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