xref: /spdk/lib/blob/blobstore.c (revision c94020001adcc0d8fae89ff7902e2e0f2243e25d)
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 (desc_extent->extents[i].cluster_idx != 0) {
477 						if (!spdk_bit_array_get(blob->bs->used_clusters,
478 									desc_extent->extents[i].cluster_idx + j)) {
479 							return -EINVAL;
480 						}
481 					}
482 					cluster_count++;
483 				}
484 			}
485 
486 			if (cluster_count == 0) {
487 				return -EINVAL;
488 			}
489 			tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t));
490 			if (tmp == NULL) {
491 				return -ENOMEM;
492 			}
493 			blob->active.clusters = tmp;
494 			blob->active.cluster_array_size = cluster_count;
495 
496 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
497 				for (j = 0; j < desc_extent->extents[i].length; j++) {
498 					if (desc_extent->extents[i].cluster_idx != 0) {
499 						blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs,
500 								desc_extent->extents[i].cluster_idx + j);
501 					} else if (spdk_blob_is_thin_provisioned(blob)) {
502 						blob->active.clusters[blob->active.num_clusters++] = 0;
503 					} else {
504 						return -EINVAL;
505 					}
506 				}
507 			}
508 
509 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
510 			int rc;
511 
512 			rc = _spdk_blob_deserialize_xattr(blob,
513 							  (struct spdk_blob_md_descriptor_xattr *) desc, false);
514 			if (rc != 0) {
515 				return rc;
516 			}
517 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
518 			int rc;
519 
520 			rc = _spdk_blob_deserialize_xattr(blob,
521 							  (struct spdk_blob_md_descriptor_xattr *) desc, true);
522 			if (rc != 0) {
523 				return rc;
524 			}
525 		} else {
526 			/* Unrecognized descriptor type.  Do not fail - just continue to the
527 			 *  next descriptor.  If this descriptor is associated with some feature
528 			 *  defined in a newer version of blobstore, that version of blobstore
529 			 *  should create and set an associated feature flag to specify if this
530 			 *  blob can be loaded or not.
531 			 */
532 		}
533 
534 		/* Advance to the next descriptor */
535 		cur_desc += sizeof(*desc) + desc->length;
536 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
537 			break;
538 		}
539 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
540 	}
541 
542 	return 0;
543 }
544 
545 static int
546 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count,
547 		 struct spdk_blob *blob)
548 {
549 	const struct spdk_blob_md_page *page;
550 	uint32_t i;
551 	int rc;
552 
553 	assert(page_count > 0);
554 	assert(pages[0].sequence_num == 0);
555 	assert(blob != NULL);
556 	assert(blob->state == SPDK_BLOB_STATE_LOADING);
557 	assert(blob->active.clusters == NULL);
558 
559 	/* The blobid provided doesn't match what's in the MD, this can
560 	 * happen for example if a bogus blobid is passed in through open.
561 	 */
562 	if (blob->id != pages[0].id) {
563 		SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n",
564 			    blob->id, pages[0].id);
565 		return -ENOENT;
566 	}
567 
568 	for (i = 0; i < page_count; i++) {
569 		page = &pages[i];
570 
571 		assert(page->id == blob->id);
572 		assert(page->sequence_num == i);
573 
574 		rc = _spdk_blob_parse_page(page, blob);
575 		if (rc != 0) {
576 			return rc;
577 		}
578 	}
579 
580 	return 0;
581 }
582 
583 static int
584 _spdk_blob_serialize_add_page(const struct spdk_blob *blob,
585 			      struct spdk_blob_md_page **pages,
586 			      uint32_t *page_count,
587 			      struct spdk_blob_md_page **last_page)
588 {
589 	struct spdk_blob_md_page *page;
590 
591 	assert(pages != NULL);
592 	assert(page_count != NULL);
593 
594 	if (*page_count == 0) {
595 		assert(*pages == NULL);
596 		*page_count = 1;
597 		*pages = spdk_dma_malloc(SPDK_BS_PAGE_SIZE,
598 					 SPDK_BS_PAGE_SIZE,
599 					 NULL);
600 	} else {
601 		assert(*pages != NULL);
602 		(*page_count)++;
603 		*pages = spdk_dma_realloc(*pages,
604 					  SPDK_BS_PAGE_SIZE * (*page_count),
605 					  SPDK_BS_PAGE_SIZE,
606 					  NULL);
607 	}
608 
609 	if (*pages == NULL) {
610 		*page_count = 0;
611 		*last_page = NULL;
612 		return -ENOMEM;
613 	}
614 
615 	page = &(*pages)[*page_count - 1];
616 	memset(page, 0, sizeof(*page));
617 	page->id = blob->id;
618 	page->sequence_num = *page_count - 1;
619 	page->next = SPDK_INVALID_MD_PAGE;
620 	*last_page = page;
621 
622 	return 0;
623 }
624 
625 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor.
626  * Update required_sz on both success and failure.
627  *
628  */
629 static int
630 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr,
631 			   uint8_t *buf, size_t buf_sz,
632 			   size_t *required_sz, bool internal)
633 {
634 	struct spdk_blob_md_descriptor_xattr	*desc;
635 
636 	*required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) +
637 		       strlen(xattr->name) +
638 		       xattr->value_len;
639 
640 	if (buf_sz < *required_sz) {
641 		return -1;
642 	}
643 
644 	desc = (struct spdk_blob_md_descriptor_xattr *)buf;
645 
646 	desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR;
647 	desc->length = sizeof(desc->name_length) +
648 		       sizeof(desc->value_length) +
649 		       strlen(xattr->name) +
650 		       xattr->value_len;
651 	desc->name_length = strlen(xattr->name);
652 	desc->value_length = xattr->value_len;
653 
654 	memcpy(desc->name, xattr->name, desc->name_length);
655 	memcpy((void *)((uintptr_t)desc->name + desc->name_length),
656 	       xattr->value,
657 	       desc->value_length);
658 
659 	return 0;
660 }
661 
662 static void
663 _spdk_blob_serialize_extent(const struct spdk_blob *blob,
664 			    uint64_t start_cluster, uint64_t *next_cluster,
665 			    uint8_t *buf, size_t buf_sz)
666 {
667 	struct spdk_blob_md_descriptor_extent *desc;
668 	size_t cur_sz;
669 	uint64_t i, extent_idx;
670 	uint64_t lba, lba_per_cluster, lba_count;
671 
672 	/* The buffer must have room for at least one extent */
673 	cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]);
674 	if (buf_sz < cur_sz) {
675 		*next_cluster = start_cluster;
676 		return;
677 	}
678 
679 	desc = (struct spdk_blob_md_descriptor_extent *)buf;
680 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT;
681 
682 	lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1);
683 
684 	lba = blob->active.clusters[start_cluster];
685 	lba_count = lba_per_cluster;
686 	extent_idx = 0;
687 	for (i = start_cluster + 1; i < blob->active.num_clusters; i++) {
688 		if ((lba + lba_count) == blob->active.clusters[i]) {
689 			lba_count += lba_per_cluster;
690 			continue;
691 		} else if (lba == 0 && blob->active.clusters[i] == 0) {
692 			lba_count += lba_per_cluster;
693 			continue;
694 		}
695 		desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
696 		desc->extents[extent_idx].length = lba_count / lba_per_cluster;
697 		extent_idx++;
698 
699 		cur_sz += sizeof(desc->extents[extent_idx]);
700 
701 		if (buf_sz < cur_sz) {
702 			/* If we ran out of buffer space, return */
703 			desc->length = sizeof(desc->extents[0]) * extent_idx;
704 			*next_cluster = i;
705 			return;
706 		}
707 
708 		lba = blob->active.clusters[i];
709 		lba_count = lba_per_cluster;
710 	}
711 
712 	desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster;
713 	desc->extents[extent_idx].length = lba_count / lba_per_cluster;
714 	extent_idx++;
715 
716 	desc->length = sizeof(desc->extents[0]) * extent_idx;
717 	*next_cluster = blob->active.num_clusters;
718 
719 	return;
720 }
721 
722 static void
723 _spdk_blob_serialize_flags(const struct spdk_blob *blob,
724 			   uint8_t *buf, size_t *buf_sz)
725 {
726 	struct spdk_blob_md_descriptor_flags *desc;
727 
728 	/*
729 	 * Flags get serialized first, so we should always have room for the flags
730 	 *  descriptor.
731 	 */
732 	assert(*buf_sz >= sizeof(*desc));
733 
734 	desc = (struct spdk_blob_md_descriptor_flags *)buf;
735 	desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS;
736 	desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor);
737 	desc->invalid_flags = blob->invalid_flags;
738 	desc->data_ro_flags = blob->data_ro_flags;
739 	desc->md_ro_flags = blob->md_ro_flags;
740 
741 	*buf_sz -= sizeof(*desc);
742 }
743 
744 static int
745 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob,
746 			    const struct spdk_xattr_tailq *xattrs, bool internal,
747 			    struct spdk_blob_md_page **pages,
748 			    struct spdk_blob_md_page *cur_page,
749 			    uint32_t *page_count, uint8_t **buf,
750 			    size_t *remaining_sz)
751 {
752 	const struct spdk_xattr	*xattr;
753 	int	rc;
754 
755 	TAILQ_FOREACH(xattr, xattrs, link) {
756 		size_t required_sz = 0;
757 
758 		rc = _spdk_blob_serialize_xattr(xattr,
759 						*buf, *remaining_sz,
760 						&required_sz, internal);
761 		if (rc < 0) {
762 			/* Need to add a new page to the chain */
763 			rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
764 							   &cur_page);
765 			if (rc < 0) {
766 				spdk_dma_free(*pages);
767 				*pages = NULL;
768 				*page_count = 0;
769 				return rc;
770 			}
771 
772 			*buf = (uint8_t *)cur_page->descriptors;
773 			*remaining_sz = sizeof(cur_page->descriptors);
774 
775 			/* Try again */
776 			required_sz = 0;
777 			rc = _spdk_blob_serialize_xattr(xattr,
778 							*buf, *remaining_sz,
779 							&required_sz, internal);
780 
781 			if (rc < 0) {
782 				spdk_dma_free(*pages);
783 				*pages = NULL;
784 				*page_count = 0;
785 				return rc;
786 			}
787 		}
788 
789 		*remaining_sz -= required_sz;
790 		*buf += required_sz;
791 	}
792 
793 	return 0;
794 }
795 
796 static int
797 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages,
798 		     uint32_t *page_count)
799 {
800 	struct spdk_blob_md_page		*cur_page;
801 	int					rc;
802 	uint8_t					*buf;
803 	size_t					remaining_sz;
804 	uint64_t				last_cluster;
805 
806 	assert(pages != NULL);
807 	assert(page_count != NULL);
808 	assert(blob != NULL);
809 	assert(blob->state == SPDK_BLOB_STATE_DIRTY);
810 
811 	*pages = NULL;
812 	*page_count = 0;
813 
814 	/* A blob always has at least 1 page, even if it has no descriptors */
815 	rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page);
816 	if (rc < 0) {
817 		return rc;
818 	}
819 
820 	buf = (uint8_t *)cur_page->descriptors;
821 	remaining_sz = sizeof(cur_page->descriptors);
822 
823 	/* Serialize flags */
824 	_spdk_blob_serialize_flags(blob, buf, &remaining_sz);
825 	buf += sizeof(struct spdk_blob_md_descriptor_flags);
826 
827 	/* Serialize xattrs */
828 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false,
829 					 pages, cur_page, page_count, &buf, &remaining_sz);
830 	if (rc < 0) {
831 		return rc;
832 	}
833 
834 	/* Serialize internal xattrs */
835 	rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true,
836 					 pages, cur_page, page_count, &buf, &remaining_sz);
837 	if (rc < 0) {
838 		return rc;
839 	}
840 
841 	/* Serialize extents */
842 	last_cluster = 0;
843 	while (last_cluster < blob->active.num_clusters) {
844 		_spdk_blob_serialize_extent(blob, last_cluster, &last_cluster,
845 					    buf, remaining_sz);
846 
847 		if (last_cluster == blob->active.num_clusters) {
848 			break;
849 		}
850 
851 		rc = _spdk_blob_serialize_add_page(blob, pages, page_count,
852 						   &cur_page);
853 		if (rc < 0) {
854 			return rc;
855 		}
856 
857 		buf = (uint8_t *)cur_page->descriptors;
858 		remaining_sz = sizeof(cur_page->descriptors);
859 	}
860 
861 	return 0;
862 }
863 
864 struct spdk_blob_load_ctx {
865 	struct spdk_blob		*blob;
866 
867 	struct spdk_blob_md_page	*pages;
868 	uint32_t			num_pages;
869 	spdk_bs_sequence_t	        *seq;
870 
871 	spdk_bs_sequence_cpl		cb_fn;
872 	void				*cb_arg;
873 };
874 
875 static uint32_t
876 _spdk_blob_md_page_calc_crc(void *page)
877 {
878 	uint32_t		crc;
879 
880 	crc = BLOB_CRC32C_INITIAL;
881 	crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc);
882 	crc ^= BLOB_CRC32C_INITIAL;
883 
884 	return crc;
885 
886 }
887 
888 static void
889 _spdk_blob_load_final(void *cb_arg, int bserrno)
890 {
891 	struct spdk_blob_load_ctx	*ctx = cb_arg;
892 	struct spdk_blob		*blob = ctx->blob;
893 
894 	_spdk_blob_mark_clean(blob);
895 
896 	ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno);
897 
898 	/* Free the memory */
899 	spdk_dma_free(ctx->pages);
900 	free(ctx);
901 }
902 
903 static void
904 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno)
905 {
906 	struct spdk_blob_load_ctx	*ctx = cb_arg;
907 	struct spdk_blob		*blob = ctx->blob;
908 
909 	if (bserrno != 0) {
910 		goto error;
911 	}
912 
913 	blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot);
914 
915 	if (blob->back_bs_dev == NULL) {
916 		bserrno = -ENOMEM;
917 		goto error;
918 	}
919 
920 	_spdk_blob_load_final(ctx, bserrno);
921 	return;
922 
923 error:
924 	SPDK_ERRLOG("Snapshot fail\n");
925 	_spdk_blob_free(blob);
926 	ctx->cb_fn(ctx->seq, NULL, bserrno);
927 	spdk_dma_free(ctx->pages);
928 	free(ctx);
929 }
930 
931 static void
932 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
933 {
934 	struct spdk_blob_load_ctx	*ctx = cb_arg;
935 	struct spdk_blob		*blob = ctx->blob;
936 	struct spdk_blob_md_page	*page;
937 	const void			*value;
938 	size_t				len;
939 	int				rc;
940 	uint32_t			crc;
941 
942 	page = &ctx->pages[ctx->num_pages - 1];
943 	crc = _spdk_blob_md_page_calc_crc(page);
944 	if (crc != page->crc) {
945 		SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages);
946 		_spdk_blob_free(blob);
947 		ctx->cb_fn(seq, NULL, -EINVAL);
948 		spdk_dma_free(ctx->pages);
949 		free(ctx);
950 		return;
951 	}
952 
953 	if (page->next != SPDK_INVALID_MD_PAGE) {
954 		uint32_t next_page = page->next;
955 		uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page);
956 
957 
958 		assert(next_lba < (blob->bs->md_start + blob->bs->md_len));
959 
960 		/* Read the next page */
961 		ctx->num_pages++;
962 		ctx->pages = spdk_dma_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages),
963 					      sizeof(*page), NULL);
964 		if (ctx->pages == NULL) {
965 			ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM);
966 			free(ctx);
967 			return;
968 		}
969 
970 		spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1],
971 					  next_lba,
972 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)),
973 					  _spdk_blob_load_cpl, ctx);
974 		return;
975 	}
976 
977 	/* Parse the pages */
978 	rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob);
979 	if (rc) {
980 		_spdk_blob_free(blob);
981 		ctx->cb_fn(seq, NULL, rc);
982 		spdk_dma_free(ctx->pages);
983 		free(ctx);
984 		return;
985 	}
986 	ctx->seq = seq;
987 
988 
989 	if (spdk_blob_is_thin_provisioned(blob)) {
990 		rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true);
991 		if (rc == 0) {
992 			if (len != sizeof(spdk_blob_id)) {
993 				_spdk_blob_free(blob);
994 				ctx->cb_fn(seq, NULL, -EINVAL);
995 				spdk_dma_free(ctx->pages);
996 				free(ctx);
997 				return;
998 			}
999 			/* open snapshot blob and continue in the callback function */
1000 			blob->parent_id = *(spdk_blob_id *)value;
1001 			spdk_bs_open_blob(blob->bs, blob->parent_id,
1002 					  _spdk_blob_load_snapshot_cpl, ctx);
1003 			return;
1004 		} else {
1005 			/* add zeroes_dev for thin provisioned blob */
1006 			blob->back_bs_dev = spdk_bs_create_zeroes_dev();
1007 		}
1008 	} else {
1009 		/* standard blob */
1010 		blob->back_bs_dev = NULL;
1011 	}
1012 	_spdk_blob_load_final(ctx, bserrno);
1013 }
1014 
1015 /* Load a blob from disk given a blobid */
1016 static void
1017 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1018 		spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1019 {
1020 	struct spdk_blob_load_ctx *ctx;
1021 	struct spdk_blob_store *bs;
1022 	uint32_t page_num;
1023 	uint64_t lba;
1024 
1025 	_spdk_blob_verify_md_op(blob);
1026 
1027 	bs = blob->bs;
1028 
1029 	ctx = calloc(1, sizeof(*ctx));
1030 	if (!ctx) {
1031 		cb_fn(seq, cb_arg, -ENOMEM);
1032 		return;
1033 	}
1034 
1035 	ctx->blob = blob;
1036 	ctx->pages = spdk_dma_realloc(ctx->pages, SPDK_BS_PAGE_SIZE,
1037 				      SPDK_BS_PAGE_SIZE, NULL);
1038 	if (!ctx->pages) {
1039 		free(ctx);
1040 		cb_fn(seq, cb_arg, -ENOMEM);
1041 		return;
1042 	}
1043 	ctx->num_pages = 1;
1044 	ctx->cb_fn = cb_fn;
1045 	ctx->cb_arg = cb_arg;
1046 
1047 	page_num = _spdk_bs_blobid_to_page(blob->id);
1048 	lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num);
1049 
1050 	blob->state = SPDK_BLOB_STATE_LOADING;
1051 
1052 	spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba,
1053 				  _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE),
1054 				  _spdk_blob_load_cpl, ctx);
1055 }
1056 
1057 struct spdk_blob_persist_ctx {
1058 	struct spdk_blob		*blob;
1059 
1060 	struct spdk_bs_super_block	*super;
1061 
1062 	struct spdk_blob_md_page	*pages;
1063 
1064 	uint64_t			idx;
1065 
1066 	spdk_bs_sequence_t		*seq;
1067 	spdk_bs_sequence_cpl		cb_fn;
1068 	void				*cb_arg;
1069 };
1070 
1071 static void
1072 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1073 {
1074 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1075 	struct spdk_blob		*blob = ctx->blob;
1076 
1077 	if (bserrno == 0) {
1078 		_spdk_blob_mark_clean(blob);
1079 	}
1080 
1081 	/* Call user callback */
1082 	ctx->cb_fn(seq, ctx->cb_arg, bserrno);
1083 
1084 	/* Free the memory */
1085 	spdk_dma_free(ctx->pages);
1086 	free(ctx);
1087 }
1088 
1089 static void
1090 _spdk_blob_persist_unmap_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1091 {
1092 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1093 	struct spdk_blob		*blob = ctx->blob;
1094 	struct spdk_blob_store		*bs = blob->bs;
1095 	void				*tmp;
1096 	size_t				i;
1097 
1098 	/* Release all clusters that were truncated */
1099 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1100 		uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]);
1101 
1102 		/* Nothing to release if it was not allocated */
1103 		if (blob->active.clusters[i] != 0) {
1104 			_spdk_bs_release_cluster(bs, cluster_num);
1105 		}
1106 	}
1107 
1108 	if (blob->active.num_clusters == 0) {
1109 		free(blob->active.clusters);
1110 		blob->active.clusters = NULL;
1111 		blob->active.cluster_array_size = 0;
1112 	} else {
1113 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters);
1114 		assert(tmp != NULL);
1115 		blob->active.clusters = tmp;
1116 		blob->active.cluster_array_size = blob->active.num_clusters;
1117 	}
1118 
1119 	_spdk_blob_persist_complete(seq, ctx, bserrno);
1120 }
1121 
1122 static void
1123 _spdk_blob_persist_unmap_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1124 {
1125 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1126 	struct spdk_blob		*blob = ctx->blob;
1127 	struct spdk_blob_store		*bs = blob->bs;
1128 	spdk_bs_batch_t			*batch;
1129 	size_t				i;
1130 	uint64_t			lba;
1131 	uint32_t			lba_count;
1132 
1133 	/* Clusters don't move around in blobs. The list shrinks or grows
1134 	 * at the end, but no changes ever occur in the middle of the list.
1135 	 */
1136 
1137 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_unmap_clusters_cpl, ctx);
1138 
1139 	/* Unmap all clusters that were truncated */
1140 	lba = 0;
1141 	lba_count = 0;
1142 	for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) {
1143 		uint64_t next_lba = blob->active.clusters[i];
1144 		uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1);
1145 
1146 		if (next_lba > 0 && (lba + lba_count) == next_lba) {
1147 			/* This cluster is contiguous with the previous one. */
1148 			lba_count += next_lba_count;
1149 			continue;
1150 		}
1151 
1152 		/* This cluster is not contiguous with the previous one. */
1153 
1154 		/* If a run of LBAs previously existing, send them
1155 		 * as an unmap.
1156 		 */
1157 		if (lba_count > 0) {
1158 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1159 		}
1160 
1161 		/* Start building the next batch */
1162 		lba = next_lba;
1163 		if (next_lba > 0) {
1164 			lba_count = next_lba_count;
1165 		} else {
1166 			lba_count = 0;
1167 		}
1168 	}
1169 
1170 	/* If we ended with a contiguous set of LBAs, send the unmap now */
1171 	if (lba_count > 0) {
1172 		spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1173 	}
1174 
1175 	spdk_bs_batch_close(batch);
1176 }
1177 
1178 static void
1179 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1180 {
1181 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1182 	struct spdk_blob		*blob = ctx->blob;
1183 	struct spdk_blob_store		*bs = blob->bs;
1184 	size_t				i;
1185 
1186 	/* This loop starts at 1 because the first page is special and handled
1187 	 * below. The pages (except the first) are never written in place,
1188 	 * so any pages in the clean list must be zeroed.
1189 	 */
1190 	for (i = 1; i < blob->clean.num_pages; i++) {
1191 		spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]);
1192 	}
1193 
1194 	if (blob->active.num_pages == 0) {
1195 		uint32_t page_num;
1196 
1197 		page_num = _spdk_bs_blobid_to_page(blob->id);
1198 		spdk_bit_array_clear(bs->used_md_pages, page_num);
1199 	}
1200 
1201 	/* Move on to unmapping clusters */
1202 	_spdk_blob_persist_unmap_clusters(seq, ctx, 0);
1203 }
1204 
1205 static void
1206 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1207 {
1208 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1209 	struct spdk_blob		*blob = ctx->blob;
1210 	struct spdk_blob_store		*bs = blob->bs;
1211 	uint64_t			lba;
1212 	uint32_t			lba_count;
1213 	spdk_bs_batch_t			*batch;
1214 	size_t				i;
1215 
1216 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx);
1217 
1218 	lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE);
1219 
1220 	/* This loop starts at 1 because the first page is special and handled
1221 	 * below. The pages (except the first) are never written in place,
1222 	 * so any pages in the clean list must be zeroed.
1223 	 */
1224 	for (i = 1; i < blob->clean.num_pages; i++) {
1225 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]);
1226 
1227 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1228 	}
1229 
1230 	/* The first page will only be zeroed if this is a delete. */
1231 	if (blob->active.num_pages == 0) {
1232 		uint32_t page_num;
1233 
1234 		/* The first page in the metadata goes where the blobid indicates */
1235 		page_num = _spdk_bs_blobid_to_page(blob->id);
1236 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num);
1237 
1238 		spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1239 	}
1240 
1241 	spdk_bs_batch_close(batch);
1242 }
1243 
1244 static void
1245 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1246 {
1247 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1248 	struct spdk_blob		*blob = ctx->blob;
1249 	struct spdk_blob_store		*bs = blob->bs;
1250 	uint64_t			lba;
1251 	uint32_t			lba_count;
1252 	struct spdk_blob_md_page	*page;
1253 
1254 	if (blob->active.num_pages == 0) {
1255 		/* Move on to the next step */
1256 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1257 		return;
1258 	}
1259 
1260 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1261 
1262 	page = &ctx->pages[0];
1263 	/* The first page in the metadata goes where the blobid indicates */
1264 	lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id));
1265 
1266 	spdk_bs_sequence_write_dev(seq, page, lba, lba_count,
1267 				   _spdk_blob_persist_zero_pages, ctx);
1268 }
1269 
1270 static void
1271 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1272 {
1273 	struct spdk_blob_persist_ctx	*ctx = cb_arg;
1274 	struct spdk_blob		*blob = ctx->blob;
1275 	struct spdk_blob_store		*bs = blob->bs;
1276 	uint64_t			lba;
1277 	uint32_t			lba_count;
1278 	struct spdk_blob_md_page	*page;
1279 	spdk_bs_batch_t			*batch;
1280 	size_t				i;
1281 
1282 	/* Clusters don't move around in blobs. The list shrinks or grows
1283 	 * at the end, but no changes ever occur in the middle of the list.
1284 	 */
1285 
1286 	lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page));
1287 
1288 	batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx);
1289 
1290 	/* This starts at 1. The root page is not written until
1291 	 * all of the others are finished
1292 	 */
1293 	for (i = 1; i < blob->active.num_pages; i++) {
1294 		page = &ctx->pages[i];
1295 		assert(page->sequence_num == i);
1296 
1297 		lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]);
1298 
1299 		spdk_bs_batch_write_dev(batch, page, lba, lba_count);
1300 	}
1301 
1302 	spdk_bs_batch_close(batch);
1303 }
1304 
1305 static int
1306 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz)
1307 {
1308 	uint64_t	i;
1309 	uint64_t	*tmp;
1310 	uint64_t	lfc; /* lowest free cluster */
1311 	uint64_t	num_clusters;
1312 	struct spdk_blob_store *bs;
1313 
1314 	bs = blob->bs;
1315 
1316 	_spdk_blob_verify_md_op(blob);
1317 
1318 	if (blob->active.num_clusters == sz) {
1319 		return 0;
1320 	}
1321 
1322 	if (blob->active.num_clusters < blob->active.cluster_array_size) {
1323 		/* If this blob was resized to be larger, then smaller, then
1324 		 * larger without syncing, then the cluster array already
1325 		 * contains spare assigned clusters we can use.
1326 		 */
1327 		num_clusters = spdk_min(blob->active.cluster_array_size,
1328 					sz);
1329 	} else {
1330 		num_clusters = blob->active.num_clusters;
1331 	}
1332 
1333 	/* Do two passes - one to verify that we can obtain enough clusters
1334 	 * and another to actually claim them.
1335 	 */
1336 
1337 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1338 		lfc = 0;
1339 		for (i = num_clusters; i < sz; i++) {
1340 			lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc);
1341 			if (lfc >= bs->total_clusters) {
1342 				/* No more free clusters. Cannot satisfy the request */
1343 				return -ENOSPC;
1344 			}
1345 			lfc++;
1346 		}
1347 	}
1348 
1349 	if (sz > num_clusters) {
1350 		/* Expand the cluster array if necessary.
1351 		 * We only shrink the array when persisting.
1352 		 */
1353 		tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz);
1354 		if (sz > 0 && tmp == NULL) {
1355 			return -ENOMEM;
1356 		}
1357 		memset(tmp + blob->active.cluster_array_size, 0,
1358 		       sizeof(uint64_t) * (sz - blob->active.cluster_array_size));
1359 		blob->active.clusters = tmp;
1360 		blob->active.cluster_array_size = sz;
1361 	}
1362 
1363 	blob->state = SPDK_BLOB_STATE_DIRTY;
1364 
1365 	if (spdk_blob_is_thin_provisioned(blob) == false) {
1366 		lfc = 0;
1367 		for (i = num_clusters; i < sz; i++) {
1368 			_spdk_bs_allocate_cluster(blob, i, &lfc, true);
1369 			lfc++;
1370 		}
1371 	}
1372 
1373 	blob->active.num_clusters = sz;
1374 
1375 	return 0;
1376 }
1377 
1378 static void
1379 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx)
1380 {
1381 	spdk_bs_sequence_t *seq = ctx->seq;
1382 	struct spdk_blob *blob = ctx->blob;
1383 	struct spdk_blob_store *bs = blob->bs;
1384 	uint64_t i;
1385 	uint32_t page_num;
1386 	void *tmp;
1387 	int rc;
1388 
1389 	if (blob->active.num_pages == 0) {
1390 		/* This is the signal that the blob should be deleted.
1391 		 * Immediately jump to the clean up routine. */
1392 		assert(blob->clean.num_pages > 0);
1393 		ctx->idx = blob->clean.num_pages - 1;
1394 		blob->state = SPDK_BLOB_STATE_CLEAN;
1395 		_spdk_blob_persist_zero_pages(seq, ctx, 0);
1396 		return;
1397 
1398 	}
1399 
1400 	/* Generate the new metadata */
1401 	rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages);
1402 	if (rc < 0) {
1403 		_spdk_blob_persist_complete(seq, ctx, rc);
1404 		return;
1405 	}
1406 
1407 	assert(blob->active.num_pages >= 1);
1408 
1409 	/* Resize the cache of page indices */
1410 	tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages));
1411 	if (!tmp) {
1412 		_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1413 		return;
1414 	}
1415 	blob->active.pages = tmp;
1416 
1417 	/* Assign this metadata to pages. This requires two passes -
1418 	 * one to verify that there are enough pages and a second
1419 	 * to actually claim them. */
1420 	page_num = 0;
1421 	/* Note that this loop starts at one. The first page location is fixed by the blobid. */
1422 	for (i = 1; i < blob->active.num_pages; i++) {
1423 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1424 		if (page_num >= spdk_bit_array_capacity(bs->used_md_pages)) {
1425 			_spdk_blob_persist_complete(seq, ctx, -ENOMEM);
1426 			return;
1427 		}
1428 		page_num++;
1429 	}
1430 
1431 	page_num = 0;
1432 	blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id);
1433 	for (i = 1; i < blob->active.num_pages; i++) {
1434 		page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num);
1435 		ctx->pages[i - 1].next = page_num;
1436 		/* Now that previous metadata page is complete, calculate the crc for it. */
1437 		ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1438 		blob->active.pages[i] = page_num;
1439 		spdk_bit_array_set(bs->used_md_pages, page_num);
1440 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id);
1441 		page_num++;
1442 	}
1443 	ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]);
1444 	/* Start writing the metadata from last page to first */
1445 	ctx->idx = blob->active.num_pages - 1;
1446 	blob->state = SPDK_BLOB_STATE_CLEAN;
1447 	_spdk_blob_persist_write_page_chain(seq, ctx, 0);
1448 }
1449 
1450 static void
1451 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1452 {
1453 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1454 
1455 	ctx->blob->bs->clean = 0;
1456 
1457 	spdk_dma_free(ctx->super);
1458 
1459 	_spdk_blob_persist_start(ctx);
1460 }
1461 
1462 static void
1463 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
1464 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg);
1465 
1466 
1467 static void
1468 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1469 {
1470 	struct spdk_blob_persist_ctx *ctx = cb_arg;
1471 
1472 	ctx->super->clean = 0;
1473 	if (ctx->super->size == 0) {
1474 		ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen;
1475 	}
1476 
1477 	_spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx);
1478 }
1479 
1480 
1481 /* Write a blob to disk */
1482 static void
1483 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob,
1484 		   spdk_bs_sequence_cpl cb_fn, void *cb_arg)
1485 {
1486 	struct spdk_blob_persist_ctx *ctx;
1487 
1488 	_spdk_blob_verify_md_op(blob);
1489 
1490 	if (blob->state == SPDK_BLOB_STATE_CLEAN) {
1491 		cb_fn(seq, cb_arg, 0);
1492 		return;
1493 	}
1494 
1495 	ctx = calloc(1, sizeof(*ctx));
1496 	if (!ctx) {
1497 		cb_fn(seq, cb_arg, -ENOMEM);
1498 		return;
1499 	}
1500 	ctx->blob = blob;
1501 	ctx->seq = seq;
1502 	ctx->cb_fn = cb_fn;
1503 	ctx->cb_arg = cb_arg;
1504 
1505 	if (blob->bs->clean) {
1506 		ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
1507 		if (!ctx->super) {
1508 			cb_fn(seq, cb_arg, -ENOMEM);
1509 			free(ctx);
1510 			return;
1511 		}
1512 
1513 		spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0),
1514 					  _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)),
1515 					  _spdk_blob_persist_dirty, ctx);
1516 	} else {
1517 		_spdk_blob_persist_start(ctx);
1518 	}
1519 }
1520 
1521 struct spdk_blob_copy_cluster_ctx {
1522 	struct spdk_blob *blob;
1523 	uint8_t *buf;
1524 	uint64_t page;
1525 	uint64_t new_cluster;
1526 	spdk_bs_sequence_t *seq;
1527 };
1528 
1529 static void
1530 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno)
1531 {
1532 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1533 	struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq;
1534 	TAILQ_HEAD(, spdk_bs_request_set) requests;
1535 	spdk_bs_user_op_t *op;
1536 
1537 	TAILQ_INIT(&requests);
1538 	TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link);
1539 
1540 	while (!TAILQ_EMPTY(&requests)) {
1541 		op = TAILQ_FIRST(&requests);
1542 		TAILQ_REMOVE(&requests, op, link);
1543 		if (bserrno == 0) {
1544 			spdk_bs_user_op_execute(op);
1545 		} else {
1546 			spdk_bs_user_op_abort(op);
1547 		}
1548 	}
1549 
1550 	spdk_dma_free(ctx->buf);
1551 	free(ctx);
1552 }
1553 
1554 static void
1555 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno)
1556 {
1557 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1558 
1559 	if (bserrno) {
1560 		uint32_t cluster_number;
1561 
1562 		if (bserrno == -EEXIST) {
1563 			/* The metadata insert failed because another thread
1564 			 * allocated the cluster first. Free our cluster
1565 			 * but continue without error. */
1566 			bserrno = 0;
1567 		}
1568 
1569 		cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1570 		_spdk_bs_release_cluster(ctx->blob->bs, cluster_number);
1571 	}
1572 
1573 	spdk_bs_sequence_finish(ctx->seq, bserrno);
1574 }
1575 
1576 static void
1577 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1578 {
1579 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1580 	uint32_t cluster_number;
1581 
1582 	if (bserrno) {
1583 		/* The write failed, so jump to the final completion handler */
1584 		spdk_bs_sequence_finish(seq, bserrno);
1585 		return;
1586 	}
1587 
1588 	cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page);
1589 
1590 	_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1591 					       _spdk_blob_insert_cluster_cpl, ctx);
1592 }
1593 
1594 static void
1595 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1596 {
1597 	struct spdk_blob_copy_cluster_ctx *ctx = cb_arg;
1598 
1599 	if (bserrno != 0) {
1600 		/* The read failed, so jump to the final completion handler */
1601 		spdk_bs_sequence_finish(seq, bserrno);
1602 		return;
1603 	}
1604 
1605 	/* Write whole cluster */
1606 	spdk_bs_sequence_write_dev(seq, ctx->buf,
1607 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster),
1608 				   _spdk_bs_cluster_to_lba(ctx->blob->bs, 1),
1609 				   _spdk_blob_write_copy_cpl, ctx);
1610 }
1611 
1612 static void
1613 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob,
1614 				   struct spdk_io_channel *_ch,
1615 				   uint64_t offset, spdk_bs_user_op_t *op)
1616 {
1617 	struct spdk_bs_cpl cpl;
1618 	struct spdk_bs_channel *ch;
1619 	struct spdk_blob_copy_cluster_ctx *ctx;
1620 	uint32_t cluster_start_page;
1621 	uint32_t cluster_number;
1622 	int rc;
1623 
1624 	ch = spdk_io_channel_get_ctx(_ch);
1625 
1626 	if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) {
1627 		/* There are already operations pending. Queue this user op
1628 		 * and return because it will be re-executed when the outstanding
1629 		 * cluster allocation completes. */
1630 		TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1631 		return;
1632 	}
1633 
1634 	/* Round the page offset down to the first page in the cluster */
1635 	cluster_start_page = _spdk_bs_page_to_cluster_start(blob, offset);
1636 
1637 	/* Calculate which index in the metadata cluster array the corresponding
1638 	 * cluster is supposed to be at. */
1639 	cluster_number = _spdk_bs_page_to_cluster(blob->bs, cluster_start_page);
1640 
1641 	ctx = calloc(1, sizeof(*ctx));
1642 	if (!ctx) {
1643 		spdk_bs_user_op_abort(op);
1644 		return;
1645 	}
1646 
1647 	assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0);
1648 
1649 	ctx->blob = blob;
1650 	ctx->page = cluster_start_page;
1651 
1652 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1653 		ctx->buf = spdk_dma_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, NULL);
1654 		if (!ctx->buf) {
1655 			SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n",
1656 				    blob->bs->cluster_sz);
1657 			free(ctx);
1658 			spdk_bs_user_op_abort(op);
1659 			return;
1660 		}
1661 	}
1662 
1663 	rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false);
1664 	if (rc != 0) {
1665 		spdk_dma_free(ctx->buf);
1666 		free(ctx);
1667 		spdk_bs_user_op_abort(op);
1668 		return;
1669 	}
1670 
1671 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1672 	cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl;
1673 	cpl.u.blob_basic.cb_arg = ctx;
1674 
1675 	ctx->seq = spdk_bs_sequence_start(_ch, &cpl);
1676 	if (!ctx->seq) {
1677 		_spdk_bs_release_cluster(blob->bs, ctx->new_cluster);
1678 		spdk_dma_free(ctx->buf);
1679 		free(ctx);
1680 		spdk_bs_user_op_abort(op);
1681 		return;
1682 	}
1683 
1684 	/* Queue the user op to block other incoming operations */
1685 	TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link);
1686 
1687 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
1688 		/* Read cluster from backing device */
1689 		spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf,
1690 					     _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page),
1691 					     _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz),
1692 					     _spdk_blob_write_copy, ctx);
1693 	} else {
1694 		_spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster,
1695 						       _spdk_blob_insert_cluster_cpl, ctx);
1696 	}
1697 }
1698 
1699 static void
1700 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t page, uint64_t length,
1701 				       uint64_t *lba,	uint32_t *lba_count)
1702 {
1703 	*lba_count = _spdk_bs_page_to_lba(blob->bs, length);
1704 
1705 	if (!_spdk_bs_page_is_allocated(blob, page)) {
1706 		assert(blob->back_bs_dev != NULL);
1707 		*lba = _spdk_bs_dev_page_to_lba(blob->back_bs_dev, page);
1708 		*lba_count = _spdk_bs_blob_lba_to_back_dev_lba(blob, *lba_count);
1709 	} else {
1710 		*lba = _spdk_bs_blob_page_to_lba(blob, page);
1711 	}
1712 }
1713 
1714 struct op_split_ctx {
1715 	struct spdk_blob *blob;
1716 	struct spdk_io_channel *channel;
1717 	uint64_t page_offset;
1718 	uint64_t pages_remaining;
1719 	void *curr_payload;
1720 	enum spdk_blob_op_type op_type;
1721 	spdk_bs_sequence_t *seq;
1722 };
1723 
1724 static void
1725 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno)
1726 {
1727 	struct op_split_ctx	*ctx = cb_arg;
1728 	struct spdk_blob	*blob = ctx->blob;
1729 	struct spdk_io_channel	*ch = ctx->channel;
1730 	enum spdk_blob_op_type	op_type = ctx->op_type;
1731 	uint8_t			*buf = ctx->curr_payload;
1732 	uint64_t		offset = ctx->page_offset;
1733 	uint64_t		length = ctx->pages_remaining;
1734 	uint64_t		op_length;
1735 
1736 	if (bserrno != 0 || ctx->pages_remaining == 0) {
1737 		spdk_bs_sequence_finish(ctx->seq, bserrno);
1738 		free(ctx);
1739 		return;
1740 	}
1741 
1742 	op_length = spdk_min(length, _spdk_bs_num_pages_to_cluster_boundary(blob, offset));
1743 
1744 	/* Update length and payload for next operation */
1745 	ctx->pages_remaining -= op_length;
1746 	ctx->page_offset += op_length;
1747 	if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) {
1748 		ctx->curr_payload += (op_length * SPDK_BS_PAGE_SIZE);
1749 	}
1750 
1751 	switch (op_type) {
1752 	case SPDK_BLOB_READ:
1753 		spdk_blob_io_read(blob, ch, buf, offset, op_length,
1754 				  _spdk_blob_request_submit_op_split_next, ctx);
1755 		break;
1756 	case SPDK_BLOB_WRITE:
1757 		spdk_blob_io_write(blob, ch, buf, offset, op_length,
1758 				   _spdk_blob_request_submit_op_split_next, ctx);
1759 		break;
1760 	case SPDK_BLOB_UNMAP:
1761 		spdk_blob_io_unmap(blob, ch, offset, op_length,
1762 				   _spdk_blob_request_submit_op_split_next, ctx);
1763 		break;
1764 	case SPDK_BLOB_WRITE_ZEROES:
1765 		spdk_blob_io_write_zeroes(blob, ch, offset, op_length,
1766 					  _spdk_blob_request_submit_op_split_next, ctx);
1767 		break;
1768 	case SPDK_BLOB_READV:
1769 	case SPDK_BLOB_WRITEV:
1770 		SPDK_ERRLOG("readv/write not valid for %s\n", __func__);
1771 		spdk_bs_sequence_finish(ctx->seq, -EINVAL);
1772 		free(ctx);
1773 		break;
1774 	}
1775 }
1776 
1777 static void
1778 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob,
1779 				   void *payload, uint64_t offset, uint64_t length,
1780 				   spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1781 {
1782 	struct op_split_ctx *ctx;
1783 	spdk_bs_sequence_t *seq;
1784 	struct spdk_bs_cpl cpl;
1785 
1786 	assert(blob != NULL);
1787 
1788 	ctx = calloc(1, sizeof(struct op_split_ctx));
1789 	if (ctx == NULL) {
1790 		cb_fn(cb_arg, -ENOMEM);
1791 		return;
1792 	}
1793 
1794 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1795 	cpl.u.blob_basic.cb_fn = cb_fn;
1796 	cpl.u.blob_basic.cb_arg = cb_arg;
1797 
1798 	seq = spdk_bs_sequence_start(ch, &cpl);
1799 	if (!seq) {
1800 		free(ctx);
1801 		cb_fn(cb_arg, -ENOMEM);
1802 		return;
1803 	}
1804 
1805 	ctx->blob = blob;
1806 	ctx->channel = ch;
1807 	ctx->curr_payload = payload;
1808 	ctx->page_offset = offset;
1809 	ctx->pages_remaining = length;
1810 	ctx->op_type = op_type;
1811 	ctx->seq = seq;
1812 
1813 	_spdk_blob_request_submit_op_split_next(ctx, 0);
1814 }
1815 
1816 static void
1817 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob,
1818 				    void *payload, uint64_t offset, uint64_t length,
1819 				    spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1820 {
1821 	struct spdk_bs_cpl cpl;
1822 	uint64_t lba;
1823 	uint32_t lba_count;
1824 
1825 	assert(blob != NULL);
1826 
1827 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
1828 	cpl.u.blob_basic.cb_fn = cb_fn;
1829 	cpl.u.blob_basic.cb_arg = cb_arg;
1830 
1831 	_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
1832 
1833 	if (blob->frozen_refcnt) {
1834 		/* This blob I/O is frozen */
1835 		spdk_bs_user_op_t *op;
1836 		struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch);
1837 
1838 		op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1839 		if (!op) {
1840 			cb_fn(cb_arg, -ENOMEM);
1841 			return;
1842 		}
1843 
1844 		TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
1845 
1846 		return;
1847 	}
1848 
1849 	switch (op_type) {
1850 	case SPDK_BLOB_READ: {
1851 		spdk_bs_batch_t *batch;
1852 
1853 		batch = spdk_bs_batch_open(_ch, &cpl);
1854 		if (!batch) {
1855 			cb_fn(cb_arg, -ENOMEM);
1856 			return;
1857 		}
1858 
1859 		if (_spdk_bs_page_is_allocated(blob, offset)) {
1860 			/* Read from the blob */
1861 			spdk_bs_batch_read_dev(batch, payload, lba, lba_count);
1862 		} else {
1863 			/* Read from the backing block device */
1864 			spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count);
1865 		}
1866 
1867 		spdk_bs_batch_close(batch);
1868 		break;
1869 	}
1870 	case SPDK_BLOB_WRITE:
1871 	case SPDK_BLOB_WRITE_ZEROES: {
1872 		if (_spdk_bs_page_is_allocated(blob, offset)) {
1873 			/* Write to the blob */
1874 			spdk_bs_batch_t *batch;
1875 
1876 			if (lba_count == 0) {
1877 				cb_fn(cb_arg, 0);
1878 				return;
1879 			}
1880 
1881 			batch = spdk_bs_batch_open(_ch, &cpl);
1882 			if (!batch) {
1883 				cb_fn(cb_arg, -ENOMEM);
1884 				return;
1885 			}
1886 
1887 			if (op_type == SPDK_BLOB_WRITE) {
1888 				spdk_bs_batch_write_dev(batch, payload, lba, lba_count);
1889 			} else {
1890 				spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count);
1891 			}
1892 
1893 			spdk_bs_batch_close(batch);
1894 		} else {
1895 			/* Queue this operation and allocate the cluster */
1896 			spdk_bs_user_op_t *op;
1897 
1898 			op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length);
1899 			if (!op) {
1900 				cb_fn(cb_arg, -ENOMEM);
1901 				return;
1902 			}
1903 
1904 			_spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op);
1905 		}
1906 		break;
1907 	}
1908 	case SPDK_BLOB_UNMAP: {
1909 		spdk_bs_batch_t *batch;
1910 
1911 		batch = spdk_bs_batch_open(_ch, &cpl);
1912 		if (!batch) {
1913 			cb_fn(cb_arg, -ENOMEM);
1914 			return;
1915 		}
1916 
1917 		if (_spdk_bs_page_is_allocated(blob, offset)) {
1918 			spdk_bs_batch_unmap_dev(batch, lba, lba_count);
1919 		}
1920 
1921 		spdk_bs_batch_close(batch);
1922 		break;
1923 	}
1924 	case SPDK_BLOB_READV:
1925 	case SPDK_BLOB_WRITEV:
1926 		SPDK_ERRLOG("readv/write not valid\n");
1927 		cb_fn(cb_arg, -EINVAL);
1928 		break;
1929 	}
1930 }
1931 
1932 static void
1933 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel,
1934 			     void *payload, uint64_t offset, uint64_t length,
1935 			     spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type)
1936 {
1937 	assert(blob != NULL);
1938 
1939 	if (blob->data_ro && op_type != SPDK_BLOB_READ) {
1940 		cb_fn(cb_arg, -EPERM);
1941 		return;
1942 	}
1943 
1944 	if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) {
1945 		cb_fn(cb_arg, -EINVAL);
1946 		return;
1947 	}
1948 
1949 	if (length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset)) {
1950 		_spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length,
1951 						    cb_fn, cb_arg, op_type);
1952 	} else {
1953 		_spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length,
1954 						   cb_fn, cb_arg, op_type);
1955 	}
1956 }
1957 
1958 struct rw_iov_ctx {
1959 	struct spdk_blob *blob;
1960 	struct spdk_io_channel *channel;
1961 	spdk_blob_op_complete cb_fn;
1962 	void *cb_arg;
1963 	bool read;
1964 	int iovcnt;
1965 	struct iovec *orig_iov;
1966 	uint64_t page_offset;
1967 	uint64_t pages_remaining;
1968 	uint64_t pages_done;
1969 	struct iovec iov[0];
1970 };
1971 
1972 static void
1973 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
1974 {
1975 	assert(cb_arg == NULL);
1976 	spdk_bs_sequence_finish(seq, bserrno);
1977 }
1978 
1979 static void
1980 _spdk_rw_iov_split_next(void *cb_arg, int bserrno)
1981 {
1982 	struct rw_iov_ctx *ctx = cb_arg;
1983 	struct spdk_blob *blob = ctx->blob;
1984 	struct iovec *iov, *orig_iov;
1985 	int iovcnt;
1986 	size_t orig_iovoff;
1987 	uint64_t page_count, pages_to_boundary, page_offset;
1988 	uint64_t byte_count;
1989 
1990 	if (bserrno != 0 || ctx->pages_remaining == 0) {
1991 		ctx->cb_fn(ctx->cb_arg, bserrno);
1992 		free(ctx);
1993 		return;
1994 	}
1995 
1996 	page_offset = ctx->page_offset;
1997 	pages_to_boundary = _spdk_bs_num_pages_to_cluster_boundary(blob, page_offset);
1998 	page_count = spdk_min(ctx->pages_remaining, pages_to_boundary);
1999 
2000 	/*
2001 	 * Get index and offset into the original iov array for our current position in the I/O sequence.
2002 	 *  byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will
2003 	 *  point to the current position in the I/O sequence.
2004 	 */
2005 	byte_count = ctx->pages_done * sizeof(struct spdk_blob_md_page);
2006 	orig_iov = &ctx->orig_iov[0];
2007 	orig_iovoff = 0;
2008 	while (byte_count > 0) {
2009 		if (byte_count >= orig_iov->iov_len) {
2010 			byte_count -= orig_iov->iov_len;
2011 			orig_iov++;
2012 		} else {
2013 			orig_iovoff = byte_count;
2014 			byte_count = 0;
2015 		}
2016 	}
2017 
2018 	/*
2019 	 * Build an iov array for the next I/O in the sequence.  byte_count will keep track of how many
2020 	 *  bytes of this next I/O remain to be accounted for in the new iov array.
2021 	 */
2022 	byte_count = page_count * sizeof(struct spdk_blob_md_page);
2023 	iov = &ctx->iov[0];
2024 	iovcnt = 0;
2025 	while (byte_count > 0) {
2026 		assert(iovcnt < ctx->iovcnt);
2027 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2028 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2029 		byte_count -= iov->iov_len;
2030 		orig_iovoff = 0;
2031 		orig_iov++;
2032 		iov++;
2033 		iovcnt++;
2034 	}
2035 
2036 	ctx->page_offset += page_count;
2037 	ctx->pages_done += page_count;
2038 	ctx->pages_remaining -= page_count;
2039 	iov = &ctx->iov[0];
2040 
2041 	if (ctx->read) {
2042 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, page_offset,
2043 				   page_count, _spdk_rw_iov_split_next, ctx);
2044 	} else {
2045 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, page_offset,
2046 				    page_count, _spdk_rw_iov_split_next, ctx);
2047 	}
2048 }
2049 
2050 static void
2051 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2052 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2053 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2054 {
2055 	struct spdk_bs_cpl	cpl;
2056 
2057 	assert(blob != NULL);
2058 
2059 	if (!read && blob->data_ro) {
2060 		cb_fn(cb_arg, -EPERM);
2061 		return;
2062 	}
2063 
2064 	if (length == 0) {
2065 		cb_fn(cb_arg, 0);
2066 		return;
2067 	}
2068 
2069 	if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) {
2070 		cb_fn(cb_arg, -EINVAL);
2071 		return;
2072 	}
2073 
2074 	/*
2075 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2076 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2077 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2078 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2079 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2080 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2081 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2082 	 *
2083 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2084 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2085 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2086 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2087 	 */
2088 	if (spdk_likely(length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset))) {
2089 		uint32_t lba_count;
2090 		uint64_t lba;
2091 
2092 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2093 
2094 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2095 		cpl.u.blob_basic.cb_fn = cb_fn;
2096 		cpl.u.blob_basic.cb_arg = cb_arg;
2097 		if (blob->frozen_refcnt) {
2098 			/* This blob I/O is frozen */
2099 			spdk_bs_user_op_t *op;
2100 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2101 
2102 			op = spdk_bs_user_op_alloc(_channel, &cpl, read, blob, iov, iovcnt, offset, length);
2103 			if (!op) {
2104 				cb_fn(cb_arg, -ENOMEM);
2105 				return;
2106 			}
2107 
2108 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2109 
2110 			return;
2111 		}
2112 
2113 		if (read) {
2114 			spdk_bs_sequence_t *seq;
2115 
2116 			seq = spdk_bs_sequence_start(_channel, &cpl);
2117 			if (!seq) {
2118 				cb_fn(cb_arg, -ENOMEM);
2119 				return;
2120 			}
2121 
2122 			if (_spdk_bs_page_is_allocated(blob, offset)) {
2123 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2124 			} else {
2125 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2126 							      _spdk_rw_iov_done, NULL);
2127 			}
2128 		} else {
2129 			if (_spdk_bs_page_is_allocated(blob, offset)) {
2130 				spdk_bs_sequence_t *seq;
2131 
2132 				seq = spdk_bs_sequence_start(_channel, &cpl);
2133 				if (!seq) {
2134 					cb_fn(cb_arg, -ENOMEM);
2135 					return;
2136 				}
2137 
2138 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2139 			} else {
2140 				/* Queue this operation and allocate the cluster */
2141 				spdk_bs_user_op_t *op;
2142 
2143 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, length);
2144 				if (!op) {
2145 					cb_fn(cb_arg, -ENOMEM);
2146 					return;
2147 				}
2148 
2149 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2150 			}
2151 		}
2152 	} else {
2153 		struct rw_iov_ctx *ctx;
2154 
2155 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2156 		if (ctx == NULL) {
2157 			cb_fn(cb_arg, -ENOMEM);
2158 			return;
2159 		}
2160 
2161 		ctx->blob = blob;
2162 		ctx->channel = _channel;
2163 		ctx->cb_fn = cb_fn;
2164 		ctx->cb_arg = cb_arg;
2165 		ctx->read = read;
2166 		ctx->orig_iov = iov;
2167 		ctx->iovcnt = iovcnt;
2168 		ctx->page_offset = offset;
2169 		ctx->pages_remaining = length;
2170 		ctx->pages_done = 0;
2171 
2172 		_spdk_rw_iov_split_next(ctx, 0);
2173 	}
2174 }
2175 
2176 static struct spdk_blob *
2177 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2178 {
2179 	struct spdk_blob *blob;
2180 
2181 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2182 		if (blob->id == blobid) {
2183 			return blob;
2184 		}
2185 	}
2186 
2187 	return NULL;
2188 }
2189 
2190 static int
2191 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2192 {
2193 	struct spdk_blob_store		*bs = io_device;
2194 	struct spdk_bs_channel		*channel = ctx_buf;
2195 	struct spdk_bs_dev		*dev;
2196 	uint32_t			max_ops = bs->max_channel_ops;
2197 	uint32_t			i;
2198 
2199 	dev = bs->dev;
2200 
2201 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2202 	if (!channel->req_mem) {
2203 		return -1;
2204 	}
2205 
2206 	TAILQ_INIT(&channel->reqs);
2207 
2208 	for (i = 0; i < max_ops; i++) {
2209 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2210 	}
2211 
2212 	channel->bs = bs;
2213 	channel->dev = dev;
2214 	channel->dev_channel = dev->create_channel(dev);
2215 
2216 	if (!channel->dev_channel) {
2217 		SPDK_ERRLOG("Failed to create device channel.\n");
2218 		free(channel->req_mem);
2219 		return -1;
2220 	}
2221 
2222 	TAILQ_INIT(&channel->need_cluster_alloc);
2223 	TAILQ_INIT(&channel->queued_io);
2224 
2225 	return 0;
2226 }
2227 
2228 static void
2229 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2230 {
2231 	struct spdk_bs_channel *channel = ctx_buf;
2232 	spdk_bs_user_op_t *op;
2233 
2234 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2235 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2236 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2237 		spdk_bs_user_op_abort(op);
2238 	}
2239 
2240 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2241 		op = TAILQ_FIRST(&channel->queued_io);
2242 		TAILQ_REMOVE(&channel->queued_io, op, link);
2243 		spdk_bs_user_op_abort(op);
2244 	}
2245 
2246 	free(channel->req_mem);
2247 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2248 }
2249 
2250 static void
2251 _spdk_bs_dev_destroy(void *io_device)
2252 {
2253 	struct spdk_blob_store *bs = io_device;
2254 	struct spdk_blob	*blob, *blob_tmp;
2255 
2256 	bs->dev->destroy(bs->dev);
2257 
2258 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2259 		TAILQ_REMOVE(&bs->blobs, blob, link);
2260 		_spdk_blob_free(blob);
2261 	}
2262 
2263 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2264 
2265 	spdk_bit_array_free(&bs->used_blobids);
2266 	spdk_bit_array_free(&bs->used_md_pages);
2267 	spdk_bit_array_free(&bs->used_clusters);
2268 	/*
2269 	 * If this function is called for any reason except a successful unload,
2270 	 * the unload_cpl type will be NONE and this will be a nop.
2271 	 */
2272 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2273 
2274 	free(bs);
2275 }
2276 
2277 static int
2278 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2279 {
2280 	spdk_blob_id snapshot_id;
2281 	struct spdk_blob_list *snapshot_entry = NULL;
2282 	struct spdk_blob_list *clone_entry = NULL;
2283 
2284 	assert(blob != NULL);
2285 
2286 	snapshot_id = blob->parent_id;
2287 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2288 		return 0;
2289 	}
2290 
2291 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2292 		if (snapshot_entry->id == snapshot_id) {
2293 			break;
2294 		}
2295 	}
2296 
2297 	if (snapshot_entry == NULL) {
2298 		/* Snapshot not found */
2299 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2300 		if (snapshot_entry == NULL) {
2301 			return -ENOMEM;
2302 		}
2303 		snapshot_entry->id = snapshot_id;
2304 		TAILQ_INIT(&snapshot_entry->clones);
2305 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2306 	} else {
2307 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2308 			if (clone_entry->id == blob->id) {
2309 				break;
2310 			}
2311 		}
2312 	}
2313 
2314 	if (clone_entry == NULL) {
2315 		/* Clone not found */
2316 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2317 		if (clone_entry == NULL) {
2318 			return -ENOMEM;
2319 		}
2320 		clone_entry->id = blob->id;
2321 		TAILQ_INIT(&clone_entry->clones);
2322 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2323 		snapshot_entry->clone_count++;
2324 	}
2325 
2326 	return 0;
2327 }
2328 
2329 static int
2330 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2331 {
2332 	struct spdk_blob_list *snapshot_entry = NULL;
2333 	struct spdk_blob_list *clone_entry = NULL;
2334 	spdk_blob_id snapshot_id;
2335 
2336 	assert(blob != NULL);
2337 
2338 	snapshot_id = blob->parent_id;
2339 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2340 		return 0;
2341 	}
2342 
2343 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2344 		if (snapshot_entry->id == snapshot_id) {
2345 			break;
2346 		}
2347 	}
2348 
2349 	assert(snapshot_entry != NULL);
2350 
2351 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2352 		if (clone_entry->id == blob->id) {
2353 			break;
2354 		}
2355 	}
2356 
2357 	assert(clone_entry != NULL);
2358 
2359 	blob->parent_id = SPDK_BLOBID_INVALID;
2360 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2361 	free(clone_entry);
2362 
2363 	snapshot_entry->clone_count--;
2364 	if (snapshot_entry->clone_count == 0) {
2365 		/* Snapshot have no more clones */
2366 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
2367 		free(snapshot_entry);
2368 	}
2369 
2370 	return 0;
2371 }
2372 
2373 static int
2374 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2375 {
2376 	struct spdk_blob_list *snapshot_entry;
2377 	struct spdk_blob_list *snapshot_entry_tmp;
2378 	struct spdk_blob_list *clone_entry;
2379 	struct spdk_blob_list *clone_entry_tmp;
2380 
2381 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2382 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2383 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2384 			free(clone_entry);
2385 		}
2386 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2387 		free(snapshot_entry);
2388 	}
2389 
2390 	return 0;
2391 }
2392 
2393 static void
2394 _spdk_bs_free(struct spdk_blob_store *bs)
2395 {
2396 	_spdk_bs_blob_list_free(bs);
2397 
2398 	spdk_bs_unregister_md_thread(bs);
2399 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2400 }
2401 
2402 void
2403 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2404 {
2405 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2406 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2407 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2408 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2409 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2410 	opts->iter_cb_fn = NULL;
2411 	opts->iter_cb_arg = NULL;
2412 }
2413 
2414 static int
2415 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2416 {
2417 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2418 	    opts->max_channel_ops == 0) {
2419 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2420 		return -1;
2421 	}
2422 
2423 	return 0;
2424 }
2425 
2426 static int
2427 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2428 {
2429 	struct spdk_blob_store	*bs;
2430 	uint64_t dev_size;
2431 	int rc;
2432 
2433 	dev_size = dev->blocklen * dev->blockcnt;
2434 	if (dev_size < opts->cluster_sz) {
2435 		/* Device size cannot be smaller than cluster size of blobstore */
2436 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2437 			     dev_size, opts->cluster_sz);
2438 		return -ENOSPC;
2439 	}
2440 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2441 		/* Cluster size cannot be smaller than page size */
2442 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2443 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2444 		return -EINVAL;
2445 	}
2446 	bs = calloc(1, sizeof(struct spdk_blob_store));
2447 	if (!bs) {
2448 		return -ENOMEM;
2449 	}
2450 
2451 	TAILQ_INIT(&bs->blobs);
2452 	TAILQ_INIT(&bs->snapshots);
2453 	bs->dev = dev;
2454 	bs->md_thread = spdk_get_thread();
2455 	assert(bs->md_thread != NULL);
2456 
2457 	/*
2458 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2459 	 *  even multiple of the cluster size.
2460 	 */
2461 	bs->cluster_sz = opts->cluster_sz;
2462 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2463 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2464 	bs->num_free_clusters = bs->total_clusters;
2465 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2466 	if (bs->used_clusters == NULL) {
2467 		free(bs);
2468 		return -ENOMEM;
2469 	}
2470 
2471 	bs->max_channel_ops = opts->max_channel_ops;
2472 	bs->super_blob = SPDK_BLOBID_INVALID;
2473 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2474 
2475 	/* The metadata is assumed to be at least 1 page */
2476 	bs->used_md_pages = spdk_bit_array_create(1);
2477 	bs->used_blobids = spdk_bit_array_create(0);
2478 
2479 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2480 
2481 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2482 				sizeof(struct spdk_bs_channel), "blobstore");
2483 	rc = spdk_bs_register_md_thread(bs);
2484 	if (rc == -1) {
2485 		spdk_io_device_unregister(bs, NULL);
2486 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2487 		spdk_bit_array_free(&bs->used_blobids);
2488 		spdk_bit_array_free(&bs->used_md_pages);
2489 		spdk_bit_array_free(&bs->used_clusters);
2490 		free(bs);
2491 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2492 		return -ENOMEM;
2493 	}
2494 
2495 	*_bs = bs;
2496 	return 0;
2497 }
2498 
2499 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2500 
2501 struct spdk_bs_load_ctx {
2502 	struct spdk_blob_store		*bs;
2503 	struct spdk_bs_super_block	*super;
2504 
2505 	struct spdk_bs_md_mask		*mask;
2506 	bool				in_page_chain;
2507 	uint32_t			page_index;
2508 	uint32_t			cur_page;
2509 	struct spdk_blob_md_page	*page;
2510 	bool				is_load;
2511 
2512 	spdk_bs_sequence_t			*seq;
2513 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2514 	void					*iter_cb_arg;
2515 };
2516 
2517 static void
2518 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2519 {
2520 	assert(bserrno != 0);
2521 
2522 	spdk_dma_free(ctx->super);
2523 	spdk_bs_sequence_finish(seq, bserrno);
2524 	/*
2525 	 * Only free the blobstore when a load fails.  If an unload fails (for some reason)
2526 	 *  we want to keep the blobstore in case the caller wants to try again.
2527 	 */
2528 	if (ctx->is_load) {
2529 		_spdk_bs_free(ctx->bs);
2530 	}
2531 	free(ctx);
2532 }
2533 
2534 static void
2535 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2536 {
2537 	uint32_t i = 0;
2538 
2539 	while (true) {
2540 		i = spdk_bit_array_find_first_set(array, i);
2541 		if (i >= mask->length) {
2542 			break;
2543 		}
2544 		mask->mask[i / 8] |= 1U << (i % 8);
2545 		i++;
2546 	}
2547 }
2548 
2549 static int
2550 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2551 {
2552 	struct spdk_bit_array *array;
2553 	uint32_t i;
2554 
2555 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2556 		return -ENOMEM;
2557 	}
2558 
2559 	array = *array_ptr;
2560 	for (i = 0; i < mask->length; i++) {
2561 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2562 			spdk_bit_array_set(array, i);
2563 		}
2564 	}
2565 
2566 	return 0;
2567 }
2568 
2569 static void
2570 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2571 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2572 {
2573 	/* Update the values in the super block */
2574 	super->super_blob = bs->super_blob;
2575 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2576 	super->crc = _spdk_blob_md_page_calc_crc(super);
2577 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2578 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2579 				   cb_fn, cb_arg);
2580 }
2581 
2582 static void
2583 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2584 {
2585 	struct spdk_bs_load_ctx	*ctx = arg;
2586 	uint64_t	mask_size, lba, lba_count;
2587 
2588 	/* Write out the used clusters mask */
2589 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2590 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2591 	if (!ctx->mask) {
2592 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2593 		return;
2594 	}
2595 
2596 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2597 	ctx->mask->length = ctx->bs->total_clusters;
2598 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2599 
2600 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2601 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2602 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2603 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2604 }
2605 
2606 static void
2607 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2608 {
2609 	struct spdk_bs_load_ctx	*ctx = arg;
2610 	uint64_t	mask_size, lba, lba_count;
2611 
2612 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2613 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2614 	if (!ctx->mask) {
2615 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2616 		return;
2617 	}
2618 
2619 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2620 	ctx->mask->length = ctx->super->md_len;
2621 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2622 
2623 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2624 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2625 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2626 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2627 }
2628 
2629 static void
2630 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2631 {
2632 	struct spdk_bs_load_ctx	*ctx = arg;
2633 	uint64_t	mask_size, lba, lba_count;
2634 
2635 	if (ctx->super->used_blobid_mask_len == 0) {
2636 		/*
2637 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2638 		 *  written to disk.
2639 		 */
2640 		cb_fn(seq, arg, 0);
2641 		return;
2642 	}
2643 
2644 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2645 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2646 	if (!ctx->mask) {
2647 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2648 		return;
2649 	}
2650 
2651 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2652 	ctx->mask->length = ctx->super->md_len;
2653 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2654 
2655 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2656 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2657 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2658 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2659 }
2660 
2661 static void
2662 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2663 {
2664 	struct spdk_bs_load_ctx *ctx = arg;
2665 
2666 	if (bserrno == 0) {
2667 		if (ctx->iter_cb_fn) {
2668 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2669 		}
2670 		_spdk_bs_blob_list_add(blob);
2671 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2672 		return;
2673 	}
2674 
2675 	if (bserrno == -ENOENT) {
2676 		bserrno = 0;
2677 	} else {
2678 		/*
2679 		 * This case needs to be looked at further.  Same problem
2680 		 *  exists with applications that rely on explicit blob
2681 		 *  iteration.  We should just skip the blob that failed
2682 		 *  to load and continue on to the next one.
2683 		 */
2684 		SPDK_ERRLOG("Error in iterating blobs\n");
2685 	}
2686 
2687 	ctx->iter_cb_fn = NULL;
2688 
2689 	spdk_dma_free(ctx->super);
2690 	spdk_dma_free(ctx->mask);
2691 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2692 	free(ctx);
2693 }
2694 
2695 static void
2696 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2697 {
2698 	ctx->seq = seq;
2699 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2700 }
2701 
2702 static void
2703 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2704 {
2705 	struct spdk_bs_load_ctx *ctx = cb_arg;
2706 	int rc;
2707 
2708 	/* The type must be correct */
2709 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2710 
2711 	/* The length of the mask (in bits) must not be greater than
2712 	 * the length of the buffer (converted to bits) */
2713 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2714 
2715 	/* The length of the mask must be exactly equal to the size
2716 	 * (in pages) of the metadata region */
2717 	assert(ctx->mask->length == ctx->super->md_len);
2718 
2719 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2720 	if (rc < 0) {
2721 		spdk_dma_free(ctx->mask);
2722 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2723 		return;
2724 	}
2725 
2726 	_spdk_bs_load_complete(seq, ctx, bserrno);
2727 }
2728 
2729 static void
2730 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2731 {
2732 	struct spdk_bs_load_ctx *ctx = cb_arg;
2733 	uint64_t		lba, lba_count, mask_size;
2734 	int			rc;
2735 
2736 	/* The type must be correct */
2737 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2738 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2739 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2740 					     struct spdk_blob_md_page) * 8));
2741 	/* The length of the mask must be exactly equal to the total number of clusters */
2742 	assert(ctx->mask->length == ctx->bs->total_clusters);
2743 
2744 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2745 	if (rc < 0) {
2746 		spdk_dma_free(ctx->mask);
2747 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2748 		return;
2749 	}
2750 
2751 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2752 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2753 
2754 	spdk_dma_free(ctx->mask);
2755 
2756 	/* Read the used blobids mask */
2757 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2758 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2759 	if (!ctx->mask) {
2760 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2761 		return;
2762 	}
2763 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2764 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2765 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2766 				  _spdk_bs_load_used_blobids_cpl, ctx);
2767 }
2768 
2769 static void
2770 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2771 {
2772 	struct spdk_bs_load_ctx *ctx = cb_arg;
2773 	uint64_t		lba, lba_count, mask_size;
2774 	int			rc;
2775 
2776 	/* The type must be correct */
2777 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2778 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2779 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2780 				     8));
2781 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2782 	assert(ctx->mask->length == ctx->super->md_len);
2783 
2784 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2785 	if (rc < 0) {
2786 		spdk_dma_free(ctx->mask);
2787 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2788 		return;
2789 	}
2790 
2791 	spdk_dma_free(ctx->mask);
2792 
2793 	/* Read the used clusters mask */
2794 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2795 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2796 	if (!ctx->mask) {
2797 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2798 		return;
2799 	}
2800 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2801 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2802 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2803 				  _spdk_bs_load_used_clusters_cpl, ctx);
2804 }
2805 
2806 static void
2807 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2808 {
2809 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2810 	uint64_t lba, lba_count, mask_size;
2811 
2812 	/* Read the used pages mask */
2813 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2814 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2815 	if (!ctx->mask) {
2816 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2817 		return;
2818 	}
2819 
2820 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2821 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2822 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2823 				  _spdk_bs_load_used_pages_cpl, ctx);
2824 }
2825 
2826 static int
2827 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2828 {
2829 	struct spdk_blob_md_descriptor *desc;
2830 	size_t	cur_desc = 0;
2831 
2832 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2833 	while (cur_desc < sizeof(page->descriptors)) {
2834 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2835 			if (desc->length == 0) {
2836 				/* If padding and length are 0, this terminates the page */
2837 				break;
2838 			}
2839 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2840 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2841 			unsigned int				i, j;
2842 			unsigned int				cluster_count = 0;
2843 			uint32_t				cluster_idx;
2844 
2845 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2846 
2847 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2848 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2849 					cluster_idx = desc_extent->extents[i].cluster_idx;
2850 					/*
2851 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2852 					 * in the used cluster map.
2853 					 */
2854 					if (cluster_idx != 0) {
2855 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2856 						if (bs->num_free_clusters == 0) {
2857 							return -ENOSPC;
2858 						}
2859 						bs->num_free_clusters--;
2860 					}
2861 					cluster_count++;
2862 				}
2863 			}
2864 			if (cluster_count == 0) {
2865 				return -EINVAL;
2866 			}
2867 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2868 			/* Skip this item */
2869 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2870 			/* Skip this item */
2871 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2872 			/* Skip this item */
2873 		} else {
2874 			/* Error */
2875 			return -EINVAL;
2876 		}
2877 		/* Advance to the next descriptor */
2878 		cur_desc += sizeof(*desc) + desc->length;
2879 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2880 			break;
2881 		}
2882 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2883 	}
2884 	return 0;
2885 }
2886 
2887 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2888 {
2889 	uint32_t crc;
2890 
2891 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2892 	if (crc != ctx->page->crc) {
2893 		return false;
2894 	}
2895 
2896 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2897 		return false;
2898 	}
2899 	return true;
2900 }
2901 
2902 static void
2903 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2904 
2905 static void
2906 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2907 {
2908 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2909 
2910 	_spdk_bs_load_complete(seq, ctx, bserrno);
2911 }
2912 
2913 static void
2914 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2915 {
2916 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2917 
2918 	spdk_dma_free(ctx->mask);
2919 	ctx->mask = NULL;
2920 
2921 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2922 }
2923 
2924 static void
2925 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2926 {
2927 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2928 
2929 	spdk_dma_free(ctx->mask);
2930 	ctx->mask = NULL;
2931 
2932 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2933 }
2934 
2935 static void
2936 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2937 {
2938 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2939 }
2940 
2941 static void
2942 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2943 {
2944 	struct spdk_bs_load_ctx *ctx = cb_arg;
2945 	uint64_t num_md_clusters;
2946 	uint64_t i;
2947 	uint32_t page_num;
2948 
2949 	if (bserrno != 0) {
2950 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2951 		return;
2952 	}
2953 
2954 	page_num = ctx->cur_page;
2955 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2956 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2957 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2958 			if (ctx->page->sequence_num == 0) {
2959 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2960 			}
2961 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2962 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
2963 				return;
2964 			}
2965 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
2966 				ctx->in_page_chain = true;
2967 				ctx->cur_page = ctx->page->next;
2968 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2969 				return;
2970 			}
2971 		}
2972 	}
2973 
2974 	ctx->in_page_chain = false;
2975 
2976 	do {
2977 		ctx->page_index++;
2978 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
2979 
2980 	if (ctx->page_index < ctx->super->md_len) {
2981 		ctx->cur_page = ctx->page_index;
2982 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2983 	} else {
2984 		/* Claim all of the clusters used by the metadata */
2985 		num_md_clusters = divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
2986 		for (i = 0; i < num_md_clusters; i++) {
2987 			_spdk_bs_claim_cluster(ctx->bs, i);
2988 		}
2989 		spdk_dma_free(ctx->page);
2990 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
2991 	}
2992 }
2993 
2994 static void
2995 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
2996 {
2997 	struct spdk_bs_load_ctx *ctx = cb_arg;
2998 	uint64_t lba;
2999 
3000 	assert(ctx->cur_page < ctx->super->md_len);
3001 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3002 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3003 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3004 				  _spdk_bs_load_replay_md_cpl, ctx);
3005 }
3006 
3007 static void
3008 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3009 {
3010 	struct spdk_bs_load_ctx *ctx = cb_arg;
3011 
3012 	ctx->page_index = 0;
3013 	ctx->cur_page = 0;
3014 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3015 				     SPDK_BS_PAGE_SIZE,
3016 				     NULL);
3017 	if (!ctx->page) {
3018 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3019 		return;
3020 	}
3021 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3022 }
3023 
3024 static void
3025 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3026 {
3027 	struct spdk_bs_load_ctx *ctx = cb_arg;
3028 	int		rc;
3029 
3030 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3031 	if (rc < 0) {
3032 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3033 		return;
3034 	}
3035 
3036 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3037 	if (rc < 0) {
3038 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3039 		return;
3040 	}
3041 
3042 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3043 	if (rc < 0) {
3044 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3045 		return;
3046 	}
3047 
3048 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3049 	_spdk_bs_load_replay_md(seq, cb_arg);
3050 }
3051 
3052 static void
3053 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3054 {
3055 	struct spdk_bs_load_ctx *ctx = cb_arg;
3056 	uint32_t	crc;
3057 	int		rc;
3058 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3059 
3060 	if (ctx->super->version > SPDK_BS_VERSION ||
3061 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3062 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3063 		return;
3064 	}
3065 
3066 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3067 		   sizeof(ctx->super->signature)) != 0) {
3068 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3069 		return;
3070 	}
3071 
3072 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3073 	if (crc != ctx->super->crc) {
3074 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3075 		return;
3076 	}
3077 
3078 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3079 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3080 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3081 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3082 	} else {
3083 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3084 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3085 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3086 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3087 		return;
3088 	}
3089 
3090 	if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3091 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3092 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3093 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3094 		return;
3095 	}
3096 
3097 	if (ctx->super->size == 0) {
3098 		ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen;
3099 	}
3100 
3101 	/* Parse the super block */
3102 	ctx->bs->clean = 1;
3103 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3104 	ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size;
3105 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3106 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3107 	if (rc < 0) {
3108 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3109 		return;
3110 	}
3111 	ctx->bs->md_start = ctx->super->md_start;
3112 	ctx->bs->md_len = ctx->super->md_len;
3113 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - divide_round_up(
3114 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3115 	ctx->bs->super_blob = ctx->super->super_blob;
3116 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3117 
3118 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3119 		_spdk_bs_recover(seq, ctx);
3120 	} else {
3121 		_spdk_bs_load_read_used_pages(seq, ctx);
3122 	}
3123 }
3124 
3125 void
3126 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3127 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3128 {
3129 	struct spdk_blob_store	*bs;
3130 	struct spdk_bs_cpl	cpl;
3131 	spdk_bs_sequence_t	*seq;
3132 	struct spdk_bs_load_ctx *ctx;
3133 	struct spdk_bs_opts	opts = {};
3134 	int err;
3135 
3136 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3137 
3138 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3139 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3140 		dev->destroy(dev);
3141 		cb_fn(cb_arg, NULL, -EINVAL);
3142 		return;
3143 	}
3144 
3145 	if (o) {
3146 		opts = *o;
3147 	} else {
3148 		spdk_bs_opts_init(&opts);
3149 	}
3150 
3151 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3152 		dev->destroy(dev);
3153 		cb_fn(cb_arg, NULL, -EINVAL);
3154 		return;
3155 	}
3156 
3157 	err = _spdk_bs_alloc(dev, &opts, &bs);
3158 	if (err) {
3159 		dev->destroy(dev);
3160 		cb_fn(cb_arg, NULL, err);
3161 		return;
3162 	}
3163 
3164 	ctx = calloc(1, sizeof(*ctx));
3165 	if (!ctx) {
3166 		_spdk_bs_free(bs);
3167 		cb_fn(cb_arg, NULL, -ENOMEM);
3168 		return;
3169 	}
3170 
3171 	ctx->bs = bs;
3172 	ctx->is_load = true;
3173 	ctx->iter_cb_fn = opts.iter_cb_fn;
3174 	ctx->iter_cb_arg = opts.iter_cb_arg;
3175 
3176 	/* Allocate memory for the super block */
3177 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3178 	if (!ctx->super) {
3179 		free(ctx);
3180 		_spdk_bs_free(bs);
3181 		cb_fn(cb_arg, NULL, -ENOMEM);
3182 		return;
3183 	}
3184 
3185 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3186 	cpl.u.bs_handle.cb_fn = cb_fn;
3187 	cpl.u.bs_handle.cb_arg = cb_arg;
3188 	cpl.u.bs_handle.bs = bs;
3189 
3190 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3191 	if (!seq) {
3192 		spdk_dma_free(ctx->super);
3193 		free(ctx);
3194 		_spdk_bs_free(bs);
3195 		cb_fn(cb_arg, NULL, -ENOMEM);
3196 		return;
3197 	}
3198 
3199 	/* Read the super block */
3200 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3201 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3202 				  _spdk_bs_load_super_cpl, ctx);
3203 }
3204 
3205 /* END spdk_bs_load */
3206 
3207 /* START spdk_bs_dump */
3208 
3209 struct spdk_bs_dump_ctx {
3210 	struct spdk_blob_store		*bs;
3211 	struct spdk_bs_super_block	*super;
3212 	uint32_t			cur_page;
3213 	struct spdk_blob_md_page	*page;
3214 	spdk_bs_sequence_t		*seq;
3215 	FILE				*fp;
3216 	spdk_bs_dump_print_xattr	print_xattr_fn;
3217 	char				xattr_name[4096];
3218 };
3219 
3220 static void
3221 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3222 {
3223 	spdk_dma_free(ctx->super);
3224 
3225 	/*
3226 	 * We need to defer calling spdk_bs_call_cpl() until after
3227 	 * dev destruction, so tuck these away for later use.
3228 	 */
3229 	ctx->bs->unload_err = bserrno;
3230 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3231 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3232 
3233 	spdk_bs_sequence_finish(seq, 0);
3234 	_spdk_bs_free(ctx->bs);
3235 	free(ctx);
3236 }
3237 
3238 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3239 
3240 static void
3241 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3242 {
3243 	uint32_t page_idx = ctx->cur_page;
3244 	struct spdk_blob_md_page *page = ctx->page;
3245 	struct spdk_blob_md_descriptor *desc;
3246 	size_t cur_desc = 0;
3247 	uint32_t crc;
3248 
3249 	fprintf(ctx->fp, "=========\n");
3250 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3251 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3252 
3253 	crc = _spdk_blob_md_page_calc_crc(page);
3254 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3255 
3256 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3257 	while (cur_desc < sizeof(page->descriptors)) {
3258 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3259 			if (desc->length == 0) {
3260 				/* If padding and length are 0, this terminates the page */
3261 				break;
3262 			}
3263 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3264 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3265 			unsigned int				i;
3266 
3267 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3268 
3269 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3270 				if (desc_extent->extents[i].cluster_idx != 0) {
3271 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3272 						desc_extent->extents[i].cluster_idx);
3273 				} else {
3274 					fprintf(ctx->fp, "Unallocated Extent - ");
3275 				}
3276 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3277 				fprintf(ctx->fp, "\n");
3278 			}
3279 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3280 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3281 			uint32_t i;
3282 
3283 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3284 
3285 			if (desc_xattr->length !=
3286 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3287 			    desc_xattr->name_length + desc_xattr->value_length) {
3288 			}
3289 
3290 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3291 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3292 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3293 			fprintf(ctx->fp, "       value = \"");
3294 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3295 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3296 					    desc_xattr->value_length);
3297 			fprintf(ctx->fp, "\"\n");
3298 			for (i = 0; i < desc_xattr->value_length; i++) {
3299 				if (i % 16 == 0) {
3300 					fprintf(ctx->fp, "               ");
3301 				}
3302 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3303 				if ((i + 1) % 16 == 0) {
3304 					fprintf(ctx->fp, "\n");
3305 				}
3306 			}
3307 			if (i % 16 != 0) {
3308 				fprintf(ctx->fp, "\n");
3309 			}
3310 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3311 			/* TODO */
3312 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3313 			/* TODO */
3314 		} else {
3315 			/* Error */
3316 		}
3317 		/* Advance to the next descriptor */
3318 		cur_desc += sizeof(*desc) + desc->length;
3319 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3320 			break;
3321 		}
3322 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3323 	}
3324 }
3325 
3326 static void
3327 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3328 {
3329 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3330 
3331 	if (bserrno != 0) {
3332 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3333 		return;
3334 	}
3335 
3336 	if (ctx->page->id != 0) {
3337 		_spdk_bs_dump_print_md_page(ctx);
3338 	}
3339 
3340 	ctx->cur_page++;
3341 
3342 	if (ctx->cur_page < ctx->super->md_len) {
3343 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3344 	} else {
3345 		spdk_dma_free(ctx->page);
3346 		_spdk_bs_dump_finish(seq, ctx, 0);
3347 	}
3348 }
3349 
3350 static void
3351 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3352 {
3353 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3354 	uint64_t lba;
3355 
3356 	assert(ctx->cur_page < ctx->super->md_len);
3357 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3358 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3359 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3360 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3361 }
3362 
3363 static void
3364 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3365 {
3366 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3367 
3368 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3369 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3370 		   sizeof(ctx->super->signature)) != 0) {
3371 		fprintf(ctx->fp, "(Mismatch)\n");
3372 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3373 		return;
3374 	} else {
3375 		fprintf(ctx->fp, "(OK)\n");
3376 	}
3377 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3378 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3379 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3380 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3381 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3382 	fprintf(ctx->fp, "Super Blob ID: ");
3383 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3384 		fprintf(ctx->fp, "(None)\n");
3385 	} else {
3386 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3387 	}
3388 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3389 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3390 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3391 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3392 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3393 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3394 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3395 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3396 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3397 
3398 	ctx->cur_page = 0;
3399 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3400 				     SPDK_BS_PAGE_SIZE,
3401 				     NULL);
3402 	if (!ctx->page) {
3403 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3404 		return;
3405 	}
3406 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3407 }
3408 
3409 void
3410 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3411 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3412 {
3413 	struct spdk_blob_store	*bs;
3414 	struct spdk_bs_cpl	cpl;
3415 	spdk_bs_sequence_t	*seq;
3416 	struct spdk_bs_dump_ctx *ctx;
3417 	struct spdk_bs_opts	opts = {};
3418 	int err;
3419 
3420 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3421 
3422 	spdk_bs_opts_init(&opts);
3423 
3424 	err = _spdk_bs_alloc(dev, &opts, &bs);
3425 	if (err) {
3426 		dev->destroy(dev);
3427 		cb_fn(cb_arg, err);
3428 		return;
3429 	}
3430 
3431 	ctx = calloc(1, sizeof(*ctx));
3432 	if (!ctx) {
3433 		_spdk_bs_free(bs);
3434 		cb_fn(cb_arg, -ENOMEM);
3435 		return;
3436 	}
3437 
3438 	ctx->bs = bs;
3439 	ctx->fp = fp;
3440 	ctx->print_xattr_fn = print_xattr_fn;
3441 
3442 	/* Allocate memory for the super block */
3443 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3444 	if (!ctx->super) {
3445 		free(ctx);
3446 		_spdk_bs_free(bs);
3447 		cb_fn(cb_arg, -ENOMEM);
3448 		return;
3449 	}
3450 
3451 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3452 	cpl.u.bs_basic.cb_fn = cb_fn;
3453 	cpl.u.bs_basic.cb_arg = cb_arg;
3454 
3455 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3456 	if (!seq) {
3457 		spdk_dma_free(ctx->super);
3458 		free(ctx);
3459 		_spdk_bs_free(bs);
3460 		cb_fn(cb_arg, -ENOMEM);
3461 		return;
3462 	}
3463 
3464 	/* Read the super block */
3465 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3466 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3467 				  _spdk_bs_dump_super_cpl, ctx);
3468 }
3469 
3470 /* END spdk_bs_dump */
3471 
3472 /* START spdk_bs_init */
3473 
3474 struct spdk_bs_init_ctx {
3475 	struct spdk_blob_store		*bs;
3476 	struct spdk_bs_super_block	*super;
3477 };
3478 
3479 static void
3480 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3481 {
3482 	struct spdk_bs_init_ctx *ctx = cb_arg;
3483 
3484 	spdk_dma_free(ctx->super);
3485 	free(ctx);
3486 
3487 	spdk_bs_sequence_finish(seq, bserrno);
3488 }
3489 
3490 static void
3491 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3492 {
3493 	struct spdk_bs_init_ctx *ctx = cb_arg;
3494 
3495 	/* Write super block */
3496 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3497 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3498 				   _spdk_bs_init_persist_super_cpl, ctx);
3499 }
3500 
3501 void
3502 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3503 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3504 {
3505 	struct spdk_bs_init_ctx *ctx;
3506 	struct spdk_blob_store	*bs;
3507 	struct spdk_bs_cpl	cpl;
3508 	spdk_bs_sequence_t	*seq;
3509 	spdk_bs_batch_t		*batch;
3510 	uint64_t		num_md_lba;
3511 	uint64_t		num_md_pages;
3512 	uint64_t		num_md_clusters;
3513 	uint32_t		i;
3514 	struct spdk_bs_opts	opts = {};
3515 	int			rc;
3516 
3517 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3518 
3519 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3520 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3521 			    dev->blocklen);
3522 		dev->destroy(dev);
3523 		cb_fn(cb_arg, NULL, -EINVAL);
3524 		return;
3525 	}
3526 
3527 	if (o) {
3528 		opts = *o;
3529 	} else {
3530 		spdk_bs_opts_init(&opts);
3531 	}
3532 
3533 	if (_spdk_bs_opts_verify(&opts) != 0) {
3534 		dev->destroy(dev);
3535 		cb_fn(cb_arg, NULL, -EINVAL);
3536 		return;
3537 	}
3538 
3539 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3540 	if (rc) {
3541 		dev->destroy(dev);
3542 		cb_fn(cb_arg, NULL, rc);
3543 		return;
3544 	}
3545 
3546 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3547 		/* By default, allocate 1 page per cluster.
3548 		 * Technically, this over-allocates metadata
3549 		 * because more metadata will reduce the number
3550 		 * of usable clusters. This can be addressed with
3551 		 * more complex math in the future.
3552 		 */
3553 		bs->md_len = bs->total_clusters;
3554 	} else {
3555 		bs->md_len = opts.num_md_pages;
3556 	}
3557 
3558 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3559 	if (rc < 0) {
3560 		_spdk_bs_free(bs);
3561 		cb_fn(cb_arg, NULL, -ENOMEM);
3562 		return;
3563 	}
3564 
3565 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3566 	if (rc < 0) {
3567 		_spdk_bs_free(bs);
3568 		cb_fn(cb_arg, NULL, -ENOMEM);
3569 		return;
3570 	}
3571 
3572 	ctx = calloc(1, sizeof(*ctx));
3573 	if (!ctx) {
3574 		_spdk_bs_free(bs);
3575 		cb_fn(cb_arg, NULL, -ENOMEM);
3576 		return;
3577 	}
3578 
3579 	ctx->bs = bs;
3580 
3581 	/* Allocate memory for the super block */
3582 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3583 	if (!ctx->super) {
3584 		free(ctx);
3585 		_spdk_bs_free(bs);
3586 		cb_fn(cb_arg, NULL, -ENOMEM);
3587 		return;
3588 	}
3589 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3590 	       sizeof(ctx->super->signature));
3591 	ctx->super->version = SPDK_BS_VERSION;
3592 	ctx->super->length = sizeof(*ctx->super);
3593 	ctx->super->super_blob = bs->super_blob;
3594 	ctx->super->clean = 0;
3595 	ctx->super->cluster_size = bs->cluster_sz;
3596 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3597 
3598 	/* Calculate how many pages the metadata consumes at the front
3599 	 * of the disk.
3600 	 */
3601 
3602 	/* The super block uses 1 page */
3603 	num_md_pages = 1;
3604 
3605 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3606 	 * up to the nearest page, plus a header.
3607 	 */
3608 	ctx->super->used_page_mask_start = num_md_pages;
3609 	ctx->super->used_page_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3610 					 divide_round_up(bs->md_len, 8),
3611 					 SPDK_BS_PAGE_SIZE);
3612 	num_md_pages += ctx->super->used_page_mask_len;
3613 
3614 	/* The used_clusters mask requires 1 bit per cluster, rounded
3615 	 * up to the nearest page, plus a header.
3616 	 */
3617 	ctx->super->used_cluster_mask_start = num_md_pages;
3618 	ctx->super->used_cluster_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3619 					    divide_round_up(bs->total_clusters, 8),
3620 					    SPDK_BS_PAGE_SIZE);
3621 	num_md_pages += ctx->super->used_cluster_mask_len;
3622 
3623 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3624 	 * up to the nearest page, plus a header.
3625 	 */
3626 	ctx->super->used_blobid_mask_start = num_md_pages;
3627 	ctx->super->used_blobid_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3628 					   divide_round_up(bs->md_len, 8),
3629 					   SPDK_BS_PAGE_SIZE);
3630 	num_md_pages += ctx->super->used_blobid_mask_len;
3631 
3632 	/* The metadata region size was chosen above */
3633 	ctx->super->md_start = bs->md_start = num_md_pages;
3634 	ctx->super->md_len = bs->md_len;
3635 	num_md_pages += bs->md_len;
3636 
3637 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3638 
3639 	ctx->super->size = dev->blockcnt * dev->blocklen;
3640 
3641 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3642 
3643 	num_md_clusters = divide_round_up(num_md_pages, bs->pages_per_cluster);
3644 	if (num_md_clusters > bs->total_clusters) {
3645 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3646 			    "please decrease number of pages reserved for metadata "
3647 			    "or increase cluster size.\n");
3648 		spdk_dma_free(ctx->super);
3649 		free(ctx);
3650 		_spdk_bs_free(bs);
3651 		cb_fn(cb_arg, NULL, -ENOMEM);
3652 		return;
3653 	}
3654 	/* Claim all of the clusters used by the metadata */
3655 	for (i = 0; i < num_md_clusters; i++) {
3656 		_spdk_bs_claim_cluster(bs, i);
3657 	}
3658 
3659 	bs->total_data_clusters = bs->num_free_clusters;
3660 
3661 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3662 	cpl.u.bs_handle.cb_fn = cb_fn;
3663 	cpl.u.bs_handle.cb_arg = cb_arg;
3664 	cpl.u.bs_handle.bs = bs;
3665 
3666 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3667 	if (!seq) {
3668 		spdk_dma_free(ctx->super);
3669 		free(ctx);
3670 		_spdk_bs_free(bs);
3671 		cb_fn(cb_arg, NULL, -ENOMEM);
3672 		return;
3673 	}
3674 
3675 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3676 
3677 	/* Clear metadata space */
3678 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3679 	/* Trim data clusters */
3680 	spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3681 
3682 	spdk_bs_batch_close(batch);
3683 }
3684 
3685 /* END spdk_bs_init */
3686 
3687 /* START spdk_bs_destroy */
3688 
3689 static void
3690 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3691 {
3692 	struct spdk_bs_init_ctx *ctx = cb_arg;
3693 	struct spdk_blob_store *bs = ctx->bs;
3694 
3695 	/*
3696 	 * We need to defer calling spdk_bs_call_cpl() until after
3697 	 * dev destruction, so tuck these away for later use.
3698 	 */
3699 	bs->unload_err = bserrno;
3700 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3701 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3702 
3703 	spdk_bs_sequence_finish(seq, bserrno);
3704 
3705 	_spdk_bs_free(bs);
3706 	free(ctx);
3707 }
3708 
3709 void
3710 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3711 		void *cb_arg)
3712 {
3713 	struct spdk_bs_cpl	cpl;
3714 	spdk_bs_sequence_t	*seq;
3715 	struct spdk_bs_init_ctx *ctx;
3716 
3717 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3718 
3719 	if (!TAILQ_EMPTY(&bs->blobs)) {
3720 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3721 		cb_fn(cb_arg, -EBUSY);
3722 		return;
3723 	}
3724 
3725 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3726 	cpl.u.bs_basic.cb_fn = cb_fn;
3727 	cpl.u.bs_basic.cb_arg = cb_arg;
3728 
3729 	ctx = calloc(1, sizeof(*ctx));
3730 	if (!ctx) {
3731 		cb_fn(cb_arg, -ENOMEM);
3732 		return;
3733 	}
3734 
3735 	ctx->bs = bs;
3736 
3737 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3738 	if (!seq) {
3739 		free(ctx);
3740 		cb_fn(cb_arg, -ENOMEM);
3741 		return;
3742 	}
3743 
3744 	/* Write zeroes to the super block */
3745 	spdk_bs_sequence_write_zeroes_dev(seq,
3746 					  _spdk_bs_page_to_lba(bs, 0),
3747 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3748 					  _spdk_bs_destroy_trim_cpl, ctx);
3749 }
3750 
3751 /* END spdk_bs_destroy */
3752 
3753 /* START spdk_bs_unload */
3754 
3755 static void
3756 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3757 {
3758 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3759 
3760 	spdk_dma_free(ctx->super);
3761 
3762 	/*
3763 	 * We need to defer calling spdk_bs_call_cpl() until after
3764 	 * dev destruction, so tuck these away for later use.
3765 	 */
3766 	ctx->bs->unload_err = bserrno;
3767 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3768 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3769 
3770 	spdk_bs_sequence_finish(seq, bserrno);
3771 
3772 	_spdk_bs_free(ctx->bs);
3773 	free(ctx);
3774 }
3775 
3776 static void
3777 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3778 {
3779 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3780 
3781 	spdk_dma_free(ctx->mask);
3782 	ctx->super->clean = 1;
3783 
3784 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3785 }
3786 
3787 static void
3788 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3789 {
3790 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3791 
3792 	spdk_dma_free(ctx->mask);
3793 	ctx->mask = NULL;
3794 
3795 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3796 }
3797 
3798 static void
3799 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3800 {
3801 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3802 
3803 	spdk_dma_free(ctx->mask);
3804 	ctx->mask = NULL;
3805 
3806 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3807 }
3808 
3809 static void
3810 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3811 {
3812 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3813 }
3814 
3815 void
3816 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3817 {
3818 	struct spdk_bs_cpl	cpl;
3819 	spdk_bs_sequence_t	*seq;
3820 	struct spdk_bs_load_ctx *ctx;
3821 
3822 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3823 
3824 	if (!TAILQ_EMPTY(&bs->blobs)) {
3825 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3826 		cb_fn(cb_arg, -EBUSY);
3827 		return;
3828 	}
3829 
3830 	ctx = calloc(1, sizeof(*ctx));
3831 	if (!ctx) {
3832 		cb_fn(cb_arg, -ENOMEM);
3833 		return;
3834 	}
3835 
3836 	ctx->bs = bs;
3837 	ctx->is_load = false;
3838 
3839 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3840 	if (!ctx->super) {
3841 		free(ctx);
3842 		cb_fn(cb_arg, -ENOMEM);
3843 		return;
3844 	}
3845 
3846 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3847 	cpl.u.bs_basic.cb_fn = cb_fn;
3848 	cpl.u.bs_basic.cb_arg = cb_arg;
3849 
3850 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3851 	if (!seq) {
3852 		spdk_dma_free(ctx->super);
3853 		free(ctx);
3854 		cb_fn(cb_arg, -ENOMEM);
3855 		return;
3856 	}
3857 
3858 	/* Read super block */
3859 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3860 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3861 				  _spdk_bs_unload_read_super_cpl, ctx);
3862 }
3863 
3864 /* END spdk_bs_unload */
3865 
3866 /* START spdk_bs_set_super */
3867 
3868 struct spdk_bs_set_super_ctx {
3869 	struct spdk_blob_store		*bs;
3870 	struct spdk_bs_super_block	*super;
3871 };
3872 
3873 static void
3874 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3875 {
3876 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3877 
3878 	if (bserrno != 0) {
3879 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3880 	}
3881 
3882 	spdk_dma_free(ctx->super);
3883 
3884 	spdk_bs_sequence_finish(seq, bserrno);
3885 
3886 	free(ctx);
3887 }
3888 
3889 static void
3890 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3891 {
3892 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3893 
3894 	if (bserrno != 0) {
3895 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3896 		spdk_dma_free(ctx->super);
3897 		spdk_bs_sequence_finish(seq, bserrno);
3898 		free(ctx);
3899 		return;
3900 	}
3901 
3902 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3903 }
3904 
3905 void
3906 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3907 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3908 {
3909 	struct spdk_bs_cpl		cpl;
3910 	spdk_bs_sequence_t		*seq;
3911 	struct spdk_bs_set_super_ctx	*ctx;
3912 
3913 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3914 
3915 	ctx = calloc(1, sizeof(*ctx));
3916 	if (!ctx) {
3917 		cb_fn(cb_arg, -ENOMEM);
3918 		return;
3919 	}
3920 
3921 	ctx->bs = bs;
3922 
3923 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3924 	if (!ctx->super) {
3925 		free(ctx);
3926 		cb_fn(cb_arg, -ENOMEM);
3927 		return;
3928 	}
3929 
3930 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3931 	cpl.u.bs_basic.cb_fn = cb_fn;
3932 	cpl.u.bs_basic.cb_arg = cb_arg;
3933 
3934 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3935 	if (!seq) {
3936 		spdk_dma_free(ctx->super);
3937 		free(ctx);
3938 		cb_fn(cb_arg, -ENOMEM);
3939 		return;
3940 	}
3941 
3942 	bs->super_blob = blobid;
3943 
3944 	/* Read super block */
3945 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3946 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3947 				  _spdk_bs_set_super_read_cpl, ctx);
3948 }
3949 
3950 /* END spdk_bs_set_super */
3951 
3952 void
3953 spdk_bs_get_super(struct spdk_blob_store *bs,
3954 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
3955 {
3956 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
3957 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
3958 	} else {
3959 		cb_fn(cb_arg, bs->super_blob, 0);
3960 	}
3961 }
3962 
3963 uint64_t
3964 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
3965 {
3966 	return bs->cluster_sz;
3967 }
3968 
3969 uint64_t
3970 spdk_bs_get_page_size(struct spdk_blob_store *bs)
3971 {
3972 	return SPDK_BS_PAGE_SIZE;
3973 }
3974 
3975 uint64_t
3976 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
3977 {
3978 	return bs->num_free_clusters;
3979 }
3980 
3981 uint64_t
3982 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
3983 {
3984 	return bs->total_data_clusters;
3985 }
3986 
3987 static int
3988 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
3989 {
3990 	bs->md_channel = spdk_get_io_channel(bs);
3991 	if (!bs->md_channel) {
3992 		SPDK_ERRLOG("Failed to get IO channel.\n");
3993 		return -1;
3994 	}
3995 
3996 	return 0;
3997 }
3998 
3999 static int
4000 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
4001 {
4002 	spdk_put_io_channel(bs->md_channel);
4003 
4004 	return 0;
4005 }
4006 
4007 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4008 {
4009 	assert(blob != NULL);
4010 
4011 	return blob->id;
4012 }
4013 
4014 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4015 {
4016 	assert(blob != NULL);
4017 
4018 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4019 }
4020 
4021 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4022 {
4023 	assert(blob != NULL);
4024 
4025 	return blob->active.num_clusters;
4026 }
4027 
4028 /* START spdk_bs_create_blob */
4029 
4030 static void
4031 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4032 {
4033 	struct spdk_blob *blob = cb_arg;
4034 
4035 	_spdk_blob_free(blob);
4036 
4037 	spdk_bs_sequence_finish(seq, bserrno);
4038 }
4039 
4040 static int
4041 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4042 		      bool internal)
4043 {
4044 	uint64_t i;
4045 	size_t value_len = 0;
4046 	int rc;
4047 	const void *value = NULL;
4048 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4049 		return -EINVAL;
4050 	}
4051 	for (i = 0; i < xattrs->count; i++) {
4052 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4053 		if (value == NULL || value_len == 0) {
4054 			return -EINVAL;
4055 		}
4056 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4057 		if (rc < 0) {
4058 			return rc;
4059 		}
4060 	}
4061 	return 0;
4062 }
4063 
4064 static void
4065 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4066 {
4067 	_spdk_blob_verify_md_op(blob);
4068 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4069 	blob->state = SPDK_BLOB_STATE_DIRTY;
4070 }
4071 
4072 static void
4073 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4074 		     const struct spdk_blob_opts *opts,
4075 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4076 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4077 {
4078 	struct spdk_blob	*blob;
4079 	uint32_t		page_idx;
4080 	struct spdk_bs_cpl	cpl;
4081 	struct spdk_blob_opts	opts_default;
4082 	struct spdk_blob_xattr_opts internal_xattrs_default;
4083 	spdk_bs_sequence_t	*seq;
4084 	spdk_blob_id		id;
4085 	int rc;
4086 
4087 	assert(spdk_get_thread() == bs->md_thread);
4088 
4089 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4090 	if (page_idx >= spdk_bit_array_capacity(bs->used_md_pages)) {
4091 		cb_fn(cb_arg, 0, -ENOMEM);
4092 		return;
4093 	}
4094 	spdk_bit_array_set(bs->used_blobids, page_idx);
4095 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4096 
4097 	id = _spdk_bs_page_to_blobid(page_idx);
4098 
4099 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4100 
4101 	blob = _spdk_blob_alloc(bs, id);
4102 	if (!blob) {
4103 		cb_fn(cb_arg, 0, -ENOMEM);
4104 		return;
4105 	}
4106 
4107 	if (!opts) {
4108 		spdk_blob_opts_init(&opts_default);
4109 		opts = &opts_default;
4110 	}
4111 	if (!internal_xattrs) {
4112 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4113 		internal_xattrs = &internal_xattrs_default;
4114 	}
4115 
4116 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4117 	if (rc < 0) {
4118 		_spdk_blob_free(blob);
4119 		cb_fn(cb_arg, 0, rc);
4120 		return;
4121 	}
4122 
4123 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4124 	if (rc < 0) {
4125 		_spdk_blob_free(blob);
4126 		cb_fn(cb_arg, 0, rc);
4127 		return;
4128 	}
4129 
4130 	if (opts->thin_provision) {
4131 		_spdk_blob_set_thin_provision(blob);
4132 	}
4133 
4134 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4135 	if (rc < 0) {
4136 		_spdk_blob_free(blob);
4137 		cb_fn(cb_arg, 0, rc);
4138 		return;
4139 	}
4140 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4141 	cpl.u.blobid.cb_fn = cb_fn;
4142 	cpl.u.blobid.cb_arg = cb_arg;
4143 	cpl.u.blobid.blobid = blob->id;
4144 
4145 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4146 	if (!seq) {
4147 		_spdk_blob_free(blob);
4148 		cb_fn(cb_arg, 0, -ENOMEM);
4149 		return;
4150 	}
4151 
4152 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4153 }
4154 
4155 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4156 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4157 {
4158 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4159 }
4160 
4161 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4162 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4163 {
4164 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4165 }
4166 
4167 /* END spdk_bs_create_blob */
4168 
4169 /* START blob_cleanup */
4170 
4171 struct spdk_clone_snapshot_ctx {
4172 	struct spdk_bs_cpl      cpl;
4173 	int bserrno;
4174 	bool frozen;
4175 
4176 	struct spdk_io_channel *channel;
4177 
4178 	/* Current cluster for inflate operation */
4179 	uint64_t cluster;
4180 
4181 	/* For inflation force allocation of all unallocated clusters and remove
4182 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4183 	bool allocate_all;
4184 
4185 	struct {
4186 		spdk_blob_id id;
4187 		struct spdk_blob *blob;
4188 	} original;
4189 	struct {
4190 		spdk_blob_id id;
4191 		struct spdk_blob *blob;
4192 	} new;
4193 
4194 	/* xattrs specified for snapshot/clones only. They have no impact on
4195 	 * the original blobs xattrs. */
4196 	const struct spdk_blob_xattr_opts *xattrs;
4197 };
4198 
4199 static void
4200 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4201 {
4202 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4203 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4204 
4205 	if (bserrno != 0) {
4206 		if (ctx->bserrno != 0) {
4207 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4208 		} else {
4209 			ctx->bserrno = bserrno;
4210 		}
4211 	}
4212 
4213 	switch (cpl->type) {
4214 	case SPDK_BS_CPL_TYPE_BLOBID:
4215 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4216 		break;
4217 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4218 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4219 		break;
4220 	default:
4221 		SPDK_UNREACHABLE();
4222 		break;
4223 	}
4224 
4225 	free(ctx);
4226 }
4227 
4228 static void
4229 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4230 {
4231 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4232 	struct spdk_blob *origblob = ctx->original.blob;
4233 
4234 	if (bserrno != 0) {
4235 		if (ctx->bserrno != 0) {
4236 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4237 		} else {
4238 			ctx->bserrno = bserrno;
4239 		}
4240 	}
4241 
4242 	ctx->original.id = origblob->id;
4243 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4244 }
4245 
4246 static void
4247 _spdk_bs_clone_snapshot_origblob_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 *origblob = ctx->original.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 	if (ctx->frozen) {
4261 		/* Unfreeze any outstanding I/O */
4262 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4263 	} else {
4264 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4265 	}
4266 
4267 }
4268 
4269 static void
4270 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4271 {
4272 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4273 	struct spdk_blob *newblob = ctx->new.blob;
4274 
4275 	if (bserrno != 0) {
4276 		if (ctx->bserrno != 0) {
4277 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4278 		} else {
4279 			ctx->bserrno = bserrno;
4280 		}
4281 	}
4282 
4283 	ctx->new.id = newblob->id;
4284 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4285 }
4286 
4287 /* END blob_cleanup */
4288 
4289 /* START spdk_bs_create_snapshot */
4290 
4291 static void
4292 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4293 {
4294 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4295 	struct spdk_blob *newblob = ctx->new.blob;
4296 
4297 	if (bserrno != 0) {
4298 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4299 		return;
4300 	}
4301 
4302 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4303 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4304 	if (bserrno != 0) {
4305 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4306 		return;
4307 	}
4308 
4309 	_spdk_bs_blob_list_add(ctx->original.blob);
4310 
4311 	spdk_blob_set_read_only(newblob);
4312 
4313 	/* sync snapshot metadata */
4314 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4315 }
4316 
4317 static void
4318 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4319 {
4320 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4321 	struct spdk_blob *origblob = ctx->original.blob;
4322 	struct spdk_blob *newblob = ctx->new.blob;
4323 
4324 	if (bserrno != 0) {
4325 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4326 		return;
4327 	}
4328 
4329 	/* Set internal xattr for snapshot id */
4330 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4331 	if (bserrno != 0) {
4332 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4333 		return;
4334 	}
4335 
4336 	_spdk_bs_blob_list_remove(origblob);
4337 	origblob->parent_id = newblob->id;
4338 
4339 	/* Create new back_bs_dev for snapshot */
4340 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4341 	if (origblob->back_bs_dev == NULL) {
4342 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4343 		return;
4344 	}
4345 
4346 	/* set clone blob as thin provisioned */
4347 	_spdk_blob_set_thin_provision(origblob);
4348 
4349 	_spdk_bs_blob_list_add(newblob);
4350 
4351 	/* Zero out origblob cluster map */
4352 	memset(origblob->active.clusters, 0,
4353 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4354 
4355 	/* sync clone metadata */
4356 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4357 }
4358 
4359 static void
4360 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4361 {
4362 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4363 	struct spdk_blob *origblob = ctx->original.blob;
4364 	struct spdk_blob *newblob = ctx->new.blob;
4365 	int bserrno;
4366 
4367 	if (rc != 0) {
4368 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4369 		return;
4370 	}
4371 
4372 	ctx->frozen = true;
4373 
4374 	/* set new back_bs_dev for snapshot */
4375 	newblob->back_bs_dev = origblob->back_bs_dev;
4376 	/* Set invalid flags from origblob */
4377 	newblob->invalid_flags = origblob->invalid_flags;
4378 
4379 	/* inherit parent from original blob if set */
4380 	newblob->parent_id = origblob->parent_id;
4381 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4382 		/* Set internal xattr for snapshot id */
4383 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4384 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4385 		if (bserrno != 0) {
4386 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4387 			return;
4388 		}
4389 	}
4390 
4391 	/* Copy cluster map to snapshot */
4392 	memcpy(newblob->active.clusters, origblob->active.clusters,
4393 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4394 
4395 	/* sync snapshot metadata */
4396 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4397 }
4398 
4399 static void
4400 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4401 {
4402 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4403 	struct spdk_blob *origblob = ctx->original.blob;
4404 	struct spdk_blob *newblob = _blob;
4405 
4406 	if (bserrno != 0) {
4407 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4408 		return;
4409 	}
4410 
4411 	ctx->new.blob = newblob;
4412 
4413 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4414 }
4415 
4416 static void
4417 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4418 {
4419 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4420 	struct spdk_blob *origblob = ctx->original.blob;
4421 
4422 	if (bserrno != 0) {
4423 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4424 		return;
4425 	}
4426 
4427 	ctx->new.id = blobid;
4428 	ctx->cpl.u.blobid.blobid = blobid;
4429 
4430 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4431 }
4432 
4433 
4434 static void
4435 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4436 			const void **value, size_t *value_len)
4437 {
4438 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4439 
4440 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4441 	*value = &blob->id;
4442 	*value_len = sizeof(blob->id);
4443 }
4444 
4445 static void
4446 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4447 {
4448 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4449 	struct spdk_blob_opts opts;
4450 	struct spdk_blob_xattr_opts internal_xattrs;
4451 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4452 
4453 	if (bserrno != 0) {
4454 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4455 		return;
4456 	}
4457 
4458 	ctx->original.blob = _blob;
4459 
4460 	if (_blob->data_ro || _blob->md_ro) {
4461 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4462 			      _blob->id);
4463 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4464 		return;
4465 	}
4466 
4467 	spdk_blob_opts_init(&opts);
4468 	_spdk_blob_xattrs_init(&internal_xattrs);
4469 
4470 	/* Change the size of new blob to the same as in original blob,
4471 	 * but do not allocate clusters */
4472 	opts.thin_provision = true;
4473 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4474 
4475 	/* If there are any xattrs specified for snapshot, set them now */
4476 	if (ctx->xattrs) {
4477 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4478 	}
4479 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4480 	internal_xattrs.count = 1;
4481 	internal_xattrs.ctx = _blob;
4482 	internal_xattrs.names = xattrs_names;
4483 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4484 
4485 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4486 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4487 }
4488 
4489 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4490 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4491 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4492 {
4493 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4494 
4495 	if (!ctx) {
4496 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4497 		return;
4498 	}
4499 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4500 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4501 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4502 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4503 	ctx->bserrno = 0;
4504 	ctx->frozen = false;
4505 	ctx->original.id = blobid;
4506 	ctx->xattrs = snapshot_xattrs;
4507 
4508 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4509 }
4510 /* END spdk_bs_create_snapshot */
4511 
4512 /* START spdk_bs_create_clone */
4513 
4514 static void
4515 _spdk_bs_xattr_clone(void *arg, const char *name,
4516 		     const void **value, size_t *value_len)
4517 {
4518 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4519 
4520 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4521 	*value = &blob->id;
4522 	*value_len = sizeof(blob->id);
4523 }
4524 
4525 static void
4526 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4527 {
4528 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4529 	struct spdk_blob *clone = _blob;
4530 
4531 	ctx->new.blob = clone;
4532 	_spdk_bs_blob_list_add(clone);
4533 
4534 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4535 }
4536 
4537 static void
4538 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4539 {
4540 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4541 
4542 	ctx->cpl.u.blobid.blobid = blobid;
4543 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4544 }
4545 
4546 static void
4547 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4548 {
4549 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4550 	struct spdk_blob_opts		opts;
4551 	struct spdk_blob_xattr_opts internal_xattrs;
4552 	char *xattr_names[] = { BLOB_SNAPSHOT };
4553 
4554 	if (bserrno != 0) {
4555 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4556 		return;
4557 	}
4558 
4559 	ctx->original.blob = _blob;
4560 
4561 	if (!_blob->data_ro || !_blob->md_ro) {
4562 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4563 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4564 		return;
4565 	}
4566 
4567 	spdk_blob_opts_init(&opts);
4568 	_spdk_blob_xattrs_init(&internal_xattrs);
4569 
4570 	opts.thin_provision = true;
4571 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4572 	if (ctx->xattrs) {
4573 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4574 	}
4575 
4576 	/* Set internal xattr BLOB_SNAPSHOT */
4577 	internal_xattrs.count = 1;
4578 	internal_xattrs.ctx = _blob;
4579 	internal_xattrs.names = xattr_names;
4580 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4581 
4582 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4583 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4584 }
4585 
4586 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4587 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4588 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4589 {
4590 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4591 
4592 	if (!ctx) {
4593 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4594 		return;
4595 	}
4596 
4597 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4598 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4599 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4600 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4601 	ctx->bserrno = 0;
4602 	ctx->xattrs = clone_xattrs;
4603 	ctx->original.id = blobid;
4604 
4605 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4606 }
4607 
4608 /* END spdk_bs_create_clone */
4609 
4610 /* START spdk_bs_inflate_blob */
4611 
4612 static void
4613 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4614 {
4615 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4616 	struct spdk_blob *_blob = ctx->original.blob;
4617 
4618 	if (bserrno != 0) {
4619 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4620 		return;
4621 	}
4622 
4623 	assert(_parent != NULL);
4624 
4625 	_spdk_bs_blob_list_remove(_blob);
4626 	_blob->parent_id = _parent->id;
4627 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4628 			     sizeof(spdk_blob_id), true);
4629 
4630 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4631 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4632 	_spdk_bs_blob_list_add(_blob);
4633 
4634 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4635 }
4636 
4637 static void
4638 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4639 {
4640 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4641 	struct spdk_blob *_blob = ctx->original.blob;
4642 	struct spdk_blob *_parent;
4643 
4644 	if (bserrno != 0) {
4645 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4646 		return;
4647 	}
4648 
4649 	if (ctx->allocate_all) {
4650 		/* remove thin provisioning */
4651 		_spdk_bs_blob_list_remove(_blob);
4652 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4653 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4654 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4655 		_blob->back_bs_dev = NULL;
4656 		_blob->parent_id = SPDK_BLOBID_INVALID;
4657 	} else {
4658 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4659 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4660 			/* We must change the parent of the inflated blob */
4661 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4662 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4663 			return;
4664 		}
4665 
4666 		_spdk_bs_blob_list_remove(_blob);
4667 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4668 		_blob->parent_id = SPDK_BLOBID_INVALID;
4669 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4670 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4671 	}
4672 
4673 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4674 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4675 }
4676 
4677 /* Check if cluster needs allocation */
4678 static inline bool
4679 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4680 {
4681 	struct spdk_blob_bs_dev *b;
4682 
4683 	assert(blob != NULL);
4684 
4685 	if (blob->active.clusters[cluster] != 0) {
4686 		/* Cluster is already allocated */
4687 		return false;
4688 	}
4689 
4690 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4691 		/* Blob have no parent blob */
4692 		return allocate_all;
4693 	}
4694 
4695 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4696 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4697 }
4698 
4699 static void
4700 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4701 {
4702 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4703 	struct spdk_blob *_blob = ctx->original.blob;
4704 	uint64_t offset;
4705 
4706 	if (bserrno != 0) {
4707 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4708 		return;
4709 	}
4710 
4711 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4712 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4713 			break;
4714 		}
4715 	}
4716 
4717 	if (ctx->cluster < _blob->active.num_clusters) {
4718 		offset = _spdk_bs_cluster_to_page(_blob->bs, ctx->cluster);
4719 
4720 		/* We may safely increment a cluster before write */
4721 		ctx->cluster++;
4722 
4723 		/* Use zero length write to touch a cluster */
4724 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4725 				   _spdk_bs_inflate_blob_touch_next, ctx);
4726 	} else {
4727 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4728 	}
4729 }
4730 
4731 static void
4732 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4733 {
4734 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4735 	uint64_t lfc; /* lowest free cluster */
4736 	uint64_t i;
4737 
4738 	if (bserrno != 0) {
4739 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4740 		return;
4741 	}
4742 	ctx->original.blob = _blob;
4743 
4744 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4745 		/* This blob have no parent, so we cannot decouple it. */
4746 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4747 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4748 		return;
4749 	}
4750 
4751 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4752 		/* This is not thin provisioned blob. No need to inflate. */
4753 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4754 		return;
4755 	}
4756 
4757 	/* Do two passes - one to verify that we can obtain enough clusters
4758 	 * and another to actually claim them.
4759 	 */
4760 	lfc = 0;
4761 	for (i = 0; i < _blob->active.num_clusters; i++) {
4762 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4763 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4764 			if (lfc >= _blob->bs->total_clusters) {
4765 				/* No more free clusters. Cannot satisfy the request */
4766 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4767 				return;
4768 			}
4769 			lfc++;
4770 		}
4771 	}
4772 
4773 	ctx->cluster = 0;
4774 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4775 }
4776 
4777 static void
4778 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4779 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4780 {
4781 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4782 
4783 	if (!ctx) {
4784 		cb_fn(cb_arg, -ENOMEM);
4785 		return;
4786 	}
4787 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4788 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4789 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4790 	ctx->bserrno = 0;
4791 	ctx->original.id = blobid;
4792 	ctx->channel = channel;
4793 	ctx->allocate_all = allocate_all;
4794 
4795 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4796 }
4797 
4798 void
4799 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4800 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4801 {
4802 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4803 }
4804 
4805 void
4806 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4807 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4808 {
4809 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
4810 }
4811 /* END spdk_bs_inflate_blob */
4812 
4813 /* START spdk_blob_resize */
4814 struct spdk_bs_resize_ctx {
4815 	spdk_blob_op_complete cb_fn;
4816 	void *cb_arg;
4817 	struct spdk_blob *blob;
4818 	uint64_t sz;
4819 	int rc;
4820 };
4821 
4822 static void
4823 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4824 {
4825 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4826 
4827 	if (rc != 0) {
4828 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4829 	}
4830 
4831 	if (ctx->rc != 0) {
4832 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4833 		rc = ctx->rc;
4834 	}
4835 
4836 	ctx->blob->resize_in_progress = false;
4837 
4838 	ctx->cb_fn(ctx->cb_arg, rc);
4839 	free(ctx);
4840 }
4841 
4842 static void
4843 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4844 {
4845 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4846 
4847 	if (rc != 0) {
4848 		ctx->blob->resize_in_progress = false;
4849 		ctx->cb_fn(ctx->cb_arg, rc);
4850 		free(ctx);
4851 		return;
4852 	}
4853 
4854 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4855 
4856 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4857 }
4858 
4859 void
4860 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4861 {
4862 	struct spdk_bs_resize_ctx *ctx;
4863 
4864 	_spdk_blob_verify_md_op(blob);
4865 
4866 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4867 
4868 	if (blob->md_ro) {
4869 		cb_fn(cb_arg, -EPERM);
4870 		return;
4871 	}
4872 
4873 	if (sz == blob->active.num_clusters) {
4874 		cb_fn(cb_arg, 0);
4875 		return;
4876 	}
4877 
4878 	if (blob->resize_in_progress) {
4879 		cb_fn(cb_arg, -EBUSY);
4880 		return;
4881 	}
4882 
4883 	ctx = calloc(1, sizeof(*ctx));
4884 	if (!ctx) {
4885 		cb_fn(cb_arg, -ENOMEM);
4886 		return;
4887 	}
4888 
4889 	blob->resize_in_progress = true;
4890 	ctx->cb_fn = cb_fn;
4891 	ctx->cb_arg = cb_arg;
4892 	ctx->blob = blob;
4893 	ctx->sz = sz;
4894 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
4895 }
4896 
4897 /* END spdk_blob_resize */
4898 
4899 
4900 /* START spdk_bs_delete_blob */
4901 
4902 static void
4903 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
4904 {
4905 	spdk_bs_sequence_t *seq = cb_arg;
4906 
4907 	spdk_bs_sequence_finish(seq, bserrno);
4908 }
4909 
4910 static void
4911 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4912 {
4913 	struct spdk_blob *blob = cb_arg;
4914 
4915 	if (bserrno != 0) {
4916 		/*
4917 		 * We already removed this blob from the blobstore tailq, so
4918 		 *  we need to free it here since this is the last reference
4919 		 *  to it.
4920 		 */
4921 		_spdk_blob_free(blob);
4922 		_spdk_bs_delete_close_cpl(seq, bserrno);
4923 		return;
4924 	}
4925 
4926 	/*
4927 	 * This will immediately decrement the ref_count and call
4928 	 *  the completion routine since the metadata state is clean.
4929 	 *  By calling spdk_blob_close, we reduce the number of call
4930 	 *  points into code that touches the blob->open_ref count
4931 	 *  and the blobstore's blob list.
4932 	 */
4933 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
4934 }
4935 
4936 static void
4937 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
4938 {
4939 	spdk_bs_sequence_t *seq = cb_arg;
4940 	uint32_t page_num;
4941 
4942 	if (bserrno != 0) {
4943 		spdk_bs_sequence_finish(seq, bserrno);
4944 		return;
4945 	}
4946 
4947 	_spdk_blob_verify_md_op(blob);
4948 
4949 	if (blob->open_ref > 1) {
4950 		/*
4951 		 * Someone has this blob open (besides this delete context).
4952 		 *  Decrement the ref count directly and return -EBUSY.
4953 		 */
4954 		blob->open_ref--;
4955 		spdk_bs_sequence_finish(seq, -EBUSY);
4956 		return;
4957 	}
4958 
4959 	bserrno = _spdk_bs_blob_list_remove(blob);
4960 	if (bserrno != 0) {
4961 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Remove blob #%" PRIu64 " from a list\n", blob->id);
4962 		spdk_bs_sequence_finish(seq, bserrno);
4963 		return;
4964 	}
4965 
4966 	/*
4967 	 * Remove the blob from the blob_store list now, to ensure it does not
4968 	 *  get returned after this point by _spdk_blob_lookup().
4969 	 */
4970 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
4971 	page_num = _spdk_bs_blobid_to_page(blob->id);
4972 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
4973 	blob->state = SPDK_BLOB_STATE_DIRTY;
4974 	blob->active.num_pages = 0;
4975 	_spdk_blob_resize(blob, 0);
4976 
4977 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
4978 }
4979 
4980 void
4981 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
4982 		    spdk_blob_op_complete cb_fn, void *cb_arg)
4983 {
4984 	struct spdk_bs_cpl	cpl;
4985 	spdk_bs_sequence_t	*seq;
4986 	struct spdk_blob_list	*snapshot_entry = NULL;
4987 
4988 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
4989 
4990 	assert(spdk_get_thread() == bs->md_thread);
4991 
4992 	/* Check if this is a snapshot with clones */
4993 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
4994 		if (snapshot_entry->id == blobid) {
4995 			break;
4996 		}
4997 	}
4998 	if (snapshot_entry != NULL) {
4999 		/* If snapshot have clones, we cannot remove it */
5000 		if (!TAILQ_EMPTY(&snapshot_entry->clones)) {
5001 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
5002 			cb_fn(cb_arg, -EBUSY);
5003 			return;
5004 		}
5005 	}
5006 
5007 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5008 	cpl.u.blob_basic.cb_fn = cb_fn;
5009 	cpl.u.blob_basic.cb_arg = cb_arg;
5010 
5011 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5012 	if (!seq) {
5013 		cb_fn(cb_arg, -ENOMEM);
5014 		return;
5015 	}
5016 
5017 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5018 }
5019 
5020 /* END spdk_bs_delete_blob */
5021 
5022 /* START spdk_bs_open_blob */
5023 
5024 static void
5025 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5026 {
5027 	struct spdk_blob *blob = cb_arg;
5028 
5029 	/* If the blob have crc error, we just return NULL. */
5030 	if (blob == NULL) {
5031 		seq->cpl.u.blob_handle.blob = NULL;
5032 		spdk_bs_sequence_finish(seq, bserrno);
5033 		return;
5034 	}
5035 
5036 	blob->open_ref++;
5037 
5038 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5039 
5040 	spdk_bs_sequence_finish(seq, bserrno);
5041 }
5042 
5043 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5044 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5045 {
5046 	struct spdk_blob		*blob;
5047 	struct spdk_bs_cpl		cpl;
5048 	spdk_bs_sequence_t		*seq;
5049 	uint32_t			page_num;
5050 
5051 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5052 	assert(spdk_get_thread() == bs->md_thread);
5053 
5054 	page_num = _spdk_bs_blobid_to_page(blobid);
5055 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5056 		/* Invalid blobid */
5057 		cb_fn(cb_arg, NULL, -ENOENT);
5058 		return;
5059 	}
5060 
5061 	blob = _spdk_blob_lookup(bs, blobid);
5062 	if (blob) {
5063 		blob->open_ref++;
5064 		cb_fn(cb_arg, blob, 0);
5065 		return;
5066 	}
5067 
5068 	blob = _spdk_blob_alloc(bs, blobid);
5069 	if (!blob) {
5070 		cb_fn(cb_arg, NULL, -ENOMEM);
5071 		return;
5072 	}
5073 
5074 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5075 	cpl.u.blob_handle.cb_fn = cb_fn;
5076 	cpl.u.blob_handle.cb_arg = cb_arg;
5077 	cpl.u.blob_handle.blob = blob;
5078 
5079 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5080 	if (!seq) {
5081 		_spdk_blob_free(blob);
5082 		cb_fn(cb_arg, NULL, -ENOMEM);
5083 		return;
5084 	}
5085 
5086 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5087 }
5088 /* END spdk_bs_open_blob */
5089 
5090 /* START spdk_blob_set_read_only */
5091 int spdk_blob_set_read_only(struct spdk_blob *blob)
5092 {
5093 	_spdk_blob_verify_md_op(blob);
5094 
5095 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5096 
5097 	blob->state = SPDK_BLOB_STATE_DIRTY;
5098 	return 0;
5099 }
5100 /* END spdk_blob_set_read_only */
5101 
5102 /* START spdk_blob_sync_md */
5103 
5104 static void
5105 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5106 {
5107 	struct spdk_blob *blob = cb_arg;
5108 
5109 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5110 		blob->data_ro = true;
5111 		blob->md_ro = true;
5112 	}
5113 
5114 	spdk_bs_sequence_finish(seq, bserrno);
5115 }
5116 
5117 static void
5118 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5119 {
5120 	struct spdk_bs_cpl	cpl;
5121 	spdk_bs_sequence_t	*seq;
5122 
5123 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5124 	cpl.u.blob_basic.cb_fn = cb_fn;
5125 	cpl.u.blob_basic.cb_arg = cb_arg;
5126 
5127 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5128 	if (!seq) {
5129 		cb_fn(cb_arg, -ENOMEM);
5130 		return;
5131 	}
5132 
5133 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5134 }
5135 
5136 void
5137 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5138 {
5139 	_spdk_blob_verify_md_op(blob);
5140 
5141 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5142 
5143 	if (blob->md_ro) {
5144 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5145 		cb_fn(cb_arg, 0);
5146 		return;
5147 	}
5148 
5149 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5150 }
5151 
5152 /* END spdk_blob_sync_md */
5153 
5154 struct spdk_blob_insert_cluster_ctx {
5155 	struct spdk_thread	*thread;
5156 	struct spdk_blob	*blob;
5157 	uint32_t		cluster_num;	/* cluster index in blob */
5158 	uint32_t		cluster;	/* cluster on disk */
5159 	int			rc;
5160 	spdk_blob_op_complete	cb_fn;
5161 	void			*cb_arg;
5162 };
5163 
5164 static void
5165 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5166 {
5167 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5168 
5169 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5170 	free(ctx);
5171 }
5172 
5173 static void
5174 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5175 {
5176 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5177 
5178 	ctx->rc = bserrno;
5179 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5180 }
5181 
5182 static void
5183 _spdk_blob_insert_cluster_msg(void *arg)
5184 {
5185 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5186 
5187 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5188 	if (ctx->rc != 0) {
5189 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5190 		return;
5191 	}
5192 
5193 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5194 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5195 }
5196 
5197 static void
5198 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5199 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5200 {
5201 	struct spdk_blob_insert_cluster_ctx *ctx;
5202 
5203 	ctx = calloc(1, sizeof(*ctx));
5204 	if (ctx == NULL) {
5205 		cb_fn(cb_arg, -ENOMEM);
5206 		return;
5207 	}
5208 
5209 	ctx->thread = spdk_get_thread();
5210 	ctx->blob = blob;
5211 	ctx->cluster_num = cluster_num;
5212 	ctx->cluster = cluster;
5213 	ctx->cb_fn = cb_fn;
5214 	ctx->cb_arg = cb_arg;
5215 
5216 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5217 }
5218 
5219 /* START spdk_blob_close */
5220 
5221 static void
5222 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5223 {
5224 	struct spdk_blob *blob = cb_arg;
5225 
5226 	if (bserrno == 0) {
5227 		blob->open_ref--;
5228 		if (blob->open_ref == 0) {
5229 			/*
5230 			 * Blobs with active.num_pages == 0 are deleted blobs.
5231 			 *  these blobs are removed from the blob_store list
5232 			 *  when the deletion process starts - so don't try to
5233 			 *  remove them again.
5234 			 */
5235 			if (blob->active.num_pages > 0) {
5236 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5237 			}
5238 			_spdk_blob_free(blob);
5239 		}
5240 	}
5241 
5242 	spdk_bs_sequence_finish(seq, bserrno);
5243 }
5244 
5245 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5246 {
5247 	struct spdk_bs_cpl	cpl;
5248 	spdk_bs_sequence_t	*seq;
5249 
5250 	_spdk_blob_verify_md_op(blob);
5251 
5252 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5253 
5254 	if (blob->open_ref == 0) {
5255 		cb_fn(cb_arg, -EBADF);
5256 		return;
5257 	}
5258 
5259 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5260 	cpl.u.blob_basic.cb_fn = cb_fn;
5261 	cpl.u.blob_basic.cb_arg = cb_arg;
5262 
5263 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5264 	if (!seq) {
5265 		cb_fn(cb_arg, -ENOMEM);
5266 		return;
5267 	}
5268 
5269 	/* Sync metadata */
5270 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5271 }
5272 
5273 /* END spdk_blob_close */
5274 
5275 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5276 {
5277 	return spdk_get_io_channel(bs);
5278 }
5279 
5280 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5281 {
5282 	spdk_put_io_channel(channel);
5283 }
5284 
5285 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5286 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5287 {
5288 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5289 				     SPDK_BLOB_UNMAP);
5290 }
5291 
5292 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5293 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5294 {
5295 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5296 				     SPDK_BLOB_WRITE_ZEROES);
5297 }
5298 
5299 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5300 			void *payload, uint64_t offset, uint64_t length,
5301 			spdk_blob_op_complete cb_fn, void *cb_arg)
5302 {
5303 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5304 				     SPDK_BLOB_WRITE);
5305 }
5306 
5307 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5308 		       void *payload, uint64_t offset, uint64_t length,
5309 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5310 {
5311 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5312 				     SPDK_BLOB_READ);
5313 }
5314 
5315 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5316 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5317 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5318 {
5319 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5320 }
5321 
5322 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5323 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5324 			spdk_blob_op_complete cb_fn, void *cb_arg)
5325 {
5326 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5327 }
5328 
5329 struct spdk_bs_iter_ctx {
5330 	int64_t page_num;
5331 	struct spdk_blob_store *bs;
5332 
5333 	spdk_blob_op_with_handle_complete cb_fn;
5334 	void *cb_arg;
5335 };
5336 
5337 static void
5338 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5339 {
5340 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5341 	struct spdk_blob_store *bs = ctx->bs;
5342 	spdk_blob_id id;
5343 
5344 	if (bserrno == 0) {
5345 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5346 		free(ctx);
5347 		return;
5348 	}
5349 
5350 	ctx->page_num++;
5351 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5352 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5353 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5354 		free(ctx);
5355 		return;
5356 	}
5357 
5358 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5359 
5360 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5361 }
5362 
5363 void
5364 spdk_bs_iter_first(struct spdk_blob_store *bs,
5365 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5366 {
5367 	struct spdk_bs_iter_ctx *ctx;
5368 
5369 	ctx = calloc(1, sizeof(*ctx));
5370 	if (!ctx) {
5371 		cb_fn(cb_arg, NULL, -ENOMEM);
5372 		return;
5373 	}
5374 
5375 	ctx->page_num = -1;
5376 	ctx->bs = bs;
5377 	ctx->cb_fn = cb_fn;
5378 	ctx->cb_arg = cb_arg;
5379 
5380 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5381 }
5382 
5383 static void
5384 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5385 {
5386 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5387 
5388 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5389 }
5390 
5391 void
5392 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5393 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5394 {
5395 	struct spdk_bs_iter_ctx *ctx;
5396 
5397 	assert(blob != NULL);
5398 
5399 	ctx = calloc(1, sizeof(*ctx));
5400 	if (!ctx) {
5401 		cb_fn(cb_arg, NULL, -ENOMEM);
5402 		return;
5403 	}
5404 
5405 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5406 	ctx->bs = bs;
5407 	ctx->cb_fn = cb_fn;
5408 	ctx->cb_arg = cb_arg;
5409 
5410 	/* Close the existing blob */
5411 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5412 }
5413 
5414 static int
5415 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5416 		     uint16_t value_len, bool internal)
5417 {
5418 	struct spdk_xattr_tailq *xattrs;
5419 	struct spdk_xattr	*xattr;
5420 
5421 	_spdk_blob_verify_md_op(blob);
5422 
5423 	if (blob->md_ro) {
5424 		return -EPERM;
5425 	}
5426 
5427 	if (internal) {
5428 		xattrs = &blob->xattrs_internal;
5429 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5430 	} else {
5431 		xattrs = &blob->xattrs;
5432 	}
5433 
5434 	TAILQ_FOREACH(xattr, xattrs, link) {
5435 		if (!strcmp(name, xattr->name)) {
5436 			free(xattr->value);
5437 			xattr->value_len = value_len;
5438 			xattr->value = malloc(value_len);
5439 			memcpy(xattr->value, value, value_len);
5440 
5441 			blob->state = SPDK_BLOB_STATE_DIRTY;
5442 
5443 			return 0;
5444 		}
5445 	}
5446 
5447 	xattr = calloc(1, sizeof(*xattr));
5448 	if (!xattr) {
5449 		return -ENOMEM;
5450 	}
5451 	xattr->name = strdup(name);
5452 	xattr->value_len = value_len;
5453 	xattr->value = malloc(value_len);
5454 	memcpy(xattr->value, value, value_len);
5455 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5456 
5457 	blob->state = SPDK_BLOB_STATE_DIRTY;
5458 
5459 	return 0;
5460 }
5461 
5462 int
5463 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5464 		    uint16_t value_len)
5465 {
5466 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5467 }
5468 
5469 static int
5470 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5471 {
5472 	struct spdk_xattr_tailq *xattrs;
5473 	struct spdk_xattr	*xattr;
5474 
5475 	_spdk_blob_verify_md_op(blob);
5476 
5477 	if (blob->md_ro) {
5478 		return -EPERM;
5479 	}
5480 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5481 
5482 	TAILQ_FOREACH(xattr, xattrs, link) {
5483 		if (!strcmp(name, xattr->name)) {
5484 			TAILQ_REMOVE(xattrs, xattr, link);
5485 			free(xattr->value);
5486 			free(xattr->name);
5487 			free(xattr);
5488 
5489 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5490 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5491 			}
5492 			blob->state = SPDK_BLOB_STATE_DIRTY;
5493 
5494 			return 0;
5495 		}
5496 	}
5497 
5498 	return -ENOENT;
5499 }
5500 
5501 int
5502 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5503 {
5504 	return _spdk_blob_remove_xattr(blob, name, false);
5505 }
5506 
5507 static int
5508 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5509 			   const void **value, size_t *value_len, bool internal)
5510 {
5511 	struct spdk_xattr	*xattr;
5512 	struct spdk_xattr_tailq *xattrs;
5513 
5514 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5515 
5516 	TAILQ_FOREACH(xattr, xattrs, link) {
5517 		if (!strcmp(name, xattr->name)) {
5518 			*value = xattr->value;
5519 			*value_len = xattr->value_len;
5520 			return 0;
5521 		}
5522 	}
5523 	return -ENOENT;
5524 }
5525 
5526 int
5527 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5528 			  const void **value, size_t *value_len)
5529 {
5530 	_spdk_blob_verify_md_op(blob);
5531 
5532 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5533 }
5534 
5535 struct spdk_xattr_names {
5536 	uint32_t	count;
5537 	const char	*names[0];
5538 };
5539 
5540 static int
5541 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5542 {
5543 	struct spdk_xattr	*xattr;
5544 	int			count = 0;
5545 
5546 	TAILQ_FOREACH(xattr, xattrs, link) {
5547 		count++;
5548 	}
5549 
5550 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5551 	if (*names == NULL) {
5552 		return -ENOMEM;
5553 	}
5554 
5555 	TAILQ_FOREACH(xattr, xattrs, link) {
5556 		(*names)->names[(*names)->count++] = xattr->name;
5557 	}
5558 
5559 	return 0;
5560 }
5561 
5562 int
5563 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5564 {
5565 	_spdk_blob_verify_md_op(blob);
5566 
5567 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5568 }
5569 
5570 uint32_t
5571 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5572 {
5573 	assert(names != NULL);
5574 
5575 	return names->count;
5576 }
5577 
5578 const char *
5579 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5580 {
5581 	if (index >= names->count) {
5582 		return NULL;
5583 	}
5584 
5585 	return names->names[index];
5586 }
5587 
5588 void
5589 spdk_xattr_names_free(struct spdk_xattr_names *names)
5590 {
5591 	free(names);
5592 }
5593 
5594 struct spdk_bs_type
5595 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5596 {
5597 	return bs->bstype;
5598 }
5599 
5600 void
5601 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5602 {
5603 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5604 }
5605 
5606 bool
5607 spdk_blob_is_read_only(struct spdk_blob *blob)
5608 {
5609 	assert(blob != NULL);
5610 	return (blob->data_ro || blob->md_ro);
5611 }
5612 
5613 bool
5614 spdk_blob_is_snapshot(struct spdk_blob *blob)
5615 {
5616 	struct spdk_blob_list *snapshot_entry;
5617 
5618 	assert(blob != NULL);
5619 
5620 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
5621 		if (snapshot_entry->id == blob->id) {
5622 			break;
5623 		}
5624 	}
5625 
5626 	if (snapshot_entry == NULL) {
5627 		return false;
5628 	}
5629 
5630 	return true;
5631 }
5632 
5633 bool
5634 spdk_blob_is_clone(struct spdk_blob *blob)
5635 {
5636 	assert(blob != NULL);
5637 
5638 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5639 		assert(spdk_blob_is_thin_provisioned(blob));
5640 		return true;
5641 	}
5642 
5643 	return false;
5644 }
5645 
5646 bool
5647 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5648 {
5649 	assert(blob != NULL);
5650 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5651 }
5652 
5653 spdk_blob_id
5654 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5655 {
5656 	struct spdk_blob_list *snapshot_entry = NULL;
5657 	struct spdk_blob_list *clone_entry = NULL;
5658 
5659 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5660 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5661 			if (clone_entry->id == blob_id) {
5662 				return snapshot_entry->id;
5663 			}
5664 		}
5665 	}
5666 
5667 	return SPDK_BLOBID_INVALID;
5668 }
5669 
5670 int
5671 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5672 		     size_t *count)
5673 {
5674 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5675 	size_t n;
5676 
5677 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5678 		if (snapshot_entry->id == blobid) {
5679 			break;
5680 		}
5681 	}
5682 	if (snapshot_entry == NULL) {
5683 		*count = 0;
5684 		return 0;
5685 	}
5686 
5687 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5688 		*count = snapshot_entry->clone_count;
5689 		return -ENOMEM;
5690 	}
5691 	*count = snapshot_entry->clone_count;
5692 
5693 	n = 0;
5694 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5695 		ids[n++] = clone_entry->id;
5696 	}
5697 
5698 	return 0;
5699 }
5700 
5701 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5702