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/log.h" 45 46 #include "blobstore.h" 47 48 #define BLOB_CRC32C_INITIAL 0xffffffffUL 49 50 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs); 51 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs); 52 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno); 53 void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 54 uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg); 55 56 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 57 uint16_t value_len, bool internal); 58 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 59 const void **value, size_t *value_len, bool internal); 60 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal); 61 62 static void 63 _spdk_blob_verify_md_op(struct spdk_blob *blob) 64 { 65 assert(blob != NULL); 66 assert(spdk_get_thread() == blob->bs->md_thread); 67 assert(blob->state != SPDK_BLOB_STATE_LOADING); 68 } 69 70 static inline size_t 71 divide_round_up(size_t num, size_t divisor) 72 { 73 return (num + divisor - 1) / divisor; 74 } 75 76 static void 77 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 78 { 79 assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters)); 80 assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false); 81 assert(bs->num_free_clusters > 0); 82 83 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num); 84 85 spdk_bit_array_set(bs->used_clusters, cluster_num); 86 bs->num_free_clusters--; 87 } 88 89 static int 90 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster) 91 { 92 uint64_t *cluster_lba = &blob->active.clusters[cluster_num]; 93 94 _spdk_blob_verify_md_op(blob); 95 96 if (*cluster_lba != 0) { 97 return -EEXIST; 98 } 99 100 *cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster); 101 return 0; 102 } 103 104 static int 105 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num, 106 uint64_t *lowest_free_cluster, bool update_map) 107 { 108 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 109 *lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters, 110 *lowest_free_cluster); 111 if (*lowest_free_cluster >= blob->bs->total_clusters) { 112 /* No more free clusters. Cannot satisfy the request */ 113 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 114 return -ENOSPC; 115 } 116 117 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id); 118 _spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster); 119 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 120 121 if (update_map) { 122 _spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster); 123 } 124 125 return 0; 126 } 127 128 static void 129 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 130 { 131 assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters)); 132 assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true); 133 assert(bs->num_free_clusters < bs->total_clusters); 134 135 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num); 136 137 pthread_mutex_lock(&bs->used_clusters_mutex); 138 spdk_bit_array_clear(bs->used_clusters, cluster_num); 139 bs->num_free_clusters++; 140 pthread_mutex_unlock(&bs->used_clusters_mutex); 141 } 142 143 void 144 spdk_blob_opts_init(struct spdk_blob_opts *opts) 145 { 146 opts->num_clusters = 0; 147 opts->thin_provision = false; 148 opts->xattrs.count = 0; 149 opts->xattrs.names = NULL; 150 opts->xattrs.ctx = NULL; 151 opts->xattrs.get_value = NULL; 152 } 153 154 static struct spdk_blob * 155 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id) 156 { 157 struct spdk_blob *blob; 158 159 blob = calloc(1, sizeof(*blob)); 160 if (!blob) { 161 return NULL; 162 } 163 164 blob->id = id; 165 blob->bs = bs; 166 167 blob->state = SPDK_BLOB_STATE_DIRTY; 168 blob->active.num_pages = 1; 169 blob->active.pages = calloc(1, sizeof(*blob->active.pages)); 170 if (!blob->active.pages) { 171 free(blob); 172 return NULL; 173 } 174 175 blob->active.pages[0] = _spdk_bs_blobid_to_page(id); 176 177 TAILQ_INIT(&blob->xattrs); 178 TAILQ_INIT(&blob->xattrs_internal); 179 180 return blob; 181 } 182 183 static void 184 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs) 185 { 186 struct spdk_xattr *xattr, *xattr_tmp; 187 188 TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) { 189 TAILQ_REMOVE(xattrs, xattr, link); 190 free(xattr->name); 191 free(xattr->value); 192 free(xattr); 193 } 194 } 195 196 static void 197 _spdk_blob_free(struct spdk_blob *blob) 198 { 199 assert(blob != NULL); 200 201 free(blob->active.clusters); 202 free(blob->clean.clusters); 203 free(blob->active.pages); 204 free(blob->clean.pages); 205 206 _spdk_xattrs_free(&blob->xattrs); 207 _spdk_xattrs_free(&blob->xattrs_internal); 208 209 free(blob); 210 } 211 212 static int 213 _spdk_blob_mark_clean(struct spdk_blob *blob) 214 { 215 uint64_t *clusters = NULL; 216 uint32_t *pages = NULL; 217 218 assert(blob != NULL); 219 220 if (blob->active.num_clusters) { 221 assert(blob->active.clusters); 222 clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters)); 223 if (!clusters) { 224 return -1; 225 } 226 memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*clusters)); 227 } 228 229 if (blob->active.num_pages) { 230 assert(blob->active.pages); 231 pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages)); 232 if (!pages) { 233 free(clusters); 234 return -1; 235 } 236 memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*pages)); 237 } 238 239 free(blob->clean.clusters); 240 free(blob->clean.pages); 241 242 blob->clean.num_clusters = blob->active.num_clusters; 243 blob->clean.clusters = blob->active.clusters; 244 blob->clean.num_pages = blob->active.num_pages; 245 blob->clean.pages = blob->active.pages; 246 247 blob->active.clusters = clusters; 248 blob->active.pages = pages; 249 250 /* If the metadata was dirtied again while the metadata was being written to disk, 251 * we do not want to revert the DIRTY state back to CLEAN here. 252 */ 253 if (blob->state == SPDK_BLOB_STATE_LOADING) { 254 blob->state = SPDK_BLOB_STATE_CLEAN; 255 } 256 257 return 0; 258 } 259 260 static int 261 _spdk_blob_deserialize_xattr(struct spdk_blob *blob, 262 struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal) 263 { 264 struct spdk_xattr *xattr; 265 266 if (desc_xattr->length != sizeof(desc_xattr->name_length) + 267 sizeof(desc_xattr->value_length) + 268 desc_xattr->name_length + desc_xattr->value_length) { 269 return -EINVAL; 270 } 271 272 xattr = calloc(1, sizeof(*xattr)); 273 if (xattr == NULL) { 274 return -ENOMEM; 275 } 276 277 xattr->name = malloc(desc_xattr->name_length + 1); 278 if (xattr->name == NULL) { 279 free(xattr); 280 return -ENOMEM; 281 } 282 strncpy(xattr->name, desc_xattr->name, desc_xattr->name_length); 283 xattr->name[desc_xattr->name_length] = '\0'; 284 285 xattr->value = malloc(desc_xattr->value_length); 286 if (xattr->value == NULL) { 287 free(xattr->name); 288 free(xattr); 289 return -ENOMEM; 290 } 291 xattr->value_len = desc_xattr->value_length; 292 memcpy(xattr->value, 293 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 294 desc_xattr->value_length); 295 296 TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link); 297 298 return 0; 299 } 300 301 302 static int 303 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob) 304 { 305 struct spdk_blob_md_descriptor *desc; 306 size_t cur_desc = 0; 307 void *tmp; 308 309 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 310 while (cur_desc < sizeof(page->descriptors)) { 311 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 312 if (desc->length == 0) { 313 /* If padding and length are 0, this terminates the page */ 314 break; 315 } 316 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 317 struct spdk_blob_md_descriptor_flags *desc_flags; 318 319 desc_flags = (struct spdk_blob_md_descriptor_flags *)desc; 320 321 if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) { 322 return -EINVAL; 323 } 324 325 if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) != 326 SPDK_BLOB_INVALID_FLAGS_MASK) { 327 return -EINVAL; 328 } 329 330 if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) != 331 SPDK_BLOB_DATA_RO_FLAGS_MASK) { 332 blob->data_ro = true; 333 blob->md_ro = true; 334 } 335 336 if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) != 337 SPDK_BLOB_MD_RO_FLAGS_MASK) { 338 blob->md_ro = true; 339 } 340 341 if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 342 blob->data_ro = true; 343 blob->md_ro = true; 344 } 345 346 blob->invalid_flags = desc_flags->invalid_flags; 347 blob->data_ro_flags = desc_flags->data_ro_flags; 348 blob->md_ro_flags = desc_flags->md_ro_flags; 349 350 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 351 struct spdk_blob_md_descriptor_extent *desc_extent; 352 unsigned int i, j; 353 unsigned int cluster_count = blob->active.num_clusters; 354 355 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 356 357 if (desc_extent->length == 0 || 358 (desc_extent->length % sizeof(desc_extent->extents[0]) != 0)) { 359 return -EINVAL; 360 } 361 362 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 363 for (j = 0; j < desc_extent->extents[i].length; j++) { 364 if (!spdk_bit_array_get(blob->bs->used_clusters, 365 desc_extent->extents[i].cluster_idx + j)) { 366 return -EINVAL; 367 } 368 cluster_count++; 369 } 370 } 371 372 if (cluster_count == 0) { 373 return -EINVAL; 374 } 375 tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t)); 376 if (tmp == NULL) { 377 return -ENOMEM; 378 } 379 blob->active.clusters = tmp; 380 blob->active.cluster_array_size = cluster_count; 381 382 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 383 for (j = 0; j < desc_extent->extents[i].length; j++) { 384 if (desc_extent->extents[i].cluster_idx != 0) { 385 blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs, 386 desc_extent->extents[i].cluster_idx + j); 387 } else if (spdk_blob_is_thin_provisioned(blob)) { 388 blob->active.clusters[blob->active.num_clusters++] = 0; 389 } else { 390 return -EINVAL; 391 } 392 } 393 } 394 395 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 396 int rc; 397 398 rc = _spdk_blob_deserialize_xattr(blob, 399 (struct spdk_blob_md_descriptor_xattr *) desc, false); 400 if (rc != 0) { 401 return rc; 402 } 403 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 404 int rc; 405 406 rc = _spdk_blob_deserialize_xattr(blob, 407 (struct spdk_blob_md_descriptor_xattr *) desc, true); 408 if (rc != 0) { 409 return rc; 410 } 411 } else { 412 /* Unrecognized descriptor type. Do not fail - just continue to the 413 * next descriptor. If this descriptor is associated with some feature 414 * defined in a newer version of blobstore, that version of blobstore 415 * should create and set an associated feature flag to specify if this 416 * blob can be loaded or not. 417 */ 418 } 419 420 /* Advance to the next descriptor */ 421 cur_desc += sizeof(*desc) + desc->length; 422 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 423 break; 424 } 425 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 426 } 427 428 return 0; 429 } 430 431 static int 432 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count, 433 struct spdk_blob *blob) 434 { 435 const struct spdk_blob_md_page *page; 436 uint32_t i; 437 int rc; 438 439 assert(page_count > 0); 440 assert(pages[0].sequence_num == 0); 441 assert(blob != NULL); 442 assert(blob->state == SPDK_BLOB_STATE_LOADING); 443 assert(blob->active.clusters == NULL); 444 assert(blob->state == SPDK_BLOB_STATE_LOADING); 445 446 /* The blobid provided doesn't match what's in the MD, this can 447 * happen for example if a bogus blobid is passed in through open. 448 */ 449 if (blob->id != pages[0].id) { 450 SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n", 451 blob->id, pages[0].id); 452 return -ENOENT; 453 } 454 455 for (i = 0; i < page_count; i++) { 456 page = &pages[i]; 457 458 assert(page->id == blob->id); 459 assert(page->sequence_num == i); 460 461 rc = _spdk_blob_parse_page(page, blob); 462 if (rc != 0) { 463 return rc; 464 } 465 } 466 467 return 0; 468 } 469 470 static int 471 _spdk_blob_serialize_add_page(const struct spdk_blob *blob, 472 struct spdk_blob_md_page **pages, 473 uint32_t *page_count, 474 struct spdk_blob_md_page **last_page) 475 { 476 struct spdk_blob_md_page *page; 477 478 assert(pages != NULL); 479 assert(page_count != NULL); 480 481 if (*page_count == 0) { 482 assert(*pages == NULL); 483 *page_count = 1; 484 *pages = spdk_dma_malloc(SPDK_BS_PAGE_SIZE, 485 SPDK_BS_PAGE_SIZE, 486 NULL); 487 } else { 488 assert(*pages != NULL); 489 (*page_count)++; 490 *pages = spdk_dma_realloc(*pages, 491 SPDK_BS_PAGE_SIZE * (*page_count), 492 SPDK_BS_PAGE_SIZE, 493 NULL); 494 } 495 496 if (*pages == NULL) { 497 *page_count = 0; 498 *last_page = NULL; 499 return -ENOMEM; 500 } 501 502 page = &(*pages)[*page_count - 1]; 503 memset(page, 0, sizeof(*page)); 504 page->id = blob->id; 505 page->sequence_num = *page_count - 1; 506 page->next = SPDK_INVALID_MD_PAGE; 507 *last_page = page; 508 509 return 0; 510 } 511 512 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor. 513 * Update required_sz on both success and failure. 514 * 515 */ 516 static int 517 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr, 518 uint8_t *buf, size_t buf_sz, 519 size_t *required_sz, bool internal) 520 { 521 struct spdk_blob_md_descriptor_xattr *desc; 522 523 *required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) + 524 strlen(xattr->name) + 525 xattr->value_len; 526 527 if (buf_sz < *required_sz) { 528 return -1; 529 } 530 531 desc = (struct spdk_blob_md_descriptor_xattr *)buf; 532 533 desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR; 534 desc->length = sizeof(desc->name_length) + 535 sizeof(desc->value_length) + 536 strlen(xattr->name) + 537 xattr->value_len; 538 desc->name_length = strlen(xattr->name); 539 desc->value_length = xattr->value_len; 540 541 memcpy(desc->name, xattr->name, desc->name_length); 542 memcpy((void *)((uintptr_t)desc->name + desc->name_length), 543 xattr->value, 544 desc->value_length); 545 546 return 0; 547 } 548 549 static void 550 _spdk_blob_serialize_extent(const struct spdk_blob *blob, 551 uint64_t start_cluster, uint64_t *next_cluster, 552 uint8_t *buf, size_t buf_sz) 553 { 554 struct spdk_blob_md_descriptor_extent *desc; 555 size_t cur_sz; 556 uint64_t i, extent_idx; 557 uint32_t lba, lba_per_cluster, lba_count; 558 559 /* The buffer must have room for at least one extent */ 560 cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]); 561 if (buf_sz < cur_sz) { 562 *next_cluster = start_cluster; 563 return; 564 } 565 566 desc = (struct spdk_blob_md_descriptor_extent *)buf; 567 desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT; 568 569 lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1); 570 571 lba = blob->active.clusters[start_cluster]; 572 lba_count = lba_per_cluster; 573 extent_idx = 0; 574 for (i = start_cluster + 1; i < blob->active.num_clusters; i++) { 575 if ((lba + lba_count) == blob->active.clusters[i]) { 576 lba_count += lba_per_cluster; 577 continue; 578 } 579 desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 580 desc->extents[extent_idx].length = lba_count / lba_per_cluster; 581 extent_idx++; 582 583 cur_sz += sizeof(desc->extents[extent_idx]); 584 585 if (buf_sz < cur_sz) { 586 /* If we ran out of buffer space, return */ 587 desc->length = sizeof(desc->extents[0]) * extent_idx; 588 *next_cluster = i; 589 return; 590 } 591 592 lba = blob->active.clusters[i]; 593 lba_count = lba_per_cluster; 594 } 595 596 desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 597 desc->extents[extent_idx].length = lba_count / lba_per_cluster; 598 extent_idx++; 599 600 desc->length = sizeof(desc->extents[0]) * extent_idx; 601 *next_cluster = blob->active.num_clusters; 602 603 return; 604 } 605 606 static void 607 _spdk_blob_serialize_flags(const struct spdk_blob *blob, 608 uint8_t *buf, size_t *buf_sz) 609 { 610 struct spdk_blob_md_descriptor_flags *desc; 611 612 /* 613 * Flags get serialized first, so we should always have room for the flags 614 * descriptor. 615 */ 616 assert(*buf_sz >= sizeof(*desc)); 617 618 desc = (struct spdk_blob_md_descriptor_flags *)buf; 619 desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS; 620 desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor); 621 desc->invalid_flags = blob->invalid_flags; 622 desc->data_ro_flags = blob->data_ro_flags; 623 desc->md_ro_flags = blob->md_ro_flags; 624 625 *buf_sz -= sizeof(*desc); 626 } 627 628 static int 629 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob, 630 const struct spdk_xattr_tailq *xattrs, bool internal, 631 struct spdk_blob_md_page **pages, 632 struct spdk_blob_md_page *cur_page, 633 uint32_t *page_count, uint8_t **buf, 634 size_t *remaining_sz) 635 { 636 const struct spdk_xattr *xattr; 637 int rc; 638 639 TAILQ_FOREACH(xattr, xattrs, link) { 640 size_t required_sz = 0; 641 642 rc = _spdk_blob_serialize_xattr(xattr, 643 *buf, *remaining_sz, 644 &required_sz, internal); 645 if (rc < 0) { 646 /* Need to add a new page to the chain */ 647 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, 648 &cur_page); 649 if (rc < 0) { 650 spdk_dma_free(*pages); 651 *pages = NULL; 652 *page_count = 0; 653 return rc; 654 } 655 656 *buf = (uint8_t *)cur_page->descriptors; 657 *remaining_sz = sizeof(cur_page->descriptors); 658 659 /* Try again */ 660 required_sz = 0; 661 rc = _spdk_blob_serialize_xattr(xattr, 662 *buf, *remaining_sz, 663 &required_sz, internal); 664 665 if (rc < 0) { 666 spdk_dma_free(*pages); 667 *pages = NULL; 668 *page_count = 0; 669 return -1; 670 } 671 } 672 673 *remaining_sz -= required_sz; 674 *buf += required_sz; 675 } 676 677 return 0; 678 } 679 680 static int 681 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages, 682 uint32_t *page_count) 683 { 684 struct spdk_blob_md_page *cur_page; 685 int rc; 686 uint8_t *buf; 687 size_t remaining_sz; 688 uint64_t last_cluster; 689 690 assert(pages != NULL); 691 assert(page_count != NULL); 692 assert(blob != NULL); 693 assert(blob->state == SPDK_BLOB_STATE_DIRTY); 694 695 *pages = NULL; 696 *page_count = 0; 697 698 /* A blob always has at least 1 page, even if it has no descriptors */ 699 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page); 700 if (rc < 0) { 701 return rc; 702 } 703 704 buf = (uint8_t *)cur_page->descriptors; 705 remaining_sz = sizeof(cur_page->descriptors); 706 707 /* Serialize flags */ 708 _spdk_blob_serialize_flags(blob, buf, &remaining_sz); 709 buf += sizeof(struct spdk_blob_md_descriptor_flags); 710 711 /* Serialize xattrs */ 712 rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false, 713 pages, cur_page, page_count, &buf, &remaining_sz); 714 if (rc < 0) { 715 return rc; 716 } 717 718 /* Serialize internal xattrs */ 719 rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true, 720 pages, cur_page, page_count, &buf, &remaining_sz); 721 if (rc < 0) { 722 return rc; 723 } 724 725 /* Serialize extents */ 726 last_cluster = 0; 727 while (last_cluster < blob->active.num_clusters) { 728 _spdk_blob_serialize_extent(blob, last_cluster, &last_cluster, 729 buf, remaining_sz); 730 731 if (last_cluster == blob->active.num_clusters) { 732 break; 733 } 734 735 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, 736 &cur_page); 737 if (rc < 0) { 738 return rc; 739 } 740 741 buf = (uint8_t *)cur_page->descriptors; 742 remaining_sz = sizeof(cur_page->descriptors); 743 } 744 745 return 0; 746 } 747 748 struct spdk_blob_load_ctx { 749 struct spdk_blob *blob; 750 751 struct spdk_blob_md_page *pages; 752 uint32_t num_pages; 753 754 spdk_bs_sequence_cpl cb_fn; 755 void *cb_arg; 756 }; 757 758 static uint32_t 759 _spdk_blob_md_page_calc_crc(void *page) 760 { 761 uint32_t crc; 762 763 crc = BLOB_CRC32C_INITIAL; 764 crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc); 765 crc ^= BLOB_CRC32C_INITIAL; 766 767 return crc; 768 769 } 770 771 static void 772 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 773 { 774 struct spdk_blob_load_ctx *ctx = cb_arg; 775 struct spdk_blob *blob = ctx->blob; 776 struct spdk_blob_md_page *page; 777 int rc; 778 uint32_t crc; 779 780 page = &ctx->pages[ctx->num_pages - 1]; 781 crc = _spdk_blob_md_page_calc_crc(page); 782 if (crc != page->crc) { 783 SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages); 784 _spdk_blob_free(blob); 785 ctx->cb_fn(seq, NULL, -EINVAL); 786 spdk_dma_free(ctx->pages); 787 free(ctx); 788 return; 789 } 790 791 if (page->next != SPDK_INVALID_MD_PAGE) { 792 uint32_t next_page = page->next; 793 uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page); 794 795 796 assert(next_lba < (blob->bs->md_start + blob->bs->md_len)); 797 798 /* Read the next page */ 799 ctx->num_pages++; 800 ctx->pages = spdk_dma_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages), 801 sizeof(*page), NULL); 802 if (ctx->pages == NULL) { 803 ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM); 804 free(ctx); 805 return; 806 } 807 808 spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1], 809 next_lba, 810 _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)), 811 _spdk_blob_load_cpl, ctx); 812 return; 813 } 814 815 /* Parse the pages */ 816 rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob); 817 if (rc) { 818 _spdk_blob_free(blob); 819 ctx->cb_fn(seq, NULL, rc); 820 spdk_dma_free(ctx->pages); 821 free(ctx); 822 return; 823 } 824 825 if (spdk_blob_is_thin_provisioned(blob) == true) { 826 blob->back_bs_dev = spdk_bs_create_zeroes_dev(); 827 } 828 829 _spdk_blob_mark_clean(blob); 830 831 ctx->cb_fn(seq, ctx->cb_arg, rc); 832 833 /* Free the memory */ 834 spdk_dma_free(ctx->pages); 835 free(ctx); 836 } 837 838 /* Load a blob from disk given a blobid */ 839 static void 840 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 841 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 842 { 843 struct spdk_blob_load_ctx *ctx; 844 struct spdk_blob_store *bs; 845 uint32_t page_num; 846 uint64_t lba; 847 848 assert(blob != NULL); 849 assert(blob->state == SPDK_BLOB_STATE_CLEAN || 850 blob->state == SPDK_BLOB_STATE_DIRTY); 851 852 bs = blob->bs; 853 854 ctx = calloc(1, sizeof(*ctx)); 855 if (!ctx) { 856 cb_fn(seq, cb_arg, -ENOMEM); 857 return; 858 } 859 860 ctx->blob = blob; 861 ctx->pages = spdk_dma_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, 862 SPDK_BS_PAGE_SIZE, NULL); 863 if (!ctx->pages) { 864 free(ctx); 865 cb_fn(seq, cb_arg, -ENOMEM); 866 return; 867 } 868 ctx->num_pages = 1; 869 ctx->cb_fn = cb_fn; 870 ctx->cb_arg = cb_arg; 871 872 page_num = _spdk_bs_blobid_to_page(blob->id); 873 lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num); 874 875 blob->state = SPDK_BLOB_STATE_LOADING; 876 877 spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba, 878 _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE), 879 _spdk_blob_load_cpl, ctx); 880 } 881 882 struct spdk_blob_persist_ctx { 883 struct spdk_blob *blob; 884 885 struct spdk_blob_md_page *pages; 886 887 uint64_t idx; 888 889 spdk_bs_sequence_cpl cb_fn; 890 void *cb_arg; 891 }; 892 893 static void 894 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 895 { 896 struct spdk_blob_persist_ctx *ctx = cb_arg; 897 struct spdk_blob *blob = ctx->blob; 898 899 if (bserrno == 0) { 900 _spdk_blob_mark_clean(blob); 901 } 902 903 /* Call user callback */ 904 ctx->cb_fn(seq, ctx->cb_arg, bserrno); 905 906 /* Free the memory */ 907 spdk_dma_free(ctx->pages); 908 free(ctx); 909 } 910 911 static void 912 _spdk_blob_persist_unmap_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 913 { 914 struct spdk_blob_persist_ctx *ctx = cb_arg; 915 struct spdk_blob *blob = ctx->blob; 916 struct spdk_blob_store *bs = blob->bs; 917 void *tmp; 918 size_t i; 919 920 /* Release all clusters that were truncated */ 921 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 922 uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]); 923 924 /* Nothing to release if it was not allocated */ 925 if (blob->active.clusters[i] != 0) { 926 _spdk_bs_release_cluster(bs, cluster_num); 927 } 928 } 929 930 if (blob->active.num_clusters == 0) { 931 free(blob->active.clusters); 932 blob->active.clusters = NULL; 933 blob->active.cluster_array_size = 0; 934 } else { 935 tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters); 936 assert(tmp != NULL); 937 blob->active.clusters = tmp; 938 blob->active.cluster_array_size = blob->active.num_clusters; 939 } 940 941 _spdk_blob_persist_complete(seq, ctx, bserrno); 942 } 943 944 static void 945 _spdk_blob_persist_unmap_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 946 { 947 struct spdk_blob_persist_ctx *ctx = cb_arg; 948 struct spdk_blob *blob = ctx->blob; 949 struct spdk_blob_store *bs = blob->bs; 950 spdk_bs_batch_t *batch; 951 size_t i; 952 uint64_t lba; 953 uint32_t lba_count; 954 955 /* Clusters don't move around in blobs. The list shrinks or grows 956 * at the end, but no changes ever occur in the middle of the list. 957 */ 958 959 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_unmap_clusters_cpl, ctx); 960 961 /* Unmap all clusters that were truncated */ 962 lba = 0; 963 lba_count = 0; 964 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 965 uint64_t next_lba = blob->active.clusters[i]; 966 uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1); 967 968 if (next_lba > 0 && (lba + lba_count) == next_lba) { 969 /* This cluster is contiguous with the previous one. */ 970 lba_count += next_lba_count; 971 continue; 972 } 973 974 /* This cluster is not contiguous with the previous one. */ 975 976 /* If a run of LBAs previously existing, send them 977 * as an unmap. 978 */ 979 if (lba_count > 0) { 980 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 981 } 982 983 /* Start building the next batch */ 984 lba = next_lba; 985 if (next_lba > 0) { 986 lba_count = next_lba_count; 987 } else { 988 lba_count = 0; 989 } 990 } 991 992 /* If we ended with a contiguous set of LBAs, send the unmap now */ 993 if (lba_count > 0) { 994 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 995 } 996 997 spdk_bs_batch_close(batch); 998 } 999 1000 static void 1001 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1002 { 1003 struct spdk_blob_persist_ctx *ctx = cb_arg; 1004 struct spdk_blob *blob = ctx->blob; 1005 struct spdk_blob_store *bs = blob->bs; 1006 size_t i; 1007 1008 /* This loop starts at 1 because the first page is special and handled 1009 * below. The pages (except the first) are never written in place, 1010 * so any pages in the clean list must be zeroed. 1011 */ 1012 for (i = 1; i < blob->clean.num_pages; i++) { 1013 spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]); 1014 } 1015 1016 if (blob->active.num_pages == 0) { 1017 uint32_t page_num; 1018 1019 page_num = _spdk_bs_blobid_to_page(blob->id); 1020 spdk_bit_array_clear(bs->used_md_pages, page_num); 1021 } 1022 1023 /* Move on to unmapping clusters */ 1024 _spdk_blob_persist_unmap_clusters(seq, ctx, 0); 1025 } 1026 1027 static void 1028 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1029 { 1030 struct spdk_blob_persist_ctx *ctx = cb_arg; 1031 struct spdk_blob *blob = ctx->blob; 1032 struct spdk_blob_store *bs = blob->bs; 1033 uint64_t lba; 1034 uint32_t lba_count; 1035 spdk_bs_batch_t *batch; 1036 size_t i; 1037 1038 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx); 1039 1040 lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE); 1041 1042 /* This loop starts at 1 because the first page is special and handled 1043 * below. The pages (except the first) are never written in place, 1044 * so any pages in the clean list must be zeroed. 1045 */ 1046 for (i = 1; i < blob->clean.num_pages; i++) { 1047 lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]); 1048 1049 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1050 } 1051 1052 /* The first page will only be zeroed if this is a delete. */ 1053 if (blob->active.num_pages == 0) { 1054 uint32_t page_num; 1055 1056 /* The first page in the metadata goes where the blobid indicates */ 1057 page_num = _spdk_bs_blobid_to_page(blob->id); 1058 lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num); 1059 1060 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1061 } 1062 1063 spdk_bs_batch_close(batch); 1064 } 1065 1066 static void 1067 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1068 { 1069 struct spdk_blob_persist_ctx *ctx = cb_arg; 1070 struct spdk_blob *blob = ctx->blob; 1071 struct spdk_blob_store *bs = blob->bs; 1072 uint64_t lba; 1073 uint32_t lba_count; 1074 struct spdk_blob_md_page *page; 1075 1076 if (blob->active.num_pages == 0) { 1077 /* Move on to the next step */ 1078 _spdk_blob_persist_zero_pages(seq, ctx, 0); 1079 return; 1080 } 1081 1082 lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page)); 1083 1084 page = &ctx->pages[0]; 1085 /* The first page in the metadata goes where the blobid indicates */ 1086 lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id)); 1087 1088 spdk_bs_sequence_write_dev(seq, page, lba, lba_count, 1089 _spdk_blob_persist_zero_pages, ctx); 1090 } 1091 1092 static void 1093 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1094 { 1095 struct spdk_blob_persist_ctx *ctx = cb_arg; 1096 struct spdk_blob *blob = ctx->blob; 1097 struct spdk_blob_store *bs = blob->bs; 1098 uint64_t lba; 1099 uint32_t lba_count; 1100 struct spdk_blob_md_page *page; 1101 spdk_bs_batch_t *batch; 1102 size_t i; 1103 1104 /* Clusters don't move around in blobs. The list shrinks or grows 1105 * at the end, but no changes ever occur in the middle of the list. 1106 */ 1107 1108 lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page)); 1109 1110 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx); 1111 1112 /* This starts at 1. The root page is not written until 1113 * all of the others are finished 1114 */ 1115 for (i = 1; i < blob->active.num_pages; i++) { 1116 page = &ctx->pages[i]; 1117 assert(page->sequence_num == i); 1118 1119 lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]); 1120 1121 spdk_bs_batch_write_dev(batch, page, lba, lba_count); 1122 } 1123 1124 spdk_bs_batch_close(batch); 1125 } 1126 1127 static int 1128 _spdk_resize_blob(struct spdk_blob *blob, uint64_t sz) 1129 { 1130 uint64_t i; 1131 uint64_t *tmp; 1132 uint64_t lfc; /* lowest free cluster */ 1133 uint64_t num_clusters; 1134 struct spdk_blob_store *bs; 1135 1136 bs = blob->bs; 1137 1138 assert(blob->state != SPDK_BLOB_STATE_LOADING); 1139 1140 if (blob->active.num_clusters == sz) { 1141 return 0; 1142 } 1143 1144 if (blob->active.num_clusters < blob->active.cluster_array_size) { 1145 /* If this blob was resized to be larger, then smaller, then 1146 * larger without syncing, then the cluster array already 1147 * contains spare assigned clusters we can use. 1148 */ 1149 num_clusters = spdk_min(blob->active.cluster_array_size, 1150 sz); 1151 } else { 1152 num_clusters = blob->active.num_clusters; 1153 } 1154 1155 /* Do two passes - one to verify that we can obtain enough clusters 1156 * and another to actually claim them. 1157 */ 1158 1159 if (spdk_blob_is_thin_provisioned(blob) == false) { 1160 lfc = 0; 1161 for (i = num_clusters; i < sz; i++) { 1162 lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc); 1163 if (lfc >= bs->total_clusters) { 1164 /* No more free clusters. Cannot satisfy the request */ 1165 return -ENOSPC; 1166 } 1167 lfc++; 1168 } 1169 } 1170 1171 if (sz > num_clusters) { 1172 /* Expand the cluster array if necessary. 1173 * We only shrink the array when persisting. 1174 */ 1175 tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz); 1176 if (sz > 0 && tmp == NULL) { 1177 return -ENOMEM; 1178 } 1179 memset(tmp + blob->active.cluster_array_size, 0, 1180 sizeof(uint64_t) * (sz - blob->active.cluster_array_size)); 1181 blob->active.clusters = tmp; 1182 blob->active.cluster_array_size = sz; 1183 } 1184 1185 blob->state = SPDK_BLOB_STATE_DIRTY; 1186 1187 if (spdk_blob_is_thin_provisioned(blob) == false) { 1188 lfc = 0; 1189 for (i = num_clusters; i < sz; i++) { 1190 _spdk_bs_allocate_cluster(blob, i, &lfc, true); 1191 lfc++; 1192 } 1193 } 1194 1195 blob->active.num_clusters = sz; 1196 1197 return 0; 1198 } 1199 1200 /* Write a blob to disk */ 1201 static void 1202 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 1203 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 1204 { 1205 struct spdk_blob_persist_ctx *ctx; 1206 int rc; 1207 uint64_t i; 1208 uint32_t page_num; 1209 struct spdk_blob_store *bs; 1210 1211 assert(blob != NULL); 1212 assert(blob->state == SPDK_BLOB_STATE_CLEAN || 1213 blob->state == SPDK_BLOB_STATE_DIRTY); 1214 1215 if (blob->state == SPDK_BLOB_STATE_CLEAN) { 1216 cb_fn(seq, cb_arg, 0); 1217 return; 1218 } 1219 1220 bs = blob->bs; 1221 1222 ctx = calloc(1, sizeof(*ctx)); 1223 if (!ctx) { 1224 cb_fn(seq, cb_arg, -ENOMEM); 1225 return; 1226 } 1227 ctx->blob = blob; 1228 ctx->cb_fn = cb_fn; 1229 ctx->cb_arg = cb_arg; 1230 1231 if (blob->active.num_pages == 0) { 1232 /* This is the signal that the blob should be deleted. 1233 * Immediately jump to the clean up routine. */ 1234 assert(blob->clean.num_pages > 0); 1235 ctx->idx = blob->clean.num_pages - 1; 1236 blob->state = SPDK_BLOB_STATE_CLEAN; 1237 _spdk_blob_persist_zero_pages(seq, ctx, 0); 1238 return; 1239 1240 } 1241 1242 /* Generate the new metadata */ 1243 rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages); 1244 if (rc < 0) { 1245 free(ctx); 1246 cb_fn(seq, cb_arg, rc); 1247 return; 1248 } 1249 1250 assert(blob->active.num_pages >= 1); 1251 1252 /* Resize the cache of page indices */ 1253 blob->active.pages = realloc(blob->active.pages, 1254 blob->active.num_pages * sizeof(*blob->active.pages)); 1255 if (!blob->active.pages) { 1256 free(ctx); 1257 cb_fn(seq, cb_arg, -ENOMEM); 1258 return; 1259 } 1260 1261 /* Assign this metadata to pages. This requires two passes - 1262 * one to verify that there are enough pages and a second 1263 * to actually claim them. */ 1264 page_num = 0; 1265 /* Note that this loop starts at one. The first page location is fixed by the blobid. */ 1266 for (i = 1; i < blob->active.num_pages; i++) { 1267 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 1268 if (page_num >= spdk_bit_array_capacity(bs->used_md_pages)) { 1269 spdk_dma_free(ctx->pages); 1270 free(ctx); 1271 cb_fn(seq, cb_arg, -ENOMEM); 1272 return; 1273 } 1274 page_num++; 1275 } 1276 1277 page_num = 0; 1278 blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id); 1279 for (i = 1; i < blob->active.num_pages; i++) { 1280 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 1281 ctx->pages[i - 1].next = page_num; 1282 /* Now that previous metadata page is complete, calculate the crc for it. */ 1283 ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]); 1284 blob->active.pages[i] = page_num; 1285 spdk_bit_array_set(bs->used_md_pages, page_num); 1286 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id); 1287 page_num++; 1288 } 1289 ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]); 1290 /* Start writing the metadata from last page to first */ 1291 ctx->idx = blob->active.num_pages - 1; 1292 blob->state = SPDK_BLOB_STATE_CLEAN; 1293 _spdk_blob_persist_write_page_chain(seq, ctx, 0); 1294 } 1295 1296 struct spdk_blob_copy_cluster_ctx { 1297 struct spdk_blob *blob; 1298 uint8_t *buf; 1299 uint64_t page; 1300 uint64_t new_cluster; 1301 spdk_bs_sequence_t *seq; 1302 }; 1303 1304 static void 1305 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno) 1306 { 1307 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1308 struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq; 1309 TAILQ_HEAD(, spdk_bs_request_set) requests; 1310 spdk_bs_user_op_t *op; 1311 1312 TAILQ_INIT(&requests); 1313 TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link); 1314 1315 while (!TAILQ_EMPTY(&requests)) { 1316 op = TAILQ_FIRST(&requests); 1317 TAILQ_REMOVE(&requests, op, link); 1318 if (bserrno == 0) { 1319 spdk_bs_user_op_execute(op); 1320 } else { 1321 spdk_bs_user_op_abort(op); 1322 } 1323 } 1324 1325 spdk_dma_free(ctx->buf); 1326 free(ctx); 1327 } 1328 1329 static void 1330 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno) 1331 { 1332 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1333 1334 if (bserrno) { 1335 uint32_t cluster_number; 1336 1337 if (bserrno == -EEXIST) { 1338 /* The metadata insert failed because another thread 1339 * allocated the cluster first. Free our cluster 1340 * but continue without error. */ 1341 bserrno = 0; 1342 } 1343 1344 cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page); 1345 _spdk_bs_release_cluster(ctx->blob->bs, cluster_number); 1346 } 1347 1348 spdk_bs_sequence_finish(ctx->seq, bserrno); 1349 } 1350 1351 static void 1352 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1353 { 1354 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1355 uint32_t cluster_number; 1356 1357 if (bserrno) { 1358 /* The write failed, so jump to the final completion handler */ 1359 spdk_bs_sequence_finish(seq, bserrno); 1360 return; 1361 } 1362 1363 cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page); 1364 1365 _spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 1366 _spdk_blob_insert_cluster_cpl, ctx); 1367 } 1368 1369 static void 1370 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1371 { 1372 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1373 1374 if (bserrno != 0) { 1375 /* The read failed, so jump to the final completion handler */ 1376 spdk_bs_sequence_finish(seq, bserrno); 1377 return; 1378 } 1379 1380 /* Write whole cluster */ 1381 spdk_bs_sequence_write_dev(seq, ctx->buf, 1382 _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster), 1383 _spdk_bs_cluster_to_lba(ctx->blob->bs, 1), 1384 _spdk_blob_write_copy_cpl, ctx); 1385 } 1386 1387 static void 1388 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob, 1389 struct spdk_io_channel *_ch, 1390 uint64_t offset, spdk_bs_user_op_t *op) 1391 { 1392 struct spdk_bs_cpl cpl; 1393 struct spdk_bs_channel *ch; 1394 struct spdk_blob_copy_cluster_ctx *ctx; 1395 uint32_t cluster_start_page; 1396 uint32_t cluster_number; 1397 int rc; 1398 1399 ch = spdk_io_channel_get_ctx(_ch); 1400 1401 if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) { 1402 /* There are already operations pending. Queue this user op 1403 * and return because it will be re-executed when the outstanding 1404 * cluster allocation completes. */ 1405 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 1406 return; 1407 } 1408 1409 /* Round the page offset down to the first page in the cluster */ 1410 cluster_start_page = _spdk_bs_page_to_cluster_start(blob, offset); 1411 1412 /* Calculate which index in the metadata cluster array the corresponding 1413 * cluster is supposed to be at. */ 1414 cluster_number = _spdk_bs_page_to_cluster(blob->bs, cluster_start_page); 1415 1416 ctx = calloc(1, sizeof(*ctx)); 1417 if (!ctx) { 1418 spdk_bs_user_op_abort(op); 1419 return; 1420 } 1421 1422 assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0); 1423 1424 ctx->blob = blob; 1425 ctx->page = cluster_start_page; 1426 1427 ctx->buf = spdk_dma_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, NULL); 1428 if (!ctx->buf) { 1429 SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n", 1430 blob->bs->cluster_sz); 1431 free(ctx); 1432 spdk_bs_user_op_abort(op); 1433 return; 1434 } 1435 1436 rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false); 1437 if (rc != 0) { 1438 spdk_dma_free(ctx->buf); 1439 free(ctx); 1440 spdk_bs_user_op_abort(op); 1441 return; 1442 } 1443 1444 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1445 cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl; 1446 cpl.u.blob_basic.cb_arg = ctx; 1447 1448 ctx->seq = spdk_bs_sequence_start(_ch, &cpl); 1449 if (!ctx->seq) { 1450 _spdk_bs_release_cluster(blob->bs, ctx->new_cluster); 1451 spdk_dma_free(ctx->buf); 1452 free(ctx); 1453 spdk_bs_user_op_abort(op); 1454 return; 1455 } 1456 1457 /* Queue the user op to block other incoming operations */ 1458 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 1459 1460 /* Read cluster from backing device */ 1461 spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf, 1462 _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page), 1463 _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz), 1464 _spdk_blob_write_copy, ctx); 1465 } 1466 1467 static void 1468 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t page, uint64_t length, 1469 uint64_t *lba, uint32_t *lba_count) 1470 { 1471 *lba_count = _spdk_bs_page_to_lba(blob->bs, length); 1472 1473 if (!_spdk_bs_page_is_allocated(blob, page)) { 1474 assert(blob->back_bs_dev != NULL); 1475 *lba = _spdk_bs_dev_page_to_lba(blob->back_bs_dev, page); 1476 *lba_count = _spdk_bs_blob_lba_to_back_dev_lba(blob, *lba_count); 1477 } else { 1478 *lba = _spdk_bs_blob_page_to_lba(blob, page); 1479 } 1480 } 1481 1482 static void 1483 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob, 1484 void *payload, uint64_t offset, uint64_t length, 1485 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1486 { 1487 spdk_bs_batch_t *batch; 1488 struct spdk_bs_cpl cpl; 1489 uint64_t op_length; 1490 uint8_t *buf; 1491 1492 assert(blob != NULL); 1493 1494 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1495 cpl.u.blob_basic.cb_fn = cb_fn; 1496 cpl.u.blob_basic.cb_arg = cb_arg; 1497 1498 batch = spdk_bs_batch_open(ch, &cpl); 1499 if (!batch) { 1500 cb_fn(cb_arg, -ENOMEM); 1501 return; 1502 } 1503 1504 buf = payload; 1505 while (length > 0) { 1506 op_length = spdk_min(length, _spdk_bs_num_pages_to_cluster_boundary(blob, offset)); 1507 1508 switch (op_type) { 1509 case SPDK_BLOB_READ: 1510 spdk_bs_batch_read_blob(batch, blob, buf, offset, op_length); 1511 break; 1512 case SPDK_BLOB_WRITE: 1513 spdk_bs_batch_write_blob(batch, blob, buf, offset, op_length); 1514 break; 1515 case SPDK_BLOB_UNMAP: 1516 spdk_bs_batch_unmap_blob(batch, blob, offset, op_length); 1517 break; 1518 case SPDK_BLOB_WRITE_ZEROES: 1519 spdk_bs_batch_write_zeroes_blob(batch, blob, offset, op_length); 1520 break; 1521 case SPDK_BLOB_READV: 1522 case SPDK_BLOB_WRITEV: 1523 SPDK_ERRLOG("readv/write not valid for %s\n", __func__); 1524 break; 1525 } 1526 1527 length -= op_length; 1528 offset += op_length; 1529 if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) { 1530 buf += op_length * SPDK_BS_PAGE_SIZE; 1531 } 1532 } 1533 1534 spdk_bs_batch_close(batch); 1535 } 1536 1537 static void 1538 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob, 1539 void *payload, uint64_t offset, uint64_t length, 1540 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1541 { 1542 struct spdk_bs_cpl cpl; 1543 uint64_t lba; 1544 uint32_t lba_count; 1545 1546 assert(blob != NULL); 1547 1548 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1549 cpl.u.blob_basic.cb_fn = cb_fn; 1550 cpl.u.blob_basic.cb_arg = cb_arg; 1551 1552 _spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 1553 1554 switch (op_type) { 1555 case SPDK_BLOB_READ: { 1556 spdk_bs_batch_t *batch; 1557 1558 batch = spdk_bs_batch_open(_ch, &cpl); 1559 if (!batch) { 1560 cb_fn(cb_arg, -ENOMEM); 1561 return; 1562 } 1563 1564 if (_spdk_bs_page_is_allocated(blob, offset)) { 1565 /* Read from the blob */ 1566 spdk_bs_batch_read_dev(batch, payload, lba, lba_count); 1567 } else { 1568 /* Read from the backing block device */ 1569 spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count); 1570 } 1571 1572 spdk_bs_batch_close(batch); 1573 break; 1574 } 1575 case SPDK_BLOB_WRITE: 1576 case SPDK_BLOB_WRITE_ZEROES: { 1577 if (_spdk_bs_page_is_allocated(blob, offset)) { 1578 /* Write to the blob */ 1579 spdk_bs_batch_t *batch; 1580 1581 batch = spdk_bs_batch_open(_ch, &cpl); 1582 if (!batch) { 1583 cb_fn(cb_arg, -ENOMEM); 1584 return; 1585 } 1586 1587 if (op_type == SPDK_BLOB_WRITE) { 1588 spdk_bs_batch_write_dev(batch, payload, lba, lba_count); 1589 } else { 1590 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1591 } 1592 1593 spdk_bs_batch_close(batch); 1594 } else { 1595 /* Queue this operation and allocate the cluster */ 1596 spdk_bs_user_op_t *op; 1597 1598 op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 1599 if (!op) { 1600 cb_fn(cb_arg, -ENOMEM); 1601 return; 1602 } 1603 1604 _spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op); 1605 } 1606 break; 1607 } 1608 case SPDK_BLOB_UNMAP: { 1609 spdk_bs_batch_t *batch; 1610 1611 batch = spdk_bs_batch_open(_ch, &cpl); 1612 if (!batch) { 1613 cb_fn(cb_arg, -ENOMEM); 1614 return; 1615 } 1616 1617 if (_spdk_bs_page_is_allocated(blob, offset)) { 1618 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 1619 } 1620 1621 spdk_bs_batch_close(batch); 1622 break; 1623 } 1624 case SPDK_BLOB_READV: 1625 case SPDK_BLOB_WRITEV: 1626 SPDK_ERRLOG("readv/write not valid\n"); 1627 cb_fn(cb_arg, -EINVAL); 1628 break; 1629 } 1630 } 1631 1632 static void 1633 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel, 1634 void *payload, uint64_t offset, uint64_t length, 1635 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1636 { 1637 assert(blob != NULL); 1638 1639 if (blob->data_ro && op_type != SPDK_BLOB_READ) { 1640 cb_fn(cb_arg, -EPERM); 1641 return; 1642 } 1643 1644 if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) { 1645 cb_fn(cb_arg, -EINVAL); 1646 return; 1647 } 1648 1649 if (length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset)) { 1650 _spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length, 1651 cb_fn, cb_arg, op_type); 1652 } else { 1653 _spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length, 1654 cb_fn, cb_arg, op_type); 1655 } 1656 } 1657 1658 struct rw_iov_ctx { 1659 struct spdk_blob *blob; 1660 struct spdk_io_channel *channel; 1661 spdk_blob_op_complete cb_fn; 1662 void *cb_arg; 1663 bool read; 1664 int iovcnt; 1665 struct iovec *orig_iov; 1666 uint64_t page_offset; 1667 uint64_t pages_remaining; 1668 uint64_t pages_done; 1669 struct iovec iov[0]; 1670 }; 1671 1672 static void 1673 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1674 { 1675 assert(cb_arg == NULL); 1676 spdk_bs_sequence_finish(seq, bserrno); 1677 } 1678 1679 static void 1680 _spdk_rw_iov_split_next(void *cb_arg, int bserrno) 1681 { 1682 struct rw_iov_ctx *ctx = cb_arg; 1683 struct spdk_blob *blob = ctx->blob; 1684 struct iovec *iov, *orig_iov; 1685 int iovcnt; 1686 size_t orig_iovoff; 1687 uint64_t page_count, pages_to_boundary, page_offset; 1688 uint64_t byte_count; 1689 1690 if (bserrno != 0 || ctx->pages_remaining == 0) { 1691 ctx->cb_fn(ctx->cb_arg, bserrno); 1692 free(ctx); 1693 return; 1694 } 1695 1696 page_offset = ctx->page_offset; 1697 pages_to_boundary = _spdk_bs_num_pages_to_cluster_boundary(blob, page_offset); 1698 page_count = spdk_min(ctx->pages_remaining, pages_to_boundary); 1699 1700 /* 1701 * Get index and offset into the original iov array for our current position in the I/O sequence. 1702 * byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will 1703 * point to the current position in the I/O sequence. 1704 */ 1705 byte_count = ctx->pages_done * sizeof(struct spdk_blob_md_page); 1706 orig_iov = &ctx->orig_iov[0]; 1707 orig_iovoff = 0; 1708 while (byte_count > 0) { 1709 if (byte_count >= orig_iov->iov_len) { 1710 byte_count -= orig_iov->iov_len; 1711 orig_iov++; 1712 } else { 1713 orig_iovoff = byte_count; 1714 byte_count = 0; 1715 } 1716 } 1717 1718 /* 1719 * Build an iov array for the next I/O in the sequence. byte_count will keep track of how many 1720 * bytes of this next I/O remain to be accounted for in the new iov array. 1721 */ 1722 byte_count = page_count * sizeof(struct spdk_blob_md_page); 1723 iov = &ctx->iov[0]; 1724 iovcnt = 0; 1725 while (byte_count > 0) { 1726 iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff); 1727 iov->iov_base = orig_iov->iov_base + orig_iovoff; 1728 byte_count -= iov->iov_len; 1729 orig_iovoff = 0; 1730 orig_iov++; 1731 iov++; 1732 iovcnt++; 1733 } 1734 1735 ctx->page_offset += page_count; 1736 ctx->pages_done += page_count; 1737 ctx->pages_remaining -= page_count; 1738 iov = &ctx->iov[0]; 1739 1740 if (ctx->read) { 1741 spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, page_offset, 1742 page_count, _spdk_rw_iov_split_next, ctx); 1743 } else { 1744 spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, page_offset, 1745 page_count, _spdk_rw_iov_split_next, ctx); 1746 } 1747 } 1748 1749 static void 1750 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel, 1751 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 1752 spdk_blob_op_complete cb_fn, void *cb_arg, bool read) 1753 { 1754 struct spdk_bs_cpl cpl; 1755 1756 assert(blob != NULL); 1757 1758 if (!read && blob->data_ro) { 1759 cb_fn(cb_arg, -EPERM); 1760 return; 1761 } 1762 1763 if (length == 0) { 1764 cb_fn(cb_arg, 0); 1765 return; 1766 } 1767 1768 if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) { 1769 cb_fn(cb_arg, -EINVAL); 1770 return; 1771 } 1772 1773 /* 1774 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having 1775 * to split a request that spans a cluster boundary. For I/O that do not span a cluster boundary, 1776 * there will be no noticeable difference compared to using a batch. For I/O that do span a cluster 1777 * boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need 1778 * to allocate a separate iov array and split the I/O such that none of the resulting 1779 * smaller I/O cross a cluster boundary. These smaller I/O will be issued in sequence (not in parallel) 1780 * but since this case happens very infrequently, any performance impact will be negligible. 1781 * 1782 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs 1783 * for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them 1784 * in a batch. That would also require creating an intermediate spdk_bs_cpl that would get called 1785 * when the batch was completed, to allow for freeing the memory for the iov arrays. 1786 */ 1787 if (spdk_likely(length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset))) { 1788 uint32_t lba_count; 1789 uint64_t lba; 1790 1791 _spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 1792 1793 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1794 cpl.u.blob_basic.cb_fn = cb_fn; 1795 cpl.u.blob_basic.cb_arg = cb_arg; 1796 1797 if (read) { 1798 spdk_bs_sequence_t *seq; 1799 1800 seq = spdk_bs_sequence_start(_channel, &cpl); 1801 if (!seq) { 1802 cb_fn(cb_arg, -ENOMEM); 1803 return; 1804 } 1805 1806 if (_spdk_bs_page_is_allocated(blob, offset)) { 1807 spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL); 1808 } else { 1809 spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count, 1810 _spdk_rw_iov_done, NULL); 1811 } 1812 } else { 1813 if (_spdk_bs_page_is_allocated(blob, offset)) { 1814 spdk_bs_sequence_t *seq; 1815 1816 seq = spdk_bs_sequence_start(_channel, &cpl); 1817 if (!seq) { 1818 cb_fn(cb_arg, -ENOMEM); 1819 return; 1820 } 1821 1822 spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL); 1823 } else { 1824 /* Queue this operation and allocate the cluster */ 1825 spdk_bs_user_op_t *op; 1826 1827 op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, length); 1828 if (!op) { 1829 cb_fn(cb_arg, -ENOMEM); 1830 return; 1831 } 1832 1833 _spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op); 1834 } 1835 } 1836 } else { 1837 struct rw_iov_ctx *ctx; 1838 1839 ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec)); 1840 if (ctx == NULL) { 1841 cb_fn(cb_arg, -ENOMEM); 1842 return; 1843 } 1844 1845 ctx->blob = blob; 1846 ctx->channel = _channel; 1847 ctx->cb_fn = cb_fn; 1848 ctx->cb_arg = cb_arg; 1849 ctx->read = read; 1850 ctx->orig_iov = iov; 1851 ctx->iovcnt = iovcnt; 1852 ctx->page_offset = offset; 1853 ctx->pages_remaining = length; 1854 ctx->pages_done = 0; 1855 1856 _spdk_rw_iov_split_next(ctx, 0); 1857 } 1858 } 1859 1860 static struct spdk_blob * 1861 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid) 1862 { 1863 struct spdk_blob *blob; 1864 1865 TAILQ_FOREACH(blob, &bs->blobs, link) { 1866 if (blob->id == blobid) { 1867 return blob; 1868 } 1869 } 1870 1871 return NULL; 1872 } 1873 1874 static int 1875 _spdk_bs_channel_create(void *io_device, void *ctx_buf) 1876 { 1877 struct spdk_blob_store *bs = io_device; 1878 struct spdk_bs_channel *channel = ctx_buf; 1879 struct spdk_bs_dev *dev; 1880 uint32_t max_ops = bs->max_channel_ops; 1881 uint32_t i; 1882 1883 dev = bs->dev; 1884 1885 channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set)); 1886 if (!channel->req_mem) { 1887 return -1; 1888 } 1889 1890 TAILQ_INIT(&channel->reqs); 1891 1892 for (i = 0; i < max_ops; i++) { 1893 TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link); 1894 } 1895 1896 channel->bs = bs; 1897 channel->dev = dev; 1898 channel->dev_channel = dev->create_channel(dev); 1899 1900 if (!channel->dev_channel) { 1901 SPDK_ERRLOG("Failed to create device channel.\n"); 1902 free(channel->req_mem); 1903 return -1; 1904 } 1905 1906 TAILQ_INIT(&channel->need_cluster_alloc); 1907 1908 return 0; 1909 } 1910 1911 static void 1912 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf) 1913 { 1914 struct spdk_bs_channel *channel = ctx_buf; 1915 spdk_bs_user_op_t *op; 1916 1917 while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) { 1918 op = TAILQ_FIRST(&channel->need_cluster_alloc); 1919 TAILQ_REMOVE(&channel->need_cluster_alloc, op, link); 1920 spdk_bs_user_op_abort(op); 1921 } 1922 1923 free(channel->req_mem); 1924 channel->dev->destroy_channel(channel->dev, channel->dev_channel); 1925 } 1926 1927 static void 1928 _spdk_bs_dev_destroy(void *io_device) 1929 { 1930 struct spdk_blob_store *bs = io_device; 1931 struct spdk_blob *blob, *blob_tmp; 1932 1933 bs->dev->destroy(bs->dev); 1934 1935 TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) { 1936 TAILQ_REMOVE(&bs->blobs, blob, link); 1937 _spdk_blob_free(blob); 1938 } 1939 1940 pthread_mutex_destroy(&bs->used_clusters_mutex); 1941 1942 spdk_bit_array_free(&bs->used_blobids); 1943 spdk_bit_array_free(&bs->used_md_pages); 1944 spdk_bit_array_free(&bs->used_clusters); 1945 /* 1946 * If this function is called for any reason except a successful unload, 1947 * the unload_cpl type will be NONE and this will be a nop. 1948 */ 1949 spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err); 1950 1951 free(bs); 1952 } 1953 1954 static void 1955 _spdk_bs_free(struct spdk_blob_store *bs) 1956 { 1957 spdk_bs_unregister_md_thread(bs); 1958 spdk_io_device_unregister(bs, _spdk_bs_dev_destroy); 1959 } 1960 1961 void 1962 spdk_bs_opts_init(struct spdk_bs_opts *opts) 1963 { 1964 opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ; 1965 opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES; 1966 opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS; 1967 opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS; 1968 memset(&opts->bstype, 0, sizeof(opts->bstype)); 1969 } 1970 1971 static int 1972 _spdk_bs_opts_verify(struct spdk_bs_opts *opts) 1973 { 1974 if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 || 1975 opts->max_channel_ops == 0) { 1976 SPDK_ERRLOG("Blobstore options cannot be set to 0\n"); 1977 return -1; 1978 } 1979 1980 return 0; 1981 } 1982 1983 static struct spdk_blob_store * 1984 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts) 1985 { 1986 struct spdk_blob_store *bs; 1987 uint64_t dev_size; 1988 int rc; 1989 1990 dev_size = dev->blocklen * dev->blockcnt; 1991 if (dev_size < opts->cluster_sz) { 1992 /* Device size cannot be smaller than cluster size of blobstore */ 1993 SPDK_ERRLOG("Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n", 1994 dev_size, opts->cluster_sz); 1995 return NULL; 1996 } 1997 if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) { 1998 /* Cluster size cannot be smaller than page size */ 1999 SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n", 2000 opts->cluster_sz, SPDK_BS_PAGE_SIZE); 2001 return NULL; 2002 } 2003 bs = calloc(1, sizeof(struct spdk_blob_store)); 2004 if (!bs) { 2005 return NULL; 2006 } 2007 2008 TAILQ_INIT(&bs->blobs); 2009 bs->dev = dev; 2010 bs->md_thread = spdk_get_thread(); 2011 assert(bs->md_thread != NULL); 2012 2013 /* 2014 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an 2015 * even multiple of the cluster size. 2016 */ 2017 bs->cluster_sz = opts->cluster_sz; 2018 bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen); 2019 bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE; 2020 bs->num_free_clusters = bs->total_clusters; 2021 bs->used_clusters = spdk_bit_array_create(bs->total_clusters); 2022 if (bs->used_clusters == NULL) { 2023 free(bs); 2024 return NULL; 2025 } 2026 2027 bs->max_channel_ops = opts->max_channel_ops; 2028 bs->super_blob = SPDK_BLOBID_INVALID; 2029 memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype)); 2030 2031 /* The metadata is assumed to be at least 1 page */ 2032 bs->used_md_pages = spdk_bit_array_create(1); 2033 bs->used_blobids = spdk_bit_array_create(0); 2034 2035 pthread_mutex_init(&bs->used_clusters_mutex, NULL); 2036 2037 spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy, 2038 sizeof(struct spdk_bs_channel)); 2039 rc = spdk_bs_register_md_thread(bs); 2040 if (rc == -1) { 2041 spdk_io_device_unregister(bs, NULL); 2042 pthread_mutex_destroy(&bs->used_clusters_mutex); 2043 spdk_bit_array_free(&bs->used_blobids); 2044 spdk_bit_array_free(&bs->used_md_pages); 2045 spdk_bit_array_free(&bs->used_clusters); 2046 free(bs); 2047 return NULL; 2048 } 2049 2050 return bs; 2051 } 2052 2053 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */ 2054 2055 struct spdk_bs_load_ctx { 2056 struct spdk_blob_store *bs; 2057 struct spdk_bs_super_block *super; 2058 2059 struct spdk_bs_md_mask *mask; 2060 bool in_page_chain; 2061 uint32_t page_index; 2062 uint32_t cur_page; 2063 struct spdk_blob_md_page *page; 2064 bool is_load; 2065 }; 2066 2067 static void 2068 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 2069 { 2070 assert(bserrno != 0); 2071 2072 spdk_dma_free(ctx->super); 2073 /* 2074 * Only free the blobstore when a load fails. If an unload fails (for some reason) 2075 * we want to keep the blobstore in case the caller wants to try again. 2076 */ 2077 if (ctx->is_load) { 2078 _spdk_bs_free(ctx->bs); 2079 } 2080 free(ctx); 2081 spdk_bs_sequence_finish(seq, bserrno); 2082 } 2083 2084 static void 2085 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask) 2086 { 2087 uint32_t i = 0; 2088 2089 while (true) { 2090 i = spdk_bit_array_find_first_set(array, i); 2091 if (i >= mask->length) { 2092 break; 2093 } 2094 mask->mask[i / 8] |= 1U << (i % 8); 2095 i++; 2096 } 2097 } 2098 2099 static void 2100 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 2101 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg) 2102 { 2103 /* Update the values in the super block */ 2104 super->super_blob = bs->super_blob; 2105 memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype)); 2106 super->crc = _spdk_blob_md_page_calc_crc(super); 2107 spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0), 2108 _spdk_bs_byte_to_lba(bs, sizeof(*super)), 2109 cb_fn, cb_arg); 2110 } 2111 2112 static void 2113 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2114 { 2115 struct spdk_bs_load_ctx *ctx = arg; 2116 uint64_t mask_size, lba, lba_count; 2117 2118 /* Write out the used clusters mask */ 2119 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 2120 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2121 if (!ctx->mask) { 2122 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2123 return; 2124 } 2125 2126 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS; 2127 ctx->mask->length = ctx->bs->total_clusters; 2128 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters)); 2129 2130 _spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask); 2131 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 2132 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 2133 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2134 } 2135 2136 static void 2137 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2138 { 2139 struct spdk_bs_load_ctx *ctx = arg; 2140 uint64_t mask_size, lba, lba_count; 2141 2142 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 2143 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2144 if (!ctx->mask) { 2145 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2146 return; 2147 } 2148 2149 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES; 2150 ctx->mask->length = ctx->super->md_len; 2151 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages)); 2152 2153 _spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask); 2154 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 2155 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 2156 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2157 } 2158 2159 static void 2160 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2161 { 2162 struct spdk_bs_load_ctx *ctx = arg; 2163 uint64_t mask_size, lba, lba_count; 2164 2165 if (ctx->super->used_blobid_mask_len == 0) { 2166 /* 2167 * This is a pre-v3 on-disk format where the blobid mask does not get 2168 * written to disk. 2169 */ 2170 cb_fn(seq, arg, 0); 2171 return; 2172 } 2173 2174 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 2175 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2176 if (!ctx->mask) { 2177 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2178 return; 2179 } 2180 2181 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS; 2182 ctx->mask->length = ctx->super->md_len; 2183 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids)); 2184 2185 _spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask); 2186 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 2187 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 2188 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2189 } 2190 2191 static void 2192 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 2193 { 2194 spdk_dma_free(ctx->super); 2195 spdk_dma_free(ctx->mask); 2196 free(ctx); 2197 spdk_bs_sequence_finish(seq, bserrno); 2198 } 2199 2200 static void 2201 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2202 { 2203 struct spdk_bs_load_ctx *ctx = cb_arg; 2204 uint32_t i, j; 2205 int rc; 2206 2207 /* The type must be correct */ 2208 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS); 2209 2210 /* The length of the mask (in bits) must not be greater than 2211 * the length of the buffer (converted to bits) */ 2212 assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8)); 2213 2214 /* The length of the mask must be exactly equal to the size 2215 * (in pages) of the metadata region */ 2216 assert(ctx->mask->length == ctx->super->md_len); 2217 2218 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->mask->length); 2219 if (rc < 0) { 2220 spdk_dma_free(ctx->mask); 2221 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2222 return; 2223 } 2224 2225 for (i = 0; i < ctx->mask->length / 8; i++) { 2226 uint8_t segment = ctx->mask->mask[i]; 2227 for (j = 0; segment; j++) { 2228 if (segment & 1U) { 2229 spdk_bit_array_set(ctx->bs->used_blobids, (i * 8) + j); 2230 } 2231 segment >>= 1U; 2232 } 2233 } 2234 2235 _spdk_bs_load_complete(seq, ctx, bserrno); 2236 } 2237 2238 static void 2239 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2240 { 2241 struct spdk_bs_load_ctx *ctx = cb_arg; 2242 uint64_t lba, lba_count, mask_size; 2243 uint32_t i, j; 2244 int rc; 2245 2246 /* The type must be correct */ 2247 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS); 2248 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 2249 assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof( 2250 struct spdk_blob_md_page) * 8)); 2251 /* The length of the mask must be exactly equal to the total number of clusters */ 2252 assert(ctx->mask->length == ctx->bs->total_clusters); 2253 2254 rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters); 2255 if (rc < 0) { 2256 spdk_dma_free(ctx->mask); 2257 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2258 return; 2259 } 2260 2261 ctx->bs->num_free_clusters = ctx->bs->total_clusters; 2262 for (i = 0; i < ctx->mask->length / 8; i++) { 2263 uint8_t segment = ctx->mask->mask[i]; 2264 for (j = 0; segment && (j < 8); j++) { 2265 if (segment & 1U) { 2266 spdk_bit_array_set(ctx->bs->used_clusters, (i * 8) + j); 2267 assert(ctx->bs->num_free_clusters > 0); 2268 ctx->bs->num_free_clusters--; 2269 } 2270 segment >>= 1U; 2271 } 2272 } 2273 2274 spdk_dma_free(ctx->mask); 2275 2276 /* Read the used blobids mask */ 2277 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 2278 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2279 if (!ctx->mask) { 2280 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2281 return; 2282 } 2283 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 2284 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 2285 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2286 _spdk_bs_load_used_blobids_cpl, ctx); 2287 } 2288 2289 static void 2290 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2291 { 2292 struct spdk_bs_load_ctx *ctx = cb_arg; 2293 uint64_t lba, lba_count, mask_size; 2294 uint32_t i, j; 2295 int rc; 2296 2297 /* The type must be correct */ 2298 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES); 2299 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 2300 assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE * 2301 8)); 2302 /* The length of the mask must be exactly equal to the size (in pages) of the metadata region */ 2303 assert(ctx->mask->length == ctx->super->md_len); 2304 2305 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->mask->length); 2306 if (rc < 0) { 2307 spdk_dma_free(ctx->mask); 2308 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2309 return; 2310 } 2311 2312 for (i = 0; i < ctx->mask->length / 8; i++) { 2313 uint8_t segment = ctx->mask->mask[i]; 2314 for (j = 0; segment && (j < 8); j++) { 2315 if (segment & 1U) { 2316 spdk_bit_array_set(ctx->bs->used_md_pages, (i * 8) + j); 2317 } 2318 segment >>= 1U; 2319 } 2320 } 2321 spdk_dma_free(ctx->mask); 2322 2323 /* Read the used clusters mask */ 2324 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 2325 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2326 if (!ctx->mask) { 2327 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2328 return; 2329 } 2330 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 2331 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 2332 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2333 _spdk_bs_load_used_clusters_cpl, ctx); 2334 } 2335 2336 static void 2337 _spdk_bs_load_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2338 { 2339 struct spdk_bs_load_ctx *ctx = cb_arg; 2340 uint64_t lba, lba_count, mask_size; 2341 2342 /* Read the used pages mask */ 2343 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 2344 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2345 if (!ctx->mask) { 2346 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2347 return; 2348 } 2349 2350 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 2351 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 2352 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2353 _spdk_bs_load_used_pages_cpl, ctx); 2354 } 2355 2356 static int 2357 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs) 2358 { 2359 struct spdk_blob_md_descriptor *desc; 2360 size_t cur_desc = 0; 2361 2362 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 2363 while (cur_desc < sizeof(page->descriptors)) { 2364 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 2365 if (desc->length == 0) { 2366 /* If padding and length are 0, this terminates the page */ 2367 break; 2368 } 2369 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 2370 struct spdk_blob_md_descriptor_extent *desc_extent; 2371 unsigned int i, j; 2372 unsigned int cluster_count = 0; 2373 2374 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 2375 2376 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 2377 for (j = 0; j < desc_extent->extents[i].length; j++) { 2378 spdk_bit_array_set(bs->used_clusters, desc_extent->extents[i].cluster_idx + j); 2379 if (bs->num_free_clusters == 0) { 2380 return -1; 2381 } 2382 bs->num_free_clusters--; 2383 cluster_count++; 2384 } 2385 } 2386 if (cluster_count == 0) { 2387 return -1; 2388 } 2389 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 2390 /* Skip this item */ 2391 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 2392 /* Skip this item */ 2393 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 2394 /* Skip this item */ 2395 } else { 2396 /* Error */ 2397 return -1; 2398 } 2399 /* Advance to the next descriptor */ 2400 cur_desc += sizeof(*desc) + desc->length; 2401 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 2402 break; 2403 } 2404 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 2405 } 2406 return 0; 2407 } 2408 2409 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx) 2410 { 2411 uint32_t crc; 2412 2413 crc = _spdk_blob_md_page_calc_crc(ctx->page); 2414 if (crc != ctx->page->crc) { 2415 return false; 2416 } 2417 2418 if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) { 2419 return false; 2420 } 2421 return true; 2422 } 2423 2424 static void 2425 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg); 2426 2427 static void 2428 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2429 { 2430 struct spdk_bs_load_ctx *ctx = cb_arg; 2431 2432 _spdk_bs_load_complete(seq, ctx, bserrno); 2433 } 2434 2435 static void 2436 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2437 { 2438 struct spdk_bs_load_ctx *ctx = cb_arg; 2439 2440 spdk_dma_free(ctx->mask); 2441 ctx->mask = NULL; 2442 2443 _spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl); 2444 } 2445 2446 static void 2447 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2448 { 2449 struct spdk_bs_load_ctx *ctx = cb_arg; 2450 2451 spdk_dma_free(ctx->mask); 2452 ctx->mask = NULL; 2453 2454 _spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl); 2455 } 2456 2457 static void 2458 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2459 { 2460 _spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl); 2461 } 2462 2463 static void 2464 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2465 { 2466 struct spdk_bs_load_ctx *ctx = cb_arg; 2467 uint64_t num_md_clusters; 2468 uint64_t i; 2469 uint32_t page_num; 2470 2471 if (bserrno != 0) { 2472 _spdk_bs_load_ctx_fail(seq, ctx, bserrno); 2473 return; 2474 } 2475 2476 page_num = ctx->cur_page; 2477 if (_spdk_bs_load_cur_md_page_valid(ctx) == true) { 2478 if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) { 2479 spdk_bit_array_set(ctx->bs->used_md_pages, page_num); 2480 if (ctx->page->sequence_num == 0) { 2481 spdk_bit_array_set(ctx->bs->used_blobids, page_num); 2482 } 2483 if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) { 2484 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 2485 return; 2486 } 2487 if (ctx->page->next != SPDK_INVALID_MD_PAGE) { 2488 ctx->in_page_chain = true; 2489 ctx->cur_page = ctx->page->next; 2490 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 2491 return; 2492 } 2493 } 2494 } 2495 2496 ctx->in_page_chain = false; 2497 2498 do { 2499 ctx->page_index++; 2500 } while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true); 2501 2502 if (ctx->page_index < ctx->super->md_len) { 2503 ctx->cur_page = ctx->page_index; 2504 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 2505 } else { 2506 /* Claim all of the clusters used by the metadata */ 2507 num_md_clusters = divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster); 2508 for (i = 0; i < num_md_clusters; i++) { 2509 _spdk_bs_claim_cluster(ctx->bs, i); 2510 } 2511 spdk_dma_free(ctx->page); 2512 _spdk_bs_load_write_used_md(seq, ctx, bserrno); 2513 } 2514 } 2515 2516 static void 2517 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg) 2518 { 2519 struct spdk_bs_load_ctx *ctx = cb_arg; 2520 uint64_t lba; 2521 2522 assert(ctx->cur_page < ctx->super->md_len); 2523 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page); 2524 spdk_bs_sequence_read_dev(seq, ctx->page, lba, 2525 _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 2526 _spdk_bs_load_replay_md_cpl, ctx); 2527 } 2528 2529 static void 2530 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg) 2531 { 2532 struct spdk_bs_load_ctx *ctx = cb_arg; 2533 2534 ctx->page_index = 0; 2535 ctx->cur_page = 0; 2536 ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE, 2537 SPDK_BS_PAGE_SIZE, 2538 NULL); 2539 if (!ctx->page) { 2540 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2541 return; 2542 } 2543 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 2544 } 2545 2546 static void 2547 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2548 { 2549 struct spdk_bs_load_ctx *ctx = cb_arg; 2550 int rc; 2551 2552 if (bserrno != 0) { 2553 _spdk_bs_load_ctx_fail(seq, ctx, -EIO); 2554 return; 2555 } 2556 2557 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len); 2558 if (rc < 0) { 2559 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2560 return; 2561 } 2562 2563 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len); 2564 if (rc < 0) { 2565 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2566 return; 2567 } 2568 2569 rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters); 2570 if (rc < 0) { 2571 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2572 return; 2573 } 2574 2575 ctx->bs->num_free_clusters = ctx->bs->total_clusters; 2576 _spdk_bs_load_replay_md(seq, cb_arg); 2577 } 2578 2579 static void 2580 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2581 { 2582 struct spdk_bs_load_ctx *ctx = cb_arg; 2583 uint32_t crc; 2584 static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH]; 2585 2586 if (ctx->super->version > SPDK_BS_VERSION || 2587 ctx->super->version < SPDK_BS_INITIAL_VERSION) { 2588 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 2589 return; 2590 } 2591 2592 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 2593 sizeof(ctx->super->signature)) != 0) { 2594 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 2595 return; 2596 } 2597 2598 crc = _spdk_blob_md_page_calc_crc(ctx->super); 2599 if (crc != ctx->super->crc) { 2600 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 2601 return; 2602 } 2603 2604 if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 2605 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n"); 2606 } else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 2607 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n"); 2608 } else { 2609 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n"); 2610 SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 2611 SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 2612 _spdk_bs_load_ctx_fail(seq, ctx, -ENXIO); 2613 return; 2614 } 2615 2616 /* Parse the super block */ 2617 ctx->bs->cluster_sz = ctx->super->cluster_size; 2618 ctx->bs->total_clusters = ctx->bs->dev->blockcnt / (ctx->bs->cluster_sz / ctx->bs->dev->blocklen); 2619 ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE; 2620 ctx->bs->md_start = ctx->super->md_start; 2621 ctx->bs->md_len = ctx->super->md_len; 2622 ctx->bs->total_data_clusters = ctx->bs->total_clusters - divide_round_up( 2623 ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster); 2624 ctx->bs->super_blob = ctx->super->super_blob; 2625 memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype)); 2626 2627 if (ctx->super->clean == 0) { 2628 _spdk_bs_recover(seq, ctx, 0); 2629 } else if (ctx->super->used_blobid_mask_len == 0) { 2630 /* 2631 * Metadata is clean, but this is an old metadata format without 2632 * a blobid mask. Clear the clean bit and then build the masks 2633 * using _spdk_bs_recover. 2634 */ 2635 ctx->super->clean = 0; 2636 _spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_recover, ctx); 2637 } else { 2638 ctx->super->clean = 0; 2639 _spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_load_write_super_cpl, ctx); 2640 } 2641 } 2642 2643 void 2644 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 2645 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 2646 { 2647 struct spdk_blob_store *bs; 2648 struct spdk_bs_cpl cpl; 2649 spdk_bs_sequence_t *seq; 2650 struct spdk_bs_load_ctx *ctx; 2651 struct spdk_bs_opts opts = {}; 2652 2653 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev); 2654 2655 if (o) { 2656 opts = *o; 2657 } else { 2658 spdk_bs_opts_init(&opts); 2659 } 2660 2661 if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) { 2662 cb_fn(cb_arg, NULL, -EINVAL); 2663 return; 2664 } 2665 2666 bs = _spdk_bs_alloc(dev, &opts); 2667 if (!bs) { 2668 cb_fn(cb_arg, NULL, -ENOMEM); 2669 return; 2670 } 2671 2672 ctx = calloc(1, sizeof(*ctx)); 2673 if (!ctx) { 2674 _spdk_bs_free(bs); 2675 cb_fn(cb_arg, NULL, -ENOMEM); 2676 return; 2677 } 2678 2679 ctx->bs = bs; 2680 ctx->is_load = true; 2681 2682 /* Allocate memory for the super block */ 2683 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 2684 if (!ctx->super) { 2685 free(ctx); 2686 _spdk_bs_free(bs); 2687 return; 2688 } 2689 2690 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 2691 cpl.u.bs_handle.cb_fn = cb_fn; 2692 cpl.u.bs_handle.cb_arg = cb_arg; 2693 cpl.u.bs_handle.bs = bs; 2694 2695 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 2696 if (!seq) { 2697 spdk_dma_free(ctx->super); 2698 free(ctx); 2699 _spdk_bs_free(bs); 2700 cb_fn(cb_arg, NULL, -ENOMEM); 2701 return; 2702 } 2703 2704 /* Read the super block */ 2705 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 2706 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 2707 _spdk_bs_load_super_cpl, ctx); 2708 } 2709 2710 /* END spdk_bs_load */ 2711 2712 /* START spdk_bs_init */ 2713 2714 struct spdk_bs_init_ctx { 2715 struct spdk_blob_store *bs; 2716 struct spdk_bs_super_block *super; 2717 }; 2718 2719 static void 2720 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2721 { 2722 struct spdk_bs_init_ctx *ctx = cb_arg; 2723 2724 spdk_dma_free(ctx->super); 2725 free(ctx); 2726 2727 spdk_bs_sequence_finish(seq, bserrno); 2728 } 2729 2730 static void 2731 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2732 { 2733 struct spdk_bs_init_ctx *ctx = cb_arg; 2734 2735 /* Write super block */ 2736 spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0), 2737 _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)), 2738 _spdk_bs_init_persist_super_cpl, ctx); 2739 } 2740 2741 void 2742 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 2743 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 2744 { 2745 struct spdk_bs_init_ctx *ctx; 2746 struct spdk_blob_store *bs; 2747 struct spdk_bs_cpl cpl; 2748 spdk_bs_sequence_t *seq; 2749 spdk_bs_batch_t *batch; 2750 uint64_t num_md_lba; 2751 uint64_t num_md_pages; 2752 uint64_t num_md_clusters; 2753 uint32_t i; 2754 struct spdk_bs_opts opts = {}; 2755 int rc; 2756 2757 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev); 2758 2759 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 2760 SPDK_ERRLOG("unsupported dev block length of %d\n", 2761 dev->blocklen); 2762 dev->destroy(dev); 2763 cb_fn(cb_arg, NULL, -EINVAL); 2764 return; 2765 } 2766 2767 if (o) { 2768 opts = *o; 2769 } else { 2770 spdk_bs_opts_init(&opts); 2771 } 2772 2773 if (_spdk_bs_opts_verify(&opts) != 0) { 2774 dev->destroy(dev); 2775 cb_fn(cb_arg, NULL, -EINVAL); 2776 return; 2777 } 2778 2779 bs = _spdk_bs_alloc(dev, &opts); 2780 if (!bs) { 2781 dev->destroy(dev); 2782 cb_fn(cb_arg, NULL, -ENOMEM); 2783 return; 2784 } 2785 2786 if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) { 2787 /* By default, allocate 1 page per cluster. 2788 * Technically, this over-allocates metadata 2789 * because more metadata will reduce the number 2790 * of usable clusters. This can be addressed with 2791 * more complex math in the future. 2792 */ 2793 bs->md_len = bs->total_clusters; 2794 } else { 2795 bs->md_len = opts.num_md_pages; 2796 } 2797 2798 rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len); 2799 if (rc < 0) { 2800 _spdk_bs_free(bs); 2801 cb_fn(cb_arg, NULL, -ENOMEM); 2802 return; 2803 } 2804 2805 rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len); 2806 if (rc < 0) { 2807 _spdk_bs_free(bs); 2808 cb_fn(cb_arg, NULL, -ENOMEM); 2809 return; 2810 } 2811 2812 ctx = calloc(1, sizeof(*ctx)); 2813 if (!ctx) { 2814 _spdk_bs_free(bs); 2815 cb_fn(cb_arg, NULL, -ENOMEM); 2816 return; 2817 } 2818 2819 ctx->bs = bs; 2820 2821 /* Allocate memory for the super block */ 2822 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 2823 if (!ctx->super) { 2824 free(ctx); 2825 _spdk_bs_free(bs); 2826 return; 2827 } 2828 memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 2829 sizeof(ctx->super->signature)); 2830 ctx->super->version = SPDK_BS_VERSION; 2831 ctx->super->length = sizeof(*ctx->super); 2832 ctx->super->super_blob = bs->super_blob; 2833 ctx->super->clean = 0; 2834 ctx->super->cluster_size = bs->cluster_sz; 2835 memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype)); 2836 2837 /* Calculate how many pages the metadata consumes at the front 2838 * of the disk. 2839 */ 2840 2841 /* The super block uses 1 page */ 2842 num_md_pages = 1; 2843 2844 /* The used_md_pages mask requires 1 bit per metadata page, rounded 2845 * up to the nearest page, plus a header. 2846 */ 2847 ctx->super->used_page_mask_start = num_md_pages; 2848 ctx->super->used_page_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) + 2849 divide_round_up(bs->md_len, 8), 2850 SPDK_BS_PAGE_SIZE); 2851 num_md_pages += ctx->super->used_page_mask_len; 2852 2853 /* The used_clusters mask requires 1 bit per cluster, rounded 2854 * up to the nearest page, plus a header. 2855 */ 2856 ctx->super->used_cluster_mask_start = num_md_pages; 2857 ctx->super->used_cluster_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) + 2858 divide_round_up(bs->total_clusters, 8), 2859 SPDK_BS_PAGE_SIZE); 2860 num_md_pages += ctx->super->used_cluster_mask_len; 2861 2862 /* The used_blobids mask requires 1 bit per metadata page, rounded 2863 * up to the nearest page, plus a header. 2864 */ 2865 ctx->super->used_blobid_mask_start = num_md_pages; 2866 ctx->super->used_blobid_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) + 2867 divide_round_up(bs->md_len, 8), 2868 SPDK_BS_PAGE_SIZE); 2869 num_md_pages += ctx->super->used_blobid_mask_len; 2870 2871 /* The metadata region size was chosen above */ 2872 ctx->super->md_start = bs->md_start = num_md_pages; 2873 ctx->super->md_len = bs->md_len; 2874 num_md_pages += bs->md_len; 2875 2876 num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages); 2877 2878 ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super); 2879 2880 num_md_clusters = divide_round_up(num_md_pages, bs->pages_per_cluster); 2881 if (num_md_clusters > bs->total_clusters) { 2882 SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, " 2883 "please decrease number of pages reserved for metadata " 2884 "or increase cluster size.\n"); 2885 spdk_dma_free(ctx->super); 2886 free(ctx); 2887 _spdk_bs_free(bs); 2888 cb_fn(cb_arg, NULL, -ENOMEM); 2889 return; 2890 } 2891 /* Claim all of the clusters used by the metadata */ 2892 for (i = 0; i < num_md_clusters; i++) { 2893 _spdk_bs_claim_cluster(bs, i); 2894 } 2895 2896 bs->total_data_clusters = bs->num_free_clusters; 2897 2898 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 2899 cpl.u.bs_handle.cb_fn = cb_fn; 2900 cpl.u.bs_handle.cb_arg = cb_arg; 2901 cpl.u.bs_handle.bs = bs; 2902 2903 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 2904 if (!seq) { 2905 spdk_dma_free(ctx->super); 2906 free(ctx); 2907 _spdk_bs_free(bs); 2908 cb_fn(cb_arg, NULL, -ENOMEM); 2909 return; 2910 } 2911 2912 batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx); 2913 2914 /* Clear metadata space */ 2915 spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba); 2916 /* Trim data clusters */ 2917 spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba); 2918 2919 spdk_bs_batch_close(batch); 2920 } 2921 2922 /* END spdk_bs_init */ 2923 2924 /* START spdk_bs_destroy */ 2925 2926 static void 2927 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2928 { 2929 struct spdk_bs_init_ctx *ctx = cb_arg; 2930 struct spdk_blob_store *bs = ctx->bs; 2931 2932 /* 2933 * We need to defer calling spdk_bs_call_cpl() until after 2934 * dev destruction, so tuck these away for later use. 2935 */ 2936 bs->unload_err = bserrno; 2937 memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 2938 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 2939 2940 spdk_bs_sequence_finish(seq, bserrno); 2941 2942 _spdk_bs_free(bs); 2943 free(ctx); 2944 } 2945 2946 void 2947 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, 2948 void *cb_arg) 2949 { 2950 struct spdk_bs_cpl cpl; 2951 spdk_bs_sequence_t *seq; 2952 struct spdk_bs_init_ctx *ctx; 2953 2954 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n"); 2955 2956 if (!TAILQ_EMPTY(&bs->blobs)) { 2957 SPDK_ERRLOG("Blobstore still has open blobs\n"); 2958 cb_fn(cb_arg, -EBUSY); 2959 return; 2960 } 2961 2962 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 2963 cpl.u.bs_basic.cb_fn = cb_fn; 2964 cpl.u.bs_basic.cb_arg = cb_arg; 2965 2966 ctx = calloc(1, sizeof(*ctx)); 2967 if (!ctx) { 2968 cb_fn(cb_arg, -ENOMEM); 2969 return; 2970 } 2971 2972 ctx->bs = bs; 2973 2974 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 2975 if (!seq) { 2976 free(ctx); 2977 cb_fn(cb_arg, -ENOMEM); 2978 return; 2979 } 2980 2981 /* Write zeroes to the super block */ 2982 spdk_bs_sequence_write_zeroes_dev(seq, 2983 _spdk_bs_page_to_lba(bs, 0), 2984 _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)), 2985 _spdk_bs_destroy_trim_cpl, ctx); 2986 } 2987 2988 /* END spdk_bs_destroy */ 2989 2990 /* START spdk_bs_unload */ 2991 2992 static void 2993 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2994 { 2995 struct spdk_bs_load_ctx *ctx = cb_arg; 2996 2997 spdk_dma_free(ctx->super); 2998 2999 /* 3000 * We need to defer calling spdk_bs_call_cpl() until after 3001 * dev destuction, so tuck these away for later use. 3002 */ 3003 ctx->bs->unload_err = bserrno; 3004 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 3005 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 3006 3007 spdk_bs_sequence_finish(seq, bserrno); 3008 3009 _spdk_bs_free(ctx->bs); 3010 free(ctx); 3011 } 3012 3013 static void 3014 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3015 { 3016 struct spdk_bs_load_ctx *ctx = cb_arg; 3017 3018 spdk_dma_free(ctx->mask); 3019 ctx->super->clean = 1; 3020 3021 _spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx); 3022 } 3023 3024 static void 3025 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3026 { 3027 struct spdk_bs_load_ctx *ctx = cb_arg; 3028 3029 spdk_dma_free(ctx->mask); 3030 ctx->mask = NULL; 3031 3032 _spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl); 3033 } 3034 3035 static void 3036 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3037 { 3038 struct spdk_bs_load_ctx *ctx = cb_arg; 3039 3040 spdk_dma_free(ctx->mask); 3041 ctx->mask = NULL; 3042 3043 _spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl); 3044 } 3045 3046 static void 3047 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3048 { 3049 _spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl); 3050 } 3051 3052 void 3053 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg) 3054 { 3055 struct spdk_bs_cpl cpl; 3056 spdk_bs_sequence_t *seq; 3057 struct spdk_bs_load_ctx *ctx; 3058 3059 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n"); 3060 3061 if (!TAILQ_EMPTY(&bs->blobs)) { 3062 SPDK_ERRLOG("Blobstore still has open blobs\n"); 3063 cb_fn(cb_arg, -EBUSY); 3064 return; 3065 } 3066 3067 ctx = calloc(1, sizeof(*ctx)); 3068 if (!ctx) { 3069 cb_fn(cb_arg, -ENOMEM); 3070 return; 3071 } 3072 3073 ctx->bs = bs; 3074 ctx->is_load = false; 3075 3076 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 3077 if (!ctx->super) { 3078 free(ctx); 3079 cb_fn(cb_arg, -ENOMEM); 3080 return; 3081 } 3082 3083 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3084 cpl.u.bs_basic.cb_fn = cb_fn; 3085 cpl.u.bs_basic.cb_arg = cb_arg; 3086 3087 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3088 if (!seq) { 3089 spdk_dma_free(ctx->super); 3090 free(ctx); 3091 cb_fn(cb_arg, -ENOMEM); 3092 return; 3093 } 3094 3095 /* Read super block */ 3096 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3097 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3098 _spdk_bs_unload_read_super_cpl, ctx); 3099 } 3100 3101 /* END spdk_bs_unload */ 3102 3103 void 3104 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid, 3105 spdk_bs_op_complete cb_fn, void *cb_arg) 3106 { 3107 bs->super_blob = blobid; 3108 cb_fn(cb_arg, 0); 3109 } 3110 3111 void 3112 spdk_bs_get_super(struct spdk_blob_store *bs, 3113 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 3114 { 3115 if (bs->super_blob == SPDK_BLOBID_INVALID) { 3116 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT); 3117 } else { 3118 cb_fn(cb_arg, bs->super_blob, 0); 3119 } 3120 } 3121 3122 uint64_t 3123 spdk_bs_get_cluster_size(struct spdk_blob_store *bs) 3124 { 3125 return bs->cluster_sz; 3126 } 3127 3128 uint64_t 3129 spdk_bs_get_page_size(struct spdk_blob_store *bs) 3130 { 3131 return SPDK_BS_PAGE_SIZE; 3132 } 3133 3134 uint64_t 3135 spdk_bs_free_cluster_count(struct spdk_blob_store *bs) 3136 { 3137 return bs->num_free_clusters; 3138 } 3139 3140 uint64_t 3141 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs) 3142 { 3143 return bs->total_data_clusters; 3144 } 3145 3146 static int 3147 spdk_bs_register_md_thread(struct spdk_blob_store *bs) 3148 { 3149 bs->md_channel = spdk_get_io_channel(bs); 3150 if (!bs->md_channel) { 3151 SPDK_ERRLOG("Failed to get IO channel.\n"); 3152 return -1; 3153 } 3154 3155 return 0; 3156 } 3157 3158 static int 3159 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs) 3160 { 3161 spdk_put_io_channel(bs->md_channel); 3162 3163 return 0; 3164 } 3165 3166 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob) 3167 { 3168 assert(blob != NULL); 3169 3170 return blob->id; 3171 } 3172 3173 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob) 3174 { 3175 assert(blob != NULL); 3176 3177 return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters); 3178 } 3179 3180 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob) 3181 { 3182 assert(blob != NULL); 3183 3184 return blob->active.num_clusters; 3185 } 3186 3187 /* START spdk_bs_create_blob */ 3188 3189 static void 3190 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3191 { 3192 struct spdk_blob *blob = cb_arg; 3193 3194 _spdk_blob_free(blob); 3195 3196 spdk_bs_sequence_finish(seq, bserrno); 3197 } 3198 3199 static int 3200 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs, 3201 bool internal) 3202 { 3203 uint64_t i; 3204 size_t value_len = 0; 3205 int rc; 3206 const void *value = NULL; 3207 if (xattrs->count > 0 && xattrs->get_value == NULL) { 3208 return -EINVAL; 3209 } 3210 for (i = 0; i < xattrs->count; i++) { 3211 xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len); 3212 if (value == NULL || value_len == 0) { 3213 return -EINVAL; 3214 } 3215 rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal); 3216 if (rc < 0) { 3217 return rc; 3218 } 3219 } 3220 return 0; 3221 } 3222 3223 static void 3224 _spdk_blob_set_thin_provision(struct spdk_blob *blob) 3225 { 3226 _spdk_blob_verify_md_op(blob); 3227 blob->invalid_flags |= SPDK_BLOB_THIN_PROV; 3228 blob->state = SPDK_BLOB_STATE_DIRTY; 3229 } 3230 3231 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts, 3232 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 3233 { 3234 struct spdk_blob *blob; 3235 uint32_t page_idx; 3236 struct spdk_bs_cpl cpl; 3237 struct spdk_blob_opts opts_default; 3238 spdk_bs_sequence_t *seq; 3239 spdk_blob_id id; 3240 int rc; 3241 3242 assert(spdk_get_thread() == bs->md_thread); 3243 3244 page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0); 3245 if (page_idx >= spdk_bit_array_capacity(bs->used_md_pages)) { 3246 cb_fn(cb_arg, 0, -ENOMEM); 3247 return; 3248 } 3249 spdk_bit_array_set(bs->used_blobids, page_idx); 3250 spdk_bit_array_set(bs->used_md_pages, page_idx); 3251 3252 id = _spdk_bs_page_to_blobid(page_idx); 3253 3254 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx); 3255 3256 blob = _spdk_blob_alloc(bs, id); 3257 if (!blob) { 3258 cb_fn(cb_arg, 0, -ENOMEM); 3259 return; 3260 } 3261 3262 if (!opts) { 3263 spdk_blob_opts_init(&opts_default); 3264 opts = &opts_default; 3265 } 3266 3267 rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false); 3268 if (rc < 0) { 3269 _spdk_blob_free(blob); 3270 cb_fn(cb_arg, 0, rc); 3271 return; 3272 } 3273 if (opts->thin_provision) { 3274 _spdk_blob_set_thin_provision(blob); 3275 } 3276 3277 rc = spdk_blob_resize(blob, opts->num_clusters); 3278 if (rc < 0) { 3279 _spdk_blob_free(blob); 3280 cb_fn(cb_arg, 0, rc); 3281 return; 3282 } 3283 cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 3284 cpl.u.blobid.cb_fn = cb_fn; 3285 cpl.u.blobid.cb_arg = cb_arg; 3286 cpl.u.blobid.blobid = blob->id; 3287 3288 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3289 if (!seq) { 3290 _spdk_blob_free(blob); 3291 cb_fn(cb_arg, 0, -ENOMEM); 3292 return; 3293 } 3294 3295 _spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob); 3296 } 3297 3298 void spdk_bs_create_blob(struct spdk_blob_store *bs, 3299 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 3300 { 3301 spdk_bs_create_blob_ext(bs, NULL, cb_fn, cb_arg); 3302 } 3303 3304 /* END spdk_bs_create_blob */ 3305 3306 /* START spdk_blob_resize */ 3307 int 3308 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz) 3309 { 3310 int rc; 3311 3312 _spdk_blob_verify_md_op(blob); 3313 3314 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz); 3315 3316 if (blob->md_ro) { 3317 return -EPERM; 3318 } 3319 3320 if (sz == blob->active.num_clusters) { 3321 return 0; 3322 } 3323 3324 rc = _spdk_resize_blob(blob, sz); 3325 if (rc < 0) { 3326 return rc; 3327 } 3328 3329 return 0; 3330 } 3331 3332 /* END spdk_blob_resize */ 3333 3334 3335 /* START spdk_bs_delete_blob */ 3336 3337 static void 3338 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno) 3339 { 3340 spdk_bs_sequence_t *seq = cb_arg; 3341 3342 spdk_bs_sequence_finish(seq, bserrno); 3343 } 3344 3345 static void 3346 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3347 { 3348 struct spdk_blob *blob = cb_arg; 3349 3350 if (bserrno != 0) { 3351 /* 3352 * We already removed this blob from the blobstore tailq, so 3353 * we need to free it here since this is the last reference 3354 * to it. 3355 */ 3356 _spdk_blob_free(blob); 3357 _spdk_bs_delete_close_cpl(seq, bserrno); 3358 return; 3359 } 3360 3361 /* 3362 * This will immediately decrement the ref_count and call 3363 * the completion routine since the metadata state is clean. 3364 * By calling spdk_blob_close, we reduce the number of call 3365 * points into code that touches the blob->open_ref count 3366 * and the blobstore's blob list. 3367 */ 3368 spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq); 3369 } 3370 3371 static void 3372 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno) 3373 { 3374 spdk_bs_sequence_t *seq = cb_arg; 3375 uint32_t page_num; 3376 3377 if (bserrno != 0) { 3378 spdk_bs_sequence_finish(seq, bserrno); 3379 return; 3380 } 3381 3382 _spdk_blob_verify_md_op(blob); 3383 3384 if (blob->open_ref > 1) { 3385 /* 3386 * Someone has this blob open (besides this delete context). 3387 * Decrement the ref count directly and return -EBUSY. 3388 */ 3389 blob->open_ref--; 3390 spdk_bs_sequence_finish(seq, -EBUSY); 3391 return; 3392 } 3393 3394 /* 3395 * Remove the blob from the blob_store list now, to ensure it does not 3396 * get returned after this point by _spdk_blob_lookup(). 3397 */ 3398 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 3399 page_num = _spdk_bs_blobid_to_page(blob->id); 3400 spdk_bit_array_clear(blob->bs->used_blobids, page_num); 3401 blob->state = SPDK_BLOB_STATE_DIRTY; 3402 blob->active.num_pages = 0; 3403 _spdk_resize_blob(blob, 0); 3404 3405 _spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob); 3406 } 3407 3408 void 3409 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 3410 spdk_blob_op_complete cb_fn, void *cb_arg) 3411 { 3412 struct spdk_bs_cpl cpl; 3413 spdk_bs_sequence_t *seq; 3414 3415 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid); 3416 3417 assert(spdk_get_thread() == bs->md_thread); 3418 3419 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 3420 cpl.u.blob_basic.cb_fn = cb_fn; 3421 cpl.u.blob_basic.cb_arg = cb_arg; 3422 3423 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3424 if (!seq) { 3425 cb_fn(cb_arg, -ENOMEM); 3426 return; 3427 } 3428 3429 spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq); 3430 } 3431 3432 /* END spdk_bs_delete_blob */ 3433 3434 /* START spdk_bs_open_blob */ 3435 3436 static void 3437 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3438 { 3439 struct spdk_blob *blob = cb_arg; 3440 3441 /* If the blob have crc error, we just return NULL. */ 3442 if (blob == NULL) { 3443 seq->cpl.u.blob_handle.blob = NULL; 3444 spdk_bs_sequence_finish(seq, bserrno); 3445 return; 3446 } 3447 3448 blob->open_ref++; 3449 3450 TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link); 3451 3452 spdk_bs_sequence_finish(seq, bserrno); 3453 } 3454 3455 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 3456 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 3457 { 3458 struct spdk_blob *blob; 3459 struct spdk_bs_cpl cpl; 3460 spdk_bs_sequence_t *seq; 3461 uint32_t page_num; 3462 3463 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid); 3464 assert(spdk_get_thread() == bs->md_thread); 3465 3466 page_num = _spdk_bs_blobid_to_page(blobid); 3467 if (spdk_bit_array_get(bs->used_blobids, page_num) == false) { 3468 /* Invalid blobid */ 3469 cb_fn(cb_arg, NULL, -ENOENT); 3470 return; 3471 } 3472 3473 blob = _spdk_blob_lookup(bs, blobid); 3474 if (blob) { 3475 blob->open_ref++; 3476 cb_fn(cb_arg, blob, 0); 3477 return; 3478 } 3479 3480 blob = _spdk_blob_alloc(bs, blobid); 3481 if (!blob) { 3482 cb_fn(cb_arg, NULL, -ENOMEM); 3483 return; 3484 } 3485 3486 cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE; 3487 cpl.u.blob_handle.cb_fn = cb_fn; 3488 cpl.u.blob_handle.cb_arg = cb_arg; 3489 cpl.u.blob_handle.blob = blob; 3490 3491 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3492 if (!seq) { 3493 _spdk_blob_free(blob); 3494 cb_fn(cb_arg, NULL, -ENOMEM); 3495 return; 3496 } 3497 3498 _spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob); 3499 } 3500 /* END spdk_bs_open_blob */ 3501 3502 /* START spdk_blob_set_read_only */ 3503 int spdk_blob_set_read_only(struct spdk_blob *blob) 3504 { 3505 _spdk_blob_verify_md_op(blob); 3506 3507 blob->data_ro_flags |= SPDK_BLOB_READ_ONLY; 3508 3509 blob->state = SPDK_BLOB_STATE_DIRTY; 3510 return 0; 3511 } 3512 /* END spdk_blob_set_read_only */ 3513 3514 /* START spdk_blob_sync_md */ 3515 3516 static void 3517 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3518 { 3519 struct spdk_blob *blob = cb_arg; 3520 3521 if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 3522 blob->data_ro = true; 3523 blob->md_ro = true; 3524 } 3525 3526 spdk_bs_sequence_finish(seq, bserrno); 3527 } 3528 3529 static void 3530 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 3531 { 3532 struct spdk_bs_cpl cpl; 3533 spdk_bs_sequence_t *seq; 3534 3535 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 3536 cpl.u.blob_basic.cb_fn = cb_fn; 3537 cpl.u.blob_basic.cb_arg = cb_arg; 3538 3539 seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl); 3540 if (!seq) { 3541 cb_fn(cb_arg, -ENOMEM); 3542 return; 3543 } 3544 3545 _spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob); 3546 } 3547 3548 void 3549 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 3550 { 3551 _spdk_blob_verify_md_op(blob); 3552 3553 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id); 3554 3555 if (blob->md_ro) { 3556 assert(blob->state == SPDK_BLOB_STATE_CLEAN); 3557 cb_fn(cb_arg, 0); 3558 return; 3559 } 3560 3561 _spdk_blob_sync_md(blob, cb_fn, cb_arg); 3562 } 3563 3564 /* END spdk_blob_sync_md */ 3565 3566 struct spdk_blob_insert_cluster_ctx { 3567 struct spdk_thread *thread; 3568 struct spdk_blob *blob; 3569 uint32_t cluster_num; /* cluster index in blob */ 3570 uint32_t cluster; /* cluster on disk */ 3571 int rc; 3572 spdk_blob_op_complete cb_fn; 3573 void *cb_arg; 3574 }; 3575 3576 static void 3577 _spdk_blob_insert_cluster_msg_cpl(void *arg) 3578 { 3579 struct spdk_blob_insert_cluster_ctx *ctx = arg; 3580 3581 ctx->cb_fn(ctx->cb_arg, ctx->rc); 3582 free(ctx); 3583 } 3584 3585 static void 3586 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno) 3587 { 3588 struct spdk_blob_insert_cluster_ctx *ctx = arg; 3589 3590 ctx->rc = bserrno; 3591 spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx); 3592 } 3593 3594 static void 3595 _spdk_blob_insert_cluster_msg(void *arg) 3596 { 3597 struct spdk_blob_insert_cluster_ctx *ctx = arg; 3598 3599 ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster); 3600 if (ctx->rc != 0) { 3601 spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx); 3602 return; 3603 } 3604 3605 ctx->blob->state = SPDK_BLOB_STATE_DIRTY; 3606 _spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx); 3607 } 3608 3609 void 3610 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 3611 uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg) 3612 { 3613 struct spdk_blob_insert_cluster_ctx *ctx; 3614 3615 ctx = calloc(1, sizeof(*ctx)); 3616 if (ctx == NULL) { 3617 cb_fn(cb_arg, -ENOMEM); 3618 return; 3619 } 3620 3621 ctx->thread = spdk_get_thread(); 3622 ctx->blob = blob; 3623 ctx->cluster_num = cluster_num; 3624 ctx->cluster = cluster; 3625 ctx->cb_fn = cb_fn; 3626 ctx->cb_arg = cb_arg; 3627 3628 spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx); 3629 } 3630 3631 /* START spdk_blob_close */ 3632 3633 static void 3634 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3635 { 3636 struct spdk_blob *blob = cb_arg; 3637 3638 if (bserrno == 0) { 3639 blob->open_ref--; 3640 if (blob->open_ref == 0) { 3641 /* 3642 * Blobs with active.num_pages == 0 are deleted blobs. 3643 * these blobs are removed from the blob_store list 3644 * when the deletion process starts - so don't try to 3645 * remove them again. 3646 */ 3647 if (blob->active.num_pages > 0) { 3648 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 3649 } 3650 _spdk_blob_free(blob); 3651 } 3652 } 3653 3654 spdk_bs_sequence_finish(seq, bserrno); 3655 } 3656 3657 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 3658 { 3659 struct spdk_bs_cpl cpl; 3660 spdk_bs_sequence_t *seq; 3661 3662 _spdk_blob_verify_md_op(blob); 3663 3664 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id); 3665 3666 if (blob->open_ref == 0) { 3667 cb_fn(cb_arg, -EBADF); 3668 return; 3669 } 3670 3671 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 3672 cpl.u.blob_basic.cb_fn = cb_fn; 3673 cpl.u.blob_basic.cb_arg = cb_arg; 3674 3675 seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl); 3676 if (!seq) { 3677 cb_fn(cb_arg, -ENOMEM); 3678 return; 3679 } 3680 3681 /* Sync metadata */ 3682 _spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob); 3683 } 3684 3685 /* END spdk_blob_close */ 3686 3687 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs) 3688 { 3689 return spdk_get_io_channel(bs); 3690 } 3691 3692 void spdk_bs_free_io_channel(struct spdk_io_channel *channel) 3693 { 3694 spdk_put_io_channel(channel); 3695 } 3696 3697 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel, 3698 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 3699 { 3700 _spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 3701 SPDK_BLOB_UNMAP); 3702 } 3703 3704 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel, 3705 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 3706 { 3707 _spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 3708 SPDK_BLOB_WRITE_ZEROES); 3709 } 3710 3711 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel, 3712 void *payload, uint64_t offset, uint64_t length, 3713 spdk_blob_op_complete cb_fn, void *cb_arg) 3714 { 3715 _spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 3716 SPDK_BLOB_WRITE); 3717 } 3718 3719 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel, 3720 void *payload, uint64_t offset, uint64_t length, 3721 spdk_blob_op_complete cb_fn, void *cb_arg) 3722 { 3723 _spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 3724 SPDK_BLOB_READ); 3725 } 3726 3727 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel, 3728 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 3729 spdk_blob_op_complete cb_fn, void *cb_arg) 3730 { 3731 _spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false); 3732 } 3733 3734 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel, 3735 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 3736 spdk_blob_op_complete cb_fn, void *cb_arg) 3737 { 3738 _spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true); 3739 } 3740 3741 void spdk_bs_io_unmap_blob(struct spdk_blob *blob, struct spdk_io_channel *channel, 3742 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 3743 { 3744 spdk_blob_io_unmap(blob, channel, offset, length, cb_fn, cb_arg); 3745 } 3746 3747 void spdk_bs_io_write_zeroes_blob(struct spdk_blob *blob, struct spdk_io_channel *channel, 3748 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 3749 { 3750 spdk_blob_io_write_zeroes(blob, channel, offset, length, cb_fn, cb_arg); 3751 } 3752 3753 void spdk_bs_io_write_blob(struct spdk_blob *blob, struct spdk_io_channel *channel, 3754 void *payload, uint64_t offset, uint64_t length, 3755 spdk_blob_op_complete cb_fn, void *cb_arg) 3756 { 3757 spdk_blob_io_write(blob, channel, payload, offset, length, cb_fn, cb_arg); 3758 } 3759 3760 void spdk_bs_io_read_blob(struct spdk_blob *blob, struct spdk_io_channel *channel, 3761 void *payload, uint64_t offset, uint64_t length, 3762 spdk_blob_op_complete cb_fn, void *cb_arg) 3763 { 3764 spdk_blob_io_read(blob, channel, payload, offset, length, cb_fn, cb_arg); 3765 } 3766 3767 void spdk_bs_io_writev_blob(struct spdk_blob *blob, struct spdk_io_channel *channel, 3768 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 3769 spdk_blob_op_complete cb_fn, void *cb_arg) 3770 { 3771 spdk_blob_io_writev(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg); 3772 } 3773 3774 void spdk_bs_io_readv_blob(struct spdk_blob *blob, struct spdk_io_channel *channel, 3775 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 3776 spdk_blob_op_complete cb_fn, void *cb_arg) 3777 { 3778 spdk_blob_io_readv(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg); 3779 } 3780 3781 struct spdk_bs_iter_ctx { 3782 int64_t page_num; 3783 struct spdk_blob_store *bs; 3784 3785 spdk_blob_op_with_handle_complete cb_fn; 3786 void *cb_arg; 3787 }; 3788 3789 static void 3790 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 3791 { 3792 struct spdk_bs_iter_ctx *ctx = cb_arg; 3793 struct spdk_blob_store *bs = ctx->bs; 3794 spdk_blob_id id; 3795 3796 if (bserrno == 0) { 3797 ctx->cb_fn(ctx->cb_arg, _blob, bserrno); 3798 free(ctx); 3799 return; 3800 } 3801 3802 ctx->page_num++; 3803 ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num); 3804 if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) { 3805 ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT); 3806 free(ctx); 3807 return; 3808 } 3809 3810 id = _spdk_bs_page_to_blobid(ctx->page_num); 3811 3812 spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx); 3813 } 3814 3815 void 3816 spdk_bs_iter_first(struct spdk_blob_store *bs, 3817 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 3818 { 3819 struct spdk_bs_iter_ctx *ctx; 3820 3821 ctx = calloc(1, sizeof(*ctx)); 3822 if (!ctx) { 3823 cb_fn(cb_arg, NULL, -ENOMEM); 3824 return; 3825 } 3826 3827 ctx->page_num = -1; 3828 ctx->bs = bs; 3829 ctx->cb_fn = cb_fn; 3830 ctx->cb_arg = cb_arg; 3831 3832 _spdk_bs_iter_cpl(ctx, NULL, -1); 3833 } 3834 3835 static void 3836 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno) 3837 { 3838 struct spdk_bs_iter_ctx *ctx = cb_arg; 3839 3840 _spdk_bs_iter_cpl(ctx, NULL, -1); 3841 } 3842 3843 void 3844 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob, 3845 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 3846 { 3847 struct spdk_bs_iter_ctx *ctx; 3848 3849 assert(blob != NULL); 3850 3851 ctx = calloc(1, sizeof(*ctx)); 3852 if (!ctx) { 3853 cb_fn(cb_arg, NULL, -ENOMEM); 3854 return; 3855 } 3856 3857 ctx->page_num = _spdk_bs_blobid_to_page(blob->id); 3858 ctx->bs = bs; 3859 ctx->cb_fn = cb_fn; 3860 ctx->cb_arg = cb_arg; 3861 3862 /* Close the existing blob */ 3863 spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx); 3864 } 3865 3866 static int 3867 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 3868 uint16_t value_len, bool internal) 3869 { 3870 struct spdk_xattr_tailq *xattrs; 3871 struct spdk_xattr *xattr; 3872 3873 _spdk_blob_verify_md_op(blob); 3874 3875 if (blob->md_ro) { 3876 return -EPERM; 3877 } 3878 3879 if (internal) { 3880 xattrs = &blob->xattrs_internal; 3881 blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR; 3882 } else { 3883 xattrs = &blob->xattrs; 3884 } 3885 3886 TAILQ_FOREACH(xattr, xattrs, link) { 3887 if (!strcmp(name, xattr->name)) { 3888 free(xattr->value); 3889 xattr->value_len = value_len; 3890 xattr->value = malloc(value_len); 3891 memcpy(xattr->value, value, value_len); 3892 3893 blob->state = SPDK_BLOB_STATE_DIRTY; 3894 3895 return 0; 3896 } 3897 } 3898 3899 xattr = calloc(1, sizeof(*xattr)); 3900 if (!xattr) { 3901 return -1; 3902 } 3903 xattr->name = strdup(name); 3904 xattr->value_len = value_len; 3905 xattr->value = malloc(value_len); 3906 memcpy(xattr->value, value, value_len); 3907 TAILQ_INSERT_TAIL(xattrs, xattr, link); 3908 3909 blob->state = SPDK_BLOB_STATE_DIRTY; 3910 3911 return 0; 3912 } 3913 3914 int 3915 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 3916 uint16_t value_len) 3917 { 3918 return _spdk_blob_set_xattr(blob, name, value, value_len, false); 3919 } 3920 3921 static int 3922 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal) 3923 { 3924 struct spdk_xattr_tailq *xattrs; 3925 struct spdk_xattr *xattr; 3926 3927 _spdk_blob_verify_md_op(blob); 3928 3929 if (blob->md_ro) { 3930 return -EPERM; 3931 } 3932 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 3933 3934 TAILQ_FOREACH(xattr, xattrs, link) { 3935 if (!strcmp(name, xattr->name)) { 3936 TAILQ_REMOVE(xattrs, xattr, link); 3937 free(xattr->value); 3938 free(xattr->name); 3939 free(xattr); 3940 3941 if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) { 3942 blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR; 3943 } 3944 blob->state = SPDK_BLOB_STATE_DIRTY; 3945 3946 return 0; 3947 } 3948 } 3949 3950 return -ENOENT; 3951 } 3952 3953 int 3954 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name) 3955 { 3956 return _spdk_blob_remove_xattr(blob, name, false); 3957 } 3958 3959 static int 3960 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 3961 const void **value, size_t *value_len, bool internal) 3962 { 3963 struct spdk_xattr *xattr; 3964 struct spdk_xattr_tailq *xattrs; 3965 3966 _spdk_blob_verify_md_op(blob); 3967 3968 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 3969 3970 TAILQ_FOREACH(xattr, xattrs, link) { 3971 if (!strcmp(name, xattr->name)) { 3972 *value = xattr->value; 3973 *value_len = xattr->value_len; 3974 return 0; 3975 } 3976 } 3977 return -ENOENT; 3978 } 3979 3980 int 3981 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 3982 const void **value, size_t *value_len) 3983 { 3984 return _spdk_blob_get_xattr_value(blob, name, value, value_len, false); 3985 } 3986 3987 struct spdk_xattr_names { 3988 uint32_t count; 3989 const char *names[0]; 3990 }; 3991 3992 static int 3993 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names) 3994 { 3995 struct spdk_xattr *xattr; 3996 int count = 0; 3997 3998 TAILQ_FOREACH(xattr, xattrs, link) { 3999 count++; 4000 } 4001 4002 *names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *)); 4003 if (*names == NULL) { 4004 return -ENOMEM; 4005 } 4006 4007 TAILQ_FOREACH(xattr, xattrs, link) { 4008 (*names)->names[(*names)->count++] = xattr->name; 4009 } 4010 4011 return 0; 4012 } 4013 4014 int 4015 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names) 4016 { 4017 _spdk_blob_verify_md_op(blob); 4018 4019 return _spdk_blob_get_xattr_names(&blob->xattrs, names); 4020 } 4021 4022 uint32_t 4023 spdk_xattr_names_get_count(struct spdk_xattr_names *names) 4024 { 4025 assert(names != NULL); 4026 4027 return names->count; 4028 } 4029 4030 const char * 4031 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index) 4032 { 4033 if (index >= names->count) { 4034 return NULL; 4035 } 4036 4037 return names->names[index]; 4038 } 4039 4040 void 4041 spdk_xattr_names_free(struct spdk_xattr_names *names) 4042 { 4043 free(names); 4044 } 4045 4046 struct spdk_bs_type 4047 spdk_bs_get_bstype(struct spdk_blob_store *bs) 4048 { 4049 return bs->bstype; 4050 } 4051 4052 void 4053 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype) 4054 { 4055 memcpy(&bs->bstype, &bstype, sizeof(bstype)); 4056 } 4057 4058 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB) 4059