xref: /spdk/lib/blob/blobstore.c (revision 0de89b36173569630eea4269398179082c030a6f)
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 struct spdk_blob_store *
2426 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts)
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_ERRLOG("Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n",
2436 			    dev_size, opts->cluster_sz);
2437 		return NULL;
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 NULL;
2444 	}
2445 	bs = calloc(1, sizeof(struct spdk_blob_store));
2446 	if (!bs) {
2447 		return NULL;
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 NULL;
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 		return NULL;
2491 	}
2492 
2493 	return bs;
2494 }
2495 
2496 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */
2497 
2498 struct spdk_bs_load_ctx {
2499 	struct spdk_blob_store		*bs;
2500 	struct spdk_bs_super_block	*super;
2501 
2502 	struct spdk_bs_md_mask		*mask;
2503 	bool				in_page_chain;
2504 	uint32_t			page_index;
2505 	uint32_t			cur_page;
2506 	struct spdk_blob_md_page	*page;
2507 	bool				is_load;
2508 
2509 	spdk_bs_sequence_t			*seq;
2510 	spdk_blob_op_with_handle_complete	iter_cb_fn;
2511 	void					*iter_cb_arg;
2512 };
2513 
2514 static void
2515 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2516 {
2517 	assert(bserrno != 0);
2518 
2519 	spdk_dma_free(ctx->super);
2520 	spdk_bs_sequence_finish(seq, bserrno);
2521 	/*
2522 	 * Only free the blobstore when a load fails.  If an unload fails (for some reason)
2523 	 *  we want to keep the blobstore in case the caller wants to try again.
2524 	 */
2525 	if (ctx->is_load) {
2526 		_spdk_bs_free(ctx->bs);
2527 	}
2528 	free(ctx);
2529 }
2530 
2531 static void
2532 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask)
2533 {
2534 	uint32_t i = 0;
2535 
2536 	while (true) {
2537 		i = spdk_bit_array_find_first_set(array, i);
2538 		if (i >= mask->length) {
2539 			break;
2540 		}
2541 		mask->mask[i / 8] |= 1U << (i % 8);
2542 		i++;
2543 	}
2544 }
2545 
2546 static int
2547 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask)
2548 {
2549 	struct spdk_bit_array *array;
2550 	uint32_t i;
2551 
2552 	if (spdk_bit_array_resize(array_ptr, mask->length) < 0) {
2553 		return -ENOMEM;
2554 	}
2555 
2556 	array = *array_ptr;
2557 	for (i = 0; i < mask->length; i++) {
2558 		if (mask->mask[i / 8] & (1U << (i % 8))) {
2559 			spdk_bit_array_set(array, i);
2560 		}
2561 	}
2562 
2563 	return 0;
2564 }
2565 
2566 static void
2567 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs,
2568 		     struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg)
2569 {
2570 	/* Update the values in the super block */
2571 	super->super_blob = bs->super_blob;
2572 	memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype));
2573 	super->crc = _spdk_blob_md_page_calc_crc(super);
2574 	spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0),
2575 				   _spdk_bs_byte_to_lba(bs, sizeof(*super)),
2576 				   cb_fn, cb_arg);
2577 }
2578 
2579 static void
2580 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2581 {
2582 	struct spdk_bs_load_ctx	*ctx = arg;
2583 	uint64_t	mask_size, lba, lba_count;
2584 
2585 	/* Write out the used clusters mask */
2586 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2587 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2588 	if (!ctx->mask) {
2589 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2590 		return;
2591 	}
2592 
2593 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS;
2594 	ctx->mask->length = ctx->bs->total_clusters;
2595 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters));
2596 
2597 	_spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask);
2598 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2599 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2600 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2601 }
2602 
2603 static void
2604 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2605 {
2606 	struct spdk_bs_load_ctx	*ctx = arg;
2607 	uint64_t	mask_size, lba, lba_count;
2608 
2609 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2610 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2611 	if (!ctx->mask) {
2612 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2613 		return;
2614 	}
2615 
2616 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES;
2617 	ctx->mask->length = ctx->super->md_len;
2618 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages));
2619 
2620 	_spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask);
2621 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2622 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2623 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2624 }
2625 
2626 static void
2627 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn)
2628 {
2629 	struct spdk_bs_load_ctx	*ctx = arg;
2630 	uint64_t	mask_size, lba, lba_count;
2631 
2632 	if (ctx->super->used_blobid_mask_len == 0) {
2633 		/*
2634 		 * This is a pre-v3 on-disk format where the blobid mask does not get
2635 		 *  written to disk.
2636 		 */
2637 		cb_fn(seq, arg, 0);
2638 		return;
2639 	}
2640 
2641 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2642 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2643 	if (!ctx->mask) {
2644 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2645 		return;
2646 	}
2647 
2648 	ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS;
2649 	ctx->mask->length = ctx->super->md_len;
2650 	assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids));
2651 
2652 	_spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask);
2653 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2654 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2655 	spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg);
2656 }
2657 
2658 static void
2659 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno)
2660 {
2661 	struct spdk_bs_load_ctx *ctx = arg;
2662 
2663 	if (bserrno == 0) {
2664 		if (ctx->iter_cb_fn) {
2665 			ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0);
2666 		}
2667 		_spdk_bs_blob_list_add(blob);
2668 		spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx);
2669 		return;
2670 	}
2671 
2672 	if (bserrno == -ENOENT) {
2673 		bserrno = 0;
2674 	} else {
2675 		/*
2676 		 * This case needs to be looked at further.  Same problem
2677 		 *  exists with applications that rely on explicit blob
2678 		 *  iteration.  We should just skip the blob that failed
2679 		 *  to load and coontinue on to the next one.
2680 		 */
2681 		SPDK_ERRLOG("Error in iterating blobs\n");
2682 	}
2683 
2684 	ctx->iter_cb_fn = NULL;
2685 
2686 	spdk_dma_free(ctx->super);
2687 	spdk_dma_free(ctx->mask);
2688 	spdk_bs_sequence_finish(ctx->seq, bserrno);
2689 	free(ctx);
2690 }
2691 
2692 static void
2693 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno)
2694 {
2695 	ctx->seq = seq;
2696 	spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx);
2697 }
2698 
2699 static void
2700 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2701 {
2702 	struct spdk_bs_load_ctx *ctx = cb_arg;
2703 	int rc;
2704 
2705 	/* The type must be correct */
2706 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS);
2707 
2708 	/* The length of the mask (in bits) must not be greater than
2709 	 * the length of the buffer (converted to bits) */
2710 	assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8));
2711 
2712 	/* The length of the mask must be exactly equal to the size
2713 	 * (in pages) of the metadata region */
2714 	assert(ctx->mask->length == ctx->super->md_len);
2715 
2716 	rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask);
2717 	if (rc < 0) {
2718 		spdk_dma_free(ctx->mask);
2719 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2720 		return;
2721 	}
2722 
2723 	_spdk_bs_load_complete(seq, ctx, bserrno);
2724 }
2725 
2726 static void
2727 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2728 {
2729 	struct spdk_bs_load_ctx *ctx = cb_arg;
2730 	uint64_t		lba, lba_count, mask_size;
2731 	int			rc;
2732 
2733 	/* The type must be correct */
2734 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS);
2735 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2736 	assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof(
2737 					     struct spdk_blob_md_page) * 8));
2738 	/* The length of the mask must be exactly equal to the total number of clusters */
2739 	assert(ctx->mask->length == ctx->bs->total_clusters);
2740 
2741 	rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask);
2742 	if (rc < 0) {
2743 		spdk_dma_free(ctx->mask);
2744 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2745 		return;
2746 	}
2747 
2748 	ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters);
2749 	assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters);
2750 
2751 	spdk_dma_free(ctx->mask);
2752 
2753 	/* Read the used blobids mask */
2754 	mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE;
2755 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2756 	if (!ctx->mask) {
2757 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2758 		return;
2759 	}
2760 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start);
2761 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len);
2762 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2763 				  _spdk_bs_load_used_blobids_cpl, ctx);
2764 }
2765 
2766 static void
2767 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2768 {
2769 	struct spdk_bs_load_ctx *ctx = cb_arg;
2770 	uint64_t		lba, lba_count, mask_size;
2771 	int			rc;
2772 
2773 	/* The type must be correct */
2774 	assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES);
2775 	/* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */
2776 	assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE *
2777 				     8));
2778 	/* The length of the mask must be exactly equal to the size (in pages) of the metadata region */
2779 	assert(ctx->mask->length == ctx->super->md_len);
2780 
2781 	rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask);
2782 	if (rc < 0) {
2783 		spdk_dma_free(ctx->mask);
2784 		_spdk_bs_load_ctx_fail(seq, ctx, rc);
2785 		return;
2786 	}
2787 
2788 	spdk_dma_free(ctx->mask);
2789 
2790 	/* Read the used clusters mask */
2791 	mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE;
2792 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2793 	if (!ctx->mask) {
2794 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2795 		return;
2796 	}
2797 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start);
2798 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len);
2799 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2800 				  _spdk_bs_load_used_clusters_cpl, ctx);
2801 }
2802 
2803 static void
2804 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg)
2805 {
2806 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2807 	uint64_t lba, lba_count, mask_size;
2808 
2809 	/* Read the used pages mask */
2810 	mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE;
2811 	ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL);
2812 	if (!ctx->mask) {
2813 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
2814 		return;
2815 	}
2816 
2817 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start);
2818 	lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len);
2819 	spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count,
2820 				  _spdk_bs_load_used_pages_cpl, ctx);
2821 }
2822 
2823 static int
2824 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs)
2825 {
2826 	struct spdk_blob_md_descriptor *desc;
2827 	size_t	cur_desc = 0;
2828 
2829 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
2830 	while (cur_desc < sizeof(page->descriptors)) {
2831 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
2832 			if (desc->length == 0) {
2833 				/* If padding and length are 0, this terminates the page */
2834 				break;
2835 			}
2836 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
2837 			struct spdk_blob_md_descriptor_extent	*desc_extent;
2838 			unsigned int				i, j;
2839 			unsigned int				cluster_count = 0;
2840 			uint32_t				cluster_idx;
2841 
2842 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
2843 
2844 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
2845 				for (j = 0; j < desc_extent->extents[i].length; j++) {
2846 					cluster_idx = desc_extent->extents[i].cluster_idx;
2847 					/*
2848 					 * cluster_idx = 0 means an unallocated cluster - don't mark that
2849 					 * in the used cluster map.
2850 					 */
2851 					if (cluster_idx != 0) {
2852 						spdk_bit_array_set(bs->used_clusters, cluster_idx + j);
2853 						if (bs->num_free_clusters == 0) {
2854 							return -ENOSPC;
2855 						}
2856 						bs->num_free_clusters--;
2857 					}
2858 					cluster_count++;
2859 				}
2860 			}
2861 			if (cluster_count == 0) {
2862 				return -EINVAL;
2863 			}
2864 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
2865 			/* Skip this item */
2866 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
2867 			/* Skip this item */
2868 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
2869 			/* Skip this item */
2870 		} else {
2871 			/* Error */
2872 			return -EINVAL;
2873 		}
2874 		/* Advance to the next descriptor */
2875 		cur_desc += sizeof(*desc) + desc->length;
2876 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
2877 			break;
2878 		}
2879 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
2880 	}
2881 	return 0;
2882 }
2883 
2884 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx)
2885 {
2886 	uint32_t crc;
2887 
2888 	crc = _spdk_blob_md_page_calc_crc(ctx->page);
2889 	if (crc != ctx->page->crc) {
2890 		return false;
2891 	}
2892 
2893 	if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) {
2894 		return false;
2895 	}
2896 	return true;
2897 }
2898 
2899 static void
2900 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
2901 
2902 static void
2903 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2904 {
2905 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2906 
2907 	_spdk_bs_load_complete(seq, ctx, bserrno);
2908 }
2909 
2910 static void
2911 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2912 {
2913 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2914 
2915 	spdk_dma_free(ctx->mask);
2916 	ctx->mask = NULL;
2917 
2918 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl);
2919 }
2920 
2921 static void
2922 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2923 {
2924 	struct spdk_bs_load_ctx	*ctx = cb_arg;
2925 
2926 	spdk_dma_free(ctx->mask);
2927 	ctx->mask = NULL;
2928 
2929 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl);
2930 }
2931 
2932 static void
2933 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2934 {
2935 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl);
2936 }
2937 
2938 static void
2939 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
2940 {
2941 	struct spdk_bs_load_ctx *ctx = cb_arg;
2942 	uint64_t num_md_clusters;
2943 	uint64_t i;
2944 	uint32_t page_num;
2945 
2946 	if (bserrno != 0) {
2947 		_spdk_bs_load_ctx_fail(seq, ctx, bserrno);
2948 		return;
2949 	}
2950 
2951 	page_num = ctx->cur_page;
2952 	if (_spdk_bs_load_cur_md_page_valid(ctx) == true) {
2953 		if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) {
2954 			spdk_bit_array_set(ctx->bs->used_md_pages, page_num);
2955 			if (ctx->page->sequence_num == 0) {
2956 				spdk_bit_array_set(ctx->bs->used_blobids, page_num);
2957 			}
2958 			if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) {
2959 				_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
2960 				return;
2961 			}
2962 			if (ctx->page->next != SPDK_INVALID_MD_PAGE) {
2963 				ctx->in_page_chain = true;
2964 				ctx->cur_page = ctx->page->next;
2965 				_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2966 				return;
2967 			}
2968 		}
2969 	}
2970 
2971 	ctx->in_page_chain = false;
2972 
2973 	do {
2974 		ctx->page_index++;
2975 	} while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true);
2976 
2977 	if (ctx->page_index < ctx->super->md_len) {
2978 		ctx->cur_page = ctx->page_index;
2979 		_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
2980 	} else {
2981 		/* Claim all of the clusters used by the metadata */
2982 		num_md_clusters = divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster);
2983 		for (i = 0; i < num_md_clusters; i++) {
2984 			_spdk_bs_claim_cluster(ctx->bs, i);
2985 		}
2986 		spdk_dma_free(ctx->page);
2987 		_spdk_bs_load_write_used_md(seq, ctx, bserrno);
2988 	}
2989 }
2990 
2991 static void
2992 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
2993 {
2994 	struct spdk_bs_load_ctx *ctx = cb_arg;
2995 	uint64_t lba;
2996 
2997 	assert(ctx->cur_page < ctx->super->md_len);
2998 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
2999 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3000 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3001 				  _spdk_bs_load_replay_md_cpl, ctx);
3002 }
3003 
3004 static void
3005 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg)
3006 {
3007 	struct spdk_bs_load_ctx *ctx = cb_arg;
3008 
3009 	ctx->page_index = 0;
3010 	ctx->cur_page = 0;
3011 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3012 				     SPDK_BS_PAGE_SIZE,
3013 				     NULL);
3014 	if (!ctx->page) {
3015 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3016 		return;
3017 	}
3018 	_spdk_bs_load_replay_cur_md_page(seq, cb_arg);
3019 }
3020 
3021 static void
3022 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg)
3023 {
3024 	struct spdk_bs_load_ctx *ctx = cb_arg;
3025 	int		rc;
3026 
3027 	rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len);
3028 	if (rc < 0) {
3029 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3030 		return;
3031 	}
3032 
3033 	rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len);
3034 	if (rc < 0) {
3035 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3036 		return;
3037 	}
3038 
3039 	rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters);
3040 	if (rc < 0) {
3041 		_spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM);
3042 		return;
3043 	}
3044 
3045 	ctx->bs->num_free_clusters = ctx->bs->total_clusters;
3046 	_spdk_bs_load_replay_md(seq, cb_arg);
3047 }
3048 
3049 static void
3050 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3051 {
3052 	struct spdk_bs_load_ctx *ctx = cb_arg;
3053 	uint32_t	crc;
3054 	static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH];
3055 
3056 	if (ctx->super->version > SPDK_BS_VERSION ||
3057 	    ctx->super->version < SPDK_BS_INITIAL_VERSION) {
3058 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3059 		return;
3060 	}
3061 
3062 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3063 		   sizeof(ctx->super->signature)) != 0) {
3064 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3065 		return;
3066 	}
3067 
3068 	crc = _spdk_blob_md_page_calc_crc(ctx->super);
3069 	if (crc != ctx->super->crc) {
3070 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3071 		return;
3072 	}
3073 
3074 	if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3075 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n");
3076 	} else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) {
3077 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n");
3078 	} else {
3079 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n");
3080 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3081 		SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH);
3082 		_spdk_bs_load_ctx_fail(seq, ctx, -ENXIO);
3083 		return;
3084 	}
3085 
3086 	if (ctx->super->size == 0) {
3087 		/* Update number of blocks for blobstore */
3088 		ctx->bs->total_clusters = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen / ctx->bs->cluster_sz;
3089 	} else if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) {
3090 		SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n",
3091 			       ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size);
3092 		_spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ);
3093 		return;
3094 	} else {
3095 		ctx->bs->total_clusters = ctx->super->size / ctx->bs->cluster_sz;
3096 	}
3097 
3098 	/* Parse the super block */
3099 	ctx->bs->clean = 1;
3100 	ctx->bs->cluster_sz = ctx->super->cluster_size;
3101 	ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE;
3102 	ctx->bs->md_start = ctx->super->md_start;
3103 	ctx->bs->md_len = ctx->super->md_len;
3104 	ctx->bs->total_data_clusters = ctx->bs->total_clusters - divide_round_up(
3105 					       ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster);
3106 	ctx->bs->super_blob = ctx->super->super_blob;
3107 	memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype));
3108 
3109 	if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) {
3110 		_spdk_bs_recover(seq, ctx);
3111 	} else {
3112 		_spdk_bs_load_read_used_pages(seq, ctx);
3113 	}
3114 }
3115 
3116 void
3117 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3118 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3119 {
3120 	struct spdk_blob_store	*bs;
3121 	struct spdk_bs_cpl	cpl;
3122 	spdk_bs_sequence_t	*seq;
3123 	struct spdk_bs_load_ctx *ctx;
3124 	struct spdk_bs_opts	opts = {};
3125 
3126 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev);
3127 
3128 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3129 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen);
3130 		dev->destroy(dev);
3131 		cb_fn(cb_arg, NULL, -EINVAL);
3132 		return;
3133 	}
3134 
3135 	if (o) {
3136 		opts = *o;
3137 	} else {
3138 		spdk_bs_opts_init(&opts);
3139 	}
3140 
3141 	if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) {
3142 		dev->destroy(dev);
3143 		cb_fn(cb_arg, NULL, -EINVAL);
3144 		return;
3145 	}
3146 
3147 	bs = _spdk_bs_alloc(dev, &opts);
3148 	if (!bs) {
3149 		dev->destroy(dev);
3150 		cb_fn(cb_arg, NULL, -ENOMEM);
3151 		return;
3152 	}
3153 
3154 	ctx = calloc(1, sizeof(*ctx));
3155 	if (!ctx) {
3156 		_spdk_bs_free(bs);
3157 		cb_fn(cb_arg, NULL, -ENOMEM);
3158 		return;
3159 	}
3160 
3161 	ctx->bs = bs;
3162 	ctx->is_load = true;
3163 	ctx->iter_cb_fn = opts.iter_cb_fn;
3164 	ctx->iter_cb_arg = opts.iter_cb_arg;
3165 
3166 	/* Allocate memory for the super block */
3167 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3168 	if (!ctx->super) {
3169 		free(ctx);
3170 		_spdk_bs_free(bs);
3171 		cb_fn(cb_arg, NULL, -ENOMEM);
3172 		return;
3173 	}
3174 
3175 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3176 	cpl.u.bs_handle.cb_fn = cb_fn;
3177 	cpl.u.bs_handle.cb_arg = cb_arg;
3178 	cpl.u.bs_handle.bs = bs;
3179 
3180 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3181 	if (!seq) {
3182 		spdk_dma_free(ctx->super);
3183 		free(ctx);
3184 		_spdk_bs_free(bs);
3185 		cb_fn(cb_arg, NULL, -ENOMEM);
3186 		return;
3187 	}
3188 
3189 	/* Read the super block */
3190 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3191 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3192 				  _spdk_bs_load_super_cpl, ctx);
3193 }
3194 
3195 /* END spdk_bs_load */
3196 
3197 /* START spdk_bs_dump */
3198 
3199 struct spdk_bs_dump_ctx {
3200 	struct spdk_blob_store		*bs;
3201 	struct spdk_bs_super_block	*super;
3202 	uint32_t			cur_page;
3203 	struct spdk_blob_md_page	*page;
3204 	spdk_bs_sequence_t		*seq;
3205 	FILE				*fp;
3206 	spdk_bs_dump_print_xattr	print_xattr_fn;
3207 	char				xattr_name[4096];
3208 };
3209 
3210 static void
3211 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno)
3212 {
3213 	spdk_dma_free(ctx->super);
3214 
3215 	/*
3216 	 * We need to defer calling spdk_bs_call_cpl() until after
3217 	 * dev destuction, so tuck these away for later use.
3218 	 */
3219 	ctx->bs->unload_err = bserrno;
3220 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3221 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3222 
3223 	spdk_bs_sequence_finish(seq, 0);
3224 	_spdk_bs_free(ctx->bs);
3225 	free(ctx);
3226 }
3227 
3228 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg);
3229 
3230 static void
3231 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx)
3232 {
3233 	uint32_t page_idx = ctx->cur_page;
3234 	struct spdk_blob_md_page *page = ctx->page;
3235 	struct spdk_blob_md_descriptor *desc;
3236 	size_t cur_desc = 0;
3237 	uint32_t crc;
3238 
3239 	fprintf(ctx->fp, "=========\n");
3240 	fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx);
3241 	fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id);
3242 
3243 	crc = _spdk_blob_md_page_calc_crc(page);
3244 	fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch");
3245 
3246 	desc = (struct spdk_blob_md_descriptor *)page->descriptors;
3247 	while (cur_desc < sizeof(page->descriptors)) {
3248 		if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) {
3249 			if (desc->length == 0) {
3250 				/* If padding and length are 0, this terminates the page */
3251 				break;
3252 			}
3253 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) {
3254 			struct spdk_blob_md_descriptor_extent	*desc_extent;
3255 			unsigned int				i;
3256 
3257 			desc_extent = (struct spdk_blob_md_descriptor_extent *)desc;
3258 
3259 			for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) {
3260 				if (desc_extent->extents[i].cluster_idx != 0) {
3261 					fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32,
3262 						desc_extent->extents[i].cluster_idx);
3263 				} else {
3264 					fprintf(ctx->fp, "Unallocated Extent - ");
3265 				}
3266 				fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length);
3267 				fprintf(ctx->fp, "\n");
3268 			}
3269 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) {
3270 			struct spdk_blob_md_descriptor_xattr *desc_xattr;
3271 			uint32_t i;
3272 
3273 			desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc;
3274 
3275 			if (desc_xattr->length !=
3276 			    sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) +
3277 			    desc_xattr->name_length + desc_xattr->value_length) {
3278 			}
3279 
3280 			memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length);
3281 			ctx->xattr_name[desc_xattr->name_length] = '\0';
3282 			fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name);
3283 			fprintf(ctx->fp, "       value = \"");
3284 			ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name,
3285 					    (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length),
3286 					    desc_xattr->value_length);
3287 			fprintf(ctx->fp, "\"\n");
3288 			for (i = 0; i < desc_xattr->value_length; i++) {
3289 				if (i % 16 == 0) {
3290 					fprintf(ctx->fp, "               ");
3291 				}
3292 				fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i));
3293 				if ((i + 1) % 16 == 0) {
3294 					fprintf(ctx->fp, "\n");
3295 				}
3296 			}
3297 			if (i % 16 != 0) {
3298 				fprintf(ctx->fp, "\n");
3299 			}
3300 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) {
3301 			/* TODO */
3302 		} else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) {
3303 			/* TODO */
3304 		} else {
3305 			/* Error */
3306 		}
3307 		/* Advance to the next descriptor */
3308 		cur_desc += sizeof(*desc) + desc->length;
3309 		if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) {
3310 			break;
3311 		}
3312 		desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc);
3313 	}
3314 }
3315 
3316 static void
3317 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3318 {
3319 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3320 
3321 	if (bserrno != 0) {
3322 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3323 		return;
3324 	}
3325 
3326 	if (ctx->page->id != 0) {
3327 		_spdk_bs_dump_print_md_page(ctx);
3328 	}
3329 
3330 	ctx->cur_page++;
3331 
3332 	if (ctx->cur_page < ctx->super->md_len) {
3333 		_spdk_bs_dump_read_md_page(seq, cb_arg);
3334 	} else {
3335 		spdk_dma_free(ctx->page);
3336 		_spdk_bs_dump_finish(seq, ctx, 0);
3337 	}
3338 }
3339 
3340 static void
3341 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg)
3342 {
3343 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3344 	uint64_t lba;
3345 
3346 	assert(ctx->cur_page < ctx->super->md_len);
3347 	lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page);
3348 	spdk_bs_sequence_read_dev(seq, ctx->page, lba,
3349 				  _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE),
3350 				  _spdk_bs_dump_read_md_page_cpl, ctx);
3351 }
3352 
3353 static void
3354 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3355 {
3356 	struct spdk_bs_dump_ctx *ctx = cb_arg;
3357 
3358 	fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature);
3359 	if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3360 		   sizeof(ctx->super->signature)) != 0) {
3361 		fprintf(ctx->fp, "(Mismatch)\n");
3362 		_spdk_bs_dump_finish(seq, ctx, bserrno);
3363 		return;
3364 	} else {
3365 		fprintf(ctx->fp, "(OK)\n");
3366 	}
3367 	fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version);
3368 	fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc,
3369 		(ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch");
3370 	fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype);
3371 	fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size);
3372 	fprintf(ctx->fp, "Super Blob ID: ");
3373 	if (ctx->super->super_blob == SPDK_BLOBID_INVALID) {
3374 		fprintf(ctx->fp, "(None)\n");
3375 	} else {
3376 		fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob);
3377 	}
3378 	fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean);
3379 	fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start);
3380 	fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len);
3381 	fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start);
3382 	fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len);
3383 	fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start);
3384 	fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len);
3385 	fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start);
3386 	fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len);
3387 
3388 	ctx->cur_page = 0;
3389 	ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE,
3390 				     SPDK_BS_PAGE_SIZE,
3391 				     NULL);
3392 	if (!ctx->page) {
3393 		_spdk_bs_dump_finish(seq, ctx, -ENOMEM);
3394 		return;
3395 	}
3396 	_spdk_bs_dump_read_md_page(seq, cb_arg);
3397 }
3398 
3399 void
3400 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn,
3401 	     spdk_bs_op_complete cb_fn, void *cb_arg)
3402 {
3403 	struct spdk_blob_store	*bs;
3404 	struct spdk_bs_cpl	cpl;
3405 	spdk_bs_sequence_t	*seq;
3406 	struct spdk_bs_dump_ctx *ctx;
3407 	struct spdk_bs_opts	opts = {};
3408 
3409 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev);
3410 
3411 	spdk_bs_opts_init(&opts);
3412 
3413 	bs = _spdk_bs_alloc(dev, &opts);
3414 	if (!bs) {
3415 		dev->destroy(dev);
3416 		cb_fn(cb_arg, -ENOMEM);
3417 		return;
3418 	}
3419 
3420 	ctx = calloc(1, sizeof(*ctx));
3421 	if (!ctx) {
3422 		_spdk_bs_free(bs);
3423 		cb_fn(cb_arg, -ENOMEM);
3424 		return;
3425 	}
3426 
3427 	ctx->bs = bs;
3428 	ctx->fp = fp;
3429 	ctx->print_xattr_fn = print_xattr_fn;
3430 
3431 	/* Allocate memory for the super block */
3432 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3433 	if (!ctx->super) {
3434 		free(ctx);
3435 		_spdk_bs_free(bs);
3436 		cb_fn(cb_arg, -ENOMEM);
3437 		return;
3438 	}
3439 
3440 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3441 	cpl.u.bs_basic.cb_fn = cb_fn;
3442 	cpl.u.bs_basic.cb_arg = cb_arg;
3443 
3444 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3445 	if (!seq) {
3446 		spdk_dma_free(ctx->super);
3447 		free(ctx);
3448 		_spdk_bs_free(bs);
3449 		cb_fn(cb_arg, -ENOMEM);
3450 		return;
3451 	}
3452 
3453 	/* Read the super block */
3454 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3455 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3456 				  _spdk_bs_dump_super_cpl, ctx);
3457 }
3458 
3459 /* END spdk_bs_dump */
3460 
3461 /* START spdk_bs_init */
3462 
3463 struct spdk_bs_init_ctx {
3464 	struct spdk_blob_store		*bs;
3465 	struct spdk_bs_super_block	*super;
3466 };
3467 
3468 static void
3469 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3470 {
3471 	struct spdk_bs_init_ctx *ctx = cb_arg;
3472 
3473 	spdk_dma_free(ctx->super);
3474 	free(ctx);
3475 
3476 	spdk_bs_sequence_finish(seq, bserrno);
3477 }
3478 
3479 static void
3480 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3481 {
3482 	struct spdk_bs_init_ctx *ctx = cb_arg;
3483 
3484 	/* Write super block */
3485 	spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0),
3486 				   _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)),
3487 				   _spdk_bs_init_persist_super_cpl, ctx);
3488 }
3489 
3490 void
3491 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o,
3492 	     spdk_bs_op_with_handle_complete cb_fn, void *cb_arg)
3493 {
3494 	struct spdk_bs_init_ctx *ctx;
3495 	struct spdk_blob_store	*bs;
3496 	struct spdk_bs_cpl	cpl;
3497 	spdk_bs_sequence_t	*seq;
3498 	spdk_bs_batch_t		*batch;
3499 	uint64_t		num_md_lba;
3500 	uint64_t		num_md_pages;
3501 	uint64_t		num_md_clusters;
3502 	uint32_t		i;
3503 	struct spdk_bs_opts	opts = {};
3504 	int			rc;
3505 
3506 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev);
3507 
3508 	if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) {
3509 		SPDK_ERRLOG("unsupported dev block length of %d\n",
3510 			    dev->blocklen);
3511 		dev->destroy(dev);
3512 		cb_fn(cb_arg, NULL, -EINVAL);
3513 		return;
3514 	}
3515 
3516 	if (o) {
3517 		opts = *o;
3518 	} else {
3519 		spdk_bs_opts_init(&opts);
3520 	}
3521 
3522 	if (_spdk_bs_opts_verify(&opts) != 0) {
3523 		dev->destroy(dev);
3524 		cb_fn(cb_arg, NULL, -EINVAL);
3525 		return;
3526 	}
3527 
3528 	bs = _spdk_bs_alloc(dev, &opts);
3529 	if (!bs) {
3530 		dev->destroy(dev);
3531 		cb_fn(cb_arg, NULL, -ENOMEM);
3532 		return;
3533 	}
3534 
3535 	if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) {
3536 		/* By default, allocate 1 page per cluster.
3537 		 * Technically, this over-allocates metadata
3538 		 * because more metadata will reduce the number
3539 		 * of usable clusters. This can be addressed with
3540 		 * more complex math in the future.
3541 		 */
3542 		bs->md_len = bs->total_clusters;
3543 	} else {
3544 		bs->md_len = opts.num_md_pages;
3545 	}
3546 
3547 	rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len);
3548 	if (rc < 0) {
3549 		_spdk_bs_free(bs);
3550 		cb_fn(cb_arg, NULL, -ENOMEM);
3551 		return;
3552 	}
3553 
3554 	rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len);
3555 	if (rc < 0) {
3556 		_spdk_bs_free(bs);
3557 		cb_fn(cb_arg, NULL, -ENOMEM);
3558 		return;
3559 	}
3560 
3561 	ctx = calloc(1, sizeof(*ctx));
3562 	if (!ctx) {
3563 		_spdk_bs_free(bs);
3564 		cb_fn(cb_arg, NULL, -ENOMEM);
3565 		return;
3566 	}
3567 
3568 	ctx->bs = bs;
3569 
3570 	/* Allocate memory for the super block */
3571 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3572 	if (!ctx->super) {
3573 		free(ctx);
3574 		_spdk_bs_free(bs);
3575 		cb_fn(cb_arg, NULL, -ENOMEM);
3576 		return;
3577 	}
3578 	memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG,
3579 	       sizeof(ctx->super->signature));
3580 	ctx->super->version = SPDK_BS_VERSION;
3581 	ctx->super->length = sizeof(*ctx->super);
3582 	ctx->super->super_blob = bs->super_blob;
3583 	ctx->super->clean = 0;
3584 	ctx->super->cluster_size = bs->cluster_sz;
3585 	memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype));
3586 
3587 	/* Calculate how many pages the metadata consumes at the front
3588 	 * of the disk.
3589 	 */
3590 
3591 	/* The super block uses 1 page */
3592 	num_md_pages = 1;
3593 
3594 	/* The used_md_pages mask requires 1 bit per metadata page, rounded
3595 	 * up to the nearest page, plus a header.
3596 	 */
3597 	ctx->super->used_page_mask_start = num_md_pages;
3598 	ctx->super->used_page_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3599 					 divide_round_up(bs->md_len, 8),
3600 					 SPDK_BS_PAGE_SIZE);
3601 	num_md_pages += ctx->super->used_page_mask_len;
3602 
3603 	/* The used_clusters mask requires 1 bit per cluster, rounded
3604 	 * up to the nearest page, plus a header.
3605 	 */
3606 	ctx->super->used_cluster_mask_start = num_md_pages;
3607 	ctx->super->used_cluster_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3608 					    divide_round_up(bs->total_clusters, 8),
3609 					    SPDK_BS_PAGE_SIZE);
3610 	num_md_pages += ctx->super->used_cluster_mask_len;
3611 
3612 	/* The used_blobids mask requires 1 bit per metadata page, rounded
3613 	 * up to the nearest page, plus a header.
3614 	 */
3615 	ctx->super->used_blobid_mask_start = num_md_pages;
3616 	ctx->super->used_blobid_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) +
3617 					   divide_round_up(bs->md_len, 8),
3618 					   SPDK_BS_PAGE_SIZE);
3619 	num_md_pages += ctx->super->used_blobid_mask_len;
3620 
3621 	/* The metadata region size was chosen above */
3622 	ctx->super->md_start = bs->md_start = num_md_pages;
3623 	ctx->super->md_len = bs->md_len;
3624 	num_md_pages += bs->md_len;
3625 
3626 	num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages);
3627 
3628 	ctx->super->size = dev->blockcnt * dev->blocklen;
3629 
3630 	ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super);
3631 
3632 	num_md_clusters = divide_round_up(num_md_pages, bs->pages_per_cluster);
3633 	if (num_md_clusters > bs->total_clusters) {
3634 		SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, "
3635 			    "please decrease number of pages reserved for metadata "
3636 			    "or increase cluster size.\n");
3637 		spdk_dma_free(ctx->super);
3638 		free(ctx);
3639 		_spdk_bs_free(bs);
3640 		cb_fn(cb_arg, NULL, -ENOMEM);
3641 		return;
3642 	}
3643 	/* Claim all of the clusters used by the metadata */
3644 	for (i = 0; i < num_md_clusters; i++) {
3645 		_spdk_bs_claim_cluster(bs, i);
3646 	}
3647 
3648 	bs->total_data_clusters = bs->num_free_clusters;
3649 
3650 	cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE;
3651 	cpl.u.bs_handle.cb_fn = cb_fn;
3652 	cpl.u.bs_handle.cb_arg = cb_arg;
3653 	cpl.u.bs_handle.bs = bs;
3654 
3655 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3656 	if (!seq) {
3657 		spdk_dma_free(ctx->super);
3658 		free(ctx);
3659 		_spdk_bs_free(bs);
3660 		cb_fn(cb_arg, NULL, -ENOMEM);
3661 		return;
3662 	}
3663 
3664 	batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx);
3665 
3666 	/* Clear metadata space */
3667 	spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba);
3668 	/* Trim data clusters */
3669 	spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba);
3670 
3671 	spdk_bs_batch_close(batch);
3672 }
3673 
3674 /* END spdk_bs_init */
3675 
3676 /* START spdk_bs_destroy */
3677 
3678 static void
3679 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3680 {
3681 	struct spdk_bs_init_ctx *ctx = cb_arg;
3682 	struct spdk_blob_store *bs = ctx->bs;
3683 
3684 	/*
3685 	 * We need to defer calling spdk_bs_call_cpl() until after
3686 	 * dev destruction, so tuck these away for later use.
3687 	 */
3688 	bs->unload_err = bserrno;
3689 	memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3690 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3691 
3692 	spdk_bs_sequence_finish(seq, bserrno);
3693 
3694 	_spdk_bs_free(bs);
3695 	free(ctx);
3696 }
3697 
3698 void
3699 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn,
3700 		void *cb_arg)
3701 {
3702 	struct spdk_bs_cpl	cpl;
3703 	spdk_bs_sequence_t	*seq;
3704 	struct spdk_bs_init_ctx *ctx;
3705 
3706 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n");
3707 
3708 	if (!TAILQ_EMPTY(&bs->blobs)) {
3709 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3710 		cb_fn(cb_arg, -EBUSY);
3711 		return;
3712 	}
3713 
3714 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3715 	cpl.u.bs_basic.cb_fn = cb_fn;
3716 	cpl.u.bs_basic.cb_arg = cb_arg;
3717 
3718 	ctx = calloc(1, sizeof(*ctx));
3719 	if (!ctx) {
3720 		cb_fn(cb_arg, -ENOMEM);
3721 		return;
3722 	}
3723 
3724 	ctx->bs = bs;
3725 
3726 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3727 	if (!seq) {
3728 		free(ctx);
3729 		cb_fn(cb_arg, -ENOMEM);
3730 		return;
3731 	}
3732 
3733 	/* Write zeroes to the super block */
3734 	spdk_bs_sequence_write_zeroes_dev(seq,
3735 					  _spdk_bs_page_to_lba(bs, 0),
3736 					  _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)),
3737 					  _spdk_bs_destroy_trim_cpl, ctx);
3738 }
3739 
3740 /* END spdk_bs_destroy */
3741 
3742 /* START spdk_bs_unload */
3743 
3744 static void
3745 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3746 {
3747 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3748 
3749 	spdk_dma_free(ctx->super);
3750 
3751 	/*
3752 	 * We need to defer calling spdk_bs_call_cpl() until after
3753 	 * dev destuction, so tuck these away for later use.
3754 	 */
3755 	ctx->bs->unload_err = bserrno;
3756 	memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl));
3757 	seq->cpl.type = SPDK_BS_CPL_TYPE_NONE;
3758 
3759 	spdk_bs_sequence_finish(seq, bserrno);
3760 
3761 	_spdk_bs_free(ctx->bs);
3762 	free(ctx);
3763 }
3764 
3765 static void
3766 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3767 {
3768 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3769 
3770 	spdk_dma_free(ctx->mask);
3771 	ctx->super->clean = 1;
3772 
3773 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx);
3774 }
3775 
3776 static void
3777 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3778 {
3779 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3780 
3781 	spdk_dma_free(ctx->mask);
3782 	ctx->mask = NULL;
3783 
3784 	_spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl);
3785 }
3786 
3787 static void
3788 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3789 {
3790 	struct spdk_bs_load_ctx	*ctx = cb_arg;
3791 
3792 	spdk_dma_free(ctx->mask);
3793 	ctx->mask = NULL;
3794 
3795 	_spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl);
3796 }
3797 
3798 static void
3799 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3800 {
3801 	_spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl);
3802 }
3803 
3804 void
3805 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg)
3806 {
3807 	struct spdk_bs_cpl	cpl;
3808 	spdk_bs_sequence_t	*seq;
3809 	struct spdk_bs_load_ctx *ctx;
3810 
3811 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n");
3812 
3813 	if (!TAILQ_EMPTY(&bs->blobs)) {
3814 		SPDK_ERRLOG("Blobstore still has open blobs\n");
3815 		cb_fn(cb_arg, -EBUSY);
3816 		return;
3817 	}
3818 
3819 	ctx = calloc(1, sizeof(*ctx));
3820 	if (!ctx) {
3821 		cb_fn(cb_arg, -ENOMEM);
3822 		return;
3823 	}
3824 
3825 	ctx->bs = bs;
3826 	ctx->is_load = false;
3827 
3828 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3829 	if (!ctx->super) {
3830 		free(ctx);
3831 		cb_fn(cb_arg, -ENOMEM);
3832 		return;
3833 	}
3834 
3835 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3836 	cpl.u.bs_basic.cb_fn = cb_fn;
3837 	cpl.u.bs_basic.cb_arg = cb_arg;
3838 
3839 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3840 	if (!seq) {
3841 		spdk_dma_free(ctx->super);
3842 		free(ctx);
3843 		cb_fn(cb_arg, -ENOMEM);
3844 		return;
3845 	}
3846 
3847 	/* Read super block */
3848 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3849 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3850 				  _spdk_bs_unload_read_super_cpl, ctx);
3851 }
3852 
3853 /* END spdk_bs_unload */
3854 
3855 /* START spdk_bs_set_super */
3856 
3857 struct spdk_bs_set_super_ctx {
3858 	struct spdk_blob_store		*bs;
3859 	struct spdk_bs_super_block	*super;
3860 };
3861 
3862 static void
3863 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3864 {
3865 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3866 
3867 	if (bserrno != 0) {
3868 		SPDK_ERRLOG("Unable to write to super block of blobstore\n");
3869 	}
3870 
3871 	spdk_dma_free(ctx->super);
3872 
3873 	spdk_bs_sequence_finish(seq, bserrno);
3874 
3875 	free(ctx);
3876 }
3877 
3878 static void
3879 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
3880 {
3881 	struct spdk_bs_set_super_ctx	*ctx = cb_arg;
3882 
3883 	if (bserrno != 0) {
3884 		SPDK_ERRLOG("Unable to read super block of blobstore\n");
3885 		spdk_dma_free(ctx->super);
3886 		spdk_bs_sequence_finish(seq, bserrno);
3887 		free(ctx);
3888 		return;
3889 	}
3890 
3891 	_spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx);
3892 }
3893 
3894 void
3895 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid,
3896 		  spdk_bs_op_complete cb_fn, void *cb_arg)
3897 {
3898 	struct spdk_bs_cpl		cpl;
3899 	spdk_bs_sequence_t		*seq;
3900 	struct spdk_bs_set_super_ctx	*ctx;
3901 
3902 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n");
3903 
3904 	ctx = calloc(1, sizeof(*ctx));
3905 	if (!ctx) {
3906 		cb_fn(cb_arg, -ENOMEM);
3907 		return;
3908 	}
3909 
3910 	ctx->bs = bs;
3911 
3912 	ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL);
3913 	if (!ctx->super) {
3914 		free(ctx);
3915 		cb_fn(cb_arg, -ENOMEM);
3916 		return;
3917 	}
3918 
3919 	cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC;
3920 	cpl.u.bs_basic.cb_fn = cb_fn;
3921 	cpl.u.bs_basic.cb_arg = cb_arg;
3922 
3923 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
3924 	if (!seq) {
3925 		spdk_dma_free(ctx->super);
3926 		free(ctx);
3927 		cb_fn(cb_arg, -ENOMEM);
3928 		return;
3929 	}
3930 
3931 	bs->super_blob = blobid;
3932 
3933 	/* Read super block */
3934 	spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0),
3935 				  _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)),
3936 				  _spdk_bs_set_super_read_cpl, ctx);
3937 }
3938 
3939 /* END spdk_bs_set_super */
3940 
3941 void
3942 spdk_bs_get_super(struct spdk_blob_store *bs,
3943 		  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
3944 {
3945 	if (bs->super_blob == SPDK_BLOBID_INVALID) {
3946 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT);
3947 	} else {
3948 		cb_fn(cb_arg, bs->super_blob, 0);
3949 	}
3950 }
3951 
3952 uint64_t
3953 spdk_bs_get_cluster_size(struct spdk_blob_store *bs)
3954 {
3955 	return bs->cluster_sz;
3956 }
3957 
3958 uint64_t
3959 spdk_bs_get_page_size(struct spdk_blob_store *bs)
3960 {
3961 	return SPDK_BS_PAGE_SIZE;
3962 }
3963 
3964 uint64_t
3965 spdk_bs_free_cluster_count(struct spdk_blob_store *bs)
3966 {
3967 	return bs->num_free_clusters;
3968 }
3969 
3970 uint64_t
3971 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs)
3972 {
3973 	return bs->total_data_clusters;
3974 }
3975 
3976 static int
3977 spdk_bs_register_md_thread(struct spdk_blob_store *bs)
3978 {
3979 	bs->md_channel = spdk_get_io_channel(bs);
3980 	if (!bs->md_channel) {
3981 		SPDK_ERRLOG("Failed to get IO channel.\n");
3982 		return -1;
3983 	}
3984 
3985 	return 0;
3986 }
3987 
3988 static int
3989 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs)
3990 {
3991 	spdk_put_io_channel(bs->md_channel);
3992 
3993 	return 0;
3994 }
3995 
3996 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob)
3997 {
3998 	assert(blob != NULL);
3999 
4000 	return blob->id;
4001 }
4002 
4003 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob)
4004 {
4005 	assert(blob != NULL);
4006 
4007 	return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters);
4008 }
4009 
4010 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob)
4011 {
4012 	assert(blob != NULL);
4013 
4014 	return blob->active.num_clusters;
4015 }
4016 
4017 /* START spdk_bs_create_blob */
4018 
4019 static void
4020 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4021 {
4022 	struct spdk_blob *blob = cb_arg;
4023 
4024 	_spdk_blob_free(blob);
4025 
4026 	spdk_bs_sequence_finish(seq, bserrno);
4027 }
4028 
4029 static int
4030 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs,
4031 		      bool internal)
4032 {
4033 	uint64_t i;
4034 	size_t value_len = 0;
4035 	int rc;
4036 	const void *value = NULL;
4037 	if (xattrs->count > 0 && xattrs->get_value == NULL) {
4038 		return -EINVAL;
4039 	}
4040 	for (i = 0; i < xattrs->count; i++) {
4041 		xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len);
4042 		if (value == NULL || value_len == 0) {
4043 			return -EINVAL;
4044 		}
4045 		rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal);
4046 		if (rc < 0) {
4047 			return rc;
4048 		}
4049 	}
4050 	return 0;
4051 }
4052 
4053 static void
4054 _spdk_blob_set_thin_provision(struct spdk_blob *blob)
4055 {
4056 	_spdk_blob_verify_md_op(blob);
4057 	blob->invalid_flags |= SPDK_BLOB_THIN_PROV;
4058 	blob->state = SPDK_BLOB_STATE_DIRTY;
4059 }
4060 
4061 static void
4062 _spdk_bs_create_blob(struct spdk_blob_store *bs,
4063 		     const struct spdk_blob_opts *opts,
4064 		     const struct spdk_blob_xattr_opts *internal_xattrs,
4065 		     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4066 {
4067 	struct spdk_blob	*blob;
4068 	uint32_t		page_idx;
4069 	struct spdk_bs_cpl	cpl;
4070 	struct spdk_blob_opts	opts_default;
4071 	struct spdk_blob_xattr_opts internal_xattrs_default;
4072 	spdk_bs_sequence_t	*seq;
4073 	spdk_blob_id		id;
4074 	int rc;
4075 
4076 	assert(spdk_get_thread() == bs->md_thread);
4077 
4078 	page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0);
4079 	if (page_idx >= spdk_bit_array_capacity(bs->used_md_pages)) {
4080 		cb_fn(cb_arg, 0, -ENOMEM);
4081 		return;
4082 	}
4083 	spdk_bit_array_set(bs->used_blobids, page_idx);
4084 	spdk_bit_array_set(bs->used_md_pages, page_idx);
4085 
4086 	id = _spdk_bs_page_to_blobid(page_idx);
4087 
4088 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx);
4089 
4090 	blob = _spdk_blob_alloc(bs, id);
4091 	if (!blob) {
4092 		cb_fn(cb_arg, 0, -ENOMEM);
4093 		return;
4094 	}
4095 
4096 	if (!opts) {
4097 		spdk_blob_opts_init(&opts_default);
4098 		opts = &opts_default;
4099 	}
4100 	if (!internal_xattrs) {
4101 		_spdk_blob_xattrs_init(&internal_xattrs_default);
4102 		internal_xattrs = &internal_xattrs_default;
4103 	}
4104 
4105 	rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false);
4106 	if (rc < 0) {
4107 		_spdk_blob_free(blob);
4108 		cb_fn(cb_arg, 0, rc);
4109 		return;
4110 	}
4111 
4112 	rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true);
4113 	if (rc < 0) {
4114 		_spdk_blob_free(blob);
4115 		cb_fn(cb_arg, 0, rc);
4116 		return;
4117 	}
4118 
4119 	if (opts->thin_provision) {
4120 		_spdk_blob_set_thin_provision(blob);
4121 	}
4122 
4123 	rc = _spdk_blob_resize(blob, opts->num_clusters);
4124 	if (rc < 0) {
4125 		_spdk_blob_free(blob);
4126 		cb_fn(cb_arg, 0, rc);
4127 		return;
4128 	}
4129 	cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4130 	cpl.u.blobid.cb_fn = cb_fn;
4131 	cpl.u.blobid.cb_arg = cb_arg;
4132 	cpl.u.blobid.blobid = blob->id;
4133 
4134 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
4135 	if (!seq) {
4136 		_spdk_blob_free(blob);
4137 		cb_fn(cb_arg, 0, -ENOMEM);
4138 		return;
4139 	}
4140 
4141 	_spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob);
4142 }
4143 
4144 void spdk_bs_create_blob(struct spdk_blob_store *bs,
4145 			 spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4146 {
4147 	_spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg);
4148 }
4149 
4150 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts,
4151 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4152 {
4153 	_spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg);
4154 }
4155 
4156 /* END spdk_bs_create_blob */
4157 
4158 /* START blob_cleanup */
4159 
4160 struct spdk_clone_snapshot_ctx {
4161 	struct spdk_bs_cpl      cpl;
4162 	int bserrno;
4163 	bool frozen;
4164 
4165 	struct spdk_io_channel *channel;
4166 
4167 	/* Current cluster for inflate operation */
4168 	uint64_t cluster;
4169 
4170 	/* For inflation force allocation of all unallocated clusters and remove
4171 	 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */
4172 	bool allocate_all;
4173 
4174 	struct {
4175 		spdk_blob_id id;
4176 		struct spdk_blob *blob;
4177 	} original;
4178 	struct {
4179 		spdk_blob_id id;
4180 		struct spdk_blob *blob;
4181 	} new;
4182 
4183 	/* xattrs specified for snapshot/clones only. They have no impact on
4184 	 * the original blobs xattrs. */
4185 	const struct spdk_blob_xattr_opts *xattrs;
4186 };
4187 
4188 static void
4189 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno)
4190 {
4191 	struct spdk_clone_snapshot_ctx *ctx = cb_arg;
4192 	struct spdk_bs_cpl *cpl = &ctx->cpl;
4193 
4194 	if (bserrno != 0) {
4195 		if (ctx->bserrno != 0) {
4196 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4197 		} else {
4198 			ctx->bserrno = bserrno;
4199 		}
4200 	}
4201 
4202 	switch (cpl->type) {
4203 	case SPDK_BS_CPL_TYPE_BLOBID:
4204 		cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno);
4205 		break;
4206 	case SPDK_BS_CPL_TYPE_BLOB_BASIC:
4207 		cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno);
4208 		break;
4209 	default:
4210 		SPDK_UNREACHABLE();
4211 		break;
4212 	}
4213 
4214 	free(ctx);
4215 }
4216 
4217 static void
4218 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno)
4219 {
4220 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4221 	struct spdk_blob *origblob = ctx->original.blob;
4222 
4223 	if (bserrno != 0) {
4224 		if (ctx->bserrno != 0) {
4225 			SPDK_ERRLOG("Unfreeze error %d\n", bserrno);
4226 		} else {
4227 			ctx->bserrno = bserrno;
4228 		}
4229 	}
4230 
4231 	ctx->original.id = origblob->id;
4232 	spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx);
4233 }
4234 
4235 static void
4236 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno)
4237 {
4238 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4239 	struct spdk_blob *origblob = ctx->original.blob;
4240 
4241 	if (bserrno != 0) {
4242 		if (ctx->bserrno != 0) {
4243 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4244 		} else {
4245 			ctx->bserrno = bserrno;
4246 		}
4247 	}
4248 
4249 	if (ctx->frozen) {
4250 		/* Unfreeze any outstanding I/O */
4251 		_spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx);
4252 	} else {
4253 		_spdk_bs_snapshot_unfreeze_cpl(ctx, 0);
4254 	}
4255 
4256 }
4257 
4258 static void
4259 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno)
4260 {
4261 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4262 	struct spdk_blob *newblob = ctx->new.blob;
4263 
4264 	if (bserrno != 0) {
4265 		if (ctx->bserrno != 0) {
4266 			SPDK_ERRLOG("Cleanup error %d\n", bserrno);
4267 		} else {
4268 			ctx->bserrno = bserrno;
4269 		}
4270 	}
4271 
4272 	ctx->new.id = newblob->id;
4273 	spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4274 }
4275 
4276 /* END blob_cleanup */
4277 
4278 /* START spdk_bs_create_snapshot */
4279 
4280 static void
4281 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno)
4282 {
4283 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4284 	struct spdk_blob *newblob = ctx->new.blob;
4285 
4286 	if (bserrno != 0) {
4287 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4288 		return;
4289 	}
4290 
4291 	/* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */
4292 	bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true);
4293 	if (bserrno != 0) {
4294 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4295 		return;
4296 	}
4297 
4298 	_spdk_bs_blob_list_add(ctx->original.blob);
4299 
4300 	spdk_blob_set_read_only(newblob);
4301 
4302 	/* sync snapshot metadata */
4303 	spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg);
4304 }
4305 
4306 static void
4307 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno)
4308 {
4309 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4310 	struct spdk_blob *origblob = ctx->original.blob;
4311 	struct spdk_blob *newblob = ctx->new.blob;
4312 
4313 	if (bserrno != 0) {
4314 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4315 		return;
4316 	}
4317 
4318 	/* Set internal xattr for snapshot id */
4319 	bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true);
4320 	if (bserrno != 0) {
4321 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4322 		return;
4323 	}
4324 
4325 	_spdk_bs_blob_list_remove(origblob);
4326 	origblob->parent_id = newblob->id;
4327 
4328 	/* Create new back_bs_dev for snapshot */
4329 	origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob);
4330 	if (origblob->back_bs_dev == NULL) {
4331 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL);
4332 		return;
4333 	}
4334 
4335 	/* set clone blob as thin provisioned */
4336 	_spdk_blob_set_thin_provision(origblob);
4337 
4338 	_spdk_bs_blob_list_add(newblob);
4339 
4340 	/* Zero out origblob cluster map */
4341 	memset(origblob->active.clusters, 0,
4342 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4343 
4344 	/* sync clone metadata */
4345 	spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx);
4346 }
4347 
4348 static void
4349 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc)
4350 {
4351 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4352 	struct spdk_blob *origblob = ctx->original.blob;
4353 	struct spdk_blob *newblob = ctx->new.blob;
4354 	int bserrno;
4355 
4356 	if (rc != 0) {
4357 		_spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc);
4358 		return;
4359 	}
4360 
4361 	ctx->frozen = true;
4362 
4363 	/* set new back_bs_dev for snapshot */
4364 	newblob->back_bs_dev = origblob->back_bs_dev;
4365 	/* Set invalid flags from origblob */
4366 	newblob->invalid_flags = origblob->invalid_flags;
4367 
4368 	/* inherit parent from original blob if set */
4369 	newblob->parent_id = origblob->parent_id;
4370 	if (origblob->parent_id != SPDK_BLOBID_INVALID) {
4371 		/* Set internal xattr for snapshot id */
4372 		bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT,
4373 					       &origblob->parent_id, sizeof(spdk_blob_id), true);
4374 		if (bserrno != 0) {
4375 			_spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno);
4376 			return;
4377 		}
4378 	}
4379 
4380 	/* Copy cluster map to snapshot */
4381 	memcpy(newblob->active.clusters, origblob->active.clusters,
4382 	       origblob->active.num_clusters * sizeof(origblob->active.clusters));
4383 
4384 	/* sync snapshot metadata */
4385 	spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx);
4386 }
4387 
4388 static void
4389 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4390 {
4391 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4392 	struct spdk_blob *origblob = ctx->original.blob;
4393 	struct spdk_blob *newblob = _blob;
4394 
4395 	if (bserrno != 0) {
4396 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4397 		return;
4398 	}
4399 
4400 	ctx->new.blob = newblob;
4401 
4402 	_spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx);
4403 }
4404 
4405 static void
4406 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4407 {
4408 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4409 	struct spdk_blob *origblob = ctx->original.blob;
4410 
4411 	if (bserrno != 0) {
4412 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4413 		return;
4414 	}
4415 
4416 	ctx->new.id = blobid;
4417 	ctx->cpl.u.blobid.blobid = blobid;
4418 
4419 	spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx);
4420 }
4421 
4422 
4423 static void
4424 _spdk_bs_xattr_snapshot(void *arg, const char *name,
4425 			const void **value, size_t *value_len)
4426 {
4427 	assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0);
4428 
4429 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4430 	*value = &blob->id;
4431 	*value_len = sizeof(blob->id);
4432 }
4433 
4434 static void
4435 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4436 {
4437 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4438 	struct spdk_blob_opts opts;
4439 	struct spdk_blob_xattr_opts internal_xattrs;
4440 	char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS };
4441 
4442 	if (bserrno != 0) {
4443 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4444 		return;
4445 	}
4446 
4447 	ctx->original.blob = _blob;
4448 
4449 	if (_blob->data_ro || _blob->md_ro) {
4450 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n",
4451 			      _blob->id);
4452 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4453 		return;
4454 	}
4455 
4456 	spdk_blob_opts_init(&opts);
4457 	_spdk_blob_xattrs_init(&internal_xattrs);
4458 
4459 	/* Change the size of new blob to the same as in original blob,
4460 	 * but do not allocate clusters */
4461 	opts.thin_provision = true;
4462 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4463 
4464 	/* If there are any xattrs specified for snapshot, set them now */
4465 	if (ctx->xattrs) {
4466 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4467 	}
4468 	/* Set internal xattr SNAPSHOT_IN_PROGRESS */
4469 	internal_xattrs.count = 1;
4470 	internal_xattrs.ctx = _blob;
4471 	internal_xattrs.names = xattrs_names;
4472 	internal_xattrs.get_value = _spdk_bs_xattr_snapshot;
4473 
4474 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4475 			     _spdk_bs_snapshot_newblob_create_cpl, ctx);
4476 }
4477 
4478 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid,
4479 			     const struct spdk_blob_xattr_opts *snapshot_xattrs,
4480 			     spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4481 {
4482 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4483 
4484 	if (!ctx) {
4485 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4486 		return;
4487 	}
4488 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4489 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4490 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4491 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4492 	ctx->bserrno = 0;
4493 	ctx->frozen = false;
4494 	ctx->original.id = blobid;
4495 	ctx->xattrs = snapshot_xattrs;
4496 
4497 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx);
4498 }
4499 /* END spdk_bs_create_snapshot */
4500 
4501 /* START spdk_bs_create_clone */
4502 
4503 static void
4504 _spdk_bs_xattr_clone(void *arg, const char *name,
4505 		     const void **value, size_t *value_len)
4506 {
4507 	assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0);
4508 
4509 	struct spdk_blob *blob = (struct spdk_blob *)arg;
4510 	*value = &blob->id;
4511 	*value_len = sizeof(blob->id);
4512 }
4513 
4514 static void
4515 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4516 {
4517 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4518 	struct spdk_blob *clone = _blob;
4519 
4520 	ctx->new.blob = clone;
4521 	_spdk_bs_blob_list_add(clone);
4522 
4523 	spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4524 }
4525 
4526 static void
4527 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno)
4528 {
4529 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4530 
4531 	ctx->cpl.u.blobid.blobid = blobid;
4532 	spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx);
4533 }
4534 
4535 static void
4536 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4537 {
4538 	struct spdk_clone_snapshot_ctx	*ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4539 	struct spdk_blob_opts		opts;
4540 	struct spdk_blob_xattr_opts internal_xattrs;
4541 	char *xattr_names[] = { BLOB_SNAPSHOT };
4542 
4543 	if (bserrno != 0) {
4544 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4545 		return;
4546 	}
4547 
4548 	ctx->original.blob = _blob;
4549 
4550 	if (!_blob->data_ro || !_blob->md_ro) {
4551 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n");
4552 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4553 		return;
4554 	}
4555 
4556 	spdk_blob_opts_init(&opts);
4557 	_spdk_blob_xattrs_init(&internal_xattrs);
4558 
4559 	opts.thin_provision = true;
4560 	opts.num_clusters = spdk_blob_get_num_clusters(_blob);
4561 	if (ctx->xattrs) {
4562 		memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs));
4563 	}
4564 
4565 	/* Set internal xattr BLOB_SNAPSHOT */
4566 	internal_xattrs.count = 1;
4567 	internal_xattrs.ctx = _blob;
4568 	internal_xattrs.names = xattr_names;
4569 	internal_xattrs.get_value = _spdk_bs_xattr_clone;
4570 
4571 	_spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs,
4572 			     _spdk_bs_clone_newblob_create_cpl, ctx);
4573 }
4574 
4575 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid,
4576 			  const struct spdk_blob_xattr_opts *clone_xattrs,
4577 			  spdk_blob_op_with_id_complete cb_fn, void *cb_arg)
4578 {
4579 	struct spdk_clone_snapshot_ctx	*ctx = calloc(1, sizeof(*ctx));
4580 
4581 	if (!ctx) {
4582 		cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM);
4583 		return;
4584 	}
4585 
4586 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID;
4587 	ctx->cpl.u.blobid.cb_fn = cb_fn;
4588 	ctx->cpl.u.blobid.cb_arg = cb_arg;
4589 	ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID;
4590 	ctx->bserrno = 0;
4591 	ctx->xattrs = clone_xattrs;
4592 	ctx->original.id = blobid;
4593 
4594 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx);
4595 }
4596 
4597 /* END spdk_bs_create_clone */
4598 
4599 /* START spdk_bs_inflate_blob */
4600 
4601 static void
4602 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno)
4603 {
4604 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4605 	struct spdk_blob *_blob = ctx->original.blob;
4606 
4607 	if (bserrno != 0) {
4608 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4609 		return;
4610 	}
4611 
4612 	assert(_parent != NULL);
4613 
4614 	_spdk_bs_blob_list_remove(_blob);
4615 	_blob->parent_id = _parent->id;
4616 	_spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id,
4617 			     sizeof(spdk_blob_id), true);
4618 
4619 	_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4620 	_blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent);
4621 	_spdk_bs_blob_list_add(_blob);
4622 
4623 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4624 }
4625 
4626 static void
4627 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno)
4628 {
4629 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4630 	struct spdk_blob *_blob = ctx->original.blob;
4631 	struct spdk_blob *_parent;
4632 
4633 	if (bserrno != 0) {
4634 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4635 		return;
4636 	}
4637 
4638 	if (ctx->allocate_all) {
4639 		/* remove thin provisioning */
4640 		_spdk_bs_blob_list_remove(_blob);
4641 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4642 		_blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV;
4643 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4644 		_blob->back_bs_dev = NULL;
4645 		_blob->parent_id = SPDK_BLOBID_INVALID;
4646 	} else {
4647 		_parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob;
4648 		if (_parent->parent_id != SPDK_BLOBID_INVALID) {
4649 			/* We must change the parent of the inflated blob */
4650 			spdk_bs_open_blob(_blob->bs, _parent->parent_id,
4651 					  _spdk_bs_inflate_blob_set_parent_cpl, ctx);
4652 			return;
4653 		}
4654 
4655 		_spdk_bs_blob_list_remove(_blob);
4656 		_spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true);
4657 		_blob->parent_id = SPDK_BLOBID_INVALID;
4658 		_blob->back_bs_dev->destroy(_blob->back_bs_dev);
4659 		_blob->back_bs_dev = spdk_bs_create_zeroes_dev();
4660 	}
4661 
4662 	_blob->state = SPDK_BLOB_STATE_DIRTY;
4663 	spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx);
4664 }
4665 
4666 /* Check if cluster needs allocation */
4667 static inline bool
4668 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all)
4669 {
4670 	struct spdk_blob_bs_dev *b;
4671 
4672 	assert(blob != NULL);
4673 
4674 	if (blob->active.clusters[cluster] != 0) {
4675 		/* Cluster is already allocated */
4676 		return false;
4677 	}
4678 
4679 	if (blob->parent_id == SPDK_BLOBID_INVALID) {
4680 		/* Blob have no parent blob */
4681 		return allocate_all;
4682 	}
4683 
4684 	b = (struct spdk_blob_bs_dev *)blob->back_bs_dev;
4685 	return (allocate_all || b->blob->active.clusters[cluster] != 0);
4686 }
4687 
4688 static void
4689 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno)
4690 {
4691 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4692 	struct spdk_blob *_blob = ctx->original.blob;
4693 	uint64_t offset;
4694 
4695 	if (bserrno != 0) {
4696 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno);
4697 		return;
4698 	}
4699 
4700 	for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) {
4701 		if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) {
4702 			break;
4703 		}
4704 	}
4705 
4706 	if (ctx->cluster < _blob->active.num_clusters) {
4707 		offset = _spdk_bs_cluster_to_page(_blob->bs, ctx->cluster);
4708 
4709 		/* We may safely increment a cluster before write */
4710 		ctx->cluster++;
4711 
4712 		/* Use zero length write to touch a cluster */
4713 		spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0,
4714 				   _spdk_bs_inflate_blob_touch_next, ctx);
4715 	} else {
4716 		_spdk_bs_inflate_blob_done(cb_arg, bserrno);
4717 	}
4718 }
4719 
4720 static void
4721 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
4722 {
4723 	struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg;
4724 	uint64_t lfc; /* lowest free cluster */
4725 	uint64_t i;
4726 
4727 	if (bserrno != 0) {
4728 		_spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno);
4729 		return;
4730 	}
4731 	ctx->original.blob = _blob;
4732 
4733 	if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) {
4734 		/* This blob have no parent, so we cannot decouple it. */
4735 		SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n");
4736 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL);
4737 		return;
4738 	}
4739 
4740 	if (spdk_blob_is_thin_provisioned(_blob) == false) {
4741 		/* This is not thin provisioned blob. No need to inflate. */
4742 		_spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0);
4743 		return;
4744 	}
4745 
4746 	/* Do two passes - one to verify that we can obtain enough clusters
4747 	 * and another to actually claim them.
4748 	 */
4749 	lfc = 0;
4750 	for (i = 0; i < _blob->active.num_clusters; i++) {
4751 		if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) {
4752 			lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc);
4753 			if (lfc >= _blob->bs->total_clusters) {
4754 				/* No more free clusters. Cannot satisfy the request */
4755 				_spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC);
4756 				return;
4757 			}
4758 			lfc++;
4759 		}
4760 	}
4761 
4762 	ctx->cluster = 0;
4763 	_spdk_bs_inflate_blob_touch_next(ctx, 0);
4764 }
4765 
4766 static void
4767 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4768 		      spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg)
4769 {
4770 	struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx));
4771 
4772 	if (!ctx) {
4773 		cb_fn(cb_arg, -ENOMEM);
4774 		return;
4775 	}
4776 	ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4777 	ctx->cpl.u.bs_basic.cb_fn = cb_fn;
4778 	ctx->cpl.u.bs_basic.cb_arg = cb_arg;
4779 	ctx->bserrno = 0;
4780 	ctx->original.id = blobid;
4781 	ctx->channel = channel;
4782 	ctx->allocate_all = allocate_all;
4783 
4784 	spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx);
4785 }
4786 
4787 void
4788 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4789 		     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4790 {
4791 	_spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg);
4792 }
4793 
4794 void
4795 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel,
4796 			     spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg)
4797 {
4798 	_spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg);
4799 }
4800 /* END spdk_bs_inflate_blob */
4801 
4802 /* START spdk_blob_resize */
4803 struct spdk_bs_resize_ctx {
4804 	spdk_blob_op_complete cb_fn;
4805 	void *cb_arg;
4806 	struct spdk_blob *blob;
4807 	uint64_t sz;
4808 	int rc;
4809 };
4810 
4811 static void
4812 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc)
4813 {
4814 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4815 
4816 	if (rc != 0) {
4817 		SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc);
4818 	}
4819 
4820 	if (ctx->rc != 0) {
4821 		SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc);
4822 		rc = ctx->rc;
4823 	}
4824 
4825 	ctx->blob->resize_in_progress = false;
4826 
4827 	ctx->cb_fn(ctx->cb_arg, rc);
4828 	free(ctx);
4829 }
4830 
4831 static void
4832 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc)
4833 {
4834 	struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg;
4835 
4836 	if (rc != 0) {
4837 		ctx->blob->resize_in_progress = false;
4838 		ctx->cb_fn(ctx->cb_arg, rc);
4839 		free(ctx);
4840 		return;
4841 	}
4842 
4843 	ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz);
4844 
4845 	_spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx);
4846 }
4847 
4848 void
4849 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg)
4850 {
4851 	struct spdk_bs_resize_ctx *ctx;
4852 
4853 	_spdk_blob_verify_md_op(blob);
4854 
4855 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz);
4856 
4857 	if (blob->md_ro) {
4858 		cb_fn(cb_arg, -EPERM);
4859 		return;
4860 	}
4861 
4862 	if (sz == blob->active.num_clusters) {
4863 		cb_fn(cb_arg, 0);
4864 		return;
4865 	}
4866 
4867 	if (blob->resize_in_progress) {
4868 		cb_fn(cb_arg, -EBUSY);
4869 		return;
4870 	}
4871 
4872 	ctx = calloc(1, sizeof(*ctx));
4873 	if (!ctx) {
4874 		cb_fn(cb_arg, -ENOMEM);
4875 		return;
4876 	}
4877 
4878 	blob->resize_in_progress = true;
4879 	ctx->cb_fn = cb_fn;
4880 	ctx->cb_arg = cb_arg;
4881 	ctx->blob = blob;
4882 	ctx->sz = sz;
4883 	_spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx);
4884 }
4885 
4886 /* END spdk_blob_resize */
4887 
4888 
4889 /* START spdk_bs_delete_blob */
4890 
4891 static void
4892 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno)
4893 {
4894 	spdk_bs_sequence_t *seq = cb_arg;
4895 
4896 	spdk_bs_sequence_finish(seq, bserrno);
4897 }
4898 
4899 static void
4900 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
4901 {
4902 	struct spdk_blob *blob = cb_arg;
4903 
4904 	if (bserrno != 0) {
4905 		/*
4906 		 * We already removed this blob from the blobstore tailq, so
4907 		 *  we need to free it here since this is the last reference
4908 		 *  to it.
4909 		 */
4910 		_spdk_blob_free(blob);
4911 		_spdk_bs_delete_close_cpl(seq, bserrno);
4912 		return;
4913 	}
4914 
4915 	/*
4916 	 * This will immediately decrement the ref_count and call
4917 	 *  the completion routine since the metadata state is clean.
4918 	 *  By calling spdk_blob_close, we reduce the number of call
4919 	 *  points into code that touches the blob->open_ref count
4920 	 *  and the blobstore's blob list.
4921 	 */
4922 	spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq);
4923 }
4924 
4925 static void
4926 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno)
4927 {
4928 	spdk_bs_sequence_t *seq = cb_arg;
4929 	uint32_t page_num;
4930 
4931 	if (bserrno != 0) {
4932 		spdk_bs_sequence_finish(seq, bserrno);
4933 		return;
4934 	}
4935 
4936 	_spdk_blob_verify_md_op(blob);
4937 
4938 	if (blob->open_ref > 1) {
4939 		/*
4940 		 * Someone has this blob open (besides this delete context).
4941 		 *  Decrement the ref count directly and return -EBUSY.
4942 		 */
4943 		blob->open_ref--;
4944 		spdk_bs_sequence_finish(seq, -EBUSY);
4945 		return;
4946 	}
4947 
4948 	bserrno = _spdk_bs_blob_list_remove(blob);
4949 	if (bserrno != 0) {
4950 		SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Remove blob #%" PRIu64 " from a list\n", blob->id);
4951 		spdk_bs_sequence_finish(seq, bserrno);
4952 		return;
4953 	}
4954 
4955 	/*
4956 	 * Remove the blob from the blob_store list now, to ensure it does not
4957 	 *  get returned after this point by _spdk_blob_lookup().
4958 	 */
4959 	TAILQ_REMOVE(&blob->bs->blobs, blob, link);
4960 	page_num = _spdk_bs_blobid_to_page(blob->id);
4961 	spdk_bit_array_clear(blob->bs->used_blobids, page_num);
4962 	blob->state = SPDK_BLOB_STATE_DIRTY;
4963 	blob->active.num_pages = 0;
4964 	_spdk_blob_resize(blob, 0);
4965 
4966 	_spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob);
4967 }
4968 
4969 void
4970 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
4971 		    spdk_blob_op_complete cb_fn, void *cb_arg)
4972 {
4973 	struct spdk_bs_cpl	cpl;
4974 	spdk_bs_sequence_t	*seq;
4975 	struct spdk_blob_list	*snapshot_entry = NULL;
4976 
4977 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid);
4978 
4979 	assert(spdk_get_thread() == bs->md_thread);
4980 
4981 	/* Check if this is a snapshot with clones */
4982 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
4983 		if (snapshot_entry->id == blobid) {
4984 			break;
4985 		}
4986 	}
4987 	if (snapshot_entry != NULL) {
4988 		/* If snapshot have clones, we cannot remove it */
4989 		if (!TAILQ_EMPTY(&snapshot_entry->clones)) {
4990 			SPDK_ERRLOG("Cannot remove snapshot with clones\n");
4991 			cb_fn(cb_arg, -EBUSY);
4992 			return;
4993 		}
4994 	}
4995 
4996 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
4997 	cpl.u.blob_basic.cb_fn = cb_fn;
4998 	cpl.u.blob_basic.cb_arg = cb_arg;
4999 
5000 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5001 	if (!seq) {
5002 		cb_fn(cb_arg, -ENOMEM);
5003 		return;
5004 	}
5005 
5006 	spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq);
5007 }
5008 
5009 /* END spdk_bs_delete_blob */
5010 
5011 /* START spdk_bs_open_blob */
5012 
5013 static void
5014 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5015 {
5016 	struct spdk_blob *blob = cb_arg;
5017 
5018 	/* If the blob have crc error, we just return NULL. */
5019 	if (blob == NULL) {
5020 		seq->cpl.u.blob_handle.blob = NULL;
5021 		spdk_bs_sequence_finish(seq, bserrno);
5022 		return;
5023 	}
5024 
5025 	blob->open_ref++;
5026 
5027 	TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link);
5028 
5029 	spdk_bs_sequence_finish(seq, bserrno);
5030 }
5031 
5032 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid,
5033 		       spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5034 {
5035 	struct spdk_blob		*blob;
5036 	struct spdk_bs_cpl		cpl;
5037 	spdk_bs_sequence_t		*seq;
5038 	uint32_t			page_num;
5039 
5040 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid);
5041 	assert(spdk_get_thread() == bs->md_thread);
5042 
5043 	page_num = _spdk_bs_blobid_to_page(blobid);
5044 	if (spdk_bit_array_get(bs->used_blobids, page_num) == false) {
5045 		/* Invalid blobid */
5046 		cb_fn(cb_arg, NULL, -ENOENT);
5047 		return;
5048 	}
5049 
5050 	blob = _spdk_blob_lookup(bs, blobid);
5051 	if (blob) {
5052 		blob->open_ref++;
5053 		cb_fn(cb_arg, blob, 0);
5054 		return;
5055 	}
5056 
5057 	blob = _spdk_blob_alloc(bs, blobid);
5058 	if (!blob) {
5059 		cb_fn(cb_arg, NULL, -ENOMEM);
5060 		return;
5061 	}
5062 
5063 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE;
5064 	cpl.u.blob_handle.cb_fn = cb_fn;
5065 	cpl.u.blob_handle.cb_arg = cb_arg;
5066 	cpl.u.blob_handle.blob = blob;
5067 
5068 	seq = spdk_bs_sequence_start(bs->md_channel, &cpl);
5069 	if (!seq) {
5070 		_spdk_blob_free(blob);
5071 		cb_fn(cb_arg, NULL, -ENOMEM);
5072 		return;
5073 	}
5074 
5075 	_spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob);
5076 }
5077 /* END spdk_bs_open_blob */
5078 
5079 /* START spdk_blob_set_read_only */
5080 int spdk_blob_set_read_only(struct spdk_blob *blob)
5081 {
5082 	_spdk_blob_verify_md_op(blob);
5083 
5084 	blob->data_ro_flags |= SPDK_BLOB_READ_ONLY;
5085 
5086 	blob->state = SPDK_BLOB_STATE_DIRTY;
5087 	return 0;
5088 }
5089 /* END spdk_blob_set_read_only */
5090 
5091 /* START spdk_blob_sync_md */
5092 
5093 static void
5094 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5095 {
5096 	struct spdk_blob *blob = cb_arg;
5097 
5098 	if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) {
5099 		blob->data_ro = true;
5100 		blob->md_ro = true;
5101 	}
5102 
5103 	spdk_bs_sequence_finish(seq, bserrno);
5104 }
5105 
5106 static void
5107 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5108 {
5109 	struct spdk_bs_cpl	cpl;
5110 	spdk_bs_sequence_t	*seq;
5111 
5112 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5113 	cpl.u.blob_basic.cb_fn = cb_fn;
5114 	cpl.u.blob_basic.cb_arg = cb_arg;
5115 
5116 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5117 	if (!seq) {
5118 		cb_fn(cb_arg, -ENOMEM);
5119 		return;
5120 	}
5121 
5122 	_spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob);
5123 }
5124 
5125 void
5126 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5127 {
5128 	_spdk_blob_verify_md_op(blob);
5129 
5130 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id);
5131 
5132 	if (blob->md_ro) {
5133 		assert(blob->state == SPDK_BLOB_STATE_CLEAN);
5134 		cb_fn(cb_arg, 0);
5135 		return;
5136 	}
5137 
5138 	_spdk_blob_sync_md(blob, cb_fn, cb_arg);
5139 }
5140 
5141 /* END spdk_blob_sync_md */
5142 
5143 struct spdk_blob_insert_cluster_ctx {
5144 	struct spdk_thread	*thread;
5145 	struct spdk_blob	*blob;
5146 	uint32_t		cluster_num;	/* cluster index in blob */
5147 	uint32_t		cluster;	/* cluster on disk */
5148 	int			rc;
5149 	spdk_blob_op_complete	cb_fn;
5150 	void			*cb_arg;
5151 };
5152 
5153 static void
5154 _spdk_blob_insert_cluster_msg_cpl(void *arg)
5155 {
5156 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5157 
5158 	ctx->cb_fn(ctx->cb_arg, ctx->rc);
5159 	free(ctx);
5160 }
5161 
5162 static void
5163 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno)
5164 {
5165 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5166 
5167 	ctx->rc = bserrno;
5168 	spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5169 }
5170 
5171 static void
5172 _spdk_blob_insert_cluster_msg(void *arg)
5173 {
5174 	struct spdk_blob_insert_cluster_ctx *ctx = arg;
5175 
5176 	ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster);
5177 	if (ctx->rc != 0) {
5178 		spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx);
5179 		return;
5180 	}
5181 
5182 	ctx->blob->state = SPDK_BLOB_STATE_DIRTY;
5183 	_spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx);
5184 }
5185 
5186 static void
5187 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num,
5188 				       uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg)
5189 {
5190 	struct spdk_blob_insert_cluster_ctx *ctx;
5191 
5192 	ctx = calloc(1, sizeof(*ctx));
5193 	if (ctx == NULL) {
5194 		cb_fn(cb_arg, -ENOMEM);
5195 		return;
5196 	}
5197 
5198 	ctx->thread = spdk_get_thread();
5199 	ctx->blob = blob;
5200 	ctx->cluster_num = cluster_num;
5201 	ctx->cluster = cluster;
5202 	ctx->cb_fn = cb_fn;
5203 	ctx->cb_arg = cb_arg;
5204 
5205 	spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx);
5206 }
5207 
5208 /* START spdk_blob_close */
5209 
5210 static void
5211 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno)
5212 {
5213 	struct spdk_blob *blob = cb_arg;
5214 
5215 	if (bserrno == 0) {
5216 		blob->open_ref--;
5217 		if (blob->open_ref == 0) {
5218 			/*
5219 			 * Blobs with active.num_pages == 0 are deleted blobs.
5220 			 *  these blobs are removed from the blob_store list
5221 			 *  when the deletion process starts - so don't try to
5222 			 *  remove them again.
5223 			 */
5224 			if (blob->active.num_pages > 0) {
5225 				TAILQ_REMOVE(&blob->bs->blobs, blob, link);
5226 			}
5227 			_spdk_blob_free(blob);
5228 		}
5229 	}
5230 
5231 	spdk_bs_sequence_finish(seq, bserrno);
5232 }
5233 
5234 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg)
5235 {
5236 	struct spdk_bs_cpl	cpl;
5237 	spdk_bs_sequence_t	*seq;
5238 
5239 	_spdk_blob_verify_md_op(blob);
5240 
5241 	SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id);
5242 
5243 	if (blob->open_ref == 0) {
5244 		cb_fn(cb_arg, -EBADF);
5245 		return;
5246 	}
5247 
5248 	cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC;
5249 	cpl.u.blob_basic.cb_fn = cb_fn;
5250 	cpl.u.blob_basic.cb_arg = cb_arg;
5251 
5252 	seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl);
5253 	if (!seq) {
5254 		cb_fn(cb_arg, -ENOMEM);
5255 		return;
5256 	}
5257 
5258 	/* Sync metadata */
5259 	_spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob);
5260 }
5261 
5262 /* END spdk_blob_close */
5263 
5264 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs)
5265 {
5266 	return spdk_get_io_channel(bs);
5267 }
5268 
5269 void spdk_bs_free_io_channel(struct spdk_io_channel *channel)
5270 {
5271 	spdk_put_io_channel(channel);
5272 }
5273 
5274 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel,
5275 			uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5276 {
5277 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5278 				     SPDK_BLOB_UNMAP);
5279 }
5280 
5281 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel,
5282 			       uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg)
5283 {
5284 	_spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg,
5285 				     SPDK_BLOB_WRITE_ZEROES);
5286 }
5287 
5288 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel,
5289 			void *payload, uint64_t offset, uint64_t length,
5290 			spdk_blob_op_complete cb_fn, void *cb_arg)
5291 {
5292 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5293 				     SPDK_BLOB_WRITE);
5294 }
5295 
5296 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel,
5297 		       void *payload, uint64_t offset, uint64_t length,
5298 		       spdk_blob_op_complete cb_fn, void *cb_arg)
5299 {
5300 	_spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg,
5301 				     SPDK_BLOB_READ);
5302 }
5303 
5304 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel,
5305 			 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5306 			 spdk_blob_op_complete cb_fn, void *cb_arg)
5307 {
5308 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false);
5309 }
5310 
5311 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel,
5312 			struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length,
5313 			spdk_blob_op_complete cb_fn, void *cb_arg)
5314 {
5315 	_spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true);
5316 }
5317 
5318 struct spdk_bs_iter_ctx {
5319 	int64_t page_num;
5320 	struct spdk_blob_store *bs;
5321 
5322 	spdk_blob_op_with_handle_complete cb_fn;
5323 	void *cb_arg;
5324 };
5325 
5326 static void
5327 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno)
5328 {
5329 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5330 	struct spdk_blob_store *bs = ctx->bs;
5331 	spdk_blob_id id;
5332 
5333 	if (bserrno == 0) {
5334 		ctx->cb_fn(ctx->cb_arg, _blob, bserrno);
5335 		free(ctx);
5336 		return;
5337 	}
5338 
5339 	ctx->page_num++;
5340 	ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num);
5341 	if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) {
5342 		ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT);
5343 		free(ctx);
5344 		return;
5345 	}
5346 
5347 	id = _spdk_bs_page_to_blobid(ctx->page_num);
5348 
5349 	spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx);
5350 }
5351 
5352 void
5353 spdk_bs_iter_first(struct spdk_blob_store *bs,
5354 		   spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5355 {
5356 	struct spdk_bs_iter_ctx *ctx;
5357 
5358 	ctx = calloc(1, sizeof(*ctx));
5359 	if (!ctx) {
5360 		cb_fn(cb_arg, NULL, -ENOMEM);
5361 		return;
5362 	}
5363 
5364 	ctx->page_num = -1;
5365 	ctx->bs = bs;
5366 	ctx->cb_fn = cb_fn;
5367 	ctx->cb_arg = cb_arg;
5368 
5369 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5370 }
5371 
5372 static void
5373 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno)
5374 {
5375 	struct spdk_bs_iter_ctx *ctx = cb_arg;
5376 
5377 	_spdk_bs_iter_cpl(ctx, NULL, -1);
5378 }
5379 
5380 void
5381 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob,
5382 		  spdk_blob_op_with_handle_complete cb_fn, void *cb_arg)
5383 {
5384 	struct spdk_bs_iter_ctx *ctx;
5385 
5386 	assert(blob != NULL);
5387 
5388 	ctx = calloc(1, sizeof(*ctx));
5389 	if (!ctx) {
5390 		cb_fn(cb_arg, NULL, -ENOMEM);
5391 		return;
5392 	}
5393 
5394 	ctx->page_num = _spdk_bs_blobid_to_page(blob->id);
5395 	ctx->bs = bs;
5396 	ctx->cb_fn = cb_fn;
5397 	ctx->cb_arg = cb_arg;
5398 
5399 	/* Close the existing blob */
5400 	spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx);
5401 }
5402 
5403 static int
5404 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5405 		     uint16_t value_len, bool internal)
5406 {
5407 	struct spdk_xattr_tailq *xattrs;
5408 	struct spdk_xattr	*xattr;
5409 
5410 	_spdk_blob_verify_md_op(blob);
5411 
5412 	if (blob->md_ro) {
5413 		return -EPERM;
5414 	}
5415 
5416 	if (internal) {
5417 		xattrs = &blob->xattrs_internal;
5418 		blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR;
5419 	} else {
5420 		xattrs = &blob->xattrs;
5421 	}
5422 
5423 	TAILQ_FOREACH(xattr, xattrs, link) {
5424 		if (!strcmp(name, xattr->name)) {
5425 			free(xattr->value);
5426 			xattr->value_len = value_len;
5427 			xattr->value = malloc(value_len);
5428 			memcpy(xattr->value, value, value_len);
5429 
5430 			blob->state = SPDK_BLOB_STATE_DIRTY;
5431 
5432 			return 0;
5433 		}
5434 	}
5435 
5436 	xattr = calloc(1, sizeof(*xattr));
5437 	if (!xattr) {
5438 		return -ENOMEM;
5439 	}
5440 	xattr->name = strdup(name);
5441 	xattr->value_len = value_len;
5442 	xattr->value = malloc(value_len);
5443 	memcpy(xattr->value, value, value_len);
5444 	TAILQ_INSERT_TAIL(xattrs, xattr, link);
5445 
5446 	blob->state = SPDK_BLOB_STATE_DIRTY;
5447 
5448 	return 0;
5449 }
5450 
5451 int
5452 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value,
5453 		    uint16_t value_len)
5454 {
5455 	return _spdk_blob_set_xattr(blob, name, value, value_len, false);
5456 }
5457 
5458 static int
5459 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal)
5460 {
5461 	struct spdk_xattr_tailq *xattrs;
5462 	struct spdk_xattr	*xattr;
5463 
5464 	_spdk_blob_verify_md_op(blob);
5465 
5466 	if (blob->md_ro) {
5467 		return -EPERM;
5468 	}
5469 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5470 
5471 	TAILQ_FOREACH(xattr, xattrs, link) {
5472 		if (!strcmp(name, xattr->name)) {
5473 			TAILQ_REMOVE(xattrs, xattr, link);
5474 			free(xattr->value);
5475 			free(xattr->name);
5476 			free(xattr);
5477 
5478 			if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) {
5479 				blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR;
5480 			}
5481 			blob->state = SPDK_BLOB_STATE_DIRTY;
5482 
5483 			return 0;
5484 		}
5485 	}
5486 
5487 	return -ENOENT;
5488 }
5489 
5490 int
5491 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name)
5492 {
5493 	return _spdk_blob_remove_xattr(blob, name, false);
5494 }
5495 
5496 static int
5497 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5498 			   const void **value, size_t *value_len, bool internal)
5499 {
5500 	struct spdk_xattr	*xattr;
5501 	struct spdk_xattr_tailq *xattrs;
5502 
5503 	xattrs = internal ? &blob->xattrs_internal : &blob->xattrs;
5504 
5505 	TAILQ_FOREACH(xattr, xattrs, link) {
5506 		if (!strcmp(name, xattr->name)) {
5507 			*value = xattr->value;
5508 			*value_len = xattr->value_len;
5509 			return 0;
5510 		}
5511 	}
5512 	return -ENOENT;
5513 }
5514 
5515 int
5516 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name,
5517 			  const void **value, size_t *value_len)
5518 {
5519 	_spdk_blob_verify_md_op(blob);
5520 
5521 	return _spdk_blob_get_xattr_value(blob, name, value, value_len, false);
5522 }
5523 
5524 struct spdk_xattr_names {
5525 	uint32_t	count;
5526 	const char	*names[0];
5527 };
5528 
5529 static int
5530 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names)
5531 {
5532 	struct spdk_xattr	*xattr;
5533 	int			count = 0;
5534 
5535 	TAILQ_FOREACH(xattr, xattrs, link) {
5536 		count++;
5537 	}
5538 
5539 	*names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *));
5540 	if (*names == NULL) {
5541 		return -ENOMEM;
5542 	}
5543 
5544 	TAILQ_FOREACH(xattr, xattrs, link) {
5545 		(*names)->names[(*names)->count++] = xattr->name;
5546 	}
5547 
5548 	return 0;
5549 }
5550 
5551 int
5552 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names)
5553 {
5554 	_spdk_blob_verify_md_op(blob);
5555 
5556 	return _spdk_blob_get_xattr_names(&blob->xattrs, names);
5557 }
5558 
5559 uint32_t
5560 spdk_xattr_names_get_count(struct spdk_xattr_names *names)
5561 {
5562 	assert(names != NULL);
5563 
5564 	return names->count;
5565 }
5566 
5567 const char *
5568 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index)
5569 {
5570 	if (index >= names->count) {
5571 		return NULL;
5572 	}
5573 
5574 	return names->names[index];
5575 }
5576 
5577 void
5578 spdk_xattr_names_free(struct spdk_xattr_names *names)
5579 {
5580 	free(names);
5581 }
5582 
5583 struct spdk_bs_type
5584 spdk_bs_get_bstype(struct spdk_blob_store *bs)
5585 {
5586 	return bs->bstype;
5587 }
5588 
5589 void
5590 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype)
5591 {
5592 	memcpy(&bs->bstype, &bstype, sizeof(bstype));
5593 }
5594 
5595 bool
5596 spdk_blob_is_read_only(struct spdk_blob *blob)
5597 {
5598 	assert(blob != NULL);
5599 	return (blob->data_ro || blob->md_ro);
5600 }
5601 
5602 bool
5603 spdk_blob_is_snapshot(struct spdk_blob *blob)
5604 {
5605 	struct spdk_blob_list *snapshot_entry;
5606 
5607 	assert(blob != NULL);
5608 
5609 	TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) {
5610 		if (snapshot_entry->id == blob->id) {
5611 			break;
5612 		}
5613 	}
5614 
5615 	if (snapshot_entry == NULL) {
5616 		return false;
5617 	}
5618 
5619 	return true;
5620 }
5621 
5622 bool
5623 spdk_blob_is_clone(struct spdk_blob *blob)
5624 {
5625 	assert(blob != NULL);
5626 
5627 	if (blob->parent_id != SPDK_BLOBID_INVALID) {
5628 		assert(spdk_blob_is_thin_provisioned(blob));
5629 		return true;
5630 	}
5631 
5632 	return false;
5633 }
5634 
5635 bool
5636 spdk_blob_is_thin_provisioned(struct spdk_blob *blob)
5637 {
5638 	assert(blob != NULL);
5639 	return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV);
5640 }
5641 
5642 spdk_blob_id
5643 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id)
5644 {
5645 	struct spdk_blob_list *snapshot_entry = NULL;
5646 	struct spdk_blob_list *clone_entry = NULL;
5647 
5648 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5649 		TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5650 			if (clone_entry->id == blob_id) {
5651 				return snapshot_entry->id;
5652 			}
5653 		}
5654 	}
5655 
5656 	return SPDK_BLOBID_INVALID;
5657 }
5658 
5659 int
5660 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids,
5661 		     size_t *count)
5662 {
5663 	struct spdk_blob_list *snapshot_entry, *clone_entry;
5664 	size_t n;
5665 
5666 	TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) {
5667 		if (snapshot_entry->id == blobid) {
5668 			break;
5669 		}
5670 	}
5671 	if (snapshot_entry == NULL) {
5672 		*count = 0;
5673 		return 0;
5674 	}
5675 
5676 	if (ids == NULL || *count < snapshot_entry->clone_count) {
5677 		*count = snapshot_entry->clone_count;
5678 		return -ENOMEM;
5679 	}
5680 	*count = snapshot_entry->clone_count;
5681 
5682 	n = 0;
5683 	TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) {
5684 		ids[n++] = clone_entry->id;
5685 	}
5686 
5687 	return 0;
5688 }
5689 
5690 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB)
5691