);
5672 	return (blob->data_ro || blob->md_ro);
5673 }
5674 
5675 bool
5676 spdk_blob_is_snapshot(struct spdk_blob *blob)
5677 {
5678 	struct spdk_blob_list *snapshot_entry;
5679 
5680 	assert(blob != NULL);
5681 
5682 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
5683 		if (snapshot_entry->id == blob->id) {
5684 			break;
5685 		}
5686 	}
5687 
5688 	if (snapshot_entry == NULL) {
5689 		return false;
5690 	}
5691 
5692 	return true;
5693 }
5694 
5695 bool
5696 spdk_blob_is_clone(struct spdk_blob *blob)
5697 {
5698 	assert(blob != NULL);
5699 
5700 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5701 		assert(spdk_blob_is_thin_provisioned(blob));
5702 		return true;
5703 	}
5704 
5705 	return false;
5706 }
5707 
5708 bool
5709 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5710 {
5711 	assert(blob != NULL);
5712 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5713 }
5714 
5715 spdk_blob_id
5716 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5717 {
5718 	struct spdk_blob_list *snapshot_entry = NULL;
5719 	struct spdk_blob_list *clone_entry = NULL;
5720 
5721 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5722 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5723 			if (clone_entry->id == blob_id) {
5724 				return snapshot_entry->id;
5725 			}
5726 		}
5727 	}
5728 
5729 	return SPDK_BLOBID_INVALID;
5730 }
5731 
5732 int
5733 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5734 		     size_t *count)
5735 {
5736 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5737 	size_t n;
5738 
5739 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5740 		if (snapshot_entry->id == blobid) {
5741 			break;
5742 		}
5743 	}
5744 	if (snapshot_entry == NULL) {
5745 		*count = 0;
5746 		return 0;
5747 	}
5748 
5749 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5750 		*count = snapshot_entry->clone_count;
5751 		return -ENOMEM;
5752 	}
5753 	*count = snapshot_entry->clone_count;
5754 
5755 	n = 0;
5756 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5757 		ids[n++] = clone_entry->id;
5758 	}
5759 
5760 	return 0;
5761 }
5762 
5763 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5764