xref: /spdk/lib/blob/blobstore.c (revision 712a3f69d32632bf6c862f00200f7f437d3f7529)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright (c) Intel Corporation.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
11  *     * Redistributions of source code must retain the above copyright
12  *       notice, this list of conditions and the following disclaimer.
13  *     * Redistributions in binary form must reproduce the above copyright
14  *       notice, this list of conditions and the following disclaimer in
15  *       the documentation and/or other materials provided with the
16  *       distribution.
17  *     * Neither the name of Intel Corporation nor the names of its
18  *       contributors may be used to endorse or promote products derived
19  *       from this software without specific prior written permission.
20  *
21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32  */
33 
34 #include "spdk/stdinc.h"
35 
36 #include "spdk/blob.h"
37 #include "spdk/crc32.h"
38 #include "spdk/env.h"
39 #include "spdk/queue.h"
40 #include "spdk/thread.h"
41 #include "spdk/bit_array.h"
42 #include "spdk/likely.h"
43 #include "spdk/util.h"
44 #include "spdk/string.h"
45 
46 #include "spdk_internal/assert.h"
47 #include "spdk_internal/log.h"
48 
49 #include "blobstore.h"
50 
51 #define BLOB_CRC32C_INITIAL    0xffffffffUL
52 
53 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs);
54 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs);
55 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
56 static void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
57 		uint64_t cluster, uint32_t extent, spdk_blob_op_complete cb_fn, void *cb_arg);
58 
59 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
60 				uint16_t value_len, bool internal);
61 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
62 				      const void **value, size_t *value_len, bool internal);
63 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal);
64 
65 static void _spdk_blob_insert_extent(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num,
66 				     spdk_blob_op_complete cb_fn, void *cb_arg);
67 
68 static void
69 _spdk_blob_verify_md_op(struct spdk_blob *blob)
70 {
71 	assert(blob != NULL);
72 	assert(spdk_get_thread() == blob->bs->md_thread);
73 	assert(blob->state != SPDK_BLOB_STATE_LOADING);
74 }
75 
76 static struct spdk_blob_list *
77 _spdk_bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid)
78 {
79 	struct spdk_blob_list *snapshot_entry = NULL;
80 
81 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
82 		if (snapshot_entry->id == blobid) {
83 			break;
84 		}
85 	}
86 
87 	return snapshot_entry;
88 }
89 
90 static void
91 _spdk_bs_claim_md_page(struct spdk_blob_store *bs, uint32_t page)
92 {
93 	assert(page < spdk_bit_array_capacity(bs->used_md_pages));
94 	assert(spdk_bit_array_get(bs->used_md_pages, page) == false);
95 
96 	spdk_bit_array_set(bs->used_md_pages, page);
97 }
98 
99 static void
100 _spdk_bs_release_md_page(struct spdk_blob_store *bs, uint32_t page)
101 {
102 	assert(page < spdk_bit_array_capacity(bs->used_md_pages));
103 	assert(spdk_bit_array_get(bs->used_md_pages, page) == true);
104 
105 	spdk_bit_array_clear(bs->used_md_pages, page);
106 }
107 
108 static void
109 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
110 {
111 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
112 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false);
113 	assert(bs->num_free_clusters > 0);
114 
115 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num);
116 
117 	spdk_bit_array_set(bs->used_clusters, cluster_num);
118 	bs->num_free_clusters--;
119 }
120 
121 static int
122 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
123 {
124 	uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
125 
126 	_spdk_blob_verify_md_op(blob);
127 
128 	if (*cluster_lba != 0) {
129 		return -EEXIST;
130 	}
131 
132 	*cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster);
133 	return 0;
134 }
135 
136 static int
137 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
138 			  uint64_t *lowest_free_cluster, uint32_t *lowest_free_md_page, bool update_map)
139 {
140 	uint32_t *extent_page = 0;
141 
142 	pthread_mutex_lock(&blob->bs->used_clusters_mutex);
143 	*lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters,
144 			       *lowest_free_cluster);
145 	if (*lowest_free_cluster == UINT32_MAX) {
146 		/* No more free clusters. Cannot satisfy the request */
147 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
148 		return -ENOSPC;
149 	}
150 
151 	if (blob->use_extent_table) {
152 		extent_page = _spdk_bs_cluster_to_extent_page(blob, cluster_num);
153 		if (*extent_page == 0) {
154 			/* No extent_page is allocated for the cluster */
155 			*lowest_free_md_page = spdk_bit_array_find_first_clear(blob->bs->used_md_pages,
156 					       *lowest_free_md_page);
157 			if (*lowest_free_md_page == UINT32_MAX) {
158 				/* No more free md pages. Cannot satisfy the request */
159 				pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
160 				return -ENOSPC;
161 			}
162 			_spdk_bs_claim_md_page(blob->bs, *lowest_free_md_page);
163 		}
164 	}
165 
166 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id);
167 	_spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster);
168 
169 	pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
170 
171 	if (update_map) {
172 		_spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster);
173 		if (blob->use_extent_table && *extent_page == 0) {
174 			*extent_page = *lowest_free_md_page;
175 		}
176 	}
177 
178 	return 0;
179 }
180 
181 static void
182 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
183 {
184 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
185 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true);
186 	assert(bs->num_free_clusters < bs->total_clusters);
187 
188 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num);
189 
190 	pthread_mutex_lock(&bs->used_clusters_mutex);
191 	spdk_bit_array_clear(bs->used_clusters, cluster_num);
192 	bs->num_free_clusters++;
193 	pthread_mutex_unlock(&bs->used_clusters_mutex);
194 }
195 
196 static void
197 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
198 {
199 	xattrs->count = 0;
200 	xattrs->names = NULL;
201 	xattrs->ctx = NULL;
202 	xattrs->get_value = NULL;
203 }
204 
205 void
206 spdk_blob_opts_init(struct spdk_blob_opts *opts)
207 {
208 	opts->num_clusters = 0;
209 	opts->thin_provision = false;
210 	opts->clear_method = BLOB_CLEAR_WITH_DEFAULT;
211 	_spdk_blob_xattrs_init(&opts->xattrs);
212 	opts->use_extent_table = true;
213 }
214 
215 void
216 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts)
217 {
218 	opts->clear_method = BLOB_CLEAR_WITH_DEFAULT;
219 }
220 
221 static struct spdk_blob *
222 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
223 {
224 	struct spdk_blob *blob;
225 
226 	blob = calloc(1, sizeof(*blob));
227 	if (!blob) {
228 		return NULL;
229 	}
230 
231 	blob->id = id;
232 	blob->bs = bs;
233 
234 	blob->parent_id = SPDK_BLOBID_INVALID;
235 
236 	blob->state = SPDK_BLOB_STATE_DIRTY;
237 	blob->extent_rle_found = false;
238 	blob->extent_table_found = false;
239 	blob->active.num_pages = 1;
240 	blob->active.pages = calloc(1, sizeof(*blob->active.pages));
241 	if (!blob->active.pages) {
242 		free(blob);
243 		return NULL;
244 	}
245 
246 	blob->active.pages[0] = _spdk_bs_blobid_to_page(id);
247 
248 	TAILQ_INIT(&blob->xattrs);
249 	TAILQ_INIT(&blob->xattrs_internal);
250 
251 	return blob;
252 }
253 
254 static void
255 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs)
256 {
257 	struct spdk_xattr	*xattr, *xattr_tmp;
258 
259 	TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
260 		TAILQ_REMOVE(xattrs, xattr, link);
261 		free(xattr->name);
262 		free(xattr->value);
263 		free(xattr);
264 	}
265 }
266 
267 static void
268 _spdk_blob_free(struct spdk_blob *blob)
269 {
270 	assert(blob != NULL);
271 
272 	free(blob->active.extent_pages);
273 	free(blob->clean.extent_pages);
274 	free(blob->active.clusters);
275 	free(blob->clean.clusters);
276 	free(blob->active.pages);
277 	free(blob->clean.pages);
278 
279 	_spdk_xattrs_free(&blob->xattrs);
280 	_spdk_xattrs_free(&blob->xattrs_internal);
281 
282 	if (blob->back_bs_dev) {
283 		blob->back_bs_dev->destroy(blob->back_bs_dev);
284 	}
285 
286 	free(blob);
287 }
288 
289 struct freeze_io_ctx {
290 	struct spdk_bs_cpl cpl;
291 	struct spdk_blob *blob;
292 };
293 
294 static void
295 _spdk_blob_io_sync(struct spdk_io_channel_iter *i)
296 {
297 	spdk_for_each_channel_continue(i, 0);
298 }
299 
300 static void
301 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i)
302 {
303 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
304 	struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
305 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
306 	struct spdk_bs_request_set	*set;
307 	struct spdk_bs_user_op_args	*args;
308 	spdk_bs_user_op_t *op, *tmp;
309 
310 	TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
311 		set = (struct spdk_bs_request_set *)op;
312 		args = &set->u.user_op;
313 
314 		if (args->blob == ctx->blob) {
315 			TAILQ_REMOVE(&ch->queued_io, op, link);
316 			spdk_bs_user_op_execute(op);
317 		}
318 	}
319 
320 	spdk_for_each_channel_continue(i, 0);
321 }
322 
323 static void
324 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status)
325 {
326 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
327 
328 	ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
329 
330 	free(ctx);
331 }
332 
333 static void
334 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
335 {
336 	struct freeze_io_ctx *ctx;
337 
338 	ctx = calloc(1, sizeof(*ctx));
339 	if (!ctx) {
340 		cb_fn(cb_arg, -ENOMEM);
341 		return;
342 	}
343 
344 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
345 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
346 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
347 	ctx->blob = blob;
348 
349 	/* Freeze I/O on blob */
350 	blob->frozen_refcnt++;
351 
352 	if (blob->frozen_refcnt == 1) {
353 		spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl);
354 	} else {
355 		cb_fn(cb_arg, 0);
356 		free(ctx);
357 	}
358 }
359 
360 static void
361 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
362 {
363 	struct freeze_io_ctx *ctx;
364 
365 	ctx = calloc(1, sizeof(*ctx));
366 	if (!ctx) {
367 		cb_fn(cb_arg, -ENOMEM);
368 		return;
369 	}
370 
371 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
372 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
373 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
374 	ctx->blob = blob;
375 
376 	assert(blob->frozen_refcnt > 0);
377 
378 	blob->frozen_refcnt--;
379 
380 	if (blob->frozen_refcnt == 0) {
381 		spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl);
382 	} else {
383 		cb_fn(cb_arg, 0);
384 		free(ctx);
385 	}
386 }
387 
388 static int
389 _spdk_blob_mark_clean(struct spdk_blob *blob)
390 {
391 	uint32_t *extent_pages = NULL;
392 	uint64_t *clusters = NULL;
393 	uint32_t *pages = NULL;
394 
395 	assert(blob != NULL);
396 
397 	if (blob->active.num_extent_pages) {
398 		assert(blob->active.extent_pages);
399 		extent_pages = calloc(blob->active.num_extent_pages, sizeof(*blob->active.extent_pages));
400 		if (!extent_pages) {
401 			return -ENOMEM;
402 		}
403 		memcpy(extent_pages, blob->active.extent_pages,
404 		       blob->active.num_extent_pages * sizeof(*extent_pages));
405 	}
406 
407 	if (blob->active.num_clusters) {
408 		assert(blob->active.clusters);
409 		clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
410 		if (!clusters) {
411 			free(extent_pages);
412 			return -ENOMEM;
413 		}
414 		memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters));
415 	}
416 
417 	if (blob->active.num_pages) {
418 		assert(blob->active.pages);
419 		pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
420 		if (!pages) {
421 			free(extent_pages);
422 			free(clusters);
423 			return -ENOMEM;
424 		}
425 		memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
426 	}
427 
428 	free(blob->clean.extent_pages);
429 	free(blob->clean.clusters);
430 	free(blob->clean.pages);
431 
432 	blob->clean.num_extent_pages = blob->active.num_extent_pages;
433 	blob->clean.extent_pages = blob->active.extent_pages;
434 	blob->clean.num_clusters = blob->active.num_clusters;
435 	blob->clean.clusters = blob->active.clusters;
436 	blob->clean.num_pages = blob->active.num_pages;
437 	blob->clean.pages = blob->active.pages;
438 
439 	blob->active.extent_pages = extent_pages;
440 	blob->active.clusters = clusters;
441 	blob->active.pages = pages;
442 
443 	/* If the metadata was dirtied again while the metadata was being written to disk,
444 	 *  we do not want to revert the DIRTY state back to CLEAN here.
445 	 */
446 	if (blob->state == SPDK_BLOB_STATE_LOADING) {
447 		blob->state = SPDK_BLOB_STATE_CLEAN;
448 	}
449 
450 	return 0;
451 }
452 
453 static int
454 _spdk_blob_deserialize_xattr(struct spdk_blob *blob,
455 			     struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
456 {
457 	struct spdk_xattr                       *xattr;
458 
459 	if (desc_xattr->length != sizeof(desc_xattr->name_length) +
460 	    sizeof(desc_xattr->value_length) +
461 	    desc_xattr->name_length + desc_xattr->value_length) {
462 		return -EINVAL;
463 	}
464 
465 	xattr = calloc(1, sizeof(*xattr));
466 	if (xattr == NULL) {
467 		return -ENOMEM;
468 	}
469 
470 	xattr->name = malloc(desc_xattr->name_length + 1);
471 	if (xattr->name == NULL) {
472 		free(xattr);
473 		return -ENOMEM;
474 	}
475 	memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
476 	xattr->name[desc_xattr->name_length] = '\0';
477 
478 	xattr->value = malloc(desc_xattr->value_length);
479 	if (xattr->value == NULL) {
480 		free(xattr->name);
481 		free(xattr);
482 		return -ENOMEM;
483 	}
484 	xattr->value_len = desc_xattr->value_length;
485 	memcpy(xattr->value,
486 	       (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
487 	       desc_xattr->value_length);
488 
489 	TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
490 
491 	return 0;
492 }
493 
494 
495 static int
496 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
497 {
498 	struct spdk_blob_md_descriptor *desc;
499 	size_t	cur_desc = 0;
500 	void *tmp;
501 
502 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
503 	while (cur_desc < sizeof(page->descriptors)) {
504 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
505 			if (desc->length == 0) {
506 				/* If padding and length are 0, this terminates the page */
507 				break;
508 			}
509 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
510 			struct spdk_blob_md_descriptor_flags	*desc_flags;
511 
512 			desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
513 
514 			if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
515 				return -EINVAL;
516 			}
517 
518 			if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
519 			    SPDK_BLOB_INVALID_FLAGS_MASK) {
520 				return -EINVAL;
521 			}
522 
523 			if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
524 			    SPDK_BLOB_DATA_RO_FLAGS_MASK) {
525 				blob->data_ro = true;
526 				blob->md_ro = true;
527 			}
528 
529 			if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
530 			    SPDK_BLOB_MD_RO_FLAGS_MASK) {
531 				blob->md_ro = true;
532 			}
533 
534 			if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
535 				blob->data_ro = true;
536 				blob->md_ro = true;
537 			}
538 
539 			blob->invalid_flags = desc_flags->invalid_flags;
540 			blob->data_ro_flags = desc_flags->data_ro_flags;
541 			blob->md_ro_flags = desc_flags->md_ro_flags;
542 
543 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
544 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
545 			unsigned int				i, j;
546 			unsigned int				cluster_count = blob->active.num_clusters;
547 
548 			if (blob->extent_table_found) {
549 				/* Extent Table already present in the md,
550 				 * both descriptors should never be at the same time. */
551 				return -EINVAL;
552 			}
553 			blob->extent_rle_found = true;
554 
555 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
556 
557 			if (desc_extent_rle->length == 0 ||
558 			    (desc_extent_rle->length % sizeof(desc_extent_rle->extents[0]) != 0)) {
559 				return -EINVAL;
560 			}
561 
562 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
563 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
564 					if (desc_extent_rle->extents[i].cluster_idx != 0) {
565 						if (!spdk_bit_array_get(blob->bs->used_clusters,
566 									desc_extent_rle->extents[i].cluster_idx + j)) {
567 							return -EINVAL;
568 						}
569 					}
570 					cluster_count++;
571 				}
572 			}
573 
574 			if (cluster_count == 0) {
575 				return -EINVAL;
576 			}
577 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(*blob->active.clusters));
578 			if (tmp == NULL) {
579 				return -ENOMEM;
580 			}
581 			blob->active.clusters = tmp;
582 			blob->active.cluster_array_size = cluster_count;
583 
584 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
585 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
586 					if (desc_extent_rle->extents[i].cluster_idx != 0) {
587 						blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
588 								desc_extent_rle->extents[i].cluster_idx + j);
589 					} else if (spdk_blob_is_thin_provisioned(blob)) {
590 						blob->active.clusters[blob->active.num_clusters++] = 0;
591 					} else {
592 						return -EINVAL;
593 					}
594 				}
595 			}
596 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
597 			struct spdk_blob_md_descriptor_extent_table *desc_extent_table;
598 			uint32_t num_extent_pages = blob->active.num_extent_pages;
599 			uint32_t i, j;
600 			size_t extent_pages_length;
601 
602 			desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc;
603 			extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters);
604 
605 			if (blob->extent_rle_found) {
606 				/* This means that Extent RLE is present in MD,
607 				 * both should never be at the same time. */
608 				return -EINVAL;
609 			} else if (blob->extent_table_found &&
610 				   desc_extent_table->num_clusters != blob->remaining_clusters_in_et) {
611 				/* Number of clusters in this ET does not match number
612 				 * from previously read EXTENT_TABLE. */
613 				return -EINVAL;
614 			}
615 
616 			blob->extent_table_found = true;
617 
618 			if (desc_extent_table->length == 0 ||
619 			    (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) {
620 				return -EINVAL;
621 			}
622 
623 			for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
624 				num_extent_pages += desc_extent_table->extent_page[i].num_pages;
625 			}
626 
627 			tmp = realloc(blob->active.extent_pages, num_extent_pages * sizeof(uint32_t));
628 			if (tmp == NULL) {
629 				return -ENOMEM;
630 			}
631 			blob->active.extent_pages = tmp;
632 			blob->active.extent_pages_array_size = num_extent_pages;
633 
634 			blob->remaining_clusters_in_et = desc_extent_table->num_clusters;
635 
636 			/* Extent table entries contain md page numbers for extent pages.
637 			 * Zeroes represent unallocated extent pages, those are run-length-encoded.
638 			 */
639 			for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
640 				if (desc_extent_table->extent_page[i].page_idx != 0) {
641 					assert(desc_extent_table->extent_page[i].num_pages == 1);
642 					blob->active.extent_pages[blob->active.num_extent_pages++] =
643 						desc_extent_table->extent_page[i].page_idx;
644 				} else if (spdk_blob_is_thin_provisioned(blob)) {
645 					for (j = 0; j < desc_extent_table->extent_page[i].num_pages; j++) {
646 						blob->active.extent_pages[blob->active.num_extent_pages++] = 0;
647 					}
648 				} else {
649 					return -EINVAL;
650 				}
651 			}
652 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
653 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
654 			unsigned int					i;
655 			unsigned int					cluster_count = 0;
656 			size_t						cluster_idx_length;
657 
658 			if (blob->extent_rle_found) {
659 				/* This means that Extent RLE is present in MD,
660 				 * both should never be at the same time. */
661 				return -EINVAL;
662 			}
663 
664 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
665 			cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx);
666 
667 			if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) ||
668 			    (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) {
669 				return -EINVAL;
670 			}
671 
672 			for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
673 				if (desc_extent->cluster_idx[i] != 0) {
674 					if (!spdk_bit_array_get(blob->bs->used_clusters, desc_extent->cluster_idx[i])) {
675 						return -EINVAL;
676 					}
677 				}
678 				cluster_count++;
679 			}
680 
681 			if (cluster_count == 0) {
682 				return -EINVAL;
683 			}
684 
685 			/* When reading extent pages sequentially starting cluster idx should match
686 			 * current size of a blob.
687 			 * If changed to batch reading, this check shall be removed. */
688 			if (desc_extent->start_cluster_idx != blob->active.num_clusters) {
689 				return -EINVAL;
690 			}
691 
692 			tmp = realloc(blob->active.clusters,
693 				      (cluster_count + blob->active.num_clusters) * sizeof(*blob->active.clusters));
694 			if (tmp == NULL) {
695 				return -ENOMEM;
696 			}
697 			blob->active.clusters = tmp;
698 			blob->active.cluster_array_size = (cluster_count + blob->active.num_clusters);
699 
700 			for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
701 				if (desc_extent->cluster_idx[i] != 0) {
702 					blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
703 							desc_extent->cluster_idx[i]);
704 				} else if (spdk_blob_is_thin_provisioned(blob)) {
705 					blob->active.clusters[blob->active.num_clusters++] = 0;
706 				} else {
707 					return -EINVAL;
708 				}
709 			}
710 			assert(desc_extent->start_cluster_idx + cluster_count == blob->active.num_clusters);
711 			assert(blob->remaining_clusters_in_et >= cluster_count);
712 			blob->remaining_clusters_in_et -= cluster_count;
713 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
714 			int rc;
715 
716 			rc = _spdk_blob_deserialize_xattr(blob,
717 							  (struct spdk_blob_md_descriptor_xattr *) desc, false);
718 			if (rc != 0) {
719 				return rc;
720 			}
721 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
722 			int rc;
723 
724 			rc = _spdk_blob_deserialize_xattr(blob,
725 							  (struct spdk_blob_md_descriptor_xattr *) desc, true);
726 			if (rc != 0) {
727 				return rc;
728 			}
729 		} else {
730 			/* Unrecognized descriptor type.  Do not fail - just continue to the
731 			 *  next descriptor.  If this descriptor is associated with some feature
732 			 *  defined in a newer version of blobstore, that version of blobstore
733 			 *  should create and set an associated feature flag to specify if this
734 			 *  blob can be loaded or not.
735 			 */
736 		}
737 
738 		/* Advance to the next descriptor */
739 		cur_desc += sizeof(*desc) + desc->length;
740 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
741 			break;
742 		}
743 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
744 	}
745 
746 	return 0;
747 }
748 
749 static bool _spdk_bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page);
750 
751 static int
752 _spdk_blob_parse_extent_page(struct spdk_blob_md_page *extent_page, struct spdk_blob *blob)
753 {
754 	assert(blob != NULL);
755 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
756 
757 	if (_spdk_bs_load_cur_extent_page_valid(extent_page) == false) {
758 		return -ENOENT;
759 	}
760 
761 	return _spdk_blob_parse_page(extent_page, blob);
762 }
763 
764 static int
765 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
766 		 struct spdk_blob *blob)
767 {
768 	const struct spdk_blob_md_page *page;
769 	uint32_t i;
770 	int rc;
771 
772 	assert(page_count > 0);
773 	assert(pages[0].sequence_num == 0);
774 	assert(blob != NULL);
775 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
776 	assert(blob->active.clusters == NULL);
777 
778 	/* The blobid provided doesn't match what's in the MD, this can
779 	 * happen for example if a bogus blobid is passed in through open.
780 	 */
781 	if (blob->id != pages[0].id) {
782 		SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n",
783 			    blob->id, pages[0].id);
784 		return -ENOENT;
785 	}
786 
787 	for (i = 0; i < page_count; i++) {
788 		page = &pages[i];
789 
790 		assert(page->id == blob->id);
791 		assert(page->sequence_num == i);
792 
793 		rc = _spdk_blob_parse_page(page, blob);
794 		if (rc != 0) {
795 			return rc;
796 		}
797 	}
798 
799 	return 0;
800 }
801 
802 static int
803 _spdk_blob_serialize_add_page(const struct spdk_blob *blob,
804 			      struct spdk_blob_md_page **pages,
805 			      uint32_t *page_count,
806 			      struct spdk_blob_md_page **last_page)
807 {
808 	struct spdk_blob_md_page *page;
809 
810 	assert(pages != NULL);
811 	assert(page_count != NULL);
812 
813 	if (*page_count == 0) {
814 		assert(*pages == NULL);
815 		*page_count = 1;
816 		*pages = spdk_malloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
817 				     NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
818 	} else {
819 		assert(*pages != NULL);
820 		(*page_count)++;
821 		*pages = spdk_realloc(*pages,
822 				      SPDK_BS_PAGE_SIZE * (*page_count),
823 				      SPDK_BS_PAGE_SIZE);
824 	}
825 
826 	if (*pages == NULL) {
827 		*page_count = 0;
828 		*last_page = NULL;
829 		return -ENOMEM;
830 	}
831 
832 	page = &(*pages)[*page_count - 1];
833 	memset(page, 0, sizeof(*page));
834 	page->id = blob->id;
835 	page->sequence_num = *page_count - 1;
836 	page->next = SPDK_INVALID_MD_PAGE;
837 	*last_page = page;
838 
839 	return 0;
840 }
841 
842 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
843  * Update required_sz on both success and failure.
844  *
845  */
846 static int
847 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
848 			   uint8_t *buf, size_t buf_sz,
849 			   size_t *required_sz, bool internal)
850 {
851 	struct spdk_blob_md_descriptor_xattr	*desc;
852 
853 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
854 		       strlen(xattr->name) +
855 		       xattr->value_len;
856 
857 	if (buf_sz < *required_sz) {
858 		return -1;
859 	}
860 
861 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
862 
863 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
864 	desc->length = sizeof(desc->name_length) +
865 		       sizeof(desc->value_length) +
866 		       strlen(xattr->name) +
867 		       xattr->value_len;
868 	desc->name_length = strlen(xattr->name);
869 	desc->value_length = xattr->value_len;
870 
871 	memcpy(desc->name, xattr->name, desc->name_length);
872 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
873 	       xattr->value,
874 	       desc->value_length);
875 
876 	return 0;
877 }
878 
879 static void
880 _spdk_blob_serialize_extent_table_entry(const struct spdk_blob *blob,
881 					uint64_t start_ep, uint64_t *next_ep,
882 					uint8_t **buf, size_t *remaining_sz)
883 {
884 	struct spdk_blob_md_descriptor_extent_table *desc;
885 	size_t cur_sz;
886 	uint64_t i, et_idx;
887 	uint32_t extent_page, ep_len;
888 
889 	/* The buffer must have room for at least num_clusters entry */
890 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->num_clusters);
891 	if (*remaining_sz < cur_sz) {
892 		*next_ep = start_ep;
893 		return;
894 	}
895 
896 	desc = (struct spdk_blob_md_descriptor_extent_table *)*buf;
897 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE;
898 
899 	desc->num_clusters = blob->active.num_clusters;
900 
901 	ep_len = 1;
902 	et_idx = 0;
903 	for (i = start_ep; i < blob->active.num_extent_pages; i++) {
904 		if (*remaining_sz < cur_sz  + sizeof(desc->extent_page[0])) {
905 			/* If we ran out of buffer space, return */
906 			break;
907 		}
908 
909 		extent_page = blob->active.extent_pages[i];
910 		/* Verify that next extent_page is unallocated */
911 		if (extent_page == 0 &&
912 		    (i + 1 < blob->active.num_extent_pages && blob->active.extent_pages[i + 1] == 0)) {
913 			ep_len++;
914 			continue;
915 		}
916 		desc->extent_page[et_idx].page_idx = extent_page;
917 		desc->extent_page[et_idx].num_pages = ep_len;
918 		et_idx++;
919 
920 		ep_len = 1;
921 		cur_sz += sizeof(desc->extent_page[et_idx]);
922 	}
923 	*next_ep = i;
924 
925 	desc->length = sizeof(desc->num_clusters) + sizeof(desc->extent_page[0]) * et_idx;
926 	*remaining_sz -= sizeof(struct spdk_blob_md_descriptor) + desc->length;
927 	*buf += sizeof(struct spdk_blob_md_descriptor) + desc->length;
928 }
929 
930 static int
931 _spdk_blob_serialize_extent_table(const struct spdk_blob *blob,
932 				  struct spdk_blob_md_page **pages,
933 				  struct spdk_blob_md_page *cur_page,
934 				  uint32_t *page_count, uint8_t **buf,
935 				  size_t *remaining_sz)
936 {
937 	uint64_t				last_extent_page;
938 	int					rc;
939 
940 	last_extent_page = 0;
941 	/* At least single extent table entry has to be always persisted.
942 	 * Such case occurs with num_extent_pages == 0. */
943 	while (last_extent_page <= blob->active.num_extent_pages) {
944 		_spdk_blob_serialize_extent_table_entry(blob, last_extent_page, &last_extent_page, buf,
945 							remaining_sz);
946 
947 		if (last_extent_page == blob->active.num_extent_pages) {
948 			break;
949 		}
950 
951 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
952 		if (rc < 0) {
953 			return rc;
954 		}
955 
956 		*buf = (uint8_t *)cur_page->descriptors;
957 		*remaining_sz = sizeof(cur_page->descriptors);
958 	}
959 
960 	return 0;
961 }
962 
963 static void
964 _spdk_blob_serialize_extent_rle(const struct spdk_blob *blob,
965 				uint64_t start_cluster, uint64_t *next_cluster,
966 				uint8_t **buf, size_t *buf_sz)
967 {
968 	struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle;
969 	size_t cur_sz;
970 	uint64_t i, extent_idx;
971 	uint64_t lba, lba_per_cluster, lba_count;
972 
973 	/* The buffer must have room for at least one extent */
974 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc_extent_rle->extents[0]);
975 	if (*buf_sz < cur_sz) {
976 		*next_cluster = start_cluster;
977 		return;
978 	}
979 
980 	desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)*buf;
981 	desc_extent_rle->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE;
982 
983 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
984 
985 	lba = blob->active.clusters[start_cluster];
986 	lba_count = lba_per_cluster;
987 	extent_idx = 0;
988 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
989 		if ((lba + lba_count) == blob->active.clusters[i] && lba != 0) {
990 			/* Run-length encode sequential non-zero LBA */
991 			lba_count += lba_per_cluster;
992 			continue;
993 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
994 			/* Run-length encode unallocated clusters */
995 			lba_count += lba_per_cluster;
996 			continue;
997 		}
998 		desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
999 		desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
1000 		extent_idx++;
1001 
1002 		cur_sz += sizeof(desc_extent_rle->extents[extent_idx]);
1003 
1004 		if (*buf_sz < cur_sz) {
1005 			/* If we ran out of buffer space, return */
1006 			*next_cluster = i;
1007 			break;
1008 		}
1009 
1010 		lba = blob->active.clusters[i];
1011 		lba_count = lba_per_cluster;
1012 	}
1013 
1014 	if (*buf_sz >= cur_sz) {
1015 		desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
1016 		desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster;
1017 		extent_idx++;
1018 
1019 		*next_cluster = blob->active.num_clusters;
1020 	}
1021 
1022 	desc_extent_rle->length = sizeof(desc_extent_rle->extents[0]) * extent_idx;
1023 	*buf_sz -= sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
1024 	*buf += sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length;
1025 }
1026 
1027 static int
1028 _spdk_blob_serialize_extents_rle(const struct spdk_blob *blob,
1029 				 struct spdk_blob_md_page **pages,
1030 				 struct spdk_blob_md_page *cur_page,
1031 				 uint32_t *page_count, uint8_t **buf,
1032 				 size_t *remaining_sz)
1033 {
1034 	uint64_t				last_cluster;
1035 	int					rc;
1036 
1037 	last_cluster = 0;
1038 	while (last_cluster < blob->active.num_clusters) {
1039 		_spdk_blob_serialize_extent_rle(blob, last_cluster, &last_cluster, buf, remaining_sz);
1040 
1041 		if (last_cluster == blob->active.num_clusters) {
1042 			break;
1043 		}
1044 
1045 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
1046 		if (rc < 0) {
1047 			return rc;
1048 		}
1049 
1050 		*buf = (uint8_t *)cur_page->descriptors;
1051 		*remaining_sz = sizeof(cur_page->descriptors);
1052 	}
1053 
1054 	return 0;
1055 }
1056 
1057 static void
1058 _spdk_blob_serialize_extent_page(const struct spdk_blob *blob,
1059 				 uint64_t cluster, struct spdk_blob_md_page *page)
1060 {
1061 	struct spdk_blob_md_descriptor_extent_page *desc_extent;
1062 	uint64_t i, extent_idx;
1063 	uint64_t lba, lba_per_cluster;
1064 	uint64_t start_cluster_idx = (cluster / SPDK_EXTENTS_PER_EP) * SPDK_EXTENTS_PER_EP;
1065 
1066 	desc_extent = (struct spdk_blob_md_descriptor_extent_page *) page->descriptors;
1067 	desc_extent->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE;
1068 
1069 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
1070 
1071 	desc_extent->start_cluster_idx = start_cluster_idx;
1072 	extent_idx = 0;
1073 	for (i = start_cluster_idx; i < blob->active.num_clusters; i++) {
1074 		lba = blob->active.clusters[i];
1075 		desc_extent->cluster_idx[extent_idx++] = lba / lba_per_cluster;
1076 		if (extent_idx >= SPDK_EXTENTS_PER_EP) {
1077 			break;
1078 		}
1079 	}
1080 	desc_extent->length = sizeof(desc_extent->start_cluster_idx) +
1081 			      sizeof(desc_extent->cluster_idx[0]) * extent_idx;
1082 }
1083 
1084 static void
1085 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
1086 			   uint8_t *buf, size_t *buf_sz)
1087 {
1088 	struct spdk_blob_md_descriptor_flags *desc;
1089 
1090 	/*
1091 	 * Flags get serialized first, so we should always have room for the flags
1092 	 *  descriptor.
1093 	 */
1094 	assert(*buf_sz >= sizeof(*desc));
1095 
1096 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
1097 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
1098 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
1099 	desc->invalid_flags = blob->invalid_flags;
1100 	desc->data_ro_flags = blob->data_ro_flags;
1101 	desc->md_ro_flags = blob->md_ro_flags;
1102 
1103 	*buf_sz -= sizeof(*desc);
1104 }
1105 
1106 static int
1107 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
1108 			    const struct spdk_xattr_tailq *xattrs, bool internal,
1109 			    struct spdk_blob_md_page **pages,
1110 			    struct spdk_blob_md_page *cur_page,
1111 			    uint32_t *page_count, uint8_t **buf,
1112 			    size_t *remaining_sz)
1113 {
1114 	const struct spdk_xattr	*xattr;
1115 	int	rc;
1116 
1117 	TAILQ_FOREACH(xattr, xattrs, link) {
1118 		size_t required_sz = 0;
1119 
1120 		rc = _spdk_blob_serialize_xattr(xattr,
1121 						*buf, *remaining_sz,
1122 						&required_sz, internal);
1123 		if (rc < 0) {
1124 			/* Need to add a new page to the chain */
1125 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
1126 							   &cur_page);
1127 			if (rc < 0) {
1128 				spdk_free(*pages);
1129 				*pages = NULL;
1130 				*page_count = 0;
1131 				return rc;
1132 			}
1133 
1134 			*buf = (uint8_t *)cur_page->descriptors;
1135 			*remaining_sz = sizeof(cur_page->descriptors);
1136 
1137 			/* Try again */
1138 			required_sz = 0;
1139 			rc = _spdk_blob_serialize_xattr(xattr,
1140 							*buf, *remaining_sz,
1141 							&required_sz, internal);
1142 
1143 			if (rc < 0) {
1144 				spdk_free(*pages);
1145 				*pages = NULL;
1146 				*page_count = 0;
1147 				return rc;
1148 			}
1149 		}
1150 
1151 		*remaining_sz -= required_sz;
1152 		*buf += required_sz;
1153 	}
1154 
1155 	return 0;
1156 }
1157 
1158 static int
1159 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
1160 		     uint32_t *page_count)
1161 {
1162 	struct spdk_blob_md_page		*cur_page;
1163 	int					rc;
1164 	uint8_t					*buf;
1165 	size_t					remaining_sz;
1166 
1167 	assert(pages != NULL);
1168 	assert(page_count != NULL);
1169 	assert(blob != NULL);
1170 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
1171 
1172 	*pages = NULL;
1173 	*page_count = 0;
1174 
1175 	/* A blob always has at least 1 page, even if it has no descriptors */
1176 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
1177 	if (rc < 0) {
1178 		return rc;
1179 	}
1180 
1181 	buf = (uint8_t *)cur_page->descriptors;
1182 	remaining_sz = sizeof(cur_page->descriptors);
1183 
1184 	/* Serialize flags */
1185 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
1186 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
1187 
1188 	/* Serialize xattrs */
1189 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
1190 					 pages, cur_page, page_count, &buf, &remaining_sz);
1191 	if (rc < 0) {
1192 		return rc;
1193 	}
1194 
1195 	/* Serialize internal xattrs */
1196 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
1197 					 pages, cur_page, page_count, &buf, &remaining_sz);
1198 	if (rc < 0) {
1199 		return rc;
1200 	}
1201 
1202 	if (blob->use_extent_table) {
1203 		/* Serialize extent table */
1204 		rc = _spdk_blob_serialize_extent_table(blob, pages, cur_page, page_count, &buf, &remaining_sz);
1205 	} else {
1206 		/* Serialize extents */
1207 		rc = _spdk_blob_serialize_extents_rle(blob, pages, cur_page, page_count, &buf, &remaining_sz);
1208 	}
1209 
1210 	return rc;
1211 }
1212 
1213 struct spdk_blob_load_ctx {
1214 	struct spdk_blob		*blob;
1215 
1216 	struct spdk_blob_md_page	*pages;
1217 	uint32_t			num_pages;
1218 	uint32_t			next_extent_page;
1219 	spdk_bs_sequence_t	        *seq;
1220 
1221 	spdk_bs_sequence_cpl		cb_fn;
1222 	void				*cb_arg;
1223 };
1224 
1225 static uint32_t
1226 _spdk_blob_md_page_calc_crc(void *page)
1227 {
1228 	uint32_t		crc;
1229 
1230 	crc = BLOB_CRC32C_INITIAL;
1231 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
1232 	crc ^= BLOB_CRC32C_INITIAL;
1233 
1234 	return crc;
1235 
1236 }
1237 
1238 static void
1239 _spdk_blob_load_final(void *cb_arg, int bserrno)
1240 {
1241 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1242 	struct spdk_blob		*blob = ctx->blob;
1243 
1244 	if (bserrno == 0) {
1245 		_spdk_blob_mark_clean(blob);
1246 	}
1247 
1248 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
1249 
1250 	/* Free the memory */
1251 	spdk_free(ctx->pages);
1252 	free(ctx);
1253 }
1254 
1255 static void
1256 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
1257 {
1258 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1259 	struct spdk_blob		*blob = ctx->blob;
1260 
1261 	if (bserrno == 0) {
1262 		blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
1263 		if (blob->back_bs_dev == NULL) {
1264 			bserrno = -ENOMEM;
1265 		}
1266 	}
1267 	if (bserrno != 0) {
1268 		SPDK_ERRLOG("Snapshot fail\n");
1269 	}
1270 
1271 	_spdk_blob_load_final(ctx, bserrno);
1272 }
1273 
1274 static void _spdk_blob_update_clear_method(struct spdk_blob *blob);
1275 
1276 static void
1277 _spdk_blob_load_backing_dev(void *cb_arg)
1278 {
1279 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1280 	struct spdk_blob		*blob = ctx->blob;
1281 	const void			*value;
1282 	size_t				len;
1283 	int				rc;
1284 
1285 	if (spdk_blob_is_thin_provisioned(blob)) {
1286 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
1287 		if (rc == 0) {
1288 			if (len != sizeof(spdk_blob_id)) {
1289 				_spdk_blob_load_final(ctx, -EINVAL);
1290 				return;
1291 			}
1292 			/* open snapshot blob and continue in the callback function */
1293 			blob->parent_id = *(spdk_blob_id *)value;
1294 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1295 					  _spdk_blob_load_snapshot_cpl, ctx);
1296 			return;
1297 		} else {
1298 			/* add zeroes_dev for thin provisioned blob */
1299 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1300 		}
1301 	} else {
1302 		/* standard blob */
1303 		blob->back_bs_dev = NULL;
1304 	}
1305 	_spdk_blob_load_final(ctx, 0);
1306 }
1307 
1308 static void
1309 _spdk_blob_load_cpl_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1310 {
1311 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1312 	struct spdk_blob		*blob = ctx->blob;
1313 	struct spdk_blob_md_page	*page;
1314 	uint64_t			i;
1315 	uint32_t			crc;
1316 	uint64_t			lba;
1317 	void				*tmp;
1318 	uint64_t			sz;
1319 
1320 	if (bserrno) {
1321 		SPDK_ERRLOG("Extent page read failed: %d\n", bserrno);
1322 		_spdk_blob_load_final(ctx, bserrno);
1323 		return;
1324 	}
1325 
1326 	if (ctx->pages == NULL) {
1327 		/* First iteration of this function, allocate buffer for single EXTENT_PAGE */
1328 		ctx->pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE, NULL, SPDK_ENV_SOCKET_ID_ANY,
1329 					  SPDK_MALLOC_DMA);
1330 		if (!ctx->pages) {
1331 			_spdk_blob_load_final(ctx, -ENOMEM);
1332 			return;
1333 		}
1334 		ctx->num_pages = 1;
1335 		ctx->next_extent_page = 0;
1336 	} else {
1337 		page = &ctx->pages[0];
1338 		crc = _spdk_blob_md_page_calc_crc(page);
1339 		if (crc != page->crc) {
1340 			_spdk_blob_load_final(ctx, -EINVAL);
1341 			return;
1342 		}
1343 
1344 		if (page->next != SPDK_INVALID_MD_PAGE) {
1345 			_spdk_blob_load_final(ctx, -EINVAL);
1346 			return;
1347 		}
1348 
1349 		bserrno = _spdk_blob_parse_extent_page(page, blob);
1350 		if (bserrno) {
1351 			_spdk_blob_load_final(ctx, bserrno);
1352 			return;
1353 		}
1354 	}
1355 
1356 	for (i = ctx->next_extent_page; i < blob->active.num_extent_pages; i++) {
1357 		if (blob->active.extent_pages[i] != 0) {
1358 			/* Extent page was allocated, read and parse it. */
1359 			lba = _spdk_bs_md_page_to_lba(blob->bs, blob->active.extent_pages[i]);
1360 			ctx->next_extent_page = i + 1;
1361 
1362 			spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1363 						  _spdk_bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE),
1364 						  _spdk_blob_load_cpl_extents_cpl, ctx);
1365 			return;
1366 		} else {
1367 			/* Thin provisioned blobs can point to unallocated extent pages.
1368 			 * In this case blob size should be increased by up to the amount left in remaining_clusters_in_et. */
1369 
1370 			sz = spdk_min(blob->remaining_clusters_in_et, SPDK_EXTENTS_PER_EP);
1371 			blob->active.num_clusters += sz;
1372 			blob->remaining_clusters_in_et -= sz;
1373 
1374 			assert(spdk_blob_is_thin_provisioned(blob));
1375 			assert(i + 1 < blob->active.num_extent_pages || blob->remaining_clusters_in_et == 0);
1376 
1377 			tmp = realloc(blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters));
1378 			if (tmp == NULL) {
1379 				_spdk_blob_load_final(ctx, -ENOMEM);
1380 				return;
1381 			}
1382 			memset(tmp + blob->active.cluster_array_size, 0,
1383 			       sizeof(*blob->active.clusters) * (blob->active.num_clusters - blob->active.cluster_array_size));
1384 			blob->active.clusters = tmp;
1385 			blob->active.cluster_array_size = blob->active.num_clusters;
1386 		}
1387 	}
1388 
1389 	_spdk_blob_load_backing_dev(ctx);
1390 }
1391 
1392 static void
1393 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1394 {
1395 	struct spdk_blob_load_ctx	*ctx = cb_arg;
1396 	struct spdk_blob		*blob = ctx->blob;
1397 	struct spdk_blob_md_page	*page;
1398 	int				rc;
1399 	uint32_t			crc;
1400 
1401 	if (bserrno) {
1402 		SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno);
1403 		_spdk_blob_load_final(ctx, bserrno);
1404 		return;
1405 	}
1406 
1407 	page = &ctx->pages[ctx->num_pages - 1];
1408 	crc = _spdk_blob_md_page_calc_crc(page);
1409 	if (crc != page->crc) {
1410 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
1411 		_spdk_blob_load_final(ctx, -EINVAL);
1412 		return;
1413 	}
1414 
1415 	if (page->next != SPDK_INVALID_MD_PAGE) {
1416 		uint32_t next_page = page->next;
1417 		uint64_t next_lba = _spdk_bs_md_page_to_lba(blob->bs, next_page);
1418 
1419 		/* Read the next page */
1420 		ctx->num_pages++;
1421 		ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
1422 					  sizeof(*page));
1423 		if (ctx->pages == NULL) {
1424 			_spdk_blob_load_final(ctx, -ENOMEM);
1425 			return;
1426 		}
1427 
1428 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
1429 					  next_lba,
1430 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
1431 					  _spdk_blob_load_cpl, ctx);
1432 		return;
1433 	}
1434 
1435 	/* Parse the pages */
1436 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
1437 	if (rc) {
1438 		_spdk_blob_load_final(ctx, rc);
1439 		return;
1440 	}
1441 
1442 	if (blob->extent_table_found == true) {
1443 		/* If EXTENT_TABLE was found, that means support for it should be enabled. */
1444 		assert(blob->extent_rle_found == false);
1445 		blob->use_extent_table = true;
1446 	} else {
1447 		/* If EXTENT_RLE or no extent_* descriptor was found disable support
1448 		 * for extent table. No extent_* descriptors means that blob has length of 0
1449 		 * and no extent_rle descriptors were persisted for it.
1450 		 * EXTENT_TABLE if used, is always present in metadata regardless of length. */
1451 		blob->use_extent_table = false;
1452 	}
1453 
1454 	/* Check the clear_method stored in metadata vs what may have been passed
1455 	 * via spdk_bs_open_blob_ext() and update accordingly.
1456 	 */
1457 	_spdk_blob_update_clear_method(blob);
1458 
1459 	spdk_free(ctx->pages);
1460 	ctx->pages = NULL;
1461 
1462 	if (blob->extent_table_found) {
1463 		_spdk_blob_load_cpl_extents_cpl(seq, ctx, 0);
1464 	} else {
1465 		_spdk_blob_load_backing_dev(ctx);
1466 	}
1467 }
1468 
1469 /* Load a blob from disk given a blobid */
1470 static void
1471 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1472 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1473 {
1474 	struct spdk_blob_load_ctx *ctx;
1475 	struct spdk_blob_store *bs;
1476 	uint32_t page_num;
1477 	uint64_t lba;
1478 
1479 	_spdk_blob_verify_md_op(blob);
1480 
1481 	bs = blob->bs;
1482 
1483 	ctx = calloc(1, sizeof(*ctx));
1484 	if (!ctx) {
1485 		cb_fn(seq, cb_arg, -ENOMEM);
1486 		return;
1487 	}
1488 
1489 	ctx->blob = blob;
1490 	ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE);
1491 	if (!ctx->pages) {
1492 		free(ctx);
1493 		cb_fn(seq, cb_arg, -ENOMEM);
1494 		return;
1495 	}
1496 	ctx->num_pages = 1;
1497 	ctx->cb_fn = cb_fn;
1498 	ctx->cb_arg = cb_arg;
1499 	ctx->seq = seq;
1500 
1501 	page_num = _spdk_bs_blobid_to_page(blob->id);
1502 	lba = _spdk_bs_md_page_to_lba(blob->bs, page_num);
1503 
1504 	blob->state = SPDK_BLOB_STATE_LOADING;
1505 
1506 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1507 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1508 				  _spdk_blob_load_cpl, ctx);
1509 }
1510 
1511 struct spdk_blob_persist_ctx {
1512 	struct spdk_blob		*blob;
1513 
1514 	struct spdk_bs_super_block	*super;
1515 
1516 	struct spdk_blob_md_page	*pages;
1517 	uint32_t			next_extent_page;
1518 	struct spdk_blob_md_page	*extent_page;
1519 
1520 	spdk_bs_sequence_t		*seq;
1521 	spdk_bs_sequence_cpl		cb_fn;
1522 	void				*cb_arg;
1523 };
1524 
1525 static void
1526 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba,
1527 			uint32_t lba_count)
1528 {
1529 	switch (ctx->blob->clear_method) {
1530 	case BLOB_CLEAR_WITH_DEFAULT:
1531 	case BLOB_CLEAR_WITH_UNMAP:
1532 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1533 		break;
1534 	case BLOB_CLEAR_WITH_WRITE_ZEROES:
1535 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1536 		break;
1537 	case BLOB_CLEAR_WITH_NONE:
1538 	default:
1539 		break;
1540 	}
1541 }
1542 
1543 static void
1544 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1545 {
1546 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1547 	struct spdk_blob		*blob = ctx->blob;
1548 
1549 	if (bserrno == 0) {
1550 		_spdk_blob_mark_clean(blob);
1551 	}
1552 
1553 	/* Call user callback */
1554 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1555 
1556 	/* Free the memory */
1557 	spdk_free(ctx->pages);
1558 	free(ctx);
1559 }
1560 
1561 static void
1562 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1563 {
1564 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1565 	struct spdk_blob		*blob = ctx->blob;
1566 	struct spdk_blob_store		*bs = blob->bs;
1567 	size_t				i;
1568 
1569 	/* Release all clusters that were truncated */
1570 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1571 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1572 
1573 		/* Nothing to release if it was not allocated */
1574 		if (blob->active.clusters[i] != 0) {
1575 			_spdk_bs_release_cluster(bs, cluster_num);
1576 		}
1577 	}
1578 
1579 	if (blob->active.num_clusters == 0) {
1580 		free(blob->active.clusters);
1581 		blob->active.clusters = NULL;
1582 		blob->active.cluster_array_size = 0;
1583 	} else if (blob->active.num_clusters != blob->active.cluster_array_size) {
1584 #ifndef __clang_analyzer__
1585 		void *tmp;
1586 
1587 		/* scan-build really can't figure reallocs, workaround it */
1588 		tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * blob->active.num_clusters);
1589 		assert(tmp != NULL);
1590 		blob->active.clusters = tmp;
1591 
1592 		tmp = realloc(blob->active.extent_pages, sizeof(uint32_t) * blob->active.num_extent_pages);
1593 		assert(tmp != NULL);
1594 		blob->active.extent_pages = tmp;
1595 #endif
1596 		blob->active.extent_pages_array_size = blob->active.num_extent_pages;
1597 		blob->active.cluster_array_size = blob->active.num_clusters;
1598 	}
1599 
1600 	/* TODO: Add path to persist clear extent pages. */
1601 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1602 }
1603 
1604 static void
1605 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1606 {
1607 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1608 	struct spdk_blob		*blob = ctx->blob;
1609 	struct spdk_blob_store		*bs = blob->bs;
1610 	spdk_bs_batch_t			*batch;
1611 	size_t				i;
1612 	uint64_t			lba;
1613 	uint32_t			lba_count;
1614 
1615 	/* Clusters don't move around in blobs. The list shrinks or grows
1616 	 * at the end, but no changes ever occur in the middle of the list.
1617 	 */
1618 
1619 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx);
1620 
1621 	/* Clear all clusters that were truncated */
1622 	lba = 0;
1623 	lba_count = 0;
1624 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1625 		uint64_t next_lba = blob->active.clusters[i];
1626 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1627 
1628 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1629 			/* This cluster is contiguous with the previous one. */
1630 			lba_count += next_lba_count;
1631 			continue;
1632 		}
1633 
1634 		/* This cluster is not contiguous with the previous one. */
1635 
1636 		/* If a run of LBAs previously existing, clear them now */
1637 		if (lba_count > 0) {
1638 			spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1639 		}
1640 
1641 		/* Start building the next batch */
1642 		lba = next_lba;
1643 		if (next_lba > 0) {
1644 			lba_count = next_lba_count;
1645 		} else {
1646 			lba_count = 0;
1647 		}
1648 	}
1649 
1650 	/* If we ended with a contiguous set of LBAs, clear them now */
1651 	if (lba_count > 0) {
1652 		spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count);
1653 	}
1654 
1655 	spdk_bs_batch_close(batch);
1656 }
1657 
1658 static void
1659 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1660 {
1661 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1662 	struct spdk_blob		*blob = ctx->blob;
1663 	struct spdk_blob_store		*bs = blob->bs;
1664 	size_t				i;
1665 
1666 	/* This loop starts at 1 because the first page is special and handled
1667 	 * below. The pages (except the first) are never written in place,
1668 	 * so any pages in the clean list must be zeroed.
1669 	 */
1670 	for (i = 1; i < blob->clean.num_pages; i++) {
1671 		_spdk_bs_release_md_page(bs, blob->clean.pages[i]);
1672 	}
1673 
1674 	if (blob->active.num_pages == 0) {
1675 		uint32_t page_num;
1676 
1677 		page_num = _spdk_bs_blobid_to_page(blob->id);
1678 		_spdk_bs_release_md_page(bs, page_num);
1679 	}
1680 
1681 	/* Move on to clearing clusters */
1682 	_spdk_blob_persist_clear_clusters(seq, ctx, 0);
1683 }
1684 
1685 static void
1686 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1687 {
1688 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1689 	struct spdk_blob		*blob = ctx->blob;
1690 	struct spdk_blob_store		*bs = blob->bs;
1691 	uint64_t			lba;
1692 	uint32_t			lba_count;
1693 	spdk_bs_batch_t			*batch;
1694 	size_t				i;
1695 
1696 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1697 
1698 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1699 
1700 	/* This loop starts at 1 because the first page is special and handled
1701 	 * below. The pages (except the first) are never written in place,
1702 	 * so any pages in the clean list must be zeroed.
1703 	 */
1704 	for (i = 1; i < blob->clean.num_pages; i++) {
1705 		lba = _spdk_bs_md_page_to_lba(bs, blob->clean.pages[i]);
1706 
1707 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1708 	}
1709 
1710 	/* The first page will only be zeroed if this is a delete. */
1711 	if (blob->active.num_pages == 0) {
1712 		uint32_t page_num;
1713 
1714 		/* The first page in the metadata goes where the blobid indicates */
1715 		page_num = _spdk_bs_blobid_to_page(blob->id);
1716 		lba = _spdk_bs_md_page_to_lba(bs, page_num);
1717 
1718 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1719 	}
1720 
1721 	spdk_bs_batch_close(batch);
1722 }
1723 
1724 static void
1725 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1726 {
1727 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1728 	struct spdk_blob		*blob = ctx->blob;
1729 	struct spdk_blob_store		*bs = blob->bs;
1730 	uint64_t			lba;
1731 	uint32_t			lba_count;
1732 	struct spdk_blob_md_page	*page;
1733 
1734 	if (blob->active.num_pages == 0) {
1735 		/* Move on to the next step */
1736 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1737 		return;
1738 	}
1739 
1740 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1741 
1742 	page = &ctx->pages[0];
1743 	/* The first page in the metadata goes where the blobid indicates */
1744 	lba = _spdk_bs_md_page_to_lba(bs, _spdk_bs_blobid_to_page(blob->id));
1745 
1746 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1747 				   _spdk_blob_persist_zero_pages, ctx);
1748 }
1749 
1750 static void
1751 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1752 {
1753 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1754 	struct spdk_blob		*blob = ctx->blob;
1755 	struct spdk_blob_store		*bs = blob->bs;
1756 	uint64_t			lba;
1757 	uint32_t			lba_count;
1758 	struct spdk_blob_md_page	*page;
1759 	spdk_bs_batch_t			*batch;
1760 	size_t				i;
1761 
1762 	/* Clusters don't move around in blobs. The list shrinks or grows
1763 	 * at the end, but no changes ever occur in the middle of the list.
1764 	 */
1765 
1766 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1767 
1768 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1769 
1770 	/* This starts at 1. The root page is not written until
1771 	 * all of the others are finished
1772 	 */
1773 	for (i = 1; i < blob->active.num_pages; i++) {
1774 		page = &ctx->pages[i];
1775 		assert(page->sequence_num == i);
1776 
1777 		lba = _spdk_bs_md_page_to_lba(bs, blob->active.pages[i]);
1778 
1779 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1780 	}
1781 
1782 	spdk_bs_batch_close(batch);
1783 }
1784 
1785 static int
1786 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1787 {
1788 	uint64_t	i;
1789 	uint64_t	*tmp;
1790 	uint64_t	lfc; /* lowest free cluster */
1791 	uint32_t	lfmd; /*  lowest free md page */
1792 	uint64_t	num_clusters;
1793 	uint32_t	*ep_tmp;
1794 	uint64_t	new_num_ep = 0, current_num_ep = 0;
1795 	struct spdk_blob_store *bs;
1796 
1797 	bs = blob->bs;
1798 
1799 	_spdk_blob_verify_md_op(blob);
1800 
1801 	if (blob->active.num_clusters == sz) {
1802 		return 0;
1803 	}
1804 
1805 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1806 		/* If this blob was resized to be larger, then smaller, then
1807 		 * larger without syncing, then the cluster array already
1808 		 * contains spare assigned clusters we can use.
1809 		 */
1810 		num_clusters = spdk_min(blob->active.cluster_array_size,
1811 					sz);
1812 	} else {
1813 		num_clusters = blob->active.num_clusters;
1814 	}
1815 
1816 	if (blob->use_extent_table) {
1817 		/* Round up since every cluster beyond current Extent Table size,
1818 		 * requires new extent page. */
1819 		new_num_ep = spdk_divide_round_up(sz, SPDK_EXTENTS_PER_EP);
1820 		current_num_ep = spdk_divide_round_up(num_clusters, SPDK_EXTENTS_PER_EP);
1821 	}
1822 
1823 	/* Do two passes - one to verify that we can obtain enough clusters
1824 	 * and md pages, another to actually claim them.
1825 	 */
1826 
1827 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1828 		lfc = 0;
1829 		for (i = num_clusters; i < sz; i++) {
1830 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1831 			if (lfc == UINT32_MAX) {
1832 				/* No more free clusters. Cannot satisfy the request */
1833 				return -ENOSPC;
1834 			}
1835 			lfc++;
1836 		}
1837 		lfmd = 0;
1838 		for (i = current_num_ep; i < new_num_ep ; i++) {
1839 			lfmd = spdk_bit_array_find_first_clear(blob->bs->used_md_pages, lfmd);
1840 			if (lfmd == UINT32_MAX) {
1841 				/* No more free md pages. Cannot satisfy the request */
1842 				return -ENOSPC;
1843 			}
1844 		}
1845 	}
1846 
1847 	if (sz > num_clusters) {
1848 		/* Expand the cluster array if necessary.
1849 		 * We only shrink the array when persisting.
1850 		 */
1851 		tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * sz);
1852 		if (sz > 0 && tmp == NULL) {
1853 			return -ENOMEM;
1854 		}
1855 		memset(tmp + blob->active.cluster_array_size, 0,
1856 		       sizeof(*blob->active.clusters) * (sz - blob->active.cluster_array_size));
1857 		blob->active.clusters = tmp;
1858 		blob->active.cluster_array_size = sz;
1859 
1860 		/* Expand the extents table, only if enough clusters were added */
1861 		if (new_num_ep > current_num_ep && blob->use_extent_table) {
1862 			ep_tmp = realloc(blob->active.extent_pages, sizeof(*blob->active.extent_pages) * new_num_ep);
1863 			if (new_num_ep > 0 && ep_tmp == NULL) {
1864 				return -ENOMEM;
1865 			}
1866 			memset(ep_tmp + blob->active.extent_pages_array_size, 0,
1867 			       sizeof(*blob->active.extent_pages) * (new_num_ep - blob->active.extent_pages_array_size));
1868 			blob->active.extent_pages = ep_tmp;
1869 			blob->active.extent_pages_array_size = new_num_ep;
1870 		}
1871 	}
1872 
1873 	blob->state = SPDK_BLOB_STATE_DIRTY;
1874 
1875 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1876 		lfc = 0;
1877 		lfmd = 0;
1878 		for (i = num_clusters; i < sz; i++) {
1879 			_spdk_bs_allocate_cluster(blob, i, &lfc, &lfmd, true);
1880 			lfc++;
1881 			lfmd++;
1882 		}
1883 	}
1884 
1885 	blob->active.num_clusters = sz;
1886 	blob->active.num_extent_pages = new_num_ep;
1887 
1888 	return 0;
1889 }
1890 
1891 static void
1892 _spdk_blob_persist_generate_new_md(struct spdk_blob_persist_ctx *ctx)
1893 {
1894 	spdk_bs_sequence_t *seq = ctx->seq;
1895 	struct spdk_blob *blob = ctx->blob;
1896 	struct spdk_blob_store *bs = blob->bs;
1897 	uint64_t i;
1898 	uint32_t page_num;
1899 	void *tmp;
1900 	int rc;
1901 
1902 	/* Generate the new metadata */
1903 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1904 	if (rc < 0) {
1905 		_spdk_blob_persist_complete(seq, ctx, rc);
1906 		return;
1907 	}
1908 
1909 	assert(blob->active.num_pages >= 1);
1910 
1911 	/* Resize the cache of page indices */
1912 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1913 	if (!tmp) {
1914 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1915 		return;
1916 	}
1917 	blob->active.pages = tmp;
1918 
1919 	/* Assign this metadata to pages. This requires two passes -
1920 	 * one to verify that there are enough pages and a second
1921 	 * to actually claim them. */
1922 	page_num = 0;
1923 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1924 	for (i = 1; i < blob->active.num_pages; i++) {
1925 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1926 		if (page_num == UINT32_MAX) {
1927 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1928 			return;
1929 		}
1930 		page_num++;
1931 	}
1932 
1933 	page_num = 0;
1934 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1935 	for (i = 1; i < blob->active.num_pages; i++) {
1936 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1937 		ctx->pages[i - 1].next = page_num;
1938 		/* Now that previous metadata page is complete, calculate the crc for it. */
1939 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1940 		blob->active.pages[i] = page_num;
1941 		_spdk_bs_claim_md_page(bs, page_num);
1942 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1943 		page_num++;
1944 	}
1945 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1946 	/* Start writing the metadata from last page to first */
1947 	blob->state = SPDK_BLOB_STATE_CLEAN;
1948 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1949 }
1950 
1951 static void _spdk_blob_persist_write_extent_pages(spdk_bs_sequence_t *seq, void *cb_arg,
1952 		int bserrno);
1953 
1954 static void
1955 _spdk_blob_persist_write_extent_page(uint32_t extent, uint64_t cluster_num,
1956 				     struct spdk_blob_persist_ctx *ctx)
1957 {
1958 	spdk_bs_sequence_t		*seq = ctx->seq;
1959 	uint32_t                        page_count = 0;
1960 	struct spdk_blob		*blob = ctx->blob;
1961 	int				rc;
1962 
1963 	rc = _spdk_blob_serialize_add_page(blob, &ctx->extent_page, &page_count, &ctx->extent_page);
1964 	if (rc < 0) {
1965 		assert(false);
1966 		return;
1967 	}
1968 
1969 	_spdk_blob_serialize_extent_page(blob, cluster_num, ctx->extent_page);
1970 
1971 	ctx->extent_page->crc = _spdk_blob_md_page_calc_crc(ctx->extent_page);
1972 
1973 	spdk_bs_sequence_write_dev(seq, ctx->extent_page, _spdk_bs_md_page_to_lba(blob->bs, extent),
1974 				   _spdk_bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE),
1975 				   _spdk_blob_persist_write_extent_pages, ctx);
1976 }
1977 
1978 static void
1979 _spdk_blob_persist_write_extent_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1980 {
1981 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1982 	struct spdk_blob		*blob = ctx->blob;
1983 	size_t				i;
1984 	uint32_t			extent_page_id;
1985 
1986 	if (ctx->extent_page != NULL) {
1987 		spdk_free(ctx->extent_page);
1988 		ctx->extent_page = NULL;
1989 	}
1990 
1991 	/* Only write out changed extent pages */
1992 	for (i = ctx->next_extent_page; i < blob->active.num_extent_pages; i++) {
1993 		extent_page_id = blob->active.extent_pages[i];
1994 		if (extent_page_id == 0) {
1995 			/* No Extent Page to persist */
1996 			assert(spdk_blob_is_thin_provisioned(blob));
1997 			continue;
1998 		}
1999 		/* Writing out new extent page for the first time. Either active extent pages is larger
2000 		 * than clean extent pages or there was no extent page assigned due to thin provisioning. */
2001 		if (i >= blob->clean.extent_pages_array_size || blob->clean.extent_pages[i] == 0) {
2002 			blob->state = SPDK_BLOB_STATE_DIRTY;
2003 			assert(spdk_bit_array_get(blob->bs->used_md_pages, extent_page_id));
2004 			ctx->next_extent_page = i + 1;
2005 			_spdk_blob_persist_write_extent_page(extent_page_id, i * SPDK_EXTENTS_PER_EP, ctx);
2006 			return;
2007 		}
2008 		assert(blob->clean.extent_pages[i] != 0);
2009 	}
2010 
2011 	_spdk_blob_persist_generate_new_md(ctx);
2012 }
2013 
2014 static void
2015 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
2016 {
2017 	spdk_bs_sequence_t *seq = ctx->seq;
2018 	struct spdk_blob *blob = ctx->blob;
2019 
2020 	if (blob->active.num_pages == 0) {
2021 		/* This is the signal that the blob should be deleted.
2022 		 * Immediately jump to the clean up routine. */
2023 		assert(blob->clean.num_pages > 0);
2024 		blob->state = SPDK_BLOB_STATE_CLEAN;
2025 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
2026 		return;
2027 
2028 	}
2029 
2030 	_spdk_blob_persist_write_extent_pages(seq, ctx, 0);
2031 }
2032 
2033 static void
2034 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2035 {
2036 	struct spdk_blob_persist_ctx *ctx = cb_arg;
2037 
2038 	ctx->blob->bs->clean = 0;
2039 
2040 	spdk_free(ctx->super);
2041 
2042 	_spdk_blob_persist_start(ctx);
2043 }
2044 
2045 static void
2046 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2047 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
2048 
2049 
2050 static void
2051 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2052 {
2053 	struct spdk_blob_persist_ctx *ctx = cb_arg;
2054 
2055 	ctx->super->clean = 0;
2056 	if (ctx->super->size == 0) {
2057 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
2058 	}
2059 
2060 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
2061 }
2062 
2063 
2064 /* Write a blob to disk */
2065 static void
2066 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
2067 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2068 {
2069 	struct spdk_blob_persist_ctx *ctx;
2070 
2071 	_spdk_blob_verify_md_op(blob);
2072 
2073 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
2074 		cb_fn(seq, cb_arg, 0);
2075 		return;
2076 	}
2077 
2078 	ctx = calloc(1, sizeof(*ctx));
2079 	if (!ctx) {
2080 		cb_fn(seq, cb_arg, -ENOMEM);
2081 		return;
2082 	}
2083 	ctx->blob = blob;
2084 	ctx->seq = seq;
2085 	ctx->cb_fn = cb_fn;
2086 	ctx->cb_arg = cb_arg;
2087 	ctx->next_extent_page = 0;
2088 
2089 	if (blob->bs->clean) {
2090 		ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
2091 					  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2092 		if (!ctx->super) {
2093 			cb_fn(seq, cb_arg, -ENOMEM);
2094 			free(ctx);
2095 			return;
2096 		}
2097 
2098 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
2099 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
2100 					  _spdk_blob_persist_dirty, ctx);
2101 	} else {
2102 		_spdk_blob_persist_start(ctx);
2103 	}
2104 }
2105 
2106 struct spdk_blob_copy_cluster_ctx {
2107 	struct spdk_blob *blob;
2108 	uint8_t *buf;
2109 	uint64_t page;
2110 	uint64_t new_cluster;
2111 	uint32_t new_extent_page;
2112 	spdk_bs_sequence_t *seq;
2113 };
2114 
2115 static void
2116 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
2117 {
2118 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2119 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
2120 	TAILQ_HEAD(, spdk_bs_request_set) requests;
2121 	spdk_bs_user_op_t *op;
2122 
2123 	TAILQ_INIT(&requests);
2124 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
2125 
2126 	while (!TAILQ_EMPTY(&requests)) {
2127 		op = TAILQ_FIRST(&requests);
2128 		TAILQ_REMOVE(&requests, op, link);
2129 		if (bserrno == 0) {
2130 			spdk_bs_user_op_execute(op);
2131 		} else {
2132 			spdk_bs_user_op_abort(op);
2133 		}
2134 	}
2135 
2136 	spdk_free(ctx->buf);
2137 	free(ctx);
2138 }
2139 
2140 static void
2141 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
2142 {
2143 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2144 
2145 	if (bserrno) {
2146 		if (bserrno == -EEXIST) {
2147 			/* The metadata insert failed because another thread
2148 			 * allocated the cluster first. Free our cluster
2149 			 * but continue without error. */
2150 			bserrno = 0;
2151 		}
2152 		_spdk_bs_release_cluster(ctx->blob->bs, ctx->new_cluster);
2153 		if (ctx->new_extent_page != 0) {
2154 			_spdk_bs_release_md_page(ctx->blob->bs, ctx->new_extent_page);
2155 		}
2156 	}
2157 
2158 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2159 }
2160 
2161 static void
2162 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2163 {
2164 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2165 	uint32_t cluster_number;
2166 
2167 	if (bserrno) {
2168 		/* The write failed, so jump to the final completion handler */
2169 		spdk_bs_sequence_finish(seq, bserrno);
2170 		return;
2171 	}
2172 
2173 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
2174 
2175 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
2176 					       ctx->new_extent_page, _spdk_blob_insert_cluster_cpl, ctx);
2177 }
2178 
2179 static void
2180 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2181 {
2182 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
2183 
2184 	if (bserrno != 0) {
2185 		/* The read failed, so jump to the final completion handler */
2186 		spdk_bs_sequence_finish(seq, bserrno);
2187 		return;
2188 	}
2189 
2190 	/* Write whole cluster */
2191 	spdk_bs_sequence_write_dev(seq, ctx->buf,
2192 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
2193 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
2194 				   _spdk_blob_write_copy_cpl, ctx);
2195 }
2196 
2197 static void
2198 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
2199 				   struct spdk_io_channel *_ch,
2200 				   uint64_t io_unit, spdk_bs_user_op_t *op)
2201 {
2202 	struct spdk_bs_cpl cpl;
2203 	struct spdk_bs_channel *ch;
2204 	struct spdk_blob_copy_cluster_ctx *ctx;
2205 	uint32_t cluster_start_page;
2206 	uint32_t cluster_number;
2207 	int rc;
2208 
2209 	ch = spdk_io_channel_get_ctx(_ch);
2210 
2211 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
2212 		/* There are already operations pending. Queue this user op
2213 		 * and return because it will be re-executed when the outstanding
2214 		 * cluster allocation completes. */
2215 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
2216 		return;
2217 	}
2218 
2219 	/* Round the io_unit offset down to the first page in the cluster */
2220 	cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit);
2221 
2222 	/* Calculate which index in the metadata cluster array the corresponding
2223 	 * cluster is supposed to be at. */
2224 	cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit);
2225 
2226 	ctx = calloc(1, sizeof(*ctx));
2227 	if (!ctx) {
2228 		spdk_bs_user_op_abort(op);
2229 		return;
2230 	}
2231 
2232 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
2233 
2234 	ctx->blob = blob;
2235 	ctx->page = cluster_start_page;
2236 
2237 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
2238 		ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen,
2239 				       NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
2240 		if (!ctx->buf) {
2241 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
2242 				    blob->bs->cluster_sz);
2243 			free(ctx);
2244 			spdk_bs_user_op_abort(op);
2245 			return;
2246 		}
2247 	}
2248 
2249 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, &ctx->new_extent_page,
2250 				       false);
2251 	if (rc != 0) {
2252 		spdk_free(ctx->buf);
2253 		free(ctx);
2254 		spdk_bs_user_op_abort(op);
2255 		return;
2256 	}
2257 
2258 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2259 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
2260 	cpl.u.blob_basic.cb_arg = ctx;
2261 
2262 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
2263 	if (!ctx->seq) {
2264 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
2265 		spdk_free(ctx->buf);
2266 		free(ctx);
2267 		spdk_bs_user_op_abort(op);
2268 		return;
2269 	}
2270 
2271 	/* Queue the user op to block other incoming operations */
2272 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
2273 
2274 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
2275 		/* Read cluster from backing device */
2276 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
2277 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
2278 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
2279 					     _spdk_blob_write_copy, ctx);
2280 	} else {
2281 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
2282 						       ctx->new_extent_page, _spdk_blob_insert_cluster_cpl, ctx);
2283 	}
2284 }
2285 
2286 static void
2287 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length,
2288 				       uint64_t *lba,	uint32_t *lba_count)
2289 {
2290 	*lba_count = length;
2291 
2292 	if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) {
2293 		assert(blob->back_bs_dev != NULL);
2294 		*lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit);
2295 		*lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count);
2296 	} else {
2297 		*lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit);
2298 	}
2299 }
2300 
2301 struct op_split_ctx {
2302 	struct spdk_blob *blob;
2303 	struct spdk_io_channel *channel;
2304 	uint64_t io_unit_offset;
2305 	uint64_t io_units_remaining;
2306 	void *curr_payload;
2307 	enum spdk_blob_op_type op_type;
2308 	spdk_bs_sequence_t *seq;
2309 };
2310 
2311 static void
2312 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
2313 {
2314 	struct op_split_ctx	*ctx = cb_arg;
2315 	struct spdk_blob	*blob = ctx->blob;
2316 	struct spdk_io_channel	*ch = ctx->channel;
2317 	enum spdk_blob_op_type	op_type = ctx->op_type;
2318 	uint8_t			*buf = ctx->curr_payload;
2319 	uint64_t		offset = ctx->io_unit_offset;
2320 	uint64_t		length = ctx->io_units_remaining;
2321 	uint64_t		op_length;
2322 
2323 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2324 		spdk_bs_sequence_finish(ctx->seq, bserrno);
2325 		free(ctx);
2326 		return;
2327 	}
2328 
2329 	op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob,
2330 			     offset));
2331 
2332 	/* Update length and payload for next operation */
2333 	ctx->io_units_remaining -= op_length;
2334 	ctx->io_unit_offset += op_length;
2335 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
2336 		ctx->curr_payload += op_length * blob->bs->io_unit_size;
2337 	}
2338 
2339 	switch (op_type) {
2340 	case SPDK_BLOB_READ:
2341 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
2342 				  _spdk_blob_request_submit_op_split_next, ctx);
2343 		break;
2344 	case SPDK_BLOB_WRITE:
2345 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
2346 				   _spdk_blob_request_submit_op_split_next, ctx);
2347 		break;
2348 	case SPDK_BLOB_UNMAP:
2349 		spdk_blob_io_unmap(blob, ch, offset, op_length,
2350 				   _spdk_blob_request_submit_op_split_next, ctx);
2351 		break;
2352 	case SPDK_BLOB_WRITE_ZEROES:
2353 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
2354 					  _spdk_blob_request_submit_op_split_next, ctx);
2355 		break;
2356 	case SPDK_BLOB_READV:
2357 	case SPDK_BLOB_WRITEV:
2358 		SPDK_ERRLOG("readv/write not valid\n");
2359 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
2360 		free(ctx);
2361 		break;
2362 	}
2363 }
2364 
2365 static void
2366 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
2367 				   void *payload, uint64_t offset, uint64_t length,
2368 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2369 {
2370 	struct op_split_ctx *ctx;
2371 	spdk_bs_sequence_t *seq;
2372 	struct spdk_bs_cpl cpl;
2373 
2374 	assert(blob != NULL);
2375 
2376 	ctx = calloc(1, sizeof(struct op_split_ctx));
2377 	if (ctx == NULL) {
2378 		cb_fn(cb_arg, -ENOMEM);
2379 		return;
2380 	}
2381 
2382 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2383 	cpl.u.blob_basic.cb_fn = cb_fn;
2384 	cpl.u.blob_basic.cb_arg = cb_arg;
2385 
2386 	seq = spdk_bs_sequence_start(ch, &cpl);
2387 	if (!seq) {
2388 		free(ctx);
2389 		cb_fn(cb_arg, -ENOMEM);
2390 		return;
2391 	}
2392 
2393 	ctx->blob = blob;
2394 	ctx->channel = ch;
2395 	ctx->curr_payload = payload;
2396 	ctx->io_unit_offset = offset;
2397 	ctx->io_units_remaining = length;
2398 	ctx->op_type = op_type;
2399 	ctx->seq = seq;
2400 
2401 	_spdk_blob_request_submit_op_split_next(ctx, 0);
2402 }
2403 
2404 static void
2405 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
2406 				    void *payload, uint64_t offset, uint64_t length,
2407 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2408 {
2409 	struct spdk_bs_cpl cpl;
2410 	uint64_t lba;
2411 	uint32_t lba_count;
2412 
2413 	assert(blob != NULL);
2414 
2415 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2416 	cpl.u.blob_basic.cb_fn = cb_fn;
2417 	cpl.u.blob_basic.cb_arg = cb_arg;
2418 
2419 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2420 
2421 	if (blob->frozen_refcnt) {
2422 		/* This blob I/O is frozen */
2423 		spdk_bs_user_op_t *op;
2424 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
2425 
2426 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
2427 		if (!op) {
2428 			cb_fn(cb_arg, -ENOMEM);
2429 			return;
2430 		}
2431 
2432 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2433 
2434 		return;
2435 	}
2436 
2437 	switch (op_type) {
2438 	case SPDK_BLOB_READ: {
2439 		spdk_bs_batch_t *batch;
2440 
2441 		batch = spdk_bs_batch_open(_ch, &cpl);
2442 		if (!batch) {
2443 			cb_fn(cb_arg, -ENOMEM);
2444 			return;
2445 		}
2446 
2447 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2448 			/* Read from the blob */
2449 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
2450 		} else {
2451 			/* Read from the backing block device */
2452 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
2453 		}
2454 
2455 		spdk_bs_batch_close(batch);
2456 		break;
2457 	}
2458 	case SPDK_BLOB_WRITE:
2459 	case SPDK_BLOB_WRITE_ZEROES: {
2460 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2461 			/* Write to the blob */
2462 			spdk_bs_batch_t *batch;
2463 
2464 			if (lba_count == 0) {
2465 				cb_fn(cb_arg, 0);
2466 				return;
2467 			}
2468 
2469 			batch = spdk_bs_batch_open(_ch, &cpl);
2470 			if (!batch) {
2471 				cb_fn(cb_arg, -ENOMEM);
2472 				return;
2473 			}
2474 
2475 			if (op_type == SPDK_BLOB_WRITE) {
2476 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
2477 			} else {
2478 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
2479 			}
2480 
2481 			spdk_bs_batch_close(batch);
2482 		} else {
2483 			/* Queue this operation and allocate the cluster */
2484 			spdk_bs_user_op_t *op;
2485 
2486 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
2487 			if (!op) {
2488 				cb_fn(cb_arg, -ENOMEM);
2489 				return;
2490 			}
2491 
2492 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
2493 		}
2494 		break;
2495 	}
2496 	case SPDK_BLOB_UNMAP: {
2497 		spdk_bs_batch_t *batch;
2498 
2499 		batch = spdk_bs_batch_open(_ch, &cpl);
2500 		if (!batch) {
2501 			cb_fn(cb_arg, -ENOMEM);
2502 			return;
2503 		}
2504 
2505 		if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2506 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
2507 		}
2508 
2509 		spdk_bs_batch_close(batch);
2510 		break;
2511 	}
2512 	case SPDK_BLOB_READV:
2513 	case SPDK_BLOB_WRITEV:
2514 		SPDK_ERRLOG("readv/write not valid\n");
2515 		cb_fn(cb_arg, -EINVAL);
2516 		break;
2517 	}
2518 }
2519 
2520 static void
2521 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2522 			     void *payload, uint64_t offset, uint64_t length,
2523 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
2524 {
2525 	assert(blob != NULL);
2526 
2527 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
2528 		cb_fn(cb_arg, -EPERM);
2529 		return;
2530 	}
2531 
2532 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2533 		cb_fn(cb_arg, -EINVAL);
2534 		return;
2535 	}
2536 	if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) {
2537 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
2538 						    cb_fn, cb_arg, op_type);
2539 	} else {
2540 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
2541 						   cb_fn, cb_arg, op_type);
2542 	}
2543 }
2544 
2545 struct rw_iov_ctx {
2546 	struct spdk_blob *blob;
2547 	struct spdk_io_channel *channel;
2548 	spdk_blob_op_complete cb_fn;
2549 	void *cb_arg;
2550 	bool read;
2551 	int iovcnt;
2552 	struct iovec *orig_iov;
2553 	uint64_t io_unit_offset;
2554 	uint64_t io_units_remaining;
2555 	uint64_t io_units_done;
2556 	struct iovec iov[0];
2557 };
2558 
2559 static void
2560 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2561 {
2562 	assert(cb_arg == NULL);
2563 	spdk_bs_sequence_finish(seq, bserrno);
2564 }
2565 
2566 static void
2567 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
2568 {
2569 	struct rw_iov_ctx *ctx = cb_arg;
2570 	struct spdk_blob *blob = ctx->blob;
2571 	struct iovec *iov, *orig_iov;
2572 	int iovcnt;
2573 	size_t orig_iovoff;
2574 	uint64_t io_units_count, io_units_to_boundary, io_unit_offset;
2575 	uint64_t byte_count;
2576 
2577 	if (bserrno != 0 || ctx->io_units_remaining == 0) {
2578 		ctx->cb_fn(ctx->cb_arg, bserrno);
2579 		free(ctx);
2580 		return;
2581 	}
2582 
2583 	io_unit_offset = ctx->io_unit_offset;
2584 	io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset);
2585 	io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary);
2586 	/*
2587 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2588 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2589 	 *  point to the current position in the I/O sequence.
2590 	 */
2591 	byte_count = ctx->io_units_done * blob->bs->io_unit_size;
2592 	orig_iov = &ctx->orig_iov[0];
2593 	orig_iovoff = 0;
2594 	while (byte_count > 0) {
2595 		if (byte_count >= orig_iov->iov_len) {
2596 			byte_count -= orig_iov->iov_len;
2597 			orig_iov++;
2598 		} else {
2599 			orig_iovoff = byte_count;
2600 			byte_count = 0;
2601 		}
2602 	}
2603 
2604 	/*
2605 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2606 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2607 	 */
2608 	byte_count = io_units_count * blob->bs->io_unit_size;
2609 	iov = &ctx->iov[0];
2610 	iovcnt = 0;
2611 	while (byte_count > 0) {
2612 		assert(iovcnt < ctx->iovcnt);
2613 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2614 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2615 		byte_count -= iov->iov_len;
2616 		orig_iovoff = 0;
2617 		orig_iov++;
2618 		iov++;
2619 		iovcnt++;
2620 	}
2621 
2622 	ctx->io_unit_offset += io_units_count;
2623 	ctx->io_units_remaining -= io_units_count;
2624 	ctx->io_units_done += io_units_count;
2625 	iov = &ctx->iov[0];
2626 
2627 	if (ctx->read) {
2628 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2629 				   io_units_count, _spdk_rw_iov_split_next, ctx);
2630 	} else {
2631 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset,
2632 				    io_units_count, _spdk_rw_iov_split_next, ctx);
2633 	}
2634 }
2635 
2636 static void
2637 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2638 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2639 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2640 {
2641 	struct spdk_bs_cpl	cpl;
2642 
2643 	assert(blob != NULL);
2644 
2645 	if (!read && blob->data_ro) {
2646 		cb_fn(cb_arg, -EPERM);
2647 		return;
2648 	}
2649 
2650 	if (length == 0) {
2651 		cb_fn(cb_arg, 0);
2652 		return;
2653 	}
2654 
2655 	if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) {
2656 		cb_fn(cb_arg, -EINVAL);
2657 		return;
2658 	}
2659 
2660 	/*
2661 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2662 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2663 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2664 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2665 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2666 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2667 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2668 	 *
2669 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2670 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2671 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2672 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2673 	 */
2674 	if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) {
2675 		uint32_t lba_count;
2676 		uint64_t lba;
2677 
2678 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2679 		cpl.u.blob_basic.cb_fn = cb_fn;
2680 		cpl.u.blob_basic.cb_arg = cb_arg;
2681 
2682 		if (blob->frozen_refcnt) {
2683 			/* This blob I/O is frozen */
2684 			enum spdk_blob_op_type op_type;
2685 			spdk_bs_user_op_t *op;
2686 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2687 
2688 			op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV;
2689 			op = spdk_bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length);
2690 			if (!op) {
2691 				cb_fn(cb_arg, -ENOMEM);
2692 				return;
2693 			}
2694 
2695 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2696 
2697 			return;
2698 		}
2699 
2700 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2701 
2702 		if (read) {
2703 			spdk_bs_sequence_t *seq;
2704 
2705 			seq = spdk_bs_sequence_start(_channel, &cpl);
2706 			if (!seq) {
2707 				cb_fn(cb_arg, -ENOMEM);
2708 				return;
2709 			}
2710 
2711 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2712 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2713 			} else {
2714 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2715 							      _spdk_rw_iov_done, NULL);
2716 			}
2717 		} else {
2718 			if (_spdk_bs_io_unit_is_allocated(blob, offset)) {
2719 				spdk_bs_sequence_t *seq;
2720 
2721 				seq = spdk_bs_sequence_start(_channel, &cpl);
2722 				if (!seq) {
2723 					cb_fn(cb_arg, -ENOMEM);
2724 					return;
2725 				}
2726 
2727 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2728 			} else {
2729 				/* Queue this operation and allocate the cluster */
2730 				spdk_bs_user_op_t *op;
2731 
2732 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset,
2733 							   length);
2734 				if (!op) {
2735 					cb_fn(cb_arg, -ENOMEM);
2736 					return;
2737 				}
2738 
2739 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2740 			}
2741 		}
2742 	} else {
2743 		struct rw_iov_ctx *ctx;
2744 
2745 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2746 		if (ctx == NULL) {
2747 			cb_fn(cb_arg, -ENOMEM);
2748 			return;
2749 		}
2750 
2751 		ctx->blob = blob;
2752 		ctx->channel = _channel;
2753 		ctx->cb_fn = cb_fn;
2754 		ctx->cb_arg = cb_arg;
2755 		ctx->read = read;
2756 		ctx->orig_iov = iov;
2757 		ctx->iovcnt = iovcnt;
2758 		ctx->io_unit_offset = offset;
2759 		ctx->io_units_remaining = length;
2760 		ctx->io_units_done = 0;
2761 
2762 		_spdk_rw_iov_split_next(ctx, 0);
2763 	}
2764 }
2765 
2766 static struct spdk_blob *
2767 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2768 {
2769 	struct spdk_blob *blob;
2770 
2771 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2772 		if (blob->id == blobid) {
2773 			return blob;
2774 		}
2775 	}
2776 
2777 	return NULL;
2778 }
2779 
2780 static void
2781 _spdk_blob_get_snapshot_and_clone_entries(struct spdk_blob *blob,
2782 		struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry)
2783 {
2784 	assert(blob != NULL);
2785 	*snapshot_entry = NULL;
2786 	*clone_entry = NULL;
2787 
2788 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
2789 		return;
2790 	}
2791 
2792 	TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) {
2793 		if ((*snapshot_entry)->id == blob->parent_id) {
2794 			break;
2795 		}
2796 	}
2797 
2798 	if (*snapshot_entry != NULL) {
2799 		TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) {
2800 			if ((*clone_entry)->id == blob->id) {
2801 				break;
2802 			}
2803 		}
2804 
2805 		assert(clone_entry != NULL);
2806 	}
2807 }
2808 
2809 static int
2810 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2811 {
2812 	struct spdk_blob_store		*bs = io_device;
2813 	struct spdk_bs_channel		*channel = ctx_buf;
2814 	struct spdk_bs_dev		*dev;
2815 	uint32_t			max_ops = bs->max_channel_ops;
2816 	uint32_t			i;
2817 
2818 	dev = bs->dev;
2819 
2820 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2821 	if (!channel->req_mem) {
2822 		return -1;
2823 	}
2824 
2825 	TAILQ_INIT(&channel->reqs);
2826 
2827 	for (i = 0; i < max_ops; i++) {
2828 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2829 	}
2830 
2831 	channel->bs = bs;
2832 	channel->dev = dev;
2833 	channel->dev_channel = dev->create_channel(dev);
2834 
2835 	if (!channel->dev_channel) {
2836 		SPDK_ERRLOG("Failed to create device channel.\n");
2837 		free(channel->req_mem);
2838 		return -1;
2839 	}
2840 
2841 	TAILQ_INIT(&channel->need_cluster_alloc);
2842 	TAILQ_INIT(&channel->queued_io);
2843 
2844 	return 0;
2845 }
2846 
2847 static void
2848 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2849 {
2850 	struct spdk_bs_channel *channel = ctx_buf;
2851 	spdk_bs_user_op_t *op;
2852 
2853 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2854 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2855 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2856 		spdk_bs_user_op_abort(op);
2857 	}
2858 
2859 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2860 		op = TAILQ_FIRST(&channel->queued_io);
2861 		TAILQ_REMOVE(&channel->queued_io, op, link);
2862 		spdk_bs_user_op_abort(op);
2863 	}
2864 
2865 	free(channel->req_mem);
2866 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2867 }
2868 
2869 static void
2870 _spdk_bs_dev_destroy(void *io_device)
2871 {
2872 	struct spdk_blob_store *bs = io_device;
2873 	struct spdk_blob	*blob, *blob_tmp;
2874 
2875 	bs->dev->destroy(bs->dev);
2876 
2877 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2878 		TAILQ_REMOVE(&bs->blobs, blob, link);
2879 		_spdk_blob_free(blob);
2880 	}
2881 
2882 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2883 
2884 	spdk_bit_array_free(&bs->used_blobids);
2885 	spdk_bit_array_free(&bs->used_md_pages);
2886 	spdk_bit_array_free(&bs->used_clusters);
2887 	/*
2888 	 * If this function is called for any reason except a successful unload,
2889 	 * the unload_cpl type will be NONE and this will be a nop.
2890 	 */
2891 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2892 
2893 	free(bs);
2894 }
2895 
2896 static int
2897 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2898 {
2899 	spdk_blob_id snapshot_id;
2900 	struct spdk_blob_list *snapshot_entry = NULL;
2901 	struct spdk_blob_list *clone_entry = NULL;
2902 
2903 	assert(blob != NULL);
2904 
2905 	snapshot_id = blob->parent_id;
2906 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2907 		return 0;
2908 	}
2909 
2910 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, snapshot_id);
2911 	if (snapshot_entry == NULL) {
2912 		/* Snapshot not found */
2913 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2914 		if (snapshot_entry == NULL) {
2915 			return -ENOMEM;
2916 		}
2917 		snapshot_entry->id = snapshot_id;
2918 		TAILQ_INIT(&snapshot_entry->clones);
2919 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2920 	} else {
2921 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2922 			if (clone_entry->id == blob->id) {
2923 				break;
2924 			}
2925 		}
2926 	}
2927 
2928 	if (clone_entry == NULL) {
2929 		/* Clone not found */
2930 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2931 		if (clone_entry == NULL) {
2932 			return -ENOMEM;
2933 		}
2934 		clone_entry->id = blob->id;
2935 		TAILQ_INIT(&clone_entry->clones);
2936 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2937 		snapshot_entry->clone_count++;
2938 	}
2939 
2940 	return 0;
2941 }
2942 
2943 static void
2944 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2945 {
2946 	struct spdk_blob_list *snapshot_entry = NULL;
2947 	struct spdk_blob_list *clone_entry = NULL;
2948 
2949 	_spdk_blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry);
2950 
2951 	if (snapshot_entry == NULL) {
2952 		return;
2953 	}
2954 
2955 	blob->parent_id = SPDK_BLOBID_INVALID;
2956 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2957 	free(clone_entry);
2958 
2959 	snapshot_entry->clone_count--;
2960 }
2961 
2962 static int
2963 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2964 {
2965 	struct spdk_blob_list *snapshot_entry;
2966 	struct spdk_blob_list *snapshot_entry_tmp;
2967 	struct spdk_blob_list *clone_entry;
2968 	struct spdk_blob_list *clone_entry_tmp;
2969 
2970 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2971 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2972 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2973 			free(clone_entry);
2974 		}
2975 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2976 		free(snapshot_entry);
2977 	}
2978 
2979 	return 0;
2980 }
2981 
2982 static void
2983 _spdk_bs_free(struct spdk_blob_store *bs)
2984 {
2985 	_spdk_bs_blob_list_free(bs);
2986 
2987 	spdk_bs_unregister_md_thread(bs);
2988 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2989 }
2990 
2991 void
2992 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2993 {
2994 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2995 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2996 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2997 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2998 	opts->clear_method = BS_CLEAR_WITH_UNMAP;
2999 	memset(&opts->bstype, 0, sizeof(opts->bstype));
3000 	opts->iter_cb_fn = NULL;
3001 	opts->iter_cb_arg = NULL;
3002 }
3003 
3004 static int
3005 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
3006 {
3007 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
3008 	    opts->max_channel_ops == 0) {
3009 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
3010 		return -1;
3011 	}
3012 
3013 	return 0;
3014 }
3015 
3016 static int
3017 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
3018 {
3019 	struct spdk_blob_store	*bs;
3020 	uint64_t dev_size;
3021 	int rc;
3022 
3023 	dev_size = dev->blocklen * dev->blockcnt;
3024 	if (dev_size < opts->cluster_sz) {
3025 		/* Device size cannot be smaller than cluster size of blobstore */
3026 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
3027 			     dev_size, opts->cluster_sz);
3028 		return -ENOSPC;
3029 	}
3030 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
3031 		/* Cluster size cannot be smaller than page size */
3032 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
3033 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
3034 		return -EINVAL;
3035 	}
3036 	bs = calloc(1, sizeof(struct spdk_blob_store));
3037 	if (!bs) {
3038 		return -ENOMEM;
3039 	}
3040 
3041 	TAILQ_INIT(&bs->blobs);
3042 	TAILQ_INIT(&bs->snapshots);
3043 	bs->dev = dev;
3044 	bs->md_thread = spdk_get_thread();
3045 	assert(bs->md_thread != NULL);
3046 
3047 	/*
3048 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
3049 	 *  even multiple of the cluster size.
3050 	 */
3051 	bs->cluster_sz = opts->cluster_sz;
3052 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
3053 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3054 	bs->num_free_clusters = bs->total_clusters;
3055 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
3056 	bs->io_unit_size = dev->blocklen;
3057 	if (bs->used_clusters == NULL) {
3058 		free(bs);
3059 		return -ENOMEM;
3060 	}
3061 
3062 	bs->max_channel_ops = opts->max_channel_ops;
3063 	bs->super_blob = SPDK_BLOBID_INVALID;
3064 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
3065 
3066 	/* The metadata is assumed to be at least 1 page */
3067 	bs->used_md_pages = spdk_bit_array_create(1);
3068 	bs->used_blobids = spdk_bit_array_create(0);
3069 
3070 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
3071 
3072 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
3073 				sizeof(struct spdk_bs_channel), "blobstore");
3074 	rc = spdk_bs_register_md_thread(bs);
3075 	if (rc == -1) {
3076 		spdk_io_device_unregister(bs, NULL);
3077 		pthread_mutex_destroy(&bs->used_clusters_mutex);
3078 		spdk_bit_array_free(&bs->used_blobids);
3079 		spdk_bit_array_free(&bs->used_md_pages);
3080 		spdk_bit_array_free(&bs->used_clusters);
3081 		free(bs);
3082 		/* FIXME: this is a lie but don't know how to get a proper error code here */
3083 		return -ENOMEM;
3084 	}
3085 
3086 	*_bs = bs;
3087 	return 0;
3088 }
3089 
3090 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
3091 
3092 struct spdk_bs_load_ctx {
3093 	struct spdk_blob_store		*bs;
3094 	struct spdk_bs_super_block	*super;
3095 
3096 	struct spdk_bs_md_mask		*mask;
3097 	bool				in_page_chain;
3098 	uint32_t			page_index;
3099 	uint32_t			cur_page;
3100 	struct spdk_blob_md_page	*page;
3101 
3102 	uint64_t			num_extent_pages;
3103 	uint32_t			*extent_pages;
3104 
3105 	spdk_bs_sequence_t			*seq;
3106 	spdk_blob_op_with_handle_complete	iter_cb_fn;
3107 	void					*iter_cb_arg;
3108 	struct spdk_blob			*blob;
3109 	spdk_blob_id				blobid;
3110 };
3111 
3112 static void
3113 _spdk_bs_load_ctx_fail(struct spdk_bs_load_ctx *ctx, int bserrno)
3114 {
3115 	assert(bserrno != 0);
3116 
3117 	spdk_free(ctx->super);
3118 	spdk_bs_sequence_finish(ctx->seq, bserrno);
3119 	_spdk_bs_free(ctx->bs);
3120 	free(ctx);
3121 }
3122 
3123 static void
3124 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
3125 {
3126 	uint32_t i = 0;
3127 
3128 	while (true) {
3129 		i = spdk_bit_array_find_first_set(array, i);
3130 		if (i >= mask->length) {
3131 			break;
3132 		}
3133 		mask->mask[i / 8] |= 1U << (i % 8);
3134 		i++;
3135 	}
3136 }
3137 
3138 static int
3139 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
3140 {
3141 	struct spdk_bit_array *array;
3142 	uint32_t i;
3143 
3144 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
3145 		return -ENOMEM;
3146 	}
3147 
3148 	array = *array_ptr;
3149 	for (i = 0; i < mask->length; i++) {
3150 		if (mask->mask[i / 8] & (1U << (i % 8))) {
3151 			spdk_bit_array_set(array, i);
3152 		}
3153 	}
3154 
3155 	return 0;
3156 }
3157 
3158 static void
3159 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
3160 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
3161 {
3162 	/* Update the values in the super block */
3163 	super->super_blob = bs->super_blob;
3164 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
3165 	super->crc = _spdk_blob_md_page_calc_crc(super);
3166 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
3167 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
3168 				   cb_fn, cb_arg);
3169 }
3170 
3171 static void
3172 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3173 {
3174 	struct spdk_bs_load_ctx	*ctx = arg;
3175 	uint64_t	mask_size, lba, lba_count;
3176 
3177 	/* Write out the used clusters mask */
3178 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
3179 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3180 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3181 	if (!ctx->mask) {
3182 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3183 		return;
3184 	}
3185 
3186 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
3187 	ctx->mask->length = ctx->bs->total_clusters;
3188 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
3189 
3190 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
3191 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
3192 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
3193 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3194 }
3195 
3196 static void
3197 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3198 {
3199 	struct spdk_bs_load_ctx	*ctx = arg;
3200 	uint64_t	mask_size, lba, lba_count;
3201 
3202 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
3203 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3204 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3205 	if (!ctx->mask) {
3206 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3207 		return;
3208 	}
3209 
3210 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
3211 	ctx->mask->length = ctx->super->md_len;
3212 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
3213 
3214 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
3215 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3216 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3217 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3218 }
3219 
3220 static void
3221 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
3222 {
3223 	struct spdk_bs_load_ctx	*ctx = arg;
3224 	uint64_t	mask_size, lba, lba_count;
3225 
3226 	if (ctx->super->used_blobid_mask_len == 0) {
3227 		/*
3228 		 * This is a pre-v3 on-disk format where the blobid mask does not get
3229 		 *  written to disk.
3230 		 */
3231 		cb_fn(seq, arg, 0);
3232 		return;
3233 	}
3234 
3235 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
3236 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3237 				 SPDK_MALLOC_DMA);
3238 	if (!ctx->mask) {
3239 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3240 		return;
3241 	}
3242 
3243 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
3244 	ctx->mask->length = ctx->super->md_len;
3245 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
3246 
3247 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
3248 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
3249 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
3250 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
3251 }
3252 
3253 static void
3254 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
3255 {
3256 	_spdk_blob_verify_md_op(blob);
3257 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
3258 	blob->state = SPDK_BLOB_STATE_DIRTY;
3259 }
3260 
3261 static void
3262 _spdk_blob_set_clear_method(struct spdk_blob *blob, enum blob_clear_method clear_method)
3263 {
3264 	_spdk_blob_verify_md_op(blob);
3265 	blob->clear_method = clear_method;
3266 	blob->md_ro_flags |= (clear_method << SPDK_BLOB_CLEAR_METHOD_SHIFT);
3267 	blob->state = SPDK_BLOB_STATE_DIRTY;
3268 }
3269 
3270 static void _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno);
3271 
3272 static void
3273 _spdk_bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno)
3274 {
3275 	struct spdk_bs_load_ctx *ctx = cb_arg;
3276 	spdk_blob_id id;
3277 	int64_t page_num;
3278 
3279 	/* Iterate to next blob (we can't use spdk_bs_iter_next function as our
3280 	 * last blob has been removed */
3281 	page_num = _spdk_bs_blobid_to_page(ctx->blobid);
3282 	page_num++;
3283 	page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num);
3284 	if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) {
3285 		_spdk_bs_load_iter(ctx, NULL, -ENOENT);
3286 		return;
3287 	}
3288 
3289 	id = _spdk_bs_page_to_blobid(page_num);
3290 
3291 	spdk_bs_open_blob(ctx->bs, id, _spdk_bs_load_iter, ctx);
3292 }
3293 
3294 static void
3295 _spdk_bs_delete_corrupted_close_cb(void *cb_arg, int bserrno)
3296 {
3297 	struct spdk_bs_load_ctx *ctx = cb_arg;
3298 
3299 	if (bserrno != 0) {
3300 		SPDK_ERRLOG("Failed to close corrupted blob\n");
3301 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
3302 		return;
3303 	}
3304 
3305 	spdk_bs_delete_blob(ctx->bs, ctx->blobid, _spdk_bs_delete_corrupted_blob_cpl, ctx);
3306 }
3307 
3308 static void
3309 _spdk_bs_delete_corrupted_blob(void *cb_arg, int bserrno)
3310 {
3311 	struct spdk_bs_load_ctx *ctx = cb_arg;
3312 	uint64_t i;
3313 
3314 	if (bserrno != 0) {
3315 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
3316 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
3317 		return;
3318 	}
3319 
3320 	/* Snapshot and clone have the same copy of cluster map and extent pages
3321 	 * at this point. Let's clear both for snpashot now,
3322 	 * so that it won't be cleared for clone later when we remove snapshot.
3323 	 * Also set thin provision to pass data corruption check */
3324 	for (i = 0; i < ctx->blob->active.num_clusters; i++) {
3325 		ctx->blob->active.clusters[i] = 0;
3326 	}
3327 	for (i = 0; i < ctx->blob->active.num_extent_pages; i++) {
3328 		ctx->blob->active.extent_pages[i] = 0;
3329 	}
3330 
3331 	ctx->blob->md_ro = false;
3332 
3333 	_spdk_blob_set_thin_provision(ctx->blob);
3334 
3335 	ctx->blobid = ctx->blob->id;
3336 
3337 	spdk_blob_close(ctx->blob, _spdk_bs_delete_corrupted_close_cb, ctx);
3338 }
3339 
3340 static void
3341 _spdk_bs_update_corrupted_blob(void *cb_arg, int bserrno)
3342 {
3343 	struct spdk_bs_load_ctx *ctx = cb_arg;
3344 
3345 	if (bserrno != 0) {
3346 		SPDK_ERRLOG("Failed to close clone of a corrupted blob\n");
3347 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
3348 		return;
3349 	}
3350 
3351 	ctx->blob->md_ro = false;
3352 	_spdk_blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true);
3353 	_spdk_blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true);
3354 	spdk_blob_set_read_only(ctx->blob);
3355 
3356 	if (ctx->iter_cb_fn) {
3357 		ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0);
3358 	}
3359 	_spdk_bs_blob_list_add(ctx->blob);
3360 
3361 	spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
3362 }
3363 
3364 static void
3365 _spdk_bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno)
3366 {
3367 	struct spdk_bs_load_ctx *ctx = cb_arg;
3368 
3369 	if (bserrno != 0) {
3370 		SPDK_ERRLOG("Failed to open clone of a corrupted blob\n");
3371 		spdk_bs_iter_next(ctx->bs, ctx->blob, _spdk_bs_load_iter, ctx);
3372 		return;
3373 	}
3374 
3375 	if (blob->parent_id == ctx->blob->id) {
3376 		/* Power failure occured before updating clone (snapshot delete case)
3377 		 * or after updating clone (creating snapshot case) - keep snapshot */
3378 		spdk_blob_close(blob, _spdk_bs_update_corrupted_blob, ctx);
3379 	} else {
3380 		/* Power failure occured after updating clone (snapshot delete case)
3381 		 * or before updating clone (creating snapshot case) - remove snapshot */
3382 		spdk_blob_close(blob, _spdk_bs_delete_corrupted_blob, ctx);
3383 	}
3384 }
3385 
3386 static void
3387 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
3388 {
3389 	struct spdk_bs_load_ctx *ctx = arg;
3390 	const void *value;
3391 	size_t len;
3392 	int rc = 0;
3393 
3394 	if (bserrno == 0) {
3395 		/* Examine blob if it is corrupted after power failure. Fix
3396 		 * the ones that can be fixed and remove any other corrupted
3397 		 * ones. If it is not corrupted just process it */
3398 		rc = _spdk_blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true);
3399 		if (rc != 0) {
3400 			rc = _spdk_blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true);
3401 			if (rc != 0) {
3402 				/* Not corrupted - process it and continue with iterating through blobs */
3403 				if (ctx->iter_cb_fn) {
3404 					ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
3405 				}
3406 				_spdk_bs_blob_list_add(blob);
3407 				spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
3408 				return;
3409 			}
3410 
3411 		}
3412 
3413 		assert(len == sizeof(spdk_blob_id));
3414 
3415 		ctx->blob = blob;
3416 
3417 		/* Open clone to check if we are able to fix this blob or should we remove it */
3418 		spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, _spdk_bs_examine_clone, ctx);
3419 		return;
3420 	} else if (bserrno == -ENOENT) {
3421 		bserrno = 0;
3422 	} else {
3423 		/*
3424 		 * This case needs to be looked at further.  Same problem
3425 		 *  exists with applications that rely on explicit blob
3426 		 *  iteration.  We should just skip the blob that failed
3427 		 *  to load and continue on to the next one.
3428 		 */
3429 		SPDK_ERRLOG("Error in iterating blobs\n");
3430 	}
3431 
3432 	ctx->iter_cb_fn = NULL;
3433 
3434 	spdk_free(ctx->super);
3435 	spdk_free(ctx->mask);
3436 	spdk_bs_sequence_finish(ctx->seq, bserrno);
3437 	free(ctx);
3438 }
3439 
3440 static void
3441 _spdk_bs_load_complete(struct spdk_bs_load_ctx *ctx)
3442 {
3443 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
3444 }
3445 
3446 static void
3447 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3448 {
3449 	struct spdk_bs_load_ctx *ctx = cb_arg;
3450 	int rc;
3451 
3452 	/* The type must be correct */
3453 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
3454 
3455 	/* The length of the mask (in bits) must not be greater than
3456 	 * the length of the buffer (converted to bits) */
3457 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
3458 
3459 	/* The length of the mask must be exactly equal to the size
3460 	 * (in pages) of the metadata region */
3461 	assert(ctx->mask->length == ctx->super->md_len);
3462 
3463 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
3464 	if (rc < 0) {
3465 		spdk_free(ctx->mask);
3466 		_spdk_bs_load_ctx_fail(ctx, rc);
3467 		return;
3468 	}
3469 
3470 	_spdk_bs_load_complete(ctx);
3471 }
3472 
3473 static void
3474 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3475 {
3476 	struct spdk_bs_load_ctx *ctx = cb_arg;
3477 	uint64_t		lba, lba_count, mask_size;
3478 	int			rc;
3479 
3480 	if (bserrno != 0) {
3481 		_spdk_bs_load_ctx_fail(ctx, bserrno);
3482 		return;
3483 	}
3484 
3485 	/* The type must be correct */
3486 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
3487 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
3488 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
3489 					     struct spdk_blob_md_page) * 8));
3490 	/* The length of the mask must be exactly equal to the total number of clusters */
3491 	assert(ctx->mask->length == ctx->bs->total_clusters);
3492 
3493 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
3494 	if (rc < 0) {
3495 		spdk_free(ctx->mask);
3496 		_spdk_bs_load_ctx_fail(ctx, rc);
3497 		return;
3498 	}
3499 
3500 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
3501 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
3502 
3503 	spdk_free(ctx->mask);
3504 
3505 	/* Read the used blobids mask */
3506 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
3507 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3508 				 SPDK_MALLOC_DMA);
3509 	if (!ctx->mask) {
3510 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3511 		return;
3512 	}
3513 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
3514 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
3515 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3516 				  _spdk_bs_load_used_blobids_cpl, ctx);
3517 }
3518 
3519 static void
3520 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3521 {
3522 	struct spdk_bs_load_ctx *ctx = cb_arg;
3523 	uint64_t		lba, lba_count, mask_size;
3524 	int			rc;
3525 
3526 	if (bserrno != 0) {
3527 		_spdk_bs_load_ctx_fail(ctx, bserrno);
3528 		return;
3529 	}
3530 
3531 	/* The type must be correct */
3532 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
3533 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
3534 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
3535 				     8));
3536 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
3537 	assert(ctx->mask->length == ctx->super->md_len);
3538 
3539 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
3540 	if (rc < 0) {
3541 		spdk_free(ctx->mask);
3542 		_spdk_bs_load_ctx_fail(ctx, rc);
3543 		return;
3544 	}
3545 
3546 	spdk_free(ctx->mask);
3547 
3548 	/* Read the used clusters mask */
3549 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
3550 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY,
3551 				 SPDK_MALLOC_DMA);
3552 	if (!ctx->mask) {
3553 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3554 		return;
3555 	}
3556 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
3557 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
3558 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
3559 				  _spdk_bs_load_used_clusters_cpl, ctx);
3560 }
3561 
3562 static void
3563 _spdk_bs_load_read_used_pages(struct spdk_bs_load_ctx *ctx)
3564 {
3565 	uint64_t lba, lba_count, mask_size;
3566 
3567 	/* Read the used pages mask */
3568 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
3569 	ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL,
3570 				 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3571 	if (!ctx->mask) {
3572 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3573 		return;
3574 	}
3575 
3576 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
3577 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
3578 	spdk_bs_sequence_read_dev(ctx->seq, ctx->mask, lba, lba_count,
3579 				  _spdk_bs_load_used_pages_cpl, ctx);
3580 }
3581 
3582 static int
3583 _spdk_bs_load_replay_md_parse_page(struct spdk_bs_load_ctx *ctx)
3584 {
3585 	struct spdk_blob_store *bs = ctx->bs;
3586 	struct spdk_blob_md_page *page = ctx->page;
3587 	struct spdk_blob_md_descriptor *desc;
3588 	size_t	cur_desc = 0;
3589 
3590 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3591 	while (cur_desc < sizeof(page->descriptors)) {
3592 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3593 			if (desc->length == 0) {
3594 				/* If padding and length are 0, this terminates the page */
3595 				break;
3596 			}
3597 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
3598 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
3599 			unsigned int				i, j;
3600 			unsigned int				cluster_count = 0;
3601 			uint32_t				cluster_idx;
3602 
3603 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
3604 
3605 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
3606 				for (j = 0; j < desc_extent_rle->extents[i].length; j++) {
3607 					cluster_idx = desc_extent_rle->extents[i].cluster_idx;
3608 					/*
3609 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
3610 					 * in the used cluster map.
3611 					 */
3612 					if (cluster_idx != 0) {
3613 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
3614 						if (bs->num_free_clusters == 0) {
3615 							return -ENOSPC;
3616 						}
3617 						bs->num_free_clusters--;
3618 					}
3619 					cluster_count++;
3620 				}
3621 			}
3622 			if (cluster_count == 0) {
3623 				return -EINVAL;
3624 			}
3625 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
3626 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
3627 			uint32_t					i;
3628 			uint32_t					cluster_count = 0;
3629 			uint32_t					cluster_idx;
3630 			size_t						cluster_idx_length;
3631 
3632 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
3633 			cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx);
3634 
3635 			if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) ||
3636 			    (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) {
3637 				return -EINVAL;
3638 			}
3639 
3640 			for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) {
3641 				cluster_idx = desc_extent->cluster_idx[i];
3642 				/*
3643 				 * cluster_idx = 0 means an unallocated cluster - don't mark that
3644 				 * in the used cluster map.
3645 				 */
3646 				if (cluster_idx != 0) {
3647 					if (cluster_idx < desc_extent->start_cluster_idx &&
3648 					    cluster_idx >= desc_extent->start_cluster_idx + cluster_count) {
3649 						return -EINVAL;
3650 					}
3651 					spdk_bit_array_set(bs->used_clusters, cluster_idx);
3652 					if (bs->num_free_clusters == 0) {
3653 						return -ENOSPC;
3654 					}
3655 					bs->num_free_clusters--;
3656 				}
3657 				cluster_count++;
3658 			}
3659 
3660 			if (cluster_count == 0) {
3661 				return -EINVAL;
3662 			}
3663 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3664 			/* Skip this item */
3665 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3666 			/* Skip this item */
3667 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3668 			/* Skip this item */
3669 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) {
3670 			struct spdk_blob_md_descriptor_extent_table *desc_extent_table;
3671 			uint32_t num_extent_pages = ctx->num_extent_pages;
3672 			uint32_t i;
3673 			size_t extent_pages_length;
3674 			void *tmp;
3675 
3676 			desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc;
3677 			extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters);
3678 
3679 			if (desc_extent_table->length == 0 ||
3680 			    (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) {
3681 				return -EINVAL;
3682 			}
3683 
3684 			for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
3685 				if (desc_extent_table->extent_page[i].page_idx != 0) {
3686 					if (desc_extent_table->extent_page[i].num_pages != 1) {
3687 						return -EINVAL;
3688 					}
3689 					num_extent_pages += 1;
3690 				}
3691 			}
3692 
3693 			if (num_extent_pages > 0) {
3694 				tmp = realloc(ctx->extent_pages, num_extent_pages * sizeof(uint32_t));
3695 				if (tmp == NULL) {
3696 					return -ENOMEM;
3697 				}
3698 				ctx->extent_pages = tmp;
3699 
3700 				/* Extent table entries contain md page numbers for extent pages.
3701 				 * Zeroes represent unallocated extent pages, those are run-length-encoded.
3702 				 */
3703 				for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) {
3704 					if (desc_extent_table->extent_page[i].page_idx != 0) {
3705 						ctx->extent_pages[ctx->num_extent_pages] = desc_extent_table->extent_page[i].page_idx;
3706 						ctx->num_extent_pages += 1;
3707 					}
3708 				}
3709 			}
3710 		} else {
3711 			/* Error */
3712 			return -EINVAL;
3713 		}
3714 		/* Advance to the next descriptor */
3715 		cur_desc += sizeof(*desc) + desc->length;
3716 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3717 			break;
3718 		}
3719 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3720 	}
3721 	return 0;
3722 }
3723 
3724 static bool _spdk_bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page)
3725 {
3726 	uint32_t crc;
3727 	struct spdk_blob_md_descriptor *desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3728 	size_t desc_len;
3729 
3730 	crc = _spdk_blob_md_page_calc_crc(page);
3731 	if (crc != page->crc) {
3732 		return false;
3733 	}
3734 
3735 	/* Extent page should always be of sequence num 0. */
3736 	if (page->sequence_num != 0) {
3737 		return false;
3738 	}
3739 
3740 	/* Descriptor type must be EXTENT_PAGE. */
3741 	if (desc->type != SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
3742 		return false;
3743 	}
3744 
3745 	/* Descriptor length cannot exceed the page. */
3746 	desc_len = sizeof(*desc) + desc->length;
3747 	if (desc_len > sizeof(page->descriptors)) {
3748 		return false;
3749 	}
3750 
3751 	/* It has to be the only descriptor in the page. */
3752 	if (desc_len + sizeof(*desc) <= sizeof(page->descriptors)) {
3753 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + desc_len);
3754 		if (desc->length != 0) {
3755 			return false;
3756 		}
3757 	}
3758 
3759 	return true;
3760 }
3761 
3762 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
3763 {
3764 	uint32_t crc;
3765 
3766 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
3767 	if (crc != ctx->page->crc) {
3768 		return false;
3769 	}
3770 
3771 	/* First page of a sequence should match the blobid. */
3772 	if (ctx->page->sequence_num == 0 &&
3773 	    _spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
3774 		return false;
3775 	}
3776 	return true;
3777 }
3778 
3779 static void
3780 _spdk_bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx);
3781 
3782 static void
3783 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3784 {
3785 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3786 
3787 	if (bserrno != 0) {
3788 		_spdk_bs_load_ctx_fail(ctx, bserrno);
3789 		return;
3790 	}
3791 
3792 	_spdk_bs_load_complete(ctx);
3793 }
3794 
3795 static void
3796 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3797 {
3798 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3799 
3800 	spdk_free(ctx->mask);
3801 	ctx->mask = NULL;
3802 
3803 	if (bserrno != 0) {
3804 		_spdk_bs_load_ctx_fail(ctx, bserrno);
3805 		return;
3806 	}
3807 
3808 	_spdk_bs_write_used_clusters(seq, ctx, _spdk_bs_load_write_used_clusters_cpl);
3809 }
3810 
3811 static void
3812 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3813 {
3814 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3815 
3816 	spdk_free(ctx->mask);
3817 	ctx->mask = NULL;
3818 
3819 	if (bserrno != 0) {
3820 		_spdk_bs_load_ctx_fail(ctx, bserrno);
3821 		return;
3822 	}
3823 
3824 	_spdk_bs_write_used_blobids(seq, ctx, _spdk_bs_load_write_used_blobids_cpl);
3825 }
3826 
3827 static void
3828 _spdk_bs_load_write_used_md(struct spdk_bs_load_ctx *ctx)
3829 {
3830 	_spdk_bs_write_used_md(ctx->seq, ctx, _spdk_bs_load_write_used_pages_cpl);
3831 }
3832 
3833 static void
3834 _spdk_bs_load_replay_md_chain_cpl(struct spdk_bs_load_ctx *ctx)
3835 {
3836 	uint64_t num_md_clusters;
3837 	uint64_t i;
3838 
3839 	ctx->in_page_chain = false;
3840 
3841 	do {
3842 		ctx->page_index++;
3843 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
3844 
3845 	if (ctx->page_index < ctx->super->md_len) {
3846 		ctx->cur_page = ctx->page_index;
3847 		_spdk_bs_load_replay_cur_md_page(ctx);
3848 	} else {
3849 		/* Claim all of the clusters used by the metadata */
3850 		num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
3851 		for (i = 0; i < num_md_clusters; i++) {
3852 			_spdk_bs_claim_cluster(ctx->bs, i);
3853 		}
3854 		spdk_free(ctx->page);
3855 		_spdk_bs_load_write_used_md(ctx);
3856 	}
3857 }
3858 
3859 static void _spdk_bs_load_replay_extent_page(spdk_bs_sequence_t *seq, uint32_t page, void *cb_arg);
3860 
3861 static void
3862 _spdk_bs_load_replay_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3863 {
3864 	struct spdk_bs_load_ctx *ctx = cb_arg;
3865 	uint32_t page_num;
3866 
3867 	if (bserrno != 0) {
3868 		_spdk_bs_load_ctx_fail(ctx, bserrno);
3869 		return;
3870 	}
3871 
3872 	/* Extent pages are only read when present within in chain md.
3873 	 * Integrity of md is not right if that page was not a valid extent page. */
3874 	if (_spdk_bs_load_cur_extent_page_valid(ctx->page) != true) {
3875 		_spdk_bs_load_ctx_fail(ctx, -EILSEQ);
3876 		return;
3877 	}
3878 
3879 	page_num = ctx->extent_pages[ctx->num_extent_pages - 1];
3880 	spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
3881 	if (_spdk_bs_load_replay_md_parse_page(ctx)) {
3882 		_spdk_bs_load_ctx_fail(ctx, -EILSEQ);
3883 		return;
3884 	}
3885 
3886 	ctx->num_extent_pages--;
3887 	if (ctx->num_extent_pages > 0) {
3888 		_spdk_bs_load_replay_extent_page(seq, ctx->extent_pages[ctx->num_extent_pages - 1], ctx);
3889 		return;
3890 	}
3891 
3892 	free(ctx->extent_pages);
3893 	ctx->extent_pages = NULL;
3894 
3895 	_spdk_bs_load_replay_md_chain_cpl(ctx);
3896 }
3897 
3898 static void
3899 _spdk_bs_load_replay_extent_page(spdk_bs_sequence_t *seq, uint32_t page, void *cb_arg)
3900 {
3901 	struct spdk_bs_load_ctx *ctx = cb_arg;
3902 	uint64_t lba;
3903 
3904 	assert(page < ctx->super->md_len);
3905 	lba = _spdk_bs_md_page_to_lba(ctx->bs, page);
3906 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3907 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3908 				  _spdk_bs_load_replay_extent_page_cpl, ctx);
3909 }
3910 
3911 static void
3912 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3913 {
3914 	struct spdk_bs_load_ctx *ctx = cb_arg;
3915 	uint32_t page_num;
3916 
3917 	if (bserrno != 0) {
3918 		_spdk_bs_load_ctx_fail(ctx, bserrno);
3919 		return;
3920 	}
3921 
3922 	page_num = ctx->cur_page;
3923 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
3924 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
3925 			_spdk_bs_claim_md_page(ctx->bs, page_num);
3926 			if (ctx->page->sequence_num == 0) {
3927 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
3928 			}
3929 			if (_spdk_bs_load_replay_md_parse_page(ctx)) {
3930 				_spdk_bs_load_ctx_fail(ctx, -EILSEQ);
3931 				return;
3932 			}
3933 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
3934 				ctx->in_page_chain = true;
3935 				ctx->cur_page = ctx->page->next;
3936 				_spdk_bs_load_replay_cur_md_page(ctx);
3937 				return;
3938 			}
3939 			if (ctx->num_extent_pages != 0) {
3940 				/* Extent pages are read from last to first,
3941 				 * decreasing the num_extent_pages as they are read. */
3942 				_spdk_bs_load_replay_extent_page(seq, ctx->extent_pages[ctx->num_extent_pages - 1], ctx);
3943 				return;
3944 			}
3945 		}
3946 	}
3947 	_spdk_bs_load_replay_md_chain_cpl(ctx);
3948 }
3949 
3950 static void
3951 _spdk_bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx)
3952 {
3953 	uint64_t lba;
3954 
3955 	assert(ctx->cur_page < ctx->super->md_len);
3956 	lba = _spdk_bs_md_page_to_lba(ctx->bs, ctx->cur_page);
3957 	spdk_bs_sequence_read_dev(ctx->seq, ctx->page, lba,
3958 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3959 				  _spdk_bs_load_replay_md_cpl, ctx);
3960 }
3961 
3962 static void
3963 _spdk_bs_load_replay_md(struct spdk_bs_load_ctx *ctx)
3964 {
3965 	ctx->page_index = 0;
3966 	ctx->cur_page = 0;
3967 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
3968 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
3969 	if (!ctx->page) {
3970 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3971 		return;
3972 	}
3973 	_spdk_bs_load_replay_cur_md_page(ctx);
3974 }
3975 
3976 static void
3977 _spdk_bs_recover(struct spdk_bs_load_ctx *ctx)
3978 {
3979 	int		rc;
3980 
3981 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3982 	if (rc < 0) {
3983 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3984 		return;
3985 	}
3986 
3987 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3988 	if (rc < 0) {
3989 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3990 		return;
3991 	}
3992 
3993 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3994 	if (rc < 0) {
3995 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
3996 		return;
3997 	}
3998 
3999 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
4000 	_spdk_bs_load_replay_md(ctx);
4001 }
4002 
4003 static void
4004 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4005 {
4006 	struct spdk_bs_load_ctx *ctx = cb_arg;
4007 	uint32_t	crc;
4008 	int		rc;
4009 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
4010 
4011 	if (ctx->super->version > SPDK_BS_VERSION ||
4012 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
4013 		_spdk_bs_load_ctx_fail(ctx, -EILSEQ);
4014 		return;
4015 	}
4016 
4017 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4018 		   sizeof(ctx->super->signature)) != 0) {
4019 		_spdk_bs_load_ctx_fail(ctx, -EILSEQ);
4020 		return;
4021 	}
4022 
4023 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
4024 	if (crc != ctx->super->crc) {
4025 		_spdk_bs_load_ctx_fail(ctx, -EILSEQ);
4026 		return;
4027 	}
4028 
4029 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
4030 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
4031 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
4032 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
4033 	} else {
4034 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
4035 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
4036 		SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
4037 		_spdk_bs_load_ctx_fail(ctx, -ENXIO);
4038 		return;
4039 	}
4040 
4041 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
4042 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
4043 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
4044 		_spdk_bs_load_ctx_fail(ctx, -EILSEQ);
4045 		return;
4046 	}
4047 
4048 	if (ctx->super->size == 0) {
4049 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
4050 	}
4051 
4052 	if (ctx->super->io_unit_size == 0) {
4053 		ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE;
4054 	}
4055 
4056 	/* Parse the super block */
4057 	ctx->bs->clean = 1;
4058 	ctx->bs->cluster_sz = ctx->super->cluster_size;
4059 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
4060 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
4061 	ctx->bs->io_unit_size = ctx->super->io_unit_size;
4062 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
4063 	if (rc < 0) {
4064 		_spdk_bs_load_ctx_fail(ctx, -ENOMEM);
4065 		return;
4066 	}
4067 	ctx->bs->md_start = ctx->super->md_start;
4068 	ctx->bs->md_len = ctx->super->md_len;
4069 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up(
4070 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
4071 	ctx->bs->super_blob = ctx->super->super_blob;
4072 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
4073 
4074 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
4075 		_spdk_bs_recover(ctx);
4076 	} else {
4077 		_spdk_bs_load_read_used_pages(ctx);
4078 	}
4079 }
4080 
4081 void
4082 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
4083 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
4084 {
4085 	struct spdk_blob_store	*bs;
4086 	struct spdk_bs_cpl	cpl;
4087 	struct spdk_bs_load_ctx *ctx;
4088 	struct spdk_bs_opts	opts = {};
4089 	int err;
4090 
4091 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
4092 
4093 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
4094 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
4095 		dev->destroy(dev);
4096 		cb_fn(cb_arg, NULL, -EINVAL);
4097 		return;
4098 	}
4099 
4100 	if (o) {
4101 		opts = *o;
4102 	} else {
4103 		spdk_bs_opts_init(&opts);
4104 	}
4105 
4106 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
4107 		dev->destroy(dev);
4108 		cb_fn(cb_arg, NULL, -EINVAL);
4109 		return;
4110 	}
4111 
4112 	err = _spdk_bs_alloc(dev, &opts, &bs);
4113 	if (err) {
4114 		dev->destroy(dev);
4115 		cb_fn(cb_arg, NULL, err);
4116 		return;
4117 	}
4118 
4119 	ctx = calloc(1, sizeof(*ctx));
4120 	if (!ctx) {
4121 		_spdk_bs_free(bs);
4122 		cb_fn(cb_arg, NULL, -ENOMEM);
4123 		return;
4124 	}
4125 
4126 	ctx->bs = bs;
4127 	ctx->iter_cb_fn = opts.iter_cb_fn;
4128 	ctx->iter_cb_arg = opts.iter_cb_arg;
4129 
4130 	/* Allocate memory for the super block */
4131 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4132 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4133 	if (!ctx->super) {
4134 		free(ctx);
4135 		_spdk_bs_free(bs);
4136 		cb_fn(cb_arg, NULL, -ENOMEM);
4137 		return;
4138 	}
4139 
4140 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
4141 	cpl.u.bs_handle.cb_fn = cb_fn;
4142 	cpl.u.bs_handle.cb_arg = cb_arg;
4143 	cpl.u.bs_handle.bs = bs;
4144 
4145 	ctx->seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4146 	if (!ctx->seq) {
4147 		spdk_free(ctx->super);
4148 		free(ctx);
4149 		_spdk_bs_free(bs);
4150 		cb_fn(cb_arg, NULL, -ENOMEM);
4151 		return;
4152 	}
4153 
4154 	/* Read the super block */
4155 	spdk_bs_sequence_read_dev(ctx->seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4156 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4157 				  _spdk_bs_load_super_cpl, ctx);
4158 }
4159 
4160 /* END spdk_bs_load */
4161 
4162 /* START spdk_bs_dump */
4163 
4164 struct spdk_bs_dump_ctx {
4165 	struct spdk_blob_store		*bs;
4166 	struct spdk_bs_super_block	*super;
4167 	uint32_t			cur_page;
4168 	struct spdk_blob_md_page	*page;
4169 	spdk_bs_sequence_t		*seq;
4170 	FILE				*fp;
4171 	spdk_bs_dump_print_xattr	print_xattr_fn;
4172 	char				xattr_name[4096];
4173 };
4174 
4175 static void
4176 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
4177 {
4178 	spdk_free(ctx->super);
4179 
4180 	/*
4181 	 * We need to defer calling spdk_bs_call_cpl() until after
4182 	 * dev destruction, so tuck these away for later use.
4183 	 */
4184 	ctx->bs->unload_err = bserrno;
4185 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
4186 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
4187 
4188 	spdk_bs_sequence_finish(seq, 0);
4189 	_spdk_bs_free(ctx->bs);
4190 	free(ctx);
4191 }
4192 
4193 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
4194 
4195 static void
4196 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
4197 {
4198 	uint32_t page_idx = ctx->cur_page;
4199 	struct spdk_blob_md_page *page = ctx->page;
4200 	struct spdk_blob_md_descriptor *desc;
4201 	size_t cur_desc = 0;
4202 	uint32_t crc;
4203 
4204 	fprintf(ctx->fp, "=========\n");
4205 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
4206 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
4207 
4208 	crc = _spdk_blob_md_page_calc_crc(page);
4209 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
4210 
4211 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
4212 	while (cur_desc < sizeof(page->descriptors)) {
4213 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
4214 			if (desc->length == 0) {
4215 				/* If padding and length are 0, this terminates the page */
4216 				break;
4217 			}
4218 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) {
4219 			struct spdk_blob_md_descriptor_extent_rle	*desc_extent_rle;
4220 			unsigned int				i;
4221 
4222 			desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc;
4223 
4224 			for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) {
4225 				if (desc_extent_rle->extents[i].cluster_idx != 0) {
4226 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
4227 						desc_extent_rle->extents[i].cluster_idx);
4228 				} else {
4229 					fprintf(ctx->fp, "Unallocated Extent - ");
4230 				}
4231 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent_rle->extents[i].length);
4232 				fprintf(ctx->fp, "\n");
4233 			}
4234 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) {
4235 			struct spdk_blob_md_descriptor_extent_page	*desc_extent;
4236 			unsigned int					i;
4237 
4238 			desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc;
4239 
4240 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->cluster_idx[0]); i++) {
4241 				if (desc_extent->cluster_idx[i] != 0) {
4242 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
4243 						desc_extent->cluster_idx[i]);
4244 				} else {
4245 					fprintf(ctx->fp, "Unallocated Extent");
4246 				}
4247 				fprintf(ctx->fp, "\n");
4248 			}
4249 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
4250 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
4251 			uint32_t i;
4252 
4253 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
4254 
4255 			if (desc_xattr->length !=
4256 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
4257 			    desc_xattr->name_length + desc_xattr->value_length) {
4258 			}
4259 
4260 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
4261 			ctx->xattr_name[desc_xattr->name_length] = '\0';
4262 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
4263 			fprintf(ctx->fp, "       value = \"");
4264 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
4265 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
4266 					    desc_xattr->value_length);
4267 			fprintf(ctx->fp, "\"\n");
4268 			for (i = 0; i < desc_xattr->value_length; i++) {
4269 				if (i % 16 == 0) {
4270 					fprintf(ctx->fp, "               ");
4271 				}
4272 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
4273 				if ((i + 1) % 16 == 0) {
4274 					fprintf(ctx->fp, "\n");
4275 				}
4276 			}
4277 			if (i % 16 != 0) {
4278 				fprintf(ctx->fp, "\n");
4279 			}
4280 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
4281 			/* TODO */
4282 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
4283 			/* TODO */
4284 		} else {
4285 			/* Error */
4286 		}
4287 		/* Advance to the next descriptor */
4288 		cur_desc += sizeof(*desc) + desc->length;
4289 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
4290 			break;
4291 		}
4292 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
4293 	}
4294 }
4295 
4296 static void
4297 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4298 {
4299 	struct spdk_bs_dump_ctx *ctx = cb_arg;
4300 
4301 	if (bserrno != 0) {
4302 		_spdk_bs_dump_finish(seq, ctx, bserrno);
4303 		return;
4304 	}
4305 
4306 	if (ctx->page->id != 0) {
4307 		_spdk_bs_dump_print_md_page(ctx);
4308 	}
4309 
4310 	ctx->cur_page++;
4311 
4312 	if (ctx->cur_page < ctx->super->md_len) {
4313 		_spdk_bs_dump_read_md_page(seq, ctx);
4314 	} else {
4315 		spdk_free(ctx->page);
4316 		_spdk_bs_dump_finish(seq, ctx, 0);
4317 	}
4318 }
4319 
4320 static void
4321 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
4322 {
4323 	struct spdk_bs_dump_ctx *ctx = cb_arg;
4324 	uint64_t lba;
4325 
4326 	assert(ctx->cur_page < ctx->super->md_len);
4327 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
4328 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
4329 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
4330 				  _spdk_bs_dump_read_md_page_cpl, ctx);
4331 }
4332 
4333 static void
4334 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4335 {
4336 	struct spdk_bs_dump_ctx *ctx = cb_arg;
4337 
4338 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
4339 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4340 		   sizeof(ctx->super->signature)) != 0) {
4341 		fprintf(ctx->fp, "(Mismatch)\n");
4342 		_spdk_bs_dump_finish(seq, ctx, bserrno);
4343 		return;
4344 	} else {
4345 		fprintf(ctx->fp, "(OK)\n");
4346 	}
4347 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
4348 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
4349 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
4350 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
4351 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
4352 	fprintf(ctx->fp, "Super Blob ID: ");
4353 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
4354 		fprintf(ctx->fp, "(None)\n");
4355 	} else {
4356 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
4357 	}
4358 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
4359 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
4360 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
4361 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
4362 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
4363 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
4364 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
4365 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
4366 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
4367 
4368 	ctx->cur_page = 0;
4369 	ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE,
4370 				 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4371 	if (!ctx->page) {
4372 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
4373 		return;
4374 	}
4375 	_spdk_bs_dump_read_md_page(seq, ctx);
4376 }
4377 
4378 void
4379 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
4380 	     spdk_bs_op_complete cb_fn, void *cb_arg)
4381 {
4382 	struct spdk_blob_store	*bs;
4383 	struct spdk_bs_cpl	cpl;
4384 	spdk_bs_sequence_t	*seq;
4385 	struct spdk_bs_dump_ctx *ctx;
4386 	struct spdk_bs_opts	opts = {};
4387 	int err;
4388 
4389 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
4390 
4391 	spdk_bs_opts_init(&opts);
4392 
4393 	err = _spdk_bs_alloc(dev, &opts, &bs);
4394 	if (err) {
4395 		dev->destroy(dev);
4396 		cb_fn(cb_arg, err);
4397 		return;
4398 	}
4399 
4400 	ctx = calloc(1, sizeof(*ctx));
4401 	if (!ctx) {
4402 		_spdk_bs_free(bs);
4403 		cb_fn(cb_arg, -ENOMEM);
4404 		return;
4405 	}
4406 
4407 	ctx->bs = bs;
4408 	ctx->fp = fp;
4409 	ctx->print_xattr_fn = print_xattr_fn;
4410 
4411 	/* Allocate memory for the super block */
4412 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4413 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4414 	if (!ctx->super) {
4415 		free(ctx);
4416 		_spdk_bs_free(bs);
4417 		cb_fn(cb_arg, -ENOMEM);
4418 		return;
4419 	}
4420 
4421 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4422 	cpl.u.bs_basic.cb_fn = cb_fn;
4423 	cpl.u.bs_basic.cb_arg = cb_arg;
4424 
4425 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4426 	if (!seq) {
4427 		spdk_free(ctx->super);
4428 		free(ctx);
4429 		_spdk_bs_free(bs);
4430 		cb_fn(cb_arg, -ENOMEM);
4431 		return;
4432 	}
4433 
4434 	/* Read the super block */
4435 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4436 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4437 				  _spdk_bs_dump_super_cpl, ctx);
4438 }
4439 
4440 /* END spdk_bs_dump */
4441 
4442 /* START spdk_bs_init */
4443 
4444 struct spdk_bs_init_ctx {
4445 	struct spdk_blob_store		*bs;
4446 	struct spdk_bs_super_block	*super;
4447 };
4448 
4449 static void
4450 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4451 {
4452 	struct spdk_bs_init_ctx *ctx = cb_arg;
4453 
4454 	spdk_free(ctx->super);
4455 	free(ctx);
4456 
4457 	spdk_bs_sequence_finish(seq, bserrno);
4458 }
4459 
4460 static void
4461 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4462 {
4463 	struct spdk_bs_init_ctx *ctx = cb_arg;
4464 
4465 	/* Write super block */
4466 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
4467 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
4468 				   _spdk_bs_init_persist_super_cpl, ctx);
4469 }
4470 
4471 void
4472 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
4473 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
4474 {
4475 	struct spdk_bs_init_ctx *ctx;
4476 	struct spdk_blob_store	*bs;
4477 	struct spdk_bs_cpl	cpl;
4478 	spdk_bs_sequence_t	*seq;
4479 	spdk_bs_batch_t		*batch;
4480 	uint64_t		num_md_lba;
4481 	uint64_t		num_md_pages;
4482 	uint64_t		num_md_clusters;
4483 	uint32_t		i;
4484 	struct spdk_bs_opts	opts = {};
4485 	int			rc;
4486 
4487 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
4488 
4489 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
4490 		SPDK_ERRLOG("unsupported dev block length of %d\n",
4491 			    dev->blocklen);
4492 		dev->destroy(dev);
4493 		cb_fn(cb_arg, NULL, -EINVAL);
4494 		return;
4495 	}
4496 
4497 	if (o) {
4498 		opts = *o;
4499 	} else {
4500 		spdk_bs_opts_init(&opts);
4501 	}
4502 
4503 	if (_spdk_bs_opts_verify(&opts) != 0) {
4504 		dev->destroy(dev);
4505 		cb_fn(cb_arg, NULL, -EINVAL);
4506 		return;
4507 	}
4508 
4509 	rc = _spdk_bs_alloc(dev, &opts, &bs);
4510 	if (rc) {
4511 		dev->destroy(dev);
4512 		cb_fn(cb_arg, NULL, rc);
4513 		return;
4514 	}
4515 
4516 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
4517 		/* By default, allocate 1 page per cluster.
4518 		 * Technically, this over-allocates metadata
4519 		 * because more metadata will reduce the number
4520 		 * of usable clusters. This can be addressed with
4521 		 * more complex math in the future.
4522 		 */
4523 		bs->md_len = bs->total_clusters;
4524 	} else {
4525 		bs->md_len = opts.num_md_pages;
4526 	}
4527 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
4528 	if (rc < 0) {
4529 		_spdk_bs_free(bs);
4530 		cb_fn(cb_arg, NULL, -ENOMEM);
4531 		return;
4532 	}
4533 
4534 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
4535 	if (rc < 0) {
4536 		_spdk_bs_free(bs);
4537 		cb_fn(cb_arg, NULL, -ENOMEM);
4538 		return;
4539 	}
4540 
4541 	ctx = calloc(1, sizeof(*ctx));
4542 	if (!ctx) {
4543 		_spdk_bs_free(bs);
4544 		cb_fn(cb_arg, NULL, -ENOMEM);
4545 		return;
4546 	}
4547 
4548 	ctx->bs = bs;
4549 
4550 	/* Allocate memory for the super block */
4551 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4552 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4553 	if (!ctx->super) {
4554 		free(ctx);
4555 		_spdk_bs_free(bs);
4556 		cb_fn(cb_arg, NULL, -ENOMEM);
4557 		return;
4558 	}
4559 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
4560 	       sizeof(ctx->super->signature));
4561 	ctx->super->version = SPDK_BS_VERSION;
4562 	ctx->super->length = sizeof(*ctx->super);
4563 	ctx->super->super_blob = bs->super_blob;
4564 	ctx->super->clean = 0;
4565 	ctx->super->cluster_size = bs->cluster_sz;
4566 	ctx->super->io_unit_size = bs->io_unit_size;
4567 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
4568 
4569 	/* Calculate how many pages the metadata consumes at the front
4570 	 * of the disk.
4571 	 */
4572 
4573 	/* The super block uses 1 page */
4574 	num_md_pages = 1;
4575 
4576 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
4577 	 * up to the nearest page, plus a header.
4578 	 */
4579 	ctx->super->used_page_mask_start = num_md_pages;
4580 	ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
4581 					 spdk_divide_round_up(bs->md_len, 8),
4582 					 SPDK_BS_PAGE_SIZE);
4583 	num_md_pages += ctx->super->used_page_mask_len;
4584 
4585 	/* The used_clusters mask requires 1 bit per cluster, rounded
4586 	 * up to the nearest page, plus a header.
4587 	 */
4588 	ctx->super->used_cluster_mask_start = num_md_pages;
4589 	ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
4590 					    spdk_divide_round_up(bs->total_clusters, 8),
4591 					    SPDK_BS_PAGE_SIZE);
4592 	num_md_pages += ctx->super->used_cluster_mask_len;
4593 
4594 	/* The used_blobids mask requires 1 bit per metadata page, rounded
4595 	 * up to the nearest page, plus a header.
4596 	 */
4597 	ctx->super->used_blobid_mask_start = num_md_pages;
4598 	ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) +
4599 					   spdk_divide_round_up(bs->md_len, 8),
4600 					   SPDK_BS_PAGE_SIZE);
4601 	num_md_pages += ctx->super->used_blobid_mask_len;
4602 
4603 	/* The metadata region size was chosen above */
4604 	ctx->super->md_start = bs->md_start = num_md_pages;
4605 	ctx->super->md_len = bs->md_len;
4606 	num_md_pages += bs->md_len;
4607 
4608 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
4609 
4610 	ctx->super->size = dev->blockcnt * dev->blocklen;
4611 
4612 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
4613 
4614 	num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster);
4615 	if (num_md_clusters > bs->total_clusters) {
4616 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
4617 			    "please decrease number of pages reserved for metadata "
4618 			    "or increase cluster size.\n");
4619 		spdk_free(ctx->super);
4620 		free(ctx);
4621 		_spdk_bs_free(bs);
4622 		cb_fn(cb_arg, NULL, -ENOMEM);
4623 		return;
4624 	}
4625 	/* Claim all of the clusters used by the metadata */
4626 	for (i = 0; i < num_md_clusters; i++) {
4627 		_spdk_bs_claim_cluster(bs, i);
4628 	}
4629 
4630 	bs->total_data_clusters = bs->num_free_clusters;
4631 
4632 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
4633 	cpl.u.bs_handle.cb_fn = cb_fn;
4634 	cpl.u.bs_handle.cb_arg = cb_arg;
4635 	cpl.u.bs_handle.bs = bs;
4636 
4637 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4638 	if (!seq) {
4639 		spdk_free(ctx->super);
4640 		free(ctx);
4641 		_spdk_bs_free(bs);
4642 		cb_fn(cb_arg, NULL, -ENOMEM);
4643 		return;
4644 	}
4645 
4646 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
4647 
4648 	/* Clear metadata space */
4649 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
4650 
4651 	switch (opts.clear_method) {
4652 	case BS_CLEAR_WITH_UNMAP:
4653 		/* Trim data clusters */
4654 		spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
4655 		break;
4656 	case BS_CLEAR_WITH_WRITE_ZEROES:
4657 		/* Write_zeroes to data clusters */
4658 		spdk_bs_batch_write_zeroes_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
4659 		break;
4660 	case BS_CLEAR_WITH_NONE:
4661 	default:
4662 		break;
4663 	}
4664 
4665 	spdk_bs_batch_close(batch);
4666 }
4667 
4668 /* END spdk_bs_init */
4669 
4670 /* START spdk_bs_destroy */
4671 
4672 static void
4673 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4674 {
4675 	struct spdk_bs_init_ctx *ctx = cb_arg;
4676 	struct spdk_blob_store *bs = ctx->bs;
4677 
4678 	/*
4679 	 * We need to defer calling spdk_bs_call_cpl() until after
4680 	 * dev destruction, so tuck these away for later use.
4681 	 */
4682 	bs->unload_err = bserrno;
4683 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
4684 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
4685 
4686 	spdk_bs_sequence_finish(seq, bserrno);
4687 
4688 	_spdk_bs_free(bs);
4689 	free(ctx);
4690 }
4691 
4692 void
4693 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
4694 		void *cb_arg)
4695 {
4696 	struct spdk_bs_cpl	cpl;
4697 	spdk_bs_sequence_t	*seq;
4698 	struct spdk_bs_init_ctx *ctx;
4699 
4700 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
4701 
4702 	if (!TAILQ_EMPTY(&bs->blobs)) {
4703 		SPDK_ERRLOG("Blobstore still has open blobs\n");
4704 		cb_fn(cb_arg, -EBUSY);
4705 		return;
4706 	}
4707 
4708 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4709 	cpl.u.bs_basic.cb_fn = cb_fn;
4710 	cpl.u.bs_basic.cb_arg = cb_arg;
4711 
4712 	ctx = calloc(1, sizeof(*ctx));
4713 	if (!ctx) {
4714 		cb_fn(cb_arg, -ENOMEM);
4715 		return;
4716 	}
4717 
4718 	ctx->bs = bs;
4719 
4720 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4721 	if (!seq) {
4722 		free(ctx);
4723 		cb_fn(cb_arg, -ENOMEM);
4724 		return;
4725 	}
4726 
4727 	/* Write zeroes to the super block */
4728 	spdk_bs_sequence_write_zeroes_dev(seq,
4729 					  _spdk_bs_page_to_lba(bs, 0),
4730 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
4731 					  _spdk_bs_destroy_trim_cpl, ctx);
4732 }
4733 
4734 /* END spdk_bs_destroy */
4735 
4736 /* START spdk_bs_unload */
4737 
4738 static void
4739 _spdk_bs_unload_finish(struct spdk_bs_load_ctx *ctx, int bserrno)
4740 {
4741 	spdk_bs_sequence_t *seq = ctx->seq;
4742 
4743 	spdk_free(ctx->super);
4744 
4745 	/*
4746 	 * We need to defer calling spdk_bs_call_cpl() until after
4747 	 * dev destruction, so tuck these away for later use.
4748 	 */
4749 	ctx->bs->unload_err = bserrno;
4750 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
4751 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
4752 
4753 	spdk_bs_sequence_finish(seq, bserrno);
4754 
4755 	_spdk_bs_free(ctx->bs);
4756 	free(ctx);
4757 }
4758 
4759 static void
4760 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4761 {
4762 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4763 
4764 	_spdk_bs_unload_finish(ctx, bserrno);
4765 }
4766 
4767 static void
4768 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4769 {
4770 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4771 
4772 	spdk_free(ctx->mask);
4773 
4774 	if (bserrno != 0) {
4775 		_spdk_bs_unload_finish(ctx, bserrno);
4776 		return;
4777 	}
4778 
4779 	ctx->super->clean = 1;
4780 
4781 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
4782 }
4783 
4784 static void
4785 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4786 {
4787 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4788 
4789 	spdk_free(ctx->mask);
4790 	ctx->mask = NULL;
4791 
4792 	if (bserrno != 0) {
4793 		_spdk_bs_unload_finish(ctx, bserrno);
4794 		return;
4795 	}
4796 
4797 	_spdk_bs_write_used_clusters(seq, ctx, _spdk_bs_unload_write_used_clusters_cpl);
4798 }
4799 
4800 static void
4801 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4802 {
4803 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4804 
4805 	spdk_free(ctx->mask);
4806 	ctx->mask = NULL;
4807 
4808 	if (bserrno != 0) {
4809 		_spdk_bs_unload_finish(ctx, bserrno);
4810 		return;
4811 	}
4812 
4813 	_spdk_bs_write_used_blobids(seq, ctx, _spdk_bs_unload_write_used_blobids_cpl);
4814 }
4815 
4816 static void
4817 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4818 {
4819 	struct spdk_bs_load_ctx	*ctx = cb_arg;
4820 
4821 	if (bserrno != 0) {
4822 		_spdk_bs_unload_finish(ctx, bserrno);
4823 		return;
4824 	}
4825 
4826 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
4827 }
4828 
4829 void
4830 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
4831 {
4832 	struct spdk_bs_cpl	cpl;
4833 	struct spdk_bs_load_ctx *ctx;
4834 
4835 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
4836 
4837 	if (!TAILQ_EMPTY(&bs->blobs)) {
4838 		SPDK_ERRLOG("Blobstore still has open blobs\n");
4839 		cb_fn(cb_arg, -EBUSY);
4840 		return;
4841 	}
4842 
4843 	ctx = calloc(1, sizeof(*ctx));
4844 	if (!ctx) {
4845 		cb_fn(cb_arg, -ENOMEM);
4846 		return;
4847 	}
4848 
4849 	ctx->bs = bs;
4850 
4851 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4852 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4853 	if (!ctx->super) {
4854 		free(ctx);
4855 		cb_fn(cb_arg, -ENOMEM);
4856 		return;
4857 	}
4858 
4859 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4860 	cpl.u.bs_basic.cb_fn = cb_fn;
4861 	cpl.u.bs_basic.cb_arg = cb_arg;
4862 
4863 	ctx->seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4864 	if (!ctx->seq) {
4865 		spdk_free(ctx->super);
4866 		free(ctx);
4867 		cb_fn(cb_arg, -ENOMEM);
4868 		return;
4869 	}
4870 
4871 	/* Read super block */
4872 	spdk_bs_sequence_read_dev(ctx->seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4873 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4874 				  _spdk_bs_unload_read_super_cpl, ctx);
4875 }
4876 
4877 /* END spdk_bs_unload */
4878 
4879 /* START spdk_bs_set_super */
4880 
4881 struct spdk_bs_set_super_ctx {
4882 	struct spdk_blob_store		*bs;
4883 	struct spdk_bs_super_block	*super;
4884 };
4885 
4886 static void
4887 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4888 {
4889 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
4890 
4891 	if (bserrno != 0) {
4892 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
4893 	}
4894 
4895 	spdk_free(ctx->super);
4896 
4897 	spdk_bs_sequence_finish(seq, bserrno);
4898 
4899 	free(ctx);
4900 }
4901 
4902 static void
4903 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4904 {
4905 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
4906 
4907 	if (bserrno != 0) {
4908 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
4909 		spdk_free(ctx->super);
4910 		spdk_bs_sequence_finish(seq, bserrno);
4911 		free(ctx);
4912 		return;
4913 	}
4914 
4915 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
4916 }
4917 
4918 void
4919 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
4920 		  spdk_bs_op_complete cb_fn, void *cb_arg)
4921 {
4922 	struct spdk_bs_cpl		cpl;
4923 	spdk_bs_sequence_t		*seq;
4924 	struct spdk_bs_set_super_ctx	*ctx;
4925 
4926 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
4927 
4928 	ctx = calloc(1, sizeof(*ctx));
4929 	if (!ctx) {
4930 		cb_fn(cb_arg, -ENOMEM);
4931 		return;
4932 	}
4933 
4934 	ctx->bs = bs;
4935 
4936 	ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL,
4937 				  SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA);
4938 	if (!ctx->super) {
4939 		free(ctx);
4940 		cb_fn(cb_arg, -ENOMEM);
4941 		return;
4942 	}
4943 
4944 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
4945 	cpl.u.bs_basic.cb_fn = cb_fn;
4946 	cpl.u.bs_basic.cb_arg = cb_arg;
4947 
4948 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4949 	if (!seq) {
4950 		spdk_free(ctx->super);
4951 		free(ctx);
4952 		cb_fn(cb_arg, -ENOMEM);
4953 		return;
4954 	}
4955 
4956 	bs->super_blob = blobid;
4957 
4958 	/* Read super block */
4959 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
4960 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
4961 				  _spdk_bs_set_super_read_cpl, ctx);
4962 }
4963 
4964 /* END spdk_bs_set_super */
4965 
4966 void
4967 spdk_bs_get_super(struct spdk_blob_store *bs,
4968 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4969 {
4970 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
4971 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
4972 	} else {
4973 		cb_fn(cb_arg, bs->super_blob, 0);
4974 	}
4975 }
4976 
4977 uint64_t
4978 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
4979 {
4980 	return bs->cluster_sz;
4981 }
4982 
4983 uint64_t
4984 spdk_bs_get_page_size(struct spdk_blob_store *bs)
4985 {
4986 	return SPDK_BS_PAGE_SIZE;
4987 }
4988 
4989 uint64_t
4990 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs)
4991 {
4992 	return bs->io_unit_size;
4993 }
4994 
4995 uint64_t
4996 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
4997 {
4998 	return bs->num_free_clusters;
4999 }
5000 
5001 uint64_t
5002 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
5003 {
5004 	return bs->total_data_clusters;
5005 }
5006 
5007 static int
5008 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
5009 {
5010 	bs->md_channel = spdk_get_io_channel(bs);
5011 	if (!bs->md_channel) {
5012 		SPDK_ERRLOG("Failed to get IO channel.\n");
5013 		return -1;
5014 	}
5015 
5016 	return 0;
5017 }
5018 
5019 static int
5020 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
5021 {
5022 	spdk_put_io_channel(bs->md_channel);
5023 
5024 	return 0;
5025 }
5026 
5027 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
5028 {
5029 	assert(blob != NULL);
5030 
5031 	return blob->id;
5032 }
5033 
5034 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
5035 {
5036 	assert(blob != NULL);
5037 
5038 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
5039 }
5040 
5041 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob)
5042 {
5043 	assert(blob != NULL);
5044 
5045 	return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs);
5046 }
5047 
5048 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
5049 {
5050 	assert(blob != NULL);
5051 
5052 	return blob->active.num_clusters;
5053 }
5054 
5055 /* START spdk_bs_create_blob */
5056 
5057 static void
5058 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5059 {
5060 	struct spdk_blob *blob = cb_arg;
5061 
5062 	_spdk_blob_free(blob);
5063 
5064 	spdk_bs_sequence_finish(seq, bserrno);
5065 }
5066 
5067 static int
5068 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
5069 		      bool internal)
5070 {
5071 	uint64_t i;
5072 	size_t value_len = 0;
5073 	int rc;
5074 	const void *value = NULL;
5075 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
5076 		return -EINVAL;
5077 	}
5078 	for (i = 0; i < xattrs->count; i++) {
5079 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
5080 		if (value == NULL || value_len == 0) {
5081 			return -EINVAL;
5082 		}
5083 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
5084 		if (rc < 0) {
5085 			return rc;
5086 		}
5087 	}
5088 	return 0;
5089 }
5090 
5091 static void
5092 _spdk_bs_create_blob(struct spdk_blob_store *bs,
5093 		     const struct spdk_blob_opts *opts,
5094 		     const struct spdk_blob_xattr_opts *internal_xattrs,
5095 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5096 {
5097 	struct spdk_blob	*blob;
5098 	uint32_t		page_idx;
5099 	struct spdk_bs_cpl	cpl;
5100 	struct spdk_blob_opts	opts_default;
5101 	struct spdk_blob_xattr_opts internal_xattrs_default;
5102 	spdk_bs_sequence_t	*seq;
5103 	spdk_blob_id		id;
5104 	int rc;
5105 
5106 	assert(spdk_get_thread() == bs->md_thread);
5107 
5108 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
5109 	if (page_idx == UINT32_MAX) {
5110 		cb_fn(cb_arg, 0, -ENOMEM);
5111 		return;
5112 	}
5113 	spdk_bit_array_set(bs->used_blobids, page_idx);
5114 	_spdk_bs_claim_md_page(bs, page_idx);
5115 
5116 	id = _spdk_bs_page_to_blobid(page_idx);
5117 
5118 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
5119 
5120 	blob = _spdk_blob_alloc(bs, id);
5121 	if (!blob) {
5122 		cb_fn(cb_arg, 0, -ENOMEM);
5123 		return;
5124 	}
5125 
5126 	if (!opts) {
5127 		spdk_blob_opts_init(&opts_default);
5128 		opts = &opts_default;
5129 	}
5130 
5131 	blob->use_extent_table = opts->use_extent_table;
5132 	if (blob->use_extent_table) {
5133 		blob->invalid_flags |= SPDK_BLOB_EXTENT_TABLE;
5134 	}
5135 
5136 	if (!internal_xattrs) {
5137 		_spdk_blob_xattrs_init(&internal_xattrs_default);
5138 		internal_xattrs = &internal_xattrs_default;
5139 	}
5140 
5141 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
5142 	if (rc < 0) {
5143 		_spdk_blob_free(blob);
5144 		cb_fn(cb_arg, 0, rc);
5145 		return;
5146 	}
5147 
5148 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
5149 	if (rc < 0) {
5150 		_spdk_blob_free(blob);
5151 		cb_fn(cb_arg, 0, rc);
5152 		return;
5153 	}
5154 
5155 	if (opts->thin_provision) {
5156 		_spdk_blob_set_thin_provision(blob);
5157 	}
5158 
5159 	_spdk_blob_set_clear_method(blob, opts->clear_method);
5160 
5161 	rc = _spdk_blob_resize(blob, opts->num_clusters);
5162 	if (rc < 0) {
5163 		_spdk_blob_free(blob);
5164 		cb_fn(cb_arg, 0, rc);
5165 		return;
5166 	}
5167 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
5168 	cpl.u.blobid.cb_fn = cb_fn;
5169 	cpl.u.blobid.cb_arg = cb_arg;
5170 	cpl.u.blobid.blobid = blob->id;
5171 
5172 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5173 	if (!seq) {
5174 		_spdk_blob_free(blob);
5175 		cb_fn(cb_arg, 0, -ENOMEM);
5176 		return;
5177 	}
5178 
5179 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
5180 }
5181 
5182 void spdk_bs_create_blob(struct spdk_blob_store *bs,
5183 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5184 {
5185 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
5186 }
5187 
5188 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
5189 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5190 {
5191 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
5192 }
5193 
5194 /* END spdk_bs_create_blob */
5195 
5196 /* START blob_cleanup */
5197 
5198 struct spdk_clone_snapshot_ctx {
5199 	struct spdk_bs_cpl      cpl;
5200 	int bserrno;
5201 	bool frozen;
5202 
5203 	struct spdk_io_channel *channel;
5204 
5205 	/* Current cluster for inflate operation */
5206 	uint64_t cluster;
5207 
5208 	/* For inflation force allocation of all unallocated clusters and remove
5209 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
5210 	bool allocate_all;
5211 
5212 	struct {
5213 		spdk_blob_id id;
5214 		struct spdk_blob *blob;
5215 	} original;
5216 	struct {
5217 		spdk_blob_id id;
5218 		struct spdk_blob *blob;
5219 	} new;
5220 
5221 	/* xattrs specified for snapshot/clones only. They have no impact on
5222 	 * the original blobs xattrs. */
5223 	const struct spdk_blob_xattr_opts *xattrs;
5224 };
5225 
5226 static void
5227 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
5228 {
5229 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
5230 	struct spdk_bs_cpl *cpl = &ctx->cpl;
5231 
5232 	if (bserrno != 0) {
5233 		if (ctx->bserrno != 0) {
5234 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5235 		} else {
5236 			ctx->bserrno = bserrno;
5237 		}
5238 	}
5239 
5240 	switch (cpl->type) {
5241 	case SPDK_BS_CPL_TYPE_BLOBID:
5242 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
5243 		break;
5244 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
5245 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
5246 		break;
5247 	default:
5248 		SPDK_UNREACHABLE();
5249 		break;
5250 	}
5251 
5252 	free(ctx);
5253 }
5254 
5255 static void
5256 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
5257 {
5258 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5259 	struct spdk_blob *origblob = ctx->original.blob;
5260 
5261 	if (bserrno != 0) {
5262 		if (ctx->bserrno != 0) {
5263 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
5264 		} else {
5265 			ctx->bserrno = bserrno;
5266 		}
5267 	}
5268 
5269 	ctx->original.id = origblob->id;
5270 	origblob->locked_operation_in_progress = false;
5271 
5272 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
5273 }
5274 
5275 static void
5276 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
5277 {
5278 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5279 	struct spdk_blob *origblob = ctx->original.blob;
5280 
5281 	if (bserrno != 0) {
5282 		if (ctx->bserrno != 0) {
5283 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5284 		} else {
5285 			ctx->bserrno = bserrno;
5286 		}
5287 	}
5288 
5289 	if (ctx->frozen) {
5290 		/* Unfreeze any outstanding I/O */
5291 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
5292 	} else {
5293 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
5294 	}
5295 
5296 }
5297 
5298 static void
5299 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
5300 {
5301 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5302 	struct spdk_blob *newblob = ctx->new.blob;
5303 
5304 	if (bserrno != 0) {
5305 		if (ctx->bserrno != 0) {
5306 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
5307 		} else {
5308 			ctx->bserrno = bserrno;
5309 		}
5310 	}
5311 
5312 	ctx->new.id = newblob->id;
5313 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
5314 }
5315 
5316 /* END blob_cleanup */
5317 
5318 /* START spdk_bs_create_snapshot */
5319 
5320 static void
5321 _spdk_bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2)
5322 {
5323 	uint64_t *cluster_temp;
5324 	uint32_t *extent_page_temp;
5325 
5326 	cluster_temp = blob1->active.clusters;
5327 	blob1->active.clusters = blob2->active.clusters;
5328 	blob2->active.clusters = cluster_temp;
5329 
5330 	extent_page_temp = blob1->active.extent_pages;
5331 	blob1->active.extent_pages = blob2->active.extent_pages;
5332 	blob2->active.extent_pages = extent_page_temp;
5333 }
5334 
5335 static void
5336 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
5337 {
5338 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5339 	struct spdk_blob *origblob = ctx->original.blob;
5340 	struct spdk_blob *newblob = ctx->new.blob;
5341 
5342 	if (bserrno != 0) {
5343 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
5344 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5345 		return;
5346 	}
5347 
5348 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
5349 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
5350 	if (bserrno != 0) {
5351 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5352 		return;
5353 	}
5354 
5355 	_spdk_bs_blob_list_add(ctx->original.blob);
5356 
5357 	spdk_blob_set_read_only(newblob);
5358 
5359 	/* sync snapshot metadata */
5360 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
5361 }
5362 
5363 static void
5364 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
5365 {
5366 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5367 	struct spdk_blob *origblob = ctx->original.blob;
5368 	struct spdk_blob *newblob = ctx->new.blob;
5369 
5370 	if (bserrno != 0) {
5371 		/* return cluster map back to original */
5372 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
5373 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5374 		return;
5375 	}
5376 
5377 	/* Set internal xattr for snapshot id */
5378 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
5379 	if (bserrno != 0) {
5380 		/* return cluster map back to original */
5381 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
5382 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5383 		return;
5384 	}
5385 
5386 	_spdk_bs_blob_list_remove(origblob);
5387 	origblob->parent_id = newblob->id;
5388 
5389 	/* Create new back_bs_dev for snapshot */
5390 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
5391 	if (origblob->back_bs_dev == NULL) {
5392 		/* return cluster map back to original */
5393 		_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
5394 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
5395 		return;
5396 	}
5397 
5398 	/* set clone blob as thin provisioned */
5399 	_spdk_blob_set_thin_provision(origblob);
5400 
5401 	_spdk_bs_blob_list_add(newblob);
5402 
5403 	/* sync clone metadata */
5404 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
5405 }
5406 
5407 static void
5408 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
5409 {
5410 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5411 	struct spdk_blob *origblob = ctx->original.blob;
5412 	struct spdk_blob *newblob = ctx->new.blob;
5413 	int bserrno;
5414 
5415 	if (rc != 0) {
5416 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
5417 		return;
5418 	}
5419 
5420 	ctx->frozen = true;
5421 
5422 	/* set new back_bs_dev for snapshot */
5423 	newblob->back_bs_dev = origblob->back_bs_dev;
5424 	/* Set invalid flags from origblob */
5425 	newblob->invalid_flags = origblob->invalid_flags;
5426 
5427 	/* inherit parent from original blob if set */
5428 	newblob->parent_id = origblob->parent_id;
5429 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
5430 		/* Set internal xattr for snapshot id */
5431 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
5432 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
5433 		if (bserrno != 0) {
5434 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
5435 			return;
5436 		}
5437 	}
5438 
5439 	/* swap cluster maps */
5440 	_spdk_bs_snapshot_swap_cluster_maps(newblob, origblob);
5441 
5442 	/* Set the clear method on the new blob to match the original. */
5443 	_spdk_blob_set_clear_method(newblob, origblob->clear_method);
5444 
5445 	/* sync snapshot metadata */
5446 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
5447 }
5448 
5449 static void
5450 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5451 {
5452 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5453 	struct spdk_blob *origblob = ctx->original.blob;
5454 	struct spdk_blob *newblob = _blob;
5455 
5456 	if (bserrno != 0) {
5457 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5458 		return;
5459 	}
5460 
5461 	ctx->new.blob = newblob;
5462 	assert(spdk_blob_is_thin_provisioned(newblob));
5463 	assert(spdk_mem_all_zero(newblob->active.clusters,
5464 				 newblob->active.num_clusters * sizeof(*newblob->active.clusters)));
5465 	assert(spdk_mem_all_zero(newblob->active.extent_pages,
5466 				 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages)));
5467 
5468 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
5469 }
5470 
5471 static void
5472 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
5473 {
5474 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5475 	struct spdk_blob *origblob = ctx->original.blob;
5476 
5477 	if (bserrno != 0) {
5478 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5479 		return;
5480 	}
5481 
5482 	ctx->new.id = blobid;
5483 	ctx->cpl.u.blobid.blobid = blobid;
5484 
5485 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
5486 }
5487 
5488 
5489 static void
5490 _spdk_bs_xattr_snapshot(void *arg, const char *name,
5491 			const void **value, size_t *value_len)
5492 {
5493 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
5494 
5495 	struct spdk_blob *blob = (struct spdk_blob *)arg;
5496 	*value = &blob->id;
5497 	*value_len = sizeof(blob->id);
5498 }
5499 
5500 static void
5501 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5502 {
5503 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5504 	struct spdk_blob_opts opts;
5505 	struct spdk_blob_xattr_opts internal_xattrs;
5506 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
5507 
5508 	if (bserrno != 0) {
5509 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
5510 		return;
5511 	}
5512 
5513 	ctx->original.blob = _blob;
5514 
5515 	if (_blob->data_ro || _blob->md_ro) {
5516 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
5517 			      _blob->id);
5518 		ctx->bserrno = -EINVAL;
5519 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
5520 		return;
5521 	}
5522 
5523 	if (_blob->locked_operation_in_progress) {
5524 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot - another operation in progress\n");
5525 		ctx->bserrno = -EBUSY;
5526 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
5527 		return;
5528 	}
5529 
5530 	_blob->locked_operation_in_progress = true;
5531 
5532 	spdk_blob_opts_init(&opts);
5533 	_spdk_blob_xattrs_init(&internal_xattrs);
5534 
5535 	/* Change the size of new blob to the same as in original blob,
5536 	 * but do not allocate clusters */
5537 	opts.thin_provision = true;
5538 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
5539 	opts.use_extent_table = _blob->use_extent_table;
5540 
5541 	/* If there are any xattrs specified for snapshot, set them now */
5542 	if (ctx->xattrs) {
5543 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
5544 	}
5545 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
5546 	internal_xattrs.count = 1;
5547 	internal_xattrs.ctx = _blob;
5548 	internal_xattrs.names = xattrs_names;
5549 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
5550 
5551 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
5552 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
5553 }
5554 
5555 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
5556 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
5557 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5558 {
5559 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
5560 
5561 	if (!ctx) {
5562 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
5563 		return;
5564 	}
5565 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
5566 	ctx->cpl.u.blobid.cb_fn = cb_fn;
5567 	ctx->cpl.u.blobid.cb_arg = cb_arg;
5568 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
5569 	ctx->bserrno = 0;
5570 	ctx->frozen = false;
5571 	ctx->original.id = blobid;
5572 	ctx->xattrs = snapshot_xattrs;
5573 
5574 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
5575 }
5576 /* END spdk_bs_create_snapshot */
5577 
5578 /* START spdk_bs_create_clone */
5579 
5580 static void
5581 _spdk_bs_xattr_clone(void *arg, const char *name,
5582 		     const void **value, size_t *value_len)
5583 {
5584 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
5585 
5586 	struct spdk_blob *blob = (struct spdk_blob *)arg;
5587 	*value = &blob->id;
5588 	*value_len = sizeof(blob->id);
5589 }
5590 
5591 static void
5592 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5593 {
5594 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5595 	struct spdk_blob *clone = _blob;
5596 
5597 	ctx->new.blob = clone;
5598 	_spdk_bs_blob_list_add(clone);
5599 
5600 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
5601 }
5602 
5603 static void
5604 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
5605 {
5606 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5607 
5608 	ctx->cpl.u.blobid.blobid = blobid;
5609 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
5610 }
5611 
5612 static void
5613 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5614 {
5615 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5616 	struct spdk_blob_opts		opts;
5617 	struct spdk_blob_xattr_opts internal_xattrs;
5618 	char *xattr_names[] = { BLOB_SNAPSHOT };
5619 
5620 	if (bserrno != 0) {
5621 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
5622 		return;
5623 	}
5624 
5625 	ctx->original.blob = _blob;
5626 
5627 	if (!_blob->data_ro || !_blob->md_ro) {
5628 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
5629 		ctx->bserrno = -EINVAL;
5630 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
5631 		return;
5632 	}
5633 
5634 	if (_blob->locked_operation_in_progress) {
5635 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create clone - another operation in progress\n");
5636 		ctx->bserrno = -EBUSY;
5637 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
5638 		return;
5639 	}
5640 
5641 	_blob->locked_operation_in_progress = true;
5642 
5643 	spdk_blob_opts_init(&opts);
5644 	_spdk_blob_xattrs_init(&internal_xattrs);
5645 
5646 	opts.thin_provision = true;
5647 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
5648 	opts.use_extent_table = _blob->use_extent_table;
5649 	if (ctx->xattrs) {
5650 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
5651 	}
5652 
5653 	/* Set internal xattr BLOB_SNAPSHOT */
5654 	internal_xattrs.count = 1;
5655 	internal_xattrs.ctx = _blob;
5656 	internal_xattrs.names = xattr_names;
5657 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
5658 
5659 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
5660 			     _spdk_bs_clone_newblob_create_cpl, ctx);
5661 }
5662 
5663 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
5664 			  const struct spdk_blob_xattr_opts *clone_xattrs,
5665 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
5666 {
5667 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
5668 
5669 	if (!ctx) {
5670 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
5671 		return;
5672 	}
5673 
5674 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
5675 	ctx->cpl.u.blobid.cb_fn = cb_fn;
5676 	ctx->cpl.u.blobid.cb_arg = cb_arg;
5677 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
5678 	ctx->bserrno = 0;
5679 	ctx->xattrs = clone_xattrs;
5680 	ctx->original.id = blobid;
5681 
5682 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
5683 }
5684 
5685 /* END spdk_bs_create_clone */
5686 
5687 /* START spdk_bs_inflate_blob */
5688 
5689 static void
5690 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
5691 {
5692 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5693 	struct spdk_blob *_blob = ctx->original.blob;
5694 
5695 	if (bserrno != 0) {
5696 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5697 		return;
5698 	}
5699 
5700 	assert(_parent != NULL);
5701 
5702 	_spdk_bs_blob_list_remove(_blob);
5703 	_blob->parent_id = _parent->id;
5704 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
5705 			     sizeof(spdk_blob_id), true);
5706 
5707 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
5708 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
5709 	_spdk_bs_blob_list_add(_blob);
5710 
5711 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
5712 }
5713 
5714 static void
5715 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
5716 {
5717 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5718 	struct spdk_blob *_blob = ctx->original.blob;
5719 	struct spdk_blob *_parent;
5720 
5721 	if (bserrno != 0) {
5722 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5723 		return;
5724 	}
5725 
5726 	if (ctx->allocate_all) {
5727 		/* remove thin provisioning */
5728 		_spdk_bs_blob_list_remove(_blob);
5729 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
5730 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
5731 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
5732 		_blob->back_bs_dev = NULL;
5733 		_blob->parent_id = SPDK_BLOBID_INVALID;
5734 	} else {
5735 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
5736 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
5737 			/* We must change the parent of the inflated blob */
5738 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
5739 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
5740 			return;
5741 		}
5742 
5743 		_spdk_bs_blob_list_remove(_blob);
5744 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
5745 		_blob->parent_id = SPDK_BLOBID_INVALID;
5746 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
5747 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
5748 	}
5749 
5750 	_blob->state = SPDK_BLOB_STATE_DIRTY;
5751 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
5752 }
5753 
5754 /* Check if cluster needs allocation */
5755 static inline bool
5756 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
5757 {
5758 	struct spdk_blob_bs_dev *b;
5759 
5760 	assert(blob != NULL);
5761 
5762 	if (blob->active.clusters[cluster] != 0) {
5763 		/* Cluster is already allocated */
5764 		return false;
5765 	}
5766 
5767 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
5768 		/* Blob have no parent blob */
5769 		return allocate_all;
5770 	}
5771 
5772 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
5773 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
5774 }
5775 
5776 static void
5777 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
5778 {
5779 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5780 	struct spdk_blob *_blob = ctx->original.blob;
5781 	uint64_t offset;
5782 
5783 	if (bserrno != 0) {
5784 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
5785 		return;
5786 	}
5787 
5788 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
5789 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
5790 			break;
5791 		}
5792 	}
5793 
5794 	if (ctx->cluster < _blob->active.num_clusters) {
5795 		offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster);
5796 
5797 		/* We may safely increment a cluster before write */
5798 		ctx->cluster++;
5799 
5800 		/* Use zero length write to touch a cluster */
5801 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
5802 				   _spdk_bs_inflate_blob_touch_next, ctx);
5803 	} else {
5804 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
5805 	}
5806 }
5807 
5808 static void
5809 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5810 {
5811 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
5812 	uint64_t lfc; /* lowest free cluster */
5813 	uint64_t i;
5814 
5815 	if (bserrno != 0) {
5816 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
5817 		return;
5818 	}
5819 
5820 	ctx->original.blob = _blob;
5821 
5822 	if (_blob->locked_operation_in_progress) {
5823 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot inflate blob - another operation in progress\n");
5824 		ctx->bserrno = -EBUSY;
5825 		spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
5826 		return;
5827 	}
5828 
5829 	_blob->locked_operation_in_progress = true;
5830 
5831 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
5832 		/* This blob have no parent, so we cannot decouple it. */
5833 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
5834 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
5835 		return;
5836 	}
5837 
5838 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
5839 		/* This is not thin provisioned blob. No need to inflate. */
5840 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
5841 		return;
5842 	}
5843 
5844 	/* Do two passes - one to verify that we can obtain enough clusters
5845 	 * and another to actually claim them.
5846 	 */
5847 	lfc = 0;
5848 	for (i = 0; i < _blob->active.num_clusters; i++) {
5849 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
5850 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
5851 			if (lfc == UINT32_MAX) {
5852 				/* No more free clusters. Cannot satisfy the request */
5853 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
5854 				return;
5855 			}
5856 			lfc++;
5857 		}
5858 	}
5859 
5860 	ctx->cluster = 0;
5861 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
5862 }
5863 
5864 static void
5865 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5866 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
5867 {
5868 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
5869 
5870 	if (!ctx) {
5871 		cb_fn(cb_arg, -ENOMEM);
5872 		return;
5873 	}
5874 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5875 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
5876 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
5877 	ctx->bserrno = 0;
5878 	ctx->original.id = blobid;
5879 	ctx->channel = channel;
5880 	ctx->allocate_all = allocate_all;
5881 
5882 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
5883 }
5884 
5885 void
5886 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5887 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
5888 {
5889 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
5890 }
5891 
5892 void
5893 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
5894 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
5895 {
5896 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
5897 }
5898 /* END spdk_bs_inflate_blob */
5899 
5900 /* START spdk_blob_resize */
5901 struct spdk_bs_resize_ctx {
5902 	spdk_blob_op_complete cb_fn;
5903 	void *cb_arg;
5904 	struct spdk_blob *blob;
5905 	uint64_t sz;
5906 	int rc;
5907 };
5908 
5909 static void
5910 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
5911 {
5912 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
5913 
5914 	if (rc != 0) {
5915 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
5916 	}
5917 
5918 	if (ctx->rc != 0) {
5919 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
5920 		rc = ctx->rc;
5921 	}
5922 
5923 	ctx->blob->locked_operation_in_progress = false;
5924 
5925 	ctx->cb_fn(ctx->cb_arg, rc);
5926 	free(ctx);
5927 }
5928 
5929 static void
5930 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
5931 {
5932 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
5933 
5934 	if (rc != 0) {
5935 		ctx->blob->locked_operation_in_progress = false;
5936 		ctx->cb_fn(ctx->cb_arg, rc);
5937 		free(ctx);
5938 		return;
5939 	}
5940 
5941 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
5942 
5943 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
5944 }
5945 
5946 void
5947 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
5948 {
5949 	struct spdk_bs_resize_ctx *ctx;
5950 
5951 	_spdk_blob_verify_md_op(blob);
5952 
5953 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
5954 
5955 	if (blob->md_ro) {
5956 		cb_fn(cb_arg, -EPERM);
5957 		return;
5958 	}
5959 
5960 	if (sz == blob->active.num_clusters) {
5961 		cb_fn(cb_arg, 0);
5962 		return;
5963 	}
5964 
5965 	if (blob->locked_operation_in_progress) {
5966 		cb_fn(cb_arg, -EBUSY);
5967 		return;
5968 	}
5969 
5970 	ctx = calloc(1, sizeof(*ctx));
5971 	if (!ctx) {
5972 		cb_fn(cb_arg, -ENOMEM);
5973 		return;
5974 	}
5975 
5976 	blob->locked_operation_in_progress = true;
5977 	ctx->cb_fn = cb_fn;
5978 	ctx->cb_arg = cb_arg;
5979 	ctx->blob = blob;
5980 	ctx->sz = sz;
5981 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
5982 }
5983 
5984 /* END spdk_blob_resize */
5985 
5986 
5987 /* START spdk_bs_delete_blob */
5988 
5989 static void
5990 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
5991 {
5992 	spdk_bs_sequence_t *seq = cb_arg;
5993 
5994 	spdk_bs_sequence_finish(seq, bserrno);
5995 }
5996 
5997 static void
5998 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5999 {
6000 	struct spdk_blob *blob = cb_arg;
6001 
6002 	if (bserrno != 0) {
6003 		/*
6004 		 * We already removed this blob from the blobstore tailq, so
6005 		 *  we need to free it here since this is the last reference
6006 		 *  to it.
6007 		 */
6008 		_spdk_blob_free(blob);
6009 		_spdk_bs_delete_close_cpl(seq, bserrno);
6010 		return;
6011 	}
6012 
6013 	/*
6014 	 * This will immediately decrement the ref_count and call
6015 	 *  the completion routine since the metadata state is clean.
6016 	 *  By calling spdk_blob_close, we reduce the number of call
6017 	 *  points into code that touches the blob->open_ref count
6018 	 *  and the blobstore's blob list.
6019 	 */
6020 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
6021 }
6022 
6023 struct delete_snapshot_ctx {
6024 	struct spdk_blob_list *parent_snapshot_entry;
6025 	struct spdk_blob *snapshot;
6026 	bool snapshot_md_ro;
6027 	struct spdk_blob *clone;
6028 	bool clone_md_ro;
6029 	spdk_blob_op_with_handle_complete cb_fn;
6030 	void *cb_arg;
6031 	int bserrno;
6032 };
6033 
6034 static void
6035 _spdk_delete_blob_cleanup_finish(void *cb_arg, int bserrno)
6036 {
6037 	struct delete_snapshot_ctx *ctx = cb_arg;
6038 
6039 	if (bserrno != 0) {
6040 		SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno);
6041 	}
6042 
6043 	assert(ctx != NULL);
6044 
6045 	if (bserrno != 0 && ctx->bserrno == 0) {
6046 		ctx->bserrno = bserrno;
6047 	}
6048 
6049 	ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno);
6050 	free(ctx);
6051 }
6052 
6053 static void
6054 _spdk_delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno)
6055 {
6056 	struct delete_snapshot_ctx *ctx = cb_arg;
6057 
6058 	if (bserrno != 0) {
6059 		ctx->bserrno = bserrno;
6060 		SPDK_ERRLOG("Clone cleanup error %d\n", bserrno);
6061 	}
6062 
6063 	/* open_ref == 1 menas that only deletion context has opened this snapshot
6064 	 * open_ref == 2 menas that clone has opened this snapshot as well,
6065 	 * so we have to add it back to the blobs list */
6066 	if (ctx->snapshot->open_ref == 2) {
6067 		TAILQ_INSERT_HEAD(&ctx->snapshot->bs->blobs, ctx->snapshot, link);
6068 	}
6069 
6070 	ctx->snapshot->locked_operation_in_progress = false;
6071 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
6072 
6073 	spdk_blob_close(ctx->snapshot, _spdk_delete_blob_cleanup_finish, ctx);
6074 }
6075 
6076 static void
6077 _spdk_delete_snapshot_cleanup_clone(void *cb_arg, int bserrno)
6078 {
6079 	struct delete_snapshot_ctx *ctx = cb_arg;
6080 
6081 	ctx->clone->locked_operation_in_progress = false;
6082 	ctx->clone->md_ro = ctx->clone_md_ro;
6083 
6084 	spdk_blob_close(ctx->clone, _spdk_delete_snapshot_cleanup_snapshot, ctx);
6085 }
6086 
6087 static void
6088 _spdk_delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
6089 {
6090 	struct delete_snapshot_ctx *ctx = cb_arg;
6091 
6092 	if (bserrno) {
6093 		ctx->bserrno = bserrno;
6094 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
6095 		return;
6096 	}
6097 
6098 	ctx->clone->locked_operation_in_progress = false;
6099 	spdk_blob_close(ctx->clone, _spdk_delete_blob_cleanup_finish, ctx);
6100 }
6101 
6102 static void
6103 _spdk_delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno)
6104 {
6105 	struct delete_snapshot_ctx *ctx = cb_arg;
6106 	struct spdk_blob_list *parent_snapshot_entry = NULL;
6107 	struct spdk_blob_list *snapshot_entry = NULL;
6108 	struct spdk_blob_list *clone_entry = NULL;
6109 	struct spdk_blob_list *snapshot_clone_entry = NULL;
6110 
6111 	if (bserrno) {
6112 		SPDK_ERRLOG("Failed to sync MD on blob\n");
6113 		ctx->bserrno = bserrno;
6114 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
6115 		return;
6116 	}
6117 
6118 	/* Get snapshot entry for the snapshot we want to remove */
6119 	snapshot_entry = _spdk_bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id);
6120 
6121 	assert(snapshot_entry != NULL);
6122 
6123 	/* Remove clone entry in this snapshot (at this point there can be only one clone) */
6124 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6125 	assert(clone_entry != NULL);
6126 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
6127 	snapshot_entry->clone_count--;
6128 	assert(TAILQ_EMPTY(&snapshot_entry->clones));
6129 
6130 	if (ctx->snapshot->parent_id != SPDK_BLOBID_INVALID) {
6131 		/* This snapshot is at the same time a clone of another snapshot - we need to
6132 		 * update parent snapshot (remove current clone, add new one inherited from
6133 		 * the snapshot that is being removed) */
6134 
6135 		/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
6136 		 * snapshot that we are removing */
6137 		_spdk_blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry,
6138 				&snapshot_clone_entry);
6139 
6140 		/* Switch clone entry in parent snapshot */
6141 		TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link);
6142 		TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link);
6143 		free(snapshot_clone_entry);
6144 	} else {
6145 		/* No parent snapshot - just remove clone entry */
6146 		free(clone_entry);
6147 	}
6148 
6149 	/* Restore md_ro flags */
6150 	ctx->clone->md_ro = ctx->clone_md_ro;
6151 	ctx->snapshot->md_ro = ctx->snapshot_md_ro;
6152 
6153 	_spdk_blob_unfreeze_io(ctx->clone, _spdk_delete_snapshot_unfreeze_cpl, ctx);
6154 }
6155 
6156 static void
6157 _spdk_delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno)
6158 {
6159 	struct delete_snapshot_ctx *ctx = cb_arg;
6160 	uint64_t i;
6161 
6162 	ctx->snapshot->md_ro = false;
6163 
6164 	if (bserrno) {
6165 		SPDK_ERRLOG("Failed to sync MD on clone\n");
6166 		ctx->bserrno = bserrno;
6167 
6168 		/* Restore snapshot to previous state */
6169 		bserrno = _spdk_blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true);
6170 		if (bserrno != 0) {
6171 			_spdk_delete_snapshot_cleanup_clone(ctx, bserrno);
6172 			return;
6173 		}
6174 
6175 		spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_cleanup_clone, ctx);
6176 		return;
6177 	}
6178 
6179 	/* Clear cluster map entries for snapshot */
6180 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
6181 		if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) {
6182 			ctx->snapshot->active.clusters[i] = 0;
6183 		}
6184 	}
6185 
6186 	ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY;
6187 
6188 	if (ctx->parent_snapshot_entry != NULL) {
6189 		ctx->snapshot->back_bs_dev = NULL;
6190 	}
6191 
6192 	spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_sync_snapshot_cpl, ctx);
6193 }
6194 
6195 static void
6196 _spdk_delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno)
6197 {
6198 	struct delete_snapshot_ctx *ctx = cb_arg;
6199 	uint64_t i;
6200 
6201 	/* Temporarily override md_ro flag for clone for MD modification */
6202 	ctx->clone_md_ro = ctx->clone->md_ro;
6203 	ctx->clone->md_ro = false;
6204 
6205 	if (bserrno) {
6206 		SPDK_ERRLOG("Failed to sync MD with xattr on blob\n");
6207 		ctx->bserrno = bserrno;
6208 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
6209 		return;
6210 	}
6211 
6212 	/* Copy snapshot map to clone map (only unallocated clusters in clone) */
6213 	for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) {
6214 		if (ctx->clone->active.clusters[i] == 0) {
6215 			ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i];
6216 		}
6217 	}
6218 
6219 	/* Delete old backing bs_dev from clone (related to snapshot that will be removed) */
6220 	ctx->clone->back_bs_dev->destroy(ctx->clone->back_bs_dev);
6221 
6222 	/* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */
6223 	if (ctx->parent_snapshot_entry != NULL) {
6224 		/* ...to parent snapshot */
6225 		ctx->clone->parent_id = ctx->parent_snapshot_entry->id;
6226 		ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev;
6227 		_spdk_blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id,
6228 				     sizeof(spdk_blob_id),
6229 				     true);
6230 	} else {
6231 		/* ...to blobid invalid and zeroes dev */
6232 		ctx->clone->parent_id = SPDK_BLOBID_INVALID;
6233 		ctx->clone->back_bs_dev = spdk_bs_create_zeroes_dev();
6234 		_spdk_blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true);
6235 	}
6236 
6237 	spdk_blob_sync_md(ctx->clone, _spdk_delete_snapshot_sync_clone_cpl, ctx);
6238 }
6239 
6240 static void
6241 _spdk_delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno)
6242 {
6243 	struct delete_snapshot_ctx *ctx = cb_arg;
6244 
6245 	if (bserrno) {
6246 		SPDK_ERRLOG("Failed to freeze I/O on clone\n");
6247 		ctx->bserrno = bserrno;
6248 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
6249 		return;
6250 	}
6251 
6252 	/* Temporarily override md_ro flag for snapshot for MD modification */
6253 	ctx->snapshot_md_ro = ctx->snapshot->md_ro;
6254 	ctx->snapshot->md_ro = false;
6255 
6256 	/* Mark blob as pending for removal for power failure safety, use clone id for recovery */
6257 	ctx->bserrno = _spdk_blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id,
6258 					    sizeof(spdk_blob_id), true);
6259 	if (ctx->bserrno != 0) {
6260 		_spdk_delete_snapshot_cleanup_clone(ctx, 0);
6261 		return;
6262 	}
6263 
6264 	spdk_blob_sync_md(ctx->snapshot, _spdk_delete_snapshot_sync_snapshot_xattr_cpl, ctx);
6265 }
6266 
6267 static void
6268 _spdk_delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno)
6269 {
6270 	struct delete_snapshot_ctx *ctx = cb_arg;
6271 
6272 	if (bserrno) {
6273 		SPDK_ERRLOG("Failed to open clone\n");
6274 		ctx->bserrno = bserrno;
6275 		_spdk_delete_snapshot_cleanup_snapshot(ctx, 0);
6276 		return;
6277 	}
6278 
6279 	ctx->clone = clone;
6280 
6281 	if (clone->locked_operation_in_progress) {
6282 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress on its clone\n");
6283 		ctx->bserrno = -EBUSY;
6284 		spdk_blob_close(ctx->clone, _spdk_delete_snapshot_cleanup_snapshot, ctx);
6285 		return;
6286 	}
6287 
6288 	clone->locked_operation_in_progress = true;
6289 
6290 	_spdk_blob_freeze_io(clone, _spdk_delete_snapshot_freeze_io_cb, ctx);
6291 }
6292 
6293 static void
6294 _spdk_update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx)
6295 {
6296 	struct spdk_blob_list *snapshot_entry = NULL;
6297 	struct spdk_blob_list *clone_entry = NULL;
6298 	struct spdk_blob_list *snapshot_clone_entry = NULL;
6299 
6300 	/* Get snapshot entry for the snapshot we want to remove */
6301 	snapshot_entry = _spdk_bs_get_snapshot_entry(snapshot->bs, snapshot->id);
6302 
6303 	assert(snapshot_entry != NULL);
6304 
6305 	/* Get clone of the snapshot (at this point there can be only one clone) */
6306 	clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6307 	assert(snapshot_entry->clone_count == 1);
6308 	assert(clone_entry != NULL);
6309 
6310 	/* Get snapshot entry for parent snapshot and clone entry within that snapshot for
6311 	 * snapshot that we are removing */
6312 	_spdk_blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry,
6313 			&snapshot_clone_entry);
6314 
6315 	spdk_bs_open_blob(snapshot->bs, clone_entry->id, _spdk_delete_snapshot_open_clone_cb, ctx);
6316 }
6317 
6318 static void
6319 _spdk_bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno)
6320 {
6321 	spdk_bs_sequence_t *seq = cb_arg;
6322 	struct spdk_blob_list *snapshot_entry = NULL;
6323 	uint32_t page_num;
6324 
6325 	if (bserrno) {
6326 		SPDK_ERRLOG("Failed to remove blob\n");
6327 		spdk_bs_sequence_finish(seq, bserrno);
6328 		return;
6329 	}
6330 
6331 	/* Remove snapshot from the list */
6332 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
6333 	if (snapshot_entry != NULL) {
6334 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
6335 		free(snapshot_entry);
6336 	}
6337 
6338 	page_num = _spdk_bs_blobid_to_page(blob->id);
6339 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
6340 	blob->state = SPDK_BLOB_STATE_DIRTY;
6341 	blob->active.num_pages = 0;
6342 	_spdk_blob_resize(blob, 0);
6343 
6344 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
6345 }
6346 
6347 static int
6348 _spdk_bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone)
6349 {
6350 	struct spdk_blob_list *snapshot_entry = NULL;
6351 	struct spdk_blob_list *clone_entry = NULL;
6352 	struct spdk_blob *clone = NULL;
6353 	bool has_one_clone = false;
6354 
6355 	/* Check if this is a snapshot with clones */
6356 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
6357 	if (snapshot_entry != NULL) {
6358 		if (snapshot_entry->clone_count > 1) {
6359 			SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n");
6360 			return -EBUSY;
6361 		} else if (snapshot_entry->clone_count == 1) {
6362 			has_one_clone = true;
6363 		}
6364 	}
6365 
6366 	/* Check if someone has this blob open (besides this delete context):
6367 	 * - open_ref = 1 - only this context opened blob, so it is ok to remove it
6368 	 * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot
6369 	 *	and that is ok, because we will update it accordingly */
6370 	if (blob->open_ref <= 2 && has_one_clone) {
6371 		clone_entry = TAILQ_FIRST(&snapshot_entry->clones);
6372 		assert(clone_entry != NULL);
6373 		clone = _spdk_blob_lookup(blob->bs, clone_entry->id);
6374 
6375 		if (blob->open_ref == 2 && clone == NULL) {
6376 			/* Clone is closed and someone else opened this blob */
6377 			SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
6378 			return -EBUSY;
6379 		}
6380 
6381 		*update_clone = true;
6382 		return 0;
6383 	}
6384 
6385 	if (blob->open_ref > 1) {
6386 		SPDK_ERRLOG("Cannot remove snapshot because it is open\n");
6387 		return -EBUSY;
6388 	}
6389 
6390 	assert(has_one_clone == false);
6391 	*update_clone = false;
6392 	return 0;
6393 }
6394 
6395 static void
6396 _spdk_bs_delete_enomem_close_cpl(void *cb_arg, int bserrno)
6397 {
6398 	spdk_bs_sequence_t *seq = cb_arg;
6399 
6400 	spdk_bs_sequence_finish(seq, -ENOMEM);
6401 }
6402 
6403 static void
6404 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
6405 {
6406 	spdk_bs_sequence_t *seq = cb_arg;
6407 	struct delete_snapshot_ctx *ctx;
6408 	bool update_clone = false;
6409 
6410 	if (bserrno != 0) {
6411 		spdk_bs_sequence_finish(seq, bserrno);
6412 		return;
6413 	}
6414 
6415 	_spdk_blob_verify_md_op(blob);
6416 
6417 	ctx = calloc(1, sizeof(*ctx));
6418 	if (ctx == NULL) {
6419 		spdk_blob_close(blob, _spdk_bs_delete_enomem_close_cpl, seq);
6420 		return;
6421 	}
6422 
6423 	ctx->snapshot = blob;
6424 	ctx->cb_fn = _spdk_bs_delete_blob_finish;
6425 	ctx->cb_arg = seq;
6426 
6427 	/* Check if blob can be removed and if it is a snapshot with clone on top of it */
6428 	ctx->bserrno = _spdk_bs_is_blob_deletable(blob, &update_clone);
6429 	if (ctx->bserrno) {
6430 		spdk_blob_close(blob, _spdk_delete_blob_cleanup_finish, ctx);
6431 		return;
6432 	}
6433 
6434 	if (blob->locked_operation_in_progress) {
6435 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress\n");
6436 		ctx->bserrno = -EBUSY;
6437 		spdk_blob_close(blob, _spdk_delete_blob_cleanup_finish, ctx);
6438 		return;
6439 	}
6440 
6441 	blob->locked_operation_in_progress = true;
6442 
6443 	/*
6444 	 * Remove the blob from the blob_store list now, to ensure it does not
6445 	 *  get returned after this point by _spdk_blob_lookup().
6446 	 */
6447 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
6448 
6449 	if (update_clone) {
6450 		/* This blob is a snapshot with active clone - update clone first */
6451 		_spdk_update_clone_on_snapshot_deletion(blob, ctx);
6452 	} else {
6453 		/* This blob does not have any clones - just remove it */
6454 		_spdk_bs_blob_list_remove(blob);
6455 		_spdk_bs_delete_blob_finish(seq, blob, 0);
6456 		free(ctx);
6457 	}
6458 }
6459 
6460 void
6461 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
6462 		    spdk_blob_op_complete cb_fn, void *cb_arg)
6463 {
6464 	struct spdk_bs_cpl	cpl;
6465 	spdk_bs_sequence_t	*seq;
6466 
6467 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
6468 
6469 	assert(spdk_get_thread() == bs->md_thread);
6470 
6471 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6472 	cpl.u.blob_basic.cb_fn = cb_fn;
6473 	cpl.u.blob_basic.cb_arg = cb_arg;
6474 
6475 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
6476 	if (!seq) {
6477 		cb_fn(cb_arg, -ENOMEM);
6478 		return;
6479 	}
6480 
6481 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
6482 }
6483 
6484 /* END spdk_bs_delete_blob */
6485 
6486 /* START spdk_bs_open_blob */
6487 
6488 static void
6489 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6490 {
6491 	struct spdk_blob *blob = cb_arg;
6492 
6493 	if (bserrno != 0) {
6494 		_spdk_blob_free(blob);
6495 		seq->cpl.u.blob_handle.blob = NULL;
6496 		spdk_bs_sequence_finish(seq, bserrno);
6497 		return;
6498 	}
6499 
6500 	blob->open_ref++;
6501 
6502 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
6503 
6504 	spdk_bs_sequence_finish(seq, bserrno);
6505 }
6506 
6507 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
6508 			       struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6509 {
6510 	struct spdk_blob		*blob;
6511 	struct spdk_bs_cpl		cpl;
6512 	struct spdk_blob_open_opts	opts_default;
6513 	spdk_bs_sequence_t		*seq;
6514 	uint32_t			page_num;
6515 
6516 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
6517 	assert(spdk_get_thread() == bs->md_thread);
6518 
6519 	page_num = _spdk_bs_blobid_to_page(blobid);
6520 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
6521 		/* Invalid blobid */
6522 		cb_fn(cb_arg, NULL, -ENOENT);
6523 		return;
6524 	}
6525 
6526 	blob = _spdk_blob_lookup(bs, blobid);
6527 	if (blob) {
6528 		blob->open_ref++;
6529 		cb_fn(cb_arg, blob, 0);
6530 		return;
6531 	}
6532 
6533 	blob = _spdk_blob_alloc(bs, blobid);
6534 	if (!blob) {
6535 		cb_fn(cb_arg, NULL, -ENOMEM);
6536 		return;
6537 	}
6538 
6539 	if (!opts) {
6540 		spdk_blob_open_opts_init(&opts_default);
6541 		opts = &opts_default;
6542 	}
6543 
6544 	blob->clear_method = opts->clear_method;
6545 
6546 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
6547 	cpl.u.blob_handle.cb_fn = cb_fn;
6548 	cpl.u.blob_handle.cb_arg = cb_arg;
6549 	cpl.u.blob_handle.blob = blob;
6550 
6551 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
6552 	if (!seq) {
6553 		_spdk_blob_free(blob);
6554 		cb_fn(cb_arg, NULL, -ENOMEM);
6555 		return;
6556 	}
6557 
6558 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
6559 }
6560 
6561 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
6562 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6563 {
6564 	_spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg);
6565 }
6566 
6567 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid,
6568 			   struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6569 {
6570 	_spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg);
6571 }
6572 
6573 /* END spdk_bs_open_blob */
6574 
6575 /* START spdk_blob_set_read_only */
6576 int spdk_blob_set_read_only(struct spdk_blob *blob)
6577 {
6578 	_spdk_blob_verify_md_op(blob);
6579 
6580 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
6581 
6582 	blob->state = SPDK_BLOB_STATE_DIRTY;
6583 	return 0;
6584 }
6585 /* END spdk_blob_set_read_only */
6586 
6587 /* START spdk_blob_sync_md */
6588 
6589 static void
6590 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6591 {
6592 	struct spdk_blob *blob = cb_arg;
6593 
6594 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
6595 		blob->data_ro = true;
6596 		blob->md_ro = true;
6597 	}
6598 
6599 	spdk_bs_sequence_finish(seq, bserrno);
6600 }
6601 
6602 static void
6603 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
6604 {
6605 	struct spdk_bs_cpl	cpl;
6606 	spdk_bs_sequence_t	*seq;
6607 
6608 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6609 	cpl.u.blob_basic.cb_fn = cb_fn;
6610 	cpl.u.blob_basic.cb_arg = cb_arg;
6611 
6612 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
6613 	if (!seq) {
6614 		cb_fn(cb_arg, -ENOMEM);
6615 		return;
6616 	}
6617 
6618 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
6619 }
6620 
6621 void
6622 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
6623 {
6624 	_spdk_blob_verify_md_op(blob);
6625 
6626 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
6627 
6628 	if (blob->md_ro) {
6629 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
6630 		cb_fn(cb_arg, 0);
6631 		return;
6632 	}
6633 
6634 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
6635 }
6636 
6637 /* END spdk_blob_sync_md */
6638 
6639 struct spdk_blob_insert_cluster_ctx {
6640 	struct spdk_thread	*thread;
6641 	struct spdk_blob	*blob;
6642 	uint32_t		cluster_num;	/* cluster index in blob */
6643 	uint32_t		cluster;	/* cluster on disk */
6644 	uint32_t		extent_page;	/* extent page on disk */
6645 	int			rc;
6646 	spdk_blob_op_complete	cb_fn;
6647 	void			*cb_arg;
6648 };
6649 
6650 static void
6651 _spdk_blob_insert_cluster_msg_cpl(void *arg)
6652 {
6653 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
6654 
6655 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
6656 	free(ctx);
6657 }
6658 
6659 static void
6660 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
6661 {
6662 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
6663 
6664 	ctx->rc = bserrno;
6665 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
6666 }
6667 
6668 static void
6669 _spdk_blob_persist_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6670 {
6671 	struct spdk_blob_md_page        *page = cb_arg;
6672 
6673 	spdk_bs_sequence_finish(seq, bserrno);
6674 	spdk_free(page);
6675 }
6676 
6677 static void
6678 _spdk_blob_insert_extent(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num,
6679 			 spdk_blob_op_complete cb_fn, void *cb_arg)
6680 {
6681 	spdk_bs_sequence_t		*seq;
6682 	struct spdk_bs_cpl		cpl;
6683 	struct spdk_blob_md_page	*page = NULL;
6684 	uint32_t			page_count = 0;
6685 	int				rc;
6686 
6687 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6688 	cpl.u.blob_basic.cb_fn = cb_fn;
6689 	cpl.u.blob_basic.cb_arg = cb_arg;
6690 
6691 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
6692 	if (!seq) {
6693 		cb_fn(cb_arg, -ENOMEM);
6694 		return;
6695 	}
6696 	rc = _spdk_blob_serialize_add_page(blob, &page, &page_count, &page);
6697 	if (rc < 0) {
6698 		spdk_bs_sequence_finish(seq, rc);
6699 		return;
6700 	}
6701 
6702 	_spdk_blob_serialize_extent_page(blob, cluster_num, page);
6703 
6704 	page->crc = _spdk_blob_md_page_calc_crc(page);
6705 
6706 	assert(spdk_bit_array_get(blob->bs->used_md_pages, extent) == true);
6707 
6708 	spdk_bs_sequence_write_dev(seq, page, _spdk_bs_md_page_to_lba(blob->bs, extent),
6709 				   _spdk_bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE),
6710 				   _spdk_blob_persist_extent_page_cpl, page);
6711 }
6712 
6713 static void
6714 _spdk_blob_insert_cluster_msg(void *arg)
6715 {
6716 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
6717 	uint32_t *extent_page;
6718 
6719 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
6720 	if (ctx->rc != 0) {
6721 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
6722 		return;
6723 	}
6724 
6725 	if (ctx->blob->use_extent_table == false) {
6726 		/* Extent table is not used, proceed with sync of md that will only use extents_rle. */
6727 		ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
6728 		_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
6729 		return;
6730 	}
6731 
6732 	extent_page = _spdk_bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num);
6733 	if (*extent_page == 0) {
6734 		/* Extent page requires allocation.
6735 		 * It was already claimed in the used_md_pages map and placed in ctx.
6736 		 * Blob persist will take care of writing out new extent page on disk. */
6737 		assert(ctx->extent_page != 0);
6738 		assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
6739 		*extent_page = ctx->extent_page;
6740 		ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
6741 		_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
6742 	} else {
6743 		/* It is possible for original thread to allocate extent page for
6744 		 * different cluster in the same extent page. In such case proceed with
6745 		 * updating the existing extent page, but release the additional one. */
6746 		if (ctx->extent_page != 0) {
6747 			assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true);
6748 			_spdk_bs_release_md_page(ctx->blob->bs, ctx->extent_page);
6749 		}
6750 		/* Extent page already allocated.
6751 		 * Every cluster allocation, requires just an update of single extent page. */
6752 		_spdk_blob_insert_extent(ctx->blob, *extent_page, ctx->cluster_num,
6753 					 _spdk_blob_insert_cluster_msg_cb, ctx);
6754 	}
6755 }
6756 
6757 static void
6758 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
6759 				       uint64_t cluster, uint32_t extent_page, spdk_blob_op_complete cb_fn, void *cb_arg)
6760 {
6761 	struct spdk_blob_insert_cluster_ctx *ctx;
6762 
6763 	ctx = calloc(1, sizeof(*ctx));
6764 	if (ctx == NULL) {
6765 		cb_fn(cb_arg, -ENOMEM);
6766 		return;
6767 	}
6768 
6769 	ctx->thread = spdk_get_thread();
6770 	ctx->blob = blob;
6771 	ctx->cluster_num = cluster_num;
6772 	ctx->cluster = cluster;
6773 	ctx->extent_page = extent_page;
6774 	ctx->cb_fn = cb_fn;
6775 	ctx->cb_arg = cb_arg;
6776 
6777 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
6778 }
6779 
6780 /* START spdk_blob_close */
6781 
6782 static void
6783 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
6784 {
6785 	struct spdk_blob *blob = cb_arg;
6786 
6787 	if (bserrno == 0) {
6788 		blob->open_ref--;
6789 		if (blob->open_ref == 0) {
6790 			/*
6791 			 * Blobs with active.num_pages == 0 are deleted blobs.
6792 			 *  these blobs are removed from the blob_store list
6793 			 *  when the deletion process starts - so don't try to
6794 			 *  remove them again.
6795 			 */
6796 			if (blob->active.num_pages > 0) {
6797 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
6798 			}
6799 			_spdk_blob_free(blob);
6800 		}
6801 	}
6802 
6803 	spdk_bs_sequence_finish(seq, bserrno);
6804 }
6805 
6806 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
6807 {
6808 	struct spdk_bs_cpl	cpl;
6809 	spdk_bs_sequence_t	*seq;
6810 
6811 	_spdk_blob_verify_md_op(blob);
6812 
6813 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
6814 
6815 	if (blob->open_ref == 0) {
6816 		cb_fn(cb_arg, -EBADF);
6817 		return;
6818 	}
6819 
6820 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
6821 	cpl.u.blob_basic.cb_fn = cb_fn;
6822 	cpl.u.blob_basic.cb_arg = cb_arg;
6823 
6824 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
6825 	if (!seq) {
6826 		cb_fn(cb_arg, -ENOMEM);
6827 		return;
6828 	}
6829 
6830 	/* Sync metadata */
6831 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
6832 }
6833 
6834 /* END spdk_blob_close */
6835 
6836 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
6837 {
6838 	return spdk_get_io_channel(bs);
6839 }
6840 
6841 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
6842 {
6843 	spdk_put_io_channel(channel);
6844 }
6845 
6846 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
6847 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
6848 {
6849 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
6850 				     SPDK_BLOB_UNMAP);
6851 }
6852 
6853 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
6854 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
6855 {
6856 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
6857 				     SPDK_BLOB_WRITE_ZEROES);
6858 }
6859 
6860 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
6861 			void *payload, uint64_t offset, uint64_t length,
6862 			spdk_blob_op_complete cb_fn, void *cb_arg)
6863 {
6864 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
6865 				     SPDK_BLOB_WRITE);
6866 }
6867 
6868 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
6869 		       void *payload, uint64_t offset, uint64_t length,
6870 		       spdk_blob_op_complete cb_fn, void *cb_arg)
6871 {
6872 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
6873 				     SPDK_BLOB_READ);
6874 }
6875 
6876 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
6877 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
6878 			 spdk_blob_op_complete cb_fn, void *cb_arg)
6879 {
6880 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
6881 }
6882 
6883 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
6884 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
6885 			spdk_blob_op_complete cb_fn, void *cb_arg)
6886 {
6887 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
6888 }
6889 
6890 struct spdk_bs_iter_ctx {
6891 	int64_t page_num;
6892 	struct spdk_blob_store *bs;
6893 
6894 	spdk_blob_op_with_handle_complete cb_fn;
6895 	void *cb_arg;
6896 };
6897 
6898 static void
6899 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
6900 {
6901 	struct spdk_bs_iter_ctx *ctx = cb_arg;
6902 	struct spdk_blob_store *bs = ctx->bs;
6903 	spdk_blob_id id;
6904 
6905 	if (bserrno == 0) {
6906 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
6907 		free(ctx);
6908 		return;
6909 	}
6910 
6911 	ctx->page_num++;
6912 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
6913 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
6914 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
6915 		free(ctx);
6916 		return;
6917 	}
6918 
6919 	id = _spdk_bs_page_to_blobid(ctx->page_num);
6920 
6921 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
6922 }
6923 
6924 void
6925 spdk_bs_iter_first(struct spdk_blob_store *bs,
6926 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6927 {
6928 	struct spdk_bs_iter_ctx *ctx;
6929 
6930 	ctx = calloc(1, sizeof(*ctx));
6931 	if (!ctx) {
6932 		cb_fn(cb_arg, NULL, -ENOMEM);
6933 		return;
6934 	}
6935 
6936 	ctx->page_num = -1;
6937 	ctx->bs = bs;
6938 	ctx->cb_fn = cb_fn;
6939 	ctx->cb_arg = cb_arg;
6940 
6941 	_spdk_bs_iter_cpl(ctx, NULL, -1);
6942 }
6943 
6944 static void
6945 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
6946 {
6947 	struct spdk_bs_iter_ctx *ctx = cb_arg;
6948 
6949 	_spdk_bs_iter_cpl(ctx, NULL, -1);
6950 }
6951 
6952 void
6953 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
6954 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
6955 {
6956 	struct spdk_bs_iter_ctx *ctx;
6957 
6958 	assert(blob != NULL);
6959 
6960 	ctx = calloc(1, sizeof(*ctx));
6961 	if (!ctx) {
6962 		cb_fn(cb_arg, NULL, -ENOMEM);
6963 		return;
6964 	}
6965 
6966 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
6967 	ctx->bs = bs;
6968 	ctx->cb_fn = cb_fn;
6969 	ctx->cb_arg = cb_arg;
6970 
6971 	/* Close the existing blob */
6972 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
6973 }
6974 
6975 static int
6976 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
6977 		     uint16_t value_len, bool internal)
6978 {
6979 	struct spdk_xattr_tailq *xattrs;
6980 	struct spdk_xattr	*xattr;
6981 	size_t			desc_size;
6982 
6983 	_spdk_blob_verify_md_op(blob);
6984 
6985 	if (blob->md_ro) {
6986 		return -EPERM;
6987 	}
6988 
6989 	desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len;
6990 	if (desc_size > SPDK_BS_MAX_DESC_SIZE) {
6991 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Xattr '%s' of size %ld does not fix into single page %ld\n", name,
6992 			      desc_size, SPDK_BS_MAX_DESC_SIZE);
6993 		return -ENOMEM;
6994 	}
6995 
6996 	if (internal) {
6997 		xattrs = &blob->xattrs_internal;
6998 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
6999 	} else {
7000 		xattrs = &blob->xattrs;
7001 	}
7002 
7003 	TAILQ_FOREACH(xattr, xattrs, link) {
7004 		if (!strcmp(name, xattr->name)) {
7005 			free(xattr->value);
7006 			xattr->value_len = value_len;
7007 			xattr->value = malloc(value_len);
7008 			memcpy(xattr->value, value, value_len);
7009 
7010 			blob->state = SPDK_BLOB_STATE_DIRTY;
7011 
7012 			return 0;
7013 		}
7014 	}
7015 
7016 	xattr = calloc(1, sizeof(*xattr));
7017 	if (!xattr) {
7018 		return -ENOMEM;
7019 	}
7020 	xattr->name = strdup(name);
7021 	xattr->value_len = value_len;
7022 	xattr->value = malloc(value_len);
7023 	memcpy(xattr->value, value, value_len);
7024 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
7025 
7026 	blob->state = SPDK_BLOB_STATE_DIRTY;
7027 
7028 	return 0;
7029 }
7030 
7031 int
7032 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
7033 		    uint16_t value_len)
7034 {
7035 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
7036 }
7037 
7038 static int
7039 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
7040 {
7041 	struct spdk_xattr_tailq *xattrs;
7042 	struct spdk_xattr	*xattr;
7043 
7044 	_spdk_blob_verify_md_op(blob);
7045 
7046 	if (blob->md_ro) {
7047 		return -EPERM;
7048 	}
7049 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
7050 
7051 	TAILQ_FOREACH(xattr, xattrs, link) {
7052 		if (!strcmp(name, xattr->name)) {
7053 			TAILQ_REMOVE(xattrs, xattr, link);
7054 			free(xattr->value);
7055 			free(xattr->name);
7056 			free(xattr);
7057 
7058 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
7059 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
7060 			}
7061 			blob->state = SPDK_BLOB_STATE_DIRTY;
7062 
7063 			return 0;
7064 		}
7065 	}
7066 
7067 	return -ENOENT;
7068 }
7069 
7070 int
7071 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
7072 {
7073 	return _spdk_blob_remove_xattr(blob, name, false);
7074 }
7075 
7076 static int
7077 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
7078 			   const void **value, size_t *value_len, bool internal)
7079 {
7080 	struct spdk_xattr	*xattr;
7081 	struct spdk_xattr_tailq *xattrs;
7082 
7083 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
7084 
7085 	TAILQ_FOREACH(xattr, xattrs, link) {
7086 		if (!strcmp(name, xattr->name)) {
7087 			*value = xattr->value;
7088 			*value_len = xattr->value_len;
7089 			return 0;
7090 		}
7091 	}
7092 	return -ENOENT;
7093 }
7094 
7095 int
7096 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
7097 			  const void **value, size_t *value_len)
7098 {
7099 	_spdk_blob_verify_md_op(blob);
7100 
7101 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
7102 }
7103 
7104 struct spdk_xattr_names {
7105 	uint32_t	count;
7106 	const char	*names[0];
7107 };
7108 
7109 static int
7110 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
7111 {
7112 	struct spdk_xattr	*xattr;
7113 	int			count = 0;
7114 
7115 	TAILQ_FOREACH(xattr, xattrs, link) {
7116 		count++;
7117 	}
7118 
7119 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
7120 	if (*names == NULL) {
7121 		return -ENOMEM;
7122 	}
7123 
7124 	TAILQ_FOREACH(xattr, xattrs, link) {
7125 		(*names)->names[(*names)->count++] = xattr->name;
7126 	}
7127 
7128 	return 0;
7129 }
7130 
7131 int
7132 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
7133 {
7134 	_spdk_blob_verify_md_op(blob);
7135 
7136 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
7137 }
7138 
7139 uint32_t
7140 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
7141 {
7142 	assert(names != NULL);
7143 
7144 	return names->count;
7145 }
7146 
7147 const char *
7148 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
7149 {
7150 	if (index >= names->count) {
7151 		return NULL;
7152 	}
7153 
7154 	return names->names[index];
7155 }
7156 
7157 void
7158 spdk_xattr_names_free(struct spdk_xattr_names *names)
7159 {
7160 	free(names);
7161 }
7162 
7163 struct spdk_bs_type
7164 spdk_bs_get_bstype(struct spdk_blob_store *bs)
7165 {
7166 	return bs->bstype;
7167 }
7168 
7169 void
7170 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
7171 {
7172 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
7173 }
7174 
7175 bool
7176 spdk_blob_is_read_only(struct spdk_blob *blob)
7177 {
7178 	assert(blob != NULL);
7179 	return (blob->data_ro || blob->md_ro);
7180 }
7181 
7182 bool
7183 spdk_blob_is_snapshot(struct spdk_blob *blob)
7184 {
7185 	struct spdk_blob_list *snapshot_entry;
7186 
7187 	assert(blob != NULL);
7188 
7189 	snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id);
7190 	if (snapshot_entry == NULL) {
7191 		return false;
7192 	}
7193 
7194 	return true;
7195 }
7196 
7197 bool
7198 spdk_blob_is_clone(struct spdk_blob *blob)
7199 {
7200 	assert(blob != NULL);
7201 
7202 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
7203 		assert(spdk_blob_is_thin_provisioned(blob));
7204 		return true;
7205 	}
7206 
7207 	return false;
7208 }
7209 
7210 bool
7211 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
7212 {
7213 	assert(blob != NULL);
7214 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
7215 }
7216 
7217 static void
7218 _spdk_blob_update_clear_method(struct spdk_blob *blob)
7219 {
7220 	enum blob_clear_method stored_cm;
7221 
7222 	assert(blob != NULL);
7223 
7224 	/* If BLOB_CLEAR_WITH_DEFAULT was passed in, use the setting stored
7225 	 * in metadata previously.  If something other than the default was
7226 	 * specified, ignore stored value and used what was passed in.
7227 	 */
7228 	stored_cm = ((blob->md_ro_flags & SPDK_BLOB_CLEAR_METHOD) >> SPDK_BLOB_CLEAR_METHOD_SHIFT);
7229 
7230 	if (blob->clear_method == BLOB_CLEAR_WITH_DEFAULT) {
7231 		blob->clear_method = stored_cm;
7232 	} else if (blob->clear_method != stored_cm) {
7233 		SPDK_WARNLOG("Using passed in clear method 0x%x instead of stored value of 0x%x\n",
7234 			     blob->clear_method, stored_cm);
7235 	}
7236 }
7237 
7238 spdk_blob_id
7239 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
7240 {
7241 	struct spdk_blob_list *snapshot_entry = NULL;
7242 	struct spdk_blob_list *clone_entry = NULL;
7243 
7244 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
7245 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
7246 			if (clone_entry->id == blob_id) {
7247 				return snapshot_entry->id;
7248 			}
7249 		}
7250 	}
7251 
7252 	return SPDK_BLOBID_INVALID;
7253 }
7254 
7255 int
7256 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
7257 		     size_t *count)
7258 {
7259 	struct spdk_blob_list *snapshot_entry, *clone_entry;
7260 	size_t n;
7261 
7262 	snapshot_entry = _spdk_bs_get_snapshot_entry(bs, blobid);
7263 	if (snapshot_entry == NULL) {
7264 		*count = 0;
7265 		return 0;
7266 	}
7267 
7268 	if (ids == NULL || *count < snapshot_entry->clone_count) {
7269 		*count = snapshot_entry->clone_count;
7270 		return -ENOMEM;
7271 	}
7272 	*count = snapshot_entry->clone_count;
7273 
7274 	n = 0;
7275 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
7276 		ids[n++] = clone_entry->id;
7277 	}
7278 
7279 	return 0;
7280 }
7281 
7282 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
7283