xref: /spdk/lib/blob/blobstore.c (revision 0d2745c94b03b159020b6812c6caddb4922e4449)
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 
44 #include "spdk_internal/assert.h"
45 #include "spdk_internal/log.h"
46 
47 #include "blobstore.h"
48 
49 #define BLOB_CRC32C_INITIAL    0xffffffffUL
50 
51 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs);
52 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs);
53 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno);
54 static void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
55 		uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg);
56 
57 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
58 				uint16_t value_len, bool internal);
59 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
60 				      const void **value, size_t *value_len, bool internal);
61 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal);
62 
63 static void
64 _spdk_blob_verify_md_op(struct spdk_blob *blob)
65 {
66 	assert(blob != NULL);
67 	assert(spdk_get_thread() == blob->bs->md_thread);
68 	assert(blob->state != SPDK_BLOB_STATE_LOADING);
69 }
70 
71 static inline size_t
72 divide_round_up(size_t num, size_t divisor)
73 {
74 	return (num + divisor - 1) / divisor;
75 }
76 
77 static void
78 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
79 {
80 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
81 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false);
82 	assert(bs->num_free_clusters > 0);
83 
84 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num);
85 
86 	spdk_bit_array_set(bs->used_clusters, cluster_num);
87 	bs->num_free_clusters--;
88 }
89 
90 static int
91 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster)
92 {
93 	uint64_t *cluster_lba = &blob->active.clusters[cluster_num];
94 
95 	_spdk_blob_verify_md_op(blob);
96 
97 	if (*cluster_lba != 0) {
98 		return -EEXIST;
99 	}
100 
101 	*cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster);
102 	return 0;
103 }
104 
105 static int
106 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num,
107 			  uint64_t *lowest_free_cluster, bool update_map)
108 {
109 	pthread_mutex_lock(&blob->bs->used_clusters_mutex);
110 	*lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters,
111 			       *lowest_free_cluster);
112 	if (*lowest_free_cluster >= blob->bs->total_clusters) {
113 		/* No more free clusters. Cannot satisfy the request */
114 		pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
115 		return -ENOSPC;
116 	}
117 
118 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id);
119 	_spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster);
120 	pthread_mutex_unlock(&blob->bs->used_clusters_mutex);
121 
122 	if (update_map) {
123 		_spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster);
124 	}
125 
126 	return 0;
127 }
128 
129 static void
130 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num)
131 {
132 	assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters));
133 	assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true);
134 	assert(bs->num_free_clusters < bs->total_clusters);
135 
136 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num);
137 
138 	pthread_mutex_lock(&bs->used_clusters_mutex);
139 	spdk_bit_array_clear(bs->used_clusters, cluster_num);
140 	bs->num_free_clusters++;
141 	pthread_mutex_unlock(&bs->used_clusters_mutex);
142 }
143 
144 static void
145 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs)
146 {
147 	xattrs->count = 0;
148 	xattrs->names = NULL;
149 	xattrs->ctx = NULL;
150 	xattrs->get_value = NULL;
151 }
152 
153 void
154 spdk_blob_opts_init(struct spdk_blob_opts *opts)
155 {
156 	opts->num_clusters = 0;
157 	opts->thin_provision = false;
158 	_spdk_blob_xattrs_init(&opts->xattrs);
159 }
160 
161 static struct spdk_blob *
162 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id)
163 {
164 	struct spdk_blob *blob;
165 
166 	blob = calloc(1, sizeof(*blob));
167 	if (!blob) {
168 		return NULL;
169 	}
170 
171 	blob->id = id;
172 	blob->bs = bs;
173 
174 	blob->parent_id = SPDK_BLOBID_INVALID;
175 
176 	blob->state = SPDK_BLOB_STATE_DIRTY;
177 	blob->active.num_pages = 1;
178 	blob->active.pages = calloc(1, sizeof(*blob->active.pages));
179 	if (!blob->active.pages) {
180 		free(blob);
181 		return NULL;
182 	}
183 
184 	blob->active.pages[0] = _spdk_bs_blobid_to_page(id);
185 
186 	TAILQ_INIT(&blob->xattrs);
187 	TAILQ_INIT(&blob->xattrs_internal);
188 
189 	return blob;
190 }
191 
192 static void
193 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs)
194 {
195 	struct spdk_xattr	*xattr, *xattr_tmp;
196 
197 	TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) {
198 		TAILQ_REMOVE(xattrs, xattr, link);
199 		free(xattr->name);
200 		free(xattr->value);
201 		free(xattr);
202 	}
203 }
204 
205 static void
206 _spdk_blob_free(struct spdk_blob *blob)
207 {
208 	assert(blob != NULL);
209 
210 	free(blob->active.clusters);
211 	free(blob->clean.clusters);
212 	free(blob->active.pages);
213 	free(blob->clean.pages);
214 
215 	_spdk_xattrs_free(&blob->xattrs);
216 	_spdk_xattrs_free(&blob->xattrs_internal);
217 
218 	if (blob->back_bs_dev) {
219 		blob->back_bs_dev->destroy(blob->back_bs_dev);
220 	}
221 
222 	free(blob);
223 }
224 
225 struct freeze_io_ctx {
226 	struct spdk_bs_cpl cpl;
227 	struct spdk_blob *blob;
228 };
229 
230 static void
231 _spdk_blob_io_sync(struct spdk_io_channel_iter *i)
232 {
233 	spdk_for_each_channel_continue(i, 0);
234 }
235 
236 static void
237 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i)
238 {
239 	struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i);
240 	struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch);
241 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
242 	struct spdk_bs_request_set	*set;
243 	struct spdk_bs_user_op_args	*args;
244 	spdk_bs_user_op_t *op, *tmp;
245 
246 	TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) {
247 		set = (struct spdk_bs_request_set *)op;
248 		args = &set->u.user_op;
249 
250 		if (args->blob == ctx->blob) {
251 			TAILQ_REMOVE(&ch->queued_io, op, link);
252 			spdk_bs_user_op_execute(op);
253 		}
254 	}
255 
256 	spdk_for_each_channel_continue(i, 0);
257 }
258 
259 static void
260 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status)
261 {
262 	struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i);
263 
264 	ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0);
265 
266 	free(ctx);
267 }
268 
269 static void
270 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
271 {
272 	struct freeze_io_ctx *ctx;
273 
274 	ctx = calloc(1, sizeof(*ctx));
275 	if (!ctx) {
276 		cb_fn(cb_arg, -ENOMEM);
277 		return;
278 	}
279 
280 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
281 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
282 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
283 	ctx->blob = blob;
284 
285 	/* Freeze I/O on blob */
286 	blob->frozen_refcnt++;
287 
288 	if (blob->frozen_refcnt == 1) {
289 		spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl);
290 	} else {
291 		cb_fn(cb_arg, 0);
292 		free(ctx);
293 	}
294 }
295 
296 static void
297 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
298 {
299 	struct freeze_io_ctx *ctx;
300 
301 	ctx = calloc(1, sizeof(*ctx));
302 	if (!ctx) {
303 		cb_fn(cb_arg, -ENOMEM);
304 		return;
305 	}
306 
307 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
308 	ctx->cpl.u.blob_basic.cb_fn = cb_fn;
309 	ctx->cpl.u.blob_basic.cb_arg = cb_arg;
310 	ctx->blob = blob;
311 
312 	assert(blob->frozen_refcnt > 0);
313 
314 	blob->frozen_refcnt--;
315 
316 	if (blob->frozen_refcnt == 0) {
317 		spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl);
318 	} else {
319 		cb_fn(cb_arg, 0);
320 		free(ctx);
321 	}
322 }
323 
324 static int
325 _spdk_blob_mark_clean(struct spdk_blob *blob)
326 {
327 	uint64_t *clusters = NULL;
328 	uint32_t *pages = NULL;
329 
330 	assert(blob != NULL);
331 
332 	if (blob->active.num_clusters) {
333 		assert(blob->active.clusters);
334 		clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters));
335 		if (!clusters) {
336 			return -ENOMEM;
337 		}
338 		memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*clusters));
339 	}
340 
341 	if (blob->active.num_pages) {
342 		assert(blob->active.pages);
343 		pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages));
344 		if (!pages) {
345 			free(clusters);
346 			return -ENOMEM;
347 		}
348 		memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*pages));
349 	}
350 
351 	free(blob->clean.clusters);
352 	free(blob->clean.pages);
353 
354 	blob->clean.num_clusters = blob->active.num_clusters;
355 	blob->clean.clusters = blob->active.clusters;
356 	blob->clean.num_pages = blob->active.num_pages;
357 	blob->clean.pages = blob->active.pages;
358 
359 	blob->active.clusters = clusters;
360 	blob->active.pages = pages;
361 
362 	/* If the metadata was dirtied again while the metadata was being written to disk,
363 	 *  we do not want to revert the DIRTY state back to CLEAN here.
364 	 */
365 	if (blob->state == SPDK_BLOB_STATE_LOADING) {
366 		blob->state = SPDK_BLOB_STATE_CLEAN;
367 	}
368 
369 	return 0;
370 }
371 
372 static int
373 _spdk_blob_deserialize_xattr(struct spdk_blob *blob,
374 			     struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal)
375 {
376 	struct spdk_xattr                       *xattr;
377 
378 	if (desc_xattr->length != sizeof(desc_xattr->name_length) +
379 	    sizeof(desc_xattr->value_length) +
380 	    desc_xattr->name_length + desc_xattr->value_length) {
381 		return -EINVAL;
382 	}
383 
384 	xattr = calloc(1, sizeof(*xattr));
385 	if (xattr == NULL) {
386 		return -ENOMEM;
387 	}
388 
389 	xattr->name = malloc(desc_xattr->name_length + 1);
390 	if (xattr->name == NULL) {
391 		free(xattr);
392 		return -ENOMEM;
393 	}
394 	memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length);
395 	xattr->name[desc_xattr->name_length] = '\0';
396 
397 	xattr->value = malloc(desc_xattr->value_length);
398 	if (xattr->value == NULL) {
399 		free(xattr->name);
400 		free(xattr);
401 		return -ENOMEM;
402 	}
403 	xattr->value_len = desc_xattr->value_length;
404 	memcpy(xattr->value,
405 	       (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
406 	       desc_xattr->value_length);
407 
408 	TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link);
409 
410 	return 0;
411 }
412 
413 
414 static int
415 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob)
416 {
417 	struct spdk_blob_md_descriptor *desc;
418 	size_t	cur_desc = 0;
419 	void *tmp;
420 
421 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
422 	while (cur_desc < sizeof(page->descriptors)) {
423 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
424 			if (desc->length == 0) {
425 				/* If padding and length are 0, this terminates the page */
426 				break;
427 			}
428 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
429 			struct spdk_blob_md_descriptor_flags	*desc_flags;
430 
431 			desc_flags = (struct spdk_blob_md_descriptor_flags *)desc;
432 
433 			if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) {
434 				return -EINVAL;
435 			}
436 
437 			if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) !=
438 			    SPDK_BLOB_INVALID_FLAGS_MASK) {
439 				return -EINVAL;
440 			}
441 
442 			if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) !=
443 			    SPDK_BLOB_DATA_RO_FLAGS_MASK) {
444 				blob->data_ro = true;
445 				blob->md_ro = true;
446 			}
447 
448 			if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) !=
449 			    SPDK_BLOB_MD_RO_FLAGS_MASK) {
450 				blob->md_ro = true;
451 			}
452 
453 			if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
454 				blob->data_ro = true;
455 				blob->md_ro = true;
456 			}
457 
458 			blob->invalid_flags = desc_flags->invalid_flags;
459 			blob->data_ro_flags = desc_flags->data_ro_flags;
460 			blob->md_ro_flags = desc_flags->md_ro_flags;
461 
462 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
463 			struct spdk_blob_md_descriptor_extent	*desc_extent;
464 			unsigned int				i, j;
465 			unsigned int				cluster_count = blob->active.num_clusters;
466 
467 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
468 
469 			if (desc_extent->length == 0 ||
470 			    (desc_extent->length % sizeof(desc_extent->extents[0]) != 0)) {
471 				return -EINVAL;
472 			}
473 
474 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
475 				for (j = 0; j < desc_extent->extents[i].length; j++) {
476 					if (!spdk_bit_array_get(blob->bs->used_clusters,
477 								desc_extent->extents[i].cluster_idx + j)) {
478 						return -EINVAL;
479 					}
480 					cluster_count++;
481 				}
482 			}
483 
484 			if (cluster_count == 0) {
485 				return -EINVAL;
486 			}
487 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t));
488 			if (tmp == NULL) {
489 				return -ENOMEM;
490 			}
491 			blob->active.clusters = tmp;
492 			blob->active.cluster_array_size = cluster_count;
493 
494 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
495 				for (j = 0; j < desc_extent->extents[i].length; j++) {
496 					if (desc_extent->extents[i].cluster_idx != 0) {
497 						blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
498 								desc_extent->extents[i].cluster_idx + j);
499 					} else if (spdk_blob_is_thin_provisioned(blob)) {
500 						blob->active.clusters[blob->active.num_clusters++] = 0;
501 					} else {
502 						return -EINVAL;
503 					}
504 				}
505 			}
506 
507 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
508 			int rc;
509 
510 			rc = _spdk_blob_deserialize_xattr(blob,
511 							  (struct spdk_blob_md_descriptor_xattr *) desc, false);
512 			if (rc != 0) {
513 				return rc;
514 			}
515 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
516 			int rc;
517 
518 			rc = _spdk_blob_deserialize_xattr(blob,
519 							  (struct spdk_blob_md_descriptor_xattr *) desc, true);
520 			if (rc != 0) {
521 				return rc;
522 			}
523 		} else {
524 			/* Unrecognized descriptor type.  Do not fail - just continue to the
525 			 *  next descriptor.  If this descriptor is associated with some feature
526 			 *  defined in a newer version of blobstore, that version of blobstore
527 			 *  should create and set an associated feature flag to specify if this
528 			 *  blob can be loaded or not.
529 			 */
530 		}
531 
532 		/* Advance to the next descriptor */
533 		cur_desc += sizeof(*desc) + desc->length;
534 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
535 			break;
536 		}
537 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
538 	}
539 
540 	return 0;
541 }
542 
543 static int
544 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
545 		 struct spdk_blob *blob)
546 {
547 	const struct spdk_blob_md_page *page;
548 	uint32_t i;
549 	int rc;
550 
551 	assert(page_count > 0);
552 	assert(pages[0].sequence_num == 0);
553 	assert(blob != NULL);
554 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
555 	assert(blob->active.clusters == NULL);
556 
557 	/* The blobid provided doesn't match what's in the MD, this can
558 	 * happen for example if a bogus blobid is passed in through open.
559 	 */
560 	if (blob->id != pages[0].id) {
561 		SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n",
562 			    blob->id, pages[0].id);
563 		return -ENOENT;
564 	}
565 
566 	for (i = 0; i < page_count; i++) {
567 		page = &pages[i];
568 
569 		assert(page->id == blob->id);
570 		assert(page->sequence_num == i);
571 
572 		rc = _spdk_blob_parse_page(page, blob);
573 		if (rc != 0) {
574 			return rc;
575 		}
576 	}
577 
578 	return 0;
579 }
580 
581 static int
582 _spdk_blob_serialize_add_page(const struct spdk_blob *blob,
583 			      struct spdk_blob_md_page **pages,
584 			      uint32_t *page_count,
585 			      struct spdk_blob_md_page **last_page)
586 {
587 	struct spdk_blob_md_page *page;
588 
589 	assert(pages != NULL);
590 	assert(page_count != NULL);
591 
592 	if (*page_count == 0) {
593 		assert(*pages == NULL);
594 		*page_count = 1;
595 		*pages = spdk_dma_malloc(SPDK_BS_PAGE_SIZE,
596 					 SPDK_BS_PAGE_SIZE,
597 					 NULL);
598 	} else {
599 		assert(*pages != NULL);
600 		(*page_count)++;
601 		*pages = spdk_dma_realloc(*pages,
602 					  SPDK_BS_PAGE_SIZE * (*page_count),
603 					  SPDK_BS_PAGE_SIZE,
604 					  NULL);
605 	}
606 
607 	if (*pages == NULL) {
608 		*page_count = 0;
609 		*last_page = NULL;
610 		return -ENOMEM;
611 	}
612 
613 	page = &(*pages)[*page_count - 1];
614 	memset(page, 0, sizeof(*page));
615 	page->id = blob->id;
616 	page->sequence_num = *page_count - 1;
617 	page->next = SPDK_INVALID_MD_PAGE;
618 	*last_page = page;
619 
620 	return 0;
621 }
622 
623 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
624  * Update required_sz on both success and failure.
625  *
626  */
627 static int
628 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
629 			   uint8_t *buf, size_t buf_sz,
630 			   size_t *required_sz, bool internal)
631 {
632 	struct spdk_blob_md_descriptor_xattr	*desc;
633 
634 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
635 		       strlen(xattr->name) +
636 		       xattr->value_len;
637 
638 	if (buf_sz < *required_sz) {
639 		return -1;
640 	}
641 
642 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
643 
644 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
645 	desc->length = sizeof(desc->name_length) +
646 		       sizeof(desc->value_length) +
647 		       strlen(xattr->name) +
648 		       xattr->value_len;
649 	desc->name_length = strlen(xattr->name);
650 	desc->value_length = xattr->value_len;
651 
652 	memcpy(desc->name, xattr->name, desc->name_length);
653 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
654 	       xattr->value,
655 	       desc->value_length);
656 
657 	return 0;
658 }
659 
660 static void
661 _spdk_blob_serialize_extent(const struct spdk_blob *blob,
662 			    uint64_t start_cluster, uint64_t *next_cluster,
663 			    uint8_t *buf, size_t buf_sz)
664 {
665 	struct spdk_blob_md_descriptor_extent *desc;
666 	size_t cur_sz;
667 	uint64_t i, extent_idx;
668 	uint64_t lba, lba_per_cluster, lba_count;
669 
670 	/* The buffer must have room for at least one extent */
671 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]);
672 	if (buf_sz < cur_sz) {
673 		*next_cluster = start_cluster;
674 		return;
675 	}
676 
677 	desc = (struct spdk_blob_md_descriptor_extent *)buf;
678 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT;
679 
680 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
681 
682 	lba = blob->active.clusters[start_cluster];
683 	lba_count = lba_per_cluster;
684 	extent_idx = 0;
685 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
686 		if ((lba + lba_count) == blob->active.clusters[i]) {
687 			lba_count += lba_per_cluster;
688 			continue;
689 		}
690 		desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
691 		desc->extents[extent_idx].length = lba_count / lba_per_cluster;
692 		extent_idx++;
693 
694 		cur_sz += sizeof(desc->extents[extent_idx]);
695 
696 		if (buf_sz < cur_sz) {
697 			/* If we ran out of buffer space, return */
698 			desc->length = sizeof(desc->extents[0]) * extent_idx;
699 			*next_cluster = i;
700 			return;
701 		}
702 
703 		lba = blob->active.clusters[i];
704 		lba_count = lba_per_cluster;
705 	}
706 
707 	desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
708 	desc->extents[extent_idx].length = lba_count / lba_per_cluster;
709 	extent_idx++;
710 
711 	desc->length = sizeof(desc->extents[0]) * extent_idx;
712 	*next_cluster = blob->active.num_clusters;
713 
714 	return;
715 }
716 
717 static void
718 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
719 			   uint8_t *buf, size_t *buf_sz)
720 {
721 	struct spdk_blob_md_descriptor_flags *desc;
722 
723 	/*
724 	 * Flags get serialized first, so we should always have room for the flags
725 	 *  descriptor.
726 	 */
727 	assert(*buf_sz >= sizeof(*desc));
728 
729 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
730 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
731 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
732 	desc->invalid_flags = blob->invalid_flags;
733 	desc->data_ro_flags = blob->data_ro_flags;
734 	desc->md_ro_flags = blob->md_ro_flags;
735 
736 	*buf_sz -= sizeof(*desc);
737 }
738 
739 static int
740 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
741 			    const struct spdk_xattr_tailq *xattrs, bool internal,
742 			    struct spdk_blob_md_page **pages,
743 			    struct spdk_blob_md_page *cur_page,
744 			    uint32_t *page_count, uint8_t **buf,
745 			    size_t *remaining_sz)
746 {
747 	const struct spdk_xattr	*xattr;
748 	int	rc;
749 
750 	TAILQ_FOREACH(xattr, xattrs, link) {
751 		size_t required_sz = 0;
752 
753 		rc = _spdk_blob_serialize_xattr(xattr,
754 						*buf, *remaining_sz,
755 						&required_sz, internal);
756 		if (rc < 0) {
757 			/* Need to add a new page to the chain */
758 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
759 							   &cur_page);
760 			if (rc < 0) {
761 				spdk_dma_free(*pages);
762 				*pages = NULL;
763 				*page_count = 0;
764 				return rc;
765 			}
766 
767 			*buf = (uint8_t *)cur_page->descriptors;
768 			*remaining_sz = sizeof(cur_page->descriptors);
769 
770 			/* Try again */
771 			required_sz = 0;
772 			rc = _spdk_blob_serialize_xattr(xattr,
773 							*buf, *remaining_sz,
774 							&required_sz, internal);
775 
776 			if (rc < 0) {
777 				spdk_dma_free(*pages);
778 				*pages = NULL;
779 				*page_count = 0;
780 				return rc;
781 			}
782 		}
783 
784 		*remaining_sz -= required_sz;
785 		*buf += required_sz;
786 	}
787 
788 	return 0;
789 }
790 
791 static int
792 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
793 		     uint32_t *page_count)
794 {
795 	struct spdk_blob_md_page		*cur_page;
796 	int					rc;
797 	uint8_t					*buf;
798 	size_t					remaining_sz;
799 	uint64_t				last_cluster;
800 
801 	assert(pages != NULL);
802 	assert(page_count != NULL);
803 	assert(blob != NULL);
804 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
805 
806 	*pages = NULL;
807 	*page_count = 0;
808 
809 	/* A blob always has at least 1 page, even if it has no descriptors */
810 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
811 	if (rc < 0) {
812 		return rc;
813 	}
814 
815 	buf = (uint8_t *)cur_page->descriptors;
816 	remaining_sz = sizeof(cur_page->descriptors);
817 
818 	/* Serialize flags */
819 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
820 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
821 
822 	/* Serialize xattrs */
823 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
824 					 pages, cur_page, page_count, &buf, &remaining_sz);
825 	if (rc < 0) {
826 		return rc;
827 	}
828 
829 	/* Serialize internal xattrs */
830 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
831 					 pages, cur_page, page_count, &buf, &remaining_sz);
832 	if (rc < 0) {
833 		return rc;
834 	}
835 
836 	/* Serialize extents */
837 	last_cluster = 0;
838 	while (last_cluster < blob->active.num_clusters) {
839 		_spdk_blob_serialize_extent(blob, last_cluster, &last_cluster,
840 					    buf, remaining_sz);
841 
842 		if (last_cluster == blob->active.num_clusters) {
843 			break;
844 		}
845 
846 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
847 						   &cur_page);
848 		if (rc < 0) {
849 			return rc;
850 		}
851 
852 		buf = (uint8_t *)cur_page->descriptors;
853 		remaining_sz = sizeof(cur_page->descriptors);
854 	}
855 
856 	return 0;
857 }
858 
859 struct spdk_blob_load_ctx {
860 	struct spdk_blob		*blob;
861 
862 	struct spdk_blob_md_page	*pages;
863 	uint32_t			num_pages;
864 	spdk_bs_sequence_t	        *seq;
865 
866 	spdk_bs_sequence_cpl		cb_fn;
867 	void				*cb_arg;
868 };
869 
870 static uint32_t
871 _spdk_blob_md_page_calc_crc(void *page)
872 {
873 	uint32_t		crc;
874 
875 	crc = BLOB_CRC32C_INITIAL;
876 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
877 	crc ^= BLOB_CRC32C_INITIAL;
878 
879 	return crc;
880 
881 }
882 
883 static void
884 _spdk_blob_load_final(void *cb_arg, int bserrno)
885 {
886 	struct spdk_blob_load_ctx	*ctx = cb_arg;
887 	struct spdk_blob		*blob = ctx->blob;
888 
889 	_spdk_blob_mark_clean(blob);
890 
891 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
892 
893 	/* Free the memory */
894 	spdk_dma_free(ctx->pages);
895 	free(ctx);
896 }
897 
898 static void
899 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
900 {
901 	struct spdk_blob_load_ctx	*ctx = cb_arg;
902 	struct spdk_blob		*blob = ctx->blob;
903 
904 	if (bserrno != 0) {
905 		goto error;
906 	}
907 
908 	blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
909 
910 	if (blob->back_bs_dev == NULL) {
911 		bserrno = -ENOMEM;
912 		goto error;
913 	}
914 
915 	_spdk_blob_load_final(ctx, bserrno);
916 	return;
917 
918 error:
919 	SPDK_ERRLOG("Snapshot fail\n");
920 	_spdk_blob_free(blob);
921 	ctx->cb_fn(ctx->seq, NULL, bserrno);
922 	spdk_dma_free(ctx->pages);
923 	free(ctx);
924 }
925 
926 static void
927 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
928 {
929 	struct spdk_blob_load_ctx	*ctx = cb_arg;
930 	struct spdk_blob		*blob = ctx->blob;
931 	struct spdk_blob_md_page	*page;
932 	const void			*value;
933 	size_t				len;
934 	int				rc;
935 	uint32_t			crc;
936 
937 	page = &ctx->pages[ctx->num_pages - 1];
938 	crc = _spdk_blob_md_page_calc_crc(page);
939 	if (crc != page->crc) {
940 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
941 		_spdk_blob_free(blob);
942 		ctx->cb_fn(seq, NULL, -EINVAL);
943 		spdk_dma_free(ctx->pages);
944 		free(ctx);
945 		return;
946 	}
947 
948 	if (page->next != SPDK_INVALID_MD_PAGE) {
949 		uint32_t next_page = page->next;
950 		uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page);
951 
952 
953 		assert(next_lba < (blob->bs->md_start + blob->bs->md_len));
954 
955 		/* Read the next page */
956 		ctx->num_pages++;
957 		ctx->pages = spdk_dma_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
958 					      sizeof(*page), NULL);
959 		if (ctx->pages == NULL) {
960 			ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM);
961 			free(ctx);
962 			return;
963 		}
964 
965 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
966 					  next_lba,
967 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
968 					  _spdk_blob_load_cpl, ctx);
969 		return;
970 	}
971 
972 	/* Parse the pages */
973 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
974 	if (rc) {
975 		_spdk_blob_free(blob);
976 		ctx->cb_fn(seq, NULL, rc);
977 		spdk_dma_free(ctx->pages);
978 		free(ctx);
979 		return;
980 	}
981 	ctx->seq = seq;
982 
983 
984 	if (spdk_blob_is_thin_provisioned(blob)) {
985 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
986 		if (rc == 0) {
987 			if (len != sizeof(spdk_blob_id)) {
988 				_spdk_blob_free(blob);
989 				ctx->cb_fn(seq, NULL, -EINVAL);
990 				spdk_dma_free(ctx->pages);
991 				free(ctx);
992 				return;
993 			}
994 			/* open snapshot blob and continue in the callback function */
995 			blob->parent_id = *(spdk_blob_id *)value;
996 			spdk_bs_open_blob(blob->bs, blob->parent_id,
997 					  _spdk_blob_load_snapshot_cpl, ctx);
998 			return;
999 		} else {
1000 			/* add zeroes_dev for thin provisioned blob */
1001 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1002 		}
1003 	} else {
1004 		/* standard blob */
1005 		blob->back_bs_dev = NULL;
1006 	}
1007 	_spdk_blob_load_final(ctx, bserrno);
1008 }
1009 
1010 /* Load a blob from disk given a blobid */
1011 static void
1012 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1013 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1014 {
1015 	struct spdk_blob_load_ctx *ctx;
1016 	struct spdk_blob_store *bs;
1017 	uint32_t page_num;
1018 	uint64_t lba;
1019 
1020 	_spdk_blob_verify_md_op(blob);
1021 
1022 	bs = blob->bs;
1023 
1024 	ctx = calloc(1, sizeof(*ctx));
1025 	if (!ctx) {
1026 		cb_fn(seq, cb_arg, -ENOMEM);
1027 		return;
1028 	}
1029 
1030 	ctx->blob = blob;
1031 	ctx->pages = spdk_dma_realloc(ctx->pages, SPDK_BS_PAGE_SIZE,
1032 				      SPDK_BS_PAGE_SIZE, NULL);
1033 	if (!ctx->pages) {
1034 		free(ctx);
1035 		cb_fn(seq, cb_arg, -ENOMEM);
1036 		return;
1037 	}
1038 	ctx->num_pages = 1;
1039 	ctx->cb_fn = cb_fn;
1040 	ctx->cb_arg = cb_arg;
1041 
1042 	page_num = _spdk_bs_blobid_to_page(blob->id);
1043 	lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num);
1044 
1045 	blob->state = SPDK_BLOB_STATE_LOADING;
1046 
1047 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1048 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1049 				  _spdk_blob_load_cpl, ctx);
1050 }
1051 
1052 struct spdk_blob_persist_ctx {
1053 	struct spdk_blob		*blob;
1054 
1055 	struct spdk_bs_super_block	*super;
1056 
1057 	struct spdk_blob_md_page	*pages;
1058 
1059 	uint64_t			idx;
1060 
1061 	spdk_bs_sequence_t		*seq;
1062 	spdk_bs_sequence_cpl		cb_fn;
1063 	void				*cb_arg;
1064 };
1065 
1066 static void
1067 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1068 {
1069 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1070 	struct spdk_blob		*blob = ctx->blob;
1071 
1072 	if (bserrno == 0) {
1073 		_spdk_blob_mark_clean(blob);
1074 	}
1075 
1076 	/* Call user callback */
1077 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1078 
1079 	/* Free the memory */
1080 	spdk_dma_free(ctx->pages);
1081 	free(ctx);
1082 }
1083 
1084 static void
1085 _spdk_blob_persist_unmap_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1086 {
1087 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1088 	struct spdk_blob		*blob = ctx->blob;
1089 	struct spdk_blob_store		*bs = blob->bs;
1090 	void				*tmp;
1091 	size_t				i;
1092 
1093 	/* Release all clusters that were truncated */
1094 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1095 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1096 
1097 		/* Nothing to release if it was not allocated */
1098 		if (blob->active.clusters[i] != 0) {
1099 			_spdk_bs_release_cluster(bs, cluster_num);
1100 		}
1101 	}
1102 
1103 	if (blob->active.num_clusters == 0) {
1104 		free(blob->active.clusters);
1105 		blob->active.clusters = NULL;
1106 		blob->active.cluster_array_size = 0;
1107 	} else {
1108 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters);
1109 		assert(tmp != NULL);
1110 		blob->active.clusters = tmp;
1111 		blob->active.cluster_array_size = blob->active.num_clusters;
1112 	}
1113 
1114 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1115 }
1116 
1117 static void
1118 _spdk_blob_persist_unmap_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1119 {
1120 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1121 	struct spdk_blob		*blob = ctx->blob;
1122 	struct spdk_blob_store		*bs = blob->bs;
1123 	spdk_bs_batch_t			*batch;
1124 	size_t				i;
1125 	uint64_t			lba;
1126 	uint32_t			lba_count;
1127 
1128 	/* Clusters don't move around in blobs. The list shrinks or grows
1129 	 * at the end, but no changes ever occur in the middle of the list.
1130 	 */
1131 
1132 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_unmap_clusters_cpl, ctx);
1133 
1134 	/* Unmap all clusters that were truncated */
1135 	lba = 0;
1136 	lba_count = 0;
1137 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1138 		uint64_t next_lba = blob->active.clusters[i];
1139 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1140 
1141 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1142 			/* This cluster is contiguous with the previous one. */
1143 			lba_count += next_lba_count;
1144 			continue;
1145 		}
1146 
1147 		/* This cluster is not contiguous with the previous one. */
1148 
1149 		/* If a run of LBAs previously existing, send them
1150 		 * as an unmap.
1151 		 */
1152 		if (lba_count > 0) {
1153 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1154 		}
1155 
1156 		/* Start building the next batch */
1157 		lba = next_lba;
1158 		if (next_lba > 0) {
1159 			lba_count = next_lba_count;
1160 		} else {
1161 			lba_count = 0;
1162 		}
1163 	}
1164 
1165 	/* If we ended with a contiguous set of LBAs, send the unmap now */
1166 	if (lba_count > 0) {
1167 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1168 	}
1169 
1170 	spdk_bs_batch_close(batch);
1171 }
1172 
1173 static void
1174 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1175 {
1176 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1177 	struct spdk_blob		*blob = ctx->blob;
1178 	struct spdk_blob_store		*bs = blob->bs;
1179 	size_t				i;
1180 
1181 	/* This loop starts at 1 because the first page is special and handled
1182 	 * below. The pages (except the first) are never written in place,
1183 	 * so any pages in the clean list must be zeroed.
1184 	 */
1185 	for (i = 1; i < blob->clean.num_pages; i++) {
1186 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1187 	}
1188 
1189 	if (blob->active.num_pages == 0) {
1190 		uint32_t page_num;
1191 
1192 		page_num = _spdk_bs_blobid_to_page(blob->id);
1193 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1194 	}
1195 
1196 	/* Move on to unmapping clusters */
1197 	_spdk_blob_persist_unmap_clusters(seq, ctx, 0);
1198 }
1199 
1200 static void
1201 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1202 {
1203 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1204 	struct spdk_blob		*blob = ctx->blob;
1205 	struct spdk_blob_store		*bs = blob->bs;
1206 	uint64_t			lba;
1207 	uint32_t			lba_count;
1208 	spdk_bs_batch_t			*batch;
1209 	size_t				i;
1210 
1211 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1212 
1213 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1214 
1215 	/* This loop starts at 1 because the first page is special and handled
1216 	 * below. The pages (except the first) are never written in place,
1217 	 * so any pages in the clean list must be zeroed.
1218 	 */
1219 	for (i = 1; i < blob->clean.num_pages; i++) {
1220 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);
1221 
1222 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1223 	}
1224 
1225 	/* The first page will only be zeroed if this is a delete. */
1226 	if (blob->active.num_pages == 0) {
1227 		uint32_t page_num;
1228 
1229 		/* The first page in the metadata goes where the blobid indicates */
1230 		page_num = _spdk_bs_blobid_to_page(blob->id);
1231 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);
1232 
1233 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1234 	}
1235 
1236 	spdk_bs_batch_close(batch);
1237 }
1238 
1239 static void
1240 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1241 {
1242 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1243 	struct spdk_blob		*blob = ctx->blob;
1244 	struct spdk_blob_store		*bs = blob->bs;
1245 	uint64_t			lba;
1246 	uint32_t			lba_count;
1247 	struct spdk_blob_md_page	*page;
1248 
1249 	if (blob->active.num_pages == 0) {
1250 		/* Move on to the next step */
1251 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1252 		return;
1253 	}
1254 
1255 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1256 
1257 	page = &ctx->pages[0];
1258 	/* The first page in the metadata goes where the blobid indicates */
1259 	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));
1260 
1261 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1262 				   _spdk_blob_persist_zero_pages, ctx);
1263 }
1264 
1265 static void
1266 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1267 {
1268 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1269 	struct spdk_blob		*blob = ctx->blob;
1270 	struct spdk_blob_store		*bs = blob->bs;
1271 	uint64_t			lba;
1272 	uint32_t			lba_count;
1273 	struct spdk_blob_md_page	*page;
1274 	spdk_bs_batch_t			*batch;
1275 	size_t				i;
1276 
1277 	/* Clusters don't move around in blobs. The list shrinks or grows
1278 	 * at the end, but no changes ever occur in the middle of the list.
1279 	 */
1280 
1281 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1282 
1283 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1284 
1285 	/* This starts at 1. The root page is not written until
1286 	 * all of the others are finished
1287 	 */
1288 	for (i = 1; i < blob->active.num_pages; i++) {
1289 		page = &ctx->pages[i];
1290 		assert(page->sequence_num == i);
1291 
1292 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]);
1293 
1294 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1295 	}
1296 
1297 	spdk_bs_batch_close(batch);
1298 }
1299 
1300 static int
1301 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1302 {
1303 	uint64_t	i;
1304 	uint64_t	*tmp;
1305 	uint64_t	lfc; /* lowest free cluster */
1306 	uint64_t	num_clusters;
1307 	struct spdk_blob_store *bs;
1308 
1309 	bs = blob->bs;
1310 
1311 	_spdk_blob_verify_md_op(blob);
1312 
1313 	if (blob->active.num_clusters == sz) {
1314 		return 0;
1315 	}
1316 
1317 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1318 		/* If this blob was resized to be larger, then smaller, then
1319 		 * larger without syncing, then the cluster array already
1320 		 * contains spare assigned clusters we can use.
1321 		 */
1322 		num_clusters = spdk_min(blob->active.cluster_array_size,
1323 					sz);
1324 	} else {
1325 		num_clusters = blob->active.num_clusters;
1326 	}
1327 
1328 	/* Do two passes - one to verify that we can obtain enough clusters
1329 	 * and another to actually claim them.
1330 	 */
1331 
1332 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1333 		lfc = 0;
1334 		for (i = num_clusters; i < sz; i++) {
1335 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1336 			if (lfc >= bs->total_clusters) {
1337 				/* No more free clusters. Cannot satisfy the request */
1338 				return -ENOSPC;
1339 			}
1340 			lfc++;
1341 		}
1342 	}
1343 
1344 	if (sz > num_clusters) {
1345 		/* Expand the cluster array if necessary.
1346 		 * We only shrink the array when persisting.
1347 		 */
1348 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz);
1349 		if (sz > 0 && tmp == NULL) {
1350 			return -ENOMEM;
1351 		}
1352 		memset(tmp + blob->active.cluster_array_size, 0,
1353 		       sizeof(uint64_t) * (sz - blob->active.cluster_array_size));
1354 		blob->active.clusters = tmp;
1355 		blob->active.cluster_array_size = sz;
1356 	}
1357 
1358 	blob->state = SPDK_BLOB_STATE_DIRTY;
1359 
1360 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1361 		lfc = 0;
1362 		for (i = num_clusters; i < sz; i++) {
1363 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1364 			lfc++;
1365 		}
1366 	}
1367 
1368 	blob->active.num_clusters = sz;
1369 
1370 	return 0;
1371 }
1372 
1373 static void
1374 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1375 {
1376 	spdk_bs_sequence_t *seq = ctx->seq;
1377 	struct spdk_blob *blob = ctx->blob;
1378 	struct spdk_blob_store *bs = blob->bs;
1379 	uint64_t i;
1380 	uint32_t page_num;
1381 	void *tmp;
1382 	int rc;
1383 
1384 	if (blob->active.num_pages == 0) {
1385 		/* This is the signal that the blob should be deleted.
1386 		 * Immediately jump to the clean up routine. */
1387 		assert(blob->clean.num_pages > 0);
1388 		ctx->idx = blob->clean.num_pages - 1;
1389 		blob->state = SPDK_BLOB_STATE_CLEAN;
1390 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1391 		return;
1392 
1393 	}
1394 
1395 	/* Generate the new metadata */
1396 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1397 	if (rc < 0) {
1398 		_spdk_blob_persist_complete(seq, ctx, rc);
1399 		return;
1400 	}
1401 
1402 	assert(blob->active.num_pages >= 1);
1403 
1404 	/* Resize the cache of page indices */
1405 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1406 	if (!tmp) {
1407 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1408 		return;
1409 	}
1410 	blob->active.pages = tmp;
1411 
1412 	/* Assign this metadata to pages. This requires two passes -
1413 	 * one to verify that there are enough pages and a second
1414 	 * to actually claim them. */
1415 	page_num = 0;
1416 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1417 	for (i = 1; i < blob->active.num_pages; i++) {
1418 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1419 		if (page_num >= spdk_bit_array_capacity(bs->used_md_pages)) {
1420 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1421 			return;
1422 		}
1423 		page_num++;
1424 	}
1425 
1426 	page_num = 0;
1427 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1428 	for (i = 1; i < blob->active.num_pages; i++) {
1429 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1430 		ctx->pages[i - 1].next = page_num;
1431 		/* Now that previous metadata page is complete, calculate the crc for it. */
1432 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1433 		blob->active.pages[i] = page_num;
1434 		spdk_bit_array_set(bs->used_md_pages, page_num);
1435 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1436 		page_num++;
1437 	}
1438 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1439 	/* Start writing the metadata from last page to first */
1440 	ctx->idx = blob->active.num_pages - 1;
1441 	blob->state = SPDK_BLOB_STATE_CLEAN;
1442 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1443 }
1444 
1445 static void
1446 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1447 {
1448 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1449 
1450 	ctx->blob->bs->clean = 0;
1451 
1452 	spdk_dma_free(ctx->super);
1453 
1454 	_spdk_blob_persist_start(ctx);
1455 }
1456 
1457 static void
1458 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1459 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1460 
1461 
1462 static void
1463 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1464 {
1465 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1466 
1467 	ctx->super->clean = 0;
1468 	if (ctx->super->size == 0) {
1469 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1470 	}
1471 
1472 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1473 }
1474 
1475 
1476 /* Write a blob to disk */
1477 static void
1478 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1479 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1480 {
1481 	struct spdk_blob_persist_ctx *ctx;
1482 
1483 	_spdk_blob_verify_md_op(blob);
1484 
1485 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1486 		cb_fn(seq, cb_arg, 0);
1487 		return;
1488 	}
1489 
1490 	ctx = calloc(1, sizeof(*ctx));
1491 	if (!ctx) {
1492 		cb_fn(seq, cb_arg, -ENOMEM);
1493 		return;
1494 	}
1495 	ctx->blob = blob;
1496 	ctx->seq = seq;
1497 	ctx->cb_fn = cb_fn;
1498 	ctx->cb_arg = cb_arg;
1499 
1500 	if (blob->bs->clean) {
1501 		ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
1502 		if (!ctx->super) {
1503 			cb_fn(seq, cb_arg, -ENOMEM);
1504 			free(ctx);
1505 			return;
1506 		}
1507 
1508 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1509 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1510 					  _spdk_blob_persist_dirty, ctx);
1511 	} else {
1512 		_spdk_blob_persist_start(ctx);
1513 	}
1514 }
1515 
1516 struct spdk_blob_copy_cluster_ctx {
1517 	struct spdk_blob *blob;
1518 	uint8_t *buf;
1519 	uint64_t page;
1520 	uint64_t new_cluster;
1521 	spdk_bs_sequence_t *seq;
1522 };
1523 
1524 static void
1525 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1526 {
1527 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1528 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1529 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1530 	spdk_bs_user_op_t *op;
1531 
1532 	TAILQ_INIT(&requests);
1533 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1534 
1535 	while (!TAILQ_EMPTY(&requests)) {
1536 		op = TAILQ_FIRST(&requests);
1537 		TAILQ_REMOVE(&requests, op, link);
1538 		if (bserrno == 0) {
1539 			spdk_bs_user_op_execute(op);
1540 		} else {
1541 			spdk_bs_user_op_abort(op);
1542 		}
1543 	}
1544 
1545 	spdk_dma_free(ctx->buf);
1546 	free(ctx);
1547 }
1548 
1549 static void
1550 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1551 {
1552 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1553 
1554 	if (bserrno) {
1555 		uint32_t cluster_number;
1556 
1557 		if (bserrno == -EEXIST) {
1558 			/* The metadata insert failed because another thread
1559 			 * allocated the cluster first. Free our cluster
1560 			 * but continue without error. */
1561 			bserrno = 0;
1562 		}
1563 
1564 		cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1565 		_spdk_bs_release_cluster(ctx->blob->bs, cluster_number);
1566 	}
1567 
1568 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1569 }
1570 
1571 static void
1572 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1573 {
1574 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1575 	uint32_t cluster_number;
1576 
1577 	if (bserrno) {
1578 		/* The write failed, so jump to the final completion handler */
1579 		spdk_bs_sequence_finish(seq, bserrno);
1580 		return;
1581 	}
1582 
1583 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1584 
1585 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1586 					       _spdk_blob_insert_cluster_cpl, ctx);
1587 }
1588 
1589 static void
1590 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1591 {
1592 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1593 
1594 	if (bserrno != 0) {
1595 		/* The read failed, so jump to the final completion handler */
1596 		spdk_bs_sequence_finish(seq, bserrno);
1597 		return;
1598 	}
1599 
1600 	/* Write whole cluster */
1601 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1602 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1603 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1604 				   _spdk_blob_write_copy_cpl, ctx);
1605 }
1606 
1607 static void
1608 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1609 				   struct spdk_io_channel *_ch,
1610 				   uint64_t offset, spdk_bs_user_op_t *op)
1611 {
1612 	struct spdk_bs_cpl cpl;
1613 	struct spdk_bs_channel *ch;
1614 	struct spdk_blob_copy_cluster_ctx *ctx;
1615 	uint32_t cluster_start_page;
1616 	uint32_t cluster_number;
1617 	int rc;
1618 
1619 	ch = spdk_io_channel_get_ctx(_ch);
1620 
1621 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1622 		/* There are already operations pending. Queue this user op
1623 		 * and return because it will be re-executed when the outstanding
1624 		 * cluster allocation completes. */
1625 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1626 		return;
1627 	}
1628 
1629 	/* Round the page offset down to the first page in the cluster */
1630 	cluster_start_page = _spdk_bs_page_to_cluster_start(blob, offset);
1631 
1632 	/* Calculate which index in the metadata cluster array the corresponding
1633 	 * cluster is supposed to be at. */
1634 	cluster_number = _spdk_bs_page_to_cluster(blob->bs, cluster_start_page);
1635 
1636 	ctx = calloc(1, sizeof(*ctx));
1637 	if (!ctx) {
1638 		spdk_bs_user_op_abort(op);
1639 		return;
1640 	}
1641 
1642 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1643 
1644 	ctx->blob = blob;
1645 	ctx->page = cluster_start_page;
1646 
1647 	ctx->buf = spdk_dma_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, NULL);
1648 	if (!ctx->buf) {
1649 		SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1650 			    blob->bs->cluster_sz);
1651 		free(ctx);
1652 		spdk_bs_user_op_abort(op);
1653 		return;
1654 	}
1655 
1656 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1657 	if (rc != 0) {
1658 		spdk_dma_free(ctx->buf);
1659 		free(ctx);
1660 		spdk_bs_user_op_abort(op);
1661 		return;
1662 	}
1663 
1664 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1665 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1666 	cpl.u.blob_basic.cb_arg = ctx;
1667 
1668 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1669 	if (!ctx->seq) {
1670 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1671 		spdk_dma_free(ctx->buf);
1672 		free(ctx);
1673 		spdk_bs_user_op_abort(op);
1674 		return;
1675 	}
1676 
1677 	/* Queue the user op to block other incoming operations */
1678 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1679 
1680 	/* Read cluster from backing device */
1681 	spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1682 				     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1683 				     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1684 				     _spdk_blob_write_copy, ctx);
1685 }
1686 
1687 static void
1688 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t page, uint64_t length,
1689 				       uint64_t *lba,	uint32_t *lba_count)
1690 {
1691 	*lba_count = _spdk_bs_page_to_lba(blob->bs, length);
1692 
1693 	if (!_spdk_bs_page_is_allocated(blob, page)) {
1694 		assert(blob->back_bs_dev != NULL);
1695 		*lba = _spdk_bs_dev_page_to_lba(blob->back_bs_dev, page);
1696 		*lba_count = _spdk_bs_blob_lba_to_back_dev_lba(blob, *lba_count);
1697 	} else {
1698 		*lba = _spdk_bs_blob_page_to_lba(blob, page);
1699 	}
1700 }
1701 
1702 struct op_split_ctx {
1703 	struct spdk_blob *blob;
1704 	struct spdk_io_channel *channel;
1705 	uint64_t page_offset;
1706 	uint64_t pages_remaining;
1707 	void *curr_payload;
1708 	enum spdk_blob_op_type op_type;
1709 	spdk_bs_sequence_t *seq;
1710 };
1711 
1712 static void
1713 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1714 {
1715 	struct op_split_ctx	*ctx = cb_arg;
1716 	struct spdk_blob	*blob = ctx->blob;
1717 	struct spdk_io_channel	*ch = ctx->channel;
1718 	enum spdk_blob_op_type	op_type = ctx->op_type;
1719 	uint8_t			*buf = ctx->curr_payload;
1720 	uint64_t		offset = ctx->page_offset;
1721 	uint64_t		length = ctx->pages_remaining;
1722 	uint64_t		op_length;
1723 
1724 	if (bserrno != 0 || ctx->pages_remaining == 0) {
1725 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1726 		free(ctx);
1727 		return;
1728 	}
1729 
1730 	op_length = spdk_min(length, _spdk_bs_num_pages_to_cluster_boundary(blob, offset));
1731 
1732 	/* Update length and payload for next operation */
1733 	ctx->pages_remaining -= op_length;
1734 	ctx->page_offset += op_length;
1735 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1736 		ctx->curr_payload += (op_length * SPDK_BS_PAGE_SIZE);
1737 	}
1738 
1739 	switch (op_type) {
1740 	case SPDK_BLOB_READ:
1741 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1742 				  _spdk_blob_request_submit_op_split_next, ctx);
1743 		break;
1744 	case SPDK_BLOB_WRITE:
1745 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1746 				   _spdk_blob_request_submit_op_split_next, ctx);
1747 		break;
1748 	case SPDK_BLOB_UNMAP:
1749 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1750 				   _spdk_blob_request_submit_op_split_next, ctx);
1751 		break;
1752 	case SPDK_BLOB_WRITE_ZEROES:
1753 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1754 					  _spdk_blob_request_submit_op_split_next, ctx);
1755 		break;
1756 	case SPDK_BLOB_READV:
1757 	case SPDK_BLOB_WRITEV:
1758 		SPDK_ERRLOG("readv/write not valid for %s\n", __func__);
1759 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1760 		free(ctx);
1761 		break;
1762 	}
1763 }
1764 
1765 static void
1766 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1767 				   void *payload, uint64_t offset, uint64_t length,
1768 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1769 {
1770 	struct op_split_ctx *ctx;
1771 	spdk_bs_sequence_t *seq;
1772 	struct spdk_bs_cpl cpl;
1773 
1774 	assert(blob != NULL);
1775 
1776 	ctx = calloc(1, sizeof(struct op_split_ctx));
1777 	if (ctx == NULL) {
1778 		cb_fn(cb_arg, -ENOMEM);
1779 		return;
1780 	}
1781 
1782 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1783 	cpl.u.blob_basic.cb_fn = cb_fn;
1784 	cpl.u.blob_basic.cb_arg = cb_arg;
1785 
1786 	seq = spdk_bs_sequence_start(ch, &cpl);
1787 	if (!seq) {
1788 		free(ctx);
1789 		cb_fn(cb_arg, -ENOMEM);
1790 		return;
1791 	}
1792 
1793 	ctx->blob = blob;
1794 	ctx->channel = ch;
1795 	ctx->curr_payload = payload;
1796 	ctx->page_offset = offset;
1797 	ctx->pages_remaining = length;
1798 	ctx->op_type = op_type;
1799 	ctx->seq = seq;
1800 
1801 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1802 }
1803 
1804 static void
1805 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1806 				    void *payload, uint64_t offset, uint64_t length,
1807 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1808 {
1809 	struct spdk_bs_cpl cpl;
1810 	uint64_t lba;
1811 	uint32_t lba_count;
1812 
1813 	assert(blob != NULL);
1814 
1815 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1816 	cpl.u.blob_basic.cb_fn = cb_fn;
1817 	cpl.u.blob_basic.cb_arg = cb_arg;
1818 
1819 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1820 
1821 	if (blob->frozen_refcnt) {
1822 		/* This blob I/O is frozen */
1823 		spdk_bs_user_op_t *op;
1824 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1825 
1826 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1827 		if (!op) {
1828 			cb_fn(cb_arg, -ENOMEM);
1829 			return;
1830 		}
1831 
1832 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1833 
1834 		return;
1835 	}
1836 
1837 	switch (op_type) {
1838 	case SPDK_BLOB_READ: {
1839 		spdk_bs_batch_t *batch;
1840 
1841 		batch = spdk_bs_batch_open(_ch, &cpl);
1842 		if (!batch) {
1843 			cb_fn(cb_arg, -ENOMEM);
1844 			return;
1845 		}
1846 
1847 		if (_spdk_bs_page_is_allocated(blob, offset)) {
1848 			/* Read from the blob */
1849 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1850 		} else {
1851 			/* Read from the backing block device */
1852 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1853 		}
1854 
1855 		spdk_bs_batch_close(batch);
1856 		break;
1857 	}
1858 	case SPDK_BLOB_WRITE:
1859 	case SPDK_BLOB_WRITE_ZEROES: {
1860 		if (_spdk_bs_page_is_allocated(blob, offset)) {
1861 			/* Write to the blob */
1862 			spdk_bs_batch_t *batch;
1863 
1864 			if (lba_count == 0) {
1865 				cb_fn(cb_arg, 0);
1866 				return;
1867 			}
1868 
1869 			batch = spdk_bs_batch_open(_ch, &cpl);
1870 			if (!batch) {
1871 				cb_fn(cb_arg, -ENOMEM);
1872 				return;
1873 			}
1874 
1875 			if (op_type == SPDK_BLOB_WRITE) {
1876 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1877 			} else {
1878 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1879 			}
1880 
1881 			spdk_bs_batch_close(batch);
1882 		} else {
1883 			/* Queue this operation and allocate the cluster */
1884 			spdk_bs_user_op_t *op;
1885 
1886 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1887 			if (!op) {
1888 				cb_fn(cb_arg, -ENOMEM);
1889 				return;
1890 			}
1891 
1892 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1893 		}
1894 		break;
1895 	}
1896 	case SPDK_BLOB_UNMAP: {
1897 		spdk_bs_batch_t *batch;
1898 
1899 		batch = spdk_bs_batch_open(_ch, &cpl);
1900 		if (!batch) {
1901 			cb_fn(cb_arg, -ENOMEM);
1902 			return;
1903 		}
1904 
1905 		if (_spdk_bs_page_is_allocated(blob, offset)) {
1906 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1907 		}
1908 
1909 		spdk_bs_batch_close(batch);
1910 		break;
1911 	}
1912 	case SPDK_BLOB_READV:
1913 	case SPDK_BLOB_WRITEV:
1914 		SPDK_ERRLOG("readv/write not valid\n");
1915 		cb_fn(cb_arg, -EINVAL);
1916 		break;
1917 	}
1918 }
1919 
1920 static void
1921 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1922 			     void *payload, uint64_t offset, uint64_t length,
1923 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1924 {
1925 	assert(blob != NULL);
1926 
1927 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1928 		cb_fn(cb_arg, -EPERM);
1929 		return;
1930 	}
1931 
1932 	if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) {
1933 		cb_fn(cb_arg, -EINVAL);
1934 		return;
1935 	}
1936 
1937 	if (length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset)) {
1938 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1939 						    cb_fn, cb_arg, op_type);
1940 	} else {
1941 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1942 						   cb_fn, cb_arg, op_type);
1943 	}
1944 }
1945 
1946 struct rw_iov_ctx {
1947 	struct spdk_blob *blob;
1948 	struct spdk_io_channel *channel;
1949 	spdk_blob_op_complete cb_fn;
1950 	void *cb_arg;
1951 	bool read;
1952 	int iovcnt;
1953 	struct iovec *orig_iov;
1954 	uint64_t page_offset;
1955 	uint64_t pages_remaining;
1956 	uint64_t pages_done;
1957 	struct iovec iov[0];
1958 };
1959 
1960 static void
1961 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1962 {
1963 	assert(cb_arg == NULL);
1964 	spdk_bs_sequence_finish(seq, bserrno);
1965 }
1966 
1967 static void
1968 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
1969 {
1970 	struct rw_iov_ctx *ctx = cb_arg;
1971 	struct spdk_blob *blob = ctx->blob;
1972 	struct iovec *iov, *orig_iov;
1973 	int iovcnt;
1974 	size_t orig_iovoff;
1975 	uint64_t page_count, pages_to_boundary, page_offset;
1976 	uint64_t byte_count;
1977 
1978 	if (bserrno != 0 || ctx->pages_remaining == 0) {
1979 		ctx->cb_fn(ctx->cb_arg, bserrno);
1980 		free(ctx);
1981 		return;
1982 	}
1983 
1984 	page_offset = ctx->page_offset;
1985 	pages_to_boundary = _spdk_bs_num_pages_to_cluster_boundary(blob, page_offset);
1986 	page_count = spdk_min(ctx->pages_remaining, pages_to_boundary);
1987 
1988 	/*
1989 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
1990 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
1991 	 *  point to the current position in the I/O sequence.
1992 	 */
1993 	byte_count = ctx->pages_done * sizeof(struct spdk_blob_md_page);
1994 	orig_iov = &ctx->orig_iov[0];
1995 	orig_iovoff = 0;
1996 	while (byte_count > 0) {
1997 		if (byte_count >= orig_iov->iov_len) {
1998 			byte_count -= orig_iov->iov_len;
1999 			orig_iov++;
2000 		} else {
2001 			orig_iovoff = byte_count;
2002 			byte_count = 0;
2003 		}
2004 	}
2005 
2006 	/*
2007 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2008 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2009 	 */
2010 	byte_count = page_count * sizeof(struct spdk_blob_md_page);
2011 	iov = &ctx->iov[0];
2012 	iovcnt = 0;
2013 	while (byte_count > 0) {
2014 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2015 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2016 		byte_count -= iov->iov_len;
2017 		orig_iovoff = 0;
2018 		orig_iov++;
2019 		iov++;
2020 		iovcnt++;
2021 	}
2022 
2023 	ctx->page_offset += page_count;
2024 	ctx->pages_done += page_count;
2025 	ctx->pages_remaining -= page_count;
2026 	iov = &ctx->iov[0];
2027 
2028 	if (ctx->read) {
2029 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, page_offset,
2030 				   page_count, _spdk_rw_iov_split_next, ctx);
2031 	} else {
2032 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, page_offset,
2033 				    page_count, _spdk_rw_iov_split_next, ctx);
2034 	}
2035 }
2036 
2037 static void
2038 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2039 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2040 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2041 {
2042 	struct spdk_bs_cpl	cpl;
2043 
2044 	assert(blob != NULL);
2045 
2046 	if (!read && blob->data_ro) {
2047 		cb_fn(cb_arg, -EPERM);
2048 		return;
2049 	}
2050 
2051 	if (length == 0) {
2052 		cb_fn(cb_arg, 0);
2053 		return;
2054 	}
2055 
2056 	if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) {
2057 		cb_fn(cb_arg, -EINVAL);
2058 		return;
2059 	}
2060 
2061 	/*
2062 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2063 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2064 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2065 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2066 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2067 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2068 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2069 	 *
2070 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2071 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2072 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2073 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2074 	 */
2075 	if (spdk_likely(length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset))) {
2076 		uint32_t lba_count;
2077 		uint64_t lba;
2078 
2079 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2080 
2081 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2082 		cpl.u.blob_basic.cb_fn = cb_fn;
2083 		cpl.u.blob_basic.cb_arg = cb_arg;
2084 		if (blob->frozen_refcnt) {
2085 			/* This blob I/O is frozen */
2086 			spdk_bs_user_op_t *op;
2087 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2088 
2089 			op = spdk_bs_user_op_alloc(_channel, &cpl, read, blob, iov, iovcnt, offset, length);
2090 			if (!op) {
2091 				cb_fn(cb_arg, -ENOMEM);
2092 				return;
2093 			}
2094 
2095 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2096 
2097 			return;
2098 		}
2099 
2100 		if (read) {
2101 			spdk_bs_sequence_t *seq;
2102 
2103 			seq = spdk_bs_sequence_start(_channel, &cpl);
2104 			if (!seq) {
2105 				cb_fn(cb_arg, -ENOMEM);
2106 				return;
2107 			}
2108 
2109 			if (_spdk_bs_page_is_allocated(blob, offset)) {
2110 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2111 			} else {
2112 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2113 							      _spdk_rw_iov_done, NULL);
2114 			}
2115 		} else {
2116 			if (_spdk_bs_page_is_allocated(blob, offset)) {
2117 				spdk_bs_sequence_t *seq;
2118 
2119 				seq = spdk_bs_sequence_start(_channel, &cpl);
2120 				if (!seq) {
2121 					cb_fn(cb_arg, -ENOMEM);
2122 					return;
2123 				}
2124 
2125 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2126 			} else {
2127 				/* Queue this operation and allocate the cluster */
2128 				spdk_bs_user_op_t *op;
2129 
2130 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, length);
2131 				if (!op) {
2132 					cb_fn(cb_arg, -ENOMEM);
2133 					return;
2134 				}
2135 
2136 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2137 			}
2138 		}
2139 	} else {
2140 		struct rw_iov_ctx *ctx;
2141 
2142 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2143 		if (ctx == NULL) {
2144 			cb_fn(cb_arg, -ENOMEM);
2145 			return;
2146 		}
2147 
2148 		ctx->blob = blob;
2149 		ctx->channel = _channel;
2150 		ctx->cb_fn = cb_fn;
2151 		ctx->cb_arg = cb_arg;
2152 		ctx->read = read;
2153 		ctx->orig_iov = iov;
2154 		ctx->iovcnt = iovcnt;
2155 		ctx->page_offset = offset;
2156 		ctx->pages_remaining = length;
2157 		ctx->pages_done = 0;
2158 
2159 		_spdk_rw_iov_split_next(ctx, 0);
2160 	}
2161 }
2162 
2163 static struct spdk_blob *
2164 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2165 {
2166 	struct spdk_blob *blob;
2167 
2168 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2169 		if (blob->id == blobid) {
2170 			return blob;
2171 		}
2172 	}
2173 
2174 	return NULL;
2175 }
2176 
2177 static int
2178 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2179 {
2180 	struct spdk_blob_store		*bs = io_device;
2181 	struct spdk_bs_channel		*channel = ctx_buf;
2182 	struct spdk_bs_dev		*dev;
2183 	uint32_t			max_ops = bs->max_channel_ops;
2184 	uint32_t			i;
2185 
2186 	dev = bs->dev;
2187 
2188 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2189 	if (!channel->req_mem) {
2190 		return -1;
2191 	}
2192 
2193 	TAILQ_INIT(&channel->reqs);
2194 
2195 	for (i = 0; i < max_ops; i++) {
2196 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2197 	}
2198 
2199 	channel->bs = bs;
2200 	channel->dev = dev;
2201 	channel->dev_channel = dev->create_channel(dev);
2202 
2203 	if (!channel->dev_channel) {
2204 		SPDK_ERRLOG("Failed to create device channel.\n");
2205 		free(channel->req_mem);
2206 		return -1;
2207 	}
2208 
2209 	TAILQ_INIT(&channel->need_cluster_alloc);
2210 	TAILQ_INIT(&channel->queued_io);
2211 
2212 	return 0;
2213 }
2214 
2215 static void
2216 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2217 {
2218 	struct spdk_bs_channel *channel = ctx_buf;
2219 	spdk_bs_user_op_t *op;
2220 
2221 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2222 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2223 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2224 		spdk_bs_user_op_abort(op);
2225 	}
2226 
2227 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2228 		op = TAILQ_FIRST(&channel->queued_io);
2229 		TAILQ_REMOVE(&channel->queued_io, op, link);
2230 		spdk_bs_user_op_abort(op);
2231 	}
2232 
2233 	free(channel->req_mem);
2234 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2235 }
2236 
2237 static void
2238 _spdk_bs_dev_destroy(void *io_device)
2239 {
2240 	struct spdk_blob_store *bs = io_device;
2241 	struct spdk_blob	*blob, *blob_tmp;
2242 
2243 	bs->dev->destroy(bs->dev);
2244 
2245 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2246 		TAILQ_REMOVE(&bs->blobs, blob, link);
2247 		_spdk_blob_free(blob);
2248 	}
2249 
2250 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2251 
2252 	spdk_bit_array_free(&bs->used_blobids);
2253 	spdk_bit_array_free(&bs->used_md_pages);
2254 	spdk_bit_array_free(&bs->used_clusters);
2255 	/*
2256 	 * If this function is called for any reason except a successful unload,
2257 	 * the unload_cpl type will be NONE and this will be a nop.
2258 	 */
2259 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2260 
2261 	free(bs);
2262 }
2263 
2264 static int
2265 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2266 {
2267 	spdk_blob_id snapshot_id;
2268 	struct spdk_blob_list *snapshot_entry = NULL;
2269 	struct spdk_blob_list *clone_entry = NULL;
2270 
2271 	assert(blob != NULL);
2272 
2273 	snapshot_id = blob->parent_id;
2274 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2275 		return 0;
2276 	}
2277 
2278 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2279 		if (snapshot_entry->id == snapshot_id) {
2280 			break;
2281 		}
2282 	}
2283 
2284 	if (snapshot_entry == NULL) {
2285 		/* Snapshot not found */
2286 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2287 		if (snapshot_entry == NULL) {
2288 			return -ENOMEM;
2289 		}
2290 		snapshot_entry->id = snapshot_id;
2291 		TAILQ_INIT(&snapshot_entry->clones);
2292 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2293 	} else {
2294 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2295 			if (clone_entry->id == blob->id) {
2296 				break;
2297 			}
2298 		}
2299 	}
2300 
2301 	if (clone_entry == NULL) {
2302 		/* Clone not found */
2303 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2304 		if (clone_entry == NULL) {
2305 			return -ENOMEM;
2306 		}
2307 		clone_entry->id = blob->id;
2308 		TAILQ_INIT(&clone_entry->clones);
2309 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2310 		snapshot_entry->clone_count++;
2311 	}
2312 
2313 	return 0;
2314 }
2315 
2316 static int
2317 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2318 {
2319 	struct spdk_blob_list *snapshot_entry = NULL;
2320 	struct spdk_blob_list *clone_entry = NULL;
2321 	spdk_blob_id snapshot_id;
2322 
2323 	assert(blob != NULL);
2324 
2325 	snapshot_id = blob->parent_id;
2326 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2327 		return 0;
2328 	}
2329 
2330 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2331 		if (snapshot_entry->id == snapshot_id) {
2332 			break;
2333 		}
2334 	}
2335 
2336 	assert(snapshot_entry != NULL);
2337 
2338 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2339 		if (clone_entry->id == blob->id) {
2340 			break;
2341 		}
2342 	}
2343 
2344 	assert(clone_entry != NULL);
2345 
2346 	blob->parent_id = SPDK_BLOBID_INVALID;
2347 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2348 	free(clone_entry);
2349 
2350 	snapshot_entry->clone_count--;
2351 	if (snapshot_entry->clone_count == 0) {
2352 		/* Snapshot have no more clones */
2353 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
2354 		free(snapshot_entry);
2355 	}
2356 
2357 	return 0;
2358 }
2359 
2360 static int
2361 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2362 {
2363 	struct spdk_blob_list *snapshot_entry;
2364 	struct spdk_blob_list *snapshot_entry_tmp;
2365 	struct spdk_blob_list *clone_entry;
2366 	struct spdk_blob_list *clone_entry_tmp;
2367 
2368 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2369 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2370 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2371 			free(clone_entry);
2372 		}
2373 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2374 		free(snapshot_entry);
2375 	}
2376 
2377 	return 0;
2378 }
2379 
2380 static void
2381 _spdk_bs_free(struct spdk_blob_store *bs)
2382 {
2383 	_spdk_bs_blob_list_free(bs);
2384 
2385 	spdk_bs_unregister_md_thread(bs);
2386 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2387 }
2388 
2389 void
2390 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2391 {
2392 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2393 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2394 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2395 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2396 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2397 	opts->iter_cb_fn = NULL;
2398 	opts->iter_cb_arg = NULL;
2399 }
2400 
2401 static int
2402 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2403 {
2404 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2405 	    opts->max_channel_ops == 0) {
2406 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2407 		return -1;
2408 	}
2409 
2410 	return 0;
2411 }
2412 
2413 static struct spdk_blob_store *
2414 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts)
2415 {
2416 	struct spdk_blob_store	*bs;
2417 	uint64_t dev_size;
2418 	int rc;
2419 
2420 	dev_size = dev->blocklen * dev->blockcnt;
2421 	if (dev_size < opts->cluster_sz) {
2422 		/* Device size cannot be smaller than cluster size of blobstore */
2423 		SPDK_ERRLOG("Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2424 			    dev_size, opts->cluster_sz);
2425 		return NULL;
2426 	}
2427 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2428 		/* Cluster size cannot be smaller than page size */
2429 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2430 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2431 		return NULL;
2432 	}
2433 	bs = calloc(1, sizeof(struct spdk_blob_store));
2434 	if (!bs) {
2435 		return NULL;
2436 	}
2437 
2438 	TAILQ_INIT(&bs->blobs);
2439 	TAILQ_INIT(&bs->snapshots);
2440 	bs->dev = dev;
2441 	bs->md_thread = spdk_get_thread();
2442 	assert(bs->md_thread != NULL);
2443 
2444 	/*
2445 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2446 	 *  even multiple of the cluster size.
2447 	 */
2448 	bs->cluster_sz = opts->cluster_sz;
2449 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2450 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2451 	bs->num_free_clusters = bs->total_clusters;
2452 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2453 	if (bs->used_clusters == NULL) {
2454 		free(bs);
2455 		return NULL;
2456 	}
2457 
2458 	bs->max_channel_ops = opts->max_channel_ops;
2459 	bs->super_blob = SPDK_BLOBID_INVALID;
2460 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2461 
2462 	/* The metadata is assumed to be at least 1 page */
2463 	bs->used_md_pages = spdk_bit_array_create(1);
2464 	bs->used_blobids = spdk_bit_array_create(0);
2465 
2466 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2467 
2468 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2469 				sizeof(struct spdk_bs_channel));
2470 	rc = spdk_bs_register_md_thread(bs);
2471 	if (rc == -1) {
2472 		spdk_io_device_unregister(bs, NULL);
2473 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2474 		spdk_bit_array_free(&bs->used_blobids);
2475 		spdk_bit_array_free(&bs->used_md_pages);
2476 		spdk_bit_array_free(&bs->used_clusters);
2477 		free(bs);
2478 		return NULL;
2479 	}
2480 
2481 	return bs;
2482 }
2483 
2484 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2485 
2486 struct spdk_bs_load_ctx {
2487 	struct spdk_blob_store		*bs;
2488 	struct spdk_bs_super_block	*super;
2489 
2490 	struct spdk_bs_md_mask		*mask;
2491 	bool				in_page_chain;
2492 	uint32_t			page_index;
2493 	uint32_t			cur_page;
2494 	struct spdk_blob_md_page	*page;
2495 	bool				is_load;
2496 
2497 	spdk_bs_sequence_t			*seq;
2498 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2499 	void					*iter_cb_arg;
2500 };
2501 
2502 static void
2503 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2504 {
2505 	assert(bserrno != 0);
2506 
2507 	spdk_dma_free(ctx->super);
2508 	spdk_bs_sequence_finish(seq, bserrno);
2509 	/*
2510 	 * Only free the blobstore when a load fails.  If an unload fails (for some reason)
2511 	 *  we want to keep the blobstore in case the caller wants to try again.
2512 	 */
2513 	if (ctx->is_load) {
2514 		_spdk_bs_free(ctx->bs);
2515 	}
2516 	free(ctx);
2517 }
2518 
2519 static void
2520 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2521 {
2522 	uint32_t i = 0;
2523 
2524 	while (true) {
2525 		i = spdk_bit_array_find_first_set(array, i);
2526 		if (i >= mask->length) {
2527 			break;
2528 		}
2529 		mask->mask[i / 8] |= 1U << (i % 8);
2530 		i++;
2531 	}
2532 }
2533 
2534 static int
2535 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2536 {
2537 	struct spdk_bit_array *array;
2538 	uint32_t i;
2539 
2540 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2541 		return -ENOMEM;
2542 	}
2543 
2544 	array = *array_ptr;
2545 	for (i = 0; i < mask->length; i++) {
2546 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2547 			spdk_bit_array_set(array, i);
2548 		}
2549 	}
2550 
2551 	return 0;
2552 }
2553 
2554 static void
2555 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2556 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2557 {
2558 	/* Update the values in the super block */
2559 	super->super_blob = bs->super_blob;
2560 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2561 	super->crc = _spdk_blob_md_page_calc_crc(super);
2562 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2563 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2564 				   cb_fn, cb_arg);
2565 }
2566 
2567 static void
2568 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2569 {
2570 	struct spdk_bs_load_ctx	*ctx = arg;
2571 	uint64_t	mask_size, lba, lba_count;
2572 
2573 	/* Write out the used clusters mask */
2574 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2575 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2576 	if (!ctx->mask) {
2577 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2578 		return;
2579 	}
2580 
2581 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2582 	ctx->mask->length = ctx->bs->total_clusters;
2583 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2584 
2585 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2586 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2587 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2588 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2589 }
2590 
2591 static void
2592 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2593 {
2594 	struct spdk_bs_load_ctx	*ctx = arg;
2595 	uint64_t	mask_size, lba, lba_count;
2596 
2597 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2598 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2599 	if (!ctx->mask) {
2600 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2601 		return;
2602 	}
2603 
2604 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2605 	ctx->mask->length = ctx->super->md_len;
2606 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2607 
2608 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2609 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2610 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2611 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2612 }
2613 
2614 static void
2615 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2616 {
2617 	struct spdk_bs_load_ctx	*ctx = arg;
2618 	uint64_t	mask_size, lba, lba_count;
2619 
2620 	if (ctx->super->used_blobid_mask_len == 0) {
2621 		/*
2622 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2623 		 *  written to disk.
2624 		 */
2625 		cb_fn(seq, arg, 0);
2626 		return;
2627 	}
2628 
2629 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2630 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2631 	if (!ctx->mask) {
2632 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2633 		return;
2634 	}
2635 
2636 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2637 	ctx->mask->length = ctx->super->md_len;
2638 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2639 
2640 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2641 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2642 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2643 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2644 }
2645 
2646 static void
2647 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2648 {
2649 	struct spdk_bs_load_ctx *ctx = arg;
2650 
2651 	if (bserrno == 0) {
2652 		if (ctx->iter_cb_fn) {
2653 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2654 		}
2655 		_spdk_bs_blob_list_add(blob);
2656 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2657 		return;
2658 	}
2659 
2660 	if (bserrno == -ENOENT) {
2661 		bserrno = 0;
2662 	} else {
2663 		/*
2664 		 * This case needs to be looked at further.  Same problem
2665 		 *  exists with applications that rely on explicit blob
2666 		 *  iteration.  We should just skip the blob that failed
2667 		 *  to load and coontinue on to the next one.
2668 		 */
2669 		SPDK_ERRLOG("Error in iterating blobs\n");
2670 	}
2671 
2672 	ctx->iter_cb_fn = NULL;
2673 
2674 	spdk_dma_free(ctx->super);
2675 	spdk_dma_free(ctx->mask);
2676 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2677 	free(ctx);
2678 }
2679 
2680 static void
2681 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2682 {
2683 	ctx->seq = seq;
2684 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2685 }
2686 
2687 static void
2688 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2689 {
2690 	struct spdk_bs_load_ctx *ctx = cb_arg;
2691 	int rc;
2692 
2693 	/* The type must be correct */
2694 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2695 
2696 	/* The length of the mask (in bits) must not be greater than
2697 	 * the length of the buffer (converted to bits) */
2698 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2699 
2700 	/* The length of the mask must be exactly equal to the size
2701 	 * (in pages) of the metadata region */
2702 	assert(ctx->mask->length == ctx->super->md_len);
2703 
2704 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2705 	if (rc < 0) {
2706 		spdk_dma_free(ctx->mask);
2707 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2708 		return;
2709 	}
2710 
2711 	_spdk_bs_load_complete(seq, ctx, bserrno);
2712 }
2713 
2714 static void
2715 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2716 {
2717 	struct spdk_bs_load_ctx *ctx = cb_arg;
2718 	uint64_t		lba, lba_count, mask_size;
2719 	int			rc;
2720 
2721 	/* The type must be correct */
2722 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2723 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2724 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2725 					     struct spdk_blob_md_page) * 8));
2726 	/* The length of the mask must be exactly equal to the total number of clusters */
2727 	assert(ctx->mask->length == ctx->bs->total_clusters);
2728 
2729 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2730 	if (rc < 0) {
2731 		spdk_dma_free(ctx->mask);
2732 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2733 		return;
2734 	}
2735 
2736 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2737 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2738 
2739 	spdk_dma_free(ctx->mask);
2740 
2741 	/* Read the used blobids mask */
2742 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2743 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2744 	if (!ctx->mask) {
2745 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2746 		return;
2747 	}
2748 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2749 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2750 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2751 				  _spdk_bs_load_used_blobids_cpl, ctx);
2752 }
2753 
2754 static void
2755 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2756 {
2757 	struct spdk_bs_load_ctx *ctx = cb_arg;
2758 	uint64_t		lba, lba_count, mask_size;
2759 	int			rc;
2760 
2761 	/* The type must be correct */
2762 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2763 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2764 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2765 				     8));
2766 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2767 	assert(ctx->mask->length == ctx->super->md_len);
2768 
2769 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2770 	if (rc < 0) {
2771 		spdk_dma_free(ctx->mask);
2772 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2773 		return;
2774 	}
2775 
2776 	spdk_dma_free(ctx->mask);
2777 
2778 	/* Read the used clusters mask */
2779 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2780 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2781 	if (!ctx->mask) {
2782 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2783 		return;
2784 	}
2785 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2786 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2787 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2788 				  _spdk_bs_load_used_clusters_cpl, ctx);
2789 }
2790 
2791 static void
2792 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2793 {
2794 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2795 	uint64_t lba, lba_count, mask_size;
2796 
2797 	/* Read the used pages mask */
2798 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2799 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2800 	if (!ctx->mask) {
2801 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2802 		return;
2803 	}
2804 
2805 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2806 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2807 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2808 				  _spdk_bs_load_used_pages_cpl, ctx);
2809 }
2810 
2811 static int
2812 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2813 {
2814 	struct spdk_blob_md_descriptor *desc;
2815 	size_t	cur_desc = 0;
2816 
2817 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2818 	while (cur_desc < sizeof(page->descriptors)) {
2819 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2820 			if (desc->length == 0) {
2821 				/* If padding and length are 0, this terminates the page */
2822 				break;
2823 			}
2824 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2825 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2826 			unsigned int				i, j;
2827 			unsigned int				cluster_count = 0;
2828 			uint32_t				cluster_idx;
2829 
2830 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2831 
2832 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2833 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2834 					cluster_idx = desc_extent->extents[i].cluster_idx;
2835 					/*
2836 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2837 					 * in the used cluster map.
2838 					 */
2839 					if (cluster_idx != 0) {
2840 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2841 						if (bs->num_free_clusters == 0) {
2842 							return -ENOSPC;
2843 						}
2844 						bs->num_free_clusters--;
2845 					}
2846 					cluster_count++;
2847 				}
2848 			}
2849 			if (cluster_count == 0) {
2850 				return -EINVAL;
2851 			}
2852 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2853 			/* Skip this item */
2854 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2855 			/* Skip this item */
2856 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2857 			/* Skip this item */
2858 		} else {
2859 			/* Error */
2860 			return -EINVAL;
2861 		}
2862 		/* Advance to the next descriptor */
2863 		cur_desc += sizeof(*desc) + desc->length;
2864 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2865 			break;
2866 		}
2867 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2868 	}
2869 	return 0;
2870 }
2871 
2872 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2873 {
2874 	uint32_t crc;
2875 
2876 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2877 	if (crc != ctx->page->crc) {
2878 		return false;
2879 	}
2880 
2881 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2882 		return false;
2883 	}
2884 	return true;
2885 }
2886 
2887 static void
2888 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2889 
2890 static void
2891 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2892 {
2893 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2894 
2895 	_spdk_bs_load_complete(seq, ctx, bserrno);
2896 }
2897 
2898 static void
2899 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2900 {
2901 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2902 
2903 	spdk_dma_free(ctx->mask);
2904 	ctx->mask = NULL;
2905 
2906 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2907 }
2908 
2909 static void
2910 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2911 {
2912 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2913 
2914 	spdk_dma_free(ctx->mask);
2915 	ctx->mask = NULL;
2916 
2917 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2918 }
2919 
2920 static void
2921 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2922 {
2923 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2924 }
2925 
2926 static void
2927 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2928 {
2929 	struct spdk_bs_load_ctx *ctx = cb_arg;
2930 	uint64_t num_md_clusters;
2931 	uint64_t i;
2932 	uint32_t page_num;
2933 
2934 	if (bserrno != 0) {
2935 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2936 		return;
2937 	}
2938 
2939 	page_num = ctx->cur_page;
2940 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2941 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2942 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2943 			if (ctx->page->sequence_num == 0) {
2944 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2945 			}
2946 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2947 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
2948 				return;
2949 			}
2950 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
2951 				ctx->in_page_chain = true;
2952 				ctx->cur_page = ctx->page->next;
2953 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2954 				return;
2955 			}
2956 		}
2957 	}
2958 
2959 	ctx->in_page_chain = false;
2960 
2961 	do {
2962 		ctx->page_index++;
2963 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
2964 
2965 	if (ctx->page_index < ctx->super->md_len) {
2966 		ctx->cur_page = ctx->page_index;
2967 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2968 	} else {
2969 		/* Claim all of the clusters used by the metadata */
2970 		num_md_clusters = divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
2971 		for (i = 0; i < num_md_clusters; i++) {
2972 			_spdk_bs_claim_cluster(ctx->bs, i);
2973 		}
2974 		spdk_dma_free(ctx->page);
2975 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
2976 	}
2977 }
2978 
2979 static void
2980 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
2981 {
2982 	struct spdk_bs_load_ctx *ctx = cb_arg;
2983 	uint64_t lba;
2984 
2985 	assert(ctx->cur_page < ctx->super->md_len);
2986 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
2987 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
2988 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
2989 				  _spdk_bs_load_replay_md_cpl, ctx);
2990 }
2991 
2992 static void
2993 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
2994 {
2995 	struct spdk_bs_load_ctx *ctx = cb_arg;
2996 
2997 	ctx->page_index = 0;
2998 	ctx->cur_page = 0;
2999 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3000 				     SPDK_BS_PAGE_SIZE,
3001 				     NULL);
3002 	if (!ctx->page) {
3003 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3004 		return;
3005 	}
3006 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3007 }
3008 
3009 static void
3010 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3011 {
3012 	struct spdk_bs_load_ctx *ctx = cb_arg;
3013 	int		rc;
3014 
3015 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3016 	if (rc < 0) {
3017 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3018 		return;
3019 	}
3020 
3021 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3022 	if (rc < 0) {
3023 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3024 		return;
3025 	}
3026 
3027 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3028 	if (rc < 0) {
3029 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3030 		return;
3031 	}
3032 
3033 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3034 	_spdk_bs_load_replay_md(seq, cb_arg);
3035 }
3036 
3037 static void
3038 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3039 {
3040 	struct spdk_bs_load_ctx *ctx = cb_arg;
3041 	uint32_t	crc;
3042 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3043 
3044 	if (ctx->super->version > SPDK_BS_VERSION ||
3045 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3046 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3047 		return;
3048 	}
3049 
3050 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3051 		   sizeof(ctx->super->signature)) != 0) {
3052 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3053 		return;
3054 	}
3055 
3056 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3057 	if (crc != ctx->super->crc) {
3058 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3059 		return;
3060 	}
3061 
3062 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3063 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3064 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3065 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3066 	} else {
3067 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3068 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3069 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3070 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3071 		return;
3072 	}
3073 
3074 	if (ctx->super->size == 0) {
3075 		/* Update number of blocks for blobstore */
3076 		ctx->bs->total_clusters = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen / ctx->bs->cluster_sz;
3077 	} else if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3078 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3079 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3080 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3081 		return;
3082 	} else {
3083 		ctx->bs->total_clusters = ctx->super->size / ctx->bs->cluster_sz;
3084 	}
3085 
3086 	/* Parse the super block */
3087 	ctx->bs->clean = 1;
3088 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3089 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3090 	ctx->bs->md_start = ctx->super->md_start;
3091 	ctx->bs->md_len = ctx->super->md_len;
3092 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - divide_round_up(
3093 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3094 	ctx->bs->super_blob = ctx->super->super_blob;
3095 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3096 
3097 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3098 		_spdk_bs_recover(seq, ctx);
3099 	} else {
3100 		_spdk_bs_load_read_used_pages(seq, ctx);
3101 	}
3102 }
3103 
3104 void
3105 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3106 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3107 {
3108 	struct spdk_blob_store	*bs;
3109 	struct spdk_bs_cpl	cpl;
3110 	spdk_bs_sequence_t	*seq;
3111 	struct spdk_bs_load_ctx *ctx;
3112 	struct spdk_bs_opts	opts = {};
3113 
3114 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3115 
3116 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3117 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3118 		dev->destroy(dev);
3119 		cb_fn(cb_arg, NULL, -EINVAL);
3120 		return;
3121 	}
3122 
3123 	if (o) {
3124 		opts = *o;
3125 	} else {
3126 		spdk_bs_opts_init(&opts);
3127 	}
3128 
3129 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3130 		dev->destroy(dev);
3131 		cb_fn(cb_arg, NULL, -EINVAL);
3132 		return;
3133 	}
3134 
3135 	bs = _spdk_bs_alloc(dev, &opts);
3136 	if (!bs) {
3137 		dev->destroy(dev);
3138 		cb_fn(cb_arg, NULL, -ENOMEM);
3139 		return;
3140 	}
3141 
3142 	ctx = calloc(1, sizeof(*ctx));
3143 	if (!ctx) {
3144 		_spdk_bs_free(bs);
3145 		cb_fn(cb_arg, NULL, -ENOMEM);
3146 		return;
3147 	}
3148 
3149 	ctx->bs = bs;
3150 	ctx->is_load = true;
3151 	ctx->iter_cb_fn = opts.iter_cb_fn;
3152 	ctx->iter_cb_arg = opts.iter_cb_arg;
3153 
3154 	/* Allocate memory for the super block */
3155 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3156 	if (!ctx->super) {
3157 		free(ctx);
3158 		_spdk_bs_free(bs);
3159 		cb_fn(cb_arg, NULL, -ENOMEM);
3160 		return;
3161 	}
3162 
3163 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3164 	cpl.u.bs_handle.cb_fn = cb_fn;
3165 	cpl.u.bs_handle.cb_arg = cb_arg;
3166 	cpl.u.bs_handle.bs = bs;
3167 
3168 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3169 	if (!seq) {
3170 		spdk_dma_free(ctx->super);
3171 		free(ctx);
3172 		_spdk_bs_free(bs);
3173 		cb_fn(cb_arg, NULL, -ENOMEM);
3174 		return;
3175 	}
3176 
3177 	/* Read the super block */
3178 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3179 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3180 				  _spdk_bs_load_super_cpl, ctx);
3181 }
3182 
3183 /* END spdk_bs_load */
3184 
3185 /* START spdk_bs_dump */
3186 
3187 struct spdk_bs_dump_ctx {
3188 	struct spdk_blob_store		*bs;
3189 	struct spdk_bs_super_block	*super;
3190 	uint32_t			cur_page;
3191 	struct spdk_blob_md_page	*page;
3192 	spdk_bs_sequence_t		*seq;
3193 	FILE				*fp;
3194 	spdk_bs_dump_print_xattr	print_xattr_fn;
3195 	char				xattr_name[4096];
3196 };
3197 
3198 static void
3199 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3200 {
3201 	spdk_dma_free(ctx->super);
3202 
3203 	/*
3204 	 * We need to defer calling spdk_bs_call_cpl() until after
3205 	 * dev destuction, so tuck these away for later use.
3206 	 */
3207 	ctx->bs->unload_err = bserrno;
3208 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3209 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3210 
3211 	spdk_bs_sequence_finish(seq, 0);
3212 	_spdk_bs_free(ctx->bs);
3213 	free(ctx);
3214 }
3215 
3216 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3217 
3218 static void
3219 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3220 {
3221 	uint32_t page_idx = ctx->cur_page;
3222 	struct spdk_blob_md_page *page = ctx->page;
3223 	struct spdk_blob_md_descriptor *desc;
3224 	size_t cur_desc = 0;
3225 	uint32_t crc;
3226 
3227 	fprintf(ctx->fp, "=========\n");
3228 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3229 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3230 
3231 	crc = _spdk_blob_md_page_calc_crc(page);
3232 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3233 
3234 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3235 	while (cur_desc < sizeof(page->descriptors)) {
3236 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3237 			if (desc->length == 0) {
3238 				/* If padding and length are 0, this terminates the page */
3239 				break;
3240 			}
3241 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3242 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3243 			unsigned int				i;
3244 
3245 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3246 
3247 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3248 				if (desc_extent->extents[i].cluster_idx != 0) {
3249 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3250 						desc_extent->extents[i].cluster_idx);
3251 				} else {
3252 					fprintf(ctx->fp, "Unallocated Extent - ");
3253 				}
3254 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3255 				fprintf(ctx->fp, "\n");
3256 			}
3257 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3258 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3259 			uint32_t i;
3260 
3261 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3262 
3263 			if (desc_xattr->length !=
3264 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3265 			    desc_xattr->name_length + desc_xattr->value_length) {
3266 			}
3267 
3268 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3269 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3270 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3271 			fprintf(ctx->fp, "       value = \"");
3272 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3273 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3274 					    desc_xattr->value_length);
3275 			fprintf(ctx->fp, "\"\n");
3276 			for (i = 0; i < desc_xattr->value_length; i++) {
3277 				if (i % 16 == 0) {
3278 					fprintf(ctx->fp, "               ");
3279 				}
3280 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3281 				if ((i + 1) % 16 == 0) {
3282 					fprintf(ctx->fp, "\n");
3283 				}
3284 			}
3285 			if (i % 16 != 0) {
3286 				fprintf(ctx->fp, "\n");
3287 			}
3288 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3289 			/* TODO */
3290 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3291 			/* TODO */
3292 		} else {
3293 			/* Error */
3294 		}
3295 		/* Advance to the next descriptor */
3296 		cur_desc += sizeof(*desc) + desc->length;
3297 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3298 			break;
3299 		}
3300 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3301 	}
3302 }
3303 
3304 static void
3305 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3306 {
3307 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3308 
3309 	if (bserrno != 0) {
3310 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3311 		return;
3312 	}
3313 
3314 	if (ctx->page->id != 0) {
3315 		_spdk_bs_dump_print_md_page(ctx);
3316 	}
3317 
3318 	ctx->cur_page++;
3319 
3320 	if (ctx->cur_page < ctx->super->md_len) {
3321 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3322 	} else {
3323 		spdk_dma_free(ctx->page);
3324 		_spdk_bs_dump_finish(seq, ctx, 0);
3325 	}
3326 }
3327 
3328 static void
3329 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3330 {
3331 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3332 	uint64_t lba;
3333 
3334 	assert(ctx->cur_page < ctx->super->md_len);
3335 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3336 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3337 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3338 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3339 }
3340 
3341 static void
3342 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3343 {
3344 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3345 
3346 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3347 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3348 		   sizeof(ctx->super->signature)) != 0) {
3349 		fprintf(ctx->fp, "(Mismatch)\n");
3350 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3351 		return;
3352 	} else {
3353 		fprintf(ctx->fp, "(OK)\n");
3354 	}
3355 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3356 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3357 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3358 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3359 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3360 	fprintf(ctx->fp, "Super Blob ID: ");
3361 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3362 		fprintf(ctx->fp, "(None)\n");
3363 	} else {
3364 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3365 	}
3366 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3367 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3368 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3369 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3370 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3371 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3372 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3373 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3374 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3375 
3376 	ctx->cur_page = 0;
3377 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3378 				     SPDK_BS_PAGE_SIZE,
3379 				     NULL);
3380 	if (!ctx->page) {
3381 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3382 		return;
3383 	}
3384 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3385 }
3386 
3387 void
3388 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3389 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3390 {
3391 	struct spdk_blob_store	*bs;
3392 	struct spdk_bs_cpl	cpl;
3393 	spdk_bs_sequence_t	*seq;
3394 	struct spdk_bs_dump_ctx *ctx;
3395 	struct spdk_bs_opts	opts = {};
3396 
3397 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3398 
3399 	spdk_bs_opts_init(&opts);
3400 
3401 	bs = _spdk_bs_alloc(dev, &opts);
3402 	if (!bs) {
3403 		dev->destroy(dev);
3404 		cb_fn(cb_arg, -ENOMEM);
3405 		return;
3406 	}
3407 
3408 	ctx = calloc(1, sizeof(*ctx));
3409 	if (!ctx) {
3410 		_spdk_bs_free(bs);
3411 		cb_fn(cb_arg, -ENOMEM);
3412 		return;
3413 	}
3414 
3415 	ctx->bs = bs;
3416 	ctx->fp = fp;
3417 	ctx->print_xattr_fn = print_xattr_fn;
3418 
3419 	/* Allocate memory for the super block */
3420 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3421 	if (!ctx->super) {
3422 		free(ctx);
3423 		_spdk_bs_free(bs);
3424 		cb_fn(cb_arg, -ENOMEM);
3425 		return;
3426 	}
3427 
3428 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3429 	cpl.u.bs_basic.cb_fn = cb_fn;
3430 	cpl.u.bs_basic.cb_arg = cb_arg;
3431 
3432 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3433 	if (!seq) {
3434 		spdk_dma_free(ctx->super);
3435 		free(ctx);
3436 		_spdk_bs_free(bs);
3437 		cb_fn(cb_arg, -ENOMEM);
3438 		return;
3439 	}
3440 
3441 	/* Read the super block */
3442 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3443 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3444 				  _spdk_bs_dump_super_cpl, ctx);
3445 }
3446 
3447 /* END spdk_bs_dump */
3448 
3449 /* START spdk_bs_init */
3450 
3451 struct spdk_bs_init_ctx {
3452 	struct spdk_blob_store		*bs;
3453 	struct spdk_bs_super_block	*super;
3454 };
3455 
3456 static void
3457 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3458 {
3459 	struct spdk_bs_init_ctx *ctx = cb_arg;
3460 
3461 	spdk_dma_free(ctx->super);
3462 	free(ctx);
3463 
3464 	spdk_bs_sequence_finish(seq, bserrno);
3465 }
3466 
3467 static void
3468 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3469 {
3470 	struct spdk_bs_init_ctx *ctx = cb_arg;
3471 
3472 	/* Write super block */
3473 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3474 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3475 				   _spdk_bs_init_persist_super_cpl, ctx);
3476 }
3477 
3478 void
3479 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3480 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3481 {
3482 	struct spdk_bs_init_ctx *ctx;
3483 	struct spdk_blob_store	*bs;
3484 	struct spdk_bs_cpl	cpl;
3485 	spdk_bs_sequence_t	*seq;
3486 	spdk_bs_batch_t		*batch;
3487 	uint64_t		num_md_lba;
3488 	uint64_t		num_md_pages;
3489 	uint64_t		num_md_clusters;
3490 	uint32_t		i;
3491 	struct spdk_bs_opts	opts = {};
3492 	int			rc;
3493 
3494 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3495 
3496 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3497 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3498 			    dev->blocklen);
3499 		dev->destroy(dev);
3500 		cb_fn(cb_arg, NULL, -EINVAL);
3501 		return;
3502 	}
3503 
3504 	if (o) {
3505 		opts = *o;
3506 	} else {
3507 		spdk_bs_opts_init(&opts);
3508 	}
3509 
3510 	if (_spdk_bs_opts_verify(&opts) != 0) {
3511 		dev->destroy(dev);
3512 		cb_fn(cb_arg, NULL, -EINVAL);
3513 		return;
3514 	}
3515 
3516 	bs = _spdk_bs_alloc(dev, &opts);
3517 	if (!bs) {
3518 		dev->destroy(dev);
3519 		cb_fn(cb_arg, NULL, -ENOMEM);
3520 		return;
3521 	}
3522 
3523 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3524 		/* By default, allocate 1 page per cluster.
3525 		 * Technically, this over-allocates metadata
3526 		 * because more metadata will reduce the number
3527 		 * of usable clusters. This can be addressed with
3528 		 * more complex math in the future.
3529 		 */
3530 		bs->md_len = bs->total_clusters;
3531 	} else {
3532 		bs->md_len = opts.num_md_pages;
3533 	}
3534 
3535 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3536 	if (rc < 0) {
3537 		_spdk_bs_free(bs);
3538 		cb_fn(cb_arg, NULL, -ENOMEM);
3539 		return;
3540 	}
3541 
3542 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3543 	if (rc < 0) {
3544 		_spdk_bs_free(bs);
3545 		cb_fn(cb_arg, NULL, -ENOMEM);
3546 		return;
3547 	}
3548 
3549 	ctx = calloc(1, sizeof(*ctx));
3550 	if (!ctx) {
3551 		_spdk_bs_free(bs);
3552 		cb_fn(cb_arg, NULL, -ENOMEM);
3553 		return;
3554 	}
3555 
3556 	ctx->bs = bs;
3557 
3558 	/* Allocate memory for the super block */
3559 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3560 	if (!ctx->super) {
3561 		free(ctx);
3562 		_spdk_bs_free(bs);
3563 		cb_fn(cb_arg, NULL, -ENOMEM);
3564 		return;
3565 	}
3566 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3567 	       sizeof(ctx->super->signature));
3568 	ctx->super->version = SPDK_BS_VERSION;
3569 	ctx->super->length = sizeof(*ctx->super);
3570 	ctx->super->super_blob = bs->super_blob;
3571 	ctx->super->clean = 0;
3572 	ctx->super->cluster_size = bs->cluster_sz;
3573 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3574 
3575 	/* Calculate how many pages the metadata consumes at the front
3576 	 * of the disk.
3577 	 */
3578 
3579 	/* The super block uses 1 page */
3580 	num_md_pages = 1;
3581 
3582 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3583 	 * up to the nearest page, plus a header.
3584 	 */
3585 	ctx->super->used_page_mask_start = num_md_pages;
3586 	ctx->super->used_page_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3587 					 divide_round_up(bs->md_len, 8),
3588 					 SPDK_BS_PAGE_SIZE);
3589 	num_md_pages += ctx->super->used_page_mask_len;
3590 
3591 	/* The used_clusters mask requires 1 bit per cluster, rounded
3592 	 * up to the nearest page, plus a header.
3593 	 */
3594 	ctx->super->used_cluster_mask_start = num_md_pages;
3595 	ctx->super->used_cluster_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3596 					    divide_round_up(bs->total_clusters, 8),
3597 					    SPDK_BS_PAGE_SIZE);
3598 	num_md_pages += ctx->super->used_cluster_mask_len;
3599 
3600 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3601 	 * up to the nearest page, plus a header.
3602 	 */
3603 	ctx->super->used_blobid_mask_start = num_md_pages;
3604 	ctx->super->used_blobid_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3605 					   divide_round_up(bs->md_len, 8),
3606 					   SPDK_BS_PAGE_SIZE);
3607 	num_md_pages += ctx->super->used_blobid_mask_len;
3608 
3609 	/* The metadata region size was chosen above */
3610 	ctx->super->md_start = bs->md_start = num_md_pages;
3611 	ctx->super->md_len = bs->md_len;
3612 	num_md_pages += bs->md_len;
3613 
3614 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3615 
3616 	ctx->super->size = dev->blockcnt * dev->blocklen;
3617 
3618 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3619 
3620 	num_md_clusters = divide_round_up(num_md_pages, bs->pages_per_cluster);
3621 	if (num_md_clusters > bs->total_clusters) {
3622 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3623 			    "please decrease number of pages reserved for metadata "
3624 			    "or increase cluster size.\n");
3625 		spdk_dma_free(ctx->super);
3626 		free(ctx);
3627 		_spdk_bs_free(bs);
3628 		cb_fn(cb_arg, NULL, -ENOMEM);
3629 		return;
3630 	}
3631 	/* Claim all of the clusters used by the metadata */
3632 	for (i = 0; i < num_md_clusters; i++) {
3633 		_spdk_bs_claim_cluster(bs, i);
3634 	}
3635 
3636 	bs->total_data_clusters = bs->num_free_clusters;
3637 
3638 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3639 	cpl.u.bs_handle.cb_fn = cb_fn;
3640 	cpl.u.bs_handle.cb_arg = cb_arg;
3641 	cpl.u.bs_handle.bs = bs;
3642 
3643 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3644 	if (!seq) {
3645 		spdk_dma_free(ctx->super);
3646 		free(ctx);
3647 		_spdk_bs_free(bs);
3648 		cb_fn(cb_arg, NULL, -ENOMEM);
3649 		return;
3650 	}
3651 
3652 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3653 
3654 	/* Clear metadata space */
3655 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3656 	/* Trim data clusters */
3657 	spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3658 
3659 	spdk_bs_batch_close(batch);
3660 }
3661 
3662 /* END spdk_bs_init */
3663 
3664 /* START spdk_bs_destroy */
3665 
3666 static void
3667 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3668 {
3669 	struct spdk_bs_init_ctx *ctx = cb_arg;
3670 	struct spdk_blob_store *bs = ctx->bs;
3671 
3672 	/*
3673 	 * We need to defer calling spdk_bs_call_cpl() until after
3674 	 * dev destruction, so tuck these away for later use.
3675 	 */
3676 	bs->unload_err = bserrno;
3677 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3678 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3679 
3680 	spdk_bs_sequence_finish(seq, bserrno);
3681 
3682 	_spdk_bs_free(bs);
3683 	free(ctx);
3684 }
3685 
3686 void
3687 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3688 		void *cb_arg)
3689 {
3690 	struct spdk_bs_cpl	cpl;
3691 	spdk_bs_sequence_t	*seq;
3692 	struct spdk_bs_init_ctx *ctx;
3693 
3694 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3695 
3696 	if (!TAILQ_EMPTY(&bs->blobs)) {
3697 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3698 		cb_fn(cb_arg, -EBUSY);
3699 		return;
3700 	}
3701 
3702 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3703 	cpl.u.bs_basic.cb_fn = cb_fn;
3704 	cpl.u.bs_basic.cb_arg = cb_arg;
3705 
3706 	ctx = calloc(1, sizeof(*ctx));
3707 	if (!ctx) {
3708 		cb_fn(cb_arg, -ENOMEM);
3709 		return;
3710 	}
3711 
3712 	ctx->bs = bs;
3713 
3714 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3715 	if (!seq) {
3716 		free(ctx);
3717 		cb_fn(cb_arg, -ENOMEM);
3718 		return;
3719 	}
3720 
3721 	/* Write zeroes to the super block */
3722 	spdk_bs_sequence_write_zeroes_dev(seq,
3723 					  _spdk_bs_page_to_lba(bs, 0),
3724 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3725 					  _spdk_bs_destroy_trim_cpl, ctx);
3726 }
3727 
3728 /* END spdk_bs_destroy */
3729 
3730 /* START spdk_bs_unload */
3731 
3732 static void
3733 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3734 {
3735 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3736 
3737 	spdk_dma_free(ctx->super);
3738 
3739 	/*
3740 	 * We need to defer calling spdk_bs_call_cpl() until after
3741 	 * dev destuction, so tuck these away for later use.
3742 	 */
3743 	ctx->bs->unload_err = bserrno;
3744 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3745 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3746 
3747 	spdk_bs_sequence_finish(seq, bserrno);
3748 
3749 	_spdk_bs_free(ctx->bs);
3750 	free(ctx);
3751 }
3752 
3753 static void
3754 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3755 {
3756 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3757 
3758 	spdk_dma_free(ctx->mask);
3759 	ctx->super->clean = 1;
3760 
3761 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3762 }
3763 
3764 static void
3765 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3766 {
3767 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3768 
3769 	spdk_dma_free(ctx->mask);
3770 	ctx->mask = NULL;
3771 
3772 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3773 }
3774 
3775 static void
3776 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3777 {
3778 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3779 
3780 	spdk_dma_free(ctx->mask);
3781 	ctx->mask = NULL;
3782 
3783 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3784 }
3785 
3786 static void
3787 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3788 {
3789 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3790 }
3791 
3792 void
3793 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3794 {
3795 	struct spdk_bs_cpl	cpl;
3796 	spdk_bs_sequence_t	*seq;
3797 	struct spdk_bs_load_ctx *ctx;
3798 
3799 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3800 
3801 	if (!TAILQ_EMPTY(&bs->blobs)) {
3802 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3803 		cb_fn(cb_arg, -EBUSY);
3804 		return;
3805 	}
3806 
3807 	ctx = calloc(1, sizeof(*ctx));
3808 	if (!ctx) {
3809 		cb_fn(cb_arg, -ENOMEM);
3810 		return;
3811 	}
3812 
3813 	ctx->bs = bs;
3814 	ctx->is_load = false;
3815 
3816 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3817 	if (!ctx->super) {
3818 		free(ctx);
3819 		cb_fn(cb_arg, -ENOMEM);
3820 		return;
3821 	}
3822 
3823 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3824 	cpl.u.bs_basic.cb_fn = cb_fn;
3825 	cpl.u.bs_basic.cb_arg = cb_arg;
3826 
3827 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3828 	if (!seq) {
3829 		spdk_dma_free(ctx->super);
3830 		free(ctx);
3831 		cb_fn(cb_arg, -ENOMEM);
3832 		return;
3833 	}
3834 
3835 	/* Read super block */
3836 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3837 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3838 				  _spdk_bs_unload_read_super_cpl, ctx);
3839 }
3840 
3841 /* END spdk_bs_unload */
3842 
3843 /* START spdk_bs_set_super */
3844 
3845 struct spdk_bs_set_super_ctx {
3846 	struct spdk_blob_store		*bs;
3847 	struct spdk_bs_super_block	*super;
3848 };
3849 
3850 static void
3851 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3852 {
3853 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3854 
3855 	if (bserrno != 0) {
3856 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3857 	}
3858 
3859 	spdk_dma_free(ctx->super);
3860 
3861 	spdk_bs_sequence_finish(seq, bserrno);
3862 
3863 	free(ctx);
3864 }
3865 
3866 static void
3867 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3868 {
3869 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3870 
3871 	if (bserrno != 0) {
3872 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3873 		spdk_dma_free(ctx->super);
3874 		spdk_bs_sequence_finish(seq, bserrno);
3875 		free(ctx);
3876 		return;
3877 	}
3878 
3879 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3880 }
3881 
3882 void
3883 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3884 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3885 {
3886 	struct spdk_bs_cpl		cpl;
3887 	spdk_bs_sequence_t		*seq;
3888 	struct spdk_bs_set_super_ctx	*ctx;
3889 
3890 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3891 
3892 	ctx = calloc(1, sizeof(*ctx));
3893 	if (!ctx) {
3894 		cb_fn(cb_arg, -ENOMEM);
3895 		return;
3896 	}
3897 
3898 	ctx->bs = bs;
3899 
3900 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3901 	if (!ctx->super) {
3902 		free(ctx);
3903 		cb_fn(cb_arg, -ENOMEM);
3904 		return;
3905 	}
3906 
3907 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3908 	cpl.u.bs_basic.cb_fn = cb_fn;
3909 	cpl.u.bs_basic.cb_arg = cb_arg;
3910 
3911 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3912 	if (!seq) {
3913 		spdk_dma_free(ctx->super);
3914 		free(ctx);
3915 		cb_fn(cb_arg, -ENOMEM);
3916 		return;
3917 	}
3918 
3919 	bs->super_blob = blobid;
3920 
3921 	/* Read super block */
3922 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3923 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3924 				  _spdk_bs_set_super_read_cpl, ctx);
3925 }
3926 
3927 /* END spdk_bs_set_super */
3928 
3929 void
3930 spdk_bs_get_super(struct spdk_blob_store *bs,
3931 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
3932 {
3933 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
3934 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
3935 	} else {
3936 		cb_fn(cb_arg, bs->super_blob, 0);
3937 	}
3938 }
3939 
3940 uint64_t
3941 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
3942 {
3943 	return bs->cluster_sz;
3944 }
3945 
3946 uint64_t
3947 spdk_bs_get_page_size(struct spdk_blob_store *bs)
3948 {
3949 	return SPDK_BS_PAGE_SIZE;
3950 }
3951 
3952 uint64_t
3953 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
3954 {
3955 	return bs->num_free_clusters;
3956 }
3957 
3958 uint64_t
3959 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
3960 {
3961 	return bs->total_data_clusters;
3962 }
3963 
3964 static int
3965 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
3966 {
3967 	bs->md_channel = spdk_get_io_channel(bs);
3968 	if (!bs->md_channel) {
3969 		SPDK_ERRLOG("Failed to get IO channel.\n");
3970 		return -1;
3971 	}
3972 
3973 	return 0;
3974 }
3975 
3976 static int
3977 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
3978 {
3979 	spdk_put_io_channel(bs->md_channel);
3980 
3981 	return 0;
3982 }
3983 
3984 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
3985 {
3986 	assert(blob != NULL);
3987 
3988 	return blob->id;
3989 }
3990 
3991 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
3992 {
3993 	assert(blob != NULL);
3994 
3995 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
3996 }
3997 
3998 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
3999 {
4000 	assert(blob != NULL);
4001 
4002 	return blob->active.num_clusters;
4003 }
4004 
4005 /* START spdk_bs_create_blob */
4006 
4007 static void
4008 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4009 {
4010 	struct spdk_blob *blob = cb_arg;
4011 
4012 	_spdk_blob_free(blob);
4013 
4014 	spdk_bs_sequence_finish(seq, bserrno);
4015 }
4016 
4017 static int
4018 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4019 		      bool internal)
4020 {
4021 	uint64_t i;
4022 	size_t value_len = 0;
4023 	int rc;
4024 	const void *value = NULL;
4025 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4026 		return -EINVAL;
4027 	}
4028 	for (i = 0; i < xattrs->count; i++) {
4029 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4030 		if (value == NULL || value_len == 0) {
4031 			return -EINVAL;
4032 		}
4033 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4034 		if (rc < 0) {
4035 			return rc;
4036 		}
4037 	}
4038 	return 0;
4039 }
4040 
4041 static void
4042 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4043 {
4044 	_spdk_blob_verify_md_op(blob);
4045 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4046 	blob->state = SPDK_BLOB_STATE_DIRTY;
4047 }
4048 
4049 static void
4050 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4051 		     const struct spdk_blob_opts *opts,
4052 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4053 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4054 {
4055 	struct spdk_blob	*blob;
4056 	uint32_t		page_idx;
4057 	struct spdk_bs_cpl	cpl;
4058 	struct spdk_blob_opts	opts_default;
4059 	struct spdk_blob_xattr_opts internal_xattrs_default;
4060 	spdk_bs_sequence_t	*seq;
4061 	spdk_blob_id		id;
4062 	int rc;
4063 
4064 	assert(spdk_get_thread() == bs->md_thread);
4065 
4066 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4067 	if (page_idx >= spdk_bit_array_capacity(bs->used_md_pages)) {
4068 		cb_fn(cb_arg, 0, -ENOMEM);
4069 		return;
4070 	}
4071 	spdk_bit_array_set(bs->used_blobids, page_idx);
4072 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4073 
4074 	id = _spdk_bs_page_to_blobid(page_idx);
4075 
4076 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4077 
4078 	blob = _spdk_blob_alloc(bs, id);
4079 	if (!blob) {
4080 		cb_fn(cb_arg, 0, -ENOMEM);
4081 		return;
4082 	}
4083 
4084 	if (!opts) {
4085 		spdk_blob_opts_init(&opts_default);
4086 		opts = &opts_default;
4087 	}
4088 	if (!internal_xattrs) {
4089 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4090 		internal_xattrs = &internal_xattrs_default;
4091 	}
4092 
4093 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4094 	if (rc < 0) {
4095 		_spdk_blob_free(blob);
4096 		cb_fn(cb_arg, 0, rc);
4097 		return;
4098 	}
4099 
4100 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4101 	if (rc < 0) {
4102 		_spdk_blob_free(blob);
4103 		cb_fn(cb_arg, 0, rc);
4104 		return;
4105 	}
4106 
4107 	if (opts->thin_provision) {
4108 		_spdk_blob_set_thin_provision(blob);
4109 	}
4110 
4111 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4112 	if (rc < 0) {
4113 		_spdk_blob_free(blob);
4114 		cb_fn(cb_arg, 0, rc);
4115 		return;
4116 	}
4117 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4118 	cpl.u.blobid.cb_fn = cb_fn;
4119 	cpl.u.blobid.cb_arg = cb_arg;
4120 	cpl.u.blobid.blobid = blob->id;
4121 
4122 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4123 	if (!seq) {
4124 		_spdk_blob_free(blob);
4125 		cb_fn(cb_arg, 0, -ENOMEM);
4126 		return;
4127 	}
4128 
4129 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4130 }
4131 
4132 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4133 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4134 {
4135 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4136 }
4137 
4138 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4139 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4140 {
4141 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4142 }
4143 
4144 /* END spdk_bs_create_blob */
4145 
4146 /* START blob_cleanup */
4147 
4148 struct spdk_clone_snapshot_ctx {
4149 	struct spdk_bs_cpl      cpl;
4150 	int bserrno;
4151 	bool frozen;
4152 
4153 	struct spdk_io_channel *channel;
4154 
4155 	/* Current cluster for inflate operation */
4156 	uint64_t cluster;
4157 
4158 	/* For inflation force allocation of all unallocated clusters and remove
4159 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4160 	bool allocate_all;
4161 
4162 	struct {
4163 		spdk_blob_id id;
4164 		struct spdk_blob *blob;
4165 	} original;
4166 	struct {
4167 		spdk_blob_id id;
4168 		struct spdk_blob *blob;
4169 	} new;
4170 
4171 	/* xattrs specified for snapshot/clones only. They have no impact on
4172 	 * the original blobs xattrs. */
4173 	const struct spdk_blob_xattr_opts *xattrs;
4174 };
4175 
4176 static void
4177 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4178 {
4179 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4180 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4181 
4182 	if (bserrno != 0) {
4183 		if (ctx->bserrno != 0) {
4184 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4185 		} else {
4186 			ctx->bserrno = bserrno;
4187 		}
4188 	}
4189 
4190 	switch (cpl->type) {
4191 	case SPDK_BS_CPL_TYPE_BLOBID:
4192 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4193 		break;
4194 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4195 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4196 		break;
4197 	default:
4198 		SPDK_UNREACHABLE();
4199 		break;
4200 	}
4201 
4202 	free(ctx);
4203 }
4204 
4205 static void
4206 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4207 {
4208 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4209 	struct spdk_blob *origblob = ctx->original.blob;
4210 
4211 	if (bserrno != 0) {
4212 		if (ctx->bserrno != 0) {
4213 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4214 		} else {
4215 			ctx->bserrno = bserrno;
4216 		}
4217 	}
4218 
4219 	ctx->original.id = origblob->id;
4220 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4221 }
4222 
4223 static void
4224 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4225 {
4226 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4227 	struct spdk_blob *origblob = ctx->original.blob;
4228 
4229 	if (bserrno != 0) {
4230 		if (ctx->bserrno != 0) {
4231 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4232 		} else {
4233 			ctx->bserrno = bserrno;
4234 		}
4235 	}
4236 
4237 	if (ctx->frozen) {
4238 		/* Unfreeze any outstanding I/O */
4239 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4240 	} else {
4241 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4242 	}
4243 
4244 }
4245 
4246 static void
4247 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4248 {
4249 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4250 	struct spdk_blob *newblob = ctx->new.blob;
4251 
4252 	if (bserrno != 0) {
4253 		if (ctx->bserrno != 0) {
4254 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4255 		} else {
4256 			ctx->bserrno = bserrno;
4257 		}
4258 	}
4259 
4260 	ctx->new.id = newblob->id;
4261 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4262 }
4263 
4264 /* END blob_cleanup */
4265 
4266 /* START spdk_bs_create_snapshot */
4267 
4268 static void
4269 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4270 {
4271 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4272 	struct spdk_blob *newblob = ctx->new.blob;
4273 
4274 	if (bserrno != 0) {
4275 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4276 		return;
4277 	}
4278 
4279 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4280 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4281 	if (bserrno != 0) {
4282 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4283 		return;
4284 	}
4285 
4286 	_spdk_bs_blob_list_add(ctx->original.blob);
4287 
4288 	spdk_blob_set_read_only(newblob);
4289 
4290 	/* sync snapshot metadata */
4291 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4292 }
4293 
4294 static void
4295 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4296 {
4297 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4298 	struct spdk_blob *origblob = ctx->original.blob;
4299 	struct spdk_blob *newblob = ctx->new.blob;
4300 
4301 	if (bserrno != 0) {
4302 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4303 		return;
4304 	}
4305 
4306 	/* Set internal xattr for snapshot id */
4307 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4308 	if (bserrno != 0) {
4309 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4310 		return;
4311 	}
4312 
4313 	_spdk_bs_blob_list_remove(origblob);
4314 	origblob->parent_id = newblob->id;
4315 
4316 	/* Create new back_bs_dev for snapshot */
4317 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4318 	if (origblob->back_bs_dev == NULL) {
4319 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4320 		return;
4321 	}
4322 
4323 	/* set clone blob as thin provisioned */
4324 	_spdk_blob_set_thin_provision(origblob);
4325 
4326 	_spdk_bs_blob_list_add(newblob);
4327 
4328 	/* Zero out origblob cluster map */
4329 	memset(origblob->active.clusters, 0,
4330 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4331 
4332 	/* sync clone metadata */
4333 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4334 }
4335 
4336 static void
4337 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4338 {
4339 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4340 	struct spdk_blob *origblob = ctx->original.blob;
4341 	struct spdk_blob *newblob = ctx->new.blob;
4342 	int bserrno;
4343 
4344 	if (rc != 0) {
4345 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4346 		return;
4347 	}
4348 
4349 	ctx->frozen = true;
4350 
4351 	/* set new back_bs_dev for snapshot */
4352 	newblob->back_bs_dev = origblob->back_bs_dev;
4353 	/* Set invalid flags from origblob */
4354 	newblob->invalid_flags = origblob->invalid_flags;
4355 
4356 	/* inherit parent from original blob if set */
4357 	newblob->parent_id = origblob->parent_id;
4358 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4359 		/* Set internal xattr for snapshot id */
4360 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4361 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4362 		if (bserrno != 0) {
4363 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4364 			return;
4365 		}
4366 	}
4367 
4368 	/* Copy cluster map to snapshot */
4369 	memcpy(newblob->active.clusters, origblob->active.clusters,
4370 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4371 
4372 	/* sync snapshot metadata */
4373 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4374 }
4375 
4376 static void
4377 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4378 {
4379 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4380 	struct spdk_blob *origblob = ctx->original.blob;
4381 	struct spdk_blob *newblob = _blob;
4382 
4383 	if (bserrno != 0) {
4384 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4385 		return;
4386 	}
4387 
4388 	ctx->new.blob = newblob;
4389 
4390 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4391 }
4392 
4393 static void
4394 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4395 {
4396 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4397 	struct spdk_blob *origblob = ctx->original.blob;
4398 
4399 	if (bserrno != 0) {
4400 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4401 		return;
4402 	}
4403 
4404 	ctx->new.id = blobid;
4405 	ctx->cpl.u.blobid.blobid = blobid;
4406 
4407 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4408 }
4409 
4410 
4411 static void
4412 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4413 			const void **value, size_t *value_len)
4414 {
4415 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4416 
4417 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4418 	*value = &blob->id;
4419 	*value_len = sizeof(blob->id);
4420 }
4421 
4422 static void
4423 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4424 {
4425 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4426 	struct spdk_blob_opts opts;
4427 	struct spdk_blob_xattr_opts internal_xattrs;
4428 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4429 
4430 	if (bserrno != 0) {
4431 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4432 		return;
4433 	}
4434 
4435 	ctx->original.blob = _blob;
4436 
4437 	if (_blob->data_ro || _blob->md_ro) {
4438 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4439 			      _blob->id);
4440 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4441 		return;
4442 	}
4443 
4444 	spdk_blob_opts_init(&opts);
4445 	_spdk_blob_xattrs_init(&internal_xattrs);
4446 
4447 	/* Change the size of new blob to the same as in original blob,
4448 	 * but do not allocate clusters */
4449 	opts.thin_provision = true;
4450 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4451 
4452 	/* If there are any xattrs specified for snapshot, set them now */
4453 	if (ctx->xattrs) {
4454 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4455 	}
4456 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4457 	internal_xattrs.count = 1;
4458 	internal_xattrs.ctx = _blob;
4459 	internal_xattrs.names = xattrs_names;
4460 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4461 
4462 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4463 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4464 }
4465 
4466 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4467 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4468 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4469 {
4470 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4471 
4472 	if (!ctx) {
4473 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4474 		return;
4475 	}
4476 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4477 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4478 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4479 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4480 	ctx->bserrno = 0;
4481 	ctx->frozen = false;
4482 	ctx->original.id = blobid;
4483 	ctx->xattrs = snapshot_xattrs;
4484 
4485 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4486 }
4487 /* END spdk_bs_create_snapshot */
4488 
4489 /* START spdk_bs_create_clone */
4490 
4491 static void
4492 _spdk_bs_xattr_clone(void *arg, const char *name,
4493 		     const void **value, size_t *value_len)
4494 {
4495 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4496 
4497 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4498 	*value = &blob->id;
4499 	*value_len = sizeof(blob->id);
4500 }
4501 
4502 static void
4503 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4504 {
4505 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4506 	struct spdk_blob *clone = _blob;
4507 
4508 	ctx->new.blob = clone;
4509 	_spdk_bs_blob_list_add(clone);
4510 
4511 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4512 }
4513 
4514 static void
4515 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4516 {
4517 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4518 
4519 	ctx->cpl.u.blobid.blobid = blobid;
4520 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4521 }
4522 
4523 static void
4524 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4525 {
4526 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4527 	struct spdk_blob_opts		opts;
4528 	struct spdk_blob_xattr_opts internal_xattrs;
4529 	char *xattr_names[] = { BLOB_SNAPSHOT };
4530 
4531 	if (bserrno != 0) {
4532 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4533 		return;
4534 	}
4535 
4536 	ctx->original.blob = _blob;
4537 
4538 	if (!_blob->data_ro || !_blob->md_ro) {
4539 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4540 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4541 		return;
4542 	}
4543 
4544 	spdk_blob_opts_init(&opts);
4545 	_spdk_blob_xattrs_init(&internal_xattrs);
4546 
4547 	opts.thin_provision = true;
4548 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4549 	if (ctx->xattrs) {
4550 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4551 	}
4552 
4553 	/* Set internal xattr BLOB_SNAPSHOT */
4554 	internal_xattrs.count = 1;
4555 	internal_xattrs.ctx = _blob;
4556 	internal_xattrs.names = xattr_names;
4557 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4558 
4559 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4560 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4561 }
4562 
4563 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4564 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4565 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4566 {
4567 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4568 
4569 	if (!ctx) {
4570 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4571 		return;
4572 	}
4573 
4574 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4575 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4576 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4577 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4578 	ctx->bserrno = 0;
4579 	ctx->xattrs = clone_xattrs;
4580 	ctx->original.id = blobid;
4581 
4582 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4583 }
4584 
4585 /* END spdk_bs_create_clone */
4586 
4587 /* START spdk_bs_inflate_blob */
4588 
4589 static void
4590 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4591 {
4592 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4593 	struct spdk_blob *_blob = ctx->original.blob;
4594 
4595 	if (bserrno != 0) {
4596 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4597 		return;
4598 	}
4599 
4600 	assert(_parent != NULL);
4601 
4602 	_spdk_bs_blob_list_remove(_blob);
4603 	_blob->parent_id = _parent->id;
4604 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4605 			     sizeof(spdk_blob_id), true);
4606 
4607 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4608 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4609 	_spdk_bs_blob_list_add(_blob);
4610 
4611 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4612 }
4613 
4614 static void
4615 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4616 {
4617 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4618 	struct spdk_blob *_blob = ctx->original.blob;
4619 	struct spdk_blob *_parent;
4620 
4621 	if (bserrno != 0) {
4622 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4623 		return;
4624 	}
4625 
4626 	if (ctx->allocate_all) {
4627 		/* remove thin provisioning */
4628 		_spdk_bs_blob_list_remove(_blob);
4629 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4630 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4631 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4632 		_blob->back_bs_dev = NULL;
4633 		_blob->parent_id = SPDK_BLOBID_INVALID;
4634 	} else {
4635 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4636 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4637 			/* We must change the parent of the inflated blob */
4638 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4639 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4640 			return;
4641 		}
4642 
4643 		_spdk_bs_blob_list_remove(_blob);
4644 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4645 		_blob->parent_id = SPDK_BLOBID_INVALID;
4646 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4647 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4648 	}
4649 
4650 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4651 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4652 }
4653 
4654 /* Check if cluster needs allocation */
4655 static inline bool
4656 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4657 {
4658 	struct spdk_blob_bs_dev *b;
4659 
4660 	assert(blob != NULL);
4661 
4662 	if (blob->active.clusters[cluster] != 0) {
4663 		/* Cluster is already allocated */
4664 		return false;
4665 	}
4666 
4667 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4668 		/* Blob have no parent blob */
4669 		return allocate_all;
4670 	}
4671 
4672 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4673 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4674 }
4675 
4676 static void
4677 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4678 {
4679 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4680 	struct spdk_blob *_blob = ctx->original.blob;
4681 	uint64_t offset;
4682 
4683 	if (bserrno != 0) {
4684 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4685 		return;
4686 	}
4687 
4688 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4689 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4690 			break;
4691 		}
4692 	}
4693 
4694 	if (ctx->cluster < _blob->active.num_clusters) {
4695 		offset = _spdk_bs_cluster_to_page(_blob->bs, ctx->cluster);
4696 
4697 		/* We may safely increment a cluster before write */
4698 		ctx->cluster++;
4699 
4700 		/* Use zero length write to touch a cluster */
4701 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4702 				   _spdk_bs_inflate_blob_touch_next, ctx);
4703 	} else {
4704 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4705 	}
4706 }
4707 
4708 static void
4709 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4710 {
4711 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4712 	uint64_t lfc; /* lowest free cluster */
4713 	uint64_t i;
4714 
4715 	if (bserrno != 0) {
4716 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4717 		return;
4718 	}
4719 	ctx->original.blob = _blob;
4720 
4721 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4722 		/* This blob have no parent, so we cannot decouple it. */
4723 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4724 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4725 		return;
4726 	}
4727 
4728 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4729 		/* This is not thin provisioned blob. No need to inflate. */
4730 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4731 		return;
4732 	}
4733 
4734 	/* Do two passes - one to verify that we can obtain enough clusters
4735 	 * and another to actually claim them.
4736 	 */
4737 	lfc = 0;
4738 	for (i = 0; i < _blob->active.num_clusters; i++) {
4739 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4740 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4741 			if (lfc >= _blob->bs->total_clusters) {
4742 				/* No more free clusters. Cannot satisfy the request */
4743 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4744 				return;
4745 			}
4746 			lfc++;
4747 		}
4748 	}
4749 
4750 	ctx->cluster = 0;
4751 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4752 }
4753 
4754 static void
4755 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4756 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4757 {
4758 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4759 
4760 	if (!ctx) {
4761 		cb_fn(cb_arg, -ENOMEM);
4762 		return;
4763 	}
4764 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4765 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4766 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4767 	ctx->bserrno = 0;
4768 	ctx->original.id = blobid;
4769 	ctx->channel = channel;
4770 	ctx->allocate_all = allocate_all;
4771 
4772 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4773 }
4774 
4775 void
4776 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4777 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4778 {
4779 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4780 }
4781 
4782 void
4783 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4784 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4785 {
4786 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
4787 }
4788 /* END spdk_bs_inflate_blob */
4789 
4790 /* START spdk_blob_resize */
4791 struct spdk_bs_resize_ctx {
4792 	spdk_blob_op_complete cb_fn;
4793 	void *cb_arg;
4794 	struct spdk_blob *blob;
4795 	uint64_t sz;
4796 	int rc;
4797 };
4798 
4799 static void
4800 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4801 {
4802 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4803 
4804 	if (rc != 0) {
4805 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4806 	}
4807 
4808 	if (ctx->rc != 0) {
4809 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4810 		rc = ctx->rc;
4811 	}
4812 
4813 	ctx->blob->resize_in_progress = false;
4814 
4815 	ctx->cb_fn(ctx->cb_arg, rc);
4816 	free(ctx);
4817 }
4818 
4819 static void
4820 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4821 {
4822 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4823 
4824 	if (rc != 0) {
4825 		ctx->blob->resize_in_progress = false;
4826 		ctx->cb_fn(ctx->cb_arg, rc);
4827 		free(ctx);
4828 		return;
4829 	}
4830 
4831 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4832 
4833 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4834 }
4835 
4836 void
4837 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4838 {
4839 	struct spdk_bs_resize_ctx *ctx;
4840 
4841 	_spdk_blob_verify_md_op(blob);
4842 
4843 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4844 
4845 	if (blob->md_ro) {
4846 		cb_fn(cb_arg, -EPERM);
4847 		return;
4848 	}
4849 
4850 	if (sz == blob->active.num_clusters) {
4851 		cb_fn(cb_arg, 0);
4852 		return;
4853 	}
4854 
4855 	if (blob->resize_in_progress) {
4856 		cb_fn(cb_arg, -EBUSY);
4857 		return;
4858 	}
4859 
4860 	ctx = calloc(1, sizeof(*ctx));
4861 	if (!ctx) {
4862 		cb_fn(cb_arg, -ENOMEM);
4863 		return;
4864 	}
4865 
4866 	blob->resize_in_progress = true;
4867 	ctx->cb_fn = cb_fn;
4868 	ctx->cb_arg = cb_arg;
4869 	ctx->blob = blob;
4870 	ctx->sz = sz;
4871 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
4872 }
4873 
4874 /* END spdk_blob_resize */
4875 
4876 
4877 /* START spdk_bs_delete_blob */
4878 
4879 static void
4880 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
4881 {
4882 	spdk_bs_sequence_t *seq = cb_arg;
4883 
4884 	spdk_bs_sequence_finish(seq, bserrno);
4885 }
4886 
4887 static void
4888 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4889 {
4890 	struct spdk_blob *blob = cb_arg;
4891 
4892 	if (bserrno != 0) {
4893 		/*
4894 		 * We already removed this blob from the blobstore tailq, so
4895 		 *  we need to free it here since this is the last reference
4896 		 *  to it.
4897 		 */
4898 		_spdk_blob_free(blob);
4899 		_spdk_bs_delete_close_cpl(seq, bserrno);
4900 		return;
4901 	}
4902 
4903 	/*
4904 	 * This will immediately decrement the ref_count and call
4905 	 *  the completion routine since the metadata state is clean.
4906 	 *  By calling spdk_blob_close, we reduce the number of call
4907 	 *  points into code that touches the blob->open_ref count
4908 	 *  and the blobstore's blob list.
4909 	 */
4910 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
4911 }
4912 
4913 static void
4914 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
4915 {
4916 	spdk_bs_sequence_t *seq = cb_arg;
4917 	uint32_t page_num;
4918 
4919 	if (bserrno != 0) {
4920 		spdk_bs_sequence_finish(seq, bserrno);
4921 		return;
4922 	}
4923 
4924 	_spdk_blob_verify_md_op(blob);
4925 
4926 	if (blob->open_ref > 1) {
4927 		/*
4928 		 * Someone has this blob open (besides this delete context).
4929 		 *  Decrement the ref count directly and return -EBUSY.
4930 		 */
4931 		blob->open_ref--;
4932 		spdk_bs_sequence_finish(seq, -EBUSY);
4933 		return;
4934 	}
4935 
4936 	bserrno = _spdk_bs_blob_list_remove(blob);
4937 	if (bserrno != 0) {
4938 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Remove blob #%" PRIu64 " from a list\n", blob->id);
4939 		spdk_bs_sequence_finish(seq, bserrno);
4940 		return;
4941 	}
4942 
4943 	/*
4944 	 * Remove the blob from the blob_store list now, to ensure it does not
4945 	 *  get returned after this point by _spdk_blob_lookup().
4946 	 */
4947 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
4948 	page_num = _spdk_bs_blobid_to_page(blob->id);
4949 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
4950 	blob->state = SPDK_BLOB_STATE_DIRTY;
4951 	blob->active.num_pages = 0;
4952 	_spdk_blob_resize(blob, 0);
4953 
4954 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
4955 }
4956 
4957 void
4958 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
4959 		    spdk_blob_op_complete cb_fn, void *cb_arg)
4960 {
4961 	struct spdk_bs_cpl	cpl;
4962 	spdk_bs_sequence_t	*seq;
4963 	struct spdk_blob_list	*snapshot_entry = NULL;
4964 
4965 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
4966 
4967 	assert(spdk_get_thread() == bs->md_thread);
4968 
4969 	/* Check if this is a snapshot with clones */
4970 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
4971 		if (snapshot_entry->id == blobid) {
4972 			break;
4973 		}
4974 	}
4975 	if (snapshot_entry != NULL) {
4976 		/* If snapshot have clones, we cannot remove it */
4977 		if (!TAILQ_EMPTY(&snapshot_entry->clones)) {
4978 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
4979 			cb_fn(cb_arg, -EBUSY);
4980 			return;
4981 		}
4982 	}
4983 
4984 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4985 	cpl.u.blob_basic.cb_fn = cb_fn;
4986 	cpl.u.blob_basic.cb_arg = cb_arg;
4987 
4988 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4989 	if (!seq) {
4990 		cb_fn(cb_arg, -ENOMEM);
4991 		return;
4992 	}
4993 
4994 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
4995 }
4996 
4997 /* END spdk_bs_delete_blob */
4998 
4999 /* START spdk_bs_open_blob */
5000 
5001 static void
5002 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5003 {
5004 	struct spdk_blob *blob = cb_arg;
5005 
5006 	/* If the blob have crc error, we just return NULL. */
5007 	if (blob == NULL) {
5008 		seq->cpl.u.blob_handle.blob = NULL;
5009 		spdk_bs_sequence_finish(seq, bserrno);
5010 		return;
5011 	}
5012 
5013 	blob->open_ref++;
5014 
5015 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5016 
5017 	spdk_bs_sequence_finish(seq, bserrno);
5018 }
5019 
5020 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5021 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5022 {
5023 	struct spdk_blob		*blob;
5024 	struct spdk_bs_cpl		cpl;
5025 	spdk_bs_sequence_t		*seq;
5026 	uint32_t			page_num;
5027 
5028 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5029 	assert(spdk_get_thread() == bs->md_thread);
5030 
5031 	page_num = _spdk_bs_blobid_to_page(blobid);
5032 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5033 		/* Invalid blobid */
5034 		cb_fn(cb_arg, NULL, -ENOENT);
5035 		return;
5036 	}
5037 
5038 	blob = _spdk_blob_lookup(bs, blobid);
5039 	if (blob) {
5040 		blob->open_ref++;
5041 		cb_fn(cb_arg, blob, 0);
5042 		return;
5043 	}
5044 
5045 	blob = _spdk_blob_alloc(bs, blobid);
5046 	if (!blob) {
5047 		cb_fn(cb_arg, NULL, -ENOMEM);
5048 		return;
5049 	}
5050 
5051 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5052 	cpl.u.blob_handle.cb_fn = cb_fn;
5053 	cpl.u.blob_handle.cb_arg = cb_arg;
5054 	cpl.u.blob_handle.blob = blob;
5055 
5056 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5057 	if (!seq) {
5058 		_spdk_blob_free(blob);
5059 		cb_fn(cb_arg, NULL, -ENOMEM);
5060 		return;
5061 	}
5062 
5063 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5064 }
5065 /* END spdk_bs_open_blob */
5066 
5067 /* START spdk_blob_set_read_only */
5068 int spdk_blob_set_read_only(struct spdk_blob *blob)
5069 {
5070 	_spdk_blob_verify_md_op(blob);
5071 
5072 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5073 
5074 	blob->state = SPDK_BLOB_STATE_DIRTY;
5075 	return 0;
5076 }
5077 /* END spdk_blob_set_read_only */
5078 
5079 /* START spdk_blob_sync_md */
5080 
5081 static void
5082 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5083 {
5084 	struct spdk_blob *blob = cb_arg;
5085 
5086 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5087 		blob->data_ro = true;
5088 		blob->md_ro = true;
5089 	}
5090 
5091 	spdk_bs_sequence_finish(seq, bserrno);
5092 }
5093 
5094 static void
5095 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5096 {
5097 	struct spdk_bs_cpl	cpl;
5098 	spdk_bs_sequence_t	*seq;
5099 
5100 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5101 	cpl.u.blob_basic.cb_fn = cb_fn;
5102 	cpl.u.blob_basic.cb_arg = cb_arg;
5103 
5104 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5105 	if (!seq) {
5106 		cb_fn(cb_arg, -ENOMEM);
5107 		return;
5108 	}
5109 
5110 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5111 }
5112 
5113 void
5114 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5115 {
5116 	_spdk_blob_verify_md_op(blob);
5117 
5118 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5119 
5120 	if (blob->md_ro) {
5121 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5122 		cb_fn(cb_arg, 0);
5123 		return;
5124 	}
5125 
5126 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5127 }
5128 
5129 /* END spdk_blob_sync_md */
5130 
5131 struct spdk_blob_insert_cluster_ctx {
5132 	struct spdk_thread	*thread;
5133 	struct spdk_blob	*blob;
5134 	uint32_t		cluster_num;	/* cluster index in blob */
5135 	uint32_t		cluster;	/* cluster on disk */
5136 	int			rc;
5137 	spdk_blob_op_complete	cb_fn;
5138 	void			*cb_arg;
5139 };
5140 
5141 static void
5142 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5143 {
5144 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5145 
5146 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5147 	free(ctx);
5148 }
5149 
5150 static void
5151 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5152 {
5153 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5154 
5155 	ctx->rc = bserrno;
5156 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5157 }
5158 
5159 static void
5160 _spdk_blob_insert_cluster_msg(void *arg)
5161 {
5162 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5163 
5164 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5165 	if (ctx->rc != 0) {
5166 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5167 		return;
5168 	}
5169 
5170 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5171 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5172 }
5173 
5174 static void
5175 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5176 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5177 {
5178 	struct spdk_blob_insert_cluster_ctx *ctx;
5179 
5180 	ctx = calloc(1, sizeof(*ctx));
5181 	if (ctx == NULL) {
5182 		cb_fn(cb_arg, -ENOMEM);
5183 		return;
5184 	}
5185 
5186 	ctx->thread = spdk_get_thread();
5187 	ctx->blob = blob;
5188 	ctx->cluster_num = cluster_num;
5189 	ctx->cluster = cluster;
5190 	ctx->cb_fn = cb_fn;
5191 	ctx->cb_arg = cb_arg;
5192 
5193 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5194 }
5195 
5196 /* START spdk_blob_close */
5197 
5198 static void
5199 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5200 {
5201 	struct spdk_blob *blob = cb_arg;
5202 
5203 	if (bserrno == 0) {
5204 		blob->open_ref--;
5205 		if (blob->open_ref == 0) {
5206 			/*
5207 			 * Blobs with active.num_pages == 0 are deleted blobs.
5208 			 *  these blobs are removed from the blob_store list
5209 			 *  when the deletion process starts - so don't try to
5210 			 *  remove them again.
5211 			 */
5212 			if (blob->active.num_pages > 0) {
5213 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5214 			}
5215 			_spdk_blob_free(blob);
5216 		}
5217 	}
5218 
5219 	spdk_bs_sequence_finish(seq, bserrno);
5220 }
5221 
5222 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5223 {
5224 	struct spdk_bs_cpl	cpl;
5225 	spdk_bs_sequence_t	*seq;
5226 
5227 	_spdk_blob_verify_md_op(blob);
5228 
5229 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5230 
5231 	if (blob->open_ref == 0) {
5232 		cb_fn(cb_arg, -EBADF);
5233 		return;
5234 	}
5235 
5236 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5237 	cpl.u.blob_basic.cb_fn = cb_fn;
5238 	cpl.u.blob_basic.cb_arg = cb_arg;
5239 
5240 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5241 	if (!seq) {
5242 		cb_fn(cb_arg, -ENOMEM);
5243 		return;
5244 	}
5245 
5246 	/* Sync metadata */
5247 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5248 }
5249 
5250 /* END spdk_blob_close */
5251 
5252 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5253 {
5254 	return spdk_get_io_channel(bs);
5255 }
5256 
5257 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5258 {
5259 	spdk_put_io_channel(channel);
5260 }
5261 
5262 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5263 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5264 {
5265 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5266 				     SPDK_BLOB_UNMAP);
5267 }
5268 
5269 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5270 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5271 {
5272 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5273 				     SPDK_BLOB_WRITE_ZEROES);
5274 }
5275 
5276 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5277 			void *payload, uint64_t offset, uint64_t length,
5278 			spdk_blob_op_complete cb_fn, void *cb_arg)
5279 {
5280 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5281 				     SPDK_BLOB_WRITE);
5282 }
5283 
5284 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5285 		       void *payload, uint64_t offset, uint64_t length,
5286 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5287 {
5288 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5289 				     SPDK_BLOB_READ);
5290 }
5291 
5292 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5293 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5294 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5295 {
5296 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5297 }
5298 
5299 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5300 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5301 			spdk_blob_op_complete cb_fn, void *cb_arg)
5302 {
5303 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5304 }
5305 
5306 struct spdk_bs_iter_ctx {
5307 	int64_t page_num;
5308 	struct spdk_blob_store *bs;
5309 
5310 	spdk_blob_op_with_handle_complete cb_fn;
5311 	void *cb_arg;
5312 };
5313 
5314 static void
5315 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5316 {
5317 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5318 	struct spdk_blob_store *bs = ctx->bs;
5319 	spdk_blob_id id;
5320 
5321 	if (bserrno == 0) {
5322 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5323 		free(ctx);
5324 		return;
5325 	}
5326 
5327 	ctx->page_num++;
5328 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5329 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5330 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5331 		free(ctx);
5332 		return;
5333 	}
5334 
5335 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5336 
5337 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5338 }
5339 
5340 void
5341 spdk_bs_iter_first(struct spdk_blob_store *bs,
5342 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5343 {
5344 	struct spdk_bs_iter_ctx *ctx;
5345 
5346 	ctx = calloc(1, sizeof(*ctx));
5347 	if (!ctx) {
5348 		cb_fn(cb_arg, NULL, -ENOMEM);
5349 		return;
5350 	}
5351 
5352 	ctx->page_num = -1;
5353 	ctx->bs = bs;
5354 	ctx->cb_fn = cb_fn;
5355 	ctx->cb_arg = cb_arg;
5356 
5357 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5358 }
5359 
5360 static void
5361 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5362 {
5363 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5364 
5365 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5366 }
5367 
5368 void
5369 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5370 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5371 {
5372 	struct spdk_bs_iter_ctx *ctx;
5373 
5374 	assert(blob != NULL);
5375 
5376 	ctx = calloc(1, sizeof(*ctx));
5377 	if (!ctx) {
5378 		cb_fn(cb_arg, NULL, -ENOMEM);
5379 		return;
5380 	}
5381 
5382 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5383 	ctx->bs = bs;
5384 	ctx->cb_fn = cb_fn;
5385 	ctx->cb_arg = cb_arg;
5386 
5387 	/* Close the existing blob */
5388 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5389 }
5390 
5391 static int
5392 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5393 		     uint16_t value_len, bool internal)
5394 {
5395 	struct spdk_xattr_tailq *xattrs;
5396 	struct spdk_xattr	*xattr;
5397 
5398 	_spdk_blob_verify_md_op(blob);
5399 
5400 	if (blob->md_ro) {
5401 		return -EPERM;
5402 	}
5403 
5404 	if (internal) {
5405 		xattrs = &blob->xattrs_internal;
5406 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5407 	} else {
5408 		xattrs = &blob->xattrs;
5409 	}
5410 
5411 	TAILQ_FOREACH(xattr, xattrs, link) {
5412 		if (!strcmp(name, xattr->name)) {
5413 			free(xattr->value);
5414 			xattr->value_len = value_len;
5415 			xattr->value = malloc(value_len);
5416 			memcpy(xattr->value, value, value_len);
5417 
5418 			blob->state = SPDK_BLOB_STATE_DIRTY;
5419 
5420 			return 0;
5421 		}
5422 	}
5423 
5424 	xattr = calloc(1, sizeof(*xattr));
5425 	if (!xattr) {
5426 		return -ENOMEM;
5427 	}
5428 	xattr->name = strdup(name);
5429 	xattr->value_len = value_len;
5430 	xattr->value = malloc(value_len);
5431 	memcpy(xattr->value, value, value_len);
5432 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5433 
5434 	blob->state = SPDK_BLOB_STATE_DIRTY;
5435 
5436 	return 0;
5437 }
5438 
5439 int
5440 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5441 		    uint16_t value_len)
5442 {
5443 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5444 }
5445 
5446 static int
5447 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5448 {
5449 	struct spdk_xattr_tailq *xattrs;
5450 	struct spdk_xattr	*xattr;
5451 
5452 	_spdk_blob_verify_md_op(blob);
5453 
5454 	if (blob->md_ro) {
5455 		return -EPERM;
5456 	}
5457 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5458 
5459 	TAILQ_FOREACH(xattr, xattrs, link) {
5460 		if (!strcmp(name, xattr->name)) {
5461 			TAILQ_REMOVE(xattrs, xattr, link);
5462 			free(xattr->value);
5463 			free(xattr->name);
5464 			free(xattr);
5465 
5466 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5467 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5468 			}
5469 			blob->state = SPDK_BLOB_STATE_DIRTY;
5470 
5471 			return 0;
5472 		}
5473 	}
5474 
5475 	return -ENOENT;
5476 }
5477 
5478 int
5479 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5480 {
5481 	return _spdk_blob_remove_xattr(blob, name, false);
5482 }
5483 
5484 static int
5485 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5486 			   const void **value, size_t *value_len, bool internal)
5487 {
5488 	struct spdk_xattr	*xattr;
5489 	struct spdk_xattr_tailq *xattrs;
5490 
5491 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5492 
5493 	TAILQ_FOREACH(xattr, xattrs, link) {
5494 		if (!strcmp(name, xattr->name)) {
5495 			*value = xattr->value;
5496 			*value_len = xattr->value_len;
5497 			return 0;
5498 		}
5499 	}
5500 	return -ENOENT;
5501 }
5502 
5503 int
5504 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5505 			  const void **value, size_t *value_len)
5506 {
5507 	_spdk_blob_verify_md_op(blob);
5508 
5509 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5510 }
5511 
5512 struct spdk_xattr_names {
5513 	uint32_t	count;
5514 	const char	*names[0];
5515 };
5516 
5517 static int
5518 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5519 {
5520 	struct spdk_xattr	*xattr;
5521 	int			count = 0;
5522 
5523 	TAILQ_FOREACH(xattr, xattrs, link) {
5524 		count++;
5525 	}
5526 
5527 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5528 	if (*names == NULL) {
5529 		return -ENOMEM;
5530 	}
5531 
5532 	TAILQ_FOREACH(xattr, xattrs, link) {
5533 		(*names)->names[(*names)->count++] = xattr->name;
5534 	}
5535 
5536 	return 0;
5537 }
5538 
5539 int
5540 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5541 {
5542 	_spdk_blob_verify_md_op(blob);
5543 
5544 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5545 }
5546 
5547 uint32_t
5548 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5549 {
5550 	assert(names != NULL);
5551 
5552 	return names->count;
5553 }
5554 
5555 const char *
5556 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5557 {
5558 	if (index >= names->count) {
5559 		return NULL;
5560 	}
5561 
5562 	return names->names[index];
5563 }
5564 
5565 void
5566 spdk_xattr_names_free(struct spdk_xattr_names *names)
5567 {
5568 	free(names);
5569 }
5570 
5571 struct spdk_bs_type
5572 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5573 {
5574 	return bs->bstype;
5575 }
5576 
5577 void
5578 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5579 {
5580 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5581 }
5582 
5583 bool
5584 spdk_blob_is_read_only(struct spdk_blob *blob)
5585 {
5586 	assert(blob != NULL);
5587 	return (blob->data_ro || blob->md_ro);
5588 }
5589 
5590 bool
5591 spdk_blob_is_snapshot(struct spdk_blob *blob)
5592 {
5593 	struct spdk_blob_list *snapshot_entry;
5594 
5595 	assert(blob != NULL);
5596 
5597 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
5598 		if (snapshot_entry->id == blob->id) {
5599 			break;
5600 		}
5601 	}
5602 
5603 	if (snapshot_entry == NULL) {
5604 		return false;
5605 	}
5606 
5607 	return true;
5608 }
5609 
5610 bool
5611 spdk_blob_is_clone(struct spdk_blob *blob)
5612 {
5613 	assert(blob != NULL);
5614 
5615 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5616 		assert(spdk_blob_is_thin_provisioned(blob));
5617 		return true;
5618 	}
5619 
5620 	return false;
5621 }
5622 
5623 bool
5624 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5625 {
5626 	assert(blob != NULL);
5627 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5628 }
5629 
5630 spdk_blob_id
5631 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5632 {
5633 	struct spdk_blob_list *snapshot_entry = NULL;
5634 	struct spdk_blob_list *clone_entry = NULL;
5635 
5636 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5637 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5638 			if (clone_entry->id == blob_id) {
5639 				return snapshot_entry->id;
5640 			}
5641 		}
5642 	}
5643 
5644 	return SPDK_BLOBID_INVALID;
5645 }
5646 
5647 int
5648 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5649 		     size_t *count)
5650 {
5651 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5652 	size_t n;
5653 
5654 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5655 		if (snapshot_entry->id == blobid) {
5656 			break;
5657 		}
5658 	}
5659 	if (snapshot_entry == NULL) {
5660 		*count = 0;
5661 		return 0;
5662 	}
5663 
5664 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5665 		*count = snapshot_entry->clone_count;
5666 		return -ENOMEM;
5667 	}
5668 	*count = snapshot_entry->clone_count;
5669 
5670 	n = 0;
5671 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5672 		ids[n++] = clone_entry->id;
5673 	}
5674 
5675 	return 0;
5676 }
5677 
5678 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5679