xref: /spdk/lib/blob/blobstore.c (revision f6c5e40aed498e983c84a75122315a35a74c8ac1)
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 		iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff);
2027 		iov->iov_base = orig_iov->iov_base + orig_iovoff;
2028 		byte_count -= iov->iov_len;
2029 		orig_iovoff = 0;
2030 		orig_iov++;
2031 		iov++;
2032 		iovcnt++;
2033 	}
2034 
2035 	ctx->page_offset += page_count;
2036 	ctx->pages_done += page_count;
2037 	ctx->pages_remaining -= page_count;
2038 	iov = &ctx->iov[0];
2039 
2040 	if (ctx->read) {
2041 		spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, page_offset,
2042 				   page_count, _spdk_rw_iov_split_next, ctx);
2043 	} else {
2044 		spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, page_offset,
2045 				    page_count, _spdk_rw_iov_split_next, ctx);
2046 	}
2047 }
2048 
2049 static void
2050 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel,
2051 				 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
2052 				 spdk_blob_op_complete cb_fn, void *cb_arg, bool read)
2053 {
2054 	struct spdk_bs_cpl	cpl;
2055 
2056 	assert(blob != NULL);
2057 
2058 	if (!read && blob->data_ro) {
2059 		cb_fn(cb_arg, -EPERM);
2060 		return;
2061 	}
2062 
2063 	if (length == 0) {
2064 		cb_fn(cb_arg, 0);
2065 		return;
2066 	}
2067 
2068 	if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) {
2069 		cb_fn(cb_arg, -EINVAL);
2070 		return;
2071 	}
2072 
2073 	/*
2074 	 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having
2075 	 *  to split a request that spans a cluster boundary.  For I/O that do not span a cluster boundary,
2076 	 *  there will be no noticeable difference compared to using a batch.  For I/O that do span a cluster
2077 	 *  boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need
2078 	 *  to allocate a separate iov array and split the I/O such that none of the resulting
2079 	 *  smaller I/O cross a cluster boundary.  These smaller I/O will be issued in sequence (not in parallel)
2080 	 *  but since this case happens very infrequently, any performance impact will be negligible.
2081 	 *
2082 	 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs
2083 	 *  for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them
2084 	 *  in a batch.  That would also require creating an intermediate spdk_bs_cpl that would get called
2085 	 *  when the batch was completed, to allow for freeing the memory for the iov arrays.
2086 	 */
2087 	if (spdk_likely(length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset))) {
2088 		uint32_t lba_count;
2089 		uint64_t lba;
2090 
2091 		_spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count);
2092 
2093 		cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
2094 		cpl.u.blob_basic.cb_fn = cb_fn;
2095 		cpl.u.blob_basic.cb_arg = cb_arg;
2096 		if (blob->frozen_refcnt) {
2097 			/* This blob I/O is frozen */
2098 			spdk_bs_user_op_t *op;
2099 			struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel);
2100 
2101 			op = spdk_bs_user_op_alloc(_channel, &cpl, read, blob, iov, iovcnt, offset, length);
2102 			if (!op) {
2103 				cb_fn(cb_arg, -ENOMEM);
2104 				return;
2105 			}
2106 
2107 			TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link);
2108 
2109 			return;
2110 		}
2111 
2112 		if (read) {
2113 			spdk_bs_sequence_t *seq;
2114 
2115 			seq = spdk_bs_sequence_start(_channel, &cpl);
2116 			if (!seq) {
2117 				cb_fn(cb_arg, -ENOMEM);
2118 				return;
2119 			}
2120 
2121 			if (_spdk_bs_page_is_allocated(blob, offset)) {
2122 				spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2123 			} else {
2124 				spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count,
2125 							      _spdk_rw_iov_done, NULL);
2126 			}
2127 		} else {
2128 			if (_spdk_bs_page_is_allocated(blob, offset)) {
2129 				spdk_bs_sequence_t *seq;
2130 
2131 				seq = spdk_bs_sequence_start(_channel, &cpl);
2132 				if (!seq) {
2133 					cb_fn(cb_arg, -ENOMEM);
2134 					return;
2135 				}
2136 
2137 				spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL);
2138 			} else {
2139 				/* Queue this operation and allocate the cluster */
2140 				spdk_bs_user_op_t *op;
2141 
2142 				op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, length);
2143 				if (!op) {
2144 					cb_fn(cb_arg, -ENOMEM);
2145 					return;
2146 				}
2147 
2148 				_spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op);
2149 			}
2150 		}
2151 	} else {
2152 		struct rw_iov_ctx *ctx;
2153 
2154 		ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec));
2155 		if (ctx == NULL) {
2156 			cb_fn(cb_arg, -ENOMEM);
2157 			return;
2158 		}
2159 
2160 		ctx->blob = blob;
2161 		ctx->channel = _channel;
2162 		ctx->cb_fn = cb_fn;
2163 		ctx->cb_arg = cb_arg;
2164 		ctx->read = read;
2165 		ctx->orig_iov = iov;
2166 		ctx->iovcnt = iovcnt;
2167 		ctx->page_offset = offset;
2168 		ctx->pages_remaining = length;
2169 		ctx->pages_done = 0;
2170 
2171 		_spdk_rw_iov_split_next(ctx, 0);
2172 	}
2173 }
2174 
2175 static struct spdk_blob *
2176 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid)
2177 {
2178 	struct spdk_blob *blob;
2179 
2180 	TAILQ_FOREACH(blob, &bs->blobs, link) {
2181 		if (blob->id == blobid) {
2182 			return blob;
2183 		}
2184 	}
2185 
2186 	return NULL;
2187 }
2188 
2189 static int
2190 _spdk_bs_channel_create(void *io_device, void *ctx_buf)
2191 {
2192 	struct spdk_blob_store		*bs = io_device;
2193 	struct spdk_bs_channel		*channel = ctx_buf;
2194 	struct spdk_bs_dev		*dev;
2195 	uint32_t			max_ops = bs->max_channel_ops;
2196 	uint32_t			i;
2197 
2198 	dev = bs->dev;
2199 
2200 	channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set));
2201 	if (!channel->req_mem) {
2202 		return -1;
2203 	}
2204 
2205 	TAILQ_INIT(&channel->reqs);
2206 
2207 	for (i = 0; i < max_ops; i++) {
2208 		TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link);
2209 	}
2210 
2211 	channel->bs = bs;
2212 	channel->dev = dev;
2213 	channel->dev_channel = dev->create_channel(dev);
2214 
2215 	if (!channel->dev_channel) {
2216 		SPDK_ERRLOG("Failed to create device channel.\n");
2217 		free(channel->req_mem);
2218 		return -1;
2219 	}
2220 
2221 	TAILQ_INIT(&channel->need_cluster_alloc);
2222 	TAILQ_INIT(&channel->queued_io);
2223 
2224 	return 0;
2225 }
2226 
2227 static void
2228 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf)
2229 {
2230 	struct spdk_bs_channel *channel = ctx_buf;
2231 	spdk_bs_user_op_t *op;
2232 
2233 	while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) {
2234 		op = TAILQ_FIRST(&channel->need_cluster_alloc);
2235 		TAILQ_REMOVE(&channel->need_cluster_alloc, op, link);
2236 		spdk_bs_user_op_abort(op);
2237 	}
2238 
2239 	while (!TAILQ_EMPTY(&channel->queued_io)) {
2240 		op = TAILQ_FIRST(&channel->queued_io);
2241 		TAILQ_REMOVE(&channel->queued_io, op, link);
2242 		spdk_bs_user_op_abort(op);
2243 	}
2244 
2245 	free(channel->req_mem);
2246 	channel->dev->destroy_channel(channel->dev, channel->dev_channel);
2247 }
2248 
2249 static void
2250 _spdk_bs_dev_destroy(void *io_device)
2251 {
2252 	struct spdk_blob_store *bs = io_device;
2253 	struct spdk_blob	*blob, *blob_tmp;
2254 
2255 	bs->dev->destroy(bs->dev);
2256 
2257 	TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) {
2258 		TAILQ_REMOVE(&bs->blobs, blob, link);
2259 		_spdk_blob_free(blob);
2260 	}
2261 
2262 	pthread_mutex_destroy(&bs->used_clusters_mutex);
2263 
2264 	spdk_bit_array_free(&bs->used_blobids);
2265 	spdk_bit_array_free(&bs->used_md_pages);
2266 	spdk_bit_array_free(&bs->used_clusters);
2267 	/*
2268 	 * If this function is called for any reason except a successful unload,
2269 	 * the unload_cpl type will be NONE and this will be a nop.
2270 	 */
2271 	spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err);
2272 
2273 	free(bs);
2274 }
2275 
2276 static int
2277 _spdk_bs_blob_list_add(struct spdk_blob *blob)
2278 {
2279 	spdk_blob_id snapshot_id;
2280 	struct spdk_blob_list *snapshot_entry = NULL;
2281 	struct spdk_blob_list *clone_entry = NULL;
2282 
2283 	assert(blob != NULL);
2284 
2285 	snapshot_id = blob->parent_id;
2286 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2287 		return 0;
2288 	}
2289 
2290 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2291 		if (snapshot_entry->id == snapshot_id) {
2292 			break;
2293 		}
2294 	}
2295 
2296 	if (snapshot_entry == NULL) {
2297 		/* Snapshot not found */
2298 		snapshot_entry = calloc(1, sizeof(struct spdk_blob_list));
2299 		if (snapshot_entry == NULL) {
2300 			return -ENOMEM;
2301 		}
2302 		snapshot_entry->id = snapshot_id;
2303 		TAILQ_INIT(&snapshot_entry->clones);
2304 		TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link);
2305 	} else {
2306 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2307 			if (clone_entry->id == blob->id) {
2308 				break;
2309 			}
2310 		}
2311 	}
2312 
2313 	if (clone_entry == NULL) {
2314 		/* Clone not found */
2315 		clone_entry = calloc(1, sizeof(struct spdk_blob_list));
2316 		if (clone_entry == NULL) {
2317 			return -ENOMEM;
2318 		}
2319 		clone_entry->id = blob->id;
2320 		TAILQ_INIT(&clone_entry->clones);
2321 		TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link);
2322 		snapshot_entry->clone_count++;
2323 	}
2324 
2325 	return 0;
2326 }
2327 
2328 static int
2329 _spdk_bs_blob_list_remove(struct spdk_blob *blob)
2330 {
2331 	struct spdk_blob_list *snapshot_entry = NULL;
2332 	struct spdk_blob_list *clone_entry = NULL;
2333 	spdk_blob_id snapshot_id;
2334 
2335 	assert(blob != NULL);
2336 
2337 	snapshot_id = blob->parent_id;
2338 	if (snapshot_id == SPDK_BLOBID_INVALID) {
2339 		return 0;
2340 	}
2341 
2342 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
2343 		if (snapshot_entry->id == snapshot_id) {
2344 			break;
2345 		}
2346 	}
2347 
2348 	assert(snapshot_entry != NULL);
2349 
2350 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
2351 		if (clone_entry->id == blob->id) {
2352 			break;
2353 		}
2354 	}
2355 
2356 	assert(clone_entry != NULL);
2357 
2358 	blob->parent_id = SPDK_BLOBID_INVALID;
2359 	TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2360 	free(clone_entry);
2361 
2362 	snapshot_entry->clone_count--;
2363 	if (snapshot_entry->clone_count == 0) {
2364 		/* Snapshot have no more clones */
2365 		TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link);
2366 		free(snapshot_entry);
2367 	}
2368 
2369 	return 0;
2370 }
2371 
2372 static int
2373 _spdk_bs_blob_list_free(struct spdk_blob_store *bs)
2374 {
2375 	struct spdk_blob_list *snapshot_entry;
2376 	struct spdk_blob_list *snapshot_entry_tmp;
2377 	struct spdk_blob_list *clone_entry;
2378 	struct spdk_blob_list *clone_entry_tmp;
2379 
2380 	TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) {
2381 		TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) {
2382 			TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link);
2383 			free(clone_entry);
2384 		}
2385 		TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link);
2386 		free(snapshot_entry);
2387 	}
2388 
2389 	return 0;
2390 }
2391 
2392 static void
2393 _spdk_bs_free(struct spdk_blob_store *bs)
2394 {
2395 	_spdk_bs_blob_list_free(bs);
2396 
2397 	spdk_bs_unregister_md_thread(bs);
2398 	spdk_io_device_unregister(bs, _spdk_bs_dev_destroy);
2399 }
2400 
2401 void
2402 spdk_bs_opts_init(struct spdk_bs_opts *opts)
2403 {
2404 	opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ;
2405 	opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES;
2406 	opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS;
2407 	opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS;
2408 	memset(&opts->bstype, 0, sizeof(opts->bstype));
2409 	opts->iter_cb_fn = NULL;
2410 	opts->iter_cb_arg = NULL;
2411 }
2412 
2413 static int
2414 _spdk_bs_opts_verify(struct spdk_bs_opts *opts)
2415 {
2416 	if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 ||
2417 	    opts->max_channel_ops == 0) {
2418 		SPDK_ERRLOG("Blobstore options cannot be set to 0\n");
2419 		return -1;
2420 	}
2421 
2422 	return 0;
2423 }
2424 
2425 static int
2426 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs)
2427 {
2428 	struct spdk_blob_store	*bs;
2429 	uint64_t dev_size;
2430 	int rc;
2431 
2432 	dev_size = dev->blocklen * dev->blockcnt;
2433 	if (dev_size < opts->cluster_sz) {
2434 		/* Device size cannot be smaller than cluster size of blobstore */
2435 		SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2436 			     dev_size, opts->cluster_sz);
2437 		return -ENOSPC;
2438 	}
2439 	if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) {
2440 		/* Cluster size cannot be smaller than page size */
2441 		SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n",
2442 			    opts->cluster_sz, SPDK_BS_PAGE_SIZE);
2443 		return -EINVAL;
2444 	}
2445 	bs = calloc(1, sizeof(struct spdk_blob_store));
2446 	if (!bs) {
2447 		return -ENOMEM;
2448 	}
2449 
2450 	TAILQ_INIT(&bs->blobs);
2451 	TAILQ_INIT(&bs->snapshots);
2452 	bs->dev = dev;
2453 	bs->md_thread = spdk_get_thread();
2454 	assert(bs->md_thread != NULL);
2455 
2456 	/*
2457 	 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an
2458 	 *  even multiple of the cluster size.
2459 	 */
2460 	bs->cluster_sz = opts->cluster_sz;
2461 	bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen);
2462 	bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE;
2463 	bs->num_free_clusters = bs->total_clusters;
2464 	bs->used_clusters = spdk_bit_array_create(bs->total_clusters);
2465 	if (bs->used_clusters == NULL) {
2466 		free(bs);
2467 		return -ENOMEM;
2468 	}
2469 
2470 	bs->max_channel_ops = opts->max_channel_ops;
2471 	bs->super_blob = SPDK_BLOBID_INVALID;
2472 	memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype));
2473 
2474 	/* The metadata is assumed to be at least 1 page */
2475 	bs->used_md_pages = spdk_bit_array_create(1);
2476 	bs->used_blobids = spdk_bit_array_create(0);
2477 
2478 	pthread_mutex_init(&bs->used_clusters_mutex, NULL);
2479 
2480 	spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy,
2481 				sizeof(struct spdk_bs_channel));
2482 	rc = spdk_bs_register_md_thread(bs);
2483 	if (rc == -1) {
2484 		spdk_io_device_unregister(bs, NULL);
2485 		pthread_mutex_destroy(&bs->used_clusters_mutex);
2486 		spdk_bit_array_free(&bs->used_blobids);
2487 		spdk_bit_array_free(&bs->used_md_pages);
2488 		spdk_bit_array_free(&bs->used_clusters);
2489 		free(bs);
2490 		/* FIXME: this is a lie but don't know how to get a proper error code here */
2491 		return -ENOMEM;
2492 	}
2493 
2494 	*_bs = bs;
2495 	return 0;
2496 }
2497 
2498 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2499 
2500 struct spdk_bs_load_ctx {
2501 	struct spdk_blob_store		*bs;
2502 	struct spdk_bs_super_block	*super;
2503 
2504 	struct spdk_bs_md_mask		*mask;
2505 	bool				in_page_chain;
2506 	uint32_t			page_index;
2507 	uint32_t			cur_page;
2508 	struct spdk_blob_md_page	*page;
2509 	bool				is_load;
2510 
2511 	spdk_bs_sequence_t			*seq;
2512 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2513 	void					*iter_cb_arg;
2514 };
2515 
2516 static void
2517 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2518 {
2519 	assert(bserrno != 0);
2520 
2521 	spdk_dma_free(ctx->super);
2522 	spdk_bs_sequence_finish(seq, bserrno);
2523 	/*
2524 	 * Only free the blobstore when a load fails.  If an unload fails (for some reason)
2525 	 *  we want to keep the blobstore in case the caller wants to try again.
2526 	 */
2527 	if (ctx->is_load) {
2528 		_spdk_bs_free(ctx->bs);
2529 	}
2530 	free(ctx);
2531 }
2532 
2533 static void
2534 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2535 {
2536 	uint32_t i = 0;
2537 
2538 	while (true) {
2539 		i = spdk_bit_array_find_first_set(array, i);
2540 		if (i >= mask->length) {
2541 			break;
2542 		}
2543 		mask->mask[i / 8] |= 1U << (i % 8);
2544 		i++;
2545 	}
2546 }
2547 
2548 static int
2549 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2550 {
2551 	struct spdk_bit_array *array;
2552 	uint32_t i;
2553 
2554 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2555 		return -ENOMEM;
2556 	}
2557 
2558 	array = *array_ptr;
2559 	for (i = 0; i < mask->length; i++) {
2560 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2561 			spdk_bit_array_set(array, i);
2562 		}
2563 	}
2564 
2565 	return 0;
2566 }
2567 
2568 static void
2569 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2570 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2571 {
2572 	/* Update the values in the super block */
2573 	super->super_blob = bs->super_blob;
2574 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2575 	super->crc = _spdk_blob_md_page_calc_crc(super);
2576 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2577 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2578 				   cb_fn, cb_arg);
2579 }
2580 
2581 static void
2582 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2583 {
2584 	struct spdk_bs_load_ctx	*ctx = arg;
2585 	uint64_t	mask_size, lba, lba_count;
2586 
2587 	/* Write out the used clusters mask */
2588 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2589 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2590 	if (!ctx->mask) {
2591 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2592 		return;
2593 	}
2594 
2595 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2596 	ctx->mask->length = ctx->bs->total_clusters;
2597 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2598 
2599 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2600 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2601 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2602 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2603 }
2604 
2605 static void
2606 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2607 {
2608 	struct spdk_bs_load_ctx	*ctx = arg;
2609 	uint64_t	mask_size, lba, lba_count;
2610 
2611 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2612 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2613 	if (!ctx->mask) {
2614 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2615 		return;
2616 	}
2617 
2618 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2619 	ctx->mask->length = ctx->super->md_len;
2620 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2621 
2622 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2623 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2624 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2625 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2626 }
2627 
2628 static void
2629 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2630 {
2631 	struct spdk_bs_load_ctx	*ctx = arg;
2632 	uint64_t	mask_size, lba, lba_count;
2633 
2634 	if (ctx->super->used_blobid_mask_len == 0) {
2635 		/*
2636 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2637 		 *  written to disk.
2638 		 */
2639 		cb_fn(seq, arg, 0);
2640 		return;
2641 	}
2642 
2643 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2644 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2645 	if (!ctx->mask) {
2646 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2647 		return;
2648 	}
2649 
2650 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2651 	ctx->mask->length = ctx->super->md_len;
2652 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2653 
2654 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2655 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2656 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2657 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2658 }
2659 
2660 static void
2661 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2662 {
2663 	struct spdk_bs_load_ctx *ctx = arg;
2664 
2665 	if (bserrno == 0) {
2666 		if (ctx->iter_cb_fn) {
2667 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2668 		}
2669 		_spdk_bs_blob_list_add(blob);
2670 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2671 		return;
2672 	}
2673 
2674 	if (bserrno == -ENOENT) {
2675 		bserrno = 0;
2676 	} else {
2677 		/*
2678 		 * This case needs to be looked at further.  Same problem
2679 		 *  exists with applications that rely on explicit blob
2680 		 *  iteration.  We should just skip the blob that failed
2681 		 *  to load and coontinue on to the next one.
2682 		 */
2683 		SPDK_ERRLOG("Error in iterating blobs\n");
2684 	}
2685 
2686 	ctx->iter_cb_fn = NULL;
2687 
2688 	spdk_dma_free(ctx->super);
2689 	spdk_dma_free(ctx->mask);
2690 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2691 	free(ctx);
2692 }
2693 
2694 static void
2695 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2696 {
2697 	ctx->seq = seq;
2698 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2699 }
2700 
2701 static void
2702 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2703 {
2704 	struct spdk_bs_load_ctx *ctx = cb_arg;
2705 	int rc;
2706 
2707 	/* The type must be correct */
2708 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2709 
2710 	/* The length of the mask (in bits) must not be greater than
2711 	 * the length of the buffer (converted to bits) */
2712 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2713 
2714 	/* The length of the mask must be exactly equal to the size
2715 	 * (in pages) of the metadata region */
2716 	assert(ctx->mask->length == ctx->super->md_len);
2717 
2718 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2719 	if (rc < 0) {
2720 		spdk_dma_free(ctx->mask);
2721 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2722 		return;
2723 	}
2724 
2725 	_spdk_bs_load_complete(seq, ctx, bserrno);
2726 }
2727 
2728 static void
2729 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2730 {
2731 	struct spdk_bs_load_ctx *ctx = cb_arg;
2732 	uint64_t		lba, lba_count, mask_size;
2733 	int			rc;
2734 
2735 	/* The type must be correct */
2736 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2737 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2738 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2739 					     struct spdk_blob_md_page) * 8));
2740 	/* The length of the mask must be exactly equal to the total number of clusters */
2741 	assert(ctx->mask->length == ctx->bs->total_clusters);
2742 
2743 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2744 	if (rc < 0) {
2745 		spdk_dma_free(ctx->mask);
2746 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2747 		return;
2748 	}
2749 
2750 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2751 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2752 
2753 	spdk_dma_free(ctx->mask);
2754 
2755 	/* Read the used blobids mask */
2756 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2757 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2758 	if (!ctx->mask) {
2759 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2760 		return;
2761 	}
2762 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2763 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2764 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2765 				  _spdk_bs_load_used_blobids_cpl, ctx);
2766 }
2767 
2768 static void
2769 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2770 {
2771 	struct spdk_bs_load_ctx *ctx = cb_arg;
2772 	uint64_t		lba, lba_count, mask_size;
2773 	int			rc;
2774 
2775 	/* The type must be correct */
2776 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2777 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2778 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2779 				     8));
2780 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2781 	assert(ctx->mask->length == ctx->super->md_len);
2782 
2783 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2784 	if (rc < 0) {
2785 		spdk_dma_free(ctx->mask);
2786 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2787 		return;
2788 	}
2789 
2790 	spdk_dma_free(ctx->mask);
2791 
2792 	/* Read the used clusters mask */
2793 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2794 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2795 	if (!ctx->mask) {
2796 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2797 		return;
2798 	}
2799 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2800 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2801 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2802 				  _spdk_bs_load_used_clusters_cpl, ctx);
2803 }
2804 
2805 static void
2806 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2807 {
2808 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2809 	uint64_t lba, lba_count, mask_size;
2810 
2811 	/* Read the used pages mask */
2812 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2813 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2814 	if (!ctx->mask) {
2815 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2816 		return;
2817 	}
2818 
2819 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2820 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2821 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2822 				  _spdk_bs_load_used_pages_cpl, ctx);
2823 }
2824 
2825 static int
2826 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2827 {
2828 	struct spdk_blob_md_descriptor *desc;
2829 	size_t	cur_desc = 0;
2830 
2831 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2832 	while (cur_desc < sizeof(page->descriptors)) {
2833 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2834 			if (desc->length == 0) {
2835 				/* If padding and length are 0, this terminates the page */
2836 				break;
2837 			}
2838 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2839 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2840 			unsigned int				i, j;
2841 			unsigned int				cluster_count = 0;
2842 			uint32_t				cluster_idx;
2843 
2844 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2845 
2846 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2847 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2848 					cluster_idx = desc_extent->extents[i].cluster_idx;
2849 					/*
2850 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2851 					 * in the used cluster map.
2852 					 */
2853 					if (cluster_idx != 0) {
2854 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2855 						if (bs->num_free_clusters == 0) {
2856 							return -ENOSPC;
2857 						}
2858 						bs->num_free_clusters--;
2859 					}
2860 					cluster_count++;
2861 				}
2862 			}
2863 			if (cluster_count == 0) {
2864 				return -EINVAL;
2865 			}
2866 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2867 			/* Skip this item */
2868 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2869 			/* Skip this item */
2870 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2871 			/* Skip this item */
2872 		} else {
2873 			/* Error */
2874 			return -EINVAL;
2875 		}
2876 		/* Advance to the next descriptor */
2877 		cur_desc += sizeof(*desc) + desc->length;
2878 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2879 			break;
2880 		}
2881 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2882 	}
2883 	return 0;
2884 }
2885 
2886 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2887 {
2888 	uint32_t crc;
2889 
2890 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2891 	if (crc != ctx->page->crc) {
2892 		return false;
2893 	}
2894 
2895 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2896 		return false;
2897 	}
2898 	return true;
2899 }
2900 
2901 static void
2902 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2903 
2904 static void
2905 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2906 {
2907 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2908 
2909 	_spdk_bs_load_complete(seq, ctx, bserrno);
2910 }
2911 
2912 static void
2913 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2914 {
2915 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2916 
2917 	spdk_dma_free(ctx->mask);
2918 	ctx->mask = NULL;
2919 
2920 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2921 }
2922 
2923 static void
2924 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2925 {
2926 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2927 
2928 	spdk_dma_free(ctx->mask);
2929 	ctx->mask = NULL;
2930 
2931 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2932 }
2933 
2934 static void
2935 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2936 {
2937 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2938 }
2939 
2940 static void
2941 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2942 {
2943 	struct spdk_bs_load_ctx *ctx = cb_arg;
2944 	uint64_t num_md_clusters;
2945 	uint64_t i;
2946 	uint32_t page_num;
2947 
2948 	if (bserrno != 0) {
2949 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2950 		return;
2951 	}
2952 
2953 	page_num = ctx->cur_page;
2954 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2955 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2956 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2957 			if (ctx->page->sequence_num == 0) {
2958 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2959 			}
2960 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2961 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
2962 				return;
2963 			}
2964 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
2965 				ctx->in_page_chain = true;
2966 				ctx->cur_page = ctx->page->next;
2967 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2968 				return;
2969 			}
2970 		}
2971 	}
2972 
2973 	ctx->in_page_chain = false;
2974 
2975 	do {
2976 		ctx->page_index++;
2977 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
2978 
2979 	if (ctx->page_index < ctx->super->md_len) {
2980 		ctx->cur_page = ctx->page_index;
2981 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2982 	} else {
2983 		/* Claim all of the clusters used by the metadata */
2984 		num_md_clusters = divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
2985 		for (i = 0; i < num_md_clusters; i++) {
2986 			_spdk_bs_claim_cluster(ctx->bs, i);
2987 		}
2988 		spdk_dma_free(ctx->page);
2989 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
2990 	}
2991 }
2992 
2993 static void
2994 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
2995 {
2996 	struct spdk_bs_load_ctx *ctx = cb_arg;
2997 	uint64_t lba;
2998 
2999 	assert(ctx->cur_page < ctx->super->md_len);
3000 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3001 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3002 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3003 				  _spdk_bs_load_replay_md_cpl, ctx);
3004 }
3005 
3006 static void
3007 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3008 {
3009 	struct spdk_bs_load_ctx *ctx = cb_arg;
3010 
3011 	ctx->page_index = 0;
3012 	ctx->cur_page = 0;
3013 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3014 				     SPDK_BS_PAGE_SIZE,
3015 				     NULL);
3016 	if (!ctx->page) {
3017 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3018 		return;
3019 	}
3020 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3021 }
3022 
3023 static void
3024 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3025 {
3026 	struct spdk_bs_load_ctx *ctx = cb_arg;
3027 	int		rc;
3028 
3029 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3030 	if (rc < 0) {
3031 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3032 		return;
3033 	}
3034 
3035 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3036 	if (rc < 0) {
3037 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3038 		return;
3039 	}
3040 
3041 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3042 	if (rc < 0) {
3043 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3044 		return;
3045 	}
3046 
3047 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3048 	_spdk_bs_load_replay_md(seq, cb_arg);
3049 }
3050 
3051 static void
3052 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3053 {
3054 	struct spdk_bs_load_ctx *ctx = cb_arg;
3055 	uint32_t	crc;
3056 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3057 
3058 	if (ctx->super->version > SPDK_BS_VERSION ||
3059 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3060 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3061 		return;
3062 	}
3063 
3064 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3065 		   sizeof(ctx->super->signature)) != 0) {
3066 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3067 		return;
3068 	}
3069 
3070 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3071 	if (crc != ctx->super->crc) {
3072 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3073 		return;
3074 	}
3075 
3076 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3077 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3078 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3079 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3080 	} else {
3081 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3082 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3083 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3084 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3085 		return;
3086 	}
3087 
3088 	if (ctx->super->size == 0) {
3089 		/* Update number of blocks for blobstore */
3090 		ctx->bs->total_clusters = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen / ctx->bs->cluster_sz;
3091 	} else if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3092 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3093 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3094 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3095 		return;
3096 	} else {
3097 		ctx->bs->total_clusters = ctx->super->size / ctx->bs->cluster_sz;
3098 	}
3099 
3100 	/* Parse the super block */
3101 	ctx->bs->clean = 1;
3102 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3103 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3104 	ctx->bs->md_start = ctx->super->md_start;
3105 	ctx->bs->md_len = ctx->super->md_len;
3106 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - divide_round_up(
3107 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3108 	ctx->bs->super_blob = ctx->super->super_blob;
3109 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3110 
3111 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3112 		_spdk_bs_recover(seq, ctx);
3113 	} else {
3114 		_spdk_bs_load_read_used_pages(seq, ctx);
3115 	}
3116 }
3117 
3118 void
3119 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3120 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3121 {
3122 	struct spdk_blob_store	*bs;
3123 	struct spdk_bs_cpl	cpl;
3124 	spdk_bs_sequence_t	*seq;
3125 	struct spdk_bs_load_ctx *ctx;
3126 	struct spdk_bs_opts	opts = {};
3127 	int err;
3128 
3129 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3130 
3131 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3132 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3133 		dev->destroy(dev);
3134 		cb_fn(cb_arg, NULL, -EINVAL);
3135 		return;
3136 	}
3137 
3138 	if (o) {
3139 		opts = *o;
3140 	} else {
3141 		spdk_bs_opts_init(&opts);
3142 	}
3143 
3144 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3145 		dev->destroy(dev);
3146 		cb_fn(cb_arg, NULL, -EINVAL);
3147 		return;
3148 	}
3149 
3150 	err = _spdk_bs_alloc(dev, &opts, &bs);
3151 	if (err) {
3152 		dev->destroy(dev);
3153 		cb_fn(cb_arg, NULL, err);
3154 		return;
3155 	}
3156 
3157 	ctx = calloc(1, sizeof(*ctx));
3158 	if (!ctx) {
3159 		_spdk_bs_free(bs);
3160 		cb_fn(cb_arg, NULL, -ENOMEM);
3161 		return;
3162 	}
3163 
3164 	ctx->bs = bs;
3165 	ctx->is_load = true;
3166 	ctx->iter_cb_fn = opts.iter_cb_fn;
3167 	ctx->iter_cb_arg = opts.iter_cb_arg;
3168 
3169 	/* Allocate memory for the super block */
3170 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3171 	if (!ctx->super) {
3172 		free(ctx);
3173 		_spdk_bs_free(bs);
3174 		cb_fn(cb_arg, NULL, -ENOMEM);
3175 		return;
3176 	}
3177 
3178 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3179 	cpl.u.bs_handle.cb_fn = cb_fn;
3180 	cpl.u.bs_handle.cb_arg = cb_arg;
3181 	cpl.u.bs_handle.bs = bs;
3182 
3183 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3184 	if (!seq) {
3185 		spdk_dma_free(ctx->super);
3186 		free(ctx);
3187 		_spdk_bs_free(bs);
3188 		cb_fn(cb_arg, NULL, -ENOMEM);
3189 		return;
3190 	}
3191 
3192 	/* Read the super block */
3193 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3194 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3195 				  _spdk_bs_load_super_cpl, ctx);
3196 }
3197 
3198 /* END spdk_bs_load */
3199 
3200 /* START spdk_bs_dump */
3201 
3202 struct spdk_bs_dump_ctx {
3203 	struct spdk_blob_store		*bs;
3204 	struct spdk_bs_super_block	*super;
3205 	uint32_t			cur_page;
3206 	struct spdk_blob_md_page	*page;
3207 	spdk_bs_sequence_t		*seq;
3208 	FILE				*fp;
3209 	spdk_bs_dump_print_xattr	print_xattr_fn;
3210 	char				xattr_name[4096];
3211 };
3212 
3213 static void
3214 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3215 {
3216 	spdk_dma_free(ctx->super);
3217 
3218 	/*
3219 	 * We need to defer calling spdk_bs_call_cpl() until after
3220 	 * dev destuction, so tuck these away for later use.
3221 	 */
3222 	ctx->bs->unload_err = bserrno;
3223 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3224 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3225 
3226 	spdk_bs_sequence_finish(seq, 0);
3227 	_spdk_bs_free(ctx->bs);
3228 	free(ctx);
3229 }
3230 
3231 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3232 
3233 static void
3234 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3235 {
3236 	uint32_t page_idx = ctx->cur_page;
3237 	struct spdk_blob_md_page *page = ctx->page;
3238 	struct spdk_blob_md_descriptor *desc;
3239 	size_t cur_desc = 0;
3240 	uint32_t crc;
3241 
3242 	fprintf(ctx->fp, "=========\n");
3243 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3244 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3245 
3246 	crc = _spdk_blob_md_page_calc_crc(page);
3247 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3248 
3249 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3250 	while (cur_desc < sizeof(page->descriptors)) {
3251 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3252 			if (desc->length == 0) {
3253 				/* If padding and length are 0, this terminates the page */
3254 				break;
3255 			}
3256 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3257 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3258 			unsigned int				i;
3259 
3260 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3261 
3262 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3263 				if (desc_extent->extents[i].cluster_idx != 0) {
3264 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3265 						desc_extent->extents[i].cluster_idx);
3266 				} else {
3267 					fprintf(ctx->fp, "Unallocated Extent - ");
3268 				}
3269 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3270 				fprintf(ctx->fp, "\n");
3271 			}
3272 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3273 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3274 			uint32_t i;
3275 
3276 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3277 
3278 			if (desc_xattr->length !=
3279 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3280 			    desc_xattr->name_length + desc_xattr->value_length) {
3281 			}
3282 
3283 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3284 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3285 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3286 			fprintf(ctx->fp, "       value = \"");
3287 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3288 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3289 					    desc_xattr->value_length);
3290 			fprintf(ctx->fp, "\"\n");
3291 			for (i = 0; i < desc_xattr->value_length; i++) {
3292 				if (i % 16 == 0) {
3293 					fprintf(ctx->fp, "               ");
3294 				}
3295 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3296 				if ((i + 1) % 16 == 0) {
3297 					fprintf(ctx->fp, "\n");
3298 				}
3299 			}
3300 			if (i % 16 != 0) {
3301 				fprintf(ctx->fp, "\n");
3302 			}
3303 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3304 			/* TODO */
3305 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3306 			/* TODO */
3307 		} else {
3308 			/* Error */
3309 		}
3310 		/* Advance to the next descriptor */
3311 		cur_desc += sizeof(*desc) + desc->length;
3312 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3313 			break;
3314 		}
3315 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3316 	}
3317 }
3318 
3319 static void
3320 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3321 {
3322 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3323 
3324 	if (bserrno != 0) {
3325 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3326 		return;
3327 	}
3328 
3329 	if (ctx->page->id != 0) {
3330 		_spdk_bs_dump_print_md_page(ctx);
3331 	}
3332 
3333 	ctx->cur_page++;
3334 
3335 	if (ctx->cur_page < ctx->super->md_len) {
3336 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3337 	} else {
3338 		spdk_dma_free(ctx->page);
3339 		_spdk_bs_dump_finish(seq, ctx, 0);
3340 	}
3341 }
3342 
3343 static void
3344 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3345 {
3346 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3347 	uint64_t lba;
3348 
3349 	assert(ctx->cur_page < ctx->super->md_len);
3350 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3351 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3352 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3353 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3354 }
3355 
3356 static void
3357 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3358 {
3359 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3360 
3361 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3362 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3363 		   sizeof(ctx->super->signature)) != 0) {
3364 		fprintf(ctx->fp, "(Mismatch)\n");
3365 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3366 		return;
3367 	} else {
3368 		fprintf(ctx->fp, "(OK)\n");
3369 	}
3370 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3371 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3372 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3373 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3374 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3375 	fprintf(ctx->fp, "Super Blob ID: ");
3376 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3377 		fprintf(ctx->fp, "(None)\n");
3378 	} else {
3379 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3380 	}
3381 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3382 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3383 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3384 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3385 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3386 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3387 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3388 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3389 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3390 
3391 	ctx->cur_page = 0;
3392 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3393 				     SPDK_BS_PAGE_SIZE,
3394 				     NULL);
3395 	if (!ctx->page) {
3396 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3397 		return;
3398 	}
3399 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3400 }
3401 
3402 void
3403 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3404 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3405 {
3406 	struct spdk_blob_store	*bs;
3407 	struct spdk_bs_cpl	cpl;
3408 	spdk_bs_sequence_t	*seq;
3409 	struct spdk_bs_dump_ctx *ctx;
3410 	struct spdk_bs_opts	opts = {};
3411 	int err;
3412 
3413 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3414 
3415 	spdk_bs_opts_init(&opts);
3416 
3417 	err = _spdk_bs_alloc(dev, &opts, &bs);
3418 	if (err) {
3419 		dev->destroy(dev);
3420 		cb_fn(cb_arg, err);
3421 		return;
3422 	}
3423 
3424 	ctx = calloc(1, sizeof(*ctx));
3425 	if (!ctx) {
3426 		_spdk_bs_free(bs);
3427 		cb_fn(cb_arg, -ENOMEM);
3428 		return;
3429 	}
3430 
3431 	ctx->bs = bs;
3432 	ctx->fp = fp;
3433 	ctx->print_xattr_fn = print_xattr_fn;
3434 
3435 	/* Allocate memory for the super block */
3436 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3437 	if (!ctx->super) {
3438 		free(ctx);
3439 		_spdk_bs_free(bs);
3440 		cb_fn(cb_arg, -ENOMEM);
3441 		return;
3442 	}
3443 
3444 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3445 	cpl.u.bs_basic.cb_fn = cb_fn;
3446 	cpl.u.bs_basic.cb_arg = cb_arg;
3447 
3448 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3449 	if (!seq) {
3450 		spdk_dma_free(ctx->super);
3451 		free(ctx);
3452 		_spdk_bs_free(bs);
3453 		cb_fn(cb_arg, -ENOMEM);
3454 		return;
3455 	}
3456 
3457 	/* Read the super block */
3458 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3459 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3460 				  _spdk_bs_dump_super_cpl, ctx);
3461 }
3462 
3463 /* END spdk_bs_dump */
3464 
3465 /* START spdk_bs_init */
3466 
3467 struct spdk_bs_init_ctx {
3468 	struct spdk_blob_store		*bs;
3469 	struct spdk_bs_super_block	*super;
3470 };
3471 
3472 static void
3473 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3474 {
3475 	struct spdk_bs_init_ctx *ctx = cb_arg;
3476 
3477 	spdk_dma_free(ctx->super);
3478 	free(ctx);
3479 
3480 	spdk_bs_sequence_finish(seq, bserrno);
3481 }
3482 
3483 static void
3484 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3485 {
3486 	struct spdk_bs_init_ctx *ctx = cb_arg;
3487 
3488 	/* Write super block */
3489 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3490 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3491 				   _spdk_bs_init_persist_super_cpl, ctx);
3492 }
3493 
3494 void
3495 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3496 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3497 {
3498 	struct spdk_bs_init_ctx *ctx;
3499 	struct spdk_blob_store	*bs;
3500 	struct spdk_bs_cpl	cpl;
3501 	spdk_bs_sequence_t	*seq;
3502 	spdk_bs_batch_t		*batch;
3503 	uint64_t		num_md_lba;
3504 	uint64_t		num_md_pages;
3505 	uint64_t		num_md_clusters;
3506 	uint32_t		i;
3507 	struct spdk_bs_opts	opts = {};
3508 	int			rc;
3509 
3510 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3511 
3512 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3513 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3514 			    dev->blocklen);
3515 		dev->destroy(dev);
3516 		cb_fn(cb_arg, NULL, -EINVAL);
3517 		return;
3518 	}
3519 
3520 	if (o) {
3521 		opts = *o;
3522 	} else {
3523 		spdk_bs_opts_init(&opts);
3524 	}
3525 
3526 	if (_spdk_bs_opts_verify(&opts) != 0) {
3527 		dev->destroy(dev);
3528 		cb_fn(cb_arg, NULL, -EINVAL);
3529 		return;
3530 	}
3531 
3532 	rc = _spdk_bs_alloc(dev, &opts, &bs);
3533 	if (rc) {
3534 		dev->destroy(dev);
3535 		cb_fn(cb_arg, NULL, rc);
3536 		return;
3537 	}
3538 
3539 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3540 		/* By default, allocate 1 page per cluster.
3541 		 * Technically, this over-allocates metadata
3542 		 * because more metadata will reduce the number
3543 		 * of usable clusters. This can be addressed with
3544 		 * more complex math in the future.
3545 		 */
3546 		bs->md_len = bs->total_clusters;
3547 	} else {
3548 		bs->md_len = opts.num_md_pages;
3549 	}
3550 
3551 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3552 	if (rc < 0) {
3553 		_spdk_bs_free(bs);
3554 		cb_fn(cb_arg, NULL, -ENOMEM);
3555 		return;
3556 	}
3557 
3558 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3559 	if (rc < 0) {
3560 		_spdk_bs_free(bs);
3561 		cb_fn(cb_arg, NULL, -ENOMEM);
3562 		return;
3563 	}
3564 
3565 	ctx = calloc(1, sizeof(*ctx));
3566 	if (!ctx) {
3567 		_spdk_bs_free(bs);
3568 		cb_fn(cb_arg, NULL, -ENOMEM);
3569 		return;
3570 	}
3571 
3572 	ctx->bs = bs;
3573 
3574 	/* Allocate memory for the super block */
3575 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3576 	if (!ctx->super) {
3577 		free(ctx);
3578 		_spdk_bs_free(bs);
3579 		cb_fn(cb_arg, NULL, -ENOMEM);
3580 		return;
3581 	}
3582 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3583 	       sizeof(ctx->super->signature));
3584 	ctx->super->version = SPDK_BS_VERSION;
3585 	ctx->super->length = sizeof(*ctx->super);
3586 	ctx->super->super_blob = bs->super_blob;
3587 	ctx->super->clean = 0;
3588 	ctx->super->cluster_size = bs->cluster_sz;
3589 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3590 
3591 	/* Calculate how many pages the metadata consumes at the front
3592 	 * of the disk.
3593 	 */
3594 
3595 	/* The super block uses 1 page */
3596 	num_md_pages = 1;
3597 
3598 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3599 	 * up to the nearest page, plus a header.
3600 	 */
3601 	ctx->super->used_page_mask_start = num_md_pages;
3602 	ctx->super->used_page_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3603 					 divide_round_up(bs->md_len, 8),
3604 					 SPDK_BS_PAGE_SIZE);
3605 	num_md_pages += ctx->super->used_page_mask_len;
3606 
3607 	/* The used_clusters mask requires 1 bit per cluster, rounded
3608 	 * up to the nearest page, plus a header.
3609 	 */
3610 	ctx->super->used_cluster_mask_start = num_md_pages;
3611 	ctx->super->used_cluster_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3612 					    divide_round_up(bs->total_clusters, 8),
3613 					    SPDK_BS_PAGE_SIZE);
3614 	num_md_pages += ctx->super->used_cluster_mask_len;
3615 
3616 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3617 	 * up to the nearest page, plus a header.
3618 	 */
3619 	ctx->super->used_blobid_mask_start = num_md_pages;
3620 	ctx->super->used_blobid_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3621 					   divide_round_up(bs->md_len, 8),
3622 					   SPDK_BS_PAGE_SIZE);
3623 	num_md_pages += ctx->super->used_blobid_mask_len;
3624 
3625 	/* The metadata region size was chosen above */
3626 	ctx->super->md_start = bs->md_start = num_md_pages;
3627 	ctx->super->md_len = bs->md_len;
3628 	num_md_pages += bs->md_len;
3629 
3630 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3631 
3632 	ctx->super->size = dev->blockcnt * dev->blocklen;
3633 
3634 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3635 
3636 	num_md_clusters = divide_round_up(num_md_pages, bs->pages_per_cluster);
3637 	if (num_md_clusters > bs->total_clusters) {
3638 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3639 			    "please decrease number of pages reserved for metadata "
3640 			    "or increase cluster size.\n");
3641 		spdk_dma_free(ctx->super);
3642 		free(ctx);
3643 		_spdk_bs_free(bs);
3644 		cb_fn(cb_arg, NULL, -ENOMEM);
3645 		return;
3646 	}
3647 	/* Claim all of the clusters used by the metadata */
3648 	for (i = 0; i < num_md_clusters; i++) {
3649 		_spdk_bs_claim_cluster(bs, i);
3650 	}
3651 
3652 	bs->total_data_clusters = bs->num_free_clusters;
3653 
3654 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3655 	cpl.u.bs_handle.cb_fn = cb_fn;
3656 	cpl.u.bs_handle.cb_arg = cb_arg;
3657 	cpl.u.bs_handle.bs = bs;
3658 
3659 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3660 	if (!seq) {
3661 		spdk_dma_free(ctx->super);
3662 		free(ctx);
3663 		_spdk_bs_free(bs);
3664 		cb_fn(cb_arg, NULL, -ENOMEM);
3665 		return;
3666 	}
3667 
3668 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3669 
3670 	/* Clear metadata space */
3671 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3672 	/* Trim data clusters */
3673 	spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3674 
3675 	spdk_bs_batch_close(batch);
3676 }
3677 
3678 /* END spdk_bs_init */
3679 
3680 /* START spdk_bs_destroy */
3681 
3682 static void
3683 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3684 {
3685 	struct spdk_bs_init_ctx *ctx = cb_arg;
3686 	struct spdk_blob_store *bs = ctx->bs;
3687 
3688 	/*
3689 	 * We need to defer calling spdk_bs_call_cpl() until after
3690 	 * dev destruction, so tuck these away for later use.
3691 	 */
3692 	bs->unload_err = bserrno;
3693 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3694 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3695 
3696 	spdk_bs_sequence_finish(seq, bserrno);
3697 
3698 	_spdk_bs_free(bs);
3699 	free(ctx);
3700 }
3701 
3702 void
3703 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3704 		void *cb_arg)
3705 {
3706 	struct spdk_bs_cpl	cpl;
3707 	spdk_bs_sequence_t	*seq;
3708 	struct spdk_bs_init_ctx *ctx;
3709 
3710 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3711 
3712 	if (!TAILQ_EMPTY(&bs->blobs)) {
3713 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3714 		cb_fn(cb_arg, -EBUSY);
3715 		return;
3716 	}
3717 
3718 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3719 	cpl.u.bs_basic.cb_fn = cb_fn;
3720 	cpl.u.bs_basic.cb_arg = cb_arg;
3721 
3722 	ctx = calloc(1, sizeof(*ctx));
3723 	if (!ctx) {
3724 		cb_fn(cb_arg, -ENOMEM);
3725 		return;
3726 	}
3727 
3728 	ctx->bs = bs;
3729 
3730 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3731 	if (!seq) {
3732 		free(ctx);
3733 		cb_fn(cb_arg, -ENOMEM);
3734 		return;
3735 	}
3736 
3737 	/* Write zeroes to the super block */
3738 	spdk_bs_sequence_write_zeroes_dev(seq,
3739 					  _spdk_bs_page_to_lba(bs, 0),
3740 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3741 					  _spdk_bs_destroy_trim_cpl, ctx);
3742 }
3743 
3744 /* END spdk_bs_destroy */
3745 
3746 /* START spdk_bs_unload */
3747 
3748 static void
3749 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3750 {
3751 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3752 
3753 	spdk_dma_free(ctx->super);
3754 
3755 	/*
3756 	 * We need to defer calling spdk_bs_call_cpl() until after
3757 	 * dev destuction, so tuck these away for later use.
3758 	 */
3759 	ctx->bs->unload_err = bserrno;
3760 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3761 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3762 
3763 	spdk_bs_sequence_finish(seq, bserrno);
3764 
3765 	_spdk_bs_free(ctx->bs);
3766 	free(ctx);
3767 }
3768 
3769 static void
3770 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3771 {
3772 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3773 
3774 	spdk_dma_free(ctx->mask);
3775 	ctx->super->clean = 1;
3776 
3777 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3778 }
3779 
3780 static void
3781 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3782 {
3783 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3784 
3785 	spdk_dma_free(ctx->mask);
3786 	ctx->mask = NULL;
3787 
3788 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3789 }
3790 
3791 static void
3792 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3793 {
3794 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3795 
3796 	spdk_dma_free(ctx->mask);
3797 	ctx->mask = NULL;
3798 
3799 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3800 }
3801 
3802 static void
3803 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3804 {
3805 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3806 }
3807 
3808 void
3809 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3810 {
3811 	struct spdk_bs_cpl	cpl;
3812 	spdk_bs_sequence_t	*seq;
3813 	struct spdk_bs_load_ctx *ctx;
3814 
3815 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3816 
3817 	if (!TAILQ_EMPTY(&bs->blobs)) {
3818 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3819 		cb_fn(cb_arg, -EBUSY);
3820 		return;
3821 	}
3822 
3823 	ctx = calloc(1, sizeof(*ctx));
3824 	if (!ctx) {
3825 		cb_fn(cb_arg, -ENOMEM);
3826 		return;
3827 	}
3828 
3829 	ctx->bs = bs;
3830 	ctx->is_load = false;
3831 
3832 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3833 	if (!ctx->super) {
3834 		free(ctx);
3835 		cb_fn(cb_arg, -ENOMEM);
3836 		return;
3837 	}
3838 
3839 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3840 	cpl.u.bs_basic.cb_fn = cb_fn;
3841 	cpl.u.bs_basic.cb_arg = cb_arg;
3842 
3843 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3844 	if (!seq) {
3845 		spdk_dma_free(ctx->super);
3846 		free(ctx);
3847 		cb_fn(cb_arg, -ENOMEM);
3848 		return;
3849 	}
3850 
3851 	/* Read super block */
3852 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3853 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3854 				  _spdk_bs_unload_read_super_cpl, ctx);
3855 }
3856 
3857 /* END spdk_bs_unload */
3858 
3859 /* START spdk_bs_set_super */
3860 
3861 struct spdk_bs_set_super_ctx {
3862 	struct spdk_blob_store		*bs;
3863 	struct spdk_bs_super_block	*super;
3864 };
3865 
3866 static void
3867 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3868 {
3869 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3870 
3871 	if (bserrno != 0) {
3872 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3873 	}
3874 
3875 	spdk_dma_free(ctx->super);
3876 
3877 	spdk_bs_sequence_finish(seq, bserrno);
3878 
3879 	free(ctx);
3880 }
3881 
3882 static void
3883 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3884 {
3885 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3886 
3887 	if (bserrno != 0) {
3888 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3889 		spdk_dma_free(ctx->super);
3890 		spdk_bs_sequence_finish(seq, bserrno);
3891 		free(ctx);
3892 		return;
3893 	}
3894 
3895 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3896 }
3897 
3898 void
3899 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3900 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3901 {
3902 	struct spdk_bs_cpl		cpl;
3903 	spdk_bs_sequence_t		*seq;
3904 	struct spdk_bs_set_super_ctx	*ctx;
3905 
3906 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3907 
3908 	ctx = calloc(1, sizeof(*ctx));
3909 	if (!ctx) {
3910 		cb_fn(cb_arg, -ENOMEM);
3911 		return;
3912 	}
3913 
3914 	ctx->bs = bs;
3915 
3916 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3917 	if (!ctx->super) {
3918 		free(ctx);
3919 		cb_fn(cb_arg, -ENOMEM);
3920 		return;
3921 	}
3922 
3923 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3924 	cpl.u.bs_basic.cb_fn = cb_fn;
3925 	cpl.u.bs_basic.cb_arg = cb_arg;
3926 
3927 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3928 	if (!seq) {
3929 		spdk_dma_free(ctx->super);
3930 		free(ctx);
3931 		cb_fn(cb_arg, -ENOMEM);
3932 		return;
3933 	}
3934 
3935 	bs->super_blob = blobid;
3936 
3937 	/* Read super block */
3938 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3939 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3940 				  _spdk_bs_set_super_read_cpl, ctx);
3941 }
3942 
3943 /* END spdk_bs_set_super */
3944 
3945 void
3946 spdk_bs_get_super(struct spdk_blob_store *bs,
3947 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
3948 {
3949 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
3950 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
3951 	} else {
3952 		cb_fn(cb_arg, bs->super_blob, 0);
3953 	}
3954 }
3955 
3956 uint64_t
3957 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
3958 {
3959 	return bs->cluster_sz;
3960 }
3961 
3962 uint64_t
3963 spdk_bs_get_page_size(struct spdk_blob_store *bs)
3964 {
3965 	return SPDK_BS_PAGE_SIZE;
3966 }
3967 
3968 uint64_t
3969 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
3970 {
3971 	return bs->num_free_clusters;
3972 }
3973 
3974 uint64_t
3975 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
3976 {
3977 	return bs->total_data_clusters;
3978 }
3979 
3980 static int
3981 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
3982 {
3983 	bs->md_channel = spdk_get_io_channel(bs);
3984 	if (!bs->md_channel) {
3985 		SPDK_ERRLOG("Failed to get IO channel.\n");
3986 		return -1;
3987 	}
3988 
3989 	return 0;
3990 }
3991 
3992 static int
3993 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
3994 {
3995 	spdk_put_io_channel(bs->md_channel);
3996 
3997 	return 0;
3998 }
3999 
4000 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
4001 {
4002 	assert(blob != NULL);
4003 
4004 	return blob->id;
4005 }
4006 
4007 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4008 {
4009 	assert(blob != NULL);
4010 
4011 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4012 }
4013 
4014 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4015 {
4016 	assert(blob != NULL);
4017 
4018 	return blob->active.num_clusters;
4019 }
4020 
4021 /* START spdk_bs_create_blob */
4022 
4023 static void
4024 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4025 {
4026 	struct spdk_blob *blob = cb_arg;
4027 
4028 	_spdk_blob_free(blob);
4029 
4030 	spdk_bs_sequence_finish(seq, bserrno);
4031 }
4032 
4033 static int
4034 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4035 		      bool internal)
4036 {
4037 	uint64_t i;
4038 	size_t value_len = 0;
4039 	int rc;
4040 	const void *value = NULL;
4041 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4042 		return -EINVAL;
4043 	}
4044 	for (i = 0; i < xattrs->count; i++) {
4045 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4046 		if (value == NULL || value_len == 0) {
4047 			return -EINVAL;
4048 		}
4049 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4050 		if (rc < 0) {
4051 			return rc;
4052 		}
4053 	}
4054 	return 0;
4055 }
4056 
4057 static void
4058 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4059 {
4060 	_spdk_blob_verify_md_op(blob);
4061 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4062 	blob->state = SPDK_BLOB_STATE_DIRTY;
4063 }
4064 
4065 static void
4066 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4067 		     const struct spdk_blob_opts *opts,
4068 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4069 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4070 {
4071 	struct spdk_blob	*blob;
4072 	uint32_t		page_idx;
4073 	struct spdk_bs_cpl	cpl;
4074 	struct spdk_blob_opts	opts_default;
4075 	struct spdk_blob_xattr_opts internal_xattrs_default;
4076 	spdk_bs_sequence_t	*seq;
4077 	spdk_blob_id		id;
4078 	int rc;
4079 
4080 	assert(spdk_get_thread() == bs->md_thread);
4081 
4082 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4083 	if (page_idx >= spdk_bit_array_capacity(bs->used_md_pages)) {
4084 		cb_fn(cb_arg, 0, -ENOMEM);
4085 		return;
4086 	}
4087 	spdk_bit_array_set(bs->used_blobids, page_idx);
4088 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4089 
4090 	id = _spdk_bs_page_to_blobid(page_idx);
4091 
4092 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4093 
4094 	blob = _spdk_blob_alloc(bs, id);
4095 	if (!blob) {
4096 		cb_fn(cb_arg, 0, -ENOMEM);
4097 		return;
4098 	}
4099 
4100 	if (!opts) {
4101 		spdk_blob_opts_init(&opts_default);
4102 		opts = &opts_default;
4103 	}
4104 	if (!internal_xattrs) {
4105 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4106 		internal_xattrs = &internal_xattrs_default;
4107 	}
4108 
4109 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4110 	if (rc < 0) {
4111 		_spdk_blob_free(blob);
4112 		cb_fn(cb_arg, 0, rc);
4113 		return;
4114 	}
4115 
4116 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4117 	if (rc < 0) {
4118 		_spdk_blob_free(blob);
4119 		cb_fn(cb_arg, 0, rc);
4120 		return;
4121 	}
4122 
4123 	if (opts->thin_provision) {
4124 		_spdk_blob_set_thin_provision(blob);
4125 	}
4126 
4127 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4128 	if (rc < 0) {
4129 		_spdk_blob_free(blob);
4130 		cb_fn(cb_arg, 0, rc);
4131 		return;
4132 	}
4133 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4134 	cpl.u.blobid.cb_fn = cb_fn;
4135 	cpl.u.blobid.cb_arg = cb_arg;
4136 	cpl.u.blobid.blobid = blob->id;
4137 
4138 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4139 	if (!seq) {
4140 		_spdk_blob_free(blob);
4141 		cb_fn(cb_arg, 0, -ENOMEM);
4142 		return;
4143 	}
4144 
4145 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4146 }
4147 
4148 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4149 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4150 {
4151 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4152 }
4153 
4154 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4155 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4156 {
4157 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4158 }
4159 
4160 /* END spdk_bs_create_blob */
4161 
4162 /* START blob_cleanup */
4163 
4164 struct spdk_clone_snapshot_ctx {
4165 	struct spdk_bs_cpl      cpl;
4166 	int bserrno;
4167 	bool frozen;
4168 
4169 	struct spdk_io_channel *channel;
4170 
4171 	/* Current cluster for inflate operation */
4172 	uint64_t cluster;
4173 
4174 	/* For inflation force allocation of all unallocated clusters and remove
4175 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4176 	bool allocate_all;
4177 
4178 	struct {
4179 		spdk_blob_id id;
4180 		struct spdk_blob *blob;
4181 	} original;
4182 	struct {
4183 		spdk_blob_id id;
4184 		struct spdk_blob *blob;
4185 	} new;
4186 
4187 	/* xattrs specified for snapshot/clones only. They have no impact on
4188 	 * the original blobs xattrs. */
4189 	const struct spdk_blob_xattr_opts *xattrs;
4190 };
4191 
4192 static void
4193 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4194 {
4195 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4196 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4197 
4198 	if (bserrno != 0) {
4199 		if (ctx->bserrno != 0) {
4200 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4201 		} else {
4202 			ctx->bserrno = bserrno;
4203 		}
4204 	}
4205 
4206 	switch (cpl->type) {
4207 	case SPDK_BS_CPL_TYPE_BLOBID:
4208 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4209 		break;
4210 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4211 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4212 		break;
4213 	default:
4214 		SPDK_UNREACHABLE();
4215 		break;
4216 	}
4217 
4218 	free(ctx);
4219 }
4220 
4221 static void
4222 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4223 {
4224 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4225 	struct spdk_blob *origblob = ctx->original.blob;
4226 
4227 	if (bserrno != 0) {
4228 		if (ctx->bserrno != 0) {
4229 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4230 		} else {
4231 			ctx->bserrno = bserrno;
4232 		}
4233 	}
4234 
4235 	ctx->original.id = origblob->id;
4236 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4237 }
4238 
4239 static void
4240 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4241 {
4242 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4243 	struct spdk_blob *origblob = ctx->original.blob;
4244 
4245 	if (bserrno != 0) {
4246 		if (ctx->bserrno != 0) {
4247 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4248 		} else {
4249 			ctx->bserrno = bserrno;
4250 		}
4251 	}
4252 
4253 	if (ctx->frozen) {
4254 		/* Unfreeze any outstanding I/O */
4255 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4256 	} else {
4257 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4258 	}
4259 
4260 }
4261 
4262 static void
4263 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4264 {
4265 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4266 	struct spdk_blob *newblob = ctx->new.blob;
4267 
4268 	if (bserrno != 0) {
4269 		if (ctx->bserrno != 0) {
4270 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4271 		} else {
4272 			ctx->bserrno = bserrno;
4273 		}
4274 	}
4275 
4276 	ctx->new.id = newblob->id;
4277 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4278 }
4279 
4280 /* END blob_cleanup */
4281 
4282 /* START spdk_bs_create_snapshot */
4283 
4284 static void
4285 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4286 {
4287 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4288 	struct spdk_blob *newblob = ctx->new.blob;
4289 
4290 	if (bserrno != 0) {
4291 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4292 		return;
4293 	}
4294 
4295 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4296 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4297 	if (bserrno != 0) {
4298 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4299 		return;
4300 	}
4301 
4302 	_spdk_bs_blob_list_add(ctx->original.blob);
4303 
4304 	spdk_blob_set_read_only(newblob);
4305 
4306 	/* sync snapshot metadata */
4307 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4308 }
4309 
4310 static void
4311 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4312 {
4313 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4314 	struct spdk_blob *origblob = ctx->original.blob;
4315 	struct spdk_blob *newblob = ctx->new.blob;
4316 
4317 	if (bserrno != 0) {
4318 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4319 		return;
4320 	}
4321 
4322 	/* Set internal xattr for snapshot id */
4323 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4324 	if (bserrno != 0) {
4325 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4326 		return;
4327 	}
4328 
4329 	_spdk_bs_blob_list_remove(origblob);
4330 	origblob->parent_id = newblob->id;
4331 
4332 	/* Create new back_bs_dev for snapshot */
4333 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4334 	if (origblob->back_bs_dev == NULL) {
4335 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4336 		return;
4337 	}
4338 
4339 	/* set clone blob as thin provisioned */
4340 	_spdk_blob_set_thin_provision(origblob);
4341 
4342 	_spdk_bs_blob_list_add(newblob);
4343 
4344 	/* Zero out origblob cluster map */
4345 	memset(origblob->active.clusters, 0,
4346 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4347 
4348 	/* sync clone metadata */
4349 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4350 }
4351 
4352 static void
4353 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4354 {
4355 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4356 	struct spdk_blob *origblob = ctx->original.blob;
4357 	struct spdk_blob *newblob = ctx->new.blob;
4358 	int bserrno;
4359 
4360 	if (rc != 0) {
4361 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4362 		return;
4363 	}
4364 
4365 	ctx->frozen = true;
4366 
4367 	/* set new back_bs_dev for snapshot */
4368 	newblob->back_bs_dev = origblob->back_bs_dev;
4369 	/* Set invalid flags from origblob */
4370 	newblob->invalid_flags = origblob->invalid_flags;
4371 
4372 	/* inherit parent from original blob if set */
4373 	newblob->parent_id = origblob->parent_id;
4374 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4375 		/* Set internal xattr for snapshot id */
4376 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4377 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4378 		if (bserrno != 0) {
4379 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4380 			return;
4381 		}
4382 	}
4383 
4384 	/* Copy cluster map to snapshot */
4385 	memcpy(newblob->active.clusters, origblob->active.clusters,
4386 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4387 
4388 	/* sync snapshot metadata */
4389 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4390 }
4391 
4392 static void
4393 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4394 {
4395 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4396 	struct spdk_blob *origblob = ctx->original.blob;
4397 	struct spdk_blob *newblob = _blob;
4398 
4399 	if (bserrno != 0) {
4400 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4401 		return;
4402 	}
4403 
4404 	ctx->new.blob = newblob;
4405 
4406 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4407 }
4408 
4409 static void
4410 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4411 {
4412 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4413 	struct spdk_blob *origblob = ctx->original.blob;
4414 
4415 	if (bserrno != 0) {
4416 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4417 		return;
4418 	}
4419 
4420 	ctx->new.id = blobid;
4421 	ctx->cpl.u.blobid.blobid = blobid;
4422 
4423 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4424 }
4425 
4426 
4427 static void
4428 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4429 			const void **value, size_t *value_len)
4430 {
4431 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4432 
4433 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4434 	*value = &blob->id;
4435 	*value_len = sizeof(blob->id);
4436 }
4437 
4438 static void
4439 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4440 {
4441 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4442 	struct spdk_blob_opts opts;
4443 	struct spdk_blob_xattr_opts internal_xattrs;
4444 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4445 
4446 	if (bserrno != 0) {
4447 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4448 		return;
4449 	}
4450 
4451 	ctx->original.blob = _blob;
4452 
4453 	if (_blob->data_ro || _blob->md_ro) {
4454 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4455 			      _blob->id);
4456 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4457 		return;
4458 	}
4459 
4460 	spdk_blob_opts_init(&opts);
4461 	_spdk_blob_xattrs_init(&internal_xattrs);
4462 
4463 	/* Change the size of new blob to the same as in original blob,
4464 	 * but do not allocate clusters */
4465 	opts.thin_provision = true;
4466 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4467 
4468 	/* If there are any xattrs specified for snapshot, set them now */
4469 	if (ctx->xattrs) {
4470 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4471 	}
4472 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4473 	internal_xattrs.count = 1;
4474 	internal_xattrs.ctx = _blob;
4475 	internal_xattrs.names = xattrs_names;
4476 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4477 
4478 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4479 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4480 }
4481 
4482 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4483 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4484 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4485 {
4486 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4487 
4488 	if (!ctx) {
4489 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4490 		return;
4491 	}
4492 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4493 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4494 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4495 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4496 	ctx->bserrno = 0;
4497 	ctx->frozen = false;
4498 	ctx->original.id = blobid;
4499 	ctx->xattrs = snapshot_xattrs;
4500 
4501 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4502 }
4503 /* END spdk_bs_create_snapshot */
4504 
4505 /* START spdk_bs_create_clone */
4506 
4507 static void
4508 _spdk_bs_xattr_clone(void *arg, const char *name,
4509 		     const void **value, size_t *value_len)
4510 {
4511 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4512 
4513 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4514 	*value = &blob->id;
4515 	*value_len = sizeof(blob->id);
4516 }
4517 
4518 static void
4519 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4520 {
4521 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4522 	struct spdk_blob *clone = _blob;
4523 
4524 	ctx->new.blob = clone;
4525 	_spdk_bs_blob_list_add(clone);
4526 
4527 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4528 }
4529 
4530 static void
4531 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4532 {
4533 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4534 
4535 	ctx->cpl.u.blobid.blobid = blobid;
4536 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4537 }
4538 
4539 static void
4540 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4541 {
4542 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4543 	struct spdk_blob_opts		opts;
4544 	struct spdk_blob_xattr_opts internal_xattrs;
4545 	char *xattr_names[] = { BLOB_SNAPSHOT };
4546 
4547 	if (bserrno != 0) {
4548 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4549 		return;
4550 	}
4551 
4552 	ctx->original.blob = _blob;
4553 
4554 	if (!_blob->data_ro || !_blob->md_ro) {
4555 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4556 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4557 		return;
4558 	}
4559 
4560 	spdk_blob_opts_init(&opts);
4561 	_spdk_blob_xattrs_init(&internal_xattrs);
4562 
4563 	opts.thin_provision = true;
4564 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4565 	if (ctx->xattrs) {
4566 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4567 	}
4568 
4569 	/* Set internal xattr BLOB_SNAPSHOT */
4570 	internal_xattrs.count = 1;
4571 	internal_xattrs.ctx = _blob;
4572 	internal_xattrs.names = xattr_names;
4573 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4574 
4575 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4576 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4577 }
4578 
4579 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4580 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4581 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4582 {
4583 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4584 
4585 	if (!ctx) {
4586 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4587 		return;
4588 	}
4589 
4590 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4591 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4592 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4593 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4594 	ctx->bserrno = 0;
4595 	ctx->xattrs = clone_xattrs;
4596 	ctx->original.id = blobid;
4597 
4598 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4599 }
4600 
4601 /* END spdk_bs_create_clone */
4602 
4603 /* START spdk_bs_inflate_blob */
4604 
4605 static void
4606 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4607 {
4608 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4609 	struct spdk_blob *_blob = ctx->original.blob;
4610 
4611 	if (bserrno != 0) {
4612 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4613 		return;
4614 	}
4615 
4616 	assert(_parent != NULL);
4617 
4618 	_spdk_bs_blob_list_remove(_blob);
4619 	_blob->parent_id = _parent->id;
4620 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4621 			     sizeof(spdk_blob_id), true);
4622 
4623 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4624 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4625 	_spdk_bs_blob_list_add(_blob);
4626 
4627 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4628 }
4629 
4630 static void
4631 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4632 {
4633 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4634 	struct spdk_blob *_blob = ctx->original.blob;
4635 	struct spdk_blob *_parent;
4636 
4637 	if (bserrno != 0) {
4638 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4639 		return;
4640 	}
4641 
4642 	if (ctx->allocate_all) {
4643 		/* remove thin provisioning */
4644 		_spdk_bs_blob_list_remove(_blob);
4645 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4646 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4647 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4648 		_blob->back_bs_dev = NULL;
4649 		_blob->parent_id = SPDK_BLOBID_INVALID;
4650 	} else {
4651 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4652 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4653 			/* We must change the parent of the inflated blob */
4654 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4655 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4656 			return;
4657 		}
4658 
4659 		_spdk_bs_blob_list_remove(_blob);
4660 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4661 		_blob->parent_id = SPDK_BLOBID_INVALID;
4662 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4663 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4664 	}
4665 
4666 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4667 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4668 }
4669 
4670 /* Check if cluster needs allocation */
4671 static inline bool
4672 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4673 {
4674 	struct spdk_blob_bs_dev *b;
4675 
4676 	assert(blob != NULL);
4677 
4678 	if (blob->active.clusters[cluster] != 0) {
4679 		/* Cluster is already allocated */
4680 		return false;
4681 	}
4682 
4683 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4684 		/* Blob have no parent blob */
4685 		return allocate_all;
4686 	}
4687 
4688 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4689 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4690 }
4691 
4692 static void
4693 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4694 {
4695 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4696 	struct spdk_blob *_blob = ctx->original.blob;
4697 	uint64_t offset;
4698 
4699 	if (bserrno != 0) {
4700 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4701 		return;
4702 	}
4703 
4704 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4705 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4706 			break;
4707 		}
4708 	}
4709 
4710 	if (ctx->cluster < _blob->active.num_clusters) {
4711 		offset = _spdk_bs_cluster_to_page(_blob->bs, ctx->cluster);
4712 
4713 		/* We may safely increment a cluster before write */
4714 		ctx->cluster++;
4715 
4716 		/* Use zero length write to touch a cluster */
4717 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4718 				   _spdk_bs_inflate_blob_touch_next, ctx);
4719 	} else {
4720 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4721 	}
4722 }
4723 
4724 static void
4725 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4726 {
4727 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4728 	uint64_t lfc; /* lowest free cluster */
4729 	uint64_t i;
4730 
4731 	if (bserrno != 0) {
4732 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4733 		return;
4734 	}
4735 	ctx->original.blob = _blob;
4736 
4737 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4738 		/* This blob have no parent, so we cannot decouple it. */
4739 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4740 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4741 		return;
4742 	}
4743 
4744 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4745 		/* This is not thin provisioned blob. No need to inflate. */
4746 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4747 		return;
4748 	}
4749 
4750 	/* Do two passes - one to verify that we can obtain enough clusters
4751 	 * and another to actually claim them.
4752 	 */
4753 	lfc = 0;
4754 	for (i = 0; i < _blob->active.num_clusters; i++) {
4755 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4756 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4757 			if (lfc >= _blob->bs->total_clusters) {
4758 				/* No more free clusters. Cannot satisfy the request */
4759 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4760 				return;
4761 			}
4762 			lfc++;
4763 		}
4764 	}
4765 
4766 	ctx->cluster = 0;
4767 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4768 }
4769 
4770 static void
4771 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4772 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4773 {
4774 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4775 
4776 	if (!ctx) {
4777 		cb_fn(cb_arg, -ENOMEM);
4778 		return;
4779 	}
4780 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4781 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4782 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4783 	ctx->bserrno = 0;
4784 	ctx->original.id = blobid;
4785 	ctx->channel = channel;
4786 	ctx->allocate_all = allocate_all;
4787 
4788 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4789 }
4790 
4791 void
4792 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4793 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4794 {
4795 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4796 }
4797 
4798 void
4799 spdk_bs_blob_decouple_parent(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, false, cb_fn, cb_arg);
4803 }
4804 /* END spdk_bs_inflate_blob */
4805 
4806 /* START spdk_blob_resize */
4807 struct spdk_bs_resize_ctx {
4808 	spdk_blob_op_complete cb_fn;
4809 	void *cb_arg;
4810 	struct spdk_blob *blob;
4811 	uint64_t sz;
4812 	int rc;
4813 };
4814 
4815 static void
4816 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4817 {
4818 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4819 
4820 	if (rc != 0) {
4821 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4822 	}
4823 
4824 	if (ctx->rc != 0) {
4825 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4826 		rc = ctx->rc;
4827 	}
4828 
4829 	ctx->blob->resize_in_progress = false;
4830 
4831 	ctx->cb_fn(ctx->cb_arg, rc);
4832 	free(ctx);
4833 }
4834 
4835 static void
4836 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4837 {
4838 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4839 
4840 	if (rc != 0) {
4841 		ctx->blob->resize_in_progress = false;
4842 		ctx->cb_fn(ctx->cb_arg, rc);
4843 		free(ctx);
4844 		return;
4845 	}
4846 
4847 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4848 
4849 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4850 }
4851 
4852 void
4853 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4854 {
4855 	struct spdk_bs_resize_ctx *ctx;
4856 
4857 	_spdk_blob_verify_md_op(blob);
4858 
4859 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4860 
4861 	if (blob->md_ro) {
4862 		cb_fn(cb_arg, -EPERM);
4863 		return;
4864 	}
4865 
4866 	if (sz == blob->active.num_clusters) {
4867 		cb_fn(cb_arg, 0);
4868 		return;
4869 	}
4870 
4871 	if (blob->resize_in_progress) {
4872 		cb_fn(cb_arg, -EBUSY);
4873 		return;
4874 	}
4875 
4876 	ctx = calloc(1, sizeof(*ctx));
4877 	if (!ctx) {
4878 		cb_fn(cb_arg, -ENOMEM);
4879 		return;
4880 	}
4881 
4882 	blob->resize_in_progress = true;
4883 	ctx->cb_fn = cb_fn;
4884 	ctx->cb_arg = cb_arg;
4885 	ctx->blob = blob;
4886 	ctx->sz = sz;
4887 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
4888 }
4889 
4890 /* END spdk_blob_resize */
4891 
4892 
4893 /* START spdk_bs_delete_blob */
4894 
4895 static void
4896 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
4897 {
4898 	spdk_bs_sequence_t *seq = cb_arg;
4899 
4900 	spdk_bs_sequence_finish(seq, bserrno);
4901 }
4902 
4903 static void
4904 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4905 {
4906 	struct spdk_blob *blob = cb_arg;
4907 
4908 	if (bserrno != 0) {
4909 		/*
4910 		 * We already removed this blob from the blobstore tailq, so
4911 		 *  we need to free it here since this is the last reference
4912 		 *  to it.
4913 		 */
4914 		_spdk_blob_free(blob);
4915 		_spdk_bs_delete_close_cpl(seq, bserrno);
4916 		return;
4917 	}
4918 
4919 	/*
4920 	 * This will immediately decrement the ref_count and call
4921 	 *  the completion routine since the metadata state is clean.
4922 	 *  By calling spdk_blob_close, we reduce the number of call
4923 	 *  points into code that touches the blob->open_ref count
4924 	 *  and the blobstore's blob list.
4925 	 */
4926 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
4927 }
4928 
4929 static void
4930 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
4931 {
4932 	spdk_bs_sequence_t *seq = cb_arg;
4933 	uint32_t page_num;
4934 
4935 	if (bserrno != 0) {
4936 		spdk_bs_sequence_finish(seq, bserrno);
4937 		return;
4938 	}
4939 
4940 	_spdk_blob_verify_md_op(blob);
4941 
4942 	if (blob->open_ref > 1) {
4943 		/*
4944 		 * Someone has this blob open (besides this delete context).
4945 		 *  Decrement the ref count directly and return -EBUSY.
4946 		 */
4947 		blob->open_ref--;
4948 		spdk_bs_sequence_finish(seq, -EBUSY);
4949 		return;
4950 	}
4951 
4952 	bserrno = _spdk_bs_blob_list_remove(blob);
4953 	if (bserrno != 0) {
4954 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Remove blob #%" PRIu64 " from a list\n", blob->id);
4955 		spdk_bs_sequence_finish(seq, bserrno);
4956 		return;
4957 	}
4958 
4959 	/*
4960 	 * Remove the blob from the blob_store list now, to ensure it does not
4961 	 *  get returned after this point by _spdk_blob_lookup().
4962 	 */
4963 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
4964 	page_num = _spdk_bs_blobid_to_page(blob->id);
4965 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
4966 	blob->state = SPDK_BLOB_STATE_DIRTY;
4967 	blob->active.num_pages = 0;
4968 	_spdk_blob_resize(blob, 0);
4969 
4970 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
4971 }
4972 
4973 void
4974 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
4975 		    spdk_blob_op_complete cb_fn, void *cb_arg)
4976 {
4977 	struct spdk_bs_cpl	cpl;
4978 	spdk_bs_sequence_t	*seq;
4979 	struct spdk_blob_list	*snapshot_entry = NULL;
4980 
4981 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
4982 
4983 	assert(spdk_get_thread() == bs->md_thread);
4984 
4985 	/* Check if this is a snapshot with clones */
4986 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
4987 		if (snapshot_entry->id == blobid) {
4988 			break;
4989 		}
4990 	}
4991 	if (snapshot_entry != NULL) {
4992 		/* If snapshot have clones, we cannot remove it */
4993 		if (!TAILQ_EMPTY(&snapshot_entry->clones)) {
4994 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
4995 			cb_fn(cb_arg, -EBUSY);
4996 			return;
4997 		}
4998 	}
4999 
5000 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5001 	cpl.u.blob_basic.cb_fn = cb_fn;
5002 	cpl.u.blob_basic.cb_arg = cb_arg;
5003 
5004 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5005 	if (!seq) {
5006 		cb_fn(cb_arg, -ENOMEM);
5007 		return;
5008 	}
5009 
5010 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5011 }
5012 
5013 /* END spdk_bs_delete_blob */
5014 
5015 /* START spdk_bs_open_blob */
5016 
5017 static void
5018 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5019 {
5020 	struct spdk_blob *blob = cb_arg;
5021 
5022 	/* If the blob have crc error, we just return NULL. */
5023 	if (blob == NULL) {
5024 		seq->cpl.u.blob_handle.blob = NULL;
5025 		spdk_bs_sequence_finish(seq, bserrno);
5026 		return;
5027 	}
5028 
5029 	blob->open_ref++;
5030 
5031 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5032 
5033 	spdk_bs_sequence_finish(seq, bserrno);
5034 }
5035 
5036 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5037 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5038 {
5039 	struct spdk_blob		*blob;
5040 	struct spdk_bs_cpl		cpl;
5041 	spdk_bs_sequence_t		*seq;
5042 	uint32_t			page_num;
5043 
5044 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5045 	assert(spdk_get_thread() == bs->md_thread);
5046 
5047 	page_num = _spdk_bs_blobid_to_page(blobid);
5048 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5049 		/* Invalid blobid */
5050 		cb_fn(cb_arg, NULL, -ENOENT);
5051 		return;
5052 	}
5053 
5054 	blob = _spdk_blob_lookup(bs, blobid);
5055 	if (blob) {
5056 		blob->open_ref++;
5057 		cb_fn(cb_arg, blob, 0);
5058 		return;
5059 	}
5060 
5061 	blob = _spdk_blob_alloc(bs, blobid);
5062 	if (!blob) {
5063 		cb_fn(cb_arg, NULL, -ENOMEM);
5064 		return;
5065 	}
5066 
5067 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5068 	cpl.u.blob_handle.cb_fn = cb_fn;
5069 	cpl.u.blob_handle.cb_arg = cb_arg;
5070 	cpl.u.blob_handle.blob = blob;
5071 
5072 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5073 	if (!seq) {
5074 		_spdk_blob_free(blob);
5075 		cb_fn(cb_arg, NULL, -ENOMEM);
5076 		return;
5077 	}
5078 
5079 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5080 }
5081 /* END spdk_bs_open_blob */
5082 
5083 /* START spdk_blob_set_read_only */
5084 int spdk_blob_set_read_only(struct spdk_blob *blob)
5085 {
5086 	_spdk_blob_verify_md_op(blob);
5087 
5088 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5089 
5090 	blob->state = SPDK_BLOB_STATE_DIRTY;
5091 	return 0;
5092 }
5093 /* END spdk_blob_set_read_only */
5094 
5095 /* START spdk_blob_sync_md */
5096 
5097 static void
5098 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5099 {
5100 	struct spdk_blob *blob = cb_arg;
5101 
5102 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5103 		blob->data_ro = true;
5104 		blob->md_ro = true;
5105 	}
5106 
5107 	spdk_bs_sequence_finish(seq, bserrno);
5108 }
5109 
5110 static void
5111 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5112 {
5113 	struct spdk_bs_cpl	cpl;
5114 	spdk_bs_sequence_t	*seq;
5115 
5116 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5117 	cpl.u.blob_basic.cb_fn = cb_fn;
5118 	cpl.u.blob_basic.cb_arg = cb_arg;
5119 
5120 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5121 	if (!seq) {
5122 		cb_fn(cb_arg, -ENOMEM);
5123 		return;
5124 	}
5125 
5126 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5127 }
5128 
5129 void
5130 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5131 {
5132 	_spdk_blob_verify_md_op(blob);
5133 
5134 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5135 
5136 	if (blob->md_ro) {
5137 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5138 		cb_fn(cb_arg, 0);
5139 		return;
5140 	}
5141 
5142 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5143 }
5144 
5145 /* END spdk_blob_sync_md */
5146 
5147 struct spdk_blob_insert_cluster_ctx {
5148 	struct spdk_thread	*thread;
5149 	struct spdk_blob	*blob;
5150 	uint32_t		cluster_num;	/* cluster index in blob */
5151 	uint32_t		cluster;	/* cluster on disk */
5152 	int			rc;
5153 	spdk_blob_op_complete	cb_fn;
5154 	void			*cb_arg;
5155 };
5156 
5157 static void
5158 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5159 {
5160 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5161 
5162 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5163 	free(ctx);
5164 }
5165 
5166 static void
5167 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5168 {
5169 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5170 
5171 	ctx->rc = bserrno;
5172 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5173 }
5174 
5175 static void
5176 _spdk_blob_insert_cluster_msg(void *arg)
5177 {
5178 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5179 
5180 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5181 	if (ctx->rc != 0) {
5182 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5183 		return;
5184 	}
5185 
5186 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5187 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5188 }
5189 
5190 static void
5191 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5192 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5193 {
5194 	struct spdk_blob_insert_cluster_ctx *ctx;
5195 
5196 	ctx = calloc(1, sizeof(*ctx));
5197 	if (ctx == NULL) {
5198 		cb_fn(cb_arg, -ENOMEM);
5199 		return;
5200 	}
5201 
5202 	ctx->thread = spdk_get_thread();
5203 	ctx->blob = blob;
5204 	ctx->cluster_num = cluster_num;
5205 	ctx->cluster = cluster;
5206 	ctx->cb_fn = cb_fn;
5207 	ctx->cb_arg = cb_arg;
5208 
5209 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5210 }
5211 
5212 /* START spdk_blob_close */
5213 
5214 static void
5215 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5216 {
5217 	struct spdk_blob *blob = cb_arg;
5218 
5219 	if (bserrno == 0) {
5220 		blob->open_ref--;
5221 		if (blob->open_ref == 0) {
5222 			/*
5223 			 * Blobs with active.num_pages == 0 are deleted blobs.
5224 			 *  these blobs are removed from the blob_store list
5225 			 *  when the deletion process starts - so don't try to
5226 			 *  remove them again.
5227 			 */
5228 			if (blob->active.num_pages > 0) {
5229 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5230 			}
5231 			_spdk_blob_free(blob);
5232 		}
5233 	}
5234 
5235 	spdk_bs_sequence_finish(seq, bserrno);
5236 }
5237 
5238 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5239 {
5240 	struct spdk_bs_cpl	cpl;
5241 	spdk_bs_sequence_t	*seq;
5242 
5243 	_spdk_blob_verify_md_op(blob);
5244 
5245 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5246 
5247 	if (blob->open_ref == 0) {
5248 		cb_fn(cb_arg, -EBADF);
5249 		return;
5250 	}
5251 
5252 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5253 	cpl.u.blob_basic.cb_fn = cb_fn;
5254 	cpl.u.blob_basic.cb_arg = cb_arg;
5255 
5256 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5257 	if (!seq) {
5258 		cb_fn(cb_arg, -ENOMEM);
5259 		return;
5260 	}
5261 
5262 	/* Sync metadata */
5263 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5264 }
5265 
5266 /* END spdk_blob_close */
5267 
5268 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5269 {
5270 	return spdk_get_io_channel(bs);
5271 }
5272 
5273 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5274 {
5275 	spdk_put_io_channel(channel);
5276 }
5277 
5278 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5279 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5280 {
5281 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5282 				     SPDK_BLOB_UNMAP);
5283 }
5284 
5285 void spdk_blob_io_write_zeroes(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_WRITE_ZEROES);
5290 }
5291 
5292 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5293 			void *payload, uint64_t offset, uint64_t length,
5294 			spdk_blob_op_complete cb_fn, void *cb_arg)
5295 {
5296 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5297 				     SPDK_BLOB_WRITE);
5298 }
5299 
5300 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5301 		       void *payload, uint64_t offset, uint64_t length,
5302 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5303 {
5304 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5305 				     SPDK_BLOB_READ);
5306 }
5307 
5308 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5309 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5310 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5311 {
5312 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5313 }
5314 
5315 void spdk_blob_io_readv(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, true);
5320 }
5321 
5322 struct spdk_bs_iter_ctx {
5323 	int64_t page_num;
5324 	struct spdk_blob_store *bs;
5325 
5326 	spdk_blob_op_with_handle_complete cb_fn;
5327 	void *cb_arg;
5328 };
5329 
5330 static void
5331 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5332 {
5333 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5334 	struct spdk_blob_store *bs = ctx->bs;
5335 	spdk_blob_id id;
5336 
5337 	if (bserrno == 0) {
5338 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5339 		free(ctx);
5340 		return;
5341 	}
5342 
5343 	ctx->page_num++;
5344 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5345 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5346 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5347 		free(ctx);
5348 		return;
5349 	}
5350 
5351 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5352 
5353 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5354 }
5355 
5356 void
5357 spdk_bs_iter_first(struct spdk_blob_store *bs,
5358 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5359 {
5360 	struct spdk_bs_iter_ctx *ctx;
5361 
5362 	ctx = calloc(1, sizeof(*ctx));
5363 	if (!ctx) {
5364 		cb_fn(cb_arg, NULL, -ENOMEM);
5365 		return;
5366 	}
5367 
5368 	ctx->page_num = -1;
5369 	ctx->bs = bs;
5370 	ctx->cb_fn = cb_fn;
5371 	ctx->cb_arg = cb_arg;
5372 
5373 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5374 }
5375 
5376 static void
5377 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5378 {
5379 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5380 
5381 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5382 }
5383 
5384 void
5385 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5386 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5387 {
5388 	struct spdk_bs_iter_ctx *ctx;
5389 
5390 	assert(blob != NULL);
5391 
5392 	ctx = calloc(1, sizeof(*ctx));
5393 	if (!ctx) {
5394 		cb_fn(cb_arg, NULL, -ENOMEM);
5395 		return;
5396 	}
5397 
5398 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5399 	ctx->bs = bs;
5400 	ctx->cb_fn = cb_fn;
5401 	ctx->cb_arg = cb_arg;
5402 
5403 	/* Close the existing blob */
5404 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5405 }
5406 
5407 static int
5408 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5409 		     uint16_t value_len, bool internal)
5410 {
5411 	struct spdk_xattr_tailq *xattrs;
5412 	struct spdk_xattr	*xattr;
5413 
5414 	_spdk_blob_verify_md_op(blob);
5415 
5416 	if (blob->md_ro) {
5417 		return -EPERM;
5418 	}
5419 
5420 	if (internal) {
5421 		xattrs = &blob->xattrs_internal;
5422 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5423 	} else {
5424 		xattrs = &blob->xattrs;
5425 	}
5426 
5427 	TAILQ_FOREACH(xattr, xattrs, link) {
5428 		if (!strcmp(name, xattr->name)) {
5429 			free(xattr->value);
5430 			xattr->value_len = value_len;
5431 			xattr->value = malloc(value_len);
5432 			memcpy(xattr->value, value, value_len);
5433 
5434 			blob->state = SPDK_BLOB_STATE_DIRTY;
5435 
5436 			return 0;
5437 		}
5438 	}
5439 
5440 	xattr = calloc(1, sizeof(*xattr));
5441 	if (!xattr) {
5442 		return -ENOMEM;
5443 	}
5444 	xattr->name = strdup(name);
5445 	xattr->value_len = value_len;
5446 	xattr->value = malloc(value_len);
5447 	memcpy(xattr->value, value, value_len);
5448 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5449 
5450 	blob->state = SPDK_BLOB_STATE_DIRTY;
5451 
5452 	return 0;
5453 }
5454 
5455 int
5456 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5457 		    uint16_t value_len)
5458 {
5459 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5460 }
5461 
5462 static int
5463 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5464 {
5465 	struct spdk_xattr_tailq *xattrs;
5466 	struct spdk_xattr	*xattr;
5467 
5468 	_spdk_blob_verify_md_op(blob);
5469 
5470 	if (blob->md_ro) {
5471 		return -EPERM;
5472 	}
5473 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5474 
5475 	TAILQ_FOREACH(xattr, xattrs, link) {
5476 		if (!strcmp(name, xattr->name)) {
5477 			TAILQ_REMOVE(xattrs, xattr, link);
5478 			free(xattr->value);
5479 			free(xattr->name);
5480 			free(xattr);
5481 
5482 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5483 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5484 			}
5485 			blob->state = SPDK_BLOB_STATE_DIRTY;
5486 
5487 			return 0;
5488 		}
5489 	}
5490 
5491 	return -ENOENT;
5492 }
5493 
5494 int
5495 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5496 {
5497 	return _spdk_blob_remove_xattr(blob, name, false);
5498 }
5499 
5500 static int
5501 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5502 			   const void **value, size_t *value_len, bool internal)
5503 {
5504 	struct spdk_xattr	*xattr;
5505 	struct spdk_xattr_tailq *xattrs;
5506 
5507 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5508 
5509 	TAILQ_FOREACH(xattr, xattrs, link) {
5510 		if (!strcmp(name, xattr->name)) {
5511 			*value = xattr->value;
5512 			*value_len = xattr->value_len;
5513 			return 0;
5514 		}
5515 	}
5516 	return -ENOENT;
5517 }
5518 
5519 int
5520 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5521 			  const void **value, size_t *value_len)
5522 {
5523 	_spdk_blob_verify_md_op(blob);
5524 
5525 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5526 }
5527 
5528 struct spdk_xattr_names {
5529 	uint32_t	count;
5530 	const char	*names[0];
5531 };
5532 
5533 static int
5534 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5535 {
5536 	struct spdk_xattr	*xattr;
5537 	int			count = 0;
5538 
5539 	TAILQ_FOREACH(xattr, xattrs, link) {
5540 		count++;
5541 	}
5542 
5543 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5544 	if (*names == NULL) {
5545 		return -ENOMEM;
5546 	}
5547 
5548 	TAILQ_FOREACH(xattr, xattrs, link) {
5549 		(*names)->names[(*names)->count++] = xattr->name;
5550 	}
5551 
5552 	return 0;
5553 }
5554 
5555 int
5556 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5557 {
5558 	_spdk_blob_verify_md_op(blob);
5559 
5560 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5561 }
5562 
5563 uint32_t
5564 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5565 {
5566 	assert(names != NULL);
5567 
5568 	return names->count;
5569 }
5570 
5571 const char *
5572 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5573 {
5574 	if (index >= names->count) {
5575 		return NULL;
5576 	}
5577 
5578 	return names->names[index];
5579 }
5580 
5581 void
5582 spdk_xattr_names_free(struct spdk_xattr_names *names)
5583 {
5584 	free(names);
5585 }
5586 
5587 struct spdk_bs_type
5588 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5589 {
5590 	return bs->bstype;
5591 }
5592 
5593 void
5594 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5595 {
5596 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5597 }
5598 
5599 bool
5600 spdk_blob_is_read_only(struct spdk_blob *blob)
5601 {
5602 	assert(blob != NULL);
5603 	return (blob->data_ro || blob->md_ro);
5604 }
5605 
5606 bool
5607 spdk_blob_is_snapshot(struct spdk_blob *blob)
5608 {
5609 	struct spdk_blob_list *snapshot_entry;
5610 
5611 	assert(blob != NULL);
5612 
5613 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
5614 		if (snapshot_entry->id == blob->id) {
5615 			break;
5616 		}
5617 	}
5618 
5619 	if (snapshot_entry == NULL) {
5620 		return false;
5621 	}
5622 
5623 	return true;
5624 }
5625 
5626 bool
5627 spdk_blob_is_clone(struct spdk_blob *blob)
5628 {
5629 	assert(blob != NULL);
5630 
5631 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5632 		assert(spdk_blob_is_thin_provisioned(blob));
5633 		return true;
5634 	}
5635 
5636 	return false;
5637 }
5638 
5639 bool
5640 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5641 {
5642 	assert(blob != NULL);
5643 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5644 }
5645 
5646 spdk_blob_id
5647 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5648 {
5649 	struct spdk_blob_list *snapshot_entry = NULL;
5650 	struct spdk_blob_list *clone_entry = NULL;
5651 
5652 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5653 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5654 			if (clone_entry->id == blob_id) {
5655 				return snapshot_entry->id;
5656 			}
5657 		}
5658 	}
5659 
5660 	return SPDK_BLOBID_INVALID;
5661 }
5662 
5663 int
5664 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5665 		     size_t *count)
5666 {
5667 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5668 	size_t n;
5669 
5670 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5671 		if (snapshot_entry->id == blobid) {
5672 			break;
5673 		}
5674 	}
5675 	if (snapshot_entry == NULL) {
5676 		*count = 0;
5677 		return 0;
5678 	}
5679 
5680 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5681 		*count = snapshot_entry->clone_count;
5682 		return -ENOMEM;
5683 	}
5684 	*count = snapshot_entry->clone_count;
5685 
5686 	n = 0;
5687 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5688 		ids[n++] = clone_entry->id;
5689 	}
5690 
5691 	return 0;
5692 }
5693 
5694 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5695