xref: /spdk/lib/blob/blobstore.c (revision 5c29449f4ed63752f891f92b10f7e61d9587c197)
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 	bool in_submit_ctx;
2537 	bool completed_in_submit_ctx;
2538 	bool done;
2539 };
2540 
2541 static void
2542 blob_request_submit_op_split_next(void *cb_arg, int bserrno)
2543 {
2544 	struct op_split_ctx	*ctx = cb_arg;
2545 	struct spdk_blob	*blob = ctx->blob;
2546 	struct spdk_io_channel	*ch = ctx->channel;
2547 	enum spdk_blob_op_type	op_type = ctx->op_type;
2548 	uint8_t			*buf;
2549 	uint64_t		offset;
2550 	uint64_t		length;
2551 	uint64_t		op_length;
2552 
2553 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2554 		bs_sequence_finish(ctx->seq, bserrno);
2555 		if (ctx->in_submit_ctx) {
2556 			/* Defer freeing of the ctx object, since it will be
2557 			 * accessed when this unwinds back to the submisison
2558 			 * context.
2559 			 */
2560 			ctx->done = true;
2561 		} else {
2562 			free(ctx);
2563 		}
2564 		return;
2565 	}
2566 
2567 	if (ctx->in_submit_ctx) {
2568 		/* If this split operation completed in the context
2569 		 * of its submission, mark the flag and return immediately
2570 		 * to avoid recursion.
2571 		 */
2572 		ctx->completed_in_submit_ctx = true;
2573 		return;
2574 	}
2575 
2576 	while (true) {
2577 		ctx->completed_in_submit_ctx = false;
2578 
2579 		offset = ctx->io_unit_offset;
2580 		length = ctx->io_units_remaining;
2581 		buf = ctx->curr_payload;
2582 		op_length = spdk_min(length, bs_num_io_units_to_cluster_boundary(blob,
2583 				     offset));
2584 
2585 		/* Update length and payload for next operation */
2586 		ctx->io_units_remaining -= op_length;
2587 		ctx->io_unit_offset += op_length;
2588 		if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
2589 			ctx->curr_payload += op_length * blob->bs->io_unit_size;
2590 		}
2591 
2592 		assert(!ctx->in_submit_ctx);
2593 		ctx->in_submit_ctx = true;
2594 
2595 		switch (op_type) {
2596 		case SPDK_BLOB_READ:
2597 			spdk_blob_io_read(blob, ch, buf, offset, op_length,
2598 					  blob_request_submit_op_split_next, ctx);
2599 			break;
2600 		case SPDK_BLOB_WRITE:
2601 			spdk_blob_io_write(blob, ch, buf, offset, op_length,
2602 					   blob_request_submit_op_split_next, ctx);
2603 			break;
2604 		case SPDK_BLOB_UNMAP:
2605 			spdk_blob_io_unmap(blob, ch, offset, op_length,
2606 					   blob_request_submit_op_split_next, ctx);
2607 			break;
2608 		case SPDK_BLOB_WRITE_ZEROES:
2609 			spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
2610 						  blob_request_submit_op_split_next, ctx);
2611 			break;
2612 		case SPDK_BLOB_READV:
2613 		case SPDK_BLOB_WRITEV:
2614 			SPDK_ERRLOG("readv/write not valid\n");
2615 			bs_sequence_finish(ctx->seq, -EINVAL);
2616 			free(ctx);
2617 			return;
2618 		}
2619 
2620 #ifndef __clang_analyzer__
2621 		/* scan-build reports a false positive around accessing the ctx here. It
2622 		 * forms a path that recursively calls this function, but then says
2623 		 * "assuming ctx->in_submit_ctx is false", when that isn't possible.
2624 		 * This path does free(ctx), returns to here, and reports a use-after-free
2625 		 * bug.  Wrapping this bit of code so that scan-build doesn't see it
2626 		 * works around the scan-build bug.
2627 		 */
2628 		assert(ctx->in_submit_ctx);
2629 		ctx->in_submit_ctx = false;
2630 
2631 		/* If the operation completed immediately, loop back and submit the
2632 		 * next operation.  Otherwise we can return and the next split
2633 		 * operation will get submitted when this current operation is
2634 		 * later completed asynchronously.
2635 		 */
2636 		if (ctx->completed_in_submit_ctx) {
2637 			continue;
2638 		} else if (ctx->done) {
2639 			free(ctx);
2640 		}
2641 #endif
2642 		break;
2643 	}
2644 }
2645 
2646 static void
2647 blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
2648 			     void *payload, uint64_t offset, uint64_t length,
2649 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2650 {
2651 	struct op_split_ctx *ctx;
2652 	spdk_bs_sequence_t *seq;
2653 	struct spdk_bs_cpl cpl;
2654 
2655 	assert(blob != NULL);
2656 
2657 	ctx = calloc(1, sizeof(struct op_split_ctx));
2658 	if (ctx == NULL) {
2659 		cb_fn(cb_arg, -ENOMEM);
2660 		return;
2661 	}
2662 
2663 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2664 	cpl.u.blob_basic.cb_fn = cb_fn;
2665 	cpl.u.blob_basic.cb_arg = cb_arg;
2666 
2667 	seq = bs_sequence_start(ch, &cpl);
2668 	if (!seq) {
2669 		free(ctx);
2670 		cb_fn(cb_arg, -ENOMEM);
2671 		return;
2672 	}
2673 
2674 	ctx->blob = blob;
2675 	ctx->channel = ch;
2676 	ctx->curr_payload = payload;
2677 	ctx->io_unit_offset = offset;
2678 	ctx->io_units_remaining = length;
2679 	ctx->op_type = op_type;
2680 	ctx->seq = seq;
2681 
2682 	blob_request_submit_op_split_next(ctx, 0);
2683 }
2684 
2685 static void
2686 blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
2687 			      void *payload, uint64_t offset, uint64_t length,
2688 			      spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2689 {
2690 	struct spdk_bs_cpl cpl;
2691 	uint64_t lba;
2692 	uint64_t lba_count;
2693 	bool is_allocated;
2694 
2695 	assert(blob != NULL);
2696 
2697 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2698 	cpl.u.blob_basic.cb_fn = cb_fn;
2699 	cpl.u.blob_basic.cb_arg = cb_arg;
2700 
2701 	if (blob->frozen_refcnt) {
2702 		/* This blob I/O is frozen */
2703 		spdk_bs_user_op_t *op;
2704 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
2705 
2706 		op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
2707 		if (!op) {
2708 			cb_fn(cb_arg, -ENOMEM);
2709 			return;
2710 		}
2711 
2712 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2713 
2714 		return;
2715 	}
2716 
2717 	is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2718 
2719 	switch (op_type) {
2720 	case SPDK_BLOB_READ: {
2721 		spdk_bs_batch_t *batch;
2722 
2723 		batch = bs_batch_open(_ch, &cpl);
2724 		if (!batch) {
2725 			cb_fn(cb_arg, -ENOMEM);
2726 			return;
2727 		}
2728 
2729 		if (is_allocated) {
2730 			/* Read from the blob */
2731 			bs_batch_read_dev(batch, payload, lba, lba_count);
2732 		} else {
2733 			/* Read from the backing block device */
2734 			bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
2735 		}
2736 
2737 		bs_batch_close(batch);
2738 		break;
2739 	}
2740 	case SPDK_BLOB_WRITE:
2741 	case SPDK_BLOB_WRITE_ZEROES: {
2742 		if (is_allocated) {
2743 			/* Write to the blob */
2744 			spdk_bs_batch_t *batch;
2745 
2746 			if (lba_count == 0) {
2747 				cb_fn(cb_arg, 0);
2748 				return;
2749 			}
2750 
2751 			batch = bs_batch_open(_ch, &cpl);
2752 			if (!batch) {
2753 				cb_fn(cb_arg, -ENOMEM);
2754 				return;
2755 			}
2756 
2757 			if (op_type == SPDK_BLOB_WRITE) {
2758 				bs_batch_write_dev(batch, payload, lba, lba_count);
2759 			} else {
2760 				bs_batch_write_zeroes_dev(batch, lba, lba_count);
2761 			}
2762 
2763 			bs_batch_close(batch);
2764 		} else {
2765 			/* Queue this operation and allocate the cluster */
2766 			spdk_bs_user_op_t *op;
2767 
2768 			op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
2769 			if (!op) {
2770 				cb_fn(cb_arg, -ENOMEM);
2771 				return;
2772 			}
2773 
2774 			bs_allocate_and_copy_cluster(blob, _ch, offset, op);
2775 		}
2776 		break;
2777 	}
2778 	case SPDK_BLOB_UNMAP: {
2779 		spdk_bs_batch_t *batch;
2780 
2781 		batch = bs_batch_open(_ch, &cpl);
2782 		if (!batch) {
2783 			cb_fn(cb_arg, -ENOMEM);
2784 			return;
2785 		}
2786 
2787 		if (is_allocated) {
2788 			bs_batch_unmap_dev(batch, lba, lba_count);
2789 		}
2790 
2791 		bs_batch_close(batch);
2792 		break;
2793 	}
2794 	case SPDK_BLOB_READV:
2795 	case SPDK_BLOB_WRITEV:
2796 		SPDK_ERRLOG("readv/write not valid\n");
2797 		cb_fn(cb_arg, -EINVAL);
2798 		break;
2799 	}
2800 }
2801 
2802 static void
2803 blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2804 		       void *payload, uint64_t offset, uint64_t length,
2805 		       spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2806 {
2807 	assert(blob != NULL);
2808 
2809 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
2810 		cb_fn(cb_arg, -EPERM);
2811 		return;
2812 	}
2813 
2814 	if (length == 0) {
2815 		cb_fn(cb_arg, 0);
2816 		return;
2817 	}
2818 
2819 	if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2820 		cb_fn(cb_arg, -EINVAL);
2821 		return;
2822 	}
2823 	if (length <= bs_num_io_units_to_cluster_boundary(blob, offset)) {
2824 		blob_request_submit_op_single(_channel, blob, payload, offset, length,
2825 					      cb_fn, cb_arg, op_type);
2826 	} else {
2827 		blob_request_submit_op_split(_channel, blob, payload, offset, length,
2828 					     cb_fn, cb_arg, op_type);
2829 	}
2830 }
2831 
2832 struct rw_iov_ctx {
2833 	struct spdk_blob *blob;
2834 	struct spdk_io_channel *channel;
2835 	spdk_blob_op_complete cb_fn;
2836 	void *cb_arg;
2837 	bool read;
2838 	int iovcnt;
2839 	struct iovec *orig_iov;
2840 	uint64_t io_unit_offset;
2841 	uint64_t io_units_remaining;
2842 	uint64_t io_units_done;
2843 	struct iovec iov[0];
2844 };
2845 
2846 static void
2847 rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2848 {
2849 	assert(cb_arg == NULL);
2850 	bs_sequence_finish(seq, bserrno);
2851 }
2852 
2853 static void
2854 rw_iov_split_next(void *cb_arg, int bserrno)
2855 {
2856 	struct rw_iov_ctx *ctx = cb_arg;
2857 	struct spdk_blob *blob = ctx->blob;
2858 	struct iovec *iov, *orig_iov;
2859 	int iovcnt;
2860 	size_t orig_iovoff;
2861 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2862 	uint64_t byte_count;
2863 
2864 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2865 		ctx->cb_fn(ctx->cb_arg, bserrno);
2866 		free(ctx);
2867 		return;
2868 	}
2869 
2870 	io_unit_offset = ctx->io_unit_offset;
2871 	io_units_to_boundary = bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2872 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2873 	/*
2874 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2875 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2876 	 *  point to the current position in the I/O sequence.
2877 	 */
2878 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2879 	orig_iov = &ctx->orig_iov[0];
2880 	orig_iovoff = 0;
2881 	while (byte_count > 0) {
2882 		if (byte_count >= orig_iov->iov_len) {
2883 			byte_count -= orig_iov->iov_len;
2884 			orig_iov++;
2885 		} else {
2886 			orig_iovoff = byte_count;
2887 			byte_count = 0;
2888 		}
2889 	}
2890 
2891 	/*
2892 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2893 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2894 	 */
2895 	byte_count = io_units_count * blob->bs->io_unit_size;
2896 	iov = &ctx->iov[0];
2897 	iovcnt = 0;
2898 	while (byte_count > 0) {
2899 		assert(iovcnt < ctx->iovcnt);
2900 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2901 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2902 		byte_count -= iov->iov_len;
2903 		orig_iovoff = 0;
2904 		orig_iov++;
2905 		iov++;
2906 		iovcnt++;
2907 	}
2908 
2909 	ctx->io_unit_offset += io_units_count;
2910 	ctx->io_units_remaining -= io_units_count;
2911 	ctx->io_units_done += io_units_count;
2912 	iov = &ctx->iov[0];
2913 
2914 	if (ctx->read) {
2915 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2916 				   io_units_count, rw_iov_split_next, ctx);
2917 	} else {
2918 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2919 				    io_units_count, rw_iov_split_next, ctx);
2920 	}
2921 }
2922 
2923 static void
2924 blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2925 			   struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2926 			   spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2927 {
2928 	struct spdk_bs_cpl	cpl;
2929 
2930 	assert(blob != NULL);
2931 
2932 	if (!read && blob->data_ro) {
2933 		cb_fn(cb_arg, -EPERM);
2934 		return;
2935 	}
2936 
2937 	if (length == 0) {
2938 		cb_fn(cb_arg, 0);
2939 		return;
2940 	}
2941 
2942 	if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2943 		cb_fn(cb_arg, -EINVAL);
2944 		return;
2945 	}
2946 
2947 	/*
2948 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2949 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2950 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2951 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2952 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2953 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2954 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2955 	 *
2956 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2957 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2958 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2959 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2960 	 */
2961 	if (spdk_likely(length <= bs_num_io_units_to_cluster_boundary(blob, offset))) {
2962 		uint64_t lba_count;
2963 		uint64_t lba;
2964 		bool is_allocated;
2965 
2966 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2967 		cpl.u.blob_basic.cb_fn = cb_fn;
2968 		cpl.u.blob_basic.cb_arg = cb_arg;
2969 
2970 		if (blob->frozen_refcnt) {
2971 			/* This blob I/O is frozen */
2972 			enum spdk_blob_op_type op_type;
2973 			spdk_bs_user_op_t *op;
2974 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2975 
2976 			op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV;
2977 			op = bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length);
2978 			if (!op) {
2979 				cb_fn(cb_arg, -ENOMEM);
2980 				return;
2981 			}
2982 
2983 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2984 
2985 			return;
2986 		}
2987 
2988 		is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2989 
2990 		if (read) {
2991 			spdk_bs_sequence_t *seq;
2992 
2993 			seq = bs_sequence_start(_channel, &cpl);
2994 			if (!seq) {
2995 				cb_fn(cb_arg, -ENOMEM);
2996 				return;
2997 			}
2998 
2999 			if (is_allocated) {
3000 				bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL);
3001 			} else {
3002 				bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
3003 							 rw_iov_done, NULL);
3004 			}
3005 		} else {
3006 			if (is_allocated) {
3007 				spdk_bs_sequence_t *seq;
3008 
3009 				seq = bs_sequence_start(_channel, &cpl);
3010 				if (!seq) {
3011 					cb_fn(cb_arg, -ENOMEM);
3012 					return;
3013 				}
3014 
3015 				bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL);
3016 			} else {
3017 				/* Queue this operation and allocate the cluster */
3018 				spdk_bs_user_op_t *op;
3019 
3020 				op = bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
3021 						      length);
3022 				if (!op) {
3023 					cb_fn(cb_arg, -ENOMEM);
3024 					return;
3025 				}
3026 
3027 				bs_allocate_and_copy_cluster(blob, _channel, offset, op);
3028 			}
3029 		}
3030 	} else {
3031 		struct rw_iov_ctx *ctx;
3032 
3033 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
3034 		if (ctx == NULL) {
3035 			cb_fn(cb_arg, -ENOMEM);
3036 			return;
3037 		}
3038 
3039 		ctx->blob = blob;
3040 		ctx->channel = _channel;
3041 		ctx->cb_fn = cb_fn;
3042 		ctx->cb_arg = cb_arg;
3043 		ctx->read = read;
3044 		ctx->orig_iov = iov;
3045 		ctx->iovcnt = iovcnt;
3046 		ctx->io_unit_offset = offset;
3047 		ctx->io_units_remaining = length;
3048 		ctx->io_units_done = 0;
3049 
3050 		rw_iov_split_next(ctx, 0);
3051 	}
3052 }
3053 
3054 static struct spdk_blob *
3055 blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
3056 {
3057 	struct spdk_blob find;
3058 
3059 	if (spdk_bit_array_get(bs->open_blobids, blobid) == 0) {
3060 		return NULL;
3061 	}
3062 
3063 	find.id = blobid;
3064 	return RB_FIND(spdk_blob_tree, &bs->open_blobs, &find);
3065 }
3066 
3067 static void
3068 blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
3069 				    struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
3070 {
3071 	assert(blob != NULL);
3072 	*snapshot_entry = NULL;
3073 	*clone_entry = NULL;
3074 
3075 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
3076 		return;
3077 	}
3078 
3079 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
3080 		if ((*snapshot_entry)->id == blob->parent_id) {
3081 			break;
3082 		}
3083 	}
3084 
3085 	if (*snapshot_entry != NULL) {
3086 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
3087 			if ((*clone_entry)->id == blob->id) {
3088 				break;
3089 			}
3090 		}
3091 
3092 		assert(*clone_entry != NULL);
3093 	}
3094 }
3095 
3096 static int
3097 bs_channel_create(void *io_device, void *ctx_buf)
3098 {
3099 	struct spdk_blob_store		*bs = io_device;
3100 	struct spdk_bs_channel		*channel = ctx_buf;
3101 	struct spdk_bs_dev		*dev;
3102 	uint32_t			max_ops = bs->max_channel_ops;
3103 	uint32_t			i;
3104 
3105 	dev = bs->dev;
3106 
3107 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
3108 	if (!channel->req_mem) {
3109 		return -1;
3110 	}
3111 
3112 	TAILQ_INIT(&channel->reqs);
3113 
3114 	for (i = 0; i < max_ops; i++) {
3115 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
3116 	}
3117 
3118 	channel->bs = bs;
3119 	channel->dev = dev;
3120 	channel->dev_channel = dev->create_channel(dev);
3121 
3122 	if (!channel->dev_channel) {
3123 		SPDK_ERRLOG("Failed to create device channel.\n");
3124 		free(channel->req_mem);
3125 		return -1;
3126 	}
3127 
3128 	TAILQ_INIT(&channel->need_cluster_alloc);
3129 	TAILQ_INIT(&channel->queued_io);
3130 
3131 	return 0;
3132 }
3133 
3134 static void
3135 bs_channel_destroy(void *io_device, void *ctx_buf)
3136 {
3137 	struct spdk_bs_channel *channel = ctx_buf;
3138 	spdk_bs_user_op_t *op;
3139 
3140 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
3141 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
3142 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
3143 		bs_user_op_abort(op);
3144 	}
3145 
3146 	while (!TAILQ_EMPTY(&channel->queued_io)) {
3147 		op = TAILQ_FIRST(&channel->queued_io);
3148 		TAILQ_REMOVE(&channel->queued_io, op, link);
3149 		bs_user_op_abort(op);
3150 	}
3151 
3152 	free(channel->req_mem);
3153 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
3154 }
3155 
3156 static void
3157 bs_dev_destroy(void *io_device)
3158 {
3159 	struct spdk_blob_store *bs = io_device;
3160 	struct spdk_blob	*blob, *blob_tmp;
3161 
3162 	bs->dev->destroy(bs->dev);
3163 
3164 	RB_FOREACH_SAFE(blob, spdk_blob_tree, &bs->open_blobs, blob_tmp) {
3165 		RB_REMOVE(spdk_blob_tree, &bs->open_blobs, blob);
3166 		spdk_bit_array_clear(bs->open_blobids, blob->id);
3167 		blob_free(blob);
3168 	}
3169 
3170 	pthread_mutex_destroy(&bs->used_clusters_mutex);
3171 
3172 	spdk_bit_array_free(&bs->open_blobids);
3173 	spdk_bit_array_free(&bs->used_blobids);
3174 	spdk_bit_array_free(&bs->used_md_pages);
3175 	spdk_bit_pool_free(&bs->used_clusters);
3176 	/*
3177 	 * If this function is called for any reason except a successful unload,
3178 	 * the unload_cpl type will be NONE and this will be a nop.
3179 	 */
3180 	bs_call_cpl(&bs->unload_cpl, bs->unload_err);
3181 
3182 	free(bs);
3183 }
3184 
3185 static int
3186 bs_blob_list_add(struct spdk_blob *blob)
3187 {
3188 	spdk_blob_id snapshot_id;
3189 	struct spdk_blob_list *snapshot_entry = NULL;
3190 	struct spdk_blob_list *clone_entry = NULL;
3191 
3192 	assert(blob != NULL);
3193 
3194 	snapshot_id = blob->parent_id;
3195 	if (snapshot_id == SPDK_BLOBID_INVALID) {
3196 		return 0;
3197 	}
3198 
3199 	snapshot_entry = bs_get_snapshot_entry(blob->bs, snapshot_id);
3200 	if (snapshot_entry == NULL) {
3201 		/* Snapshot not found */
3202 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
3203 		if (snapshot_entry == NULL) {
3204 			return -ENOMEM;
3205 		}
3206 		snapshot_entry->id = snapshot_id;
3207 		TAILQ_INIT(&snapshot_entry->clones);
3208 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
3209 	} else {
3210 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
3211 			if (clone_entry->id == blob->id) {
3212 				break;
3213 			}
3214 		}
3215 	}
3216 
3217 	if (clone_entry == NULL) {
3218 		/* Clone not found */
3219 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
3220 		if (clone_entry == NULL) {
3221 			return -ENOMEM;
3222 		}
3223 		clone_entry->id = blob->id;
3224 		TAILQ_INIT(&clone_entry->clones);
3225 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
3226 		snapshot_entry->clone_count++;
3227 	}
3228 
3229 	return 0;
3230 }
3231 
3232 static void
3233 bs_blob_list_remove(struct spdk_blob *blob)
3234 {
3235 	struct spdk_blob_list *snapshot_entry = NULL;
3236 	struct spdk_blob_list *clone_entry = NULL;
3237 
3238 	blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
3239 
3240 	if (snapshot_entry == NULL) {
3241 		return;
3242 	}
3243 
3244 	blob->parent_id = SPDK_BLOBID_INVALID;
3245 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
3246 	free(clone_entry);
3247 
3248 	snapshot_entry->clone_count--;
3249 }
3250 
3251 static int
3252 bs_blob_list_free(struct spdk_blob_store *bs)
3253 {
3254 	struct spdk_blob_list *snapshot_entry;
3255 	struct spdk_blob_list *snapshot_entry_tmp;
3256 	struct spdk_blob_list *clone_entry;
3257 	struct spdk_blob_list *clone_entry_tmp;
3258 
3259 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
3260 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
3261 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
3262 			free(clone_entry);
3263 		}
3264 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
3265 		free(snapshot_entry);
3266 	}
3267 
3268 	return 0;
3269 }
3270 
3271 static void
3272 bs_free(struct spdk_blob_store *bs)
3273 {
3274 	bs_blob_list_free(bs);
3275 
3276 	bs_unregister_md_thread(bs);
3277 	spdk_io_device_unregister(bs, bs_dev_destroy);
3278 }
3279 
3280 void
3281 spdk_bs_opts_init(struct spdk_bs_opts *opts, size_t opts_size)
3282 {
3283 
3284 	if (!opts) {
3285 		SPDK_ERRLOG("opts should not be NULL\n");
3286 		return;
3287 	}
3288 
3289 	if (!opts_size) {
3290 		SPDK_ERRLOG("opts_size should not be zero value\n");
3291 		return;
3292 	}
3293 
3294 	memset(opts, 0, opts_size);
3295 	opts->opts_size = opts_size;
3296 
3297 #define FIELD_OK(field) \
3298 	offsetof(struct spdk_bs_opts, field) + sizeof(opts->field) <= opts_size
3299 
3300 #define SET_FIELD(field, value) \
3301 	if (FIELD_OK(field)) { \
3302 		opts->field = value; \
3303 	} \
3304 
3305 	SET_FIELD(cluster_sz, SPDK_BLOB_OPTS_CLUSTER_SZ);
3306 	SET_FIELD(num_md_pages, SPDK_BLOB_OPTS_NUM_MD_PAGES);
3307 	SET_FIELD(max_md_ops, SPDK_BLOB_OPTS_NUM_MD_PAGES);
3308 	SET_FIELD(max_channel_ops, SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS);
3309 	SET_FIELD(clear_method,  BS_CLEAR_WITH_UNMAP);
3310 
3311 	if (FIELD_OK(bstype)) {
3312 		memset(&opts->bstype, 0, sizeof(opts->bstype));
3313 	}
3314 
3315 	SET_FIELD(iter_cb_fn, NULL);
3316 	SET_FIELD(iter_cb_arg, NULL);
3317 	SET_FIELD(force_recover, false);
3318 
3319 #undef FIELD_OK
3320 #undef SET_FIELD
3321 }
3322 
3323 static int
3324 bs_opts_verify(struct spdk_bs_opts *opts)
3325 {
3326 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
3327 	    opts->max_channel_ops == 0) {
3328 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
3329 		return -1;
3330 	}
3331 
3332 	return 0;
3333 }
3334 
3335 /* START spdk_bs_load */
3336 
3337 /* spdk_bs_load_ctx is used for init, load, unload and dump code paths. */
3338 
3339 struct spdk_bs_load_ctx {
3340 	struct spdk_blob_store		*bs;
3341 	struct spdk_bs_super_block	*super;
3342 
3343 	struct spdk_bs_md_mask		*mask;
3344 	bool				in_page_chain;
3345 	uint32_t			page_index;
3346 	uint32_t			cur_page;
3347 	struct spdk_blob_md_page	*page;
3348 
3349 	uint64_t			num_extent_pages;
3350 	uint32_t			*extent_page_num;
3351 	struct spdk_blob_md_page	*extent_pages;
3352 	struct spdk_bit_array		*used_clusters;
3353 
3354 	spdk_bs_sequence_t			*seq;
3355 	spdk_blob_op_with_handle_complete	iter_cb_fn;
3356 	void					*iter_cb_arg;
3357 	struct spdk_blob			*blob;
3358 	spdk_blob_id				blobid;
3359 
3360 	bool					force_recover;
3361 
3362 	/* These fields are used in the spdk_bs_dump path. */
3363 	bool					dumping;
3364 	FILE					*fp;
3365 	spdk_bs_dump_print_xattr		print_xattr_fn;
3366 	char					xattr_name[4096];
3367 };
3368 
3369 static int
3370 bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs,
3371 	 struct spdk_bs_load_ctx **_ctx)
3372 {
3373 	struct spdk_blob_store	*bs;
3374 	struct spdk_bs_load_ctx	*ctx;
3375 	uint64_t dev_size;
3376 	int rc;
3377 
3378 	dev_size = dev->blocklen * dev->blockcnt;
3379 	if (dev_size < opts->cluster_sz) {
3380 		/* Device size cannot be smaller than cluster size of blobstore */
3381 		SPDK_INFOLOG(blob, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
3382 			     dev_size, opts->cluster_sz);
3383 		return -ENOSPC;
3384 	}
3385 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
3386 		/* Cluster size cannot be smaller than page size */
3387 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
3388 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
3389 		return -EINVAL;
3390 	}
3391 	bs = calloc(1, sizeof(struct spdk_blob_store));
3392 	if (!bs) {
3393 		return -ENOMEM;
3394 	}
3395 
3396 	ctx = calloc(1, sizeof(struct spdk_bs_load_ctx));
3397 	if (!ctx) {
3398 		free(bs);
3399 		return -ENOMEM;
3400 	}
3401 
3402 	ctx->bs = bs;
3403 	ctx->iter_cb_fn = opts->iter_cb_fn;
3404 	ctx->iter_cb_arg = opts->iter_cb_arg;
3405 	ctx->force_recover = opts->force_recover;
3406 
3407 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
3408 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3409 	if (!ctx->super) {
3410 		free(ctx);
3411 		free(bs);
3412 		return -ENOMEM;
3413 	}
3414 
3415 	RB_INIT(&bs->open_blobs);
3416 	TAILQ_INIT(&bs->snapshots);
3417 	bs->dev = dev;
3418 	bs->md_thread = spdk_get_thread();
3419 	assert(bs->md_thread != NULL);
3420 
3421 	/*
3422 	 * Do not use bs_lba_to_cluster() here since blockcnt may not be an
3423 	 *  even multiple of the cluster size.
3424 	 */
3425 	bs->cluster_sz = opts->cluster_sz;
3426 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
3427 	ctx->used_clusters = spdk_bit_array_create(bs->total_clusters);
3428 	if (!ctx->used_clusters) {
3429 		spdk_free(ctx->super);
3430 		free(ctx);
3431 		free(bs);
3432 		return -ENOMEM;
3433 	}
3434 
3435 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3436 	if (spdk_u32_is_pow2(bs->pages_per_cluster)) {
3437 		bs->pages_per_cluster_shift = spdk_u32log2(bs->pages_per_cluster);
3438 	}
3439 	bs->num_free_clusters = bs->total_clusters;
3440 	bs->io_unit_size = dev->blocklen;
3441 
3442 	bs->max_channel_ops = opts->max_channel_ops;
3443 	bs->super_blob = SPDK_BLOBID_INVALID;
3444 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
3445 
3446 	/* The metadata is assumed to be at least 1 page */
3447 	bs->used_md_pages = spdk_bit_array_create(1);
3448 	bs->used_blobids = spdk_bit_array_create(0);
3449 	bs->open_blobids = spdk_bit_array_create(0);
3450 
3451 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
3452 
3453 	spdk_io_device_register(bs, bs_channel_create, bs_channel_destroy,
3454 				sizeof(struct spdk_bs_channel), "blobstore");
3455 	rc = bs_register_md_thread(bs);
3456 	if (rc == -1) {
3457 		spdk_io_device_unregister(bs, NULL);
3458 		pthread_mutex_destroy(&bs->used_clusters_mutex);
3459 		spdk_bit_array_free(&bs->open_blobids);
3460 		spdk_bit_array_free(&bs->used_blobids);
3461 		spdk_bit_array_free(&bs->used_md_pages);
3462 		spdk_bit_array_free(&ctx->used_clusters);
3463 		spdk_free(ctx->super);
3464 		free(ctx);
3465 		free(bs);
3466 		/* FIXME: this is a lie but don't know how to get a proper error code here */
3467 		return -ENOMEM;
3468 	}
3469 
3470 	*_ctx = ctx;
3471 	*_bs = bs;
3472 	return 0;
3473 }
3474 
3475 static void
3476 bs_load_ctx_fail(struct spdk_bs_load_ctx *ctx, int bserrno)
3477 {
3478 	assert(bserrno != 0);
3479 
3480 	spdk_free(ctx->super);
3481 	bs_sequence_finish(ctx->seq, bserrno);
3482 	bs_free(ctx->bs);
3483 	spdk_bit_array_free(&ctx->used_clusters);
3484 	free(ctx);
3485 }
3486 
3487 static void
3488 bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
3489 	       struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
3490 {
3491 	/* Update the values in the super block */
3492 	super->super_blob = bs->super_blob;
3493 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
3494 	super->crc = blob_md_page_calc_crc(super);
3495 	bs_sequence_write_dev(seq, super, bs_page_to_lba(bs, 0),
3496 			      bs_byte_to_lba(bs, sizeof(*super)),
3497 			      cb_fn, cb_arg);
3498 }
3499 
3500 static void
3501 bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3502 {
3503 	struct spdk_bs_load_ctx	*ctx = arg;
3504 	uint64_t	mask_size, lba, lba_count;
3505 
3506 	/* Write out the used clusters mask */
3507 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
3508 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3509 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3510 	if (!ctx->mask) {
3511 		bs_load_ctx_fail(ctx, -ENOMEM);
3512 		return;
3513 	}
3514 
3515 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
3516 	ctx->mask->length = ctx->bs->total_clusters;
3517 	/* We could get here through the normal unload path, or through dirty
3518 	 * shutdown recovery.  For the normal unload path, we use the mask from
3519 	 * the bit pool.  For dirty shutdown recovery, we don't have a bit pool yet -
3520 	 * only the bit array from the load ctx.
3521 	 */
3522 	if (ctx->bs->used_clusters) {
3523 		assert(ctx->mask->length == spdk_bit_pool_capacity(ctx->bs->used_clusters));
3524 		spdk_bit_pool_store_mask(ctx->bs->used_clusters, ctx->mask->mask);
3525 	} else {
3526 		assert(ctx->mask->length == spdk_bit_array_capacity(ctx->used_clusters));
3527 		spdk_bit_array_store_mask(ctx->used_clusters, ctx->mask->mask);
3528 	}
3529 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
3530 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
3531 	bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3532 }
3533 
3534 static void
3535 bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3536 {
3537 	struct spdk_bs_load_ctx	*ctx = arg;
3538 	uint64_t	mask_size, lba, lba_count;
3539 
3540 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
3541 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3542 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3543 	if (!ctx->mask) {
3544 		bs_load_ctx_fail(ctx, -ENOMEM);
3545 		return;
3546 	}
3547 
3548 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
3549 	ctx->mask->length = ctx->super->md_len;
3550 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
3551 
3552 	spdk_bit_array_store_mask(ctx->bs->used_md_pages, ctx->mask->mask);
3553 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3554 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3555 	bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3556 }
3557 
3558 static void
3559 bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3560 {
3561 	struct spdk_bs_load_ctx	*ctx = arg;
3562 	uint64_t	mask_size, lba, lba_count;
3563 
3564 	if (ctx->super->used_blobid_mask_len == 0) {
3565 		/*
3566 		 * This is a pre-v3 on-disk format where the blobid mask does not get
3567 		 *  written to disk.
3568 		 */
3569 		cb_fn(seq, arg, 0);
3570 		return;
3571 	}
3572 
3573 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
3574 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3575 				 SPDK_MALLOC_DMA);
3576 	if (!ctx->mask) {
3577 		bs_load_ctx_fail(ctx, -ENOMEM);
3578 		return;
3579 	}
3580 
3581 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
3582 	ctx->mask->length = ctx->super->md_len;
3583 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
3584 
3585 	spdk_bit_array_store_mask(ctx->bs->used_blobids, ctx->mask->mask);
3586 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
3587 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
3588 	bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3589 }
3590 
3591 static void
3592 blob_set_thin_provision(struct spdk_blob *blob)
3593 {
3594 	blob_verify_md_op(blob);
3595 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
3596 	blob->state = SPDK_BLOB_STATE_DIRTY;
3597 }
3598 
3599 static void
3600 blob_set_clear_method(struct spdk_blob *blob, enum blob_clear_method clear_method)
3601 {
3602 	blob_verify_md_op(blob);
3603 	blob->clear_method = clear_method;
3604 	blob->md_ro_flags |= (clear_method << SPDK_BLOB_CLEAR_METHOD_SHIFT);
3605 	blob->state = SPDK_BLOB_STATE_DIRTY;
3606 }
3607 
3608 static void bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno);
3609 
3610 static void
3611 bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno)
3612 {
3613 	struct spdk_bs_load_ctx *ctx = cb_arg;
3614 	spdk_blob_id id;
3615 	int64_t page_num;
3616 
3617 	/* Iterate to next blob (we can't use spdk_bs_iter_next function as our
3618 	 * last blob has been removed */
3619 	page_num = bs_blobid_to_page(ctx->blobid);
3620 	page_num++;
3621 	page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num);
3622 	if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) {
3623 		bs_load_iter(ctx, NULL, -ENOENT);
3624 		return;
3625 	}
3626 
3627 	id = bs_page_to_blobid(page_num);
3628 
3629 	spdk_bs_open_blob(ctx->bs, id, bs_load_iter, ctx);
3630 }
3631 
3632 static void
3633 bs_delete_corrupted_close_cb(void *cb_arg, int bserrno)
3634 {
3635 	struct spdk_bs_load_ctx *ctx = cb_arg;
3636 
3637 	if (bserrno != 0) {
3638 		SPDK_ERRLOG("Failed to close corrupted blob\n");
3639 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3640 		return;
3641 	}
3642 
3643 	spdk_bs_delete_blob(ctx->bs, ctx->blobid, bs_delete_corrupted_blob_cpl, ctx);
3644 }
3645 
3646 static void
3647 bs_delete_corrupted_blob(void *cb_arg, int bserrno)
3648 {
3649 	struct spdk_bs_load_ctx *ctx = cb_arg;
3650 	uint64_t i;
3651 
3652 	if (bserrno != 0) {
3653 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
3654 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3655 		return;
3656 	}
3657 
3658 	/* Snapshot and clone have the same copy of cluster map and extent pages
3659 	 * at this point. Let's clear both for snapshot now,
3660 	 * so that it won't be cleared for clone later when we remove snapshot.
3661 	 * Also set thin provision to pass data corruption check */
3662 	for (i = 0; i < ctx->blob->active.num_clusters; i++) {
3663 		ctx->blob->active.clusters[i] = 0;
3664 	}
3665 	for (i = 0; i < ctx->blob->active.num_extent_pages; i++) {
3666 		ctx->blob->active.extent_pages[i] = 0;
3667 	}
3668 
3669 	ctx->blob->md_ro = false;
3670 
3671 	blob_set_thin_provision(ctx->blob);
3672 
3673 	ctx->blobid = ctx->blob->id;
3674 
3675 	spdk_blob_close(ctx->blob, bs_delete_corrupted_close_cb, ctx);
3676 }
3677 
3678 static void
3679 bs_update_corrupted_blob(void *cb_arg, int bserrno)
3680 {
3681 	struct spdk_bs_load_ctx *ctx = cb_arg;
3682 
3683 	if (bserrno != 0) {
3684 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
3685 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3686 		return;
3687 	}
3688 
3689 	ctx->blob->md_ro = false;
3690 	blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true);
3691 	blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true);
3692 	spdk_blob_set_read_only(ctx->blob);
3693 
3694 	if (ctx->iter_cb_fn) {
3695 		ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0);
3696 	}
3697 	bs_blob_list_add(ctx->blob);
3698 
3699 	spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3700 }
3701 
3702 static void
3703 bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno)
3704 {
3705 	struct spdk_bs_load_ctx *ctx = cb_arg;
3706 
3707 	if (bserrno != 0) {
3708 		SPDK_ERRLOG("Failed to open clone of a corrupted blob\n");
3709 		spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx);
3710 		return;
3711 	}
3712 
3713 	if (blob->parent_id == ctx->blob->id) {
3714 		/* Power failure occurred before updating clone (snapshot delete case)
3715 		 * or after updating clone (creating snapshot case) - keep snapshot */
3716 		spdk_blob_close(blob, bs_update_corrupted_blob, ctx);
3717 	} else {
3718 		/* Power failure occurred after updating clone (snapshot delete case)
3719 		 * or before updating clone (creating snapshot case) - remove snapshot */
3720 		spdk_blob_close(blob, bs_delete_corrupted_blob, ctx);
3721 	}
3722 }
3723 
3724 static void
3725 bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
3726 {
3727 	struct spdk_bs_load_ctx *ctx = arg;
3728 	const void *value;
3729 	size_t len;
3730 	int rc = 0;
3731 
3732 	if (bserrno == 0) {
3733 		/* Examine blob if it is corrupted after power failure. Fix
3734 		 * the ones that can be fixed and remove any other corrupted
3735 		 * ones. If it is not corrupted just process it */
3736 		rc = blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true);
3737 		if (rc != 0) {
3738 			rc = blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true);
3739 			if (rc != 0) {
3740 				/* Not corrupted - process it and continue with iterating through blobs */
3741 				if (ctx->iter_cb_fn) {
3742 					ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
3743 				}
3744 				bs_blob_list_add(blob);
3745 				spdk_bs_iter_next(ctx->bs, blob, bs_load_iter, ctx);
3746 				return;
3747 			}
3748 
3749 		}
3750 
3751 		assert(len == sizeof(spdk_blob_id));
3752 
3753 		ctx->blob = blob;
3754 
3755 		/* Open clone to check if we are able to fix this blob or should we remove it */
3756 		spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, bs_examine_clone, ctx);
3757 		return;
3758 	} else if (bserrno == -ENOENT) {
3759 		bserrno = 0;
3760 	} else {
3761 		/*
3762 		 * This case needs to be looked at further.  Same problem
3763 		 *  exists with applications that rely on explicit blob
3764 		 *  iteration.  We should just skip the blob that failed
3765 		 *  to load and continue on to the next one.
3766 		 */
3767 		SPDK_ERRLOG("Error in iterating blobs\n");
3768 	}
3769 
3770 	ctx->iter_cb_fn = NULL;
3771 
3772 	spdk_free(ctx->super);
3773 	spdk_free(ctx->mask);
3774 	bs_sequence_finish(ctx->seq, bserrno);
3775 	free(ctx);
3776 }
3777 
3778 static void bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3779 
3780 static void
3781 bs_load_complete(struct spdk_bs_load_ctx *ctx)
3782 {
3783 	ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters);
3784 	if (ctx->dumping) {
3785 		bs_dump_read_md_page(ctx->seq, ctx);
3786 		return;
3787 	}
3788 	spdk_bs_iter_first(ctx->bs, bs_load_iter, ctx);
3789 }
3790 
3791 static void
3792 bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3793 {
3794 	struct spdk_bs_load_ctx *ctx = cb_arg;
3795 	int rc;
3796 
3797 	/* The type must be correct */
3798 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
3799 
3800 	/* The length of the mask (in bits) must not be greater than
3801 	 * the length of the buffer (converted to bits) */
3802 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
3803 
3804 	/* The length of the mask must be exactly equal to the size
3805 	 * (in pages) of the metadata region */
3806 	assert(ctx->mask->length == ctx->super->md_len);
3807 
3808 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->mask->length);
3809 	if (rc < 0) {
3810 		spdk_free(ctx->mask);
3811 		bs_load_ctx_fail(ctx, rc);
3812 		return;
3813 	}
3814 
3815 	spdk_bit_array_load_mask(ctx->bs->used_blobids, ctx->mask->mask);
3816 	bs_load_complete(ctx);
3817 }
3818 
3819 static void
3820 bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3821 {
3822 	struct spdk_bs_load_ctx *ctx = cb_arg;
3823 	uint64_t		lba, lba_count, mask_size;
3824 	int			rc;
3825 
3826 	if (bserrno != 0) {
3827 		bs_load_ctx_fail(ctx, bserrno);
3828 		return;
3829 	}
3830 
3831 	/* The type must be correct */
3832 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
3833 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
3834 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
3835 					     struct spdk_blob_md_page) * 8));
3836 	/* The length of the mask must be exactly equal to the total number of clusters */
3837 	assert(ctx->mask->length == ctx->bs->total_clusters);
3838 
3839 	rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->mask->length);
3840 	if (rc < 0) {
3841 		spdk_free(ctx->mask);
3842 		bs_load_ctx_fail(ctx, rc);
3843 		return;
3844 	}
3845 
3846 	spdk_bit_array_load_mask(ctx->used_clusters, ctx->mask->mask);
3847 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->used_clusters);
3848 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
3849 
3850 	spdk_free(ctx->mask);
3851 
3852 	/* Read the used blobids mask */
3853 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
3854 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3855 				 SPDK_MALLOC_DMA);
3856 	if (!ctx->mask) {
3857 		bs_load_ctx_fail(ctx, -ENOMEM);
3858 		return;
3859 	}
3860 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
3861 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
3862 	bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3863 			     bs_load_used_blobids_cpl, ctx);
3864 }
3865 
3866 static void
3867 bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3868 {
3869 	struct spdk_bs_load_ctx *ctx = cb_arg;
3870 	uint64_t		lba, lba_count, mask_size;
3871 	int			rc;
3872 
3873 	if (bserrno != 0) {
3874 		bs_load_ctx_fail(ctx, bserrno);
3875 		return;
3876 	}
3877 
3878 	/* The type must be correct */
3879 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
3880 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
3881 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
3882 				     8));
3883 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
3884 	if (ctx->mask->length != ctx->super->md_len) {
3885 		SPDK_ERRLOG("mismatched md_len in used_pages mask: "
3886 			    "mask->length=%" PRIu32 " super->md_len=%" PRIu32 "\n",
3887 			    ctx->mask->length, ctx->super->md_len);
3888 		assert(false);
3889 	}
3890 
3891 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->mask->length);
3892 	if (rc < 0) {
3893 		spdk_free(ctx->mask);
3894 		bs_load_ctx_fail(ctx, rc);
3895 		return;
3896 	}
3897 
3898 	spdk_bit_array_load_mask(ctx->bs->used_md_pages, ctx->mask->mask);
3899 	spdk_free(ctx->mask);
3900 
3901 	/* Read the used clusters mask */
3902 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
3903 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3904 				 SPDK_MALLOC_DMA);
3905 	if (!ctx->mask) {
3906 		bs_load_ctx_fail(ctx, -ENOMEM);
3907 		return;
3908 	}
3909 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
3910 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
3911 	bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3912 			     bs_load_used_clusters_cpl, ctx);
3913 }
3914 
3915 static void
3916 bs_load_read_used_pages(struct spdk_bs_load_ctx *ctx)
3917 {
3918 	uint64_t lba, lba_count, mask_size;
3919 
3920 	/* Read the used pages mask */
3921 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
3922 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3923 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3924 	if (!ctx->mask) {
3925 		bs_load_ctx_fail(ctx, -ENOMEM);
3926 		return;
3927 	}
3928 
3929 	lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3930 	lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3931 	bs_sequence_read_dev(ctx->seq, ctx->mask, lba, lba_count,
3932 			     bs_load_used_pages_cpl, ctx);
3933 }
3934 
3935 static int
3936 bs_load_replay_md_parse_page(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_page *page)
3937 {
3938 	struct spdk_blob_store *bs = ctx->bs;
3939 	struct spdk_blob_md_descriptor *desc;
3940 	size_t	cur_desc = 0;
3941 
3942 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3943 	while (cur_desc < sizeof(page->descriptors)) {
3944 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3945 			if (desc->length == 0) {
3946 				/* If padding and length are 0, this terminates the page */
3947 				break;
3948 			}
3949 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
3950 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
3951 			unsigned int				i, j;
3952 			unsigned int				cluster_count = 0;
3953 			uint32_t				cluster_idx;
3954 
3955 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
3956 
3957 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
3958 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
3959 					cluster_idx = desc_extent_rle->extents[i].cluster_idx;
3960 					/*
3961 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
3962 					 * in the used cluster map.
3963 					 */
3964 					if (cluster_idx != 0) {
3965 						SPDK_NOTICELOG("Recover: cluster %" PRIu32 "\n", cluster_idx + j);
3966 						spdk_bit_array_set(ctx->used_clusters, cluster_idx + j);
3967 						if (bs->num_free_clusters == 0) {
3968 							return -ENOSPC;
3969 						}
3970 						bs->num_free_clusters--;
3971 					}
3972 					cluster_count++;
3973 				}
3974 			}
3975 			if (cluster_count == 0) {
3976 				return -EINVAL;
3977 			}
3978 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
3979 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
3980 			uint32_t					i;
3981 			uint32_t					cluster_count = 0;
3982 			uint32_t					cluster_idx;
3983 			size_t						cluster_idx_length;
3984 
3985 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
3986 			cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx);
3987 
3988 			if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) ||
3989 			    (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) {
3990 				return -EINVAL;
3991 			}
3992 
3993 			for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
3994 				cluster_idx = desc_extent->cluster_idx[i];
3995 				/*
3996 				 * cluster_idx = 0 means an unallocated cluster - don't mark that
3997 				 * in the used cluster map.
3998 				 */
3999 				if (cluster_idx != 0) {
4000 					if (cluster_idx < desc_extent->start_cluster_idx &&
4001 					    cluster_idx >= desc_extent->start_cluster_idx + cluster_count) {
4002 						return -EINVAL;
4003 					}
4004 					spdk_bit_array_set(ctx->used_clusters, cluster_idx);
4005 					if (bs->num_free_clusters == 0) {
4006 						return -ENOSPC;
4007 					}
4008 					bs->num_free_clusters--;
4009 				}
4010 				cluster_count++;
4011 			}
4012 
4013 			if (cluster_count == 0) {
4014 				return -EINVAL;
4015 			}
4016 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
4017 			/* Skip this item */
4018 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
4019 			/* Skip this item */
4020 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
4021 			/* Skip this item */
4022 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
4023 			struct spdk_blob_md_descriptor_extent_table *desc_extent_table;
4024 			uint32_t num_extent_pages = ctx->num_extent_pages;
4025 			uint32_t i;
4026 			size_t extent_pages_length;
4027 			void *tmp;
4028 
4029 			desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc;
4030 			extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters);
4031 
4032 			if (desc_extent_table->length == 0 ||
4033 			    (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) {
4034 				return -EINVAL;
4035 			}
4036 
4037 			for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
4038 				if (desc_extent_table->extent_page[i].page_idx != 0) {
4039 					if (desc_extent_table->extent_page[i].num_pages != 1) {
4040 						return -EINVAL;
4041 					}
4042 					num_extent_pages += 1;
4043 				}
4044 			}
4045 
4046 			if (num_extent_pages > 0) {
4047 				tmp = realloc(ctx->extent_page_num, num_extent_pages * sizeof(uint32_t));
4048 				if (tmp == NULL) {
4049 					return -ENOMEM;
4050 				}
4051 				ctx->extent_page_num = tmp;
4052 
4053 				/* Extent table entries contain md page numbers for extent pages.
4054 				 * Zeroes represent unallocated extent pages, those are run-length-encoded.
4055 				 */
4056 				for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
4057 					if (desc_extent_table->extent_page[i].page_idx != 0) {
4058 						ctx->extent_page_num[ctx->num_extent_pages] = desc_extent_table->extent_page[i].page_idx;
4059 						ctx->num_extent_pages += 1;
4060 					}
4061 				}
4062 			}
4063 		} else {
4064 			/* Error */
4065 			return -EINVAL;
4066 		}
4067 		/* Advance to the next descriptor */
4068 		cur_desc += sizeof(*desc) + desc->length;
4069 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
4070 			break;
4071 		}
4072 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
4073 	}
4074 	return 0;
4075 }
4076 
4077 static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page)
4078 {
4079 	uint32_t crc;
4080 	struct spdk_blob_md_descriptor *desc = (struct spdk_blob_md_descriptor *)page->descriptors;
4081 	size_t desc_len;
4082 
4083 	crc = blob_md_page_calc_crc(page);
4084 	if (crc != page->crc) {
4085 		return false;
4086 	}
4087 
4088 	/* Extent page should always be of sequence num 0. */
4089 	if (page->sequence_num != 0) {
4090 		return false;
4091 	}
4092 
4093 	/* Descriptor type must be EXTENT_PAGE. */
4094 	if (desc->type != SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
4095 		return false;
4096 	}
4097 
4098 	/* Descriptor length cannot exceed the page. */
4099 	desc_len = sizeof(*desc) + desc->length;
4100 	if (desc_len > sizeof(page->descriptors)) {
4101 		return false;
4102 	}
4103 
4104 	/* It has to be the only descriptor in the page. */
4105 	if (desc_len + sizeof(*desc) <= sizeof(page->descriptors)) {
4106 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + desc_len);
4107 		if (desc->length != 0) {
4108 			return false;
4109 		}
4110 	}
4111 
4112 	return true;
4113 }
4114 
4115 static bool bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
4116 {
4117 	uint32_t crc;
4118 	struct spdk_blob_md_page *page = ctx->page;
4119 
4120 	crc = blob_md_page_calc_crc(page);
4121 	if (crc != page->crc) {
4122 		return false;
4123 	}
4124 
4125 	/* First page of a sequence should match the blobid. */
4126 	if (page->sequence_num == 0 &&
4127 	    bs_page_to_blobid(ctx->cur_page) != page->id) {
4128 		return false;
4129 	}
4130 	assert(bs_load_cur_extent_page_valid(page) == false);
4131 
4132 	return true;
4133 }
4134 
4135 static void
4136 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx);
4137 
4138 static void
4139 bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4140 {
4141 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4142 
4143 	if (bserrno != 0) {
4144 		bs_load_ctx_fail(ctx, bserrno);
4145 		return;
4146 	}
4147 
4148 	bs_load_complete(ctx);
4149 }
4150 
4151 static void
4152 bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4153 {
4154 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4155 
4156 	spdk_free(ctx->mask);
4157 	ctx->mask = NULL;
4158 
4159 	if (bserrno != 0) {
4160 		bs_load_ctx_fail(ctx, bserrno);
4161 		return;
4162 	}
4163 
4164 	bs_write_used_clusters(seq, ctx, bs_load_write_used_clusters_cpl);
4165 }
4166 
4167 static void
4168 bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4169 {
4170 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4171 
4172 	spdk_free(ctx->mask);
4173 	ctx->mask = NULL;
4174 
4175 	if (bserrno != 0) {
4176 		bs_load_ctx_fail(ctx, bserrno);
4177 		return;
4178 	}
4179 
4180 	bs_write_used_blobids(seq, ctx, bs_load_write_used_blobids_cpl);
4181 }
4182 
4183 static void
4184 bs_load_write_used_md(struct spdk_bs_load_ctx *ctx)
4185 {
4186 	bs_write_used_md(ctx->seq, ctx, bs_load_write_used_pages_cpl);
4187 }
4188 
4189 static void
4190 bs_load_replay_md_chain_cpl(struct spdk_bs_load_ctx *ctx)
4191 {
4192 	uint64_t num_md_clusters;
4193 	uint64_t i;
4194 
4195 	ctx->in_page_chain = false;
4196 
4197 	do {
4198 		ctx->page_index++;
4199 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
4200 
4201 	if (ctx->page_index < ctx->super->md_len) {
4202 		ctx->cur_page = ctx->page_index;
4203 		bs_load_replay_cur_md_page(ctx);
4204 	} else {
4205 		/* Claim all of the clusters used by the metadata */
4206 		num_md_clusters = spdk_divide_round_up(
4207 					  ctx->super->md_start + ctx->super->md_len, ctx->bs->pages_per_cluster);
4208 		for (i = 0; i < num_md_clusters; i++) {
4209 			spdk_bit_array_set(ctx->used_clusters, i);
4210 		}
4211 		ctx->bs->num_free_clusters -= num_md_clusters;
4212 		spdk_free(ctx->page);
4213 		bs_load_write_used_md(ctx);
4214 	}
4215 }
4216 
4217 static void
4218 bs_load_replay_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4219 {
4220 	struct spdk_bs_load_ctx *ctx = cb_arg;
4221 	uint32_t page_num;
4222 	uint64_t i;
4223 
4224 	if (bserrno != 0) {
4225 		spdk_free(ctx->extent_pages);
4226 		bs_load_ctx_fail(ctx, bserrno);
4227 		return;
4228 	}
4229 
4230 	for (i = 0; i < ctx->num_extent_pages; i++) {
4231 		/* Extent pages are only read when present within in chain md.
4232 		 * Integrity of md is not right if that page was not a valid extent page. */
4233 		if (bs_load_cur_extent_page_valid(&ctx->extent_pages[i]) != true) {
4234 			spdk_free(ctx->extent_pages);
4235 			bs_load_ctx_fail(ctx, -EILSEQ);
4236 			return;
4237 		}
4238 
4239 		page_num = ctx->extent_page_num[i];
4240 		spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
4241 		if (bs_load_replay_md_parse_page(ctx, &ctx->extent_pages[i])) {
4242 			spdk_free(ctx->extent_pages);
4243 			bs_load_ctx_fail(ctx, -EILSEQ);
4244 			return;
4245 		}
4246 	}
4247 
4248 	spdk_free(ctx->extent_pages);
4249 	free(ctx->extent_page_num);
4250 	ctx->extent_page_num = NULL;
4251 	ctx->num_extent_pages = 0;
4252 
4253 	bs_load_replay_md_chain_cpl(ctx);
4254 }
4255 
4256 static void
4257 bs_load_replay_extent_pages(struct spdk_bs_load_ctx *ctx)
4258 {
4259 	spdk_bs_batch_t *batch;
4260 	uint32_t page;
4261 	uint64_t lba;
4262 	uint64_t i;
4263 
4264 	ctx->extent_pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE * ctx->num_extent_pages, 0,
4265 					 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4266 	if (!ctx->extent_pages) {
4267 		bs_load_ctx_fail(ctx, -ENOMEM);
4268 		return;
4269 	}
4270 
4271 	batch = bs_sequence_to_batch(ctx->seq, bs_load_replay_extent_page_cpl, ctx);
4272 
4273 	for (i = 0; i < ctx->num_extent_pages; i++) {
4274 		page = ctx->extent_page_num[i];
4275 		assert(page < ctx->super->md_len);
4276 		lba = bs_md_page_to_lba(ctx->bs, page);
4277 		bs_batch_read_dev(batch, &ctx->extent_pages[i], lba,
4278 				  bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE));
4279 	}
4280 
4281 	bs_batch_close(batch);
4282 }
4283 
4284 static void
4285 bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4286 {
4287 	struct spdk_bs_load_ctx *ctx = cb_arg;
4288 	uint32_t page_num;
4289 	struct spdk_blob_md_page *page;
4290 
4291 	if (bserrno != 0) {
4292 		bs_load_ctx_fail(ctx, bserrno);
4293 		return;
4294 	}
4295 
4296 	page_num = ctx->cur_page;
4297 	page = ctx->page;
4298 	if (bs_load_cur_md_page_valid(ctx) == true) {
4299 		if (page->sequence_num == 0 || ctx->in_page_chain == true) {
4300 			bs_claim_md_page(ctx->bs, page_num);
4301 			if (page->sequence_num == 0) {
4302 				SPDK_NOTICELOG("Recover: blob %" PRIu32 "\n", page_num);
4303 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
4304 			}
4305 			if (bs_load_replay_md_parse_page(ctx, page)) {
4306 				bs_load_ctx_fail(ctx, -EILSEQ);
4307 				return;
4308 			}
4309 			if (page->next != SPDK_INVALID_MD_PAGE) {
4310 				ctx->in_page_chain = true;
4311 				ctx->cur_page = page->next;
4312 				bs_load_replay_cur_md_page(ctx);
4313 				return;
4314 			}
4315 			if (ctx->num_extent_pages != 0) {
4316 				bs_load_replay_extent_pages(ctx);
4317 				return;
4318 			}
4319 		}
4320 	}
4321 	bs_load_replay_md_chain_cpl(ctx);
4322 }
4323 
4324 static void
4325 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx)
4326 {
4327 	uint64_t lba;
4328 
4329 	assert(ctx->cur_page < ctx->super->md_len);
4330 	lba = bs_md_page_to_lba(ctx->bs, ctx->cur_page);
4331 	bs_sequence_read_dev(ctx->seq, ctx->page, lba,
4332 			     bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
4333 			     bs_load_replay_md_cpl, ctx);
4334 }
4335 
4336 static void
4337 bs_load_replay_md(struct spdk_bs_load_ctx *ctx)
4338 {
4339 	ctx->page_index = 0;
4340 	ctx->cur_page = 0;
4341 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0,
4342 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4343 	if (!ctx->page) {
4344 		bs_load_ctx_fail(ctx, -ENOMEM);
4345 		return;
4346 	}
4347 	bs_load_replay_cur_md_page(ctx);
4348 }
4349 
4350 static void
4351 bs_recover(struct spdk_bs_load_ctx *ctx)
4352 {
4353 	int		rc;
4354 
4355 	SPDK_NOTICELOG("Performing recovery on blobstore\n");
4356 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
4357 	if (rc < 0) {
4358 		bs_load_ctx_fail(ctx, -ENOMEM);
4359 		return;
4360 	}
4361 
4362 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
4363 	if (rc < 0) {
4364 		bs_load_ctx_fail(ctx, -ENOMEM);
4365 		return;
4366 	}
4367 
4368 	rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters);
4369 	if (rc < 0) {
4370 		bs_load_ctx_fail(ctx, -ENOMEM);
4371 		return;
4372 	}
4373 
4374 	rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->super->md_len);
4375 	if (rc < 0) {
4376 		bs_load_ctx_fail(ctx, -ENOMEM);
4377 		return;
4378 	}
4379 
4380 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
4381 	bs_load_replay_md(ctx);
4382 }
4383 
4384 static int
4385 bs_parse_super(struct spdk_bs_load_ctx *ctx)
4386 {
4387 	int rc;
4388 
4389 	if (ctx->super->size == 0) {
4390 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
4391 	}
4392 
4393 	if (ctx->super->io_unit_size == 0) {
4394 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
4395 	}
4396 
4397 	ctx->bs->clean = 1;
4398 	ctx->bs->cluster_sz = ctx->super->cluster_size;
4399 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
4400 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
4401 	if (spdk_u32_is_pow2(ctx->bs->pages_per_cluster)) {
4402 		ctx->bs->pages_per_cluster_shift = spdk_u32log2(ctx->bs->pages_per_cluster);
4403 	}
4404 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
4405 	rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters);
4406 	if (rc < 0) {
4407 		return -ENOMEM;
4408 	}
4409 	ctx->bs->md_start = ctx->super->md_start;
4410 	ctx->bs->md_len = ctx->super->md_len;
4411 	rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->bs->md_len);
4412 	if (rc < 0) {
4413 		return -ENOMEM;
4414 	}
4415 
4416 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
4417 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
4418 	ctx->bs->super_blob = ctx->super->super_blob;
4419 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
4420 
4421 	return 0;
4422 }
4423 
4424 static void
4425 bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4426 {
4427 	struct spdk_bs_load_ctx *ctx = cb_arg;
4428 	uint32_t	crc;
4429 	int		rc;
4430 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
4431 
4432 	if (ctx->super->version > SPDK_BS_VERSION ||
4433 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
4434 		bs_load_ctx_fail(ctx, -EILSEQ);
4435 		return;
4436 	}
4437 
4438 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4439 		   sizeof(ctx->super->signature)) != 0) {
4440 		bs_load_ctx_fail(ctx, -EILSEQ);
4441 		return;
4442 	}
4443 
4444 	crc = blob_md_page_calc_crc(ctx->super);
4445 	if (crc != ctx->super->crc) {
4446 		bs_load_ctx_fail(ctx, -EILSEQ);
4447 		return;
4448 	}
4449 
4450 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
4451 		SPDK_DEBUGLOG(blob, "Bstype matched - loading blobstore\n");
4452 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
4453 		SPDK_DEBUGLOG(blob, "Bstype wildcard used - loading blobstore regardless bstype\n");
4454 	} else {
4455 		SPDK_DEBUGLOG(blob, "Unexpected bstype\n");
4456 		SPDK_LOGDUMP(blob, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
4457 		SPDK_LOGDUMP(blob, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
4458 		bs_load_ctx_fail(ctx, -ENXIO);
4459 		return;
4460 	}
4461 
4462 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
4463 		SPDK_NOTICELOG("Size mismatch, dev size: %" PRIu64 ", blobstore size: %" PRIu64 "\n",
4464 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
4465 		bs_load_ctx_fail(ctx, -EILSEQ);
4466 		return;
4467 	}
4468 
4469 	rc = bs_parse_super(ctx);
4470 	if (rc < 0) {
4471 		bs_load_ctx_fail(ctx, rc);
4472 		return;
4473 	}
4474 
4475 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0 || ctx->force_recover) {
4476 		bs_recover(ctx);
4477 	} else {
4478 		bs_load_read_used_pages(ctx);
4479 	}
4480 }
4481 
4482 static inline int
4483 bs_opts_copy(struct spdk_bs_opts *src, struct spdk_bs_opts *dst)
4484 {
4485 
4486 	if (!src->opts_size) {
4487 		SPDK_ERRLOG("opts_size should not be zero value\n");
4488 		return -1;
4489 	}
4490 
4491 #define FIELD_OK(field) \
4492         offsetof(struct spdk_bs_opts, field) + sizeof(src->field) <= src->opts_size
4493 
4494 #define SET_FIELD(field) \
4495         if (FIELD_OK(field)) { \
4496                 dst->field = src->field; \
4497         } \
4498 
4499 	SET_FIELD(cluster_sz);
4500 	SET_FIELD(num_md_pages);
4501 	SET_FIELD(max_md_ops);
4502 	SET_FIELD(max_channel_ops);
4503 	SET_FIELD(clear_method);
4504 
4505 	if (FIELD_OK(bstype)) {
4506 		memcpy(&dst->bstype, &src->bstype, sizeof(dst->bstype));
4507 	}
4508 	SET_FIELD(iter_cb_fn);
4509 	SET_FIELD(iter_cb_arg);
4510 	SET_FIELD(force_recover);
4511 
4512 	dst->opts_size = src->opts_size;
4513 
4514 	/* You should not remove this statement, but need to update the assert statement
4515 	 * if you add a new field, and also add a corresponding SET_FIELD statement */
4516 	SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_opts) == 72, "Incorrect size");
4517 
4518 #undef FIELD_OK
4519 #undef SET_FIELD
4520 
4521 	return 0;
4522 }
4523 
4524 void
4525 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
4526 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
4527 {
4528 	struct spdk_blob_store	*bs;
4529 	struct spdk_bs_cpl	cpl;
4530 	struct spdk_bs_load_ctx *ctx;
4531 	struct spdk_bs_opts	opts = {};
4532 	int err;
4533 
4534 	SPDK_DEBUGLOG(blob, "Loading blobstore from dev %p\n", dev);
4535 
4536 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
4537 		SPDK_DEBUGLOG(blob, "unsupported dev block length of %d\n", dev->blocklen);
4538 		dev->destroy(dev);
4539 		cb_fn(cb_arg, NULL, -EINVAL);
4540 		return;
4541 	}
4542 
4543 	spdk_bs_opts_init(&opts, sizeof(opts));
4544 	if (o) {
4545 		if (bs_opts_copy(o, &opts)) {
4546 			return;
4547 		}
4548 	}
4549 
4550 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
4551 		dev->destroy(dev);
4552 		cb_fn(cb_arg, NULL, -EINVAL);
4553 		return;
4554 	}
4555 
4556 	err = bs_alloc(dev, &opts, &bs, &ctx);
4557 	if (err) {
4558 		dev->destroy(dev);
4559 		cb_fn(cb_arg, NULL, err);
4560 		return;
4561 	}
4562 
4563 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
4564 	cpl.u.bs_handle.cb_fn = cb_fn;
4565 	cpl.u.bs_handle.cb_arg = cb_arg;
4566 	cpl.u.bs_handle.bs = bs;
4567 
4568 	ctx->seq = bs_sequence_start(bs->md_channel, &cpl);
4569 	if (!ctx->seq) {
4570 		spdk_free(ctx->super);
4571 		free(ctx);
4572 		bs_free(bs);
4573 		cb_fn(cb_arg, NULL, -ENOMEM);
4574 		return;
4575 	}
4576 
4577 	/* Read the super block */
4578 	bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
4579 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
4580 			     bs_load_super_cpl, ctx);
4581 }
4582 
4583 /* END spdk_bs_load */
4584 
4585 /* START spdk_bs_dump */
4586 
4587 static void
4588 bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
4589 {
4590 	spdk_free(ctx->super);
4591 
4592 	/*
4593 	 * We need to defer calling bs_call_cpl() until after
4594 	 * dev destruction, so tuck these away for later use.
4595 	 */
4596 	ctx->bs->unload_err = bserrno;
4597 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
4598 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
4599 
4600 	bs_sequence_finish(seq, 0);
4601 	bs_free(ctx->bs);
4602 	free(ctx);
4603 }
4604 
4605 static void
4606 bs_dump_print_xattr(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_descriptor *desc)
4607 {
4608 	struct spdk_blob_md_descriptor_xattr *desc_xattr;
4609 	uint32_t i;
4610 	const char *type;
4611 
4612 	desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
4613 
4614 	if (desc_xattr->length !=
4615 	    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
4616 	    desc_xattr->name_length + desc_xattr->value_length) {
4617 	}
4618 
4619 	memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
4620 	ctx->xattr_name[desc_xattr->name_length] = '\0';
4621 	if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
4622 		type = "XATTR";
4623 	} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
4624 		type = "XATTR_INTERNAL";
4625 	} else {
4626 		assert(false);
4627 		type = "XATTR_?";
4628 	}
4629 	fprintf(ctx->fp, "%s: name = \"%s\"\n", type, ctx->xattr_name);
4630 	fprintf(ctx->fp, "       value = \"");
4631 	ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
4632 			    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
4633 			    desc_xattr->value_length);
4634 	fprintf(ctx->fp, "\"\n");
4635 	for (i = 0; i < desc_xattr->value_length; i++) {
4636 		if (i % 16 == 0) {
4637 			fprintf(ctx->fp, "               ");
4638 		}
4639 		fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
4640 		if ((i + 1) % 16 == 0) {
4641 			fprintf(ctx->fp, "\n");
4642 		}
4643 	}
4644 	if (i % 16 != 0) {
4645 		fprintf(ctx->fp, "\n");
4646 	}
4647 }
4648 
4649 struct type_flag_desc {
4650 	uint64_t mask;
4651 	uint64_t val;
4652 	const char *name;
4653 };
4654 
4655 static void
4656 bs_dump_print_type_bits(struct spdk_bs_load_ctx *ctx, uint64_t flags,
4657 			struct type_flag_desc *desc, size_t numflags)
4658 {
4659 	uint64_t covered = 0;
4660 	size_t i;
4661 
4662 	for (i = 0; i < numflags; i++) {
4663 		if ((desc[i].mask & flags) != desc[i].val) {
4664 			continue;
4665 		}
4666 		fprintf(ctx->fp, "\t\t 0x%016" PRIx64 " %s", desc[i].val, desc[i].name);
4667 		if (desc[i].mask != desc[i].val) {
4668 			fprintf(ctx->fp, " (mask 0x%" PRIx64 " value 0x%" PRIx64 ")",
4669 				desc[i].mask, desc[i].val);
4670 		}
4671 		fprintf(ctx->fp, "\n");
4672 		covered |= desc[i].mask;
4673 	}
4674 	if ((flags & ~covered) != 0) {
4675 		fprintf(ctx->fp, "\t\t 0x%016" PRIx64 " Unknown\n", flags & ~covered);
4676 	}
4677 }
4678 
4679 static void
4680 bs_dump_print_type_flags(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_descriptor *desc)
4681 {
4682 	struct spdk_blob_md_descriptor_flags *type_desc;
4683 #define ADD_FLAG(f) { f, f, #f }
4684 #define ADD_MASK_VAL(m, v) { m, v, #v }
4685 	static struct type_flag_desc invalid[] = {
4686 		ADD_FLAG(SPDK_BLOB_THIN_PROV),
4687 		ADD_FLAG(SPDK_BLOB_INTERNAL_XATTR),
4688 		ADD_FLAG(SPDK_BLOB_EXTENT_TABLE),
4689 	};
4690 	static struct type_flag_desc data_ro[] = {
4691 		ADD_FLAG(SPDK_BLOB_READ_ONLY),
4692 	};
4693 	static struct type_flag_desc md_ro[] = {
4694 		ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_DEFAULT),
4695 		ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_NONE),
4696 		ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_UNMAP),
4697 		ADD_MASK_VAL(SPDK_BLOB_MD_RO_FLAGS_MASK, BLOB_CLEAR_WITH_WRITE_ZEROES),
4698 	};
4699 #undef ADD_FLAG
4700 #undef ADD_MASK_VAL
4701 
4702 	type_desc = (struct spdk_blob_md_descriptor_flags *)desc;
4703 	fprintf(ctx->fp, "Flags:\n");
4704 	fprintf(ctx->fp, "\tinvalid: 0x%016" PRIx64 "\n", type_desc->invalid_flags);
4705 	bs_dump_print_type_bits(ctx, type_desc->invalid_flags, invalid,
4706 				SPDK_COUNTOF(invalid));
4707 	fprintf(ctx->fp, "\tdata_ro: 0x%016" PRIx64 "\n", type_desc->data_ro_flags);
4708 	bs_dump_print_type_bits(ctx, type_desc->data_ro_flags, data_ro,
4709 				SPDK_COUNTOF(data_ro));
4710 	fprintf(ctx->fp, "\t  md_ro: 0x%016" PRIx64 "\n", type_desc->md_ro_flags);
4711 	bs_dump_print_type_bits(ctx, type_desc->md_ro_flags, md_ro,
4712 				SPDK_COUNTOF(md_ro));
4713 }
4714 
4715 static void
4716 bs_dump_print_extent_table(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_descriptor *desc)
4717 {
4718 	struct spdk_blob_md_descriptor_extent_table *et_desc;
4719 	uint64_t num_extent_pages;
4720 	uint32_t et_idx;
4721 
4722 	et_desc = (struct spdk_blob_md_descriptor_extent_table *)desc;
4723 	num_extent_pages = (et_desc->length - sizeof(et_desc->num_clusters)) /
4724 			   sizeof(et_desc->extent_page[0]);
4725 
4726 	fprintf(ctx->fp, "Extent table:\n");
4727 	for (et_idx = 0; et_idx < num_extent_pages; et_idx++) {
4728 		if (et_desc->extent_page[et_idx].page_idx == 0) {
4729 			/* Zeroes represent unallocated extent pages. */
4730 			continue;
4731 		}
4732 		fprintf(ctx->fp, "\tExtent page: %5" PRIu32 " length %3" PRIu32
4733 			" at LBA %" PRIu64 "\n", et_desc->extent_page[et_idx].page_idx,
4734 			et_desc->extent_page[et_idx].num_pages,
4735 			bs_md_page_to_lba(ctx->bs, et_desc->extent_page[et_idx].page_idx));
4736 	}
4737 }
4738 
4739 static void
4740 bs_dump_print_md_page(struct spdk_bs_load_ctx *ctx)
4741 {
4742 	uint32_t page_idx = ctx->cur_page;
4743 	struct spdk_blob_md_page *page = ctx->page;
4744 	struct spdk_blob_md_descriptor *desc;
4745 	size_t cur_desc = 0;
4746 	uint32_t crc;
4747 
4748 	fprintf(ctx->fp, "=========\n");
4749 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
4750 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
4751 	fprintf(ctx->fp, "Sequence: %" PRIu32 "\n", page->sequence_num);
4752 	if (page->next == SPDK_INVALID_MD_PAGE) {
4753 		fprintf(ctx->fp, "Next: None\n");
4754 	} else {
4755 		fprintf(ctx->fp, "Next: %" PRIu32 "\n", page->next);
4756 	}
4757 	fprintf(ctx->fp, "In used bit array%s:", ctx->super->clean ? "" : " (not clean: dubious)");
4758 	if (spdk_bit_array_get(ctx->bs->used_md_pages, page_idx)) {
4759 		fprintf(ctx->fp, " md");
4760 	}
4761 	if (spdk_bit_array_get(ctx->bs->used_blobids, page_idx)) {
4762 		fprintf(ctx->fp, " blob");
4763 	}
4764 	fprintf(ctx->fp, "\n");
4765 
4766 	crc = blob_md_page_calc_crc(page);
4767 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
4768 
4769 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
4770 	while (cur_desc < sizeof(page->descriptors)) {
4771 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
4772 			if (desc->length == 0) {
4773 				/* If padding and length are 0, this terminates the page */
4774 				break;
4775 			}
4776 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
4777 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
4778 			unsigned int				i;
4779 
4780 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
4781 
4782 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
4783 				if (desc_extent_rle->extents[i].cluster_idx != 0) {
4784 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
4785 						desc_extent_rle->extents[i].cluster_idx);
4786 				} else {
4787 					fprintf(ctx->fp, "Unallocated Extent - ");
4788 				}
4789 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent_rle->extents[i].length);
4790 				fprintf(ctx->fp, "\n");
4791 			}
4792 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
4793 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
4794 			unsigned int					i;
4795 
4796 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
4797 
4798 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->cluster_idx[0]); i++) {
4799 				if (desc_extent->cluster_idx[i] != 0) {
4800 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
4801 						desc_extent->cluster_idx[i]);
4802 				} else {
4803 					fprintf(ctx->fp, "Unallocated Extent");
4804 				}
4805 				fprintf(ctx->fp, "\n");
4806 			}
4807 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
4808 			bs_dump_print_xattr(ctx, desc);
4809 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
4810 			bs_dump_print_xattr(ctx, desc);
4811 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
4812 			bs_dump_print_type_flags(ctx, desc);
4813 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
4814 			bs_dump_print_extent_table(ctx, desc);
4815 		} else {
4816 			/* Error */
4817 			fprintf(ctx->fp, "Unknown descriptor type %" PRIu8 "\n", desc->type);
4818 		}
4819 		/* Advance to the next descriptor */
4820 		cur_desc += sizeof(*desc) + desc->length;
4821 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
4822 			break;
4823 		}
4824 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
4825 	}
4826 }
4827 
4828 static void
4829 bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4830 {
4831 	struct spdk_bs_load_ctx *ctx = cb_arg;
4832 
4833 	if (bserrno != 0) {
4834 		bs_dump_finish(seq, ctx, bserrno);
4835 		return;
4836 	}
4837 
4838 	if (ctx->page->id != 0) {
4839 		bs_dump_print_md_page(ctx);
4840 	}
4841 
4842 	ctx->cur_page++;
4843 
4844 	if (ctx->cur_page < ctx->super->md_len) {
4845 		bs_dump_read_md_page(seq, ctx);
4846 	} else {
4847 		spdk_free(ctx->page);
4848 		bs_dump_finish(seq, ctx, 0);
4849 	}
4850 }
4851 
4852 static void
4853 bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
4854 {
4855 	struct spdk_bs_load_ctx *ctx = cb_arg;
4856 	uint64_t lba;
4857 
4858 	assert(ctx->cur_page < ctx->super->md_len);
4859 	lba = bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
4860 	bs_sequence_read_dev(seq, ctx->page, lba,
4861 			     bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
4862 			     bs_dump_read_md_page_cpl, ctx);
4863 }
4864 
4865 static void
4866 bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4867 {
4868 	struct spdk_bs_load_ctx *ctx = cb_arg;
4869 	int rc;
4870 
4871 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
4872 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4873 		   sizeof(ctx->super->signature)) != 0) {
4874 		fprintf(ctx->fp, "(Mismatch)\n");
4875 		bs_dump_finish(seq, ctx, bserrno);
4876 		return;
4877 	} else {
4878 		fprintf(ctx->fp, "(OK)\n");
4879 	}
4880 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
4881 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
4882 		(ctx->super->crc == blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
4883 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
4884 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
4885 	fprintf(ctx->fp, "Super Blob ID: ");
4886 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
4887 		fprintf(ctx->fp, "(None)\n");
4888 	} else {
4889 		fprintf(ctx->fp, "0x%" PRIx64 "\n", ctx->super->super_blob);
4890 	}
4891 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
4892 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
4893 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
4894 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
4895 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
4896 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
4897 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
4898 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
4899 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
4900 
4901 	ctx->cur_page = 0;
4902 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0,
4903 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4904 	if (!ctx->page) {
4905 		bs_dump_finish(seq, ctx, -ENOMEM);
4906 		return;
4907 	}
4908 
4909 	rc = bs_parse_super(ctx);
4910 	if (rc < 0) {
4911 		bs_load_ctx_fail(ctx, rc);
4912 		return;
4913 	}
4914 
4915 	bs_load_read_used_pages(ctx);
4916 }
4917 
4918 void
4919 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
4920 	     spdk_bs_op_complete cb_fn, void *cb_arg)
4921 {
4922 	struct spdk_blob_store	*bs;
4923 	struct spdk_bs_cpl	cpl;
4924 	struct spdk_bs_load_ctx *ctx;
4925 	struct spdk_bs_opts	opts = {};
4926 	int err;
4927 
4928 	SPDK_DEBUGLOG(blob, "Dumping blobstore from dev %p\n", dev);
4929 
4930 	spdk_bs_opts_init(&opts, sizeof(opts));
4931 
4932 	err = bs_alloc(dev, &opts, &bs, &ctx);
4933 	if (err) {
4934 		dev->destroy(dev);
4935 		cb_fn(cb_arg, err);
4936 		return;
4937 	}
4938 
4939 	ctx->dumping = true;
4940 	ctx->fp = fp;
4941 	ctx->print_xattr_fn = print_xattr_fn;
4942 
4943 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4944 	cpl.u.bs_basic.cb_fn = cb_fn;
4945 	cpl.u.bs_basic.cb_arg = cb_arg;
4946 
4947 	ctx->seq = bs_sequence_start(bs->md_channel, &cpl);
4948 	if (!ctx->seq) {
4949 		spdk_free(ctx->super);
4950 		free(ctx);
4951 		bs_free(bs);
4952 		cb_fn(cb_arg, -ENOMEM);
4953 		return;
4954 	}
4955 
4956 	/* Read the super block */
4957 	bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
4958 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
4959 			     bs_dump_super_cpl, ctx);
4960 }
4961 
4962 /* END spdk_bs_dump */
4963 
4964 /* START spdk_bs_init */
4965 
4966 static void
4967 bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4968 {
4969 	struct spdk_bs_load_ctx *ctx = cb_arg;
4970 
4971 	ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters);
4972 	spdk_free(ctx->super);
4973 	free(ctx);
4974 
4975 	bs_sequence_finish(seq, bserrno);
4976 }
4977 
4978 static void
4979 bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4980 {
4981 	struct spdk_bs_load_ctx *ctx = cb_arg;
4982 
4983 	/* Write super block */
4984 	bs_sequence_write_dev(seq, ctx->super, bs_page_to_lba(ctx->bs, 0),
4985 			      bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
4986 			      bs_init_persist_super_cpl, ctx);
4987 }
4988 
4989 void
4990 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
4991 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
4992 {
4993 	struct spdk_bs_load_ctx *ctx;
4994 	struct spdk_blob_store	*bs;
4995 	struct spdk_bs_cpl	cpl;
4996 	spdk_bs_sequence_t	*seq;
4997 	spdk_bs_batch_t		*batch;
4998 	uint64_t		num_md_lba;
4999 	uint64_t		num_md_pages;
5000 	uint64_t		num_md_clusters;
5001 	uint32_t		i;
5002 	struct spdk_bs_opts	opts = {};
5003 	int			rc;
5004 	uint64_t		lba, lba_count;
5005 
5006 	SPDK_DEBUGLOG(blob, "Initializing blobstore on dev %p\n", dev);
5007 
5008 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
5009 		SPDK_ERRLOG("unsupported dev block length of %d\n",
5010 			    dev->blocklen);
5011 		dev->destroy(dev);
5012 		cb_fn(cb_arg, NULL, -EINVAL);
5013 		return;
5014 	}
5015 
5016 	spdk_bs_opts_init(&opts, sizeof(opts));
5017 	if (o) {
5018 		if (bs_opts_copy(o, &opts)) {
5019 			return;
5020 		}
5021 	}
5022 
5023 	if (bs_opts_verify(&opts) != 0) {
5024 		dev->destroy(dev);
5025 		cb_fn(cb_arg, NULL, -EINVAL);
5026 		return;
5027 	}
5028 
5029 	rc = bs_alloc(dev, &opts, &bs, &ctx);
5030 	if (rc) {
5031 		dev->destroy(dev);
5032 		cb_fn(cb_arg, NULL, rc);
5033 		return;
5034 	}
5035 
5036 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
5037 		/* By default, allocate 1 page per cluster.
5038 		 * Technically, this over-allocates metadata
5039 		 * because more metadata will reduce the number
5040 		 * of usable clusters. This can be addressed with
5041 		 * more complex math in the future.
5042 		 */
5043 		bs->md_len = bs->total_clusters;
5044 	} else {
5045 		bs->md_len = opts.num_md_pages;
5046 	}
5047 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
5048 	if (rc < 0) {
5049 		spdk_free(ctx->super);
5050 		free(ctx);
5051 		bs_free(bs);
5052 		cb_fn(cb_arg, NULL, -ENOMEM);
5053 		return;
5054 	}
5055 
5056 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
5057 	if (rc < 0) {
5058 		spdk_free(ctx->super);
5059 		free(ctx);
5060 		bs_free(bs);
5061 		cb_fn(cb_arg, NULL, -ENOMEM);
5062 		return;
5063 	}
5064 
5065 	rc = spdk_bit_array_resize(&bs->open_blobids, bs->md_len);
5066 	if (rc < 0) {
5067 		spdk_free(ctx->super);
5068 		free(ctx);
5069 		bs_free(bs);
5070 		cb_fn(cb_arg, NULL, -ENOMEM);
5071 		return;
5072 	}
5073 
5074 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
5075 	       sizeof(ctx->super->signature));
5076 	ctx->super->version = SPDK_BS_VERSION;
5077 	ctx->super->length = sizeof(*ctx->super);
5078 	ctx->super->super_blob = bs->super_blob;
5079 	ctx->super->clean = 0;
5080 	ctx->super->cluster_size = bs->cluster_sz;
5081 	ctx->super->io_unit_size = bs->io_unit_size;
5082 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
5083 
5084 	/* Calculate how many pages the metadata consumes at the front
5085 	 * of the disk.
5086 	 */
5087 
5088 	/* The super block uses 1 page */
5089 	num_md_pages = 1;
5090 
5091 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
5092 	 * up to the nearest page, plus a header.
5093 	 */
5094 	ctx->super->used_page_mask_start = num_md_pages;
5095 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
5096 					 spdk_divide_round_up(bs->md_len, 8),
5097 					 SPDK_BS_PAGE_SIZE);
5098 	num_md_pages += ctx->super->used_page_mask_len;
5099 
5100 	/* The used_clusters mask requires 1 bit per cluster, rounded
5101 	 * up to the nearest page, plus a header.
5102 	 */
5103 	ctx->super->used_cluster_mask_start = num_md_pages;
5104 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
5105 					    spdk_divide_round_up(bs->total_clusters, 8),
5106 					    SPDK_BS_PAGE_SIZE);
5107 	num_md_pages += ctx->super->used_cluster_mask_len;
5108 
5109 	/* The used_blobids mask requires 1 bit per metadata page, rounded
5110 	 * up to the nearest page, plus a header.
5111 	 */
5112 	ctx->super->used_blobid_mask_start = num_md_pages;
5113 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
5114 					   spdk_divide_round_up(bs->md_len, 8),
5115 					   SPDK_BS_PAGE_SIZE);
5116 	num_md_pages += ctx->super->used_blobid_mask_len;
5117 
5118 	/* The metadata region size was chosen above */
5119 	ctx->super->md_start = bs->md_start = num_md_pages;
5120 	ctx->super->md_len = bs->md_len;
5121 	num_md_pages += bs->md_len;
5122 
5123 	num_md_lba = bs_page_to_lba(bs, num_md_pages);
5124 
5125 	ctx->super->size = dev->blockcnt * dev->blocklen;
5126 
5127 	ctx->super->crc = blob_md_page_calc_crc(ctx->super);
5128 
5129 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
5130 	if (num_md_clusters > bs->total_clusters) {
5131 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
5132 			    "please decrease number of pages reserved for metadata "
5133 			    "or increase cluster size.\n");
5134 		spdk_free(ctx->super);
5135 		spdk_bit_array_free(&ctx->used_clusters);
5136 		free(ctx);
5137 		bs_free(bs);
5138 		cb_fn(cb_arg, NULL, -ENOMEM);
5139 		return;
5140 	}
5141 	/* Claim all of the clusters used by the metadata */
5142 	for (i = 0; i < num_md_clusters; i++) {
5143 		spdk_bit_array_set(ctx->used_clusters, i);
5144 	}
5145 
5146 	bs->num_free_clusters -= num_md_clusters;
5147 	bs->total_data_clusters = bs->num_free_clusters;
5148 
5149 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
5150 	cpl.u.bs_handle.cb_fn = cb_fn;
5151 	cpl.u.bs_handle.cb_arg = cb_arg;
5152 	cpl.u.bs_handle.bs = bs;
5153 
5154 	seq = bs_sequence_start(bs->md_channel, &cpl);
5155 	if (!seq) {
5156 		spdk_free(ctx->super);
5157 		free(ctx);
5158 		bs_free(bs);
5159 		cb_fn(cb_arg, NULL, -ENOMEM);
5160 		return;
5161 	}
5162 
5163 	batch = bs_sequence_to_batch(seq, bs_init_trim_cpl, ctx);
5164 
5165 	/* Clear metadata space */
5166 	bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
5167 
5168 	lba = num_md_lba;
5169 	lba_count = ctx->bs->dev->blockcnt - lba;
5170 	switch (opts.clear_method) {
5171 	case BS_CLEAR_WITH_UNMAP:
5172 		/* Trim data clusters */
5173 		bs_batch_unmap_dev(batch, lba, lba_count);
5174 		break;
5175 	case BS_CLEAR_WITH_WRITE_ZEROES:
5176 		/* Write_zeroes to data clusters */
5177 		bs_batch_write_zeroes_dev(batch, lba, lba_count);
5178 		break;
5179 	case BS_CLEAR_WITH_NONE:
5180 	default:
5181 		break;
5182 	}
5183 
5184 	bs_batch_close(batch);
5185 }
5186 
5187 /* END spdk_bs_init */
5188 
5189 /* START spdk_bs_destroy */
5190 
5191 static void
5192 bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5193 {
5194 	struct spdk_bs_load_ctx *ctx = cb_arg;
5195 	struct spdk_blob_store *bs = ctx->bs;
5196 
5197 	/*
5198 	 * We need to defer calling bs_call_cpl() until after
5199 	 * dev destruction, so tuck these away for later use.
5200 	 */
5201 	bs->unload_err = bserrno;
5202 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
5203 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
5204 
5205 	bs_sequence_finish(seq, bserrno);
5206 
5207 	bs_free(bs);
5208 	free(ctx);
5209 }
5210 
5211 void
5212 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
5213 		void *cb_arg)
5214 {
5215 	struct spdk_bs_cpl	cpl;
5216 	spdk_bs_sequence_t	*seq;
5217 	struct spdk_bs_load_ctx *ctx;
5218 
5219 	SPDK_DEBUGLOG(blob, "Destroying blobstore\n");
5220 
5221 	if (!RB_EMPTY(&bs->open_blobs)) {
5222 		SPDK_ERRLOG("Blobstore still has open blobs\n");
5223 		cb_fn(cb_arg, -EBUSY);
5224 		return;
5225 	}
5226 
5227 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5228 	cpl.u.bs_basic.cb_fn = cb_fn;
5229 	cpl.u.bs_basic.cb_arg = cb_arg;
5230 
5231 	ctx = calloc(1, sizeof(*ctx));
5232 	if (!ctx) {
5233 		cb_fn(cb_arg, -ENOMEM);
5234 		return;
5235 	}
5236 
5237 	ctx->bs = bs;
5238 
5239 	seq = bs_sequence_start(bs->md_channel, &cpl);
5240 	if (!seq) {
5241 		free(ctx);
5242 		cb_fn(cb_arg, -ENOMEM);
5243 		return;
5244 	}
5245 
5246 	/* Write zeroes to the super block */
5247 	bs_sequence_write_zeroes_dev(seq,
5248 				     bs_page_to_lba(bs, 0),
5249 				     bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
5250 				     bs_destroy_trim_cpl, ctx);
5251 }
5252 
5253 /* END spdk_bs_destroy */
5254 
5255 /* START spdk_bs_unload */
5256 
5257 static void
5258 bs_unload_finish(struct spdk_bs_load_ctx *ctx, int bserrno)
5259 {
5260 	spdk_bs_sequence_t *seq = ctx->seq;
5261 
5262 	spdk_free(ctx->super);
5263 
5264 	/*
5265 	 * We need to defer calling bs_call_cpl() until after
5266 	 * dev destruction, so tuck these away for later use.
5267 	 */
5268 	ctx->bs->unload_err = bserrno;
5269 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
5270 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
5271 
5272 	bs_sequence_finish(seq, bserrno);
5273 
5274 	bs_free(ctx->bs);
5275 	free(ctx);
5276 }
5277 
5278 static void
5279 bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5280 {
5281 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5282 
5283 	bs_unload_finish(ctx, bserrno);
5284 }
5285 
5286 static void
5287 bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5288 {
5289 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5290 
5291 	spdk_free(ctx->mask);
5292 
5293 	if (bserrno != 0) {
5294 		bs_unload_finish(ctx, bserrno);
5295 		return;
5296 	}
5297 
5298 	ctx->super->clean = 1;
5299 
5300 	bs_write_super(seq, ctx->bs, ctx->super, bs_unload_write_super_cpl, ctx);
5301 }
5302 
5303 static void
5304 bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5305 {
5306 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5307 
5308 	spdk_free(ctx->mask);
5309 	ctx->mask = NULL;
5310 
5311 	if (bserrno != 0) {
5312 		bs_unload_finish(ctx, bserrno);
5313 		return;
5314 	}
5315 
5316 	bs_write_used_clusters(seq, ctx, bs_unload_write_used_clusters_cpl);
5317 }
5318 
5319 static void
5320 bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5321 {
5322 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5323 
5324 	spdk_free(ctx->mask);
5325 	ctx->mask = NULL;
5326 
5327 	if (bserrno != 0) {
5328 		bs_unload_finish(ctx, bserrno);
5329 		return;
5330 	}
5331 
5332 	bs_write_used_blobids(seq, ctx, bs_unload_write_used_blobids_cpl);
5333 }
5334 
5335 static void
5336 bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5337 {
5338 	struct spdk_bs_load_ctx	*ctx = cb_arg;
5339 
5340 	if (bserrno != 0) {
5341 		bs_unload_finish(ctx, bserrno);
5342 		return;
5343 	}
5344 
5345 	bs_write_used_md(seq, cb_arg, bs_unload_write_used_pages_cpl);
5346 }
5347 
5348 void
5349 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
5350 {
5351 	struct spdk_bs_cpl	cpl;
5352 	struct spdk_bs_load_ctx *ctx;
5353 
5354 	SPDK_DEBUGLOG(blob, "Syncing blobstore\n");
5355 
5356 	if (!RB_EMPTY(&bs->open_blobs)) {
5357 		SPDK_ERRLOG("Blobstore still has open blobs\n");
5358 		cb_fn(cb_arg, -EBUSY);
5359 		return;
5360 	}
5361 
5362 	ctx = calloc(1, sizeof(*ctx));
5363 	if (!ctx) {
5364 		cb_fn(cb_arg, -ENOMEM);
5365 		return;
5366 	}
5367 
5368 	ctx->bs = bs;
5369 
5370 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
5371 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
5372 	if (!ctx->super) {
5373 		free(ctx);
5374 		cb_fn(cb_arg, -ENOMEM);
5375 		return;
5376 	}
5377 
5378 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5379 	cpl.u.bs_basic.cb_fn = cb_fn;
5380 	cpl.u.bs_basic.cb_arg = cb_arg;
5381 
5382 	ctx->seq = bs_sequence_start(bs->md_channel, &cpl);
5383 	if (!ctx->seq) {
5384 		spdk_free(ctx->super);
5385 		free(ctx);
5386 		cb_fn(cb_arg, -ENOMEM);
5387 		return;
5388 	}
5389 
5390 	/* Read super block */
5391 	bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0),
5392 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
5393 			     bs_unload_read_super_cpl, ctx);
5394 }
5395 
5396 /* END spdk_bs_unload */
5397 
5398 /* START spdk_bs_set_super */
5399 
5400 struct spdk_bs_set_super_ctx {
5401 	struct spdk_blob_store		*bs;
5402 	struct spdk_bs_super_block	*super;
5403 };
5404 
5405 static void
5406 bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5407 {
5408 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
5409 
5410 	if (bserrno != 0) {
5411 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
5412 	}
5413 
5414 	spdk_free(ctx->super);
5415 
5416 	bs_sequence_finish(seq, bserrno);
5417 
5418 	free(ctx);
5419 }
5420 
5421 static void
5422 bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5423 {
5424 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
5425 
5426 	if (bserrno != 0) {
5427 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
5428 		spdk_free(ctx->super);
5429 		bs_sequence_finish(seq, bserrno);
5430 		free(ctx);
5431 		return;
5432 	}
5433 
5434 	bs_write_super(seq, ctx->bs, ctx->super, bs_set_super_write_cpl, ctx);
5435 }
5436 
5437 void
5438 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
5439 		  spdk_bs_op_complete cb_fn, void *cb_arg)
5440 {
5441 	struct spdk_bs_cpl		cpl;
5442 	spdk_bs_sequence_t		*seq;
5443 	struct spdk_bs_set_super_ctx	*ctx;
5444 
5445 	SPDK_DEBUGLOG(blob, "Setting super blob id on blobstore\n");
5446 
5447 	ctx = calloc(1, sizeof(*ctx));
5448 	if (!ctx) {
5449 		cb_fn(cb_arg, -ENOMEM);
5450 		return;
5451 	}
5452 
5453 	ctx->bs = bs;
5454 
5455 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
5456 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
5457 	if (!ctx->super) {
5458 		free(ctx);
5459 		cb_fn(cb_arg, -ENOMEM);
5460 		return;
5461 	}
5462 
5463 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
5464 	cpl.u.bs_basic.cb_fn = cb_fn;
5465 	cpl.u.bs_basic.cb_arg = cb_arg;
5466 
5467 	seq = bs_sequence_start(bs->md_channel, &cpl);
5468 	if (!seq) {
5469 		spdk_free(ctx->super);
5470 		free(ctx);
5471 		cb_fn(cb_arg, -ENOMEM);
5472 		return;
5473 	}
5474 
5475 	bs->super_blob = blobid;
5476 
5477 	/* Read super block */
5478 	bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0),
5479 			     bs_byte_to_lba(bs, sizeof(*ctx->super)),
5480 			     bs_set_super_read_cpl, ctx);
5481 }
5482 
5483 /* END spdk_bs_set_super */
5484 
5485 void
5486 spdk_bs_get_super(struct spdk_blob_store *bs,
5487 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5488 {
5489 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
5490 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
5491 	} else {
5492 		cb_fn(cb_arg, bs->super_blob, 0);
5493 	}
5494 }
5495 
5496 uint64_t
5497 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
5498 {
5499 	return bs->cluster_sz;
5500 }
5501 
5502 uint64_t
5503 spdk_bs_get_page_size(struct spdk_blob_store *bs)
5504 {
5505 	return SPDK_BS_PAGE_SIZE;
5506 }
5507 
5508 uint64_t
5509 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
5510 {
5511 	return bs->io_unit_size;
5512 }
5513 
5514 uint64_t
5515 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
5516 {
5517 	return bs->num_free_clusters;
5518 }
5519 
5520 uint64_t
5521 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
5522 {
5523 	return bs->total_data_clusters;
5524 }
5525 
5526 static int
5527 bs_register_md_thread(struct spdk_blob_store *bs)
5528 {
5529 	bs->md_channel = spdk_get_io_channel(bs);
5530 	if (!bs->md_channel) {
5531 		SPDK_ERRLOG("Failed to get IO channel.\n");
5532 		return -1;
5533 	}
5534 
5535 	return 0;
5536 }
5537 
5538 static int
5539 bs_unregister_md_thread(struct spdk_blob_store *bs)
5540 {
5541 	spdk_put_io_channel(bs->md_channel);
5542 
5543 	return 0;
5544 }
5545 
5546 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
5547 {
5548 	assert(blob != NULL);
5549 
5550 	return blob->id;
5551 }
5552 
5553 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
5554 {
5555 	assert(blob != NULL);
5556 
5557 	return bs_cluster_to_page(blob->bs, blob->active.num_clusters);
5558 }
5559 
5560 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
5561 {
5562 	assert(blob != NULL);
5563 
5564 	return spdk_blob_get_num_pages(blob) * bs_io_unit_per_page(blob->bs);
5565 }
5566 
5567 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
5568 {
5569 	assert(blob != NULL);
5570 
5571 	return blob->active.num_clusters;
5572 }
5573 
5574 /* START spdk_bs_create_blob */
5575 
5576 static void
5577 bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5578 {
5579 	struct spdk_blob *blob = cb_arg;
5580 	uint32_t page_idx = bs_blobid_to_page(blob->id);
5581 
5582 	if (bserrno != 0) {
5583 		spdk_bit_array_clear(blob->bs->used_blobids, page_idx);
5584 		bs_release_md_page(blob->bs, page_idx);
5585 	}
5586 
5587 	blob_free(blob);
5588 
5589 	bs_sequence_finish(seq, bserrno);
5590 }
5591 
5592 static int
5593 blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
5594 		bool internal)
5595 {
5596 	uint64_t i;
5597 	size_t value_len = 0;
5598 	int rc;
5599 	const void *value = NULL;
5600 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
5601 		return -EINVAL;
5602 	}
5603 	for (i = 0; i < xattrs->count; i++) {
5604 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
5605 		if (value == NULL || value_len == 0) {
5606 			return -EINVAL;
5607 		}
5608 		rc = blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
5609 		if (rc < 0) {
5610 			return rc;
5611 		}
5612 	}
5613 	return 0;
5614 }
5615 
5616 static void
5617 blob_opts_copy(const struct spdk_blob_opts *src, struct spdk_blob_opts *dst)
5618 {
5619 #define FIELD_OK(field) \
5620         offsetof(struct spdk_blob_opts, field) + sizeof(src->field) <= src->opts_size
5621 
5622 #define SET_FIELD(field) \
5623         if (FIELD_OK(field)) { \
5624                 dst->field = src->field; \
5625         } \
5626 
5627 	SET_FIELD(num_clusters);
5628 	SET_FIELD(thin_provision);
5629 	SET_FIELD(clear_method);
5630 
5631 	if (FIELD_OK(xattrs)) {
5632 		memcpy(&dst->xattrs, &src->xattrs, sizeof(src->xattrs));
5633 	}
5634 
5635 	SET_FIELD(use_extent_table);
5636 
5637 	dst->opts_size = src->opts_size;
5638 
5639 	/* You should not remove this statement, but need to update the assert statement
5640 	 * if you add a new field, and also add a corresponding SET_FIELD statement */
5641 	SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_opts) == 64, "Incorrect size");
5642 
5643 #undef FIELD_OK
5644 #undef SET_FIELD
5645 }
5646 
5647 static void
5648 bs_create_blob(struct spdk_blob_store *bs,
5649 	       const struct spdk_blob_opts *opts,
5650 	       const struct spdk_blob_xattr_opts *internal_xattrs,
5651 	       spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5652 {
5653 	struct spdk_blob	*blob;
5654 	uint32_t		page_idx;
5655 	struct spdk_bs_cpl	cpl;
5656 	struct spdk_blob_opts	opts_local;
5657 	struct spdk_blob_xattr_opts internal_xattrs_default;
5658 	spdk_bs_sequence_t	*seq;
5659 	spdk_blob_id		id;
5660 	int rc;
5661 
5662 	assert(spdk_get_thread() == bs->md_thread);
5663 
5664 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
5665 	if (page_idx == UINT32_MAX) {
5666 		cb_fn(cb_arg, 0, -ENOMEM);
5667 		return;
5668 	}
5669 	spdk_bit_array_set(bs->used_blobids, page_idx);
5670 	bs_claim_md_page(bs, page_idx);
5671 
5672 	id = bs_page_to_blobid(page_idx);
5673 
5674 	SPDK_DEBUGLOG(blob, "Creating blob with id %" PRIu64 " at page %u\n", id, page_idx);
5675 
5676 	blob = blob_alloc(bs, id);
5677 	if (!blob) {
5678 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5679 		bs_release_md_page(bs, page_idx);
5680 		cb_fn(cb_arg, 0, -ENOMEM);
5681 		return;
5682 	}
5683 
5684 	spdk_blob_opts_init(&opts_local, sizeof(opts_local));
5685 	if (opts) {
5686 		blob_opts_copy(opts, &opts_local);
5687 	}
5688 
5689 	blob->use_extent_table = opts_local.use_extent_table;
5690 	if (blob->use_extent_table) {
5691 		blob->invalid_flags |= SPDK_BLOB_EXTENT_TABLE;
5692 	}
5693 
5694 	if (!internal_xattrs) {
5695 		blob_xattrs_init(&internal_xattrs_default);
5696 		internal_xattrs = &internal_xattrs_default;
5697 	}
5698 
5699 	rc = blob_set_xattrs(blob, &opts_local.xattrs, false);
5700 	if (rc < 0) {
5701 		blob_free(blob);
5702 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5703 		bs_release_md_page(bs, page_idx);
5704 		cb_fn(cb_arg, 0, rc);
5705 		return;
5706 	}
5707 
5708 	rc = blob_set_xattrs(blob, internal_xattrs, true);
5709 	if (rc < 0) {
5710 		blob_free(blob);
5711 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5712 		bs_release_md_page(bs, page_idx);
5713 		cb_fn(cb_arg, 0, rc);
5714 		return;
5715 	}
5716 
5717 	if (opts_local.thin_provision) {
5718 		blob_set_thin_provision(blob);
5719 	}
5720 
5721 	blob_set_clear_method(blob, opts_local.clear_method);
5722 
5723 	rc = blob_resize(blob, opts_local.num_clusters);
5724 	if (rc < 0) {
5725 		blob_free(blob);
5726 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5727 		bs_release_md_page(bs, page_idx);
5728 		cb_fn(cb_arg, 0, rc);
5729 		return;
5730 	}
5731 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
5732 	cpl.u.blobid.cb_fn = cb_fn;
5733 	cpl.u.blobid.cb_arg = cb_arg;
5734 	cpl.u.blobid.blobid = blob->id;
5735 
5736 	seq = bs_sequence_start(bs->md_channel, &cpl);
5737 	if (!seq) {
5738 		blob_free(blob);
5739 		spdk_bit_array_clear(bs->used_blobids, page_idx);
5740 		bs_release_md_page(bs, page_idx);
5741 		cb_fn(cb_arg, 0, -ENOMEM);
5742 		return;
5743 	}
5744 
5745 	blob_persist(seq, blob, bs_create_blob_cpl, blob);
5746 }
5747 
5748 void spdk_bs_create_blob(struct spdk_blob_store *bs,
5749 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5750 {
5751 	bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
5752 }
5753 
5754 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
5755 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5756 {
5757 	bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
5758 }
5759 
5760 /* END spdk_bs_create_blob */
5761 
5762 /* START blob_cleanup */
5763 
5764 struct spdk_clone_snapshot_ctx {
5765 	struct spdk_bs_cpl      cpl;
5766 	int bserrno;
5767 	bool frozen;
5768 
5769 	struct spdk_io_channel *channel;
5770 
5771 	/* Current cluster for inflate operation */
5772 	uint64_t cluster;
5773 
5774 	/* For inflation force allocation of all unallocated clusters and remove
5775 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
5776 	bool allocate_all;
5777 
5778 	struct {
5779 		spdk_blob_id id;
5780 		struct spdk_blob *blob;
5781 		bool md_ro;
5782 	} original;
5783 	struct {
5784 		spdk_blob_id id;
5785 		struct spdk_blob *blob;
5786 	} new;
5787 
5788 	/* xattrs specified for snapshot/clones only. They have no impact on
5789 	 * the original blobs xattrs. */
5790 	const struct spdk_blob_xattr_opts *xattrs;
5791 };
5792 
5793 static void
5794 bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
5795 {
5796 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
5797 	struct spdk_bs_cpl *cpl = &ctx->cpl;
5798 
5799 	if (bserrno != 0) {
5800 		if (ctx->bserrno != 0) {
5801 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5802 		} else {
5803 			ctx->bserrno = bserrno;
5804 		}
5805 	}
5806 
5807 	switch (cpl->type) {
5808 	case SPDK_BS_CPL_TYPE_BLOBID:
5809 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
5810 		break;
5811 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
5812 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
5813 		break;
5814 	default:
5815 		SPDK_UNREACHABLE();
5816 		break;
5817 	}
5818 
5819 	free(ctx);
5820 }
5821 
5822 static void
5823 bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
5824 {
5825 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5826 	struct spdk_blob *origblob = ctx->original.blob;
5827 
5828 	if (bserrno != 0) {
5829 		if (ctx->bserrno != 0) {
5830 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
5831 		} else {
5832 			ctx->bserrno = bserrno;
5833 		}
5834 	}
5835 
5836 	ctx->original.id = origblob->id;
5837 	origblob->locked_operation_in_progress = false;
5838 
5839 	/* Revert md_ro to original state */
5840 	origblob->md_ro = ctx->original.md_ro;
5841 
5842 	spdk_blob_close(origblob, bs_clone_snapshot_cleanup_finish, ctx);
5843 }
5844 
5845 static void
5846 bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
5847 {
5848 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5849 	struct spdk_blob *origblob = ctx->original.blob;
5850 
5851 	if (bserrno != 0) {
5852 		if (ctx->bserrno != 0) {
5853 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5854 		} else {
5855 			ctx->bserrno = bserrno;
5856 		}
5857 	}
5858 
5859 	if (ctx->frozen) {
5860 		/* Unfreeze any outstanding I/O */
5861 		blob_unfreeze_io(origblob, bs_snapshot_unfreeze_cpl, ctx);
5862 	} else {
5863 		bs_snapshot_unfreeze_cpl(ctx, 0);
5864 	}
5865 
5866 }
5867 
5868 static void
5869 bs_clone_snapshot_newblob_cleanup(struct spdk_clone_snapshot_ctx *ctx, int bserrno)
5870 {
5871 	struct spdk_blob *newblob = ctx->new.blob;
5872 
5873 	if (bserrno != 0) {
5874 		if (ctx->bserrno != 0) {
5875 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5876 		} else {
5877 			ctx->bserrno = bserrno;
5878 		}
5879 	}
5880 
5881 	ctx->new.id = newblob->id;
5882 	spdk_blob_close(newblob, bs_clone_snapshot_origblob_cleanup, ctx);
5883 }
5884 
5885 /* END blob_cleanup */
5886 
5887 /* START spdk_bs_create_snapshot */
5888 
5889 static void
5890 bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
5891 {
5892 	uint64_t *cluster_temp;
5893 	uint32_t *extent_page_temp;
5894 
5895 	cluster_temp = blob1->active.clusters;
5896 	blob1->active.clusters = blob2->active.clusters;
5897 	blob2->active.clusters = cluster_temp;
5898 
5899 	extent_page_temp = blob1->active.extent_pages;
5900 	blob1->active.extent_pages = blob2->active.extent_pages;
5901 	blob2->active.extent_pages = extent_page_temp;
5902 }
5903 
5904 static void
5905 bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
5906 {
5907 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5908 	struct spdk_blob *origblob = ctx->original.blob;
5909 	struct spdk_blob *newblob = ctx->new.blob;
5910 
5911 	if (bserrno != 0) {
5912 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5913 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5914 		return;
5915 	}
5916 
5917 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
5918 	bserrno = blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
5919 	if (bserrno != 0) {
5920 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5921 		return;
5922 	}
5923 
5924 	bs_blob_list_add(ctx->original.blob);
5925 
5926 	spdk_blob_set_read_only(newblob);
5927 
5928 	/* sync snapshot metadata */
5929 	spdk_blob_sync_md(newblob, bs_clone_snapshot_origblob_cleanup, ctx);
5930 }
5931 
5932 static void
5933 bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
5934 {
5935 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5936 	struct spdk_blob *origblob = ctx->original.blob;
5937 	struct spdk_blob *newblob = ctx->new.blob;
5938 
5939 	if (bserrno != 0) {
5940 		/* return cluster map back to original */
5941 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5942 
5943 		/* Newblob md sync failed. Valid clusters are only present in origblob.
5944 		 * Since I/O is frozen on origblob, not changes to zeroed out cluster map should have occurred.
5945 		 * Newblob needs to be reverted to thin_provisioned state at creation to properly close. */
5946 		blob_set_thin_provision(newblob);
5947 		assert(spdk_mem_all_zero(newblob->active.clusters,
5948 					 newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
5949 		assert(spdk_mem_all_zero(newblob->active.extent_pages,
5950 					 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages)));
5951 
5952 		bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5953 		return;
5954 	}
5955 
5956 	/* Set internal xattr for snapshot id */
5957 	bserrno = blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
5958 	if (bserrno != 0) {
5959 		/* return cluster map back to original */
5960 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5961 		blob_set_thin_provision(newblob);
5962 		bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5963 		return;
5964 	}
5965 
5966 	/* Create new back_bs_dev for snapshot */
5967 	origblob->back_bs_dev = bs_create_blob_bs_dev(newblob);
5968 	if (origblob->back_bs_dev == NULL) {
5969 		/* return cluster map back to original */
5970 		bs_snapshot_swap_cluster_maps(newblob, origblob);
5971 		blob_set_thin_provision(newblob);
5972 		bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
5973 		return;
5974 	}
5975 
5976 	bs_blob_list_remove(origblob);
5977 	origblob->parent_id = newblob->id;
5978 	/* set clone blob as thin provisioned */
5979 	blob_set_thin_provision(origblob);
5980 
5981 	bs_blob_list_add(newblob);
5982 
5983 	/* sync clone metadata */
5984 	spdk_blob_sync_md(origblob, bs_snapshot_origblob_sync_cpl, ctx);
5985 }
5986 
5987 static void
5988 bs_snapshot_freeze_cpl(void *cb_arg, int rc)
5989 {
5990 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5991 	struct spdk_blob *origblob = ctx->original.blob;
5992 	struct spdk_blob *newblob = ctx->new.blob;
5993 	int bserrno;
5994 
5995 	if (rc != 0) {
5996 		bs_clone_snapshot_newblob_cleanup(ctx, rc);
5997 		return;
5998 	}
5999 
6000 	ctx->frozen = true;
6001 
6002 	/* set new back_bs_dev for snapshot */
6003 	newblob->back_bs_dev = origblob->back_bs_dev;
6004 	/* Set invalid flags from origblob */
6005 	newblob->invalid_flags = origblob->invalid_flags;
6006 
6007 	/* inherit parent from original blob if set */
6008 	newblob->parent_id = origblob->parent_id;
6009 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
6010 		/* Set internal xattr for snapshot id */
6011 		bserrno = blob_set_xattr(newblob, BLOB_SNAPSHOT,
6012 					 &origblob->parent_id, sizeof(spdk_blob_id), true);
6013 		if (bserrno != 0) {
6014 			bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
6015 			return;
6016 		}
6017 	}
6018 
6019 	/* swap cluster maps */
6020 	bs_snapshot_swap_cluster_maps(newblob, origblob);
6021 
6022 	/* Set the clear method on the new blob to match the original. */
6023 	blob_set_clear_method(newblob, origblob->clear_method);
6024 
6025 	/* sync snapshot metadata */
6026 	spdk_blob_sync_md(newblob, bs_snapshot_newblob_sync_cpl, ctx);
6027 }
6028 
6029 static void
6030 bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6031 {
6032 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6033 	struct spdk_blob *origblob = ctx->original.blob;
6034 	struct spdk_blob *newblob = _blob;
6035 
6036 	if (bserrno != 0) {
6037 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6038 		return;
6039 	}
6040 
6041 	ctx->new.blob = newblob;
6042 	assert(spdk_blob_is_thin_provisioned(newblob));
6043 	assert(spdk_mem_all_zero(newblob->active.clusters,
6044 				 newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
6045 	assert(spdk_mem_all_zero(newblob->active.extent_pages,
6046 				 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages)));
6047 
6048 	blob_freeze_io(origblob, bs_snapshot_freeze_cpl, ctx);
6049 }
6050 
6051 static void
6052 bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
6053 {
6054 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6055 	struct spdk_blob *origblob = ctx->original.blob;
6056 
6057 	if (bserrno != 0) {
6058 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6059 		return;
6060 	}
6061 
6062 	ctx->new.id = blobid;
6063 	ctx->cpl.u.blobid.blobid = blobid;
6064 
6065 	spdk_bs_open_blob(origblob->bs, ctx->new.id, bs_snapshot_newblob_open_cpl, ctx);
6066 }
6067 
6068 
6069 static void
6070 bs_xattr_snapshot(void *arg, const char *name,
6071 		  const void **value, size_t *value_len)
6072 {
6073 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
6074 
6075 	struct spdk_blob *blob = (struct spdk_blob *)arg;
6076 	*value = &blob->id;
6077 	*value_len = sizeof(blob->id);
6078 }
6079 
6080 static void
6081 bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6082 {
6083 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6084 	struct spdk_blob_opts opts;
6085 	struct spdk_blob_xattr_opts internal_xattrs;
6086 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
6087 
6088 	if (bserrno != 0) {
6089 		bs_clone_snapshot_cleanup_finish(ctx, bserrno);
6090 		return;
6091 	}
6092 
6093 	ctx->original.blob = _blob;
6094 
6095 	if (_blob->data_ro || _blob->md_ro) {
6096 		SPDK_DEBUGLOG(blob, "Cannot create snapshot from read only blob with id %" PRIu64 "\n",
6097 			      _blob->id);
6098 		ctx->bserrno = -EINVAL;
6099 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6100 		return;
6101 	}
6102 
6103 	if (_blob->locked_operation_in_progress) {
6104 		SPDK_DEBUGLOG(blob, "Cannot create snapshot - another operation in progress\n");
6105 		ctx->bserrno = -EBUSY;
6106 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6107 		return;
6108 	}
6109 
6110 	_blob->locked_operation_in_progress = true;
6111 
6112 	spdk_blob_opts_init(&opts, sizeof(opts));
6113 	blob_xattrs_init(&internal_xattrs);
6114 
6115 	/* Change the size of new blob to the same as in original blob,
6116 	 * but do not allocate clusters */
6117 	opts.thin_provision = true;
6118 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
6119 	opts.use_extent_table = _blob->use_extent_table;
6120 
6121 	/* If there are any xattrs specified for snapshot, set them now */
6122 	if (ctx->xattrs) {
6123 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
6124 	}
6125 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
6126 	internal_xattrs.count = 1;
6127 	internal_xattrs.ctx = _blob;
6128 	internal_xattrs.names = xattrs_names;
6129 	internal_xattrs.get_value = bs_xattr_snapshot;
6130 
6131 	bs_create_blob(_blob->bs, &opts, &internal_xattrs,
6132 		       bs_snapshot_newblob_create_cpl, ctx);
6133 }
6134 
6135 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
6136 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
6137 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6138 {
6139 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
6140 
6141 	if (!ctx) {
6142 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
6143 		return;
6144 	}
6145 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
6146 	ctx->cpl.u.blobid.cb_fn = cb_fn;
6147 	ctx->cpl.u.blobid.cb_arg = cb_arg;
6148 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
6149 	ctx->bserrno = 0;
6150 	ctx->frozen = false;
6151 	ctx->original.id = blobid;
6152 	ctx->xattrs = snapshot_xattrs;
6153 
6154 	spdk_bs_open_blob(bs, ctx->original.id, bs_snapshot_origblob_open_cpl, ctx);
6155 }
6156 /* END spdk_bs_create_snapshot */
6157 
6158 /* START spdk_bs_create_clone */
6159 
6160 static void
6161 bs_xattr_clone(void *arg, const char *name,
6162 	       const void **value, size_t *value_len)
6163 {
6164 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
6165 
6166 	struct spdk_blob *blob = (struct spdk_blob *)arg;
6167 	*value = &blob->id;
6168 	*value_len = sizeof(blob->id);
6169 }
6170 
6171 static void
6172 bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6173 {
6174 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6175 	struct spdk_blob *clone = _blob;
6176 
6177 	ctx->new.blob = clone;
6178 	bs_blob_list_add(clone);
6179 
6180 	spdk_blob_close(clone, bs_clone_snapshot_origblob_cleanup, ctx);
6181 }
6182 
6183 static void
6184 bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
6185 {
6186 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6187 
6188 	ctx->cpl.u.blobid.blobid = blobid;
6189 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, bs_clone_newblob_open_cpl, ctx);
6190 }
6191 
6192 static void
6193 bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6194 {
6195 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6196 	struct spdk_blob_opts		opts;
6197 	struct spdk_blob_xattr_opts internal_xattrs;
6198 	char *xattr_names[] = { BLOB_SNAPSHOT };
6199 
6200 	if (bserrno != 0) {
6201 		bs_clone_snapshot_cleanup_finish(ctx, bserrno);
6202 		return;
6203 	}
6204 
6205 	ctx->original.blob = _blob;
6206 	ctx->original.md_ro = _blob->md_ro;
6207 
6208 	if (!_blob->data_ro || !_blob->md_ro) {
6209 		SPDK_DEBUGLOG(blob, "Clone not from read-only blob\n");
6210 		ctx->bserrno = -EINVAL;
6211 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6212 		return;
6213 	}
6214 
6215 	if (_blob->locked_operation_in_progress) {
6216 		SPDK_DEBUGLOG(blob, "Cannot create clone - another operation in progress\n");
6217 		ctx->bserrno = -EBUSY;
6218 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6219 		return;
6220 	}
6221 
6222 	_blob->locked_operation_in_progress = true;
6223 
6224 	spdk_blob_opts_init(&opts, sizeof(opts));
6225 	blob_xattrs_init(&internal_xattrs);
6226 
6227 	opts.thin_provision = true;
6228 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
6229 	opts.use_extent_table = _blob->use_extent_table;
6230 	if (ctx->xattrs) {
6231 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
6232 	}
6233 
6234 	/* Set internal xattr BLOB_SNAPSHOT */
6235 	internal_xattrs.count = 1;
6236 	internal_xattrs.ctx = _blob;
6237 	internal_xattrs.names = xattr_names;
6238 	internal_xattrs.get_value = bs_xattr_clone;
6239 
6240 	bs_create_blob(_blob->bs, &opts, &internal_xattrs,
6241 		       bs_clone_newblob_create_cpl, ctx);
6242 }
6243 
6244 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
6245 			  const struct spdk_blob_xattr_opts *clone_xattrs,
6246 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
6247 {
6248 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
6249 
6250 	if (!ctx) {
6251 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
6252 		return;
6253 	}
6254 
6255 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
6256 	ctx->cpl.u.blobid.cb_fn = cb_fn;
6257 	ctx->cpl.u.blobid.cb_arg = cb_arg;
6258 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
6259 	ctx->bserrno = 0;
6260 	ctx->xattrs = clone_xattrs;
6261 	ctx->original.id = blobid;
6262 
6263 	spdk_bs_open_blob(bs, ctx->original.id, bs_clone_origblob_open_cpl, ctx);
6264 }
6265 
6266 /* END spdk_bs_create_clone */
6267 
6268 /* START spdk_bs_inflate_blob */
6269 
6270 static void
6271 bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
6272 {
6273 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6274 	struct spdk_blob *_blob = ctx->original.blob;
6275 
6276 	if (bserrno != 0) {
6277 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6278 		return;
6279 	}
6280 
6281 	/* Temporarily override md_ro flag for MD modification */
6282 	_blob->md_ro = false;
6283 
6284 	bserrno = blob_set_xattr(_blob, BLOB_SNAPSHOT, &_parent->id, sizeof(spdk_blob_id), true);
6285 	if (bserrno != 0) {
6286 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6287 		return;
6288 	}
6289 
6290 	assert(_parent != NULL);
6291 
6292 	bs_blob_list_remove(_blob);
6293 	_blob->parent_id = _parent->id;
6294 
6295 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
6296 	_blob->back_bs_dev = bs_create_blob_bs_dev(_parent);
6297 	bs_blob_list_add(_blob);
6298 
6299 	spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx);
6300 }
6301 
6302 static void
6303 bs_inflate_blob_done(struct spdk_clone_snapshot_ctx *ctx)
6304 {
6305 	struct spdk_blob *_blob = ctx->original.blob;
6306 	struct spdk_blob *_parent;
6307 
6308 	if (ctx->allocate_all) {
6309 		/* remove thin provisioning */
6310 		bs_blob_list_remove(_blob);
6311 		blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
6312 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
6313 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
6314 		_blob->back_bs_dev = NULL;
6315 		_blob->parent_id = SPDK_BLOBID_INVALID;
6316 	} else {
6317 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
6318 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
6319 			/* We must change the parent of the inflated blob */
6320 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
6321 					  bs_inflate_blob_set_parent_cpl, ctx);
6322 			return;
6323 		}
6324 
6325 		bs_blob_list_remove(_blob);
6326 		blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
6327 		_blob->parent_id = SPDK_BLOBID_INVALID;
6328 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
6329 		_blob->back_bs_dev = bs_create_zeroes_dev();
6330 	}
6331 
6332 	/* Temporarily override md_ro flag for MD modification */
6333 	_blob->md_ro = false;
6334 	_blob->state = SPDK_BLOB_STATE_DIRTY;
6335 
6336 	spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx);
6337 }
6338 
6339 /* Check if cluster needs allocation */
6340 static inline bool
6341 bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
6342 {
6343 	struct spdk_blob_bs_dev *b;
6344 
6345 	assert(blob != NULL);
6346 
6347 	if (blob->active.clusters[cluster] != 0) {
6348 		/* Cluster is already allocated */
6349 		return false;
6350 	}
6351 
6352 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
6353 		/* Blob have no parent blob */
6354 		return allocate_all;
6355 	}
6356 
6357 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
6358 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
6359 }
6360 
6361 static void
6362 bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
6363 {
6364 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6365 	struct spdk_blob *_blob = ctx->original.blob;
6366 	struct spdk_bs_cpl cpl;
6367 	spdk_bs_user_op_t *op;
6368 	uint64_t offset;
6369 
6370 	if (bserrno != 0) {
6371 		bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
6372 		return;
6373 	}
6374 
6375 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
6376 		if (bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
6377 			break;
6378 		}
6379 	}
6380 
6381 	if (ctx->cluster < _blob->active.num_clusters) {
6382 		offset = bs_cluster_to_lba(_blob->bs, ctx->cluster);
6383 
6384 		/* We may safely increment a cluster before copying */
6385 		ctx->cluster++;
6386 
6387 		/* Use a dummy 0B read as a context for cluster copy */
6388 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6389 		cpl.u.blob_basic.cb_fn = bs_inflate_blob_touch_next;
6390 		cpl.u.blob_basic.cb_arg = ctx;
6391 
6392 		op = bs_user_op_alloc(ctx->channel, &cpl, SPDK_BLOB_READ, _blob,
6393 				      NULL, 0, offset, 0);
6394 		if (!op) {
6395 			bs_clone_snapshot_origblob_cleanup(ctx, -ENOMEM);
6396 			return;
6397 		}
6398 
6399 		bs_allocate_and_copy_cluster(_blob, ctx->channel, offset, op);
6400 	} else {
6401 		bs_inflate_blob_done(ctx);
6402 	}
6403 }
6404 
6405 static void
6406 bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6407 {
6408 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
6409 	uint64_t clusters_needed;
6410 	uint64_t i;
6411 
6412 	if (bserrno != 0) {
6413 		bs_clone_snapshot_cleanup_finish(ctx, bserrno);
6414 		return;
6415 	}
6416 
6417 	ctx->original.blob = _blob;
6418 	ctx->original.md_ro = _blob->md_ro;
6419 
6420 	if (_blob->locked_operation_in_progress) {
6421 		SPDK_DEBUGLOG(blob, "Cannot inflate blob - another operation in progress\n");
6422 		ctx->bserrno = -EBUSY;
6423 		spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx);
6424 		return;
6425 	}
6426 
6427 	_blob->locked_operation_in_progress = true;
6428 
6429 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
6430 		/* This blob have no parent, so we cannot decouple it. */
6431 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
6432 		bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
6433 		return;
6434 	}
6435 
6436 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
6437 		/* This is not thin provisioned blob. No need to inflate. */
6438 		bs_clone_snapshot_origblob_cleanup(ctx, 0);
6439 		return;
6440 	}
6441 
6442 	/* Do two passes - one to verify that we can obtain enough clusters
6443 	 * and another to actually claim them.
6444 	 */
6445 	clusters_needed = 0;
6446 	for (i = 0; i < _blob->active.num_clusters; i++) {
6447 		if (bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
6448 			clusters_needed++;
6449 		}
6450 	}
6451 
6452 	if (clusters_needed > _blob->bs->num_free_clusters) {
6453 		/* Not enough free clusters. Cannot satisfy the request. */
6454 		bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
6455 		return;
6456 	}
6457 
6458 	ctx->cluster = 0;
6459 	bs_inflate_blob_touch_next(ctx, 0);
6460 }
6461 
6462 static void
6463 bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
6464 		spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
6465 {
6466 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
6467 
6468 	if (!ctx) {
6469 		cb_fn(cb_arg, -ENOMEM);
6470 		return;
6471 	}
6472 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6473 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
6474 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
6475 	ctx->bserrno = 0;
6476 	ctx->original.id = blobid;
6477 	ctx->channel = channel;
6478 	ctx->allocate_all = allocate_all;
6479 
6480 	spdk_bs_open_blob(bs, ctx->original.id, bs_inflate_blob_open_cpl, ctx);
6481 }
6482 
6483 void
6484 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
6485 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
6486 {
6487 	bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
6488 }
6489 
6490 void
6491 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
6492 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
6493 {
6494 	bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
6495 }
6496 /* END spdk_bs_inflate_blob */
6497 
6498 /* START spdk_blob_resize */
6499 struct spdk_bs_resize_ctx {
6500 	spdk_blob_op_complete cb_fn;
6501 	void *cb_arg;
6502 	struct spdk_blob *blob;
6503 	uint64_t sz;
6504 	int rc;
6505 };
6506 
6507 static void
6508 bs_resize_unfreeze_cpl(void *cb_arg, int rc)
6509 {
6510 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
6511 
6512 	if (rc != 0) {
6513 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
6514 	}
6515 
6516 	if (ctx->rc != 0) {
6517 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
6518 		rc = ctx->rc;
6519 	}
6520 
6521 	ctx->blob->locked_operation_in_progress = false;
6522 
6523 	ctx->cb_fn(ctx->cb_arg, rc);
6524 	free(ctx);
6525 }
6526 
6527 static void
6528 bs_resize_freeze_cpl(void *cb_arg, int rc)
6529 {
6530 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
6531 
6532 	if (rc != 0) {
6533 		ctx->blob->locked_operation_in_progress = false;
6534 		ctx->cb_fn(ctx->cb_arg, rc);
6535 		free(ctx);
6536 		return;
6537 	}
6538 
6539 	ctx->rc = blob_resize(ctx->blob, ctx->sz);
6540 
6541 	blob_unfreeze_io(ctx->blob, bs_resize_unfreeze_cpl, ctx);
6542 }
6543 
6544 void
6545 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
6546 {
6547 	struct spdk_bs_resize_ctx *ctx;
6548 
6549 	blob_verify_md_op(blob);
6550 
6551 	SPDK_DEBUGLOG(blob, "Resizing blob %" PRIu64 " to %" PRIu64 " clusters\n", blob->id, sz);
6552 
6553 	if (blob->md_ro) {
6554 		cb_fn(cb_arg, -EPERM);
6555 		return;
6556 	}
6557 
6558 	if (sz == blob->active.num_clusters) {
6559 		cb_fn(cb_arg, 0);
6560 		return;
6561 	}
6562 
6563 	if (blob->locked_operation_in_progress) {
6564 		cb_fn(cb_arg, -EBUSY);
6565 		return;
6566 	}
6567 
6568 	ctx = calloc(1, sizeof(*ctx));
6569 	if (!ctx) {
6570 		cb_fn(cb_arg, -ENOMEM);
6571 		return;
6572 	}
6573 
6574 	blob->locked_operation_in_progress = true;
6575 	ctx->cb_fn = cb_fn;
6576 	ctx->cb_arg = cb_arg;
6577 	ctx->blob = blob;
6578 	ctx->sz = sz;
6579 	blob_freeze_io(blob, bs_resize_freeze_cpl, ctx);
6580 }
6581 
6582 /* END spdk_blob_resize */
6583 
6584 
6585 /* START spdk_bs_delete_blob */
6586 
6587 static void
6588 bs_delete_close_cpl(void *cb_arg, int bserrno)
6589 {
6590 	spdk_bs_sequence_t *seq = cb_arg;
6591 
6592 	bs_sequence_finish(seq, bserrno);
6593 }
6594 
6595 static void
6596 bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6597 {
6598 	struct spdk_blob *blob = cb_arg;
6599 
6600 	if (bserrno != 0) {
6601 		/*
6602 		 * We already removed this blob from the blobstore tailq, so
6603 		 *  we need to free it here since this is the last reference
6604 		 *  to it.
6605 		 */
6606 		blob_free(blob);
6607 		bs_delete_close_cpl(seq, bserrno);
6608 		return;
6609 	}
6610 
6611 	/*
6612 	 * This will immediately decrement the ref_count and call
6613 	 *  the completion routine since the metadata state is clean.
6614 	 *  By calling spdk_blob_close, we reduce the number of call
6615 	 *  points into code that touches the blob->open_ref count
6616 	 *  and the blobstore's blob list.
6617 	 */
6618 	spdk_blob_close(blob, bs_delete_close_cpl, seq);
6619 }
6620 
6621 struct delete_snapshot_ctx {
6622 	struct spdk_blob_list *parent_snapshot_entry;
6623 	struct spdk_blob *snapshot;
6624 	bool snapshot_md_ro;
6625 	struct spdk_blob *clone;
6626 	bool clone_md_ro;
6627 	spdk_blob_op_with_handle_complete cb_fn;
6628 	void *cb_arg;
6629 	int bserrno;
6630 	uint32_t next_extent_page;
6631 };
6632 
6633 static void
6634 delete_blob_cleanup_finish(void *cb_arg, int bserrno)
6635 {
6636 	struct delete_snapshot_ctx *ctx = cb_arg;
6637 
6638 	if (bserrno != 0) {
6639 		SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno);
6640 	}
6641 
6642 	assert(ctx != NULL);
6643 
6644 	if (bserrno != 0 && ctx->bserrno == 0) {
6645 		ctx->bserrno = bserrno;
6646 	}
6647 
6648 	ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno);
6649 	free(ctx);
6650 }
6651 
6652 static void
6653 delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno)
6654 {
6655 	struct delete_snapshot_ctx *ctx = cb_arg;
6656 
6657 	if (bserrno != 0) {
6658 		ctx->bserrno = bserrno;
6659 		SPDK_ERRLOG("Clone cleanup error %d\n", bserrno);
6660 	}
6661 
6662 	if (ctx->bserrno != 0) {
6663 		assert(blob_lookup(ctx->snapshot->bs, ctx->snapshot->id) == NULL);
6664 		RB_INSERT(spdk_blob_tree, &ctx->snapshot->bs->open_blobs, ctx->snapshot);
6665 		spdk_bit_array_set(ctx->snapshot->bs->open_blobids, ctx->snapshot->id);
6666 	}
6667 
6668 	ctx->snapshot->locked_operation_in_progress = false;
6669 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
6670 
6671 	spdk_blob_close(ctx->snapshot, delete_blob_cleanup_finish, ctx);
6672 }
6673 
6674 static void
6675 delete_snapshot_cleanup_clone(void *cb_arg, int bserrno)
6676 {
6677 	struct delete_snapshot_ctx *ctx = cb_arg;
6678 
6679 	ctx->clone->locked_operation_in_progress = false;
6680 	ctx->clone->md_ro = ctx->clone_md_ro;
6681 
6682 	spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx);
6683 }
6684 
6685 static void
6686 delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
6687 {
6688 	struct delete_snapshot_ctx *ctx = cb_arg;
6689 
6690 	if (bserrno) {
6691 		ctx->bserrno = bserrno;
6692 		delete_snapshot_cleanup_clone(ctx, 0);
6693 		return;
6694 	}
6695 
6696 	ctx->clone->locked_operation_in_progress = false;
6697 	spdk_blob_close(ctx->clone, delete_blob_cleanup_finish, ctx);
6698 }
6699 
6700 static void
6701 delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno)
6702 {
6703 	struct delete_snapshot_ctx *ctx = cb_arg;
6704 	struct spdk_blob_list *parent_snapshot_entry = NULL;
6705 	struct spdk_blob_list *snapshot_entry = NULL;
6706 	struct spdk_blob_list *clone_entry = NULL;
6707 	struct spdk_blob_list *snapshot_clone_entry = NULL;
6708 
6709 	if (bserrno) {
6710 		SPDK_ERRLOG("Failed to sync MD on blob\n");
6711 		ctx->bserrno = bserrno;
6712 		delete_snapshot_cleanup_clone(ctx, 0);
6713 		return;
6714 	}
6715 
6716 	/* Get snapshot entry for the snapshot we want to remove */
6717 	snapshot_entry = bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id);
6718 
6719 	assert(snapshot_entry != NULL);
6720 
6721 	/* Remove clone entry in this snapshot (at this point there can be only one clone) */
6722 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6723 	assert(clone_entry != NULL);
6724 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
6725 	snapshot_entry->clone_count--;
6726 	assert(TAILQ_EMPTY(&snapshot_entry->clones));
6727 
6728 	if (ctx->snapshot->parent_id != SPDK_BLOBID_INVALID) {
6729 		/* This snapshot is at the same time a clone of another snapshot - we need to
6730 		 * update parent snapshot (remove current clone, add new one inherited from
6731 		 * the snapshot that is being removed) */
6732 
6733 		/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
6734 		 * snapshot that we are removing */
6735 		blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry,
6736 						    &snapshot_clone_entry);
6737 
6738 		/* Switch clone entry in parent snapshot */
6739 		TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link);
6740 		TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link);
6741 		free(snapshot_clone_entry);
6742 	} else {
6743 		/* No parent snapshot - just remove clone entry */
6744 		free(clone_entry);
6745 	}
6746 
6747 	/* Restore md_ro flags */
6748 	ctx->clone->md_ro = ctx->clone_md_ro;
6749 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
6750 
6751 	blob_unfreeze_io(ctx->clone, delete_snapshot_unfreeze_cpl, ctx);
6752 }
6753 
6754 static void
6755 delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno)
6756 {
6757 	struct delete_snapshot_ctx *ctx = cb_arg;
6758 	uint64_t i;
6759 
6760 	ctx->snapshot->md_ro = false;
6761 
6762 	if (bserrno) {
6763 		SPDK_ERRLOG("Failed to sync MD on clone\n");
6764 		ctx->bserrno = bserrno;
6765 
6766 		/* Restore snapshot to previous state */
6767 		bserrno = blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true);
6768 		if (bserrno != 0) {
6769 			delete_snapshot_cleanup_clone(ctx, bserrno);
6770 			return;
6771 		}
6772 
6773 		spdk_blob_sync_md(ctx->snapshot, delete_snapshot_cleanup_clone, ctx);
6774 		return;
6775 	}
6776 
6777 	/* Clear cluster map entries for snapshot */
6778 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
6779 		if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) {
6780 			ctx->snapshot->active.clusters[i] = 0;
6781 		}
6782 	}
6783 	for (i = 0; i < ctx->snapshot->active.num_extent_pages &&
6784 	     i < ctx->clone->active.num_extent_pages; i++) {
6785 		if (ctx->clone->active.extent_pages[i] == ctx->snapshot->active.extent_pages[i]) {
6786 			ctx->snapshot->active.extent_pages[i] = 0;
6787 		}
6788 	}
6789 
6790 	blob_set_thin_provision(ctx->snapshot);
6791 	ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY;
6792 
6793 	if (ctx->parent_snapshot_entry != NULL) {
6794 		ctx->snapshot->back_bs_dev = NULL;
6795 	}
6796 
6797 	spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_cpl, ctx);
6798 }
6799 
6800 static void
6801 delete_snapshot_update_extent_pages_cpl(struct delete_snapshot_ctx *ctx)
6802 {
6803 	/* Delete old backing bs_dev from clone (related to snapshot that will be removed) */
6804 	ctx->clone->back_bs_dev->destroy(ctx->clone->back_bs_dev);
6805 
6806 	/* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */
6807 	if (ctx->parent_snapshot_entry != NULL) {
6808 		/* ...to parent snapshot */
6809 		ctx->clone->parent_id = ctx->parent_snapshot_entry->id;
6810 		ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev;
6811 		blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id,
6812 			       sizeof(spdk_blob_id),
6813 			       true);
6814 	} else {
6815 		/* ...to blobid invalid and zeroes dev */
6816 		ctx->clone->parent_id = SPDK_BLOBID_INVALID;
6817 		ctx->clone->back_bs_dev = bs_create_zeroes_dev();
6818 		blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true);
6819 	}
6820 
6821 	spdk_blob_sync_md(ctx->clone, delete_snapshot_sync_clone_cpl, ctx);
6822 }
6823 
6824 static void
6825 delete_snapshot_update_extent_pages(void *cb_arg, int bserrno)
6826 {
6827 	struct delete_snapshot_ctx *ctx = cb_arg;
6828 	uint32_t *extent_page;
6829 	uint64_t i;
6830 
6831 	for (i = ctx->next_extent_page; i < ctx->snapshot->active.num_extent_pages &&
6832 	     i < ctx->clone->active.num_extent_pages; i++) {
6833 		if (ctx->snapshot->active.extent_pages[i] == 0) {
6834 			/* No extent page to use from snapshot */
6835 			continue;
6836 		}
6837 
6838 		extent_page = &ctx->clone->active.extent_pages[i];
6839 		if (*extent_page == 0) {
6840 			/* Copy extent page from snapshot when clone did not have a matching one */
6841 			*extent_page = ctx->snapshot->active.extent_pages[i];
6842 			continue;
6843 		}
6844 
6845 		/* Clone and snapshot both contain partially filled matching extent pages.
6846 		 * Update the clone extent page in place with cluster map containing the mix of both. */
6847 		ctx->next_extent_page = i + 1;
6848 
6849 		blob_write_extent_page(ctx->clone, *extent_page, i * SPDK_EXTENTS_PER_EP,
6850 				       delete_snapshot_update_extent_pages, ctx);
6851 		return;
6852 	}
6853 	delete_snapshot_update_extent_pages_cpl(ctx);
6854 }
6855 
6856 static void
6857 delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno)
6858 {
6859 	struct delete_snapshot_ctx *ctx = cb_arg;
6860 	uint64_t i;
6861 
6862 	/* Temporarily override md_ro flag for clone for MD modification */
6863 	ctx->clone_md_ro = ctx->clone->md_ro;
6864 	ctx->clone->md_ro = false;
6865 
6866 	if (bserrno) {
6867 		SPDK_ERRLOG("Failed to sync MD with xattr on blob\n");
6868 		ctx->bserrno = bserrno;
6869 		delete_snapshot_cleanup_clone(ctx, 0);
6870 		return;
6871 	}
6872 
6873 	/* Copy snapshot map to clone map (only unallocated clusters in clone) */
6874 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
6875 		if (ctx->clone->active.clusters[i] == 0) {
6876 			ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i];
6877 		}
6878 	}
6879 	ctx->next_extent_page = 0;
6880 	delete_snapshot_update_extent_pages(ctx, 0);
6881 }
6882 
6883 static void
6884 delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno)
6885 {
6886 	struct delete_snapshot_ctx *ctx = cb_arg;
6887 
6888 	if (bserrno) {
6889 		SPDK_ERRLOG("Failed to freeze I/O on clone\n");
6890 		ctx->bserrno = bserrno;
6891 		delete_snapshot_cleanup_clone(ctx, 0);
6892 		return;
6893 	}
6894 
6895 	/* Temporarily override md_ro flag for snapshot for MD modification */
6896 	ctx->snapshot_md_ro = ctx->snapshot->md_ro;
6897 	ctx->snapshot->md_ro = false;
6898 
6899 	/* Mark blob as pending for removal for power failure safety, use clone id for recovery */
6900 	ctx->bserrno = blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id,
6901 				      sizeof(spdk_blob_id), true);
6902 	if (ctx->bserrno != 0) {
6903 		delete_snapshot_cleanup_clone(ctx, 0);
6904 		return;
6905 	}
6906 
6907 	spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_xattr_cpl, ctx);
6908 }
6909 
6910 static void
6911 delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno)
6912 {
6913 	struct delete_snapshot_ctx *ctx = cb_arg;
6914 
6915 	if (bserrno) {
6916 		SPDK_ERRLOG("Failed to open clone\n");
6917 		ctx->bserrno = bserrno;
6918 		delete_snapshot_cleanup_snapshot(ctx, 0);
6919 		return;
6920 	}
6921 
6922 	ctx->clone = clone;
6923 
6924 	if (clone->locked_operation_in_progress) {
6925 		SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress on its clone\n");
6926 		ctx->bserrno = -EBUSY;
6927 		spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx);
6928 		return;
6929 	}
6930 
6931 	clone->locked_operation_in_progress = true;
6932 
6933 	blob_freeze_io(clone, delete_snapshot_freeze_io_cb, ctx);
6934 }
6935 
6936 static void
6937 update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx)
6938 {
6939 	struct spdk_blob_list *snapshot_entry = NULL;
6940 	struct spdk_blob_list *clone_entry = NULL;
6941 	struct spdk_blob_list *snapshot_clone_entry = NULL;
6942 
6943 	/* Get snapshot entry for the snapshot we want to remove */
6944 	snapshot_entry = bs_get_snapshot_entry(snapshot->bs, snapshot->id);
6945 
6946 	assert(snapshot_entry != NULL);
6947 
6948 	/* Get clone of the snapshot (at this point there can be only one clone) */
6949 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6950 	assert(snapshot_entry->clone_count == 1);
6951 	assert(clone_entry != NULL);
6952 
6953 	/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
6954 	 * snapshot that we are removing */
6955 	blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry,
6956 					    &snapshot_clone_entry);
6957 
6958 	spdk_bs_open_blob(snapshot->bs, clone_entry->id, delete_snapshot_open_clone_cb, ctx);
6959 }
6960 
6961 static void
6962 bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno)
6963 {
6964 	spdk_bs_sequence_t *seq = cb_arg;
6965 	struct spdk_blob_list *snapshot_entry = NULL;
6966 	uint32_t page_num;
6967 
6968 	if (bserrno) {
6969 		SPDK_ERRLOG("Failed to remove blob\n");
6970 		bs_sequence_finish(seq, bserrno);
6971 		return;
6972 	}
6973 
6974 	/* Remove snapshot from the list */
6975 	snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
6976 	if (snapshot_entry != NULL) {
6977 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
6978 		free(snapshot_entry);
6979 	}
6980 
6981 	page_num = bs_blobid_to_page(blob->id);
6982 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
6983 	blob->state = SPDK_BLOB_STATE_DIRTY;
6984 	blob->active.num_pages = 0;
6985 	blob_resize(blob, 0);
6986 
6987 	blob_persist(seq, blob, bs_delete_persist_cpl, blob);
6988 }
6989 
6990 static int
6991 bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone)
6992 {
6993 	struct spdk_blob_list *snapshot_entry = NULL;
6994 	struct spdk_blob_list *clone_entry = NULL;
6995 	struct spdk_blob *clone = NULL;
6996 	bool has_one_clone = false;
6997 
6998 	/* Check if this is a snapshot with clones */
6999 	snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
7000 	if (snapshot_entry != NULL) {
7001 		if (snapshot_entry->clone_count > 1) {
7002 			SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n");
7003 			return -EBUSY;
7004 		} else if (snapshot_entry->clone_count == 1) {
7005 			has_one_clone = true;
7006 		}
7007 	}
7008 
7009 	/* Check if someone has this blob open (besides this delete context):
7010 	 * - open_ref = 1 - only this context opened blob, so it is ok to remove it
7011 	 * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot
7012 	 *	and that is ok, because we will update it accordingly */
7013 	if (blob->open_ref <= 2 && has_one_clone) {
7014 		clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
7015 		assert(clone_entry != NULL);
7016 		clone = blob_lookup(blob->bs, clone_entry->id);
7017 
7018 		if (blob->open_ref == 2 && clone == NULL) {
7019 			/* Clone is closed and someone else opened this blob */
7020 			SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
7021 			return -EBUSY;
7022 		}
7023 
7024 		*update_clone = true;
7025 		return 0;
7026 	}
7027 
7028 	if (blob->open_ref > 1) {
7029 		SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
7030 		return -EBUSY;
7031 	}
7032 
7033 	assert(has_one_clone == false);
7034 	*update_clone = false;
7035 	return 0;
7036 }
7037 
7038 static void
7039 bs_delete_enomem_close_cpl(void *cb_arg, int bserrno)
7040 {
7041 	spdk_bs_sequence_t *seq = cb_arg;
7042 
7043 	bs_sequence_finish(seq, -ENOMEM);
7044 }
7045 
7046 static void
7047 bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
7048 {
7049 	spdk_bs_sequence_t *seq = cb_arg;
7050 	struct delete_snapshot_ctx *ctx;
7051 	bool update_clone = false;
7052 
7053 	if (bserrno != 0) {
7054 		bs_sequence_finish(seq, bserrno);
7055 		return;
7056 	}
7057 
7058 	blob_verify_md_op(blob);
7059 
7060 	ctx = calloc(1, sizeof(*ctx));
7061 	if (ctx == NULL) {
7062 		spdk_blob_close(blob, bs_delete_enomem_close_cpl, seq);
7063 		return;
7064 	}
7065 
7066 	ctx->snapshot = blob;
7067 	ctx->cb_fn = bs_delete_blob_finish;
7068 	ctx->cb_arg = seq;
7069 
7070 	/* Check if blob can be removed and if it is a snapshot with clone on top of it */
7071 	ctx->bserrno = bs_is_blob_deletable(blob, &update_clone);
7072 	if (ctx->bserrno) {
7073 		spdk_blob_close(blob, delete_blob_cleanup_finish, ctx);
7074 		return;
7075 	}
7076 
7077 	if (blob->locked_operation_in_progress) {
7078 		SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress\n");
7079 		ctx->bserrno = -EBUSY;
7080 		spdk_blob_close(blob, delete_blob_cleanup_finish, ctx);
7081 		return;
7082 	}
7083 
7084 	blob->locked_operation_in_progress = true;
7085 
7086 	/*
7087 	 * Remove the blob from the blob_store list now, to ensure it does not
7088 	 *  get returned after this point by blob_lookup().
7089 	 */
7090 	spdk_bit_array_clear(blob->bs->open_blobids, blob->id);
7091 	RB_REMOVE(spdk_blob_tree, &blob->bs->open_blobs, blob);
7092 
7093 	if (update_clone) {
7094 		/* This blob is a snapshot with active clone - update clone first */
7095 		update_clone_on_snapshot_deletion(blob, ctx);
7096 	} else {
7097 		/* This blob does not have any clones - just remove it */
7098 		bs_blob_list_remove(blob);
7099 		bs_delete_blob_finish(seq, blob, 0);
7100 		free(ctx);
7101 	}
7102 }
7103 
7104 void
7105 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
7106 		    spdk_blob_op_complete cb_fn, void *cb_arg)
7107 {
7108 	struct spdk_bs_cpl	cpl;
7109 	spdk_bs_sequence_t	*seq;
7110 
7111 	SPDK_DEBUGLOG(blob, "Deleting blob %" PRIu64 "\n", blobid);
7112 
7113 	assert(spdk_get_thread() == bs->md_thread);
7114 
7115 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7116 	cpl.u.blob_basic.cb_fn = cb_fn;
7117 	cpl.u.blob_basic.cb_arg = cb_arg;
7118 
7119 	seq = bs_sequence_start(bs->md_channel, &cpl);
7120 	if (!seq) {
7121 		cb_fn(cb_arg, -ENOMEM);
7122 		return;
7123 	}
7124 
7125 	spdk_bs_open_blob(bs, blobid, bs_delete_open_cpl, seq);
7126 }
7127 
7128 /* END spdk_bs_delete_blob */
7129 
7130 /* START spdk_bs_open_blob */
7131 
7132 static void
7133 bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7134 {
7135 	struct spdk_blob *blob = cb_arg;
7136 	struct spdk_blob *existing;
7137 
7138 	if (bserrno != 0) {
7139 		blob_free(blob);
7140 		seq->cpl.u.blob_handle.blob = NULL;
7141 		bs_sequence_finish(seq, bserrno);
7142 		return;
7143 	}
7144 
7145 	existing = blob_lookup(blob->bs, blob->id);
7146 	if (existing) {
7147 		blob_free(blob);
7148 		existing->open_ref++;
7149 		seq->cpl.u.blob_handle.blob = existing;
7150 		bs_sequence_finish(seq, 0);
7151 		return;
7152 	}
7153 
7154 	blob->open_ref++;
7155 
7156 	spdk_bit_array_set(blob->bs->open_blobids, blob->id);
7157 	RB_INSERT(spdk_blob_tree, &blob->bs->open_blobs, blob);
7158 
7159 	bs_sequence_finish(seq, bserrno);
7160 }
7161 
7162 static inline void
7163 blob_open_opts_copy(const struct spdk_blob_open_opts *src, struct spdk_blob_open_opts *dst)
7164 {
7165 #define FIELD_OK(field) \
7166         offsetof(struct spdk_blob_opts, field) + sizeof(src->field) <= src->opts_size
7167 
7168 #define SET_FIELD(field) \
7169         if (FIELD_OK(field)) { \
7170                 dst->field = src->field; \
7171         } \
7172 
7173 	SET_FIELD(clear_method);
7174 
7175 	dst->opts_size = src->opts_size;
7176 
7177 	/* You should not remove this statement, but need to update the assert statement
7178 	 * if you add a new field, and also add a corresponding SET_FIELD statement */
7179 	SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_open_opts) == 16, "Incorrect size");
7180 
7181 #undef FIELD_OK
7182 #undef SET_FIELD
7183 }
7184 
7185 static void
7186 bs_open_blob(struct spdk_blob_store *bs,
7187 	     spdk_blob_id blobid,
7188 	     struct spdk_blob_open_opts *opts,
7189 	     spdk_blob_op_with_handle_complete cb_fn,
7190 	     void *cb_arg)
7191 {
7192 	struct spdk_blob		*blob;
7193 	struct spdk_bs_cpl		cpl;
7194 	struct spdk_blob_open_opts	opts_local;
7195 	spdk_bs_sequence_t		*seq;
7196 	uint32_t			page_num;
7197 
7198 	SPDK_DEBUGLOG(blob, "Opening blob %" PRIu64 "\n", blobid);
7199 	assert(spdk_get_thread() == bs->md_thread);
7200 
7201 	page_num = bs_blobid_to_page(blobid);
7202 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
7203 		/* Invalid blobid */
7204 		cb_fn(cb_arg, NULL, -ENOENT);
7205 		return;
7206 	}
7207 
7208 	blob = blob_lookup(bs, blobid);
7209 	if (blob) {
7210 		blob->open_ref++;
7211 		cb_fn(cb_arg, blob, 0);
7212 		return;
7213 	}
7214 
7215 	blob = blob_alloc(bs, blobid);
7216 	if (!blob) {
7217 		cb_fn(cb_arg, NULL, -ENOMEM);
7218 		return;
7219 	}
7220 
7221 	spdk_blob_open_opts_init(&opts_local, sizeof(opts_local));
7222 	if (opts) {
7223 		blob_open_opts_copy(opts, &opts_local);
7224 	}
7225 
7226 	blob->clear_method = opts_local.clear_method;
7227 
7228 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
7229 	cpl.u.blob_handle.cb_fn = cb_fn;
7230 	cpl.u.blob_handle.cb_arg = cb_arg;
7231 	cpl.u.blob_handle.blob = blob;
7232 
7233 	seq = bs_sequence_start(bs->md_channel, &cpl);
7234 	if (!seq) {
7235 		blob_free(blob);
7236 		cb_fn(cb_arg, NULL, -ENOMEM);
7237 		return;
7238 	}
7239 
7240 	blob_load(seq, blob, bs_open_blob_cpl, blob);
7241 }
7242 
7243 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
7244 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7245 {
7246 	bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
7247 }
7248 
7249 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
7250 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7251 {
7252 	bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
7253 }
7254 
7255 /* END spdk_bs_open_blob */
7256 
7257 /* START spdk_blob_set_read_only */
7258 int spdk_blob_set_read_only(struct spdk_blob *blob)
7259 {
7260 	blob_verify_md_op(blob);
7261 
7262 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
7263 
7264 	blob->state = SPDK_BLOB_STATE_DIRTY;
7265 	return 0;
7266 }
7267 /* END spdk_blob_set_read_only */
7268 
7269 /* START spdk_blob_sync_md */
7270 
7271 static void
7272 blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7273 {
7274 	struct spdk_blob *blob = cb_arg;
7275 
7276 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
7277 		blob->data_ro = true;
7278 		blob->md_ro = true;
7279 	}
7280 
7281 	bs_sequence_finish(seq, bserrno);
7282 }
7283 
7284 static void
7285 blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
7286 {
7287 	struct spdk_bs_cpl	cpl;
7288 	spdk_bs_sequence_t	*seq;
7289 
7290 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7291 	cpl.u.blob_basic.cb_fn = cb_fn;
7292 	cpl.u.blob_basic.cb_arg = cb_arg;
7293 
7294 	seq = bs_sequence_start(blob->bs->md_channel, &cpl);
7295 	if (!seq) {
7296 		cb_fn(cb_arg, -ENOMEM);
7297 		return;
7298 	}
7299 
7300 	blob_persist(seq, blob, blob_sync_md_cpl, blob);
7301 }
7302 
7303 void
7304 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
7305 {
7306 	blob_verify_md_op(blob);
7307 
7308 	SPDK_DEBUGLOG(blob, "Syncing blob %" PRIu64 "\n", blob->id);
7309 
7310 	if (blob->md_ro) {
7311 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
7312 		cb_fn(cb_arg, 0);
7313 		return;
7314 	}
7315 
7316 	blob_sync_md(blob, cb_fn, cb_arg);
7317 }
7318 
7319 /* END spdk_blob_sync_md */
7320 
7321 struct spdk_blob_insert_cluster_ctx {
7322 	struct spdk_thread	*thread;
7323 	struct spdk_blob	*blob;
7324 	uint32_t		cluster_num;	/* cluster index in blob */
7325 	uint32_t		cluster;	/* cluster on disk */
7326 	uint32_t		extent_page;	/* extent page on disk */
7327 	int			rc;
7328 	spdk_blob_op_complete	cb_fn;
7329 	void			*cb_arg;
7330 };
7331 
7332 static void
7333 blob_insert_cluster_msg_cpl(void *arg)
7334 {
7335 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7336 
7337 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
7338 	free(ctx);
7339 }
7340 
7341 static void
7342 blob_insert_cluster_msg_cb(void *arg, int bserrno)
7343 {
7344 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7345 
7346 	ctx->rc = bserrno;
7347 	spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx);
7348 }
7349 
7350 static void
7351 blob_insert_new_ep_cb(void *arg, int bserrno)
7352 {
7353 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7354 	uint32_t *extent_page;
7355 
7356 	extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
7357 	*extent_page = ctx->extent_page;
7358 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
7359 	blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx);
7360 }
7361 
7362 static void
7363 blob_persist_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7364 {
7365 	struct spdk_blob_md_page        *page = cb_arg;
7366 
7367 	bs_sequence_finish(seq, bserrno);
7368 	spdk_free(page);
7369 }
7370 
7371 static void
7372 blob_write_extent_page(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num,
7373 		       spdk_blob_op_complete cb_fn, void *cb_arg)
7374 {
7375 	spdk_bs_sequence_t		*seq;
7376 	struct spdk_bs_cpl		cpl;
7377 	struct spdk_blob_md_page	*page = NULL;
7378 	uint32_t			page_count = 0;
7379 	int				rc;
7380 
7381 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7382 	cpl.u.blob_basic.cb_fn = cb_fn;
7383 	cpl.u.blob_basic.cb_arg = cb_arg;
7384 
7385 	seq = bs_sequence_start(blob->bs->md_channel, &cpl);
7386 	if (!seq) {
7387 		cb_fn(cb_arg, -ENOMEM);
7388 		return;
7389 	}
7390 	rc = blob_serialize_add_page(blob, &page, &page_count, &page);
7391 	if (rc < 0) {
7392 		bs_sequence_finish(seq, rc);
7393 		return;
7394 	}
7395 
7396 	blob_serialize_extent_page(blob, cluster_num, page);
7397 
7398 	page->crc = blob_md_page_calc_crc(page);
7399 
7400 	assert(spdk_bit_array_get(blob->bs->used_md_pages, extent) == true);
7401 
7402 	bs_sequence_write_dev(seq, page, bs_md_page_to_lba(blob->bs, extent),
7403 			      bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE),
7404 			      blob_persist_extent_page_cpl, page);
7405 }
7406 
7407 static void
7408 blob_insert_cluster_msg(void *arg)
7409 {
7410 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
7411 	uint32_t *extent_page;
7412 
7413 	ctx->rc = blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
7414 	if (ctx->rc != 0) {
7415 		spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx);
7416 		return;
7417 	}
7418 
7419 	if (ctx->blob->use_extent_table == false) {
7420 		/* Extent table is not used, proceed with sync of md that will only use extents_rle. */
7421 		ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
7422 		blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx);
7423 		return;
7424 	}
7425 
7426 	extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
7427 	if (*extent_page == 0) {
7428 		/* Extent page requires allocation.
7429 		 * It was already claimed in the used_md_pages map and placed in ctx. */
7430 		assert(ctx->extent_page != 0);
7431 		assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
7432 		blob_write_extent_page(ctx->blob, ctx->extent_page, ctx->cluster_num,
7433 				       blob_insert_new_ep_cb, ctx);
7434 	} else {
7435 		/* It is possible for original thread to allocate extent page for
7436 		 * different cluster in the same extent page. In such case proceed with
7437 		 * updating the existing extent page, but release the additional one. */
7438 		if (ctx->extent_page != 0) {
7439 			assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
7440 			bs_release_md_page(ctx->blob->bs, ctx->extent_page);
7441 			ctx->extent_page = 0;
7442 		}
7443 		/* Extent page already allocated.
7444 		 * Every cluster allocation, requires just an update of single extent page. */
7445 		blob_write_extent_page(ctx->blob, *extent_page, ctx->cluster_num,
7446 				       blob_insert_cluster_msg_cb, ctx);
7447 	}
7448 }
7449 
7450 static void
7451 blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
7452 				 uint64_t cluster, uint32_t extent_page, spdk_blob_op_complete cb_fn, void *cb_arg)
7453 {
7454 	struct spdk_blob_insert_cluster_ctx *ctx;
7455 
7456 	ctx = calloc(1, sizeof(*ctx));
7457 	if (ctx == NULL) {
7458 		cb_fn(cb_arg, -ENOMEM);
7459 		return;
7460 	}
7461 
7462 	ctx->thread = spdk_get_thread();
7463 	ctx->blob = blob;
7464 	ctx->cluster_num = cluster_num;
7465 	ctx->cluster = cluster;
7466 	ctx->extent_page = extent_page;
7467 	ctx->cb_fn = cb_fn;
7468 	ctx->cb_arg = cb_arg;
7469 
7470 	spdk_thread_send_msg(blob->bs->md_thread, blob_insert_cluster_msg, ctx);
7471 }
7472 
7473 /* START spdk_blob_close */
7474 
7475 static void
7476 blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
7477 {
7478 	struct spdk_blob *blob = cb_arg;
7479 
7480 	if (bserrno == 0) {
7481 		blob->open_ref--;
7482 		if (blob->open_ref == 0) {
7483 			/*
7484 			 * Blobs with active.num_pages == 0 are deleted blobs.
7485 			 *  these blobs are removed from the blob_store list
7486 			 *  when the deletion process starts - so don't try to
7487 			 *  remove them again.
7488 			 */
7489 			if (blob->active.num_pages > 0) {
7490 				spdk_bit_array_clear(blob->bs->open_blobids, blob->id);
7491 				RB_REMOVE(spdk_blob_tree, &blob->bs->open_blobs, blob);
7492 			}
7493 			blob_free(blob);
7494 		}
7495 	}
7496 
7497 	bs_sequence_finish(seq, bserrno);
7498 }
7499 
7500 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
7501 {
7502 	struct spdk_bs_cpl	cpl;
7503 	spdk_bs_sequence_t	*seq;
7504 
7505 	blob_verify_md_op(blob);
7506 
7507 	SPDK_DEBUGLOG(blob, "Closing blob %" PRIu64 "\n", blob->id);
7508 
7509 	if (blob->open_ref == 0) {
7510 		cb_fn(cb_arg, -EBADF);
7511 		return;
7512 	}
7513 
7514 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
7515 	cpl.u.blob_basic.cb_fn = cb_fn;
7516 	cpl.u.blob_basic.cb_arg = cb_arg;
7517 
7518 	seq = bs_sequence_start(blob->bs->md_channel, &cpl);
7519 	if (!seq) {
7520 		cb_fn(cb_arg, -ENOMEM);
7521 		return;
7522 	}
7523 
7524 	/* Sync metadata */
7525 	blob_persist(seq, blob, blob_close_cpl, blob);
7526 }
7527 
7528 /* END spdk_blob_close */
7529 
7530 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
7531 {
7532 	return spdk_get_io_channel(bs);
7533 }
7534 
7535 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
7536 {
7537 	spdk_put_io_channel(channel);
7538 }
7539 
7540 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
7541 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
7542 {
7543 	blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
7544 			       SPDK_BLOB_UNMAP);
7545 }
7546 
7547 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
7548 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
7549 {
7550 	blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
7551 			       SPDK_BLOB_WRITE_ZEROES);
7552 }
7553 
7554 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
7555 			void *payload, uint64_t offset, uint64_t length,
7556 			spdk_blob_op_complete cb_fn, void *cb_arg)
7557 {
7558 	blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
7559 			       SPDK_BLOB_WRITE);
7560 }
7561 
7562 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
7563 		       void *payload, uint64_t offset, uint64_t length,
7564 		       spdk_blob_op_complete cb_fn, void *cb_arg)
7565 {
7566 	blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
7567 			       SPDK_BLOB_READ);
7568 }
7569 
7570 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
7571 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
7572 			 spdk_blob_op_complete cb_fn, void *cb_arg)
7573 {
7574 	blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
7575 }
7576 
7577 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
7578 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
7579 			spdk_blob_op_complete cb_fn, void *cb_arg)
7580 {
7581 	blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
7582 }
7583 
7584 struct spdk_bs_iter_ctx {
7585 	int64_t page_num;
7586 	struct spdk_blob_store *bs;
7587 
7588 	spdk_blob_op_with_handle_complete cb_fn;
7589 	void *cb_arg;
7590 };
7591 
7592 static void
7593 bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
7594 {
7595 	struct spdk_bs_iter_ctx *ctx = cb_arg;
7596 	struct spdk_blob_store *bs = ctx->bs;
7597 	spdk_blob_id id;
7598 
7599 	if (bserrno == 0) {
7600 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
7601 		free(ctx);
7602 		return;
7603 	}
7604 
7605 	ctx->page_num++;
7606 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
7607 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
7608 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
7609 		free(ctx);
7610 		return;
7611 	}
7612 
7613 	id = bs_page_to_blobid(ctx->page_num);
7614 
7615 	spdk_bs_open_blob(bs, id, bs_iter_cpl, ctx);
7616 }
7617 
7618 void
7619 spdk_bs_iter_first(struct spdk_blob_store *bs,
7620 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7621 {
7622 	struct spdk_bs_iter_ctx *ctx;
7623 
7624 	ctx = calloc(1, sizeof(*ctx));
7625 	if (!ctx) {
7626 		cb_fn(cb_arg, NULL, -ENOMEM);
7627 		return;
7628 	}
7629 
7630 	ctx->page_num = -1;
7631 	ctx->bs = bs;
7632 	ctx->cb_fn = cb_fn;
7633 	ctx->cb_arg = cb_arg;
7634 
7635 	bs_iter_cpl(ctx, NULL, -1);
7636 }
7637 
7638 static void
7639 bs_iter_close_cpl(void *cb_arg, int bserrno)
7640 {
7641 	struct spdk_bs_iter_ctx *ctx = cb_arg;
7642 
7643 	bs_iter_cpl(ctx, NULL, -1);
7644 }
7645 
7646 void
7647 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
7648 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
7649 {
7650 	struct spdk_bs_iter_ctx *ctx;
7651 
7652 	assert(blob != NULL);
7653 
7654 	ctx = calloc(1, sizeof(*ctx));
7655 	if (!ctx) {
7656 		cb_fn(cb_arg, NULL, -ENOMEM);
7657 		return;
7658 	}
7659 
7660 	ctx->page_num = bs_blobid_to_page(blob->id);
7661 	ctx->bs = bs;
7662 	ctx->cb_fn = cb_fn;
7663 	ctx->cb_arg = cb_arg;
7664 
7665 	/* Close the existing blob */
7666 	spdk_blob_close(blob, bs_iter_close_cpl, ctx);
7667 }
7668 
7669 static int
7670 blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
7671 	       uint16_t value_len, bool internal)
7672 {
7673 	struct spdk_xattr_tailq *xattrs;
7674 	struct spdk_xattr	*xattr;
7675 	size_t			desc_size;
7676 	void			*tmp;
7677 
7678 	blob_verify_md_op(blob);
7679 
7680 	if (blob->md_ro) {
7681 		return -EPERM;
7682 	}
7683 
7684 	desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len;
7685 	if (desc_size > SPDK_BS_MAX_DESC_SIZE) {
7686 		SPDK_DEBUGLOG(blob, "Xattr '%s' of size %zu does not fix into single page %zu\n", name,
7687 			      desc_size, SPDK_BS_MAX_DESC_SIZE);
7688 		return -ENOMEM;
7689 	}
7690 
7691 	if (internal) {
7692 		xattrs = &blob->xattrs_internal;
7693 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
7694 	} else {
7695 		xattrs = &blob->xattrs;
7696 	}
7697 
7698 	TAILQ_FOREACH(xattr, xattrs, link) {
7699 		if (!strcmp(name, xattr->name)) {
7700 			tmp = malloc(value_len);
7701 			if (!tmp) {
7702 				return -ENOMEM;
7703 			}
7704 
7705 			free(xattr->value);
7706 			xattr->value_len = value_len;
7707 			xattr->value = tmp;
7708 			memcpy(xattr->value, value, value_len);
7709 
7710 			blob->state = SPDK_BLOB_STATE_DIRTY;
7711 
7712 			return 0;
7713 		}
7714 	}
7715 
7716 	xattr = calloc(1, sizeof(*xattr));
7717 	if (!xattr) {
7718 		return -ENOMEM;
7719 	}
7720 
7721 	xattr->name = strdup(name);
7722 	if (!xattr->name) {
7723 		free(xattr);
7724 		return -ENOMEM;
7725 	}
7726 
7727 	xattr->value_len = value_len;
7728 	xattr->value = malloc(value_len);
7729 	if (!xattr->value) {
7730 		free(xattr->name);
7731 		free(xattr);
7732 		return -ENOMEM;
7733 	}
7734 	memcpy(xattr->value, value, value_len);
7735 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
7736 
7737 	blob->state = SPDK_BLOB_STATE_DIRTY;
7738 
7739 	return 0;
7740 }
7741 
7742 int
7743 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
7744 		    uint16_t value_len)
7745 {
7746 	return blob_set_xattr(blob, name, value, value_len, false);
7747 }
7748 
7749 static int
7750 blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
7751 {
7752 	struct spdk_xattr_tailq *xattrs;
7753 	struct spdk_xattr	*xattr;
7754 
7755 	blob_verify_md_op(blob);
7756 
7757 	if (blob->md_ro) {
7758 		return -EPERM;
7759 	}
7760 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
7761 
7762 	TAILQ_FOREACH(xattr, xattrs, link) {
7763 		if (!strcmp(name, xattr->name)) {
7764 			TAILQ_REMOVE(xattrs, xattr, link);
7765 			free(xattr->value);
7766 			free(xattr->name);
7767 			free(xattr);
7768 
7769 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
7770 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
7771 			}
7772 			blob->state = SPDK_BLOB_STATE_DIRTY;
7773 
7774 			return 0;
7775 		}
7776 	}
7777 
7778 	return -ENOENT;
7779 }
7780 
7781 int
7782 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
7783 {
7784 	return blob_remove_xattr(blob, name, false);
7785 }
7786 
7787 static int
7788 blob_get_xattr_value(struct spdk_blob *blob, const char *name,
7789 		     const void **value, size_t *value_len, bool internal)
7790 {
7791 	struct spdk_xattr	*xattr;
7792 	struct spdk_xattr_tailq *xattrs;
7793 
7794 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
7795 
7796 	TAILQ_FOREACH(xattr, xattrs, link) {
7797 		if (!strcmp(name, xattr->name)) {
7798 			*value = xattr->value;
7799 			*value_len = xattr->value_len;
7800 			return 0;
7801 		}
7802 	}
7803 	return -ENOENT;
7804 }
7805 
7806 int
7807 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
7808 			  const void **value, size_t *value_len)
7809 {
7810 	blob_verify_md_op(blob);
7811 
7812 	return blob_get_xattr_value(blob, name, value, value_len, false);
7813 }
7814 
7815 struct spdk_xattr_names {
7816 	uint32_t	count;
7817 	const char	*names[0];
7818 };
7819 
7820 static int
7821 blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
7822 {
7823 	struct spdk_xattr	*xattr;
7824 	int			count = 0;
7825 
7826 	TAILQ_FOREACH(xattr, xattrs, link) {
7827 		count++;
7828 	}
7829 
7830 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
7831 	if (*names == NULL) {
7832 		return -ENOMEM;
7833 	}
7834 
7835 	TAILQ_FOREACH(xattr, xattrs, link) {
7836 		(*names)->names[(*names)->count++] = xattr->name;
7837 	}
7838 
7839 	return 0;
7840 }
7841 
7842 int
7843 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
7844 {
7845 	blob_verify_md_op(blob);
7846 
7847 	return blob_get_xattr_names(&blob->xattrs, names);
7848 }
7849 
7850 uint32_t
7851 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
7852 {
7853 	assert(names != NULL);
7854 
7855 	return names->count;
7856 }
7857 
7858 const char *
7859 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
7860 {
7861 	if (index >= names->count) {
7862 		return NULL;
7863 	}
7864 
7865 	return names->names[index];
7866 }
7867 
7868 void
7869 spdk_xattr_names_free(struct spdk_xattr_names *names)
7870 {
7871 	free(names);
7872 }
7873 
7874 struct spdk_bs_type
7875 spdk_bs_get_bstype(struct spdk_blob_store *bs)
7876 {
7877 	return bs->bstype;
7878 }
7879 
7880 void
7881 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
7882 {
7883 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
7884 }
7885 
7886 bool
7887 spdk_blob_is_read_only(struct spdk_blob *blob)
7888 {
7889 	assert(blob != NULL);
7890 	return (blob->data_ro || blob->md_ro);
7891 }
7892 
7893 bool
7894 spdk_blob_is_snapshot(struct spdk_blob *blob)
7895 {
7896 	struct spdk_blob_list *snapshot_entry;
7897 
7898 	assert(blob != NULL);
7899 
7900 	snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id);
7901 	if (snapshot_entry == NULL) {
7902 		return false;
7903 	}
7904 
7905 	return true;
7906 }
7907 
7908 bool
7909 spdk_blob_is_clone(struct spdk_blob *blob)
7910 {
7911 	assert(blob != NULL);
7912 
7913 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
7914 		assert(spdk_blob_is_thin_provisioned(blob));
7915 		return true;
7916 	}
7917 
7918 	return false;
7919 }
7920 
7921 bool
7922 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
7923 {
7924 	assert(blob != NULL);
7925 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
7926 }
7927 
7928 static void
7929 blob_update_clear_method(struct spdk_blob *blob)
7930 {
7931 	enum blob_clear_method stored_cm;
7932 
7933 	assert(blob != NULL);
7934 
7935 	/* If BLOB_CLEAR_WITH_DEFAULT was passed in, use the setting stored
7936 	 * in metadata previously.  If something other than the default was
7937 	 * specified, ignore stored value and used what was passed in.
7938 	 */
7939 	stored_cm = ((blob->md_ro_flags & SPDK_BLOB_CLEAR_METHOD) >> SPDK_BLOB_CLEAR_METHOD_SHIFT);
7940 
7941 	if (blob->clear_method == BLOB_CLEAR_WITH_DEFAULT) {
7942 		blob->clear_method = stored_cm;
7943 	} else if (blob->clear_method != stored_cm) {
7944 		SPDK_WARNLOG("Using passed in clear method 0x%x instead of stored value of 0x%x\n",
7945 			     blob->clear_method, stored_cm);
7946 	}
7947 }
7948 
7949 spdk_blob_id
7950 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
7951 {
7952 	struct spdk_blob_list *snapshot_entry = NULL;
7953 	struct spdk_blob_list *clone_entry = NULL;
7954 
7955 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
7956 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
7957 			if (clone_entry->id == blob_id) {
7958 				return snapshot_entry->id;
7959 			}
7960 		}
7961 	}
7962 
7963 	return SPDK_BLOBID_INVALID;
7964 }
7965 
7966 int
7967 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
7968 		     size_t *count)
7969 {
7970 	struct spdk_blob_list *snapshot_entry, *clone_entry;
7971 	size_t n;
7972 
7973 	snapshot_entry = bs_get_snapshot_entry(bs, blobid);
7974 	if (snapshot_entry == NULL) {
7975 		*count = 0;
7976 		return 0;
7977 	}
7978 
7979 	if (ids == NULL || *count < snapshot_entry->clone_count) {
7980 		*count = snapshot_entry->clone_count;
7981 		return -ENOMEM;
7982 	}
7983 	*count = snapshot_entry->clone_count;
7984 
7985 	n = 0;
7986 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
7987 		ids[n++] = clone_entry->id;
7988 	}
7989 
7990 	return 0;
7991 }
7992 
7993 SPDK_LOG_REGISTER_COMPONENT(blob)
7994