xref: /spdk/lib/blob/blobstore.c (revision bb488d2829a9b7863daab45917dd2174905cc0ae)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "spdk/blob.h"
37 #include "spdk/crc32.h"
38 #include "spdk/env.h"
39 #include "spdk/queue.h"
40 #include "spdk/thread.h"
41 #include "spdk/bit_array.h"
42 #include "spdk/likely.h"
43 #include "spdk/util.h"
44 
45 #include "spdk_internal/assert.h"
46 #include "spdk_internal/log.h"
47 
48 #include "blobstore.h"
49 
50 #define BLOB_CRC32C_INITIAL    0xffffffffUL
51 
52 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs);
53 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs);
54 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
55 static void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
56 		uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg);
57 
58 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
59 				uint16_t value_len, bool internal);
60 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
61 				      const void **value, size_t *value_len, bool internal);
62 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal);
63 
64 static void
65 _spdk_blob_verify_md_op(struct spdk_blob *blob)
66 {
67 	assert(blob != NULL);
68 	assert(spdk_get_thread() == blob->bs->md_thread);
69 	assert(blob->state != SPDK_BLOB_STATE_LOADING);
70 }
71 
72 static void
73 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
74 {
75 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
76 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false);
77 	assert(bs->num_free_clusters > 0);
78 
79 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num);
80 
81 	spdk_bit_array_set(bs->used_clusters, cluster_num);
82 	bs->num_free_clusters--;
83 }
84 
85 static int
86 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
87 {
88 	uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
89 
90 	_spdk_blob_verify_md_op(blob);
91 
92 	if (*cluster_lba != 0) {
93 		return -EEXIST;
94 	}
95 
96 	*cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster);
97 	return 0;
98 }
99 
100 static int
101 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
102 			  uint64_t *lowest_free_cluster, bool update_map)
103 {
104 	pthread_mutex_lock(&blob->bs->used_clusters_mutex);
105 	*lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters,
106 			       *lowest_free_cluster);
107 	if (*lowest_free_cluster == UINT32_MAX) {
108 		/* No more free clusters. Cannot satisfy the request */
109 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
110 		return -ENOSPC;
111 	}
112 
113 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id);
114 	_spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster);
115 	pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
116 
117 	if (update_map) {
118 		_spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster);
119 	}
120 
121 	return 0;
122 }
123 
124 static void
125 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
126 {
127 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
128 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true);
129 	assert(bs->num_free_clusters < bs->total_clusters);
130 
131 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num);
132 
133 	pthread_mutex_lock(&bs->used_clusters_mutex);
134 	spdk_bit_array_clear(bs->used_clusters, cluster_num);
135 	bs->num_free_clusters++;
136 	pthread_mutex_unlock(&bs->used_clusters_mutex);
137 }
138 
139 static void
140 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
141 {
142 	xattrs->count = 0;
143 	xattrs->names = NULL;
144 	xattrs->ctx = NULL;
145 	xattrs->get_value = NULL;
146 }
147 
148 void
149 spdk_blob_opts_init(struct spdk_blob_opts *opts)
150 {
151 	opts->num_clusters = 0;
152 	opts->thin_provision = false;
153 	_spdk_blob_xattrs_init(&opts->xattrs);
154 }
155 
156 void
157 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts)
158 {
159 	opts->clear_method = BLOB_CLEAR_WITH_UNMAP;
160 }
161 
162 static struct spdk_blob *
163 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
164 {
165 	struct spdk_blob *blob;
166 
167 	blob = calloc(1, sizeof(*blob));
168 	if (!blob) {
169 		return NULL;
170 	}
171 
172 	blob->id = id;
173 	blob->bs = bs;
174 
175 	blob->parent_id = SPDK_BLOBID_INVALID;
176 
177 	blob->state = SPDK_BLOB_STATE_DIRTY;
178 	blob->active.num_pages = 1;
179 	blob->active.pages = calloc(1, sizeof(*blob->active.pages));
180 	if (!blob->active.pages) {
181 		free(blob);
182 		return NULL;
183 	}
184 
185 	blob->active.pages[0] = _spdk_bs_blobid_to_page(id);
186 
187 	TAILQ_INIT(&blob->xattrs);
188 	TAILQ_INIT(&blob->xattrs_internal);
189 
190 	return blob;
191 }
192 
193 static void
194 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs)
195 {
196 	struct spdk_xattr	*xattr, *xattr_tmp;
197 
198 	TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
199 		TAILQ_REMOVE(xattrs, xattr, link);
200 		free(xattr->name);
201 		free(xattr->value);
202 		free(xattr);
203 	}
204 }
205 
206 static void
207 _spdk_blob_free(struct spdk_blob *blob)
208 {
209 	assert(blob != NULL);
210 
211 	free(blob->active.clusters);
212 	free(blob->clean.clusters);
213 	free(blob->active.pages);
214 	free(blob->clean.pages);
215 
216 	_spdk_xattrs_free(&blob->xattrs);
217 	_spdk_xattrs_free(&blob->xattrs_internal);
218 
219 	if (blob->back_bs_dev) {
220 		blob->back_bs_dev->destroy(blob->back_bs_dev);
221 	}
222 
223 	free(blob);
224 }
225 
226 struct freeze_io_ctx {
227 	struct spdk_bs_cpl cpl;
228 	struct spdk_blob *blob;
229 };
230 
231 static void
232 _spdk_blob_io_sync(struct spdk_io_channel_iter *i)
233 {
234 	spdk_for_each_channel_continue(i, 0);
235 }
236 
237 static void
238 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i)
239 {
240 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
241 	struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
242 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
243 	struct spdk_bs_request_set	*set;
244 	struct spdk_bs_user_op_args	*args;
245 	spdk_bs_user_op_t *op, *tmp;
246 
247 	TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
248 		set = (struct spdk_bs_request_set *)op;
249 		args = &set->u.user_op;
250 
251 		if (args->blob == ctx->blob) {
252 			TAILQ_REMOVE(&ch->queued_io, op, link);
253 			spdk_bs_user_op_execute(op);
254 		}
255 	}
256 
257 	spdk_for_each_channel_continue(i, 0);
258 }
259 
260 static void
261 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status)
262 {
263 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
264 
265 	ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
266 
267 	free(ctx);
268 }
269 
270 static void
271 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
272 {
273 	struct freeze_io_ctx *ctx;
274 
275 	ctx = calloc(1, sizeof(*ctx));
276 	if (!ctx) {
277 		cb_fn(cb_arg, -ENOMEM);
278 		return;
279 	}
280 
281 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
282 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
283 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
284 	ctx->blob = blob;
285 
286 	/* Freeze I/O on blob */
287 	blob->frozen_refcnt++;
288 
289 	if (blob->frozen_refcnt == 1) {
290 		spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl);
291 	} else {
292 		cb_fn(cb_arg, 0);
293 		free(ctx);
294 	}
295 }
296 
297 static void
298 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
299 {
300 	struct freeze_io_ctx *ctx;
301 
302 	ctx = calloc(1, sizeof(*ctx));
303 	if (!ctx) {
304 		cb_fn(cb_arg, -ENOMEM);
305 		return;
306 	}
307 
308 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
309 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
310 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
311 	ctx->blob = blob;
312 
313 	assert(blob->frozen_refcnt > 0);
314 
315 	blob->frozen_refcnt--;
316 
317 	if (blob->frozen_refcnt == 0) {
318 		spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl);
319 	} else {
320 		cb_fn(cb_arg, 0);
321 		free(ctx);
322 	}
323 }
324 
325 static int
326 _spdk_blob_mark_clean(struct spdk_blob *blob)
327 {
328 	uint64_t *clusters = NULL;
329 	uint32_t *pages = NULL;
330 
331 	assert(blob != NULL);
332 
333 	if (blob->active.num_clusters) {
334 		assert(blob->active.clusters);
335 		clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
336 		if (!clusters) {
337 			return -ENOMEM;
338 		}
339 		memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*clusters));
340 	}
341 
342 	if (blob->active.num_pages) {
343 		assert(blob->active.pages);
344 		pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
345 		if (!pages) {
346 			free(clusters);
347 			return -ENOMEM;
348 		}
349 		memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*pages));
350 	}
351 
352 	free(blob->clean.clusters);
353 	free(blob->clean.pages);
354 
355 	blob->clean.num_clusters = blob->active.num_clusters;
356 	blob->clean.clusters = blob->active.clusters;
357 	blob->clean.num_pages = blob->active.num_pages;
358 	blob->clean.pages = blob->active.pages;
359 
360 	blob->active.clusters = clusters;
361 	blob->active.pages = pages;
362 
363 	/* If the metadata was dirtied again while the metadata was being written to disk,
364 	 *  we do not want to revert the DIRTY state back to CLEAN here.
365 	 */
366 	if (blob->state == SPDK_BLOB_STATE_LOADING) {
367 		blob->state = SPDK_BLOB_STATE_CLEAN;
368 	}
369 
370 	return 0;
371 }
372 
373 static int
374 _spdk_blob_deserialize_xattr(struct spdk_blob *blob,
375 			     struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
376 {
377 	struct spdk_xattr                       *xattr;
378 
379 	if (desc_xattr->length != sizeof(desc_xattr->name_length) +
380 	    sizeof(desc_xattr->value_length) +
381 	    desc_xattr->name_length + desc_xattr->value_length) {
382 		return -EINVAL;
383 	}
384 
385 	xattr = calloc(1, sizeof(*xattr));
386 	if (xattr == NULL) {
387 		return -ENOMEM;
388 	}
389 
390 	xattr->name = malloc(desc_xattr->name_length + 1);
391 	if (xattr->name == NULL) {
392 		free(xattr);
393 		return -ENOMEM;
394 	}
395 	memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
396 	xattr->name[desc_xattr->name_length] = '\0';
397 
398 	xattr->value = malloc(desc_xattr->value_length);
399 	if (xattr->value == NULL) {
400 		free(xattr->name);
401 		free(xattr);
402 		return -ENOMEM;
403 	}
404 	xattr->value_len = desc_xattr->value_length;
405 	memcpy(xattr->value,
406 	       (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
407 	       desc_xattr->value_length);
408 
409 	TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
410 
411 	return 0;
412 }
413 
414 
415 static int
416 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
417 {
418 	struct spdk_blob_md_descriptor *desc;
419 	size_t	cur_desc = 0;
420 	void *tmp;
421 
422 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
423 	while (cur_desc < sizeof(page->descriptors)) {
424 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
425 			if (desc->length == 0) {
426 				/* If padding and length are 0, this terminates the page */
427 				break;
428 			}
429 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
430 			struct spdk_blob_md_descriptor_flags	*desc_flags;
431 
432 			desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
433 
434 			if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
435 				return -EINVAL;
436 			}
437 
438 			if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
439 			    SPDK_BLOB_INVALID_FLAGS_MASK) {
440 				return -EINVAL;
441 			}
442 
443 			if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
444 			    SPDK_BLOB_DATA_RO_FLAGS_MASK) {
445 				blob->data_ro = true;
446 				blob->md_ro = true;
447 			}
448 
449 			if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
450 			    SPDK_BLOB_MD_RO_FLAGS_MASK) {
451 				blob->md_ro = true;
452 			}
453 
454 			if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
455 				blob->data_ro = true;
456 				blob->md_ro = true;
457 			}
458 
459 			blob->invalid_flags = desc_flags->invalid_flags;
460 			blob->data_ro_flags = desc_flags->data_ro_flags;
461 			blob->md_ro_flags = desc_flags->md_ro_flags;
462 
463 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
464 			struct spdk_blob_md_descriptor_extent	*desc_extent;
465 			unsigned int				i, j;
466 			unsigned int				cluster_count = blob->active.num_clusters;
467 
468 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
469 
470 			if (desc_extent->length == 0 ||
471 			    (desc_extent->length % sizeof(desc_extent->extents[0]) != 0)) {
472 				return -EINVAL;
473 			}
474 
475 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
476 				for (j = 0; j < desc_extent->extents[i].length; j++) {
477 					if (desc_extent->extents[i].cluster_idx != 0) {
478 						if (!spdk_bit_array_get(blob->bs->used_clusters,
479 									desc_extent->extents[i].cluster_idx + j)) {
480 							return -EINVAL;
481 						}
482 					}
483 					cluster_count++;
484 				}
485 			}
486 
487 			if (cluster_count == 0) {
488 				return -EINVAL;
489 			}
490 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t));
491 			if (tmp == NULL) {
492 				return -ENOMEM;
493 			}
494 			blob->active.clusters = tmp;
495 			blob->active.cluster_array_size = cluster_count;
496 
497 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
498 				for (j = 0; j < desc_extent->extents[i].length; j++) {
499 					if (desc_extent->extents[i].cluster_idx != 0) {
500 						blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
501 								desc_extent->extents[i].cluster_idx + j);
502 					} else if (spdk_blob_is_thin_provisioned(blob)) {
503 						blob->active.clusters[blob->active.num_clusters++] = 0;
504 					} else {
505 						return -EINVAL;
506 					}
507 				}
508 			}
509 
510 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
511 			int rc;
512 
513 			rc = _spdk_blob_deserialize_xattr(blob,
514 							  (struct spdk_blob_md_descriptor_xattr *) desc, false);
515 			if (rc != 0) {
516 				return rc;
517 			}
518 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
519 			int rc;
520 
521 			rc = _spdk_blob_deserialize_xattr(blob,
522 							  (struct spdk_blob_md_descriptor_xattr *) desc, true);
523 			if (rc != 0) {
524 				return rc;
525 			}
526 		} else {
527 			/* Unrecognized descriptor type.  Do not fail - just continue to the
528 			 *  next descriptor.  If this descriptor is associated with some feature
529 			 *  defined in a newer version of blobstore, that version of blobstore
530 			 *  should create and set an associated feature flag to specify if this
531 			 *  blob can be loaded or not.
532 			 */
533 		}
534 
535 		/* Advance to the next descriptor */
536 		cur_desc += sizeof(*desc) + desc->length;
537 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
538 			break;
539 		}
540 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
541 	}
542 
543 	return 0;
544 }
545 
546 static int
547 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
548 		 struct spdk_blob *blob)
549 {
550 	const struct spdk_blob_md_page *page;
551 	uint32_t i;
552 	int rc;
553 
554 	assert(page_count > 0);
555 	assert(pages[0].sequence_num == 0);
556 	assert(blob != NULL);
557 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
558 	assert(blob->active.clusters == NULL);
559 
560 	/* The blobid provided doesn't match what's in the MD, this can
561 	 * happen for example if a bogus blobid is passed in through open.
562 	 */
563 	if (blob->id != pages[0].id) {
564 		SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n",
565 			    blob->id, pages[0].id);
566 		return -ENOENT;
567 	}
568 
569 	for (i = 0; i < page_count; i++) {
570 		page = &pages[i];
571 
572 		assert(page->id == blob->id);
573 		assert(page->sequence_num == i);
574 
575 		rc = _spdk_blob_parse_page(page, blob);
576 		if (rc != 0) {
577 			return rc;
578 		}
579 	}
580 
581 	return 0;
582 }
583 
584 static int
585 _spdk_blob_serialize_add_page(const struct spdk_blob *blob,
586 			      struct spdk_blob_md_page **pages,
587 			      uint32_t *page_count,
588 			      struct spdk_blob_md_page **last_page)
589 {
590 	struct spdk_blob_md_page *page;
591 
592 	assert(pages != NULL);
593 	assert(page_count != NULL);
594 
595 	if (*page_count == 0) {
596 		assert(*pages == NULL);
597 		*page_count = 1;
598 		*pages = spdk_malloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
599 				     NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
600 	} else {
601 		assert(*pages != NULL);
602 		(*page_count)++;
603 		*pages = spdk_realloc(*pages,
604 				      SPDK_BS_PAGE_SIZE * (*page_count),
605 				      SPDK_BS_PAGE_SIZE);
606 	}
607 
608 	if (*pages == NULL) {
609 		*page_count = 0;
610 		*last_page = NULL;
611 		return -ENOMEM;
612 	}
613 
614 	page = &(*pages)[*page_count - 1];
615 	memset(page, 0, sizeof(*page));
616 	page->id = blob->id;
617 	page->sequence_num = *page_count - 1;
618 	page->next = SPDK_INVALID_MD_PAGE;
619 	*last_page = page;
620 
621 	return 0;
622 }
623 
624 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
625  * Update required_sz on both success and failure.
626  *
627  */
628 static int
629 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
630 			   uint8_t *buf, size_t buf_sz,
631 			   size_t *required_sz, bool internal)
632 {
633 	struct spdk_blob_md_descriptor_xattr	*desc;
634 
635 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
636 		       strlen(xattr->name) +
637 		       xattr->value_len;
638 
639 	if (buf_sz < *required_sz) {
640 		return -1;
641 	}
642 
643 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
644 
645 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
646 	desc->length = sizeof(desc->name_length) +
647 		       sizeof(desc->value_length) +
648 		       strlen(xattr->name) +
649 		       xattr->value_len;
650 	desc->name_length = strlen(xattr->name);
651 	desc->value_length = xattr->value_len;
652 
653 	memcpy(desc->name, xattr->name, desc->name_length);
654 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
655 	       xattr->value,
656 	       desc->value_length);
657 
658 	return 0;
659 }
660 
661 static void
662 _spdk_blob_serialize_extent(const struct spdk_blob *blob,
663 			    uint64_t start_cluster, uint64_t *next_cluster,
664 			    uint8_t *buf, size_t buf_sz)
665 {
666 	struct spdk_blob_md_descriptor_extent *desc;
667 	size_t cur_sz;
668 	uint64_t i, extent_idx;
669 	uint64_t lba, lba_per_cluster, lba_count;
670 
671 	/* The buffer must have room for at least one extent */
672 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]);
673 	if (buf_sz < cur_sz) {
674 		*next_cluster = start_cluster;
675 		return;
676 	}
677 
678 	desc = (struct spdk_blob_md_descriptor_extent *)buf;
679 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT;
680 
681 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
682 
683 	lba = blob->active.clusters[start_cluster];
684 	lba_count = lba_per_cluster;
685 	extent_idx = 0;
686 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
687 		if ((lba + lba_count) == blob->active.clusters[i]) {
688 			lba_count += lba_per_cluster;
689 			continue;
690 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
691 			lba_count += lba_per_cluster;
692 			continue;
693 		}
694 		desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
695 		desc->extents[extent_idx].length = lba_count / lba_per_cluster;
696 		extent_idx++;
697 
698 		cur_sz += sizeof(desc->extents[extent_idx]);
699 
700 		if (buf_sz < cur_sz) {
701 			/* If we ran out of buffer space, return */
702 			desc->length = sizeof(desc->extents[0]) * extent_idx;
703 			*next_cluster = i;
704 			return;
705 		}
706 
707 		lba = blob->active.clusters[i];
708 		lba_count = lba_per_cluster;
709 	}
710 
711 	desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
712 	desc->extents[extent_idx].length = lba_count / lba_per_cluster;
713 	extent_idx++;
714 
715 	desc->length = sizeof(desc->extents[0]) * extent_idx;
716 	*next_cluster = blob->active.num_clusters;
717 
718 	return;
719 }
720 
721 static void
722 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
723 			   uint8_t *buf, size_t *buf_sz)
724 {
725 	struct spdk_blob_md_descriptor_flags *desc;
726 
727 	/*
728 	 * Flags get serialized first, so we should always have room for the flags
729 	 *  descriptor.
730 	 */
731 	assert(*buf_sz >= sizeof(*desc));
732 
733 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
734 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
735 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
736 	desc->invalid_flags = blob->invalid_flags;
737 	desc->data_ro_flags = blob->data_ro_flags;
738 	desc->md_ro_flags = blob->md_ro_flags;
739 
740 	*buf_sz -= sizeof(*desc);
741 }
742 
743 static int
744 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
745 			    const struct spdk_xattr_tailq *xattrs, bool internal,
746 			    struct spdk_blob_md_page **pages,
747 			    struct spdk_blob_md_page *cur_page,
748 			    uint32_t *page_count, uint8_t **buf,
749 			    size_t *remaining_sz)
750 {
751 	const struct spdk_xattr	*xattr;
752 	int	rc;
753 
754 	TAILQ_FOREACH(xattr, xattrs, link) {
755 		size_t required_sz = 0;
756 
757 		rc = _spdk_blob_serialize_xattr(xattr,
758 						*buf, *remaining_sz,
759 						&required_sz, internal);
760 		if (rc < 0) {
761 			/* Need to add a new page to the chain */
762 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
763 							   &cur_page);
764 			if (rc < 0) {
765 				spdk_free(*pages);
766 				*pages = NULL;
767 				*page_count = 0;
768 				return rc;
769 			}
770 
771 			*buf = (uint8_t *)cur_page->descriptors;
772 			*remaining_sz = sizeof(cur_page->descriptors);
773 
774 			/* Try again */
775 			required_sz = 0;
776 			rc = _spdk_blob_serialize_xattr(xattr,
777 							*buf, *remaining_sz,
778 							&required_sz, internal);
779 
780 			if (rc < 0) {
781 				spdk_free(*pages);
782 				*pages = NULL;
783 				*page_count = 0;
784 				return rc;
785 			}
786 		}
787 
788 		*remaining_sz -= required_sz;
789 		*buf += required_sz;
790 	}
791 
792 	return 0;
793 }
794 
795 static int
796 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
797 		     uint32_t *page_count)
798 {
799 	struct spdk_blob_md_page		*cur_page;
800 	int					rc;
801 	uint8_t					*buf;
802 	size_t					remaining_sz;
803 	uint64_t				last_cluster;
804 
805 	assert(pages != NULL);
806 	assert(page_count != NULL);
807 	assert(blob != NULL);
808 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
809 
810 	*pages = NULL;
811 	*page_count = 0;
812 
813 	/* A blob always has at least 1 page, even if it has no descriptors */
814 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
815 	if (rc < 0) {
816 		return rc;
817 	}
818 
819 	buf = (uint8_t *)cur_page->descriptors;
820 	remaining_sz = sizeof(cur_page->descriptors);
821 
822 	/* Serialize flags */
823 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
824 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
825 
826 	/* Serialize xattrs */
827 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
828 					 pages, cur_page, page_count, &buf, &remaining_sz);
829 	if (rc < 0) {
830 		return rc;
831 	}
832 
833 	/* Serialize internal xattrs */
834 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
835 					 pages, cur_page, page_count, &buf, &remaining_sz);
836 	if (rc < 0) {
837 		return rc;
838 	}
839 
840 	/* Serialize extents */
841 	last_cluster = 0;
842 	while (last_cluster < blob->active.num_clusters) {
843 		_spdk_blob_serialize_extent(blob, last_cluster, &last_cluster,
844 					    buf, remaining_sz);
845 
846 		if (last_cluster == blob->active.num_clusters) {
847 			break;
848 		}
849 
850 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
851 						   &cur_page);
852 		if (rc < 0) {
853 			return rc;
854 		}
855 
856 		buf = (uint8_t *)cur_page->descriptors;
857 		remaining_sz = sizeof(cur_page->descriptors);
858 	}
859 
860 	return 0;
861 }
862 
863 struct spdk_blob_load_ctx {
864 	struct spdk_blob		*blob;
865 
866 	struct spdk_blob_md_page	*pages;
867 	uint32_t			num_pages;
868 	spdk_bs_sequence_t	        *seq;
869 
870 	spdk_bs_sequence_cpl		cb_fn;
871 	void				*cb_arg;
872 };
873 
874 static uint32_t
875 _spdk_blob_md_page_calc_crc(void *page)
876 {
877 	uint32_t		crc;
878 
879 	crc = BLOB_CRC32C_INITIAL;
880 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
881 	crc ^= BLOB_CRC32C_INITIAL;
882 
883 	return crc;
884 
885 }
886 
887 static void
888 _spdk_blob_load_final(void *cb_arg, int bserrno)
889 {
890 	struct spdk_blob_load_ctx	*ctx = cb_arg;
891 	struct spdk_blob		*blob = ctx->blob;
892 
893 	_spdk_blob_mark_clean(blob);
894 
895 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
896 
897 	/* Free the memory */
898 	spdk_free(ctx->pages);
899 	free(ctx);
900 }
901 
902 static void
903 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
904 {
905 	struct spdk_blob_load_ctx	*ctx = cb_arg;
906 	struct spdk_blob		*blob = ctx->blob;
907 
908 	if (bserrno != 0) {
909 		goto error;
910 	}
911 
912 	blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
913 
914 	if (blob->back_bs_dev == NULL) {
915 		bserrno = -ENOMEM;
916 		goto error;
917 	}
918 
919 	_spdk_blob_load_final(ctx, bserrno);
920 	return;
921 
922 error:
923 	SPDK_ERRLOG("Snapshot fail\n");
924 	_spdk_blob_free(blob);
925 	ctx->cb_fn(ctx->seq, NULL, bserrno);
926 	spdk_free(ctx->pages);
927 	free(ctx);
928 }
929 
930 static void
931 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
932 {
933 	struct spdk_blob_load_ctx	*ctx = cb_arg;
934 	struct spdk_blob		*blob = ctx->blob;
935 	struct spdk_blob_md_page	*page;
936 	const void			*value;
937 	size_t				len;
938 	int				rc;
939 	uint32_t			crc;
940 
941 	if (bserrno) {
942 		SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno);
943 		_spdk_blob_free(blob);
944 		ctx->cb_fn(seq, NULL, bserrno);
945 		spdk_free(ctx->pages);
946 		free(ctx);
947 		return;
948 	}
949 
950 	page = &ctx->pages[ctx->num_pages - 1];
951 	crc = _spdk_blob_md_page_calc_crc(page);
952 	if (crc != page->crc) {
953 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
954 		_spdk_blob_free(blob);
955 		ctx->cb_fn(seq, NULL, -EINVAL);
956 		spdk_free(ctx->pages);
957 		free(ctx);
958 		return;
959 	}
960 
961 	if (page->next != SPDK_INVALID_MD_PAGE) {
962 		uint32_t next_page = page->next;
963 		uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page);
964 
965 
966 		assert(next_lba < (blob->bs->md_start + blob->bs->md_len));
967 
968 		/* Read the next page */
969 		ctx->num_pages++;
970 		ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
971 					  sizeof(*page));
972 		if (ctx->pages == NULL) {
973 			ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM);
974 			free(ctx);
975 			return;
976 		}
977 
978 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
979 					  next_lba,
980 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
981 					  _spdk_blob_load_cpl, ctx);
982 		return;
983 	}
984 
985 	/* Parse the pages */
986 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
987 	if (rc) {
988 		_spdk_blob_free(blob);
989 		ctx->cb_fn(seq, NULL, rc);
990 		spdk_free(ctx->pages);
991 		free(ctx);
992 		return;
993 	}
994 	ctx->seq = seq;
995 
996 
997 	if (spdk_blob_is_thin_provisioned(blob)) {
998 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
999 		if (rc == 0) {
1000 			if (len != sizeof(spdk_blob_id)) {
1001 				_spdk_blob_free(blob);
1002 				ctx->cb_fn(seq, NULL, -EINVAL);
1003 				spdk_free(ctx->pages);
1004 				free(ctx);
1005 				return;
1006 			}
1007 			/* open snapshot blob and continue in the callback function */
1008 			blob->parent_id = *(spdk_blob_id *)value;
1009 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1010 					  _spdk_blob_load_snapshot_cpl, ctx);
1011 			return;
1012 		} else {
1013 			/* add zeroes_dev for thin provisioned blob */
1014 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1015 		}
1016 	} else {
1017 		/* standard blob */
1018 		blob->back_bs_dev = NULL;
1019 	}
1020 	_spdk_blob_load_final(ctx, bserrno);
1021 }
1022 
1023 /* Load a blob from disk given a blobid */
1024 static void
1025 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1026 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1027 {
1028 	struct spdk_blob_load_ctx *ctx;
1029 	struct spdk_blob_store *bs;
1030 	uint32_t page_num;
1031 	uint64_t lba;
1032 
1033 	_spdk_blob_verify_md_op(blob);
1034 
1035 	bs = blob->bs;
1036 
1037 	ctx = calloc(1, sizeof(*ctx));
1038 	if (!ctx) {
1039 		cb_fn(seq, cb_arg, -ENOMEM);
1040 		return;
1041 	}
1042 
1043 	ctx->blob = blob;
1044 	ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE);
1045 	if (!ctx->pages) {
1046 		free(ctx);
1047 		cb_fn(seq, cb_arg, -ENOMEM);
1048 		return;
1049 	}
1050 	ctx->num_pages = 1;
1051 	ctx->cb_fn = cb_fn;
1052 	ctx->cb_arg = cb_arg;
1053 
1054 	page_num = _spdk_bs_blobid_to_page(blob->id);
1055 	lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num);
1056 
1057 	blob->state = SPDK_BLOB_STATE_LOADING;
1058 
1059 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1060 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1061 				  _spdk_blob_load_cpl, ctx);
1062 }
1063 
1064 struct spdk_blob_persist_ctx {
1065 	struct spdk_blob		*blob;
1066 
1067 	struct spdk_bs_super_block	*super;
1068 
1069 	struct spdk_blob_md_page	*pages;
1070 
1071 	uint64_t			idx;
1072 
1073 	spdk_bs_sequence_t		*seq;
1074 	spdk_bs_sequence_cpl		cb_fn;
1075 	void				*cb_arg;
1076 };
1077 
1078 static void
1079 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1080 			uint32_t lba_count)
1081 {
1082 	if (ctx->blob->clear_method == BLOB_CLEAR_WITH_DEFAULT ||
1083 	    ctx->blob->clear_method == BLOB_CLEAR_WITH_UNMAP) {
1084 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1085 	} else if (ctx->blob->clear_method == BLOB_CLEAR_WITH_WRITE_ZEROES) {
1086 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1087 	}
1088 }
1089 
1090 static void
1091 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1092 {
1093 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1094 	struct spdk_blob		*blob = ctx->blob;
1095 
1096 	if (bserrno == 0) {
1097 		_spdk_blob_mark_clean(blob);
1098 	}
1099 
1100 	/* Call user callback */
1101 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1102 
1103 	/* Free the memory */
1104 	spdk_free(ctx->pages);
1105 	free(ctx);
1106 }
1107 
1108 static void
1109 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1110 {
1111 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1112 	struct spdk_blob		*blob = ctx->blob;
1113 	struct spdk_blob_store		*bs = blob->bs;
1114 	void				*tmp;
1115 	size_t				i;
1116 
1117 	/* Release all clusters that were truncated */
1118 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1119 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1120 
1121 		/* Nothing to release if it was not allocated */
1122 		if (blob->active.clusters[i] != 0) {
1123 			_spdk_bs_release_cluster(bs, cluster_num);
1124 		}
1125 	}
1126 
1127 	if (blob->active.num_clusters == 0) {
1128 		free(blob->active.clusters);
1129 		blob->active.clusters = NULL;
1130 		blob->active.cluster_array_size = 0;
1131 	} else {
1132 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters);
1133 		assert(tmp != NULL);
1134 		blob->active.clusters = tmp;
1135 		blob->active.cluster_array_size = blob->active.num_clusters;
1136 	}
1137 
1138 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1139 }
1140 
1141 static void
1142 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1143 {
1144 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1145 	struct spdk_blob		*blob = ctx->blob;
1146 	struct spdk_blob_store		*bs = blob->bs;
1147 	spdk_bs_batch_t			*batch;
1148 	size_t				i;
1149 	uint64_t			lba;
1150 	uint32_t			lba_count;
1151 
1152 	/* Clusters don't move around in blobs. The list shrinks or grows
1153 	 * at the end, but no changes ever occur in the middle of the list.
1154 	 */
1155 
1156 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx);
1157 
1158 	/* Clear all clusters that were truncated */
1159 	lba = 0;
1160 	lba_count = 0;
1161 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1162 		uint64_t next_lba = blob->active.clusters[i];
1163 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1164 
1165 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1166 			/* This cluster is contiguous with the previous one. */
1167 			lba_count += next_lba_count;
1168 			continue;
1169 		}
1170 
1171 		/* This cluster is not contiguous with the previous one. */
1172 
1173 		/* If a run of LBAs previously existing, clear them now */
1174 		if (lba_count > 0) {
1175 			spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1176 		}
1177 
1178 		/* Start building the next batch */
1179 		lba = next_lba;
1180 		if (next_lba > 0) {
1181 			lba_count = next_lba_count;
1182 		} else {
1183 			lba_count = 0;
1184 		}
1185 	}
1186 
1187 	/* If we ended with a contiguous set of LBAs, clear them now */
1188 	if (lba_count > 0) {
1189 		spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1190 	}
1191 
1192 	spdk_bs_batch_close(batch);
1193 }
1194 
1195 static void
1196 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1197 {
1198 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1199 	struct spdk_blob		*blob = ctx->blob;
1200 	struct spdk_blob_store		*bs = blob->bs;
1201 	size_t				i;
1202 
1203 	/* This loop starts at 1 because the first page is special and handled
1204 	 * below. The pages (except the first) are never written in place,
1205 	 * so any pages in the clean list must be zeroed.
1206 	 */
1207 	for (i = 1; i < blob->clean.num_pages; i++) {
1208 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1209 	}
1210 
1211 	if (blob->active.num_pages == 0) {
1212 		uint32_t page_num;
1213 
1214 		page_num = _spdk_bs_blobid_to_page(blob->id);
1215 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1216 	}
1217 
1218 	/* Move on to clearing clusters */
1219 	_spdk_blob_persist_clear_clusters(seq, ctx, 0);
1220 }
1221 
1222 static void
1223 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1224 {
1225 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1226 	struct spdk_blob		*blob = ctx->blob;
1227 	struct spdk_blob_store		*bs = blob->bs;
1228 	uint64_t			lba;
1229 	uint32_t			lba_count;
1230 	spdk_bs_batch_t			*batch;
1231 	size_t				i;
1232 
1233 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1234 
1235 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1236 
1237 	/* This loop starts at 1 because the first page is special and handled
1238 	 * below. The pages (except the first) are never written in place,
1239 	 * so any pages in the clean list must be zeroed.
1240 	 */
1241 	for (i = 1; i < blob->clean.num_pages; i++) {
1242 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);
1243 
1244 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1245 	}
1246 
1247 	/* The first page will only be zeroed if this is a delete. */
1248 	if (blob->active.num_pages == 0) {
1249 		uint32_t page_num;
1250 
1251 		/* The first page in the metadata goes where the blobid indicates */
1252 		page_num = _spdk_bs_blobid_to_page(blob->id);
1253 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);
1254 
1255 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1256 	}
1257 
1258 	spdk_bs_batch_close(batch);
1259 }
1260 
1261 static void
1262 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1263 {
1264 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1265 	struct spdk_blob		*blob = ctx->blob;
1266 	struct spdk_blob_store		*bs = blob->bs;
1267 	uint64_t			lba;
1268 	uint32_t			lba_count;
1269 	struct spdk_blob_md_page	*page;
1270 
1271 	if (blob->active.num_pages == 0) {
1272 		/* Move on to the next step */
1273 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1274 		return;
1275 	}
1276 
1277 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1278 
1279 	page = &ctx->pages[0];
1280 	/* The first page in the metadata goes where the blobid indicates */
1281 	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));
1282 
1283 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1284 				   _spdk_blob_persist_zero_pages, ctx);
1285 }
1286 
1287 static void
1288 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1289 {
1290 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1291 	struct spdk_blob		*blob = ctx->blob;
1292 	struct spdk_blob_store		*bs = blob->bs;
1293 	uint64_t			lba;
1294 	uint32_t			lba_count;
1295 	struct spdk_blob_md_page	*page;
1296 	spdk_bs_batch_t			*batch;
1297 	size_t				i;
1298 
1299 	/* Clusters don't move around in blobs. The list shrinks or grows
1300 	 * at the end, but no changes ever occur in the middle of the list.
1301 	 */
1302 
1303 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1304 
1305 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1306 
1307 	/* This starts at 1. The root page is not written until
1308 	 * all of the others are finished
1309 	 */
1310 	for (i = 1; i < blob->active.num_pages; i++) {
1311 		page = &ctx->pages[i];
1312 		assert(page->sequence_num == i);
1313 
1314 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]);
1315 
1316 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1317 	}
1318 
1319 	spdk_bs_batch_close(batch);
1320 }
1321 
1322 static int
1323 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1324 {
1325 	uint64_t	i;
1326 	uint64_t	*tmp;
1327 	uint64_t	lfc; /* lowest free cluster */
1328 	uint64_t	num_clusters;
1329 	struct spdk_blob_store *bs;
1330 
1331 	bs = blob->bs;
1332 
1333 	_spdk_blob_verify_md_op(blob);
1334 
1335 	if (blob->active.num_clusters == sz) {
1336 		return 0;
1337 	}
1338 
1339 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1340 		/* If this blob was resized to be larger, then smaller, then
1341 		 * larger without syncing, then the cluster array already
1342 		 * contains spare assigned clusters we can use.
1343 		 */
1344 		num_clusters = spdk_min(blob->active.cluster_array_size,
1345 					sz);
1346 	} else {
1347 		num_clusters = blob->active.num_clusters;
1348 	}
1349 
1350 	/* Do two passes - one to verify that we can obtain enough clusters
1351 	 * and another to actually claim them.
1352 	 */
1353 
1354 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1355 		lfc = 0;
1356 		for (i = num_clusters; i < sz; i++) {
1357 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1358 			if (lfc == UINT32_MAX) {
1359 				/* No more free clusters. Cannot satisfy the request */
1360 				return -ENOSPC;
1361 			}
1362 			lfc++;
1363 		}
1364 	}
1365 
1366 	if (sz > num_clusters) {
1367 		/* Expand the cluster array if necessary.
1368 		 * We only shrink the array when persisting.
1369 		 */
1370 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz);
1371 		if (sz > 0 && tmp == NULL) {
1372 			return -ENOMEM;
1373 		}
1374 		memset(tmp + blob->active.cluster_array_size, 0,
1375 		       sizeof(uint64_t) * (sz - blob->active.cluster_array_size));
1376 		blob->active.clusters = tmp;
1377 		blob->active.cluster_array_size = sz;
1378 	}
1379 
1380 	blob->state = SPDK_BLOB_STATE_DIRTY;
1381 
1382 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1383 		lfc = 0;
1384 		for (i = num_clusters; i < sz; i++) {
1385 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1386 			lfc++;
1387 		}
1388 	}
1389 
1390 	blob->active.num_clusters = sz;
1391 
1392 	return 0;
1393 }
1394 
1395 static void
1396 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1397 {
1398 	spdk_bs_sequence_t *seq = ctx->seq;
1399 	struct spdk_blob *blob = ctx->blob;
1400 	struct spdk_blob_store *bs = blob->bs;
1401 	uint64_t i;
1402 	uint32_t page_num;
1403 	void *tmp;
1404 	int rc;
1405 
1406 	if (blob->active.num_pages == 0) {
1407 		/* This is the signal that the blob should be deleted.
1408 		 * Immediately jump to the clean up routine. */
1409 		assert(blob->clean.num_pages > 0);
1410 		ctx->idx = blob->clean.num_pages - 1;
1411 		blob->state = SPDK_BLOB_STATE_CLEAN;
1412 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1413 		return;
1414 
1415 	}
1416 
1417 	/* Generate the new metadata */
1418 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1419 	if (rc < 0) {
1420 		_spdk_blob_persist_complete(seq, ctx, rc);
1421 		return;
1422 	}
1423 
1424 	assert(blob->active.num_pages >= 1);
1425 
1426 	/* Resize the cache of page indices */
1427 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1428 	if (!tmp) {
1429 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1430 		return;
1431 	}
1432 	blob->active.pages = tmp;
1433 
1434 	/* Assign this metadata to pages. This requires two passes -
1435 	 * one to verify that there are enough pages and a second
1436 	 * to actually claim them. */
1437 	page_num = 0;
1438 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1439 	for (i = 1; i < blob->active.num_pages; i++) {
1440 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1441 		if (page_num == UINT32_MAX) {
1442 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1443 			return;
1444 		}
1445 		page_num++;
1446 	}
1447 
1448 	page_num = 0;
1449 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1450 	for (i = 1; i < blob->active.num_pages; i++) {
1451 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1452 		ctx->pages[i - 1].next = page_num;
1453 		/* Now that previous metadata page is complete, calculate the crc for it. */
1454 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1455 		blob->active.pages[i] = page_num;
1456 		spdk_bit_array_set(bs->used_md_pages, page_num);
1457 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1458 		page_num++;
1459 	}
1460 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1461 	/* Start writing the metadata from last page to first */
1462 	ctx->idx = blob->active.num_pages - 1;
1463 	blob->state = SPDK_BLOB_STATE_CLEAN;
1464 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1465 }
1466 
1467 static void
1468 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1469 {
1470 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1471 
1472 	ctx->blob->bs->clean = 0;
1473 
1474 	spdk_free(ctx->super);
1475 
1476 	_spdk_blob_persist_start(ctx);
1477 }
1478 
1479 static void
1480 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1481 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1482 
1483 
1484 static void
1485 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1486 {
1487 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1488 
1489 	ctx->super->clean = 0;
1490 	if (ctx->super->size == 0) {
1491 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1492 	}
1493 
1494 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1495 }
1496 
1497 
1498 /* Write a blob to disk */
1499 static void
1500 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1501 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1502 {
1503 	struct spdk_blob_persist_ctx *ctx;
1504 
1505 	_spdk_blob_verify_md_op(blob);
1506 
1507 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1508 		cb_fn(seq, cb_arg, 0);
1509 		return;
1510 	}
1511 
1512 	ctx = calloc(1, sizeof(*ctx));
1513 	if (!ctx) {
1514 		cb_fn(seq, cb_arg, -ENOMEM);
1515 		return;
1516 	}
1517 	ctx->blob = blob;
1518 	ctx->seq = seq;
1519 	ctx->cb_fn = cb_fn;
1520 	ctx->cb_arg = cb_arg;
1521 
1522 	if (blob->bs->clean) {
1523 		ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
1524 					  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1525 		if (!ctx->super) {
1526 			cb_fn(seq, cb_arg, -ENOMEM);
1527 			free(ctx);
1528 			return;
1529 		}
1530 
1531 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1532 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1533 					  _spdk_blob_persist_dirty, ctx);
1534 	} else {
1535 		_spdk_blob_persist_start(ctx);
1536 	}
1537 }
1538 
1539 struct spdk_blob_copy_cluster_ctx {
1540 	struct spdk_blob *blob;
1541 	uint8_t *buf;
1542 	uint64_t page;
1543 	uint64_t new_cluster;
1544 	spdk_bs_sequence_t *seq;
1545 };
1546 
1547 static void
1548 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1549 {
1550 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1551 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1552 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1553 	spdk_bs_user_op_t *op;
1554 
1555 	TAILQ_INIT(&requests);
1556 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1557 
1558 	while (!TAILQ_EMPTY(&requests)) {
1559 		op = TAILQ_FIRST(&requests);
1560 		TAILQ_REMOVE(&requests, op, link);
1561 		if (bserrno == 0) {
1562 			spdk_bs_user_op_execute(op);
1563 		} else {
1564 			spdk_bs_user_op_abort(op);
1565 		}
1566 	}
1567 
1568 	spdk_free(ctx->buf);
1569 	free(ctx);
1570 }
1571 
1572 static void
1573 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1574 {
1575 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1576 
1577 	if (bserrno) {
1578 		uint32_t cluster_number;
1579 
1580 		if (bserrno == -EEXIST) {
1581 			/* The metadata insert failed because another thread
1582 			 * allocated the cluster first. Free our cluster
1583 			 * but continue without error. */
1584 			bserrno = 0;
1585 		}
1586 
1587 		cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1588 		_spdk_bs_release_cluster(ctx->blob->bs, cluster_number);
1589 	}
1590 
1591 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1592 }
1593 
1594 static void
1595 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1596 {
1597 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1598 	uint32_t cluster_number;
1599 
1600 	if (bserrno) {
1601 		/* The write failed, so jump to the final completion handler */
1602 		spdk_bs_sequence_finish(seq, bserrno);
1603 		return;
1604 	}
1605 
1606 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1607 
1608 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1609 					       _spdk_blob_insert_cluster_cpl, ctx);
1610 }
1611 
1612 static void
1613 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1614 {
1615 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1616 
1617 	if (bserrno != 0) {
1618 		/* The read failed, so jump to the final completion handler */
1619 		spdk_bs_sequence_finish(seq, bserrno);
1620 		return;
1621 	}
1622 
1623 	/* Write whole cluster */
1624 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1625 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1626 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1627 				   _spdk_blob_write_copy_cpl, ctx);
1628 }
1629 
1630 static void
1631 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1632 				   struct spdk_io_channel *_ch,
1633 				   uint64_t io_unit, spdk_bs_user_op_t *op)
1634 {
1635 	struct spdk_bs_cpl cpl;
1636 	struct spdk_bs_channel *ch;
1637 	struct spdk_blob_copy_cluster_ctx *ctx;
1638 	uint32_t cluster_start_page;
1639 	uint32_t cluster_number;
1640 	int rc;
1641 
1642 	ch = spdk_io_channel_get_ctx(_ch);
1643 
1644 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1645 		/* There are already operations pending. Queue this user op
1646 		 * and return because it will be re-executed when the outstanding
1647 		 * cluster allocation completes. */
1648 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1649 		return;
1650 	}
1651 
1652 	/* Round the io_unit offset down to the first page in the cluster */
1653 	cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit);
1654 
1655 	/* Calculate which index in the metadata cluster array the corresponding
1656 	 * cluster is supposed to be at. */
1657 	cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit);
1658 
1659 	ctx = calloc(1, sizeof(*ctx));
1660 	if (!ctx) {
1661 		spdk_bs_user_op_abort(op);
1662 		return;
1663 	}
1664 
1665 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1666 
1667 	ctx->blob = blob;
1668 	ctx->page = cluster_start_page;
1669 
1670 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1671 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
1672 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1673 		if (!ctx->buf) {
1674 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1675 				    blob->bs->cluster_sz);
1676 			free(ctx);
1677 			spdk_bs_user_op_abort(op);
1678 			return;
1679 		}
1680 	}
1681 
1682 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1683 	if (rc != 0) {
1684 		spdk_free(ctx->buf);
1685 		free(ctx);
1686 		spdk_bs_user_op_abort(op);
1687 		return;
1688 	}
1689 
1690 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1691 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1692 	cpl.u.blob_basic.cb_arg = ctx;
1693 
1694 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1695 	if (!ctx->seq) {
1696 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1697 		spdk_free(ctx->buf);
1698 		free(ctx);
1699 		spdk_bs_user_op_abort(op);
1700 		return;
1701 	}
1702 
1703 	/* Queue the user op to block other incoming operations */
1704 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1705 
1706 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1707 		/* Read cluster from backing device */
1708 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1709 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1710 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1711 					     _spdk_blob_write_copy, ctx);
1712 	} else {
1713 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1714 						       _spdk_blob_insert_cluster_cpl, ctx);
1715 	}
1716 }
1717 
1718 static void
1719 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
1720 				       uint64_t *lba,	uint32_t *lba_count)
1721 {
1722 	*lba_count = length;
1723 
1724 	if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) {
1725 		assert(blob->back_bs_dev != NULL);
1726 		*lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit);
1727 		*lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count);
1728 	} else {
1729 		*lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit);
1730 	}
1731 }
1732 
1733 struct op_split_ctx {
1734 	struct spdk_blob *blob;
1735 	struct spdk_io_channel *channel;
1736 	uint64_t io_unit_offset;
1737 	uint64_t io_units_remaining;
1738 	void *curr_payload;
1739 	enum spdk_blob_op_type op_type;
1740 	spdk_bs_sequence_t *seq;
1741 };
1742 
1743 static void
1744 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1745 {
1746 	struct op_split_ctx	*ctx = cb_arg;
1747 	struct spdk_blob	*blob = ctx->blob;
1748 	struct spdk_io_channel	*ch = ctx->channel;
1749 	enum spdk_blob_op_type	op_type = ctx->op_type;
1750 	uint8_t			*buf = ctx->curr_payload;
1751 	uint64_t		offset = ctx->io_unit_offset;
1752 	uint64_t		length = ctx->io_units_remaining;
1753 	uint64_t		op_length;
1754 
1755 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
1756 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1757 		free(ctx);
1758 		return;
1759 	}
1760 
1761 	op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob,
1762 			     offset));
1763 
1764 	/* Update length and payload for next operation */
1765 	ctx->io_units_remaining -= op_length;
1766 	ctx->io_unit_offset += op_length;
1767 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1768 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
1769 	}
1770 
1771 	switch (op_type) {
1772 	case SPDK_BLOB_READ:
1773 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1774 				  _spdk_blob_request_submit_op_split_next, ctx);
1775 		break;
1776 	case SPDK_BLOB_WRITE:
1777 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1778 				   _spdk_blob_request_submit_op_split_next, ctx);
1779 		break;
1780 	case SPDK_BLOB_UNMAP:
1781 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1782 				   _spdk_blob_request_submit_op_split_next, ctx);
1783 		break;
1784 	case SPDK_BLOB_WRITE_ZEROES:
1785 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1786 					  _spdk_blob_request_submit_op_split_next, ctx);
1787 		break;
1788 	case SPDK_BLOB_READV:
1789 	case SPDK_BLOB_WRITEV:
1790 		SPDK_ERRLOG("readv/write not valid for %s\n", __func__);
1791 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1792 		free(ctx);
1793 		break;
1794 	}
1795 }
1796 
1797 static void
1798 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1799 				   void *payload, uint64_t offset, uint64_t length,
1800 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1801 {
1802 	struct op_split_ctx *ctx;
1803 	spdk_bs_sequence_t *seq;
1804 	struct spdk_bs_cpl cpl;
1805 
1806 	assert(blob != NULL);
1807 
1808 	ctx = calloc(1, sizeof(struct op_split_ctx));
1809 	if (ctx == NULL) {
1810 		cb_fn(cb_arg, -ENOMEM);
1811 		return;
1812 	}
1813 
1814 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1815 	cpl.u.blob_basic.cb_fn = cb_fn;
1816 	cpl.u.blob_basic.cb_arg = cb_arg;
1817 
1818 	seq = spdk_bs_sequence_start(ch, &cpl);
1819 	if (!seq) {
1820 		free(ctx);
1821 		cb_fn(cb_arg, -ENOMEM);
1822 		return;
1823 	}
1824 
1825 	ctx->blob = blob;
1826 	ctx->channel = ch;
1827 	ctx->curr_payload = payload;
1828 	ctx->io_unit_offset = offset;
1829 	ctx->io_units_remaining = length;
1830 	ctx->op_type = op_type;
1831 	ctx->seq = seq;
1832 
1833 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1834 }
1835 
1836 static void
1837 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1838 				    void *payload, uint64_t offset, uint64_t length,
1839 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1840 {
1841 	struct spdk_bs_cpl cpl;
1842 	uint64_t lba;
1843 	uint32_t lba_count;
1844 
1845 	assert(blob != NULL);
1846 
1847 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1848 	cpl.u.blob_basic.cb_fn = cb_fn;
1849 	cpl.u.blob_basic.cb_arg = cb_arg;
1850 
1851 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1852 
1853 	if (blob->frozen_refcnt) {
1854 		/* This blob I/O is frozen */
1855 		spdk_bs_user_op_t *op;
1856 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1857 
1858 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1859 		if (!op) {
1860 			cb_fn(cb_arg, -ENOMEM);
1861 			return;
1862 		}
1863 
1864 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1865 
1866 		return;
1867 	}
1868 
1869 	switch (op_type) {
1870 	case SPDK_BLOB_READ: {
1871 		spdk_bs_batch_t *batch;
1872 
1873 		batch = spdk_bs_batch_open(_ch, &cpl);
1874 		if (!batch) {
1875 			cb_fn(cb_arg, -ENOMEM);
1876 			return;
1877 		}
1878 
1879 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1880 			/* Read from the blob */
1881 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1882 		} else {
1883 			/* Read from the backing block device */
1884 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1885 		}
1886 
1887 		spdk_bs_batch_close(batch);
1888 		break;
1889 	}
1890 	case SPDK_BLOB_WRITE:
1891 	case SPDK_BLOB_WRITE_ZEROES: {
1892 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1893 			/* Write to the blob */
1894 			spdk_bs_batch_t *batch;
1895 
1896 			if (lba_count == 0) {
1897 				cb_fn(cb_arg, 0);
1898 				return;
1899 			}
1900 
1901 			batch = spdk_bs_batch_open(_ch, &cpl);
1902 			if (!batch) {
1903 				cb_fn(cb_arg, -ENOMEM);
1904 				return;
1905 			}
1906 
1907 			if (op_type == SPDK_BLOB_WRITE) {
1908 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1909 			} else {
1910 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1911 			}
1912 
1913 			spdk_bs_batch_close(batch);
1914 		} else {
1915 			/* Queue this operation and allocate the cluster */
1916 			spdk_bs_user_op_t *op;
1917 
1918 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1919 			if (!op) {
1920 				cb_fn(cb_arg, -ENOMEM);
1921 				return;
1922 			}
1923 
1924 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1925 		}
1926 		break;
1927 	}
1928 	case SPDK_BLOB_UNMAP: {
1929 		spdk_bs_batch_t *batch;
1930 
1931 		batch = spdk_bs_batch_open(_ch, &cpl);
1932 		if (!batch) {
1933 			cb_fn(cb_arg, -ENOMEM);
1934 			return;
1935 		}
1936 
1937 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1938 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1939 		}
1940 
1941 		spdk_bs_batch_close(batch);
1942 		break;
1943 	}
1944 	case SPDK_BLOB_READV:
1945 	case SPDK_BLOB_WRITEV:
1946 		SPDK_ERRLOG("readv/write not valid\n");
1947 		cb_fn(cb_arg, -EINVAL);
1948 		break;
1949 	}
1950 }
1951 
1952 static void
1953 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1954 			     void *payload, uint64_t offset, uint64_t length,
1955 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1956 {
1957 	assert(blob != NULL);
1958 
1959 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1960 		cb_fn(cb_arg, -EPERM);
1961 		return;
1962 	}
1963 
1964 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
1965 		cb_fn(cb_arg, -EINVAL);
1966 		return;
1967 	}
1968 	if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) {
1969 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1970 						    cb_fn, cb_arg, op_type);
1971 	} else {
1972 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1973 						   cb_fn, cb_arg, op_type);
1974 	}
1975 }
1976 
1977 struct rw_iov_ctx {
1978 	struct spdk_blob *blob;
1979 	struct spdk_io_channel *channel;
1980 	spdk_blob_op_complete cb_fn;
1981 	void *cb_arg;
1982 	bool read;
1983 	int iovcnt;
1984 	struct iovec *orig_iov;
1985 	uint64_t io_unit_offset;
1986 	uint64_t io_units_remaining;
1987 	uint64_t io_units_done;
1988 	struct iovec iov[0];
1989 };
1990 
1991 static void
1992 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1993 {
1994 	assert(cb_arg == NULL);
1995 	spdk_bs_sequence_finish(seq, bserrno);
1996 }
1997 
1998 static void
1999 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
2000 {
2001 	struct rw_iov_ctx *ctx = cb_arg;
2002 	struct spdk_blob *blob = ctx->blob;
2003 	struct iovec *iov, *orig_iov;
2004 	int iovcnt;
2005 	size_t orig_iovoff;
2006 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2007 	uint64_t byte_count;
2008 
2009 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2010 		ctx->cb_fn(ctx->cb_arg, bserrno);
2011 		free(ctx);
2012 		return;
2013 	}
2014 
2015 	io_unit_offset = ctx->io_unit_offset;
2016 	io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2017 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2018 	/*
2019 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2020 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2021 	 *  point to the current position in the I/O sequence.
2022 	 */
2023 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2024 	orig_iov = &ctx->orig_iov[0];
2025 	orig_iovoff = 0;
2026 	while (byte_count > 0) {
2027 		if (byte_count >= orig_iov->iov_len) {
2028 			byte_count -= orig_iov->iov_len;
2029 			orig_iov++;
2030 		} else {
2031 			orig_iovoff = byte_count;
2032 			byte_count = 0;
2033 		}
2034 	}
2035 
2036 	/*
2037 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2038 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2039 	 */
2040 	byte_count = io_units_count * blob->bs->io_unit_size;
2041 	iov = &ctx->iov[0];
2042 	iovcnt = 0;
2043 	while (byte_count > 0) {
2044 		assert(iovcnt < ctx->iovcnt);
2045 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2046 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2047 		byte_count -= iov->iov_len;
2048 		orig_iovoff = 0;
2049 		orig_iov++;
2050 		iov++;
2051 		iovcnt++;
2052 	}
2053 
2054 	ctx->io_unit_offset += io_units_count;
2055 	ctx->io_units_remaining -= io_units_count;
2056 	ctx->io_units_done += io_units_count;
2057 	iov = &ctx->iov[0];
2058 
2059 	if (ctx->read) {
2060 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2061 				   io_units_count, _spdk_rw_iov_split_next, ctx);
2062 	} else {
2063 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2064 				    io_units_count, _spdk_rw_iov_split_next, ctx);
2065 	}
2066 }
2067 
2068 static void
2069 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2070 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2071 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2072 {
2073 	struct spdk_bs_cpl	cpl;
2074 
2075 	assert(blob != NULL);
2076 
2077 	if (!read && blob->data_ro) {
2078 		cb_fn(cb_arg, -EPERM);
2079 		return;
2080 	}
2081 
2082 	if (length == 0) {
2083 		cb_fn(cb_arg, 0);
2084 		return;
2085 	}
2086 
2087 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2088 		cb_fn(cb_arg, -EINVAL);
2089 		return;
2090 	}
2091 
2092 	/*
2093 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2094 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2095 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2096 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2097 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2098 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2099 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2100 	 *
2101 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2102 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2103 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2104 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2105 	 */
2106 	if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) {
2107 		uint32_t lba_count;
2108 		uint64_t lba;
2109 
2110 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2111 
2112 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2113 		cpl.u.blob_basic.cb_fn = cb_fn;
2114 		cpl.u.blob_basic.cb_arg = cb_arg;
2115 		if (blob->frozen_refcnt) {
2116 			/* This blob I/O is frozen */
2117 			spdk_bs_user_op_t *op;
2118 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2119 
2120 			op = spdk_bs_user_op_alloc(_channel, &cpl, read, blob, iov, iovcnt, offset, length);
2121 			if (!op) {
2122 				cb_fn(cb_arg, -ENOMEM);
2123 				return;
2124 			}
2125 
2126 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2127 
2128 			return;
2129 		}
2130 
2131 		if (read) {
2132 			spdk_bs_sequence_t *seq;
2133 
2134 			seq = spdk_bs_sequence_start(_channel, &cpl);
2135 			if (!seq) {
2136 				cb_fn(cb_arg, -ENOMEM);
2137 				return;
2138 			}
2139 
2140 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2141 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2142 			} else {
2143 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2144 							      _spdk_rw_iov_done, NULL);
2145 			}
2146 		} else {
2147 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2148 				spdk_bs_sequence_t *seq;
2149 
2150 				seq = spdk_bs_sequence_start(_channel, &cpl);
2151 				if (!seq) {
2152 					cb_fn(cb_arg, -ENOMEM);
2153 					return;
2154 				}
2155 
2156 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2157 			} else {
2158 				/* Queue this operation and allocate the cluster */
2159 				spdk_bs_user_op_t *op;
2160 
2161 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2162 							   length);
2163 				if (!op) {
2164 					cb_fn(cb_arg, -ENOMEM);
2165 					return;
2166 				}
2167 
2168 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2169 			}
2170 		}
2171 	} else {
2172 		struct rw_iov_ctx *ctx;
2173 
2174 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2175 		if (ctx == NULL) {
2176 			cb_fn(cb_arg, -ENOMEM);
2177 			return;
2178 		}
2179 
2180 		ctx->blob = blob;
2181 		ctx->channel = _channel;
2182 		ctx->cb_fn = cb_fn;
2183 		ctx->cb_arg = cb_arg;
2184 		ctx->read = read;
2185 		ctx->orig_iov = iov;
2186 		ctx->iovcnt = iovcnt;
2187 		ctx->io_unit_offset = offset;
2188 		ctx->io_units_remaining = length;
2189 		ctx->io_units_done = 0;
2190 
2191 		_spdk_rw_iov_split_next(ctx, 0);
2192 	}
2193 }
2194 
2195 static struct spdk_blob *
2196 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2197 {
2198 	struct spdk_blob *blob;
2199 
2200 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2201 		if (blob->id == blobid) {
2202 			return blob;
2203 		}
2204 	}
2205 
2206 	return NULL;
2207 }
2208 
2209 static void
2210 _spdk_blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
2211 		struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
2212 {
2213 	assert(blob != NULL);
2214 	*snapshot_entry = NULL;
2215 	*clone_entry = NULL;
2216 
2217 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
2218 		return;
2219 	}
2220 
2221 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
2222 		if ((*snapshot_entry)->id == blob->parent_id) {
2223 			break;
2224 		}
2225 	}
2226 
2227 	if (*snapshot_entry != NULL) {
2228 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
2229 			if ((*clone_entry)->id == blob->id) {
2230 				break;
2231 			}
2232 		}
2233 
2234 		assert(clone_entry != NULL);
2235 	}
2236 }
2237 
2238 static int
2239 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2240 {
2241 	struct spdk_blob_store		*bs = io_device;
2242 	struct spdk_bs_channel		*channel = ctx_buf;
2243 	struct spdk_bs_dev		*dev;
2244 	uint32_t			max_ops = bs->max_channel_ops;
2245 	uint32_t			i;
2246 
2247 	dev = bs->dev;
2248 
2249 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2250 	if (!channel->req_mem) {
2251 		return -1;
2252 	}
2253 
2254 	TAILQ_INIT(&channel->reqs);
2255 
2256 	for (i = 0; i < max_ops; i++) {
2257 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2258 	}
2259 
2260 	channel->bs = bs;
2261 	channel->dev = dev;
2262 	channel->dev_channel = dev->create_channel(dev);
2263 
2264 	if (!channel->dev_channel) {
2265 		SPDK_ERRLOG("Failed to create device channel.\n");
2266 		free(channel->req_mem);
2267 		return -1;
2268 	}
2269 
2270 	TAILQ_INIT(&channel->need_cluster_alloc);
2271 	TAILQ_INIT(&channel->queued_io);
2272 
2273 	return 0;
2274 }
2275 
2276 static void
2277 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2278 {
2279 	struct spdk_bs_channel *channel = ctx_buf;
2280 	spdk_bs_user_op_t *op;
2281 
2282 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2283 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2284 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2285 		spdk_bs_user_op_abort(op);
2286 	}
2287 
2288 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2289 		op = TAILQ_FIRST(&channel->queued_io);
2290 		TAILQ_REMOVE(&channel->queued_io, op, link);
2291 		spdk_bs_user_op_abort(op);
2292 	}
2293 
2294 	free(channel->req_mem);
2295 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2296 }
2297 
2298 static void
2299 _spdk_bs_dev_destroy(void *io_device)
2300 {
2301 	struct spdk_blob_store *bs = io_device;
2302 	struct spdk_blob	*blob, *blob_tmp;
2303 
2304 	bs->dev->destroy(bs->dev);
2305 
2306 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2307 		TAILQ_REMOVE(&bs->blobs, blob, link);
2308 		_spdk_blob_free(blob);
2309 	}
2310 
2311 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2312 
2313 	spdk_bit_array_free(&bs->used_blobids);
2314 	spdk_bit_array_free(&bs->used_md_pages);
2315 	spdk_bit_array_free(&bs->used_clusters);
2316 	/*
2317 	 * If this function is called for any reason except a successful unload,
2318 	 * the unload_cpl type will be NONE and this will be a nop.
2319 	 */
2320 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2321 
2322 	free(bs);
2323 }
2324 
2325 static int
2326 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2327 {
2328 	spdk_blob_id snapshot_id;
2329 	struct spdk_blob_list *snapshot_entry = NULL;
2330 	struct spdk_blob_list *clone_entry = NULL;
2331 
2332 	assert(blob != NULL);
2333 
2334 	snapshot_id = blob->parent_id;
2335 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2336 		return 0;
2337 	}
2338 
2339 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2340 		if (snapshot_entry->id == snapshot_id) {
2341 			break;
2342 		}
2343 	}
2344 
2345 	if (snapshot_entry == NULL) {
2346 		/* Snapshot not found */
2347 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2348 		if (snapshot_entry == NULL) {
2349 			return -ENOMEM;
2350 		}
2351 		snapshot_entry->id = snapshot_id;
2352 		TAILQ_INIT(&snapshot_entry->clones);
2353 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2354 	} else {
2355 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2356 			if (clone_entry->id == blob->id) {
2357 				break;
2358 			}
2359 		}
2360 	}
2361 
2362 	if (clone_entry == NULL) {
2363 		/* Clone not found */
2364 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2365 		if (clone_entry == NULL) {
2366 			return -ENOMEM;
2367 		}
2368 		clone_entry->id = blob->id;
2369 		TAILQ_INIT(&clone_entry->clones);
2370 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2371 		snapshot_entry->clone_count++;
2372 	}
2373 
2374 	return 0;
2375 }
2376 
2377 static int
2378 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2379 {
2380 	struct spdk_blob_list *snapshot_entry = NULL;
2381 	struct spdk_blob_list *clone_entry = NULL;
2382 
2383 	_spdk_blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
2384 
2385 	if (snapshot_entry == NULL) {
2386 		return 0;
2387 	}
2388 
2389 	blob->parent_id = SPDK_BLOBID_INVALID;
2390 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2391 	free(clone_entry);
2392 
2393 	snapshot_entry->clone_count--;
2394 
2395 	return 0;
2396 }
2397 
2398 static int
2399 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2400 {
2401 	struct spdk_blob_list *snapshot_entry;
2402 	struct spdk_blob_list *snapshot_entry_tmp;
2403 	struct spdk_blob_list *clone_entry;
2404 	struct spdk_blob_list *clone_entry_tmp;
2405 
2406 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2407 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2408 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2409 			free(clone_entry);
2410 		}
2411 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2412 		free(snapshot_entry);
2413 	}
2414 
2415 	return 0;
2416 }
2417 
2418 static void
2419 _spdk_bs_free(struct spdk_blob_store *bs)
2420 {
2421 	_spdk_bs_blob_list_free(bs);
2422 
2423 	spdk_bs_unregister_md_thread(bs);
2424 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2425 }
2426 
2427 void
2428 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2429 {
2430 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2431 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2432 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2433 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2434 	opts->clear_method = BS_CLEAR_WITH_UNMAP;
2435 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2436 	opts->iter_cb_fn = NULL;
2437 	opts->iter_cb_arg = NULL;
2438 }
2439 
2440 static int
2441 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2442 {
2443 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2444 	    opts->max_channel_ops == 0) {
2445 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2446 		return -1;
2447 	}
2448 
2449 	return 0;
2450 }
2451 
2452 static int
2453 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2454 {
2455 	struct spdk_blob_store	*bs;
2456 	uint64_t dev_size;
2457 	int rc;
2458 
2459 	dev_size = dev->blocklen * dev->blockcnt;
2460 	if (dev_size < opts->cluster_sz) {
2461 		/* Device size cannot be smaller than cluster size of blobstore */
2462 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2463 			     dev_size, opts->cluster_sz);
2464 		return -ENOSPC;
2465 	}
2466 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2467 		/* Cluster size cannot be smaller than page size */
2468 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2469 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2470 		return -EINVAL;
2471 	}
2472 	bs = calloc(1, sizeof(struct spdk_blob_store));
2473 	if (!bs) {
2474 		return -ENOMEM;
2475 	}
2476 
2477 	TAILQ_INIT(&bs->blobs);
2478 	TAILQ_INIT(&bs->snapshots);
2479 	bs->dev = dev;
2480 	bs->md_thread = spdk_get_thread();
2481 	assert(bs->md_thread != NULL);
2482 
2483 	/*
2484 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2485 	 *  even multiple of the cluster size.
2486 	 */
2487 	bs->cluster_sz = opts->cluster_sz;
2488 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2489 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2490 	bs->num_free_clusters = bs->total_clusters;
2491 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2492 	bs->io_unit_size = dev->blocklen;
2493 	if (bs->used_clusters == NULL) {
2494 		free(bs);
2495 		return -ENOMEM;
2496 	}
2497 
2498 	bs->max_channel_ops = opts->max_channel_ops;
2499 	bs->super_blob = SPDK_BLOBID_INVALID;
2500 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2501 
2502 	/* The metadata is assumed to be at least 1 page */
2503 	bs->used_md_pages = spdk_bit_array_create(1);
2504 	bs->used_blobids = spdk_bit_array_create(0);
2505 
2506 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2507 
2508 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2509 				sizeof(struct spdk_bs_channel), "blobstore");
2510 	rc = spdk_bs_register_md_thread(bs);
2511 	if (rc == -1) {
2512 		spdk_io_device_unregister(bs, NULL);
2513 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2514 		spdk_bit_array_free(&bs->used_blobids);
2515 		spdk_bit_array_free(&bs->used_md_pages);
2516 		spdk_bit_array_free(&bs->used_clusters);
2517 		free(bs);
2518 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2519 		return -ENOMEM;
2520 	}
2521 
2522 	*_bs = bs;
2523 	return 0;
2524 }
2525 
2526 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2527 
2528 struct spdk_bs_load_ctx {
2529 	struct spdk_blob_store		*bs;
2530 	struct spdk_bs_super_block	*super;
2531 
2532 	struct spdk_bs_md_mask		*mask;
2533 	bool				in_page_chain;
2534 	uint32_t			page_index;
2535 	uint32_t			cur_page;
2536 	struct spdk_blob_md_page	*page;
2537 
2538 	spdk_bs_sequence_t			*seq;
2539 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2540 	void					*iter_cb_arg;
2541 };
2542 
2543 static void
2544 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2545 {
2546 	assert(bserrno != 0);
2547 
2548 	spdk_free(ctx->super);
2549 	spdk_bs_sequence_finish(seq, bserrno);
2550 	_spdk_bs_free(ctx->bs);
2551 	free(ctx);
2552 }
2553 
2554 static void
2555 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2556 {
2557 	uint32_t i = 0;
2558 
2559 	while (true) {
2560 		i = spdk_bit_array_find_first_set(array, i);
2561 		if (i >= mask->length) {
2562 			break;
2563 		}
2564 		mask->mask[i / 8] |= 1U << (i % 8);
2565 		i++;
2566 	}
2567 }
2568 
2569 static int
2570 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2571 {
2572 	struct spdk_bit_array *array;
2573 	uint32_t i;
2574 
2575 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2576 		return -ENOMEM;
2577 	}
2578 
2579 	array = *array_ptr;
2580 	for (i = 0; i < mask->length; i++) {
2581 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2582 			spdk_bit_array_set(array, i);
2583 		}
2584 	}
2585 
2586 	return 0;
2587 }
2588 
2589 static void
2590 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2591 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2592 {
2593 	/* Update the values in the super block */
2594 	super->super_blob = bs->super_blob;
2595 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2596 	super->crc = _spdk_blob_md_page_calc_crc(super);
2597 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2598 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2599 				   cb_fn, cb_arg);
2600 }
2601 
2602 static void
2603 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2604 {
2605 	struct spdk_bs_load_ctx	*ctx = arg;
2606 	uint64_t	mask_size, lba, lba_count;
2607 
2608 	/* Write out the used clusters mask */
2609 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2610 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2611 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2612 	if (!ctx->mask) {
2613 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2614 		return;
2615 	}
2616 
2617 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2618 	ctx->mask->length = ctx->bs->total_clusters;
2619 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2620 
2621 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2622 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2623 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2624 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2625 }
2626 
2627 static void
2628 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2629 {
2630 	struct spdk_bs_load_ctx	*ctx = arg;
2631 	uint64_t	mask_size, lba, lba_count;
2632 
2633 	if (seq->bserrno) {
2634 		_spdk_bs_load_ctx_fail(seq, ctx, seq->bserrno);
2635 		return;
2636 	}
2637 
2638 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2639 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2640 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2641 	if (!ctx->mask) {
2642 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2643 		return;
2644 	}
2645 
2646 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2647 	ctx->mask->length = ctx->super->md_len;
2648 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2649 
2650 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2651 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2652 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2653 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2654 }
2655 
2656 static void
2657 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2658 {
2659 	struct spdk_bs_load_ctx	*ctx = arg;
2660 	uint64_t	mask_size, lba, lba_count;
2661 
2662 	if (ctx->super->used_blobid_mask_len == 0) {
2663 		/*
2664 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2665 		 *  written to disk.
2666 		 */
2667 		cb_fn(seq, arg, 0);
2668 		return;
2669 	}
2670 
2671 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2672 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2673 				 SPDK_MALLOC_DMA);
2674 	if (!ctx->mask) {
2675 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2676 		return;
2677 	}
2678 
2679 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2680 	ctx->mask->length = ctx->super->md_len;
2681 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2682 
2683 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2684 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2685 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2686 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2687 }
2688 
2689 static void
2690 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2691 {
2692 	struct spdk_bs_load_ctx *ctx = arg;
2693 
2694 	if (bserrno == 0) {
2695 		if (ctx->iter_cb_fn) {
2696 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2697 		}
2698 		_spdk_bs_blob_list_add(blob);
2699 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2700 		return;
2701 	}
2702 
2703 	if (bserrno == -ENOENT) {
2704 		bserrno = 0;
2705 	} else {
2706 		/*
2707 		 * This case needs to be looked at further.  Same problem
2708 		 *  exists with applications that rely on explicit blob
2709 		 *  iteration.  We should just skip the blob that failed
2710 		 *  to load and continue on to the next one.
2711 		 */
2712 		SPDK_ERRLOG("Error in iterating blobs\n");
2713 	}
2714 
2715 	ctx->iter_cb_fn = NULL;
2716 
2717 	spdk_free(ctx->super);
2718 	spdk_free(ctx->mask);
2719 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2720 	free(ctx);
2721 }
2722 
2723 static void
2724 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2725 {
2726 	ctx->seq = seq;
2727 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2728 }
2729 
2730 static void
2731 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2732 {
2733 	struct spdk_bs_load_ctx *ctx = cb_arg;
2734 	int rc;
2735 
2736 	/* The type must be correct */
2737 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2738 
2739 	/* The length of the mask (in bits) must not be greater than
2740 	 * the length of the buffer (converted to bits) */
2741 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2742 
2743 	/* The length of the mask must be exactly equal to the size
2744 	 * (in pages) of the metadata region */
2745 	assert(ctx->mask->length == ctx->super->md_len);
2746 
2747 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2748 	if (rc < 0) {
2749 		spdk_free(ctx->mask);
2750 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2751 		return;
2752 	}
2753 
2754 	_spdk_bs_load_complete(seq, ctx, bserrno);
2755 }
2756 
2757 static void
2758 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2759 {
2760 	struct spdk_bs_load_ctx *ctx = cb_arg;
2761 	uint64_t		lba, lba_count, mask_size;
2762 	int			rc;
2763 
2764 	/* The type must be correct */
2765 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2766 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2767 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2768 					     struct spdk_blob_md_page) * 8));
2769 	/* The length of the mask must be exactly equal to the total number of clusters */
2770 	assert(ctx->mask->length == ctx->bs->total_clusters);
2771 
2772 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2773 	if (rc < 0) {
2774 		spdk_free(ctx->mask);
2775 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2776 		return;
2777 	}
2778 
2779 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2780 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2781 
2782 	spdk_free(ctx->mask);
2783 
2784 	/* Read the used blobids mask */
2785 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2786 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2787 				 SPDK_MALLOC_DMA);
2788 	if (!ctx->mask) {
2789 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2790 		return;
2791 	}
2792 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2793 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2794 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2795 				  _spdk_bs_load_used_blobids_cpl, ctx);
2796 }
2797 
2798 static void
2799 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2800 {
2801 	struct spdk_bs_load_ctx *ctx = cb_arg;
2802 	uint64_t		lba, lba_count, mask_size;
2803 	int			rc;
2804 
2805 	/* The type must be correct */
2806 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2807 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2808 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2809 				     8));
2810 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2811 	assert(ctx->mask->length == ctx->super->md_len);
2812 
2813 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2814 	if (rc < 0) {
2815 		spdk_free(ctx->mask);
2816 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2817 		return;
2818 	}
2819 
2820 	spdk_free(ctx->mask);
2821 
2822 	/* Read the used clusters mask */
2823 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2824 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2825 				 SPDK_MALLOC_DMA);
2826 	if (!ctx->mask) {
2827 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2828 		return;
2829 	}
2830 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2831 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2832 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2833 				  _spdk_bs_load_used_clusters_cpl, ctx);
2834 }
2835 
2836 static void
2837 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2838 {
2839 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2840 	uint64_t lba, lba_count, mask_size;
2841 
2842 	/* Read the used pages mask */
2843 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2844 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2845 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2846 	if (!ctx->mask) {
2847 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2848 		return;
2849 	}
2850 
2851 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2852 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2853 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2854 				  _spdk_bs_load_used_pages_cpl, ctx);
2855 }
2856 
2857 static int
2858 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2859 {
2860 	struct spdk_blob_md_descriptor *desc;
2861 	size_t	cur_desc = 0;
2862 
2863 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2864 	while (cur_desc < sizeof(page->descriptors)) {
2865 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2866 			if (desc->length == 0) {
2867 				/* If padding and length are 0, this terminates the page */
2868 				break;
2869 			}
2870 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2871 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2872 			unsigned int				i, j;
2873 			unsigned int				cluster_count = 0;
2874 			uint32_t				cluster_idx;
2875 
2876 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2877 
2878 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2879 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2880 					cluster_idx = desc_extent->extents[i].cluster_idx;
2881 					/*
2882 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2883 					 * in the used cluster map.
2884 					 */
2885 					if (cluster_idx != 0) {
2886 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2887 						if (bs->num_free_clusters == 0) {
2888 							return -ENOSPC;
2889 						}
2890 						bs->num_free_clusters--;
2891 					}
2892 					cluster_count++;
2893 				}
2894 			}
2895 			if (cluster_count == 0) {
2896 				return -EINVAL;
2897 			}
2898 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2899 			/* Skip this item */
2900 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2901 			/* Skip this item */
2902 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2903 			/* Skip this item */
2904 		} else {
2905 			/* Error */
2906 			return -EINVAL;
2907 		}
2908 		/* Advance to the next descriptor */
2909 		cur_desc += sizeof(*desc) + desc->length;
2910 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2911 			break;
2912 		}
2913 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2914 	}
2915 	return 0;
2916 }
2917 
2918 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2919 {
2920 	uint32_t crc;
2921 
2922 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2923 	if (crc != ctx->page->crc) {
2924 		return false;
2925 	}
2926 
2927 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2928 		return false;
2929 	}
2930 	return true;
2931 }
2932 
2933 static void
2934 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2935 
2936 static void
2937 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2938 {
2939 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2940 
2941 	_spdk_bs_load_complete(seq, ctx, bserrno);
2942 }
2943 
2944 static void
2945 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2946 {
2947 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2948 
2949 	spdk_free(ctx->mask);
2950 	ctx->mask = NULL;
2951 
2952 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2953 }
2954 
2955 static void
2956 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2957 {
2958 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2959 
2960 	spdk_free(ctx->mask);
2961 	ctx->mask = NULL;
2962 
2963 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2964 }
2965 
2966 static void
2967 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2968 {
2969 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2970 }
2971 
2972 static void
2973 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2974 {
2975 	struct spdk_bs_load_ctx *ctx = cb_arg;
2976 	uint64_t num_md_clusters;
2977 	uint64_t i;
2978 	uint32_t page_num;
2979 
2980 	if (bserrno != 0) {
2981 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2982 		return;
2983 	}
2984 
2985 	page_num = ctx->cur_page;
2986 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2987 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2988 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2989 			if (ctx->page->sequence_num == 0) {
2990 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2991 			}
2992 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2993 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
2994 				return;
2995 			}
2996 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
2997 				ctx->in_page_chain = true;
2998 				ctx->cur_page = ctx->page->next;
2999 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3000 				return;
3001 			}
3002 		}
3003 	}
3004 
3005 	ctx->in_page_chain = false;
3006 
3007 	do {
3008 		ctx->page_index++;
3009 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
3010 
3011 	if (ctx->page_index < ctx->super->md_len) {
3012 		ctx->cur_page = ctx->page_index;
3013 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3014 	} else {
3015 		/* Claim all of the clusters used by the metadata */
3016 		num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
3017 		for (i = 0; i < num_md_clusters; i++) {
3018 			_spdk_bs_claim_cluster(ctx->bs, i);
3019 		}
3020 		spdk_free(ctx->page);
3021 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
3022 	}
3023 }
3024 
3025 static void
3026 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3027 {
3028 	struct spdk_bs_load_ctx *ctx = cb_arg;
3029 	uint64_t lba;
3030 
3031 	assert(ctx->cur_page < ctx->super->md_len);
3032 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3033 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3034 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3035 				  _spdk_bs_load_replay_md_cpl, ctx);
3036 }
3037 
3038 static void
3039 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3040 {
3041 	struct spdk_bs_load_ctx *ctx = cb_arg;
3042 
3043 	ctx->page_index = 0;
3044 	ctx->cur_page = 0;
3045 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3046 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3047 	if (!ctx->page) {
3048 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3049 		return;
3050 	}
3051 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3052 }
3053 
3054 static void
3055 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3056 {
3057 	struct spdk_bs_load_ctx *ctx = cb_arg;
3058 	int		rc;
3059 
3060 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3061 	if (rc < 0) {
3062 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3063 		return;
3064 	}
3065 
3066 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, 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_clusters, ctx->bs->total_clusters);
3073 	if (rc < 0) {
3074 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3075 		return;
3076 	}
3077 
3078 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3079 	_spdk_bs_load_replay_md(seq, cb_arg);
3080 }
3081 
3082 static void
3083 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3084 {
3085 	struct spdk_bs_load_ctx *ctx = cb_arg;
3086 	uint32_t	crc;
3087 	int		rc;
3088 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3089 
3090 	if (ctx->super->version > SPDK_BS_VERSION ||
3091 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3092 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3093 		return;
3094 	}
3095 
3096 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3097 		   sizeof(ctx->super->signature)) != 0) {
3098 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3099 		return;
3100 	}
3101 
3102 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3103 	if (crc != ctx->super->crc) {
3104 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3105 		return;
3106 	}
3107 
3108 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3109 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3110 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3111 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3112 	} else {
3113 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3114 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3115 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3116 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3117 		return;
3118 	}
3119 
3120 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3121 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3122 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3123 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3124 		return;
3125 	}
3126 
3127 	if (ctx->super->size == 0) {
3128 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
3129 	}
3130 
3131 	if (ctx->super->io_unit_size == 0) {
3132 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
3133 	}
3134 
3135 	/* Parse the super block */
3136 	ctx->bs->clean = 1;
3137 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3138 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
3139 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3140 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
3141 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3142 	if (rc < 0) {
3143 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3144 		return;
3145 	}
3146 	ctx->bs->md_start = ctx->super->md_start;
3147 	ctx->bs->md_len = ctx->super->md_len;
3148 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
3149 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3150 	ctx->bs->super_blob = ctx->super->super_blob;
3151 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3152 
3153 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3154 		_spdk_bs_recover(seq, ctx);
3155 	} else {
3156 		_spdk_bs_load_read_used_pages(seq, ctx);
3157 	}
3158 }
3159 
3160 void
3161 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3162 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3163 {
3164 	struct spdk_blob_store	*bs;
3165 	struct spdk_bs_cpl	cpl;
3166 	spdk_bs_sequence_t	*seq;
3167 	struct spdk_bs_load_ctx *ctx;
3168 	struct spdk_bs_opts	opts = {};
3169 	int err;
3170 
3171 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3172 
3173 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3174 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3175 		dev->destroy(dev);
3176 		cb_fn(cb_arg, NULL, -EINVAL);
3177 		return;
3178 	}
3179 
3180 	if (o) {
3181 		opts = *o;
3182 	} else {
3183 		spdk_bs_opts_init(&opts);
3184 	}
3185 
3186 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3187 		dev->destroy(dev);
3188 		cb_fn(cb_arg, NULL, -EINVAL);
3189 		return;
3190 	}
3191 
3192 	err = _spdk_bs_alloc(dev, &opts, &bs);
3193 	if (err) {
3194 		dev->destroy(dev);
3195 		cb_fn(cb_arg, NULL, err);
3196 		return;
3197 	}
3198 
3199 	ctx = calloc(1, sizeof(*ctx));
3200 	if (!ctx) {
3201 		_spdk_bs_free(bs);
3202 		cb_fn(cb_arg, NULL, -ENOMEM);
3203 		return;
3204 	}
3205 
3206 	ctx->bs = bs;
3207 	ctx->iter_cb_fn = opts.iter_cb_fn;
3208 	ctx->iter_cb_arg = opts.iter_cb_arg;
3209 
3210 	/* Allocate memory for the super block */
3211 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3212 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3213 	if (!ctx->super) {
3214 		free(ctx);
3215 		_spdk_bs_free(bs);
3216 		cb_fn(cb_arg, NULL, -ENOMEM);
3217 		return;
3218 	}
3219 
3220 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3221 	cpl.u.bs_handle.cb_fn = cb_fn;
3222 	cpl.u.bs_handle.cb_arg = cb_arg;
3223 	cpl.u.bs_handle.bs = bs;
3224 
3225 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3226 	if (!seq) {
3227 		spdk_free(ctx->super);
3228 		free(ctx);
3229 		_spdk_bs_free(bs);
3230 		cb_fn(cb_arg, NULL, -ENOMEM);
3231 		return;
3232 	}
3233 
3234 	/* Read the super block */
3235 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3236 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3237 				  _spdk_bs_load_super_cpl, ctx);
3238 }
3239 
3240 /* END spdk_bs_load */
3241 
3242 /* START spdk_bs_dump */
3243 
3244 struct spdk_bs_dump_ctx {
3245 	struct spdk_blob_store		*bs;
3246 	struct spdk_bs_super_block	*super;
3247 	uint32_t			cur_page;
3248 	struct spdk_blob_md_page	*page;
3249 	spdk_bs_sequence_t		*seq;
3250 	FILE				*fp;
3251 	spdk_bs_dump_print_xattr	print_xattr_fn;
3252 	char				xattr_name[4096];
3253 };
3254 
3255 static void
3256 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3257 {
3258 	spdk_free(ctx->super);
3259 
3260 	/*
3261 	 * We need to defer calling spdk_bs_call_cpl() until after
3262 	 * dev destruction, so tuck these away for later use.
3263 	 */
3264 	ctx->bs->unload_err = bserrno;
3265 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3266 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3267 
3268 	spdk_bs_sequence_finish(seq, 0);
3269 	_spdk_bs_free(ctx->bs);
3270 	free(ctx);
3271 }
3272 
3273 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3274 
3275 static void
3276 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3277 {
3278 	uint32_t page_idx = ctx->cur_page;
3279 	struct spdk_blob_md_page *page = ctx->page;
3280 	struct spdk_blob_md_descriptor *desc;
3281 	size_t cur_desc = 0;
3282 	uint32_t crc;
3283 
3284 	fprintf(ctx->fp, "=========\n");
3285 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3286 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3287 
3288 	crc = _spdk_blob_md_page_calc_crc(page);
3289 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3290 
3291 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3292 	while (cur_desc < sizeof(page->descriptors)) {
3293 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3294 			if (desc->length == 0) {
3295 				/* If padding and length are 0, this terminates the page */
3296 				break;
3297 			}
3298 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3299 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3300 			unsigned int				i;
3301 
3302 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3303 
3304 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3305 				if (desc_extent->extents[i].cluster_idx != 0) {
3306 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3307 						desc_extent->extents[i].cluster_idx);
3308 				} else {
3309 					fprintf(ctx->fp, "Unallocated Extent - ");
3310 				}
3311 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3312 				fprintf(ctx->fp, "\n");
3313 			}
3314 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3315 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3316 			uint32_t i;
3317 
3318 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3319 
3320 			if (desc_xattr->length !=
3321 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3322 			    desc_xattr->name_length + desc_xattr->value_length) {
3323 			}
3324 
3325 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3326 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3327 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3328 			fprintf(ctx->fp, "       value = \"");
3329 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3330 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3331 					    desc_xattr->value_length);
3332 			fprintf(ctx->fp, "\"\n");
3333 			for (i = 0; i < desc_xattr->value_length; i++) {
3334 				if (i % 16 == 0) {
3335 					fprintf(ctx->fp, "               ");
3336 				}
3337 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3338 				if ((i + 1) % 16 == 0) {
3339 					fprintf(ctx->fp, "\n");
3340 				}
3341 			}
3342 			if (i % 16 != 0) {
3343 				fprintf(ctx->fp, "\n");
3344 			}
3345 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3346 			/* TODO */
3347 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3348 			/* TODO */
3349 		} else {
3350 			/* Error */
3351 		}
3352 		/* Advance to the next descriptor */
3353 		cur_desc += sizeof(*desc) + desc->length;
3354 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3355 			break;
3356 		}
3357 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3358 	}
3359 }
3360 
3361 static void
3362 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3363 {
3364 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3365 
3366 	if (bserrno != 0) {
3367 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3368 		return;
3369 	}
3370 
3371 	if (ctx->page->id != 0) {
3372 		_spdk_bs_dump_print_md_page(ctx);
3373 	}
3374 
3375 	ctx->cur_page++;
3376 
3377 	if (ctx->cur_page < ctx->super->md_len) {
3378 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3379 	} else {
3380 		spdk_free(ctx->page);
3381 		_spdk_bs_dump_finish(seq, ctx, 0);
3382 	}
3383 }
3384 
3385 static void
3386 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3387 {
3388 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3389 	uint64_t lba;
3390 
3391 	assert(ctx->cur_page < ctx->super->md_len);
3392 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3393 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3394 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3395 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3396 }
3397 
3398 static void
3399 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3400 {
3401 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3402 
3403 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3404 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3405 		   sizeof(ctx->super->signature)) != 0) {
3406 		fprintf(ctx->fp, "(Mismatch)\n");
3407 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3408 		return;
3409 	} else {
3410 		fprintf(ctx->fp, "(OK)\n");
3411 	}
3412 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3413 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3414 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3415 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3416 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3417 	fprintf(ctx->fp, "Super Blob ID: ");
3418 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3419 		fprintf(ctx->fp, "(None)\n");
3420 	} else {
3421 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3422 	}
3423 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3424 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3425 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3426 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3427 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3428 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3429 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3430 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3431 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3432 
3433 	ctx->cur_page = 0;
3434 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3435 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3436 	if (!ctx->page) {
3437 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3438 		return;
3439 	}
3440 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3441 }
3442 
3443 void
3444 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3445 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3446 {
3447 	struct spdk_blob_store	*bs;
3448 	struct spdk_bs_cpl	cpl;
3449 	spdk_bs_sequence_t	*seq;
3450 	struct spdk_bs_dump_ctx *ctx;
3451 	struct spdk_bs_opts	opts = {};
3452 	int err;
3453 
3454 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3455 
3456 	spdk_bs_opts_init(&opts);
3457 
3458 	err = _spdk_bs_alloc(dev, &opts, &bs);
3459 	if (err) {
3460 		dev->destroy(dev);
3461 		cb_fn(cb_arg, err);
3462 		return;
3463 	}
3464 
3465 	ctx = calloc(1, sizeof(*ctx));
3466 	if (!ctx) {
3467 		_spdk_bs_free(bs);
3468 		cb_fn(cb_arg, -ENOMEM);
3469 		return;
3470 	}
3471 
3472 	ctx->bs = bs;
3473 	ctx->fp = fp;
3474 	ctx->print_xattr_fn = print_xattr_fn;
3475 
3476 	/* Allocate memory for the super block */
3477 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3478 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3479 	if (!ctx->super) {
3480 		free(ctx);
3481 		_spdk_bs_free(bs);
3482 		cb_fn(cb_arg, -ENOMEM);
3483 		return;
3484 	}
3485 
3486 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3487 	cpl.u.bs_basic.cb_fn = cb_fn;
3488 	cpl.u.bs_basic.cb_arg = cb_arg;
3489 
3490 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3491 	if (!seq) {
3492 		spdk_free(ctx->super);
3493 		free(ctx);
3494 		_spdk_bs_free(bs);
3495 		cb_fn(cb_arg, -ENOMEM);
3496 		return;
3497 	}
3498 
3499 	/* Read the super block */
3500 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3501 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3502 				  _spdk_bs_dump_super_cpl, ctx);
3503 }
3504 
3505 /* END spdk_bs_dump */
3506 
3507 /* START spdk_bs_init */
3508 
3509 struct spdk_bs_init_ctx {
3510 	struct spdk_blob_store		*bs;
3511 	struct spdk_bs_super_block	*super;
3512 };
3513 
3514 static void
3515 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3516 {
3517 	struct spdk_bs_init_ctx *ctx = cb_arg;
3518 
3519 	spdk_free(ctx->super);
3520 	free(ctx);
3521 
3522 	spdk_bs_sequence_finish(seq, bserrno);
3523 }
3524 
3525 static void
3526 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3527 {
3528 	struct spdk_bs_init_ctx *ctx = cb_arg;
3529 
3530 	/* Write super block */
3531 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3532 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3533 				   _spdk_bs_init_persist_super_cpl, ctx);
3534 }
3535 
3536 void
3537 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3538 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3539 {
3540 	struct spdk_bs_init_ctx *ctx;
3541 	struct spdk_blob_store	*bs;
3542 	struct spdk_bs_cpl	cpl;
3543 	spdk_bs_sequence_t	*seq;
3544 	spdk_bs_batch_t		*batch;
3545 	uint64_t		num_md_lba;
3546 	uint64_t		num_md_pages;
3547 	uint64_t		num_md_clusters;
3548 	uint32_t		i;
3549 	struct spdk_bs_opts	opts = {};
3550 	int			rc;
3551 
3552 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3553 
3554 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3555 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3556 			    dev->blocklen);
3557 		dev->destroy(dev);
3558 		cb_fn(cb_arg, NULL, -EINVAL);
3559 		return;
3560 	}
3561 
3562 	if (o) {
3563 		opts = *o;
3564 	} else {
3565 		spdk_bs_opts_init(&opts);
3566 	}
3567 
3568 	if (_spdk_bs_opts_verify(&opts) != 0) {
3569 		dev->destroy(dev);
3570 		cb_fn(cb_arg, NULL, -EINVAL);
3571 		return;
3572 	}
3573 
3574 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3575 	if (rc) {
3576 		dev->destroy(dev);
3577 		cb_fn(cb_arg, NULL, rc);
3578 		return;
3579 	}
3580 
3581 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3582 		/* By default, allocate 1 page per cluster.
3583 		 * Technically, this over-allocates metadata
3584 		 * because more metadata will reduce the number
3585 		 * of usable clusters. This can be addressed with
3586 		 * more complex math in the future.
3587 		 */
3588 		bs->md_len = bs->total_clusters;
3589 	} else {
3590 		bs->md_len = opts.num_md_pages;
3591 	}
3592 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3593 	if (rc < 0) {
3594 		_spdk_bs_free(bs);
3595 		cb_fn(cb_arg, NULL, -ENOMEM);
3596 		return;
3597 	}
3598 
3599 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3600 	if (rc < 0) {
3601 		_spdk_bs_free(bs);
3602 		cb_fn(cb_arg, NULL, -ENOMEM);
3603 		return;
3604 	}
3605 
3606 	ctx = calloc(1, sizeof(*ctx));
3607 	if (!ctx) {
3608 		_spdk_bs_free(bs);
3609 		cb_fn(cb_arg, NULL, -ENOMEM);
3610 		return;
3611 	}
3612 
3613 	ctx->bs = bs;
3614 
3615 	/* Allocate memory for the super block */
3616 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3617 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3618 	if (!ctx->super) {
3619 		free(ctx);
3620 		_spdk_bs_free(bs);
3621 		cb_fn(cb_arg, NULL, -ENOMEM);
3622 		return;
3623 	}
3624 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3625 	       sizeof(ctx->super->signature));
3626 	ctx->super->version = SPDK_BS_VERSION;
3627 	ctx->super->length = sizeof(*ctx->super);
3628 	ctx->super->super_blob = bs->super_blob;
3629 	ctx->super->clean = 0;
3630 	ctx->super->cluster_size = bs->cluster_sz;
3631 	ctx->super->io_unit_size = bs->io_unit_size;
3632 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3633 
3634 	/* Calculate how many pages the metadata consumes at the front
3635 	 * of the disk.
3636 	 */
3637 
3638 	/* The super block uses 1 page */
3639 	num_md_pages = 1;
3640 
3641 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3642 	 * up to the nearest page, plus a header.
3643 	 */
3644 	ctx->super->used_page_mask_start = num_md_pages;
3645 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3646 					 spdk_divide_round_up(bs->md_len, 8),
3647 					 SPDK_BS_PAGE_SIZE);
3648 	num_md_pages += ctx->super->used_page_mask_len;
3649 
3650 	/* The used_clusters mask requires 1 bit per cluster, rounded
3651 	 * up to the nearest page, plus a header.
3652 	 */
3653 	ctx->super->used_cluster_mask_start = num_md_pages;
3654 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3655 					    spdk_divide_round_up(bs->total_clusters, 8),
3656 					    SPDK_BS_PAGE_SIZE);
3657 	num_md_pages += ctx->super->used_cluster_mask_len;
3658 
3659 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3660 	 * up to the nearest page, plus a header.
3661 	 */
3662 	ctx->super->used_blobid_mask_start = num_md_pages;
3663 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3664 					   spdk_divide_round_up(bs->md_len, 8),
3665 					   SPDK_BS_PAGE_SIZE);
3666 	num_md_pages += ctx->super->used_blobid_mask_len;
3667 
3668 	/* The metadata region size was chosen above */
3669 	ctx->super->md_start = bs->md_start = num_md_pages;
3670 	ctx->super->md_len = bs->md_len;
3671 	num_md_pages += bs->md_len;
3672 
3673 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3674 
3675 	ctx->super->size = dev->blockcnt * dev->blocklen;
3676 
3677 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3678 
3679 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
3680 	if (num_md_clusters > bs->total_clusters) {
3681 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3682 			    "please decrease number of pages reserved for metadata "
3683 			    "or increase cluster size.\n");
3684 		spdk_free(ctx->super);
3685 		free(ctx);
3686 		_spdk_bs_free(bs);
3687 		cb_fn(cb_arg, NULL, -ENOMEM);
3688 		return;
3689 	}
3690 	/* Claim all of the clusters used by the metadata */
3691 	for (i = 0; i < num_md_clusters; i++) {
3692 		_spdk_bs_claim_cluster(bs, i);
3693 	}
3694 
3695 	bs->total_data_clusters = bs->num_free_clusters;
3696 
3697 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3698 	cpl.u.bs_handle.cb_fn = cb_fn;
3699 	cpl.u.bs_handle.cb_arg = cb_arg;
3700 	cpl.u.bs_handle.bs = bs;
3701 
3702 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3703 	if (!seq) {
3704 		spdk_free(ctx->super);
3705 		free(ctx);
3706 		_spdk_bs_free(bs);
3707 		cb_fn(cb_arg, NULL, -ENOMEM);
3708 		return;
3709 	}
3710 
3711 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3712 
3713 	/* Clear metadata space */
3714 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3715 
3716 	if (opts.clear_method == BS_CLEAR_WITH_UNMAP) {
3717 		/* Trim data clusters */
3718 		spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3719 	} else if (opts.clear_method == BS_CLEAR_WITH_WRITE_ZEROES) {
3720 		/* Write_zeroes to data clusters */
3721 		spdk_bs_batch_write_zeroes_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3722 	}
3723 
3724 	spdk_bs_batch_close(batch);
3725 }
3726 
3727 /* END spdk_bs_init */
3728 
3729 /* START spdk_bs_destroy */
3730 
3731 static void
3732 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3733 {
3734 	struct spdk_bs_init_ctx *ctx = cb_arg;
3735 	struct spdk_blob_store *bs = ctx->bs;
3736 
3737 	/*
3738 	 * We need to defer calling spdk_bs_call_cpl() until after
3739 	 * dev destruction, so tuck these away for later use.
3740 	 */
3741 	bs->unload_err = bserrno;
3742 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3743 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3744 
3745 	spdk_bs_sequence_finish(seq, bserrno);
3746 
3747 	_spdk_bs_free(bs);
3748 	free(ctx);
3749 }
3750 
3751 void
3752 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3753 		void *cb_arg)
3754 {
3755 	struct spdk_bs_cpl	cpl;
3756 	spdk_bs_sequence_t	*seq;
3757 	struct spdk_bs_init_ctx *ctx;
3758 
3759 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3760 
3761 	if (!TAILQ_EMPTY(&bs->blobs)) {
3762 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3763 		cb_fn(cb_arg, -EBUSY);
3764 		return;
3765 	}
3766 
3767 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3768 	cpl.u.bs_basic.cb_fn = cb_fn;
3769 	cpl.u.bs_basic.cb_arg = cb_arg;
3770 
3771 	ctx = calloc(1, sizeof(*ctx));
3772 	if (!ctx) {
3773 		cb_fn(cb_arg, -ENOMEM);
3774 		return;
3775 	}
3776 
3777 	ctx->bs = bs;
3778 
3779 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3780 	if (!seq) {
3781 		free(ctx);
3782 		cb_fn(cb_arg, -ENOMEM);
3783 		return;
3784 	}
3785 
3786 	/* Write zeroes to the super block */
3787 	spdk_bs_sequence_write_zeroes_dev(seq,
3788 					  _spdk_bs_page_to_lba(bs, 0),
3789 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3790 					  _spdk_bs_destroy_trim_cpl, ctx);
3791 }
3792 
3793 /* END spdk_bs_destroy */
3794 
3795 /* START spdk_bs_unload */
3796 
3797 static void
3798 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3799 {
3800 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3801 
3802 	spdk_free(ctx->super);
3803 
3804 	/*
3805 	 * We need to defer calling spdk_bs_call_cpl() until after
3806 	 * dev destruction, so tuck these away for later use.
3807 	 */
3808 	ctx->bs->unload_err = bserrno;
3809 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3810 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3811 
3812 	spdk_bs_sequence_finish(seq, bserrno);
3813 
3814 	_spdk_bs_free(ctx->bs);
3815 	free(ctx);
3816 }
3817 
3818 static void
3819 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3820 {
3821 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3822 
3823 	spdk_free(ctx->mask);
3824 	ctx->super->clean = 1;
3825 
3826 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3827 }
3828 
3829 static void
3830 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3831 {
3832 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3833 
3834 	spdk_free(ctx->mask);
3835 	ctx->mask = NULL;
3836 
3837 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3838 }
3839 
3840 static void
3841 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3842 {
3843 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3844 
3845 	spdk_free(ctx->mask);
3846 	ctx->mask = NULL;
3847 
3848 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3849 }
3850 
3851 static void
3852 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3853 {
3854 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3855 }
3856 
3857 void
3858 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3859 {
3860 	struct spdk_bs_cpl	cpl;
3861 	spdk_bs_sequence_t	*seq;
3862 	struct spdk_bs_load_ctx *ctx;
3863 
3864 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3865 
3866 	if (!TAILQ_EMPTY(&bs->blobs)) {
3867 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3868 		cb_fn(cb_arg, -EBUSY);
3869 		return;
3870 	}
3871 
3872 	ctx = calloc(1, sizeof(*ctx));
3873 	if (!ctx) {
3874 		cb_fn(cb_arg, -ENOMEM);
3875 		return;
3876 	}
3877 
3878 	ctx->bs = bs;
3879 
3880 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3881 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3882 	if (!ctx->super) {
3883 		free(ctx);
3884 		cb_fn(cb_arg, -ENOMEM);
3885 		return;
3886 	}
3887 
3888 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3889 	cpl.u.bs_basic.cb_fn = cb_fn;
3890 	cpl.u.bs_basic.cb_arg = cb_arg;
3891 
3892 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3893 	if (!seq) {
3894 		spdk_free(ctx->super);
3895 		free(ctx);
3896 		cb_fn(cb_arg, -ENOMEM);
3897 		return;
3898 	}
3899 
3900 	/* Read super block */
3901 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3902 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3903 				  _spdk_bs_unload_read_super_cpl, ctx);
3904 }
3905 
3906 /* END spdk_bs_unload */
3907 
3908 /* START spdk_bs_set_super */
3909 
3910 struct spdk_bs_set_super_ctx {
3911 	struct spdk_blob_store		*bs;
3912 	struct spdk_bs_super_block	*super;
3913 };
3914 
3915 static void
3916 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3917 {
3918 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3919 
3920 	if (bserrno != 0) {
3921 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3922 	}
3923 
3924 	spdk_free(ctx->super);
3925 
3926 	spdk_bs_sequence_finish(seq, bserrno);
3927 
3928 	free(ctx);
3929 }
3930 
3931 static void
3932 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3933 {
3934 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3935 
3936 	if (bserrno != 0) {
3937 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3938 		spdk_free(ctx->super);
3939 		spdk_bs_sequence_finish(seq, bserrno);
3940 		free(ctx);
3941 		return;
3942 	}
3943 
3944 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3945 }
3946 
3947 void
3948 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3949 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3950 {
3951 	struct spdk_bs_cpl		cpl;
3952 	spdk_bs_sequence_t		*seq;
3953 	struct spdk_bs_set_super_ctx	*ctx;
3954 
3955 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3956 
3957 	ctx = calloc(1, sizeof(*ctx));
3958 	if (!ctx) {
3959 		cb_fn(cb_arg, -ENOMEM);
3960 		return;
3961 	}
3962 
3963 	ctx->bs = bs;
3964 
3965 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3966 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3967 	if (!ctx->super) {
3968 		free(ctx);
3969 		cb_fn(cb_arg, -ENOMEM);
3970 		return;
3971 	}
3972 
3973 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3974 	cpl.u.bs_basic.cb_fn = cb_fn;
3975 	cpl.u.bs_basic.cb_arg = cb_arg;
3976 
3977 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3978 	if (!seq) {
3979 		spdk_free(ctx->super);
3980 		free(ctx);
3981 		cb_fn(cb_arg, -ENOMEM);
3982 		return;
3983 	}
3984 
3985 	bs->super_blob = blobid;
3986 
3987 	/* Read super block */
3988 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3989 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3990 				  _spdk_bs_set_super_read_cpl, ctx);
3991 }
3992 
3993 /* END spdk_bs_set_super */
3994 
3995 void
3996 spdk_bs_get_super(struct spdk_blob_store *bs,
3997 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
3998 {
3999 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
4000 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
4001 	} else {
4002 		cb_fn(cb_arg, bs->super_blob, 0);
4003 	}
4004 }
4005 
4006 uint64_t
4007 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
4008 {
4009 	return bs->cluster_sz;
4010 }
4011 
4012 uint64_t
4013 spdk_bs_get_page_size(struct spdk_blob_store *bs)
4014 {
4015 	return SPDK_BS_PAGE_SIZE;
4016 }
4017 
4018 uint64_t
4019 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
4020 {
4021 	return bs->io_unit_size;
4022 }
4023 
4024 uint64_t
4025 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
4026 {
4027 	return bs->num_free_clusters;
4028 }
4029 
4030 uint64_t
4031 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
4032 {
4033 	return bs->total_data_clusters;
4034 }
4035 
4036 static int
4037 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
4038 {
4039 	bs->md_channel = spdk_get_io_channel(bs);
4040 	if (!bs->md_channel) {
4041 		SPDK_ERRLOG("Failed to get IO channel.\n");
4042 		return -1;
4043 	}
4044 
4045 	return 0;
4046 }
4047 
4048 static int
4049 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
4050 {
4051 	spdk_put_io_channel(bs->md_channel);
4052 
4053 	return 0;
4054 }
4055 
4056 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4057 {
4058 	assert(blob != NULL);
4059 
4060 	return blob->id;
4061 }
4062 
4063 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4064 {
4065 	assert(blob != NULL);
4066 
4067 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4068 }
4069 
4070 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
4071 {
4072 	assert(blob != NULL);
4073 
4074 	return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs);
4075 }
4076 
4077 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4078 {
4079 	assert(blob != NULL);
4080 
4081 	return blob->active.num_clusters;
4082 }
4083 
4084 /* START spdk_bs_create_blob */
4085 
4086 static void
4087 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4088 {
4089 	struct spdk_blob *blob = cb_arg;
4090 
4091 	_spdk_blob_free(blob);
4092 
4093 	spdk_bs_sequence_finish(seq, bserrno);
4094 }
4095 
4096 static int
4097 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4098 		      bool internal)
4099 {
4100 	uint64_t i;
4101 	size_t value_len = 0;
4102 	int rc;
4103 	const void *value = NULL;
4104 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4105 		return -EINVAL;
4106 	}
4107 	for (i = 0; i < xattrs->count; i++) {
4108 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4109 		if (value == NULL || value_len == 0) {
4110 			return -EINVAL;
4111 		}
4112 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4113 		if (rc < 0) {
4114 			return rc;
4115 		}
4116 	}
4117 	return 0;
4118 }
4119 
4120 static void
4121 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4122 {
4123 	_spdk_blob_verify_md_op(blob);
4124 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4125 	blob->state = SPDK_BLOB_STATE_DIRTY;
4126 }
4127 
4128 static void
4129 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4130 		     const struct spdk_blob_opts *opts,
4131 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4132 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4133 {
4134 	struct spdk_blob	*blob;
4135 	uint32_t		page_idx;
4136 	struct spdk_bs_cpl	cpl;
4137 	struct spdk_blob_opts	opts_default;
4138 	struct spdk_blob_xattr_opts internal_xattrs_default;
4139 	spdk_bs_sequence_t	*seq;
4140 	spdk_blob_id		id;
4141 	int rc;
4142 
4143 	assert(spdk_get_thread() == bs->md_thread);
4144 
4145 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4146 	if (page_idx == UINT32_MAX) {
4147 		cb_fn(cb_arg, 0, -ENOMEM);
4148 		return;
4149 	}
4150 	spdk_bit_array_set(bs->used_blobids, page_idx);
4151 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4152 
4153 	id = _spdk_bs_page_to_blobid(page_idx);
4154 
4155 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4156 
4157 	blob = _spdk_blob_alloc(bs, id);
4158 	if (!blob) {
4159 		cb_fn(cb_arg, 0, -ENOMEM);
4160 		return;
4161 	}
4162 
4163 	if (!opts) {
4164 		spdk_blob_opts_init(&opts_default);
4165 		opts = &opts_default;
4166 	}
4167 	if (!internal_xattrs) {
4168 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4169 		internal_xattrs = &internal_xattrs_default;
4170 	}
4171 
4172 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4173 	if (rc < 0) {
4174 		_spdk_blob_free(blob);
4175 		cb_fn(cb_arg, 0, rc);
4176 		return;
4177 	}
4178 
4179 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4180 	if (rc < 0) {
4181 		_spdk_blob_free(blob);
4182 		cb_fn(cb_arg, 0, rc);
4183 		return;
4184 	}
4185 
4186 	if (opts->thin_provision) {
4187 		_spdk_blob_set_thin_provision(blob);
4188 	}
4189 
4190 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4191 	if (rc < 0) {
4192 		_spdk_blob_free(blob);
4193 		cb_fn(cb_arg, 0, rc);
4194 		return;
4195 	}
4196 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4197 	cpl.u.blobid.cb_fn = cb_fn;
4198 	cpl.u.blobid.cb_arg = cb_arg;
4199 	cpl.u.blobid.blobid = blob->id;
4200 
4201 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4202 	if (!seq) {
4203 		_spdk_blob_free(blob);
4204 		cb_fn(cb_arg, 0, -ENOMEM);
4205 		return;
4206 	}
4207 
4208 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4209 }
4210 
4211 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4212 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4213 {
4214 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4215 }
4216 
4217 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4218 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4219 {
4220 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4221 }
4222 
4223 /* END spdk_bs_create_blob */
4224 
4225 /* START blob_cleanup */
4226 
4227 struct spdk_clone_snapshot_ctx {
4228 	struct spdk_bs_cpl      cpl;
4229 	int bserrno;
4230 	bool frozen;
4231 
4232 	struct spdk_io_channel *channel;
4233 
4234 	/* Current cluster for inflate operation */
4235 	uint64_t cluster;
4236 
4237 	/* For inflation force allocation of all unallocated clusters and remove
4238 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4239 	bool allocate_all;
4240 
4241 	struct {
4242 		spdk_blob_id id;
4243 		struct spdk_blob *blob;
4244 	} original;
4245 	struct {
4246 		spdk_blob_id id;
4247 		struct spdk_blob *blob;
4248 	} new;
4249 
4250 	/* xattrs specified for snapshot/clones only. They have no impact on
4251 	 * the original blobs xattrs. */
4252 	const struct spdk_blob_xattr_opts *xattrs;
4253 };
4254 
4255 static void
4256 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4257 {
4258 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4259 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4260 
4261 	if (bserrno != 0) {
4262 		if (ctx->bserrno != 0) {
4263 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4264 		} else {
4265 			ctx->bserrno = bserrno;
4266 		}
4267 	}
4268 
4269 	switch (cpl->type) {
4270 	case SPDK_BS_CPL_TYPE_BLOBID:
4271 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4272 		break;
4273 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4274 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4275 		break;
4276 	default:
4277 		SPDK_UNREACHABLE();
4278 		break;
4279 	}
4280 
4281 	free(ctx);
4282 }
4283 
4284 static void
4285 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4286 {
4287 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4288 	struct spdk_blob *origblob = ctx->original.blob;
4289 
4290 	if (bserrno != 0) {
4291 		if (ctx->bserrno != 0) {
4292 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4293 		} else {
4294 			ctx->bserrno = bserrno;
4295 		}
4296 	}
4297 
4298 	ctx->original.id = origblob->id;
4299 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4300 }
4301 
4302 static void
4303 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4304 {
4305 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4306 	struct spdk_blob *origblob = ctx->original.blob;
4307 
4308 	if (bserrno != 0) {
4309 		if (ctx->bserrno != 0) {
4310 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4311 		} else {
4312 			ctx->bserrno = bserrno;
4313 		}
4314 	}
4315 
4316 	if (ctx->frozen) {
4317 		/* Unfreeze any outstanding I/O */
4318 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4319 	} else {
4320 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4321 	}
4322 
4323 }
4324 
4325 static void
4326 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4327 {
4328 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4329 	struct spdk_blob *newblob = ctx->new.blob;
4330 
4331 	if (bserrno != 0) {
4332 		if (ctx->bserrno != 0) {
4333 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4334 		} else {
4335 			ctx->bserrno = bserrno;
4336 		}
4337 	}
4338 
4339 	ctx->new.id = newblob->id;
4340 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4341 }
4342 
4343 /* END blob_cleanup */
4344 
4345 /* START spdk_bs_create_snapshot */
4346 
4347 static void
4348 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4349 {
4350 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4351 	struct spdk_blob *newblob = ctx->new.blob;
4352 
4353 	if (bserrno != 0) {
4354 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4355 		return;
4356 	}
4357 
4358 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4359 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4360 	if (bserrno != 0) {
4361 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4362 		return;
4363 	}
4364 
4365 	_spdk_bs_blob_list_add(ctx->original.blob);
4366 
4367 	spdk_blob_set_read_only(newblob);
4368 
4369 	/* sync snapshot metadata */
4370 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4371 }
4372 
4373 static void
4374 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4375 {
4376 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4377 	struct spdk_blob *origblob = ctx->original.blob;
4378 	struct spdk_blob *newblob = ctx->new.blob;
4379 
4380 	if (bserrno != 0) {
4381 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4382 		return;
4383 	}
4384 
4385 	/* Set internal xattr for snapshot id */
4386 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4387 	if (bserrno != 0) {
4388 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4389 		return;
4390 	}
4391 
4392 	_spdk_bs_blob_list_remove(origblob);
4393 	origblob->parent_id = newblob->id;
4394 
4395 	/* Create new back_bs_dev for snapshot */
4396 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4397 	if (origblob->back_bs_dev == NULL) {
4398 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4399 		return;
4400 	}
4401 
4402 	/* set clone blob as thin provisioned */
4403 	_spdk_blob_set_thin_provision(origblob);
4404 
4405 	_spdk_bs_blob_list_add(newblob);
4406 
4407 	/* Zero out origblob cluster map */
4408 	memset(origblob->active.clusters, 0,
4409 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4410 
4411 	/* sync clone metadata */
4412 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4413 }
4414 
4415 static void
4416 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4417 {
4418 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4419 	struct spdk_blob *origblob = ctx->original.blob;
4420 	struct spdk_blob *newblob = ctx->new.blob;
4421 	int bserrno;
4422 
4423 	if (rc != 0) {
4424 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4425 		return;
4426 	}
4427 
4428 	ctx->frozen = true;
4429 
4430 	/* set new back_bs_dev for snapshot */
4431 	newblob->back_bs_dev = origblob->back_bs_dev;
4432 	/* Set invalid flags from origblob */
4433 	newblob->invalid_flags = origblob->invalid_flags;
4434 
4435 	/* inherit parent from original blob if set */
4436 	newblob->parent_id = origblob->parent_id;
4437 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4438 		/* Set internal xattr for snapshot id */
4439 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4440 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4441 		if (bserrno != 0) {
4442 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4443 			return;
4444 		}
4445 	}
4446 
4447 	/* Copy cluster map to snapshot */
4448 	memcpy(newblob->active.clusters, origblob->active.clusters,
4449 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4450 
4451 	/* sync snapshot metadata */
4452 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4453 }
4454 
4455 static void
4456 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4457 {
4458 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4459 	struct spdk_blob *origblob = ctx->original.blob;
4460 	struct spdk_blob *newblob = _blob;
4461 
4462 	if (bserrno != 0) {
4463 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4464 		return;
4465 	}
4466 
4467 	ctx->new.blob = newblob;
4468 
4469 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4470 }
4471 
4472 static void
4473 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4474 {
4475 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4476 	struct spdk_blob *origblob = ctx->original.blob;
4477 
4478 	if (bserrno != 0) {
4479 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4480 		return;
4481 	}
4482 
4483 	ctx->new.id = blobid;
4484 	ctx->cpl.u.blobid.blobid = blobid;
4485 
4486 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4487 }
4488 
4489 
4490 static void
4491 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4492 			const void **value, size_t *value_len)
4493 {
4494 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4495 
4496 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4497 	*value = &blob->id;
4498 	*value_len = sizeof(blob->id);
4499 }
4500 
4501 static void
4502 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4503 {
4504 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4505 	struct spdk_blob_opts opts;
4506 	struct spdk_blob_xattr_opts internal_xattrs;
4507 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4508 
4509 	if (bserrno != 0) {
4510 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4511 		return;
4512 	}
4513 
4514 	ctx->original.blob = _blob;
4515 
4516 	if (_blob->data_ro || _blob->md_ro) {
4517 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4518 			      _blob->id);
4519 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4520 		return;
4521 	}
4522 
4523 	spdk_blob_opts_init(&opts);
4524 	_spdk_blob_xattrs_init(&internal_xattrs);
4525 
4526 	/* Change the size of new blob to the same as in original blob,
4527 	 * but do not allocate clusters */
4528 	opts.thin_provision = true;
4529 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4530 
4531 	/* If there are any xattrs specified for snapshot, set them now */
4532 	if (ctx->xattrs) {
4533 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4534 	}
4535 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4536 	internal_xattrs.count = 1;
4537 	internal_xattrs.ctx = _blob;
4538 	internal_xattrs.names = xattrs_names;
4539 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4540 
4541 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4542 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4543 }
4544 
4545 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4546 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4547 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4548 {
4549 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4550 
4551 	if (!ctx) {
4552 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4553 		return;
4554 	}
4555 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4556 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4557 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4558 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4559 	ctx->bserrno = 0;
4560 	ctx->frozen = false;
4561 	ctx->original.id = blobid;
4562 	ctx->xattrs = snapshot_xattrs;
4563 
4564 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4565 }
4566 /* END spdk_bs_create_snapshot */
4567 
4568 /* START spdk_bs_create_clone */
4569 
4570 static void
4571 _spdk_bs_xattr_clone(void *arg, const char *name,
4572 		     const void **value, size_t *value_len)
4573 {
4574 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4575 
4576 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4577 	*value = &blob->id;
4578 	*value_len = sizeof(blob->id);
4579 }
4580 
4581 static void
4582 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4583 {
4584 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4585 	struct spdk_blob *clone = _blob;
4586 
4587 	ctx->new.blob = clone;
4588 	_spdk_bs_blob_list_add(clone);
4589 
4590 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4591 }
4592 
4593 static void
4594 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4595 {
4596 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4597 
4598 	ctx->cpl.u.blobid.blobid = blobid;
4599 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4600 }
4601 
4602 static void
4603 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4604 {
4605 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4606 	struct spdk_blob_opts		opts;
4607 	struct spdk_blob_xattr_opts internal_xattrs;
4608 	char *xattr_names[] = { BLOB_SNAPSHOT };
4609 
4610 	if (bserrno != 0) {
4611 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4612 		return;
4613 	}
4614 
4615 	ctx->original.blob = _blob;
4616 
4617 	if (!_blob->data_ro || !_blob->md_ro) {
4618 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4619 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4620 		return;
4621 	}
4622 
4623 	spdk_blob_opts_init(&opts);
4624 	_spdk_blob_xattrs_init(&internal_xattrs);
4625 
4626 	opts.thin_provision = true;
4627 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4628 	if (ctx->xattrs) {
4629 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4630 	}
4631 
4632 	/* Set internal xattr BLOB_SNAPSHOT */
4633 	internal_xattrs.count = 1;
4634 	internal_xattrs.ctx = _blob;
4635 	internal_xattrs.names = xattr_names;
4636 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4637 
4638 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4639 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4640 }
4641 
4642 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4643 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4644 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4645 {
4646 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4647 
4648 	if (!ctx) {
4649 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4650 		return;
4651 	}
4652 
4653 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4654 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4655 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4656 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4657 	ctx->bserrno = 0;
4658 	ctx->xattrs = clone_xattrs;
4659 	ctx->original.id = blobid;
4660 
4661 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4662 }
4663 
4664 /* END spdk_bs_create_clone */
4665 
4666 /* START spdk_bs_inflate_blob */
4667 
4668 static void
4669 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4670 {
4671 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4672 	struct spdk_blob *_blob = ctx->original.blob;
4673 
4674 	if (bserrno != 0) {
4675 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4676 		return;
4677 	}
4678 
4679 	assert(_parent != NULL);
4680 
4681 	_spdk_bs_blob_list_remove(_blob);
4682 	_blob->parent_id = _parent->id;
4683 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4684 			     sizeof(spdk_blob_id), true);
4685 
4686 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4687 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4688 	_spdk_bs_blob_list_add(_blob);
4689 
4690 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4691 }
4692 
4693 static void
4694 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4695 {
4696 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4697 	struct spdk_blob *_blob = ctx->original.blob;
4698 	struct spdk_blob *_parent;
4699 
4700 	if (bserrno != 0) {
4701 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4702 		return;
4703 	}
4704 
4705 	if (ctx->allocate_all) {
4706 		/* remove thin provisioning */
4707 		_spdk_bs_blob_list_remove(_blob);
4708 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4709 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4710 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4711 		_blob->back_bs_dev = NULL;
4712 		_blob->parent_id = SPDK_BLOBID_INVALID;
4713 	} else {
4714 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4715 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4716 			/* We must change the parent of the inflated blob */
4717 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4718 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4719 			return;
4720 		}
4721 
4722 		_spdk_bs_blob_list_remove(_blob);
4723 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4724 		_blob->parent_id = SPDK_BLOBID_INVALID;
4725 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4726 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4727 	}
4728 
4729 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4730 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4731 }
4732 
4733 /* Check if cluster needs allocation */
4734 static inline bool
4735 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4736 {
4737 	struct spdk_blob_bs_dev *b;
4738 
4739 	assert(blob != NULL);
4740 
4741 	if (blob->active.clusters[cluster] != 0) {
4742 		/* Cluster is already allocated */
4743 		return false;
4744 	}
4745 
4746 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4747 		/* Blob have no parent blob */
4748 		return allocate_all;
4749 	}
4750 
4751 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4752 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4753 }
4754 
4755 static void
4756 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4757 {
4758 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4759 	struct spdk_blob *_blob = ctx->original.blob;
4760 	uint64_t offset;
4761 
4762 	if (bserrno != 0) {
4763 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4764 		return;
4765 	}
4766 
4767 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4768 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4769 			break;
4770 		}
4771 	}
4772 
4773 	if (ctx->cluster < _blob->active.num_clusters) {
4774 		offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster);
4775 
4776 		/* We may safely increment a cluster before write */
4777 		ctx->cluster++;
4778 
4779 		/* Use zero length write to touch a cluster */
4780 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4781 				   _spdk_bs_inflate_blob_touch_next, ctx);
4782 	} else {
4783 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4784 	}
4785 }
4786 
4787 static void
4788 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4789 {
4790 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4791 	uint64_t lfc; /* lowest free cluster */
4792 	uint64_t i;
4793 
4794 	if (bserrno != 0) {
4795 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4796 		return;
4797 	}
4798 	ctx->original.blob = _blob;
4799 
4800 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4801 		/* This blob have no parent, so we cannot decouple it. */
4802 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4803 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4804 		return;
4805 	}
4806 
4807 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4808 		/* This is not thin provisioned blob. No need to inflate. */
4809 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4810 		return;
4811 	}
4812 
4813 	/* Do two passes - one to verify that we can obtain enough clusters
4814 	 * and another to actually claim them.
4815 	 */
4816 	lfc = 0;
4817 	for (i = 0; i < _blob->active.num_clusters; i++) {
4818 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4819 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4820 			if (lfc == UINT32_MAX) {
4821 				/* No more free clusters. Cannot satisfy the request */
4822 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4823 				return;
4824 			}
4825 			lfc++;
4826 		}
4827 	}
4828 
4829 	ctx->cluster = 0;
4830 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4831 }
4832 
4833 static void
4834 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4835 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4836 {
4837 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4838 
4839 	if (!ctx) {
4840 		cb_fn(cb_arg, -ENOMEM);
4841 		return;
4842 	}
4843 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4844 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4845 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4846 	ctx->bserrno = 0;
4847 	ctx->original.id = blobid;
4848 	ctx->channel = channel;
4849 	ctx->allocate_all = allocate_all;
4850 
4851 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4852 }
4853 
4854 void
4855 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4856 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4857 {
4858 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4859 }
4860 
4861 void
4862 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4863 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4864 {
4865 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
4866 }
4867 /* END spdk_bs_inflate_blob */
4868 
4869 /* START spdk_blob_resize */
4870 struct spdk_bs_resize_ctx {
4871 	spdk_blob_op_complete cb_fn;
4872 	void *cb_arg;
4873 	struct spdk_blob *blob;
4874 	uint64_t sz;
4875 	int rc;
4876 };
4877 
4878 static void
4879 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4880 {
4881 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4882 
4883 	if (rc != 0) {
4884 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4885 	}
4886 
4887 	if (ctx->rc != 0) {
4888 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4889 		rc = ctx->rc;
4890 	}
4891 
4892 	ctx->blob->locked_operation_in_progress = false;
4893 
4894 	ctx->cb_fn(ctx->cb_arg, rc);
4895 	free(ctx);
4896 }
4897 
4898 static void
4899 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4900 {
4901 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4902 
4903 	if (rc != 0) {
4904 		ctx->blob->locked_operation_in_progress = false;
4905 		ctx->cb_fn(ctx->cb_arg, rc);
4906 		free(ctx);
4907 		return;
4908 	}
4909 
4910 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4911 
4912 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4913 }
4914 
4915 void
4916 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4917 {
4918 	struct spdk_bs_resize_ctx *ctx;
4919 
4920 	_spdk_blob_verify_md_op(blob);
4921 
4922 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4923 
4924 	if (blob->md_ro) {
4925 		cb_fn(cb_arg, -EPERM);
4926 		return;
4927 	}
4928 
4929 	if (sz == blob->active.num_clusters) {
4930 		cb_fn(cb_arg, 0);
4931 		return;
4932 	}
4933 
4934 	if (blob->locked_operation_in_progress) {
4935 		cb_fn(cb_arg, -EBUSY);
4936 		return;
4937 	}
4938 
4939 	ctx = calloc(1, sizeof(*ctx));
4940 	if (!ctx) {
4941 		cb_fn(cb_arg, -ENOMEM);
4942 		return;
4943 	}
4944 
4945 	blob->locked_operation_in_progress = true;
4946 	ctx->cb_fn = cb_fn;
4947 	ctx->cb_arg = cb_arg;
4948 	ctx->blob = blob;
4949 	ctx->sz = sz;
4950 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
4951 }
4952 
4953 /* END spdk_blob_resize */
4954 
4955 
4956 /* START spdk_bs_delete_blob */
4957 
4958 static void
4959 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
4960 {
4961 	spdk_bs_sequence_t *seq = cb_arg;
4962 
4963 	spdk_bs_sequence_finish(seq, bserrno);
4964 }
4965 
4966 static void
4967 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4968 {
4969 	struct spdk_blob *blob = cb_arg;
4970 
4971 	if (bserrno != 0) {
4972 		/*
4973 		 * We already removed this blob from the blobstore tailq, so
4974 		 *  we need to free it here since this is the last reference
4975 		 *  to it.
4976 		 */
4977 		_spdk_blob_free(blob);
4978 		_spdk_bs_delete_close_cpl(seq, bserrno);
4979 		return;
4980 	}
4981 
4982 	/*
4983 	 * This will immediately decrement the ref_count and call
4984 	 *  the completion routine since the metadata state is clean.
4985 	 *  By calling spdk_blob_close, we reduce the number of call
4986 	 *  points into code that touches the blob->open_ref count
4987 	 *  and the blobstore's blob list.
4988 	 */
4989 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
4990 }
4991 
4992 static void
4993 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
4994 {
4995 	spdk_bs_sequence_t *seq = cb_arg;
4996 	struct spdk_blob_list *snapshot = NULL;
4997 	uint32_t page_num;
4998 
4999 	if (bserrno != 0) {
5000 		spdk_bs_sequence_finish(seq, bserrno);
5001 		return;
5002 	}
5003 
5004 	_spdk_blob_verify_md_op(blob);
5005 
5006 	if (blob->open_ref > 1) {
5007 		/*
5008 		 * Someone has this blob open (besides this delete context).
5009 		 *  Decrement the ref count directly and return -EBUSY.
5010 		 */
5011 		blob->open_ref--;
5012 		spdk_bs_sequence_finish(seq, -EBUSY);
5013 		return;
5014 	}
5015 
5016 	bserrno = _spdk_bs_blob_list_remove(blob);
5017 	if (bserrno != 0) {
5018 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Remove blob #%" PRIu64 " from a list\n", blob->id);
5019 		spdk_bs_sequence_finish(seq, bserrno);
5020 		return;
5021 	}
5022 
5023 	/*
5024 	 * Remove the blob from the blob_store list now, to ensure it does not
5025 	 *  get returned after this point by _spdk_blob_lookup().
5026 	 */
5027 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5028 
5029 	/* If blob is a snapshot then remove it from the list */
5030 	TAILQ_FOREACH(snapshot, &blob->bs->snapshots, link) {
5031 		if (snapshot->id == blob->id) {
5032 			TAILQ_REMOVE(&blob->bs->snapshots, snapshot, link);
5033 			free(snapshot);
5034 			break;
5035 		}
5036 	}
5037 
5038 	page_num = _spdk_bs_blobid_to_page(blob->id);
5039 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
5040 	blob->state = SPDK_BLOB_STATE_DIRTY;
5041 	blob->active.num_pages = 0;
5042 	_spdk_blob_resize(blob, 0);
5043 
5044 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
5045 }
5046 
5047 void
5048 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5049 		    spdk_blob_op_complete cb_fn, void *cb_arg)
5050 {
5051 	struct spdk_bs_cpl	cpl;
5052 	spdk_bs_sequence_t	*seq;
5053 	struct spdk_blob_list	*snapshot_entry = NULL;
5054 
5055 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
5056 
5057 	assert(spdk_get_thread() == bs->md_thread);
5058 
5059 	/* Check if this is a snapshot with clones */
5060 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5061 		if (snapshot_entry->id == blobid) {
5062 			break;
5063 		}
5064 	}
5065 	if (snapshot_entry != NULL) {
5066 		/* If snapshot have clones, we cannot remove it */
5067 		if (!TAILQ_EMPTY(&snapshot_entry->clones)) {
5068 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
5069 			cb_fn(cb_arg, -EBUSY);
5070 			return;
5071 		}
5072 	}
5073 
5074 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5075 	cpl.u.blob_basic.cb_fn = cb_fn;
5076 	cpl.u.blob_basic.cb_arg = cb_arg;
5077 
5078 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5079 	if (!seq) {
5080 		cb_fn(cb_arg, -ENOMEM);
5081 		return;
5082 	}
5083 
5084 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5085 }
5086 
5087 /* END spdk_bs_delete_blob */
5088 
5089 /* START spdk_bs_open_blob */
5090 
5091 static void
5092 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5093 {
5094 	struct spdk_blob *blob = cb_arg;
5095 
5096 	/* If the blob have crc error, we just return NULL. */
5097 	if (blob == NULL) {
5098 		seq->cpl.u.blob_handle.blob = NULL;
5099 		spdk_bs_sequence_finish(seq, bserrno);
5100 		return;
5101 	}
5102 
5103 	blob->open_ref++;
5104 
5105 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5106 
5107 	spdk_bs_sequence_finish(seq, bserrno);
5108 }
5109 
5110 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5111 			       struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5112 {
5113 	struct spdk_blob		*blob;
5114 	struct spdk_bs_cpl		cpl;
5115 	struct spdk_blob_open_opts	opts_default;
5116 	spdk_bs_sequence_t		*seq;
5117 	uint32_t			page_num;
5118 
5119 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5120 	assert(spdk_get_thread() == bs->md_thread);
5121 
5122 	page_num = _spdk_bs_blobid_to_page(blobid);
5123 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5124 		/* Invalid blobid */
5125 		cb_fn(cb_arg, NULL, -ENOENT);
5126 		return;
5127 	}
5128 
5129 	blob = _spdk_blob_lookup(bs, blobid);
5130 	if (blob) {
5131 		blob->open_ref++;
5132 		cb_fn(cb_arg, blob, 0);
5133 		return;
5134 	}
5135 
5136 	blob = _spdk_blob_alloc(bs, blobid);
5137 	if (!blob) {
5138 		cb_fn(cb_arg, NULL, -ENOMEM);
5139 		return;
5140 	}
5141 
5142 	if (!opts) {
5143 		spdk_blob_open_opts_init(&opts_default);
5144 		opts = &opts_default;
5145 	}
5146 
5147 	blob->clear_method = opts->clear_method;
5148 
5149 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5150 	cpl.u.blob_handle.cb_fn = cb_fn;
5151 	cpl.u.blob_handle.cb_arg = cb_arg;
5152 	cpl.u.blob_handle.blob = blob;
5153 
5154 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5155 	if (!seq) {
5156 		_spdk_blob_free(blob);
5157 		cb_fn(cb_arg, NULL, -ENOMEM);
5158 		return;
5159 	}
5160 
5161 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5162 }
5163 
5164 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5165 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5166 {
5167 	_spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
5168 }
5169 
5170 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
5171 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5172 {
5173 	_spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
5174 }
5175 
5176 /* END spdk_bs_open_blob */
5177 
5178 /* START spdk_blob_set_read_only */
5179 int spdk_blob_set_read_only(struct spdk_blob *blob)
5180 {
5181 	_spdk_blob_verify_md_op(blob);
5182 
5183 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5184 
5185 	blob->state = SPDK_BLOB_STATE_DIRTY;
5186 	return 0;
5187 }
5188 /* END spdk_blob_set_read_only */
5189 
5190 /* START spdk_blob_sync_md */
5191 
5192 static void
5193 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5194 {
5195 	struct spdk_blob *blob = cb_arg;
5196 
5197 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5198 		blob->data_ro = true;
5199 		blob->md_ro = true;
5200 	}
5201 
5202 	spdk_bs_sequence_finish(seq, bserrno);
5203 }
5204 
5205 static void
5206 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5207 {
5208 	struct spdk_bs_cpl	cpl;
5209 	spdk_bs_sequence_t	*seq;
5210 
5211 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5212 	cpl.u.blob_basic.cb_fn = cb_fn;
5213 	cpl.u.blob_basic.cb_arg = cb_arg;
5214 
5215 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5216 	if (!seq) {
5217 		cb_fn(cb_arg, -ENOMEM);
5218 		return;
5219 	}
5220 
5221 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5222 }
5223 
5224 void
5225 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5226 {
5227 	_spdk_blob_verify_md_op(blob);
5228 
5229 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5230 
5231 	if (blob->md_ro) {
5232 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5233 		cb_fn(cb_arg, 0);
5234 		return;
5235 	}
5236 
5237 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5238 }
5239 
5240 /* END spdk_blob_sync_md */
5241 
5242 struct spdk_blob_insert_cluster_ctx {
5243 	struct spdk_thread	*thread;
5244 	struct spdk_blob	*blob;
5245 	uint32_t		cluster_num;	/* cluster index in blob */
5246 	uint32_t		cluster;	/* cluster on disk */
5247 	int			rc;
5248 	spdk_blob_op_complete	cb_fn;
5249 	void			*cb_arg;
5250 };
5251 
5252 static void
5253 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5254 {
5255 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5256 
5257 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5258 	free(ctx);
5259 }
5260 
5261 static void
5262 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5263 {
5264 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5265 
5266 	ctx->rc = bserrno;
5267 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5268 }
5269 
5270 static void
5271 _spdk_blob_insert_cluster_msg(void *arg)
5272 {
5273 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5274 
5275 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5276 	if (ctx->rc != 0) {
5277 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5278 		return;
5279 	}
5280 
5281 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5282 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5283 }
5284 
5285 static void
5286 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5287 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5288 {
5289 	struct spdk_blob_insert_cluster_ctx *ctx;
5290 
5291 	ctx = calloc(1, sizeof(*ctx));
5292 	if (ctx == NULL) {
5293 		cb_fn(cb_arg, -ENOMEM);
5294 		return;
5295 	}
5296 
5297 	ctx->thread = spdk_get_thread();
5298 	ctx->blob = blob;
5299 	ctx->cluster_num = cluster_num;
5300 	ctx->cluster = cluster;
5301 	ctx->cb_fn = cb_fn;
5302 	ctx->cb_arg = cb_arg;
5303 
5304 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5305 }
5306 
5307 /* START spdk_blob_close */
5308 
5309 static void
5310 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5311 {
5312 	struct spdk_blob *blob = cb_arg;
5313 
5314 	if (bserrno == 0) {
5315 		blob->open_ref--;
5316 		if (blob->open_ref == 0) {
5317 			/*
5318 			 * Blobs with active.num_pages == 0 are deleted blobs.
5319 			 *  these blobs are removed from the blob_store list
5320 			 *  when the deletion process starts - so don't try to
5321 			 *  remove them again.
5322 			 */
5323 			if (blob->active.num_pages > 0) {
5324 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5325 			}
5326 			_spdk_blob_free(blob);
5327 		}
5328 	}
5329 
5330 	spdk_bs_sequence_finish(seq, bserrno);
5331 }
5332 
5333 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5334 {
5335 	struct spdk_bs_cpl	cpl;
5336 	spdk_bs_sequence_t	*seq;
5337 
5338 	_spdk_blob_verify_md_op(blob);
5339 
5340 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5341 
5342 	if (blob->open_ref == 0) {
5343 		cb_fn(cb_arg, -EBADF);
5344 		return;
5345 	}
5346 
5347 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5348 	cpl.u.blob_basic.cb_fn = cb_fn;
5349 	cpl.u.blob_basic.cb_arg = cb_arg;
5350 
5351 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5352 	if (!seq) {
5353 		cb_fn(cb_arg, -ENOMEM);
5354 		return;
5355 	}
5356 
5357 	/* Sync metadata */
5358 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5359 }
5360 
5361 /* END spdk_blob_close */
5362 
5363 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5364 {
5365 	return spdk_get_io_channel(bs);
5366 }
5367 
5368 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5369 {
5370 	spdk_put_io_channel(channel);
5371 }
5372 
5373 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5374 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5375 {
5376 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5377 				     SPDK_BLOB_UNMAP);
5378 }
5379 
5380 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5381 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5382 {
5383 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5384 				     SPDK_BLOB_WRITE_ZEROES);
5385 }
5386 
5387 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5388 			void *payload, uint64_t offset, uint64_t length,
5389 			spdk_blob_op_complete cb_fn, void *cb_arg)
5390 {
5391 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5392 				     SPDK_BLOB_WRITE);
5393 }
5394 
5395 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5396 		       void *payload, uint64_t offset, uint64_t length,
5397 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5398 {
5399 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5400 				     SPDK_BLOB_READ);
5401 }
5402 
5403 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5404 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5405 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5406 {
5407 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5408 }
5409 
5410 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5411 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5412 			spdk_blob_op_complete cb_fn, void *cb_arg)
5413 {
5414 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5415 }
5416 
5417 struct spdk_bs_iter_ctx {
5418 	int64_t page_num;
5419 	struct spdk_blob_store *bs;
5420 
5421 	spdk_blob_op_with_handle_complete cb_fn;
5422 	void *cb_arg;
5423 };
5424 
5425 static void
5426 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5427 {
5428 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5429 	struct spdk_blob_store *bs = ctx->bs;
5430 	spdk_blob_id id;
5431 
5432 	if (bserrno == 0) {
5433 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5434 		free(ctx);
5435 		return;
5436 	}
5437 
5438 	ctx->page_num++;
5439 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5440 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5441 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5442 		free(ctx);
5443 		return;
5444 	}
5445 
5446 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5447 
5448 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5449 }
5450 
5451 void
5452 spdk_bs_iter_first(struct spdk_blob_store *bs,
5453 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5454 {
5455 	struct spdk_bs_iter_ctx *ctx;
5456 
5457 	ctx = calloc(1, sizeof(*ctx));
5458 	if (!ctx) {
5459 		cb_fn(cb_arg, NULL, -ENOMEM);
5460 		return;
5461 	}
5462 
5463 	ctx->page_num = -1;
5464 	ctx->bs = bs;
5465 	ctx->cb_fn = cb_fn;
5466 	ctx->cb_arg = cb_arg;
5467 
5468 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5469 }
5470 
5471 static void
5472 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5473 {
5474 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5475 
5476 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5477 }
5478 
5479 void
5480 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5481 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5482 {
5483 	struct spdk_bs_iter_ctx *ctx;
5484 
5485 	assert(blob != NULL);
5486 
5487 	ctx = calloc(1, sizeof(*ctx));
5488 	if (!ctx) {
5489 		cb_fn(cb_arg, NULL, -ENOMEM);
5490 		return;
5491 	}
5492 
5493 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5494 	ctx->bs = bs;
5495 	ctx->cb_fn = cb_fn;
5496 	ctx->cb_arg = cb_arg;
5497 
5498 	/* Close the existing blob */
5499 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5500 }
5501 
5502 static int
5503 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5504 		     uint16_t value_len, bool internal)
5505 {
5506 	struct spdk_xattr_tailq *xattrs;
5507 	struct spdk_xattr	*xattr;
5508 
5509 	_spdk_blob_verify_md_op(blob);
5510 
5511 	if (blob->md_ro) {
5512 		return -EPERM;
5513 	}
5514 
5515 	if (internal) {
5516 		xattrs = &blob->xattrs_internal;
5517 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5518 	} else {
5519 		xattrs = &blob->xattrs;
5520 	}
5521 
5522 	TAILQ_FOREACH(xattr, xattrs, link) {
5523 		if (!strcmp(name, xattr->name)) {
5524 			free(xattr->value);
5525 			xattr->value_len = value_len;
5526 			xattr->value = malloc(value_len);
5527 			memcpy(xattr->value, value, value_len);
5528 
5529 			blob->state = SPDK_BLOB_STATE_DIRTY;
5530 
5531 			return 0;
5532 		}
5533 	}
5534 
5535 	xattr = calloc(1, sizeof(*xattr));
5536 	if (!xattr) {
5537 		return -ENOMEM;
5538 	}
5539 	xattr->name = strdup(name);
5540 	xattr->value_len = value_len;
5541 	xattr->value = malloc(value_len);
5542 	memcpy(xattr->value, value, value_len);
5543 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5544 
5545 	blob->state = SPDK_BLOB_STATE_DIRTY;
5546 
5547 	return 0;
5548 }
5549 
5550 int
5551 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5552 		    uint16_t value_len)
5553 {
5554 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5555 }
5556 
5557 static int
5558 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5559 {
5560 	struct spdk_xattr_tailq *xattrs;
5561 	struct spdk_xattr	*xattr;
5562 
5563 	_spdk_blob_verify_md_op(blob);
5564 
5565 	if (blob->md_ro) {
5566 		return -EPERM;
5567 	}
5568 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5569 
5570 	TAILQ_FOREACH(xattr, xattrs, link) {
5571 		if (!strcmp(name, xattr->name)) {
5572 			TAILQ_REMOVE(xattrs, xattr, link);
5573 			free(xattr->value);
5574 			free(xattr->name);
5575 			free(xattr);
5576 
5577 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5578 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5579 			}
5580 			blob->state = SPDK_BLOB_STATE_DIRTY;
5581 
5582 			return 0;
5583 		}
5584 	}
5585 
5586 	return -ENOENT;
5587 }
5588 
5589 int
5590 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5591 {
5592 	return _spdk_blob_remove_xattr(blob, name, false);
5593 }
5594 
5595 static int
5596 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5597 			   const void **value, size_t *value_len, bool internal)
5598 {
5599 	struct spdk_xattr	*xattr;
5600 	struct spdk_xattr_tailq *xattrs;
5601 
5602 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5603 
5604 	TAILQ_FOREACH(xattr, xattrs, link) {
5605 		if (!strcmp(name, xattr->name)) {
5606 			*value = xattr->value;
5607 			*value_len = xattr->value_len;
5608 			return 0;
5609 		}
5610 	}
5611 	return -ENOENT;
5612 }
5613 
5614 int
5615 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5616 			  const void **value, size_t *value_len)
5617 {
5618 	_spdk_blob_verify_md_op(blob);
5619 
5620 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5621 }
5622 
5623 struct spdk_xattr_names {
5624 	uint32_t	count;
5625 	const char	*names[0];
5626 };
5627 
5628 static int
5629 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5630 {
5631 	struct spdk_xattr	*xattr;
5632 	int			count = 0;
5633 
5634 	TAILQ_FOREACH(xattr, xattrs, link) {
5635 		count++;
5636 	}
5637 
5638 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5639 	if (*names == NULL) {
5640 		return -ENOMEM;
5641 	}
5642 
5643 	TAILQ_FOREACH(xattr, xattrs, link) {
5644 		(*names)->names[(*names)->count++] = xattr->name;
5645 	}
5646 
5647 	return 0;
5648 }
5649 
5650 int
5651 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5652 {
5653 	_spdk_blob_verify_md_op(blob);
5654 
5655 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5656 }
5657 
5658 uint32_t
5659 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5660 {
5661 	assert(names != NULL);
5662 
5663 	return names->count;
5664 }
5665 
5666 const char *
5667 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5668 {
5669 	if (index >= names->count) {
5670 		return NULL;
5671 	}
5672 
5673 	return names->names[index];
5674 }
5675 
5676 void
5677 spdk_xattr_names_free(struct spdk_xattr_names *names)
5678 {
5679 	free(names);
5680 }
5681 
5682 struct spdk_bs_type
5683 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5684 {
5685 	return bs->bstype;
5686 }
5687 
5688 void
5689 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5690 {
5691 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5692 }
5693 
5694 bool
5695 spdk_blob_is_read_only(struct spdk_blob *blob)
5696 {
5697 	assert(blob != NULL);
5698 	return (blob->data_ro || blob->md_ro);
5699 }
5700 
5701 bool
5702 spdk_blob_is_snapshot(struct spdk_blob *blob)
5703 {
5704 	struct spdk_blob_list *snapshot_entry;
5705 
5706 	assert(blob != NULL);
5707 
5708 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
5709 		if (snapshot_entry->id == blob->id) {
5710 			break;
5711 		}
5712 	}
5713 
5714 	if (snapshot_entry == NULL) {
5715 		return false;
5716 	}
5717 
5718 	return true;
5719 }
5720 
5721 bool
5722 spdk_blob_is_clone(struct spdk_blob *blob)
5723 {
5724 	assert(blob != NULL);
5725 
5726 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5727 		assert(spdk_blob_is_thin_provisioned(blob));
5728 		return true;
5729 	}
5730 
5731 	return false;
5732 }
5733 
5734 bool
5735 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5736 {
5737 	assert(blob != NULL);
5738 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5739 }
5740 
5741 spdk_blob_id
5742 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5743 {
5744 	struct spdk_blob_list *snapshot_entry = NULL;
5745 	struct spdk_blob_list *clone_entry = NULL;
5746 
5747 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5748 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5749 			if (clone_entry->id == blob_id) {
5750 				return snapshot_entry->id;
5751 			}
5752 		}
5753 	}
5754 
5755 	return SPDK_BLOBID_INVALID;
5756 }
5757 
5758 int
5759 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5760 		     size_t *count)
5761 {
5762 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5763 	size_t n;
5764 
5765 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5766 		if (snapshot_entry->id == blobid) {
5767 			break;
5768 		}
5769 	}
5770 	if (snapshot_entry == NULL) {
5771 		*count = 0;
5772 		return 0;
5773 	}
5774 
5775 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5776 		*count = snapshot_entry->clone_count;
5777 		return -ENOMEM;
5778 	}
5779 	*count = snapshot_entry->clone_count;
5780 
5781 	n = 0;
5782 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5783 		ids[n++] = clone_entry->id;
5784 	}
5785 
5786 	return 0;
5787 }
5788 
5789 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5790