xref: /spdk/lib/blob/blobstore.c (revision 41f2d0e44806bf4d0cb6d83391296b12f58c3e5b)
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 struct spdk_blob_list *
73 _spdk_bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid)
74 {
75 	struct spdk_blob_list *snapshot_entry = NULL;
76 
77 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
78 		if (snapshot_entry->id == blobid) {
79 			break;
80 		}
81 	}
82 
83 	return snapshot_entry;
84 }
85 
86 static void
87 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
88 {
89 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
90 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false);
91 	assert(bs->num_free_clusters > 0);
92 
93 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num);
94 
95 	spdk_bit_array_set(bs->used_clusters, cluster_num);
96 	bs->num_free_clusters--;
97 }
98 
99 static int
100 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
101 {
102 	uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
103 
104 	_spdk_blob_verify_md_op(blob);
105 
106 	if (*cluster_lba != 0) {
107 		return -EEXIST;
108 	}
109 
110 	*cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster);
111 	return 0;
112 }
113 
114 static int
115 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
116 			  uint64_t *lowest_free_cluster, bool update_map)
117 {
118 	pthread_mutex_lock(&blob->bs->used_clusters_mutex);
119 	*lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters,
120 			       *lowest_free_cluster);
121 	if (*lowest_free_cluster == UINT32_MAX) {
122 		/* No more free clusters. Cannot satisfy the request */
123 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
124 		return -ENOSPC;
125 	}
126 
127 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id);
128 	_spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster);
129 	pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
130 
131 	if (update_map) {
132 		_spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster);
133 	}
134 
135 	return 0;
136 }
137 
138 static void
139 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
140 {
141 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
142 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true);
143 	assert(bs->num_free_clusters < bs->total_clusters);
144 
145 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num);
146 
147 	pthread_mutex_lock(&bs->used_clusters_mutex);
148 	spdk_bit_array_clear(bs->used_clusters, cluster_num);
149 	bs->num_free_clusters++;
150 	pthread_mutex_unlock(&bs->used_clusters_mutex);
151 }
152 
153 static void
154 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
155 {
156 	xattrs->count = 0;
157 	xattrs->names = NULL;
158 	xattrs->ctx = NULL;
159 	xattrs->get_value = NULL;
160 }
161 
162 void
163 spdk_blob_opts_init(struct spdk_blob_opts *opts)
164 {
165 	opts->num_clusters = 0;
166 	opts->thin_provision = false;
167 	_spdk_blob_xattrs_init(&opts->xattrs);
168 }
169 
170 void
171 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts)
172 {
173 	opts->clear_method = BLOB_CLEAR_WITH_UNMAP;
174 }
175 
176 static struct spdk_blob *
177 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
178 {
179 	struct spdk_blob *blob;
180 
181 	blob = calloc(1, sizeof(*blob));
182 	if (!blob) {
183 		return NULL;
184 	}
185 
186 	blob->id = id;
187 	blob->bs = bs;
188 
189 	blob->parent_id = SPDK_BLOBID_INVALID;
190 
191 	blob->state = SPDK_BLOB_STATE_DIRTY;
192 	blob->active.num_pages = 1;
193 	blob->active.pages = calloc(1, sizeof(*blob->active.pages));
194 	if (!blob->active.pages) {
195 		free(blob);
196 		return NULL;
197 	}
198 
199 	blob->active.pages[0] = _spdk_bs_blobid_to_page(id);
200 
201 	TAILQ_INIT(&blob->xattrs);
202 	TAILQ_INIT(&blob->xattrs_internal);
203 
204 	return blob;
205 }
206 
207 static void
208 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs)
209 {
210 	struct spdk_xattr	*xattr, *xattr_tmp;
211 
212 	TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
213 		TAILQ_REMOVE(xattrs, xattr, link);
214 		free(xattr->name);
215 		free(xattr->value);
216 		free(xattr);
217 	}
218 }
219 
220 static void
221 _spdk_blob_free(struct spdk_blob *blob)
222 {
223 	assert(blob != NULL);
224 
225 	free(blob->active.clusters);
226 	free(blob->clean.clusters);
227 	free(blob->active.pages);
228 	free(blob->clean.pages);
229 
230 	_spdk_xattrs_free(&blob->xattrs);
231 	_spdk_xattrs_free(&blob->xattrs_internal);
232 
233 	if (blob->back_bs_dev) {
234 		blob->back_bs_dev->destroy(blob->back_bs_dev);
235 	}
236 
237 	free(blob);
238 }
239 
240 struct freeze_io_ctx {
241 	struct spdk_bs_cpl cpl;
242 	struct spdk_blob *blob;
243 };
244 
245 static void
246 _spdk_blob_io_sync(struct spdk_io_channel_iter *i)
247 {
248 	spdk_for_each_channel_continue(i, 0);
249 }
250 
251 static void
252 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i)
253 {
254 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
255 	struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
256 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
257 	struct spdk_bs_request_set	*set;
258 	struct spdk_bs_user_op_args	*args;
259 	spdk_bs_user_op_t *op, *tmp;
260 
261 	TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
262 		set = (struct spdk_bs_request_set *)op;
263 		args = &set->u.user_op;
264 
265 		if (args->blob == ctx->blob) {
266 			TAILQ_REMOVE(&ch->queued_io, op, link);
267 			spdk_bs_user_op_execute(op);
268 		}
269 	}
270 
271 	spdk_for_each_channel_continue(i, 0);
272 }
273 
274 static void
275 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status)
276 {
277 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
278 
279 	ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
280 
281 	free(ctx);
282 }
283 
284 static void
285 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
286 {
287 	struct freeze_io_ctx *ctx;
288 
289 	ctx = calloc(1, sizeof(*ctx));
290 	if (!ctx) {
291 		cb_fn(cb_arg, -ENOMEM);
292 		return;
293 	}
294 
295 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
296 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
297 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
298 	ctx->blob = blob;
299 
300 	/* Freeze I/O on blob */
301 	blob->frozen_refcnt++;
302 
303 	if (blob->frozen_refcnt == 1) {
304 		spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl);
305 	} else {
306 		cb_fn(cb_arg, 0);
307 		free(ctx);
308 	}
309 }
310 
311 static void
312 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
313 {
314 	struct freeze_io_ctx *ctx;
315 
316 	ctx = calloc(1, sizeof(*ctx));
317 	if (!ctx) {
318 		cb_fn(cb_arg, -ENOMEM);
319 		return;
320 	}
321 
322 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
323 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
324 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
325 	ctx->blob = blob;
326 
327 	assert(blob->frozen_refcnt > 0);
328 
329 	blob->frozen_refcnt--;
330 
331 	if (blob->frozen_refcnt == 0) {
332 		spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl);
333 	} else {
334 		cb_fn(cb_arg, 0);
335 		free(ctx);
336 	}
337 }
338 
339 static int
340 _spdk_blob_mark_clean(struct spdk_blob *blob)
341 {
342 	uint64_t *clusters = NULL;
343 	uint32_t *pages = NULL;
344 
345 	assert(blob != NULL);
346 
347 	if (blob->active.num_clusters) {
348 		assert(blob->active.clusters);
349 		clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
350 		if (!clusters) {
351 			return -ENOMEM;
352 		}
353 		memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*clusters));
354 	}
355 
356 	if (blob->active.num_pages) {
357 		assert(blob->active.pages);
358 		pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
359 		if (!pages) {
360 			free(clusters);
361 			return -ENOMEM;
362 		}
363 		memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*pages));
364 	}
365 
366 	free(blob->clean.clusters);
367 	free(blob->clean.pages);
368 
369 	blob->clean.num_clusters = blob->active.num_clusters;
370 	blob->clean.clusters = blob->active.clusters;
371 	blob->clean.num_pages = blob->active.num_pages;
372 	blob->clean.pages = blob->active.pages;
373 
374 	blob->active.clusters = clusters;
375 	blob->active.pages = pages;
376 
377 	/* If the metadata was dirtied again while the metadata was being written to disk,
378 	 *  we do not want to revert the DIRTY state back to CLEAN here.
379 	 */
380 	if (blob->state == SPDK_BLOB_STATE_LOADING) {
381 		blob->state = SPDK_BLOB_STATE_CLEAN;
382 	}
383 
384 	return 0;
385 }
386 
387 static int
388 _spdk_blob_deserialize_xattr(struct spdk_blob *blob,
389 			     struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
390 {
391 	struct spdk_xattr                       *xattr;
392 
393 	if (desc_xattr->length != sizeof(desc_xattr->name_length) +
394 	    sizeof(desc_xattr->value_length) +
395 	    desc_xattr->name_length + desc_xattr->value_length) {
396 		return -EINVAL;
397 	}
398 
399 	xattr = calloc(1, sizeof(*xattr));
400 	if (xattr == NULL) {
401 		return -ENOMEM;
402 	}
403 
404 	xattr->name = malloc(desc_xattr->name_length + 1);
405 	if (xattr->name == NULL) {
406 		free(xattr);
407 		return -ENOMEM;
408 	}
409 	memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
410 	xattr->name[desc_xattr->name_length] = '\0';
411 
412 	xattr->value = malloc(desc_xattr->value_length);
413 	if (xattr->value == NULL) {
414 		free(xattr->name);
415 		free(xattr);
416 		return -ENOMEM;
417 	}
418 	xattr->value_len = desc_xattr->value_length;
419 	memcpy(xattr->value,
420 	       (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
421 	       desc_xattr->value_length);
422 
423 	TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
424 
425 	return 0;
426 }
427 
428 
429 static int
430 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
431 {
432 	struct spdk_blob_md_descriptor *desc;
433 	size_t	cur_desc = 0;
434 	void *tmp;
435 
436 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
437 	while (cur_desc < sizeof(page->descriptors)) {
438 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
439 			if (desc->length == 0) {
440 				/* If padding and length are 0, this terminates the page */
441 				break;
442 			}
443 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
444 			struct spdk_blob_md_descriptor_flags	*desc_flags;
445 
446 			desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
447 
448 			if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
449 				return -EINVAL;
450 			}
451 
452 			if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
453 			    SPDK_BLOB_INVALID_FLAGS_MASK) {
454 				return -EINVAL;
455 			}
456 
457 			if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
458 			    SPDK_BLOB_DATA_RO_FLAGS_MASK) {
459 				blob->data_ro = true;
460 				blob->md_ro = true;
461 			}
462 
463 			if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
464 			    SPDK_BLOB_MD_RO_FLAGS_MASK) {
465 				blob->md_ro = true;
466 			}
467 
468 			if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
469 				blob->data_ro = true;
470 				blob->md_ro = true;
471 			}
472 
473 			blob->invalid_flags = desc_flags->invalid_flags;
474 			blob->data_ro_flags = desc_flags->data_ro_flags;
475 			blob->md_ro_flags = desc_flags->md_ro_flags;
476 
477 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
478 			struct spdk_blob_md_descriptor_extent	*desc_extent;
479 			unsigned int				i, j;
480 			unsigned int				cluster_count = blob->active.num_clusters;
481 
482 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
483 
484 			if (desc_extent->length == 0 ||
485 			    (desc_extent->length % sizeof(desc_extent->extents[0]) != 0)) {
486 				return -EINVAL;
487 			}
488 
489 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
490 				for (j = 0; j < desc_extent->extents[i].length; j++) {
491 					if (desc_extent->extents[i].cluster_idx != 0) {
492 						if (!spdk_bit_array_get(blob->bs->used_clusters,
493 									desc_extent->extents[i].cluster_idx + j)) {
494 							return -EINVAL;
495 						}
496 					}
497 					cluster_count++;
498 				}
499 			}
500 
501 			if (cluster_count == 0) {
502 				return -EINVAL;
503 			}
504 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t));
505 			if (tmp == NULL) {
506 				return -ENOMEM;
507 			}
508 			blob->active.clusters = tmp;
509 			blob->active.cluster_array_size = cluster_count;
510 
511 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
512 				for (j = 0; j < desc_extent->extents[i].length; j++) {
513 					if (desc_extent->extents[i].cluster_idx != 0) {
514 						blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
515 								desc_extent->extents[i].cluster_idx + j);
516 					} else if (spdk_blob_is_thin_provisioned(blob)) {
517 						blob->active.clusters[blob->active.num_clusters++] = 0;
518 					} else {
519 						return -EINVAL;
520 					}
521 				}
522 			}
523 
524 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
525 			int rc;
526 
527 			rc = _spdk_blob_deserialize_xattr(blob,
528 							  (struct spdk_blob_md_descriptor_xattr *) desc, false);
529 			if (rc != 0) {
530 				return rc;
531 			}
532 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
533 			int rc;
534 
535 			rc = _spdk_blob_deserialize_xattr(blob,
536 							  (struct spdk_blob_md_descriptor_xattr *) desc, true);
537 			if (rc != 0) {
538 				return rc;
539 			}
540 		} else {
541 			/* Unrecognized descriptor type.  Do not fail - just continue to the
542 			 *  next descriptor.  If this descriptor is associated with some feature
543 			 *  defined in a newer version of blobstore, that version of blobstore
544 			 *  should create and set an associated feature flag to specify if this
545 			 *  blob can be loaded or not.
546 			 */
547 		}
548 
549 		/* Advance to the next descriptor */
550 		cur_desc += sizeof(*desc) + desc->length;
551 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
552 			break;
553 		}
554 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
555 	}
556 
557 	return 0;
558 }
559 
560 static int
561 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
562 		 struct spdk_blob *blob)
563 {
564 	const struct spdk_blob_md_page *page;
565 	uint32_t i;
566 	int rc;
567 
568 	assert(page_count > 0);
569 	assert(pages[0].sequence_num == 0);
570 	assert(blob != NULL);
571 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
572 	assert(blob->active.clusters == NULL);
573 
574 	/* The blobid provided doesn't match what's in the MD, this can
575 	 * happen for example if a bogus blobid is passed in through open.
576 	 */
577 	if (blob->id != pages[0].id) {
578 		SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n",
579 			    blob->id, pages[0].id);
580 		return -ENOENT;
581 	}
582 
583 	for (i = 0; i < page_count; i++) {
584 		page = &pages[i];
585 
586 		assert(page->id == blob->id);
587 		assert(page->sequence_num == i);
588 
589 		rc = _spdk_blob_parse_page(page, blob);
590 		if (rc != 0) {
591 			return rc;
592 		}
593 	}
594 
595 	return 0;
596 }
597 
598 static int
599 _spdk_blob_serialize_add_page(const struct spdk_blob *blob,
600 			      struct spdk_blob_md_page **pages,
601 			      uint32_t *page_count,
602 			      struct spdk_blob_md_page **last_page)
603 {
604 	struct spdk_blob_md_page *page;
605 
606 	assert(pages != NULL);
607 	assert(page_count != NULL);
608 
609 	if (*page_count == 0) {
610 		assert(*pages == NULL);
611 		*page_count = 1;
612 		*pages = spdk_malloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
613 				     NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
614 	} else {
615 		assert(*pages != NULL);
616 		(*page_count)++;
617 		*pages = spdk_realloc(*pages,
618 				      SPDK_BS_PAGE_SIZE * (*page_count),
619 				      SPDK_BS_PAGE_SIZE);
620 	}
621 
622 	if (*pages == NULL) {
623 		*page_count = 0;
624 		*last_page = NULL;
625 		return -ENOMEM;
626 	}
627 
628 	page = &(*pages)[*page_count - 1];
629 	memset(page, 0, sizeof(*page));
630 	page->id = blob->id;
631 	page->sequence_num = *page_count - 1;
632 	page->next = SPDK_INVALID_MD_PAGE;
633 	*last_page = page;
634 
635 	return 0;
636 }
637 
638 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
639  * Update required_sz on both success and failure.
640  *
641  */
642 static int
643 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
644 			   uint8_t *buf, size_t buf_sz,
645 			   size_t *required_sz, bool internal)
646 {
647 	struct spdk_blob_md_descriptor_xattr	*desc;
648 
649 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
650 		       strlen(xattr->name) +
651 		       xattr->value_len;
652 
653 	if (buf_sz < *required_sz) {
654 		return -1;
655 	}
656 
657 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
658 
659 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
660 	desc->length = sizeof(desc->name_length) +
661 		       sizeof(desc->value_length) +
662 		       strlen(xattr->name) +
663 		       xattr->value_len;
664 	desc->name_length = strlen(xattr->name);
665 	desc->value_length = xattr->value_len;
666 
667 	memcpy(desc->name, xattr->name, desc->name_length);
668 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
669 	       xattr->value,
670 	       desc->value_length);
671 
672 	return 0;
673 }
674 
675 static void
676 _spdk_blob_serialize_extent(const struct spdk_blob *blob,
677 			    uint64_t start_cluster, uint64_t *next_cluster,
678 			    uint8_t *buf, size_t buf_sz)
679 {
680 	struct spdk_blob_md_descriptor_extent *desc;
681 	size_t cur_sz;
682 	uint64_t i, extent_idx;
683 	uint64_t lba, lba_per_cluster, lba_count;
684 
685 	/* The buffer must have room for at least one extent */
686 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]);
687 	if (buf_sz < cur_sz) {
688 		*next_cluster = start_cluster;
689 		return;
690 	}
691 
692 	desc = (struct spdk_blob_md_descriptor_extent *)buf;
693 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT;
694 
695 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
696 
697 	lba = blob->active.clusters[start_cluster];
698 	lba_count = lba_per_cluster;
699 	extent_idx = 0;
700 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
701 		if ((lba + lba_count) == blob->active.clusters[i]) {
702 			lba_count += lba_per_cluster;
703 			continue;
704 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
705 			lba_count += lba_per_cluster;
706 			continue;
707 		}
708 		desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
709 		desc->extents[extent_idx].length = lba_count / lba_per_cluster;
710 		extent_idx++;
711 
712 		cur_sz += sizeof(desc->extents[extent_idx]);
713 
714 		if (buf_sz < cur_sz) {
715 			/* If we ran out of buffer space, return */
716 			desc->length = sizeof(desc->extents[0]) * extent_idx;
717 			*next_cluster = i;
718 			return;
719 		}
720 
721 		lba = blob->active.clusters[i];
722 		lba_count = lba_per_cluster;
723 	}
724 
725 	desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
726 	desc->extents[extent_idx].length = lba_count / lba_per_cluster;
727 	extent_idx++;
728 
729 	desc->length = sizeof(desc->extents[0]) * extent_idx;
730 	*next_cluster = blob->active.num_clusters;
731 
732 	return;
733 }
734 
735 static int
736 _spdk_blob_serialize_extents(const struct spdk_blob *blob,
737 			     struct spdk_blob_md_page **pages,
738 			     struct spdk_blob_md_page *cur_page,
739 			     uint32_t *page_count, uint8_t **buf,
740 			     size_t *remaining_sz)
741 {
742 	uint64_t				last_cluster;
743 	int					rc;
744 
745 	last_cluster = 0;
746 	while (last_cluster < blob->active.num_clusters) {
747 		_spdk_blob_serialize_extent(blob, last_cluster, &last_cluster, *buf, *remaining_sz);
748 
749 		if (last_cluster == blob->active.num_clusters) {
750 			break;
751 		}
752 
753 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
754 		if (rc < 0) {
755 			return rc;
756 		}
757 
758 		*buf = (uint8_t *)cur_page->descriptors;
759 		*remaining_sz = sizeof(cur_page->descriptors);
760 	}
761 
762 	return 0;
763 }
764 
765 static void
766 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
767 			   uint8_t *buf, size_t *buf_sz)
768 {
769 	struct spdk_blob_md_descriptor_flags *desc;
770 
771 	/*
772 	 * Flags get serialized first, so we should always have room for the flags
773 	 *  descriptor.
774 	 */
775 	assert(*buf_sz >= sizeof(*desc));
776 
777 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
778 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
779 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
780 	desc->invalid_flags = blob->invalid_flags;
781 	desc->data_ro_flags = blob->data_ro_flags;
782 	desc->md_ro_flags = blob->md_ro_flags;
783 
784 	*buf_sz -= sizeof(*desc);
785 }
786 
787 static int
788 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
789 			    const struct spdk_xattr_tailq *xattrs, bool internal,
790 			    struct spdk_blob_md_page **pages,
791 			    struct spdk_blob_md_page *cur_page,
792 			    uint32_t *page_count, uint8_t **buf,
793 			    size_t *remaining_sz)
794 {
795 	const struct spdk_xattr	*xattr;
796 	int	rc;
797 
798 	TAILQ_FOREACH(xattr, xattrs, link) {
799 		size_t required_sz = 0;
800 
801 		rc = _spdk_blob_serialize_xattr(xattr,
802 						*buf, *remaining_sz,
803 						&required_sz, internal);
804 		if (rc < 0) {
805 			/* Need to add a new page to the chain */
806 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
807 							   &cur_page);
808 			if (rc < 0) {
809 				spdk_free(*pages);
810 				*pages = NULL;
811 				*page_count = 0;
812 				return rc;
813 			}
814 
815 			*buf = (uint8_t *)cur_page->descriptors;
816 			*remaining_sz = sizeof(cur_page->descriptors);
817 
818 			/* Try again */
819 			required_sz = 0;
820 			rc = _spdk_blob_serialize_xattr(xattr,
821 							*buf, *remaining_sz,
822 							&required_sz, internal);
823 
824 			if (rc < 0) {
825 				spdk_free(*pages);
826 				*pages = NULL;
827 				*page_count = 0;
828 				return rc;
829 			}
830 		}
831 
832 		*remaining_sz -= required_sz;
833 		*buf += required_sz;
834 	}
835 
836 	return 0;
837 }
838 
839 static int
840 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
841 		     uint32_t *page_count)
842 {
843 	struct spdk_blob_md_page		*cur_page;
844 	int					rc;
845 	uint8_t					*buf;
846 	size_t					remaining_sz;
847 
848 	assert(pages != NULL);
849 	assert(page_count != NULL);
850 	assert(blob != NULL);
851 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
852 
853 	*pages = NULL;
854 	*page_count = 0;
855 
856 	/* A blob always has at least 1 page, even if it has no descriptors */
857 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
858 	if (rc < 0) {
859 		return rc;
860 	}
861 
862 	buf = (uint8_t *)cur_page->descriptors;
863 	remaining_sz = sizeof(cur_page->descriptors);
864 
865 	/* Serialize flags */
866 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
867 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
868 
869 	/* Serialize xattrs */
870 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
871 					 pages, cur_page, page_count, &buf, &remaining_sz);
872 	if (rc < 0) {
873 		return rc;
874 	}
875 
876 	/* Serialize internal xattrs */
877 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
878 					 pages, cur_page, page_count, &buf, &remaining_sz);
879 	if (rc < 0) {
880 		return rc;
881 	}
882 
883 	/* Serialize extents */
884 	rc = _spdk_blob_serialize_extents(blob, pages, cur_page, page_count, &buf, &remaining_sz);
885 
886 	return rc;
887 }
888 
889 struct spdk_blob_load_ctx {
890 	struct spdk_blob		*blob;
891 
892 	struct spdk_blob_md_page	*pages;
893 	uint32_t			num_pages;
894 	spdk_bs_sequence_t	        *seq;
895 
896 	spdk_bs_sequence_cpl		cb_fn;
897 	void				*cb_arg;
898 };
899 
900 static uint32_t
901 _spdk_blob_md_page_calc_crc(void *page)
902 {
903 	uint32_t		crc;
904 
905 	crc = BLOB_CRC32C_INITIAL;
906 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
907 	crc ^= BLOB_CRC32C_INITIAL;
908 
909 	return crc;
910 
911 }
912 
913 static void
914 _spdk_blob_load_final(void *cb_arg, int bserrno)
915 {
916 	struct spdk_blob_load_ctx	*ctx = cb_arg;
917 	struct spdk_blob		*blob = ctx->blob;
918 
919 	_spdk_blob_mark_clean(blob);
920 
921 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
922 
923 	/* Free the memory */
924 	spdk_free(ctx->pages);
925 	free(ctx);
926 }
927 
928 static void
929 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
930 {
931 	struct spdk_blob_load_ctx	*ctx = cb_arg;
932 	struct spdk_blob		*blob = ctx->blob;
933 
934 	if (bserrno != 0) {
935 		goto error;
936 	}
937 
938 	blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
939 
940 	if (blob->back_bs_dev == NULL) {
941 		bserrno = -ENOMEM;
942 		goto error;
943 	}
944 
945 	_spdk_blob_load_final(ctx, bserrno);
946 	return;
947 
948 error:
949 	SPDK_ERRLOG("Snapshot fail\n");
950 	_spdk_blob_free(blob);
951 	ctx->cb_fn(ctx->seq, NULL, bserrno);
952 	spdk_free(ctx->pages);
953 	free(ctx);
954 }
955 
956 static void
957 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
958 {
959 	struct spdk_blob_load_ctx	*ctx = cb_arg;
960 	struct spdk_blob		*blob = ctx->blob;
961 	struct spdk_blob_md_page	*page;
962 	const void			*value;
963 	size_t				len;
964 	int				rc;
965 	uint32_t			crc;
966 
967 	if (bserrno) {
968 		SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno);
969 		_spdk_blob_free(blob);
970 		ctx->cb_fn(seq, NULL, bserrno);
971 		spdk_free(ctx->pages);
972 		free(ctx);
973 		return;
974 	}
975 
976 	page = &ctx->pages[ctx->num_pages - 1];
977 	crc = _spdk_blob_md_page_calc_crc(page);
978 	if (crc != page->crc) {
979 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
980 		_spdk_blob_free(blob);
981 		ctx->cb_fn(seq, NULL, -EINVAL);
982 		spdk_free(ctx->pages);
983 		free(ctx);
984 		return;
985 	}
986 
987 	if (page->next != SPDK_INVALID_MD_PAGE) {
988 		uint32_t next_page = page->next;
989 		uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page);
990 		uint64_t max_md_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + blob->bs->md_len);
991 
992 		if (next_lba >= max_md_lba) {
993 			assert(false);
994 		}
995 
996 		/* Read the next page */
997 		ctx->num_pages++;
998 		ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
999 					  sizeof(*page));
1000 		if (ctx->pages == NULL) {
1001 			ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM);
1002 			free(ctx);
1003 			return;
1004 		}
1005 
1006 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
1007 					  next_lba,
1008 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
1009 					  _spdk_blob_load_cpl, ctx);
1010 		return;
1011 	}
1012 
1013 	/* Parse the pages */
1014 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
1015 	if (rc) {
1016 		_spdk_blob_free(blob);
1017 		ctx->cb_fn(seq, NULL, rc);
1018 		spdk_free(ctx->pages);
1019 		free(ctx);
1020 		return;
1021 	}
1022 	ctx->seq = seq;
1023 
1024 
1025 	if (spdk_blob_is_thin_provisioned(blob)) {
1026 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
1027 		if (rc == 0) {
1028 			if (len != sizeof(spdk_blob_id)) {
1029 				_spdk_blob_free(blob);
1030 				ctx->cb_fn(seq, NULL, -EINVAL);
1031 				spdk_free(ctx->pages);
1032 				free(ctx);
1033 				return;
1034 			}
1035 			/* open snapshot blob and continue in the callback function */
1036 			blob->parent_id = *(spdk_blob_id *)value;
1037 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1038 					  _spdk_blob_load_snapshot_cpl, ctx);
1039 			return;
1040 		} else {
1041 			/* add zeroes_dev for thin provisioned blob */
1042 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1043 		}
1044 	} else {
1045 		/* standard blob */
1046 		blob->back_bs_dev = NULL;
1047 	}
1048 	_spdk_blob_load_final(ctx, bserrno);
1049 }
1050 
1051 /* Load a blob from disk given a blobid */
1052 static void
1053 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1054 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1055 {
1056 	struct spdk_blob_load_ctx *ctx;
1057 	struct spdk_blob_store *bs;
1058 	uint32_t page_num;
1059 	uint64_t lba;
1060 
1061 	_spdk_blob_verify_md_op(blob);
1062 
1063 	bs = blob->bs;
1064 
1065 	ctx = calloc(1, sizeof(*ctx));
1066 	if (!ctx) {
1067 		cb_fn(seq, cb_arg, -ENOMEM);
1068 		return;
1069 	}
1070 
1071 	ctx->blob = blob;
1072 	ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE);
1073 	if (!ctx->pages) {
1074 		free(ctx);
1075 		cb_fn(seq, cb_arg, -ENOMEM);
1076 		return;
1077 	}
1078 	ctx->num_pages = 1;
1079 	ctx->cb_fn = cb_fn;
1080 	ctx->cb_arg = cb_arg;
1081 
1082 	page_num = _spdk_bs_blobid_to_page(blob->id);
1083 	lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num);
1084 
1085 	blob->state = SPDK_BLOB_STATE_LOADING;
1086 
1087 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1088 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1089 				  _spdk_blob_load_cpl, ctx);
1090 }
1091 
1092 struct spdk_blob_persist_ctx {
1093 	struct spdk_blob		*blob;
1094 
1095 	struct spdk_bs_super_block	*super;
1096 
1097 	struct spdk_blob_md_page	*pages;
1098 
1099 	spdk_bs_sequence_t		*seq;
1100 	spdk_bs_sequence_cpl		cb_fn;
1101 	void				*cb_arg;
1102 };
1103 
1104 static void
1105 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1106 			uint32_t lba_count)
1107 {
1108 	if (ctx->blob->clear_method == BLOB_CLEAR_WITH_DEFAULT ||
1109 	    ctx->blob->clear_method == BLOB_CLEAR_WITH_UNMAP) {
1110 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1111 	} else if (ctx->blob->clear_method == BLOB_CLEAR_WITH_WRITE_ZEROES) {
1112 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1113 	}
1114 }
1115 
1116 static void
1117 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1118 {
1119 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1120 	struct spdk_blob		*blob = ctx->blob;
1121 
1122 	if (bserrno == 0) {
1123 		_spdk_blob_mark_clean(blob);
1124 	}
1125 
1126 	/* Call user callback */
1127 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1128 
1129 	/* Free the memory */
1130 	spdk_free(ctx->pages);
1131 	free(ctx);
1132 }
1133 
1134 static void
1135 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1136 {
1137 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1138 	struct spdk_blob		*blob = ctx->blob;
1139 	struct spdk_blob_store		*bs = blob->bs;
1140 	size_t				i;
1141 
1142 	/* Release all clusters that were truncated */
1143 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1144 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1145 
1146 		/* Nothing to release if it was not allocated */
1147 		if (blob->active.clusters[i] != 0) {
1148 			_spdk_bs_release_cluster(bs, cluster_num);
1149 		}
1150 	}
1151 
1152 	if (blob->active.num_clusters == 0) {
1153 		free(blob->active.clusters);
1154 		blob->active.clusters = NULL;
1155 		blob->active.cluster_array_size = 0;
1156 	} else if (blob->active.num_clusters != blob->active.cluster_array_size) {
1157 #ifndef __clang_analyzer__
1158 		void *tmp;
1159 
1160 		/* scan-build really can't figure reallocs, workaround it */
1161 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters);
1162 		assert(tmp != NULL);
1163 		blob->active.clusters = tmp;
1164 #endif
1165 		blob->active.cluster_array_size = blob->active.num_clusters;
1166 	}
1167 
1168 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1169 }
1170 
1171 static void
1172 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1173 {
1174 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1175 	struct spdk_blob		*blob = ctx->blob;
1176 	struct spdk_blob_store		*bs = blob->bs;
1177 	spdk_bs_batch_t			*batch;
1178 	size_t				i;
1179 	uint64_t			lba;
1180 	uint32_t			lba_count;
1181 
1182 	/* Clusters don't move around in blobs. The list shrinks or grows
1183 	 * at the end, but no changes ever occur in the middle of the list.
1184 	 */
1185 
1186 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx);
1187 
1188 	/* Clear all clusters that were truncated */
1189 	lba = 0;
1190 	lba_count = 0;
1191 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1192 		uint64_t next_lba = blob->active.clusters[i];
1193 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1194 
1195 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1196 			/* This cluster is contiguous with the previous one. */
1197 			lba_count += next_lba_count;
1198 			continue;
1199 		}
1200 
1201 		/* This cluster is not contiguous with the previous one. */
1202 
1203 		/* If a run of LBAs previously existing, clear them now */
1204 		if (lba_count > 0) {
1205 			spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1206 		}
1207 
1208 		/* Start building the next batch */
1209 		lba = next_lba;
1210 		if (next_lba > 0) {
1211 			lba_count = next_lba_count;
1212 		} else {
1213 			lba_count = 0;
1214 		}
1215 	}
1216 
1217 	/* If we ended with a contiguous set of LBAs, clear them now */
1218 	if (lba_count > 0) {
1219 		spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1220 	}
1221 
1222 	spdk_bs_batch_close(batch);
1223 }
1224 
1225 static void
1226 _spdk_blob_persist_zero_pages_cpl(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 	size_t				i;
1232 
1233 	/* This loop starts at 1 because the first page is special and handled
1234 	 * below. The pages (except the first) are never written in place,
1235 	 * so any pages in the clean list must be zeroed.
1236 	 */
1237 	for (i = 1; i < blob->clean.num_pages; i++) {
1238 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1239 	}
1240 
1241 	if (blob->active.num_pages == 0) {
1242 		uint32_t page_num;
1243 
1244 		page_num = _spdk_bs_blobid_to_page(blob->id);
1245 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1246 	}
1247 
1248 	/* Move on to clearing clusters */
1249 	_spdk_blob_persist_clear_clusters(seq, ctx, 0);
1250 }
1251 
1252 static void
1253 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1254 {
1255 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1256 	struct spdk_blob		*blob = ctx->blob;
1257 	struct spdk_blob_store		*bs = blob->bs;
1258 	uint64_t			lba;
1259 	uint32_t			lba_count;
1260 	spdk_bs_batch_t			*batch;
1261 	size_t				i;
1262 
1263 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1264 
1265 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1266 
1267 	/* This loop starts at 1 because the first page is special and handled
1268 	 * below. The pages (except the first) are never written in place,
1269 	 * so any pages in the clean list must be zeroed.
1270 	 */
1271 	for (i = 1; i < blob->clean.num_pages; i++) {
1272 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);
1273 
1274 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1275 	}
1276 
1277 	/* The first page will only be zeroed if this is a delete. */
1278 	if (blob->active.num_pages == 0) {
1279 		uint32_t page_num;
1280 
1281 		/* The first page in the metadata goes where the blobid indicates */
1282 		page_num = _spdk_bs_blobid_to_page(blob->id);
1283 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);
1284 
1285 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1286 	}
1287 
1288 	spdk_bs_batch_close(batch);
1289 }
1290 
1291 static void
1292 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1293 {
1294 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1295 	struct spdk_blob		*blob = ctx->blob;
1296 	struct spdk_blob_store		*bs = blob->bs;
1297 	uint64_t			lba;
1298 	uint32_t			lba_count;
1299 	struct spdk_blob_md_page	*page;
1300 
1301 	if (blob->active.num_pages == 0) {
1302 		/* Move on to the next step */
1303 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1304 		return;
1305 	}
1306 
1307 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1308 
1309 	page = &ctx->pages[0];
1310 	/* The first page in the metadata goes where the blobid indicates */
1311 	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));
1312 
1313 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1314 				   _spdk_blob_persist_zero_pages, ctx);
1315 }
1316 
1317 static void
1318 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1319 {
1320 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1321 	struct spdk_blob		*blob = ctx->blob;
1322 	struct spdk_blob_store		*bs = blob->bs;
1323 	uint64_t			lba;
1324 	uint32_t			lba_count;
1325 	struct spdk_blob_md_page	*page;
1326 	spdk_bs_batch_t			*batch;
1327 	size_t				i;
1328 
1329 	/* Clusters don't move around in blobs. The list shrinks or grows
1330 	 * at the end, but no changes ever occur in the middle of the list.
1331 	 */
1332 
1333 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1334 
1335 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1336 
1337 	/* This starts at 1. The root page is not written until
1338 	 * all of the others are finished
1339 	 */
1340 	for (i = 1; i < blob->active.num_pages; i++) {
1341 		page = &ctx->pages[i];
1342 		assert(page->sequence_num == i);
1343 
1344 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]);
1345 
1346 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1347 	}
1348 
1349 	spdk_bs_batch_close(batch);
1350 }
1351 
1352 static int
1353 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1354 {
1355 	uint64_t	i;
1356 	uint64_t	*tmp;
1357 	uint64_t	lfc; /* lowest free cluster */
1358 	uint64_t	num_clusters;
1359 	struct spdk_blob_store *bs;
1360 
1361 	bs = blob->bs;
1362 
1363 	_spdk_blob_verify_md_op(blob);
1364 
1365 	if (blob->active.num_clusters == sz) {
1366 		return 0;
1367 	}
1368 
1369 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1370 		/* If this blob was resized to be larger, then smaller, then
1371 		 * larger without syncing, then the cluster array already
1372 		 * contains spare assigned clusters we can use.
1373 		 */
1374 		num_clusters = spdk_min(blob->active.cluster_array_size,
1375 					sz);
1376 	} else {
1377 		num_clusters = blob->active.num_clusters;
1378 	}
1379 
1380 	/* Do two passes - one to verify that we can obtain enough clusters
1381 	 * and another to actually claim them.
1382 	 */
1383 
1384 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1385 		lfc = 0;
1386 		for (i = num_clusters; i < sz; i++) {
1387 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1388 			if (lfc == UINT32_MAX) {
1389 				/* No more free clusters. Cannot satisfy the request */
1390 				return -ENOSPC;
1391 			}
1392 			lfc++;
1393 		}
1394 	}
1395 
1396 	if (sz > num_clusters) {
1397 		/* Expand the cluster array if necessary.
1398 		 * We only shrink the array when persisting.
1399 		 */
1400 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz);
1401 		if (sz > 0 && tmp == NULL) {
1402 			return -ENOMEM;
1403 		}
1404 		memset(tmp + blob->active.cluster_array_size, 0,
1405 		       sizeof(uint64_t) * (sz - blob->active.cluster_array_size));
1406 		blob->active.clusters = tmp;
1407 		blob->active.cluster_array_size = sz;
1408 	}
1409 
1410 	blob->state = SPDK_BLOB_STATE_DIRTY;
1411 
1412 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1413 		lfc = 0;
1414 		for (i = num_clusters; i < sz; i++) {
1415 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1416 			lfc++;
1417 		}
1418 	}
1419 
1420 	blob->active.num_clusters = sz;
1421 
1422 	return 0;
1423 }
1424 
1425 static void
1426 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1427 {
1428 	spdk_bs_sequence_t *seq = ctx->seq;
1429 	struct spdk_blob *blob = ctx->blob;
1430 	struct spdk_blob_store *bs = blob->bs;
1431 	uint64_t i;
1432 	uint32_t page_num;
1433 	void *tmp;
1434 	int rc;
1435 
1436 	if (blob->active.num_pages == 0) {
1437 		/* This is the signal that the blob should be deleted.
1438 		 * Immediately jump to the clean up routine. */
1439 		assert(blob->clean.num_pages > 0);
1440 		blob->state = SPDK_BLOB_STATE_CLEAN;
1441 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1442 		return;
1443 
1444 	}
1445 
1446 	/* Generate the new metadata */
1447 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1448 	if (rc < 0) {
1449 		_spdk_blob_persist_complete(seq, ctx, rc);
1450 		return;
1451 	}
1452 
1453 	assert(blob->active.num_pages >= 1);
1454 
1455 	/* Resize the cache of page indices */
1456 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1457 	if (!tmp) {
1458 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1459 		return;
1460 	}
1461 	blob->active.pages = tmp;
1462 
1463 	/* Assign this metadata to pages. This requires two passes -
1464 	 * one to verify that there are enough pages and a second
1465 	 * to actually claim them. */
1466 	page_num = 0;
1467 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1468 	for (i = 1; i < blob->active.num_pages; i++) {
1469 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1470 		if (page_num == UINT32_MAX) {
1471 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1472 			return;
1473 		}
1474 		page_num++;
1475 	}
1476 
1477 	page_num = 0;
1478 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1479 	for (i = 1; i < blob->active.num_pages; i++) {
1480 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1481 		ctx->pages[i - 1].next = page_num;
1482 		/* Now that previous metadata page is complete, calculate the crc for it. */
1483 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1484 		blob->active.pages[i] = page_num;
1485 		spdk_bit_array_set(bs->used_md_pages, page_num);
1486 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1487 		page_num++;
1488 	}
1489 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1490 	/* Start writing the metadata from last page to first */
1491 	blob->state = SPDK_BLOB_STATE_CLEAN;
1492 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1493 }
1494 
1495 static void
1496 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1497 {
1498 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1499 
1500 	ctx->blob->bs->clean = 0;
1501 
1502 	spdk_free(ctx->super);
1503 
1504 	_spdk_blob_persist_start(ctx);
1505 }
1506 
1507 static void
1508 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1509 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1510 
1511 
1512 static void
1513 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1514 {
1515 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1516 
1517 	ctx->super->clean = 0;
1518 	if (ctx->super->size == 0) {
1519 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1520 	}
1521 
1522 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1523 }
1524 
1525 
1526 /* Write a blob to disk */
1527 static void
1528 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1529 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1530 {
1531 	struct spdk_blob_persist_ctx *ctx;
1532 
1533 	_spdk_blob_verify_md_op(blob);
1534 
1535 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1536 		cb_fn(seq, cb_arg, 0);
1537 		return;
1538 	}
1539 
1540 	ctx = calloc(1, sizeof(*ctx));
1541 	if (!ctx) {
1542 		cb_fn(seq, cb_arg, -ENOMEM);
1543 		return;
1544 	}
1545 	ctx->blob = blob;
1546 	ctx->seq = seq;
1547 	ctx->cb_fn = cb_fn;
1548 	ctx->cb_arg = cb_arg;
1549 
1550 	if (blob->bs->clean) {
1551 		ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
1552 					  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1553 		if (!ctx->super) {
1554 			cb_fn(seq, cb_arg, -ENOMEM);
1555 			free(ctx);
1556 			return;
1557 		}
1558 
1559 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1560 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1561 					  _spdk_blob_persist_dirty, ctx);
1562 	} else {
1563 		_spdk_blob_persist_start(ctx);
1564 	}
1565 }
1566 
1567 struct spdk_blob_copy_cluster_ctx {
1568 	struct spdk_blob *blob;
1569 	uint8_t *buf;
1570 	uint64_t page;
1571 	uint64_t new_cluster;
1572 	spdk_bs_sequence_t *seq;
1573 };
1574 
1575 static void
1576 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1577 {
1578 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1579 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1580 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1581 	spdk_bs_user_op_t *op;
1582 
1583 	TAILQ_INIT(&requests);
1584 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1585 
1586 	while (!TAILQ_EMPTY(&requests)) {
1587 		op = TAILQ_FIRST(&requests);
1588 		TAILQ_REMOVE(&requests, op, link);
1589 		if (bserrno == 0) {
1590 			spdk_bs_user_op_execute(op);
1591 		} else {
1592 			spdk_bs_user_op_abort(op);
1593 		}
1594 	}
1595 
1596 	spdk_free(ctx->buf);
1597 	free(ctx);
1598 }
1599 
1600 static void
1601 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1602 {
1603 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1604 
1605 	if (bserrno) {
1606 		if (bserrno == -EEXIST) {
1607 			/* The metadata insert failed because another thread
1608 			 * allocated the cluster first. Free our cluster
1609 			 * but continue without error. */
1610 			bserrno = 0;
1611 		}
1612 		_spdk_bs_release_cluster(ctx->blob->bs, ctx->new_cluster);
1613 	}
1614 
1615 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1616 }
1617 
1618 static void
1619 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1620 {
1621 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1622 	uint32_t cluster_number;
1623 
1624 	if (bserrno) {
1625 		/* The write failed, so jump to the final completion handler */
1626 		spdk_bs_sequence_finish(seq, bserrno);
1627 		return;
1628 	}
1629 
1630 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1631 
1632 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1633 					       _spdk_blob_insert_cluster_cpl, ctx);
1634 }
1635 
1636 static void
1637 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1638 {
1639 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1640 
1641 	if (bserrno != 0) {
1642 		/* The read failed, so jump to the final completion handler */
1643 		spdk_bs_sequence_finish(seq, bserrno);
1644 		return;
1645 	}
1646 
1647 	/* Write whole cluster */
1648 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1649 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1650 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1651 				   _spdk_blob_write_copy_cpl, ctx);
1652 }
1653 
1654 static void
1655 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1656 				   struct spdk_io_channel *_ch,
1657 				   uint64_t io_unit, spdk_bs_user_op_t *op)
1658 {
1659 	struct spdk_bs_cpl cpl;
1660 	struct spdk_bs_channel *ch;
1661 	struct spdk_blob_copy_cluster_ctx *ctx;
1662 	uint32_t cluster_start_page;
1663 	uint32_t cluster_number;
1664 	int rc;
1665 
1666 	ch = spdk_io_channel_get_ctx(_ch);
1667 
1668 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1669 		/* There are already operations pending. Queue this user op
1670 		 * and return because it will be re-executed when the outstanding
1671 		 * cluster allocation completes. */
1672 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1673 		return;
1674 	}
1675 
1676 	/* Round the io_unit offset down to the first page in the cluster */
1677 	cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit);
1678 
1679 	/* Calculate which index in the metadata cluster array the corresponding
1680 	 * cluster is supposed to be at. */
1681 	cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit);
1682 
1683 	ctx = calloc(1, sizeof(*ctx));
1684 	if (!ctx) {
1685 		spdk_bs_user_op_abort(op);
1686 		return;
1687 	}
1688 
1689 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1690 
1691 	ctx->blob = blob;
1692 	ctx->page = cluster_start_page;
1693 
1694 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1695 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
1696 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1697 		if (!ctx->buf) {
1698 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1699 				    blob->bs->cluster_sz);
1700 			free(ctx);
1701 			spdk_bs_user_op_abort(op);
1702 			return;
1703 		}
1704 	}
1705 
1706 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1707 	if (rc != 0) {
1708 		spdk_free(ctx->buf);
1709 		free(ctx);
1710 		spdk_bs_user_op_abort(op);
1711 		return;
1712 	}
1713 
1714 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1715 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1716 	cpl.u.blob_basic.cb_arg = ctx;
1717 
1718 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1719 	if (!ctx->seq) {
1720 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1721 		spdk_free(ctx->buf);
1722 		free(ctx);
1723 		spdk_bs_user_op_abort(op);
1724 		return;
1725 	}
1726 
1727 	/* Queue the user op to block other incoming operations */
1728 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1729 
1730 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1731 		/* Read cluster from backing device */
1732 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1733 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1734 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1735 					     _spdk_blob_write_copy, ctx);
1736 	} else {
1737 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1738 						       _spdk_blob_insert_cluster_cpl, ctx);
1739 	}
1740 }
1741 
1742 static void
1743 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
1744 				       uint64_t *lba,	uint32_t *lba_count)
1745 {
1746 	*lba_count = length;
1747 
1748 	if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) {
1749 		assert(blob->back_bs_dev != NULL);
1750 		*lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit);
1751 		*lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count);
1752 	} else {
1753 		*lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit);
1754 	}
1755 }
1756 
1757 struct op_split_ctx {
1758 	struct spdk_blob *blob;
1759 	struct spdk_io_channel *channel;
1760 	uint64_t io_unit_offset;
1761 	uint64_t io_units_remaining;
1762 	void *curr_payload;
1763 	enum spdk_blob_op_type op_type;
1764 	spdk_bs_sequence_t *seq;
1765 };
1766 
1767 static void
1768 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1769 {
1770 	struct op_split_ctx	*ctx = cb_arg;
1771 	struct spdk_blob	*blob = ctx->blob;
1772 	struct spdk_io_channel	*ch = ctx->channel;
1773 	enum spdk_blob_op_type	op_type = ctx->op_type;
1774 	uint8_t			*buf = ctx->curr_payload;
1775 	uint64_t		offset = ctx->io_unit_offset;
1776 	uint64_t		length = ctx->io_units_remaining;
1777 	uint64_t		op_length;
1778 
1779 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
1780 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1781 		free(ctx);
1782 		return;
1783 	}
1784 
1785 	op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob,
1786 			     offset));
1787 
1788 	/* Update length and payload for next operation */
1789 	ctx->io_units_remaining -= op_length;
1790 	ctx->io_unit_offset += op_length;
1791 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1792 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
1793 	}
1794 
1795 	switch (op_type) {
1796 	case SPDK_BLOB_READ:
1797 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1798 				  _spdk_blob_request_submit_op_split_next, ctx);
1799 		break;
1800 	case SPDK_BLOB_WRITE:
1801 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1802 				   _spdk_blob_request_submit_op_split_next, ctx);
1803 		break;
1804 	case SPDK_BLOB_UNMAP:
1805 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1806 				   _spdk_blob_request_submit_op_split_next, ctx);
1807 		break;
1808 	case SPDK_BLOB_WRITE_ZEROES:
1809 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1810 					  _spdk_blob_request_submit_op_split_next, ctx);
1811 		break;
1812 	case SPDK_BLOB_READV:
1813 	case SPDK_BLOB_WRITEV:
1814 		SPDK_ERRLOG("readv/write not valid\n");
1815 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1816 		free(ctx);
1817 		break;
1818 	}
1819 }
1820 
1821 static void
1822 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1823 				   void *payload, uint64_t offset, uint64_t length,
1824 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1825 {
1826 	struct op_split_ctx *ctx;
1827 	spdk_bs_sequence_t *seq;
1828 	struct spdk_bs_cpl cpl;
1829 
1830 	assert(blob != NULL);
1831 
1832 	ctx = calloc(1, sizeof(struct op_split_ctx));
1833 	if (ctx == NULL) {
1834 		cb_fn(cb_arg, -ENOMEM);
1835 		return;
1836 	}
1837 
1838 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1839 	cpl.u.blob_basic.cb_fn = cb_fn;
1840 	cpl.u.blob_basic.cb_arg = cb_arg;
1841 
1842 	seq = spdk_bs_sequence_start(ch, &cpl);
1843 	if (!seq) {
1844 		free(ctx);
1845 		cb_fn(cb_arg, -ENOMEM);
1846 		return;
1847 	}
1848 
1849 	ctx->blob = blob;
1850 	ctx->channel = ch;
1851 	ctx->curr_payload = payload;
1852 	ctx->io_unit_offset = offset;
1853 	ctx->io_units_remaining = length;
1854 	ctx->op_type = op_type;
1855 	ctx->seq = seq;
1856 
1857 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1858 }
1859 
1860 static void
1861 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1862 				    void *payload, uint64_t offset, uint64_t length,
1863 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1864 {
1865 	struct spdk_bs_cpl cpl;
1866 	uint64_t lba;
1867 	uint32_t lba_count;
1868 
1869 	assert(blob != NULL);
1870 
1871 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1872 	cpl.u.blob_basic.cb_fn = cb_fn;
1873 	cpl.u.blob_basic.cb_arg = cb_arg;
1874 
1875 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1876 
1877 	if (blob->frozen_refcnt) {
1878 		/* This blob I/O is frozen */
1879 		spdk_bs_user_op_t *op;
1880 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1881 
1882 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1883 		if (!op) {
1884 			cb_fn(cb_arg, -ENOMEM);
1885 			return;
1886 		}
1887 
1888 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1889 
1890 		return;
1891 	}
1892 
1893 	switch (op_type) {
1894 	case SPDK_BLOB_READ: {
1895 		spdk_bs_batch_t *batch;
1896 
1897 		batch = spdk_bs_batch_open(_ch, &cpl);
1898 		if (!batch) {
1899 			cb_fn(cb_arg, -ENOMEM);
1900 			return;
1901 		}
1902 
1903 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1904 			/* Read from the blob */
1905 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1906 		} else {
1907 			/* Read from the backing block device */
1908 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1909 		}
1910 
1911 		spdk_bs_batch_close(batch);
1912 		break;
1913 	}
1914 	case SPDK_BLOB_WRITE:
1915 	case SPDK_BLOB_WRITE_ZEROES: {
1916 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1917 			/* Write to the blob */
1918 			spdk_bs_batch_t *batch;
1919 
1920 			if (lba_count == 0) {
1921 				cb_fn(cb_arg, 0);
1922 				return;
1923 			}
1924 
1925 			batch = spdk_bs_batch_open(_ch, &cpl);
1926 			if (!batch) {
1927 				cb_fn(cb_arg, -ENOMEM);
1928 				return;
1929 			}
1930 
1931 			if (op_type == SPDK_BLOB_WRITE) {
1932 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1933 			} else {
1934 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1935 			}
1936 
1937 			spdk_bs_batch_close(batch);
1938 		} else {
1939 			/* Queue this operation and allocate the cluster */
1940 			spdk_bs_user_op_t *op;
1941 
1942 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1943 			if (!op) {
1944 				cb_fn(cb_arg, -ENOMEM);
1945 				return;
1946 			}
1947 
1948 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1949 		}
1950 		break;
1951 	}
1952 	case SPDK_BLOB_UNMAP: {
1953 		spdk_bs_batch_t *batch;
1954 
1955 		batch = spdk_bs_batch_open(_ch, &cpl);
1956 		if (!batch) {
1957 			cb_fn(cb_arg, -ENOMEM);
1958 			return;
1959 		}
1960 
1961 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1962 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1963 		}
1964 
1965 		spdk_bs_batch_close(batch);
1966 		break;
1967 	}
1968 	case SPDK_BLOB_READV:
1969 	case SPDK_BLOB_WRITEV:
1970 		SPDK_ERRLOG("readv/write not valid\n");
1971 		cb_fn(cb_arg, -EINVAL);
1972 		break;
1973 	}
1974 }
1975 
1976 static void
1977 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1978 			     void *payload, uint64_t offset, uint64_t length,
1979 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1980 {
1981 	assert(blob != NULL);
1982 
1983 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1984 		cb_fn(cb_arg, -EPERM);
1985 		return;
1986 	}
1987 
1988 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
1989 		cb_fn(cb_arg, -EINVAL);
1990 		return;
1991 	}
1992 	if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) {
1993 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1994 						    cb_fn, cb_arg, op_type);
1995 	} else {
1996 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1997 						   cb_fn, cb_arg, op_type);
1998 	}
1999 }
2000 
2001 struct rw_iov_ctx {
2002 	struct spdk_blob *blob;
2003 	struct spdk_io_channel *channel;
2004 	spdk_blob_op_complete cb_fn;
2005 	void *cb_arg;
2006 	bool read;
2007 	int iovcnt;
2008 	struct iovec *orig_iov;
2009 	uint64_t io_unit_offset;
2010 	uint64_t io_units_remaining;
2011 	uint64_t io_units_done;
2012 	struct iovec iov[0];
2013 };
2014 
2015 static void
2016 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2017 {
2018 	assert(cb_arg == NULL);
2019 	spdk_bs_sequence_finish(seq, bserrno);
2020 }
2021 
2022 static void
2023 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
2024 {
2025 	struct rw_iov_ctx *ctx = cb_arg;
2026 	struct spdk_blob *blob = ctx->blob;
2027 	struct iovec *iov, *orig_iov;
2028 	int iovcnt;
2029 	size_t orig_iovoff;
2030 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2031 	uint64_t byte_count;
2032 
2033 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2034 		ctx->cb_fn(ctx->cb_arg, bserrno);
2035 		free(ctx);
2036 		return;
2037 	}
2038 
2039 	io_unit_offset = ctx->io_unit_offset;
2040 	io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2041 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2042 	/*
2043 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2044 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2045 	 *  point to the current position in the I/O sequence.
2046 	 */
2047 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2048 	orig_iov = &ctx->orig_iov[0];
2049 	orig_iovoff = 0;
2050 	while (byte_count > 0) {
2051 		if (byte_count >= orig_iov->iov_len) {
2052 			byte_count -= orig_iov->iov_len;
2053 			orig_iov++;
2054 		} else {
2055 			orig_iovoff = byte_count;
2056 			byte_count = 0;
2057 		}
2058 	}
2059 
2060 	/*
2061 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2062 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2063 	 */
2064 	byte_count = io_units_count * blob->bs->io_unit_size;
2065 	iov = &ctx->iov[0];
2066 	iovcnt = 0;
2067 	while (byte_count > 0) {
2068 		assert(iovcnt < ctx->iovcnt);
2069 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2070 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2071 		byte_count -= iov->iov_len;
2072 		orig_iovoff = 0;
2073 		orig_iov++;
2074 		iov++;
2075 		iovcnt++;
2076 	}
2077 
2078 	ctx->io_unit_offset += io_units_count;
2079 	ctx->io_units_remaining -= io_units_count;
2080 	ctx->io_units_done += io_units_count;
2081 	iov = &ctx->iov[0];
2082 
2083 	if (ctx->read) {
2084 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2085 				   io_units_count, _spdk_rw_iov_split_next, ctx);
2086 	} else {
2087 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2088 				    io_units_count, _spdk_rw_iov_split_next, ctx);
2089 	}
2090 }
2091 
2092 static void
2093 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2094 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2095 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2096 {
2097 	struct spdk_bs_cpl	cpl;
2098 
2099 	assert(blob != NULL);
2100 
2101 	if (!read && blob->data_ro) {
2102 		cb_fn(cb_arg, -EPERM);
2103 		return;
2104 	}
2105 
2106 	if (length == 0) {
2107 		cb_fn(cb_arg, 0);
2108 		return;
2109 	}
2110 
2111 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2112 		cb_fn(cb_arg, -EINVAL);
2113 		return;
2114 	}
2115 
2116 	/*
2117 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2118 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2119 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2120 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2121 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2122 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2123 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2124 	 *
2125 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2126 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2127 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2128 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2129 	 */
2130 	if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) {
2131 		uint32_t lba_count;
2132 		uint64_t lba;
2133 
2134 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2135 		cpl.u.blob_basic.cb_fn = cb_fn;
2136 		cpl.u.blob_basic.cb_arg = cb_arg;
2137 
2138 		if (blob->frozen_refcnt) {
2139 			/* This blob I/O is frozen */
2140 			enum spdk_blob_op_type op_type;
2141 			spdk_bs_user_op_t *op;
2142 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2143 
2144 			op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV;
2145 			op = spdk_bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length);
2146 			if (!op) {
2147 				cb_fn(cb_arg, -ENOMEM);
2148 				return;
2149 			}
2150 
2151 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2152 
2153 			return;
2154 		}
2155 
2156 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2157 
2158 		if (read) {
2159 			spdk_bs_sequence_t *seq;
2160 
2161 			seq = spdk_bs_sequence_start(_channel, &cpl);
2162 			if (!seq) {
2163 				cb_fn(cb_arg, -ENOMEM);
2164 				return;
2165 			}
2166 
2167 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2168 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2169 			} else {
2170 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2171 							      _spdk_rw_iov_done, NULL);
2172 			}
2173 		} else {
2174 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2175 				spdk_bs_sequence_t *seq;
2176 
2177 				seq = spdk_bs_sequence_start(_channel, &cpl);
2178 				if (!seq) {
2179 					cb_fn(cb_arg, -ENOMEM);
2180 					return;
2181 				}
2182 
2183 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2184 			} else {
2185 				/* Queue this operation and allocate the cluster */
2186 				spdk_bs_user_op_t *op;
2187 
2188 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2189 							   length);
2190 				if (!op) {
2191 					cb_fn(cb_arg, -ENOMEM);
2192 					return;
2193 				}
2194 
2195 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2196 			}
2197 		}
2198 	} else {
2199 		struct rw_iov_ctx *ctx;
2200 
2201 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2202 		if (ctx == NULL) {
2203 			cb_fn(cb_arg, -ENOMEM);
2204 			return;
2205 		}
2206 
2207 		ctx->blob = blob;
2208 		ctx->channel = _channel;
2209 		ctx->cb_fn = cb_fn;
2210 		ctx->cb_arg = cb_arg;
2211 		ctx->read = read;
2212 		ctx->orig_iov = iov;
2213 		ctx->iovcnt = iovcnt;
2214 		ctx->io_unit_offset = offset;
2215 		ctx->io_units_remaining = length;
2216 		ctx->io_units_done = 0;
2217 
2218 		_spdk_rw_iov_split_next(ctx, 0);
2219 	}
2220 }
2221 
2222 static struct spdk_blob *
2223 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2224 {
2225 	struct spdk_blob *blob;
2226 
2227 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2228 		if (blob->id == blobid) {
2229 			return blob;
2230 		}
2231 	}
2232 
2233 	return NULL;
2234 }
2235 
2236 static void
2237 _spdk_blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
2238 		struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
2239 {
2240 	assert(blob != NULL);
2241 	*snapshot_entry = NULL;
2242 	*clone_entry = NULL;
2243 
2244 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
2245 		return;
2246 	}
2247 
2248 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
2249 		if ((*snapshot_entry)->id == blob->parent_id) {
2250 			break;
2251 		}
2252 	}
2253 
2254 	if (*snapshot_entry != NULL) {
2255 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
2256 			if ((*clone_entry)->id == blob->id) {
2257 				break;
2258 			}
2259 		}
2260 
2261 		assert(clone_entry != NULL);
2262 	}
2263 }
2264 
2265 static int
2266 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2267 {
2268 	struct spdk_blob_store		*bs = io_device;
2269 	struct spdk_bs_channel		*channel = ctx_buf;
2270 	struct spdk_bs_dev		*dev;
2271 	uint32_t			max_ops = bs->max_channel_ops;
2272 	uint32_t			i;
2273 
2274 	dev = bs->dev;
2275 
2276 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2277 	if (!channel->req_mem) {
2278 		return -1;
2279 	}
2280 
2281 	TAILQ_INIT(&channel->reqs);
2282 
2283 	for (i = 0; i < max_ops; i++) {
2284 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2285 	}
2286 
2287 	channel->bs = bs;
2288 	channel->dev = dev;
2289 	channel->dev_channel = dev->create_channel(dev);
2290 
2291 	if (!channel->dev_channel) {
2292 		SPDK_ERRLOG("Failed to create device channel.\n");
2293 		free(channel->req_mem);
2294 		return -1;
2295 	}
2296 
2297 	TAILQ_INIT(&channel->need_cluster_alloc);
2298 	TAILQ_INIT(&channel->queued_io);
2299 
2300 	return 0;
2301 }
2302 
2303 static void
2304 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2305 {
2306 	struct spdk_bs_channel *channel = ctx_buf;
2307 	spdk_bs_user_op_t *op;
2308 
2309 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2310 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2311 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2312 		spdk_bs_user_op_abort(op);
2313 	}
2314 
2315 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2316 		op = TAILQ_FIRST(&channel->queued_io);
2317 		TAILQ_REMOVE(&channel->queued_io, op, link);
2318 		spdk_bs_user_op_abort(op);
2319 	}
2320 
2321 	free(channel->req_mem);
2322 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2323 }
2324 
2325 static void
2326 _spdk_bs_dev_destroy(void *io_device)
2327 {
2328 	struct spdk_blob_store *bs = io_device;
2329 	struct spdk_blob	*blob, *blob_tmp;
2330 
2331 	bs->dev->destroy(bs->dev);
2332 
2333 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2334 		TAILQ_REMOVE(&bs->blobs, blob, link);
2335 		_spdk_blob_free(blob);
2336 	}
2337 
2338 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2339 
2340 	spdk_bit_array_free(&bs->used_blobids);
2341 	spdk_bit_array_free(&bs->used_md_pages);
2342 	spdk_bit_array_free(&bs->used_clusters);
2343 	/*
2344 	 * If this function is called for any reason except a successful unload,
2345 	 * the unload_cpl type will be NONE and this will be a nop.
2346 	 */
2347 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2348 
2349 	free(bs);
2350 }
2351 
2352 static int
2353 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2354 {
2355 	spdk_blob_id snapshot_id;
2356 	struct spdk_blob_list *snapshot_entry = NULL;
2357 	struct spdk_blob_list *clone_entry = NULL;
2358 
2359 	assert(blob != NULL);
2360 
2361 	snapshot_id = blob->parent_id;
2362 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2363 		return 0;
2364 	}
2365 
2366 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, snapshot_id);
2367 	if (snapshot_entry == NULL) {
2368 		/* Snapshot not found */
2369 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2370 		if (snapshot_entry == NULL) {
2371 			return -ENOMEM;
2372 		}
2373 		snapshot_entry->id = snapshot_id;
2374 		TAILQ_INIT(&snapshot_entry->clones);
2375 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2376 	} else {
2377 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2378 			if (clone_entry->id == blob->id) {
2379 				break;
2380 			}
2381 		}
2382 	}
2383 
2384 	if (clone_entry == NULL) {
2385 		/* Clone not found */
2386 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2387 		if (clone_entry == NULL) {
2388 			return -ENOMEM;
2389 		}
2390 		clone_entry->id = blob->id;
2391 		TAILQ_INIT(&clone_entry->clones);
2392 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2393 		snapshot_entry->clone_count++;
2394 	}
2395 
2396 	return 0;
2397 }
2398 
2399 static void
2400 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2401 {
2402 	struct spdk_blob_list *snapshot_entry = NULL;
2403 	struct spdk_blob_list *clone_entry = NULL;
2404 
2405 	_spdk_blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
2406 
2407 	if (snapshot_entry == NULL) {
2408 		return;
2409 	}
2410 
2411 	blob->parent_id = SPDK_BLOBID_INVALID;
2412 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2413 	free(clone_entry);
2414 
2415 	snapshot_entry->clone_count--;
2416 }
2417 
2418 static int
2419 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2420 {
2421 	struct spdk_blob_list *snapshot_entry;
2422 	struct spdk_blob_list *snapshot_entry_tmp;
2423 	struct spdk_blob_list *clone_entry;
2424 	struct spdk_blob_list *clone_entry_tmp;
2425 
2426 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2427 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2428 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2429 			free(clone_entry);
2430 		}
2431 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2432 		free(snapshot_entry);
2433 	}
2434 
2435 	return 0;
2436 }
2437 
2438 static void
2439 _spdk_bs_free(struct spdk_blob_store *bs)
2440 {
2441 	_spdk_bs_blob_list_free(bs);
2442 
2443 	spdk_bs_unregister_md_thread(bs);
2444 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2445 }
2446 
2447 void
2448 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2449 {
2450 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2451 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2452 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2453 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2454 	opts->clear_method = BS_CLEAR_WITH_UNMAP;
2455 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2456 	opts->iter_cb_fn = NULL;
2457 	opts->iter_cb_arg = NULL;
2458 }
2459 
2460 static int
2461 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2462 {
2463 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2464 	    opts->max_channel_ops == 0) {
2465 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2466 		return -1;
2467 	}
2468 
2469 	return 0;
2470 }
2471 
2472 static int
2473 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2474 {
2475 	struct spdk_blob_store	*bs;
2476 	uint64_t dev_size;
2477 	int rc;
2478 
2479 	dev_size = dev->blocklen * dev->blockcnt;
2480 	if (dev_size < opts->cluster_sz) {
2481 		/* Device size cannot be smaller than cluster size of blobstore */
2482 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2483 			     dev_size, opts->cluster_sz);
2484 		return -ENOSPC;
2485 	}
2486 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2487 		/* Cluster size cannot be smaller than page size */
2488 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2489 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2490 		return -EINVAL;
2491 	}
2492 	bs = calloc(1, sizeof(struct spdk_blob_store));
2493 	if (!bs) {
2494 		return -ENOMEM;
2495 	}
2496 
2497 	TAILQ_INIT(&bs->blobs);
2498 	TAILQ_INIT(&bs->snapshots);
2499 	bs->dev = dev;
2500 	bs->md_thread = spdk_get_thread();
2501 	assert(bs->md_thread != NULL);
2502 
2503 	/*
2504 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2505 	 *  even multiple of the cluster size.
2506 	 */
2507 	bs->cluster_sz = opts->cluster_sz;
2508 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2509 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2510 	bs->num_free_clusters = bs->total_clusters;
2511 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2512 	bs->io_unit_size = dev->blocklen;
2513 	if (bs->used_clusters == NULL) {
2514 		free(bs);
2515 		return -ENOMEM;
2516 	}
2517 
2518 	bs->max_channel_ops = opts->max_channel_ops;
2519 	bs->super_blob = SPDK_BLOBID_INVALID;
2520 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2521 
2522 	/* The metadata is assumed to be at least 1 page */
2523 	bs->used_md_pages = spdk_bit_array_create(1);
2524 	bs->used_blobids = spdk_bit_array_create(0);
2525 
2526 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2527 
2528 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2529 				sizeof(struct spdk_bs_channel), "blobstore");
2530 	rc = spdk_bs_register_md_thread(bs);
2531 	if (rc == -1) {
2532 		spdk_io_device_unregister(bs, NULL);
2533 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2534 		spdk_bit_array_free(&bs->used_blobids);
2535 		spdk_bit_array_free(&bs->used_md_pages);
2536 		spdk_bit_array_free(&bs->used_clusters);
2537 		free(bs);
2538 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2539 		return -ENOMEM;
2540 	}
2541 
2542 	*_bs = bs;
2543 	return 0;
2544 }
2545 
2546 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2547 
2548 struct spdk_bs_load_ctx {
2549 	struct spdk_blob_store		*bs;
2550 	struct spdk_bs_super_block	*super;
2551 
2552 	struct spdk_bs_md_mask		*mask;
2553 	bool				in_page_chain;
2554 	uint32_t			page_index;
2555 	uint32_t			cur_page;
2556 	struct spdk_blob_md_page	*page;
2557 
2558 	spdk_bs_sequence_t			*seq;
2559 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2560 	void					*iter_cb_arg;
2561 	struct spdk_blob			*blob;
2562 	spdk_blob_id				blobid;
2563 };
2564 
2565 static void
2566 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2567 {
2568 	assert(bserrno != 0);
2569 
2570 	spdk_free(ctx->super);
2571 	spdk_bs_sequence_finish(seq, bserrno);
2572 	_spdk_bs_free(ctx->bs);
2573 	free(ctx);
2574 }
2575 
2576 static void
2577 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2578 {
2579 	uint32_t i = 0;
2580 
2581 	while (true) {
2582 		i = spdk_bit_array_find_first_set(array, i);
2583 		if (i >= mask->length) {
2584 			break;
2585 		}
2586 		mask->mask[i / 8] |= 1U << (i % 8);
2587 		i++;
2588 	}
2589 }
2590 
2591 static int
2592 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2593 {
2594 	struct spdk_bit_array *array;
2595 	uint32_t i;
2596 
2597 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2598 		return -ENOMEM;
2599 	}
2600 
2601 	array = *array_ptr;
2602 	for (i = 0; i < mask->length; i++) {
2603 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2604 			spdk_bit_array_set(array, i);
2605 		}
2606 	}
2607 
2608 	return 0;
2609 }
2610 
2611 static void
2612 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2613 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2614 {
2615 	/* Update the values in the super block */
2616 	super->super_blob = bs->super_blob;
2617 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2618 	super->crc = _spdk_blob_md_page_calc_crc(super);
2619 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2620 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2621 				   cb_fn, cb_arg);
2622 }
2623 
2624 static void
2625 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2626 {
2627 	struct spdk_bs_load_ctx	*ctx = arg;
2628 	uint64_t	mask_size, lba, lba_count;
2629 
2630 	/* Write out the used clusters mask */
2631 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2632 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2633 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2634 	if (!ctx->mask) {
2635 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2636 		return;
2637 	}
2638 
2639 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2640 	ctx->mask->length = ctx->bs->total_clusters;
2641 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2642 
2643 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2644 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2645 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2646 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2647 }
2648 
2649 static void
2650 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2651 {
2652 	struct spdk_bs_load_ctx	*ctx = arg;
2653 	uint64_t	mask_size, lba, lba_count;
2654 
2655 	if (seq->bserrno) {
2656 		_spdk_bs_load_ctx_fail(seq, ctx, seq->bserrno);
2657 		return;
2658 	}
2659 
2660 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2661 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2662 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2663 	if (!ctx->mask) {
2664 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2665 		return;
2666 	}
2667 
2668 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2669 	ctx->mask->length = ctx->super->md_len;
2670 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2671 
2672 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2673 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2674 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2675 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2676 }
2677 
2678 static void
2679 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2680 {
2681 	struct spdk_bs_load_ctx	*ctx = arg;
2682 	uint64_t	mask_size, lba, lba_count;
2683 
2684 	if (ctx->super->used_blobid_mask_len == 0) {
2685 		/*
2686 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2687 		 *  written to disk.
2688 		 */
2689 		cb_fn(seq, arg, 0);
2690 		return;
2691 	}
2692 
2693 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2694 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2695 				 SPDK_MALLOC_DMA);
2696 	if (!ctx->mask) {
2697 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2698 		return;
2699 	}
2700 
2701 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2702 	ctx->mask->length = ctx->super->md_len;
2703 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2704 
2705 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2706 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2707 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2708 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2709 }
2710 
2711 static void
2712 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
2713 {
2714 	_spdk_blob_verify_md_op(blob);
2715 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
2716 	blob->state = SPDK_BLOB_STATE_DIRTY;
2717 }
2718 
2719 static void _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno);
2720 
2721 static void
2722 _spdk_bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno)
2723 {
2724 	struct spdk_bs_load_ctx *ctx = cb_arg;
2725 	spdk_blob_id id;
2726 	int64_t page_num;
2727 
2728 	/* Iterate to next blob (we can't use spdk_bs_iter_next function as our
2729 	 * last blob has been removed */
2730 	page_num = _spdk_bs_blobid_to_page(ctx->blobid);
2731 	page_num++;
2732 	page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num);
2733 	if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) {
2734 		_spdk_bs_load_iter(ctx, NULL, -ENOENT);
2735 		return;
2736 	}
2737 
2738 	id = _spdk_bs_page_to_blobid(page_num);
2739 
2740 	spdk_bs_open_blob(ctx->bs, id, _spdk_bs_load_iter, ctx);
2741 }
2742 
2743 static void
2744 _spdk_bs_delete_corrupted_close_cb(void *cb_arg, int bserrno)
2745 {
2746 	struct spdk_bs_load_ctx *ctx = cb_arg;
2747 
2748 	if (bserrno != 0) {
2749 		SPDK_ERRLOG("Failed to close corrupted blob\n");
2750 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2751 		return;
2752 	}
2753 
2754 	spdk_bs_delete_blob(ctx->bs, ctx->blobid, _spdk_bs_delete_corrupted_blob_cpl, ctx);
2755 }
2756 
2757 static void
2758 _spdk_bs_delete_corrupted_blob(void *cb_arg, int bserrno)
2759 {
2760 	struct spdk_bs_load_ctx *ctx = cb_arg;
2761 	uint64_t i;
2762 
2763 	if (bserrno != 0) {
2764 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
2765 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2766 		return;
2767 	}
2768 
2769 	/* Snapshot and clone have the same copy of cluster map at this point.
2770 	 * Let's clear cluster map for snpashot now so that it won't be cleared
2771 	 * for clone later when we remove snapshot. Also set thin provision to
2772 	 * pass data corruption check */
2773 	for (i = 0; i < ctx->blob->active.num_clusters; i++) {
2774 		ctx->blob->active.clusters[i] = 0;
2775 	}
2776 
2777 	ctx->blob->md_ro = false;
2778 
2779 	_spdk_blob_set_thin_provision(ctx->blob);
2780 
2781 	ctx->blobid = ctx->blob->id;
2782 
2783 	spdk_blob_close(ctx->blob, _spdk_bs_delete_corrupted_close_cb, ctx);
2784 }
2785 
2786 static void
2787 _spdk_bs_update_corrupted_blob(void *cb_arg, int bserrno)
2788 {
2789 	struct spdk_bs_load_ctx *ctx = cb_arg;
2790 
2791 	if (bserrno != 0) {
2792 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
2793 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2794 		return;
2795 	}
2796 
2797 	ctx->blob->md_ro = false;
2798 	_spdk_blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true);
2799 	_spdk_blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true);
2800 	spdk_blob_set_read_only(ctx->blob);
2801 
2802 	if (ctx->iter_cb_fn) {
2803 		ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0);
2804 	}
2805 	_spdk_bs_blob_list_add(ctx->blob);
2806 
2807 	spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2808 }
2809 
2810 static void
2811 _spdk_bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno)
2812 {
2813 	struct spdk_bs_load_ctx *ctx = cb_arg;
2814 
2815 	if (bserrno != 0) {
2816 		SPDK_ERRLOG("Failed to open clone of a corrupted blob\n");
2817 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2818 		return;
2819 	}
2820 
2821 	if (blob->parent_id == ctx->blob->id) {
2822 		/* Power failure occured before updating clone (snapshot delete case)
2823 		 * or after updating clone (creating snapshot case) - keep snapshot */
2824 		spdk_blob_close(blob, _spdk_bs_update_corrupted_blob, ctx);
2825 	} else {
2826 		/* Power failure occured after updating clone (snapshot delete case)
2827 		 * or before updating clone (creating snapshot case) - remove snapshot */
2828 		spdk_blob_close(blob, _spdk_bs_delete_corrupted_blob, ctx);
2829 	}
2830 }
2831 
2832 static void
2833 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2834 {
2835 	struct spdk_bs_load_ctx *ctx = arg;
2836 	const void *value;
2837 	size_t len;
2838 	int rc = 0;
2839 
2840 	if (bserrno == 0) {
2841 		/* Examine blob if it is corrupted after power failure. Fix
2842 		 * the ones that can be fixed and remove any other corrupted
2843 		 * ones. If it is not corrupted just process it */
2844 		rc = _spdk_blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true);
2845 		if (rc != 0) {
2846 			rc = _spdk_blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true);
2847 			if (rc != 0) {
2848 				/* Not corrupted - process it and continue with iterating through blobs */
2849 				if (ctx->iter_cb_fn) {
2850 					ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2851 				}
2852 				_spdk_bs_blob_list_add(blob);
2853 				spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2854 				return;
2855 			}
2856 
2857 		}
2858 
2859 		assert(len == sizeof(spdk_blob_id));
2860 
2861 		ctx->blob = blob;
2862 
2863 		/* Open clone to check if we are able to fix this blob or should we remove it */
2864 		spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, _spdk_bs_examine_clone, ctx);
2865 		return;
2866 	} else if (bserrno == -ENOENT) {
2867 		bserrno = 0;
2868 	} else {
2869 		/*
2870 		 * This case needs to be looked at further.  Same problem
2871 		 *  exists with applications that rely on explicit blob
2872 		 *  iteration.  We should just skip the blob that failed
2873 		 *  to load and continue on to the next one.
2874 		 */
2875 		SPDK_ERRLOG("Error in iterating blobs\n");
2876 	}
2877 
2878 	ctx->iter_cb_fn = NULL;
2879 
2880 	spdk_free(ctx->super);
2881 	spdk_free(ctx->mask);
2882 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2883 	free(ctx);
2884 }
2885 
2886 static void
2887 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2888 {
2889 	ctx->seq = seq;
2890 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2891 }
2892 
2893 static void
2894 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2895 {
2896 	struct spdk_bs_load_ctx *ctx = cb_arg;
2897 	int rc;
2898 
2899 	/* The type must be correct */
2900 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2901 
2902 	/* The length of the mask (in bits) must not be greater than
2903 	 * the length of the buffer (converted to bits) */
2904 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2905 
2906 	/* The length of the mask must be exactly equal to the size
2907 	 * (in pages) of the metadata region */
2908 	assert(ctx->mask->length == ctx->super->md_len);
2909 
2910 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2911 	if (rc < 0) {
2912 		spdk_free(ctx->mask);
2913 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2914 		return;
2915 	}
2916 
2917 	_spdk_bs_load_complete(seq, ctx, bserrno);
2918 }
2919 
2920 static void
2921 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2922 {
2923 	struct spdk_bs_load_ctx *ctx = cb_arg;
2924 	uint64_t		lba, lba_count, mask_size;
2925 	int			rc;
2926 
2927 	/* The type must be correct */
2928 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2929 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2930 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2931 					     struct spdk_blob_md_page) * 8));
2932 	/* The length of the mask must be exactly equal to the total number of clusters */
2933 	assert(ctx->mask->length == ctx->bs->total_clusters);
2934 
2935 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2936 	if (rc < 0) {
2937 		spdk_free(ctx->mask);
2938 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2939 		return;
2940 	}
2941 
2942 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2943 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2944 
2945 	spdk_free(ctx->mask);
2946 
2947 	/* Read the used blobids mask */
2948 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2949 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2950 				 SPDK_MALLOC_DMA);
2951 	if (!ctx->mask) {
2952 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2953 		return;
2954 	}
2955 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2956 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2957 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2958 				  _spdk_bs_load_used_blobids_cpl, ctx);
2959 }
2960 
2961 static void
2962 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2963 {
2964 	struct spdk_bs_load_ctx *ctx = cb_arg;
2965 	uint64_t		lba, lba_count, mask_size;
2966 	int			rc;
2967 
2968 	/* The type must be correct */
2969 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2970 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2971 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2972 				     8));
2973 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2974 	assert(ctx->mask->length == ctx->super->md_len);
2975 
2976 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2977 	if (rc < 0) {
2978 		spdk_free(ctx->mask);
2979 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2980 		return;
2981 	}
2982 
2983 	spdk_free(ctx->mask);
2984 
2985 	/* Read the used clusters mask */
2986 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2987 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2988 				 SPDK_MALLOC_DMA);
2989 	if (!ctx->mask) {
2990 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2991 		return;
2992 	}
2993 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2994 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2995 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2996 				  _spdk_bs_load_used_clusters_cpl, ctx);
2997 }
2998 
2999 static void
3000 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
3001 {
3002 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3003 	uint64_t lba, lba_count, mask_size;
3004 
3005 	/* Read the used pages mask */
3006 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
3007 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3008 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3009 	if (!ctx->mask) {
3010 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3011 		return;
3012 	}
3013 
3014 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3015 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3016 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3017 				  _spdk_bs_load_used_pages_cpl, ctx);
3018 }
3019 
3020 static int
3021 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
3022 {
3023 	struct spdk_blob_md_descriptor *desc;
3024 	size_t	cur_desc = 0;
3025 
3026 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3027 	while (cur_desc < sizeof(page->descriptors)) {
3028 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3029 			if (desc->length == 0) {
3030 				/* If padding and length are 0, this terminates the page */
3031 				break;
3032 			}
3033 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3034 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3035 			unsigned int				i, j;
3036 			unsigned int				cluster_count = 0;
3037 			uint32_t				cluster_idx;
3038 
3039 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3040 
3041 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3042 				for (j = 0; j < desc_extent->extents[i].length; j++) {
3043 					cluster_idx = desc_extent->extents[i].cluster_idx;
3044 					/*
3045 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
3046 					 * in the used cluster map.
3047 					 */
3048 					if (cluster_idx != 0) {
3049 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
3050 						if (bs->num_free_clusters == 0) {
3051 							return -ENOSPC;
3052 						}
3053 						bs->num_free_clusters--;
3054 					}
3055 					cluster_count++;
3056 				}
3057 			}
3058 			if (cluster_count == 0) {
3059 				return -EINVAL;
3060 			}
3061 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3062 			/* Skip this item */
3063 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3064 			/* Skip this item */
3065 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3066 			/* Skip this item */
3067 		} else {
3068 			/* Error */
3069 			return -EINVAL;
3070 		}
3071 		/* Advance to the next descriptor */
3072 		cur_desc += sizeof(*desc) + desc->length;
3073 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3074 			break;
3075 		}
3076 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3077 	}
3078 	return 0;
3079 }
3080 
3081 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
3082 {
3083 	uint32_t crc;
3084 
3085 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
3086 	if (crc != ctx->page->crc) {
3087 		return false;
3088 	}
3089 
3090 	if (ctx->page->sequence_num == 0 &&
3091 	    _spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
3092 		return false;
3093 	}
3094 	return true;
3095 }
3096 
3097 static void
3098 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3099 
3100 static void
3101 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3102 {
3103 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3104 
3105 	_spdk_bs_load_complete(seq, ctx, bserrno);
3106 }
3107 
3108 static void
3109 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3110 {
3111 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3112 
3113 	spdk_free(ctx->mask);
3114 	ctx->mask = NULL;
3115 
3116 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
3117 }
3118 
3119 static void
3120 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3121 {
3122 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3123 
3124 	spdk_free(ctx->mask);
3125 	ctx->mask = NULL;
3126 
3127 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
3128 }
3129 
3130 static void
3131 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3132 {
3133 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
3134 }
3135 
3136 static void
3137 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3138 {
3139 	struct spdk_bs_load_ctx *ctx = cb_arg;
3140 	uint64_t num_md_clusters;
3141 	uint64_t i;
3142 	uint32_t page_num;
3143 
3144 	if (bserrno != 0) {
3145 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
3146 		return;
3147 	}
3148 
3149 	page_num = ctx->cur_page;
3150 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
3151 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
3152 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
3153 			if (ctx->page->sequence_num == 0) {
3154 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
3155 			}
3156 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
3157 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3158 				return;
3159 			}
3160 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
3161 				ctx->in_page_chain = true;
3162 				ctx->cur_page = ctx->page->next;
3163 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3164 				return;
3165 			}
3166 		}
3167 	}
3168 
3169 	ctx->in_page_chain = false;
3170 
3171 	do {
3172 		ctx->page_index++;
3173 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
3174 
3175 	if (ctx->page_index < ctx->super->md_len) {
3176 		ctx->cur_page = ctx->page_index;
3177 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3178 	} else {
3179 		/* Claim all of the clusters used by the metadata */
3180 		num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
3181 		for (i = 0; i < num_md_clusters; i++) {
3182 			_spdk_bs_claim_cluster(ctx->bs, i);
3183 		}
3184 		spdk_free(ctx->page);
3185 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
3186 	}
3187 }
3188 
3189 static void
3190 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3191 {
3192 	struct spdk_bs_load_ctx *ctx = cb_arg;
3193 	uint64_t lba;
3194 
3195 	assert(ctx->cur_page < ctx->super->md_len);
3196 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3197 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3198 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3199 				  _spdk_bs_load_replay_md_cpl, ctx);
3200 }
3201 
3202 static void
3203 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3204 {
3205 	struct spdk_bs_load_ctx *ctx = cb_arg;
3206 
3207 	ctx->page_index = 0;
3208 	ctx->cur_page = 0;
3209 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3210 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3211 	if (!ctx->page) {
3212 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3213 		return;
3214 	}
3215 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3216 }
3217 
3218 static void
3219 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3220 {
3221 	struct spdk_bs_load_ctx *ctx = cb_arg;
3222 	int		rc;
3223 
3224 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3225 	if (rc < 0) {
3226 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3227 		return;
3228 	}
3229 
3230 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3231 	if (rc < 0) {
3232 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3233 		return;
3234 	}
3235 
3236 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3237 	if (rc < 0) {
3238 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3239 		return;
3240 	}
3241 
3242 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3243 	_spdk_bs_load_replay_md(seq, cb_arg);
3244 }
3245 
3246 static void
3247 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3248 {
3249 	struct spdk_bs_load_ctx *ctx = cb_arg;
3250 	uint32_t	crc;
3251 	int		rc;
3252 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3253 
3254 	if (ctx->super->version > SPDK_BS_VERSION ||
3255 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3256 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3257 		return;
3258 	}
3259 
3260 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3261 		   sizeof(ctx->super->signature)) != 0) {
3262 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3263 		return;
3264 	}
3265 
3266 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3267 	if (crc != ctx->super->crc) {
3268 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3269 		return;
3270 	}
3271 
3272 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3273 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3274 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3275 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3276 	} else {
3277 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3278 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3279 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3280 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3281 		return;
3282 	}
3283 
3284 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3285 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3286 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3287 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3288 		return;
3289 	}
3290 
3291 	if (ctx->super->size == 0) {
3292 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
3293 	}
3294 
3295 	if (ctx->super->io_unit_size == 0) {
3296 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
3297 	}
3298 
3299 	/* Parse the super block */
3300 	ctx->bs->clean = 1;
3301 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3302 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
3303 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3304 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
3305 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3306 	if (rc < 0) {
3307 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3308 		return;
3309 	}
3310 	ctx->bs->md_start = ctx->super->md_start;
3311 	ctx->bs->md_len = ctx->super->md_len;
3312 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
3313 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3314 	ctx->bs->super_blob = ctx->super->super_blob;
3315 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3316 
3317 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3318 		_spdk_bs_recover(seq, ctx);
3319 	} else {
3320 		_spdk_bs_load_read_used_pages(seq, ctx);
3321 	}
3322 }
3323 
3324 void
3325 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3326 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3327 {
3328 	struct spdk_blob_store	*bs;
3329 	struct spdk_bs_cpl	cpl;
3330 	spdk_bs_sequence_t	*seq;
3331 	struct spdk_bs_load_ctx *ctx;
3332 	struct spdk_bs_opts	opts = {};
3333 	int err;
3334 
3335 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3336 
3337 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3338 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3339 		dev->destroy(dev);
3340 		cb_fn(cb_arg, NULL, -EINVAL);
3341 		return;
3342 	}
3343 
3344 	if (o) {
3345 		opts = *o;
3346 	} else {
3347 		spdk_bs_opts_init(&opts);
3348 	}
3349 
3350 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3351 		dev->destroy(dev);
3352 		cb_fn(cb_arg, NULL, -EINVAL);
3353 		return;
3354 	}
3355 
3356 	err = _spdk_bs_alloc(dev, &opts, &bs);
3357 	if (err) {
3358 		dev->destroy(dev);
3359 		cb_fn(cb_arg, NULL, err);
3360 		return;
3361 	}
3362 
3363 	ctx = calloc(1, sizeof(*ctx));
3364 	if (!ctx) {
3365 		_spdk_bs_free(bs);
3366 		cb_fn(cb_arg, NULL, -ENOMEM);
3367 		return;
3368 	}
3369 
3370 	ctx->bs = bs;
3371 	ctx->iter_cb_fn = opts.iter_cb_fn;
3372 	ctx->iter_cb_arg = opts.iter_cb_arg;
3373 
3374 	/* Allocate memory for the super block */
3375 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3376 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3377 	if (!ctx->super) {
3378 		free(ctx);
3379 		_spdk_bs_free(bs);
3380 		cb_fn(cb_arg, NULL, -ENOMEM);
3381 		return;
3382 	}
3383 
3384 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3385 	cpl.u.bs_handle.cb_fn = cb_fn;
3386 	cpl.u.bs_handle.cb_arg = cb_arg;
3387 	cpl.u.bs_handle.bs = bs;
3388 
3389 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3390 	if (!seq) {
3391 		spdk_free(ctx->super);
3392 		free(ctx);
3393 		_spdk_bs_free(bs);
3394 		cb_fn(cb_arg, NULL, -ENOMEM);
3395 		return;
3396 	}
3397 
3398 	/* Read the super block */
3399 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3400 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3401 				  _spdk_bs_load_super_cpl, ctx);
3402 }
3403 
3404 /* END spdk_bs_load */
3405 
3406 /* START spdk_bs_dump */
3407 
3408 struct spdk_bs_dump_ctx {
3409 	struct spdk_blob_store		*bs;
3410 	struct spdk_bs_super_block	*super;
3411 	uint32_t			cur_page;
3412 	struct spdk_blob_md_page	*page;
3413 	spdk_bs_sequence_t		*seq;
3414 	FILE				*fp;
3415 	spdk_bs_dump_print_xattr	print_xattr_fn;
3416 	char				xattr_name[4096];
3417 };
3418 
3419 static void
3420 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3421 {
3422 	spdk_free(ctx->super);
3423 
3424 	/*
3425 	 * We need to defer calling spdk_bs_call_cpl() until after
3426 	 * dev destruction, so tuck these away for later use.
3427 	 */
3428 	ctx->bs->unload_err = bserrno;
3429 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3430 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3431 
3432 	spdk_bs_sequence_finish(seq, 0);
3433 	_spdk_bs_free(ctx->bs);
3434 	free(ctx);
3435 }
3436 
3437 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3438 
3439 static void
3440 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3441 {
3442 	uint32_t page_idx = ctx->cur_page;
3443 	struct spdk_blob_md_page *page = ctx->page;
3444 	struct spdk_blob_md_descriptor *desc;
3445 	size_t cur_desc = 0;
3446 	uint32_t crc;
3447 
3448 	fprintf(ctx->fp, "=========\n");
3449 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3450 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3451 
3452 	crc = _spdk_blob_md_page_calc_crc(page);
3453 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3454 
3455 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3456 	while (cur_desc < sizeof(page->descriptors)) {
3457 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3458 			if (desc->length == 0) {
3459 				/* If padding and length are 0, this terminates the page */
3460 				break;
3461 			}
3462 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3463 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3464 			unsigned int				i;
3465 
3466 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3467 
3468 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3469 				if (desc_extent->extents[i].cluster_idx != 0) {
3470 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3471 						desc_extent->extents[i].cluster_idx);
3472 				} else {
3473 					fprintf(ctx->fp, "Unallocated Extent - ");
3474 				}
3475 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3476 				fprintf(ctx->fp, "\n");
3477 			}
3478 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3479 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3480 			uint32_t i;
3481 
3482 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3483 
3484 			if (desc_xattr->length !=
3485 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3486 			    desc_xattr->name_length + desc_xattr->value_length) {
3487 			}
3488 
3489 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3490 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3491 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3492 			fprintf(ctx->fp, "       value = \"");
3493 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3494 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3495 					    desc_xattr->value_length);
3496 			fprintf(ctx->fp, "\"\n");
3497 			for (i = 0; i < desc_xattr->value_length; i++) {
3498 				if (i % 16 == 0) {
3499 					fprintf(ctx->fp, "               ");
3500 				}
3501 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3502 				if ((i + 1) % 16 == 0) {
3503 					fprintf(ctx->fp, "\n");
3504 				}
3505 			}
3506 			if (i % 16 != 0) {
3507 				fprintf(ctx->fp, "\n");
3508 			}
3509 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3510 			/* TODO */
3511 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3512 			/* TODO */
3513 		} else {
3514 			/* Error */
3515 		}
3516 		/* Advance to the next descriptor */
3517 		cur_desc += sizeof(*desc) + desc->length;
3518 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3519 			break;
3520 		}
3521 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3522 	}
3523 }
3524 
3525 static void
3526 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3527 {
3528 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3529 
3530 	if (bserrno != 0) {
3531 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3532 		return;
3533 	}
3534 
3535 	if (ctx->page->id != 0) {
3536 		_spdk_bs_dump_print_md_page(ctx);
3537 	}
3538 
3539 	ctx->cur_page++;
3540 
3541 	if (ctx->cur_page < ctx->super->md_len) {
3542 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3543 	} else {
3544 		spdk_free(ctx->page);
3545 		_spdk_bs_dump_finish(seq, ctx, 0);
3546 	}
3547 }
3548 
3549 static void
3550 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3551 {
3552 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3553 	uint64_t lba;
3554 
3555 	assert(ctx->cur_page < ctx->super->md_len);
3556 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3557 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3558 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3559 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3560 }
3561 
3562 static void
3563 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3564 {
3565 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3566 
3567 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3568 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3569 		   sizeof(ctx->super->signature)) != 0) {
3570 		fprintf(ctx->fp, "(Mismatch)\n");
3571 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3572 		return;
3573 	} else {
3574 		fprintf(ctx->fp, "(OK)\n");
3575 	}
3576 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3577 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3578 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3579 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3580 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3581 	fprintf(ctx->fp, "Super Blob ID: ");
3582 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3583 		fprintf(ctx->fp, "(None)\n");
3584 	} else {
3585 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3586 	}
3587 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3588 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3589 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3590 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3591 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3592 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3593 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3594 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3595 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3596 
3597 	ctx->cur_page = 0;
3598 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3599 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3600 	if (!ctx->page) {
3601 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3602 		return;
3603 	}
3604 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3605 }
3606 
3607 void
3608 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3609 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3610 {
3611 	struct spdk_blob_store	*bs;
3612 	struct spdk_bs_cpl	cpl;
3613 	spdk_bs_sequence_t	*seq;
3614 	struct spdk_bs_dump_ctx *ctx;
3615 	struct spdk_bs_opts	opts = {};
3616 	int err;
3617 
3618 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3619 
3620 	spdk_bs_opts_init(&opts);
3621 
3622 	err = _spdk_bs_alloc(dev, &opts, &bs);
3623 	if (err) {
3624 		dev->destroy(dev);
3625 		cb_fn(cb_arg, err);
3626 		return;
3627 	}
3628 
3629 	ctx = calloc(1, sizeof(*ctx));
3630 	if (!ctx) {
3631 		_spdk_bs_free(bs);
3632 		cb_fn(cb_arg, -ENOMEM);
3633 		return;
3634 	}
3635 
3636 	ctx->bs = bs;
3637 	ctx->fp = fp;
3638 	ctx->print_xattr_fn = print_xattr_fn;
3639 
3640 	/* Allocate memory for the super block */
3641 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3642 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3643 	if (!ctx->super) {
3644 		free(ctx);
3645 		_spdk_bs_free(bs);
3646 		cb_fn(cb_arg, -ENOMEM);
3647 		return;
3648 	}
3649 
3650 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3651 	cpl.u.bs_basic.cb_fn = cb_fn;
3652 	cpl.u.bs_basic.cb_arg = cb_arg;
3653 
3654 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3655 	if (!seq) {
3656 		spdk_free(ctx->super);
3657 		free(ctx);
3658 		_spdk_bs_free(bs);
3659 		cb_fn(cb_arg, -ENOMEM);
3660 		return;
3661 	}
3662 
3663 	/* Read the super block */
3664 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3665 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3666 				  _spdk_bs_dump_super_cpl, ctx);
3667 }
3668 
3669 /* END spdk_bs_dump */
3670 
3671 /* START spdk_bs_init */
3672 
3673 struct spdk_bs_init_ctx {
3674 	struct spdk_blob_store		*bs;
3675 	struct spdk_bs_super_block	*super;
3676 };
3677 
3678 static void
3679 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3680 {
3681 	struct spdk_bs_init_ctx *ctx = cb_arg;
3682 
3683 	spdk_free(ctx->super);
3684 	free(ctx);
3685 
3686 	spdk_bs_sequence_finish(seq, bserrno);
3687 }
3688 
3689 static void
3690 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3691 {
3692 	struct spdk_bs_init_ctx *ctx = cb_arg;
3693 
3694 	/* Write super block */
3695 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3696 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3697 				   _spdk_bs_init_persist_super_cpl, ctx);
3698 }
3699 
3700 void
3701 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3702 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3703 {
3704 	struct spdk_bs_init_ctx *ctx;
3705 	struct spdk_blob_store	*bs;
3706 	struct spdk_bs_cpl	cpl;
3707 	spdk_bs_sequence_t	*seq;
3708 	spdk_bs_batch_t		*batch;
3709 	uint64_t		num_md_lba;
3710 	uint64_t		num_md_pages;
3711 	uint64_t		num_md_clusters;
3712 	uint32_t		i;
3713 	struct spdk_bs_opts	opts = {};
3714 	int			rc;
3715 
3716 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3717 
3718 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3719 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3720 			    dev->blocklen);
3721 		dev->destroy(dev);
3722 		cb_fn(cb_arg, NULL, -EINVAL);
3723 		return;
3724 	}
3725 
3726 	if (o) {
3727 		opts = *o;
3728 	} else {
3729 		spdk_bs_opts_init(&opts);
3730 	}
3731 
3732 	if (_spdk_bs_opts_verify(&opts) != 0) {
3733 		dev->destroy(dev);
3734 		cb_fn(cb_arg, NULL, -EINVAL);
3735 		return;
3736 	}
3737 
3738 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3739 	if (rc) {
3740 		dev->destroy(dev);
3741 		cb_fn(cb_arg, NULL, rc);
3742 		return;
3743 	}
3744 
3745 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3746 		/* By default, allocate 1 page per cluster.
3747 		 * Technically, this over-allocates metadata
3748 		 * because more metadata will reduce the number
3749 		 * of usable clusters. This can be addressed with
3750 		 * more complex math in the future.
3751 		 */
3752 		bs->md_len = bs->total_clusters;
3753 	} else {
3754 		bs->md_len = opts.num_md_pages;
3755 	}
3756 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3757 	if (rc < 0) {
3758 		_spdk_bs_free(bs);
3759 		cb_fn(cb_arg, NULL, -ENOMEM);
3760 		return;
3761 	}
3762 
3763 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3764 	if (rc < 0) {
3765 		_spdk_bs_free(bs);
3766 		cb_fn(cb_arg, NULL, -ENOMEM);
3767 		return;
3768 	}
3769 
3770 	ctx = calloc(1, sizeof(*ctx));
3771 	if (!ctx) {
3772 		_spdk_bs_free(bs);
3773 		cb_fn(cb_arg, NULL, -ENOMEM);
3774 		return;
3775 	}
3776 
3777 	ctx->bs = bs;
3778 
3779 	/* Allocate memory for the super block */
3780 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3781 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3782 	if (!ctx->super) {
3783 		free(ctx);
3784 		_spdk_bs_free(bs);
3785 		cb_fn(cb_arg, NULL, -ENOMEM);
3786 		return;
3787 	}
3788 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3789 	       sizeof(ctx->super->signature));
3790 	ctx->super->version = SPDK_BS_VERSION;
3791 	ctx->super->length = sizeof(*ctx->super);
3792 	ctx->super->super_blob = bs->super_blob;
3793 	ctx->super->clean = 0;
3794 	ctx->super->cluster_size = bs->cluster_sz;
3795 	ctx->super->io_unit_size = bs->io_unit_size;
3796 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3797 
3798 	/* Calculate how many pages the metadata consumes at the front
3799 	 * of the disk.
3800 	 */
3801 
3802 	/* The super block uses 1 page */
3803 	num_md_pages = 1;
3804 
3805 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3806 	 * up to the nearest page, plus a header.
3807 	 */
3808 	ctx->super->used_page_mask_start = num_md_pages;
3809 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3810 					 spdk_divide_round_up(bs->md_len, 8),
3811 					 SPDK_BS_PAGE_SIZE);
3812 	num_md_pages += ctx->super->used_page_mask_len;
3813 
3814 	/* The used_clusters mask requires 1 bit per cluster, rounded
3815 	 * up to the nearest page, plus a header.
3816 	 */
3817 	ctx->super->used_cluster_mask_start = num_md_pages;
3818 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3819 					    spdk_divide_round_up(bs->total_clusters, 8),
3820 					    SPDK_BS_PAGE_SIZE);
3821 	num_md_pages += ctx->super->used_cluster_mask_len;
3822 
3823 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3824 	 * up to the nearest page, plus a header.
3825 	 */
3826 	ctx->super->used_blobid_mask_start = num_md_pages;
3827 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3828 					   spdk_divide_round_up(bs->md_len, 8),
3829 					   SPDK_BS_PAGE_SIZE);
3830 	num_md_pages += ctx->super->used_blobid_mask_len;
3831 
3832 	/* The metadata region size was chosen above */
3833 	ctx->super->md_start = bs->md_start = num_md_pages;
3834 	ctx->super->md_len = bs->md_len;
3835 	num_md_pages += bs->md_len;
3836 
3837 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3838 
3839 	ctx->super->size = dev->blockcnt * dev->blocklen;
3840 
3841 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3842 
3843 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
3844 	if (num_md_clusters > bs->total_clusters) {
3845 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3846 			    "please decrease number of pages reserved for metadata "
3847 			    "or increase cluster size.\n");
3848 		spdk_free(ctx->super);
3849 		free(ctx);
3850 		_spdk_bs_free(bs);
3851 		cb_fn(cb_arg, NULL, -ENOMEM);
3852 		return;
3853 	}
3854 	/* Claim all of the clusters used by the metadata */
3855 	for (i = 0; i < num_md_clusters; i++) {
3856 		_spdk_bs_claim_cluster(bs, i);
3857 	}
3858 
3859 	bs->total_data_clusters = bs->num_free_clusters;
3860 
3861 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3862 	cpl.u.bs_handle.cb_fn = cb_fn;
3863 	cpl.u.bs_handle.cb_arg = cb_arg;
3864 	cpl.u.bs_handle.bs = bs;
3865 
3866 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3867 	if (!seq) {
3868 		spdk_free(ctx->super);
3869 		free(ctx);
3870 		_spdk_bs_free(bs);
3871 		cb_fn(cb_arg, NULL, -ENOMEM);
3872 		return;
3873 	}
3874 
3875 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3876 
3877 	/* Clear metadata space */
3878 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3879 
3880 	if (opts.clear_method == BS_CLEAR_WITH_UNMAP) {
3881 		/* Trim data clusters */
3882 		spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3883 	} else if (opts.clear_method == BS_CLEAR_WITH_WRITE_ZEROES) {
3884 		/* Write_zeroes to data clusters */
3885 		spdk_bs_batch_write_zeroes_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3886 	}
3887 
3888 	spdk_bs_batch_close(batch);
3889 }
3890 
3891 /* END spdk_bs_init */
3892 
3893 /* START spdk_bs_destroy */
3894 
3895 static void
3896 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3897 {
3898 	struct spdk_bs_init_ctx *ctx = cb_arg;
3899 	struct spdk_blob_store *bs = ctx->bs;
3900 
3901 	/*
3902 	 * We need to defer calling spdk_bs_call_cpl() until after
3903 	 * dev destruction, so tuck these away for later use.
3904 	 */
3905 	bs->unload_err = bserrno;
3906 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3907 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3908 
3909 	spdk_bs_sequence_finish(seq, bserrno);
3910 
3911 	_spdk_bs_free(bs);
3912 	free(ctx);
3913 }
3914 
3915 void
3916 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3917 		void *cb_arg)
3918 {
3919 	struct spdk_bs_cpl	cpl;
3920 	spdk_bs_sequence_t	*seq;
3921 	struct spdk_bs_init_ctx *ctx;
3922 
3923 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3924 
3925 	if (!TAILQ_EMPTY(&bs->blobs)) {
3926 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3927 		cb_fn(cb_arg, -EBUSY);
3928 		return;
3929 	}
3930 
3931 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3932 	cpl.u.bs_basic.cb_fn = cb_fn;
3933 	cpl.u.bs_basic.cb_arg = cb_arg;
3934 
3935 	ctx = calloc(1, sizeof(*ctx));
3936 	if (!ctx) {
3937 		cb_fn(cb_arg, -ENOMEM);
3938 		return;
3939 	}
3940 
3941 	ctx->bs = bs;
3942 
3943 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3944 	if (!seq) {
3945 		free(ctx);
3946 		cb_fn(cb_arg, -ENOMEM);
3947 		return;
3948 	}
3949 
3950 	/* Write zeroes to the super block */
3951 	spdk_bs_sequence_write_zeroes_dev(seq,
3952 					  _spdk_bs_page_to_lba(bs, 0),
3953 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3954 					  _spdk_bs_destroy_trim_cpl, ctx);
3955 }
3956 
3957 /* END spdk_bs_destroy */
3958 
3959 /* START spdk_bs_unload */
3960 
3961 static void
3962 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3963 {
3964 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3965 
3966 	spdk_free(ctx->super);
3967 
3968 	/*
3969 	 * We need to defer calling spdk_bs_call_cpl() until after
3970 	 * dev destruction, so tuck these away for later use.
3971 	 */
3972 	ctx->bs->unload_err = bserrno;
3973 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3974 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3975 
3976 	spdk_bs_sequence_finish(seq, bserrno);
3977 
3978 	_spdk_bs_free(ctx->bs);
3979 	free(ctx);
3980 }
3981 
3982 static void
3983 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3984 {
3985 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3986 
3987 	spdk_free(ctx->mask);
3988 	ctx->super->clean = 1;
3989 
3990 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3991 }
3992 
3993 static void
3994 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3995 {
3996 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3997 
3998 	spdk_free(ctx->mask);
3999 	ctx->mask = NULL;
4000 
4001 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
4002 }
4003 
4004 static void
4005 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4006 {
4007 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4008 
4009 	spdk_free(ctx->mask);
4010 	ctx->mask = NULL;
4011 
4012 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
4013 }
4014 
4015 static void
4016 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4017 {
4018 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
4019 }
4020 
4021 void
4022 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
4023 {
4024 	struct spdk_bs_cpl	cpl;
4025 	spdk_bs_sequence_t	*seq;
4026 	struct spdk_bs_load_ctx *ctx;
4027 
4028 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
4029 
4030 	if (!TAILQ_EMPTY(&bs->blobs)) {
4031 		SPDK_ERRLOG("Blobstore still has open blobs\n");
4032 		cb_fn(cb_arg, -EBUSY);
4033 		return;
4034 	}
4035 
4036 	ctx = calloc(1, sizeof(*ctx));
4037 	if (!ctx) {
4038 		cb_fn(cb_arg, -ENOMEM);
4039 		return;
4040 	}
4041 
4042 	ctx->bs = bs;
4043 
4044 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4045 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4046 	if (!ctx->super) {
4047 		free(ctx);
4048 		cb_fn(cb_arg, -ENOMEM);
4049 		return;
4050 	}
4051 
4052 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4053 	cpl.u.bs_basic.cb_fn = cb_fn;
4054 	cpl.u.bs_basic.cb_arg = cb_arg;
4055 
4056 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4057 	if (!seq) {
4058 		spdk_free(ctx->super);
4059 		free(ctx);
4060 		cb_fn(cb_arg, -ENOMEM);
4061 		return;
4062 	}
4063 
4064 	/* Read super block */
4065 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4066 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4067 				  _spdk_bs_unload_read_super_cpl, ctx);
4068 }
4069 
4070 /* END spdk_bs_unload */
4071 
4072 /* START spdk_bs_set_super */
4073 
4074 struct spdk_bs_set_super_ctx {
4075 	struct spdk_blob_store		*bs;
4076 	struct spdk_bs_super_block	*super;
4077 };
4078 
4079 static void
4080 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4081 {
4082 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
4083 
4084 	if (bserrno != 0) {
4085 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
4086 	}
4087 
4088 	spdk_free(ctx->super);
4089 
4090 	spdk_bs_sequence_finish(seq, bserrno);
4091 
4092 	free(ctx);
4093 }
4094 
4095 static void
4096 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4097 {
4098 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
4099 
4100 	if (bserrno != 0) {
4101 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
4102 		spdk_free(ctx->super);
4103 		spdk_bs_sequence_finish(seq, bserrno);
4104 		free(ctx);
4105 		return;
4106 	}
4107 
4108 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
4109 }
4110 
4111 void
4112 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
4113 		  spdk_bs_op_complete cb_fn, void *cb_arg)
4114 {
4115 	struct spdk_bs_cpl		cpl;
4116 	spdk_bs_sequence_t		*seq;
4117 	struct spdk_bs_set_super_ctx	*ctx;
4118 
4119 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
4120 
4121 	ctx = calloc(1, sizeof(*ctx));
4122 	if (!ctx) {
4123 		cb_fn(cb_arg, -ENOMEM);
4124 		return;
4125 	}
4126 
4127 	ctx->bs = bs;
4128 
4129 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4130 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4131 	if (!ctx->super) {
4132 		free(ctx);
4133 		cb_fn(cb_arg, -ENOMEM);
4134 		return;
4135 	}
4136 
4137 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4138 	cpl.u.bs_basic.cb_fn = cb_fn;
4139 	cpl.u.bs_basic.cb_arg = cb_arg;
4140 
4141 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4142 	if (!seq) {
4143 		spdk_free(ctx->super);
4144 		free(ctx);
4145 		cb_fn(cb_arg, -ENOMEM);
4146 		return;
4147 	}
4148 
4149 	bs->super_blob = blobid;
4150 
4151 	/* Read super block */
4152 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4153 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4154 				  _spdk_bs_set_super_read_cpl, ctx);
4155 }
4156 
4157 /* END spdk_bs_set_super */
4158 
4159 void
4160 spdk_bs_get_super(struct spdk_blob_store *bs,
4161 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4162 {
4163 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
4164 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
4165 	} else {
4166 		cb_fn(cb_arg, bs->super_blob, 0);
4167 	}
4168 }
4169 
4170 uint64_t
4171 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
4172 {
4173 	return bs->cluster_sz;
4174 }
4175 
4176 uint64_t
4177 spdk_bs_get_page_size(struct spdk_blob_store *bs)
4178 {
4179 	return SPDK_BS_PAGE_SIZE;
4180 }
4181 
4182 uint64_t
4183 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
4184 {
4185 	return bs->io_unit_size;
4186 }
4187 
4188 uint64_t
4189 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
4190 {
4191 	return bs->num_free_clusters;
4192 }
4193 
4194 uint64_t
4195 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
4196 {
4197 	return bs->total_data_clusters;
4198 }
4199 
4200 static int
4201 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
4202 {
4203 	bs->md_channel = spdk_get_io_channel(bs);
4204 	if (!bs->md_channel) {
4205 		SPDK_ERRLOG("Failed to get IO channel.\n");
4206 		return -1;
4207 	}
4208 
4209 	return 0;
4210 }
4211 
4212 static int
4213 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
4214 {
4215 	spdk_put_io_channel(bs->md_channel);
4216 
4217 	return 0;
4218 }
4219 
4220 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4221 {
4222 	assert(blob != NULL);
4223 
4224 	return blob->id;
4225 }
4226 
4227 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4228 {
4229 	assert(blob != NULL);
4230 
4231 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4232 }
4233 
4234 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
4235 {
4236 	assert(blob != NULL);
4237 
4238 	return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs);
4239 }
4240 
4241 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4242 {
4243 	assert(blob != NULL);
4244 
4245 	return blob->active.num_clusters;
4246 }
4247 
4248 /* START spdk_bs_create_blob */
4249 
4250 static void
4251 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4252 {
4253 	struct spdk_blob *blob = cb_arg;
4254 
4255 	_spdk_blob_free(blob);
4256 
4257 	spdk_bs_sequence_finish(seq, bserrno);
4258 }
4259 
4260 static int
4261 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4262 		      bool internal)
4263 {
4264 	uint64_t i;
4265 	size_t value_len = 0;
4266 	int rc;
4267 	const void *value = NULL;
4268 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4269 		return -EINVAL;
4270 	}
4271 	for (i = 0; i < xattrs->count; i++) {
4272 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4273 		if (value == NULL || value_len == 0) {
4274 			return -EINVAL;
4275 		}
4276 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4277 		if (rc < 0) {
4278 			return rc;
4279 		}
4280 	}
4281 	return 0;
4282 }
4283 
4284 static void
4285 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4286 		     const struct spdk_blob_opts *opts,
4287 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4288 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4289 {
4290 	struct spdk_blob	*blob;
4291 	uint32_t		page_idx;
4292 	struct spdk_bs_cpl	cpl;
4293 	struct spdk_blob_opts	opts_default;
4294 	struct spdk_blob_xattr_opts internal_xattrs_default;
4295 	spdk_bs_sequence_t	*seq;
4296 	spdk_blob_id		id;
4297 	int rc;
4298 
4299 	assert(spdk_get_thread() == bs->md_thread);
4300 
4301 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4302 	if (page_idx == UINT32_MAX) {
4303 		cb_fn(cb_arg, 0, -ENOMEM);
4304 		return;
4305 	}
4306 	spdk_bit_array_set(bs->used_blobids, page_idx);
4307 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4308 
4309 	id = _spdk_bs_page_to_blobid(page_idx);
4310 
4311 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4312 
4313 	blob = _spdk_blob_alloc(bs, id);
4314 	if (!blob) {
4315 		cb_fn(cb_arg, 0, -ENOMEM);
4316 		return;
4317 	}
4318 
4319 	if (!opts) {
4320 		spdk_blob_opts_init(&opts_default);
4321 		opts = &opts_default;
4322 	}
4323 	if (!internal_xattrs) {
4324 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4325 		internal_xattrs = &internal_xattrs_default;
4326 	}
4327 
4328 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4329 	if (rc < 0) {
4330 		_spdk_blob_free(blob);
4331 		cb_fn(cb_arg, 0, rc);
4332 		return;
4333 	}
4334 
4335 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4336 	if (rc < 0) {
4337 		_spdk_blob_free(blob);
4338 		cb_fn(cb_arg, 0, rc);
4339 		return;
4340 	}
4341 
4342 	if (opts->thin_provision) {
4343 		_spdk_blob_set_thin_provision(blob);
4344 	}
4345 
4346 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4347 	if (rc < 0) {
4348 		_spdk_blob_free(blob);
4349 		cb_fn(cb_arg, 0, rc);
4350 		return;
4351 	}
4352 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4353 	cpl.u.blobid.cb_fn = cb_fn;
4354 	cpl.u.blobid.cb_arg = cb_arg;
4355 	cpl.u.blobid.blobid = blob->id;
4356 
4357 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4358 	if (!seq) {
4359 		_spdk_blob_free(blob);
4360 		cb_fn(cb_arg, 0, -ENOMEM);
4361 		return;
4362 	}
4363 
4364 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4365 }
4366 
4367 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4368 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4369 {
4370 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4371 }
4372 
4373 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4374 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4375 {
4376 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4377 }
4378 
4379 /* END spdk_bs_create_blob */
4380 
4381 /* START blob_cleanup */
4382 
4383 struct spdk_clone_snapshot_ctx {
4384 	struct spdk_bs_cpl      cpl;
4385 	int bserrno;
4386 	bool frozen;
4387 
4388 	struct spdk_io_channel *channel;
4389 
4390 	/* Current cluster for inflate operation */
4391 	uint64_t cluster;
4392 
4393 	/* For inflation force allocation of all unallocated clusters and remove
4394 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4395 	bool allocate_all;
4396 
4397 	struct {
4398 		spdk_blob_id id;
4399 		struct spdk_blob *blob;
4400 	} original;
4401 	struct {
4402 		spdk_blob_id id;
4403 		struct spdk_blob *blob;
4404 	} new;
4405 
4406 	/* xattrs specified for snapshot/clones only. They have no impact on
4407 	 * the original blobs xattrs. */
4408 	const struct spdk_blob_xattr_opts *xattrs;
4409 };
4410 
4411 static void
4412 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4413 {
4414 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4415 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4416 
4417 	if (bserrno != 0) {
4418 		if (ctx->bserrno != 0) {
4419 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4420 		} else {
4421 			ctx->bserrno = bserrno;
4422 		}
4423 	}
4424 
4425 	switch (cpl->type) {
4426 	case SPDK_BS_CPL_TYPE_BLOBID:
4427 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4428 		break;
4429 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4430 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4431 		break;
4432 	default:
4433 		SPDK_UNREACHABLE();
4434 		break;
4435 	}
4436 
4437 	free(ctx);
4438 }
4439 
4440 static void
4441 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4442 {
4443 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4444 	struct spdk_blob *origblob = ctx->original.blob;
4445 
4446 	if (bserrno != 0) {
4447 		if (ctx->bserrno != 0) {
4448 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4449 		} else {
4450 			ctx->bserrno = bserrno;
4451 		}
4452 	}
4453 
4454 	ctx->original.id = origblob->id;
4455 	origblob->locked_operation_in_progress = false;
4456 
4457 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4458 }
4459 
4460 static void
4461 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4462 {
4463 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4464 	struct spdk_blob *origblob = ctx->original.blob;
4465 
4466 	if (bserrno != 0) {
4467 		if (ctx->bserrno != 0) {
4468 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4469 		} else {
4470 			ctx->bserrno = bserrno;
4471 		}
4472 	}
4473 
4474 	if (ctx->frozen) {
4475 		/* Unfreeze any outstanding I/O */
4476 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4477 	} else {
4478 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4479 	}
4480 
4481 }
4482 
4483 static void
4484 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4485 {
4486 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4487 	struct spdk_blob *newblob = ctx->new.blob;
4488 
4489 	if (bserrno != 0) {
4490 		if (ctx->bserrno != 0) {
4491 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4492 		} else {
4493 			ctx->bserrno = bserrno;
4494 		}
4495 	}
4496 
4497 	ctx->new.id = newblob->id;
4498 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4499 }
4500 
4501 /* END blob_cleanup */
4502 
4503 /* START spdk_bs_create_snapshot */
4504 
4505 static void
4506 _spdk_bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
4507 {
4508 	uint64_t *cluster_temp;
4509 
4510 	cluster_temp = blob1->active.clusters;
4511 	blob1->active.clusters = blob2->active.clusters;
4512 	blob2->active.clusters = cluster_temp;
4513 }
4514 
4515 static void
4516 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4517 {
4518 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4519 	struct spdk_blob *origblob = ctx->original.blob;
4520 	struct spdk_blob *newblob = ctx->new.blob;
4521 
4522 	if (bserrno != 0) {
4523 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4524 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4525 		return;
4526 	}
4527 
4528 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4529 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4530 	if (bserrno != 0) {
4531 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4532 		return;
4533 	}
4534 
4535 	_spdk_bs_blob_list_add(ctx->original.blob);
4536 
4537 	spdk_blob_set_read_only(newblob);
4538 
4539 	/* sync snapshot metadata */
4540 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4541 }
4542 
4543 static void
4544 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4545 {
4546 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4547 	struct spdk_blob *origblob = ctx->original.blob;
4548 	struct spdk_blob *newblob = ctx->new.blob;
4549 
4550 	if (bserrno != 0) {
4551 		/* return cluster map back to original */
4552 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4553 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4554 		return;
4555 	}
4556 
4557 	/* Set internal xattr for snapshot id */
4558 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4559 	if (bserrno != 0) {
4560 		/* return cluster map back to original */
4561 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4562 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4563 		return;
4564 	}
4565 
4566 	_spdk_bs_blob_list_remove(origblob);
4567 	origblob->parent_id = newblob->id;
4568 
4569 	/* Create new back_bs_dev for snapshot */
4570 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4571 	if (origblob->back_bs_dev == NULL) {
4572 		/* return cluster map back to original */
4573 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4574 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4575 		return;
4576 	}
4577 
4578 	/* set clone blob as thin provisioned */
4579 	_spdk_blob_set_thin_provision(origblob);
4580 
4581 	_spdk_bs_blob_list_add(newblob);
4582 
4583 	/* sync clone metadata */
4584 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4585 }
4586 
4587 static void
4588 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4589 {
4590 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4591 	struct spdk_blob *origblob = ctx->original.blob;
4592 	struct spdk_blob *newblob = ctx->new.blob;
4593 	int bserrno;
4594 
4595 	if (rc != 0) {
4596 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4597 		return;
4598 	}
4599 
4600 	ctx->frozen = true;
4601 
4602 	/* set new back_bs_dev for snapshot */
4603 	newblob->back_bs_dev = origblob->back_bs_dev;
4604 	/* Set invalid flags from origblob */
4605 	newblob->invalid_flags = origblob->invalid_flags;
4606 
4607 	/* inherit parent from original blob if set */
4608 	newblob->parent_id = origblob->parent_id;
4609 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4610 		/* Set internal xattr for snapshot id */
4611 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4612 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4613 		if (bserrno != 0) {
4614 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4615 			return;
4616 		}
4617 	}
4618 
4619 	/* swap cluster maps */
4620 	_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4621 
4622 	/* sync snapshot metadata */
4623 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4624 }
4625 
4626 static void
4627 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4628 {
4629 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4630 	struct spdk_blob *origblob = ctx->original.blob;
4631 	struct spdk_blob *newblob = _blob;
4632 
4633 	if (bserrno != 0) {
4634 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4635 		return;
4636 	}
4637 
4638 	ctx->new.blob = newblob;
4639 
4640 	/* Zero out newblob cluster map */
4641 	memset(newblob->active.clusters, 0,
4642 	       newblob->active.num_clusters * sizeof(newblob->active.clusters));
4643 
4644 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4645 }
4646 
4647 static void
4648 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4649 {
4650 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4651 	struct spdk_blob *origblob = ctx->original.blob;
4652 
4653 	if (bserrno != 0) {
4654 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4655 		return;
4656 	}
4657 
4658 	ctx->new.id = blobid;
4659 	ctx->cpl.u.blobid.blobid = blobid;
4660 
4661 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4662 }
4663 
4664 
4665 static void
4666 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4667 			const void **value, size_t *value_len)
4668 {
4669 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4670 
4671 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4672 	*value = &blob->id;
4673 	*value_len = sizeof(blob->id);
4674 }
4675 
4676 static void
4677 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4678 {
4679 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4680 	struct spdk_blob_opts opts;
4681 	struct spdk_blob_xattr_opts internal_xattrs;
4682 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4683 
4684 	if (bserrno != 0) {
4685 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4686 		return;
4687 	}
4688 
4689 	ctx->original.blob = _blob;
4690 
4691 	if (_blob->data_ro || _blob->md_ro) {
4692 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4693 			      _blob->id);
4694 		ctx->bserrno = -EINVAL;
4695 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4696 		return;
4697 	}
4698 
4699 	if (_blob->locked_operation_in_progress) {
4700 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot - another operation in progress\n");
4701 		ctx->bserrno = -EBUSY;
4702 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4703 		return;
4704 	}
4705 
4706 	_blob->locked_operation_in_progress = true;
4707 
4708 	spdk_blob_opts_init(&opts);
4709 	_spdk_blob_xattrs_init(&internal_xattrs);
4710 
4711 	/* Change the size of new blob to the same as in original blob,
4712 	 * but do not allocate clusters */
4713 	opts.thin_provision = true;
4714 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4715 
4716 	/* If there are any xattrs specified for snapshot, set them now */
4717 	if (ctx->xattrs) {
4718 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4719 	}
4720 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4721 	internal_xattrs.count = 1;
4722 	internal_xattrs.ctx = _blob;
4723 	internal_xattrs.names = xattrs_names;
4724 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4725 
4726 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4727 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4728 }
4729 
4730 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4731 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4732 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4733 {
4734 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4735 
4736 	if (!ctx) {
4737 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4738 		return;
4739 	}
4740 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4741 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4742 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4743 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4744 	ctx->bserrno = 0;
4745 	ctx->frozen = false;
4746 	ctx->original.id = blobid;
4747 	ctx->xattrs = snapshot_xattrs;
4748 
4749 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4750 }
4751 /* END spdk_bs_create_snapshot */
4752 
4753 /* START spdk_bs_create_clone */
4754 
4755 static void
4756 _spdk_bs_xattr_clone(void *arg, const char *name,
4757 		     const void **value, size_t *value_len)
4758 {
4759 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4760 
4761 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4762 	*value = &blob->id;
4763 	*value_len = sizeof(blob->id);
4764 }
4765 
4766 static void
4767 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4768 {
4769 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4770 	struct spdk_blob *clone = _blob;
4771 
4772 	ctx->new.blob = clone;
4773 	_spdk_bs_blob_list_add(clone);
4774 
4775 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4776 }
4777 
4778 static void
4779 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4780 {
4781 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4782 
4783 	ctx->cpl.u.blobid.blobid = blobid;
4784 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4785 }
4786 
4787 static void
4788 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4789 {
4790 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4791 	struct spdk_blob_opts		opts;
4792 	struct spdk_blob_xattr_opts internal_xattrs;
4793 	char *xattr_names[] = { BLOB_SNAPSHOT };
4794 
4795 	if (bserrno != 0) {
4796 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4797 		return;
4798 	}
4799 
4800 	ctx->original.blob = _blob;
4801 
4802 	if (!_blob->data_ro || !_blob->md_ro) {
4803 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4804 		ctx->bserrno = -EINVAL;
4805 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4806 		return;
4807 	}
4808 
4809 	if (_blob->locked_operation_in_progress) {
4810 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create clone - another operation in progress\n");
4811 		ctx->bserrno = -EBUSY;
4812 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4813 		return;
4814 	}
4815 
4816 	_blob->locked_operation_in_progress = true;
4817 
4818 	spdk_blob_opts_init(&opts);
4819 	_spdk_blob_xattrs_init(&internal_xattrs);
4820 
4821 	opts.thin_provision = true;
4822 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4823 	if (ctx->xattrs) {
4824 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4825 	}
4826 
4827 	/* Set internal xattr BLOB_SNAPSHOT */
4828 	internal_xattrs.count = 1;
4829 	internal_xattrs.ctx = _blob;
4830 	internal_xattrs.names = xattr_names;
4831 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4832 
4833 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4834 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4835 }
4836 
4837 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4838 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4839 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4840 {
4841 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4842 
4843 	if (!ctx) {
4844 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4845 		return;
4846 	}
4847 
4848 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4849 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4850 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4851 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4852 	ctx->bserrno = 0;
4853 	ctx->xattrs = clone_xattrs;
4854 	ctx->original.id = blobid;
4855 
4856 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4857 }
4858 
4859 /* END spdk_bs_create_clone */
4860 
4861 /* START spdk_bs_inflate_blob */
4862 
4863 static void
4864 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4865 {
4866 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4867 	struct spdk_blob *_blob = ctx->original.blob;
4868 
4869 	if (bserrno != 0) {
4870 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4871 		return;
4872 	}
4873 
4874 	assert(_parent != NULL);
4875 
4876 	_spdk_bs_blob_list_remove(_blob);
4877 	_blob->parent_id = _parent->id;
4878 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4879 			     sizeof(spdk_blob_id), true);
4880 
4881 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4882 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4883 	_spdk_bs_blob_list_add(_blob);
4884 
4885 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4886 }
4887 
4888 static void
4889 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4890 {
4891 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4892 	struct spdk_blob *_blob = ctx->original.blob;
4893 	struct spdk_blob *_parent;
4894 
4895 	if (bserrno != 0) {
4896 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4897 		return;
4898 	}
4899 
4900 	if (ctx->allocate_all) {
4901 		/* remove thin provisioning */
4902 		_spdk_bs_blob_list_remove(_blob);
4903 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4904 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4905 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4906 		_blob->back_bs_dev = NULL;
4907 		_blob->parent_id = SPDK_BLOBID_INVALID;
4908 	} else {
4909 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4910 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4911 			/* We must change the parent of the inflated blob */
4912 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4913 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4914 			return;
4915 		}
4916 
4917 		_spdk_bs_blob_list_remove(_blob);
4918 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4919 		_blob->parent_id = SPDK_BLOBID_INVALID;
4920 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4921 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4922 	}
4923 
4924 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4925 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4926 }
4927 
4928 /* Check if cluster needs allocation */
4929 static inline bool
4930 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4931 {
4932 	struct spdk_blob_bs_dev *b;
4933 
4934 	assert(blob != NULL);
4935 
4936 	if (blob->active.clusters[cluster] != 0) {
4937 		/* Cluster is already allocated */
4938 		return false;
4939 	}
4940 
4941 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4942 		/* Blob have no parent blob */
4943 		return allocate_all;
4944 	}
4945 
4946 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4947 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4948 }
4949 
4950 static void
4951 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4952 {
4953 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4954 	struct spdk_blob *_blob = ctx->original.blob;
4955 	uint64_t offset;
4956 
4957 	if (bserrno != 0) {
4958 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4959 		return;
4960 	}
4961 
4962 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4963 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4964 			break;
4965 		}
4966 	}
4967 
4968 	if (ctx->cluster < _blob->active.num_clusters) {
4969 		offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster);
4970 
4971 		/* We may safely increment a cluster before write */
4972 		ctx->cluster++;
4973 
4974 		/* Use zero length write to touch a cluster */
4975 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4976 				   _spdk_bs_inflate_blob_touch_next, ctx);
4977 	} else {
4978 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4979 	}
4980 }
4981 
4982 static void
4983 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4984 {
4985 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4986 	uint64_t lfc; /* lowest free cluster */
4987 	uint64_t i;
4988 
4989 	if (bserrno != 0) {
4990 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4991 		return;
4992 	}
4993 
4994 	ctx->original.blob = _blob;
4995 
4996 	if (_blob->locked_operation_in_progress) {
4997 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot inflate blob - another operation in progress\n");
4998 		ctx->bserrno = -EBUSY;
4999 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
5000 		return;
5001 	}
5002 
5003 	_blob->locked_operation_in_progress = true;
5004 
5005 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
5006 		/* This blob have no parent, so we cannot decouple it. */
5007 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
5008 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
5009 		return;
5010 	}
5011 
5012 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
5013 		/* This is not thin provisioned blob. No need to inflate. */
5014 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
5015 		return;
5016 	}
5017 
5018 	/* Do two passes - one to verify that we can obtain enough clusters
5019 	 * and another to actually claim them.
5020 	 */
5021 	lfc = 0;
5022 	for (i = 0; i < _blob->active.num_clusters; i++) {
5023 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
5024 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
5025 			if (lfc == UINT32_MAX) {
5026 				/* No more free clusters. Cannot satisfy the request */
5027 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
5028 				return;
5029 			}
5030 			lfc++;
5031 		}
5032 	}
5033 
5034 	ctx->cluster = 0;
5035 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
5036 }
5037 
5038 static void
5039 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5040 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
5041 {
5042 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
5043 
5044 	if (!ctx) {
5045 		cb_fn(cb_arg, -ENOMEM);
5046 		return;
5047 	}
5048 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5049 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
5050 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
5051 	ctx->bserrno = 0;
5052 	ctx->original.id = blobid;
5053 	ctx->channel = channel;
5054 	ctx->allocate_all = allocate_all;
5055 
5056 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
5057 }
5058 
5059 void
5060 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5061 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
5062 {
5063 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
5064 }
5065 
5066 void
5067 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5068 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
5069 {
5070 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
5071 }
5072 /* END spdk_bs_inflate_blob */
5073 
5074 /* START spdk_blob_resize */
5075 struct spdk_bs_resize_ctx {
5076 	spdk_blob_op_complete cb_fn;
5077 	void *cb_arg;
5078 	struct spdk_blob *blob;
5079 	uint64_t sz;
5080 	int rc;
5081 };
5082 
5083 static void
5084 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
5085 {
5086 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
5087 
5088 	if (rc != 0) {
5089 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
5090 	}
5091 
5092 	if (ctx->rc != 0) {
5093 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
5094 		rc = ctx->rc;
5095 	}
5096 
5097 	ctx->blob->locked_operation_in_progress = false;
5098 
5099 	ctx->cb_fn(ctx->cb_arg, rc);
5100 	free(ctx);
5101 }
5102 
5103 static void
5104 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
5105 {
5106 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
5107 
5108 	if (rc != 0) {
5109 		ctx->blob->locked_operation_in_progress = false;
5110 		ctx->cb_fn(ctx->cb_arg, rc);
5111 		free(ctx);
5112 		return;
5113 	}
5114 
5115 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
5116 
5117 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
5118 }
5119 
5120 void
5121 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
5122 {
5123 	struct spdk_bs_resize_ctx *ctx;
5124 
5125 	_spdk_blob_verify_md_op(blob);
5126 
5127 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
5128 
5129 	if (blob->md_ro) {
5130 		cb_fn(cb_arg, -EPERM);
5131 		return;
5132 	}
5133 
5134 	if (sz == blob->active.num_clusters) {
5135 		cb_fn(cb_arg, 0);
5136 		return;
5137 	}
5138 
5139 	if (blob->locked_operation_in_progress) {
5140 		cb_fn(cb_arg, -EBUSY);
5141 		return;
5142 	}
5143 
5144 	ctx = calloc(1, sizeof(*ctx));
5145 	if (!ctx) {
5146 		cb_fn(cb_arg, -ENOMEM);
5147 		return;
5148 	}
5149 
5150 	blob->locked_operation_in_progress = true;
5151 	ctx->cb_fn = cb_fn;
5152 	ctx->cb_arg = cb_arg;
5153 	ctx->blob = blob;
5154 	ctx->sz = sz;
5155 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
5156 }
5157 
5158 /* END spdk_blob_resize */
5159 
5160 
5161 /* START spdk_bs_delete_blob */
5162 
5163 static void
5164 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
5165 {
5166 	spdk_bs_sequence_t *seq = cb_arg;
5167 
5168 	spdk_bs_sequence_finish(seq, bserrno);
5169 }
5170 
5171 static void
5172 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5173 {
5174 	struct spdk_blob *blob = cb_arg;
5175 
5176 	if (bserrno != 0) {
5177 		/*
5178 		 * We already removed this blob from the blobstore tailq, so
5179 		 *  we need to free it here since this is the last reference
5180 		 *  to it.
5181 		 */
5182 		_spdk_blob_free(blob);
5183 		_spdk_bs_delete_close_cpl(seq, bserrno);
5184 		return;
5185 	}
5186 
5187 	/*
5188 	 * This will immediately decrement the ref_count and call
5189 	 *  the completion routine since the metadata state is clean.
5190 	 *  By calling spdk_blob_close, we reduce the number of call
5191 	 *  points into code that touches the blob->open_ref count
5192 	 *  and the blobstore's blob list.
5193 	 */
5194 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
5195 }
5196 
5197 struct delete_snapshot_ctx {
5198 	struct spdk_blob_list *parent_snapshot_entry;
5199 	struct spdk_blob *snapshot;
5200 	bool snapshot_md_ro;
5201 	struct spdk_blob *clone;
5202 	bool clone_md_ro;
5203 	spdk_blob_op_with_handle_complete cb_fn;
5204 	void *cb_arg;
5205 	int bserrno;
5206 };
5207 
5208 static void
5209 _spdk_delete_blob_cleanup_finish(void *cb_arg, int bserrno)
5210 {
5211 	struct delete_snapshot_ctx *ctx = cb_arg;
5212 
5213 	if (bserrno != 0) {
5214 		SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno);
5215 	}
5216 
5217 	assert(ctx != NULL);
5218 
5219 	if (bserrno != 0 && ctx->bserrno == 0) {
5220 		ctx->bserrno = bserrno;
5221 	}
5222 
5223 	ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno);
5224 	free(ctx);
5225 }
5226 
5227 static void
5228 _spdk_delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno)
5229 {
5230 	struct delete_snapshot_ctx *ctx = cb_arg;
5231 
5232 	if (bserrno != 0) {
5233 		ctx->bserrno = bserrno;
5234 		SPDK_ERRLOG("Clone cleanup error %d\n", bserrno);
5235 	}
5236 
5237 	/* open_ref == 1 menas that only deletion context has opened this snapshot
5238 	 * open_ref == 2 menas that clone has opened this snapshot as well,
5239 	 * so we have to add it back to the blobs list */
5240 	if (ctx->snapshot->open_ref == 2) {
5241 		TAILQ_INSERT_HEAD(&ctx->snapshot->bs->blobs, ctx->snapshot, link);
5242 	}
5243 
5244 	ctx->snapshot->locked_operation_in_progress = false;
5245 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
5246 
5247 	spdk_blob_close(ctx->snapshot, _spdk_delete_blob_cleanup_finish, ctx);
5248 }
5249 
5250 static void
5251 _spdk_delete_snapshot_cleanup_clone(void *cb_arg, int bserrno)
5252 {
5253 	struct delete_snapshot_ctx *ctx = cb_arg;
5254 
5255 	ctx->clone->locked_operation_in_progress = false;
5256 	ctx->clone->md_ro = ctx->clone_md_ro;
5257 
5258 	spdk_blob_close(ctx->clone, _spdk_delete_snapshot_cleanup_snapshot, ctx);
5259 }
5260 
5261 static void
5262 _spdk_delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
5263 {
5264 	struct delete_snapshot_ctx *ctx = cb_arg;
5265 
5266 	if (bserrno) {
5267 		ctx->bserrno = bserrno;
5268 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5269 		return;
5270 	}
5271 
5272 	ctx->clone->locked_operation_in_progress = false;
5273 	spdk_blob_close(ctx->clone, _spdk_delete_blob_cleanup_finish, ctx);
5274 }
5275 
5276 static void
5277 _spdk_delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno)
5278 {
5279 	struct delete_snapshot_ctx *ctx = cb_arg;
5280 	struct spdk_blob_list *parent_snapshot_entry = NULL;
5281 	struct spdk_blob_list *snapshot_entry = NULL;
5282 	struct spdk_blob_list *clone_entry = NULL;
5283 	struct spdk_blob_list *snapshot_clone_entry = NULL;
5284 
5285 	if (bserrno) {
5286 		SPDK_ERRLOG("Failed to sync MD on blob\n");
5287 		ctx->bserrno = bserrno;
5288 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5289 		return;
5290 	}
5291 
5292 	/* Get snapshot entry for the snapshot we want to remove */
5293 	snapshot_entry = _spdk_bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id);
5294 
5295 	assert(snapshot_entry != NULL);
5296 
5297 	/* Remove clone entry in this snapshot (at this point there can be only one clone) */
5298 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
5299 	assert(clone_entry != NULL);
5300 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
5301 	snapshot_entry->clone_count--;
5302 	assert(TAILQ_EMPTY(&snapshot_entry->clones));
5303 
5304 	if (ctx->snapshot->parent_id != SPDK_BLOBID_INVALID) {
5305 		/* This snapshot is at the same time a clone of another snapshot - we need to
5306 		 * update parent snapshot (remove current clone, add new one inherited from
5307 		 * the snapshot that is being removed) */
5308 
5309 		/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
5310 		 * snapshot that we are removing */
5311 		_spdk_blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry,
5312 				&snapshot_clone_entry);
5313 
5314 		/* Switch clone entry in parent snapshot */
5315 		TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link);
5316 		TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link);
5317 		free(snapshot_clone_entry);
5318 	} else {
5319 		/* No parent snapshot - just remove clone entry */
5320 		free(clone_entry);
5321 	}
5322 
5323 	/* Restore md_ro flags */
5324 	ctx->clone->md_ro = ctx->clone_md_ro;
5325 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
5326 
5327 	_spdk_blob_unfreeze_io(ctx->clone, _spdk_delete_snapshot_unfreeze_cpl, ctx);
5328 }
5329 
5330 static void
5331 _spdk_delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno)
5332 {
5333 	struct delete_snapshot_ctx *ctx = cb_arg;
5334 	uint64_t i;
5335 
5336 	ctx->snapshot->md_ro = false;
5337 
5338 	if (bserrno) {
5339 		SPDK_ERRLOG("Failed to sync MD on clone\n");
5340 		ctx->bserrno = bserrno;
5341 
5342 		/* Restore snapshot to previous state */
5343 		bserrno = _spdk_blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true);
5344 		if (bserrno != 0) {
5345 			_spdk_delete_snapshot_cleanup_clone(ctx, bserrno);
5346 			return;
5347 		}
5348 
5349 		spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_cleanup_clone, ctx);
5350 		return;
5351 	}
5352 
5353 	/* Clear cluster map entries for snapshot */
5354 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
5355 		if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) {
5356 			ctx->snapshot->active.clusters[i] = 0;
5357 		}
5358 	}
5359 
5360 	ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY;
5361 
5362 	if (ctx->parent_snapshot_entry != NULL) {
5363 		ctx->snapshot->back_bs_dev = NULL;
5364 	}
5365 
5366 	spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_sync_snapshot_cpl, ctx);
5367 }
5368 
5369 static void
5370 _spdk_delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno)
5371 {
5372 	struct delete_snapshot_ctx *ctx = cb_arg;
5373 	uint64_t i;
5374 
5375 	/* Temporarily override md_ro flag for clone for MD modification */
5376 	ctx->clone_md_ro = ctx->clone->md_ro;
5377 	ctx->clone->md_ro = false;
5378 
5379 	if (bserrno) {
5380 		SPDK_ERRLOG("Failed to sync MD with xattr on blob\n");
5381 		ctx->bserrno = bserrno;
5382 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5383 		return;
5384 	}
5385 
5386 	/* Copy snapshot map to clone map (only unallocated clusters in clone) */
5387 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
5388 		if (ctx->clone->active.clusters[i] == 0) {
5389 			ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i];
5390 		}
5391 	}
5392 
5393 	/* Delete old backing bs_dev from clone (related to snapshot that will be removed) */
5394 	ctx->clone->back_bs_dev->destroy(ctx->clone->back_bs_dev);
5395 
5396 	/* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */
5397 	if (ctx->parent_snapshot_entry != NULL) {
5398 		/* ...to parent snapshot */
5399 		ctx->clone->parent_id = ctx->parent_snapshot_entry->id;
5400 		ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev;
5401 		_spdk_blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id,
5402 				     sizeof(spdk_blob_id),
5403 				     true);
5404 	} else {
5405 		/* ...to blobid invalid and zeroes dev */
5406 		ctx->clone->parent_id = SPDK_BLOBID_INVALID;
5407 		ctx->clone->back_bs_dev = spdk_bs_create_zeroes_dev();
5408 		_spdk_blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true);
5409 	}
5410 
5411 	spdk_blob_sync_md(ctx->clone, _spdk_delete_snapshot_sync_clone_cpl, ctx);
5412 }
5413 
5414 static void
5415 _spdk_delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno)
5416 {
5417 	struct delete_snapshot_ctx *ctx = cb_arg;
5418 
5419 	if (bserrno) {
5420 		SPDK_ERRLOG("Failed to freeze I/O on clone\n");
5421 		ctx->bserrno = bserrno;
5422 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5423 		return;
5424 	}
5425 
5426 	/* Temporarily override md_ro flag for snapshot for MD modification */
5427 	ctx->snapshot_md_ro = ctx->snapshot->md_ro;
5428 	ctx->snapshot->md_ro = false;
5429 
5430 	/* Mark blob as pending for removal for power failure safety, use clone id for recovery */
5431 	ctx->bserrno = _spdk_blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id,
5432 					    sizeof(spdk_blob_id), true);
5433 	if (ctx->bserrno != 0) {
5434 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5435 		return;
5436 	}
5437 
5438 	spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_sync_snapshot_xattr_cpl, ctx);
5439 }
5440 
5441 static void
5442 _spdk_delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno)
5443 {
5444 	struct delete_snapshot_ctx *ctx = cb_arg;
5445 
5446 	if (bserrno) {
5447 		SPDK_ERRLOG("Failed to open clone\n");
5448 		ctx->bserrno = bserrno;
5449 		_spdk_delete_snapshot_cleanup_snapshot(ctx, 0);
5450 		return;
5451 	}
5452 
5453 	ctx->clone = clone;
5454 
5455 	if (clone->locked_operation_in_progress) {
5456 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress on its clone\n");
5457 		ctx->bserrno = -EBUSY;
5458 		spdk_blob_close(ctx->clone, _spdk_delete_snapshot_cleanup_snapshot, ctx);
5459 		return;
5460 	}
5461 
5462 	clone->locked_operation_in_progress = true;
5463 
5464 	_spdk_blob_freeze_io(clone, _spdk_delete_snapshot_freeze_io_cb, ctx);
5465 }
5466 
5467 static void
5468 _spdk_update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx)
5469 {
5470 	struct spdk_blob_list *snapshot_entry = NULL;
5471 	struct spdk_blob_list *clone_entry = NULL;
5472 	struct spdk_blob_list *snapshot_clone_entry = NULL;
5473 
5474 	/* Get snapshot entry for the snapshot we want to remove */
5475 	snapshot_entry = _spdk_bs_get_snapshot_entry(snapshot->bs, snapshot->id);
5476 
5477 	assert(snapshot_entry != NULL);
5478 
5479 	/* Get clone of the snapshot (at this point there can be only one clone) */
5480 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
5481 	assert(snapshot_entry->clone_count == 1);
5482 	assert(clone_entry != NULL);
5483 
5484 	/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
5485 	 * snapshot that we are removing */
5486 	_spdk_blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry,
5487 			&snapshot_clone_entry);
5488 
5489 	spdk_bs_open_blob(snapshot->bs, clone_entry->id, _spdk_delete_snapshot_open_clone_cb, ctx);
5490 }
5491 
5492 static void
5493 _spdk_bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno)
5494 {
5495 	spdk_bs_sequence_t *seq = cb_arg;
5496 	struct spdk_blob_list *snapshot_entry = NULL;
5497 	uint32_t page_num;
5498 
5499 	if (bserrno) {
5500 		SPDK_ERRLOG("Failed to remove blob\n");
5501 		spdk_bs_sequence_finish(seq, bserrno);
5502 		return;
5503 	}
5504 
5505 	/* Remove snapshot from the list */
5506 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
5507 	if (snapshot_entry != NULL) {
5508 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
5509 		free(snapshot_entry);
5510 	}
5511 
5512 	page_num = _spdk_bs_blobid_to_page(blob->id);
5513 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
5514 	blob->state = SPDK_BLOB_STATE_DIRTY;
5515 	blob->active.num_pages = 0;
5516 	_spdk_blob_resize(blob, 0);
5517 
5518 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
5519 }
5520 
5521 static int
5522 _spdk_bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone)
5523 {
5524 	struct spdk_blob_list *snapshot_entry = NULL;
5525 	struct spdk_blob_list *clone_entry = NULL;
5526 	struct spdk_blob *clone = NULL;
5527 	bool has_one_clone = false;
5528 
5529 	/* Check if this is a snapshot with clones */
5530 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
5531 	if (snapshot_entry != NULL) {
5532 		if (snapshot_entry->clone_count > 1) {
5533 			SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n");
5534 			return -EBUSY;
5535 		} else if (snapshot_entry->clone_count == 1) {
5536 			has_one_clone = true;
5537 		}
5538 	}
5539 
5540 	/* Check if someone has this blob open (besides this delete context):
5541 	 * - open_ref = 1 - only this context opened blob, so it is ok to remove it
5542 	 * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot
5543 	 *	and that is ok, because we will update it accordingly */
5544 	if (blob->open_ref <= 2 && has_one_clone) {
5545 		clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
5546 		assert(clone_entry != NULL);
5547 		clone = _spdk_blob_lookup(blob->bs, clone_entry->id);
5548 
5549 		if (blob->open_ref == 2 && clone == NULL) {
5550 			/* Clone is closed and someone else opened this blob */
5551 			SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
5552 			return -EBUSY;
5553 		}
5554 
5555 		*update_clone = true;
5556 		return 0;
5557 	}
5558 
5559 	if (blob->open_ref > 1) {
5560 		SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
5561 		return -EBUSY;
5562 	}
5563 
5564 	assert(has_one_clone == false);
5565 	*update_clone = false;
5566 	return 0;
5567 }
5568 
5569 static void
5570 _spdk_bs_delete_enomem_close_cpl(void *cb_arg, int bserrno)
5571 {
5572 	spdk_bs_sequence_t *seq = cb_arg;
5573 
5574 	spdk_bs_sequence_finish(seq, -ENOMEM);
5575 }
5576 
5577 static void
5578 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
5579 {
5580 	spdk_bs_sequence_t *seq = cb_arg;
5581 	struct delete_snapshot_ctx *ctx;
5582 	bool update_clone = false;
5583 
5584 	if (bserrno != 0) {
5585 		spdk_bs_sequence_finish(seq, bserrno);
5586 		return;
5587 	}
5588 
5589 	_spdk_blob_verify_md_op(blob);
5590 
5591 	ctx = calloc(1, sizeof(*ctx));
5592 	if (ctx == NULL) {
5593 		spdk_blob_close(blob, _spdk_bs_delete_enomem_close_cpl, seq);
5594 		return;
5595 	}
5596 
5597 	ctx->snapshot = blob;
5598 	ctx->cb_fn = _spdk_bs_delete_blob_finish;
5599 	ctx->cb_arg = seq;
5600 
5601 	/* Check if blob can be removed and if it is a snapshot with clone on top of it */
5602 	ctx->bserrno = _spdk_bs_is_blob_deletable(blob, &update_clone);
5603 	if (ctx->bserrno) {
5604 		spdk_blob_close(blob, _spdk_delete_blob_cleanup_finish, ctx);
5605 		return;
5606 	}
5607 
5608 	if (blob->locked_operation_in_progress) {
5609 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress\n");
5610 		ctx->bserrno = -EBUSY;
5611 		spdk_blob_close(blob, _spdk_delete_blob_cleanup_finish, ctx);
5612 		return;
5613 	}
5614 
5615 	blob->locked_operation_in_progress = true;
5616 
5617 	/*
5618 	 * Remove the blob from the blob_store list now, to ensure it does not
5619 	 *  get returned after this point by _spdk_blob_lookup().
5620 	 */
5621 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5622 
5623 	if (update_clone) {
5624 		/* This blob is a snapshot with active clone - update clone first */
5625 		_spdk_update_clone_on_snapshot_deletion(blob, ctx);
5626 	} else {
5627 		/* This blob does not have any clones - just remove it */
5628 		_spdk_bs_blob_list_remove(blob);
5629 		_spdk_bs_delete_blob_finish(seq, blob, 0);
5630 		free(ctx);
5631 	}
5632 }
5633 
5634 void
5635 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5636 		    spdk_blob_op_complete cb_fn, void *cb_arg)
5637 {
5638 	struct spdk_bs_cpl	cpl;
5639 	spdk_bs_sequence_t	*seq;
5640 
5641 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
5642 
5643 	assert(spdk_get_thread() == bs->md_thread);
5644 
5645 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5646 	cpl.u.blob_basic.cb_fn = cb_fn;
5647 	cpl.u.blob_basic.cb_arg = cb_arg;
5648 
5649 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5650 	if (!seq) {
5651 		cb_fn(cb_arg, -ENOMEM);
5652 		return;
5653 	}
5654 
5655 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5656 }
5657 
5658 /* END spdk_bs_delete_blob */
5659 
5660 /* START spdk_bs_open_blob */
5661 
5662 static void
5663 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5664 {
5665 	struct spdk_blob *blob = cb_arg;
5666 
5667 	/* If the blob have crc error, we just return NULL. */
5668 	if (blob == NULL) {
5669 		seq->cpl.u.blob_handle.blob = NULL;
5670 		spdk_bs_sequence_finish(seq, bserrno);
5671 		return;
5672 	}
5673 
5674 	blob->open_ref++;
5675 
5676 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5677 
5678 	spdk_bs_sequence_finish(seq, bserrno);
5679 }
5680 
5681 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5682 			       struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5683 {
5684 	struct spdk_blob		*blob;
5685 	struct spdk_bs_cpl		cpl;
5686 	struct spdk_blob_open_opts	opts_default;
5687 	spdk_bs_sequence_t		*seq;
5688 	uint32_t			page_num;
5689 
5690 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5691 	assert(spdk_get_thread() == bs->md_thread);
5692 
5693 	page_num = _spdk_bs_blobid_to_page(blobid);
5694 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5695 		/* Invalid blobid */
5696 		cb_fn(cb_arg, NULL, -ENOENT);
5697 		return;
5698 	}
5699 
5700 	blob = _spdk_blob_lookup(bs, blobid);
5701 	if (blob) {
5702 		blob->open_ref++;
5703 		cb_fn(cb_arg, blob, 0);
5704 		return;
5705 	}
5706 
5707 	blob = _spdk_blob_alloc(bs, blobid);
5708 	if (!blob) {
5709 		cb_fn(cb_arg, NULL, -ENOMEM);
5710 		return;
5711 	}
5712 
5713 	if (!opts) {
5714 		spdk_blob_open_opts_init(&opts_default);
5715 		opts = &opts_default;
5716 	}
5717 
5718 	blob->clear_method = opts->clear_method;
5719 
5720 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5721 	cpl.u.blob_handle.cb_fn = cb_fn;
5722 	cpl.u.blob_handle.cb_arg = cb_arg;
5723 	cpl.u.blob_handle.blob = blob;
5724 
5725 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5726 	if (!seq) {
5727 		_spdk_blob_free(blob);
5728 		cb_fn(cb_arg, NULL, -ENOMEM);
5729 		return;
5730 	}
5731 
5732 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5733 }
5734 
5735 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5736 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5737 {
5738 	_spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
5739 }
5740 
5741 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
5742 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5743 {
5744 	_spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
5745 }
5746 
5747 /* END spdk_bs_open_blob */
5748 
5749 /* START spdk_blob_set_read_only */
5750 int spdk_blob_set_read_only(struct spdk_blob *blob)
5751 {
5752 	_spdk_blob_verify_md_op(blob);
5753 
5754 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5755 
5756 	blob->state = SPDK_BLOB_STATE_DIRTY;
5757 	return 0;
5758 }
5759 /* END spdk_blob_set_read_only */
5760 
5761 /* START spdk_blob_sync_md */
5762 
5763 static void
5764 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5765 {
5766 	struct spdk_blob *blob = cb_arg;
5767 
5768 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5769 		blob->data_ro = true;
5770 		blob->md_ro = true;
5771 	}
5772 
5773 	spdk_bs_sequence_finish(seq, bserrno);
5774 }
5775 
5776 static void
5777 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5778 {
5779 	struct spdk_bs_cpl	cpl;
5780 	spdk_bs_sequence_t	*seq;
5781 
5782 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5783 	cpl.u.blob_basic.cb_fn = cb_fn;
5784 	cpl.u.blob_basic.cb_arg = cb_arg;
5785 
5786 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5787 	if (!seq) {
5788 		cb_fn(cb_arg, -ENOMEM);
5789 		return;
5790 	}
5791 
5792 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5793 }
5794 
5795 void
5796 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5797 {
5798 	_spdk_blob_verify_md_op(blob);
5799 
5800 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5801 
5802 	if (blob->md_ro) {
5803 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5804 		cb_fn(cb_arg, 0);
5805 		return;
5806 	}
5807 
5808 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5809 }
5810 
5811 /* END spdk_blob_sync_md */
5812 
5813 struct spdk_blob_insert_cluster_ctx {
5814 	struct spdk_thread	*thread;
5815 	struct spdk_blob	*blob;
5816 	uint32_t		cluster_num;	/* cluster index in blob */
5817 	uint32_t		cluster;	/* cluster on disk */
5818 	int			rc;
5819 	spdk_blob_op_complete	cb_fn;
5820 	void			*cb_arg;
5821 };
5822 
5823 static void
5824 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5825 {
5826 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5827 
5828 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5829 	free(ctx);
5830 }
5831 
5832 static void
5833 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5834 {
5835 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5836 
5837 	ctx->rc = bserrno;
5838 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5839 }
5840 
5841 static void
5842 _spdk_blob_insert_cluster_msg(void *arg)
5843 {
5844 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5845 
5846 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5847 	if (ctx->rc != 0) {
5848 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5849 		return;
5850 	}
5851 
5852 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5853 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5854 }
5855 
5856 static void
5857 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5858 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5859 {
5860 	struct spdk_blob_insert_cluster_ctx *ctx;
5861 
5862 	ctx = calloc(1, sizeof(*ctx));
5863 	if (ctx == NULL) {
5864 		cb_fn(cb_arg, -ENOMEM);
5865 		return;
5866 	}
5867 
5868 	ctx->thread = spdk_get_thread();
5869 	ctx->blob = blob;
5870 	ctx->cluster_num = cluster_num;
5871 	ctx->cluster = cluster;
5872 	ctx->cb_fn = cb_fn;
5873 	ctx->cb_arg = cb_arg;
5874 
5875 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5876 }
5877 
5878 /* START spdk_blob_close */
5879 
5880 static void
5881 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5882 {
5883 	struct spdk_blob *blob = cb_arg;
5884 
5885 	if (bserrno == 0) {
5886 		blob->open_ref--;
5887 		if (blob->open_ref == 0) {
5888 			/*
5889 			 * Blobs with active.num_pages == 0 are deleted blobs.
5890 			 *  these blobs are removed from the blob_store list
5891 			 *  when the deletion process starts - so don't try to
5892 			 *  remove them again.
5893 			 */
5894 			if (blob->active.num_pages > 0) {
5895 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5896 			}
5897 			_spdk_blob_free(blob);
5898 		}
5899 	}
5900 
5901 	spdk_bs_sequence_finish(seq, bserrno);
5902 }
5903 
5904 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5905 {
5906 	struct spdk_bs_cpl	cpl;
5907 	spdk_bs_sequence_t	*seq;
5908 
5909 	_spdk_blob_verify_md_op(blob);
5910 
5911 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5912 
5913 	if (blob->open_ref == 0) {
5914 		cb_fn(cb_arg, -EBADF);
5915 		return;
5916 	}
5917 
5918 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5919 	cpl.u.blob_basic.cb_fn = cb_fn;
5920 	cpl.u.blob_basic.cb_arg = cb_arg;
5921 
5922 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5923 	if (!seq) {
5924 		cb_fn(cb_arg, -ENOMEM);
5925 		return;
5926 	}
5927 
5928 	/* Sync metadata */
5929 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5930 }
5931 
5932 /* END spdk_blob_close */
5933 
5934 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5935 {
5936 	return spdk_get_io_channel(bs);
5937 }
5938 
5939 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5940 {
5941 	spdk_put_io_channel(channel);
5942 }
5943 
5944 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5945 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5946 {
5947 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5948 				     SPDK_BLOB_UNMAP);
5949 }
5950 
5951 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5952 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5953 {
5954 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5955 				     SPDK_BLOB_WRITE_ZEROES);
5956 }
5957 
5958 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5959 			void *payload, uint64_t offset, uint64_t length,
5960 			spdk_blob_op_complete cb_fn, void *cb_arg)
5961 {
5962 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5963 				     SPDK_BLOB_WRITE);
5964 }
5965 
5966 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5967 		       void *payload, uint64_t offset, uint64_t length,
5968 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5969 {
5970 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5971 				     SPDK_BLOB_READ);
5972 }
5973 
5974 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5975 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5976 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5977 {
5978 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5979 }
5980 
5981 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5982 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5983 			spdk_blob_op_complete cb_fn, void *cb_arg)
5984 {
5985 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5986 }
5987 
5988 struct spdk_bs_iter_ctx {
5989 	int64_t page_num;
5990 	struct spdk_blob_store *bs;
5991 
5992 	spdk_blob_op_with_handle_complete cb_fn;
5993 	void *cb_arg;
5994 };
5995 
5996 static void
5997 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5998 {
5999 	struct spdk_bs_iter_ctx *ctx = cb_arg;
6000 	struct spdk_blob_store *bs = ctx->bs;
6001 	spdk_blob_id id;
6002 
6003 	if (bserrno == 0) {
6004 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
6005 		free(ctx);
6006 		return;
6007 	}
6008 
6009 	ctx->page_num++;
6010 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
6011 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
6012 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
6013 		free(ctx);
6014 		return;
6015 	}
6016 
6017 	id = _spdk_bs_page_to_blobid(ctx->page_num);
6018 
6019 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
6020 }
6021 
6022 void
6023 spdk_bs_iter_first(struct spdk_blob_store *bs,
6024 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6025 {
6026 	struct spdk_bs_iter_ctx *ctx;
6027 
6028 	ctx = calloc(1, sizeof(*ctx));
6029 	if (!ctx) {
6030 		cb_fn(cb_arg, NULL, -ENOMEM);
6031 		return;
6032 	}
6033 
6034 	ctx->page_num = -1;
6035 	ctx->bs = bs;
6036 	ctx->cb_fn = cb_fn;
6037 	ctx->cb_arg = cb_arg;
6038 
6039 	_spdk_bs_iter_cpl(ctx, NULL, -1);
6040 }
6041 
6042 static void
6043 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
6044 {
6045 	struct spdk_bs_iter_ctx *ctx = cb_arg;
6046 
6047 	_spdk_bs_iter_cpl(ctx, NULL, -1);
6048 }
6049 
6050 void
6051 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
6052 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6053 {
6054 	struct spdk_bs_iter_ctx *ctx;
6055 
6056 	assert(blob != NULL);
6057 
6058 	ctx = calloc(1, sizeof(*ctx));
6059 	if (!ctx) {
6060 		cb_fn(cb_arg, NULL, -ENOMEM);
6061 		return;
6062 	}
6063 
6064 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
6065 	ctx->bs = bs;
6066 	ctx->cb_fn = cb_fn;
6067 	ctx->cb_arg = cb_arg;
6068 
6069 	/* Close the existing blob */
6070 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
6071 }
6072 
6073 static int
6074 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
6075 		     uint16_t value_len, bool internal)
6076 {
6077 	struct spdk_xattr_tailq *xattrs;
6078 	struct spdk_xattr	*xattr;
6079 	size_t			desc_size;
6080 
6081 	_spdk_blob_verify_md_op(blob);
6082 
6083 	if (blob->md_ro) {
6084 		return -EPERM;
6085 	}
6086 
6087 	desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len;
6088 	if (desc_size > SPDK_BS_MAX_DESC_SIZE) {
6089 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Xattr '%s' of size %ld does not fix into single page %ld\n", name,
6090 			      desc_size, SPDK_BS_MAX_DESC_SIZE);
6091 		return -ENOMEM;
6092 	}
6093 
6094 	if (internal) {
6095 		xattrs = &blob->xattrs_internal;
6096 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
6097 	} else {
6098 		xattrs = &blob->xattrs;
6099 	}
6100 
6101 	TAILQ_FOREACH(xattr, xattrs, link) {
6102 		if (!strcmp(name, xattr->name)) {
6103 			free(xattr->value);
6104 			xattr->value_len = value_len;
6105 			xattr->value = malloc(value_len);
6106 			memcpy(xattr->value, value, value_len);
6107 
6108 			blob->state = SPDK_BLOB_STATE_DIRTY;
6109 
6110 			return 0;
6111 		}
6112 	}
6113 
6114 	xattr = calloc(1, sizeof(*xattr));
6115 	if (!xattr) {
6116 		return -ENOMEM;
6117 	}
6118 	xattr->name = strdup(name);
6119 	xattr->value_len = value_len;
6120 	xattr->value = malloc(value_len);
6121 	memcpy(xattr->value, value, value_len);
6122 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
6123 
6124 	blob->state = SPDK_BLOB_STATE_DIRTY;
6125 
6126 	return 0;
6127 }
6128 
6129 int
6130 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
6131 		    uint16_t value_len)
6132 {
6133 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
6134 }
6135 
6136 static int
6137 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
6138 {
6139 	struct spdk_xattr_tailq *xattrs;
6140 	struct spdk_xattr	*xattr;
6141 
6142 	_spdk_blob_verify_md_op(blob);
6143 
6144 	if (blob->md_ro) {
6145 		return -EPERM;
6146 	}
6147 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
6148 
6149 	TAILQ_FOREACH(xattr, xattrs, link) {
6150 		if (!strcmp(name, xattr->name)) {
6151 			TAILQ_REMOVE(xattrs, xattr, link);
6152 			free(xattr->value);
6153 			free(xattr->name);
6154 			free(xattr);
6155 
6156 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
6157 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
6158 			}
6159 			blob->state = SPDK_BLOB_STATE_DIRTY;
6160 
6161 			return 0;
6162 		}
6163 	}
6164 
6165 	return -ENOENT;
6166 }
6167 
6168 int
6169 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
6170 {
6171 	return _spdk_blob_remove_xattr(blob, name, false);
6172 }
6173 
6174 static int
6175 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
6176 			   const void **value, size_t *value_len, bool internal)
6177 {
6178 	struct spdk_xattr	*xattr;
6179 	struct spdk_xattr_tailq *xattrs;
6180 
6181 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
6182 
6183 	TAILQ_FOREACH(xattr, xattrs, link) {
6184 		if (!strcmp(name, xattr->name)) {
6185 			*value = xattr->value;
6186 			*value_len = xattr->value_len;
6187 			return 0;
6188 		}
6189 	}
6190 	return -ENOENT;
6191 }
6192 
6193 int
6194 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
6195 			  const void **value, size_t *value_len)
6196 {
6197 	_spdk_blob_verify_md_op(blob);
6198 
6199 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
6200 }
6201 
6202 struct spdk_xattr_names {
6203 	uint32_t	count;
6204 	const char	*names[0];
6205 };
6206 
6207 static int
6208 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
6209 {
6210 	struct spdk_xattr	*xattr;
6211 	int			count = 0;
6212 
6213 	TAILQ_FOREACH(xattr, xattrs, link) {
6214 		count++;
6215 	}
6216 
6217 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
6218 	if (*names == NULL) {
6219 		return -ENOMEM;
6220 	}
6221 
6222 	TAILQ_FOREACH(xattr, xattrs, link) {
6223 		(*names)->names[(*names)->count++] = xattr->name;
6224 	}
6225 
6226 	return 0;
6227 }
6228 
6229 int
6230 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
6231 {
6232 	_spdk_blob_verify_md_op(blob);
6233 
6234 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
6235 }
6236 
6237 uint32_t
6238 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
6239 {
6240 	assert(names != NULL);
6241 
6242 	return names->count;
6243 }
6244 
6245 const char *
6246 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
6247 {
6248 	if (index >= names->count) {
6249 		return NULL;
6250 	}
6251 
6252 	return names->names[index];
6253 }
6254 
6255 void
6256 spdk_xattr_names_free(struct spdk_xattr_names *names)
6257 {
6258 	free(names);
6259 }
6260 
6261 struct spdk_bs_type
6262 spdk_bs_get_bstype(struct spdk_blob_store *bs)
6263 {
6264 	return bs->bstype;
6265 }
6266 
6267 void
6268 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
6269 {
6270 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
6271 }
6272 
6273 bool
6274 spdk_blob_is_read_only(struct spdk_blob *blob)
6275 {
6276 	assert(blob != NULL);
6277 	return (blob->data_ro || blob->md_ro);
6278 }
6279 
6280 bool
6281 spdk_blob_is_snapshot(struct spdk_blob *blob)
6282 {
6283 	struct spdk_blob_list *snapshot_entry;
6284 
6285 	assert(blob != NULL);
6286 
6287 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
6288 	if (snapshot_entry == NULL) {
6289 		return false;
6290 	}
6291 
6292 	return true;
6293 }
6294 
6295 bool
6296 spdk_blob_is_clone(struct spdk_blob *blob)
6297 {
6298 	assert(blob != NULL);
6299 
6300 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
6301 		assert(spdk_blob_is_thin_provisioned(blob));
6302 		return true;
6303 	}
6304 
6305 	return false;
6306 }
6307 
6308 bool
6309 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
6310 {
6311 	assert(blob != NULL);
6312 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
6313 }
6314 
6315 spdk_blob_id
6316 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
6317 {
6318 	struct spdk_blob_list *snapshot_entry = NULL;
6319 	struct spdk_blob_list *clone_entry = NULL;
6320 
6321 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
6322 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
6323 			if (clone_entry->id == blob_id) {
6324 				return snapshot_entry->id;
6325 			}
6326 		}
6327 	}
6328 
6329 	return SPDK_BLOBID_INVALID;
6330 }
6331 
6332 int
6333 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
6334 		     size_t *count)
6335 {
6336 	struct spdk_blob_list *snapshot_entry, *clone_entry;
6337 	size_t n;
6338 
6339 	snapshot_entry = _spdk_bs_get_snapshot_entry(bs, blobid);
6340 	if (snapshot_entry == NULL) {
6341 		*count = 0;
6342 		return 0;
6343 	}
6344 
6345 	if (ids == NULL || *count < snapshot_entry->clone_count) {
6346 		*count = snapshot_entry->clone_count;
6347 		return -ENOMEM;
6348 	}
6349 	*count = snapshot_entry->clone_count;
6350 
6351 	n = 0;
6352 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
6353 		ids[n++] = clone_entry->id;
6354 	}
6355 
6356 	return 0;
6357 }
6358 
6359 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
6360