xref: /spdk/lib/blob/blobstore.c (revision ff4fd58b95445d5264080ceb6e9583a03f145879)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *   Copyright (c) 2021-2022, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
12  *     * Redistributions of source code must retain the above copyright
13  *       notice, this list of conditions and the following disclaimer.
14  *     * Redistributions in binary form must reproduce the above copyright
15  *       notice, this list of conditions and the following disclaimer in
16  *       the documentation and/or other materials provided with the
17  *       distribution.
18  *     * Neither the name of Intel Corporation nor the names of its
19  *       contributors may be used to endorse or promote products derived
20  *       from this software without specific prior written permission.
21  *
22  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 
35 #include "spdk/stdinc.h"
36 
37 #include "spdk/blob.h"
38 #include "spdk/crc32.h"
39 #include "spdk/env.h"
40 #include "spdk/queue.h"
41 #include "spdk/thread.h"
42 #include "spdk/bit_array.h"
43 #include "spdk/bit_pool.h"
44 #include "spdk/likely.h"
45 #include "spdk/util.h"
46 #include "spdk/string.h"
47 
48 #include "spdk_internal/assert.h"
49 #include "spdk/log.h"
50 
51 #include "blobstore.h"
52 
53 #define BLOB_CRC32C_INITIAL    0xffffffffUL
54 
55 static int bs_register_md_thread(struct spdk_blob_store *bs);
56 static int bs_unregister_md_thread(struct spdk_blob_store *bs);
57 static void blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
58 static void blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
59 		uint64_t cluster, uint32_t extent, spdk_blob_op_complete cb_fn, void *cb_arg);
60 
61 static int blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
62 			  uint16_t value_len, bool internal);
63 static int blob_get_xattr_value(struct spdk_blob *blob, const char *name,
64 				const void **value, size_t *value_len, bool internal);
65 static int blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal);
66 
67 static void blob_write_extent_page(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num,
68 				   spdk_blob_op_complete cb_fn, void *cb_arg);
69 
70 static int
71 blob_id_cmp(struct spdk_blob *blob1, struct spdk_blob *blob2)
72 {
73 	return (blob1->id < blob2->id ? -1 : blob1->id > blob2->id);
74 }
75 
76 RB_GENERATE_STATIC(spdk_blob_tree, spdk_blob, link, blob_id_cmp);
77 
78 static void
79 blob_verify_md_op(struct spdk_blob *blob)
80 {
81 	assert(blob != NULL);
82 	assert(spdk_get_thread() == blob->bs->md_thread);
83 	assert(blob->state != SPDK_BLOB_STATE_LOADING);
84 }
85 
86 static struct spdk_blob_list *
87 bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid)
88 {
89 	struct spdk_blob_list *snapshot_entry = NULL;
90 
91 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
92 		if (snapshot_entry->id == blobid) {
93 			break;
94 		}
95 	}
96 
97 	return snapshot_entry;
98 }
99 
100 static void
101 bs_claim_md_page(struct spdk_blob_store *bs, uint32_t page)
102 {
103 	assert(page < spdk_bit_array_capacity(bs->used_md_pages));
104 	assert(spdk_bit_array_get(bs->used_md_pages, page) == false);
105 
106 	spdk_bit_array_set(bs->used_md_pages, page);
107 }
108 
109 static void
110 bs_release_md_page(struct spdk_blob_store *bs, uint32_t page)
111 {
112 	assert(page < spdk_bit_array_capacity(bs->used_md_pages));
113 	assert(spdk_bit_array_get(bs->used_md_pages, page) == true);
114 
115 	spdk_bit_array_clear(bs->used_md_pages, page);
116 }
117 
118 static uint32_t
119 bs_claim_cluster(struct spdk_blob_store *bs)
120 {
121 	uint32_t cluster_num;
122 
123 	cluster_num = spdk_bit_pool_allocate_bit(bs->used_clusters);
124 	if (cluster_num == UINT32_MAX) {
125 		return UINT32_MAX;
126 	}
127 
128 	SPDK_DEBUGLOG(blob, "Claiming cluster %u\n", cluster_num);
129 	bs->num_free_clusters--;
130 
131 	return cluster_num;
132 }
133 
134 static void
135 bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
136 {
137 	assert(cluster_num < spdk_bit_pool_capacity(bs->used_clusters));
138 	assert(spdk_bit_pool_is_allocated(bs->used_clusters, cluster_num) == true);
139 	assert(bs->num_free_clusters < bs->total_clusters);
140 
141 	SPDK_DEBUGLOG(blob, "Releasing cluster %u\n", cluster_num);
142 
143 	spdk_bit_pool_free_bit(bs->used_clusters, cluster_num);
144 	bs->num_free_clusters++;
145 }
146 
147 static int
148 blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
149 {
150 	uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
151 
152 	blob_verify_md_op(blob);
153 
154 	if (*cluster_lba != 0) {
155 		return -EEXIST;
156 	}
157 
158 	*cluster_lba = bs_cluster_to_lba(blob->bs, cluster);
159 	return 0;
160 }
161 
162 static int
163 bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
164 		    uint64_t *cluster, uint32_t *lowest_free_md_page, bool update_map)
165 {
166 	uint32_t *extent_page = 0;
167 
168 	*cluster = bs_claim_cluster(blob->bs);
169 	if (*cluster == UINT32_MAX) {
170 		/* No more free clusters. Cannot satisfy the request */
171 		return -ENOSPC;
172 	}
173 
174 	if (blob->use_extent_table) {
175 		extent_page = bs_cluster_to_extent_page(blob, cluster_num);
176 		if (*extent_page == 0) {
177 			/* Extent page shall never occupy md_page so start the search from 1 */
178 			if (*lowest_free_md_page == 0) {
179 				*lowest_free_md_page = 1;
180 			}
181 			/* No extent_page is allocated for the cluster */
182 			*lowest_free_md_page = spdk_bit_array_find_first_clear(blob->bs->used_md_pages,
183 					       *lowest_free_md_page);
184 			if (*lowest_free_md_page == UINT32_MAX) {
185 				/* No more free md pages. Cannot satisfy the request */
186 				bs_release_cluster(blob->bs, *cluster);
187 				return -ENOSPC;
188 			}
189 			bs_claim_md_page(blob->bs, *lowest_free_md_page);
190 		}
191 	}
192 
193 	SPDK_DEBUGLOG(blob, "Claiming cluster %" PRIu64 " for blob %" PRIu64 "\n", *cluster, blob->id);
194 
195 	if (update_map) {
196 		blob_insert_cluster(blob, cluster_num, *cluster);
197 		if (blob->use_extent_table && *extent_page == 0) {
198 			*extent_page = *lowest_free_md_page;
199 		}
200 	}
201 
202 	return 0;
203 }
204 
205 static void
206 blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
207 {
208 	xattrs->count = 0;
209 	xattrs->names = NULL;
210 	xattrs->ctx = NULL;
211 	xattrs->get_value = NULL;
212 }
213 
214 void
215 spdk_blob_opts_init(struct spdk_blob_opts *opts, size_t opts_size)
216 {
217 	if (!opts) {
218 		SPDK_ERRLOG("opts should not be NULL\n");
219 		return;
220 	}
221 
222 	if (!opts_size) {
223 		SPDK_ERRLOG("opts_size should not be zero value\n");
224 		return;
225 	}
226 
227 	memset(opts, 0, opts_size);
228 	opts->opts_size = opts_size;
229 
230 #define FIELD_OK(field) \
231         offsetof(struct spdk_blob_opts, field) + sizeof(opts->field) <= opts_size
232 
233 #define SET_FIELD(field, value) \
234         if (FIELD_OK(field)) { \
235                 opts->field = value; \
236         } \
237 
238 	SET_FIELD(num_clusters, 0);
239 	SET_FIELD(thin_provision, false);
240 	SET_FIELD(clear_method, BLOB_CLEAR_WITH_DEFAULT);
241 
242 	if (FIELD_OK(xattrs)) {
243 		blob_xattrs_init(&opts->xattrs);
244 	}
245 
246 	SET_FIELD(use_extent_table, true);
247 
248 #undef FIELD_OK
249 #undef SET_FIELD
250 }
251 
252 void
253 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts, size_t opts_size)
254 {
255 	if (!opts) {
256 		SPDK_ERRLOG("opts should not be NULL\n");
257 		return;
258 	}
259 
260 	if (!opts_size) {
261 		SPDK_ERRLOG("opts_size should not be zero value\n");
262 		return;
263 	}
264 
265 	memset(opts, 0, opts_size);
266 	opts->opts_size = opts_size;
267 
268 #define FIELD_OK(field) \
269         offsetof(struct spdk_blob_open_opts, field) + sizeof(opts->field) <= opts_size
270 
271 #define SET_FIELD(field, value) \
272         if (FIELD_OK(field)) { \
273                 opts->field = value; \
274         } \
275 
276 	SET_FIELD(clear_method, BLOB_CLEAR_WITH_DEFAULT);
277 
278 #undef FIELD_OK
279 #undef SET_FILED
280 }
281 
282 static struct spdk_blob *
283 blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
284 {
285 	struct spdk_blob *blob;
286 
287 	blob = calloc(1, sizeof(*blob));
288 	if (!blob) {
289 		return NULL;
290 	}
291 
292 	blob->id = id;
293 	blob->bs = bs;
294 
295 	blob->parent_id = SPDK_BLOBID_INVALID;
296 
297 	blob->state = SPDK_BLOB_STATE_DIRTY;
298 	blob->extent_rle_found = false;
299 	blob->extent_table_found = false;
300 	blob->active.num_pages = 1;
301 	blob->active.pages = calloc(1, sizeof(*blob->active.pages));
302 	if (!blob->active.pages) {
303 		free(blob);
304 		return NULL;
305 	}
306 
307 	blob->active.pages[0] = bs_blobid_to_page(id);
308 
309 	TAILQ_INIT(&blob->xattrs);
310 	TAILQ_INIT(&blob->xattrs_internal);
311 	TAILQ_INIT(&blob->pending_persists);
312 	TAILQ_INIT(&blob->persists_to_complete);
313 
314 	return blob;
315 }
316 
317 static void
318 xattrs_free(struct spdk_xattr_tailq *xattrs)
319 {
320 	struct spdk_xattr	*xattr, *xattr_tmp;
321 
322 	TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
323 		TAILQ_REMOVE(xattrs, xattr, link);
324 		free(xattr->name);
325 		free(xattr->value);
326 		free(xattr);
327 	}
328 }
329 
330 static void
331 blob_free(struct spdk_blob *blob)
332 {
333 	assert(blob != NULL);
334 	assert(TAILQ_EMPTY(&blob->pending_persists));
335 	assert(TAILQ_EMPTY(&blob->persists_to_complete));
336 
337 	free(blob->active.extent_pages);
338 	free(blob->clean.extent_pages);
339 	free(blob->active.clusters);
340 	free(blob->clean.clusters);
341 	free(blob->active.pages);
342 	free(blob->clean.pages);
343 
344 	xattrs_free(&blob->xattrs);
345 	xattrs_free(&blob->xattrs_internal);
346 
347 	if (blob->back_bs_dev) {
348 		blob->back_bs_dev->destroy(blob->back_bs_dev);
349 	}
350 
351 	free(blob);
352 }
353 
354 struct freeze_io_ctx {
355 	struct spdk_bs_cpl cpl;
356 	struct spdk_blob *blob;
357 };
358 
359 static void
360 blob_io_sync(struct spdk_io_channel_iter *i)
361 {
362 	spdk_for_each_channel_continue(i, 0);
363 }
364 
365 static void
366 blob_execute_queued_io(struct spdk_io_channel_iter *i)
367 {
368 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
369 	struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
370 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
371 	struct spdk_bs_request_set	*set;
372 	struct spdk_bs_user_op_args	*args;
373 	spdk_bs_user_op_t *op, *tmp;
374 
375 	TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
376 		set = (struct spdk_bs_request_set *)op;
377 		args = &set->u.user_op;
378 
379 		if (args->blob == ctx->blob) {
380 			TAILQ_REMOVE(&ch->queued_io, op, link);
381 			bs_user_op_execute(op);
382 		}
383 	}
384 
385 	spdk_for_each_channel_continue(i, 0);
386 }
387 
388 static void
389 blob_io_cpl(struct spdk_io_channel_iter *i, int status)
390 {
391 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
392 
393 	ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
394 
395 	free(ctx);
396 }
397 
398 static void
399 blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
400 {
401 	struct freeze_io_ctx *ctx;
402 
403 	ctx = calloc(1, sizeof(*ctx));
404 	if (!ctx) {
405 		cb_fn(cb_arg, -ENOMEM);
406 		return;
407 	}
408 
409 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
410 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
411 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
412 	ctx->blob = blob;
413 
414 	/* Freeze I/O on blob */
415 	blob->frozen_refcnt++;
416 
417 	if (blob->frozen_refcnt == 1) {
418 		spdk_for_each_channel(blob->bs, blob_io_sync, ctx, blob_io_cpl);
419 	} else {
420 		cb_fn(cb_arg, 0);
421 		free(ctx);
422 	}
423 }
424 
425 static void
426 blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
427 {
428 	struct freeze_io_ctx *ctx;
429 
430 	ctx = calloc(1, sizeof(*ctx));
431 	if (!ctx) {
432 		cb_fn(cb_arg, -ENOMEM);
433 		return;
434 	}
435 
436 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
437 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
438 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
439 	ctx->blob = blob;
440 
441 	assert(blob->frozen_refcnt > 0);
442 
443 	blob->frozen_refcnt--;
444 
445 	if (blob->frozen_refcnt == 0) {
446 		spdk_for_each_channel(blob->bs, blob_execute_queued_io, ctx, blob_io_cpl);
447 	} else {
448 		cb_fn(cb_arg, 0);
449 		free(ctx);
450 	}
451 }
452 
453 static int
454 blob_mark_clean(struct spdk_blob *blob)
455 {
456 	uint32_t *extent_pages = NULL;
457 	uint64_t *clusters = NULL;
458 	uint32_t *pages = NULL;
459 
460 	assert(blob != NULL);
461 
462 	if (blob->active.num_extent_pages) {
463 		assert(blob->active.extent_pages);
464 		extent_pages = calloc(blob->active.num_extent_pages, sizeof(*blob->active.extent_pages));
465 		if (!extent_pages) {
466 			return -ENOMEM;
467 		}
468 		memcpy(extent_pages, blob->active.extent_pages,
469 		       blob->active.num_extent_pages * sizeof(*extent_pages));
470 	}
471 
472 	if (blob->active.num_clusters) {
473 		assert(blob->active.clusters);
474 		clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
475 		if (!clusters) {
476 			free(extent_pages);
477 			return -ENOMEM;
478 		}
479 		memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters));
480 	}
481 
482 	if (blob->active.num_pages) {
483 		assert(blob->active.pages);
484 		pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
485 		if (!pages) {
486 			free(extent_pages);
487 			free(clusters);
488 			return -ENOMEM;
489 		}
490 		memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
491 	}
492 
493 	free(blob->clean.extent_pages);
494 	free(blob->clean.clusters);
495 	free(blob->clean.pages);
496 
497 	blob->clean.num_extent_pages = blob->active.num_extent_pages;
498 	blob->clean.extent_pages = blob->active.extent_pages;
499 	blob->clean.num_clusters = blob->active.num_clusters;
500 	blob->clean.clusters = blob->active.clusters;
501 	blob->clean.num_pages = blob->active.num_pages;
502 	blob->clean.pages = blob->active.pages;
503 
504 	blob->active.extent_pages = extent_pages;
505 	blob->active.clusters = clusters;
506 	blob->active.pages = pages;
507 
508 	/* If the metadata was dirtied again while the metadata was being written to disk,
509 	 *  we do not want to revert the DIRTY state back to CLEAN here.
510 	 */
511 	if (blob->state == SPDK_BLOB_STATE_LOADING) {
512 		blob->state = SPDK_BLOB_STATE_CLEAN;
513 	}
514 
515 	return 0;
516 }
517 
518 static int
519 blob_deserialize_xattr(struct spdk_blob *blob,
520 		       struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
521 {
522 	struct spdk_xattr                       *xattr;
523 
524 	if (desc_xattr->length != sizeof(desc_xattr->name_length) +
525 	    sizeof(desc_xattr->value_length) +
526 	    desc_xattr->name_length + desc_xattr->value_length) {
527 		return -EINVAL;
528 	}
529 
530 	xattr = calloc(1, sizeof(*xattr));
531 	if (xattr == NULL) {
532 		return -ENOMEM;
533 	}
534 
535 	xattr->name = malloc(desc_xattr->name_length + 1);
536 	if (xattr->name == NULL) {
537 		free(xattr);
538 		return -ENOMEM;
539 	}
540 	memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
541 	xattr->name[desc_xattr->name_length] = '\0';
542 
543 	xattr->value = malloc(desc_xattr->value_length);
544 	if (xattr->value == NULL) {
545 		free(xattr->name);
546 		free(xattr);
547 		return -ENOMEM;
548 	}
549 	xattr->value_len = desc_xattr->value_length;
550 	memcpy(xattr->value,
551 	       (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
552 	       desc_xattr->value_length);
553 
554 	TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
555 
556 	return 0;
557 }
558 
559 
560 static int
561 blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
562 {
563 	struct spdk_blob_md_descriptor *desc;
564 	size_t	cur_desc = 0;
565 	void *tmp;
566 
567 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
568 	while (cur_desc < sizeof(page->descriptors)) {
569 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
570 			if (desc->length == 0) {
571 				/* If padding and length are 0, this terminates the page */
572 				break;
573 			}
574 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
575 			struct spdk_blob_md_descriptor_flags	*desc_flags;
576 
577 			desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
578 
579 			if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
580 				return -EINVAL;
581 			}
582 
583 			if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
584 			    SPDK_BLOB_INVALID_FLAGS_MASK) {
585 				return -EINVAL;
586 			}
587 
588 			if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
589 			    SPDK_BLOB_DATA_RO_FLAGS_MASK) {
590 				blob->data_ro = true;
591 				blob->md_ro = true;
592 			}
593 
594 			if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
595 			    SPDK_BLOB_MD_RO_FLAGS_MASK) {
596 				blob->md_ro = true;
597 			}
598 
599 			if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
600 				blob->data_ro = true;
601 				blob->md_ro = true;
602 			}
603 
604 			blob->invalid_flags = desc_flags->invalid_flags;
605 			blob->data_ro_flags = desc_flags->data_ro_flags;
606 			blob->md_ro_flags = desc_flags->md_ro_flags;
607 
608 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
609 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
610 			unsigned int				i, j;
611 			unsigned int				cluster_count = blob->active.num_clusters;
612 
613 			if (blob->extent_table_found) {
614 				/* Extent Table already present in the md,
615 				 * both descriptors should never be at the same time. */
616 				return -EINVAL;
617 			}
618 			blob->extent_rle_found = true;
619 
620 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
621 
622 			if (desc_extent_rle->length == 0 ||
623 			    (desc_extent_rle->length % sizeof(desc_extent_rle->extents[0]) != 0)) {
624 				return -EINVAL;
625 			}
626 
627 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
628 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
629 					if (desc_extent_rle->extents[i].cluster_idx != 0) {
630 						if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters,
631 										desc_extent_rle->extents[i].cluster_idx + j)) {
632 							return -EINVAL;
633 						}
634 					}
635 					cluster_count++;
636 				}
637 			}
638 
639 			if (cluster_count == 0) {
640 				return -EINVAL;
641 			}
642 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(*blob->active.clusters));
643 			if (tmp == NULL) {
644 				return -ENOMEM;
645 			}
646 			blob->active.clusters = tmp;
647 			blob->active.cluster_array_size = cluster_count;
648 
649 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
650 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
651 					if (desc_extent_rle->extents[i].cluster_idx != 0) {
652 						blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs,
653 								desc_extent_rle->extents[i].cluster_idx + j);
654 					} else if (spdk_blob_is_thin_provisioned(blob)) {
655 						blob->active.clusters[blob->active.num_clusters++] = 0;
656 					} else {
657 						return -EINVAL;
658 					}
659 				}
660 			}
661 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
662 			struct spdk_blob_md_descriptor_extent_table *desc_extent_table;
663 			uint32_t num_extent_pages = blob->active.num_extent_pages;
664 			uint32_t i, j;
665 			size_t extent_pages_length;
666 
667 			desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc;
668 			extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters);
669 
670 			if (blob->extent_rle_found) {
671 				/* This means that Extent RLE is present in MD,
672 				 * both should never be at the same time. */
673 				return -EINVAL;
674 			} else if (blob->extent_table_found &&
675 				   desc_extent_table->num_clusters != blob->remaining_clusters_in_et) {
676 				/* Number of clusters in this ET does not match number
677 				 * from previously read EXTENT_TABLE. */
678 				return -EINVAL;
679 			}
680 
681 			if (desc_extent_table->length == 0 ||
682 			    (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) {
683 				return -EINVAL;
684 			}
685 
686 			blob->extent_table_found = true;
687 
688 			for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
689 				num_extent_pages += desc_extent_table->extent_page[i].num_pages;
690 			}
691 
692 			if (num_extent_pages > 0) {
693 				tmp = realloc(blob->active.extent_pages, num_extent_pages * sizeof(uint32_t));
694 				if (tmp == NULL) {
695 					return -ENOMEM;
696 				}
697 				blob->active.extent_pages = tmp;
698 			}
699 			blob->active.extent_pages_array_size = num_extent_pages;
700 
701 			blob->remaining_clusters_in_et = desc_extent_table->num_clusters;
702 
703 			/* Extent table entries contain md page numbers for extent pages.
704 			 * Zeroes represent unallocated extent pages, those are run-length-encoded.
705 			 */
706 			for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
707 				if (desc_extent_table->extent_page[i].page_idx != 0) {
708 					assert(desc_extent_table->extent_page[i].num_pages == 1);
709 					blob->active.extent_pages[blob->active.num_extent_pages++] =
710 						desc_extent_table->extent_page[i].page_idx;
711 				} else if (spdk_blob_is_thin_provisioned(blob)) {
712 					for (j = 0; j < desc_extent_table->extent_page[i].num_pages; j++) {
713 						blob->active.extent_pages[blob->active.num_extent_pages++] = 0;
714 					}
715 				} else {
716 					return -EINVAL;
717 				}
718 			}
719 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
720 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
721 			unsigned int					i;
722 			unsigned int					cluster_count = 0;
723 			size_t						cluster_idx_length;
724 
725 			if (blob->extent_rle_found) {
726 				/* This means that Extent RLE is present in MD,
727 				 * both should never be at the same time. */
728 				return -EINVAL;
729 			}
730 
731 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
732 			cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx);
733 
734 			if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) ||
735 			    (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) {
736 				return -EINVAL;
737 			}
738 
739 			for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
740 				if (desc_extent->cluster_idx[i] != 0) {
741 					if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters, desc_extent->cluster_idx[i])) {
742 						return -EINVAL;
743 					}
744 				}
745 				cluster_count++;
746 			}
747 
748 			if (cluster_count == 0) {
749 				return -EINVAL;
750 			}
751 
752 			/* When reading extent pages sequentially starting cluster idx should match
753 			 * current size of a blob.
754 			 * If changed to batch reading, this check shall be removed. */
755 			if (desc_extent->start_cluster_idx != blob->active.num_clusters) {
756 				return -EINVAL;
757 			}
758 
759 			tmp = realloc(blob->active.clusters,
760 				      (cluster_count + blob->active.num_clusters) * sizeof(*blob->active.clusters));
761 			if (tmp == NULL) {
762 				return -ENOMEM;
763 			}
764 			blob->active.clusters = tmp;
765 			blob->active.cluster_array_size = (cluster_count + blob->active.num_clusters);
766 
767 			for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
768 				if (desc_extent->cluster_idx[i] != 0) {
769 					blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs,
770 							desc_extent->cluster_idx[i]);
771 				} else if (spdk_blob_is_thin_provisioned(blob)) {
772 					blob->active.clusters[blob->active.num_clusters++] = 0;
773 				} else {
774 					return -EINVAL;
775 				}
776 			}
777 			assert(desc_extent->start_cluster_idx + cluster_count == blob->active.num_clusters);
778 			assert(blob->remaining_clusters_in_et >= cluster_count);
779 			blob->remaining_clusters_in_et -= cluster_count;
780 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
781 			int rc;
782 
783 			rc = blob_deserialize_xattr(blob,
784 						    (struct spdk_blob_md_descriptor_xattr *) desc, false);
785 			if (rc != 0) {
786 				return rc;
787 			}
788 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
789 			int rc;
790 
791 			rc = blob_deserialize_xattr(blob,
792 						    (struct spdk_blob_md_descriptor_xattr *) desc, true);
793 			if (rc != 0) {
794 				return rc;
795 			}
796 		} else {
797 			/* Unrecognized descriptor type.  Do not fail - just continue to the
798 			 *  next descriptor.  If this descriptor is associated with some feature
799 			 *  defined in a newer version of blobstore, that version of blobstore
800 			 *  should create and set an associated feature flag to specify if this
801 			 *  blob can be loaded or not.
802 			 */
803 		}
804 
805 		/* Advance to the next descriptor */
806 		cur_desc += sizeof(*desc) + desc->length;
807 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
808 			break;
809 		}
810 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
811 	}
812 
813 	return 0;
814 }
815 
816 static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page);
817 
818 static int
819 blob_parse_extent_page(struct spdk_blob_md_page *extent_page, struct spdk_blob *blob)
820 {
821 	assert(blob != NULL);
822 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
823 
824 	if (bs_load_cur_extent_page_valid(extent_page) == false) {
825 		return -ENOENT;
826 	}
827 
828 	return blob_parse_page(extent_page, blob);
829 }
830 
831 static int
832 blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
833 	   struct spdk_blob *blob)
834 {
835 	const struct spdk_blob_md_page *page;
836 	uint32_t i;
837 	int rc;
838 	void *tmp;
839 
840 	assert(page_count > 0);
841 	assert(pages[0].sequence_num == 0);
842 	assert(blob != NULL);
843 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
844 	assert(blob->active.clusters == NULL);
845 
846 	/* The blobid provided doesn't match what's in the MD, this can
847 	 * happen for example if a bogus blobid is passed in through open.
848 	 */
849 	if (blob->id != pages[0].id) {
850 		SPDK_ERRLOG("Blobid (%" PRIu64 ") doesn't match what's in metadata (%" PRIu64 ")\n",
851 			    blob->id, pages[0].id);
852 		return -ENOENT;
853 	}
854 
855 	tmp = realloc(blob->active.pages, page_count * sizeof(*blob->active.pages));
856 	if (!tmp) {
857 		return -ENOMEM;
858 	}
859 	blob->active.pages = tmp;
860 
861 	blob->active.pages[0] = pages[0].id;
862 
863 	for (i = 1; i < page_count; i++) {
864 		assert(spdk_bit_array_get(blob->bs->used_md_pages, pages[i - 1].next));
865 		blob->active.pages[i] = pages[i - 1].next;
866 	}
867 	blob->active.num_pages = page_count;
868 
869 	for (i = 0; i < page_count; i++) {
870 		page = &pages[i];
871 
872 		assert(page->id == blob->id);
873 		assert(page->sequence_num == i);
874 
875 		rc = blob_parse_page(page, blob);
876 		if (rc != 0) {
877 			return rc;
878 		}
879 	}
880 
881 	return 0;
882 }
883 
884 static int
885 blob_serialize_add_page(const struct spdk_blob *blob,
886 			struct spdk_blob_md_page **pages,
887 			uint32_t *page_count,
888 			struct spdk_blob_md_page **last_page)
889 {
890 	struct spdk_blob_md_page *page, *tmp_pages;
891 
892 	assert(pages != NULL);
893 	assert(page_count != NULL);
894 
895 	*last_page = NULL;
896 	if (*page_count == 0) {
897 		assert(*pages == NULL);
898 		*pages = spdk_malloc(SPDK_BS_PAGE_SIZE, 0,
899 				     NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
900 		if (*pages == NULL) {
901 			return -ENOMEM;
902 		}
903 		*page_count = 1;
904 	} else {
905 		assert(*pages != NULL);
906 		tmp_pages = spdk_realloc(*pages, SPDK_BS_PAGE_SIZE * (*page_count + 1), 0);
907 		if (tmp_pages == NULL) {
908 			return -ENOMEM;
909 		}
910 		(*page_count)++;
911 		*pages = tmp_pages;
912 	}
913 
914 	page = &(*pages)[*page_count - 1];
915 	memset(page, 0, sizeof(*page));
916 	page->id = blob->id;
917 	page->sequence_num = *page_count - 1;
918 	page->next = SPDK_INVALID_MD_PAGE;
919 	*last_page = page;
920 
921 	return 0;
922 }
923 
924 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
925  * Update required_sz on both success and failure.
926  *
927  */
928 static int
929 blob_serialize_xattr(const struct spdk_xattr *xattr,
930 		     uint8_t *buf, size_t buf_sz,
931 		     size_t *required_sz, bool internal)
932 {
933 	struct spdk_blob_md_descriptor_xattr	*desc;
934 
935 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
936 		       strlen(xattr->name) +
937 		       xattr->value_len;
938 
939 	if (buf_sz < *required_sz) {
940 		return -1;
941 	}
942 
943 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
944 
945 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
946 	desc->length = sizeof(desc->name_length) +
947 		       sizeof(desc->value_length) +
948 		       strlen(xattr->name) +
949 		       xattr->value_len;
950 	desc->name_length = strlen(xattr->name);
951 	desc->value_length = xattr->value_len;
952 
953 	memcpy(desc->name, xattr->name, desc->name_length);
954 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
955 	       xattr->value,
956 	       desc->value_length);
957 
958 	return 0;
959 }
960 
961 static void
962 blob_serialize_extent_table_entry(const struct spdk_blob *blob,
963 				  uint64_t start_ep, uint64_t *next_ep,
964 				  uint8_t **buf, size_t *remaining_sz)
965 {
966 	struct spdk_blob_md_descriptor_extent_table *desc;
967 	size_t cur_sz;
968 	uint64_t i, et_idx;
969 	uint32_t extent_page, ep_len;
970 
971 	/* The buffer must have room for at least num_clusters entry */
972 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->num_clusters);
973 	if (*remaining_sz < cur_sz) {
974 		*next_ep = start_ep;
975 		return;
976 	}
977 
978 	desc = (struct spdk_blob_md_descriptor_extent_table *)*buf;
979 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE;
980 
981 	desc->num_clusters = blob->active.num_clusters;
982 
983 	ep_len = 1;
984 	et_idx = 0;
985 	for (i = start_ep; i < blob->active.num_extent_pages; i++) {
986 		if (*remaining_sz < cur_sz  + sizeof(desc->extent_page[0])) {
987 			/* If we ran out of buffer space, return */
988 			break;
989 		}
990 
991 		extent_page = blob->active.extent_pages[i];
992 		/* Verify that next extent_page is unallocated */
993 		if (extent_page == 0 &&
994 		    (i + 1 < blob->active.num_extent_pages && blob->active.extent_pages[i + 1] == 0)) {
995 			ep_len++;
996 			continue;
997 		}
998 		desc->extent_page[et_idx].page_idx = extent_page;
999 		desc->extent_page[et_idx].num_pages = ep_len;
1000 		et_idx++;
1001 
1002 		ep_len = 1;
1003 		cur_sz += sizeof(desc->extent_page[et_idx]);
1004 	}
1005 	*next_ep = i;
1006 
1007 	desc->length = sizeof(desc->num_clusters) + sizeof(desc->extent_page[0]) * et_idx;
1008 	*remaining_sz -= sizeof(struct spdk_blob_md_descriptor) + desc->length;
1009 	*buf += sizeof(struct spdk_blob_md_descriptor) + desc->length;
1010 }
1011 
1012 static int
1013 blob_serialize_extent_table(const struct spdk_blob *blob,
1014 			    struct spdk_blob_md_page **pages,
1015 			    struct spdk_blob_md_page *cur_page,
1016 			    uint32_t *page_count, uint8_t **buf,
1017 			    size_t *remaining_sz)
1018 {
1019 	uint64_t				last_extent_page;
1020 	int					rc;
1021 
1022 	last_extent_page = 0;
1023 	/* At least single extent table entry has to be always persisted.
1024 	 * Such case occurs with num_extent_pages == 0. */
1025 	while (last_extent_page <= blob->active.num_extent_pages) {
1026 		blob_serialize_extent_table_entry(blob, last_extent_page, &last_extent_page, buf,
1027 						  remaining_sz);
1028 
1029 		if (last_extent_page == blob->active.num_extent_pages) {
1030 			break;
1031 		}
1032 
1033 		rc = blob_serialize_add_page(blob, pages, page_count, &cur_page);
1034 		if (rc < 0) {
1035 			return rc;
1036 		}
1037 
1038 		*buf = (uint8_t *)cur_page->descriptors;
1039 		*remaining_sz = sizeof(cur_page->descriptors);
1040 	}
1041 
1042 	return 0;
1043 }
1044 
1045 static void
1046 blob_serialize_extent_rle(const struct spdk_blob *blob,
1047 			  uint64_t start_cluster, uint64_t *next_cluster,
1048 			  uint8_t **buf, size_t *buf_sz)
1049 {
1050 	struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle;
1051 	size_t cur_sz;
1052 	uint64_t i, extent_idx;
1053 	uint64_t lba, lba_per_cluster, lba_count;
1054 
1055 	/* The buffer must have room for at least one extent */
1056 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc_extent_rle->extents[0]);
1057 	if (*buf_sz < cur_sz) {
1058 		*next_cluster = start_cluster;
1059 		return;
1060 	}
1061 
1062 	desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)*buf;
1063 	desc_extent_rle->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE;
1064 
1065 	lba_per_cluster = bs_cluster_to_lba(blob->bs, 1);
1066 
1067 	lba = blob->active.clusters[start_cluster];
1068 	lba_count = lba_per_cluster;
1069 	extent_idx = 0;
1070 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
1071 		if ((lba + lba_count) == blob->active.clusters[i] && lba != 0) {
1072 			/* Run-length encode sequential non-zero LBA */
1073 			lba_count += lba_per_cluster;
1074 			continue;
1075 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
1076 			/* Run-length encode unallocated clusters */
1077 			lba_count += lba_per_cluster;
1078 			continue;
1079 		}
1080 		desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
1081 		desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
1082 		extent_idx++;
1083 
1084 		cur_sz += sizeof(desc_extent_rle->extents[extent_idx]);
1085 
1086 		if (*buf_sz < cur_sz) {
1087 			/* If we ran out of buffer space, return */
1088 			*next_cluster = i;
1089 			break;
1090 		}
1091 
1092 		lba = blob->active.clusters[i];
1093 		lba_count = lba_per_cluster;
1094 	}
1095 
1096 	if (*buf_sz >= cur_sz) {
1097 		desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
1098 		desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
1099 		extent_idx++;
1100 
1101 		*next_cluster = blob->active.num_clusters;
1102 	}
1103 
1104 	desc_extent_rle->length = sizeof(desc_extent_rle->extents[0]) * extent_idx;
1105 	*buf_sz -= sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
1106 	*buf += sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
1107 }
1108 
1109 static int
1110 blob_serialize_extents_rle(const struct spdk_blob *blob,
1111 			   struct spdk_blob_md_page **pages,
1112 			   struct spdk_blob_md_page *cur_page,
1113 			   uint32_t *page_count, uint8_t **buf,
1114 			   size_t *remaining_sz)
1115 {
1116 	uint64_t				last_cluster;
1117 	int					rc;
1118 
1119 	last_cluster = 0;
1120 	while (last_cluster < blob->active.num_clusters) {
1121 		blob_serialize_extent_rle(blob, last_cluster, &last_cluster, buf, remaining_sz);
1122 
1123 		if (last_cluster == blob->active.num_clusters) {
1124 			break;
1125 		}
1126 
1127 		rc = blob_serialize_add_page(blob, pages, page_count, &cur_page);
1128 		if (rc < 0) {
1129 			return rc;
1130 		}
1131 
1132 		*buf = (uint8_t *)cur_page->descriptors;
1133 		*remaining_sz = sizeof(cur_page->descriptors);
1134 	}
1135 
1136 	return 0;
1137 }
1138 
1139 static void
1140 blob_serialize_extent_page(const struct spdk_blob *blob,
1141 			   uint64_t cluster, struct spdk_blob_md_page *page)
1142 {
1143 	struct spdk_blob_md_descriptor_extent_page *desc_extent;
1144 	uint64_t i, extent_idx;
1145 	uint64_t lba, lba_per_cluster;
1146 	uint64_t start_cluster_idx = (cluster / SPDK_EXTENTS_PER_EP) * SPDK_EXTENTS_PER_EP;
1147 
1148 	desc_extent = (struct spdk_blob_md_descriptor_extent_page *) page->descriptors;
1149 	desc_extent->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE;
1150 
1151 	lba_per_cluster = bs_cluster_to_lba(blob->bs, 1);
1152 
1153 	desc_extent->start_cluster_idx = start_cluster_idx;
1154 	extent_idx = 0;
1155 	for (i = start_cluster_idx; i < blob->active.num_clusters; i++) {
1156 		lba = blob->active.clusters[i];
1157 		desc_extent->cluster_idx[extent_idx++] = lba / lba_per_cluster;
1158 		if (extent_idx >= SPDK_EXTENTS_PER_EP) {
1159 			break;
1160 		}
1161 	}
1162 	desc_extent->length = sizeof(desc_extent->start_cluster_idx) +
1163 			      sizeof(desc_extent->cluster_idx[0]) * extent_idx;
1164 }
1165 
1166 static void
1167 blob_serialize_flags(const struct spdk_blob *blob,
1168 		     uint8_t *buf, size_t *buf_sz)
1169 {
1170 	struct spdk_blob_md_descriptor_flags *desc;
1171 
1172 	/*
1173 	 * Flags get serialized first, so we should always have room for the flags
1174 	 *  descriptor.
1175 	 */
1176 	assert(*buf_sz >= sizeof(*desc));
1177 
1178 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
1179 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
1180 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
1181 	desc->invalid_flags = blob->invalid_flags;
1182 	desc->data_ro_flags = blob->data_ro_flags;
1183 	desc->md_ro_flags = blob->md_ro_flags;
1184 
1185 	*buf_sz -= sizeof(*desc);
1186 }
1187 
1188 static int
1189 blob_serialize_xattrs(const struct spdk_blob *blob,
1190 		      const struct spdk_xattr_tailq *xattrs, bool internal,
1191 		      struct spdk_blob_md_page **pages,
1192 		      struct spdk_blob_md_page *cur_page,
1193 		      uint32_t *page_count, uint8_t **buf,
1194 		      size_t *remaining_sz)
1195 {
1196 	const struct spdk_xattr	*xattr;
1197 	int	rc;
1198 
1199 	TAILQ_FOREACH(xattr, xattrs, link) {
1200 		size_t required_sz = 0;
1201 
1202 		rc = blob_serialize_xattr(xattr,
1203 					  *buf, *remaining_sz,
1204 					  &required_sz, internal);
1205 		if (rc < 0) {
1206 			/* Need to add a new page to the chain */
1207 			rc = blob_serialize_add_page(blob, pages, page_count,
1208 						     &cur_page);
1209 			if (rc < 0) {
1210 				spdk_free(*pages);
1211 				*pages = NULL;
1212 				*page_count = 0;
1213 				return rc;
1214 			}
1215 
1216 			*buf = (uint8_t *)cur_page->descriptors;
1217 			*remaining_sz = sizeof(cur_page->descriptors);
1218 
1219 			/* Try again */
1220 			required_sz = 0;
1221 			rc = blob_serialize_xattr(xattr,
1222 						  *buf, *remaining_sz,
1223 						  &required_sz, internal);
1224 
1225 			if (rc < 0) {
1226 				spdk_free(*pages);
1227 				*pages = NULL;
1228 				*page_count = 0;
1229 				return rc;
1230 			}
1231 		}
1232 
1233 		*remaining_sz -= required_sz;
1234 		*buf += required_sz;
1235 	}
1236 
1237 	return 0;
1238 }
1239 
1240 static int
1241 blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
1242 	       uint32_t *page_count)
1243 {
1244 	struct spdk_blob_md_page		*cur_page;
1245 	int					rc;
1246 	uint8_t					*buf;
1247 	size_t					remaining_sz;
1248 
1249 	assert(pages != NULL);
1250 	assert(page_count != NULL);
1251 	assert(blob != NULL);
1252 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
1253 
1254 	*pages = NULL;
1255 	*page_count = 0;
1256 
1257 	/* A blob always has at least 1 page, even if it has no descriptors */
1258 	rc = blob_serialize_add_page(blob, pages, page_count, &cur_page);
1259 	if (rc < 0) {
1260 		return rc;
1261 	}
1262 
1263 	buf = (uint8_t *)cur_page->descriptors;
1264 	remaining_sz = sizeof(cur_page->descriptors);
1265 
1266 	/* Serialize flags */
1267 	blob_serialize_flags(blob, buf, &remaining_sz);
1268 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
1269 
1270 	/* Serialize xattrs */
1271 	rc = blob_serialize_xattrs(blob, &blob->xattrs, false,
1272 				   pages, cur_page, page_count, &buf, &remaining_sz);
1273 	if (rc < 0) {
1274 		return rc;
1275 	}
1276 
1277 	/* Serialize internal xattrs */
1278 	rc = blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
1279 				   pages, cur_page, page_count, &buf, &remaining_sz);
1280 	if (rc < 0) {
1281 		return rc;
1282 	}
1283 
1284 	if (blob->use_extent_table) {
1285 		/* Serialize extent table */
1286 		rc = blob_serialize_extent_table(blob, pages, cur_page, page_count, &buf, &remaining_sz);
1287 	} else {
1288 		/* Serialize extents */
1289 		rc = blob_serialize_extents_rle(blob, pages, cur_page, page_count, &buf, &remaining_sz);
1290 	}
1291 
1292 	return rc;
1293 }
1294 
1295 struct spdk_blob_load_ctx {
1296 	struct spdk_blob		*blob;
1297 
1298 	struct spdk_blob_md_page	*pages;
1299 	uint32_t			num_pages;
1300 	uint32_t			next_extent_page;
1301 	spdk_bs_sequence_t	        *seq;
1302 
1303 	spdk_bs_sequence_cpl		cb_fn;
1304 	void				*cb_arg;
1305 };
1306 
1307 static uint32_t
1308 blob_md_page_calc_crc(void *page)
1309 {
1310 	uint32_t		crc;
1311 
1312 	crc = BLOB_CRC32C_INITIAL;
1313 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
1314 	crc ^= BLOB_CRC32C_INITIAL;
1315 
1316 	return crc;
1317 
1318 }
1319 
1320 static void
1321 blob_load_final(struct spdk_blob_load_ctx *ctx, int bserrno)
1322 {
1323 	struct spdk_blob		*blob = ctx->blob;
1324 
1325 	if (bserrno == 0) {
1326 		blob_mark_clean(blob);
1327 	}
1328 
1329 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
1330 
1331 	/* Free the memory */
1332 	spdk_free(ctx->pages);
1333 	free(ctx);
1334 }
1335 
1336 static void
1337 blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
1338 {
1339 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1340 	struct spdk_blob		*blob = ctx->blob;
1341 
1342 	if (bserrno == 0) {
1343 		blob->back_bs_dev = bs_create_blob_bs_dev(snapshot);
1344 		if (blob->back_bs_dev == NULL) {
1345 			bserrno = -ENOMEM;
1346 		}
1347 	}
1348 	if (bserrno != 0) {
1349 		SPDK_ERRLOG("Snapshot fail\n");
1350 	}
1351 
1352 	blob_load_final(ctx, bserrno);
1353 }
1354 
1355 static void blob_update_clear_method(struct spdk_blob *blob);
1356 
1357 static void
1358 blob_load_backing_dev(void *cb_arg)
1359 {
1360 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1361 	struct spdk_blob		*blob = ctx->blob;
1362 	const void			*value;
1363 	size_t				len;
1364 	int				rc;
1365 
1366 	if (spdk_blob_is_thin_provisioned(blob)) {
1367 		rc = blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
1368 		if (rc == 0) {
1369 			if (len != sizeof(spdk_blob_id)) {
1370 				blob_load_final(ctx, -EINVAL);
1371 				return;
1372 			}
1373 			/* open snapshot blob and continue in the callback function */
1374 			blob->parent_id = *(spdk_blob_id *)value;
1375 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1376 					  blob_load_snapshot_cpl, ctx);
1377 			return;
1378 		} else {
1379 			/* add zeroes_dev for thin provisioned blob */
1380 			blob->back_bs_dev = bs_create_zeroes_dev();
1381 		}
1382 	} else {
1383 		/* standard blob */
1384 		blob->back_bs_dev = NULL;
1385 	}
1386 	blob_load_final(ctx, 0);
1387 }
1388 
1389 static void
1390 blob_load_cpl_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1391 {
1392 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1393 	struct spdk_blob		*blob = ctx->blob;
1394 	struct spdk_blob_md_page	*page;
1395 	uint64_t			i;
1396 	uint32_t			crc;
1397 	uint64_t			lba;
1398 	void				*tmp;
1399 	uint64_t			sz;
1400 
1401 	if (bserrno) {
1402 		SPDK_ERRLOG("Extent page read failed: %d\n", bserrno);
1403 		blob_load_final(ctx, bserrno);
1404 		return;
1405 	}
1406 
1407 	if (ctx->pages == NULL) {
1408 		/* First iteration of this function, allocate buffer for single EXTENT_PAGE */
1409 		ctx->pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0,
1410 					  NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
1411 		if (!ctx->pages) {
1412 			blob_load_final(ctx, -ENOMEM);
1413 			return;
1414 		}
1415 		ctx->num_pages = 1;
1416 		ctx->next_extent_page = 0;
1417 	} else {
1418 		page = &ctx->pages[0];
1419 		crc = blob_md_page_calc_crc(page);
1420 		if (crc != page->crc) {
1421 			blob_load_final(ctx, -EINVAL);
1422 			return;
1423 		}
1424 
1425 		if (page->next != SPDK_INVALID_MD_PAGE) {
1426 			blob_load_final(ctx, -EINVAL);
1427 			return;
1428 		}
1429 
1430 		bserrno = blob_parse_extent_page(page, blob);
1431 		if (bserrno) {
1432 			blob_load_final(ctx, bserrno);
1433 			return;
1434 		}
1435 	}
1436 
1437 	for (i = ctx->next_extent_page; i < blob->active.num_extent_pages; i++) {
1438 		if (blob->active.extent_pages[i] != 0) {
1439 			/* Extent page was allocated, read and parse it. */
1440 			lba = bs_md_page_to_lba(blob->bs, blob->active.extent_pages[i]);
1441 			ctx->next_extent_page = i + 1;
1442 
1443 			bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1444 					     bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE),
1445 					     blob_load_cpl_extents_cpl, ctx);
1446 			return;
1447 		} else {
1448 			/* Thin provisioned blobs can point to unallocated extent pages.
1449 			 * In this case blob size should be increased by up to the amount left in remaining_clusters_in_et. */
1450 
1451 			sz = spdk_min(blob->remaining_clusters_in_et, SPDK_EXTENTS_PER_EP);
1452 			blob->active.num_clusters += sz;
1453 			blob->remaining_clusters_in_et -= sz;
1454 
1455 			assert(spdk_blob_is_thin_provisioned(blob));
1456 			assert(i + 1 < blob->active.num_extent_pages || blob->remaining_clusters_in_et == 0);
1457 
1458 			tmp = realloc(blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters));
1459 			if (tmp == NULL) {
1460 				blob_load_final(ctx, -ENOMEM);
1461 				return;
1462 			}
1463 			memset(tmp + sizeof(*blob->active.clusters) * blob->active.cluster_array_size, 0,
1464 			       sizeof(*blob->active.clusters) * (blob->active.num_clusters - blob->active.cluster_array_size));
1465 			blob->active.clusters = tmp;
1466 			blob->active.cluster_array_size = blob->active.num_clusters;
1467 		}
1468 	}
1469 
1470 	blob_load_backing_dev(ctx);
1471 }
1472 
1473 static void
1474 blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1475 {
1476 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1477 	struct spdk_blob		*blob = ctx->blob;
1478 	struct spdk_blob_md_page	*page;
1479 	int				rc;
1480 	uint32_t			crc;
1481 	uint32_t			current_page;
1482 
1483 	if (ctx->num_pages == 1) {
1484 		current_page = bs_blobid_to_page(blob->id);
1485 	} else {
1486 		assert(ctx->num_pages != 0);
1487 		page = &ctx->pages[ctx->num_pages - 2];
1488 		current_page = page->next;
1489 	}
1490 
1491 	if (bserrno) {
1492 		SPDK_ERRLOG("Metadata page %d read failed for blobid %" PRIu64 ": %d\n",
1493 			    current_page, blob->id, bserrno);
1494 		blob_load_final(ctx, bserrno);
1495 		return;
1496 	}
1497 
1498 	page = &ctx->pages[ctx->num_pages - 1];
1499 	crc = blob_md_page_calc_crc(page);
1500 	if (crc != page->crc) {
1501 		SPDK_ERRLOG("Metadata page %d crc mismatch for blobid %" PRIu64 "\n",
1502 			    current_page, blob->id);
1503 		blob_load_final(ctx, -EINVAL);
1504 		return;
1505 	}
1506 
1507 	if (page->next != SPDK_INVALID_MD_PAGE) {
1508 		struct spdk_blob_md_page *tmp_pages;
1509 		uint32_t next_page = page->next;
1510 		uint64_t next_lba = bs_md_page_to_lba(blob->bs, next_page);
1511 
1512 		/* Read the next page */
1513 		tmp_pages = spdk_realloc(ctx->pages, (sizeof(*page) * (ctx->num_pages + 1)), 0);
1514 		if (tmp_pages == NULL) {
1515 			blob_load_final(ctx, -ENOMEM);
1516 			return;
1517 		}
1518 		ctx->num_pages++;
1519 		ctx->pages = tmp_pages;
1520 
1521 		bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
1522 				     next_lba,
1523 				     bs_byte_to_lba(blob->bs, sizeof(*page)),
1524 				     blob_load_cpl, ctx);
1525 		return;
1526 	}
1527 
1528 	/* Parse the pages */
1529 	rc = blob_parse(ctx->pages, ctx->num_pages, blob);
1530 	if (rc) {
1531 		blob_load_final(ctx, rc);
1532 		return;
1533 	}
1534 
1535 	if (blob->extent_table_found == true) {
1536 		/* If EXTENT_TABLE was found, that means support for it should be enabled. */
1537 		assert(blob->extent_rle_found == false);
1538 		blob->use_extent_table = true;
1539 	} else {
1540 		/* If EXTENT_RLE or no extent_* descriptor was found disable support
1541 		 * for extent table. No extent_* descriptors means that blob has length of 0
1542 		 * and no extent_rle descriptors were persisted for it.
1543 		 * EXTENT_TABLE if used, is always present in metadata regardless of length. */
1544 		blob->use_extent_table = false;
1545 	}
1546 
1547 	/* Check the clear_method stored in metadata vs what may have been passed
1548 	 * via spdk_bs_open_blob_ext() and update accordingly.
1549 	 */
1550 	blob_update_clear_method(blob);
1551 
1552 	spdk_free(ctx->pages);
1553 	ctx->pages = NULL;
1554 
1555 	if (blob->extent_table_found) {
1556 		blob_load_cpl_extents_cpl(seq, ctx, 0);
1557 	} else {
1558 		blob_load_backing_dev(ctx);
1559 	}
1560 }
1561 
1562 /* Load a blob from disk given a blobid */
1563 static void
1564 blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1565 	  spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1566 {
1567 	struct spdk_blob_load_ctx *ctx;
1568 	struct spdk_blob_store *bs;
1569 	uint32_t page_num;
1570 	uint64_t lba;
1571 
1572 	blob_verify_md_op(blob);
1573 
1574 	bs = blob->bs;
1575 
1576 	ctx = calloc(1, sizeof(*ctx));
1577 	if (!ctx) {
1578 		cb_fn(seq, cb_arg, -ENOMEM);
1579 		return;
1580 	}
1581 
1582 	ctx->blob = blob;
1583 	ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, 0);
1584 	if (!ctx->pages) {
1585 		free(ctx);
1586 		cb_fn(seq, cb_arg, -ENOMEM);
1587 		return;
1588 	}
1589 	ctx->num_pages = 1;
1590 	ctx->cb_fn = cb_fn;
1591 	ctx->cb_arg = cb_arg;
1592 	ctx->seq = seq;
1593 
1594 	page_num = bs_blobid_to_page(blob->id);
1595 	lba = bs_md_page_to_lba(blob->bs, page_num);
1596 
1597 	blob->state = SPDK_BLOB_STATE_LOADING;
1598 
1599 	bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1600 			     bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1601 			     blob_load_cpl, ctx);
1602 }
1603 
1604 struct spdk_blob_persist_ctx {
1605 	struct spdk_blob		*blob;
1606 
1607 	struct spdk_bs_super_block	*super;
1608 
1609 	struct spdk_blob_md_page	*pages;
1610 	uint32_t			next_extent_page;
1611 	struct spdk_blob_md_page	*extent_page;
1612 
1613 	spdk_bs_sequence_t		*seq;
1614 	spdk_bs_sequence_cpl		cb_fn;
1615 	void				*cb_arg;
1616 	TAILQ_ENTRY(spdk_blob_persist_ctx) link;
1617 };
1618 
1619 static void
1620 bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1621 		   uint64_t lba_count)
1622 {
1623 	switch (ctx->blob->clear_method) {
1624 	case BLOB_CLEAR_WITH_DEFAULT:
1625 	case BLOB_CLEAR_WITH_UNMAP:
1626 		bs_batch_unmap_dev(batch, lba, lba_count);
1627 		break;
1628 	case BLOB_CLEAR_WITH_WRITE_ZEROES:
1629 		bs_batch_write_zeroes_dev(batch, lba, lba_count);
1630 		break;
1631 	case BLOB_CLEAR_WITH_NONE:
1632 	default:
1633 		break;
1634 	}
1635 }
1636 
1637 static void blob_persist_check_dirty(struct spdk_blob_persist_ctx *ctx);
1638 
1639 static void
1640 blob_persist_complete_cb(void *arg)
1641 {
1642 	struct spdk_blob_persist_ctx *ctx = arg;
1643 
1644 	/* Call user callback */
1645 	ctx->cb_fn(ctx->seq, ctx->cb_arg, 0);
1646 
1647 	/* Free the memory */
1648 	spdk_free(ctx->pages);
1649 	free(ctx);
1650 }
1651 
1652 static void
1653 blob_persist_complete(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx, int bserrno)
1654 {
1655 	struct spdk_blob_persist_ctx	*next_persist, *tmp;
1656 	struct spdk_blob		*blob = ctx->blob;
1657 
1658 	if (bserrno == 0) {
1659 		blob_mark_clean(blob);
1660 	}
1661 
1662 	assert(ctx == TAILQ_FIRST(&blob->persists_to_complete));
1663 
1664 	/* Complete all persists that were pending when the current persist started */
1665 	TAILQ_FOREACH_SAFE(next_persist, &blob->persists_to_complete, link, tmp) {
1666 		TAILQ_REMOVE(&blob->persists_to_complete, next_persist, link);
1667 		spdk_thread_send_msg(spdk_get_thread(), blob_persist_complete_cb, next_persist);
1668 	}
1669 
1670 	if (TAILQ_EMPTY(&blob->pending_persists)) {
1671 		return;
1672 	}
1673 
1674 	/* Queue up all pending persists for completion and start blob persist with first one */
1675 	TAILQ_SWAP(&blob->persists_to_complete, &blob->pending_persists, spdk_blob_persist_ctx, link);
1676 	next_persist = TAILQ_FIRST(&blob->persists_to_complete);
1677 
1678 	blob->state = SPDK_BLOB_STATE_DIRTY;
1679 	blob_persist_check_dirty(next_persist);
1680 }
1681 
1682 static void
1683 blob_persist_clear_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1684 {
1685 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1686 	struct spdk_blob		*blob = ctx->blob;
1687 	struct spdk_blob_store		*bs = blob->bs;
1688 	size_t				i;
1689 
1690 	if (bserrno != 0) {
1691 		blob_persist_complete(seq, ctx, bserrno);
1692 		return;
1693 	}
1694 
1695 	/* Release all extent_pages that were truncated */
1696 	for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) {
1697 		/* Nothing to release if it was not allocated */
1698 		if (blob->active.extent_pages[i] != 0) {
1699 			bs_release_md_page(bs, blob->active.extent_pages[i]);
1700 		}
1701 	}
1702 
1703 	if (blob->active.num_extent_pages == 0) {
1704 		free(blob->active.extent_pages);
1705 		blob->active.extent_pages = NULL;
1706 		blob->active.extent_pages_array_size = 0;
1707 	} else if (blob->active.num_extent_pages != blob->active.extent_pages_array_size) {
1708 #ifndef __clang_analyzer__
1709 		void *tmp;
1710 
1711 		/* scan-build really can't figure reallocs, workaround it */
1712 		tmp = realloc(blob->active.extent_pages, sizeof(uint32_t) * blob->active.num_extent_pages);
1713 		assert(tmp != NULL);
1714 		blob->active.extent_pages = tmp;
1715 #endif
1716 		blob->active.extent_pages_array_size = blob->active.num_extent_pages;
1717 	}
1718 
1719 	blob_persist_complete(seq, ctx, bserrno);
1720 }
1721 
1722 static void
1723 blob_persist_clear_extents(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx)
1724 {
1725 	struct spdk_blob		*blob = ctx->blob;
1726 	struct spdk_blob_store		*bs = blob->bs;
1727 	size_t				i;
1728 	uint64_t                        lba;
1729 	uint64_t                        lba_count;
1730 	spdk_bs_batch_t                 *batch;
1731 
1732 	batch = bs_sequence_to_batch(seq, blob_persist_clear_extents_cpl, ctx);
1733 	lba_count = bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1734 
1735 	/* Clear all extent_pages that were truncated */
1736 	for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) {
1737 		/* Nothing to clear if it was not allocated */
1738 		if (blob->active.extent_pages[i] != 0) {
1739 			lba = bs_md_page_to_lba(bs, blob->active.extent_pages[i]);
1740 			bs_batch_write_zeroes_dev(batch, lba, lba_count);
1741 		}
1742 	}
1743 
1744 	bs_batch_close(batch);
1745 }
1746 
1747 static void
1748 blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1749 {
1750 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1751 	struct spdk_blob		*blob = ctx->blob;
1752 	struct spdk_blob_store		*bs = blob->bs;
1753 	size_t				i;
1754 
1755 	if (bserrno != 0) {
1756 		blob_persist_complete(seq, ctx, bserrno);
1757 		return;
1758 	}
1759 
1760 	pthread_mutex_lock(&bs->used_clusters_mutex);
1761 	/* Release all clusters that were truncated */
1762 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1763 		uint32_t cluster_num = bs_lba_to_cluster(bs, blob->active.clusters[i]);
1764 
1765 		/* Nothing to release if it was not allocated */
1766 		if (blob->active.clusters[i] != 0) {
1767 			bs_release_cluster(bs, cluster_num);
1768 		}
1769 	}
1770 	pthread_mutex_unlock(&bs->used_clusters_mutex);
1771 
1772 	if (blob->active.num_clusters == 0) {
1773 		free(blob->active.clusters);
1774 		blob->active.clusters = NULL;
1775 		blob->active.cluster_array_size = 0;
1776 	} else if (blob->active.num_clusters != blob->active.cluster_array_size) {
1777 #ifndef __clang_analyzer__
1778 		void *tmp;
1779 
1780 		/* scan-build really can't figure reallocs, workaround it */
1781 		tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * blob->active.num_clusters);
1782 		assert(tmp != NULL);
1783 		blob->active.clusters = tmp;
1784 
1785 #endif
1786 		blob->active.cluster_array_size = blob->active.num_clusters;
1787 	}
1788 
1789 	/* Move on to clearing extent pages */
1790 	blob_persist_clear_extents(seq, ctx);
1791 }
1792 
1793 static void
1794 blob_persist_clear_clusters(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx)
1795 {
1796 	struct spdk_blob		*blob = ctx->blob;
1797 	struct spdk_blob_store		*bs = blob->bs;
1798 	spdk_bs_batch_t			*batch;
1799 	size_t				i;
1800 	uint64_t			lba;
1801 	uint64_t			lba_count;
1802 
1803 	/* Clusters don't move around in blobs. The list shrinks or grows
1804 	 * at the end, but no changes ever occur in the middle of the list.
1805 	 */
1806 
1807 	batch = bs_sequence_to_batch(seq, blob_persist_clear_clusters_cpl, ctx);
1808 
1809 	/* Clear all clusters that were truncated */
1810 	lba = 0;
1811 	lba_count = 0;
1812 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1813 		uint64_t next_lba = blob->active.clusters[i];
1814 		uint64_t next_lba_count = bs_cluster_to_lba(bs, 1);
1815 
1816 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1817 			/* This cluster is contiguous with the previous one. */
1818 			lba_count += next_lba_count;
1819 			continue;
1820 		} else if (next_lba == 0) {
1821 			continue;
1822 		}
1823 
1824 		/* This cluster is not contiguous with the previous one. */
1825 
1826 		/* If a run of LBAs previously existing, clear them now */
1827 		if (lba_count > 0) {
1828 			bs_batch_clear_dev(ctx, batch, lba, lba_count);
1829 		}
1830 
1831 		/* Start building the next batch */
1832 		lba = next_lba;
1833 		if (next_lba > 0) {
1834 			lba_count = next_lba_count;
1835 		} else {
1836 			lba_count = 0;
1837 		}
1838 	}
1839 
1840 	/* If we ended with a contiguous set of LBAs, clear them now */
1841 	if (lba_count > 0) {
1842 		bs_batch_clear_dev(ctx, batch, lba, lba_count);
1843 	}
1844 
1845 	bs_batch_close(batch);
1846 }
1847 
1848 static void
1849 blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1850 {
1851 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1852 	struct spdk_blob		*blob = ctx->blob;
1853 	struct spdk_blob_store		*bs = blob->bs;
1854 	size_t				i;
1855 
1856 	if (bserrno != 0) {
1857 		blob_persist_complete(seq, ctx, bserrno);
1858 		return;
1859 	}
1860 
1861 	/* This loop starts at 1 because the first page is special and handled
1862 	 * below. The pages (except the first) are never written in place,
1863 	 * so any pages in the clean list must be zeroed.
1864 	 */
1865 	for (i = 1; i < blob->clean.num_pages; i++) {
1866 		bs_release_md_page(bs, blob->clean.pages[i]);
1867 	}
1868 
1869 	if (blob->active.num_pages == 0) {
1870 		uint32_t page_num;
1871 
1872 		page_num = bs_blobid_to_page(blob->id);
1873 		bs_release_md_page(bs, page_num);
1874 	}
1875 
1876 	/* Move on to clearing clusters */
1877 	blob_persist_clear_clusters(seq, ctx);
1878 }
1879 
1880 static void
1881 blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1882 {
1883 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1884 	struct spdk_blob		*blob = ctx->blob;
1885 	struct spdk_blob_store		*bs = blob->bs;
1886 	uint64_t			lba;
1887 	uint64_t			lba_count;
1888 	spdk_bs_batch_t			*batch;
1889 	size_t				i;
1890 
1891 	if (bserrno != 0) {
1892 		blob_persist_complete(seq, ctx, bserrno);
1893 		return;
1894 	}
1895 
1896 	batch = bs_sequence_to_batch(seq, blob_persist_zero_pages_cpl, ctx);
1897 
1898 	lba_count = bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1899 
1900 	/* This loop starts at 1 because the first page is special and handled
1901 	 * below. The pages (except the first) are never written in place,
1902 	 * so any pages in the clean list must be zeroed.
1903 	 */
1904 	for (i = 1; i < blob->clean.num_pages; i++) {
1905 		lba = bs_md_page_to_lba(bs, blob->clean.pages[i]);
1906 
1907 		bs_batch_write_zeroes_dev(batch, lba, lba_count);
1908 	}
1909 
1910 	/* The first page will only be zeroed if this is a delete. */
1911 	if (blob->active.num_pages == 0) {
1912 		uint32_t page_num;
1913 
1914 		/* The first page in the metadata goes where the blobid indicates */
1915 		page_num = bs_blobid_to_page(blob->id);
1916 		lba = bs_md_page_to_lba(bs, page_num);
1917 
1918 		bs_batch_write_zeroes_dev(batch, lba, lba_count);
1919 	}
1920 
1921 	bs_batch_close(batch);
1922 }
1923 
1924 static void
1925 blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1926 {
1927 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1928 	struct spdk_blob		*blob = ctx->blob;
1929 	struct spdk_blob_store		*bs = blob->bs;
1930 	uint64_t			lba;
1931 	uint32_t			lba_count;
1932 	struct spdk_blob_md_page	*page;
1933 
1934 	if (bserrno != 0) {
1935 		blob_persist_complete(seq, ctx, bserrno);
1936 		return;
1937 	}
1938 
1939 	if (blob->active.num_pages == 0) {
1940 		/* Move on to the next step */
1941 		blob_persist_zero_pages(seq, ctx, 0);
1942 		return;
1943 	}
1944 
1945 	lba_count = bs_byte_to_lba(bs, sizeof(*page));
1946 
1947 	page = &ctx->pages[0];
1948 	/* The first page in the metadata goes where the blobid indicates */
1949 	lba = bs_md_page_to_lba(bs, bs_blobid_to_page(blob->id));
1950 
1951 	bs_sequence_write_dev(seq, page, lba, lba_count,
1952 			      blob_persist_zero_pages, ctx);
1953 }
1954 
1955 static void
1956 blob_persist_write_page_chain(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx)
1957 {
1958 	struct spdk_blob		*blob = ctx->blob;
1959 	struct spdk_blob_store		*bs = blob->bs;
1960 	uint64_t			lba;
1961 	uint32_t			lba_count;
1962 	struct spdk_blob_md_page	*page;
1963 	spdk_bs_batch_t			*batch;
1964 	size_t				i;
1965 
1966 	/* Clusters don't move around in blobs. The list shrinks or grows
1967 	 * at the end, but no changes ever occur in the middle of the list.
1968 	 */
1969 
1970 	lba_count = bs_byte_to_lba(bs, sizeof(*page));
1971 
1972 	batch = bs_sequence_to_batch(seq, blob_persist_write_page_root, ctx);
1973 
1974 	/* This starts at 1. The root page is not written until
1975 	 * all of the others are finished
1976 	 */
1977 	for (i = 1; i < blob->active.num_pages; i++) {
1978 		page = &ctx->pages[i];
1979 		assert(page->sequence_num == i);
1980 
1981 		lba = bs_md_page_to_lba(bs, blob->active.pages[i]);
1982 
1983 		bs_batch_write_dev(batch, page, lba, lba_count);
1984 	}
1985 
1986 	bs_batch_close(batch);
1987 }
1988 
1989 static int
1990 blob_resize(struct spdk_blob *blob, uint64_t sz)
1991 {
1992 	uint64_t	i;
1993 	uint64_t	*tmp;
1994 	uint64_t	cluster;
1995 	uint32_t	lfmd; /*  lowest free md page */
1996 	uint64_t	num_clusters;
1997 	uint32_t	*ep_tmp;
1998 	uint64_t	new_num_ep = 0, current_num_ep = 0;
1999 	struct spdk_blob_store *bs;
2000 
2001 	bs = blob->bs;
2002 
2003 	blob_verify_md_op(blob);
2004 
2005 	if (blob->active.num_clusters == sz) {
2006 		return 0;
2007 	}
2008 
2009 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
2010 		/* If this blob was resized to be larger, then smaller, then
2011 		 * larger without syncing, then the cluster array already
2012 		 * contains spare assigned clusters we can use.
2013 		 */
2014 		num_clusters = spdk_min(blob->active.cluster_array_size,
2015 					sz);
2016 	} else {
2017 		num_clusters = blob->active.num_clusters;
2018 	}
2019 
2020 	if (blob->use_extent_table) {
2021 		/* Round up since every cluster beyond current Extent Table size,
2022 		 * requires new extent page. */
2023 		new_num_ep = spdk_divide_round_up(sz, SPDK_EXTENTS_PER_EP);
2024 		current_num_ep = spdk_divide_round_up(num_clusters, SPDK_EXTENTS_PER_EP);
2025 	}
2026 
2027 	/* Check first that we have enough clusters and md pages before we start claiming them. */
2028 	if (sz > num_clusters && spdk_blob_is_thin_provisioned(blob) == false) {
2029 		if ((sz - num_clusters) > bs->num_free_clusters) {
2030 			return -ENOSPC;
2031 		}
2032 		lfmd = 0;
2033 		for (i = current_num_ep; i < new_num_ep ; i++) {
2034 			lfmd = spdk_bit_array_find_first_clear(blob->bs->used_md_pages, lfmd);
2035 			if (lfmd == UINT32_MAX) {
2036 				/* No more free md pages. Cannot satisfy the request */
2037 				return -ENOSPC;
2038 			}
2039 		}
2040 	}
2041 
2042 	if (sz > num_clusters) {
2043 		/* Expand the cluster array if necessary.
2044 		 * We only shrink the array when persisting.
2045 		 */
2046 		tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * sz);
2047 		if (sz > 0 && tmp == NULL) {
2048 			return -ENOMEM;
2049 		}
2050 		memset(tmp + blob->active.cluster_array_size, 0,
2051 		       sizeof(*blob->active.clusters) * (sz - blob->active.cluster_array_size));
2052 		blob->active.clusters = tmp;
2053 		blob->active.cluster_array_size = sz;
2054 
2055 		/* Expand the extents table, only if enough clusters were added */
2056 		if (new_num_ep > current_num_ep && blob->use_extent_table) {
2057 			ep_tmp = realloc(blob->active.extent_pages, sizeof(*blob->active.extent_pages) * new_num_ep);
2058 			if (new_num_ep > 0 && ep_tmp == NULL) {
2059 				return -ENOMEM;
2060 			}
2061 			memset(ep_tmp + blob->active.extent_pages_array_size, 0,
2062 			       sizeof(*blob->active.extent_pages) * (new_num_ep - blob->active.extent_pages_array_size));
2063 			blob->active.extent_pages = ep_tmp;
2064 			blob->active.extent_pages_array_size = new_num_ep;
2065 		}
2066 	}
2067 
2068 	blob->state = SPDK_BLOB_STATE_DIRTY;
2069 
2070 	if (spdk_blob_is_thin_provisioned(blob) == false) {
2071 		cluster = 0;
2072 		lfmd = 0;
2073 		pthread_mutex_lock(&blob->bs->used_clusters_mutex);
2074 		for (i = num_clusters; i < sz; i++) {
2075 			bs_allocate_cluster(blob, i, &cluster, &lfmd, true);
2076 			lfmd++;
2077 		}
2078 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
2079 	}
2080 
2081 	blob->active.num_clusters = sz;
2082 	blob->active.num_extent_pages = new_num_ep;
2083 
2084 	return 0;
2085 }
2086 
2087 static void
2088 blob_persist_generate_new_md(struct spdk_blob_persist_ctx *ctx)
2089 {
2090 	spdk_bs_sequence_t *seq = ctx->seq;
2091 	struct spdk_blob *blob = ctx->blob;
2092 	struct spdk_blob_store *bs = blob->bs;
2093 	uint64_t i;
2094 	uint32_t page_num;
2095 	void *tmp;
2096 	int rc;
2097 
2098 	/* Generate the new metadata */
2099 	rc = blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
2100 	if (rc < 0) {
2101 		blob_persist_complete(seq, ctx, rc);
2102 		return;
2103 	}
2104 
2105 	assert(blob->active.num_pages >= 1);
2106 
2107 	/* Resize the cache of page indices */
2108 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
2109 	if (!tmp) {
2110 		blob_persist_complete(seq, ctx, -ENOMEM);
2111 		return;
2112 	}
2113 	blob->active.pages = tmp;
2114 
2115 	/* Assign this metadata to pages. This requires two passes -
2116 	 * one to verify that there are enough pages and a second
2117 	 * to actually claim them. */
2118 	page_num = 0;
2119 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
2120 	for (i = 1; i < blob->active.num_pages; i++) {
2121 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
2122 		if (page_num == UINT32_MAX) {
2123 			blob_persist_complete(seq, ctx, -ENOMEM);
2124 			return;
2125 		}
2126 		page_num++;
2127 	}
2128 
2129 	page_num = 0;
2130 	blob->active.pages[0] = bs_blobid_to_page(blob->id);
2131 	for (i = 1; i < blob->active.num_pages; i++) {
2132 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
2133 		ctx->pages[i - 1].next = page_num;
2134 		/* Now that previous metadata page is complete, calculate the crc for it. */
2135 		ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]);
2136 		blob->active.pages[i] = page_num;
2137 		bs_claim_md_page(bs, page_num);
2138 		SPDK_DEBUGLOG(blob, "Claiming page %u for blob %" PRIu64 "\n", page_num, blob->id);
2139 		page_num++;
2140 	}
2141 	ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]);
2142 	/* Start writing the metadata from last page to first */
2143 	blob->state = SPDK_BLOB_STATE_CLEAN;
2144 	blob_persist_write_page_chain(seq, ctx);
2145 }
2146 
2147 static void
2148 blob_persist_write_extent_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2149 {
2150 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
2151 	struct spdk_blob		*blob = ctx->blob;
2152 	size_t				i;
2153 	uint32_t			extent_page_id;
2154 	uint32_t                        page_count = 0;
2155 	int				rc;
2156 
2157 	if (ctx->extent_page != NULL) {
2158 		spdk_free(ctx->extent_page);
2159 		ctx->extent_page = NULL;
2160 	}
2161 
2162 	if (bserrno != 0) {
2163 		blob_persist_complete(seq, ctx, bserrno);
2164 		return;
2165 	}
2166 
2167 	/* Only write out Extent Pages when blob was resized. */
2168 	for (i = ctx->next_extent_page; i < blob->active.extent_pages_array_size; i++) {
2169 		extent_page_id = blob->active.extent_pages[i];
2170 		if (extent_page_id == 0) {
2171 			/* No Extent Page to persist */
2172 			assert(spdk_blob_is_thin_provisioned(blob));
2173 			continue;
2174 		}
2175 		assert(spdk_bit_array_get(blob->bs->used_md_pages, extent_page_id));
2176 		ctx->next_extent_page = i + 1;
2177 		rc = blob_serialize_add_page(ctx->blob, &ctx->extent_page, &page_count, &ctx->extent_page);
2178 		if (rc < 0) {
2179 			blob_persist_complete(seq, ctx, rc);
2180 			return;
2181 		}
2182 
2183 		blob->state = SPDK_BLOB_STATE_DIRTY;
2184 		blob_serialize_extent_page(blob, i * SPDK_EXTENTS_PER_EP, ctx->extent_page);
2185 
2186 		ctx->extent_page->crc = blob_md_page_calc_crc(ctx->extent_page);
2187 
2188 		bs_sequence_write_dev(seq, ctx->extent_page, bs_md_page_to_lba(blob->bs, extent_page_id),
2189 				      bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE),
2190 				      blob_persist_write_extent_pages, ctx);
2191 		return;
2192 	}
2193 
2194 	blob_persist_generate_new_md(ctx);
2195 }
2196 
2197 static void
2198 blob_persist_start(struct spdk_blob_persist_ctx *ctx)
2199 {
2200 	spdk_bs_sequence_t *seq = ctx->seq;
2201 	struct spdk_blob *blob = ctx->blob;
2202 
2203 	if (blob->active.num_pages == 0) {
2204 		/* This is the signal that the blob should be deleted.
2205 		 * Immediately jump to the clean up routine. */
2206 		assert(blob->clean.num_pages > 0);
2207 		blob->state = SPDK_BLOB_STATE_CLEAN;
2208 		blob_persist_zero_pages(seq, ctx, 0);
2209 		return;
2210 
2211 	}
2212 
2213 	if (blob->clean.num_clusters < blob->active.num_clusters) {
2214 		/* Blob was resized up */
2215 		assert(blob->clean.num_extent_pages <= blob->active.num_extent_pages);
2216 		ctx->next_extent_page = spdk_max(1, blob->clean.num_extent_pages) - 1;
2217 	} else if (blob->active.num_clusters < blob->active.cluster_array_size) {
2218 		/* Blob was resized down */
2219 		assert(blob->clean.num_extent_pages >= blob->active.num_extent_pages);
2220 		ctx->next_extent_page = spdk_max(1, blob->active.num_extent_pages) - 1;
2221 	} else {
2222 		/* No change in size occurred */
2223 		blob_persist_generate_new_md(ctx);
2224 		return;
2225 	}
2226 
2227 	blob_persist_write_extent_pages(seq, ctx, 0);
2228 }
2229 
2230 static void
2231 blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2232 {
2233 	struct spdk_blob_persist_ctx *ctx = cb_arg;
2234 
2235 	spdk_free(ctx->super);
2236 
2237 	if (bserrno != 0) {
2238 		blob_persist_complete(seq, ctx, bserrno);
2239 		return;
2240 	}
2241 
2242 	ctx->blob->bs->clean = 0;
2243 
2244 	blob_persist_start(ctx);
2245 }
2246 
2247 static void
2248 bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2249 	       struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
2250 
2251 
2252 static void
2253 blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2254 {
2255 	struct spdk_blob_persist_ctx *ctx = cb_arg;
2256 
2257 	if (bserrno != 0) {
2258 		spdk_free(ctx->super);
2259 		blob_persist_complete(seq, ctx, bserrno);
2260 		return;
2261 	}
2262 
2263 	ctx->super->clean = 0;
2264 	if (ctx->super->size == 0) {
2265 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
2266 	}
2267 
2268 	bs_write_super(seq, ctx->blob->bs, ctx->super, blob_persist_dirty_cpl, ctx);
2269 }
2270 
2271 static void
2272 blob_persist_check_dirty(struct spdk_blob_persist_ctx *ctx)
2273 {
2274 	if (ctx->blob->bs->clean) {
2275 		ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
2276 					  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2277 		if (!ctx->super) {
2278 			blob_persist_complete(ctx->seq, ctx, -ENOMEM);
2279 			return;
2280 		}
2281 
2282 		bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(ctx->blob->bs, 0),
2283 				     bs_byte_to_lba(ctx->blob->bs, sizeof(*ctx->super)),
2284 				     blob_persist_dirty, ctx);
2285 	} else {
2286 		blob_persist_start(ctx);
2287 	}
2288 }
2289 
2290 /* Write a blob to disk */
2291 static void
2292 blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
2293 	     spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2294 {
2295 	struct spdk_blob_persist_ctx *ctx;
2296 
2297 	blob_verify_md_op(blob);
2298 
2299 	if (blob->state == SPDK_BLOB_STATE_CLEAN && TAILQ_EMPTY(&blob->persists_to_complete)) {
2300 		cb_fn(seq, cb_arg, 0);
2301 		return;
2302 	}
2303 
2304 	ctx = calloc(1, sizeof(*ctx));
2305 	if (!ctx) {
2306 		cb_fn(seq, cb_arg, -ENOMEM);
2307 		return;
2308 	}
2309 	ctx->blob = blob;
2310 	ctx->seq = seq;
2311 	ctx->cb_fn = cb_fn;
2312 	ctx->cb_arg = cb_arg;
2313 
2314 	/* Multiple blob persists can affect one another, via blob->state or
2315 	 * blob mutable data changes. To prevent it, queue up the persists. */
2316 	if (!TAILQ_EMPTY(&blob->persists_to_complete)) {
2317 		TAILQ_INSERT_TAIL(&blob->pending_persists, ctx, link);
2318 		return;
2319 	}
2320 	TAILQ_INSERT_HEAD(&blob->persists_to_complete, ctx, link);
2321 
2322 	blob_persist_check_dirty(ctx);
2323 }
2324 
2325 struct spdk_blob_copy_cluster_ctx {
2326 	struct spdk_blob *blob;
2327 	uint8_t *buf;
2328 	uint64_t page;
2329 	uint64_t new_cluster;
2330 	uint32_t new_extent_page;
2331 	spdk_bs_sequence_t *seq;
2332 };
2333 
2334 static void
2335 blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
2336 {
2337 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2338 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
2339 	TAILQ_HEAD(, spdk_bs_request_set) requests;
2340 	spdk_bs_user_op_t *op;
2341 
2342 	TAILQ_INIT(&requests);
2343 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
2344 
2345 	while (!TAILQ_EMPTY(&requests)) {
2346 		op = TAILQ_FIRST(&requests);
2347 		TAILQ_REMOVE(&requests, op, link);
2348 		if (bserrno == 0) {
2349 			bs_user_op_execute(op);
2350 		} else {
2351 			bs_user_op_abort(op);
2352 		}
2353 	}
2354 
2355 	spdk_free(ctx->buf);
2356 	free(ctx);
2357 }
2358 
2359 static void
2360 blob_insert_cluster_cpl(void *cb_arg, int bserrno)
2361 {
2362 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2363 
2364 	if (bserrno) {
2365 		if (bserrno == -EEXIST) {
2366 			/* The metadata insert failed because another thread
2367 			 * allocated the cluster first. Free our cluster
2368 			 * but continue without error. */
2369 			bserrno = 0;
2370 		}
2371 		pthread_mutex_lock(&ctx->blob->bs->used_clusters_mutex);
2372 		bs_release_cluster(ctx->blob->bs, ctx->new_cluster);
2373 		pthread_mutex_unlock(&ctx->blob->bs->used_clusters_mutex);
2374 		if (ctx->new_extent_page != 0) {
2375 			bs_release_md_page(ctx->blob->bs, ctx->new_extent_page);
2376 		}
2377 	}
2378 
2379 	bs_sequence_finish(ctx->seq, bserrno);
2380 }
2381 
2382 static void
2383 blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2384 {
2385 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2386 	uint32_t cluster_number;
2387 
2388 	if (bserrno) {
2389 		/* The write failed, so jump to the final completion handler */
2390 		bs_sequence_finish(seq, bserrno);
2391 		return;
2392 	}
2393 
2394 	cluster_number = bs_page_to_cluster(ctx->blob->bs, ctx->page);
2395 
2396 	blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
2397 					 ctx->new_extent_page, blob_insert_cluster_cpl, ctx);
2398 }
2399 
2400 static void
2401 blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2402 {
2403 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2404 
2405 	if (bserrno != 0) {
2406 		/* The read failed, so jump to the final completion handler */
2407 		bs_sequence_finish(seq, bserrno);
2408 		return;
2409 	}
2410 
2411 	/* Write whole cluster */
2412 	bs_sequence_write_dev(seq, ctx->buf,
2413 			      bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
2414 			      bs_cluster_to_lba(ctx->blob->bs, 1),
2415 			      blob_write_copy_cpl, ctx);
2416 }
2417 
2418 static void
2419 bs_allocate_and_copy_cluster(struct spdk_blob *blob,
2420 			     struct spdk_io_channel *_ch,
2421 			     uint64_t io_unit, spdk_bs_user_op_t *op)
2422 {
2423 	struct spdk_bs_cpl cpl;
2424 	struct spdk_bs_channel *ch;
2425 	struct spdk_blob_copy_cluster_ctx *ctx;
2426 	uint32_t cluster_start_page;
2427 	uint32_t cluster_number;
2428 	int rc;
2429 
2430 	ch = spdk_io_channel_get_ctx(_ch);
2431 
2432 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
2433 		/* There are already operations pending. Queue this user op
2434 		 * and return because it will be re-executed when the outstanding
2435 		 * cluster allocation completes. */
2436 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
2437 		return;
2438 	}
2439 
2440 	/* Round the io_unit offset down to the first page in the cluster */
2441 	cluster_start_page = bs_io_unit_to_cluster_start(blob, io_unit);
2442 
2443 	/* Calculate which index in the metadata cluster array the corresponding
2444 	 * cluster is supposed to be at. */
2445 	cluster_number = bs_io_unit_to_cluster_number(blob, io_unit);
2446 
2447 	ctx = calloc(1, sizeof(*ctx));
2448 	if (!ctx) {
2449 		bs_user_op_abort(op);
2450 		return;
2451 	}
2452 
2453 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
2454 
2455 	ctx->blob = blob;
2456 	ctx->page = cluster_start_page;
2457 
2458 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
2459 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
2460 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2461 		if (!ctx->buf) {
2462 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
2463 				    blob->bs->cluster_sz);
2464 			free(ctx);
2465 			bs_user_op_abort(op);
2466 			return;
2467 		}
2468 	}
2469 
2470 	pthread_mutex_lock(&blob->bs->used_clusters_mutex);
2471 	rc = bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, &ctx->new_extent_page,
2472 				 false);
2473 	pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
2474 	if (rc != 0) {
2475 		spdk_free(ctx->buf);
2476 		free(ctx);
2477 		bs_user_op_abort(op);
2478 		return;
2479 	}
2480 
2481 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2482 	cpl.u.blob_basic.cb_fn = blob_allocate_and_copy_cluster_cpl;
2483 	cpl.u.blob_basic.cb_arg = ctx;
2484 
2485 	ctx->seq = bs_sequence_start(_ch, &cpl);
2486 	if (!ctx->seq) {
2487 		pthread_mutex_lock(&blob->bs->used_clusters_mutex);
2488 		bs_release_cluster(blob->bs, ctx->new_cluster);
2489 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
2490 		spdk_free(ctx->buf);
2491 		free(ctx);
2492 		bs_user_op_abort(op);
2493 		return;
2494 	}
2495 
2496 	/* Queue the user op to block other incoming operations */
2497 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
2498 
2499 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
2500 		/* Read cluster from backing device */
2501 		bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
2502 					bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
2503 					bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
2504 					blob_write_copy, ctx);
2505 	} else {
2506 		blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
2507 						 ctx->new_extent_page, blob_insert_cluster_cpl, ctx);
2508 	}
2509 }
2510 
2511 static inline bool
2512 blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
2513 				 uint64_t *lba,	uint64_t *lba_count)
2514 {
2515 	*lba_count = length;
2516 
2517 	if (!bs_io_unit_is_allocated(blob, io_unit)) {
2518 		assert(blob->back_bs_dev != NULL);
2519 		*lba = bs_io_unit_to_back_dev_lba(blob, io_unit);
2520 		*lba_count = bs_io_unit_to_back_dev_lba(blob, *lba_count);
2521 		return false;
2522 	} else {
2523 		*lba = bs_blob_io_unit_to_lba(blob, io_unit);
2524 		return true;
2525 	}
2526 }
2527 
2528 struct op_split_ctx {
2529 	struct spdk_blob *blob;
2530 	struct spdk_io_channel *channel;
2531 	uint64_t io_unit_offset;
2532 	uint64_t io_units_remaining;
2533 	void *curr_payload;
2534 	enum spdk_blob_op_type op_type;
2535 	spdk_bs_sequence_t *seq;
2536 };
2537 
2538 static void
2539 blob_request_submit_op_split_next(void *cb_arg, int bserrno)
2540 {
2541 	struct op_split_ctx	*ctx = cb_arg;
2542 	struct spdk_blob	*blob = ctx->blob;
2543 	struct spdk_io_channel	*ch = ctx->channel;
2544 	enum spdk_blob_op_type	op_type = ctx->op_type;
2545 	uint8_t			*buf = ctx->curr_payload;
2546 	uint64_t		offset = ctx->io_unit_offset;
2547 	uint64_t		length = ctx->io_units_remaining;
2548 	uint64_t		op_length;
2549 
2550 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2551 		bs_sequence_finish(ctx->seq, bserrno);
2552 		free(ctx);
2553 		return;
2554 	}
2555 
2556 	op_length = spdk_min(length, bs_num_io_units_to_cluster_boundary(blob,
2557 			     offset));
2558 
2559 	/* Update length and payload for next operation */
2560 	ctx->io_units_remaining -= op_length;
2561 	ctx->io_unit_offset += op_length;
2562 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
2563 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
2564 	}
2565 
2566 	switch (op_type) {
2567 	case SPDK_BLOB_READ:
2568 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
2569 				  blob_request_submit_op_split_next, ctx);
2570 		break;
2571 	case SPDK_BLOB_WRITE:
2572 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
2573 				   blob_request_submit_op_split_next, ctx);
2574 		break;
2575 	case SPDK_BLOB_UNMAP:
2576 		spdk_blob_io_unmap(blob, ch, offset, op_length,
2577 				   blob_request_submit_op_split_next, ctx);
2578 		break;
2579 	case SPDK_BLOB_WRITE_ZEROES:
2580 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
2581 					  blob_request_submit_op_split_next, ctx);
2582 		break;
2583 	case SPDK_BLOB_READV:
2584 	case SPDK_BLOB_WRITEV:
2585 		SPDK_ERRLOG("readv/write not valid\n");
2586 		bs_sequence_finish(ctx->seq, -EINVAL);
2587 		free(ctx);
2588 		break;
2589 	}
2590 }
2591 
2592 static void
2593 blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
2594 			     void *payload, uint64_t offset, uint64_t length,
2595 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2596 {
2597 	struct op_split_ctx *ctx;
2598 	spdk_bs_sequence_t *seq;
2599 	struct spdk_bs_cpl cpl;
2600 
2601 	assert(blob != NULL);
2602 
2603 	ctx = calloc(1, sizeof(struct op_split_ctx));
2604 	if (ctx == NULL) {
2605 		cb_fn(cb_arg, -ENOMEM);
2606 		return;
2607 	}
2608 
2609 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2610 	cpl.u.blob_basic.cb_fn = cb_fn;
2611 	cpl.u.blob_basic.cb_arg = cb_arg;
2612 
2613 	seq = bs_sequence_start(ch, &cpl);
2614 	if (!seq) {
2615 		free(ctx);
2616 		cb_fn(cb_arg, -ENOMEM);
2617 		return;
2618 	}
2619 
2620 	ctx->blob = blob;
2621 	ctx->channel = ch;
2622 	ctx->curr_payload = payload;
2623 	ctx->io_unit_offset = offset;
2624 	ctx->io_units_remaining = length;
2625 	ctx->op_type = op_type;
2626 	ctx->seq = seq;
2627 
2628 	blob_request_submit_op_split_next(ctx, 0);
2629 }
2630 
2631 static void
2632 blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
2633 			      void *payload, uint64_t offset, uint64_t length,
2634 			      spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2635 {
2636 	struct spdk_bs_cpl cpl;
2637 	uint64_t lba;
2638 	uint64_t lba_count;
2639 	bool is_allocated;
2640 
2641 	assert(blob != NULL);
2642 
2643 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2644 	cpl.u.blob_basic.cb_fn = cb_fn;
2645 	cpl.u.blob_basic.cb_arg = cb_arg;
2646 
2647 	if (blob->frozen_refcnt) {
2648 		/* This blob I/O is frozen */
2649 		spdk_bs_user_op_t *op;
2650 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
2651 
2652 		op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
2653 		if (!op) {
2654 			cb_fn(cb_arg, -ENOMEM);
2655 			return;
2656 		}
2657 
2658 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2659 
2660 		return;
2661 	}
2662 
2663 	is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2664 
2665 	switch (op_type) {
2666 	case SPDK_BLOB_READ: {
2667 		spdk_bs_batch_t *batch;
2668 
2669 		batch = bs_batch_open(_ch, &cpl);
2670 		if (!batch) {
2671 			cb_fn(cb_arg, -ENOMEM);
2672 			return;
2673 		}
2674 
2675 		if (is_allocated) {
2676 			/* Read from the blob */
2677 			bs_batch_read_dev(batch, payload, lba, lba_count);
2678 		} else {
2679 			/* Read from the backing block device */
2680 			bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
2681 		}
2682 
2683 		bs_batch_close(batch);
2684 		break;
2685 	}
2686 	case SPDK_BLOB_WRITE:
2687 	case SPDK_BLOB_WRITE_ZEROES: {
2688 		if (is_allocated) {
2689 			/* Write to the blob */
2690 			spdk_bs_batch_t *batch;
2691 
2692 			if (lba_count == 0) {
2693 				cb_fn(cb_arg, 0);
2694 				return;
2695 			}
2696 
2697 			batch = bs_batch_open(_ch, &cpl);
2698 			if (!batch) {
2699 				cb_fn(cb_arg, -ENOMEM);
2700 				return;
2701 			}
2702 
2703 			if (op_type == SPDK_BLOB_WRITE) {
2704 				bs_batch_write_dev(batch, payload, lba, lba_count);
2705 			} else {
2706 				bs_batch_write_zeroes_dev(batch, lba, lba_count);
2707 			}
2708 
2709 			bs_batch_close(batch);
2710 		} else {
2711 			/* Queue this operation and allocate the cluster */
2712 			spdk_bs_user_op_t *op;
2713 
2714 			op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
2715 			if (!op) {
2716 				cb_fn(cb_arg, -ENOMEM);
2717 				return;
2718 			}
2719 
2720 			bs_allocate_and_copy_cluster(blob, _ch, offset, op);
2721 		}
2722 		break;
2723 	}
2724 	case SPDK_BLOB_UNMAP: {
2725 		spdk_bs_batch_t *batch;
2726 
2727 		batch = bs_batch_open(_ch, &cpl);
2728 		if (!batch) {
2729 			cb_fn(cb_arg, -ENOMEM);
2730 			return;
2731 		}
2732 
2733 		if (is_allocated) {
2734 			bs_batch_unmap_dev(batch, lba, lba_count);
2735 		}
2736 
2737 		bs_batch_close(batch);
2738 		break;
2739 	}
2740 	case SPDK_BLOB_READV:
2741 	case SPDK_BLOB_WRITEV:
2742 		SPDK_ERRLOG("readv/write not valid\n");
2743 		cb_fn(cb_arg, -EINVAL);
2744 		break;
2745 	}
2746 }
2747 
2748 static void
2749 blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2750 		       void *payload, uint64_t offset, uint64_t length,
2751 		       spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2752 {
2753 	assert(blob != NULL);
2754 
2755 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
2756 		cb_fn(cb_arg, -EPERM);
2757 		return;
2758 	}
2759 
2760 	if (length == 0) {
2761 		cb_fn(cb_arg, 0);
2762 		return;
2763 	}
2764 
2765 	if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2766 		cb_fn(cb_arg, -EINVAL);
2767 		return;
2768 	}
2769 	if (length <= bs_num_io_units_to_cluster_boundary(blob, offset)) {
2770 		blob_request_submit_op_single(_channel, blob, payload, offset, length,
2771 					      cb_fn, cb_arg, op_type);
2772 	} else {
2773 		blob_request_submit_op_split(_channel, blob, payload, offset, length,
2774 					     cb_fn, cb_arg, op_type);
2775 	}
2776 }
2777 
2778 struct rw_iov_ctx {
2779 	struct spdk_blob *blob;
2780 	struct spdk_io_channel *channel;
2781 	spdk_blob_op_complete cb_fn;
2782 	void *cb_arg;
2783 	bool read;
2784 	int iovcnt;
2785 	struct iovec *orig_iov;
2786 	uint64_t io_unit_offset;
2787 	uint64_t io_units_remaining;
2788 	uint64_t io_units_done;
2789 	struct iovec iov[0];
2790 };
2791 
2792 static void
2793 rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2794 {
2795 	assert(cb_arg == NULL);
2796 	bs_sequence_finish(seq, bserrno);
2797 }
2798 
2799 static void
2800 rw_iov_split_next(void *cb_arg, int bserrno)
2801 {
2802 	struct rw_iov_ctx *ctx = cb_arg;
2803 	struct spdk_blob *blob = ctx->blob;
2804 	struct iovec *iov, *orig_iov;
2805 	int iovcnt;
2806 	size_t orig_iovoff;
2807 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2808 	uint64_t byte_count;
2809 
2810 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2811 		ctx->cb_fn(ctx->cb_arg, bserrno);
2812 		free(ctx);
2813 		return;
2814 	}
2815 
2816 	io_unit_offset = ctx->io_unit_offset;
2817 	io_units_to_boundary = bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2818 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2819 	/*
2820 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2821 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2822 	 *  point to the current position in the I/O sequence.
2823 	 */
2824 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2825 	orig_iov = &ctx->orig_iov[0];
2826 	orig_iovoff = 0;
2827 	while (byte_count > 0) {
2828 		if (byte_count >= orig_iov->iov_len) {
2829 			byte_count -= orig_iov->iov_len;
2830 			orig_iov++;
2831 		} else {
2832 			orig_iovoff = byte_count;
2833 			byte_count = 0;
2834 		}
2835 	}
2836 
2837 	/*
2838 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2839 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2840 	 */
2841 	byte_count = io_units_count * blob->bs->io_unit_size;
2842 	iov = &ctx->iov[0];
2843 	iovcnt = 0;
2844 	while (byte_count > 0) {
2845 		assert(iovcnt < ctx->iovcnt);
2846 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2847 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2848 		byte_count -= iov->iov_len;
2849 		orig_iovoff = 0;
2850 		orig_iov++;
2851 		iov++;
2852 		iovcnt++;
2853 	}
2854 
2855 	ctx->io_unit_offset += io_units_count;
2856 	ctx->io_units_remaining -= io_units_count;
2857 	ctx->io_units_done += io_units_count;
2858 	iov = &ctx->iov[0];
2859 
2860 	if (ctx->read) {
2861 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2862 				   io_units_count, rw_iov_split_next, ctx);
2863 	} else {
2864 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2865 				    io_units_count, rw_iov_split_next, ctx);
2866 	}
2867 }
2868 
2869 static void
2870 blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2871 			   struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2872 			   spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2873 {
2874 	struct spdk_bs_cpl	cpl;
2875 
2876 	assert(blob != NULL);
2877 
2878 	if (!read && blob->data_ro) {
2879 		cb_fn(cb_arg, -EPERM);
2880 		return;
2881 	}
2882 
2883 	if (length == 0) {
2884 		cb_fn(cb_arg, 0);
2885 		return;
2886 	}
2887 
2888 	if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2889 		cb_fn(cb_arg, -EINVAL);
2890 		return;
2891 	}
2892 
2893 	/*
2894 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2895 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2896 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2897 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2898 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2899 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2900 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2901 	 *
2902 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2903 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2904 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2905 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2906 	 */
2907 	if (spdk_likely(length <= bs_num_io_units_to_cluster_boundary(blob, offset))) {
2908 		uint64_t lba_count;
2909 		uint64_t lba;
2910 		bool is_allocated;
2911 
2912 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2913 		cpl.u.blob_basic.cb_fn = cb_fn;
2914 		cpl.u.blob_basic.cb_arg = cb_arg;
2915 
2916 		if (blob->frozen_refcnt) {
2917 			/* This blob I/O is frozen */
2918 			enum spdk_blob_op_type op_type;
2919 			spdk_bs_user_op_t *op;
2920 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2921 
2922 			op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV;
2923 			op = bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length);
2924 			if (!op) {
2925 				cb_fn(cb_arg, -ENOMEM);
2926 				return;
2927 			}
2928 
2929 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2930 
2931 			return;
2932 		}
2933 
2934 		is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2935 
2936 		if (read) {
2937 			spdk_bs_sequence_t *seq;
2938 
2939 			seq = bs_sequence_start(_channel, &cpl);
2940 			if (!seq) {
2941 				cb_fn(cb_arg, -ENOMEM);
2942 				return;
2943 			}
2944 
2945 			if (is_allocated) {
2946 				bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL);
2947 			} else {
2948 				bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2949 							 rw_iov_done, NULL);
2950 			}
2951 		} else {
2952 			if (is_allocated) {
2953 				spdk_bs_sequence_t *seq;
2954 
2955 				seq = bs_sequence_start(_channel, &cpl);
2956 				if (!seq) {
2957 					cb_fn(cb_arg, -ENOMEM);
2958 					return;
2959 				}
2960 
2961 				bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL);
2962 			} else {
2963 				/* Queue this operation and allocate the cluster */
2964 				spdk_bs_user_op_t *op;
2965 
2966 				op = bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2967 						      length);
2968 				if (!op) {
2969 					cb_fn(cb_arg, -ENOMEM);
2970 					return;
2971 				}
2972 
2973 				bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2974 			}
2975 		}
2976 	} else {
2977 		struct rw_iov_ctx *ctx;
2978 
2979 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2980 		if (ctx == NULL) {
2981 			cb_fn(cb_arg, -ENOMEM);
2982 			return;
2983 		}
2984 
2985 		ctx->blob = blob;
2986 		ctx->channel = _channel;
2987 		ctx->cb_fn = cb_fn;
2988 		ctx->cb_arg = cb_arg;
2989 		ctx->read = read;
2990 		ctx->orig_iov = iov;
2991 		ctx->iovcnt = iovcnt;
2992 		ctx->io_unit_offset = offset;
2993 		ctx->io_units_remaining = length;
2994 		ctx->io_units_done = 0;
2995 
2996 		rw_iov_split_next(ctx, 0);
2997 	}
2998 }
2999 
3000 static struct spdk_blob *
3001 blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
3002 {
3003 	struct spdk_blob find;
3004 
3005 	if (spdk_bit_array_get(bs->open_blobids, blobid) == 0) {
3006 		return NULL;
3007 	}
3008 
3009 	find.id = blobid;
3010 	return RB_FIND(spdk_blob_tree, &bs->open_blobs, &find);
3011 }
3012 
3013 static void
3014 blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
3015 				    struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
3016 {
3017 	assert(blob != NULL);
3018 	*snapshot_entry = NULL;
3019 	*clone_entry = NULL;
3020 
3021 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
3022 		return;
3023 	}
3024 
3025 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
3026 		if ((*snapshot_entry)->id == blob->parent_id) {
3027 			break;
3028 		}
3029 	}
3030 
3031 	if (*snapshot_entry != NULL) {
3032 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
3033 			if ((*clone_entry)->id == blob->id) {
3034 				break;
3035 			}
3036 		}
3037 
3038 		assert(*clone_entry != NULL);
3039 	}
3040 }
3041 
3042 static int
3043 bs_channel_create(void *io_device, void *ctx_buf)
3044 {
3045 	struct spdk_blob_store		*bs = io_device;
3046 	struct spdk_bs_channel		*channel = ctx_buf;
3047 	struct spdk_bs_dev		*dev;
3048 	uint32_t			max_ops = bs->max_channel_ops;
3049 	uint32_t			i;
3050 
3051 	dev = bs->dev;
3052 
3053 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
3054 	if (!channel->req_mem) {
3055 		return -1;
3056 	}
3057 
3058 	TAILQ_INIT(&channel->reqs);
3059 
3060 	for (i = 0; i < max_ops; i++) {
3061 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
3062 	}
3063 
3064 	channel->bs = bs;
3065 	channel->dev = dev;
3066 	channel->dev_channel = dev->create_channel(dev);
3067 
3068 	if (!channel->dev_channel) {
3069 		SPDK_ERRLOG("Failed to create device channel.\n");
3070 		free(channel->req_mem);
3071 		return -1;
3072 	}
3073 
3074 	TAILQ_INIT(&channel->need_cluster_alloc);
3075 	TAILQ_INIT(&channel->queued_io);
3076 
3077 	return 0;
3078 }
3079 
3080 static void
3081 bs_channel_destroy(void *io_device, void *ctx_buf)
3082 {
3083 	struct spdk_bs_channel *channel = ctx_buf;
3084 	spdk_bs_user_op_t *op;
3085 
3086 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
3087 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
3088 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
3089 		bs_user_op_abort(op);
3090 	}
3091 
3092 	while (!TAILQ_EMPTY(&channel->queued_io)) {
3093 		op = TAILQ_FIRST(&channel->queued_io);
3094 		TAILQ_REMOVE(&channel->queued_io, op, link);
3095 		bs_user_op_abort(op);
3096 	}
3097 
3098 	free(channel->req_mem);
3099 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
3100 }
3101 
3102 static void
3103 bs_dev_destroy(void *io_device)
3104 {
3105 	struct spdk_blob_store *bs = io_device;
3106 	struct spdk_blob	*blob, *blob_tmp;
3107 
3108 	bs->dev->destroy(bs->dev);
3109 
3110 	RB_FOREACH_SAFE(blob, spdk_blob_tree, &bs->open_blobs, blob_tmp) {
3111 		RB_REMOVE(spdk_blob_tree, &bs->open_blobs, blob);
3112 		spdk_bit_array_clear(bs->open_blobids, blob->id);
3113 		blob_free(blob);
3114 	}
3115 
3116 	pthread_mutex_destroy(&bs->used_clusters_mutex);
3117 
3118 	spdk_bit_array_free(&bs->open_blobids);
3119 	spdk_bit_array_free(&bs->used_blobids);
3120 	spdk_bit_array_free(&bs->used_md_pages);
3121 	spdk_bit_pool_free(&bs->used_clusters);
3122 	/*
3123 	 * If this function is called for any reason except a successful unload,
3124 	 * the unload_cpl type will be NONE and this will be a nop.
3125 	 */
3126 	bs_call_cpl(&bs->unload_cpl, bs->unload_err);
3127 
3128 	free(bs);
3129 }
3130 
3131 static int
3132 bs_blob_list_add(struct spdk_blob *blob)
3133 {
3134 	spdk_blob_id snapshot_id;
3135 	struct spdk_blob_list *snapshot_entry = NULL;
3136 	struct spdk_blob_list *clone_entry = NULL;
3137 
3138 	assert(blob != NULL);
3139 
3140 	snapshot_id = blob->parent_id;
3141 	if (snapshot_id == SPDK_BLOBID_INVALID) {
3142 		return 0;
3143 	}
3144 
3145 	snapshot_entry = bs_get_snapshot_entry(blob->bs, snapshot_id);
3146 	if (snapshot_entry == NULL) {
3147 		/* Snapshot not found */
3148 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
3149 		if (snapshot_entry == NULL) {
3150 			return -ENOMEM;
3151 		}
3152 		snapshot_entry->id = snapshot_id;
3153 		TAILQ_INIT(&snapshot_entry->clones);
3154 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
3155 	} else {
3156 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
3157 			if (clone_entry->id == blob->id) {
3158 				break;
3159 			}
3160 		}
3161 	}
3162 
3163 	if (clone_entry == NULL) {
3164 		/* Clone not found */
3165 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
3166 		if (clone_entry == NULL) {
3167 			return -ENOMEM;
3168 		}
3169 		clone_entry->id = blob->id;
3170 		TAILQ_INIT(&clone_entry->clones);
3171 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
3172 		snapshot_entry->clone_count++;
3173 	}
3174 
3175 	return 0;
3176 }
3177 
3178 static void
3179 bs_blob_list_remove(struct spdk_blob *blob)
3180 {
3181 	struct spdk_blob_list *snapshot_entry = NULL;
3182 	struct spdk_blob_list *clone_entry = NULL;
3183 
3184 	blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
3185 
3186 	if (snapshot_entry == NULL) {
3187 		return;
3188 	}
3189 
3190 	blob->parent_id = SPDK_BLOBID_INVALID;
3191 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
3192 	free(clone_entry);
3193 
3194 	snapshot_entry->clone_count--;
3195 }
3196 
3197 static int
3198 bs_blob_list_free(struct spdk_blob_store *bs)
3199 {
3200 	struct spdk_blob_list *snapshot_entry;
3201 	struct spdk_blob_list *snapshot_entry_tmp;
3202 	struct spdk_blob_list *clone_entry;
3203 	struct spdk_blob_list *clone_entry_tmp;
3204 
3205 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
3206 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
3207 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
3208 			free(clone_entry);
3209 		}
3210 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
3211 		free(snapshot_entry);
3212 	}
3213 
3214 	return 0;
3215 }
3216 
3217 static void
3218 bs_free(struct spdk_blob_store *bs)
3219 {
3220 	bs_blob_list_free(bs);
3221 
3222 	bs_unregister_md_thread(bs);
3223 	spdk_io_device_unregister(bs, bs_dev_destroy);
3224 }
3225 
3226 void
3227 spdk_bs_opts_init(struct spdk_bs_opts *opts, size_t opts_size)
3228 {
3229 
3230 	if (!opts) {
3231 		SPDK_ERRLOG("opts should not be NULL\n");
3232 		return;
3233 	}
3234 
3235 	if (!opts_size) {
3236 		SPDK_ERRLOG("opts_size should not be zero value\n");
3237 		return;
3238 	}
3239 
3240 	memset(opts, 0, opts_size);
3241 	opts->opts_size = opts_size;
3242 
3243 #define FIELD_OK(field) \
3244 	offsetof(struct spdk_bs_opts, field) + sizeof(opts->field) <= opts_size
3245 
3246 #define SET_FIELD(field, value) \
3247 	if (FIELD_OK(field)) { \
3248 		opts->field = value; \
3249 	} \
3250 
3251 	SET_FIELD(cluster_sz, SPDK_BLOB_OPTS_CLUSTER_SZ);
3252 	SET_FIELD(num_md_pages, SPDK_BLOB_OPTS_NUM_MD_PAGES);
3253 	SET_FIELD(max_md_ops, SPDK_BLOB_OPTS_NUM_MD_PAGES);
3254 	SET_FIELD(max_channel_ops, SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS);
3255 	SET_FIELD(clear_method,  BS_CLEAR_WITH_UNMAP);
3256 
3257 	if (FIELD_OK(bstype)) {
3258 		memset(&opts->bstype, 0, sizeof(opts->bstype));
3259 	}
3260 
3261 	SET_FIELD(iter_cb_fn, NULL);
3262 	SET_FIELD(iter_cb_arg, NULL);
3263 	SET_FIELD(force_recover, false);
3264 
3265 #undef FIELD_OK
3266 #undef SET_FIELD
3267 }
3268 
3269 static int
3270 bs_opts_verify(struct spdk_bs_opts *opts)
3271 {
3272 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
3273 	    opts->max_channel_ops == 0) {
3274 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
3275 		return -1;
3276 	}
3277 
3278 	return 0;
3279 }
3280 
3281 /* START spdk_bs_load */
3282 
3283 /* spdk_bs_load_ctx is used for init, load, unload and dump code paths. */
3284 
3285 struct spdk_bs_load_ctx {
3286 	struct spdk_blob_store		*bs;
3287 	struct spdk_bs_super_block	*super;
3288 
3289 	struct spdk_bs_md_mask		*mask;
3290 	bool				in_page_chain;
3291 	uint32_t			page_index;
3292 	uint32_t			cur_page;
3293 	struct spdk_blob_md_page	*page;
3294 
3295 	uint64_t			num_extent_pages;
3296 	uint32_t			*extent_page_num;
3297 	struct spdk_blob_md_page	*extent_pages;
3298 	struct spdk_bit_array		*used_clusters;
3299 
3300 	spdk_bs_sequence_t			*seq;
3301 	spdk_blob_op_with_handle_complete	iter_cb_fn;
3302 	void					*iter_cb_arg;
3303 	struct spdk_blob			*blob;
3304 	spdk_blob_id				blobid;
3305 
3306 	bool					force_recover;
3307 
3308 	/* These fields are used in the spdk_bs_dump path. */
3309 	bool					dumping;
3310 	FILE					*fp;
3311 	spdk_bs_dump_print_xattr		print_xattr_fn;
3312 	char					xattr_name[4096];
3313 };
3314 
3315 static int
3316 bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs,
3317 	 struct spdk_bs_load_ctx **_ctx)
3318 {
3319 	struct spdk_blob_store	*bs;
3320 	struct spdk_bs_load_ctx	*ctx;
3321 	uint64_t dev_size;
3322 	int rc;
3323 
3324 	dev_size = dev->blocklen * dev->blockcnt;
3325 	if (dev_size < opts->cluster_sz) {
3326 		/* Device size cannot be smaller than cluster size of blobstore */
3327 		SPDK_INFOLOG(blob, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
3328 			     dev_size, opts->cluster_sz);
3329 		return -ENOSPC;
3330 	}
3331 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
3332 		/* Cluster size cannot be smaller than page size */
3333 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
3334 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
3335 		return -EINVAL;
3336 	}
3337 	bs = calloc(1, sizeof(struct spdk_blob_store));
3338 	if (!bs) {
3339 		return -ENOMEM;
3340 	}
3341 
3342 	ctx = calloc(1, sizeof(struct spdk_bs_load_ctx));
3343 	if (!ctx) {
3344 		free(bs);
3345 		return -ENOMEM;
3346 	}
3347 
3348 	ctx->bs = bs;
3349 	ctx->iter_cb_fn = opts->iter_cb_fn;
3350 	ctx->iter_cb_arg = opts->iter_cb_arg;
3351 	ctx->force_recover = opts->force_recover;
3352 
3353 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3354 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3355 	if (!ctx->super) {
3356 		free(ctx);
3357 		free(bs);
3358 		return -ENOMEM;
3359 	}
3360 
3361 	RB_INIT(&bs->open_blobs);
3362 	TAILQ_INIT(&bs->snapshots);
3363 	bs->dev = dev;
3364 	bs->md_thread = spdk_get_thread();
3365 	assert(bs->md_thread != NULL);
3366 
3367 	/*
3368 	 * Do not use bs_lba_to_cluster() here since blockcnt may not be an
3369 	 *  even multiple of the cluster size.
3370 	 */
3371 	bs->cluster_sz = opts->cluster_sz;
3372 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
3373 	ctx->used_clusters = spdk_bit_array_create(bs->total_clusters);
3374 	if (!ctx->used_clusters) {
3375 		spdk_free(ctx->super);
3376 		free(ctx);
3377 		free(bs);
3378 		return -ENOMEM;
3379 	}
3380 
3381 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3382 	if (spdk_u32_is_pow2(bs->pages_per_cluster)) {
3383 		bs->pages_per_cluster_shift = spdk_u32log2(bs->pages_per_cluster);
3384 	}
3385 	bs->num_free_clusters = bs->total_clusters;
3386 	bs->io_unit_size = dev->blocklen;
3387 
3388 	bs->max_channel_ops = opts->max_channel_ops;
3389 	bs->super_blob = SPDK_BLOBID_INVALID;
3390 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
3391 
3392 	/* The metadata is assumed to be at least 1 page */
3393 	bs->used_md_pages = spdk_bit_array_create(1);
3394 	bs->used_blobids = spdk_bit_array_create(0);
3395 	bs->open_blobids = spdk_bit_array_create(0);
3396 
3397 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
3398 
3399 	spdk_io_device_register(bs, bs_channel_create, bs_channel_destroy,
3400 				sizeof(struct spdk_bs_channel), "blobstore");
3401 	rc = bs_register_md_thread(bs);
3402 	if (rc == -1) {
3403 		spdk_io_device_unregister(bs, NULL);
3404 		pthread_mutex_destroy(&bs->used_clusters_mutex);
3405 		spdk_bit_array_free(&bs->open_blobids);
3406 		spdk_bit_array_free(&bs->used_blobids);
3407 		spdk_bit_array_free(&bs->used_md_pages);
3408 		spdk_bit_array_free(&ctx->used_clusters);
3409 		spdk_free(ctx->super);
3410 		free(ctx);
3411 		free(bs);
3412 		/* FIXME: this is a lie but don't know how to get a proper error code here */
3413 		return -ENOMEM;
3414 	}
3415 
3416 	*_ctx = ctx;
3417 	*_bs = bs;
3418 	return 0;
3419 }
3420 
3421 static void
3422 bs_load_ctx_fail(struct spdk_bs_load_ctx *ctx, int bserrno)
3423 {
3424 	assert(bserrno != 0);
3425 
3426 	spdk_free(ctx->super);
3427 	bs_sequence_finish(ctx->seq, bserrno);
3428 	bs_free(ctx->bs);
3429 	spdk_bit_array_free(&ctx->used_clusters);
3430 	free(ctx);
3431 }
3432 
3433 static void
3434 bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
3435 	       struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
3436 {
3437 	/* Update the values in the super block */
3438 	super->super_blob = bs->super_blob;
3439 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
3440 	super->crc = blob_md_page_calc_crc(super);
3441 	bs_sequence_write_dev(seq, super, bs_page_to_lba(bs, 0),
3442 			      bs_byte_to_lba(bs, sizeof(*super)),
3443 			      cb_fn, cb_arg);
3444 }
3445 
3446 static void
3447 bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3448 {
3449 	struct spdk_bs_load_ctx	*ctx = arg;
3450 	uint64_t	mask_size, lba, lba_count;
3451 
3452 	/* Write out the used clusters mask */
3453 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
3454 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3455 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3456 	if (!ctx->mask) {
3457 		bs_load_ctx_fail(ctx, -ENOMEM);
3458 		return;
3459 	}
3460 
3461 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
3462 	ctx->mask->length = ctx->bs->total_clusters;
3463 	/* We could get here through the normal unload path, or through dirty
3464 	 * shutdown recovery.  For the normal unload path, we use the mask from
3465 	 * the bit pool.  For dirty shutdown recovery, we don't have a bit pool yet -
3466 	 * only the bit array from the load ctx.
3467 	 */
3468 	if (ctx->bs->used_clusters) {
3469 		assert(ctx->mask->length == spdk_bit_pool_capacity(ctx->bs->used_clusters));
3470 		spdk_bit_pool_store_mask(ctx->bs->used_clusters, ctx->mask->mask);
3471 	} else {
3472 		assert(ctx->mask->length == spdk_bit_array_capacity(ctx->used_clusters));
3473 		spdk_bit_array_store_mask(ctx->used_clusters, ctx->mask->mask);
3474 	}
3475 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
3476 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
3477 	bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3478 }
3479 
3480 static void
3481 bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3482 {
3483 	struct spdk_bs_load_ctx	*ctx = arg;
3484 	uint64_t	mask_size, lba, lba_count;
3485 
3486 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
3487 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3488 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3489 	if (!ctx->mask) {
3490 		bs_load_ctx_fail(ctx, -ENOMEM);
3491 		return;
3492 	}
3493 
3494 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
3495 	ctx->mask->length = ctx->super->md_len;
3496 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
3497 
3498 	spdk_bit_array_store_mask(ctx->bs->used_md_pages, ctx->mask->mask);
3499 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3500 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3501 	bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3502 }
3503 
3504 static void
3505 bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3506 {
3507 	struct spdk_bs_load_ctx	*ctx = arg;
3508 	uint64_t	mask_size, lba, lba_count;
3509 
3510 	if (ctx->super->used_blobid_mask_len == 0) {
3511 		/*
3512 		 * This is a pre-v3 on-disk format where the blobid mask does not get
3513 		 *  written to disk.
3514 		 */
3515 		cb_fn(seq, arg, 0);
3516 		return;
3517 	}
3518 
3519 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
3520 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3521 				 SPDK_MALLOC_DMA);
3522 	if (!ctx->mask) {
3523 		bs_load_ctx_fail(ctx, -ENOMEM);
3524 		return;
3525 	}
3526 
3527 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
3528 	ctx->mask->length = ctx->super->md_len;
3529 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
3530 
3531 	spdk_bit_array_store_mask(ctx->bs->used_blobids, ctx->mask->mask);
3532 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
3533 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
3534 	bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3535 }
3536 
3537 static void
3538 blob_set_thin_provision(struct spdk_blob *blob)
3539 {
3540 	blob_verify_md_op(blob);
3541 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
3542 	blob->state = SPDK_BLOB_STATE_DIRTY;
3543 }
3544 
3545 static void
3546 blob_set_clear_method(struct spdk_blob *blob, enum blob_clear_method clear_method)
3547 {
3548 	blob_verify_md_op(blob);
3549 	blob->clear_method = clear_method;
3550 	blob->md_ro_flags |= (clear_method << SPDK_BLOB_CLEAR_METHOD_SHIFT);
3551 	blob->state = SPDK_BLOB_STATE_DIRTY;
3552 }
3553 
3554 static void bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno);
3555 
3556 static void
3557 bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno)
3558 {
3559 	struct spdk_bs_load_ctx *ctx = cb_arg;
3560 	spdk_blob_id id;
3561 	int64_t page_num;
3562 
3563 	/* Iterate to next blob (we can't use spdk_bs_iter_next function as our
3564 	 * last blob has been removed */
3565 	page_num = bs_blobid_to_page(ctx->blobid);
3566 	page_num++;
3567 	page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num);
3568 	if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) {
3569 		bs_load_iter(ctx, NULL, -ENOENT);
3570 		return;
3571 	}
3572 
3573 	id = bs_page_to_blobid(page_num);
3574 
3575 	spdk_bs_open_blob(ctx->bs, id, bs_load_iter, ctx);
3576 }
3577 
3578 static void
3579 bs_delete_corrupted_close_cb(void *cb_arg, int bserrno)
3580 {
3581 	struct spdk_bs_load_ctx *ctx = cb_arg;
3582 
3583 	if (bserrno != 0) {
3584 		SPDK_ERRLOG("Failed to close corrupted blob\n");
3585 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3586 		return;
3587 	}
3588 
3589 	spdk_bs_delete_blob(ctx->bs, ctx->blobid, bs_delete_corrupted_blob_cpl, ctx);
3590 }
3591 
3592 static void
3593 bs_delete_corrupted_blob(void *cb_arg, int bserrno)
3594 {
3595 	struct spdk_bs_load_ctx *ctx = cb_arg;
3596 	uint64_t i;
3597 
3598 	if (bserrno != 0) {
3599 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
3600 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3601 		return;
3602 	}
3603 
3604 	/* Snapshot and clone have the same copy of cluster map and extent pages
3605 	 * at this point. Let's clear both for snapshot now,
3606 	 * so that it won't be cleared for clone later when we remove snapshot.
3607 	 * Also set thin provision to pass data corruption check */
3608 	for (i = 0; i < ctx->blob->active.num_clusters; i++) {
3609 		ctx->blob->active.clusters[i] = 0;
3610 	}
3611 	for (i = 0; i < ctx->blob->active.num_extent_pages; i++) {
3612 		ctx->blob->active.extent_pages[i] = 0;
3613 	}
3614 
3615 	ctx->blob->md_ro = false;
3616 
3617 	blob_set_thin_provision(ctx->blob);
3618 
3619 	ctx->blobid = ctx->blob->id;
3620 
3621 	spdk_blob_close(ctx->blob, bs_delete_corrupted_close_cb, ctx);
3622 }
3623 
3624 static void
3625 bs_update_corrupted_blob(void *cb_arg, int bserrno)
3626 {
3627 	struct spdk_bs_load_ctx *ctx = cb_arg;
3628 
3629 	if (bserrno != 0) {
3630 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
3631 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3632 		return;
3633 	}
3634 
3635 	ctx->blob->md_ro = false;
3636 	blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true);
3637 	blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true);
3638 	spdk_blob_set_read_only(ctx->blob);
3639 
3640 	if (ctx->iter_cb_fn) {
3641 		ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0);
3642 	}
3643 	bs_blob_list_add(ctx->blob);
3644 
3645 	spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3646 }
3647 
3648 static void
3649 bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno)
3650 {
3651 	struct spdk_bs_load_ctx *ctx = cb_arg;
3652 
3653 	if (bserrno != 0) {
3654 		SPDK_ERRLOG("Failed to open clone of a corrupted blob\n");
3655 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3656 		return;
3657 	}
3658 
3659 	if (blob->parent_id == ctx->blob->id) {
3660 		/* Power failure occurred before updating clone (snapshot delete case)
3661 		 * or after updating clone (creating snapshot case) - keep snapshot */
3662 		spdk_blob_close(blob, bs_update_corrupted_blob, ctx);
3663 	} else {
3664 		/* Power failure occurred after updating clone (snapshot delete case)
3665 		 * or before updating clone (creating snapshot case) - remove snapshot */
3666 		spdk_blob_close(blob, bs_delete_corrupted_blob, ctx);
3667 	}
3668 }
3669 
3670 static void
3671 bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
3672 {
3673 	struct spdk_bs_load_ctx *ctx = arg;
3674 	const void *value;
3675 	size_t len;
3676 	int rc = 0;
3677 
3678 	if (bserrno == 0) {
3679 		/* Examine blob if it is corrupted after power failure. Fix
3680 		 * the ones that can be fixed and remove any other corrupted
3681 		 * ones. If it is not corrupted just process it */
3682 		rc = blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true);
3683 		if (rc != 0) {
3684 			rc = blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true);
3685 			if (rc != 0) {
3686 				/* Not corrupted - process it and continue with iterating through blobs */
3687 				if (ctx->iter_cb_fn) {
3688 					ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
3689 				}
3690 				bs_blob_list_add(blob);
3691 				spdk_bs_iter_next(ctx->bs, blob, bs_load_iter, ctx);
3692 				return;
3693 			}
3694 
3695 		}
3696 
3697 		assert(len == sizeof(spdk_blob_id));
3698 
3699 		ctx->blob = blob;
3700 
3701 		/* Open clone to check if we are able to fix this blob or should we remove it */
3702 		spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, bs_examine_clone, ctx);
3703 		return;
3704 	} else if (bserrno == -ENOENT) {
3705 		bserrno = 0;
3706 	} else {
3707 		/*
3708 		 * This case needs to be looked at further.  Same problem
3709 		 *  exists with applications that rely on explicit blob
3710 		 *  iteration.  We should just skip the blob that failed
3711 		 *  to load and continue on to the next one.
3712 		 */
3713 		SPDK_ERRLOG("Error in iterating blobs\n");
3714 	}
3715 
3716 	ctx->iter_cb_fn = NULL;
3717 
3718 	spdk_free(ctx->super);
3719 	spdk_free(ctx->mask);
3720 	bs_sequence_finish(ctx->seq, bserrno);
3721 	free(ctx);
3722 }
3723 
3724 static void bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3725 
3726 static void
3727 bs_load_complete(struct spdk_bs_load_ctx *ctx)
3728 {
3729 	ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters);
3730 	if (ctx->dumping) {
3731 		bs_dump_read_md_page(ctx->seq, ctx);
3732 		return;
3733 	}
3734 	spdk_bs_iter_first(ctx->bs, bs_load_iter, ctx);
3735 }
3736 
3737 static void
3738 bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3739 {
3740 	struct spdk_bs_load_ctx *ctx = cb_arg;
3741 	int rc;
3742 
3743 	/* The type must be correct */
3744 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
3745 
3746 	/* The length of the mask (in bits) must not be greater than
3747 	 * the length of the buffer (converted to bits) */
3748 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
3749 
3750 	/* The length of the mask must be exactly equal to the size
3751 	 * (in pages) of the metadata region */
3752 	assert(ctx->mask->length == ctx->super->md_len);
3753 
3754 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->mask->length);
3755 	if (rc < 0) {
3756 		spdk_free(ctx->mask);
3757 		bs_load_ctx_fail(ctx, rc);
3758 		return;
3759 	}
3760 
3761 	spdk_bit_array_load_mask(ctx->bs->used_blobids, ctx->mask->mask);
3762 	bs_load_complete(ctx);
3763 }
3764 
3765 static void
3766 bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3767 {
3768 	struct spdk_bs_load_ctx *ctx = cb_arg;
3769 	uint64_t		lba, lba_count, mask_size;
3770 	int			rc;
3771 
3772 	if (bserrno != 0) {
3773 		bs_load_ctx_fail(ctx, bserrno);
3774 		return;
3775 	}
3776 
3777 	/* The type must be correct */
3778 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
3779 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
3780 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
3781 					     struct spdk_blob_md_page) * 8));
3782 	/* The length of the mask must be exactly equal to the total number of clusters */
3783 	assert(ctx->mask->length == ctx->bs->total_clusters);
3784 
3785 	rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->mask->length);
3786 	if (rc < 0) {
3787 		spdk_free(ctx->mask);
3788 		bs_load_ctx_fail(ctx, rc);
3789 		return;
3790 	}
3791 
3792 	spdk_bit_array_load_mask(ctx->used_clusters, ctx->mask->mask);
3793 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->used_clusters);
3794 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
3795 
3796 	spdk_free(ctx->mask);
3797 
3798 	/* Read the used blobids mask */
3799 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
3800 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3801 				 SPDK_MALLOC_DMA);
3802 	if (!ctx->mask) {
3803 		bs_load_ctx_fail(ctx, -ENOMEM);
3804 		return;
3805 	}
3806 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
3807 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
3808 	bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3809 			     bs_load_used_blobids_cpl, ctx);
3810 }
3811 
3812 static void
3813 bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3814 {
3815 	struct spdk_bs_load_ctx *ctx = cb_arg;
3816 	uint64_t		lba, lba_count, mask_size;
3817 	int			rc;
3818 
3819 	if (bserrno != 0) {
3820 		bs_load_ctx_fail(ctx, bserrno);
3821 		return;
3822 	}
3823 
3824 	/* The type must be correct */
3825 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
3826 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
3827 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
3828 				     8));
3829 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
3830 	if (ctx->mask->length != ctx->super->md_len) {
3831 		SPDK_ERRLOG("mismatched md_len in used_pages mask: "
3832 			    "mask->length=%" PRIu32 " super->md_len=%" PRIu32 "\n",
3833 			    ctx->mask->length, ctx->super->md_len);
3834 		assert(false);
3835 	}
3836 
3837 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->mask->length);
3838 	if (rc < 0) {
3839 		spdk_free(ctx->mask);
3840 		bs_load_ctx_fail(ctx, rc);
3841 		return;
3842 	}
3843 
3844 	spdk_bit_array_load_mask(ctx->bs->used_md_pages, ctx->mask->mask);
3845 	spdk_free(ctx->mask);
3846 
3847 	/* Read the used clusters mask */
3848 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
3849 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3850 				 SPDK_MALLOC_DMA);
3851 	if (!ctx->mask) {
3852 		bs_load_ctx_fail(ctx, -ENOMEM);
3853 		return;
3854 	}
3855 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
3856 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
3857 	bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3858 			     bs_load_used_clusters_cpl, ctx);
3859 }
3860 
3861 static void
3862 bs_load_read_used_pages(struct spdk_bs_load_ctx *ctx)
3863 {
3864 	uint64_t lba, lba_count, mask_size;
3865 
3866 	/* Read the used pages mask */
3867 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
3868 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3869 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3870 	if (!ctx->mask) {
3871 		bs_load_ctx_fail(ctx, -ENOMEM);
3872 		return;
3873 	}
3874 
3875 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3876 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3877 	bs_sequence_read_dev(ctx->seq, ctx->mask, lba, lba_count,
3878 			     bs_load_used_pages_cpl, ctx);
3879 }
3880 
3881 static int
3882 bs_load_replay_md_parse_page(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_page *page)
3883 {
3884 	struct spdk_blob_store *bs = ctx->bs;
3885 	struct spdk_blob_md_descriptor *desc;
3886 	size_t	cur_desc = 0;
3887 
3888 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3889 	while (cur_desc < sizeof(page->descriptors)) {
3890 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3891 			if (desc->length == 0) {
3892 				/* If padding and length are 0, this terminates the page */
3893 				break;
3894 			}
3895 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
3896 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
3897 			unsigned int				i, j;
3898 			unsigned int				cluster_count = 0;
3899 			uint32_t				cluster_idx;
3900 
3901 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
3902 
3903 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
3904 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
3905 					cluster_idx = desc_extent_rle->extents[i].cluster_idx;
3906 					/*
3907 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
3908 					 * in the used cluster map.
3909 					 */
3910 					if (cluster_idx != 0) {
3911 						SPDK_NOTICELOG("Recover: cluster %" PRIu32 "\n", cluster_idx + j);
3912 						spdk_bit_array_set(ctx->used_clusters, cluster_idx + j);
3913 						if (bs->num_free_clusters == 0) {
3914 							return -ENOSPC;
3915 						}
3916 						bs->num_free_clusters--;
3917 					}
3918 					cluster_count++;
3919 				}
3920 			}
3921 			if (cluster_count == 0) {
3922 				return -EINVAL;
3923 			}
3924 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
3925 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
3926 			uint32_t					i;
3927 			uint32_t					cluster_count = 0;
3928 			uint32_t					cluster_idx;
3929 			size_t						cluster_idx_length;
3930 
3931 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
3932 			cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx);
3933 
3934 			if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) ||
3935 			    (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) {
3936 				return -EINVAL;
3937 			}
3938 
3939 			for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
3940 				cluster_idx = desc_extent->cluster_idx[i];
3941 				/*
3942 				 * cluster_idx = 0 means an unallocated cluster - don't mark that
3943 				 * in the used cluster map.
3944 				 */
3945 				if (cluster_idx != 0) {
3946 					if (cluster_idx < desc_extent->start_cluster_idx &&
3947 					    cluster_idx >= desc_extent->start_cluster_idx + cluster_count) {
3948 						return -EINVAL;
3949 					}
3950 					spdk_bit_array_set(ctx->used_clusters, cluster_idx);
3951 					if (bs->num_free_clusters == 0) {
3952 						return -ENOSPC;
3953 					}
3954 					bs->num_free_clusters--;
3955 				}
3956 				cluster_count++;
3957 			}
3958 
3959 			if (cluster_count == 0) {
3960 				return -EINVAL;
3961 			}
3962 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3963 			/* Skip this item */
3964 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3965 			/* Skip this item */
3966 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3967 			/* Skip this item */
3968 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
3969 			struct spdk_blob_md_descriptor_extent_table *desc_extent_table;
3970 			uint32_t num_extent_pages = ctx->num_extent_pages;
3971 			uint32_t i;
3972 			size_t extent_pages_length;
3973 			void *tmp;
3974 
3975 			desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc;
3976 			extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters);
3977 
3978 			if (desc_extent_table->length == 0 ||
3979 			    (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) {
3980 				return -EINVAL;
3981 			}
3982 
3983 			for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
3984 				if (desc_extent_table->extent_page[i].page_idx != 0) {
3985 					if (desc_extent_table->extent_page[i].num_pages != 1) {
3986 						return -EINVAL;
3987 					}
3988 					num_extent_pages += 1;
3989 				}
3990 			}
3991 
3992 			if (num_extent_pages > 0) {
3993 				tmp = realloc(ctx->extent_page_num, num_extent_pages * sizeof(uint32_t));
3994 				if (tmp == NULL) {
3995 					return -ENOMEM;
3996 				}
3997 				ctx->extent_page_num = tmp;
3998 
3999 				/* Extent table entries contain md page numbers for extent pages.
4000 				 * Zeroes represent unallocated extent pages, those are run-length-encoded.
4001 				 */
4002 				for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
4003 					if (desc_extent_table->extent_page[i].page_idx != 0) {
4004 						ctx->extent_page_num[ctx->num_extent_pages] = desc_extent_table->extent_page[i].page_idx;
4005 						ctx->num_extent_pages += 1;
4006 					}
4007 				}
4008 			}
4009 		} else {
4010 			/* Error */
4011 			return -EINVAL;
4012 		}
4013 		/* Advance to the next descriptor */
4014 		cur_desc += sizeof(*desc) + desc->length;
4015 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
4016 			break;
4017 		}
4018 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
4019 	}
4020 	return 0;
4021 }
4022 
4023 static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page)
4024 {
4025 	uint32_t crc;
4026 	struct spdk_blob_md_descriptor *desc = (struct spdk_blob_md_descriptor *)page->descriptors;
4027 	size_t desc_len;
4028 
4029 	crc = blob_md_page_calc_crc(page);
4030 	if (crc != page->crc) {
4031 		return false;
4032 	}
4033 
4034 	/* Extent page should always be of sequence num 0. */
4035 	if (page->sequence_num != 0) {
4036 		return false;
4037 	}
4038 
4039 	/* Descriptor type must be EXTENT_PAGE. */
4040 	if (desc->type != SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
4041 		return false;
4042 	}
4043 
4044 	/* Descriptor length cannot exceed the page. */
4045 	desc_len = sizeof(*desc) + desc->length;
4046 	if (desc_len > sizeof(page->descriptors)) {
4047 		return false;
4048 	}
4049 
4050 	/* It has to be the only descriptor in the page. */
4051 	if (desc_len + sizeof(*desc) <= sizeof(page->descriptors)) {
4052 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + desc_len);
4053 		if (desc->length != 0) {
4054 			return false;
4055 		}
4056 	}
4057 
4058 	return true;
4059 }
4060 
4061 static bool bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
4062 {
4063 	uint32_t crc;
4064 	struct spdk_blob_md_page *page = ctx->page;
4065 
4066 	crc = blob_md_page_calc_crc(page);
4067 	if (crc != page->crc) {
4068 		return false;
4069 	}
4070 
4071 	/* First page of a sequence should match the blobid. */
4072 	if (page->sequence_num == 0 &&
4073 	    bs_page_to_blobid(ctx->cur_page) != page->id) {
4074 		return false;
4075 	}
4076 	assert(bs_load_cur_extent_page_valid(page) == false);
4077 
4078 	return true;
4079 }
4080 
4081 static void
4082 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx);
4083 
4084 static void
4085 bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4086 {
4087 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4088 
4089 	if (bserrno != 0) {
4090 		bs_load_ctx_fail(ctx, bserrno);
4091 		return;
4092 	}
4093 
4094 	bs_load_complete(ctx);
4095 }
4096 
4097 static void
4098 bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4099 {
4100 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4101 
4102 	spdk_free(ctx->mask);
4103 	ctx->mask = NULL;
4104 
4105 	if (bserrno != 0) {
4106 		bs_load_ctx_fail(ctx, bserrno);
4107 		return;
4108 	}
4109 
4110 	bs_write_used_clusters(seq, ctx, bs_load_write_used_clusters_cpl);
4111 }
4112 
4113 static void
4114 bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4115 {
4116 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4117 
4118 	spdk_free(ctx->mask);
4119 	ctx->mask = NULL;
4120 
4121 	if (bserrno != 0) {
4122 		bs_load_ctx_fail(ctx, bserrno);
4123 		return;
4124 	}
4125 
4126 	bs_write_used_blobids(seq, ctx, bs_load_write_used_blobids_cpl);
4127 }
4128 
4129 static void
4130 bs_load_write_used_md(struct spdk_bs_load_ctx *ctx)
4131 {
4132 	bs_write_used_md(ctx->seq, ctx, bs_load_write_used_pages_cpl);
4133 }
4134 
4135 static void
4136 bs_load_replay_md_chain_cpl(struct spdk_bs_load_ctx *ctx)
4137 {
4138 	uint64_t num_md_clusters;
4139 	uint64_t i;
4140 
4141 	ctx->in_page_chain = false;
4142 
4143 	do {
4144 		ctx->page_index++;
4145 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
4146 
4147 	if (ctx->page_index < ctx->super->md_len) {
4148 		ctx->cur_page = ctx->page_index;
4149 		bs_load_replay_cur_md_page(ctx);
4150 	} else {
4151 		/* Claim all of the clusters used by the metadata */
4152 		num_md_clusters = spdk_divide_round_up(
4153 					  ctx->super->md_start + ctx->super->md_len, ctx->bs->pages_per_cluster);
4154 		for (i = 0; i < num_md_clusters; i++) {
4155 			spdk_bit_array_set(ctx->used_clusters, i);
4156 		}
4157 		ctx->bs->num_free_clusters -= num_md_clusters;
4158 		spdk_free(ctx->page);
4159 		bs_load_write_used_md(ctx);
4160 	}
4161 }
4162 
4163 static void
4164 bs_load_replay_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4165 {
4166 	struct spdk_bs_load_ctx *ctx = cb_arg;
4167 	uint32_t page_num;
4168 	uint64_t i;
4169 
4170 	if (bserrno != 0) {
4171 		spdk_free(ctx->extent_pages);
4172 		bs_load_ctx_fail(ctx, bserrno);
4173 		return;
4174 	}
4175 
4176 	for (i = 0; i < ctx->num_extent_pages; i++) {
4177 		/* Extent pages are only read when present within in chain md.
4178 		 * Integrity of md is not right if that page was not a valid extent page. */
4179 		if (bs_load_cur_extent_page_valid(&ctx->extent_pages[i]) != true) {
4180 			spdk_free(ctx->extent_pages);
4181 			bs_load_ctx_fail(ctx, -EILSEQ);
4182 			return;
4183 		}
4184 
4185 		page_num = ctx->extent_page_num[i];
4186 		spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
4187 		if (bs_load_replay_md_parse_page(ctx, &ctx->extent_pages[i])) {
4188 			spdk_free(ctx->extent_pages);
4189 			bs_load_ctx_fail(ctx, -EILSEQ);
4190 			return;
4191 		}
4192 	}
4193 
4194 	spdk_free(ctx->extent_pages);
4195 	free(ctx->extent_page_num);
4196 	ctx->extent_page_num = NULL;
4197 	ctx->num_extent_pages = 0;
4198 
4199 	bs_load_replay_md_chain_cpl(ctx);
4200 }
4201 
4202 static void
4203 bs_load_replay_extent_pages(struct spdk_bs_load_ctx *ctx)
4204 {
4205 	spdk_bs_batch_t *batch;
4206 	uint32_t page;
4207 	uint64_t lba;
4208 	uint64_t i;
4209 
4210 	ctx->extent_pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE * ctx->num_extent_pages, 0,
4211 					 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4212 	if (!ctx->extent_pages) {
4213 		bs_load_ctx_fail(ctx, -ENOMEM);
4214 		return;
4215 	}
4216 
4217 	batch = bs_sequence_to_batch(ctx->seq, bs_load_replay_extent_page_cpl, ctx);
4218 
4219 	for (i = 0; i < ctx->num_extent_pages; i++) {
4220 		page = ctx->extent_page_num[i];
4221 		assert(page < ctx->super->md_len);
4222 		lba = bs_md_page_to_lba(ctx->bs, page);
4223 		bs_batch_read_dev(batch, &ctx->extent_pages[i], lba,
4224 				  bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE));
4225 	}
4226 
4227 	bs_batch_close(batch);
4228 }
4229 
4230 static void
4231 bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4232 {
4233 	struct spdk_bs_load_ctx *ctx = cb_arg;
4234 	uint32_t page_num;
4235 	struct spdk_blob_md_page *page;
4236 
4237 	if (bserrno != 0) {
4238 		bs_load_ctx_fail(ctx, bserrno);
4239 		return;
4240 	}
4241 
4242 	page_num = ctx->cur_page;
4243 	page = ctx->page;
4244 	if (bs_load_cur_md_page_valid(ctx) == true) {
4245 		if (page->sequence_num == 0 || ctx->in_page_chain == true) {
4246 			bs_claim_md_page(ctx->bs, page_num);
4247 			if (page->sequence_num == 0) {
4248 				SPDK_NOTICELOG("Recover: blob %" PRIu32 "\n", page_num);
4249 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
4250 			}
4251 			if (bs_load_replay_md_parse_page(ctx, page)) {
4252 				bs_load_ctx_fail(ctx, -EILSEQ);
4253 				return;
4254 			}
4255 			if (page->next != SPDK_INVALID_MD_PAGE) {
4256 				ctx->in_page_chain = true;
4257 				ctx->cur_page = page->next;
4258 				bs_load_replay_cur_md_page(ctx);
4259 				return;
4260 			}
4261 			if (ctx->num_extent_pages != 0) {
4262 				bs_load_replay_extent_pages(ctx);
4263 				return;
4264 			}
4265 		}
4266 	}
4267 	bs_load_replay_md_chain_cpl(ctx);
4268 }
4269 
4270 static void
4271 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx)
4272 {
4273 	uint64_t lba;
4274 
4275 	assert(ctx->cur_page < ctx->super->md_len);
4276 	lba = bs_md_page_to_lba(ctx->bs, ctx->cur_page);
4277 	bs_sequence_read_dev(ctx->seq, ctx->page, lba,
4278 			     bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
4279 			     bs_load_replay_md_cpl, ctx);
4280 }
4281 
4282 static void
4283 bs_load_replay_md(struct spdk_bs_load_ctx *ctx)
4284 {
4285 	ctx->page_index = 0;
4286 	ctx->cur_page = 0;
4287 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0,
4288 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4289 	if (!ctx->page) {
4290 		bs_load_ctx_fail(ctx, -ENOMEM);
4291 		return;
4292 	}
4293 	bs_load_replay_cur_md_page(ctx);
4294 }
4295 
4296 static void
4297 bs_recover(struct spdk_bs_load_ctx *ctx)
4298 {
4299 	int		rc;
4300 
4301 	SPDK_NOTICELOG("Performing recovery on blobstore\n");
4302 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
4303 	if (rc < 0) {
4304 		bs_load_ctx_fail(ctx, -ENOMEM);
4305 		return;
4306 	}
4307 
4308 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
4309 	if (rc < 0) {
4310 		bs_load_ctx_fail(ctx, -ENOMEM);
4311 		return;
4312 	}
4313 
4314 	rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters);
4315 	if (rc < 0) {
4316 		bs_load_ctx_fail(ctx, -ENOMEM);
4317 		return;
4318 	}
4319 
4320 	rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->super->md_len);
4321 	if (rc < 0) {
4322 		bs_load_ctx_fail(ctx, -ENOMEM);
4323 		return;
4324 	}
4325 
4326 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
4327 	bs_load_replay_md(ctx);
4328 }
4329 
4330 static int
4331 bs_parse_super(struct spdk_bs_load_ctx *ctx)
4332 {
4333 	int rc;
4334 
4335 	if (ctx->super->size == 0) {
4336 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
4337 	}
4338 
4339 	if (ctx->super->io_unit_size == 0) {
4340 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
4341 	}
4342 
4343 	ctx->bs->clean = 1;
4344 	ctx->bs->cluster_sz = ctx->super->cluster_size;
4345 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
4346 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
4347 	if (spdk_u32_is_pow2(ctx->bs->pages_per_cluster)) {
4348 		ctx->bs->pages_per_cluster_shift = spdk_u32log2(ctx->bs->pages_per_cluster);
4349 	}
4350 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
4351 	rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters);
4352 	if (rc < 0) {
4353 		return -ENOMEM;
4354 	}
4355 	ctx->bs->md_start = ctx->super->md_start;
4356 	ctx->bs->md_len = ctx->super->md_len;
4357 	rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->bs->md_len);
4358 	if (rc < 0) {
4359 		return -ENOMEM;
4360 	}
4361 
4362 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
4363 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
4364 	ctx->bs->super_blob = ctx->super->super_blob;
4365 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
4366 
4367 	return 0;
4368 }
4369 
4370 static void
4371 bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4372 {
4373 	struct spdk_bs_load_ctx *ctx = cb_arg;
4374 	uint32_t	crc;
4375 	int		rc;
4376 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
4377 
4378 	if (ctx->super->version > SPDK_BS_VERSION ||
4379 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
4380 		bs_load_ctx_fail(ctx, -EILSEQ);
4381 		return;
4382 	}
4383 
4384 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4385 		   sizeof(ctx->super->signature)) != 0) {
4386 		bs_load_ctx_fail(ctx, -EILSEQ);
4387 		return;
4388 	}
4389 
4390 	crc = blob_md_page_calc_crc(ctx->super);
4391 	if (crc != ctx->super->crc) {
4392 		bs_load_ctx_fail(ctx, -EILSEQ);
4393 		return;
4394 	}
4395 
4396 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
4397 		SPDK_DEBUGLOG(blob, "Bstype matched - loading blobstore\n");
4398 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
4399 		SPDK_DEBUGLOG(blob, "Bstype wildcard used - loading blobstore regardless bstype\n");
4400 	} else {
4401 		SPDK_DEBUGLOG(blob, "Unexpected bstype\n");
4402 		SPDK_LOGDUMP(blob, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
4403 		SPDK_LOGDUMP(blob, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
4404 		bs_load_ctx_fail(ctx, -ENXIO);
4405 		return;
4406 	}
4407 
4408 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
4409 		SPDK_NOTICELOG("Size mismatch, dev size: %" PRIu64 ", blobstore size: %" PRIu64 "\n",
4410 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
4411 		bs_load_ctx_fail(ctx, -EILSEQ);
4412 		return;
4413 	}
4414 
4415 	rc = bs_parse_super(ctx);
4416 	if (rc < 0) {
4417 		bs_load_ctx_fail(ctx, rc);
4418 		return;
4419 	}
4420 
4421 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0 || ctx->force_recover) {
4422 		bs_recover(ctx);
4423 	} else {
4424 		bs_load_read_used_pages(ctx);
4425 	}
4426 }
4427 
4428 static inline int
4429 bs_opts_copy(struct spdk_bs_opts *src, struct spdk_bs_opts *dst)
4430 {
4431 
4432 	if (!src->opts_size) {
4433 		SPDK_ERRLOG("opts_size should not be zero value\n");
4434 		return -1;
4435 	}
4436 
4437 #define FIELD_OK(field) \
4438         offsetof(struct spdk_bs_opts, field) + sizeof(src->field) <= src->opts_size
4439 
4440 #define SET_FIELD(field) \
4441         if (FIELD_OK(field)) { \
4442                 dst->field = src->field; \
4443         } \
4444 
4445 	SET_FIELD(cluster_sz);
4446 	SET_FIELD(num_md_pages);
4447 	SET_FIELD(max_md_ops);
4448 	SET_FIELD(max_channel_ops);
4449 	SET_FIELD(clear_method);
4450 
4451 	if (FIELD_OK(bstype)) {
4452 		memcpy(&dst->bstype, &src->bstype, sizeof(dst->bstype));
4453 	}
4454 	SET_FIELD(iter_cb_fn);
4455 	SET_FIELD(iter_cb_arg);
4456 	SET_FIELD(force_recover);
4457 
4458 	dst->opts_size = src->opts_size;
4459 
4460 	/* You should not remove this statement, but need to update the assert statement
4461 	 * if you add a new field, and also add a corresponding SET_FIELD statement */
4462 	SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_opts) == 72, "Incorrect size");
4463 
4464 #undef FIELD_OK
4465 #undef SET_FIELD
4466 
4467 	return 0;
4468 }
4469 
4470 void
4471 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
4472 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
4473 {
4474 	struct spdk_blob_store	*bs;
4475 	struct spdk_bs_cpl	cpl;
4476 	struct spdk_bs_load_ctx *ctx;
4477 	struct spdk_bs_opts	opts = {};
4478 	int err;
4479 
4480 	SPDK_DEBUGLOG(blob, "Loading blobstore from dev %p\n", dev);
4481 
4482 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
4483 		SPDK_DEBUGLOG(blob, "unsupported dev block length of %d\n", dev->blocklen);
4484 		dev->destroy(dev);
4485 		cb_fn(cb_arg, NULL, -EINVAL);
4486 		return;
4487 	}
4488 
4489 	spdk_bs_opts_init(&opts, sizeof(opts));
4490 	if (o) {
4491 		if (bs_opts_copy(o, &opts)) {
4492 			return;
4493 		}
4494 	}
4495 
4496 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
4497 		dev->destroy(dev);
4498 		cb_fn(cb_arg, NULL, -EINVAL);
4499 		return;
4500 	}
4501 
4502 	err = bs_alloc(dev, &opts, &bs, &ctx);
4503 	if (err) {
4504 		dev->destroy(dev);
4505 		cb_fn(cb_arg, NULL, err);
4506 		return;
4507 	}
4508 
4509 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
4510 	cpl.u.bs_handle.cb_fn = cb_fn;
4511 	cpl.u.bs_handle.cb_arg = cb_arg;
4512 	cpl.u.bs_handle.bs = bs;
4513 
4514 	ctx->seq = bs_sequence_start(bs->md_channel, &cpl);
4515 	if (!ctx->seq) {
4516 		spdk_free(ctx->super);
4517 		free(ctx);
4518 		bs_free(bs);
4519 		cb_fn(cb_arg, NULL, -ENOMEM);
4520 		return;
4521 	}
4522 
4523 	/* Read the super block */
4524 	bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
4525 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
4526 			     bs_load_super_cpl, ctx);
4527 }
4528 
4529 /* END spdk_bs_load */
4530 
4531 /* START spdk_bs_dump */
4532 
4533 static void
4534 bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
4535 {
4536 	spdk_free(ctx->super);
4537 
4538 	/*
4539 	 * We need to defer calling bs_call_cpl() until after
4540 	 * dev destruction, so tuck these away for later use.
4541 	 */
4542 	ctx->bs->unload_err = bserrno;
4543 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
4544 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
4545 
4546 	bs_sequence_finish(seq, 0);
4547 	bs_free(ctx->bs);
4548 	free(ctx);
4549 }
4550 
4551 static void
4552 bs_dump_print_md_page(struct spdk_bs_load_ctx *ctx)
4553 {
4554 	uint32_t page_idx = ctx->cur_page;
4555 	struct spdk_blob_md_page *page = ctx->page;
4556 	struct spdk_blob_md_descriptor *desc;
4557 	size_t cur_desc = 0;
4558 	uint32_t crc;
4559 
4560 	fprintf(ctx->fp, "=========\n");
4561 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
4562 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
4563 	fprintf(ctx->fp, "Sequence: %" PRIu32 "\n", page->sequence_num);
4564 	if (page->next == SPDK_INVALID_MD_PAGE) {
4565 		fprintf(ctx->fp, "Next: None\n");
4566 	} else {
4567 		fprintf(ctx->fp, "Next: %" PRIu32 "\n", page->next);
4568 	}
4569 	fprintf(ctx->fp, "In used bit array%s:", ctx->super->clean ? "" : " (not clean: dubious)");
4570 	if (spdk_bit_array_get(ctx->bs->used_md_pages, page_idx)) {
4571 		fprintf(ctx->fp, " md");
4572 	}
4573 	if (spdk_bit_array_get(ctx->bs->used_blobids, page_idx)) {
4574 		fprintf(ctx->fp, " blob");
4575 	}
4576 	fprintf(ctx->fp, "\n");
4577 
4578 	crc = blob_md_page_calc_crc(page);
4579 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
4580 
4581 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
4582 	while (cur_desc < sizeof(page->descriptors)) {
4583 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
4584 			if (desc->length == 0) {
4585 				/* If padding and length are 0, this terminates the page */
4586 				break;
4587 			}
4588 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
4589 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
4590 			unsigned int				i;
4591 
4592 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
4593 
4594 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
4595 				if (desc_extent_rle->extents[i].cluster_idx != 0) {
4596 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
4597 						desc_extent_rle->extents[i].cluster_idx);
4598 				} else {
4599 					fprintf(ctx->fp, "Unallocated Extent - ");
4600 				}
4601 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent_rle->extents[i].length);
4602 				fprintf(ctx->fp, "\n");
4603 			}
4604 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
4605 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
4606 			unsigned int					i;
4607 
4608 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
4609 
4610 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->cluster_idx[0]); i++) {
4611 				if (desc_extent->cluster_idx[i] != 0) {
4612 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
4613 						desc_extent->cluster_idx[i]);
4614 				} else {
4615 					fprintf(ctx->fp, "Unallocated Extent");
4616 				}
4617 				fprintf(ctx->fp, "\n");
4618 			}
4619 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
4620 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
4621 			uint32_t i;
4622 
4623 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
4624 
4625 			if (desc_xattr->length !=
4626 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
4627 			    desc_xattr->name_length + desc_xattr->value_length) {
4628 			}
4629 
4630 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
4631 			ctx->xattr_name[desc_xattr->name_length] = '\0';
4632 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
4633 			fprintf(ctx->fp, "       value = \"");
4634 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
4635 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
4636 					    desc_xattr->value_length);
4637 			fprintf(ctx->fp, "\"\n");
4638 			for (i = 0; i < desc_xattr->value_length; i++) {
4639 				if (i % 16 == 0) {
4640 					fprintf(ctx->fp, "               ");
4641 				}
4642 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
4643 				if ((i + 1) % 16 == 0) {
4644 					fprintf(ctx->fp, "\n");
4645 				}
4646 			}
4647 			if (i % 16 != 0) {
4648 				fprintf(ctx->fp, "\n");
4649 			}
4650 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
4651 			/* TODO */
4652 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
4653 			/* TODO */
4654 		} else {
4655 			/* Error */
4656 		}
4657 		/* Advance to the next descriptor */
4658 		cur_desc += sizeof(*desc) + desc->length;
4659 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
4660 			break;
4661 		}
4662 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
4663 	}
4664 }
4665 
4666 static void
4667 bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4668 {
4669 	struct spdk_bs_load_ctx *ctx = cb_arg;
4670 
4671 	if (bserrno != 0) {
4672 		bs_dump_finish(seq, ctx, bserrno);
4673 		return;
4674 	}
4675 
4676 	if (ctx->page->id != 0) {
4677 		bs_dump_print_md_page(ctx);
4678 	}
4679 
4680 	ctx->cur_page++;
4681 
4682 	if (ctx->cur_page < ctx->super->md_len) {
4683 		bs_dump_read_md_page(seq, ctx);
4684 	} else {
4685 		spdk_free(ctx->page);
4686 		bs_dump_finish(seq, ctx, 0);
4687 	}
4688 }
4689 
4690 static void
4691 bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
4692 {
4693 	struct spdk_bs_load_ctx *ctx = cb_arg;
4694 	uint64_t lba;
4695 
4696 	assert(ctx->cur_page < ctx->super->md_len);
4697 	lba = bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
4698 	bs_sequence_read_dev(seq, ctx->page, lba,
4699 			     bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
4700 			     bs_dump_read_md_page_cpl, ctx);
4701 }
4702 
4703 static void
4704 bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4705 {
4706 	struct spdk_bs_load_ctx *ctx = cb_arg;
4707 	int rc;
4708 
4709 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
4710 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4711 		   sizeof(ctx->super->signature)) != 0) {
4712 		fprintf(ctx->fp, "(Mismatch)\n");
4713 		bs_dump_finish(seq, ctx, bserrno);
4714 		return;
4715 	} else {
4716 		fprintf(ctx->fp, "(OK)\n");
4717 	}
4718 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
4719 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
4720 		(ctx->super->crc == blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
4721 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
4722 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
4723 	fprintf(ctx->fp, "Super Blob ID: ");
4724 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
4725 		fprintf(ctx->fp, "(None)\n");
4726 	} else {
4727 		fprintf(ctx->fp, "0x%" PRIx64 "\n", ctx->super->super_blob);
4728 	}
4729 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
4730 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
4731 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
4732 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
4733 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
4734 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
4735 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
4736 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
4737 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
4738 
4739 	ctx->cur_page = 0;
4740 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0,
4741 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4742 	if (!ctx->page) {
4743 		bs_dump_finish(seq, ctx, -ENOMEM);
4744 		return;
4745 	}
4746 
4747 	rc = bs_parse_super(ctx);
4748 	if (rc < 0) {
4749 		bs_load_ctx_fail(ctx, rc);
4750 		return;
4751 	}
4752 
4753 	bs_load_read_used_pages(ctx);
4754 }
4755 
4756 void
4757 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
4758 	     spdk_bs_op_complete cb_fn, void *cb_arg)
4759 {
4760 	struct spdk_blob_store	*bs;
4761 	struct spdk_bs_cpl	cpl;
4762 	struct spdk_bs_load_ctx *ctx;
4763 	struct spdk_bs_opts	opts = {};
4764 	int err;
4765 
4766 	SPDK_DEBUGLOG(blob, "Dumping blobstore from dev %p\n", dev);
4767 
4768 	spdk_bs_opts_init(&opts, sizeof(opts));
4769 
4770 	err = bs_alloc(dev, &opts, &bs, &ctx);
4771 	if (err) {
4772 		dev->destroy(dev);
4773 		cb_fn(cb_arg, err);
4774 		return;
4775 	}
4776 
4777 	ctx->dumping = true;
4778 	ctx->fp = fp;
4779 	ctx->print_xattr_fn = print_xattr_fn;
4780 
4781 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4782 	cpl.u.bs_basic.cb_fn = cb_fn;
4783 	cpl.u.bs_basic.cb_arg = cb_arg;
4784 
4785 	ctx->seq = bs_sequence_start(bs->md_channel, &cpl);
4786 	if (!ctx->seq) {
4787 		spdk_free(ctx->super);
4788 		free(ctx);
4789 		bs_free(bs);
4790 		cb_fn(cb_arg, -ENOMEM);
4791 		return;
4792 	}
4793 
4794 	/* Read the super block */
4795 	bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
4796 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
4797 			     bs_dump_super_cpl, ctx);
4798 }
4799 
4800 /* END spdk_bs_dump */
4801 
4802 /* START spdk_bs_init */
4803 
4804 static void
4805 bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4806 {
4807 	struct spdk_bs_load_ctx *ctx = cb_arg;
4808 
4809 	ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters);
4810 	spdk_free(ctx->super);
4811 	free(ctx);
4812 
4813 	bs_sequence_finish(seq, bserrno);
4814 }
4815 
4816 static void
4817 bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4818 {
4819 	struct spdk_bs_load_ctx *ctx = cb_arg;
4820 
4821 	/* Write super block */
4822 	bs_sequence_write_dev(seq, ctx->super, bs_page_to_lba(ctx->bs, 0),
4823 			      bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
4824 			      bs_init_persist_super_cpl, ctx);
4825 }
4826 
4827 void
4828 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
4829 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
4830 {
4831 	struct spdk_bs_load_ctx *ctx;
4832 	struct spdk_blob_store	*bs;
4833 	struct spdk_bs_cpl	cpl;
4834 	spdk_bs_sequence_t	*seq;
4835 	spdk_bs_batch_t		*batch;
4836 	uint64_t		num_md_lba;
4837 	uint64_t		num_md_pages;
4838 	uint64_t		num_md_clusters;
4839 	uint32_t		i;
4840 	struct spdk_bs_opts	opts = {};
4841 	int			rc;
4842 	uint64_t		lba, lba_count;
4843 
4844 	SPDK_DEBUGLOG(blob, "Initializing blobstore on dev %p\n", dev);
4845 
4846 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
4847 		SPDK_ERRLOG("unsupported dev block length of %d\n",
4848 			    dev->blocklen);
4849 		dev->destroy(dev);
4850 		cb_fn(cb_arg, NULL, -EINVAL);
4851 		return;
4852 	}
4853 
4854 	spdk_bs_opts_init(&opts, sizeof(opts));
4855 	if (o) {
4856 		if (bs_opts_copy(o, &opts)) {
4857 			return;
4858 		}
4859 	}
4860 
4861 	if (bs_opts_verify(&opts) != 0) {
4862 		dev->destroy(dev);
4863 		cb_fn(cb_arg, NULL, -EINVAL);
4864 		return;
4865 	}
4866 
4867 	rc = bs_alloc(dev, &opts, &bs, &ctx);
4868 	if (rc) {
4869 		dev->destroy(dev);
4870 		cb_fn(cb_arg, NULL, rc);
4871 		return;
4872 	}
4873 
4874 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
4875 		/* By default, allocate 1 page per cluster.
4876 		 * Technically, this over-allocates metadata
4877 		 * because more metadata will reduce the number
4878 		 * of usable clusters. This can be addressed with
4879 		 * more complex math in the future.
4880 		 */
4881 		bs->md_len = bs->total_clusters;
4882 	} else {
4883 		bs->md_len = opts.num_md_pages;
4884 	}
4885 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
4886 	if (rc < 0) {
4887 		spdk_free(ctx->super);
4888 		free(ctx);
4889 		bs_free(bs);
4890 		cb_fn(cb_arg, NULL, -ENOMEM);
4891 		return;
4892 	}
4893 
4894 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
4895 	if (rc < 0) {
4896 		spdk_free(ctx->super);
4897 		free(ctx);
4898 		bs_free(bs);
4899 		cb_fn(cb_arg, NULL, -ENOMEM);
4900 		return;
4901 	}
4902 
4903 	rc = spdk_bit_array_resize(&bs->open_blobids, bs->md_len);
4904 	if (rc < 0) {
4905 		spdk_free(ctx->super);
4906 		free(ctx);
4907 		bs_free(bs);
4908 		cb_fn(cb_arg, NULL, -ENOMEM);
4909 		return;
4910 	}
4911 
4912 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4913 	       sizeof(ctx->super->signature));
4914 	ctx->super->version = SPDK_BS_VERSION;
4915 	ctx->super->length = sizeof(*ctx->super);
4916 	ctx->super->super_blob = bs->super_blob;
4917 	ctx->super->clean = 0;
4918 	ctx->super->cluster_size = bs->cluster_sz;
4919 	ctx->super->io_unit_size = bs->io_unit_size;
4920 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
4921 
4922 	/* Calculate how many pages the metadata consumes at the front
4923 	 * of the disk.
4924 	 */
4925 
4926 	/* The super block uses 1 page */
4927 	num_md_pages = 1;
4928 
4929 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
4930 	 * up to the nearest page, plus a header.
4931 	 */
4932 	ctx->super->used_page_mask_start = num_md_pages;
4933 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
4934 					 spdk_divide_round_up(bs->md_len, 8),
4935 					 SPDK_BS_PAGE_SIZE);
4936 	num_md_pages += ctx->super->used_page_mask_len;
4937 
4938 	/* The used_clusters mask requires 1 bit per cluster, rounded
4939 	 * up to the nearest page, plus a header.
4940 	 */
4941 	ctx->super->used_cluster_mask_start = num_md_pages;
4942 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
4943 					    spdk_divide_round_up(bs->total_clusters, 8),
4944 					    SPDK_BS_PAGE_SIZE);
4945 	num_md_pages += ctx->super->used_cluster_mask_len;
4946 
4947 	/* The used_blobids mask requires 1 bit per metadata page, rounded
4948 	 * up to the nearest page, plus a header.
4949 	 */
4950 	ctx->super->used_blobid_mask_start = num_md_pages;
4951 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
4952 					   spdk_divide_round_up(bs->md_len, 8),
4953 					   SPDK_BS_PAGE_SIZE);
4954 	num_md_pages += ctx->super->used_blobid_mask_len;
4955 
4956 	/* The metadata region size was chosen above */
4957 	ctx->super->md_start = bs->md_start = num_md_pages;
4958 	ctx->super->md_len = bs->md_len;
4959 	num_md_pages += bs->md_len;
4960 
4961 	num_md_lba = bs_page_to_lba(bs, num_md_pages);
4962 
4963 	ctx->super->size = dev->blockcnt * dev->blocklen;
4964 
4965 	ctx->super->crc = blob_md_page_calc_crc(ctx->super);
4966 
4967 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
4968 	if (num_md_clusters > bs->total_clusters) {
4969 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
4970 			    "please decrease number of pages reserved for metadata "
4971 			    "or increase cluster size.\n");
4972 		spdk_free(ctx->super);
4973 		spdk_bit_array_free(&ctx->used_clusters);
4974 		free(ctx);
4975 		bs_free(bs);
4976 		cb_fn(cb_arg, NULL, -ENOMEM);
4977 		return;
4978 	}
4979 	/* Claim all of the clusters used by the metadata */
4980 	for (i = 0; i < num_md_clusters; i++) {
4981 		spdk_bit_array_set(ctx->used_clusters, i);
4982 	}
4983 
4984 	bs->num_free_clusters -= num_md_clusters;
4985 	bs->total_data_clusters = bs->num_free_clusters;
4986 
4987 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
4988 	cpl.u.bs_handle.cb_fn = cb_fn;
4989 	cpl.u.bs_handle.cb_arg = cb_arg;
4990 	cpl.u.bs_handle.bs = bs;
4991 
4992 	seq = bs_sequence_start(bs->md_channel, &cpl);
4993 	if (!seq) {
4994 		spdk_free(ctx->super);
4995 		free(ctx);
4996 		bs_free(bs);
4997 		cb_fn(cb_arg, NULL, -ENOMEM);
4998 		return;
4999 	}
5000 
5001 	batch = bs_sequence_to_batch(seq, bs_init_trim_cpl, ctx);
5002 
5003 	/* Clear metadata space */
5004 	bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
5005 
5006 	lba = num_md_lba;
5007 	lba_count = ctx->bs->dev->blockcnt - lba;
5008 	switch (opts.clear_method) {
5009 	case BS_CLEAR_WITH_UNMAP:
5010 		/* Trim data clusters */
5011 		bs_batch_unmap_dev(batch, lba, lba_count);
5012 		break;
5013 	case BS_CLEAR_WITH_WRITE_ZEROES:
5014 		/* Write_zeroes to data clusters */
5015 		bs_batch_write_zeroes_dev(batch, lba, lba_count);
5016 		break;
5017 	case BS_CLEAR_WITH_NONE:
5018 	default:
5019 		break;
5020 	}
5021 
5022 	bs_batch_close(batch);
5023 }
5024 
5025 /* END spdk_bs_init */
5026 
5027 /* START spdk_bs_destroy */
5028 
5029 static void
5030 bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5031 {
5032 	struct spdk_bs_load_ctx *ctx = cb_arg;
5033 	struct spdk_blob_store *bs = ctx->bs;
5034 
5035 	/*
5036 	 * We need to defer calling bs_call_cpl() until after
5037 	 * dev destruction, so tuck these away for later use.
5038 	 */
5039 	bs->unload_err = bserrno;
5040 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
5041 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
5042 
5043 	bs_sequence_finish(seq, bserrno);
5044 
5045 	bs_free(bs);
5046 	free(ctx);
5047 }
5048 
5049 void
5050 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
5051 		void *cb_arg)
5052 {
5053 	struct spdk_bs_cpl	cpl;
5054 	spdk_bs_sequence_t	*seq;
5055 	struct spdk_bs_load_ctx *ctx;
5056 
5057 	SPDK_DEBUGLOG(blob, "Destroying blobstore\n");
5058 
5059 	if (!RB_EMPTY(&bs->open_blobs)) {
5060 		SPDK_ERRLOG("Blobstore still has open blobs\n");
5061 		cb_fn(cb_arg, -EBUSY);
5062 		return;
5063 	}
5064 
5065 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5066 	cpl.u.bs_basic.cb_fn = cb_fn;
5067 	cpl.u.bs_basic.cb_arg = cb_arg;
5068 
5069 	ctx = calloc(1, sizeof(*ctx));
5070 	if (!ctx) {
5071 		cb_fn(cb_arg, -ENOMEM);
5072 		return;
5073 	}
5074 
5075 	ctx->bs = bs;
5076 
5077 	seq = bs_sequence_start(bs->md_channel, &cpl);
5078 	if (!seq) {
5079 		free(ctx);
5080 		cb_fn(cb_arg, -ENOMEM);
5081 		return;
5082 	}
5083 
5084 	/* Write zeroes to the super block */
5085 	bs_sequence_write_zeroes_dev(seq,
5086 				     bs_page_to_lba(bs, 0),
5087 				     bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
5088 				     bs_destroy_trim_cpl, ctx);
5089 }
5090 
5091 /* END spdk_bs_destroy */
5092 
5093 /* START spdk_bs_unload */
5094 
5095 static void
5096 bs_unload_finish(struct spdk_bs_load_ctx *ctx, int bserrno)
5097 {
5098 	spdk_bs_sequence_t *seq = ctx->seq;
5099 
5100 	spdk_free(ctx->super);
5101 
5102 	/*
5103 	 * We need to defer calling bs_call_cpl() until after
5104 	 * dev destruction, so tuck these away for later use.
5105 	 */
5106 	ctx->bs->unload_err = bserrno;
5107 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
5108 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
5109 
5110 	bs_sequence_finish(seq, bserrno);
5111 
5112 	bs_free(ctx->bs);
5113 	free(ctx);
5114 }
5115 
5116 static void
5117 bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5118 {
5119 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5120 
5121 	bs_unload_finish(ctx, bserrno);
5122 }
5123 
5124 static void
5125 bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5126 {
5127 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5128 
5129 	spdk_free(ctx->mask);
5130 
5131 	if (bserrno != 0) {
5132 		bs_unload_finish(ctx, bserrno);
5133 		return;
5134 	}
5135 
5136 	ctx->super->clean = 1;
5137 
5138 	bs_write_super(seq, ctx->bs, ctx->super, bs_unload_write_super_cpl, ctx);
5139 }
5140 
5141 static void
5142 bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5143 {
5144 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5145 
5146 	spdk_free(ctx->mask);
5147 	ctx->mask = NULL;
5148 
5149 	if (bserrno != 0) {
5150 		bs_unload_finish(ctx, bserrno);
5151 		return;
5152 	}
5153 
5154 	bs_write_used_clusters(seq, ctx, bs_unload_write_used_clusters_cpl);
5155 }
5156 
5157 static void
5158 bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5159 {
5160 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5161 
5162 	spdk_free(ctx->mask);
5163 	ctx->mask = NULL;
5164 
5165 	if (bserrno != 0) {
5166 		bs_unload_finish(ctx, bserrno);
5167 		return;
5168 	}
5169 
5170 	bs_write_used_blobids(seq, ctx, bs_unload_write_used_blobids_cpl);
5171 }
5172 
5173 static void
5174 bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5175 {
5176 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5177 
5178 	if (bserrno != 0) {
5179 		bs_unload_finish(ctx, bserrno);
5180 		return;
5181 	}
5182 
5183 	bs_write_used_md(seq, cb_arg, bs_unload_write_used_pages_cpl);
5184 }
5185 
5186 void
5187 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
5188 {
5189 	struct spdk_bs_cpl	cpl;
5190 	struct spdk_bs_load_ctx *ctx;
5191 
5192 	SPDK_DEBUGLOG(blob, "Syncing blobstore\n");
5193 
5194 	if (!RB_EMPTY(&bs->open_blobs)) {
5195 		SPDK_ERRLOG("Blobstore still has open blobs\n");
5196 		cb_fn(cb_arg, -EBUSY);
5197 		return;
5198 	}
5199 
5200 	ctx = calloc(1, sizeof(*ctx));
5201 	if (!ctx) {
5202 		cb_fn(cb_arg, -ENOMEM);
5203 		return;
5204 	}
5205 
5206 	ctx->bs = bs;
5207 
5208 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
5209 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
5210 	if (!ctx->super) {
5211 		free(ctx);
5212 		cb_fn(cb_arg, -ENOMEM);
5213 		return;
5214 	}
5215 
5216 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5217 	cpl.u.bs_basic.cb_fn = cb_fn;
5218 	cpl.u.bs_basic.cb_arg = cb_arg;
5219 
5220 	ctx->seq = bs_sequence_start(bs->md_channel, &cpl);
5221 	if (!ctx->seq) {
5222 		spdk_free(ctx->super);
5223 		free(ctx);
5224 		cb_fn(cb_arg, -ENOMEM);
5225 		return;
5226 	}
5227 
5228 	/* Read super block */
5229 	bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
5230 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
5231 			     bs_unload_read_super_cpl, ctx);
5232 }
5233 
5234 /* END spdk_bs_unload */
5235 
5236 /* START spdk_bs_set_super */
5237 
5238 struct spdk_bs_set_super_ctx {
5239 	struct spdk_blob_store		*bs;
5240 	struct spdk_bs_super_block	*super;
5241 };
5242 
5243 static void
5244 bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5245 {
5246 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
5247 
5248 	if (bserrno != 0) {
5249 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
5250 	}
5251 
5252 	spdk_free(ctx->super);
5253 
5254 	bs_sequence_finish(seq, bserrno);
5255 
5256 	free(ctx);
5257 }
5258 
5259 static void
5260 bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5261 {
5262 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
5263 
5264 	if (bserrno != 0) {
5265 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
5266 		spdk_free(ctx->super);
5267 		bs_sequence_finish(seq, bserrno);
5268 		free(ctx);
5269 		return;
5270 	}
5271 
5272 	bs_write_super(seq, ctx->bs, ctx->super, bs_set_super_write_cpl, ctx);
5273 }
5274 
5275 void
5276 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
5277 		  spdk_bs_op_complete cb_fn, void *cb_arg)
5278 {
5279 	struct spdk_bs_cpl		cpl;
5280 	spdk_bs_sequence_t		*seq;
5281 	struct spdk_bs_set_super_ctx	*ctx;
5282 
5283 	SPDK_DEBUGLOG(blob, "Setting super blob id on blobstore\n");
5284 
5285 	ctx = calloc(1, sizeof(*ctx));
5286 	if (!ctx) {
5287 		cb_fn(cb_arg, -ENOMEM);
5288 		return;
5289 	}
5290 
5291 	ctx->bs = bs;
5292 
5293 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
5294 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
5295 	if (!ctx->super) {
5296 		free(ctx);
5297 		cb_fn(cb_arg, -ENOMEM);
5298 		return;
5299 	}
5300 
5301 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5302 	cpl.u.bs_basic.cb_fn = cb_fn;
5303 	cpl.u.bs_basic.cb_arg = cb_arg;
5304 
5305 	seq = bs_sequence_start(bs->md_channel, &cpl);
5306 	if (!seq) {
5307 		spdk_free(ctx->super);
5308 		free(ctx);
5309 		cb_fn(cb_arg, -ENOMEM);
5310 		return;
5311 	}
5312 
5313 	bs->super_blob = blobid;
5314 
5315 	/* Read super block */
5316 	bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0),
5317 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
5318 			     bs_set_super_read_cpl, ctx);
5319 }
5320 
5321 /* END spdk_bs_set_super */
5322 
5323 void
5324 spdk_bs_get_super(struct spdk_blob_store *bs,
5325 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5326 {
5327 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
5328 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
5329 	} else {
5330 		cb_fn(cb_arg, bs->super_blob, 0);
5331 	}
5332 }
5333 
5334 uint64_t
5335 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
5336 {
5337 	return bs->cluster_sz;
5338 }
5339 
5340 uint64_t
5341 spdk_bs_get_page_size(struct spdk_blob_store *bs)
5342 {
5343 	return SPDK_BS_PAGE_SIZE;
5344 }
5345 
5346 uint64_t
5347 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
5348 {
5349 	return bs->io_unit_size;
5350 }
5351 
5352 uint64_t
5353 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
5354 {
5355 	return bs->num_free_clusters;
5356 }
5357 
5358 uint64_t
5359 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
5360 {
5361 	return bs->total_data_clusters;
5362 }
5363 
5364 static int
5365 bs_register_md_thread(struct spdk_blob_store *bs)
5366 {
5367 	bs->md_channel = spdk_get_io_channel(bs);
5368 	if (!bs->md_channel) {
5369 		SPDK_ERRLOG("Failed to get IO channel.\n");
5370 		return -1;
5371 	}
5372 
5373 	return 0;
5374 }
5375 
5376 static int
5377 bs_unregister_md_thread(struct spdk_blob_store *bs)
5378 {
5379 	spdk_put_io_channel(bs->md_channel);
5380 
5381 	return 0;
5382 }
5383 
5384 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
5385 {
5386 	assert(blob != NULL);
5387 
5388 	return blob->id;
5389 }
5390 
5391 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
5392 {
5393 	assert(blob != NULL);
5394 
5395 	return bs_cluster_to_page(blob->bs, blob->active.num_clusters);
5396 }
5397 
5398 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
5399 {
5400 	assert(blob != NULL);
5401 
5402 	return spdk_blob_get_num_pages(blob) * bs_io_unit_per_page(blob->bs);
5403 }
5404 
5405 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
5406 {
5407 	assert(blob != NULL);
5408 
5409 	return blob->active.num_clusters;
5410 }
5411 
5412 /* START spdk_bs_create_blob */
5413 
5414 static void
5415 bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5416 {
5417 	struct spdk_blob *blob = cb_arg;
5418 	uint32_t page_idx = bs_blobid_to_page(blob->id);
5419 
5420 	if (bserrno != 0) {
5421 		spdk_bit_array_clear(blob->bs->used_blobids, page_idx);
5422 		bs_release_md_page(blob->bs, page_idx);
5423 	}
5424 
5425 	blob_free(blob);
5426 
5427 	bs_sequence_finish(seq, bserrno);
5428 }
5429 
5430 static int
5431 blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
5432 		bool internal)
5433 {
5434 	uint64_t i;
5435 	size_t value_len = 0;
5436 	int rc;
5437 	const void *value = NULL;
5438 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
5439 		return -EINVAL;
5440 	}
5441 	for (i = 0; i < xattrs->count; i++) {
5442 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
5443 		if (value == NULL || value_len == 0) {
5444 			return -EINVAL;
5445 		}
5446 		rc = blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
5447 		if (rc < 0) {
5448 			return rc;
5449 		}
5450 	}
5451 	return 0;
5452 }
5453 
5454 static void
5455 blob_opts_copy(const struct spdk_blob_opts *src, struct spdk_blob_opts *dst)
5456 {
5457 #define FIELD_OK(field) \
5458         offsetof(struct spdk_blob_opts, field) + sizeof(src->field) <= src->opts_size
5459 
5460 #define SET_FIELD(field) \
5461         if (FIELD_OK(field)) { \
5462                 dst->field = src->field; \
5463         } \
5464 
5465 	SET_FIELD(num_clusters);
5466 	SET_FIELD(thin_provision);
5467 	SET_FIELD(clear_method);
5468 
5469 	if (FIELD_OK(xattrs)) {
5470 		memcpy(&dst->xattrs, &src->xattrs, sizeof(src->xattrs));
5471 	}
5472 
5473 	SET_FIELD(use_extent_table);
5474 
5475 	dst->opts_size = src->opts_size;
5476 
5477 	/* You should not remove this statement, but need to update the assert statement
5478 	 * if you add a new field, and also add a corresponding SET_FIELD statement */
5479 	SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_opts) == 64, "Incorrect size");
5480 
5481 #undef FIELD_OK
5482 #undef SET_FIELD
5483 }
5484 
5485 static void
5486 bs_create_blob(struct spdk_blob_store *bs,
5487 	       const struct spdk_blob_opts *opts,
5488 	       const struct spdk_blob_xattr_opts *internal_xattrs,
5489 	       spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5490 {
5491 	struct spdk_blob	*blob;
5492 	uint32_t		page_idx;
5493 	struct spdk_bs_cpl	cpl;
5494 	struct spdk_blob_opts	opts_local;
5495 	struct spdk_blob_xattr_opts internal_xattrs_default;
5496 	spdk_bs_sequence_t	*seq;
5497 	spdk_blob_id		id;
5498 	int rc;
5499 
5500 	assert(spdk_get_thread() == bs->md_thread);
5501 
5502 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
5503 	if (page_idx == UINT32_MAX) {
5504 		cb_fn(cb_arg, 0, -ENOMEM);
5505 		return;
5506 	}
5507 	spdk_bit_array_set(bs->used_blobids, page_idx);
5508 	bs_claim_md_page(bs, page_idx);
5509 
5510 	id = bs_page_to_blobid(page_idx);
5511 
5512 	SPDK_DEBUGLOG(blob, "Creating blob with id %" PRIu64 " at page %u\n", id, page_idx);
5513 
5514 	blob = blob_alloc(bs, id);
5515 	if (!blob) {
5516 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5517 		bs_release_md_page(bs, page_idx);
5518 		cb_fn(cb_arg, 0, -ENOMEM);
5519 		return;
5520 	}
5521 
5522 	spdk_blob_opts_init(&opts_local, sizeof(opts_local));
5523 	if (opts) {
5524 		blob_opts_copy(opts, &opts_local);
5525 	}
5526 
5527 	blob->use_extent_table = opts_local.use_extent_table;
5528 	if (blob->use_extent_table) {
5529 		blob->invalid_flags |= SPDK_BLOB_EXTENT_TABLE;
5530 	}
5531 
5532 	if (!internal_xattrs) {
5533 		blob_xattrs_init(&internal_xattrs_default);
5534 		internal_xattrs = &internal_xattrs_default;
5535 	}
5536 
5537 	rc = blob_set_xattrs(blob, &opts_local.xattrs, false);
5538 	if (rc < 0) {
5539 		blob_free(blob);
5540 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5541 		bs_release_md_page(bs, page_idx);
5542 		cb_fn(cb_arg, 0, rc);
5543 		return;
5544 	}
5545 
5546 	rc = blob_set_xattrs(blob, internal_xattrs, true);
5547 	if (rc < 0) {
5548 		blob_free(blob);
5549 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5550 		bs_release_md_page(bs, page_idx);
5551 		cb_fn(cb_arg, 0, rc);
5552 		return;
5553 	}
5554 
5555 	if (opts_local.thin_provision) {
5556 		blob_set_thin_provision(blob);
5557 	}
5558 
5559 	blob_set_clear_method(blob, opts_local.clear_method);
5560 
5561 	rc = blob_resize(blob, opts_local.num_clusters);
5562 	if (rc < 0) {
5563 		blob_free(blob);
5564 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5565 		bs_release_md_page(bs, page_idx);
5566 		cb_fn(cb_arg, 0, rc);
5567 		return;
5568 	}
5569 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
5570 	cpl.u.blobid.cb_fn = cb_fn;
5571 	cpl.u.blobid.cb_arg = cb_arg;
5572 	cpl.u.blobid.blobid = blob->id;
5573 
5574 	seq = bs_sequence_start(bs->md_channel, &cpl);
5575 	if (!seq) {
5576 		blob_free(blob);
5577 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5578 		bs_release_md_page(bs, page_idx);
5579 		cb_fn(cb_arg, 0, -ENOMEM);
5580 		return;
5581 	}
5582 
5583 	blob_persist(seq, blob, bs_create_blob_cpl, blob);
5584 }
5585 
5586 void spdk_bs_create_blob(struct spdk_blob_store *bs,
5587 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5588 {
5589 	bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
5590 }
5591 
5592 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
5593 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5594 {
5595 	bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
5596 }
5597 
5598 /* END spdk_bs_create_blob */
5599 
5600 /* START blob_cleanup */
5601 
5602 struct spdk_clone_snapshot_ctx {
5603 	struct spdk_bs_cpl      cpl;
5604 	int bserrno;
5605 	bool frozen;
5606 
5607 	struct spdk_io_channel *channel;
5608 
5609 	/* Current cluster for inflate operation */
5610 	uint64_t cluster;
5611 
5612 	/* For inflation force allocation of all unallocated clusters and remove
5613 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
5614 	bool allocate_all;
5615 
5616 	struct {
5617 		spdk_blob_id id;
5618 		struct spdk_blob *blob;
5619 		bool md_ro;
5620 	} original;
5621 	struct {
5622 		spdk_blob_id id;
5623 		struct spdk_blob *blob;
5624 	} new;
5625 
5626 	/* xattrs specified for snapshot/clones only. They have no impact on
5627 	 * the original blobs xattrs. */
5628 	const struct spdk_blob_xattr_opts *xattrs;
5629 };
5630 
5631 static void
5632 bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
5633 {
5634 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
5635 	struct spdk_bs_cpl *cpl = &ctx->cpl;
5636 
5637 	if (bserrno != 0) {
5638 		if (ctx->bserrno != 0) {
5639 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5640 		} else {
5641 			ctx->bserrno = bserrno;
5642 		}
5643 	}
5644 
5645 	switch (cpl->type) {
5646 	case SPDK_BS_CPL_TYPE_BLOBID:
5647 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
5648 		break;
5649 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
5650 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
5651 		break;
5652 	default:
5653 		SPDK_UNREACHABLE();
5654 		break;
5655 	}
5656 
5657 	free(ctx);
5658 }
5659 
5660 static void
5661 bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
5662 {
5663 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5664 	struct spdk_blob *origblob = ctx->original.blob;
5665 
5666 	if (bserrno != 0) {
5667 		if (ctx->bserrno != 0) {
5668 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
5669 		} else {
5670 			ctx->bserrno = bserrno;
5671 		}
5672 	}
5673 
5674 	ctx->original.id = origblob->id;
5675 	origblob->locked_operation_in_progress = false;
5676 
5677 	/* Revert md_ro to original state */
5678 	origblob->md_ro = ctx->original.md_ro;
5679 
5680 	spdk_blob_close(origblob, bs_clone_snapshot_cleanup_finish, ctx);
5681 }
5682 
5683 static void
5684 bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
5685 {
5686 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5687 	struct spdk_blob *origblob = ctx->original.blob;
5688 
5689 	if (bserrno != 0) {
5690 		if (ctx->bserrno != 0) {
5691 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5692 		} else {
5693 			ctx->bserrno = bserrno;
5694 		}
5695 	}
5696 
5697 	if (ctx->frozen) {
5698 		/* Unfreeze any outstanding I/O */
5699 		blob_unfreeze_io(origblob, bs_snapshot_unfreeze_cpl, ctx);
5700 	} else {
5701 		bs_snapshot_unfreeze_cpl(ctx, 0);
5702 	}
5703 
5704 }
5705 
5706 static void
5707 bs_clone_snapshot_newblob_cleanup(struct spdk_clone_snapshot_ctx *ctx, int bserrno)
5708 {
5709 	struct spdk_blob *newblob = ctx->new.blob;
5710 
5711 	if (bserrno != 0) {
5712 		if (ctx->bserrno != 0) {
5713 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5714 		} else {
5715 			ctx->bserrno = bserrno;
5716 		}
5717 	}
5718 
5719 	ctx->new.id = newblob->id;
5720 	spdk_blob_close(newblob, bs_clone_snapshot_origblob_cleanup, ctx);
5721 }
5722 
5723 /* END blob_cleanup */
5724 
5725 /* START spdk_bs_create_snapshot */
5726 
5727 static void
5728 bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
5729 {
5730 	uint64_t *cluster_temp;
5731 	uint32_t *extent_page_temp;
5732 
5733 	cluster_temp = blob1->active.clusters;
5734 	blob1->active.clusters = blob2->active.clusters;
5735 	blob2->active.clusters = cluster_temp;
5736 
5737 	extent_page_temp = blob1->active.extent_pages;
5738 	blob1->active.extent_pages = blob2->active.extent_pages;
5739 	blob2->active.extent_pages = extent_page_temp;
5740 }
5741 
5742 static void
5743 bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
5744 {
5745 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5746 	struct spdk_blob *origblob = ctx->original.blob;
5747 	struct spdk_blob *newblob = ctx->new.blob;
5748 
5749 	if (bserrno != 0) {
5750 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5751 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5752 		return;
5753 	}
5754 
5755 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
5756 	bserrno = blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
5757 	if (bserrno != 0) {
5758 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5759 		return;
5760 	}
5761 
5762 	bs_blob_list_add(ctx->original.blob);
5763 
5764 	spdk_blob_set_read_only(newblob);
5765 
5766 	/* sync snapshot metadata */
5767 	spdk_blob_sync_md(newblob, bs_clone_snapshot_origblob_cleanup, ctx);
5768 }
5769 
5770 static void
5771 bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
5772 {
5773 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5774 	struct spdk_blob *origblob = ctx->original.blob;
5775 	struct spdk_blob *newblob = ctx->new.blob;
5776 
5777 	if (bserrno != 0) {
5778 		/* return cluster map back to original */
5779 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5780 
5781 		/* Newblob md sync failed. Valid clusters are only present in origblob.
5782 		 * Since I/O is frozen on origblob, not changes to zeroed out cluster map should have occurred.
5783 		 * Newblob needs to be reverted to thin_provisioned state at creation to properly close. */
5784 		blob_set_thin_provision(newblob);
5785 		assert(spdk_mem_all_zero(newblob->active.clusters,
5786 					 newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
5787 		assert(spdk_mem_all_zero(newblob->active.extent_pages,
5788 					 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages)));
5789 
5790 		bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5791 		return;
5792 	}
5793 
5794 	/* Set internal xattr for snapshot id */
5795 	bserrno = blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
5796 	if (bserrno != 0) {
5797 		/* return cluster map back to original */
5798 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5799 		blob_set_thin_provision(newblob);
5800 		bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5801 		return;
5802 	}
5803 
5804 	/* Create new back_bs_dev for snapshot */
5805 	origblob->back_bs_dev = bs_create_blob_bs_dev(newblob);
5806 	if (origblob->back_bs_dev == NULL) {
5807 		/* return cluster map back to original */
5808 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5809 		blob_set_thin_provision(newblob);
5810 		bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
5811 		return;
5812 	}
5813 
5814 	bs_blob_list_remove(origblob);
5815 	origblob->parent_id = newblob->id;
5816 	/* set clone blob as thin provisioned */
5817 	blob_set_thin_provision(origblob);
5818 
5819 	bs_blob_list_add(newblob);
5820 
5821 	/* sync clone metadata */
5822 	spdk_blob_sync_md(origblob, bs_snapshot_origblob_sync_cpl, ctx);
5823 }
5824 
5825 static void
5826 bs_snapshot_freeze_cpl(void *cb_arg, int rc)
5827 {
5828 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5829 	struct spdk_blob *origblob = ctx->original.blob;
5830 	struct spdk_blob *newblob = ctx->new.blob;
5831 	int bserrno;
5832 
5833 	if (rc != 0) {
5834 		bs_clone_snapshot_newblob_cleanup(ctx, rc);
5835 		return;
5836 	}
5837 
5838 	ctx->frozen = true;
5839 
5840 	/* set new back_bs_dev for snapshot */
5841 	newblob->back_bs_dev = origblob->back_bs_dev;
5842 	/* Set invalid flags from origblob */
5843 	newblob->invalid_flags = origblob->invalid_flags;
5844 
5845 	/* inherit parent from original blob if set */
5846 	newblob->parent_id = origblob->parent_id;
5847 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
5848 		/* Set internal xattr for snapshot id */
5849 		bserrno = blob_set_xattr(newblob, BLOB_SNAPSHOT,
5850 					 &origblob->parent_id, sizeof(spdk_blob_id), true);
5851 		if (bserrno != 0) {
5852 			bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5853 			return;
5854 		}
5855 	}
5856 
5857 	/* swap cluster maps */
5858 	bs_snapshot_swap_cluster_maps(newblob, origblob);
5859 
5860 	/* Set the clear method on the new blob to match the original. */
5861 	blob_set_clear_method(newblob, origblob->clear_method);
5862 
5863 	/* sync snapshot metadata */
5864 	spdk_blob_sync_md(newblob, bs_snapshot_newblob_sync_cpl, ctx);
5865 }
5866 
5867 static void
5868 bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5869 {
5870 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5871 	struct spdk_blob *origblob = ctx->original.blob;
5872 	struct spdk_blob *newblob = _blob;
5873 
5874 	if (bserrno != 0) {
5875 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5876 		return;
5877 	}
5878 
5879 	ctx->new.blob = newblob;
5880 	assert(spdk_blob_is_thin_provisioned(newblob));
5881 	assert(spdk_mem_all_zero(newblob->active.clusters,
5882 				 newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
5883 	assert(spdk_mem_all_zero(newblob->active.extent_pages,
5884 				 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages)));
5885 
5886 	blob_freeze_io(origblob, bs_snapshot_freeze_cpl, ctx);
5887 }
5888 
5889 static void
5890 bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
5891 {
5892 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5893 	struct spdk_blob *origblob = ctx->original.blob;
5894 
5895 	if (bserrno != 0) {
5896 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5897 		return;
5898 	}
5899 
5900 	ctx->new.id = blobid;
5901 	ctx->cpl.u.blobid.blobid = blobid;
5902 
5903 	spdk_bs_open_blob(origblob->bs, ctx->new.id, bs_snapshot_newblob_open_cpl, ctx);
5904 }
5905 
5906 
5907 static void
5908 bs_xattr_snapshot(void *arg, const char *name,
5909 		  const void **value, size_t *value_len)
5910 {
5911 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
5912 
5913 	struct spdk_blob *blob = (struct spdk_blob *)arg;
5914 	*value = &blob->id;
5915 	*value_len = sizeof(blob->id);
5916 }
5917 
5918 static void
5919 bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5920 {
5921 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5922 	struct spdk_blob_opts opts;
5923 	struct spdk_blob_xattr_opts internal_xattrs;
5924 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
5925 
5926 	if (bserrno != 0) {
5927 		bs_clone_snapshot_cleanup_finish(ctx, bserrno);
5928 		return;
5929 	}
5930 
5931 	ctx->original.blob = _blob;
5932 
5933 	if (_blob->data_ro || _blob->md_ro) {
5934 		SPDK_DEBUGLOG(blob, "Cannot create snapshot from read only blob with id %" PRIu64 "\n",
5935 			      _blob->id);
5936 		ctx->bserrno = -EINVAL;
5937 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
5938 		return;
5939 	}
5940 
5941 	if (_blob->locked_operation_in_progress) {
5942 		SPDK_DEBUGLOG(blob, "Cannot create snapshot - another operation in progress\n");
5943 		ctx->bserrno = -EBUSY;
5944 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
5945 		return;
5946 	}
5947 
5948 	_blob->locked_operation_in_progress = true;
5949 
5950 	spdk_blob_opts_init(&opts, sizeof(opts));
5951 	blob_xattrs_init(&internal_xattrs);
5952 
5953 	/* Change the size of new blob to the same as in original blob,
5954 	 * but do not allocate clusters */
5955 	opts.thin_provision = true;
5956 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
5957 	opts.use_extent_table = _blob->use_extent_table;
5958 
5959 	/* If there are any xattrs specified for snapshot, set them now */
5960 	if (ctx->xattrs) {
5961 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
5962 	}
5963 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
5964 	internal_xattrs.count = 1;
5965 	internal_xattrs.ctx = _blob;
5966 	internal_xattrs.names = xattrs_names;
5967 	internal_xattrs.get_value = bs_xattr_snapshot;
5968 
5969 	bs_create_blob(_blob->bs, &opts, &internal_xattrs,
5970 		       bs_snapshot_newblob_create_cpl, ctx);
5971 }
5972 
5973 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
5974 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
5975 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5976 {
5977 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
5978 
5979 	if (!ctx) {
5980 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
5981 		return;
5982 	}
5983 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
5984 	ctx->cpl.u.blobid.cb_fn = cb_fn;
5985 	ctx->cpl.u.blobid.cb_arg = cb_arg;
5986 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
5987 	ctx->bserrno = 0;
5988 	ctx->frozen = false;
5989 	ctx->original.id = blobid;
5990 	ctx->xattrs = snapshot_xattrs;
5991 
5992 	spdk_bs_open_blob(bs, ctx->original.id, bs_snapshot_origblob_open_cpl, ctx);
5993 }
5994 /* END spdk_bs_create_snapshot */
5995 
5996 /* START spdk_bs_create_clone */
5997 
5998 static void
5999 bs_xattr_clone(void *arg, const char *name,
6000 	       const void **value, size_t *value_len)
6001 {
6002 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
6003 
6004 	struct spdk_blob *blob = (struct spdk_blob *)arg;
6005 	*value = &blob->id;
6006 	*value_len = sizeof(blob->id);
6007 }
6008 
6009 static void
6010 bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6011 {
6012 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6013 	struct spdk_blob *clone = _blob;
6014 
6015 	ctx->new.blob = clone;
6016 	bs_blob_list_add(clone);
6017 
6018 	spdk_blob_close(clone, bs_clone_snapshot_origblob_cleanup, ctx);
6019 }
6020 
6021 static void
6022 bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
6023 {
6024 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6025 
6026 	ctx->cpl.u.blobid.blobid = blobid;
6027 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, bs_clone_newblob_open_cpl, ctx);
6028 }
6029 
6030 static void
6031 bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6032 {
6033 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6034 	struct spdk_blob_opts		opts;
6035 	struct spdk_blob_xattr_opts internal_xattrs;
6036 	char *xattr_names[] = { BLOB_SNAPSHOT };
6037 
6038 	if (bserrno != 0) {
6039 		bs_clone_snapshot_cleanup_finish(ctx, bserrno);
6040 		return;
6041 	}
6042 
6043 	ctx->original.blob = _blob;
6044 	ctx->original.md_ro = _blob->md_ro;
6045 
6046 	if (!_blob->data_ro || !_blob->md_ro) {
6047 		SPDK_DEBUGLOG(blob, "Clone not from read-only blob\n");
6048 		ctx->bserrno = -EINVAL;
6049 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6050 		return;
6051 	}
6052 
6053 	if (_blob->locked_operation_in_progress) {
6054 		SPDK_DEBUGLOG(blob, "Cannot create clone - another operation in progress\n");
6055 		ctx->bserrno = -EBUSY;
6056 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6057 		return;
6058 	}
6059 
6060 	_blob->locked_operation_in_progress = true;
6061 
6062 	spdk_blob_opts_init(&opts, sizeof(opts));
6063 	blob_xattrs_init(&internal_xattrs);
6064 
6065 	opts.thin_provision = true;
6066 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
6067 	opts.use_extent_table = _blob->use_extent_table;
6068 	if (ctx->xattrs) {
6069 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
6070 	}
6071 
6072 	/* Set internal xattr BLOB_SNAPSHOT */
6073 	internal_xattrs.count = 1;
6074 	internal_xattrs.ctx = _blob;
6075 	internal_xattrs.names = xattr_names;
6076 	internal_xattrs.get_value = bs_xattr_clone;
6077 
6078 	bs_create_blob(_blob->bs, &opts, &internal_xattrs,
6079 		       bs_clone_newblob_create_cpl, ctx);
6080 }
6081 
6082 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
6083 			  const struct spdk_blob_xattr_opts *clone_xattrs,
6084 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6085 {
6086 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
6087 
6088 	if (!ctx) {
6089 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
6090 		return;
6091 	}
6092 
6093 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
6094 	ctx->cpl.u.blobid.cb_fn = cb_fn;
6095 	ctx->cpl.u.blobid.cb_arg = cb_arg;
6096 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
6097 	ctx->bserrno = 0;
6098 	ctx->xattrs = clone_xattrs;
6099 	ctx->original.id = blobid;
6100 
6101 	spdk_bs_open_blob(bs, ctx->original.id, bs_clone_origblob_open_cpl, ctx);
6102 }
6103 
6104 /* END spdk_bs_create_clone */
6105 
6106 /* START spdk_bs_inflate_blob */
6107 
6108 static void
6109 bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
6110 {
6111 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6112 	struct spdk_blob *_blob = ctx->original.blob;
6113 
6114 	if (bserrno != 0) {
6115 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6116 		return;
6117 	}
6118 
6119 	/* Temporarily override md_ro flag for MD modification */
6120 	_blob->md_ro = false;
6121 
6122 	bserrno = blob_set_xattr(_blob, BLOB_SNAPSHOT, &_parent->id, sizeof(spdk_blob_id), true);
6123 	if (bserrno != 0) {
6124 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6125 		return;
6126 	}
6127 
6128 	assert(_parent != NULL);
6129 
6130 	bs_blob_list_remove(_blob);
6131 	_blob->parent_id = _parent->id;
6132 
6133 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
6134 	_blob->back_bs_dev = bs_create_blob_bs_dev(_parent);
6135 	bs_blob_list_add(_blob);
6136 
6137 	spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx);
6138 }
6139 
6140 static void
6141 bs_inflate_blob_done(struct spdk_clone_snapshot_ctx *ctx)
6142 {
6143 	struct spdk_blob *_blob = ctx->original.blob;
6144 	struct spdk_blob *_parent;
6145 
6146 	if (ctx->allocate_all) {
6147 		/* remove thin provisioning */
6148 		bs_blob_list_remove(_blob);
6149 		blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
6150 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
6151 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
6152 		_blob->back_bs_dev = NULL;
6153 		_blob->parent_id = SPDK_BLOBID_INVALID;
6154 	} else {
6155 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
6156 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
6157 			/* We must change the parent of the inflated blob */
6158 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
6159 					  bs_inflate_blob_set_parent_cpl, ctx);
6160 			return;
6161 		}
6162 
6163 		bs_blob_list_remove(_blob);
6164 		blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
6165 		_blob->parent_id = SPDK_BLOBID_INVALID;
6166 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
6167 		_blob->back_bs_dev = bs_create_zeroes_dev();
6168 	}
6169 
6170 	/* Temporarily override md_ro flag for MD modification */
6171 	_blob->md_ro = false;
6172 	_blob->state = SPDK_BLOB_STATE_DIRTY;
6173 
6174 	spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx);
6175 }
6176 
6177 /* Check if cluster needs allocation */
6178 static inline bool
6179 bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
6180 {
6181 	struct spdk_blob_bs_dev *b;
6182 
6183 	assert(blob != NULL);
6184 
6185 	if (blob->active.clusters[cluster] != 0) {
6186 		/* Cluster is already allocated */
6187 		return false;
6188 	}
6189 
6190 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
6191 		/* Blob have no parent blob */
6192 		return allocate_all;
6193 	}
6194 
6195 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
6196 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
6197 }
6198 
6199 static void
6200 bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
6201 {
6202 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6203 	struct spdk_blob *_blob = ctx->original.blob;
6204 	struct spdk_bs_cpl cpl;
6205 	spdk_bs_user_op_t *op;
6206 	uint64_t offset;
6207 
6208 	if (bserrno != 0) {
6209 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6210 		return;
6211 	}
6212 
6213 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
6214 		if (bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
6215 			break;
6216 		}
6217 	}
6218 
6219 	if (ctx->cluster < _blob->active.num_clusters) {
6220 		offset = bs_cluster_to_lba(_blob->bs, ctx->cluster);
6221 
6222 		/* We may safely increment a cluster before copying */
6223 		ctx->cluster++;
6224 
6225 		/* Use a dummy 0B read as a context for cluster copy */
6226 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6227 		cpl.u.blob_basic.cb_fn = bs_inflate_blob_touch_next;
6228 		cpl.u.blob_basic.cb_arg = ctx;
6229 
6230 		op = bs_user_op_alloc(ctx->channel, &cpl, SPDK_BLOB_READ, _blob,
6231 				      NULL, 0, offset, 0);
6232 		if (!op) {
6233 			bs_clone_snapshot_origblob_cleanup(ctx, -ENOMEM);
6234 			return;
6235 		}
6236 
6237 		bs_allocate_and_copy_cluster(_blob, ctx->channel, offset, op);
6238 	} else {
6239 		bs_inflate_blob_done(ctx);
6240 	}
6241 }
6242 
6243 static void
6244 bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6245 {
6246 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6247 	uint64_t clusters_needed;
6248 	uint64_t i;
6249 
6250 	if (bserrno != 0) {
6251 		bs_clone_snapshot_cleanup_finish(ctx, bserrno);
6252 		return;
6253 	}
6254 
6255 	ctx->original.blob = _blob;
6256 	ctx->original.md_ro = _blob->md_ro;
6257 
6258 	if (_blob->locked_operation_in_progress) {
6259 		SPDK_DEBUGLOG(blob, "Cannot inflate blob - another operation in progress\n");
6260 		ctx->bserrno = -EBUSY;
6261 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6262 		return;
6263 	}
6264 
6265 	_blob->locked_operation_in_progress = true;
6266 
6267 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
6268 		/* This blob have no parent, so we cannot decouple it. */
6269 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
6270 		bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
6271 		return;
6272 	}
6273 
6274 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
6275 		/* This is not thin provisioned blob. No need to inflate. */
6276 		bs_clone_snapshot_origblob_cleanup(ctx, 0);
6277 		return;
6278 	}
6279 
6280 	/* Do two passes - one to verify that we can obtain enough clusters
6281 	 * and another to actually claim them.
6282 	 */
6283 	clusters_needed = 0;
6284 	for (i = 0; i < _blob->active.num_clusters; i++) {
6285 		if (bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
6286 			clusters_needed++;
6287 		}
6288 	}
6289 
6290 	if (clusters_needed > _blob->bs->num_free_clusters) {
6291 		/* Not enough free clusters. Cannot satisfy the request. */
6292 		bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
6293 		return;
6294 	}
6295 
6296 	ctx->cluster = 0;
6297 	bs_inflate_blob_touch_next(ctx, 0);
6298 }
6299 
6300 static void
6301 bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
6302 		spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
6303 {
6304 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
6305 
6306 	if (!ctx) {
6307 		cb_fn(cb_arg, -ENOMEM);
6308 		return;
6309 	}
6310 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6311 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
6312 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
6313 	ctx->bserrno = 0;
6314 	ctx->original.id = blobid;
6315 	ctx->channel = channel;
6316 	ctx->allocate_all = allocate_all;
6317 
6318 	spdk_bs_open_blob(bs, ctx->original.id, bs_inflate_blob_open_cpl, ctx);
6319 }
6320 
6321 void
6322 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
6323 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
6324 {
6325 	bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
6326 }
6327 
6328 void
6329 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
6330 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
6331 {
6332 	bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
6333 }
6334 /* END spdk_bs_inflate_blob */
6335 
6336 /* START spdk_blob_resize */
6337 struct spdk_bs_resize_ctx {
6338 	spdk_blob_op_complete cb_fn;
6339 	void *cb_arg;
6340 	struct spdk_blob *blob;
6341 	uint64_t sz;
6342 	int rc;
6343 };
6344 
6345 static void
6346 bs_resize_unfreeze_cpl(void *cb_arg, int rc)
6347 {
6348 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
6349 
6350 	if (rc != 0) {
6351 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
6352 	}
6353 
6354 	if (ctx->rc != 0) {
6355 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
6356 		rc = ctx->rc;
6357 	}
6358 
6359 	ctx->blob->locked_operation_in_progress = false;
6360 
6361 	ctx->cb_fn(ctx->cb_arg, rc);
6362 	free(ctx);
6363 }
6364 
6365 static void
6366 bs_resize_freeze_cpl(void *cb_arg, int rc)
6367 {
6368 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
6369 
6370 	if (rc != 0) {
6371 		ctx->blob->locked_operation_in_progress = false;
6372 		ctx->cb_fn(ctx->cb_arg, rc);
6373 		free(ctx);
6374 		return;
6375 	}
6376 
6377 	ctx->rc = blob_resize(ctx->blob, ctx->sz);
6378 
6379 	blob_unfreeze_io(ctx->blob, bs_resize_unfreeze_cpl, ctx);
6380 }
6381 
6382 void
6383 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
6384 {
6385 	struct spdk_bs_resize_ctx *ctx;
6386 
6387 	blob_verify_md_op(blob);
6388 
6389 	SPDK_DEBUGLOG(blob, "Resizing blob %" PRIu64 " to %" PRIu64 " clusters\n", blob->id, sz);
6390 
6391 	if (blob->md_ro) {
6392 		cb_fn(cb_arg, -EPERM);
6393 		return;
6394 	}
6395 
6396 	if (sz == blob->active.num_clusters) {
6397 		cb_fn(cb_arg, 0);
6398 		return;
6399 	}
6400 
6401 	if (blob->locked_operation_in_progress) {
6402 		cb_fn(cb_arg, -EBUSY);
6403 		return;
6404 	}
6405 
6406 	ctx = calloc(1, sizeof(*ctx));
6407 	if (!ctx) {
6408 		cb_fn(cb_arg, -ENOMEM);
6409 		return;
6410 	}
6411 
6412 	blob->locked_operation_in_progress = true;
6413 	ctx->cb_fn = cb_fn;
6414 	ctx->cb_arg = cb_arg;
6415 	ctx->blob = blob;
6416 	ctx->sz = sz;
6417 	blob_freeze_io(blob, bs_resize_freeze_cpl, ctx);
6418 }
6419 
6420 /* END spdk_blob_resize */
6421 
6422 
6423 /* START spdk_bs_delete_blob */
6424 
6425 static void
6426 bs_delete_close_cpl(void *cb_arg, int bserrno)
6427 {
6428 	spdk_bs_sequence_t *seq = cb_arg;
6429 
6430 	bs_sequence_finish(seq, bserrno);
6431 }
6432 
6433 static void
6434 bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6435 {
6436 	struct spdk_blob *blob = cb_arg;
6437 
6438 	if (bserrno != 0) {
6439 		/*
6440 		 * We already removed this blob from the blobstore tailq, so
6441 		 *  we need to free it here since this is the last reference
6442 		 *  to it.
6443 		 */
6444 		blob_free(blob);
6445 		bs_delete_close_cpl(seq, bserrno);
6446 		return;
6447 	}
6448 
6449 	/*
6450 	 * This will immediately decrement the ref_count and call
6451 	 *  the completion routine since the metadata state is clean.
6452 	 *  By calling spdk_blob_close, we reduce the number of call
6453 	 *  points into code that touches the blob->open_ref count
6454 	 *  and the blobstore's blob list.
6455 	 */
6456 	spdk_blob_close(blob, bs_delete_close_cpl, seq);
6457 }
6458 
6459 struct delete_snapshot_ctx {
6460 	struct spdk_blob_list *parent_snapshot_entry;
6461 	struct spdk_blob *snapshot;
6462 	bool snapshot_md_ro;
6463 	struct spdk_blob *clone;
6464 	bool clone_md_ro;
6465 	spdk_blob_op_with_handle_complete cb_fn;
6466 	void *cb_arg;
6467 	int bserrno;
6468 	uint32_t next_extent_page;
6469 };
6470 
6471 static void
6472 delete_blob_cleanup_finish(void *cb_arg, int bserrno)
6473 {
6474 	struct delete_snapshot_ctx *ctx = cb_arg;
6475 
6476 	if (bserrno != 0) {
6477 		SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno);
6478 	}
6479 
6480 	assert(ctx != NULL);
6481 
6482 	if (bserrno != 0 && ctx->bserrno == 0) {
6483 		ctx->bserrno = bserrno;
6484 	}
6485 
6486 	ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno);
6487 	free(ctx);
6488 }
6489 
6490 static void
6491 delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno)
6492 {
6493 	struct delete_snapshot_ctx *ctx = cb_arg;
6494 
6495 	if (bserrno != 0) {
6496 		ctx->bserrno = bserrno;
6497 		SPDK_ERRLOG("Clone cleanup error %d\n", bserrno);
6498 	}
6499 
6500 	if (ctx->bserrno != 0) {
6501 		assert(blob_lookup(ctx->snapshot->bs, ctx->snapshot->id) == NULL);
6502 		RB_INSERT(spdk_blob_tree, &ctx->snapshot->bs->open_blobs, ctx->snapshot);
6503 		spdk_bit_array_set(ctx->snapshot->bs->open_blobids, ctx->snapshot->id);
6504 	}
6505 
6506 	ctx->snapshot->locked_operation_in_progress = false;
6507 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
6508 
6509 	spdk_blob_close(ctx->snapshot, delete_blob_cleanup_finish, ctx);
6510 }
6511 
6512 static void
6513 delete_snapshot_cleanup_clone(void *cb_arg, int bserrno)
6514 {
6515 	struct delete_snapshot_ctx *ctx = cb_arg;
6516 
6517 	ctx->clone->locked_operation_in_progress = false;
6518 	ctx->clone->md_ro = ctx->clone_md_ro;
6519 
6520 	spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx);
6521 }
6522 
6523 static void
6524 delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
6525 {
6526 	struct delete_snapshot_ctx *ctx = cb_arg;
6527 
6528 	if (bserrno) {
6529 		ctx->bserrno = bserrno;
6530 		delete_snapshot_cleanup_clone(ctx, 0);
6531 		return;
6532 	}
6533 
6534 	ctx->clone->locked_operation_in_progress = false;
6535 	spdk_blob_close(ctx->clone, delete_blob_cleanup_finish, ctx);
6536 }
6537 
6538 static void
6539 delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno)
6540 {
6541 	struct delete_snapshot_ctx *ctx = cb_arg;
6542 	struct spdk_blob_list *parent_snapshot_entry = NULL;
6543 	struct spdk_blob_list *snapshot_entry = NULL;
6544 	struct spdk_blob_list *clone_entry = NULL;
6545 	struct spdk_blob_list *snapshot_clone_entry = NULL;
6546 
6547 	if (bserrno) {
6548 		SPDK_ERRLOG("Failed to sync MD on blob\n");
6549 		ctx->bserrno = bserrno;
6550 		delete_snapshot_cleanup_clone(ctx, 0);
6551 		return;
6552 	}
6553 
6554 	/* Get snapshot entry for the snapshot we want to remove */
6555 	snapshot_entry = bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id);
6556 
6557 	assert(snapshot_entry != NULL);
6558 
6559 	/* Remove clone entry in this snapshot (at this point there can be only one clone) */
6560 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6561 	assert(clone_entry != NULL);
6562 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
6563 	snapshot_entry->clone_count--;
6564 	assert(TAILQ_EMPTY(&snapshot_entry->clones));
6565 
6566 	if (ctx->snapshot->parent_id != SPDK_BLOBID_INVALID) {
6567 		/* This snapshot is at the same time a clone of another snapshot - we need to
6568 		 * update parent snapshot (remove current clone, add new one inherited from
6569 		 * the snapshot that is being removed) */
6570 
6571 		/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
6572 		 * snapshot that we are removing */
6573 		blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry,
6574 						    &snapshot_clone_entry);
6575 
6576 		/* Switch clone entry in parent snapshot */
6577 		TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link);
6578 		TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link);
6579 		free(snapshot_clone_entry);
6580 	} else {
6581 		/* No parent snapshot - just remove clone entry */
6582 		free(clone_entry);
6583 	}
6584 
6585 	/* Restore md_ro flags */
6586 	ctx->clone->md_ro = ctx->clone_md_ro;
6587 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
6588 
6589 	blob_unfreeze_io(ctx->clone, delete_snapshot_unfreeze_cpl, ctx);
6590 }
6591 
6592 static void
6593 delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno)
6594 {
6595 	struct delete_snapshot_ctx *ctx = cb_arg;
6596 	uint64_t i;
6597 
6598 	ctx->snapshot->md_ro = false;
6599 
6600 	if (bserrno) {
6601 		SPDK_ERRLOG("Failed to sync MD on clone\n");
6602 		ctx->bserrno = bserrno;
6603 
6604 		/* Restore snapshot to previous state */
6605 		bserrno = blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true);
6606 		if (bserrno != 0) {
6607 			delete_snapshot_cleanup_clone(ctx, bserrno);
6608 			return;
6609 		}
6610 
6611 		spdk_blob_sync_md(ctx->snapshot, delete_snapshot_cleanup_clone, ctx);
6612 		return;
6613 	}
6614 
6615 	/* Clear cluster map entries for snapshot */
6616 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
6617 		if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) {
6618 			ctx->snapshot->active.clusters[i] = 0;
6619 		}
6620 	}
6621 	for (i = 0; i < ctx->snapshot->active.num_extent_pages &&
6622 	     i < ctx->clone->active.num_extent_pages; i++) {
6623 		if (ctx->clone->active.extent_pages[i] == ctx->snapshot->active.extent_pages[i]) {
6624 			ctx->snapshot->active.extent_pages[i] = 0;
6625 		}
6626 	}
6627 
6628 	blob_set_thin_provision(ctx->snapshot);
6629 	ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY;
6630 
6631 	if (ctx->parent_snapshot_entry != NULL) {
6632 		ctx->snapshot->back_bs_dev = NULL;
6633 	}
6634 
6635 	spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_cpl, ctx);
6636 }
6637 
6638 static void
6639 delete_snapshot_update_extent_pages_cpl(struct delete_snapshot_ctx *ctx)
6640 {
6641 	/* Delete old backing bs_dev from clone (related to snapshot that will be removed) */
6642 	ctx->clone->back_bs_dev->destroy(ctx->clone->back_bs_dev);
6643 
6644 	/* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */
6645 	if (ctx->parent_snapshot_entry != NULL) {
6646 		/* ...to parent snapshot */
6647 		ctx->clone->parent_id = ctx->parent_snapshot_entry->id;
6648 		ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev;
6649 		blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id,
6650 			       sizeof(spdk_blob_id),
6651 			       true);
6652 	} else {
6653 		/* ...to blobid invalid and zeroes dev */
6654 		ctx->clone->parent_id = SPDK_BLOBID_INVALID;
6655 		ctx->clone->back_bs_dev = bs_create_zeroes_dev();
6656 		blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true);
6657 	}
6658 
6659 	spdk_blob_sync_md(ctx->clone, delete_snapshot_sync_clone_cpl, ctx);
6660 }
6661 
6662 static void
6663 delete_snapshot_update_extent_pages(void *cb_arg, int bserrno)
6664 {
6665 	struct delete_snapshot_ctx *ctx = cb_arg;
6666 	uint32_t *extent_page;
6667 	uint64_t i;
6668 
6669 	for (i = ctx->next_extent_page; i < ctx->snapshot->active.num_extent_pages &&
6670 	     i < ctx->clone->active.num_extent_pages; i++) {
6671 		if (ctx->snapshot->active.extent_pages[i] == 0) {
6672 			/* No extent page to use from snapshot */
6673 			continue;
6674 		}
6675 
6676 		extent_page = &ctx->clone->active.extent_pages[i];
6677 		if (*extent_page == 0) {
6678 			/* Copy extent page from snapshot when clone did not have a matching one */
6679 			*extent_page = ctx->snapshot->active.extent_pages[i];
6680 			continue;
6681 		}
6682 
6683 		/* Clone and snapshot both contain partially filled matching extent pages.
6684 		 * Update the clone extent page in place with cluster map containing the mix of both. */
6685 		ctx->next_extent_page = i + 1;
6686 
6687 		blob_write_extent_page(ctx->clone, *extent_page, i * SPDK_EXTENTS_PER_EP,
6688 				       delete_snapshot_update_extent_pages, ctx);
6689 		return;
6690 	}
6691 	delete_snapshot_update_extent_pages_cpl(ctx);
6692 }
6693 
6694 static void
6695 delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno)
6696 {
6697 	struct delete_snapshot_ctx *ctx = cb_arg;
6698 	uint64_t i;
6699 
6700 	/* Temporarily override md_ro flag for clone for MD modification */
6701 	ctx->clone_md_ro = ctx->clone->md_ro;
6702 	ctx->clone->md_ro = false;
6703 
6704 	if (bserrno) {
6705 		SPDK_ERRLOG("Failed to sync MD with xattr on blob\n");
6706 		ctx->bserrno = bserrno;
6707 		delete_snapshot_cleanup_clone(ctx, 0);
6708 		return;
6709 	}
6710 
6711 	/* Copy snapshot map to clone map (only unallocated clusters in clone) */
6712 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
6713 		if (ctx->clone->active.clusters[i] == 0) {
6714 			ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i];
6715 		}
6716 	}
6717 	ctx->next_extent_page = 0;
6718 	delete_snapshot_update_extent_pages(ctx, 0);
6719 }
6720 
6721 static void
6722 delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno)
6723 {
6724 	struct delete_snapshot_ctx *ctx = cb_arg;
6725 
6726 	if (bserrno) {
6727 		SPDK_ERRLOG("Failed to freeze I/O on clone\n");
6728 		ctx->bserrno = bserrno;
6729 		delete_snapshot_cleanup_clone(ctx, 0);
6730 		return;
6731 	}
6732 
6733 	/* Temporarily override md_ro flag for snapshot for MD modification */
6734 	ctx->snapshot_md_ro = ctx->snapshot->md_ro;
6735 	ctx->snapshot->md_ro = false;
6736 
6737 	/* Mark blob as pending for removal for power failure safety, use clone id for recovery */
6738 	ctx->bserrno = blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id,
6739 				      sizeof(spdk_blob_id), true);
6740 	if (ctx->bserrno != 0) {
6741 		delete_snapshot_cleanup_clone(ctx, 0);
6742 		return;
6743 	}
6744 
6745 	spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_xattr_cpl, ctx);
6746 }
6747 
6748 static void
6749 delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno)
6750 {
6751 	struct delete_snapshot_ctx *ctx = cb_arg;
6752 
6753 	if (bserrno) {
6754 		SPDK_ERRLOG("Failed to open clone\n");
6755 		ctx->bserrno = bserrno;
6756 		delete_snapshot_cleanup_snapshot(ctx, 0);
6757 		return;
6758 	}
6759 
6760 	ctx->clone = clone;
6761 
6762 	if (clone->locked_operation_in_progress) {
6763 		SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress on its clone\n");
6764 		ctx->bserrno = -EBUSY;
6765 		spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx);
6766 		return;
6767 	}
6768 
6769 	clone->locked_operation_in_progress = true;
6770 
6771 	blob_freeze_io(clone, delete_snapshot_freeze_io_cb, ctx);
6772 }
6773 
6774 static void
6775 update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx)
6776 {
6777 	struct spdk_blob_list *snapshot_entry = NULL;
6778 	struct spdk_blob_list *clone_entry = NULL;
6779 	struct spdk_blob_list *snapshot_clone_entry = NULL;
6780 
6781 	/* Get snapshot entry for the snapshot we want to remove */
6782 	snapshot_entry = bs_get_snapshot_entry(snapshot->bs, snapshot->id);
6783 
6784 	assert(snapshot_entry != NULL);
6785 
6786 	/* Get clone of the snapshot (at this point there can be only one clone) */
6787 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6788 	assert(snapshot_entry->clone_count == 1);
6789 	assert(clone_entry != NULL);
6790 
6791 	/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
6792 	 * snapshot that we are removing */
6793 	blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry,
6794 					    &snapshot_clone_entry);
6795 
6796 	spdk_bs_open_blob(snapshot->bs, clone_entry->id, delete_snapshot_open_clone_cb, ctx);
6797 }
6798 
6799 static void
6800 bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno)
6801 {
6802 	spdk_bs_sequence_t *seq = cb_arg;
6803 	struct spdk_blob_list *snapshot_entry = NULL;
6804 	uint32_t page_num;
6805 
6806 	if (bserrno) {
6807 		SPDK_ERRLOG("Failed to remove blob\n");
6808 		bs_sequence_finish(seq, bserrno);
6809 		return;
6810 	}
6811 
6812 	/* Remove snapshot from the list */
6813 	snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
6814 	if (snapshot_entry != NULL) {
6815 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
6816 		free(snapshot_entry);
6817 	}
6818 
6819 	page_num = bs_blobid_to_page(blob->id);
6820 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
6821 	blob->state = SPDK_BLOB_STATE_DIRTY;
6822 	blob->active.num_pages = 0;
6823 	blob_resize(blob, 0);
6824 
6825 	blob_persist(seq, blob, bs_delete_persist_cpl, blob);
6826 }
6827 
6828 static int
6829 bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone)
6830 {
6831 	struct spdk_blob_list *snapshot_entry = NULL;
6832 	struct spdk_blob_list *clone_entry = NULL;
6833 	struct spdk_blob *clone = NULL;
6834 	bool has_one_clone = false;
6835 
6836 	/* Check if this is a snapshot with clones */
6837 	snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
6838 	if (snapshot_entry != NULL) {
6839 		if (snapshot_entry->clone_count > 1) {
6840 			SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n");
6841 			return -EBUSY;
6842 		} else if (snapshot_entry->clone_count == 1) {
6843 			has_one_clone = true;
6844 		}
6845 	}
6846 
6847 	/* Check if someone has this blob open (besides this delete context):
6848 	 * - open_ref = 1 - only this context opened blob, so it is ok to remove it
6849 	 * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot
6850 	 *	and that is ok, because we will update it accordingly */
6851 	if (blob->open_ref <= 2 && has_one_clone) {
6852 		clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6853 		assert(clone_entry != NULL);
6854 		clone = blob_lookup(blob->bs, clone_entry->id);
6855 
6856 		if (blob->open_ref == 2 && clone == NULL) {
6857 			/* Clone is closed and someone else opened this blob */
6858 			SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
6859 			return -EBUSY;
6860 		}
6861 
6862 		*update_clone = true;
6863 		return 0;
6864 	}
6865 
6866 	if (blob->open_ref > 1) {
6867 		SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
6868 		return -EBUSY;
6869 	}
6870 
6871 	assert(has_one_clone == false);
6872 	*update_clone = false;
6873 	return 0;
6874 }
6875 
6876 static void
6877 bs_delete_enomem_close_cpl(void *cb_arg, int bserrno)
6878 {
6879 	spdk_bs_sequence_t *seq = cb_arg;
6880 
6881 	bs_sequence_finish(seq, -ENOMEM);
6882 }
6883 
6884 static void
6885 bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
6886 {
6887 	spdk_bs_sequence_t *seq = cb_arg;
6888 	struct delete_snapshot_ctx *ctx;
6889 	bool update_clone = false;
6890 
6891 	if (bserrno != 0) {
6892 		bs_sequence_finish(seq, bserrno);
6893 		return;
6894 	}
6895 
6896 	blob_verify_md_op(blob);
6897 
6898 	ctx = calloc(1, sizeof(*ctx));
6899 	if (ctx == NULL) {
6900 		spdk_blob_close(blob, bs_delete_enomem_close_cpl, seq);
6901 		return;
6902 	}
6903 
6904 	ctx->snapshot = blob;
6905 	ctx->cb_fn = bs_delete_blob_finish;
6906 	ctx->cb_arg = seq;
6907 
6908 	/* Check if blob can be removed and if it is a snapshot with clone on top of it */
6909 	ctx->bserrno = bs_is_blob_deletable(blob, &update_clone);
6910 	if (ctx->bserrno) {
6911 		spdk_blob_close(blob, delete_blob_cleanup_finish, ctx);
6912 		return;
6913 	}
6914 
6915 	if (blob->locked_operation_in_progress) {
6916 		SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress\n");
6917 		ctx->bserrno = -EBUSY;
6918 		spdk_blob_close(blob, delete_blob_cleanup_finish, ctx);
6919 		return;
6920 	}
6921 
6922 	blob->locked_operation_in_progress = true;
6923 
6924 	/*
6925 	 * Remove the blob from the blob_store list now, to ensure it does not
6926 	 *  get returned after this point by blob_lookup().
6927 	 */
6928 	spdk_bit_array_clear(blob->bs->open_blobids, blob->id);
6929 	RB_REMOVE(spdk_blob_tree, &blob->bs->open_blobs, blob);
6930 
6931 	if (update_clone) {
6932 		/* This blob is a snapshot with active clone - update clone first */
6933 		update_clone_on_snapshot_deletion(blob, ctx);
6934 	} else {
6935 		/* This blob does not have any clones - just remove it */
6936 		bs_blob_list_remove(blob);
6937 		bs_delete_blob_finish(seq, blob, 0);
6938 		free(ctx);
6939 	}
6940 }
6941 
6942 void
6943 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
6944 		    spdk_blob_op_complete cb_fn, void *cb_arg)
6945 {
6946 	struct spdk_bs_cpl	cpl;
6947 	spdk_bs_sequence_t	*seq;
6948 
6949 	SPDK_DEBUGLOG(blob, "Deleting blob %" PRIu64 "\n", blobid);
6950 
6951 	assert(spdk_get_thread() == bs->md_thread);
6952 
6953 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6954 	cpl.u.blob_basic.cb_fn = cb_fn;
6955 	cpl.u.blob_basic.cb_arg = cb_arg;
6956 
6957 	seq = bs_sequence_start(bs->md_channel, &cpl);
6958 	if (!seq) {
6959 		cb_fn(cb_arg, -ENOMEM);
6960 		return;
6961 	}
6962 
6963 	spdk_bs_open_blob(bs, blobid, bs_delete_open_cpl, seq);
6964 }
6965 
6966 /* END spdk_bs_delete_blob */
6967 
6968 /* START spdk_bs_open_blob */
6969 
6970 static void
6971 bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6972 {
6973 	struct spdk_blob *blob = cb_arg;
6974 	struct spdk_blob *existing;
6975 
6976 	if (bserrno != 0) {
6977 		blob_free(blob);
6978 		seq->cpl.u.blob_handle.blob = NULL;
6979 		bs_sequence_finish(seq, bserrno);
6980 		return;
6981 	}
6982 
6983 	existing = blob_lookup(blob->bs, blob->id);
6984 	if (existing) {
6985 		blob_free(blob);
6986 		existing->open_ref++;
6987 		seq->cpl.u.blob_handle.blob = existing;
6988 		bs_sequence_finish(seq, 0);
6989 		return;
6990 	}
6991 
6992 	blob->open_ref++;
6993 
6994 	spdk_bit_array_set(blob->bs->open_blobids, blob->id);
6995 	RB_INSERT(spdk_blob_tree, &blob->bs->open_blobs, blob);
6996 
6997 	bs_sequence_finish(seq, bserrno);
6998 }
6999 
7000 static inline void
7001 blob_open_opts_copy(const struct spdk_blob_open_opts *src, struct spdk_blob_open_opts *dst)
7002 {
7003 #define FIELD_OK(field) \
7004         offsetof(struct spdk_blob_opts, field) + sizeof(src->field) <= src->opts_size
7005 
7006 #define SET_FIELD(field) \
7007         if (FIELD_OK(field)) { \
7008                 dst->field = src->field; \
7009         } \
7010 
7011 	SET_FIELD(clear_method);
7012 
7013 	dst->opts_size = src->opts_size;
7014 
7015 	/* You should not remove this statement, but need to update the assert statement
7016 	 * if you add a new field, and also add a corresponding SET_FIELD statement */
7017 	SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_open_opts) == 16, "Incorrect size");
7018 
7019 #undef FIELD_OK
7020 #undef SET_FIELD
7021 }
7022 
7023 static void
7024 bs_open_blob(struct spdk_blob_store *bs,
7025 	     spdk_blob_id blobid,
7026 	     struct spdk_blob_open_opts *opts,
7027 	     spdk_blob_op_with_handle_complete cb_fn,
7028 	     void *cb_arg)
7029 {
7030 	struct spdk_blob		*blob;
7031 	struct spdk_bs_cpl		cpl;
7032 	struct spdk_blob_open_opts	opts_local;
7033 	spdk_bs_sequence_t		*seq;
7034 	uint32_t			page_num;
7035 
7036 	SPDK_DEBUGLOG(blob, "Opening blob %" PRIu64 "\n", blobid);
7037 	assert(spdk_get_thread() == bs->md_thread);
7038 
7039 	page_num = bs_blobid_to_page(blobid);
7040 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
7041 		/* Invalid blobid */
7042 		cb_fn(cb_arg, NULL, -ENOENT);
7043 		return;
7044 	}
7045 
7046 	blob = blob_lookup(bs, blobid);
7047 	if (blob) {
7048 		blob->open_ref++;
7049 		cb_fn(cb_arg, blob, 0);
7050 		return;
7051 	}
7052 
7053 	blob = blob_alloc(bs, blobid);
7054 	if (!blob) {
7055 		cb_fn(cb_arg, NULL, -ENOMEM);
7056 		return;
7057 	}
7058 
7059 	spdk_blob_open_opts_init(&opts_local, sizeof(opts_local));
7060 	if (opts) {
7061 		blob_open_opts_copy(opts, &opts_local);
7062 	}
7063 
7064 	blob->clear_method = opts_local.clear_method;
7065 
7066 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
7067 	cpl.u.blob_handle.cb_fn = cb_fn;
7068 	cpl.u.blob_handle.cb_arg = cb_arg;
7069 	cpl.u.blob_handle.blob = blob;
7070 
7071 	seq = bs_sequence_start(bs->md_channel, &cpl);
7072 	if (!seq) {
7073 		blob_free(blob);
7074 		cb_fn(cb_arg, NULL, -ENOMEM);
7075 		return;
7076 	}
7077 
7078 	blob_load(seq, blob, bs_open_blob_cpl, blob);
7079 }
7080 
7081 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
7082 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7083 {
7084 	bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
7085 }
7086 
7087 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
7088 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7089 {
7090 	bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
7091 }
7092 
7093 /* END spdk_bs_open_blob */
7094 
7095 /* START spdk_blob_set_read_only */
7096 int spdk_blob_set_read_only(struct spdk_blob *blob)
7097 {
7098 	blob_verify_md_op(blob);
7099 
7100 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
7101 
7102 	blob->state = SPDK_BLOB_STATE_DIRTY;
7103 	return 0;
7104 }
7105 /* END spdk_blob_set_read_only */
7106 
7107 /* START spdk_blob_sync_md */
7108 
7109 static void
7110 blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7111 {
7112 	struct spdk_blob *blob = cb_arg;
7113 
7114 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
7115 		blob->data_ro = true;
7116 		blob->md_ro = true;
7117 	}
7118 
7119 	bs_sequence_finish(seq, bserrno);
7120 }
7121 
7122 static void
7123 blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
7124 {
7125 	struct spdk_bs_cpl	cpl;
7126 	spdk_bs_sequence_t	*seq;
7127 
7128 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7129 	cpl.u.blob_basic.cb_fn = cb_fn;
7130 	cpl.u.blob_basic.cb_arg = cb_arg;
7131 
7132 	seq = bs_sequence_start(blob->bs->md_channel, &cpl);
7133 	if (!seq) {
7134 		cb_fn(cb_arg, -ENOMEM);
7135 		return;
7136 	}
7137 
7138 	blob_persist(seq, blob, blob_sync_md_cpl, blob);
7139 }
7140 
7141 void
7142 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
7143 {
7144 	blob_verify_md_op(blob);
7145 
7146 	SPDK_DEBUGLOG(blob, "Syncing blob %" PRIu64 "\n", blob->id);
7147 
7148 	if (blob->md_ro) {
7149 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
7150 		cb_fn(cb_arg, 0);
7151 		return;
7152 	}
7153 
7154 	blob_sync_md(blob, cb_fn, cb_arg);
7155 }
7156 
7157 /* END spdk_blob_sync_md */
7158 
7159 struct spdk_blob_insert_cluster_ctx {
7160 	struct spdk_thread	*thread;
7161 	struct spdk_blob	*blob;
7162 	uint32_t		cluster_num;	/* cluster index in blob */
7163 	uint32_t		cluster;	/* cluster on disk */
7164 	uint32_t		extent_page;	/* extent page on disk */
7165 	int			rc;
7166 	spdk_blob_op_complete	cb_fn;
7167 	void			*cb_arg;
7168 };
7169 
7170 static void
7171 blob_insert_cluster_msg_cpl(void *arg)
7172 {
7173 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7174 
7175 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
7176 	free(ctx);
7177 }
7178 
7179 static void
7180 blob_insert_cluster_msg_cb(void *arg, int bserrno)
7181 {
7182 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7183 
7184 	ctx->rc = bserrno;
7185 	spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx);
7186 }
7187 
7188 static void
7189 blob_insert_new_ep_cb(void *arg, int bserrno)
7190 {
7191 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7192 	uint32_t *extent_page;
7193 
7194 	extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
7195 	*extent_page = ctx->extent_page;
7196 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
7197 	blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx);
7198 }
7199 
7200 static void
7201 blob_persist_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7202 {
7203 	struct spdk_blob_md_page        *page = cb_arg;
7204 
7205 	bs_sequence_finish(seq, bserrno);
7206 	spdk_free(page);
7207 }
7208 
7209 static void
7210 blob_write_extent_page(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num,
7211 		       spdk_blob_op_complete cb_fn, void *cb_arg)
7212 {
7213 	spdk_bs_sequence_t		*seq;
7214 	struct spdk_bs_cpl		cpl;
7215 	struct spdk_blob_md_page	*page = NULL;
7216 	uint32_t			page_count = 0;
7217 	int				rc;
7218 
7219 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7220 	cpl.u.blob_basic.cb_fn = cb_fn;
7221 	cpl.u.blob_basic.cb_arg = cb_arg;
7222 
7223 	seq = bs_sequence_start(blob->bs->md_channel, &cpl);
7224 	if (!seq) {
7225 		cb_fn(cb_arg, -ENOMEM);
7226 		return;
7227 	}
7228 	rc = blob_serialize_add_page(blob, &page, &page_count, &page);
7229 	if (rc < 0) {
7230 		bs_sequence_finish(seq, rc);
7231 		return;
7232 	}
7233 
7234 	blob_serialize_extent_page(blob, cluster_num, page);
7235 
7236 	page->crc = blob_md_page_calc_crc(page);
7237 
7238 	assert(spdk_bit_array_get(blob->bs->used_md_pages, extent) == true);
7239 
7240 	bs_sequence_write_dev(seq, page, bs_md_page_to_lba(blob->bs, extent),
7241 			      bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE),
7242 			      blob_persist_extent_page_cpl, page);
7243 }
7244 
7245 static void
7246 blob_insert_cluster_msg(void *arg)
7247 {
7248 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7249 	uint32_t *extent_page;
7250 
7251 	ctx->rc = blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
7252 	if (ctx->rc != 0) {
7253 		spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx);
7254 		return;
7255 	}
7256 
7257 	if (ctx->blob->use_extent_table == false) {
7258 		/* Extent table is not used, proceed with sync of md that will only use extents_rle. */
7259 		ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
7260 		blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx);
7261 		return;
7262 	}
7263 
7264 	extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
7265 	if (*extent_page == 0) {
7266 		/* Extent page requires allocation.
7267 		 * It was already claimed in the used_md_pages map and placed in ctx. */
7268 		assert(ctx->extent_page != 0);
7269 		assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
7270 		blob_write_extent_page(ctx->blob, ctx->extent_page, ctx->cluster_num,
7271 				       blob_insert_new_ep_cb, ctx);
7272 	} else {
7273 		/* It is possible for original thread to allocate extent page for
7274 		 * different cluster in the same extent page. In such case proceed with
7275 		 * updating the existing extent page, but release the additional one. */
7276 		if (ctx->extent_page != 0) {
7277 			assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
7278 			bs_release_md_page(ctx->blob->bs, ctx->extent_page);
7279 			ctx->extent_page = 0;
7280 		}
7281 		/* Extent page already allocated.
7282 		 * Every cluster allocation, requires just an update of single extent page. */
7283 		blob_write_extent_page(ctx->blob, *extent_page, ctx->cluster_num,
7284 				       blob_insert_cluster_msg_cb, ctx);
7285 	}
7286 }
7287 
7288 static void
7289 blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
7290 				 uint64_t cluster, uint32_t extent_page, spdk_blob_op_complete cb_fn, void *cb_arg)
7291 {
7292 	struct spdk_blob_insert_cluster_ctx *ctx;
7293 
7294 	ctx = calloc(1, sizeof(*ctx));
7295 	if (ctx == NULL) {
7296 		cb_fn(cb_arg, -ENOMEM);
7297 		return;
7298 	}
7299 
7300 	ctx->thread = spdk_get_thread();
7301 	ctx->blob = blob;
7302 	ctx->cluster_num = cluster_num;
7303 	ctx->cluster = cluster;
7304 	ctx->extent_page = extent_page;
7305 	ctx->cb_fn = cb_fn;
7306 	ctx->cb_arg = cb_arg;
7307 
7308 	spdk_thread_send_msg(blob->bs->md_thread, blob_insert_cluster_msg, ctx);
7309 }
7310 
7311 /* START spdk_blob_close */
7312 
7313 static void
7314 blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7315 {
7316 	struct spdk_blob *blob = cb_arg;
7317 
7318 	if (bserrno == 0) {
7319 		blob->open_ref--;
7320 		if (blob->open_ref == 0) {
7321 			/*
7322 			 * Blobs with active.num_pages == 0 are deleted blobs.
7323 			 *  these blobs are removed from the blob_store list
7324 			 *  when the deletion process starts - so don't try to
7325 			 *  remove them again.
7326 			 */
7327 			if (blob->active.num_pages > 0) {
7328 				spdk_bit_array_clear(blob->bs->open_blobids, blob->id);
7329 				RB_REMOVE(spdk_blob_tree, &blob->bs->open_blobs, blob);
7330 			}
7331 			blob_free(blob);
7332 		}
7333 	}
7334 
7335 	bs_sequence_finish(seq, bserrno);
7336 }
7337 
7338 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
7339 {
7340 	struct spdk_bs_cpl	cpl;
7341 	spdk_bs_sequence_t	*seq;
7342 
7343 	blob_verify_md_op(blob);
7344 
7345 	SPDK_DEBUGLOG(blob, "Closing blob %" PRIu64 "\n", blob->id);
7346 
7347 	if (blob->open_ref == 0) {
7348 		cb_fn(cb_arg, -EBADF);
7349 		return;
7350 	}
7351 
7352 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7353 	cpl.u.blob_basic.cb_fn = cb_fn;
7354 	cpl.u.blob_basic.cb_arg = cb_arg;
7355 
7356 	seq = bs_sequence_start(blob->bs->md_channel, &cpl);
7357 	if (!seq) {
7358 		cb_fn(cb_arg, -ENOMEM);
7359 		return;
7360 	}
7361 
7362 	/* Sync metadata */
7363 	blob_persist(seq, blob, blob_close_cpl, blob);
7364 }
7365 
7366 /* END spdk_blob_close */
7367 
7368 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
7369 {
7370 	return spdk_get_io_channel(bs);
7371 }
7372 
7373 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
7374 {
7375 	spdk_put_io_channel(channel);
7376 }
7377 
7378 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
7379 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
7380 {
7381 	blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
7382 			       SPDK_BLOB_UNMAP);
7383 }
7384 
7385 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
7386 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
7387 {
7388 	blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
7389 			       SPDK_BLOB_WRITE_ZEROES);
7390 }
7391 
7392 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
7393 			void *payload, uint64_t offset, uint64_t length,
7394 			spdk_blob_op_complete cb_fn, void *cb_arg)
7395 {
7396 	blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
7397 			       SPDK_BLOB_WRITE);
7398 }
7399 
7400 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
7401 		       void *payload, uint64_t offset, uint64_t length,
7402 		       spdk_blob_op_complete cb_fn, void *cb_arg)
7403 {
7404 	blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
7405 			       SPDK_BLOB_READ);
7406 }
7407 
7408 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
7409 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
7410 			 spdk_blob_op_complete cb_fn, void *cb_arg)
7411 {
7412 	blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
7413 }
7414 
7415 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
7416 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
7417 			spdk_blob_op_complete cb_fn, void *cb_arg)
7418 {
7419 	blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
7420 }
7421 
7422 struct spdk_bs_iter_ctx {
7423 	int64_t page_num;
7424 	struct spdk_blob_store *bs;
7425 
7426 	spdk_blob_op_with_handle_complete cb_fn;
7427 	void *cb_arg;
7428 };
7429 
7430 static void
7431 bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
7432 {
7433 	struct spdk_bs_iter_ctx *ctx = cb_arg;
7434 	struct spdk_blob_store *bs = ctx->bs;
7435 	spdk_blob_id id;
7436 
7437 	if (bserrno == 0) {
7438 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
7439 		free(ctx);
7440 		return;
7441 	}
7442 
7443 	ctx->page_num++;
7444 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
7445 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
7446 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
7447 		free(ctx);
7448 		return;
7449 	}
7450 
7451 	id = bs_page_to_blobid(ctx->page_num);
7452 
7453 	spdk_bs_open_blob(bs, id, bs_iter_cpl, ctx);
7454 }
7455 
7456 void
7457 spdk_bs_iter_first(struct spdk_blob_store *bs,
7458 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7459 {
7460 	struct spdk_bs_iter_ctx *ctx;
7461 
7462 	ctx = calloc(1, sizeof(*ctx));
7463 	if (!ctx) {
7464 		cb_fn(cb_arg, NULL, -ENOMEM);
7465 		return;
7466 	}
7467 
7468 	ctx->page_num = -1;
7469 	ctx->bs = bs;
7470 	ctx->cb_fn = cb_fn;
7471 	ctx->cb_arg = cb_arg;
7472 
7473 	bs_iter_cpl(ctx, NULL, -1);
7474 }
7475 
7476 static void
7477 bs_iter_close_cpl(void *cb_arg, int bserrno)
7478 {
7479 	struct spdk_bs_iter_ctx *ctx = cb_arg;
7480 
7481 	bs_iter_cpl(ctx, NULL, -1);
7482 }
7483 
7484 void
7485 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
7486 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7487 {
7488 	struct spdk_bs_iter_ctx *ctx;
7489 
7490 	assert(blob != NULL);
7491 
7492 	ctx = calloc(1, sizeof(*ctx));
7493 	if (!ctx) {
7494 		cb_fn(cb_arg, NULL, -ENOMEM);
7495 		return;
7496 	}
7497 
7498 	ctx->page_num = bs_blobid_to_page(blob->id);
7499 	ctx->bs = bs;
7500 	ctx->cb_fn = cb_fn;
7501 	ctx->cb_arg = cb_arg;
7502 
7503 	/* Close the existing blob */
7504 	spdk_blob_close(blob, bs_iter_close_cpl, ctx);
7505 }
7506 
7507 static int
7508 blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
7509 	       uint16_t value_len, bool internal)
7510 {
7511 	struct spdk_xattr_tailq *xattrs;
7512 	struct spdk_xattr	*xattr;
7513 	size_t			desc_size;
7514 	void			*tmp;
7515 
7516 	blob_verify_md_op(blob);
7517 
7518 	if (blob->md_ro) {
7519 		return -EPERM;
7520 	}
7521 
7522 	desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len;
7523 	if (desc_size > SPDK_BS_MAX_DESC_SIZE) {
7524 		SPDK_DEBUGLOG(blob, "Xattr '%s' of size %zu does not fix into single page %zu\n", name,
7525 			      desc_size, SPDK_BS_MAX_DESC_SIZE);
7526 		return -ENOMEM;
7527 	}
7528 
7529 	if (internal) {
7530 		xattrs = &blob->xattrs_internal;
7531 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
7532 	} else {
7533 		xattrs = &blob->xattrs;
7534 	}
7535 
7536 	TAILQ_FOREACH(xattr, xattrs, link) {
7537 		if (!strcmp(name, xattr->name)) {
7538 			tmp = malloc(value_len);
7539 			if (!tmp) {
7540 				return -ENOMEM;
7541 			}
7542 
7543 			free(xattr->value);
7544 			xattr->value_len = value_len;
7545 			xattr->value = tmp;
7546 			memcpy(xattr->value, value, value_len);
7547 
7548 			blob->state = SPDK_BLOB_STATE_DIRTY;
7549 
7550 			return 0;
7551 		}
7552 	}
7553 
7554 	xattr = calloc(1, sizeof(*xattr));
7555 	if (!xattr) {
7556 		return -ENOMEM;
7557 	}
7558 
7559 	xattr->name = strdup(name);
7560 	if (!xattr->name) {
7561 		free(xattr);
7562 		return -ENOMEM;
7563 	}
7564 
7565 	xattr->value_len = value_len;
7566 	xattr->value = malloc(value_len);
7567 	if (!xattr->value) {
7568 		free(xattr->name);
7569 		free(xattr);
7570 		return -ENOMEM;
7571 	}
7572 	memcpy(xattr->value, value, value_len);
7573 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
7574 
7575 	blob->state = SPDK_BLOB_STATE_DIRTY;
7576 
7577 	return 0;
7578 }
7579 
7580 int
7581 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
7582 		    uint16_t value_len)
7583 {
7584 	return blob_set_xattr(blob, name, value, value_len, false);
7585 }
7586 
7587 static int
7588 blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
7589 {
7590 	struct spdk_xattr_tailq *xattrs;
7591 	struct spdk_xattr	*xattr;
7592 
7593 	blob_verify_md_op(blob);
7594 
7595 	if (blob->md_ro) {
7596 		return -EPERM;
7597 	}
7598 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
7599 
7600 	TAILQ_FOREACH(xattr, xattrs, link) {
7601 		if (!strcmp(name, xattr->name)) {
7602 			TAILQ_REMOVE(xattrs, xattr, link);
7603 			free(xattr->value);
7604 			free(xattr->name);
7605 			free(xattr);
7606 
7607 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
7608 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
7609 			}
7610 			blob->state = SPDK_BLOB_STATE_DIRTY;
7611 
7612 			return 0;
7613 		}
7614 	}
7615 
7616 	return -ENOENT;
7617 }
7618 
7619 int
7620 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
7621 {
7622 	return blob_remove_xattr(blob, name, false);
7623 }
7624 
7625 static int
7626 blob_get_xattr_value(struct spdk_blob *blob, const char *name,
7627 		     const void **value, size_t *value_len, bool internal)
7628 {
7629 	struct spdk_xattr	*xattr;
7630 	struct spdk_xattr_tailq *xattrs;
7631 
7632 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
7633 
7634 	TAILQ_FOREACH(xattr, xattrs, link) {
7635 		if (!strcmp(name, xattr->name)) {
7636 			*value = xattr->value;
7637 			*value_len = xattr->value_len;
7638 			return 0;
7639 		}
7640 	}
7641 	return -ENOENT;
7642 }
7643 
7644 int
7645 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
7646 			  const void **value, size_t *value_len)
7647 {
7648 	blob_verify_md_op(blob);
7649 
7650 	return blob_get_xattr_value(blob, name, value, value_len, false);
7651 }
7652 
7653 struct spdk_xattr_names {
7654 	uint32_t	count;
7655 	const char	*names[0];
7656 };
7657 
7658 static int
7659 blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
7660 {
7661 	struct spdk_xattr	*xattr;
7662 	int			count = 0;
7663 
7664 	TAILQ_FOREACH(xattr, xattrs, link) {
7665 		count++;
7666 	}
7667 
7668 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
7669 	if (*names == NULL) {
7670 		return -ENOMEM;
7671 	}
7672 
7673 	TAILQ_FOREACH(xattr, xattrs, link) {
7674 		(*names)->names[(*names)->count++] = xattr->name;
7675 	}
7676 
7677 	return 0;
7678 }
7679 
7680 int
7681 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
7682 {
7683 	blob_verify_md_op(blob);
7684 
7685 	return blob_get_xattr_names(&blob->xattrs, names);
7686 }
7687 
7688 uint32_t
7689 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
7690 {
7691 	assert(names != NULL);
7692 
7693 	return names->count;
7694 }
7695 
7696 const char *
7697 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
7698 {
7699 	if (index >= names->count) {
7700 		return NULL;
7701 	}
7702 
7703 	return names->names[index];
7704 }
7705 
7706 void
7707 spdk_xattr_names_free(struct spdk_xattr_names *names)
7708 {
7709 	free(names);
7710 }
7711 
7712 struct spdk_bs_type
7713 spdk_bs_get_bstype(struct spdk_blob_store *bs)
7714 {
7715 	return bs->bstype;
7716 }
7717 
7718 void
7719 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
7720 {
7721 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
7722 }
7723 
7724 bool
7725 spdk_blob_is_read_only(struct spdk_blob *blob)
7726 {
7727 	assert(blob != NULL);
7728 	return (blob->data_ro || blob->md_ro);
7729 }
7730 
7731 bool
7732 spdk_blob_is_snapshot(struct spdk_blob *blob)
7733 {
7734 	struct spdk_blob_list *snapshot_entry;
7735 
7736 	assert(blob != NULL);
7737 
7738 	snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
7739 	if (snapshot_entry == NULL) {
7740 		return false;
7741 	}
7742 
7743 	return true;
7744 }
7745 
7746 bool
7747 spdk_blob_is_clone(struct spdk_blob *blob)
7748 {
7749 	assert(blob != NULL);
7750 
7751 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
7752 		assert(spdk_blob_is_thin_provisioned(blob));
7753 		return true;
7754 	}
7755 
7756 	return false;
7757 }
7758 
7759 bool
7760 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
7761 {
7762 	assert(blob != NULL);
7763 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
7764 }
7765 
7766 static void
7767 blob_update_clear_method(struct spdk_blob *blob)
7768 {
7769 	enum blob_clear_method stored_cm;
7770 
7771 	assert(blob != NULL);
7772 
7773 	/* If BLOB_CLEAR_WITH_DEFAULT was passed in, use the setting stored
7774 	 * in metadata previously.  If something other than the default was
7775 	 * specified, ignore stored value and used what was passed in.
7776 	 */
7777 	stored_cm = ((blob->md_ro_flags & SPDK_BLOB_CLEAR_METHOD) >> SPDK_BLOB_CLEAR_METHOD_SHIFT);
7778 
7779 	if (blob->clear_method == BLOB_CLEAR_WITH_DEFAULT) {
7780 		blob->clear_method = stored_cm;
7781 	} else if (blob->clear_method != stored_cm) {
7782 		SPDK_WARNLOG("Using passed in clear method 0x%x instead of stored value of 0x%x\n",
7783 			     blob->clear_method, stored_cm);
7784 	}
7785 }
7786 
7787 spdk_blob_id
7788 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
7789 {
7790 	struct spdk_blob_list *snapshot_entry = NULL;
7791 	struct spdk_blob_list *clone_entry = NULL;
7792 
7793 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
7794 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
7795 			if (clone_entry->id == blob_id) {
7796 				return snapshot_entry->id;
7797 			}
7798 		}
7799 	}
7800 
7801 	return SPDK_BLOBID_INVALID;
7802 }
7803 
7804 int
7805 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
7806 		     size_t *count)
7807 {
7808 	struct spdk_blob_list *snapshot_entry, *clone_entry;
7809 	size_t n;
7810 
7811 	snapshot_entry = bs_get_snapshot_entry(bs, blobid);
7812 	if (snapshot_entry == NULL) {
7813 		*count = 0;
7814 		return 0;
7815 	}
7816 
7817 	if (ids == NULL || *count < snapshot_entry->clone_count) {
7818 		*count = snapshot_entry->clone_count;
7819 		return -ENOMEM;
7820 	}
7821 	*count = snapshot_entry->clone_count;
7822 
7823 	n = 0;
7824 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
7825 		ids[n++] = clone_entry->id;
7826 	}
7827 
7828 	return 0;
7829 }
7830 
7831 SPDK_LOG_REGISTER_COMPONENT(blob)
7832