xref: /spdk/lib/blob/blobstore.c (revision c4d9daeb7bf491bc0eb6e8d417b75d44773cb009)
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 void
736 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
737 			   uint8_t *buf, size_t *buf_sz)
738 {
739 	struct spdk_blob_md_descriptor_flags *desc;
740 
741 	/*
742 	 * Flags get serialized first, so we should always have room for the flags
743 	 *  descriptor.
744 	 */
745 	assert(*buf_sz >= sizeof(*desc));
746 
747 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
748 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
749 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
750 	desc->invalid_flags = blob->invalid_flags;
751 	desc->data_ro_flags = blob->data_ro_flags;
752 	desc->md_ro_flags = blob->md_ro_flags;
753 
754 	*buf_sz -= sizeof(*desc);
755 }
756 
757 static int
758 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
759 			    const struct spdk_xattr_tailq *xattrs, bool internal,
760 			    struct spdk_blob_md_page **pages,
761 			    struct spdk_blob_md_page *cur_page,
762 			    uint32_t *page_count, uint8_t **buf,
763 			    size_t *remaining_sz)
764 {
765 	const struct spdk_xattr	*xattr;
766 	int	rc;
767 
768 	TAILQ_FOREACH(xattr, xattrs, link) {
769 		size_t required_sz = 0;
770 
771 		rc = _spdk_blob_serialize_xattr(xattr,
772 						*buf, *remaining_sz,
773 						&required_sz, internal);
774 		if (rc < 0) {
775 			/* Need to add a new page to the chain */
776 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
777 							   &cur_page);
778 			if (rc < 0) {
779 				spdk_free(*pages);
780 				*pages = NULL;
781 				*page_count = 0;
782 				return rc;
783 			}
784 
785 			*buf = (uint8_t *)cur_page->descriptors;
786 			*remaining_sz = sizeof(cur_page->descriptors);
787 
788 			/* Try again */
789 			required_sz = 0;
790 			rc = _spdk_blob_serialize_xattr(xattr,
791 							*buf, *remaining_sz,
792 							&required_sz, internal);
793 
794 			if (rc < 0) {
795 				spdk_free(*pages);
796 				*pages = NULL;
797 				*page_count = 0;
798 				return rc;
799 			}
800 		}
801 
802 		*remaining_sz -= required_sz;
803 		*buf += required_sz;
804 	}
805 
806 	return 0;
807 }
808 
809 static int
810 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
811 		     uint32_t *page_count)
812 {
813 	struct spdk_blob_md_page		*cur_page;
814 	int					rc;
815 	uint8_t					*buf;
816 	size_t					remaining_sz;
817 	uint64_t				last_cluster;
818 
819 	assert(pages != NULL);
820 	assert(page_count != NULL);
821 	assert(blob != NULL);
822 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
823 
824 	*pages = NULL;
825 	*page_count = 0;
826 
827 	/* A blob always has at least 1 page, even if it has no descriptors */
828 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
829 	if (rc < 0) {
830 		return rc;
831 	}
832 
833 	buf = (uint8_t *)cur_page->descriptors;
834 	remaining_sz = sizeof(cur_page->descriptors);
835 
836 	/* Serialize flags */
837 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
838 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
839 
840 	/* Serialize xattrs */
841 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
842 					 pages, cur_page, page_count, &buf, &remaining_sz);
843 	if (rc < 0) {
844 		return rc;
845 	}
846 
847 	/* Serialize internal xattrs */
848 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
849 					 pages, cur_page, page_count, &buf, &remaining_sz);
850 	if (rc < 0) {
851 		return rc;
852 	}
853 
854 	/* Serialize extents */
855 	last_cluster = 0;
856 	while (last_cluster < blob->active.num_clusters) {
857 		_spdk_blob_serialize_extent(blob, last_cluster, &last_cluster,
858 					    buf, remaining_sz);
859 
860 		if (last_cluster == blob->active.num_clusters) {
861 			break;
862 		}
863 
864 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
865 						   &cur_page);
866 		if (rc < 0) {
867 			return rc;
868 		}
869 
870 		buf = (uint8_t *)cur_page->descriptors;
871 		remaining_sz = sizeof(cur_page->descriptors);
872 	}
873 
874 	return 0;
875 }
876 
877 struct spdk_blob_load_ctx {
878 	struct spdk_blob		*blob;
879 
880 	struct spdk_blob_md_page	*pages;
881 	uint32_t			num_pages;
882 	spdk_bs_sequence_t	        *seq;
883 
884 	spdk_bs_sequence_cpl		cb_fn;
885 	void				*cb_arg;
886 };
887 
888 static uint32_t
889 _spdk_blob_md_page_calc_crc(void *page)
890 {
891 	uint32_t		crc;
892 
893 	crc = BLOB_CRC32C_INITIAL;
894 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
895 	crc ^= BLOB_CRC32C_INITIAL;
896 
897 	return crc;
898 
899 }
900 
901 static void
902 _spdk_blob_load_final(void *cb_arg, int bserrno)
903 {
904 	struct spdk_blob_load_ctx	*ctx = cb_arg;
905 	struct spdk_blob		*blob = ctx->blob;
906 
907 	_spdk_blob_mark_clean(blob);
908 
909 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
910 
911 	/* Free the memory */
912 	spdk_free(ctx->pages);
913 	free(ctx);
914 }
915 
916 static void
917 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
918 {
919 	struct spdk_blob_load_ctx	*ctx = cb_arg;
920 	struct spdk_blob		*blob = ctx->blob;
921 
922 	if (bserrno != 0) {
923 		goto error;
924 	}
925 
926 	blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
927 
928 	if (blob->back_bs_dev == NULL) {
929 		bserrno = -ENOMEM;
930 		goto error;
931 	}
932 
933 	_spdk_blob_load_final(ctx, bserrno);
934 	return;
935 
936 error:
937 	SPDK_ERRLOG("Snapshot fail\n");
938 	_spdk_blob_free(blob);
939 	ctx->cb_fn(ctx->seq, NULL, bserrno);
940 	spdk_free(ctx->pages);
941 	free(ctx);
942 }
943 
944 static void
945 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
946 {
947 	struct spdk_blob_load_ctx	*ctx = cb_arg;
948 	struct spdk_blob		*blob = ctx->blob;
949 	struct spdk_blob_md_page	*page;
950 	const void			*value;
951 	size_t				len;
952 	int				rc;
953 	uint32_t			crc;
954 
955 	if (bserrno) {
956 		SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno);
957 		_spdk_blob_free(blob);
958 		ctx->cb_fn(seq, NULL, bserrno);
959 		spdk_free(ctx->pages);
960 		free(ctx);
961 		return;
962 	}
963 
964 	page = &ctx->pages[ctx->num_pages - 1];
965 	crc = _spdk_blob_md_page_calc_crc(page);
966 	if (crc != page->crc) {
967 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
968 		_spdk_blob_free(blob);
969 		ctx->cb_fn(seq, NULL, -EINVAL);
970 		spdk_free(ctx->pages);
971 		free(ctx);
972 		return;
973 	}
974 
975 	if (page->next != SPDK_INVALID_MD_PAGE) {
976 		uint32_t next_page = page->next;
977 		uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page);
978 
979 
980 		assert(next_lba < (blob->bs->md_start + blob->bs->md_len));
981 
982 		/* Read the next page */
983 		ctx->num_pages++;
984 		ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
985 					  sizeof(*page));
986 		if (ctx->pages == NULL) {
987 			ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM);
988 			free(ctx);
989 			return;
990 		}
991 
992 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
993 					  next_lba,
994 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
995 					  _spdk_blob_load_cpl, ctx);
996 		return;
997 	}
998 
999 	/* Parse the pages */
1000 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
1001 	if (rc) {
1002 		_spdk_blob_free(blob);
1003 		ctx->cb_fn(seq, NULL, rc);
1004 		spdk_free(ctx->pages);
1005 		free(ctx);
1006 		return;
1007 	}
1008 	ctx->seq = seq;
1009 
1010 
1011 	if (spdk_blob_is_thin_provisioned(blob)) {
1012 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
1013 		if (rc == 0) {
1014 			if (len != sizeof(spdk_blob_id)) {
1015 				_spdk_blob_free(blob);
1016 				ctx->cb_fn(seq, NULL, -EINVAL);
1017 				spdk_free(ctx->pages);
1018 				free(ctx);
1019 				return;
1020 			}
1021 			/* open snapshot blob and continue in the callback function */
1022 			blob->parent_id = *(spdk_blob_id *)value;
1023 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1024 					  _spdk_blob_load_snapshot_cpl, ctx);
1025 			return;
1026 		} else {
1027 			/* add zeroes_dev for thin provisioned blob */
1028 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1029 		}
1030 	} else {
1031 		/* standard blob */
1032 		blob->back_bs_dev = NULL;
1033 	}
1034 	_spdk_blob_load_final(ctx, bserrno);
1035 }
1036 
1037 /* Load a blob from disk given a blobid */
1038 static void
1039 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1040 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1041 {
1042 	struct spdk_blob_load_ctx *ctx;
1043 	struct spdk_blob_store *bs;
1044 	uint32_t page_num;
1045 	uint64_t lba;
1046 
1047 	_spdk_blob_verify_md_op(blob);
1048 
1049 	bs = blob->bs;
1050 
1051 	ctx = calloc(1, sizeof(*ctx));
1052 	if (!ctx) {
1053 		cb_fn(seq, cb_arg, -ENOMEM);
1054 		return;
1055 	}
1056 
1057 	ctx->blob = blob;
1058 	ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE);
1059 	if (!ctx->pages) {
1060 		free(ctx);
1061 		cb_fn(seq, cb_arg, -ENOMEM);
1062 		return;
1063 	}
1064 	ctx->num_pages = 1;
1065 	ctx->cb_fn = cb_fn;
1066 	ctx->cb_arg = cb_arg;
1067 
1068 	page_num = _spdk_bs_blobid_to_page(blob->id);
1069 	lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num);
1070 
1071 	blob->state = SPDK_BLOB_STATE_LOADING;
1072 
1073 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1074 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1075 				  _spdk_blob_load_cpl, ctx);
1076 }
1077 
1078 struct spdk_blob_persist_ctx {
1079 	struct spdk_blob		*blob;
1080 
1081 	struct spdk_bs_super_block	*super;
1082 
1083 	struct spdk_blob_md_page	*pages;
1084 
1085 	uint64_t			idx;
1086 
1087 	spdk_bs_sequence_t		*seq;
1088 	spdk_bs_sequence_cpl		cb_fn;
1089 	void				*cb_arg;
1090 };
1091 
1092 static void
1093 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1094 			uint32_t lba_count)
1095 {
1096 	if (ctx->blob->clear_method == BLOB_CLEAR_WITH_DEFAULT ||
1097 	    ctx->blob->clear_method == BLOB_CLEAR_WITH_UNMAP) {
1098 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1099 	} else if (ctx->blob->clear_method == BLOB_CLEAR_WITH_WRITE_ZEROES) {
1100 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1101 	}
1102 }
1103 
1104 static void
1105 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1106 {
1107 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1108 	struct spdk_blob		*blob = ctx->blob;
1109 
1110 	if (bserrno == 0) {
1111 		_spdk_blob_mark_clean(blob);
1112 	}
1113 
1114 	/* Call user callback */
1115 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1116 
1117 	/* Free the memory */
1118 	spdk_free(ctx->pages);
1119 	free(ctx);
1120 }
1121 
1122 static void
1123 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1124 {
1125 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1126 	struct spdk_blob		*blob = ctx->blob;
1127 	struct spdk_blob_store		*bs = blob->bs;
1128 	void				*tmp;
1129 	size_t				i;
1130 
1131 	/* Release all clusters that were truncated */
1132 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1133 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1134 
1135 		/* Nothing to release if it was not allocated */
1136 		if (blob->active.clusters[i] != 0) {
1137 			_spdk_bs_release_cluster(bs, cluster_num);
1138 		}
1139 	}
1140 
1141 	if (blob->active.num_clusters == 0) {
1142 		free(blob->active.clusters);
1143 		blob->active.clusters = NULL;
1144 		blob->active.cluster_array_size = 0;
1145 	} else if (blob->active.num_clusters != blob->active.cluster_array_size) {
1146 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters);
1147 		assert(tmp != NULL);
1148 		blob->active.clusters = tmp;
1149 		blob->active.cluster_array_size = blob->active.num_clusters;
1150 	}
1151 
1152 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1153 }
1154 
1155 static void
1156 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1157 {
1158 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1159 	struct spdk_blob		*blob = ctx->blob;
1160 	struct spdk_blob_store		*bs = blob->bs;
1161 	spdk_bs_batch_t			*batch;
1162 	size_t				i;
1163 	uint64_t			lba;
1164 	uint32_t			lba_count;
1165 
1166 	/* Clusters don't move around in blobs. The list shrinks or grows
1167 	 * at the end, but no changes ever occur in the middle of the list.
1168 	 */
1169 
1170 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx);
1171 
1172 	/* Clear all clusters that were truncated */
1173 	lba = 0;
1174 	lba_count = 0;
1175 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1176 		uint64_t next_lba = blob->active.clusters[i];
1177 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1178 
1179 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1180 			/* This cluster is contiguous with the previous one. */
1181 			lba_count += next_lba_count;
1182 			continue;
1183 		}
1184 
1185 		/* This cluster is not contiguous with the previous one. */
1186 
1187 		/* If a run of LBAs previously existing, clear them now */
1188 		if (lba_count > 0) {
1189 			spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1190 		}
1191 
1192 		/* Start building the next batch */
1193 		lba = next_lba;
1194 		if (next_lba > 0) {
1195 			lba_count = next_lba_count;
1196 		} else {
1197 			lba_count = 0;
1198 		}
1199 	}
1200 
1201 	/* If we ended with a contiguous set of LBAs, clear them now */
1202 	if (lba_count > 0) {
1203 		spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1204 	}
1205 
1206 	spdk_bs_batch_close(batch);
1207 }
1208 
1209 static void
1210 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1211 {
1212 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1213 	struct spdk_blob		*blob = ctx->blob;
1214 	struct spdk_blob_store		*bs = blob->bs;
1215 	size_t				i;
1216 
1217 	/* This loop starts at 1 because the first page is special and handled
1218 	 * below. The pages (except the first) are never written in place,
1219 	 * so any pages in the clean list must be zeroed.
1220 	 */
1221 	for (i = 1; i < blob->clean.num_pages; i++) {
1222 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1223 	}
1224 
1225 	if (blob->active.num_pages == 0) {
1226 		uint32_t page_num;
1227 
1228 		page_num = _spdk_bs_blobid_to_page(blob->id);
1229 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1230 	}
1231 
1232 	/* Move on to clearing clusters */
1233 	_spdk_blob_persist_clear_clusters(seq, ctx, 0);
1234 }
1235 
1236 static void
1237 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1238 {
1239 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1240 	struct spdk_blob		*blob = ctx->blob;
1241 	struct spdk_blob_store		*bs = blob->bs;
1242 	uint64_t			lba;
1243 	uint32_t			lba_count;
1244 	spdk_bs_batch_t			*batch;
1245 	size_t				i;
1246 
1247 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1248 
1249 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1250 
1251 	/* This loop starts at 1 because the first page is special and handled
1252 	 * below. The pages (except the first) are never written in place,
1253 	 * so any pages in the clean list must be zeroed.
1254 	 */
1255 	for (i = 1; i < blob->clean.num_pages; i++) {
1256 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);
1257 
1258 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1259 	}
1260 
1261 	/* The first page will only be zeroed if this is a delete. */
1262 	if (blob->active.num_pages == 0) {
1263 		uint32_t page_num;
1264 
1265 		/* The first page in the metadata goes where the blobid indicates */
1266 		page_num = _spdk_bs_blobid_to_page(blob->id);
1267 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);
1268 
1269 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1270 	}
1271 
1272 	spdk_bs_batch_close(batch);
1273 }
1274 
1275 static void
1276 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1277 {
1278 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1279 	struct spdk_blob		*blob = ctx->blob;
1280 	struct spdk_blob_store		*bs = blob->bs;
1281 	uint64_t			lba;
1282 	uint32_t			lba_count;
1283 	struct spdk_blob_md_page	*page;
1284 
1285 	if (blob->active.num_pages == 0) {
1286 		/* Move on to the next step */
1287 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1288 		return;
1289 	}
1290 
1291 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1292 
1293 	page = &ctx->pages[0];
1294 	/* The first page in the metadata goes where the blobid indicates */
1295 	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));
1296 
1297 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1298 				   _spdk_blob_persist_zero_pages, ctx);
1299 }
1300 
1301 static void
1302 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1303 {
1304 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1305 	struct spdk_blob		*blob = ctx->blob;
1306 	struct spdk_blob_store		*bs = blob->bs;
1307 	uint64_t			lba;
1308 	uint32_t			lba_count;
1309 	struct spdk_blob_md_page	*page;
1310 	spdk_bs_batch_t			*batch;
1311 	size_t				i;
1312 
1313 	/* Clusters don't move around in blobs. The list shrinks or grows
1314 	 * at the end, but no changes ever occur in the middle of the list.
1315 	 */
1316 
1317 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1318 
1319 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1320 
1321 	/* This starts at 1. The root page is not written until
1322 	 * all of the others are finished
1323 	 */
1324 	for (i = 1; i < blob->active.num_pages; i++) {
1325 		page = &ctx->pages[i];
1326 		assert(page->sequence_num == i);
1327 
1328 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]);
1329 
1330 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1331 	}
1332 
1333 	spdk_bs_batch_close(batch);
1334 }
1335 
1336 static int
1337 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1338 {
1339 	uint64_t	i;
1340 	uint64_t	*tmp;
1341 	uint64_t	lfc; /* lowest free cluster */
1342 	uint64_t	num_clusters;
1343 	struct spdk_blob_store *bs;
1344 
1345 	bs = blob->bs;
1346 
1347 	_spdk_blob_verify_md_op(blob);
1348 
1349 	if (blob->active.num_clusters == sz) {
1350 		return 0;
1351 	}
1352 
1353 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1354 		/* If this blob was resized to be larger, then smaller, then
1355 		 * larger without syncing, then the cluster array already
1356 		 * contains spare assigned clusters we can use.
1357 		 */
1358 		num_clusters = spdk_min(blob->active.cluster_array_size,
1359 					sz);
1360 	} else {
1361 		num_clusters = blob->active.num_clusters;
1362 	}
1363 
1364 	/* Do two passes - one to verify that we can obtain enough clusters
1365 	 * and another to actually claim them.
1366 	 */
1367 
1368 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1369 		lfc = 0;
1370 		for (i = num_clusters; i < sz; i++) {
1371 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1372 			if (lfc == UINT32_MAX) {
1373 				/* No more free clusters. Cannot satisfy the request */
1374 				return -ENOSPC;
1375 			}
1376 			lfc++;
1377 		}
1378 	}
1379 
1380 	if (sz > num_clusters) {
1381 		/* Expand the cluster array if necessary.
1382 		 * We only shrink the array when persisting.
1383 		 */
1384 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz);
1385 		if (sz > 0 && tmp == NULL) {
1386 			return -ENOMEM;
1387 		}
1388 		memset(tmp + blob->active.cluster_array_size, 0,
1389 		       sizeof(uint64_t) * (sz - blob->active.cluster_array_size));
1390 		blob->active.clusters = tmp;
1391 		blob->active.cluster_array_size = sz;
1392 	}
1393 
1394 	blob->state = SPDK_BLOB_STATE_DIRTY;
1395 
1396 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1397 		lfc = 0;
1398 		for (i = num_clusters; i < sz; i++) {
1399 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1400 			lfc++;
1401 		}
1402 	}
1403 
1404 	blob->active.num_clusters = sz;
1405 
1406 	return 0;
1407 }
1408 
1409 static void
1410 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1411 {
1412 	spdk_bs_sequence_t *seq = ctx->seq;
1413 	struct spdk_blob *blob = ctx->blob;
1414 	struct spdk_blob_store *bs = blob->bs;
1415 	uint64_t i;
1416 	uint32_t page_num;
1417 	void *tmp;
1418 	int rc;
1419 
1420 	if (blob->active.num_pages == 0) {
1421 		/* This is the signal that the blob should be deleted.
1422 		 * Immediately jump to the clean up routine. */
1423 		assert(blob->clean.num_pages > 0);
1424 		ctx->idx = blob->clean.num_pages - 1;
1425 		blob->state = SPDK_BLOB_STATE_CLEAN;
1426 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1427 		return;
1428 
1429 	}
1430 
1431 	/* Generate the new metadata */
1432 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1433 	if (rc < 0) {
1434 		_spdk_blob_persist_complete(seq, ctx, rc);
1435 		return;
1436 	}
1437 
1438 	assert(blob->active.num_pages >= 1);
1439 
1440 	/* Resize the cache of page indices */
1441 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1442 	if (!tmp) {
1443 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1444 		return;
1445 	}
1446 	blob->active.pages = tmp;
1447 
1448 	/* Assign this metadata to pages. This requires two passes -
1449 	 * one to verify that there are enough pages and a second
1450 	 * to actually claim them. */
1451 	page_num = 0;
1452 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1453 	for (i = 1; i < blob->active.num_pages; i++) {
1454 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1455 		if (page_num == UINT32_MAX) {
1456 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1457 			return;
1458 		}
1459 		page_num++;
1460 	}
1461 
1462 	page_num = 0;
1463 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1464 	for (i = 1; i < blob->active.num_pages; i++) {
1465 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1466 		ctx->pages[i - 1].next = page_num;
1467 		/* Now that previous metadata page is complete, calculate the crc for it. */
1468 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1469 		blob->active.pages[i] = page_num;
1470 		spdk_bit_array_set(bs->used_md_pages, page_num);
1471 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1472 		page_num++;
1473 	}
1474 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1475 	/* Start writing the metadata from last page to first */
1476 	ctx->idx = blob->active.num_pages - 1;
1477 	blob->state = SPDK_BLOB_STATE_CLEAN;
1478 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1479 }
1480 
1481 static void
1482 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1483 {
1484 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1485 
1486 	ctx->blob->bs->clean = 0;
1487 
1488 	spdk_free(ctx->super);
1489 
1490 	_spdk_blob_persist_start(ctx);
1491 }
1492 
1493 static void
1494 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1495 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1496 
1497 
1498 static void
1499 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1500 {
1501 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1502 
1503 	ctx->super->clean = 0;
1504 	if (ctx->super->size == 0) {
1505 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1506 	}
1507 
1508 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1509 }
1510 
1511 
1512 /* Write a blob to disk */
1513 static void
1514 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1515 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1516 {
1517 	struct spdk_blob_persist_ctx *ctx;
1518 
1519 	_spdk_blob_verify_md_op(blob);
1520 
1521 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1522 		cb_fn(seq, cb_arg, 0);
1523 		return;
1524 	}
1525 
1526 	ctx = calloc(1, sizeof(*ctx));
1527 	if (!ctx) {
1528 		cb_fn(seq, cb_arg, -ENOMEM);
1529 		return;
1530 	}
1531 	ctx->blob = blob;
1532 	ctx->seq = seq;
1533 	ctx->cb_fn = cb_fn;
1534 	ctx->cb_arg = cb_arg;
1535 
1536 	if (blob->bs->clean) {
1537 		ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
1538 					  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1539 		if (!ctx->super) {
1540 			cb_fn(seq, cb_arg, -ENOMEM);
1541 			free(ctx);
1542 			return;
1543 		}
1544 
1545 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1546 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1547 					  _spdk_blob_persist_dirty, ctx);
1548 	} else {
1549 		_spdk_blob_persist_start(ctx);
1550 	}
1551 }
1552 
1553 struct spdk_blob_copy_cluster_ctx {
1554 	struct spdk_blob *blob;
1555 	uint8_t *buf;
1556 	uint64_t page;
1557 	uint64_t new_cluster;
1558 	spdk_bs_sequence_t *seq;
1559 };
1560 
1561 static void
1562 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1563 {
1564 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1565 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1566 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1567 	spdk_bs_user_op_t *op;
1568 
1569 	TAILQ_INIT(&requests);
1570 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1571 
1572 	while (!TAILQ_EMPTY(&requests)) {
1573 		op = TAILQ_FIRST(&requests);
1574 		TAILQ_REMOVE(&requests, op, link);
1575 		if (bserrno == 0) {
1576 			spdk_bs_user_op_execute(op);
1577 		} else {
1578 			spdk_bs_user_op_abort(op);
1579 		}
1580 	}
1581 
1582 	spdk_free(ctx->buf);
1583 	free(ctx);
1584 }
1585 
1586 static void
1587 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1588 {
1589 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1590 
1591 	if (bserrno) {
1592 		if (bserrno == -EEXIST) {
1593 			/* The metadata insert failed because another thread
1594 			 * allocated the cluster first. Free our cluster
1595 			 * but continue without error. */
1596 			bserrno = 0;
1597 		}
1598 		_spdk_bs_release_cluster(ctx->blob->bs, ctx->new_cluster);
1599 	}
1600 
1601 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1602 }
1603 
1604 static void
1605 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1606 {
1607 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1608 	uint32_t cluster_number;
1609 
1610 	if (bserrno) {
1611 		/* The write failed, so jump to the final completion handler */
1612 		spdk_bs_sequence_finish(seq, bserrno);
1613 		return;
1614 	}
1615 
1616 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1617 
1618 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1619 					       _spdk_blob_insert_cluster_cpl, ctx);
1620 }
1621 
1622 static void
1623 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1624 {
1625 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1626 
1627 	if (bserrno != 0) {
1628 		/* The read failed, so jump to the final completion handler */
1629 		spdk_bs_sequence_finish(seq, bserrno);
1630 		return;
1631 	}
1632 
1633 	/* Write whole cluster */
1634 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1635 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1636 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1637 				   _spdk_blob_write_copy_cpl, ctx);
1638 }
1639 
1640 static void
1641 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1642 				   struct spdk_io_channel *_ch,
1643 				   uint64_t io_unit, spdk_bs_user_op_t *op)
1644 {
1645 	struct spdk_bs_cpl cpl;
1646 	struct spdk_bs_channel *ch;
1647 	struct spdk_blob_copy_cluster_ctx *ctx;
1648 	uint32_t cluster_start_page;
1649 	uint32_t cluster_number;
1650 	int rc;
1651 
1652 	ch = spdk_io_channel_get_ctx(_ch);
1653 
1654 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1655 		/* There are already operations pending. Queue this user op
1656 		 * and return because it will be re-executed when the outstanding
1657 		 * cluster allocation completes. */
1658 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1659 		return;
1660 	}
1661 
1662 	/* Round the io_unit offset down to the first page in the cluster */
1663 	cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit);
1664 
1665 	/* Calculate which index in the metadata cluster array the corresponding
1666 	 * cluster is supposed to be at. */
1667 	cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit);
1668 
1669 	ctx = calloc(1, sizeof(*ctx));
1670 	if (!ctx) {
1671 		spdk_bs_user_op_abort(op);
1672 		return;
1673 	}
1674 
1675 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1676 
1677 	ctx->blob = blob;
1678 	ctx->page = cluster_start_page;
1679 
1680 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1681 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
1682 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1683 		if (!ctx->buf) {
1684 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1685 				    blob->bs->cluster_sz);
1686 			free(ctx);
1687 			spdk_bs_user_op_abort(op);
1688 			return;
1689 		}
1690 	}
1691 
1692 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1693 	if (rc != 0) {
1694 		spdk_free(ctx->buf);
1695 		free(ctx);
1696 		spdk_bs_user_op_abort(op);
1697 		return;
1698 	}
1699 
1700 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1701 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1702 	cpl.u.blob_basic.cb_arg = ctx;
1703 
1704 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1705 	if (!ctx->seq) {
1706 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1707 		spdk_free(ctx->buf);
1708 		free(ctx);
1709 		spdk_bs_user_op_abort(op);
1710 		return;
1711 	}
1712 
1713 	/* Queue the user op to block other incoming operations */
1714 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1715 
1716 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1717 		/* Read cluster from backing device */
1718 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1719 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1720 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1721 					     _spdk_blob_write_copy, ctx);
1722 	} else {
1723 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1724 						       _spdk_blob_insert_cluster_cpl, ctx);
1725 	}
1726 }
1727 
1728 static void
1729 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
1730 				       uint64_t *lba,	uint32_t *lba_count)
1731 {
1732 	*lba_count = length;
1733 
1734 	if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) {
1735 		assert(blob->back_bs_dev != NULL);
1736 		*lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit);
1737 		*lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count);
1738 	} else {
1739 		*lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit);
1740 	}
1741 }
1742 
1743 struct op_split_ctx {
1744 	struct spdk_blob *blob;
1745 	struct spdk_io_channel *channel;
1746 	uint64_t io_unit_offset;
1747 	uint64_t io_units_remaining;
1748 	void *curr_payload;
1749 	enum spdk_blob_op_type op_type;
1750 	spdk_bs_sequence_t *seq;
1751 };
1752 
1753 static void
1754 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1755 {
1756 	struct op_split_ctx	*ctx = cb_arg;
1757 	struct spdk_blob	*blob = ctx->blob;
1758 	struct spdk_io_channel	*ch = ctx->channel;
1759 	enum spdk_blob_op_type	op_type = ctx->op_type;
1760 	uint8_t			*buf = ctx->curr_payload;
1761 	uint64_t		offset = ctx->io_unit_offset;
1762 	uint64_t		length = ctx->io_units_remaining;
1763 	uint64_t		op_length;
1764 
1765 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
1766 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1767 		free(ctx);
1768 		return;
1769 	}
1770 
1771 	op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob,
1772 			     offset));
1773 
1774 	/* Update length and payload for next operation */
1775 	ctx->io_units_remaining -= op_length;
1776 	ctx->io_unit_offset += op_length;
1777 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1778 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
1779 	}
1780 
1781 	switch (op_type) {
1782 	case SPDK_BLOB_READ:
1783 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1784 				  _spdk_blob_request_submit_op_split_next, ctx);
1785 		break;
1786 	case SPDK_BLOB_WRITE:
1787 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1788 				   _spdk_blob_request_submit_op_split_next, ctx);
1789 		break;
1790 	case SPDK_BLOB_UNMAP:
1791 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1792 				   _spdk_blob_request_submit_op_split_next, ctx);
1793 		break;
1794 	case SPDK_BLOB_WRITE_ZEROES:
1795 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1796 					  _spdk_blob_request_submit_op_split_next, ctx);
1797 		break;
1798 	case SPDK_BLOB_READV:
1799 	case SPDK_BLOB_WRITEV:
1800 		SPDK_ERRLOG("readv/write not valid for %s\n", __func__);
1801 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1802 		free(ctx);
1803 		break;
1804 	}
1805 }
1806 
1807 static void
1808 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1809 				   void *payload, uint64_t offset, uint64_t length,
1810 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1811 {
1812 	struct op_split_ctx *ctx;
1813 	spdk_bs_sequence_t *seq;
1814 	struct spdk_bs_cpl cpl;
1815 
1816 	assert(blob != NULL);
1817 
1818 	ctx = calloc(1, sizeof(struct op_split_ctx));
1819 	if (ctx == NULL) {
1820 		cb_fn(cb_arg, -ENOMEM);
1821 		return;
1822 	}
1823 
1824 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1825 	cpl.u.blob_basic.cb_fn = cb_fn;
1826 	cpl.u.blob_basic.cb_arg = cb_arg;
1827 
1828 	seq = spdk_bs_sequence_start(ch, &cpl);
1829 	if (!seq) {
1830 		free(ctx);
1831 		cb_fn(cb_arg, -ENOMEM);
1832 		return;
1833 	}
1834 
1835 	ctx->blob = blob;
1836 	ctx->channel = ch;
1837 	ctx->curr_payload = payload;
1838 	ctx->io_unit_offset = offset;
1839 	ctx->io_units_remaining = length;
1840 	ctx->op_type = op_type;
1841 	ctx->seq = seq;
1842 
1843 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1844 }
1845 
1846 static void
1847 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1848 				    void *payload, uint64_t offset, uint64_t length,
1849 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1850 {
1851 	struct spdk_bs_cpl cpl;
1852 	uint64_t lba;
1853 	uint32_t lba_count;
1854 
1855 	assert(blob != NULL);
1856 
1857 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1858 	cpl.u.blob_basic.cb_fn = cb_fn;
1859 	cpl.u.blob_basic.cb_arg = cb_arg;
1860 
1861 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1862 
1863 	if (blob->frozen_refcnt) {
1864 		/* This blob I/O is frozen */
1865 		spdk_bs_user_op_t *op;
1866 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1867 
1868 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1869 		if (!op) {
1870 			cb_fn(cb_arg, -ENOMEM);
1871 			return;
1872 		}
1873 
1874 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1875 
1876 		return;
1877 	}
1878 
1879 	switch (op_type) {
1880 	case SPDK_BLOB_READ: {
1881 		spdk_bs_batch_t *batch;
1882 
1883 		batch = spdk_bs_batch_open(_ch, &cpl);
1884 		if (!batch) {
1885 			cb_fn(cb_arg, -ENOMEM);
1886 			return;
1887 		}
1888 
1889 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1890 			/* Read from the blob */
1891 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1892 		} else {
1893 			/* Read from the backing block device */
1894 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1895 		}
1896 
1897 		spdk_bs_batch_close(batch);
1898 		break;
1899 	}
1900 	case SPDK_BLOB_WRITE:
1901 	case SPDK_BLOB_WRITE_ZEROES: {
1902 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1903 			/* Write to the blob */
1904 			spdk_bs_batch_t *batch;
1905 
1906 			if (lba_count == 0) {
1907 				cb_fn(cb_arg, 0);
1908 				return;
1909 			}
1910 
1911 			batch = spdk_bs_batch_open(_ch, &cpl);
1912 			if (!batch) {
1913 				cb_fn(cb_arg, -ENOMEM);
1914 				return;
1915 			}
1916 
1917 			if (op_type == SPDK_BLOB_WRITE) {
1918 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1919 			} else {
1920 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1921 			}
1922 
1923 			spdk_bs_batch_close(batch);
1924 		} else {
1925 			/* Queue this operation and allocate the cluster */
1926 			spdk_bs_user_op_t *op;
1927 
1928 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1929 			if (!op) {
1930 				cb_fn(cb_arg, -ENOMEM);
1931 				return;
1932 			}
1933 
1934 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1935 		}
1936 		break;
1937 	}
1938 	case SPDK_BLOB_UNMAP: {
1939 		spdk_bs_batch_t *batch;
1940 
1941 		batch = spdk_bs_batch_open(_ch, &cpl);
1942 		if (!batch) {
1943 			cb_fn(cb_arg, -ENOMEM);
1944 			return;
1945 		}
1946 
1947 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1948 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1949 		}
1950 
1951 		spdk_bs_batch_close(batch);
1952 		break;
1953 	}
1954 	case SPDK_BLOB_READV:
1955 	case SPDK_BLOB_WRITEV:
1956 		SPDK_ERRLOG("readv/write not valid\n");
1957 		cb_fn(cb_arg, -EINVAL);
1958 		break;
1959 	}
1960 }
1961 
1962 static void
1963 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1964 			     void *payload, uint64_t offset, uint64_t length,
1965 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1966 {
1967 	assert(blob != NULL);
1968 
1969 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1970 		cb_fn(cb_arg, -EPERM);
1971 		return;
1972 	}
1973 
1974 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
1975 		cb_fn(cb_arg, -EINVAL);
1976 		return;
1977 	}
1978 	if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) {
1979 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1980 						    cb_fn, cb_arg, op_type);
1981 	} else {
1982 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1983 						   cb_fn, cb_arg, op_type);
1984 	}
1985 }
1986 
1987 struct rw_iov_ctx {
1988 	struct spdk_blob *blob;
1989 	struct spdk_io_channel *channel;
1990 	spdk_blob_op_complete cb_fn;
1991 	void *cb_arg;
1992 	bool read;
1993 	int iovcnt;
1994 	struct iovec *orig_iov;
1995 	uint64_t io_unit_offset;
1996 	uint64_t io_units_remaining;
1997 	uint64_t io_units_done;
1998 	struct iovec iov[0];
1999 };
2000 
2001 static void
2002 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2003 {
2004 	assert(cb_arg == NULL);
2005 	spdk_bs_sequence_finish(seq, bserrno);
2006 }
2007 
2008 static void
2009 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
2010 {
2011 	struct rw_iov_ctx *ctx = cb_arg;
2012 	struct spdk_blob *blob = ctx->blob;
2013 	struct iovec *iov, *orig_iov;
2014 	int iovcnt;
2015 	size_t orig_iovoff;
2016 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2017 	uint64_t byte_count;
2018 
2019 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2020 		ctx->cb_fn(ctx->cb_arg, bserrno);
2021 		free(ctx);
2022 		return;
2023 	}
2024 
2025 	io_unit_offset = ctx->io_unit_offset;
2026 	io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2027 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2028 	/*
2029 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2030 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2031 	 *  point to the current position in the I/O sequence.
2032 	 */
2033 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2034 	orig_iov = &ctx->orig_iov[0];
2035 	orig_iovoff = 0;
2036 	while (byte_count > 0) {
2037 		if (byte_count >= orig_iov->iov_len) {
2038 			byte_count -= orig_iov->iov_len;
2039 			orig_iov++;
2040 		} else {
2041 			orig_iovoff = byte_count;
2042 			byte_count = 0;
2043 		}
2044 	}
2045 
2046 	/*
2047 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2048 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2049 	 */
2050 	byte_count = io_units_count * blob->bs->io_unit_size;
2051 	iov = &ctx->iov[0];
2052 	iovcnt = 0;
2053 	while (byte_count > 0) {
2054 		assert(iovcnt < ctx->iovcnt);
2055 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2056 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2057 		byte_count -= iov->iov_len;
2058 		orig_iovoff = 0;
2059 		orig_iov++;
2060 		iov++;
2061 		iovcnt++;
2062 	}
2063 
2064 	ctx->io_unit_offset += io_units_count;
2065 	ctx->io_units_remaining -= io_units_count;
2066 	ctx->io_units_done += io_units_count;
2067 	iov = &ctx->iov[0];
2068 
2069 	if (ctx->read) {
2070 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2071 				   io_units_count, _spdk_rw_iov_split_next, ctx);
2072 	} else {
2073 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2074 				    io_units_count, _spdk_rw_iov_split_next, ctx);
2075 	}
2076 }
2077 
2078 static void
2079 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2080 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2081 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2082 {
2083 	struct spdk_bs_cpl	cpl;
2084 
2085 	assert(blob != NULL);
2086 
2087 	if (!read && blob->data_ro) {
2088 		cb_fn(cb_arg, -EPERM);
2089 		return;
2090 	}
2091 
2092 	if (length == 0) {
2093 		cb_fn(cb_arg, 0);
2094 		return;
2095 	}
2096 
2097 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2098 		cb_fn(cb_arg, -EINVAL);
2099 		return;
2100 	}
2101 
2102 	/*
2103 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2104 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2105 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2106 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2107 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2108 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2109 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2110 	 *
2111 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2112 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2113 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2114 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2115 	 */
2116 	if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) {
2117 		uint32_t lba_count;
2118 		uint64_t lba;
2119 
2120 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2121 		cpl.u.blob_basic.cb_fn = cb_fn;
2122 		cpl.u.blob_basic.cb_arg = cb_arg;
2123 
2124 		if (blob->frozen_refcnt) {
2125 			/* This blob I/O is frozen */
2126 			enum spdk_blob_op_type op_type;
2127 			spdk_bs_user_op_t *op;
2128 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2129 
2130 			op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV;
2131 			op = spdk_bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length);
2132 			if (!op) {
2133 				cb_fn(cb_arg, -ENOMEM);
2134 				return;
2135 			}
2136 
2137 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2138 
2139 			return;
2140 		}
2141 
2142 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2143 
2144 		if (read) {
2145 			spdk_bs_sequence_t *seq;
2146 
2147 			seq = spdk_bs_sequence_start(_channel, &cpl);
2148 			if (!seq) {
2149 				cb_fn(cb_arg, -ENOMEM);
2150 				return;
2151 			}
2152 
2153 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2154 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2155 			} else {
2156 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2157 							      _spdk_rw_iov_done, NULL);
2158 			}
2159 		} else {
2160 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2161 				spdk_bs_sequence_t *seq;
2162 
2163 				seq = spdk_bs_sequence_start(_channel, &cpl);
2164 				if (!seq) {
2165 					cb_fn(cb_arg, -ENOMEM);
2166 					return;
2167 				}
2168 
2169 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2170 			} else {
2171 				/* Queue this operation and allocate the cluster */
2172 				spdk_bs_user_op_t *op;
2173 
2174 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2175 							   length);
2176 				if (!op) {
2177 					cb_fn(cb_arg, -ENOMEM);
2178 					return;
2179 				}
2180 
2181 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2182 			}
2183 		}
2184 	} else {
2185 		struct rw_iov_ctx *ctx;
2186 
2187 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2188 		if (ctx == NULL) {
2189 			cb_fn(cb_arg, -ENOMEM);
2190 			return;
2191 		}
2192 
2193 		ctx->blob = blob;
2194 		ctx->channel = _channel;
2195 		ctx->cb_fn = cb_fn;
2196 		ctx->cb_arg = cb_arg;
2197 		ctx->read = read;
2198 		ctx->orig_iov = iov;
2199 		ctx->iovcnt = iovcnt;
2200 		ctx->io_unit_offset = offset;
2201 		ctx->io_units_remaining = length;
2202 		ctx->io_units_done = 0;
2203 
2204 		_spdk_rw_iov_split_next(ctx, 0);
2205 	}
2206 }
2207 
2208 static struct spdk_blob *
2209 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2210 {
2211 	struct spdk_blob *blob;
2212 
2213 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2214 		if (blob->id == blobid) {
2215 			return blob;
2216 		}
2217 	}
2218 
2219 	return NULL;
2220 }
2221 
2222 static void
2223 _spdk_blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
2224 		struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
2225 {
2226 	assert(blob != NULL);
2227 	*snapshot_entry = NULL;
2228 	*clone_entry = NULL;
2229 
2230 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
2231 		return;
2232 	}
2233 
2234 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
2235 		if ((*snapshot_entry)->id == blob->parent_id) {
2236 			break;
2237 		}
2238 	}
2239 
2240 	if (*snapshot_entry != NULL) {
2241 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
2242 			if ((*clone_entry)->id == blob->id) {
2243 				break;
2244 			}
2245 		}
2246 
2247 		assert(clone_entry != NULL);
2248 	}
2249 }
2250 
2251 static int
2252 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2253 {
2254 	struct spdk_blob_store		*bs = io_device;
2255 	struct spdk_bs_channel		*channel = ctx_buf;
2256 	struct spdk_bs_dev		*dev;
2257 	uint32_t			max_ops = bs->max_channel_ops;
2258 	uint32_t			i;
2259 
2260 	dev = bs->dev;
2261 
2262 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2263 	if (!channel->req_mem) {
2264 		return -1;
2265 	}
2266 
2267 	TAILQ_INIT(&channel->reqs);
2268 
2269 	for (i = 0; i < max_ops; i++) {
2270 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2271 	}
2272 
2273 	channel->bs = bs;
2274 	channel->dev = dev;
2275 	channel->dev_channel = dev->create_channel(dev);
2276 
2277 	if (!channel->dev_channel) {
2278 		SPDK_ERRLOG("Failed to create device channel.\n");
2279 		free(channel->req_mem);
2280 		return -1;
2281 	}
2282 
2283 	TAILQ_INIT(&channel->need_cluster_alloc);
2284 	TAILQ_INIT(&channel->queued_io);
2285 
2286 	return 0;
2287 }
2288 
2289 static void
2290 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2291 {
2292 	struct spdk_bs_channel *channel = ctx_buf;
2293 	spdk_bs_user_op_t *op;
2294 
2295 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2296 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2297 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2298 		spdk_bs_user_op_abort(op);
2299 	}
2300 
2301 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2302 		op = TAILQ_FIRST(&channel->queued_io);
2303 		TAILQ_REMOVE(&channel->queued_io, op, link);
2304 		spdk_bs_user_op_abort(op);
2305 	}
2306 
2307 	free(channel->req_mem);
2308 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2309 }
2310 
2311 static void
2312 _spdk_bs_dev_destroy(void *io_device)
2313 {
2314 	struct spdk_blob_store *bs = io_device;
2315 	struct spdk_blob	*blob, *blob_tmp;
2316 
2317 	bs->dev->destroy(bs->dev);
2318 
2319 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2320 		TAILQ_REMOVE(&bs->blobs, blob, link);
2321 		_spdk_blob_free(blob);
2322 	}
2323 
2324 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2325 
2326 	spdk_bit_array_free(&bs->used_blobids);
2327 	spdk_bit_array_free(&bs->used_md_pages);
2328 	spdk_bit_array_free(&bs->used_clusters);
2329 	/*
2330 	 * If this function is called for any reason except a successful unload,
2331 	 * the unload_cpl type will be NONE and this will be a nop.
2332 	 */
2333 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2334 
2335 	free(bs);
2336 }
2337 
2338 static int
2339 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2340 {
2341 	spdk_blob_id snapshot_id;
2342 	struct spdk_blob_list *snapshot_entry = NULL;
2343 	struct spdk_blob_list *clone_entry = NULL;
2344 
2345 	assert(blob != NULL);
2346 
2347 	snapshot_id = blob->parent_id;
2348 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2349 		return 0;
2350 	}
2351 
2352 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, snapshot_id);
2353 	if (snapshot_entry == NULL) {
2354 		/* Snapshot not found */
2355 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2356 		if (snapshot_entry == NULL) {
2357 			return -ENOMEM;
2358 		}
2359 		snapshot_entry->id = snapshot_id;
2360 		TAILQ_INIT(&snapshot_entry->clones);
2361 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2362 	} else {
2363 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2364 			if (clone_entry->id == blob->id) {
2365 				break;
2366 			}
2367 		}
2368 	}
2369 
2370 	if (clone_entry == NULL) {
2371 		/* Clone not found */
2372 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2373 		if (clone_entry == NULL) {
2374 			return -ENOMEM;
2375 		}
2376 		clone_entry->id = blob->id;
2377 		TAILQ_INIT(&clone_entry->clones);
2378 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2379 		snapshot_entry->clone_count++;
2380 	}
2381 
2382 	return 0;
2383 }
2384 
2385 static void
2386 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2387 {
2388 	struct spdk_blob_list *snapshot_entry = NULL;
2389 	struct spdk_blob_list *clone_entry = NULL;
2390 
2391 	_spdk_blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
2392 
2393 	if (snapshot_entry == NULL) {
2394 		return;
2395 	}
2396 
2397 	blob->parent_id = SPDK_BLOBID_INVALID;
2398 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2399 	free(clone_entry);
2400 
2401 	snapshot_entry->clone_count--;
2402 }
2403 
2404 static int
2405 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2406 {
2407 	struct spdk_blob_list *snapshot_entry;
2408 	struct spdk_blob_list *snapshot_entry_tmp;
2409 	struct spdk_blob_list *clone_entry;
2410 	struct spdk_blob_list *clone_entry_tmp;
2411 
2412 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2413 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2414 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2415 			free(clone_entry);
2416 		}
2417 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2418 		free(snapshot_entry);
2419 	}
2420 
2421 	return 0;
2422 }
2423 
2424 static void
2425 _spdk_bs_free(struct spdk_blob_store *bs)
2426 {
2427 	_spdk_bs_blob_list_free(bs);
2428 
2429 	spdk_bs_unregister_md_thread(bs);
2430 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2431 }
2432 
2433 void
2434 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2435 {
2436 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2437 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2438 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2439 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2440 	opts->clear_method = BS_CLEAR_WITH_UNMAP;
2441 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2442 	opts->iter_cb_fn = NULL;
2443 	opts->iter_cb_arg = NULL;
2444 }
2445 
2446 static int
2447 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2448 {
2449 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2450 	    opts->max_channel_ops == 0) {
2451 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2452 		return -1;
2453 	}
2454 
2455 	return 0;
2456 }
2457 
2458 static int
2459 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2460 {
2461 	struct spdk_blob_store	*bs;
2462 	uint64_t dev_size;
2463 	int rc;
2464 
2465 	dev_size = dev->blocklen * dev->blockcnt;
2466 	if (dev_size < opts->cluster_sz) {
2467 		/* Device size cannot be smaller than cluster size of blobstore */
2468 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2469 			     dev_size, opts->cluster_sz);
2470 		return -ENOSPC;
2471 	}
2472 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2473 		/* Cluster size cannot be smaller than page size */
2474 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2475 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2476 		return -EINVAL;
2477 	}
2478 	bs = calloc(1, sizeof(struct spdk_blob_store));
2479 	if (!bs) {
2480 		return -ENOMEM;
2481 	}
2482 
2483 	TAILQ_INIT(&bs->blobs);
2484 	TAILQ_INIT(&bs->snapshots);
2485 	bs->dev = dev;
2486 	bs->md_thread = spdk_get_thread();
2487 	assert(bs->md_thread != NULL);
2488 
2489 	/*
2490 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2491 	 *  even multiple of the cluster size.
2492 	 */
2493 	bs->cluster_sz = opts->cluster_sz;
2494 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2495 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2496 	bs->num_free_clusters = bs->total_clusters;
2497 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2498 	bs->io_unit_size = dev->blocklen;
2499 	if (bs->used_clusters == NULL) {
2500 		free(bs);
2501 		return -ENOMEM;
2502 	}
2503 
2504 	bs->max_channel_ops = opts->max_channel_ops;
2505 	bs->super_blob = SPDK_BLOBID_INVALID;
2506 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2507 
2508 	/* The metadata is assumed to be at least 1 page */
2509 	bs->used_md_pages = spdk_bit_array_create(1);
2510 	bs->used_blobids = spdk_bit_array_create(0);
2511 
2512 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2513 
2514 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2515 				sizeof(struct spdk_bs_channel), "blobstore");
2516 	rc = spdk_bs_register_md_thread(bs);
2517 	if (rc == -1) {
2518 		spdk_io_device_unregister(bs, NULL);
2519 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2520 		spdk_bit_array_free(&bs->used_blobids);
2521 		spdk_bit_array_free(&bs->used_md_pages);
2522 		spdk_bit_array_free(&bs->used_clusters);
2523 		free(bs);
2524 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2525 		return -ENOMEM;
2526 	}
2527 
2528 	*_bs = bs;
2529 	return 0;
2530 }
2531 
2532 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2533 
2534 struct spdk_bs_load_ctx {
2535 	struct spdk_blob_store		*bs;
2536 	struct spdk_bs_super_block	*super;
2537 
2538 	struct spdk_bs_md_mask		*mask;
2539 	bool				in_page_chain;
2540 	uint32_t			page_index;
2541 	uint32_t			cur_page;
2542 	struct spdk_blob_md_page	*page;
2543 
2544 	spdk_bs_sequence_t			*seq;
2545 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2546 	void					*iter_cb_arg;
2547 };
2548 
2549 static void
2550 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2551 {
2552 	assert(bserrno != 0);
2553 
2554 	spdk_free(ctx->super);
2555 	spdk_bs_sequence_finish(seq, bserrno);
2556 	_spdk_bs_free(ctx->bs);
2557 	free(ctx);
2558 }
2559 
2560 static void
2561 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2562 {
2563 	uint32_t i = 0;
2564 
2565 	while (true) {
2566 		i = spdk_bit_array_find_first_set(array, i);
2567 		if (i >= mask->length) {
2568 			break;
2569 		}
2570 		mask->mask[i / 8] |= 1U << (i % 8);
2571 		i++;
2572 	}
2573 }
2574 
2575 static int
2576 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2577 {
2578 	struct spdk_bit_array *array;
2579 	uint32_t i;
2580 
2581 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2582 		return -ENOMEM;
2583 	}
2584 
2585 	array = *array_ptr;
2586 	for (i = 0; i < mask->length; i++) {
2587 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2588 			spdk_bit_array_set(array, i);
2589 		}
2590 	}
2591 
2592 	return 0;
2593 }
2594 
2595 static void
2596 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2597 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2598 {
2599 	/* Update the values in the super block */
2600 	super->super_blob = bs->super_blob;
2601 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2602 	super->crc = _spdk_blob_md_page_calc_crc(super);
2603 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2604 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2605 				   cb_fn, cb_arg);
2606 }
2607 
2608 static void
2609 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2610 {
2611 	struct spdk_bs_load_ctx	*ctx = arg;
2612 	uint64_t	mask_size, lba, lba_count;
2613 
2614 	/* Write out the used clusters mask */
2615 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2616 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2617 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2618 	if (!ctx->mask) {
2619 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2620 		return;
2621 	}
2622 
2623 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2624 	ctx->mask->length = ctx->bs->total_clusters;
2625 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2626 
2627 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2628 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2629 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2630 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2631 }
2632 
2633 static void
2634 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2635 {
2636 	struct spdk_bs_load_ctx	*ctx = arg;
2637 	uint64_t	mask_size, lba, lba_count;
2638 
2639 	if (seq->bserrno) {
2640 		_spdk_bs_load_ctx_fail(seq, ctx, seq->bserrno);
2641 		return;
2642 	}
2643 
2644 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2645 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2646 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2647 	if (!ctx->mask) {
2648 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2649 		return;
2650 	}
2651 
2652 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2653 	ctx->mask->length = ctx->super->md_len;
2654 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2655 
2656 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2657 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2658 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2659 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2660 }
2661 
2662 static void
2663 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2664 {
2665 	struct spdk_bs_load_ctx	*ctx = arg;
2666 	uint64_t	mask_size, lba, lba_count;
2667 
2668 	if (ctx->super->used_blobid_mask_len == 0) {
2669 		/*
2670 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2671 		 *  written to disk.
2672 		 */
2673 		cb_fn(seq, arg, 0);
2674 		return;
2675 	}
2676 
2677 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2678 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2679 				 SPDK_MALLOC_DMA);
2680 	if (!ctx->mask) {
2681 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2682 		return;
2683 	}
2684 
2685 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2686 	ctx->mask->length = ctx->super->md_len;
2687 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2688 
2689 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2690 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2691 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2692 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2693 }
2694 
2695 static void
2696 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2697 {
2698 	struct spdk_bs_load_ctx *ctx = arg;
2699 
2700 	if (bserrno == 0) {
2701 		if (ctx->iter_cb_fn) {
2702 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2703 		}
2704 		_spdk_bs_blob_list_add(blob);
2705 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2706 		return;
2707 	}
2708 
2709 	if (bserrno == -ENOENT) {
2710 		bserrno = 0;
2711 	} else {
2712 		/*
2713 		 * This case needs to be looked at further.  Same problem
2714 		 *  exists with applications that rely on explicit blob
2715 		 *  iteration.  We should just skip the blob that failed
2716 		 *  to load and continue on to the next one.
2717 		 */
2718 		SPDK_ERRLOG("Error in iterating blobs\n");
2719 	}
2720 
2721 	ctx->iter_cb_fn = NULL;
2722 
2723 	spdk_free(ctx->super);
2724 	spdk_free(ctx->mask);
2725 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2726 	free(ctx);
2727 }
2728 
2729 static void
2730 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2731 {
2732 	ctx->seq = seq;
2733 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2734 }
2735 
2736 static void
2737 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2738 {
2739 	struct spdk_bs_load_ctx *ctx = cb_arg;
2740 	int rc;
2741 
2742 	/* The type must be correct */
2743 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2744 
2745 	/* The length of the mask (in bits) must not be greater than
2746 	 * the length of the buffer (converted to bits) */
2747 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2748 
2749 	/* The length of the mask must be exactly equal to the size
2750 	 * (in pages) of the metadata region */
2751 	assert(ctx->mask->length == ctx->super->md_len);
2752 
2753 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2754 	if (rc < 0) {
2755 		spdk_free(ctx->mask);
2756 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2757 		return;
2758 	}
2759 
2760 	_spdk_bs_load_complete(seq, ctx, bserrno);
2761 }
2762 
2763 static void
2764 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2765 {
2766 	struct spdk_bs_load_ctx *ctx = cb_arg;
2767 	uint64_t		lba, lba_count, mask_size;
2768 	int			rc;
2769 
2770 	/* The type must be correct */
2771 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2772 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2773 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2774 					     struct spdk_blob_md_page) * 8));
2775 	/* The length of the mask must be exactly equal to the total number of clusters */
2776 	assert(ctx->mask->length == ctx->bs->total_clusters);
2777 
2778 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2779 	if (rc < 0) {
2780 		spdk_free(ctx->mask);
2781 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2782 		return;
2783 	}
2784 
2785 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2786 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2787 
2788 	spdk_free(ctx->mask);
2789 
2790 	/* Read the used blobids mask */
2791 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2792 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2793 				 SPDK_MALLOC_DMA);
2794 	if (!ctx->mask) {
2795 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2796 		return;
2797 	}
2798 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2799 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2800 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2801 				  _spdk_bs_load_used_blobids_cpl, ctx);
2802 }
2803 
2804 static void
2805 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2806 {
2807 	struct spdk_bs_load_ctx *ctx = cb_arg;
2808 	uint64_t		lba, lba_count, mask_size;
2809 	int			rc;
2810 
2811 	/* The type must be correct */
2812 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2813 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2814 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2815 				     8));
2816 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2817 	assert(ctx->mask->length == ctx->super->md_len);
2818 
2819 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2820 	if (rc < 0) {
2821 		spdk_free(ctx->mask);
2822 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2823 		return;
2824 	}
2825 
2826 	spdk_free(ctx->mask);
2827 
2828 	/* Read the used clusters mask */
2829 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2830 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2831 				 SPDK_MALLOC_DMA);
2832 	if (!ctx->mask) {
2833 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2834 		return;
2835 	}
2836 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2837 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2838 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2839 				  _spdk_bs_load_used_clusters_cpl, ctx);
2840 }
2841 
2842 static void
2843 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2844 {
2845 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2846 	uint64_t lba, lba_count, mask_size;
2847 
2848 	/* Read the used pages mask */
2849 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2850 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2851 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2852 	if (!ctx->mask) {
2853 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2854 		return;
2855 	}
2856 
2857 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2858 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2859 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2860 				  _spdk_bs_load_used_pages_cpl, ctx);
2861 }
2862 
2863 static int
2864 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2865 {
2866 	struct spdk_blob_md_descriptor *desc;
2867 	size_t	cur_desc = 0;
2868 
2869 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2870 	while (cur_desc < sizeof(page->descriptors)) {
2871 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2872 			if (desc->length == 0) {
2873 				/* If padding and length are 0, this terminates the page */
2874 				break;
2875 			}
2876 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2877 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2878 			unsigned int				i, j;
2879 			unsigned int				cluster_count = 0;
2880 			uint32_t				cluster_idx;
2881 
2882 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2883 
2884 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2885 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2886 					cluster_idx = desc_extent->extents[i].cluster_idx;
2887 					/*
2888 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2889 					 * in the used cluster map.
2890 					 */
2891 					if (cluster_idx != 0) {
2892 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2893 						if (bs->num_free_clusters == 0) {
2894 							return -ENOSPC;
2895 						}
2896 						bs->num_free_clusters--;
2897 					}
2898 					cluster_count++;
2899 				}
2900 			}
2901 			if (cluster_count == 0) {
2902 				return -EINVAL;
2903 			}
2904 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2905 			/* Skip this item */
2906 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2907 			/* Skip this item */
2908 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2909 			/* Skip this item */
2910 		} else {
2911 			/* Error */
2912 			return -EINVAL;
2913 		}
2914 		/* Advance to the next descriptor */
2915 		cur_desc += sizeof(*desc) + desc->length;
2916 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2917 			break;
2918 		}
2919 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2920 	}
2921 	return 0;
2922 }
2923 
2924 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2925 {
2926 	uint32_t crc;
2927 
2928 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2929 	if (crc != ctx->page->crc) {
2930 		return false;
2931 	}
2932 
2933 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2934 		return false;
2935 	}
2936 	return true;
2937 }
2938 
2939 static void
2940 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2941 
2942 static void
2943 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2944 {
2945 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2946 
2947 	_spdk_bs_load_complete(seq, ctx, bserrno);
2948 }
2949 
2950 static void
2951 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2952 {
2953 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2954 
2955 	spdk_free(ctx->mask);
2956 	ctx->mask = NULL;
2957 
2958 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2959 }
2960 
2961 static void
2962 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2963 {
2964 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2965 
2966 	spdk_free(ctx->mask);
2967 	ctx->mask = NULL;
2968 
2969 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2970 }
2971 
2972 static void
2973 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2974 {
2975 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2976 }
2977 
2978 static void
2979 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2980 {
2981 	struct spdk_bs_load_ctx *ctx = cb_arg;
2982 	uint64_t num_md_clusters;
2983 	uint64_t i;
2984 	uint32_t page_num;
2985 
2986 	if (bserrno != 0) {
2987 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2988 		return;
2989 	}
2990 
2991 	page_num = ctx->cur_page;
2992 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2993 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2994 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2995 			if (ctx->page->sequence_num == 0) {
2996 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2997 			}
2998 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2999 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3000 				return;
3001 			}
3002 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
3003 				ctx->in_page_chain = true;
3004 				ctx->cur_page = ctx->page->next;
3005 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3006 				return;
3007 			}
3008 		}
3009 	}
3010 
3011 	ctx->in_page_chain = false;
3012 
3013 	do {
3014 		ctx->page_index++;
3015 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
3016 
3017 	if (ctx->page_index < ctx->super->md_len) {
3018 		ctx->cur_page = ctx->page_index;
3019 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3020 	} else {
3021 		/* Claim all of the clusters used by the metadata */
3022 		num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
3023 		for (i = 0; i < num_md_clusters; i++) {
3024 			_spdk_bs_claim_cluster(ctx->bs, i);
3025 		}
3026 		spdk_free(ctx->page);
3027 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
3028 	}
3029 }
3030 
3031 static void
3032 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3033 {
3034 	struct spdk_bs_load_ctx *ctx = cb_arg;
3035 	uint64_t lba;
3036 
3037 	assert(ctx->cur_page < ctx->super->md_len);
3038 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3039 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3040 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3041 				  _spdk_bs_load_replay_md_cpl, ctx);
3042 }
3043 
3044 static void
3045 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3046 {
3047 	struct spdk_bs_load_ctx *ctx = cb_arg;
3048 
3049 	ctx->page_index = 0;
3050 	ctx->cur_page = 0;
3051 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3052 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3053 	if (!ctx->page) {
3054 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3055 		return;
3056 	}
3057 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3058 }
3059 
3060 static void
3061 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3062 {
3063 	struct spdk_bs_load_ctx *ctx = cb_arg;
3064 	int		rc;
3065 
3066 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3067 	if (rc < 0) {
3068 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3069 		return;
3070 	}
3071 
3072 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3073 	if (rc < 0) {
3074 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3075 		return;
3076 	}
3077 
3078 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3079 	if (rc < 0) {
3080 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3081 		return;
3082 	}
3083 
3084 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3085 	_spdk_bs_load_replay_md(seq, cb_arg);
3086 }
3087 
3088 static void
3089 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3090 {
3091 	struct spdk_bs_load_ctx *ctx = cb_arg;
3092 	uint32_t	crc;
3093 	int		rc;
3094 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3095 
3096 	if (ctx->super->version > SPDK_BS_VERSION ||
3097 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3098 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3099 		return;
3100 	}
3101 
3102 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3103 		   sizeof(ctx->super->signature)) != 0) {
3104 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3105 		return;
3106 	}
3107 
3108 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3109 	if (crc != ctx->super->crc) {
3110 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3111 		return;
3112 	}
3113 
3114 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3115 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3116 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3117 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3118 	} else {
3119 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3120 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3121 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3122 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3123 		return;
3124 	}
3125 
3126 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3127 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3128 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3129 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3130 		return;
3131 	}
3132 
3133 	if (ctx->super->size == 0) {
3134 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
3135 	}
3136 
3137 	if (ctx->super->io_unit_size == 0) {
3138 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
3139 	}
3140 
3141 	/* Parse the super block */
3142 	ctx->bs->clean = 1;
3143 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3144 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
3145 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3146 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
3147 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3148 	if (rc < 0) {
3149 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3150 		return;
3151 	}
3152 	ctx->bs->md_start = ctx->super->md_start;
3153 	ctx->bs->md_len = ctx->super->md_len;
3154 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
3155 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3156 	ctx->bs->super_blob = ctx->super->super_blob;
3157 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3158 
3159 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3160 		_spdk_bs_recover(seq, ctx);
3161 	} else {
3162 		_spdk_bs_load_read_used_pages(seq, ctx);
3163 	}
3164 }
3165 
3166 void
3167 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3168 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3169 {
3170 	struct spdk_blob_store	*bs;
3171 	struct spdk_bs_cpl	cpl;
3172 	spdk_bs_sequence_t	*seq;
3173 	struct spdk_bs_load_ctx *ctx;
3174 	struct spdk_bs_opts	opts = {};
3175 	int err;
3176 
3177 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3178 
3179 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3180 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3181 		dev->destroy(dev);
3182 		cb_fn(cb_arg, NULL, -EINVAL);
3183 		return;
3184 	}
3185 
3186 	if (o) {
3187 		opts = *o;
3188 	} else {
3189 		spdk_bs_opts_init(&opts);
3190 	}
3191 
3192 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3193 		dev->destroy(dev);
3194 		cb_fn(cb_arg, NULL, -EINVAL);
3195 		return;
3196 	}
3197 
3198 	err = _spdk_bs_alloc(dev, &opts, &bs);
3199 	if (err) {
3200 		dev->destroy(dev);
3201 		cb_fn(cb_arg, NULL, err);
3202 		return;
3203 	}
3204 
3205 	ctx = calloc(1, sizeof(*ctx));
3206 	if (!ctx) {
3207 		_spdk_bs_free(bs);
3208 		cb_fn(cb_arg, NULL, -ENOMEM);
3209 		return;
3210 	}
3211 
3212 	ctx->bs = bs;
3213 	ctx->iter_cb_fn = opts.iter_cb_fn;
3214 	ctx->iter_cb_arg = opts.iter_cb_arg;
3215 
3216 	/* Allocate memory for the super block */
3217 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3218 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3219 	if (!ctx->super) {
3220 		free(ctx);
3221 		_spdk_bs_free(bs);
3222 		cb_fn(cb_arg, NULL, -ENOMEM);
3223 		return;
3224 	}
3225 
3226 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3227 	cpl.u.bs_handle.cb_fn = cb_fn;
3228 	cpl.u.bs_handle.cb_arg = cb_arg;
3229 	cpl.u.bs_handle.bs = bs;
3230 
3231 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3232 	if (!seq) {
3233 		spdk_free(ctx->super);
3234 		free(ctx);
3235 		_spdk_bs_free(bs);
3236 		cb_fn(cb_arg, NULL, -ENOMEM);
3237 		return;
3238 	}
3239 
3240 	/* Read the super block */
3241 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3242 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3243 				  _spdk_bs_load_super_cpl, ctx);
3244 }
3245 
3246 /* END spdk_bs_load */
3247 
3248 /* START spdk_bs_dump */
3249 
3250 struct spdk_bs_dump_ctx {
3251 	struct spdk_blob_store		*bs;
3252 	struct spdk_bs_super_block	*super;
3253 	uint32_t			cur_page;
3254 	struct spdk_blob_md_page	*page;
3255 	spdk_bs_sequence_t		*seq;
3256 	FILE				*fp;
3257 	spdk_bs_dump_print_xattr	print_xattr_fn;
3258 	char				xattr_name[4096];
3259 };
3260 
3261 static void
3262 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3263 {
3264 	spdk_free(ctx->super);
3265 
3266 	/*
3267 	 * We need to defer calling spdk_bs_call_cpl() until after
3268 	 * dev destruction, so tuck these away for later use.
3269 	 */
3270 	ctx->bs->unload_err = bserrno;
3271 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3272 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3273 
3274 	spdk_bs_sequence_finish(seq, 0);
3275 	_spdk_bs_free(ctx->bs);
3276 	free(ctx);
3277 }
3278 
3279 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3280 
3281 static void
3282 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3283 {
3284 	uint32_t page_idx = ctx->cur_page;
3285 	struct spdk_blob_md_page *page = ctx->page;
3286 	struct spdk_blob_md_descriptor *desc;
3287 	size_t cur_desc = 0;
3288 	uint32_t crc;
3289 
3290 	fprintf(ctx->fp, "=========\n");
3291 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3292 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3293 
3294 	crc = _spdk_blob_md_page_calc_crc(page);
3295 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3296 
3297 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3298 	while (cur_desc < sizeof(page->descriptors)) {
3299 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3300 			if (desc->length == 0) {
3301 				/* If padding and length are 0, this terminates the page */
3302 				break;
3303 			}
3304 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3305 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3306 			unsigned int				i;
3307 
3308 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3309 
3310 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3311 				if (desc_extent->extents[i].cluster_idx != 0) {
3312 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3313 						desc_extent->extents[i].cluster_idx);
3314 				} else {
3315 					fprintf(ctx->fp, "Unallocated Extent - ");
3316 				}
3317 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3318 				fprintf(ctx->fp, "\n");
3319 			}
3320 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3321 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3322 			uint32_t i;
3323 
3324 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3325 
3326 			if (desc_xattr->length !=
3327 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3328 			    desc_xattr->name_length + desc_xattr->value_length) {
3329 			}
3330 
3331 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3332 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3333 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3334 			fprintf(ctx->fp, "       value = \"");
3335 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3336 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3337 					    desc_xattr->value_length);
3338 			fprintf(ctx->fp, "\"\n");
3339 			for (i = 0; i < desc_xattr->value_length; i++) {
3340 				if (i % 16 == 0) {
3341 					fprintf(ctx->fp, "               ");
3342 				}
3343 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3344 				if ((i + 1) % 16 == 0) {
3345 					fprintf(ctx->fp, "\n");
3346 				}
3347 			}
3348 			if (i % 16 != 0) {
3349 				fprintf(ctx->fp, "\n");
3350 			}
3351 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3352 			/* TODO */
3353 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3354 			/* TODO */
3355 		} else {
3356 			/* Error */
3357 		}
3358 		/* Advance to the next descriptor */
3359 		cur_desc += sizeof(*desc) + desc->length;
3360 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3361 			break;
3362 		}
3363 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3364 	}
3365 }
3366 
3367 static void
3368 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3369 {
3370 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3371 
3372 	if (bserrno != 0) {
3373 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3374 		return;
3375 	}
3376 
3377 	if (ctx->page->id != 0) {
3378 		_spdk_bs_dump_print_md_page(ctx);
3379 	}
3380 
3381 	ctx->cur_page++;
3382 
3383 	if (ctx->cur_page < ctx->super->md_len) {
3384 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3385 	} else {
3386 		spdk_free(ctx->page);
3387 		_spdk_bs_dump_finish(seq, ctx, 0);
3388 	}
3389 }
3390 
3391 static void
3392 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3393 {
3394 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3395 	uint64_t lba;
3396 
3397 	assert(ctx->cur_page < ctx->super->md_len);
3398 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3399 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3400 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3401 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3402 }
3403 
3404 static void
3405 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3406 {
3407 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3408 
3409 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3410 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3411 		   sizeof(ctx->super->signature)) != 0) {
3412 		fprintf(ctx->fp, "(Mismatch)\n");
3413 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3414 		return;
3415 	} else {
3416 		fprintf(ctx->fp, "(OK)\n");
3417 	}
3418 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3419 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3420 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3421 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3422 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3423 	fprintf(ctx->fp, "Super Blob ID: ");
3424 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3425 		fprintf(ctx->fp, "(None)\n");
3426 	} else {
3427 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3428 	}
3429 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3430 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3431 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3432 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3433 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3434 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3435 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3436 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3437 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3438 
3439 	ctx->cur_page = 0;
3440 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3441 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3442 	if (!ctx->page) {
3443 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3444 		return;
3445 	}
3446 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3447 }
3448 
3449 void
3450 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3451 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3452 {
3453 	struct spdk_blob_store	*bs;
3454 	struct spdk_bs_cpl	cpl;
3455 	spdk_bs_sequence_t	*seq;
3456 	struct spdk_bs_dump_ctx *ctx;
3457 	struct spdk_bs_opts	opts = {};
3458 	int err;
3459 
3460 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3461 
3462 	spdk_bs_opts_init(&opts);
3463 
3464 	err = _spdk_bs_alloc(dev, &opts, &bs);
3465 	if (err) {
3466 		dev->destroy(dev);
3467 		cb_fn(cb_arg, err);
3468 		return;
3469 	}
3470 
3471 	ctx = calloc(1, sizeof(*ctx));
3472 	if (!ctx) {
3473 		_spdk_bs_free(bs);
3474 		cb_fn(cb_arg, -ENOMEM);
3475 		return;
3476 	}
3477 
3478 	ctx->bs = bs;
3479 	ctx->fp = fp;
3480 	ctx->print_xattr_fn = print_xattr_fn;
3481 
3482 	/* Allocate memory for the super block */
3483 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3484 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3485 	if (!ctx->super) {
3486 		free(ctx);
3487 		_spdk_bs_free(bs);
3488 		cb_fn(cb_arg, -ENOMEM);
3489 		return;
3490 	}
3491 
3492 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3493 	cpl.u.bs_basic.cb_fn = cb_fn;
3494 	cpl.u.bs_basic.cb_arg = cb_arg;
3495 
3496 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3497 	if (!seq) {
3498 		spdk_free(ctx->super);
3499 		free(ctx);
3500 		_spdk_bs_free(bs);
3501 		cb_fn(cb_arg, -ENOMEM);
3502 		return;
3503 	}
3504 
3505 	/* Read the super block */
3506 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3507 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3508 				  _spdk_bs_dump_super_cpl, ctx);
3509 }
3510 
3511 /* END spdk_bs_dump */
3512 
3513 /* START spdk_bs_init */
3514 
3515 struct spdk_bs_init_ctx {
3516 	struct spdk_blob_store		*bs;
3517 	struct spdk_bs_super_block	*super;
3518 };
3519 
3520 static void
3521 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3522 {
3523 	struct spdk_bs_init_ctx *ctx = cb_arg;
3524 
3525 	spdk_free(ctx->super);
3526 	free(ctx);
3527 
3528 	spdk_bs_sequence_finish(seq, bserrno);
3529 }
3530 
3531 static void
3532 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3533 {
3534 	struct spdk_bs_init_ctx *ctx = cb_arg;
3535 
3536 	/* Write super block */
3537 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3538 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3539 				   _spdk_bs_init_persist_super_cpl, ctx);
3540 }
3541 
3542 void
3543 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3544 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3545 {
3546 	struct spdk_bs_init_ctx *ctx;
3547 	struct spdk_blob_store	*bs;
3548 	struct spdk_bs_cpl	cpl;
3549 	spdk_bs_sequence_t	*seq;
3550 	spdk_bs_batch_t		*batch;
3551 	uint64_t		num_md_lba;
3552 	uint64_t		num_md_pages;
3553 	uint64_t		num_md_clusters;
3554 	uint32_t		i;
3555 	struct spdk_bs_opts	opts = {};
3556 	int			rc;
3557 
3558 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3559 
3560 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3561 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3562 			    dev->blocklen);
3563 		dev->destroy(dev);
3564 		cb_fn(cb_arg, NULL, -EINVAL);
3565 		return;
3566 	}
3567 
3568 	if (o) {
3569 		opts = *o;
3570 	} else {
3571 		spdk_bs_opts_init(&opts);
3572 	}
3573 
3574 	if (_spdk_bs_opts_verify(&opts) != 0) {
3575 		dev->destroy(dev);
3576 		cb_fn(cb_arg, NULL, -EINVAL);
3577 		return;
3578 	}
3579 
3580 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3581 	if (rc) {
3582 		dev->destroy(dev);
3583 		cb_fn(cb_arg, NULL, rc);
3584 		return;
3585 	}
3586 
3587 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3588 		/* By default, allocate 1 page per cluster.
3589 		 * Technically, this over-allocates metadata
3590 		 * because more metadata will reduce the number
3591 		 * of usable clusters. This can be addressed with
3592 		 * more complex math in the future.
3593 		 */
3594 		bs->md_len = bs->total_clusters;
3595 	} else {
3596 		bs->md_len = opts.num_md_pages;
3597 	}
3598 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3599 	if (rc < 0) {
3600 		_spdk_bs_free(bs);
3601 		cb_fn(cb_arg, NULL, -ENOMEM);
3602 		return;
3603 	}
3604 
3605 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3606 	if (rc < 0) {
3607 		_spdk_bs_free(bs);
3608 		cb_fn(cb_arg, NULL, -ENOMEM);
3609 		return;
3610 	}
3611 
3612 	ctx = calloc(1, sizeof(*ctx));
3613 	if (!ctx) {
3614 		_spdk_bs_free(bs);
3615 		cb_fn(cb_arg, NULL, -ENOMEM);
3616 		return;
3617 	}
3618 
3619 	ctx->bs = bs;
3620 
3621 	/* Allocate memory for the super block */
3622 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3623 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3624 	if (!ctx->super) {
3625 		free(ctx);
3626 		_spdk_bs_free(bs);
3627 		cb_fn(cb_arg, NULL, -ENOMEM);
3628 		return;
3629 	}
3630 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3631 	       sizeof(ctx->super->signature));
3632 	ctx->super->version = SPDK_BS_VERSION;
3633 	ctx->super->length = sizeof(*ctx->super);
3634 	ctx->super->super_blob = bs->super_blob;
3635 	ctx->super->clean = 0;
3636 	ctx->super->cluster_size = bs->cluster_sz;
3637 	ctx->super->io_unit_size = bs->io_unit_size;
3638 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3639 
3640 	/* Calculate how many pages the metadata consumes at the front
3641 	 * of the disk.
3642 	 */
3643 
3644 	/* The super block uses 1 page */
3645 	num_md_pages = 1;
3646 
3647 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3648 	 * up to the nearest page, plus a header.
3649 	 */
3650 	ctx->super->used_page_mask_start = num_md_pages;
3651 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3652 					 spdk_divide_round_up(bs->md_len, 8),
3653 					 SPDK_BS_PAGE_SIZE);
3654 	num_md_pages += ctx->super->used_page_mask_len;
3655 
3656 	/* The used_clusters mask requires 1 bit per cluster, rounded
3657 	 * up to the nearest page, plus a header.
3658 	 */
3659 	ctx->super->used_cluster_mask_start = num_md_pages;
3660 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3661 					    spdk_divide_round_up(bs->total_clusters, 8),
3662 					    SPDK_BS_PAGE_SIZE);
3663 	num_md_pages += ctx->super->used_cluster_mask_len;
3664 
3665 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3666 	 * up to the nearest page, plus a header.
3667 	 */
3668 	ctx->super->used_blobid_mask_start = num_md_pages;
3669 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3670 					   spdk_divide_round_up(bs->md_len, 8),
3671 					   SPDK_BS_PAGE_SIZE);
3672 	num_md_pages += ctx->super->used_blobid_mask_len;
3673 
3674 	/* The metadata region size was chosen above */
3675 	ctx->super->md_start = bs->md_start = num_md_pages;
3676 	ctx->super->md_len = bs->md_len;
3677 	num_md_pages += bs->md_len;
3678 
3679 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3680 
3681 	ctx->super->size = dev->blockcnt * dev->blocklen;
3682 
3683 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3684 
3685 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
3686 	if (num_md_clusters > bs->total_clusters) {
3687 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3688 			    "please decrease number of pages reserved for metadata "
3689 			    "or increase cluster size.\n");
3690 		spdk_free(ctx->super);
3691 		free(ctx);
3692 		_spdk_bs_free(bs);
3693 		cb_fn(cb_arg, NULL, -ENOMEM);
3694 		return;
3695 	}
3696 	/* Claim all of the clusters used by the metadata */
3697 	for (i = 0; i < num_md_clusters; i++) {
3698 		_spdk_bs_claim_cluster(bs, i);
3699 	}
3700 
3701 	bs->total_data_clusters = bs->num_free_clusters;
3702 
3703 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3704 	cpl.u.bs_handle.cb_fn = cb_fn;
3705 	cpl.u.bs_handle.cb_arg = cb_arg;
3706 	cpl.u.bs_handle.bs = bs;
3707 
3708 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3709 	if (!seq) {
3710 		spdk_free(ctx->super);
3711 		free(ctx);
3712 		_spdk_bs_free(bs);
3713 		cb_fn(cb_arg, NULL, -ENOMEM);
3714 		return;
3715 	}
3716 
3717 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3718 
3719 	/* Clear metadata space */
3720 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3721 
3722 	if (opts.clear_method == BS_CLEAR_WITH_UNMAP) {
3723 		/* Trim data clusters */
3724 		spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3725 	} else if (opts.clear_method == BS_CLEAR_WITH_WRITE_ZEROES) {
3726 		/* Write_zeroes to data clusters */
3727 		spdk_bs_batch_write_zeroes_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3728 	}
3729 
3730 	spdk_bs_batch_close(batch);
3731 }
3732 
3733 /* END spdk_bs_init */
3734 
3735 /* START spdk_bs_destroy */
3736 
3737 static void
3738 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3739 {
3740 	struct spdk_bs_init_ctx *ctx = cb_arg;
3741 	struct spdk_blob_store *bs = ctx->bs;
3742 
3743 	/*
3744 	 * We need to defer calling spdk_bs_call_cpl() until after
3745 	 * dev destruction, so tuck these away for later use.
3746 	 */
3747 	bs->unload_err = bserrno;
3748 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3749 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3750 
3751 	spdk_bs_sequence_finish(seq, bserrno);
3752 
3753 	_spdk_bs_free(bs);
3754 	free(ctx);
3755 }
3756 
3757 void
3758 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3759 		void *cb_arg)
3760 {
3761 	struct spdk_bs_cpl	cpl;
3762 	spdk_bs_sequence_t	*seq;
3763 	struct spdk_bs_init_ctx *ctx;
3764 
3765 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3766 
3767 	if (!TAILQ_EMPTY(&bs->blobs)) {
3768 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3769 		cb_fn(cb_arg, -EBUSY);
3770 		return;
3771 	}
3772 
3773 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3774 	cpl.u.bs_basic.cb_fn = cb_fn;
3775 	cpl.u.bs_basic.cb_arg = cb_arg;
3776 
3777 	ctx = calloc(1, sizeof(*ctx));
3778 	if (!ctx) {
3779 		cb_fn(cb_arg, -ENOMEM);
3780 		return;
3781 	}
3782 
3783 	ctx->bs = bs;
3784 
3785 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3786 	if (!seq) {
3787 		free(ctx);
3788 		cb_fn(cb_arg, -ENOMEM);
3789 		return;
3790 	}
3791 
3792 	/* Write zeroes to the super block */
3793 	spdk_bs_sequence_write_zeroes_dev(seq,
3794 					  _spdk_bs_page_to_lba(bs, 0),
3795 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3796 					  _spdk_bs_destroy_trim_cpl, ctx);
3797 }
3798 
3799 /* END spdk_bs_destroy */
3800 
3801 /* START spdk_bs_unload */
3802 
3803 static void
3804 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3805 {
3806 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3807 
3808 	spdk_free(ctx->super);
3809 
3810 	/*
3811 	 * We need to defer calling spdk_bs_call_cpl() until after
3812 	 * dev destruction, so tuck these away for later use.
3813 	 */
3814 	ctx->bs->unload_err = bserrno;
3815 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3816 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3817 
3818 	spdk_bs_sequence_finish(seq, bserrno);
3819 
3820 	_spdk_bs_free(ctx->bs);
3821 	free(ctx);
3822 }
3823 
3824 static void
3825 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3826 {
3827 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3828 
3829 	spdk_free(ctx->mask);
3830 	ctx->super->clean = 1;
3831 
3832 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3833 }
3834 
3835 static void
3836 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3837 {
3838 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3839 
3840 	spdk_free(ctx->mask);
3841 	ctx->mask = NULL;
3842 
3843 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3844 }
3845 
3846 static void
3847 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3848 {
3849 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3850 
3851 	spdk_free(ctx->mask);
3852 	ctx->mask = NULL;
3853 
3854 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3855 }
3856 
3857 static void
3858 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3859 {
3860 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3861 }
3862 
3863 void
3864 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3865 {
3866 	struct spdk_bs_cpl	cpl;
3867 	spdk_bs_sequence_t	*seq;
3868 	struct spdk_bs_load_ctx *ctx;
3869 
3870 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3871 
3872 	if (!TAILQ_EMPTY(&bs->blobs)) {
3873 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3874 		cb_fn(cb_arg, -EBUSY);
3875 		return;
3876 	}
3877 
3878 	ctx = calloc(1, sizeof(*ctx));
3879 	if (!ctx) {
3880 		cb_fn(cb_arg, -ENOMEM);
3881 		return;
3882 	}
3883 
3884 	ctx->bs = bs;
3885 
3886 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3887 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3888 	if (!ctx->super) {
3889 		free(ctx);
3890 		cb_fn(cb_arg, -ENOMEM);
3891 		return;
3892 	}
3893 
3894 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3895 	cpl.u.bs_basic.cb_fn = cb_fn;
3896 	cpl.u.bs_basic.cb_arg = cb_arg;
3897 
3898 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3899 	if (!seq) {
3900 		spdk_free(ctx->super);
3901 		free(ctx);
3902 		cb_fn(cb_arg, -ENOMEM);
3903 		return;
3904 	}
3905 
3906 	/* Read super block */
3907 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3908 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3909 				  _spdk_bs_unload_read_super_cpl, ctx);
3910 }
3911 
3912 /* END spdk_bs_unload */
3913 
3914 /* START spdk_bs_set_super */
3915 
3916 struct spdk_bs_set_super_ctx {
3917 	struct spdk_blob_store		*bs;
3918 	struct spdk_bs_super_block	*super;
3919 };
3920 
3921 static void
3922 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3923 {
3924 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3925 
3926 	if (bserrno != 0) {
3927 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3928 	}
3929 
3930 	spdk_free(ctx->super);
3931 
3932 	spdk_bs_sequence_finish(seq, bserrno);
3933 
3934 	free(ctx);
3935 }
3936 
3937 static void
3938 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3939 {
3940 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3941 
3942 	if (bserrno != 0) {
3943 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3944 		spdk_free(ctx->super);
3945 		spdk_bs_sequence_finish(seq, bserrno);
3946 		free(ctx);
3947 		return;
3948 	}
3949 
3950 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3951 }
3952 
3953 void
3954 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3955 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3956 {
3957 	struct spdk_bs_cpl		cpl;
3958 	spdk_bs_sequence_t		*seq;
3959 	struct spdk_bs_set_super_ctx	*ctx;
3960 
3961 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3962 
3963 	ctx = calloc(1, sizeof(*ctx));
3964 	if (!ctx) {
3965 		cb_fn(cb_arg, -ENOMEM);
3966 		return;
3967 	}
3968 
3969 	ctx->bs = bs;
3970 
3971 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3972 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3973 	if (!ctx->super) {
3974 		free(ctx);
3975 		cb_fn(cb_arg, -ENOMEM);
3976 		return;
3977 	}
3978 
3979 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3980 	cpl.u.bs_basic.cb_fn = cb_fn;
3981 	cpl.u.bs_basic.cb_arg = cb_arg;
3982 
3983 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3984 	if (!seq) {
3985 		spdk_free(ctx->super);
3986 		free(ctx);
3987 		cb_fn(cb_arg, -ENOMEM);
3988 		return;
3989 	}
3990 
3991 	bs->super_blob = blobid;
3992 
3993 	/* Read super block */
3994 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3995 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3996 				  _spdk_bs_set_super_read_cpl, ctx);
3997 }
3998 
3999 /* END spdk_bs_set_super */
4000 
4001 void
4002 spdk_bs_get_super(struct spdk_blob_store *bs,
4003 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4004 {
4005 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
4006 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
4007 	} else {
4008 		cb_fn(cb_arg, bs->super_blob, 0);
4009 	}
4010 }
4011 
4012 uint64_t
4013 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
4014 {
4015 	return bs->cluster_sz;
4016 }
4017 
4018 uint64_t
4019 spdk_bs_get_page_size(struct spdk_blob_store *bs)
4020 {
4021 	return SPDK_BS_PAGE_SIZE;
4022 }
4023 
4024 uint64_t
4025 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
4026 {
4027 	return bs->io_unit_size;
4028 }
4029 
4030 uint64_t
4031 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
4032 {
4033 	return bs->num_free_clusters;
4034 }
4035 
4036 uint64_t
4037 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
4038 {
4039 	return bs->total_data_clusters;
4040 }
4041 
4042 static int
4043 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
4044 {
4045 	bs->md_channel = spdk_get_io_channel(bs);
4046 	if (!bs->md_channel) {
4047 		SPDK_ERRLOG("Failed to get IO channel.\n");
4048 		return -1;
4049 	}
4050 
4051 	return 0;
4052 }
4053 
4054 static int
4055 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
4056 {
4057 	spdk_put_io_channel(bs->md_channel);
4058 
4059 	return 0;
4060 }
4061 
4062 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4063 {
4064 	assert(blob != NULL);
4065 
4066 	return blob->id;
4067 }
4068 
4069 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4070 {
4071 	assert(blob != NULL);
4072 
4073 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4074 }
4075 
4076 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
4077 {
4078 	assert(blob != NULL);
4079 
4080 	return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs);
4081 }
4082 
4083 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4084 {
4085 	assert(blob != NULL);
4086 
4087 	return blob->active.num_clusters;
4088 }
4089 
4090 /* START spdk_bs_create_blob */
4091 
4092 static void
4093 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4094 {
4095 	struct spdk_blob *blob = cb_arg;
4096 
4097 	_spdk_blob_free(blob);
4098 
4099 	spdk_bs_sequence_finish(seq, bserrno);
4100 }
4101 
4102 static int
4103 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4104 		      bool internal)
4105 {
4106 	uint64_t i;
4107 	size_t value_len = 0;
4108 	int rc;
4109 	const void *value = NULL;
4110 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4111 		return -EINVAL;
4112 	}
4113 	for (i = 0; i < xattrs->count; i++) {
4114 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4115 		if (value == NULL || value_len == 0) {
4116 			return -EINVAL;
4117 		}
4118 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4119 		if (rc < 0) {
4120 			return rc;
4121 		}
4122 	}
4123 	return 0;
4124 }
4125 
4126 static void
4127 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4128 {
4129 	_spdk_blob_verify_md_op(blob);
4130 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4131 	blob->state = SPDK_BLOB_STATE_DIRTY;
4132 }
4133 
4134 static void
4135 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4136 		     const struct spdk_blob_opts *opts,
4137 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4138 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4139 {
4140 	struct spdk_blob	*blob;
4141 	uint32_t		page_idx;
4142 	struct spdk_bs_cpl	cpl;
4143 	struct spdk_blob_opts	opts_default;
4144 	struct spdk_blob_xattr_opts internal_xattrs_default;
4145 	spdk_bs_sequence_t	*seq;
4146 	spdk_blob_id		id;
4147 	int rc;
4148 
4149 	assert(spdk_get_thread() == bs->md_thread);
4150 
4151 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4152 	if (page_idx == UINT32_MAX) {
4153 		cb_fn(cb_arg, 0, -ENOMEM);
4154 		return;
4155 	}
4156 	spdk_bit_array_set(bs->used_blobids, page_idx);
4157 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4158 
4159 	id = _spdk_bs_page_to_blobid(page_idx);
4160 
4161 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4162 
4163 	blob = _spdk_blob_alloc(bs, id);
4164 	if (!blob) {
4165 		cb_fn(cb_arg, 0, -ENOMEM);
4166 		return;
4167 	}
4168 
4169 	if (!opts) {
4170 		spdk_blob_opts_init(&opts_default);
4171 		opts = &opts_default;
4172 	}
4173 	if (!internal_xattrs) {
4174 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4175 		internal_xattrs = &internal_xattrs_default;
4176 	}
4177 
4178 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4179 	if (rc < 0) {
4180 		_spdk_blob_free(blob);
4181 		cb_fn(cb_arg, 0, rc);
4182 		return;
4183 	}
4184 
4185 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4186 	if (rc < 0) {
4187 		_spdk_blob_free(blob);
4188 		cb_fn(cb_arg, 0, rc);
4189 		return;
4190 	}
4191 
4192 	if (opts->thin_provision) {
4193 		_spdk_blob_set_thin_provision(blob);
4194 	}
4195 
4196 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4197 	if (rc < 0) {
4198 		_spdk_blob_free(blob);
4199 		cb_fn(cb_arg, 0, rc);
4200 		return;
4201 	}
4202 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4203 	cpl.u.blobid.cb_fn = cb_fn;
4204 	cpl.u.blobid.cb_arg = cb_arg;
4205 	cpl.u.blobid.blobid = blob->id;
4206 
4207 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4208 	if (!seq) {
4209 		_spdk_blob_free(blob);
4210 		cb_fn(cb_arg, 0, -ENOMEM);
4211 		return;
4212 	}
4213 
4214 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4215 }
4216 
4217 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4218 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4219 {
4220 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4221 }
4222 
4223 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4224 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4225 {
4226 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4227 }
4228 
4229 /* END spdk_bs_create_blob */
4230 
4231 /* START blob_cleanup */
4232 
4233 struct spdk_clone_snapshot_ctx {
4234 	struct spdk_bs_cpl      cpl;
4235 	int bserrno;
4236 	bool frozen;
4237 
4238 	struct spdk_io_channel *channel;
4239 
4240 	/* Current cluster for inflate operation */
4241 	uint64_t cluster;
4242 
4243 	/* For inflation force allocation of all unallocated clusters and remove
4244 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4245 	bool allocate_all;
4246 
4247 	struct {
4248 		spdk_blob_id id;
4249 		struct spdk_blob *blob;
4250 	} original;
4251 	struct {
4252 		spdk_blob_id id;
4253 		struct spdk_blob *blob;
4254 	} new;
4255 
4256 	/* xattrs specified for snapshot/clones only. They have no impact on
4257 	 * the original blobs xattrs. */
4258 	const struct spdk_blob_xattr_opts *xattrs;
4259 };
4260 
4261 static void
4262 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4263 {
4264 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4265 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4266 
4267 	if (bserrno != 0) {
4268 		if (ctx->bserrno != 0) {
4269 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4270 		} else {
4271 			ctx->bserrno = bserrno;
4272 		}
4273 	}
4274 
4275 	switch (cpl->type) {
4276 	case SPDK_BS_CPL_TYPE_BLOBID:
4277 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4278 		break;
4279 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4280 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4281 		break;
4282 	default:
4283 		SPDK_UNREACHABLE();
4284 		break;
4285 	}
4286 
4287 	free(ctx);
4288 }
4289 
4290 static void
4291 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4292 {
4293 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4294 	struct spdk_blob *origblob = ctx->original.blob;
4295 
4296 	if (bserrno != 0) {
4297 		if (ctx->bserrno != 0) {
4298 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4299 		} else {
4300 			ctx->bserrno = bserrno;
4301 		}
4302 	}
4303 
4304 	ctx->original.id = origblob->id;
4305 	origblob->locked_operation_in_progress = false;
4306 
4307 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4308 }
4309 
4310 static void
4311 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4312 {
4313 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4314 	struct spdk_blob *origblob = ctx->original.blob;
4315 
4316 	if (bserrno != 0) {
4317 		if (ctx->bserrno != 0) {
4318 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4319 		} else {
4320 			ctx->bserrno = bserrno;
4321 		}
4322 	}
4323 
4324 	if (ctx->frozen) {
4325 		/* Unfreeze any outstanding I/O */
4326 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4327 	} else {
4328 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4329 	}
4330 
4331 }
4332 
4333 static void
4334 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4335 {
4336 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4337 	struct spdk_blob *newblob = ctx->new.blob;
4338 
4339 	if (bserrno != 0) {
4340 		if (ctx->bserrno != 0) {
4341 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4342 		} else {
4343 			ctx->bserrno = bserrno;
4344 		}
4345 	}
4346 
4347 	ctx->new.id = newblob->id;
4348 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4349 }
4350 
4351 /* END blob_cleanup */
4352 
4353 /* START spdk_bs_create_snapshot */
4354 
4355 static void
4356 _spdk_bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
4357 {
4358 	uint64_t *cluster_temp;
4359 
4360 	cluster_temp = blob1->active.clusters;
4361 	blob1->active.clusters = blob2->active.clusters;
4362 	blob2->active.clusters = cluster_temp;
4363 }
4364 
4365 static void
4366 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4367 {
4368 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4369 	struct spdk_blob *origblob = ctx->original.blob;
4370 	struct spdk_blob *newblob = ctx->new.blob;
4371 
4372 	if (bserrno != 0) {
4373 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4374 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4375 		return;
4376 	}
4377 
4378 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4379 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4380 	if (bserrno != 0) {
4381 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4382 		return;
4383 	}
4384 
4385 	_spdk_bs_blob_list_add(ctx->original.blob);
4386 
4387 	spdk_blob_set_read_only(newblob);
4388 
4389 	/* sync snapshot metadata */
4390 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4391 }
4392 
4393 static void
4394 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4395 {
4396 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4397 	struct spdk_blob *origblob = ctx->original.blob;
4398 	struct spdk_blob *newblob = ctx->new.blob;
4399 
4400 	if (bserrno != 0) {
4401 		/* return cluster map back to original */
4402 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4403 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4404 		return;
4405 	}
4406 
4407 	/* Set internal xattr for snapshot id */
4408 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4409 	if (bserrno != 0) {
4410 		/* return cluster map back to original */
4411 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4412 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4413 		return;
4414 	}
4415 
4416 	_spdk_bs_blob_list_remove(origblob);
4417 	origblob->parent_id = newblob->id;
4418 
4419 	/* Create new back_bs_dev for snapshot */
4420 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4421 	if (origblob->back_bs_dev == NULL) {
4422 		/* return cluster map back to original */
4423 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4424 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4425 		return;
4426 	}
4427 
4428 	/* set clone blob as thin provisioned */
4429 	_spdk_blob_set_thin_provision(origblob);
4430 
4431 	_spdk_bs_blob_list_add(newblob);
4432 
4433 	/* sync clone metadata */
4434 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4435 }
4436 
4437 static void
4438 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4439 {
4440 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4441 	struct spdk_blob *origblob = ctx->original.blob;
4442 	struct spdk_blob *newblob = ctx->new.blob;
4443 	int bserrno;
4444 
4445 	if (rc != 0) {
4446 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4447 		return;
4448 	}
4449 
4450 	ctx->frozen = true;
4451 
4452 	/* set new back_bs_dev for snapshot */
4453 	newblob->back_bs_dev = origblob->back_bs_dev;
4454 	/* Set invalid flags from origblob */
4455 	newblob->invalid_flags = origblob->invalid_flags;
4456 
4457 	/* inherit parent from original blob if set */
4458 	newblob->parent_id = origblob->parent_id;
4459 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4460 		/* Set internal xattr for snapshot id */
4461 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4462 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4463 		if (bserrno != 0) {
4464 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4465 			return;
4466 		}
4467 	}
4468 
4469 	/* swap cluster maps */
4470 	_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4471 
4472 	/* sync snapshot metadata */
4473 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4474 }
4475 
4476 static void
4477 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4478 {
4479 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4480 	struct spdk_blob *origblob = ctx->original.blob;
4481 	struct spdk_blob *newblob = _blob;
4482 
4483 	if (bserrno != 0) {
4484 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4485 		return;
4486 	}
4487 
4488 	ctx->new.blob = newblob;
4489 
4490 	/* Zero out newblob cluster map */
4491 	memset(newblob->active.clusters, 0,
4492 	       newblob->active.num_clusters * sizeof(newblob->active.clusters));
4493 
4494 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4495 }
4496 
4497 static void
4498 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4499 {
4500 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4501 	struct spdk_blob *origblob = ctx->original.blob;
4502 
4503 	if (bserrno != 0) {
4504 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4505 		return;
4506 	}
4507 
4508 	ctx->new.id = blobid;
4509 	ctx->cpl.u.blobid.blobid = blobid;
4510 
4511 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4512 }
4513 
4514 
4515 static void
4516 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4517 			const void **value, size_t *value_len)
4518 {
4519 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4520 
4521 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4522 	*value = &blob->id;
4523 	*value_len = sizeof(blob->id);
4524 }
4525 
4526 static void
4527 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4528 {
4529 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4530 	struct spdk_blob_opts opts;
4531 	struct spdk_blob_xattr_opts internal_xattrs;
4532 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4533 
4534 	if (bserrno != 0) {
4535 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4536 		return;
4537 	}
4538 
4539 	ctx->original.blob = _blob;
4540 
4541 	if (_blob->data_ro || _blob->md_ro) {
4542 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4543 			      _blob->id);
4544 		ctx->bserrno = -EINVAL;
4545 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4546 		return;
4547 	}
4548 
4549 	if (_blob->locked_operation_in_progress) {
4550 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot - another operation in progress\n");
4551 		ctx->bserrno = -EBUSY;
4552 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4553 		return;
4554 	}
4555 
4556 	_blob->locked_operation_in_progress = true;
4557 
4558 	spdk_blob_opts_init(&opts);
4559 	_spdk_blob_xattrs_init(&internal_xattrs);
4560 
4561 	/* Change the size of new blob to the same as in original blob,
4562 	 * but do not allocate clusters */
4563 	opts.thin_provision = true;
4564 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4565 
4566 	/* If there are any xattrs specified for snapshot, set them now */
4567 	if (ctx->xattrs) {
4568 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4569 	}
4570 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4571 	internal_xattrs.count = 1;
4572 	internal_xattrs.ctx = _blob;
4573 	internal_xattrs.names = xattrs_names;
4574 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4575 
4576 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4577 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4578 }
4579 
4580 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4581 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4582 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4583 {
4584 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4585 
4586 	if (!ctx) {
4587 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4588 		return;
4589 	}
4590 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4591 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4592 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4593 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4594 	ctx->bserrno = 0;
4595 	ctx->frozen = false;
4596 	ctx->original.id = blobid;
4597 	ctx->xattrs = snapshot_xattrs;
4598 
4599 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4600 }
4601 /* END spdk_bs_create_snapshot */
4602 
4603 /* START spdk_bs_create_clone */
4604 
4605 static void
4606 _spdk_bs_xattr_clone(void *arg, const char *name,
4607 		     const void **value, size_t *value_len)
4608 {
4609 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4610 
4611 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4612 	*value = &blob->id;
4613 	*value_len = sizeof(blob->id);
4614 }
4615 
4616 static void
4617 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4618 {
4619 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4620 	struct spdk_blob *clone = _blob;
4621 
4622 	ctx->new.blob = clone;
4623 	_spdk_bs_blob_list_add(clone);
4624 
4625 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4626 }
4627 
4628 static void
4629 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4630 {
4631 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4632 
4633 	ctx->cpl.u.blobid.blobid = blobid;
4634 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4635 }
4636 
4637 static void
4638 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4639 {
4640 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4641 	struct spdk_blob_opts		opts;
4642 	struct spdk_blob_xattr_opts internal_xattrs;
4643 	char *xattr_names[] = { BLOB_SNAPSHOT };
4644 
4645 	if (bserrno != 0) {
4646 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4647 		return;
4648 	}
4649 
4650 	ctx->original.blob = _blob;
4651 
4652 	if (!_blob->data_ro || !_blob->md_ro) {
4653 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4654 		ctx->bserrno = -EINVAL;
4655 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4656 		return;
4657 	}
4658 
4659 	if (_blob->locked_operation_in_progress) {
4660 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create clone - another operation in progress\n");
4661 		ctx->bserrno = -EBUSY;
4662 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4663 		return;
4664 	}
4665 
4666 	_blob->locked_operation_in_progress = true;
4667 
4668 	spdk_blob_opts_init(&opts);
4669 	_spdk_blob_xattrs_init(&internal_xattrs);
4670 
4671 	opts.thin_provision = true;
4672 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4673 	if (ctx->xattrs) {
4674 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4675 	}
4676 
4677 	/* Set internal xattr BLOB_SNAPSHOT */
4678 	internal_xattrs.count = 1;
4679 	internal_xattrs.ctx = _blob;
4680 	internal_xattrs.names = xattr_names;
4681 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4682 
4683 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4684 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4685 }
4686 
4687 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4688 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4689 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4690 {
4691 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4692 
4693 	if (!ctx) {
4694 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4695 		return;
4696 	}
4697 
4698 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4699 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4700 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4701 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4702 	ctx->bserrno = 0;
4703 	ctx->xattrs = clone_xattrs;
4704 	ctx->original.id = blobid;
4705 
4706 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4707 }
4708 
4709 /* END spdk_bs_create_clone */
4710 
4711 /* START spdk_bs_inflate_blob */
4712 
4713 static void
4714 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4715 {
4716 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4717 	struct spdk_blob *_blob = ctx->original.blob;
4718 
4719 	if (bserrno != 0) {
4720 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4721 		return;
4722 	}
4723 
4724 	assert(_parent != NULL);
4725 
4726 	_spdk_bs_blob_list_remove(_blob);
4727 	_blob->parent_id = _parent->id;
4728 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4729 			     sizeof(spdk_blob_id), true);
4730 
4731 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4732 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4733 	_spdk_bs_blob_list_add(_blob);
4734 
4735 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4736 }
4737 
4738 static void
4739 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4740 {
4741 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4742 	struct spdk_blob *_blob = ctx->original.blob;
4743 	struct spdk_blob *_parent;
4744 
4745 	if (bserrno != 0) {
4746 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4747 		return;
4748 	}
4749 
4750 	if (ctx->allocate_all) {
4751 		/* remove thin provisioning */
4752 		_spdk_bs_blob_list_remove(_blob);
4753 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4754 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4755 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4756 		_blob->back_bs_dev = NULL;
4757 		_blob->parent_id = SPDK_BLOBID_INVALID;
4758 	} else {
4759 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4760 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4761 			/* We must change the parent of the inflated blob */
4762 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4763 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4764 			return;
4765 		}
4766 
4767 		_spdk_bs_blob_list_remove(_blob);
4768 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4769 		_blob->parent_id = SPDK_BLOBID_INVALID;
4770 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4771 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4772 	}
4773 
4774 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4775 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4776 }
4777 
4778 /* Check if cluster needs allocation */
4779 static inline bool
4780 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4781 {
4782 	struct spdk_blob_bs_dev *b;
4783 
4784 	assert(blob != NULL);
4785 
4786 	if (blob->active.clusters[cluster] != 0) {
4787 		/* Cluster is already allocated */
4788 		return false;
4789 	}
4790 
4791 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4792 		/* Blob have no parent blob */
4793 		return allocate_all;
4794 	}
4795 
4796 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4797 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4798 }
4799 
4800 static void
4801 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4802 {
4803 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4804 	struct spdk_blob *_blob = ctx->original.blob;
4805 	uint64_t offset;
4806 
4807 	if (bserrno != 0) {
4808 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4809 		return;
4810 	}
4811 
4812 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4813 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4814 			break;
4815 		}
4816 	}
4817 
4818 	if (ctx->cluster < _blob->active.num_clusters) {
4819 		offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster);
4820 
4821 		/* We may safely increment a cluster before write */
4822 		ctx->cluster++;
4823 
4824 		/* Use zero length write to touch a cluster */
4825 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4826 				   _spdk_bs_inflate_blob_touch_next, ctx);
4827 	} else {
4828 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4829 	}
4830 }
4831 
4832 static void
4833 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4834 {
4835 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4836 	uint64_t lfc; /* lowest free cluster */
4837 	uint64_t i;
4838 
4839 	if (bserrno != 0) {
4840 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4841 		return;
4842 	}
4843 
4844 	ctx->original.blob = _blob;
4845 
4846 	if (_blob->locked_operation_in_progress) {
4847 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot inflate blob - another operation in progress\n");
4848 		ctx->bserrno = -EBUSY;
4849 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4850 		return;
4851 	}
4852 
4853 	_blob->locked_operation_in_progress = true;
4854 
4855 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4856 		/* This blob have no parent, so we cannot decouple it. */
4857 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4858 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4859 		return;
4860 	}
4861 
4862 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4863 		/* This is not thin provisioned blob. No need to inflate. */
4864 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4865 		return;
4866 	}
4867 
4868 	/* Do two passes - one to verify that we can obtain enough clusters
4869 	 * and another to actually claim them.
4870 	 */
4871 	lfc = 0;
4872 	for (i = 0; i < _blob->active.num_clusters; i++) {
4873 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4874 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4875 			if (lfc == UINT32_MAX) {
4876 				/* No more free clusters. Cannot satisfy the request */
4877 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4878 				return;
4879 			}
4880 			lfc++;
4881 		}
4882 	}
4883 
4884 	ctx->cluster = 0;
4885 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4886 }
4887 
4888 static void
4889 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4890 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4891 {
4892 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4893 
4894 	if (!ctx) {
4895 		cb_fn(cb_arg, -ENOMEM);
4896 		return;
4897 	}
4898 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4899 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4900 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4901 	ctx->bserrno = 0;
4902 	ctx->original.id = blobid;
4903 	ctx->channel = channel;
4904 	ctx->allocate_all = allocate_all;
4905 
4906 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4907 }
4908 
4909 void
4910 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4911 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4912 {
4913 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4914 }
4915 
4916 void
4917 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4918 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4919 {
4920 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
4921 }
4922 /* END spdk_bs_inflate_blob */
4923 
4924 /* START spdk_blob_resize */
4925 struct spdk_bs_resize_ctx {
4926 	spdk_blob_op_complete cb_fn;
4927 	void *cb_arg;
4928 	struct spdk_blob *blob;
4929 	uint64_t sz;
4930 	int rc;
4931 };
4932 
4933 static void
4934 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4935 {
4936 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4937 
4938 	if (rc != 0) {
4939 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4940 	}
4941 
4942 	if (ctx->rc != 0) {
4943 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4944 		rc = ctx->rc;
4945 	}
4946 
4947 	ctx->blob->locked_operation_in_progress = false;
4948 
4949 	ctx->cb_fn(ctx->cb_arg, rc);
4950 	free(ctx);
4951 }
4952 
4953 static void
4954 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4955 {
4956 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4957 
4958 	if (rc != 0) {
4959 		ctx->blob->locked_operation_in_progress = false;
4960 		ctx->cb_fn(ctx->cb_arg, rc);
4961 		free(ctx);
4962 		return;
4963 	}
4964 
4965 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4966 
4967 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4968 }
4969 
4970 void
4971 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4972 {
4973 	struct spdk_bs_resize_ctx *ctx;
4974 
4975 	_spdk_blob_verify_md_op(blob);
4976 
4977 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4978 
4979 	if (blob->md_ro) {
4980 		cb_fn(cb_arg, -EPERM);
4981 		return;
4982 	}
4983 
4984 	if (sz == blob->active.num_clusters) {
4985 		cb_fn(cb_arg, 0);
4986 		return;
4987 	}
4988 
4989 	if (blob->locked_operation_in_progress) {
4990 		cb_fn(cb_arg, -EBUSY);
4991 		return;
4992 	}
4993 
4994 	ctx = calloc(1, sizeof(*ctx));
4995 	if (!ctx) {
4996 		cb_fn(cb_arg, -ENOMEM);
4997 		return;
4998 	}
4999 
5000 	blob->locked_operation_in_progress = true;
5001 	ctx->cb_fn = cb_fn;
5002 	ctx->cb_arg = cb_arg;
5003 	ctx->blob = blob;
5004 	ctx->sz = sz;
5005 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
5006 }
5007 
5008 /* END spdk_blob_resize */
5009 
5010 
5011 /* START spdk_bs_delete_blob */
5012 
5013 static void
5014 _spdk_bs_delete_ebusy_close_cpl(void *cb_arg, int bserrno)
5015 {
5016 	spdk_bs_sequence_t *seq = cb_arg;
5017 
5018 	spdk_bs_sequence_finish(seq, -EBUSY);
5019 }
5020 
5021 static void
5022 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
5023 {
5024 	spdk_bs_sequence_t *seq = cb_arg;
5025 
5026 	spdk_bs_sequence_finish(seq, bserrno);
5027 }
5028 
5029 static void
5030 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5031 {
5032 	struct spdk_blob *blob = cb_arg;
5033 
5034 	if (bserrno != 0) {
5035 		/*
5036 		 * We already removed this blob from the blobstore tailq, so
5037 		 *  we need to free it here since this is the last reference
5038 		 *  to it.
5039 		 */
5040 		_spdk_blob_free(blob);
5041 		_spdk_bs_delete_close_cpl(seq, bserrno);
5042 		return;
5043 	}
5044 
5045 	/*
5046 	 * This will immediately decrement the ref_count and call
5047 	 *  the completion routine since the metadata state is clean.
5048 	 *  By calling spdk_blob_close, we reduce the number of call
5049 	 *  points into code that touches the blob->open_ref count
5050 	 *  and the blobstore's blob list.
5051 	 */
5052 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
5053 }
5054 
5055 static void
5056 _spdk_bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno)
5057 {
5058 	spdk_bs_sequence_t *seq = cb_arg;
5059 	struct spdk_blob_list *snapshot_entry = NULL;
5060 	uint32_t page_num;
5061 
5062 	if (bserrno) {
5063 		SPDK_ERRLOG("Failed to remove blob\n");
5064 		spdk_bs_sequence_finish(seq, bserrno);
5065 		return;
5066 	}
5067 
5068 	/* Remove snapshot from the list */
5069 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
5070 	if (snapshot_entry != NULL) {
5071 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
5072 		free(snapshot_entry);
5073 	}
5074 
5075 	page_num = _spdk_bs_blobid_to_page(blob->id);
5076 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
5077 	blob->state = SPDK_BLOB_STATE_DIRTY;
5078 	blob->active.num_pages = 0;
5079 	_spdk_blob_resize(blob, 0);
5080 
5081 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
5082 }
5083 
5084 static int
5085 _spdk_bs_is_blob_deletable(struct spdk_blob *blob)
5086 {
5087 	struct spdk_blob_list *snapshot_entry = NULL;
5088 
5089 	if (blob->open_ref > 1) {
5090 		/* Someone has this blob open (besides this delete context). */
5091 		return -EBUSY;
5092 	}
5093 
5094 	/* Check if this is a snapshot with clones */
5095 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
5096 	if (snapshot_entry != NULL) {
5097 		/* If snapshot have clones, we cannot remove it */
5098 		if (!TAILQ_EMPTY(&snapshot_entry->clones)) {
5099 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
5100 			return -EBUSY;
5101 		}
5102 	}
5103 
5104 	return 0;
5105 }
5106 
5107 static void
5108 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
5109 {
5110 	spdk_bs_sequence_t *seq = cb_arg;
5111 
5112 	if (bserrno != 0) {
5113 		spdk_bs_sequence_finish(seq, bserrno);
5114 		return;
5115 	}
5116 
5117 	_spdk_blob_verify_md_op(blob);
5118 
5119 	bserrno = _spdk_bs_is_blob_deletable(blob);
5120 	if (bserrno) {
5121 		spdk_blob_close(blob, _spdk_bs_delete_ebusy_close_cpl, seq);
5122 		return;
5123 	}
5124 
5125 	_spdk_bs_blob_list_remove(blob);
5126 
5127 	if (blob->locked_operation_in_progress) {
5128 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress\n");
5129 		spdk_blob_close(blob, _spdk_bs_delete_ebusy_close_cpl, seq);
5130 		return;
5131 	}
5132 
5133 	blob->locked_operation_in_progress = true;
5134 
5135 	/*
5136 	 * Remove the blob from the blob_store list now, to ensure it does not
5137 	 *  get returned after this point by _spdk_blob_lookup().
5138 	 */
5139 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5140 
5141 	_spdk_bs_delete_blob_finish(seq, blob, 0);
5142 }
5143 
5144 void
5145 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5146 		    spdk_blob_op_complete cb_fn, void *cb_arg)
5147 {
5148 	struct spdk_bs_cpl	cpl;
5149 	spdk_bs_sequence_t	*seq;
5150 
5151 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
5152 
5153 	assert(spdk_get_thread() == bs->md_thread);
5154 
5155 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5156 	cpl.u.blob_basic.cb_fn = cb_fn;
5157 	cpl.u.blob_basic.cb_arg = cb_arg;
5158 
5159 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5160 	if (!seq) {
5161 		cb_fn(cb_arg, -ENOMEM);
5162 		return;
5163 	}
5164 
5165 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5166 }
5167 
5168 /* END spdk_bs_delete_blob */
5169 
5170 /* START spdk_bs_open_blob */
5171 
5172 static void
5173 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5174 {
5175 	struct spdk_blob *blob = cb_arg;
5176 
5177 	/* If the blob have crc error, we just return NULL. */
5178 	if (blob == NULL) {
5179 		seq->cpl.u.blob_handle.blob = NULL;
5180 		spdk_bs_sequence_finish(seq, bserrno);
5181 		return;
5182 	}
5183 
5184 	blob->open_ref++;
5185 
5186 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5187 
5188 	spdk_bs_sequence_finish(seq, bserrno);
5189 }
5190 
5191 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5192 			       struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5193 {
5194 	struct spdk_blob		*blob;
5195 	struct spdk_bs_cpl		cpl;
5196 	struct spdk_blob_open_opts	opts_default;
5197 	spdk_bs_sequence_t		*seq;
5198 	uint32_t			page_num;
5199 
5200 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5201 	assert(spdk_get_thread() == bs->md_thread);
5202 
5203 	page_num = _spdk_bs_blobid_to_page(blobid);
5204 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5205 		/* Invalid blobid */
5206 		cb_fn(cb_arg, NULL, -ENOENT);
5207 		return;
5208 	}
5209 
5210 	blob = _spdk_blob_lookup(bs, blobid);
5211 	if (blob) {
5212 		blob->open_ref++;
5213 		cb_fn(cb_arg, blob, 0);
5214 		return;
5215 	}
5216 
5217 	blob = _spdk_blob_alloc(bs, blobid);
5218 	if (!blob) {
5219 		cb_fn(cb_arg, NULL, -ENOMEM);
5220 		return;
5221 	}
5222 
5223 	if (!opts) {
5224 		spdk_blob_open_opts_init(&opts_default);
5225 		opts = &opts_default;
5226 	}
5227 
5228 	blob->clear_method = opts->clear_method;
5229 
5230 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5231 	cpl.u.blob_handle.cb_fn = cb_fn;
5232 	cpl.u.blob_handle.cb_arg = cb_arg;
5233 	cpl.u.blob_handle.blob = blob;
5234 
5235 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5236 	if (!seq) {
5237 		_spdk_blob_free(blob);
5238 		cb_fn(cb_arg, NULL, -ENOMEM);
5239 		return;
5240 	}
5241 
5242 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5243 }
5244 
5245 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5246 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5247 {
5248 	_spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
5249 }
5250 
5251 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
5252 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5253 {
5254 	_spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
5255 }
5256 
5257 /* END spdk_bs_open_blob */
5258 
5259 /* START spdk_blob_set_read_only */
5260 int spdk_blob_set_read_only(struct spdk_blob *blob)
5261 {
5262 	_spdk_blob_verify_md_op(blob);
5263 
5264 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5265 
5266 	blob->state = SPDK_BLOB_STATE_DIRTY;
5267 	return 0;
5268 }
5269 /* END spdk_blob_set_read_only */
5270 
5271 /* START spdk_blob_sync_md */
5272 
5273 static void
5274 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5275 {
5276 	struct spdk_blob *blob = cb_arg;
5277 
5278 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5279 		blob->data_ro = true;
5280 		blob->md_ro = true;
5281 	}
5282 
5283 	spdk_bs_sequence_finish(seq, bserrno);
5284 }
5285 
5286 static void
5287 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5288 {
5289 	struct spdk_bs_cpl	cpl;
5290 	spdk_bs_sequence_t	*seq;
5291 
5292 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5293 	cpl.u.blob_basic.cb_fn = cb_fn;
5294 	cpl.u.blob_basic.cb_arg = cb_arg;
5295 
5296 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5297 	if (!seq) {
5298 		cb_fn(cb_arg, -ENOMEM);
5299 		return;
5300 	}
5301 
5302 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5303 }
5304 
5305 void
5306 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5307 {
5308 	_spdk_blob_verify_md_op(blob);
5309 
5310 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5311 
5312 	if (blob->md_ro) {
5313 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5314 		cb_fn(cb_arg, 0);
5315 		return;
5316 	}
5317 
5318 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5319 }
5320 
5321 /* END spdk_blob_sync_md */
5322 
5323 struct spdk_blob_insert_cluster_ctx {
5324 	struct spdk_thread	*thread;
5325 	struct spdk_blob	*blob;
5326 	uint32_t		cluster_num;	/* cluster index in blob */
5327 	uint32_t		cluster;	/* cluster on disk */
5328 	int			rc;
5329 	spdk_blob_op_complete	cb_fn;
5330 	void			*cb_arg;
5331 };
5332 
5333 static void
5334 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5335 {
5336 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5337 
5338 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5339 	free(ctx);
5340 }
5341 
5342 static void
5343 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5344 {
5345 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5346 
5347 	ctx->rc = bserrno;
5348 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5349 }
5350 
5351 static void
5352 _spdk_blob_insert_cluster_msg(void *arg)
5353 {
5354 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5355 
5356 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5357 	if (ctx->rc != 0) {
5358 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5359 		return;
5360 	}
5361 
5362 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5363 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5364 }
5365 
5366 static void
5367 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5368 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5369 {
5370 	struct spdk_blob_insert_cluster_ctx *ctx;
5371 
5372 	ctx = calloc(1, sizeof(*ctx));
5373 	if (ctx == NULL) {
5374 		cb_fn(cb_arg, -ENOMEM);
5375 		return;
5376 	}
5377 
5378 	ctx->thread = spdk_get_thread();
5379 	ctx->blob = blob;
5380 	ctx->cluster_num = cluster_num;
5381 	ctx->cluster = cluster;
5382 	ctx->cb_fn = cb_fn;
5383 	ctx->cb_arg = cb_arg;
5384 
5385 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5386 }
5387 
5388 /* START spdk_blob_close */
5389 
5390 static void
5391 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5392 {
5393 	struct spdk_blob *blob = cb_arg;
5394 
5395 	if (bserrno == 0) {
5396 		blob->open_ref--;
5397 		if (blob->open_ref == 0) {
5398 			/*
5399 			 * Blobs with active.num_pages == 0 are deleted blobs.
5400 			 *  these blobs are removed from the blob_store list
5401 			 *  when the deletion process starts - so don't try to
5402 			 *  remove them again.
5403 			 */
5404 			if (blob->active.num_pages > 0) {
5405 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5406 			}
5407 			_spdk_blob_free(blob);
5408 		}
5409 	}
5410 
5411 	spdk_bs_sequence_finish(seq, bserrno);
5412 }
5413 
5414 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5415 {
5416 	struct spdk_bs_cpl	cpl;
5417 	spdk_bs_sequence_t	*seq;
5418 
5419 	_spdk_blob_verify_md_op(blob);
5420 
5421 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5422 
5423 	if (blob->open_ref == 0) {
5424 		cb_fn(cb_arg, -EBADF);
5425 		return;
5426 	}
5427 
5428 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5429 	cpl.u.blob_basic.cb_fn = cb_fn;
5430 	cpl.u.blob_basic.cb_arg = cb_arg;
5431 
5432 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5433 	if (!seq) {
5434 		cb_fn(cb_arg, -ENOMEM);
5435 		return;
5436 	}
5437 
5438 	/* Sync metadata */
5439 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5440 }
5441 
5442 /* END spdk_blob_close */
5443 
5444 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5445 {
5446 	return spdk_get_io_channel(bs);
5447 }
5448 
5449 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5450 {
5451 	spdk_put_io_channel(channel);
5452 }
5453 
5454 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5455 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5456 {
5457 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5458 				     SPDK_BLOB_UNMAP);
5459 }
5460 
5461 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5462 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5463 {
5464 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5465 				     SPDK_BLOB_WRITE_ZEROES);
5466 }
5467 
5468 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5469 			void *payload, uint64_t offset, uint64_t length,
5470 			spdk_blob_op_complete cb_fn, void *cb_arg)
5471 {
5472 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5473 				     SPDK_BLOB_WRITE);
5474 }
5475 
5476 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5477 		       void *payload, uint64_t offset, uint64_t length,
5478 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5479 {
5480 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5481 				     SPDK_BLOB_READ);
5482 }
5483 
5484 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5485 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5486 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5487 {
5488 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5489 }
5490 
5491 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5492 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5493 			spdk_blob_op_complete cb_fn, void *cb_arg)
5494 {
5495 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5496 }
5497 
5498 struct spdk_bs_iter_ctx {
5499 	int64_t page_num;
5500 	struct spdk_blob_store *bs;
5501 
5502 	spdk_blob_op_with_handle_complete cb_fn;
5503 	void *cb_arg;
5504 };
5505 
5506 static void
5507 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5508 {
5509 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5510 	struct spdk_blob_store *bs = ctx->bs;
5511 	spdk_blob_id id;
5512 
5513 	if (bserrno == 0) {
5514 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5515 		free(ctx);
5516 		return;
5517 	}
5518 
5519 	ctx->page_num++;
5520 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5521 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5522 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5523 		free(ctx);
5524 		return;
5525 	}
5526 
5527 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5528 
5529 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5530 }
5531 
5532 void
5533 spdk_bs_iter_first(struct spdk_blob_store *bs,
5534 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5535 {
5536 	struct spdk_bs_iter_ctx *ctx;
5537 
5538 	ctx = calloc(1, sizeof(*ctx));
5539 	if (!ctx) {
5540 		cb_fn(cb_arg, NULL, -ENOMEM);
5541 		return;
5542 	}
5543 
5544 	ctx->page_num = -1;
5545 	ctx->bs = bs;
5546 	ctx->cb_fn = cb_fn;
5547 	ctx->cb_arg = cb_arg;
5548 
5549 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5550 }
5551 
5552 static void
5553 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5554 {
5555 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5556 
5557 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5558 }
5559 
5560 void
5561 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5562 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5563 {
5564 	struct spdk_bs_iter_ctx *ctx;
5565 
5566 	assert(blob != NULL);
5567 
5568 	ctx = calloc(1, sizeof(*ctx));
5569 	if (!ctx) {
5570 		cb_fn(cb_arg, NULL, -ENOMEM);
5571 		return;
5572 	}
5573 
5574 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5575 	ctx->bs = bs;
5576 	ctx->cb_fn = cb_fn;
5577 	ctx->cb_arg = cb_arg;
5578 
5579 	/* Close the existing blob */
5580 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5581 }
5582 
5583 static int
5584 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5585 		     uint16_t value_len, bool internal)
5586 {
5587 	struct spdk_xattr_tailq *xattrs;
5588 	struct spdk_xattr	*xattr;
5589 
5590 	_spdk_blob_verify_md_op(blob);
5591 
5592 	if (blob->md_ro) {
5593 		return -EPERM;
5594 	}
5595 
5596 	if (internal) {
5597 		xattrs = &blob->xattrs_internal;
5598 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5599 	} else {
5600 		xattrs = &blob->xattrs;
5601 	}
5602 
5603 	TAILQ_FOREACH(xattr, xattrs, link) {
5604 		if (!strcmp(name, xattr->name)) {
5605 			free(xattr->value);
5606 			xattr->value_len = value_len;
5607 			xattr->value = malloc(value_len);
5608 			memcpy(xattr->value, value, value_len);
5609 
5610 			blob->state = SPDK_BLOB_STATE_DIRTY;
5611 
5612 			return 0;
5613 		}
5614 	}
5615 
5616 	xattr = calloc(1, sizeof(*xattr));
5617 	if (!xattr) {
5618 		return -ENOMEM;
5619 	}
5620 	xattr->name = strdup(name);
5621 	xattr->value_len = value_len;
5622 	xattr->value = malloc(value_len);
5623 	memcpy(xattr->value, value, value_len);
5624 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5625 
5626 	blob->state = SPDK_BLOB_STATE_DIRTY;
5627 
5628 	return 0;
5629 }
5630 
5631 int
5632 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5633 		    uint16_t value_len)
5634 {
5635 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5636 }
5637 
5638 static int
5639 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5640 {
5641 	struct spdk_xattr_tailq *xattrs;
5642 	struct spdk_xattr	*xattr;
5643 
5644 	_spdk_blob_verify_md_op(blob);
5645 
5646 	if (blob->md_ro) {
5647 		return -EPERM;
5648 	}
5649 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5650 
5651 	TAILQ_FOREACH(xattr, xattrs, link) {
5652 		if (!strcmp(name, xattr->name)) {
5653 			TAILQ_REMOVE(xattrs, xattr, link);
5654 			free(xattr->value);
5655 			free(xattr->name);
5656 			free(xattr);
5657 
5658 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5659 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5660 			}
5661 			blob->state = SPDK_BLOB_STATE_DIRTY;
5662 
5663 			return 0;
5664 		}
5665 	}
5666 
5667 	return -ENOENT;
5668 }
5669 
5670 int
5671 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5672 {
5673 	return _spdk_blob_remove_xattr(blob, name, false);
5674 }
5675 
5676 static int
5677 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5678 			   const void **value, size_t *value_len, bool internal)
5679 {
5680 	struct spdk_xattr	*xattr;
5681 	struct spdk_xattr_tailq *xattrs;
5682 
5683 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5684 
5685 	TAILQ_FOREACH(xattr, xattrs, link) {
5686 		if (!strcmp(name, xattr->name)) {
5687 			*value = xattr->value;
5688 			*value_len = xattr->value_len;
5689 			return 0;
5690 		}
5691 	}
5692 	return -ENOENT;
5693 }
5694 
5695 int
5696 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5697 			  const void **value, size_t *value_len)
5698 {
5699 	_spdk_blob_verify_md_op(blob);
5700 
5701 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5702 }
5703 
5704 struct spdk_xattr_names {
5705 	uint32_t	count;
5706 	const char	*names[0];
5707 };
5708 
5709 static int
5710 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5711 {
5712 	struct spdk_xattr	*xattr;
5713 	int			count = 0;
5714 
5715 	TAILQ_FOREACH(xattr, xattrs, link) {
5716 		count++;
5717 	}
5718 
5719 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5720 	if (*names == NULL) {
5721 		return -ENOMEM;
5722 	}
5723 
5724 	TAILQ_FOREACH(xattr, xattrs, link) {
5725 		(*names)->names[(*names)->count++] = xattr->name;
5726 	}
5727 
5728 	return 0;
5729 }
5730 
5731 int
5732 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5733 {
5734 	_spdk_blob_verify_md_op(blob);
5735 
5736 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5737 }
5738 
5739 uint32_t
5740 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5741 {
5742 	assert(names != NULL);
5743 
5744 	return names->count;
5745 }
5746 
5747 const char *
5748 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5749 {
5750 	if (index >= names->count) {
5751 		return NULL;
5752 	}
5753 
5754 	return names->names[index];
5755 }
5756 
5757 void
5758 spdk_xattr_names_free(struct spdk_xattr_names *names)
5759 {
5760 	free(names);
5761 }
5762 
5763 struct spdk_bs_type
5764 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5765 {
5766 	return bs->bstype;
5767 }
5768 
5769 void
5770 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5771 {
5772 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5773 }
5774 
5775 bool
5776 spdk_blob_is_read_only(struct spdk_blob *blob)
5777 {
5778 	assert(blob != NULL);
5779 	return (blob->data_ro || blob->md_ro);
5780 }
5781 
5782 bool
5783 spdk_blob_is_snapshot(struct spdk_blob *blob)
5784 {
5785 	struct spdk_blob_list *snapshot_entry;
5786 
5787 	assert(blob != NULL);
5788 
5789 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
5790 	if (snapshot_entry == NULL) {
5791 		return false;
5792 	}
5793 
5794 	return true;
5795 }
5796 
5797 bool
5798 spdk_blob_is_clone(struct spdk_blob *blob)
5799 {
5800 	assert(blob != NULL);
5801 
5802 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5803 		assert(spdk_blob_is_thin_provisioned(blob));
5804 		return true;
5805 	}
5806 
5807 	return false;
5808 }
5809 
5810 bool
5811 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5812 {
5813 	assert(blob != NULL);
5814 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5815 }
5816 
5817 spdk_blob_id
5818 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5819 {
5820 	struct spdk_blob_list *snapshot_entry = NULL;
5821 	struct spdk_blob_list *clone_entry = NULL;
5822 
5823 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5824 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5825 			if (clone_entry->id == blob_id) {
5826 				return snapshot_entry->id;
5827 			}
5828 		}
5829 	}
5830 
5831 	return SPDK_BLOBID_INVALID;
5832 }
5833 
5834 int
5835 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5836 		     size_t *count)
5837 {
5838 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5839 	size_t n;
5840 
5841 	snapshot_entry = _spdk_bs_get_snapshot_entry(bs, blobid);
5842 	if (snapshot_entry == NULL) {
5843 		*count = 0;
5844 		return 0;
5845 	}
5846 
5847 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5848 		*count = snapshot_entry->clone_count;
5849 		return -ENOMEM;
5850 	}
5851 	*count = snapshot_entry->clone_count;
5852 
5853 	n = 0;
5854 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5855 		ids[n++] = clone_entry->id;
5856 	}
5857 
5858 	return 0;
5859 }
5860 
5861 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5862