xref: /spdk/lib/blob/blobstore.c (revision 1437b25472597b928e117988420b52ce49e2afbb)
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 #include "spdk/string.h"
45 
46 #include "spdk_internal/assert.h"
47 #include "spdk_internal/log.h"
48 
49 #include "blobstore.h"
50 
51 #define BLOB_CRC32C_INITIAL    0xffffffffUL
52 
53 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs);
54 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs);
55 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
56 static void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
57 		uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg);
58 
59 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
60 				uint16_t value_len, bool internal);
61 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
62 				      const void **value, size_t *value_len, bool internal);
63 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal);
64 
65 static void
66 _spdk_blob_verify_md_op(struct spdk_blob *blob)
67 {
68 	assert(blob != NULL);
69 	assert(spdk_get_thread() == blob->bs->md_thread);
70 	assert(blob->state != SPDK_BLOB_STATE_LOADING);
71 }
72 
73 static struct spdk_blob_list *
74 _spdk_bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid)
75 {
76 	struct spdk_blob_list *snapshot_entry = NULL;
77 
78 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
79 		if (snapshot_entry->id == blobid) {
80 			break;
81 		}
82 	}
83 
84 	return snapshot_entry;
85 }
86 
87 static void
88 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
89 {
90 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
91 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false);
92 	assert(bs->num_free_clusters > 0);
93 
94 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num);
95 
96 	spdk_bit_array_set(bs->used_clusters, cluster_num);
97 	bs->num_free_clusters--;
98 }
99 
100 static int
101 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
102 {
103 	uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
104 
105 	_spdk_blob_verify_md_op(blob);
106 
107 	if (*cluster_lba != 0) {
108 		return -EEXIST;
109 	}
110 
111 	*cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster);
112 	return 0;
113 }
114 
115 static int
116 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
117 			  uint64_t *lowest_free_cluster, bool update_map)
118 {
119 	pthread_mutex_lock(&blob->bs->used_clusters_mutex);
120 	*lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters,
121 			       *lowest_free_cluster);
122 	if (*lowest_free_cluster == UINT32_MAX) {
123 		/* No more free clusters. Cannot satisfy the request */
124 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
125 		return -ENOSPC;
126 	}
127 
128 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id);
129 	_spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster);
130 	pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
131 
132 	if (update_map) {
133 		_spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster);
134 	}
135 
136 	return 0;
137 }
138 
139 static void
140 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
141 {
142 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
143 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true);
144 	assert(bs->num_free_clusters < bs->total_clusters);
145 
146 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num);
147 
148 	pthread_mutex_lock(&bs->used_clusters_mutex);
149 	spdk_bit_array_clear(bs->used_clusters, cluster_num);
150 	bs->num_free_clusters++;
151 	pthread_mutex_unlock(&bs->used_clusters_mutex);
152 }
153 
154 static void
155 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
156 {
157 	xattrs->count = 0;
158 	xattrs->names = NULL;
159 	xattrs->ctx = NULL;
160 	xattrs->get_value = NULL;
161 }
162 
163 void
164 spdk_blob_opts_init(struct spdk_blob_opts *opts)
165 {
166 	opts->num_clusters = 0;
167 	opts->thin_provision = false;
168 	_spdk_blob_xattrs_init(&opts->xattrs);
169 }
170 
171 void
172 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts)
173 {
174 	opts->clear_method = BLOB_CLEAR_WITH_DEFAULT;
175 }
176 
177 static struct spdk_blob *
178 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
179 {
180 	struct spdk_blob *blob;
181 
182 	blob = calloc(1, sizeof(*blob));
183 	if (!blob) {
184 		return NULL;
185 	}
186 
187 	blob->id = id;
188 	blob->bs = bs;
189 
190 	blob->parent_id = SPDK_BLOBID_INVALID;
191 
192 	blob->state = SPDK_BLOB_STATE_DIRTY;
193 	blob->active.num_pages = 1;
194 	blob->active.pages = calloc(1, sizeof(*blob->active.pages));
195 	if (!blob->active.pages) {
196 		free(blob);
197 		return NULL;
198 	}
199 
200 	blob->active.pages[0] = _spdk_bs_blobid_to_page(id);
201 
202 	TAILQ_INIT(&blob->xattrs);
203 	TAILQ_INIT(&blob->xattrs_internal);
204 
205 	return blob;
206 }
207 
208 static void
209 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs)
210 {
211 	struct spdk_xattr	*xattr, *xattr_tmp;
212 
213 	TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
214 		TAILQ_REMOVE(xattrs, xattr, link);
215 		free(xattr->name);
216 		free(xattr->value);
217 		free(xattr);
218 	}
219 }
220 
221 static void
222 _spdk_blob_free(struct spdk_blob *blob)
223 {
224 	assert(blob != NULL);
225 
226 	free(blob->active.clusters);
227 	free(blob->clean.clusters);
228 	free(blob->active.pages);
229 	free(blob->clean.pages);
230 
231 	_spdk_xattrs_free(&blob->xattrs);
232 	_spdk_xattrs_free(&blob->xattrs_internal);
233 
234 	if (blob->back_bs_dev) {
235 		blob->back_bs_dev->destroy(blob->back_bs_dev);
236 	}
237 
238 	free(blob);
239 }
240 
241 struct freeze_io_ctx {
242 	struct spdk_bs_cpl cpl;
243 	struct spdk_blob *blob;
244 };
245 
246 static void
247 _spdk_blob_io_sync(struct spdk_io_channel_iter *i)
248 {
249 	spdk_for_each_channel_continue(i, 0);
250 }
251 
252 static void
253 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i)
254 {
255 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
256 	struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
257 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
258 	struct spdk_bs_request_set	*set;
259 	struct spdk_bs_user_op_args	*args;
260 	spdk_bs_user_op_t *op, *tmp;
261 
262 	TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
263 		set = (struct spdk_bs_request_set *)op;
264 		args = &set->u.user_op;
265 
266 		if (args->blob == ctx->blob) {
267 			TAILQ_REMOVE(&ch->queued_io, op, link);
268 			spdk_bs_user_op_execute(op);
269 		}
270 	}
271 
272 	spdk_for_each_channel_continue(i, 0);
273 }
274 
275 static void
276 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status)
277 {
278 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
279 
280 	ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
281 
282 	free(ctx);
283 }
284 
285 static void
286 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
287 {
288 	struct freeze_io_ctx *ctx;
289 
290 	ctx = calloc(1, sizeof(*ctx));
291 	if (!ctx) {
292 		cb_fn(cb_arg, -ENOMEM);
293 		return;
294 	}
295 
296 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
297 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
298 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
299 	ctx->blob = blob;
300 
301 	/* Freeze I/O on blob */
302 	blob->frozen_refcnt++;
303 
304 	if (blob->frozen_refcnt == 1) {
305 		spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl);
306 	} else {
307 		cb_fn(cb_arg, 0);
308 		free(ctx);
309 	}
310 }
311 
312 static void
313 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
314 {
315 	struct freeze_io_ctx *ctx;
316 
317 	ctx = calloc(1, sizeof(*ctx));
318 	if (!ctx) {
319 		cb_fn(cb_arg, -ENOMEM);
320 		return;
321 	}
322 
323 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
324 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
325 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
326 	ctx->blob = blob;
327 
328 	assert(blob->frozen_refcnt > 0);
329 
330 	blob->frozen_refcnt--;
331 
332 	if (blob->frozen_refcnt == 0) {
333 		spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl);
334 	} else {
335 		cb_fn(cb_arg, 0);
336 		free(ctx);
337 	}
338 }
339 
340 static int
341 _spdk_blob_mark_clean(struct spdk_blob *blob)
342 {
343 	uint64_t *clusters = NULL;
344 	uint32_t *pages = NULL;
345 
346 	assert(blob != NULL);
347 
348 	if (blob->active.num_clusters) {
349 		assert(blob->active.clusters);
350 		clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
351 		if (!clusters) {
352 			return -ENOMEM;
353 		}
354 		memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters));
355 	}
356 
357 	if (blob->active.num_pages) {
358 		assert(blob->active.pages);
359 		pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
360 		if (!pages) {
361 			free(clusters);
362 			return -ENOMEM;
363 		}
364 		memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
365 	}
366 
367 	free(blob->clean.clusters);
368 	free(blob->clean.pages);
369 
370 	blob->clean.num_clusters = blob->active.num_clusters;
371 	blob->clean.clusters = blob->active.clusters;
372 	blob->clean.num_pages = blob->active.num_pages;
373 	blob->clean.pages = blob->active.pages;
374 
375 	blob->active.clusters = clusters;
376 	blob->active.pages = pages;
377 
378 	/* If the metadata was dirtied again while the metadata was being written to disk,
379 	 *  we do not want to revert the DIRTY state back to CLEAN here.
380 	 */
381 	if (blob->state == SPDK_BLOB_STATE_LOADING) {
382 		blob->state = SPDK_BLOB_STATE_CLEAN;
383 	}
384 
385 	return 0;
386 }
387 
388 static int
389 _spdk_blob_deserialize_xattr(struct spdk_blob *blob,
390 			     struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
391 {
392 	struct spdk_xattr                       *xattr;
393 
394 	if (desc_xattr->length != sizeof(desc_xattr->name_length) +
395 	    sizeof(desc_xattr->value_length) +
396 	    desc_xattr->name_length + desc_xattr->value_length) {
397 		return -EINVAL;
398 	}
399 
400 	xattr = calloc(1, sizeof(*xattr));
401 	if (xattr == NULL) {
402 		return -ENOMEM;
403 	}
404 
405 	xattr->name = malloc(desc_xattr->name_length + 1);
406 	if (xattr->name == NULL) {
407 		free(xattr);
408 		return -ENOMEM;
409 	}
410 	memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
411 	xattr->name[desc_xattr->name_length] = '\0';
412 
413 	xattr->value = malloc(desc_xattr->value_length);
414 	if (xattr->value == NULL) {
415 		free(xattr->name);
416 		free(xattr);
417 		return -ENOMEM;
418 	}
419 	xattr->value_len = desc_xattr->value_length;
420 	memcpy(xattr->value,
421 	       (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
422 	       desc_xattr->value_length);
423 
424 	TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
425 
426 	return 0;
427 }
428 
429 
430 static int
431 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
432 {
433 	struct spdk_blob_md_descriptor *desc;
434 	size_t	cur_desc = 0;
435 	void *tmp;
436 
437 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
438 	while (cur_desc < sizeof(page->descriptors)) {
439 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
440 			if (desc->length == 0) {
441 				/* If padding and length are 0, this terminates the page */
442 				break;
443 			}
444 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
445 			struct spdk_blob_md_descriptor_flags	*desc_flags;
446 
447 			desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
448 
449 			if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
450 				return -EINVAL;
451 			}
452 
453 			if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
454 			    SPDK_BLOB_INVALID_FLAGS_MASK) {
455 				return -EINVAL;
456 			}
457 
458 			if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
459 			    SPDK_BLOB_DATA_RO_FLAGS_MASK) {
460 				blob->data_ro = true;
461 				blob->md_ro = true;
462 			}
463 
464 			if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
465 			    SPDK_BLOB_MD_RO_FLAGS_MASK) {
466 				blob->md_ro = true;
467 			}
468 
469 			if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
470 				blob->data_ro = true;
471 				blob->md_ro = true;
472 			}
473 
474 			blob->invalid_flags = desc_flags->invalid_flags;
475 			blob->data_ro_flags = desc_flags->data_ro_flags;
476 			blob->md_ro_flags = desc_flags->md_ro_flags;
477 
478 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
479 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
480 			unsigned int				i, j;
481 			unsigned int				cluster_count = blob->active.num_clusters;
482 
483 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
484 
485 			if (desc_extent_rle->length == 0 ||
486 			    (desc_extent_rle->length % sizeof(desc_extent_rle->extents[0]) != 0)) {
487 				return -EINVAL;
488 			}
489 
490 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
491 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
492 					if (desc_extent_rle->extents[i].cluster_idx != 0) {
493 						if (!spdk_bit_array_get(blob->bs->used_clusters,
494 									desc_extent_rle->extents[i].cluster_idx + j)) {
495 							return -EINVAL;
496 						}
497 					}
498 					cluster_count++;
499 				}
500 			}
501 
502 			if (cluster_count == 0) {
503 				return -EINVAL;
504 			}
505 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(*blob->active.clusters));
506 			if (tmp == NULL) {
507 				return -ENOMEM;
508 			}
509 			blob->active.clusters = tmp;
510 			blob->active.cluster_array_size = cluster_count;
511 
512 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
513 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
514 					if (desc_extent_rle->extents[i].cluster_idx != 0) {
515 						blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
516 								desc_extent_rle->extents[i].cluster_idx + j);
517 					} else if (spdk_blob_is_thin_provisioned(blob)) {
518 						blob->active.clusters[blob->active.num_clusters++] = 0;
519 					} else {
520 						return -EINVAL;
521 					}
522 				}
523 			}
524 
525 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
526 			int rc;
527 
528 			rc = _spdk_blob_deserialize_xattr(blob,
529 							  (struct spdk_blob_md_descriptor_xattr *) desc, false);
530 			if (rc != 0) {
531 				return rc;
532 			}
533 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
534 			int rc;
535 
536 			rc = _spdk_blob_deserialize_xattr(blob,
537 							  (struct spdk_blob_md_descriptor_xattr *) desc, true);
538 			if (rc != 0) {
539 				return rc;
540 			}
541 		} else {
542 			/* Unrecognized descriptor type.  Do not fail - just continue to the
543 			 *  next descriptor.  If this descriptor is associated with some feature
544 			 *  defined in a newer version of blobstore, that version of blobstore
545 			 *  should create and set an associated feature flag to specify if this
546 			 *  blob can be loaded or not.
547 			 */
548 		}
549 
550 		/* Advance to the next descriptor */
551 		cur_desc += sizeof(*desc) + desc->length;
552 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
553 			break;
554 		}
555 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
556 	}
557 
558 	return 0;
559 }
560 
561 static int
562 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
563 		 struct spdk_blob *blob)
564 {
565 	const struct spdk_blob_md_page *page;
566 	uint32_t i;
567 	int rc;
568 
569 	assert(page_count > 0);
570 	assert(pages[0].sequence_num == 0);
571 	assert(blob != NULL);
572 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
573 	assert(blob->active.clusters == NULL);
574 
575 	/* The blobid provided doesn't match what's in the MD, this can
576 	 * happen for example if a bogus blobid is passed in through open.
577 	 */
578 	if (blob->id != pages[0].id) {
579 		SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n",
580 			    blob->id, pages[0].id);
581 		return -ENOENT;
582 	}
583 
584 	for (i = 0; i < page_count; i++) {
585 		page = &pages[i];
586 
587 		assert(page->id == blob->id);
588 		assert(page->sequence_num == i);
589 
590 		rc = _spdk_blob_parse_page(page, blob);
591 		if (rc != 0) {
592 			return rc;
593 		}
594 	}
595 
596 	return 0;
597 }
598 
599 static int
600 _spdk_blob_serialize_add_page(const struct spdk_blob *blob,
601 			      struct spdk_blob_md_page **pages,
602 			      uint32_t *page_count,
603 			      struct spdk_blob_md_page **last_page)
604 {
605 	struct spdk_blob_md_page *page;
606 
607 	assert(pages != NULL);
608 	assert(page_count != NULL);
609 
610 	if (*page_count == 0) {
611 		assert(*pages == NULL);
612 		*page_count = 1;
613 		*pages = spdk_malloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
614 				     NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
615 	} else {
616 		assert(*pages != NULL);
617 		(*page_count)++;
618 		*pages = spdk_realloc(*pages,
619 				      SPDK_BS_PAGE_SIZE * (*page_count),
620 				      SPDK_BS_PAGE_SIZE);
621 	}
622 
623 	if (*pages == NULL) {
624 		*page_count = 0;
625 		*last_page = NULL;
626 		return -ENOMEM;
627 	}
628 
629 	page = &(*pages)[*page_count - 1];
630 	memset(page, 0, sizeof(*page));
631 	page->id = blob->id;
632 	page->sequence_num = *page_count - 1;
633 	page->next = SPDK_INVALID_MD_PAGE;
634 	*last_page = page;
635 
636 	return 0;
637 }
638 
639 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
640  * Update required_sz on both success and failure.
641  *
642  */
643 static int
644 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
645 			   uint8_t *buf, size_t buf_sz,
646 			   size_t *required_sz, bool internal)
647 {
648 	struct spdk_blob_md_descriptor_xattr	*desc;
649 
650 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
651 		       strlen(xattr->name) +
652 		       xattr->value_len;
653 
654 	if (buf_sz < *required_sz) {
655 		return -1;
656 	}
657 
658 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
659 
660 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
661 	desc->length = sizeof(desc->name_length) +
662 		       sizeof(desc->value_length) +
663 		       strlen(xattr->name) +
664 		       xattr->value_len;
665 	desc->name_length = strlen(xattr->name);
666 	desc->value_length = xattr->value_len;
667 
668 	memcpy(desc->name, xattr->name, desc->name_length);
669 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
670 	       xattr->value,
671 	       desc->value_length);
672 
673 	return 0;
674 }
675 
676 static void
677 _spdk_blob_serialize_extent_rle(const struct spdk_blob *blob,
678 				uint64_t start_cluster, uint64_t *next_cluster,
679 				uint8_t **buf, size_t *buf_sz)
680 {
681 	struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle;
682 	size_t cur_sz;
683 	uint64_t i, extent_idx;
684 	uint64_t lba, lba_per_cluster, lba_count;
685 
686 	/* The buffer must have room for at least one extent */
687 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc_extent_rle->extents[0]);
688 	if (*buf_sz < cur_sz) {
689 		*next_cluster = start_cluster;
690 		return;
691 	}
692 
693 	desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)*buf;
694 	desc_extent_rle->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE;
695 
696 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
697 
698 	lba = blob->active.clusters[start_cluster];
699 	lba_count = lba_per_cluster;
700 	extent_idx = 0;
701 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
702 		if ((lba + lba_count) == blob->active.clusters[i] && lba != 0) {
703 			/* Run-length encode sequential non-zero LBA */
704 			lba_count += lba_per_cluster;
705 			continue;
706 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
707 			/* Run-length encode unallocated clusters */
708 			lba_count += lba_per_cluster;
709 			continue;
710 		}
711 		desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
712 		desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
713 		extent_idx++;
714 
715 		cur_sz += sizeof(desc_extent_rle->extents[extent_idx]);
716 
717 		if (*buf_sz < cur_sz) {
718 			/* If we ran out of buffer space, return */
719 			*next_cluster = i;
720 			goto finish;
721 		}
722 
723 		lba = blob->active.clusters[i];
724 		lba_count = lba_per_cluster;
725 	}
726 
727 	desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
728 	desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
729 	extent_idx++;
730 
731 	*next_cluster = blob->active.num_clusters;
732 
733 finish:
734 	desc_extent_rle->length = sizeof(desc_extent_rle->extents[0]) * extent_idx;
735 	*buf_sz -= sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
736 	*buf += sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
737 
738 	return;
739 }
740 
741 static int
742 _spdk_blob_serialize_extents_rle(const struct spdk_blob *blob,
743 				 struct spdk_blob_md_page **pages,
744 				 struct spdk_blob_md_page *cur_page,
745 				 uint32_t *page_count, uint8_t **buf,
746 				 size_t *remaining_sz)
747 {
748 	uint64_t				last_cluster;
749 	int					rc;
750 
751 	last_cluster = 0;
752 	while (last_cluster < blob->active.num_clusters) {
753 		_spdk_blob_serialize_extent_rle(blob, last_cluster, &last_cluster, buf, remaining_sz);
754 
755 		if (last_cluster == blob->active.num_clusters) {
756 			break;
757 		}
758 
759 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
760 		if (rc < 0) {
761 			return rc;
762 		}
763 
764 		*buf = (uint8_t *)cur_page->descriptors;
765 		*remaining_sz = sizeof(cur_page->descriptors);
766 	}
767 
768 	return 0;
769 }
770 
771 static void
772 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
773 			   uint8_t *buf, size_t *buf_sz)
774 {
775 	struct spdk_blob_md_descriptor_flags *desc;
776 
777 	/*
778 	 * Flags get serialized first, so we should always have room for the flags
779 	 *  descriptor.
780 	 */
781 	assert(*buf_sz >= sizeof(*desc));
782 
783 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
784 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
785 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
786 	desc->invalid_flags = blob->invalid_flags;
787 	desc->data_ro_flags = blob->data_ro_flags;
788 	desc->md_ro_flags = blob->md_ro_flags;
789 
790 	*buf_sz -= sizeof(*desc);
791 }
792 
793 static int
794 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
795 			    const struct spdk_xattr_tailq *xattrs, bool internal,
796 			    struct spdk_blob_md_page **pages,
797 			    struct spdk_blob_md_page *cur_page,
798 			    uint32_t *page_count, uint8_t **buf,
799 			    size_t *remaining_sz)
800 {
801 	const struct spdk_xattr	*xattr;
802 	int	rc;
803 
804 	TAILQ_FOREACH(xattr, xattrs, link) {
805 		size_t required_sz = 0;
806 
807 		rc = _spdk_blob_serialize_xattr(xattr,
808 						*buf, *remaining_sz,
809 						&required_sz, internal);
810 		if (rc < 0) {
811 			/* Need to add a new page to the chain */
812 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
813 							   &cur_page);
814 			if (rc < 0) {
815 				spdk_free(*pages);
816 				*pages = NULL;
817 				*page_count = 0;
818 				return rc;
819 			}
820 
821 			*buf = (uint8_t *)cur_page->descriptors;
822 			*remaining_sz = sizeof(cur_page->descriptors);
823 
824 			/* Try again */
825 			required_sz = 0;
826 			rc = _spdk_blob_serialize_xattr(xattr,
827 							*buf, *remaining_sz,
828 							&required_sz, internal);
829 
830 			if (rc < 0) {
831 				spdk_free(*pages);
832 				*pages = NULL;
833 				*page_count = 0;
834 				return rc;
835 			}
836 		}
837 
838 		*remaining_sz -= required_sz;
839 		*buf += required_sz;
840 	}
841 
842 	return 0;
843 }
844 
845 static int
846 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
847 		     uint32_t *page_count)
848 {
849 	struct spdk_blob_md_page		*cur_page;
850 	int					rc;
851 	uint8_t					*buf;
852 	size_t					remaining_sz;
853 
854 	assert(pages != NULL);
855 	assert(page_count != NULL);
856 	assert(blob != NULL);
857 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
858 
859 	*pages = NULL;
860 	*page_count = 0;
861 
862 	/* A blob always has at least 1 page, even if it has no descriptors */
863 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
864 	if (rc < 0) {
865 		return rc;
866 	}
867 
868 	buf = (uint8_t *)cur_page->descriptors;
869 	remaining_sz = sizeof(cur_page->descriptors);
870 
871 	/* Serialize flags */
872 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
873 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
874 
875 	/* Serialize xattrs */
876 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
877 					 pages, cur_page, page_count, &buf, &remaining_sz);
878 	if (rc < 0) {
879 		return rc;
880 	}
881 
882 	/* Serialize internal xattrs */
883 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
884 					 pages, cur_page, page_count, &buf, &remaining_sz);
885 	if (rc < 0) {
886 		return rc;
887 	}
888 
889 	/* Serialize extents */
890 	rc = _spdk_blob_serialize_extents_rle(blob, pages, cur_page, page_count, &buf, &remaining_sz);
891 
892 	return rc;
893 }
894 
895 struct spdk_blob_load_ctx {
896 	struct spdk_blob		*blob;
897 
898 	struct spdk_blob_md_page	*pages;
899 	uint32_t			num_pages;
900 	spdk_bs_sequence_t	        *seq;
901 
902 	spdk_bs_sequence_cpl		cb_fn;
903 	void				*cb_arg;
904 };
905 
906 static uint32_t
907 _spdk_blob_md_page_calc_crc(void *page)
908 {
909 	uint32_t		crc;
910 
911 	crc = BLOB_CRC32C_INITIAL;
912 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
913 	crc ^= BLOB_CRC32C_INITIAL;
914 
915 	return crc;
916 
917 }
918 
919 static void
920 _spdk_blob_load_final(void *cb_arg, int bserrno)
921 {
922 	struct spdk_blob_load_ctx	*ctx = cb_arg;
923 	struct spdk_blob		*blob = ctx->blob;
924 
925 	if (bserrno == 0) {
926 		_spdk_blob_mark_clean(blob);
927 	}
928 
929 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
930 
931 	/* Free the memory */
932 	spdk_free(ctx->pages);
933 	free(ctx);
934 }
935 
936 static void
937 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
938 {
939 	struct spdk_blob_load_ctx	*ctx = cb_arg;
940 	struct spdk_blob		*blob = ctx->blob;
941 
942 	if (bserrno == 0) {
943 		blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
944 		if (blob->back_bs_dev == NULL) {
945 			bserrno = -ENOMEM;
946 		}
947 	}
948 	if (bserrno != 0) {
949 		SPDK_ERRLOG("Snapshot fail\n");
950 	}
951 
952 	_spdk_blob_load_final(ctx, bserrno);
953 }
954 
955 static void
956 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
957 {
958 	struct spdk_blob_load_ctx	*ctx = cb_arg;
959 	struct spdk_blob		*blob = ctx->blob;
960 	struct spdk_blob_md_page	*page;
961 	const void			*value;
962 	size_t				len;
963 	int				rc;
964 	uint32_t			crc;
965 
966 	if (bserrno) {
967 		SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno);
968 		_spdk_blob_load_final(ctx, bserrno);
969 		return;
970 	}
971 
972 	page = &ctx->pages[ctx->num_pages - 1];
973 	crc = _spdk_blob_md_page_calc_crc(page);
974 	if (crc != page->crc) {
975 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
976 		_spdk_blob_load_final(ctx, -EINVAL);
977 		return;
978 	}
979 
980 	if (page->next != SPDK_INVALID_MD_PAGE) {
981 		uint32_t next_page = page->next;
982 		uint64_t next_lba = _spdk_bs_md_page_to_lba(blob->bs, next_page);
983 
984 		/* Read the next page */
985 		ctx->num_pages++;
986 		ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
987 					  sizeof(*page));
988 		if (ctx->pages == NULL) {
989 			_spdk_blob_load_final(ctx, -ENOMEM);
990 			return;
991 		}
992 
993 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
994 					  next_lba,
995 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
996 					  _spdk_blob_load_cpl, ctx);
997 		return;
998 	}
999 
1000 	/* Parse the pages */
1001 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
1002 	if (rc) {
1003 		_spdk_blob_load_final(ctx, rc);
1004 		return;
1005 	}
1006 
1007 	if (spdk_blob_is_thin_provisioned(blob)) {
1008 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
1009 		if (rc == 0) {
1010 			if (len != sizeof(spdk_blob_id)) {
1011 				_spdk_blob_load_final(ctx, -EINVAL);
1012 				return;
1013 			}
1014 			/* open snapshot blob and continue in the callback function */
1015 			blob->parent_id = *(spdk_blob_id *)value;
1016 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1017 					  _spdk_blob_load_snapshot_cpl, ctx);
1018 			return;
1019 		} else {
1020 			/* add zeroes_dev for thin provisioned blob */
1021 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1022 		}
1023 	} else {
1024 		/* standard blob */
1025 		blob->back_bs_dev = NULL;
1026 	}
1027 	_spdk_blob_load_final(ctx, bserrno);
1028 }
1029 
1030 /* Load a blob from disk given a blobid */
1031 static void
1032 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1033 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1034 {
1035 	struct spdk_blob_load_ctx *ctx;
1036 	struct spdk_blob_store *bs;
1037 	uint32_t page_num;
1038 	uint64_t lba;
1039 
1040 	_spdk_blob_verify_md_op(blob);
1041 
1042 	bs = blob->bs;
1043 
1044 	ctx = calloc(1, sizeof(*ctx));
1045 	if (!ctx) {
1046 		cb_fn(seq, cb_arg, -ENOMEM);
1047 		return;
1048 	}
1049 
1050 	ctx->blob = blob;
1051 	ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE);
1052 	if (!ctx->pages) {
1053 		free(ctx);
1054 		cb_fn(seq, cb_arg, -ENOMEM);
1055 		return;
1056 	}
1057 	ctx->num_pages = 1;
1058 	ctx->cb_fn = cb_fn;
1059 	ctx->cb_arg = cb_arg;
1060 	ctx->seq = seq;
1061 
1062 	page_num = _spdk_bs_blobid_to_page(blob->id);
1063 	lba = _spdk_bs_md_page_to_lba(blob->bs, page_num);
1064 
1065 	blob->state = SPDK_BLOB_STATE_LOADING;
1066 
1067 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1068 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1069 				  _spdk_blob_load_cpl, ctx);
1070 }
1071 
1072 struct spdk_blob_persist_ctx {
1073 	struct spdk_blob		*blob;
1074 
1075 	struct spdk_bs_super_block	*super;
1076 
1077 	struct spdk_blob_md_page	*pages;
1078 
1079 	spdk_bs_sequence_t		*seq;
1080 	spdk_bs_sequence_cpl		cb_fn;
1081 	void				*cb_arg;
1082 };
1083 
1084 static void
1085 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1086 			uint32_t lba_count)
1087 {
1088 	switch (ctx->blob->clear_method) {
1089 	case BLOB_CLEAR_WITH_DEFAULT:
1090 	case BLOB_CLEAR_WITH_UNMAP:
1091 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1092 		break;
1093 	case BLOB_CLEAR_WITH_WRITE_ZEROES:
1094 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1095 		break;
1096 	case BLOB_CLEAR_WITH_NONE:
1097 	default:
1098 		break;
1099 	}
1100 }
1101 
1102 static void
1103 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1104 {
1105 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1106 	struct spdk_blob		*blob = ctx->blob;
1107 
1108 	if (bserrno == 0) {
1109 		_spdk_blob_mark_clean(blob);
1110 	}
1111 
1112 	/* Call user callback */
1113 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1114 
1115 	/* Free the memory */
1116 	spdk_free(ctx->pages);
1117 	free(ctx);
1118 }
1119 
1120 static void
1121 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1122 {
1123 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1124 	struct spdk_blob		*blob = ctx->blob;
1125 	struct spdk_blob_store		*bs = blob->bs;
1126 	size_t				i;
1127 
1128 	/* Release all clusters that were truncated */
1129 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1130 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1131 
1132 		/* Nothing to release if it was not allocated */
1133 		if (blob->active.clusters[i] != 0) {
1134 			_spdk_bs_release_cluster(bs, cluster_num);
1135 		}
1136 	}
1137 
1138 	if (blob->active.num_clusters == 0) {
1139 		free(blob->active.clusters);
1140 		blob->active.clusters = NULL;
1141 		blob->active.cluster_array_size = 0;
1142 	} else if (blob->active.num_clusters != blob->active.cluster_array_size) {
1143 #ifndef __clang_analyzer__
1144 		void *tmp;
1145 
1146 		/* scan-build really can't figure reallocs, workaround it */
1147 		tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * blob->active.num_clusters);
1148 		assert(tmp != NULL);
1149 		blob->active.clusters = tmp;
1150 #endif
1151 		blob->active.cluster_array_size = blob->active.num_clusters;
1152 	}
1153 
1154 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1155 }
1156 
1157 static void
1158 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1159 {
1160 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1161 	struct spdk_blob		*blob = ctx->blob;
1162 	struct spdk_blob_store		*bs = blob->bs;
1163 	spdk_bs_batch_t			*batch;
1164 	size_t				i;
1165 	uint64_t			lba;
1166 	uint32_t			lba_count;
1167 
1168 	/* Clusters don't move around in blobs. The list shrinks or grows
1169 	 * at the end, but no changes ever occur in the middle of the list.
1170 	 */
1171 
1172 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx);
1173 
1174 	/* Clear all clusters that were truncated */
1175 	lba = 0;
1176 	lba_count = 0;
1177 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1178 		uint64_t next_lba = blob->active.clusters[i];
1179 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1180 
1181 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1182 			/* This cluster is contiguous with the previous one. */
1183 			lba_count += next_lba_count;
1184 			continue;
1185 		}
1186 
1187 		/* This cluster is not contiguous with the previous one. */
1188 
1189 		/* If a run of LBAs previously existing, clear them now */
1190 		if (lba_count > 0) {
1191 			spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1192 		}
1193 
1194 		/* Start building the next batch */
1195 		lba = next_lba;
1196 		if (next_lba > 0) {
1197 			lba_count = next_lba_count;
1198 		} else {
1199 			lba_count = 0;
1200 		}
1201 	}
1202 
1203 	/* If we ended with a contiguous set of LBAs, clear them now */
1204 	if (lba_count > 0) {
1205 		spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1206 	}
1207 
1208 	spdk_bs_batch_close(batch);
1209 }
1210 
1211 static void
1212 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1213 {
1214 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1215 	struct spdk_blob		*blob = ctx->blob;
1216 	struct spdk_blob_store		*bs = blob->bs;
1217 	size_t				i;
1218 
1219 	/* This loop starts at 1 because the first page is special and handled
1220 	 * below. The pages (except the first) are never written in place,
1221 	 * so any pages in the clean list must be zeroed.
1222 	 */
1223 	for (i = 1; i < blob->clean.num_pages; i++) {
1224 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1225 	}
1226 
1227 	if (blob->active.num_pages == 0) {
1228 		uint32_t page_num;
1229 
1230 		page_num = _spdk_bs_blobid_to_page(blob->id);
1231 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1232 	}
1233 
1234 	/* Move on to clearing clusters */
1235 	_spdk_blob_persist_clear_clusters(seq, ctx, 0);
1236 }
1237 
1238 static void
1239 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1240 {
1241 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1242 	struct spdk_blob		*blob = ctx->blob;
1243 	struct spdk_blob_store		*bs = blob->bs;
1244 	uint64_t			lba;
1245 	uint32_t			lba_count;
1246 	spdk_bs_batch_t			*batch;
1247 	size_t				i;
1248 
1249 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1250 
1251 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1252 
1253 	/* This loop starts at 1 because the first page is special and handled
1254 	 * below. The pages (except the first) are never written in place,
1255 	 * so any pages in the clean list must be zeroed.
1256 	 */
1257 	for (i = 1; i < blob->clean.num_pages; i++) {
1258 		lba = _spdk_bs_md_page_to_lba(bs, blob->clean.pages[i]);
1259 
1260 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1261 	}
1262 
1263 	/* The first page will only be zeroed if this is a delete. */
1264 	if (blob->active.num_pages == 0) {
1265 		uint32_t page_num;
1266 
1267 		/* The first page in the metadata goes where the blobid indicates */
1268 		page_num = _spdk_bs_blobid_to_page(blob->id);
1269 		lba = _spdk_bs_md_page_to_lba(bs, page_num);
1270 
1271 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1272 	}
1273 
1274 	spdk_bs_batch_close(batch);
1275 }
1276 
1277 static void
1278 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1279 {
1280 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1281 	struct spdk_blob		*blob = ctx->blob;
1282 	struct spdk_blob_store		*bs = blob->bs;
1283 	uint64_t			lba;
1284 	uint32_t			lba_count;
1285 	struct spdk_blob_md_page	*page;
1286 
1287 	if (blob->active.num_pages == 0) {
1288 		/* Move on to the next step */
1289 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1290 		return;
1291 	}
1292 
1293 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1294 
1295 	page = &ctx->pages[0];
1296 	/* The first page in the metadata goes where the blobid indicates */
1297 	lba = _spdk_bs_md_page_to_lba(bs, _spdk_bs_blobid_to_page(blob->id));
1298 
1299 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1300 				   _spdk_blob_persist_zero_pages, ctx);
1301 }
1302 
1303 static void
1304 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1305 {
1306 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1307 	struct spdk_blob		*blob = ctx->blob;
1308 	struct spdk_blob_store		*bs = blob->bs;
1309 	uint64_t			lba;
1310 	uint32_t			lba_count;
1311 	struct spdk_blob_md_page	*page;
1312 	spdk_bs_batch_t			*batch;
1313 	size_t				i;
1314 
1315 	/* Clusters don't move around in blobs. The list shrinks or grows
1316 	 * at the end, but no changes ever occur in the middle of the list.
1317 	 */
1318 
1319 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1320 
1321 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1322 
1323 	/* This starts at 1. The root page is not written until
1324 	 * all of the others are finished
1325 	 */
1326 	for (i = 1; i < blob->active.num_pages; i++) {
1327 		page = &ctx->pages[i];
1328 		assert(page->sequence_num == i);
1329 
1330 		lba = _spdk_bs_md_page_to_lba(bs, blob->active.pages[i]);
1331 
1332 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1333 	}
1334 
1335 	spdk_bs_batch_close(batch);
1336 }
1337 
1338 static int
1339 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1340 {
1341 	uint64_t	i;
1342 	uint64_t	*tmp;
1343 	uint64_t	lfc; /* lowest free cluster */
1344 	uint64_t	num_clusters;
1345 	struct spdk_blob_store *bs;
1346 
1347 	bs = blob->bs;
1348 
1349 	_spdk_blob_verify_md_op(blob);
1350 
1351 	if (blob->active.num_clusters == sz) {
1352 		return 0;
1353 	}
1354 
1355 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1356 		/* If this blob was resized to be larger, then smaller, then
1357 		 * larger without syncing, then the cluster array already
1358 		 * contains spare assigned clusters we can use.
1359 		 */
1360 		num_clusters = spdk_min(blob->active.cluster_array_size,
1361 					sz);
1362 	} else {
1363 		num_clusters = blob->active.num_clusters;
1364 	}
1365 
1366 	/* Do two passes - one to verify that we can obtain enough clusters
1367 	 * and another to actually claim them.
1368 	 */
1369 
1370 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1371 		lfc = 0;
1372 		for (i = num_clusters; i < sz; i++) {
1373 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1374 			if (lfc == UINT32_MAX) {
1375 				/* No more free clusters. Cannot satisfy the request */
1376 				return -ENOSPC;
1377 			}
1378 			lfc++;
1379 		}
1380 	}
1381 
1382 	if (sz > num_clusters) {
1383 		/* Expand the cluster array if necessary.
1384 		 * We only shrink the array when persisting.
1385 		 */
1386 		tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * sz);
1387 		if (sz > 0 && tmp == NULL) {
1388 			return -ENOMEM;
1389 		}
1390 		memset(tmp + blob->active.cluster_array_size, 0,
1391 		       sizeof(*blob->active.clusters) * (sz - blob->active.cluster_array_size));
1392 		blob->active.clusters = tmp;
1393 		blob->active.cluster_array_size = sz;
1394 	}
1395 
1396 	blob->state = SPDK_BLOB_STATE_DIRTY;
1397 
1398 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1399 		lfc = 0;
1400 		for (i = num_clusters; i < sz; i++) {
1401 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1402 			lfc++;
1403 		}
1404 	}
1405 
1406 	blob->active.num_clusters = sz;
1407 
1408 	return 0;
1409 }
1410 
1411 static void
1412 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1413 {
1414 	spdk_bs_sequence_t *seq = ctx->seq;
1415 	struct spdk_blob *blob = ctx->blob;
1416 	struct spdk_blob_store *bs = blob->bs;
1417 	uint64_t i;
1418 	uint32_t page_num;
1419 	void *tmp;
1420 	int rc;
1421 
1422 	if (blob->active.num_pages == 0) {
1423 		/* This is the signal that the blob should be deleted.
1424 		 * Immediately jump to the clean up routine. */
1425 		assert(blob->clean.num_pages > 0);
1426 		blob->state = SPDK_BLOB_STATE_CLEAN;
1427 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1428 		return;
1429 
1430 	}
1431 
1432 	/* Generate the new metadata */
1433 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1434 	if (rc < 0) {
1435 		_spdk_blob_persist_complete(seq, ctx, rc);
1436 		return;
1437 	}
1438 
1439 	assert(blob->active.num_pages >= 1);
1440 
1441 	/* Resize the cache of page indices */
1442 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1443 	if (!tmp) {
1444 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1445 		return;
1446 	}
1447 	blob->active.pages = tmp;
1448 
1449 	/* Assign this metadata to pages. This requires two passes -
1450 	 * one to verify that there are enough pages and a second
1451 	 * to actually claim them. */
1452 	page_num = 0;
1453 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1454 	for (i = 1; i < blob->active.num_pages; i++) {
1455 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1456 		if (page_num == UINT32_MAX) {
1457 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1458 			return;
1459 		}
1460 		page_num++;
1461 	}
1462 
1463 	page_num = 0;
1464 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1465 	for (i = 1; i < blob->active.num_pages; i++) {
1466 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1467 		ctx->pages[i - 1].next = page_num;
1468 		/* Now that previous metadata page is complete, calculate the crc for it. */
1469 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1470 		blob->active.pages[i] = page_num;
1471 		spdk_bit_array_set(bs->used_md_pages, page_num);
1472 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1473 		page_num++;
1474 	}
1475 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1476 	/* Start writing the metadata from last page to first */
1477 	blob->state = SPDK_BLOB_STATE_CLEAN;
1478 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1479 }
1480 
1481 static void
1482 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1483 {
1484 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1485 
1486 	ctx->blob->bs->clean = 0;
1487 
1488 	spdk_free(ctx->super);
1489 
1490 	_spdk_blob_persist_start(ctx);
1491 }
1492 
1493 static void
1494 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1495 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1496 
1497 
1498 static void
1499 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1500 {
1501 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1502 
1503 	ctx->super->clean = 0;
1504 	if (ctx->super->size == 0) {
1505 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1506 	}
1507 
1508 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1509 }
1510 
1511 
1512 /* Write a blob to disk */
1513 static void
1514 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1515 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1516 {
1517 	struct spdk_blob_persist_ctx *ctx;
1518 
1519 	_spdk_blob_verify_md_op(blob);
1520 
1521 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1522 		cb_fn(seq, cb_arg, 0);
1523 		return;
1524 	}
1525 
1526 	ctx = calloc(1, sizeof(*ctx));
1527 	if (!ctx) {
1528 		cb_fn(seq, cb_arg, -ENOMEM);
1529 		return;
1530 	}
1531 	ctx->blob = blob;
1532 	ctx->seq = seq;
1533 	ctx->cb_fn = cb_fn;
1534 	ctx->cb_arg = cb_arg;
1535 
1536 	if (blob->bs->clean) {
1537 		ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
1538 					  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1539 		if (!ctx->super) {
1540 			cb_fn(seq, cb_arg, -ENOMEM);
1541 			free(ctx);
1542 			return;
1543 		}
1544 
1545 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1546 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1547 					  _spdk_blob_persist_dirty, ctx);
1548 	} else {
1549 		_spdk_blob_persist_start(ctx);
1550 	}
1551 }
1552 
1553 struct spdk_blob_copy_cluster_ctx {
1554 	struct spdk_blob *blob;
1555 	uint8_t *buf;
1556 	uint64_t page;
1557 	uint64_t new_cluster;
1558 	spdk_bs_sequence_t *seq;
1559 };
1560 
1561 static void
1562 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1563 {
1564 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1565 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1566 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1567 	spdk_bs_user_op_t *op;
1568 
1569 	TAILQ_INIT(&requests);
1570 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1571 
1572 	while (!TAILQ_EMPTY(&requests)) {
1573 		op = TAILQ_FIRST(&requests);
1574 		TAILQ_REMOVE(&requests, op, link);
1575 		if (bserrno == 0) {
1576 			spdk_bs_user_op_execute(op);
1577 		} else {
1578 			spdk_bs_user_op_abort(op);
1579 		}
1580 	}
1581 
1582 	spdk_free(ctx->buf);
1583 	free(ctx);
1584 }
1585 
1586 static void
1587 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1588 {
1589 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1590 
1591 	if (bserrno) {
1592 		if (bserrno == -EEXIST) {
1593 			/* The metadata insert failed because another thread
1594 			 * allocated the cluster first. Free our cluster
1595 			 * but continue without error. */
1596 			bserrno = 0;
1597 		}
1598 		_spdk_bs_release_cluster(ctx->blob->bs, ctx->new_cluster);
1599 	}
1600 
1601 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1602 }
1603 
1604 static void
1605 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1606 {
1607 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1608 	uint32_t cluster_number;
1609 
1610 	if (bserrno) {
1611 		/* The write failed, so jump to the final completion handler */
1612 		spdk_bs_sequence_finish(seq, bserrno);
1613 		return;
1614 	}
1615 
1616 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1617 
1618 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1619 					       _spdk_blob_insert_cluster_cpl, ctx);
1620 }
1621 
1622 static void
1623 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1624 {
1625 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1626 
1627 	if (bserrno != 0) {
1628 		/* The read failed, so jump to the final completion handler */
1629 		spdk_bs_sequence_finish(seq, bserrno);
1630 		return;
1631 	}
1632 
1633 	/* Write whole cluster */
1634 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1635 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1636 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1637 				   _spdk_blob_write_copy_cpl, ctx);
1638 }
1639 
1640 static void
1641 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1642 				   struct spdk_io_channel *_ch,
1643 				   uint64_t io_unit, spdk_bs_user_op_t *op)
1644 {
1645 	struct spdk_bs_cpl cpl;
1646 	struct spdk_bs_channel *ch;
1647 	struct spdk_blob_copy_cluster_ctx *ctx;
1648 	uint32_t cluster_start_page;
1649 	uint32_t cluster_number;
1650 	int rc;
1651 
1652 	ch = spdk_io_channel_get_ctx(_ch);
1653 
1654 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1655 		/* There are already operations pending. Queue this user op
1656 		 * and return because it will be re-executed when the outstanding
1657 		 * cluster allocation completes. */
1658 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1659 		return;
1660 	}
1661 
1662 	/* Round the io_unit offset down to the first page in the cluster */
1663 	cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit);
1664 
1665 	/* Calculate which index in the metadata cluster array the corresponding
1666 	 * cluster is supposed to be at. */
1667 	cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit);
1668 
1669 	ctx = calloc(1, sizeof(*ctx));
1670 	if (!ctx) {
1671 		spdk_bs_user_op_abort(op);
1672 		return;
1673 	}
1674 
1675 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1676 
1677 	ctx->blob = blob;
1678 	ctx->page = cluster_start_page;
1679 
1680 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1681 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
1682 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1683 		if (!ctx->buf) {
1684 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1685 				    blob->bs->cluster_sz);
1686 			free(ctx);
1687 			spdk_bs_user_op_abort(op);
1688 			return;
1689 		}
1690 	}
1691 
1692 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1693 	if (rc != 0) {
1694 		spdk_free(ctx->buf);
1695 		free(ctx);
1696 		spdk_bs_user_op_abort(op);
1697 		return;
1698 	}
1699 
1700 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1701 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1702 	cpl.u.blob_basic.cb_arg = ctx;
1703 
1704 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1705 	if (!ctx->seq) {
1706 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1707 		spdk_free(ctx->buf);
1708 		free(ctx);
1709 		spdk_bs_user_op_abort(op);
1710 		return;
1711 	}
1712 
1713 	/* Queue the user op to block other incoming operations */
1714 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1715 
1716 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1717 		/* Read cluster from backing device */
1718 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1719 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1720 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1721 					     _spdk_blob_write_copy, ctx);
1722 	} else {
1723 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1724 						       _spdk_blob_insert_cluster_cpl, ctx);
1725 	}
1726 }
1727 
1728 static void
1729 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
1730 				       uint64_t *lba,	uint32_t *lba_count)
1731 {
1732 	*lba_count = length;
1733 
1734 	if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) {
1735 		assert(blob->back_bs_dev != NULL);
1736 		*lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit);
1737 		*lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count);
1738 	} else {
1739 		*lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit);
1740 	}
1741 }
1742 
1743 struct op_split_ctx {
1744 	struct spdk_blob *blob;
1745 	struct spdk_io_channel *channel;
1746 	uint64_t io_unit_offset;
1747 	uint64_t io_units_remaining;
1748 	void *curr_payload;
1749 	enum spdk_blob_op_type op_type;
1750 	spdk_bs_sequence_t *seq;
1751 };
1752 
1753 static void
1754 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1755 {
1756 	struct op_split_ctx	*ctx = cb_arg;
1757 	struct spdk_blob	*blob = ctx->blob;
1758 	struct spdk_io_channel	*ch = ctx->channel;
1759 	enum spdk_blob_op_type	op_type = ctx->op_type;
1760 	uint8_t			*buf = ctx->curr_payload;
1761 	uint64_t		offset = ctx->io_unit_offset;
1762 	uint64_t		length = ctx->io_units_remaining;
1763 	uint64_t		op_length;
1764 
1765 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
1766 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1767 		free(ctx);
1768 		return;
1769 	}
1770 
1771 	op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob,
1772 			     offset));
1773 
1774 	/* Update length and payload for next operation */
1775 	ctx->io_units_remaining -= op_length;
1776 	ctx->io_unit_offset += op_length;
1777 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1778 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
1779 	}
1780 
1781 	switch (op_type) {
1782 	case SPDK_BLOB_READ:
1783 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1784 				  _spdk_blob_request_submit_op_split_next, ctx);
1785 		break;
1786 	case SPDK_BLOB_WRITE:
1787 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1788 				   _spdk_blob_request_submit_op_split_next, ctx);
1789 		break;
1790 	case SPDK_BLOB_UNMAP:
1791 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1792 				   _spdk_blob_request_submit_op_split_next, ctx);
1793 		break;
1794 	case SPDK_BLOB_WRITE_ZEROES:
1795 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1796 					  _spdk_blob_request_submit_op_split_next, ctx);
1797 		break;
1798 	case SPDK_BLOB_READV:
1799 	case SPDK_BLOB_WRITEV:
1800 		SPDK_ERRLOG("readv/write not valid\n");
1801 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1802 		free(ctx);
1803 		break;
1804 	}
1805 }
1806 
1807 static void
1808 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1809 				   void *payload, uint64_t offset, uint64_t length,
1810 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1811 {
1812 	struct op_split_ctx *ctx;
1813 	spdk_bs_sequence_t *seq;
1814 	struct spdk_bs_cpl cpl;
1815 
1816 	assert(blob != NULL);
1817 
1818 	ctx = calloc(1, sizeof(struct op_split_ctx));
1819 	if (ctx == NULL) {
1820 		cb_fn(cb_arg, -ENOMEM);
1821 		return;
1822 	}
1823 
1824 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1825 	cpl.u.blob_basic.cb_fn = cb_fn;
1826 	cpl.u.blob_basic.cb_arg = cb_arg;
1827 
1828 	seq = spdk_bs_sequence_start(ch, &cpl);
1829 	if (!seq) {
1830 		free(ctx);
1831 		cb_fn(cb_arg, -ENOMEM);
1832 		return;
1833 	}
1834 
1835 	ctx->blob = blob;
1836 	ctx->channel = ch;
1837 	ctx->curr_payload = payload;
1838 	ctx->io_unit_offset = offset;
1839 	ctx->io_units_remaining = length;
1840 	ctx->op_type = op_type;
1841 	ctx->seq = seq;
1842 
1843 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1844 }
1845 
1846 static void
1847 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1848 				    void *payload, uint64_t offset, uint64_t length,
1849 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1850 {
1851 	struct spdk_bs_cpl cpl;
1852 	uint64_t lba;
1853 	uint32_t lba_count;
1854 
1855 	assert(blob != NULL);
1856 
1857 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1858 	cpl.u.blob_basic.cb_fn = cb_fn;
1859 	cpl.u.blob_basic.cb_arg = cb_arg;
1860 
1861 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1862 
1863 	if (blob->frozen_refcnt) {
1864 		/* This blob I/O is frozen */
1865 		spdk_bs_user_op_t *op;
1866 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1867 
1868 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1869 		if (!op) {
1870 			cb_fn(cb_arg, -ENOMEM);
1871 			return;
1872 		}
1873 
1874 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1875 
1876 		return;
1877 	}
1878 
1879 	switch (op_type) {
1880 	case SPDK_BLOB_READ: {
1881 		spdk_bs_batch_t *batch;
1882 
1883 		batch = spdk_bs_batch_open(_ch, &cpl);
1884 		if (!batch) {
1885 			cb_fn(cb_arg, -ENOMEM);
1886 			return;
1887 		}
1888 
1889 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1890 			/* Read from the blob */
1891 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1892 		} else {
1893 			/* Read from the backing block device */
1894 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1895 		}
1896 
1897 		spdk_bs_batch_close(batch);
1898 		break;
1899 	}
1900 	case SPDK_BLOB_WRITE:
1901 	case SPDK_BLOB_WRITE_ZEROES: {
1902 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1903 			/* Write to the blob */
1904 			spdk_bs_batch_t *batch;
1905 
1906 			if (lba_count == 0) {
1907 				cb_fn(cb_arg, 0);
1908 				return;
1909 			}
1910 
1911 			batch = spdk_bs_batch_open(_ch, &cpl);
1912 			if (!batch) {
1913 				cb_fn(cb_arg, -ENOMEM);
1914 				return;
1915 			}
1916 
1917 			if (op_type == SPDK_BLOB_WRITE) {
1918 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1919 			} else {
1920 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1921 			}
1922 
1923 			spdk_bs_batch_close(batch);
1924 		} else {
1925 			/* Queue this operation and allocate the cluster */
1926 			spdk_bs_user_op_t *op;
1927 
1928 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1929 			if (!op) {
1930 				cb_fn(cb_arg, -ENOMEM);
1931 				return;
1932 			}
1933 
1934 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1935 		}
1936 		break;
1937 	}
1938 	case SPDK_BLOB_UNMAP: {
1939 		spdk_bs_batch_t *batch;
1940 
1941 		batch = spdk_bs_batch_open(_ch, &cpl);
1942 		if (!batch) {
1943 			cb_fn(cb_arg, -ENOMEM);
1944 			return;
1945 		}
1946 
1947 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
1948 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1949 		}
1950 
1951 		spdk_bs_batch_close(batch);
1952 		break;
1953 	}
1954 	case SPDK_BLOB_READV:
1955 	case SPDK_BLOB_WRITEV:
1956 		SPDK_ERRLOG("readv/write not valid\n");
1957 		cb_fn(cb_arg, -EINVAL);
1958 		break;
1959 	}
1960 }
1961 
1962 static void
1963 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1964 			     void *payload, uint64_t offset, uint64_t length,
1965 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1966 {
1967 	assert(blob != NULL);
1968 
1969 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1970 		cb_fn(cb_arg, -EPERM);
1971 		return;
1972 	}
1973 
1974 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
1975 		cb_fn(cb_arg, -EINVAL);
1976 		return;
1977 	}
1978 	if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) {
1979 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1980 						    cb_fn, cb_arg, op_type);
1981 	} else {
1982 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1983 						   cb_fn, cb_arg, op_type);
1984 	}
1985 }
1986 
1987 struct rw_iov_ctx {
1988 	struct spdk_blob *blob;
1989 	struct spdk_io_channel *channel;
1990 	spdk_blob_op_complete cb_fn;
1991 	void *cb_arg;
1992 	bool read;
1993 	int iovcnt;
1994 	struct iovec *orig_iov;
1995 	uint64_t io_unit_offset;
1996 	uint64_t io_units_remaining;
1997 	uint64_t io_units_done;
1998 	struct iovec iov[0];
1999 };
2000 
2001 static void
2002 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2003 {
2004 	assert(cb_arg == NULL);
2005 	spdk_bs_sequence_finish(seq, bserrno);
2006 }
2007 
2008 static void
2009 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
2010 {
2011 	struct rw_iov_ctx *ctx = cb_arg;
2012 	struct spdk_blob *blob = ctx->blob;
2013 	struct iovec *iov, *orig_iov;
2014 	int iovcnt;
2015 	size_t orig_iovoff;
2016 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2017 	uint64_t byte_count;
2018 
2019 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2020 		ctx->cb_fn(ctx->cb_arg, bserrno);
2021 		free(ctx);
2022 		return;
2023 	}
2024 
2025 	io_unit_offset = ctx->io_unit_offset;
2026 	io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2027 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2028 	/*
2029 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2030 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2031 	 *  point to the current position in the I/O sequence.
2032 	 */
2033 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2034 	orig_iov = &ctx->orig_iov[0];
2035 	orig_iovoff = 0;
2036 	while (byte_count > 0) {
2037 		if (byte_count >= orig_iov->iov_len) {
2038 			byte_count -= orig_iov->iov_len;
2039 			orig_iov++;
2040 		} else {
2041 			orig_iovoff = byte_count;
2042 			byte_count = 0;
2043 		}
2044 	}
2045 
2046 	/*
2047 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2048 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2049 	 */
2050 	byte_count = io_units_count * blob->bs->io_unit_size;
2051 	iov = &ctx->iov[0];
2052 	iovcnt = 0;
2053 	while (byte_count > 0) {
2054 		assert(iovcnt < ctx->iovcnt);
2055 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2056 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2057 		byte_count -= iov->iov_len;
2058 		orig_iovoff = 0;
2059 		orig_iov++;
2060 		iov++;
2061 		iovcnt++;
2062 	}
2063 
2064 	ctx->io_unit_offset += io_units_count;
2065 	ctx->io_units_remaining -= io_units_count;
2066 	ctx->io_units_done += io_units_count;
2067 	iov = &ctx->iov[0];
2068 
2069 	if (ctx->read) {
2070 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2071 				   io_units_count, _spdk_rw_iov_split_next, ctx);
2072 	} else {
2073 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2074 				    io_units_count, _spdk_rw_iov_split_next, ctx);
2075 	}
2076 }
2077 
2078 static void
2079 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2080 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2081 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2082 {
2083 	struct spdk_bs_cpl	cpl;
2084 
2085 	assert(blob != NULL);
2086 
2087 	if (!read && blob->data_ro) {
2088 		cb_fn(cb_arg, -EPERM);
2089 		return;
2090 	}
2091 
2092 	if (length == 0) {
2093 		cb_fn(cb_arg, 0);
2094 		return;
2095 	}
2096 
2097 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2098 		cb_fn(cb_arg, -EINVAL);
2099 		return;
2100 	}
2101 
2102 	/*
2103 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2104 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2105 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2106 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2107 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2108 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2109 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2110 	 *
2111 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2112 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2113 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2114 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2115 	 */
2116 	if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) {
2117 		uint32_t lba_count;
2118 		uint64_t lba;
2119 
2120 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2121 		cpl.u.blob_basic.cb_fn = cb_fn;
2122 		cpl.u.blob_basic.cb_arg = cb_arg;
2123 
2124 		if (blob->frozen_refcnt) {
2125 			/* This blob I/O is frozen */
2126 			enum spdk_blob_op_type op_type;
2127 			spdk_bs_user_op_t *op;
2128 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2129 
2130 			op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV;
2131 			op = spdk_bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length);
2132 			if (!op) {
2133 				cb_fn(cb_arg, -ENOMEM);
2134 				return;
2135 			}
2136 
2137 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2138 
2139 			return;
2140 		}
2141 
2142 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2143 
2144 		if (read) {
2145 			spdk_bs_sequence_t *seq;
2146 
2147 			seq = spdk_bs_sequence_start(_channel, &cpl);
2148 			if (!seq) {
2149 				cb_fn(cb_arg, -ENOMEM);
2150 				return;
2151 			}
2152 
2153 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2154 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2155 			} else {
2156 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2157 							      _spdk_rw_iov_done, NULL);
2158 			}
2159 		} else {
2160 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2161 				spdk_bs_sequence_t *seq;
2162 
2163 				seq = spdk_bs_sequence_start(_channel, &cpl);
2164 				if (!seq) {
2165 					cb_fn(cb_arg, -ENOMEM);
2166 					return;
2167 				}
2168 
2169 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2170 			} else {
2171 				/* Queue this operation and allocate the cluster */
2172 				spdk_bs_user_op_t *op;
2173 
2174 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2175 							   length);
2176 				if (!op) {
2177 					cb_fn(cb_arg, -ENOMEM);
2178 					return;
2179 				}
2180 
2181 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2182 			}
2183 		}
2184 	} else {
2185 		struct rw_iov_ctx *ctx;
2186 
2187 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2188 		if (ctx == NULL) {
2189 			cb_fn(cb_arg, -ENOMEM);
2190 			return;
2191 		}
2192 
2193 		ctx->blob = blob;
2194 		ctx->channel = _channel;
2195 		ctx->cb_fn = cb_fn;
2196 		ctx->cb_arg = cb_arg;
2197 		ctx->read = read;
2198 		ctx->orig_iov = iov;
2199 		ctx->iovcnt = iovcnt;
2200 		ctx->io_unit_offset = offset;
2201 		ctx->io_units_remaining = length;
2202 		ctx->io_units_done = 0;
2203 
2204 		_spdk_rw_iov_split_next(ctx, 0);
2205 	}
2206 }
2207 
2208 static struct spdk_blob *
2209 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2210 {
2211 	struct spdk_blob *blob;
2212 
2213 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2214 		if (blob->id == blobid) {
2215 			return blob;
2216 		}
2217 	}
2218 
2219 	return NULL;
2220 }
2221 
2222 static void
2223 _spdk_blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
2224 		struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
2225 {
2226 	assert(blob != NULL);
2227 	*snapshot_entry = NULL;
2228 	*clone_entry = NULL;
2229 
2230 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
2231 		return;
2232 	}
2233 
2234 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
2235 		if ((*snapshot_entry)->id == blob->parent_id) {
2236 			break;
2237 		}
2238 	}
2239 
2240 	if (*snapshot_entry != NULL) {
2241 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
2242 			if ((*clone_entry)->id == blob->id) {
2243 				break;
2244 			}
2245 		}
2246 
2247 		assert(clone_entry != NULL);
2248 	}
2249 }
2250 
2251 static int
2252 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2253 {
2254 	struct spdk_blob_store		*bs = io_device;
2255 	struct spdk_bs_channel		*channel = ctx_buf;
2256 	struct spdk_bs_dev		*dev;
2257 	uint32_t			max_ops = bs->max_channel_ops;
2258 	uint32_t			i;
2259 
2260 	dev = bs->dev;
2261 
2262 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2263 	if (!channel->req_mem) {
2264 		return -1;
2265 	}
2266 
2267 	TAILQ_INIT(&channel->reqs);
2268 
2269 	for (i = 0; i < max_ops; i++) {
2270 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2271 	}
2272 
2273 	channel->bs = bs;
2274 	channel->dev = dev;
2275 	channel->dev_channel = dev->create_channel(dev);
2276 
2277 	if (!channel->dev_channel) {
2278 		SPDK_ERRLOG("Failed to create device channel.\n");
2279 		free(channel->req_mem);
2280 		return -1;
2281 	}
2282 
2283 	TAILQ_INIT(&channel->need_cluster_alloc);
2284 	TAILQ_INIT(&channel->queued_io);
2285 
2286 	return 0;
2287 }
2288 
2289 static void
2290 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2291 {
2292 	struct spdk_bs_channel *channel = ctx_buf;
2293 	spdk_bs_user_op_t *op;
2294 
2295 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2296 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2297 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2298 		spdk_bs_user_op_abort(op);
2299 	}
2300 
2301 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2302 		op = TAILQ_FIRST(&channel->queued_io);
2303 		TAILQ_REMOVE(&channel->queued_io, op, link);
2304 		spdk_bs_user_op_abort(op);
2305 	}
2306 
2307 	free(channel->req_mem);
2308 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2309 }
2310 
2311 static void
2312 _spdk_bs_dev_destroy(void *io_device)
2313 {
2314 	struct spdk_blob_store *bs = io_device;
2315 	struct spdk_blob	*blob, *blob_tmp;
2316 
2317 	bs->dev->destroy(bs->dev);
2318 
2319 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2320 		TAILQ_REMOVE(&bs->blobs, blob, link);
2321 		_spdk_blob_free(blob);
2322 	}
2323 
2324 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2325 
2326 	spdk_bit_array_free(&bs->used_blobids);
2327 	spdk_bit_array_free(&bs->used_md_pages);
2328 	spdk_bit_array_free(&bs->used_clusters);
2329 	/*
2330 	 * If this function is called for any reason except a successful unload,
2331 	 * the unload_cpl type will be NONE and this will be a nop.
2332 	 */
2333 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2334 
2335 	free(bs);
2336 }
2337 
2338 static int
2339 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2340 {
2341 	spdk_blob_id snapshot_id;
2342 	struct spdk_blob_list *snapshot_entry = NULL;
2343 	struct spdk_blob_list *clone_entry = NULL;
2344 
2345 	assert(blob != NULL);
2346 
2347 	snapshot_id = blob->parent_id;
2348 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2349 		return 0;
2350 	}
2351 
2352 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, snapshot_id);
2353 	if (snapshot_entry == NULL) {
2354 		/* Snapshot not found */
2355 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2356 		if (snapshot_entry == NULL) {
2357 			return -ENOMEM;
2358 		}
2359 		snapshot_entry->id = snapshot_id;
2360 		TAILQ_INIT(&snapshot_entry->clones);
2361 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2362 	} else {
2363 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2364 			if (clone_entry->id == blob->id) {
2365 				break;
2366 			}
2367 		}
2368 	}
2369 
2370 	if (clone_entry == NULL) {
2371 		/* Clone not found */
2372 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2373 		if (clone_entry == NULL) {
2374 			return -ENOMEM;
2375 		}
2376 		clone_entry->id = blob->id;
2377 		TAILQ_INIT(&clone_entry->clones);
2378 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2379 		snapshot_entry->clone_count++;
2380 	}
2381 
2382 	return 0;
2383 }
2384 
2385 static void
2386 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2387 {
2388 	struct spdk_blob_list *snapshot_entry = NULL;
2389 	struct spdk_blob_list *clone_entry = NULL;
2390 
2391 	_spdk_blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
2392 
2393 	if (snapshot_entry == NULL) {
2394 		return;
2395 	}
2396 
2397 	blob->parent_id = SPDK_BLOBID_INVALID;
2398 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2399 	free(clone_entry);
2400 
2401 	snapshot_entry->clone_count--;
2402 }
2403 
2404 static int
2405 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2406 {
2407 	struct spdk_blob_list *snapshot_entry;
2408 	struct spdk_blob_list *snapshot_entry_tmp;
2409 	struct spdk_blob_list *clone_entry;
2410 	struct spdk_blob_list *clone_entry_tmp;
2411 
2412 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2413 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2414 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2415 			free(clone_entry);
2416 		}
2417 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2418 		free(snapshot_entry);
2419 	}
2420 
2421 	return 0;
2422 }
2423 
2424 static void
2425 _spdk_bs_free(struct spdk_blob_store *bs)
2426 {
2427 	_spdk_bs_blob_list_free(bs);
2428 
2429 	spdk_bs_unregister_md_thread(bs);
2430 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2431 }
2432 
2433 void
2434 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2435 {
2436 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2437 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2438 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2439 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2440 	opts->clear_method = BS_CLEAR_WITH_UNMAP;
2441 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2442 	opts->iter_cb_fn = NULL;
2443 	opts->iter_cb_arg = NULL;
2444 }
2445 
2446 static int
2447 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2448 {
2449 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2450 	    opts->max_channel_ops == 0) {
2451 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2452 		return -1;
2453 	}
2454 
2455 	return 0;
2456 }
2457 
2458 static int
2459 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2460 {
2461 	struct spdk_blob_store	*bs;
2462 	uint64_t dev_size;
2463 	int rc;
2464 
2465 	dev_size = dev->blocklen * dev->blockcnt;
2466 	if (dev_size < opts->cluster_sz) {
2467 		/* Device size cannot be smaller than cluster size of blobstore */
2468 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2469 			     dev_size, opts->cluster_sz);
2470 		return -ENOSPC;
2471 	}
2472 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2473 		/* Cluster size cannot be smaller than page size */
2474 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2475 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2476 		return -EINVAL;
2477 	}
2478 	bs = calloc(1, sizeof(struct spdk_blob_store));
2479 	if (!bs) {
2480 		return -ENOMEM;
2481 	}
2482 
2483 	TAILQ_INIT(&bs->blobs);
2484 	TAILQ_INIT(&bs->snapshots);
2485 	bs->dev = dev;
2486 	bs->md_thread = spdk_get_thread();
2487 	assert(bs->md_thread != NULL);
2488 
2489 	/*
2490 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2491 	 *  even multiple of the cluster size.
2492 	 */
2493 	bs->cluster_sz = opts->cluster_sz;
2494 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2495 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2496 	bs->num_free_clusters = bs->total_clusters;
2497 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2498 	bs->io_unit_size = dev->blocklen;
2499 	if (bs->used_clusters == NULL) {
2500 		free(bs);
2501 		return -ENOMEM;
2502 	}
2503 
2504 	bs->max_channel_ops = opts->max_channel_ops;
2505 	bs->super_blob = SPDK_BLOBID_INVALID;
2506 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2507 
2508 	/* The metadata is assumed to be at least 1 page */
2509 	bs->used_md_pages = spdk_bit_array_create(1);
2510 	bs->used_blobids = spdk_bit_array_create(0);
2511 
2512 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2513 
2514 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2515 				sizeof(struct spdk_bs_channel), "blobstore");
2516 	rc = spdk_bs_register_md_thread(bs);
2517 	if (rc == -1) {
2518 		spdk_io_device_unregister(bs, NULL);
2519 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2520 		spdk_bit_array_free(&bs->used_blobids);
2521 		spdk_bit_array_free(&bs->used_md_pages);
2522 		spdk_bit_array_free(&bs->used_clusters);
2523 		free(bs);
2524 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2525 		return -ENOMEM;
2526 	}
2527 
2528 	*_bs = bs;
2529 	return 0;
2530 }
2531 
2532 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2533 
2534 struct spdk_bs_load_ctx {
2535 	struct spdk_blob_store		*bs;
2536 	struct spdk_bs_super_block	*super;
2537 
2538 	struct spdk_bs_md_mask		*mask;
2539 	bool				in_page_chain;
2540 	uint32_t			page_index;
2541 	uint32_t			cur_page;
2542 	struct spdk_blob_md_page	*page;
2543 
2544 	spdk_bs_sequence_t			*seq;
2545 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2546 	void					*iter_cb_arg;
2547 	struct spdk_blob			*blob;
2548 	spdk_blob_id				blobid;
2549 };
2550 
2551 static void
2552 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2553 {
2554 	assert(bserrno != 0);
2555 
2556 	spdk_free(ctx->super);
2557 	spdk_bs_sequence_finish(seq, bserrno);
2558 	_spdk_bs_free(ctx->bs);
2559 	free(ctx);
2560 }
2561 
2562 static void
2563 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2564 {
2565 	uint32_t i = 0;
2566 
2567 	while (true) {
2568 		i = spdk_bit_array_find_first_set(array, i);
2569 		if (i >= mask->length) {
2570 			break;
2571 		}
2572 		mask->mask[i / 8] |= 1U << (i % 8);
2573 		i++;
2574 	}
2575 }
2576 
2577 static int
2578 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2579 {
2580 	struct spdk_bit_array *array;
2581 	uint32_t i;
2582 
2583 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2584 		return -ENOMEM;
2585 	}
2586 
2587 	array = *array_ptr;
2588 	for (i = 0; i < mask->length; i++) {
2589 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2590 			spdk_bit_array_set(array, i);
2591 		}
2592 	}
2593 
2594 	return 0;
2595 }
2596 
2597 static void
2598 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2599 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2600 {
2601 	/* Update the values in the super block */
2602 	super->super_blob = bs->super_blob;
2603 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2604 	super->crc = _spdk_blob_md_page_calc_crc(super);
2605 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2606 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2607 				   cb_fn, cb_arg);
2608 }
2609 
2610 static void
2611 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2612 {
2613 	struct spdk_bs_load_ctx	*ctx = arg;
2614 	uint64_t	mask_size, lba, lba_count;
2615 
2616 	/* Write out the used clusters mask */
2617 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2618 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2619 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2620 	if (!ctx->mask) {
2621 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2622 		return;
2623 	}
2624 
2625 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2626 	ctx->mask->length = ctx->bs->total_clusters;
2627 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2628 
2629 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2630 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2631 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2632 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2633 }
2634 
2635 static void
2636 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2637 {
2638 	struct spdk_bs_load_ctx	*ctx = arg;
2639 	uint64_t	mask_size, lba, lba_count;
2640 
2641 	if (seq->bserrno) {
2642 		_spdk_bs_load_ctx_fail(seq, ctx, seq->bserrno);
2643 		return;
2644 	}
2645 
2646 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2647 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2648 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2649 	if (!ctx->mask) {
2650 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2651 		return;
2652 	}
2653 
2654 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2655 	ctx->mask->length = ctx->super->md_len;
2656 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2657 
2658 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2659 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2660 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2661 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2662 }
2663 
2664 static void
2665 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2666 {
2667 	struct spdk_bs_load_ctx	*ctx = arg;
2668 	uint64_t	mask_size, lba, lba_count;
2669 
2670 	if (ctx->super->used_blobid_mask_len == 0) {
2671 		/*
2672 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2673 		 *  written to disk.
2674 		 */
2675 		cb_fn(seq, arg, 0);
2676 		return;
2677 	}
2678 
2679 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2680 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2681 				 SPDK_MALLOC_DMA);
2682 	if (!ctx->mask) {
2683 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2684 		return;
2685 	}
2686 
2687 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2688 	ctx->mask->length = ctx->super->md_len;
2689 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2690 
2691 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2692 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2693 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2694 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2695 }
2696 
2697 static void
2698 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
2699 {
2700 	_spdk_blob_verify_md_op(blob);
2701 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
2702 	blob->state = SPDK_BLOB_STATE_DIRTY;
2703 }
2704 
2705 static void _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno);
2706 
2707 static void
2708 _spdk_bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno)
2709 {
2710 	struct spdk_bs_load_ctx *ctx = cb_arg;
2711 	spdk_blob_id id;
2712 	int64_t page_num;
2713 
2714 	/* Iterate to next blob (we can't use spdk_bs_iter_next function as our
2715 	 * last blob has been removed */
2716 	page_num = _spdk_bs_blobid_to_page(ctx->blobid);
2717 	page_num++;
2718 	page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num);
2719 	if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) {
2720 		_spdk_bs_load_iter(ctx, NULL, -ENOENT);
2721 		return;
2722 	}
2723 
2724 	id = _spdk_bs_page_to_blobid(page_num);
2725 
2726 	spdk_bs_open_blob(ctx->bs, id, _spdk_bs_load_iter, ctx);
2727 }
2728 
2729 static void
2730 _spdk_bs_delete_corrupted_close_cb(void *cb_arg, int bserrno)
2731 {
2732 	struct spdk_bs_load_ctx *ctx = cb_arg;
2733 
2734 	if (bserrno != 0) {
2735 		SPDK_ERRLOG("Failed to close corrupted blob\n");
2736 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2737 		return;
2738 	}
2739 
2740 	spdk_bs_delete_blob(ctx->bs, ctx->blobid, _spdk_bs_delete_corrupted_blob_cpl, ctx);
2741 }
2742 
2743 static void
2744 _spdk_bs_delete_corrupted_blob(void *cb_arg, int bserrno)
2745 {
2746 	struct spdk_bs_load_ctx *ctx = cb_arg;
2747 	uint64_t i;
2748 
2749 	if (bserrno != 0) {
2750 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
2751 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2752 		return;
2753 	}
2754 
2755 	/* Snapshot and clone have the same copy of cluster map at this point.
2756 	 * Let's clear cluster map for snpashot now so that it won't be cleared
2757 	 * for clone later when we remove snapshot. Also set thin provision to
2758 	 * pass data corruption check */
2759 	for (i = 0; i < ctx->blob->active.num_clusters; i++) {
2760 		ctx->blob->active.clusters[i] = 0;
2761 	}
2762 
2763 	ctx->blob->md_ro = false;
2764 
2765 	_spdk_blob_set_thin_provision(ctx->blob);
2766 
2767 	ctx->blobid = ctx->blob->id;
2768 
2769 	spdk_blob_close(ctx->blob, _spdk_bs_delete_corrupted_close_cb, ctx);
2770 }
2771 
2772 static void
2773 _spdk_bs_update_corrupted_blob(void *cb_arg, int bserrno)
2774 {
2775 	struct spdk_bs_load_ctx *ctx = cb_arg;
2776 
2777 	if (bserrno != 0) {
2778 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
2779 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2780 		return;
2781 	}
2782 
2783 	ctx->blob->md_ro = false;
2784 	_spdk_blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true);
2785 	_spdk_blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true);
2786 	spdk_blob_set_read_only(ctx->blob);
2787 
2788 	if (ctx->iter_cb_fn) {
2789 		ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0);
2790 	}
2791 	_spdk_bs_blob_list_add(ctx->blob);
2792 
2793 	spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2794 }
2795 
2796 static void
2797 _spdk_bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno)
2798 {
2799 	struct spdk_bs_load_ctx *ctx = cb_arg;
2800 
2801 	if (bserrno != 0) {
2802 		SPDK_ERRLOG("Failed to open clone of a corrupted blob\n");
2803 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
2804 		return;
2805 	}
2806 
2807 	if (blob->parent_id == ctx->blob->id) {
2808 		/* Power failure occured before updating clone (snapshot delete case)
2809 		 * or after updating clone (creating snapshot case) - keep snapshot */
2810 		spdk_blob_close(blob, _spdk_bs_update_corrupted_blob, ctx);
2811 	} else {
2812 		/* Power failure occured after updating clone (snapshot delete case)
2813 		 * or before updating clone (creating snapshot case) - remove snapshot */
2814 		spdk_blob_close(blob, _spdk_bs_delete_corrupted_blob, ctx);
2815 	}
2816 }
2817 
2818 static void
2819 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2820 {
2821 	struct spdk_bs_load_ctx *ctx = arg;
2822 	const void *value;
2823 	size_t len;
2824 	int rc = 0;
2825 
2826 	if (bserrno == 0) {
2827 		/* Examine blob if it is corrupted after power failure. Fix
2828 		 * the ones that can be fixed and remove any other corrupted
2829 		 * ones. If it is not corrupted just process it */
2830 		rc = _spdk_blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true);
2831 		if (rc != 0) {
2832 			rc = _spdk_blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true);
2833 			if (rc != 0) {
2834 				/* Not corrupted - process it and continue with iterating through blobs */
2835 				if (ctx->iter_cb_fn) {
2836 					ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2837 				}
2838 				_spdk_bs_blob_list_add(blob);
2839 				spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2840 				return;
2841 			}
2842 
2843 		}
2844 
2845 		assert(len == sizeof(spdk_blob_id));
2846 
2847 		ctx->blob = blob;
2848 
2849 		/* Open clone to check if we are able to fix this blob or should we remove it */
2850 		spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, _spdk_bs_examine_clone, ctx);
2851 		return;
2852 	} else if (bserrno == -ENOENT) {
2853 		bserrno = 0;
2854 	} else {
2855 		/*
2856 		 * This case needs to be looked at further.  Same problem
2857 		 *  exists with applications that rely on explicit blob
2858 		 *  iteration.  We should just skip the blob that failed
2859 		 *  to load and continue on to the next one.
2860 		 */
2861 		SPDK_ERRLOG("Error in iterating blobs\n");
2862 	}
2863 
2864 	ctx->iter_cb_fn = NULL;
2865 
2866 	spdk_free(ctx->super);
2867 	spdk_free(ctx->mask);
2868 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2869 	free(ctx);
2870 }
2871 
2872 static void
2873 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2874 {
2875 	ctx->seq = seq;
2876 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2877 }
2878 
2879 static void
2880 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2881 {
2882 	struct spdk_bs_load_ctx *ctx = cb_arg;
2883 	int rc;
2884 
2885 	/* The type must be correct */
2886 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2887 
2888 	/* The length of the mask (in bits) must not be greater than
2889 	 * the length of the buffer (converted to bits) */
2890 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2891 
2892 	/* The length of the mask must be exactly equal to the size
2893 	 * (in pages) of the metadata region */
2894 	assert(ctx->mask->length == ctx->super->md_len);
2895 
2896 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2897 	if (rc < 0) {
2898 		spdk_free(ctx->mask);
2899 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2900 		return;
2901 	}
2902 
2903 	_spdk_bs_load_complete(seq, ctx, bserrno);
2904 }
2905 
2906 static void
2907 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2908 {
2909 	struct spdk_bs_load_ctx *ctx = cb_arg;
2910 	uint64_t		lba, lba_count, mask_size;
2911 	int			rc;
2912 
2913 	/* The type must be correct */
2914 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2915 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2916 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2917 					     struct spdk_blob_md_page) * 8));
2918 	/* The length of the mask must be exactly equal to the total number of clusters */
2919 	assert(ctx->mask->length == ctx->bs->total_clusters);
2920 
2921 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2922 	if (rc < 0) {
2923 		spdk_free(ctx->mask);
2924 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2925 		return;
2926 	}
2927 
2928 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2929 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2930 
2931 	spdk_free(ctx->mask);
2932 
2933 	/* Read the used blobids mask */
2934 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2935 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2936 				 SPDK_MALLOC_DMA);
2937 	if (!ctx->mask) {
2938 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2939 		return;
2940 	}
2941 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2942 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2943 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2944 				  _spdk_bs_load_used_blobids_cpl, ctx);
2945 }
2946 
2947 static void
2948 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2949 {
2950 	struct spdk_bs_load_ctx *ctx = cb_arg;
2951 	uint64_t		lba, lba_count, mask_size;
2952 	int			rc;
2953 
2954 	/* The type must be correct */
2955 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2956 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2957 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2958 				     8));
2959 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2960 	assert(ctx->mask->length == ctx->super->md_len);
2961 
2962 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2963 	if (rc < 0) {
2964 		spdk_free(ctx->mask);
2965 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2966 		return;
2967 	}
2968 
2969 	spdk_free(ctx->mask);
2970 
2971 	/* Read the used clusters mask */
2972 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2973 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
2974 				 SPDK_MALLOC_DMA);
2975 	if (!ctx->mask) {
2976 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2977 		return;
2978 	}
2979 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2980 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2981 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2982 				  _spdk_bs_load_used_clusters_cpl, ctx);
2983 }
2984 
2985 static void
2986 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2987 {
2988 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2989 	uint64_t lba, lba_count, mask_size;
2990 
2991 	/* Read the used pages mask */
2992 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2993 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
2994 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2995 	if (!ctx->mask) {
2996 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2997 		return;
2998 	}
2999 
3000 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3001 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3002 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3003 				  _spdk_bs_load_used_pages_cpl, ctx);
3004 }
3005 
3006 static int
3007 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
3008 {
3009 	struct spdk_blob_md_descriptor *desc;
3010 	size_t	cur_desc = 0;
3011 
3012 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3013 	while (cur_desc < sizeof(page->descriptors)) {
3014 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3015 			if (desc->length == 0) {
3016 				/* If padding and length are 0, this terminates the page */
3017 				break;
3018 			}
3019 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
3020 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
3021 			unsigned int				i, j;
3022 			unsigned int				cluster_count = 0;
3023 			uint32_t				cluster_idx;
3024 
3025 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
3026 
3027 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
3028 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
3029 					cluster_idx = desc_extent_rle->extents[i].cluster_idx;
3030 					/*
3031 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
3032 					 * in the used cluster map.
3033 					 */
3034 					if (cluster_idx != 0) {
3035 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
3036 						if (bs->num_free_clusters == 0) {
3037 							return -ENOSPC;
3038 						}
3039 						bs->num_free_clusters--;
3040 					}
3041 					cluster_count++;
3042 				}
3043 			}
3044 			if (cluster_count == 0) {
3045 				return -EINVAL;
3046 			}
3047 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3048 			/* Skip this item */
3049 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3050 			/* Skip this item */
3051 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3052 			/* Skip this item */
3053 		} else {
3054 			/* Error */
3055 			return -EINVAL;
3056 		}
3057 		/* Advance to the next descriptor */
3058 		cur_desc += sizeof(*desc) + desc->length;
3059 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3060 			break;
3061 		}
3062 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3063 	}
3064 	return 0;
3065 }
3066 
3067 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
3068 {
3069 	uint32_t crc;
3070 
3071 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
3072 	if (crc != ctx->page->crc) {
3073 		return false;
3074 	}
3075 
3076 	if (ctx->page->sequence_num == 0 &&
3077 	    _spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
3078 		return false;
3079 	}
3080 	return true;
3081 }
3082 
3083 static void
3084 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3085 
3086 static void
3087 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3088 {
3089 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3090 
3091 	_spdk_bs_load_complete(seq, ctx, bserrno);
3092 }
3093 
3094 static void
3095 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3096 {
3097 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3098 
3099 	spdk_free(ctx->mask);
3100 	ctx->mask = NULL;
3101 
3102 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
3103 }
3104 
3105 static void
3106 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3107 {
3108 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3109 
3110 	spdk_free(ctx->mask);
3111 	ctx->mask = NULL;
3112 
3113 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
3114 }
3115 
3116 static void
3117 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3118 {
3119 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
3120 }
3121 
3122 static void
3123 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3124 {
3125 	struct spdk_bs_load_ctx *ctx = cb_arg;
3126 	uint64_t num_md_clusters;
3127 	uint64_t i;
3128 	uint32_t page_num;
3129 
3130 	if (bserrno != 0) {
3131 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
3132 		return;
3133 	}
3134 
3135 	page_num = ctx->cur_page;
3136 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
3137 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
3138 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
3139 			if (ctx->page->sequence_num == 0) {
3140 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
3141 			}
3142 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
3143 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3144 				return;
3145 			}
3146 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
3147 				ctx->in_page_chain = true;
3148 				ctx->cur_page = ctx->page->next;
3149 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3150 				return;
3151 			}
3152 		}
3153 	}
3154 
3155 	ctx->in_page_chain = false;
3156 
3157 	do {
3158 		ctx->page_index++;
3159 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
3160 
3161 	if (ctx->page_index < ctx->super->md_len) {
3162 		ctx->cur_page = ctx->page_index;
3163 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3164 	} else {
3165 		/* Claim all of the clusters used by the metadata */
3166 		num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
3167 		for (i = 0; i < num_md_clusters; i++) {
3168 			_spdk_bs_claim_cluster(ctx->bs, i);
3169 		}
3170 		spdk_free(ctx->page);
3171 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
3172 	}
3173 }
3174 
3175 static void
3176 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3177 {
3178 	struct spdk_bs_load_ctx *ctx = cb_arg;
3179 	uint64_t lba;
3180 
3181 	assert(ctx->cur_page < ctx->super->md_len);
3182 	lba = _spdk_bs_md_page_to_lba(ctx->bs, ctx->cur_page);
3183 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3184 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3185 				  _spdk_bs_load_replay_md_cpl, ctx);
3186 }
3187 
3188 static void
3189 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3190 {
3191 	struct spdk_bs_load_ctx *ctx = cb_arg;
3192 
3193 	ctx->page_index = 0;
3194 	ctx->cur_page = 0;
3195 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3196 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3197 	if (!ctx->page) {
3198 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3199 		return;
3200 	}
3201 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3202 }
3203 
3204 static void
3205 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3206 {
3207 	struct spdk_bs_load_ctx *ctx = cb_arg;
3208 	int		rc;
3209 
3210 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3211 	if (rc < 0) {
3212 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3213 		return;
3214 	}
3215 
3216 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3217 	if (rc < 0) {
3218 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3219 		return;
3220 	}
3221 
3222 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3223 	if (rc < 0) {
3224 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3225 		return;
3226 	}
3227 
3228 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3229 	_spdk_bs_load_replay_md(seq, cb_arg);
3230 }
3231 
3232 static void
3233 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3234 {
3235 	struct spdk_bs_load_ctx *ctx = cb_arg;
3236 	uint32_t	crc;
3237 	int		rc;
3238 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3239 
3240 	if (ctx->super->version > SPDK_BS_VERSION ||
3241 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3242 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3243 		return;
3244 	}
3245 
3246 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3247 		   sizeof(ctx->super->signature)) != 0) {
3248 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3249 		return;
3250 	}
3251 
3252 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3253 	if (crc != ctx->super->crc) {
3254 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3255 		return;
3256 	}
3257 
3258 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3259 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3260 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3261 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3262 	} else {
3263 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3264 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3265 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3266 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3267 		return;
3268 	}
3269 
3270 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3271 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3272 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3273 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3274 		return;
3275 	}
3276 
3277 	if (ctx->super->size == 0) {
3278 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
3279 	}
3280 
3281 	if (ctx->super->io_unit_size == 0) {
3282 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
3283 	}
3284 
3285 	/* Parse the super block */
3286 	ctx->bs->clean = 1;
3287 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3288 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
3289 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3290 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
3291 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3292 	if (rc < 0) {
3293 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3294 		return;
3295 	}
3296 	ctx->bs->md_start = ctx->super->md_start;
3297 	ctx->bs->md_len = ctx->super->md_len;
3298 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
3299 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3300 	ctx->bs->super_blob = ctx->super->super_blob;
3301 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3302 
3303 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3304 		_spdk_bs_recover(seq, ctx);
3305 	} else {
3306 		_spdk_bs_load_read_used_pages(seq, ctx);
3307 	}
3308 }
3309 
3310 void
3311 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3312 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3313 {
3314 	struct spdk_blob_store	*bs;
3315 	struct spdk_bs_cpl	cpl;
3316 	spdk_bs_sequence_t	*seq;
3317 	struct spdk_bs_load_ctx *ctx;
3318 	struct spdk_bs_opts	opts = {};
3319 	int err;
3320 
3321 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3322 
3323 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3324 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3325 		dev->destroy(dev);
3326 		cb_fn(cb_arg, NULL, -EINVAL);
3327 		return;
3328 	}
3329 
3330 	if (o) {
3331 		opts = *o;
3332 	} else {
3333 		spdk_bs_opts_init(&opts);
3334 	}
3335 
3336 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3337 		dev->destroy(dev);
3338 		cb_fn(cb_arg, NULL, -EINVAL);
3339 		return;
3340 	}
3341 
3342 	err = _spdk_bs_alloc(dev, &opts, &bs);
3343 	if (err) {
3344 		dev->destroy(dev);
3345 		cb_fn(cb_arg, NULL, err);
3346 		return;
3347 	}
3348 
3349 	ctx = calloc(1, sizeof(*ctx));
3350 	if (!ctx) {
3351 		_spdk_bs_free(bs);
3352 		cb_fn(cb_arg, NULL, -ENOMEM);
3353 		return;
3354 	}
3355 
3356 	ctx->bs = bs;
3357 	ctx->iter_cb_fn = opts.iter_cb_fn;
3358 	ctx->iter_cb_arg = opts.iter_cb_arg;
3359 
3360 	/* Allocate memory for the super block */
3361 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3362 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3363 	if (!ctx->super) {
3364 		free(ctx);
3365 		_spdk_bs_free(bs);
3366 		cb_fn(cb_arg, NULL, -ENOMEM);
3367 		return;
3368 	}
3369 
3370 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3371 	cpl.u.bs_handle.cb_fn = cb_fn;
3372 	cpl.u.bs_handle.cb_arg = cb_arg;
3373 	cpl.u.bs_handle.bs = bs;
3374 
3375 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3376 	if (!seq) {
3377 		spdk_free(ctx->super);
3378 		free(ctx);
3379 		_spdk_bs_free(bs);
3380 		cb_fn(cb_arg, NULL, -ENOMEM);
3381 		return;
3382 	}
3383 
3384 	/* Read the super block */
3385 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3386 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3387 				  _spdk_bs_load_super_cpl, ctx);
3388 }
3389 
3390 /* END spdk_bs_load */
3391 
3392 /* START spdk_bs_dump */
3393 
3394 struct spdk_bs_dump_ctx {
3395 	struct spdk_blob_store		*bs;
3396 	struct spdk_bs_super_block	*super;
3397 	uint32_t			cur_page;
3398 	struct spdk_blob_md_page	*page;
3399 	spdk_bs_sequence_t		*seq;
3400 	FILE				*fp;
3401 	spdk_bs_dump_print_xattr	print_xattr_fn;
3402 	char				xattr_name[4096];
3403 };
3404 
3405 static void
3406 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3407 {
3408 	spdk_free(ctx->super);
3409 
3410 	/*
3411 	 * We need to defer calling spdk_bs_call_cpl() until after
3412 	 * dev destruction, so tuck these away for later use.
3413 	 */
3414 	ctx->bs->unload_err = bserrno;
3415 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3416 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3417 
3418 	spdk_bs_sequence_finish(seq, 0);
3419 	_spdk_bs_free(ctx->bs);
3420 	free(ctx);
3421 }
3422 
3423 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3424 
3425 static void
3426 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3427 {
3428 	uint32_t page_idx = ctx->cur_page;
3429 	struct spdk_blob_md_page *page = ctx->page;
3430 	struct spdk_blob_md_descriptor *desc;
3431 	size_t cur_desc = 0;
3432 	uint32_t crc;
3433 
3434 	fprintf(ctx->fp, "=========\n");
3435 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3436 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3437 
3438 	crc = _spdk_blob_md_page_calc_crc(page);
3439 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3440 
3441 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3442 	while (cur_desc < sizeof(page->descriptors)) {
3443 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3444 			if (desc->length == 0) {
3445 				/* If padding and length are 0, this terminates the page */
3446 				break;
3447 			}
3448 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
3449 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
3450 			unsigned int				i;
3451 
3452 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
3453 
3454 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
3455 				if (desc_extent_rle->extents[i].cluster_idx != 0) {
3456 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3457 						desc_extent_rle->extents[i].cluster_idx);
3458 				} else {
3459 					fprintf(ctx->fp, "Unallocated Extent - ");
3460 				}
3461 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent_rle->extents[i].length);
3462 				fprintf(ctx->fp, "\n");
3463 			}
3464 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3465 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3466 			uint32_t i;
3467 
3468 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3469 
3470 			if (desc_xattr->length !=
3471 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3472 			    desc_xattr->name_length + desc_xattr->value_length) {
3473 			}
3474 
3475 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3476 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3477 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3478 			fprintf(ctx->fp, "       value = \"");
3479 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3480 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3481 					    desc_xattr->value_length);
3482 			fprintf(ctx->fp, "\"\n");
3483 			for (i = 0; i < desc_xattr->value_length; i++) {
3484 				if (i % 16 == 0) {
3485 					fprintf(ctx->fp, "               ");
3486 				}
3487 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3488 				if ((i + 1) % 16 == 0) {
3489 					fprintf(ctx->fp, "\n");
3490 				}
3491 			}
3492 			if (i % 16 != 0) {
3493 				fprintf(ctx->fp, "\n");
3494 			}
3495 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3496 			/* TODO */
3497 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3498 			/* TODO */
3499 		} else {
3500 			/* Error */
3501 		}
3502 		/* Advance to the next descriptor */
3503 		cur_desc += sizeof(*desc) + desc->length;
3504 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3505 			break;
3506 		}
3507 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3508 	}
3509 }
3510 
3511 static void
3512 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3513 {
3514 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3515 
3516 	if (bserrno != 0) {
3517 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3518 		return;
3519 	}
3520 
3521 	if (ctx->page->id != 0) {
3522 		_spdk_bs_dump_print_md_page(ctx);
3523 	}
3524 
3525 	ctx->cur_page++;
3526 
3527 	if (ctx->cur_page < ctx->super->md_len) {
3528 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3529 	} else {
3530 		spdk_free(ctx->page);
3531 		_spdk_bs_dump_finish(seq, ctx, 0);
3532 	}
3533 }
3534 
3535 static void
3536 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3537 {
3538 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3539 	uint64_t lba;
3540 
3541 	assert(ctx->cur_page < ctx->super->md_len);
3542 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3543 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3544 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3545 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3546 }
3547 
3548 static void
3549 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3550 {
3551 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3552 
3553 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3554 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3555 		   sizeof(ctx->super->signature)) != 0) {
3556 		fprintf(ctx->fp, "(Mismatch)\n");
3557 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3558 		return;
3559 	} else {
3560 		fprintf(ctx->fp, "(OK)\n");
3561 	}
3562 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3563 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3564 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3565 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3566 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3567 	fprintf(ctx->fp, "Super Blob ID: ");
3568 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3569 		fprintf(ctx->fp, "(None)\n");
3570 	} else {
3571 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3572 	}
3573 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3574 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3575 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3576 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3577 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3578 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3579 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3580 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3581 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3582 
3583 	ctx->cur_page = 0;
3584 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3585 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3586 	if (!ctx->page) {
3587 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3588 		return;
3589 	}
3590 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3591 }
3592 
3593 void
3594 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3595 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3596 {
3597 	struct spdk_blob_store	*bs;
3598 	struct spdk_bs_cpl	cpl;
3599 	spdk_bs_sequence_t	*seq;
3600 	struct spdk_bs_dump_ctx *ctx;
3601 	struct spdk_bs_opts	opts = {};
3602 	int err;
3603 
3604 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3605 
3606 	spdk_bs_opts_init(&opts);
3607 
3608 	err = _spdk_bs_alloc(dev, &opts, &bs);
3609 	if (err) {
3610 		dev->destroy(dev);
3611 		cb_fn(cb_arg, err);
3612 		return;
3613 	}
3614 
3615 	ctx = calloc(1, sizeof(*ctx));
3616 	if (!ctx) {
3617 		_spdk_bs_free(bs);
3618 		cb_fn(cb_arg, -ENOMEM);
3619 		return;
3620 	}
3621 
3622 	ctx->bs = bs;
3623 	ctx->fp = fp;
3624 	ctx->print_xattr_fn = print_xattr_fn;
3625 
3626 	/* Allocate memory for the super block */
3627 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3628 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3629 	if (!ctx->super) {
3630 		free(ctx);
3631 		_spdk_bs_free(bs);
3632 		cb_fn(cb_arg, -ENOMEM);
3633 		return;
3634 	}
3635 
3636 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3637 	cpl.u.bs_basic.cb_fn = cb_fn;
3638 	cpl.u.bs_basic.cb_arg = cb_arg;
3639 
3640 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3641 	if (!seq) {
3642 		spdk_free(ctx->super);
3643 		free(ctx);
3644 		_spdk_bs_free(bs);
3645 		cb_fn(cb_arg, -ENOMEM);
3646 		return;
3647 	}
3648 
3649 	/* Read the super block */
3650 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3651 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3652 				  _spdk_bs_dump_super_cpl, ctx);
3653 }
3654 
3655 /* END spdk_bs_dump */
3656 
3657 /* START spdk_bs_init */
3658 
3659 struct spdk_bs_init_ctx {
3660 	struct spdk_blob_store		*bs;
3661 	struct spdk_bs_super_block	*super;
3662 };
3663 
3664 static void
3665 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3666 {
3667 	struct spdk_bs_init_ctx *ctx = cb_arg;
3668 
3669 	spdk_free(ctx->super);
3670 	free(ctx);
3671 
3672 	spdk_bs_sequence_finish(seq, bserrno);
3673 }
3674 
3675 static void
3676 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3677 {
3678 	struct spdk_bs_init_ctx *ctx = cb_arg;
3679 
3680 	/* Write super block */
3681 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3682 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3683 				   _spdk_bs_init_persist_super_cpl, ctx);
3684 }
3685 
3686 void
3687 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3688 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3689 {
3690 	struct spdk_bs_init_ctx *ctx;
3691 	struct spdk_blob_store	*bs;
3692 	struct spdk_bs_cpl	cpl;
3693 	spdk_bs_sequence_t	*seq;
3694 	spdk_bs_batch_t		*batch;
3695 	uint64_t		num_md_lba;
3696 	uint64_t		num_md_pages;
3697 	uint64_t		num_md_clusters;
3698 	uint32_t		i;
3699 	struct spdk_bs_opts	opts = {};
3700 	int			rc;
3701 
3702 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3703 
3704 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3705 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3706 			    dev->blocklen);
3707 		dev->destroy(dev);
3708 		cb_fn(cb_arg, NULL, -EINVAL);
3709 		return;
3710 	}
3711 
3712 	if (o) {
3713 		opts = *o;
3714 	} else {
3715 		spdk_bs_opts_init(&opts);
3716 	}
3717 
3718 	if (_spdk_bs_opts_verify(&opts) != 0) {
3719 		dev->destroy(dev);
3720 		cb_fn(cb_arg, NULL, -EINVAL);
3721 		return;
3722 	}
3723 
3724 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3725 	if (rc) {
3726 		dev->destroy(dev);
3727 		cb_fn(cb_arg, NULL, rc);
3728 		return;
3729 	}
3730 
3731 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3732 		/* By default, allocate 1 page per cluster.
3733 		 * Technically, this over-allocates metadata
3734 		 * because more metadata will reduce the number
3735 		 * of usable clusters. This can be addressed with
3736 		 * more complex math in the future.
3737 		 */
3738 		bs->md_len = bs->total_clusters;
3739 	} else {
3740 		bs->md_len = opts.num_md_pages;
3741 	}
3742 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3743 	if (rc < 0) {
3744 		_spdk_bs_free(bs);
3745 		cb_fn(cb_arg, NULL, -ENOMEM);
3746 		return;
3747 	}
3748 
3749 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3750 	if (rc < 0) {
3751 		_spdk_bs_free(bs);
3752 		cb_fn(cb_arg, NULL, -ENOMEM);
3753 		return;
3754 	}
3755 
3756 	ctx = calloc(1, sizeof(*ctx));
3757 	if (!ctx) {
3758 		_spdk_bs_free(bs);
3759 		cb_fn(cb_arg, NULL, -ENOMEM);
3760 		return;
3761 	}
3762 
3763 	ctx->bs = bs;
3764 
3765 	/* Allocate memory for the super block */
3766 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3767 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3768 	if (!ctx->super) {
3769 		free(ctx);
3770 		_spdk_bs_free(bs);
3771 		cb_fn(cb_arg, NULL, -ENOMEM);
3772 		return;
3773 	}
3774 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3775 	       sizeof(ctx->super->signature));
3776 	ctx->super->version = SPDK_BS_VERSION;
3777 	ctx->super->length = sizeof(*ctx->super);
3778 	ctx->super->super_blob = bs->super_blob;
3779 	ctx->super->clean = 0;
3780 	ctx->super->cluster_size = bs->cluster_sz;
3781 	ctx->super->io_unit_size = bs->io_unit_size;
3782 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3783 
3784 	/* Calculate how many pages the metadata consumes at the front
3785 	 * of the disk.
3786 	 */
3787 
3788 	/* The super block uses 1 page */
3789 	num_md_pages = 1;
3790 
3791 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3792 	 * up to the nearest page, plus a header.
3793 	 */
3794 	ctx->super->used_page_mask_start = num_md_pages;
3795 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3796 					 spdk_divide_round_up(bs->md_len, 8),
3797 					 SPDK_BS_PAGE_SIZE);
3798 	num_md_pages += ctx->super->used_page_mask_len;
3799 
3800 	/* The used_clusters mask requires 1 bit per cluster, rounded
3801 	 * up to the nearest page, plus a header.
3802 	 */
3803 	ctx->super->used_cluster_mask_start = num_md_pages;
3804 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3805 					    spdk_divide_round_up(bs->total_clusters, 8),
3806 					    SPDK_BS_PAGE_SIZE);
3807 	num_md_pages += ctx->super->used_cluster_mask_len;
3808 
3809 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3810 	 * up to the nearest page, plus a header.
3811 	 */
3812 	ctx->super->used_blobid_mask_start = num_md_pages;
3813 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
3814 					   spdk_divide_round_up(bs->md_len, 8),
3815 					   SPDK_BS_PAGE_SIZE);
3816 	num_md_pages += ctx->super->used_blobid_mask_len;
3817 
3818 	/* The metadata region size was chosen above */
3819 	ctx->super->md_start = bs->md_start = num_md_pages;
3820 	ctx->super->md_len = bs->md_len;
3821 	num_md_pages += bs->md_len;
3822 
3823 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3824 
3825 	ctx->super->size = dev->blockcnt * dev->blocklen;
3826 
3827 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3828 
3829 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
3830 	if (num_md_clusters > bs->total_clusters) {
3831 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3832 			    "please decrease number of pages reserved for metadata "
3833 			    "or increase cluster size.\n");
3834 		spdk_free(ctx->super);
3835 		free(ctx);
3836 		_spdk_bs_free(bs);
3837 		cb_fn(cb_arg, NULL, -ENOMEM);
3838 		return;
3839 	}
3840 	/* Claim all of the clusters used by the metadata */
3841 	for (i = 0; i < num_md_clusters; i++) {
3842 		_spdk_bs_claim_cluster(bs, i);
3843 	}
3844 
3845 	bs->total_data_clusters = bs->num_free_clusters;
3846 
3847 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3848 	cpl.u.bs_handle.cb_fn = cb_fn;
3849 	cpl.u.bs_handle.cb_arg = cb_arg;
3850 	cpl.u.bs_handle.bs = bs;
3851 
3852 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3853 	if (!seq) {
3854 		spdk_free(ctx->super);
3855 		free(ctx);
3856 		_spdk_bs_free(bs);
3857 		cb_fn(cb_arg, NULL, -ENOMEM);
3858 		return;
3859 	}
3860 
3861 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3862 
3863 	/* Clear metadata space */
3864 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3865 
3866 	switch (opts.clear_method) {
3867 	case BS_CLEAR_WITH_UNMAP:
3868 		/* Trim data clusters */
3869 		spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3870 		break;
3871 	case BS_CLEAR_WITH_WRITE_ZEROES:
3872 		/* Write_zeroes to data clusters */
3873 		spdk_bs_batch_write_zeroes_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3874 		break;
3875 	case BS_CLEAR_WITH_NONE:
3876 	default:
3877 		break;
3878 	}
3879 
3880 	spdk_bs_batch_close(batch);
3881 }
3882 
3883 /* END spdk_bs_init */
3884 
3885 /* START spdk_bs_destroy */
3886 
3887 static void
3888 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3889 {
3890 	struct spdk_bs_init_ctx *ctx = cb_arg;
3891 	struct spdk_blob_store *bs = ctx->bs;
3892 
3893 	/*
3894 	 * We need to defer calling spdk_bs_call_cpl() until after
3895 	 * dev destruction, so tuck these away for later use.
3896 	 */
3897 	bs->unload_err = bserrno;
3898 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3899 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3900 
3901 	spdk_bs_sequence_finish(seq, bserrno);
3902 
3903 	_spdk_bs_free(bs);
3904 	free(ctx);
3905 }
3906 
3907 void
3908 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3909 		void *cb_arg)
3910 {
3911 	struct spdk_bs_cpl	cpl;
3912 	spdk_bs_sequence_t	*seq;
3913 	struct spdk_bs_init_ctx *ctx;
3914 
3915 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3916 
3917 	if (!TAILQ_EMPTY(&bs->blobs)) {
3918 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3919 		cb_fn(cb_arg, -EBUSY);
3920 		return;
3921 	}
3922 
3923 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3924 	cpl.u.bs_basic.cb_fn = cb_fn;
3925 	cpl.u.bs_basic.cb_arg = cb_arg;
3926 
3927 	ctx = calloc(1, sizeof(*ctx));
3928 	if (!ctx) {
3929 		cb_fn(cb_arg, -ENOMEM);
3930 		return;
3931 	}
3932 
3933 	ctx->bs = bs;
3934 
3935 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3936 	if (!seq) {
3937 		free(ctx);
3938 		cb_fn(cb_arg, -ENOMEM);
3939 		return;
3940 	}
3941 
3942 	/* Write zeroes to the super block */
3943 	spdk_bs_sequence_write_zeroes_dev(seq,
3944 					  _spdk_bs_page_to_lba(bs, 0),
3945 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3946 					  _spdk_bs_destroy_trim_cpl, ctx);
3947 }
3948 
3949 /* END spdk_bs_destroy */
3950 
3951 /* START spdk_bs_unload */
3952 
3953 static void
3954 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3955 {
3956 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3957 
3958 	spdk_free(ctx->super);
3959 
3960 	/*
3961 	 * We need to defer calling spdk_bs_call_cpl() until after
3962 	 * dev destruction, so tuck these away for later use.
3963 	 */
3964 	ctx->bs->unload_err = bserrno;
3965 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3966 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3967 
3968 	spdk_bs_sequence_finish(seq, bserrno);
3969 
3970 	_spdk_bs_free(ctx->bs);
3971 	free(ctx);
3972 }
3973 
3974 static void
3975 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3976 {
3977 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3978 
3979 	spdk_free(ctx->mask);
3980 	ctx->super->clean = 1;
3981 
3982 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3983 }
3984 
3985 static void
3986 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3987 {
3988 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3989 
3990 	spdk_free(ctx->mask);
3991 	ctx->mask = NULL;
3992 
3993 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3994 }
3995 
3996 static void
3997 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3998 {
3999 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4000 
4001 	spdk_free(ctx->mask);
4002 	ctx->mask = NULL;
4003 
4004 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
4005 }
4006 
4007 static void
4008 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4009 {
4010 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
4011 }
4012 
4013 void
4014 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
4015 {
4016 	struct spdk_bs_cpl	cpl;
4017 	spdk_bs_sequence_t	*seq;
4018 	struct spdk_bs_load_ctx *ctx;
4019 
4020 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
4021 
4022 	if (!TAILQ_EMPTY(&bs->blobs)) {
4023 		SPDK_ERRLOG("Blobstore still has open blobs\n");
4024 		cb_fn(cb_arg, -EBUSY);
4025 		return;
4026 	}
4027 
4028 	ctx = calloc(1, sizeof(*ctx));
4029 	if (!ctx) {
4030 		cb_fn(cb_arg, -ENOMEM);
4031 		return;
4032 	}
4033 
4034 	ctx->bs = bs;
4035 
4036 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4037 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4038 	if (!ctx->super) {
4039 		free(ctx);
4040 		cb_fn(cb_arg, -ENOMEM);
4041 		return;
4042 	}
4043 
4044 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4045 	cpl.u.bs_basic.cb_fn = cb_fn;
4046 	cpl.u.bs_basic.cb_arg = cb_arg;
4047 
4048 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4049 	if (!seq) {
4050 		spdk_free(ctx->super);
4051 		free(ctx);
4052 		cb_fn(cb_arg, -ENOMEM);
4053 		return;
4054 	}
4055 
4056 	/* Read super block */
4057 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4058 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4059 				  _spdk_bs_unload_read_super_cpl, ctx);
4060 }
4061 
4062 /* END spdk_bs_unload */
4063 
4064 /* START spdk_bs_set_super */
4065 
4066 struct spdk_bs_set_super_ctx {
4067 	struct spdk_blob_store		*bs;
4068 	struct spdk_bs_super_block	*super;
4069 };
4070 
4071 static void
4072 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4073 {
4074 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
4075 
4076 	if (bserrno != 0) {
4077 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
4078 	}
4079 
4080 	spdk_free(ctx->super);
4081 
4082 	spdk_bs_sequence_finish(seq, bserrno);
4083 
4084 	free(ctx);
4085 }
4086 
4087 static void
4088 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4089 {
4090 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
4091 
4092 	if (bserrno != 0) {
4093 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
4094 		spdk_free(ctx->super);
4095 		spdk_bs_sequence_finish(seq, bserrno);
4096 		free(ctx);
4097 		return;
4098 	}
4099 
4100 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
4101 }
4102 
4103 void
4104 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
4105 		  spdk_bs_op_complete cb_fn, void *cb_arg)
4106 {
4107 	struct spdk_bs_cpl		cpl;
4108 	spdk_bs_sequence_t		*seq;
4109 	struct spdk_bs_set_super_ctx	*ctx;
4110 
4111 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
4112 
4113 	ctx = calloc(1, sizeof(*ctx));
4114 	if (!ctx) {
4115 		cb_fn(cb_arg, -ENOMEM);
4116 		return;
4117 	}
4118 
4119 	ctx->bs = bs;
4120 
4121 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4122 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4123 	if (!ctx->super) {
4124 		free(ctx);
4125 		cb_fn(cb_arg, -ENOMEM);
4126 		return;
4127 	}
4128 
4129 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4130 	cpl.u.bs_basic.cb_fn = cb_fn;
4131 	cpl.u.bs_basic.cb_arg = cb_arg;
4132 
4133 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4134 	if (!seq) {
4135 		spdk_free(ctx->super);
4136 		free(ctx);
4137 		cb_fn(cb_arg, -ENOMEM);
4138 		return;
4139 	}
4140 
4141 	bs->super_blob = blobid;
4142 
4143 	/* Read super block */
4144 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4145 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4146 				  _spdk_bs_set_super_read_cpl, ctx);
4147 }
4148 
4149 /* END spdk_bs_set_super */
4150 
4151 void
4152 spdk_bs_get_super(struct spdk_blob_store *bs,
4153 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4154 {
4155 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
4156 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
4157 	} else {
4158 		cb_fn(cb_arg, bs->super_blob, 0);
4159 	}
4160 }
4161 
4162 uint64_t
4163 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
4164 {
4165 	return bs->cluster_sz;
4166 }
4167 
4168 uint64_t
4169 spdk_bs_get_page_size(struct spdk_blob_store *bs)
4170 {
4171 	return SPDK_BS_PAGE_SIZE;
4172 }
4173 
4174 uint64_t
4175 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
4176 {
4177 	return bs->io_unit_size;
4178 }
4179 
4180 uint64_t
4181 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
4182 {
4183 	return bs->num_free_clusters;
4184 }
4185 
4186 uint64_t
4187 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
4188 {
4189 	return bs->total_data_clusters;
4190 }
4191 
4192 static int
4193 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
4194 {
4195 	bs->md_channel = spdk_get_io_channel(bs);
4196 	if (!bs->md_channel) {
4197 		SPDK_ERRLOG("Failed to get IO channel.\n");
4198 		return -1;
4199 	}
4200 
4201 	return 0;
4202 }
4203 
4204 static int
4205 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
4206 {
4207 	spdk_put_io_channel(bs->md_channel);
4208 
4209 	return 0;
4210 }
4211 
4212 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4213 {
4214 	assert(blob != NULL);
4215 
4216 	return blob->id;
4217 }
4218 
4219 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4220 {
4221 	assert(blob != NULL);
4222 
4223 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4224 }
4225 
4226 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
4227 {
4228 	assert(blob != NULL);
4229 
4230 	return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs);
4231 }
4232 
4233 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4234 {
4235 	assert(blob != NULL);
4236 
4237 	return blob->active.num_clusters;
4238 }
4239 
4240 /* START spdk_bs_create_blob */
4241 
4242 static void
4243 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4244 {
4245 	struct spdk_blob *blob = cb_arg;
4246 
4247 	_spdk_blob_free(blob);
4248 
4249 	spdk_bs_sequence_finish(seq, bserrno);
4250 }
4251 
4252 static int
4253 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4254 		      bool internal)
4255 {
4256 	uint64_t i;
4257 	size_t value_len = 0;
4258 	int rc;
4259 	const void *value = NULL;
4260 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4261 		return -EINVAL;
4262 	}
4263 	for (i = 0; i < xattrs->count; i++) {
4264 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4265 		if (value == NULL || value_len == 0) {
4266 			return -EINVAL;
4267 		}
4268 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4269 		if (rc < 0) {
4270 			return rc;
4271 		}
4272 	}
4273 	return 0;
4274 }
4275 
4276 static void
4277 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4278 		     const struct spdk_blob_opts *opts,
4279 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4280 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4281 {
4282 	struct spdk_blob	*blob;
4283 	uint32_t		page_idx;
4284 	struct spdk_bs_cpl	cpl;
4285 	struct spdk_blob_opts	opts_default;
4286 	struct spdk_blob_xattr_opts internal_xattrs_default;
4287 	spdk_bs_sequence_t	*seq;
4288 	spdk_blob_id		id;
4289 	int rc;
4290 
4291 	assert(spdk_get_thread() == bs->md_thread);
4292 
4293 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4294 	if (page_idx == UINT32_MAX) {
4295 		cb_fn(cb_arg, 0, -ENOMEM);
4296 		return;
4297 	}
4298 	spdk_bit_array_set(bs->used_blobids, page_idx);
4299 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4300 
4301 	id = _spdk_bs_page_to_blobid(page_idx);
4302 
4303 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4304 
4305 	blob = _spdk_blob_alloc(bs, id);
4306 	if (!blob) {
4307 		cb_fn(cb_arg, 0, -ENOMEM);
4308 		return;
4309 	}
4310 
4311 	if (!opts) {
4312 		spdk_blob_opts_init(&opts_default);
4313 		opts = &opts_default;
4314 	}
4315 	if (!internal_xattrs) {
4316 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4317 		internal_xattrs = &internal_xattrs_default;
4318 	}
4319 
4320 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4321 	if (rc < 0) {
4322 		_spdk_blob_free(blob);
4323 		cb_fn(cb_arg, 0, rc);
4324 		return;
4325 	}
4326 
4327 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4328 	if (rc < 0) {
4329 		_spdk_blob_free(blob);
4330 		cb_fn(cb_arg, 0, rc);
4331 		return;
4332 	}
4333 
4334 	if (opts->thin_provision) {
4335 		_spdk_blob_set_thin_provision(blob);
4336 	}
4337 
4338 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4339 	if (rc < 0) {
4340 		_spdk_blob_free(blob);
4341 		cb_fn(cb_arg, 0, rc);
4342 		return;
4343 	}
4344 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4345 	cpl.u.blobid.cb_fn = cb_fn;
4346 	cpl.u.blobid.cb_arg = cb_arg;
4347 	cpl.u.blobid.blobid = blob->id;
4348 
4349 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4350 	if (!seq) {
4351 		_spdk_blob_free(blob);
4352 		cb_fn(cb_arg, 0, -ENOMEM);
4353 		return;
4354 	}
4355 
4356 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4357 }
4358 
4359 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4360 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4361 {
4362 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4363 }
4364 
4365 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4366 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4367 {
4368 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4369 }
4370 
4371 /* END spdk_bs_create_blob */
4372 
4373 /* START blob_cleanup */
4374 
4375 struct spdk_clone_snapshot_ctx {
4376 	struct spdk_bs_cpl      cpl;
4377 	int bserrno;
4378 	bool frozen;
4379 
4380 	struct spdk_io_channel *channel;
4381 
4382 	/* Current cluster for inflate operation */
4383 	uint64_t cluster;
4384 
4385 	/* For inflation force allocation of all unallocated clusters and remove
4386 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4387 	bool allocate_all;
4388 
4389 	struct {
4390 		spdk_blob_id id;
4391 		struct spdk_blob *blob;
4392 	} original;
4393 	struct {
4394 		spdk_blob_id id;
4395 		struct spdk_blob *blob;
4396 	} new;
4397 
4398 	/* xattrs specified for snapshot/clones only. They have no impact on
4399 	 * the original blobs xattrs. */
4400 	const struct spdk_blob_xattr_opts *xattrs;
4401 };
4402 
4403 static void
4404 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4405 {
4406 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4407 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4408 
4409 	if (bserrno != 0) {
4410 		if (ctx->bserrno != 0) {
4411 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4412 		} else {
4413 			ctx->bserrno = bserrno;
4414 		}
4415 	}
4416 
4417 	switch (cpl->type) {
4418 	case SPDK_BS_CPL_TYPE_BLOBID:
4419 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4420 		break;
4421 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4422 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4423 		break;
4424 	default:
4425 		SPDK_UNREACHABLE();
4426 		break;
4427 	}
4428 
4429 	free(ctx);
4430 }
4431 
4432 static void
4433 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4434 {
4435 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4436 	struct spdk_blob *origblob = ctx->original.blob;
4437 
4438 	if (bserrno != 0) {
4439 		if (ctx->bserrno != 0) {
4440 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4441 		} else {
4442 			ctx->bserrno = bserrno;
4443 		}
4444 	}
4445 
4446 	ctx->original.id = origblob->id;
4447 	origblob->locked_operation_in_progress = false;
4448 
4449 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4450 }
4451 
4452 static void
4453 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4454 {
4455 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4456 	struct spdk_blob *origblob = ctx->original.blob;
4457 
4458 	if (bserrno != 0) {
4459 		if (ctx->bserrno != 0) {
4460 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4461 		} else {
4462 			ctx->bserrno = bserrno;
4463 		}
4464 	}
4465 
4466 	if (ctx->frozen) {
4467 		/* Unfreeze any outstanding I/O */
4468 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4469 	} else {
4470 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4471 	}
4472 
4473 }
4474 
4475 static void
4476 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4477 {
4478 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4479 	struct spdk_blob *newblob = ctx->new.blob;
4480 
4481 	if (bserrno != 0) {
4482 		if (ctx->bserrno != 0) {
4483 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4484 		} else {
4485 			ctx->bserrno = bserrno;
4486 		}
4487 	}
4488 
4489 	ctx->new.id = newblob->id;
4490 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4491 }
4492 
4493 /* END blob_cleanup */
4494 
4495 /* START spdk_bs_create_snapshot */
4496 
4497 static void
4498 _spdk_bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
4499 {
4500 	uint64_t *cluster_temp;
4501 
4502 	cluster_temp = blob1->active.clusters;
4503 	blob1->active.clusters = blob2->active.clusters;
4504 	blob2->active.clusters = cluster_temp;
4505 }
4506 
4507 static void
4508 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4509 {
4510 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4511 	struct spdk_blob *origblob = ctx->original.blob;
4512 	struct spdk_blob *newblob = ctx->new.blob;
4513 
4514 	if (bserrno != 0) {
4515 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4516 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4517 		return;
4518 	}
4519 
4520 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4521 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4522 	if (bserrno != 0) {
4523 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4524 		return;
4525 	}
4526 
4527 	_spdk_bs_blob_list_add(ctx->original.blob);
4528 
4529 	spdk_blob_set_read_only(newblob);
4530 
4531 	/* sync snapshot metadata */
4532 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4533 }
4534 
4535 static void
4536 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4537 {
4538 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4539 	struct spdk_blob *origblob = ctx->original.blob;
4540 	struct spdk_blob *newblob = ctx->new.blob;
4541 
4542 	if (bserrno != 0) {
4543 		/* return cluster map back to original */
4544 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4545 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4546 		return;
4547 	}
4548 
4549 	/* Set internal xattr for snapshot id */
4550 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4551 	if (bserrno != 0) {
4552 		/* return cluster map back to original */
4553 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4554 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4555 		return;
4556 	}
4557 
4558 	_spdk_bs_blob_list_remove(origblob);
4559 	origblob->parent_id = newblob->id;
4560 
4561 	/* Create new back_bs_dev for snapshot */
4562 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4563 	if (origblob->back_bs_dev == NULL) {
4564 		/* return cluster map back to original */
4565 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4566 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4567 		return;
4568 	}
4569 
4570 	/* set clone blob as thin provisioned */
4571 	_spdk_blob_set_thin_provision(origblob);
4572 
4573 	_spdk_bs_blob_list_add(newblob);
4574 
4575 	/* sync clone metadata */
4576 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4577 }
4578 
4579 static void
4580 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4581 {
4582 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4583 	struct spdk_blob *origblob = ctx->original.blob;
4584 	struct spdk_blob *newblob = ctx->new.blob;
4585 	int bserrno;
4586 
4587 	if (rc != 0) {
4588 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4589 		return;
4590 	}
4591 
4592 	ctx->frozen = true;
4593 
4594 	/* set new back_bs_dev for snapshot */
4595 	newblob->back_bs_dev = origblob->back_bs_dev;
4596 	/* Set invalid flags from origblob */
4597 	newblob->invalid_flags = origblob->invalid_flags;
4598 
4599 	/* inherit parent from original blob if set */
4600 	newblob->parent_id = origblob->parent_id;
4601 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4602 		/* Set internal xattr for snapshot id */
4603 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4604 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4605 		if (bserrno != 0) {
4606 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4607 			return;
4608 		}
4609 	}
4610 
4611 	/* swap cluster maps */
4612 	_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
4613 
4614 	/* sync snapshot metadata */
4615 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4616 }
4617 
4618 static void
4619 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4620 {
4621 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4622 	struct spdk_blob *origblob = ctx->original.blob;
4623 	struct spdk_blob *newblob = _blob;
4624 
4625 	if (bserrno != 0) {
4626 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4627 		return;
4628 	}
4629 
4630 	ctx->new.blob = newblob;
4631 	assert(spdk_blob_is_thin_provisioned(newblob));
4632 	assert(spdk_mem_all_zero(newblob->active.clusters,
4633 				 newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
4634 
4635 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4636 }
4637 
4638 static void
4639 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4640 {
4641 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4642 	struct spdk_blob *origblob = ctx->original.blob;
4643 
4644 	if (bserrno != 0) {
4645 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4646 		return;
4647 	}
4648 
4649 	ctx->new.id = blobid;
4650 	ctx->cpl.u.blobid.blobid = blobid;
4651 
4652 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4653 }
4654 
4655 
4656 static void
4657 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4658 			const void **value, size_t *value_len)
4659 {
4660 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4661 
4662 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4663 	*value = &blob->id;
4664 	*value_len = sizeof(blob->id);
4665 }
4666 
4667 static void
4668 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4669 {
4670 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4671 	struct spdk_blob_opts opts;
4672 	struct spdk_blob_xattr_opts internal_xattrs;
4673 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4674 
4675 	if (bserrno != 0) {
4676 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4677 		return;
4678 	}
4679 
4680 	ctx->original.blob = _blob;
4681 
4682 	if (_blob->data_ro || _blob->md_ro) {
4683 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4684 			      _blob->id);
4685 		ctx->bserrno = -EINVAL;
4686 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4687 		return;
4688 	}
4689 
4690 	if (_blob->locked_operation_in_progress) {
4691 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot - another operation in progress\n");
4692 		ctx->bserrno = -EBUSY;
4693 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4694 		return;
4695 	}
4696 
4697 	_blob->locked_operation_in_progress = true;
4698 
4699 	spdk_blob_opts_init(&opts);
4700 	_spdk_blob_xattrs_init(&internal_xattrs);
4701 
4702 	/* Change the size of new blob to the same as in original blob,
4703 	 * but do not allocate clusters */
4704 	opts.thin_provision = true;
4705 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4706 
4707 	/* If there are any xattrs specified for snapshot, set them now */
4708 	if (ctx->xattrs) {
4709 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4710 	}
4711 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4712 	internal_xattrs.count = 1;
4713 	internal_xattrs.ctx = _blob;
4714 	internal_xattrs.names = xattrs_names;
4715 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4716 
4717 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4718 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4719 }
4720 
4721 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4722 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4723 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4724 {
4725 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4726 
4727 	if (!ctx) {
4728 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4729 		return;
4730 	}
4731 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4732 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4733 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4734 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4735 	ctx->bserrno = 0;
4736 	ctx->frozen = false;
4737 	ctx->original.id = blobid;
4738 	ctx->xattrs = snapshot_xattrs;
4739 
4740 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4741 }
4742 /* END spdk_bs_create_snapshot */
4743 
4744 /* START spdk_bs_create_clone */
4745 
4746 static void
4747 _spdk_bs_xattr_clone(void *arg, const char *name,
4748 		     const void **value, size_t *value_len)
4749 {
4750 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4751 
4752 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4753 	*value = &blob->id;
4754 	*value_len = sizeof(blob->id);
4755 }
4756 
4757 static void
4758 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4759 {
4760 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4761 	struct spdk_blob *clone = _blob;
4762 
4763 	ctx->new.blob = clone;
4764 	_spdk_bs_blob_list_add(clone);
4765 
4766 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4767 }
4768 
4769 static void
4770 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4771 {
4772 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4773 
4774 	ctx->cpl.u.blobid.blobid = blobid;
4775 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4776 }
4777 
4778 static void
4779 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4780 {
4781 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4782 	struct spdk_blob_opts		opts;
4783 	struct spdk_blob_xattr_opts internal_xattrs;
4784 	char *xattr_names[] = { BLOB_SNAPSHOT };
4785 
4786 	if (bserrno != 0) {
4787 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4788 		return;
4789 	}
4790 
4791 	ctx->original.blob = _blob;
4792 
4793 	if (!_blob->data_ro || !_blob->md_ro) {
4794 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4795 		ctx->bserrno = -EINVAL;
4796 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4797 		return;
4798 	}
4799 
4800 	if (_blob->locked_operation_in_progress) {
4801 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create clone - another operation in progress\n");
4802 		ctx->bserrno = -EBUSY;
4803 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4804 		return;
4805 	}
4806 
4807 	_blob->locked_operation_in_progress = true;
4808 
4809 	spdk_blob_opts_init(&opts);
4810 	_spdk_blob_xattrs_init(&internal_xattrs);
4811 
4812 	opts.thin_provision = true;
4813 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4814 	if (ctx->xattrs) {
4815 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4816 	}
4817 
4818 	/* Set internal xattr BLOB_SNAPSHOT */
4819 	internal_xattrs.count = 1;
4820 	internal_xattrs.ctx = _blob;
4821 	internal_xattrs.names = xattr_names;
4822 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4823 
4824 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4825 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4826 }
4827 
4828 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4829 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4830 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4831 {
4832 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4833 
4834 	if (!ctx) {
4835 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4836 		return;
4837 	}
4838 
4839 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4840 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4841 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4842 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4843 	ctx->bserrno = 0;
4844 	ctx->xattrs = clone_xattrs;
4845 	ctx->original.id = blobid;
4846 
4847 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4848 }
4849 
4850 /* END spdk_bs_create_clone */
4851 
4852 /* START spdk_bs_inflate_blob */
4853 
4854 static void
4855 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4856 {
4857 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4858 	struct spdk_blob *_blob = ctx->original.blob;
4859 
4860 	if (bserrno != 0) {
4861 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4862 		return;
4863 	}
4864 
4865 	assert(_parent != NULL);
4866 
4867 	_spdk_bs_blob_list_remove(_blob);
4868 	_blob->parent_id = _parent->id;
4869 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4870 			     sizeof(spdk_blob_id), true);
4871 
4872 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4873 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4874 	_spdk_bs_blob_list_add(_blob);
4875 
4876 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4877 }
4878 
4879 static void
4880 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4881 {
4882 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4883 	struct spdk_blob *_blob = ctx->original.blob;
4884 	struct spdk_blob *_parent;
4885 
4886 	if (bserrno != 0) {
4887 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4888 		return;
4889 	}
4890 
4891 	if (ctx->allocate_all) {
4892 		/* remove thin provisioning */
4893 		_spdk_bs_blob_list_remove(_blob);
4894 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4895 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4896 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4897 		_blob->back_bs_dev = NULL;
4898 		_blob->parent_id = SPDK_BLOBID_INVALID;
4899 	} else {
4900 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4901 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4902 			/* We must change the parent of the inflated blob */
4903 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4904 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4905 			return;
4906 		}
4907 
4908 		_spdk_bs_blob_list_remove(_blob);
4909 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4910 		_blob->parent_id = SPDK_BLOBID_INVALID;
4911 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4912 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4913 	}
4914 
4915 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4916 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4917 }
4918 
4919 /* Check if cluster needs allocation */
4920 static inline bool
4921 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4922 {
4923 	struct spdk_blob_bs_dev *b;
4924 
4925 	assert(blob != NULL);
4926 
4927 	if (blob->active.clusters[cluster] != 0) {
4928 		/* Cluster is already allocated */
4929 		return false;
4930 	}
4931 
4932 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4933 		/* Blob have no parent blob */
4934 		return allocate_all;
4935 	}
4936 
4937 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4938 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4939 }
4940 
4941 static void
4942 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4943 {
4944 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4945 	struct spdk_blob *_blob = ctx->original.blob;
4946 	uint64_t offset;
4947 
4948 	if (bserrno != 0) {
4949 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4950 		return;
4951 	}
4952 
4953 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4954 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4955 			break;
4956 		}
4957 	}
4958 
4959 	if (ctx->cluster < _blob->active.num_clusters) {
4960 		offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster);
4961 
4962 		/* We may safely increment a cluster before write */
4963 		ctx->cluster++;
4964 
4965 		/* Use zero length write to touch a cluster */
4966 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4967 				   _spdk_bs_inflate_blob_touch_next, ctx);
4968 	} else {
4969 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4970 	}
4971 }
4972 
4973 static void
4974 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4975 {
4976 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4977 	uint64_t lfc; /* lowest free cluster */
4978 	uint64_t i;
4979 
4980 	if (bserrno != 0) {
4981 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4982 		return;
4983 	}
4984 
4985 	ctx->original.blob = _blob;
4986 
4987 	if (_blob->locked_operation_in_progress) {
4988 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot inflate blob - another operation in progress\n");
4989 		ctx->bserrno = -EBUSY;
4990 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4991 		return;
4992 	}
4993 
4994 	_blob->locked_operation_in_progress = true;
4995 
4996 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4997 		/* This blob have no parent, so we cannot decouple it. */
4998 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4999 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
5000 		return;
5001 	}
5002 
5003 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
5004 		/* This is not thin provisioned blob. No need to inflate. */
5005 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
5006 		return;
5007 	}
5008 
5009 	/* Do two passes - one to verify that we can obtain enough clusters
5010 	 * and another to actually claim them.
5011 	 */
5012 	lfc = 0;
5013 	for (i = 0; i < _blob->active.num_clusters; i++) {
5014 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
5015 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
5016 			if (lfc == UINT32_MAX) {
5017 				/* No more free clusters. Cannot satisfy the request */
5018 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
5019 				return;
5020 			}
5021 			lfc++;
5022 		}
5023 	}
5024 
5025 	ctx->cluster = 0;
5026 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
5027 }
5028 
5029 static void
5030 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5031 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
5032 {
5033 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
5034 
5035 	if (!ctx) {
5036 		cb_fn(cb_arg, -ENOMEM);
5037 		return;
5038 	}
5039 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5040 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
5041 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
5042 	ctx->bserrno = 0;
5043 	ctx->original.id = blobid;
5044 	ctx->channel = channel;
5045 	ctx->allocate_all = allocate_all;
5046 
5047 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
5048 }
5049 
5050 void
5051 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5052 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
5053 {
5054 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
5055 }
5056 
5057 void
5058 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5059 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
5060 {
5061 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
5062 }
5063 /* END spdk_bs_inflate_blob */
5064 
5065 /* START spdk_blob_resize */
5066 struct spdk_bs_resize_ctx {
5067 	spdk_blob_op_complete cb_fn;
5068 	void *cb_arg;
5069 	struct spdk_blob *blob;
5070 	uint64_t sz;
5071 	int rc;
5072 };
5073 
5074 static void
5075 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
5076 {
5077 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
5078 
5079 	if (rc != 0) {
5080 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
5081 	}
5082 
5083 	if (ctx->rc != 0) {
5084 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
5085 		rc = ctx->rc;
5086 	}
5087 
5088 	ctx->blob->locked_operation_in_progress = false;
5089 
5090 	ctx->cb_fn(ctx->cb_arg, rc);
5091 	free(ctx);
5092 }
5093 
5094 static void
5095 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
5096 {
5097 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
5098 
5099 	if (rc != 0) {
5100 		ctx->blob->locked_operation_in_progress = false;
5101 		ctx->cb_fn(ctx->cb_arg, rc);
5102 		free(ctx);
5103 		return;
5104 	}
5105 
5106 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
5107 
5108 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
5109 }
5110 
5111 void
5112 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
5113 {
5114 	struct spdk_bs_resize_ctx *ctx;
5115 
5116 	_spdk_blob_verify_md_op(blob);
5117 
5118 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
5119 
5120 	if (blob->md_ro) {
5121 		cb_fn(cb_arg, -EPERM);
5122 		return;
5123 	}
5124 
5125 	if (sz == blob->active.num_clusters) {
5126 		cb_fn(cb_arg, 0);
5127 		return;
5128 	}
5129 
5130 	if (blob->locked_operation_in_progress) {
5131 		cb_fn(cb_arg, -EBUSY);
5132 		return;
5133 	}
5134 
5135 	ctx = calloc(1, sizeof(*ctx));
5136 	if (!ctx) {
5137 		cb_fn(cb_arg, -ENOMEM);
5138 		return;
5139 	}
5140 
5141 	blob->locked_operation_in_progress = true;
5142 	ctx->cb_fn = cb_fn;
5143 	ctx->cb_arg = cb_arg;
5144 	ctx->blob = blob;
5145 	ctx->sz = sz;
5146 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
5147 }
5148 
5149 /* END spdk_blob_resize */
5150 
5151 
5152 /* START spdk_bs_delete_blob */
5153 
5154 static void
5155 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
5156 {
5157 	spdk_bs_sequence_t *seq = cb_arg;
5158 
5159 	spdk_bs_sequence_finish(seq, bserrno);
5160 }
5161 
5162 static void
5163 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5164 {
5165 	struct spdk_blob *blob = cb_arg;
5166 
5167 	if (bserrno != 0) {
5168 		/*
5169 		 * We already removed this blob from the blobstore tailq, so
5170 		 *  we need to free it here since this is the last reference
5171 		 *  to it.
5172 		 */
5173 		_spdk_blob_free(blob);
5174 		_spdk_bs_delete_close_cpl(seq, bserrno);
5175 		return;
5176 	}
5177 
5178 	/*
5179 	 * This will immediately decrement the ref_count and call
5180 	 *  the completion routine since the metadata state is clean.
5181 	 *  By calling spdk_blob_close, we reduce the number of call
5182 	 *  points into code that touches the blob->open_ref count
5183 	 *  and the blobstore's blob list.
5184 	 */
5185 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
5186 }
5187 
5188 struct delete_snapshot_ctx {
5189 	struct spdk_blob_list *parent_snapshot_entry;
5190 	struct spdk_blob *snapshot;
5191 	bool snapshot_md_ro;
5192 	struct spdk_blob *clone;
5193 	bool clone_md_ro;
5194 	spdk_blob_op_with_handle_complete cb_fn;
5195 	void *cb_arg;
5196 	int bserrno;
5197 };
5198 
5199 static void
5200 _spdk_delete_blob_cleanup_finish(void *cb_arg, int bserrno)
5201 {
5202 	struct delete_snapshot_ctx *ctx = cb_arg;
5203 
5204 	if (bserrno != 0) {
5205 		SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno);
5206 	}
5207 
5208 	assert(ctx != NULL);
5209 
5210 	if (bserrno != 0 && ctx->bserrno == 0) {
5211 		ctx->bserrno = bserrno;
5212 	}
5213 
5214 	ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno);
5215 	free(ctx);
5216 }
5217 
5218 static void
5219 _spdk_delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno)
5220 {
5221 	struct delete_snapshot_ctx *ctx = cb_arg;
5222 
5223 	if (bserrno != 0) {
5224 		ctx->bserrno = bserrno;
5225 		SPDK_ERRLOG("Clone cleanup error %d\n", bserrno);
5226 	}
5227 
5228 	/* open_ref == 1 menas that only deletion context has opened this snapshot
5229 	 * open_ref == 2 menas that clone has opened this snapshot as well,
5230 	 * so we have to add it back to the blobs list */
5231 	if (ctx->snapshot->open_ref == 2) {
5232 		TAILQ_INSERT_HEAD(&ctx->snapshot->bs->blobs, ctx->snapshot, link);
5233 	}
5234 
5235 	ctx->snapshot->locked_operation_in_progress = false;
5236 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
5237 
5238 	spdk_blob_close(ctx->snapshot, _spdk_delete_blob_cleanup_finish, ctx);
5239 }
5240 
5241 static void
5242 _spdk_delete_snapshot_cleanup_clone(void *cb_arg, int bserrno)
5243 {
5244 	struct delete_snapshot_ctx *ctx = cb_arg;
5245 
5246 	ctx->clone->locked_operation_in_progress = false;
5247 	ctx->clone->md_ro = ctx->clone_md_ro;
5248 
5249 	spdk_blob_close(ctx->clone, _spdk_delete_snapshot_cleanup_snapshot, ctx);
5250 }
5251 
5252 static void
5253 _spdk_delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
5254 {
5255 	struct delete_snapshot_ctx *ctx = cb_arg;
5256 
5257 	if (bserrno) {
5258 		ctx->bserrno = bserrno;
5259 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5260 		return;
5261 	}
5262 
5263 	ctx->clone->locked_operation_in_progress = false;
5264 	spdk_blob_close(ctx->clone, _spdk_delete_blob_cleanup_finish, ctx);
5265 }
5266 
5267 static void
5268 _spdk_delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno)
5269 {
5270 	struct delete_snapshot_ctx *ctx = cb_arg;
5271 	struct spdk_blob_list *parent_snapshot_entry = NULL;
5272 	struct spdk_blob_list *snapshot_entry = NULL;
5273 	struct spdk_blob_list *clone_entry = NULL;
5274 	struct spdk_blob_list *snapshot_clone_entry = NULL;
5275 
5276 	if (bserrno) {
5277 		SPDK_ERRLOG("Failed to sync MD on blob\n");
5278 		ctx->bserrno = bserrno;
5279 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5280 		return;
5281 	}
5282 
5283 	/* Get snapshot entry for the snapshot we want to remove */
5284 	snapshot_entry = _spdk_bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id);
5285 
5286 	assert(snapshot_entry != NULL);
5287 
5288 	/* Remove clone entry in this snapshot (at this point there can be only one clone) */
5289 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
5290 	assert(clone_entry != NULL);
5291 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
5292 	snapshot_entry->clone_count--;
5293 	assert(TAILQ_EMPTY(&snapshot_entry->clones));
5294 
5295 	if (ctx->snapshot->parent_id != SPDK_BLOBID_INVALID) {
5296 		/* This snapshot is at the same time a clone of another snapshot - we need to
5297 		 * update parent snapshot (remove current clone, add new one inherited from
5298 		 * the snapshot that is being removed) */
5299 
5300 		/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
5301 		 * snapshot that we are removing */
5302 		_spdk_blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry,
5303 				&snapshot_clone_entry);
5304 
5305 		/* Switch clone entry in parent snapshot */
5306 		TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link);
5307 		TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link);
5308 		free(snapshot_clone_entry);
5309 	} else {
5310 		/* No parent snapshot - just remove clone entry */
5311 		free(clone_entry);
5312 	}
5313 
5314 	/* Restore md_ro flags */
5315 	ctx->clone->md_ro = ctx->clone_md_ro;
5316 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
5317 
5318 	_spdk_blob_unfreeze_io(ctx->clone, _spdk_delete_snapshot_unfreeze_cpl, ctx);
5319 }
5320 
5321 static void
5322 _spdk_delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno)
5323 {
5324 	struct delete_snapshot_ctx *ctx = cb_arg;
5325 	uint64_t i;
5326 
5327 	ctx->snapshot->md_ro = false;
5328 
5329 	if (bserrno) {
5330 		SPDK_ERRLOG("Failed to sync MD on clone\n");
5331 		ctx->bserrno = bserrno;
5332 
5333 		/* Restore snapshot to previous state */
5334 		bserrno = _spdk_blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true);
5335 		if (bserrno != 0) {
5336 			_spdk_delete_snapshot_cleanup_clone(ctx, bserrno);
5337 			return;
5338 		}
5339 
5340 		spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_cleanup_clone, ctx);
5341 		return;
5342 	}
5343 
5344 	/* Clear cluster map entries for snapshot */
5345 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
5346 		if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) {
5347 			ctx->snapshot->active.clusters[i] = 0;
5348 		}
5349 	}
5350 
5351 	ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY;
5352 
5353 	if (ctx->parent_snapshot_entry != NULL) {
5354 		ctx->snapshot->back_bs_dev = NULL;
5355 	}
5356 
5357 	spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_sync_snapshot_cpl, ctx);
5358 }
5359 
5360 static void
5361 _spdk_delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno)
5362 {
5363 	struct delete_snapshot_ctx *ctx = cb_arg;
5364 	uint64_t i;
5365 
5366 	/* Temporarily override md_ro flag for clone for MD modification */
5367 	ctx->clone_md_ro = ctx->clone->md_ro;
5368 	ctx->clone->md_ro = false;
5369 
5370 	if (bserrno) {
5371 		SPDK_ERRLOG("Failed to sync MD with xattr on blob\n");
5372 		ctx->bserrno = bserrno;
5373 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5374 		return;
5375 	}
5376 
5377 	/* Copy snapshot map to clone map (only unallocated clusters in clone) */
5378 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
5379 		if (ctx->clone->active.clusters[i] == 0) {
5380 			ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i];
5381 		}
5382 	}
5383 
5384 	/* Delete old backing bs_dev from clone (related to snapshot that will be removed) */
5385 	ctx->clone->back_bs_dev->destroy(ctx->clone->back_bs_dev);
5386 
5387 	/* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */
5388 	if (ctx->parent_snapshot_entry != NULL) {
5389 		/* ...to parent snapshot */
5390 		ctx->clone->parent_id = ctx->parent_snapshot_entry->id;
5391 		ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev;
5392 		_spdk_blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id,
5393 				     sizeof(spdk_blob_id),
5394 				     true);
5395 	} else {
5396 		/* ...to blobid invalid and zeroes dev */
5397 		ctx->clone->parent_id = SPDK_BLOBID_INVALID;
5398 		ctx->clone->back_bs_dev = spdk_bs_create_zeroes_dev();
5399 		_spdk_blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true);
5400 	}
5401 
5402 	spdk_blob_sync_md(ctx->clone, _spdk_delete_snapshot_sync_clone_cpl, ctx);
5403 }
5404 
5405 static void
5406 _spdk_delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno)
5407 {
5408 	struct delete_snapshot_ctx *ctx = cb_arg;
5409 
5410 	if (bserrno) {
5411 		SPDK_ERRLOG("Failed to freeze I/O on clone\n");
5412 		ctx->bserrno = bserrno;
5413 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5414 		return;
5415 	}
5416 
5417 	/* Temporarily override md_ro flag for snapshot for MD modification */
5418 	ctx->snapshot_md_ro = ctx->snapshot->md_ro;
5419 	ctx->snapshot->md_ro = false;
5420 
5421 	/* Mark blob as pending for removal for power failure safety, use clone id for recovery */
5422 	ctx->bserrno = _spdk_blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id,
5423 					    sizeof(spdk_blob_id), true);
5424 	if (ctx->bserrno != 0) {
5425 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
5426 		return;
5427 	}
5428 
5429 	spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_sync_snapshot_xattr_cpl, ctx);
5430 }
5431 
5432 static void
5433 _spdk_delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno)
5434 {
5435 	struct delete_snapshot_ctx *ctx = cb_arg;
5436 
5437 	if (bserrno) {
5438 		SPDK_ERRLOG("Failed to open clone\n");
5439 		ctx->bserrno = bserrno;
5440 		_spdk_delete_snapshot_cleanup_snapshot(ctx, 0);
5441 		return;
5442 	}
5443 
5444 	ctx->clone = clone;
5445 
5446 	if (clone->locked_operation_in_progress) {
5447 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress on its clone\n");
5448 		ctx->bserrno = -EBUSY;
5449 		spdk_blob_close(ctx->clone, _spdk_delete_snapshot_cleanup_snapshot, ctx);
5450 		return;
5451 	}
5452 
5453 	clone->locked_operation_in_progress = true;
5454 
5455 	_spdk_blob_freeze_io(clone, _spdk_delete_snapshot_freeze_io_cb, ctx);
5456 }
5457 
5458 static void
5459 _spdk_update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx)
5460 {
5461 	struct spdk_blob_list *snapshot_entry = NULL;
5462 	struct spdk_blob_list *clone_entry = NULL;
5463 	struct spdk_blob_list *snapshot_clone_entry = NULL;
5464 
5465 	/* Get snapshot entry for the snapshot we want to remove */
5466 	snapshot_entry = _spdk_bs_get_snapshot_entry(snapshot->bs, snapshot->id);
5467 
5468 	assert(snapshot_entry != NULL);
5469 
5470 	/* Get clone of the snapshot (at this point there can be only one clone) */
5471 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
5472 	assert(snapshot_entry->clone_count == 1);
5473 	assert(clone_entry != NULL);
5474 
5475 	/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
5476 	 * snapshot that we are removing */
5477 	_spdk_blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry,
5478 			&snapshot_clone_entry);
5479 
5480 	spdk_bs_open_blob(snapshot->bs, clone_entry->id, _spdk_delete_snapshot_open_clone_cb, ctx);
5481 }
5482 
5483 static void
5484 _spdk_bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno)
5485 {
5486 	spdk_bs_sequence_t *seq = cb_arg;
5487 	struct spdk_blob_list *snapshot_entry = NULL;
5488 	uint32_t page_num;
5489 
5490 	if (bserrno) {
5491 		SPDK_ERRLOG("Failed to remove blob\n");
5492 		spdk_bs_sequence_finish(seq, bserrno);
5493 		return;
5494 	}
5495 
5496 	/* Remove snapshot from the list */
5497 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
5498 	if (snapshot_entry != NULL) {
5499 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
5500 		free(snapshot_entry);
5501 	}
5502 
5503 	page_num = _spdk_bs_blobid_to_page(blob->id);
5504 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
5505 	blob->state = SPDK_BLOB_STATE_DIRTY;
5506 	blob->active.num_pages = 0;
5507 	_spdk_blob_resize(blob, 0);
5508 
5509 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
5510 }
5511 
5512 static int
5513 _spdk_bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone)
5514 {
5515 	struct spdk_blob_list *snapshot_entry = NULL;
5516 	struct spdk_blob_list *clone_entry = NULL;
5517 	struct spdk_blob *clone = NULL;
5518 	bool has_one_clone = false;
5519 
5520 	/* Check if this is a snapshot with clones */
5521 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
5522 	if (snapshot_entry != NULL) {
5523 		if (snapshot_entry->clone_count > 1) {
5524 			SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n");
5525 			return -EBUSY;
5526 		} else if (snapshot_entry->clone_count == 1) {
5527 			has_one_clone = true;
5528 		}
5529 	}
5530 
5531 	/* Check if someone has this blob open (besides this delete context):
5532 	 * - open_ref = 1 - only this context opened blob, so it is ok to remove it
5533 	 * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot
5534 	 *	and that is ok, because we will update it accordingly */
5535 	if (blob->open_ref <= 2 && has_one_clone) {
5536 		clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
5537 		assert(clone_entry != NULL);
5538 		clone = _spdk_blob_lookup(blob->bs, clone_entry->id);
5539 
5540 		if (blob->open_ref == 2 && clone == NULL) {
5541 			/* Clone is closed and someone else opened this blob */
5542 			SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
5543 			return -EBUSY;
5544 		}
5545 
5546 		*update_clone = true;
5547 		return 0;
5548 	}
5549 
5550 	if (blob->open_ref > 1) {
5551 		SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
5552 		return -EBUSY;
5553 	}
5554 
5555 	assert(has_one_clone == false);
5556 	*update_clone = false;
5557 	return 0;
5558 }
5559 
5560 static void
5561 _spdk_bs_delete_enomem_close_cpl(void *cb_arg, int bserrno)
5562 {
5563 	spdk_bs_sequence_t *seq = cb_arg;
5564 
5565 	spdk_bs_sequence_finish(seq, -ENOMEM);
5566 }
5567 
5568 static void
5569 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
5570 {
5571 	spdk_bs_sequence_t *seq = cb_arg;
5572 	struct delete_snapshot_ctx *ctx;
5573 	bool update_clone = false;
5574 
5575 	if (bserrno != 0) {
5576 		spdk_bs_sequence_finish(seq, bserrno);
5577 		return;
5578 	}
5579 
5580 	_spdk_blob_verify_md_op(blob);
5581 
5582 	ctx = calloc(1, sizeof(*ctx));
5583 	if (ctx == NULL) {
5584 		spdk_blob_close(blob, _spdk_bs_delete_enomem_close_cpl, seq);
5585 		return;
5586 	}
5587 
5588 	ctx->snapshot = blob;
5589 	ctx->cb_fn = _spdk_bs_delete_blob_finish;
5590 	ctx->cb_arg = seq;
5591 
5592 	/* Check if blob can be removed and if it is a snapshot with clone on top of it */
5593 	ctx->bserrno = _spdk_bs_is_blob_deletable(blob, &update_clone);
5594 	if (ctx->bserrno) {
5595 		spdk_blob_close(blob, _spdk_delete_blob_cleanup_finish, ctx);
5596 		return;
5597 	}
5598 
5599 	if (blob->locked_operation_in_progress) {
5600 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress\n");
5601 		ctx->bserrno = -EBUSY;
5602 		spdk_blob_close(blob, _spdk_delete_blob_cleanup_finish, ctx);
5603 		return;
5604 	}
5605 
5606 	blob->locked_operation_in_progress = true;
5607 
5608 	/*
5609 	 * Remove the blob from the blob_store list now, to ensure it does not
5610 	 *  get returned after this point by _spdk_blob_lookup().
5611 	 */
5612 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5613 
5614 	if (update_clone) {
5615 		/* This blob is a snapshot with active clone - update clone first */
5616 		_spdk_update_clone_on_snapshot_deletion(blob, ctx);
5617 	} else {
5618 		/* This blob does not have any clones - just remove it */
5619 		_spdk_bs_blob_list_remove(blob);
5620 		_spdk_bs_delete_blob_finish(seq, blob, 0);
5621 		free(ctx);
5622 	}
5623 }
5624 
5625 void
5626 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5627 		    spdk_blob_op_complete cb_fn, void *cb_arg)
5628 {
5629 	struct spdk_bs_cpl	cpl;
5630 	spdk_bs_sequence_t	*seq;
5631 
5632 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
5633 
5634 	assert(spdk_get_thread() == bs->md_thread);
5635 
5636 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5637 	cpl.u.blob_basic.cb_fn = cb_fn;
5638 	cpl.u.blob_basic.cb_arg = cb_arg;
5639 
5640 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5641 	if (!seq) {
5642 		cb_fn(cb_arg, -ENOMEM);
5643 		return;
5644 	}
5645 
5646 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5647 }
5648 
5649 /* END spdk_bs_delete_blob */
5650 
5651 /* START spdk_bs_open_blob */
5652 
5653 static void
5654 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5655 {
5656 	struct spdk_blob *blob = cb_arg;
5657 
5658 	if (bserrno != 0) {
5659 		_spdk_blob_free(blob);
5660 		seq->cpl.u.blob_handle.blob = NULL;
5661 		spdk_bs_sequence_finish(seq, bserrno);
5662 		return;
5663 	}
5664 
5665 	blob->open_ref++;
5666 
5667 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5668 
5669 	spdk_bs_sequence_finish(seq, bserrno);
5670 }
5671 
5672 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5673 			       struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5674 {
5675 	struct spdk_blob		*blob;
5676 	struct spdk_bs_cpl		cpl;
5677 	struct spdk_blob_open_opts	opts_default;
5678 	spdk_bs_sequence_t		*seq;
5679 	uint32_t			page_num;
5680 
5681 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5682 	assert(spdk_get_thread() == bs->md_thread);
5683 
5684 	page_num = _spdk_bs_blobid_to_page(blobid);
5685 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5686 		/* Invalid blobid */
5687 		cb_fn(cb_arg, NULL, -ENOENT);
5688 		return;
5689 	}
5690 
5691 	blob = _spdk_blob_lookup(bs, blobid);
5692 	if (blob) {
5693 		blob->open_ref++;
5694 		cb_fn(cb_arg, blob, 0);
5695 		return;
5696 	}
5697 
5698 	blob = _spdk_blob_alloc(bs, blobid);
5699 	if (!blob) {
5700 		cb_fn(cb_arg, NULL, -ENOMEM);
5701 		return;
5702 	}
5703 
5704 	if (!opts) {
5705 		spdk_blob_open_opts_init(&opts_default);
5706 		opts = &opts_default;
5707 	}
5708 
5709 	blob->clear_method = opts->clear_method;
5710 
5711 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5712 	cpl.u.blob_handle.cb_fn = cb_fn;
5713 	cpl.u.blob_handle.cb_arg = cb_arg;
5714 	cpl.u.blob_handle.blob = blob;
5715 
5716 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5717 	if (!seq) {
5718 		_spdk_blob_free(blob);
5719 		cb_fn(cb_arg, NULL, -ENOMEM);
5720 		return;
5721 	}
5722 
5723 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5724 }
5725 
5726 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5727 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5728 {
5729 	_spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
5730 }
5731 
5732 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
5733 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5734 {
5735 	_spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
5736 }
5737 
5738 /* END spdk_bs_open_blob */
5739 
5740 /* START spdk_blob_set_read_only */
5741 int spdk_blob_set_read_only(struct spdk_blob *blob)
5742 {
5743 	_spdk_blob_verify_md_op(blob);
5744 
5745 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5746 
5747 	blob->state = SPDK_BLOB_STATE_DIRTY;
5748 	return 0;
5749 }
5750 /* END spdk_blob_set_read_only */
5751 
5752 /* START spdk_blob_sync_md */
5753 
5754 static void
5755 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5756 {
5757 	struct spdk_blob *blob = cb_arg;
5758 
5759 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5760 		blob->data_ro = true;
5761 		blob->md_ro = true;
5762 	}
5763 
5764 	spdk_bs_sequence_finish(seq, bserrno);
5765 }
5766 
5767 static void
5768 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5769 {
5770 	struct spdk_bs_cpl	cpl;
5771 	spdk_bs_sequence_t	*seq;
5772 
5773 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5774 	cpl.u.blob_basic.cb_fn = cb_fn;
5775 	cpl.u.blob_basic.cb_arg = cb_arg;
5776 
5777 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5778 	if (!seq) {
5779 		cb_fn(cb_arg, -ENOMEM);
5780 		return;
5781 	}
5782 
5783 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5784 }
5785 
5786 void
5787 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5788 {
5789 	_spdk_blob_verify_md_op(blob);
5790 
5791 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5792 
5793 	if (blob->md_ro) {
5794 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5795 		cb_fn(cb_arg, 0);
5796 		return;
5797 	}
5798 
5799 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5800 }
5801 
5802 /* END spdk_blob_sync_md */
5803 
5804 struct spdk_blob_insert_cluster_ctx {
5805 	struct spdk_thread	*thread;
5806 	struct spdk_blob	*blob;
5807 	uint32_t		cluster_num;	/* cluster index in blob */
5808 	uint32_t		cluster;	/* cluster on disk */
5809 	int			rc;
5810 	spdk_blob_op_complete	cb_fn;
5811 	void			*cb_arg;
5812 };
5813 
5814 static void
5815 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5816 {
5817 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5818 
5819 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5820 	free(ctx);
5821 }
5822 
5823 static void
5824 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5825 {
5826 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5827 
5828 	ctx->rc = bserrno;
5829 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5830 }
5831 
5832 static void
5833 _spdk_blob_insert_cluster_msg(void *arg)
5834 {
5835 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5836 
5837 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5838 	if (ctx->rc != 0) {
5839 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5840 		return;
5841 	}
5842 
5843 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5844 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5845 }
5846 
5847 static void
5848 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5849 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5850 {
5851 	struct spdk_blob_insert_cluster_ctx *ctx;
5852 
5853 	ctx = calloc(1, sizeof(*ctx));
5854 	if (ctx == NULL) {
5855 		cb_fn(cb_arg, -ENOMEM);
5856 		return;
5857 	}
5858 
5859 	ctx->thread = spdk_get_thread();
5860 	ctx->blob = blob;
5861 	ctx->cluster_num = cluster_num;
5862 	ctx->cluster = cluster;
5863 	ctx->cb_fn = cb_fn;
5864 	ctx->cb_arg = cb_arg;
5865 
5866 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5867 }
5868 
5869 /* START spdk_blob_close */
5870 
5871 static void
5872 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5873 {
5874 	struct spdk_blob *blob = cb_arg;
5875 
5876 	if (bserrno == 0) {
5877 		blob->open_ref--;
5878 		if (blob->open_ref == 0) {
5879 			/*
5880 			 * Blobs with active.num_pages == 0 are deleted blobs.
5881 			 *  these blobs are removed from the blob_store list
5882 			 *  when the deletion process starts - so don't try to
5883 			 *  remove them again.
5884 			 */
5885 			if (blob->active.num_pages > 0) {
5886 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5887 			}
5888 			_spdk_blob_free(blob);
5889 		}
5890 	}
5891 
5892 	spdk_bs_sequence_finish(seq, bserrno);
5893 }
5894 
5895 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5896 {
5897 	struct spdk_bs_cpl	cpl;
5898 	spdk_bs_sequence_t	*seq;
5899 
5900 	_spdk_blob_verify_md_op(blob);
5901 
5902 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5903 
5904 	if (blob->open_ref == 0) {
5905 		cb_fn(cb_arg, -EBADF);
5906 		return;
5907 	}
5908 
5909 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5910 	cpl.u.blob_basic.cb_fn = cb_fn;
5911 	cpl.u.blob_basic.cb_arg = cb_arg;
5912 
5913 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5914 	if (!seq) {
5915 		cb_fn(cb_arg, -ENOMEM);
5916 		return;
5917 	}
5918 
5919 	/* Sync metadata */
5920 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5921 }
5922 
5923 /* END spdk_blob_close */
5924 
5925 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5926 {
5927 	return spdk_get_io_channel(bs);
5928 }
5929 
5930 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5931 {
5932 	spdk_put_io_channel(channel);
5933 }
5934 
5935 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5936 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5937 {
5938 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5939 				     SPDK_BLOB_UNMAP);
5940 }
5941 
5942 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5943 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5944 {
5945 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5946 				     SPDK_BLOB_WRITE_ZEROES);
5947 }
5948 
5949 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5950 			void *payload, uint64_t offset, uint64_t length,
5951 			spdk_blob_op_complete cb_fn, void *cb_arg)
5952 {
5953 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5954 				     SPDK_BLOB_WRITE);
5955 }
5956 
5957 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5958 		       void *payload, uint64_t offset, uint64_t length,
5959 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5960 {
5961 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5962 				     SPDK_BLOB_READ);
5963 }
5964 
5965 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5966 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5967 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5968 {
5969 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5970 }
5971 
5972 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5973 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5974 			spdk_blob_op_complete cb_fn, void *cb_arg)
5975 {
5976 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5977 }
5978 
5979 struct spdk_bs_iter_ctx {
5980 	int64_t page_num;
5981 	struct spdk_blob_store *bs;
5982 
5983 	spdk_blob_op_with_handle_complete cb_fn;
5984 	void *cb_arg;
5985 };
5986 
5987 static void
5988 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5989 {
5990 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5991 	struct spdk_blob_store *bs = ctx->bs;
5992 	spdk_blob_id id;
5993 
5994 	if (bserrno == 0) {
5995 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5996 		free(ctx);
5997 		return;
5998 	}
5999 
6000 	ctx->page_num++;
6001 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
6002 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
6003 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
6004 		free(ctx);
6005 		return;
6006 	}
6007 
6008 	id = _spdk_bs_page_to_blobid(ctx->page_num);
6009 
6010 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
6011 }
6012 
6013 void
6014 spdk_bs_iter_first(struct spdk_blob_store *bs,
6015 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6016 {
6017 	struct spdk_bs_iter_ctx *ctx;
6018 
6019 	ctx = calloc(1, sizeof(*ctx));
6020 	if (!ctx) {
6021 		cb_fn(cb_arg, NULL, -ENOMEM);
6022 		return;
6023 	}
6024 
6025 	ctx->page_num = -1;
6026 	ctx->bs = bs;
6027 	ctx->cb_fn = cb_fn;
6028 	ctx->cb_arg = cb_arg;
6029 
6030 	_spdk_bs_iter_cpl(ctx, NULL, -1);
6031 }
6032 
6033 static void
6034 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
6035 {
6036 	struct spdk_bs_iter_ctx *ctx = cb_arg;
6037 
6038 	_spdk_bs_iter_cpl(ctx, NULL, -1);
6039 }
6040 
6041 void
6042 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
6043 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6044 {
6045 	struct spdk_bs_iter_ctx *ctx;
6046 
6047 	assert(blob != NULL);
6048 
6049 	ctx = calloc(1, sizeof(*ctx));
6050 	if (!ctx) {
6051 		cb_fn(cb_arg, NULL, -ENOMEM);
6052 		return;
6053 	}
6054 
6055 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
6056 	ctx->bs = bs;
6057 	ctx->cb_fn = cb_fn;
6058 	ctx->cb_arg = cb_arg;
6059 
6060 	/* Close the existing blob */
6061 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
6062 }
6063 
6064 static int
6065 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
6066 		     uint16_t value_len, bool internal)
6067 {
6068 	struct spdk_xattr_tailq *xattrs;
6069 	struct spdk_xattr	*xattr;
6070 	size_t			desc_size;
6071 
6072 	_spdk_blob_verify_md_op(blob);
6073 
6074 	if (blob->md_ro) {
6075 		return -EPERM;
6076 	}
6077 
6078 	desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len;
6079 	if (desc_size > SPDK_BS_MAX_DESC_SIZE) {
6080 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Xattr '%s' of size %ld does not fix into single page %ld\n", name,
6081 			      desc_size, SPDK_BS_MAX_DESC_SIZE);
6082 		return -ENOMEM;
6083 	}
6084 
6085 	if (internal) {
6086 		xattrs = &blob->xattrs_internal;
6087 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
6088 	} else {
6089 		xattrs = &blob->xattrs;
6090 	}
6091 
6092 	TAILQ_FOREACH(xattr, xattrs, link) {
6093 		if (!strcmp(name, xattr->name)) {
6094 			free(xattr->value);
6095 			xattr->value_len = value_len;
6096 			xattr->value = malloc(value_len);
6097 			memcpy(xattr->value, value, value_len);
6098 
6099 			blob->state = SPDK_BLOB_STATE_DIRTY;
6100 
6101 			return 0;
6102 		}
6103 	}
6104 
6105 	xattr = calloc(1, sizeof(*xattr));
6106 	if (!xattr) {
6107 		return -ENOMEM;
6108 	}
6109 	xattr->name = strdup(name);
6110 	xattr->value_len = value_len;
6111 	xattr->value = malloc(value_len);
6112 	memcpy(xattr->value, value, value_len);
6113 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
6114 
6115 	blob->state = SPDK_BLOB_STATE_DIRTY;
6116 
6117 	return 0;
6118 }
6119 
6120 int
6121 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
6122 		    uint16_t value_len)
6123 {
6124 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
6125 }
6126 
6127 static int
6128 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
6129 {
6130 	struct spdk_xattr_tailq *xattrs;
6131 	struct spdk_xattr	*xattr;
6132 
6133 	_spdk_blob_verify_md_op(blob);
6134 
6135 	if (blob->md_ro) {
6136 		return -EPERM;
6137 	}
6138 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
6139 
6140 	TAILQ_FOREACH(xattr, xattrs, link) {
6141 		if (!strcmp(name, xattr->name)) {
6142 			TAILQ_REMOVE(xattrs, xattr, link);
6143 			free(xattr->value);
6144 			free(xattr->name);
6145 			free(xattr);
6146 
6147 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
6148 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
6149 			}
6150 			blob->state = SPDK_BLOB_STATE_DIRTY;
6151 
6152 			return 0;
6153 		}
6154 	}
6155 
6156 	return -ENOENT;
6157 }
6158 
6159 int
6160 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
6161 {
6162 	return _spdk_blob_remove_xattr(blob, name, false);
6163 }
6164 
6165 static int
6166 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
6167 			   const void **value, size_t *value_len, bool internal)
6168 {
6169 	struct spdk_xattr	*xattr;
6170 	struct spdk_xattr_tailq *xattrs;
6171 
6172 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
6173 
6174 	TAILQ_FOREACH(xattr, xattrs, link) {
6175 		if (!strcmp(name, xattr->name)) {
6176 			*value = xattr->value;
6177 			*value_len = xattr->value_len;
6178 			return 0;
6179 		}
6180 	}
6181 	return -ENOENT;
6182 }
6183 
6184 int
6185 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
6186 			  const void **value, size_t *value_len)
6187 {
6188 	_spdk_blob_verify_md_op(blob);
6189 
6190 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
6191 }
6192 
6193 struct spdk_xattr_names {
6194 	uint32_t	count;
6195 	const char	*names[0];
6196 };
6197 
6198 static int
6199 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
6200 {
6201 	struct spdk_xattr	*xattr;
6202 	int			count = 0;
6203 
6204 	TAILQ_FOREACH(xattr, xattrs, link) {
6205 		count++;
6206 	}
6207 
6208 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
6209 	if (*names == NULL) {
6210 		return -ENOMEM;
6211 	}
6212 
6213 	TAILQ_FOREACH(xattr, xattrs, link) {
6214 		(*names)->names[(*names)->count++] = xattr->name;
6215 	}
6216 
6217 	return 0;
6218 }
6219 
6220 int
6221 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
6222 {
6223 	_spdk_blob_verify_md_op(blob);
6224 
6225 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
6226 }
6227 
6228 uint32_t
6229 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
6230 {
6231 	assert(names != NULL);
6232 
6233 	return names->count;
6234 }
6235 
6236 const char *
6237 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
6238 {
6239 	if (index >= names->count) {
6240 		return NULL;
6241 	}
6242 
6243 	return names->names[index];
6244 }
6245 
6246 void
6247 spdk_xattr_names_free(struct spdk_xattr_names *names)
6248 {
6249 	free(names);
6250 }
6251 
6252 struct spdk_bs_type
6253 spdk_bs_get_bstype(struct spdk_blob_store *bs)
6254 {
6255 	return bs->bstype;
6256 }
6257 
6258 void
6259 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
6260 {
6261 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
6262 }
6263 
6264 bool
6265 spdk_blob_is_read_only(struct spdk_blob *blob)
6266 {
6267 	assert(blob != NULL);
6268 	return (blob->data_ro || blob->md_ro);
6269 }
6270 
6271 bool
6272 spdk_blob_is_snapshot(struct spdk_blob *blob)
6273 {
6274 	struct spdk_blob_list *snapshot_entry;
6275 
6276 	assert(blob != NULL);
6277 
6278 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
6279 	if (snapshot_entry == NULL) {
6280 		return false;
6281 	}
6282 
6283 	return true;
6284 }
6285 
6286 bool
6287 spdk_blob_is_clone(struct spdk_blob *blob)
6288 {
6289 	assert(blob != NULL);
6290 
6291 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
6292 		assert(spdk_blob_is_thin_provisioned(blob));
6293 		return true;
6294 	}
6295 
6296 	return false;
6297 }
6298 
6299 bool
6300 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
6301 {
6302 	assert(blob != NULL);
6303 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
6304 }
6305 
6306 spdk_blob_id
6307 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
6308 {
6309 	struct spdk_blob_list *snapshot_entry = NULL;
6310 	struct spdk_blob_list *clone_entry = NULL;
6311 
6312 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
6313 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
6314 			if (clone_entry->id == blob_id) {
6315 				return snapshot_entry->id;
6316 			}
6317 		}
6318 	}
6319 
6320 	return SPDK_BLOBID_INVALID;
6321 }
6322 
6323 int
6324 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
6325 		     size_t *count)
6326 {
6327 	struct spdk_blob_list *snapshot_entry, *clone_entry;
6328 	size_t n;
6329 
6330 	snapshot_entry = _spdk_bs_get_snapshot_entry(bs, blobid);
6331 	if (snapshot_entry == NULL) {
6332 		*count = 0;
6333 		return 0;
6334 	}
6335 
6336 	if (ids == NULL || *count < snapshot_entry->clone_count) {
6337 		*count = snapshot_entry->clone_count;
6338 		return -ENOMEM;
6339 	}
6340 	*count = snapshot_entry->clone_count;
6341 
6342 	n = 0;
6343 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
6344 		ids[n++] = clone_entry->id;
6345 	}
6346 
6347 	return 0;
6348 }
6349 
6350 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
6351