1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/stdinc.h" 35 36 #include "spdk/blob.h" 37 #include "spdk/crc32.h" 38 #include "spdk/env.h" 39 #include "spdk/queue.h" 40 #include "spdk/thread.h" 41 #include "spdk/bit_array.h" 42 #include "spdk/likely.h" 43 44 #include "spdk_internal/assert.h" 45 #include "spdk_internal/log.h" 46 47 #include "blobstore.h" 48 49 #define BLOB_CRC32C_INITIAL 0xffffffffUL 50 51 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs); 52 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs); 53 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno); 54 static void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 55 uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg); 56 57 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 58 uint16_t value_len, bool internal); 59 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 60 const void **value, size_t *value_len, bool internal); 61 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal); 62 63 static void 64 _spdk_blob_verify_md_op(struct spdk_blob *blob) 65 { 66 assert(blob != NULL); 67 assert(spdk_get_thread() == blob->bs->md_thread); 68 assert(blob->state != SPDK_BLOB_STATE_LOADING); 69 } 70 71 static inline size_t 72 divide_round_up(size_t num, size_t divisor) 73 { 74 return (num + divisor - 1) / divisor; 75 } 76 77 static void 78 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 79 { 80 assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters)); 81 assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false); 82 assert(bs->num_free_clusters > 0); 83 84 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num); 85 86 spdk_bit_array_set(bs->used_clusters, cluster_num); 87 bs->num_free_clusters--; 88 } 89 90 static int 91 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster) 92 { 93 uint64_t *cluster_lba = &blob->active.clusters[cluster_num]; 94 95 _spdk_blob_verify_md_op(blob); 96 97 if (*cluster_lba != 0) { 98 return -EEXIST; 99 } 100 101 *cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster); 102 return 0; 103 } 104 105 static int 106 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num, 107 uint64_t *lowest_free_cluster, bool update_map) 108 { 109 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 110 *lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters, 111 *lowest_free_cluster); 112 if (*lowest_free_cluster >= blob->bs->total_clusters) { 113 /* No more free clusters. Cannot satisfy the request */ 114 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 115 return -ENOSPC; 116 } 117 118 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id); 119 _spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster); 120 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 121 122 if (update_map) { 123 _spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster); 124 } 125 126 return 0; 127 } 128 129 static void 130 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 131 { 132 assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters)); 133 assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true); 134 assert(bs->num_free_clusters < bs->total_clusters); 135 136 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num); 137 138 pthread_mutex_lock(&bs->used_clusters_mutex); 139 spdk_bit_array_clear(bs->used_clusters, cluster_num); 140 bs->num_free_clusters++; 141 pthread_mutex_unlock(&bs->used_clusters_mutex); 142 } 143 144 static void 145 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs) 146 { 147 xattrs->count = 0; 148 xattrs->names = NULL; 149 xattrs->ctx = NULL; 150 xattrs->get_value = NULL; 151 } 152 153 void 154 spdk_blob_opts_init(struct spdk_blob_opts *opts) 155 { 156 opts->num_clusters = 0; 157 opts->thin_provision = false; 158 _spdk_blob_xattrs_init(&opts->xattrs); 159 } 160 161 static struct spdk_blob * 162 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id) 163 { 164 struct spdk_blob *blob; 165 166 blob = calloc(1, sizeof(*blob)); 167 if (!blob) { 168 return NULL; 169 } 170 171 blob->id = id; 172 blob->bs = bs; 173 174 blob->parent_id = SPDK_BLOBID_INVALID; 175 176 blob->state = SPDK_BLOB_STATE_DIRTY; 177 blob->active.num_pages = 1; 178 blob->active.pages = calloc(1, sizeof(*blob->active.pages)); 179 if (!blob->active.pages) { 180 free(blob); 181 return NULL; 182 } 183 184 blob->active.pages[0] = _spdk_bs_blobid_to_page(id); 185 186 TAILQ_INIT(&blob->xattrs); 187 TAILQ_INIT(&blob->xattrs_internal); 188 189 return blob; 190 } 191 192 static void 193 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs) 194 { 195 struct spdk_xattr *xattr, *xattr_tmp; 196 197 TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) { 198 TAILQ_REMOVE(xattrs, xattr, link); 199 free(xattr->name); 200 free(xattr->value); 201 free(xattr); 202 } 203 } 204 205 static void 206 _spdk_blob_free(struct spdk_blob *blob) 207 { 208 assert(blob != NULL); 209 210 free(blob->active.clusters); 211 free(blob->clean.clusters); 212 free(blob->active.pages); 213 free(blob->clean.pages); 214 215 _spdk_xattrs_free(&blob->xattrs); 216 _spdk_xattrs_free(&blob->xattrs_internal); 217 218 if (blob->back_bs_dev) { 219 blob->back_bs_dev->destroy(blob->back_bs_dev); 220 } 221 222 free(blob); 223 } 224 225 struct freeze_io_ctx { 226 struct spdk_bs_cpl cpl; 227 struct spdk_blob *blob; 228 }; 229 230 static void 231 _spdk_blob_io_sync(struct spdk_io_channel_iter *i) 232 { 233 spdk_for_each_channel_continue(i, 0); 234 } 235 236 static void 237 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i) 238 { 239 struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i); 240 struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch); 241 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 242 struct spdk_bs_request_set *set; 243 struct spdk_bs_user_op_args *args; 244 spdk_bs_user_op_t *op, *tmp; 245 246 TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) { 247 set = (struct spdk_bs_request_set *)op; 248 args = &set->u.user_op; 249 250 if (args->blob == ctx->blob) { 251 TAILQ_REMOVE(&ch->queued_io, op, link); 252 spdk_bs_user_op_execute(op); 253 } 254 } 255 256 spdk_for_each_channel_continue(i, 0); 257 } 258 259 static void 260 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status) 261 { 262 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 263 264 ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0); 265 266 free(ctx); 267 } 268 269 static void 270 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 271 { 272 struct freeze_io_ctx *ctx; 273 274 ctx = calloc(1, sizeof(*ctx)); 275 if (!ctx) { 276 cb_fn(cb_arg, -ENOMEM); 277 return; 278 } 279 280 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 281 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 282 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 283 ctx->blob = blob; 284 285 /* Freeze I/O on blob */ 286 blob->frozen_refcnt++; 287 288 if (blob->frozen_refcnt == 1) { 289 spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl); 290 } else { 291 cb_fn(cb_arg, 0); 292 free(ctx); 293 } 294 } 295 296 static void 297 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 298 { 299 struct freeze_io_ctx *ctx; 300 301 ctx = calloc(1, sizeof(*ctx)); 302 if (!ctx) { 303 cb_fn(cb_arg, -ENOMEM); 304 return; 305 } 306 307 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 308 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 309 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 310 ctx->blob = blob; 311 312 assert(blob->frozen_refcnt > 0); 313 314 blob->frozen_refcnt--; 315 316 if (blob->frozen_refcnt == 0) { 317 spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl); 318 } else { 319 cb_fn(cb_arg, 0); 320 free(ctx); 321 } 322 } 323 324 static int 325 _spdk_blob_mark_clean(struct spdk_blob *blob) 326 { 327 uint64_t *clusters = NULL; 328 uint32_t *pages = NULL; 329 330 assert(blob != NULL); 331 332 if (blob->active.num_clusters) { 333 assert(blob->active.clusters); 334 clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters)); 335 if (!clusters) { 336 return -ENOMEM; 337 } 338 memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*clusters)); 339 } 340 341 if (blob->active.num_pages) { 342 assert(blob->active.pages); 343 pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages)); 344 if (!pages) { 345 free(clusters); 346 return -ENOMEM; 347 } 348 memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*pages)); 349 } 350 351 free(blob->clean.clusters); 352 free(blob->clean.pages); 353 354 blob->clean.num_clusters = blob->active.num_clusters; 355 blob->clean.clusters = blob->active.clusters; 356 blob->clean.num_pages = blob->active.num_pages; 357 blob->clean.pages = blob->active.pages; 358 359 blob->active.clusters = clusters; 360 blob->active.pages = pages; 361 362 /* If the metadata was dirtied again while the metadata was being written to disk, 363 * we do not want to revert the DIRTY state back to CLEAN here. 364 */ 365 if (blob->state == SPDK_BLOB_STATE_LOADING) { 366 blob->state = SPDK_BLOB_STATE_CLEAN; 367 } 368 369 return 0; 370 } 371 372 static int 373 _spdk_blob_deserialize_xattr(struct spdk_blob *blob, 374 struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal) 375 { 376 struct spdk_xattr *xattr; 377 378 if (desc_xattr->length != sizeof(desc_xattr->name_length) + 379 sizeof(desc_xattr->value_length) + 380 desc_xattr->name_length + desc_xattr->value_length) { 381 return -EINVAL; 382 } 383 384 xattr = calloc(1, sizeof(*xattr)); 385 if (xattr == NULL) { 386 return -ENOMEM; 387 } 388 389 xattr->name = malloc(desc_xattr->name_length + 1); 390 if (xattr->name == NULL) { 391 free(xattr); 392 return -ENOMEM; 393 } 394 memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length); 395 xattr->name[desc_xattr->name_length] = '\0'; 396 397 xattr->value = malloc(desc_xattr->value_length); 398 if (xattr->value == NULL) { 399 free(xattr->name); 400 free(xattr); 401 return -ENOMEM; 402 } 403 xattr->value_len = desc_xattr->value_length; 404 memcpy(xattr->value, 405 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 406 desc_xattr->value_length); 407 408 TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link); 409 410 return 0; 411 } 412 413 414 static int 415 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob) 416 { 417 struct spdk_blob_md_descriptor *desc; 418 size_t cur_desc = 0; 419 void *tmp; 420 421 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 422 while (cur_desc < sizeof(page->descriptors)) { 423 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 424 if (desc->length == 0) { 425 /* If padding and length are 0, this terminates the page */ 426 break; 427 } 428 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 429 struct spdk_blob_md_descriptor_flags *desc_flags; 430 431 desc_flags = (struct spdk_blob_md_descriptor_flags *)desc; 432 433 if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) { 434 return -EINVAL; 435 } 436 437 if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) != 438 SPDK_BLOB_INVALID_FLAGS_MASK) { 439 return -EINVAL; 440 } 441 442 if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) != 443 SPDK_BLOB_DATA_RO_FLAGS_MASK) { 444 blob->data_ro = true; 445 blob->md_ro = true; 446 } 447 448 if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) != 449 SPDK_BLOB_MD_RO_FLAGS_MASK) { 450 blob->md_ro = true; 451 } 452 453 if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 454 blob->data_ro = true; 455 blob->md_ro = true; 456 } 457 458 blob->invalid_flags = desc_flags->invalid_flags; 459 blob->data_ro_flags = desc_flags->data_ro_flags; 460 blob->md_ro_flags = desc_flags->md_ro_flags; 461 462 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 463 struct spdk_blob_md_descriptor_extent *desc_extent; 464 unsigned int i, j; 465 unsigned int cluster_count = blob->active.num_clusters; 466 467 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 468 469 if (desc_extent->length == 0 || 470 (desc_extent->length % sizeof(desc_extent->extents[0]) != 0)) { 471 return -EINVAL; 472 } 473 474 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 475 for (j = 0; j < desc_extent->extents[i].length; j++) { 476 if (desc_extent->extents[i].cluster_idx != 0) { 477 if (!spdk_bit_array_get(blob->bs->used_clusters, 478 desc_extent->extents[i].cluster_idx + j)) { 479 return -EINVAL; 480 } 481 } 482 cluster_count++; 483 } 484 } 485 486 if (cluster_count == 0) { 487 return -EINVAL; 488 } 489 tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t)); 490 if (tmp == NULL) { 491 return -ENOMEM; 492 } 493 blob->active.clusters = tmp; 494 blob->active.cluster_array_size = cluster_count; 495 496 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 497 for (j = 0; j < desc_extent->extents[i].length; j++) { 498 if (desc_extent->extents[i].cluster_idx != 0) { 499 blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs, 500 desc_extent->extents[i].cluster_idx + j); 501 } else if (spdk_blob_is_thin_provisioned(blob)) { 502 blob->active.clusters[blob->active.num_clusters++] = 0; 503 } else { 504 return -EINVAL; 505 } 506 } 507 } 508 509 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 510 int rc; 511 512 rc = _spdk_blob_deserialize_xattr(blob, 513 (struct spdk_blob_md_descriptor_xattr *) desc, false); 514 if (rc != 0) { 515 return rc; 516 } 517 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 518 int rc; 519 520 rc = _spdk_blob_deserialize_xattr(blob, 521 (struct spdk_blob_md_descriptor_xattr *) desc, true); 522 if (rc != 0) { 523 return rc; 524 } 525 } else { 526 /* Unrecognized descriptor type. Do not fail - just continue to the 527 * next descriptor. If this descriptor is associated with some feature 528 * defined in a newer version of blobstore, that version of blobstore 529 * should create and set an associated feature flag to specify if this 530 * blob can be loaded or not. 531 */ 532 } 533 534 /* Advance to the next descriptor */ 535 cur_desc += sizeof(*desc) + desc->length; 536 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 537 break; 538 } 539 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 540 } 541 542 return 0; 543 } 544 545 static int 546 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count, 547 struct spdk_blob *blob) 548 { 549 const struct spdk_blob_md_page *page; 550 uint32_t i; 551 int rc; 552 553 assert(page_count > 0); 554 assert(pages[0].sequence_num == 0); 555 assert(blob != NULL); 556 assert(blob->state == SPDK_BLOB_STATE_LOADING); 557 assert(blob->active.clusters == NULL); 558 559 /* The blobid provided doesn't match what's in the MD, this can 560 * happen for example if a bogus blobid is passed in through open. 561 */ 562 if (blob->id != pages[0].id) { 563 SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n", 564 blob->id, pages[0].id); 565 return -ENOENT; 566 } 567 568 for (i = 0; i < page_count; i++) { 569 page = &pages[i]; 570 571 assert(page->id == blob->id); 572 assert(page->sequence_num == i); 573 574 rc = _spdk_blob_parse_page(page, blob); 575 if (rc != 0) { 576 return rc; 577 } 578 } 579 580 return 0; 581 } 582 583 static int 584 _spdk_blob_serialize_add_page(const struct spdk_blob *blob, 585 struct spdk_blob_md_page **pages, 586 uint32_t *page_count, 587 struct spdk_blob_md_page **last_page) 588 { 589 struct spdk_blob_md_page *page; 590 591 assert(pages != NULL); 592 assert(page_count != NULL); 593 594 if (*page_count == 0) { 595 assert(*pages == NULL); 596 *page_count = 1; 597 *pages = spdk_dma_malloc(SPDK_BS_PAGE_SIZE, 598 SPDK_BS_PAGE_SIZE, 599 NULL); 600 } else { 601 assert(*pages != NULL); 602 (*page_count)++; 603 *pages = spdk_dma_realloc(*pages, 604 SPDK_BS_PAGE_SIZE * (*page_count), 605 SPDK_BS_PAGE_SIZE, 606 NULL); 607 } 608 609 if (*pages == NULL) { 610 *page_count = 0; 611 *last_page = NULL; 612 return -ENOMEM; 613 } 614 615 page = &(*pages)[*page_count - 1]; 616 memset(page, 0, sizeof(*page)); 617 page->id = blob->id; 618 page->sequence_num = *page_count - 1; 619 page->next = SPDK_INVALID_MD_PAGE; 620 *last_page = page; 621 622 return 0; 623 } 624 625 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor. 626 * Update required_sz on both success and failure. 627 * 628 */ 629 static int 630 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr, 631 uint8_t *buf, size_t buf_sz, 632 size_t *required_sz, bool internal) 633 { 634 struct spdk_blob_md_descriptor_xattr *desc; 635 636 *required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) + 637 strlen(xattr->name) + 638 xattr->value_len; 639 640 if (buf_sz < *required_sz) { 641 return -1; 642 } 643 644 desc = (struct spdk_blob_md_descriptor_xattr *)buf; 645 646 desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR; 647 desc->length = sizeof(desc->name_length) + 648 sizeof(desc->value_length) + 649 strlen(xattr->name) + 650 xattr->value_len; 651 desc->name_length = strlen(xattr->name); 652 desc->value_length = xattr->value_len; 653 654 memcpy(desc->name, xattr->name, desc->name_length); 655 memcpy((void *)((uintptr_t)desc->name + desc->name_length), 656 xattr->value, 657 desc->value_length); 658 659 return 0; 660 } 661 662 static void 663 _spdk_blob_serialize_extent(const struct spdk_blob *blob, 664 uint64_t start_cluster, uint64_t *next_cluster, 665 uint8_t *buf, size_t buf_sz) 666 { 667 struct spdk_blob_md_descriptor_extent *desc; 668 size_t cur_sz; 669 uint64_t i, extent_idx; 670 uint64_t lba, lba_per_cluster, lba_count; 671 672 /* The buffer must have room for at least one extent */ 673 cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]); 674 if (buf_sz < cur_sz) { 675 *next_cluster = start_cluster; 676 return; 677 } 678 679 desc = (struct spdk_blob_md_descriptor_extent *)buf; 680 desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT; 681 682 lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1); 683 684 lba = blob->active.clusters[start_cluster]; 685 lba_count = lba_per_cluster; 686 extent_idx = 0; 687 for (i = start_cluster + 1; i < blob->active.num_clusters; i++) { 688 if ((lba + lba_count) == blob->active.clusters[i]) { 689 lba_count += lba_per_cluster; 690 continue; 691 } else if (lba == 0 && blob->active.clusters[i] == 0) { 692 lba_count += lba_per_cluster; 693 continue; 694 } 695 desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 696 desc->extents[extent_idx].length = lba_count / lba_per_cluster; 697 extent_idx++; 698 699 cur_sz += sizeof(desc->extents[extent_idx]); 700 701 if (buf_sz < cur_sz) { 702 /* If we ran out of buffer space, return */ 703 desc->length = sizeof(desc->extents[0]) * extent_idx; 704 *next_cluster = i; 705 return; 706 } 707 708 lba = blob->active.clusters[i]; 709 lba_count = lba_per_cluster; 710 } 711 712 desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 713 desc->extents[extent_idx].length = lba_count / lba_per_cluster; 714 extent_idx++; 715 716 desc->length = sizeof(desc->extents[0]) * extent_idx; 717 *next_cluster = blob->active.num_clusters; 718 719 return; 720 } 721 722 static void 723 _spdk_blob_serialize_flags(const struct spdk_blob *blob, 724 uint8_t *buf, size_t *buf_sz) 725 { 726 struct spdk_blob_md_descriptor_flags *desc; 727 728 /* 729 * Flags get serialized first, so we should always have room for the flags 730 * descriptor. 731 */ 732 assert(*buf_sz >= sizeof(*desc)); 733 734 desc = (struct spdk_blob_md_descriptor_flags *)buf; 735 desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS; 736 desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor); 737 desc->invalid_flags = blob->invalid_flags; 738 desc->data_ro_flags = blob->data_ro_flags; 739 desc->md_ro_flags = blob->md_ro_flags; 740 741 *buf_sz -= sizeof(*desc); 742 } 743 744 static int 745 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob, 746 const struct spdk_xattr_tailq *xattrs, bool internal, 747 struct spdk_blob_md_page **pages, 748 struct spdk_blob_md_page *cur_page, 749 uint32_t *page_count, uint8_t **buf, 750 size_t *remaining_sz) 751 { 752 const struct spdk_xattr *xattr; 753 int rc; 754 755 TAILQ_FOREACH(xattr, xattrs, link) { 756 size_t required_sz = 0; 757 758 rc = _spdk_blob_serialize_xattr(xattr, 759 *buf, *remaining_sz, 760 &required_sz, internal); 761 if (rc < 0) { 762 /* Need to add a new page to the chain */ 763 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, 764 &cur_page); 765 if (rc < 0) { 766 spdk_dma_free(*pages); 767 *pages = NULL; 768 *page_count = 0; 769 return rc; 770 } 771 772 *buf = (uint8_t *)cur_page->descriptors; 773 *remaining_sz = sizeof(cur_page->descriptors); 774 775 /* Try again */ 776 required_sz = 0; 777 rc = _spdk_blob_serialize_xattr(xattr, 778 *buf, *remaining_sz, 779 &required_sz, internal); 780 781 if (rc < 0) { 782 spdk_dma_free(*pages); 783 *pages = NULL; 784 *page_count = 0; 785 return rc; 786 } 787 } 788 789 *remaining_sz -= required_sz; 790 *buf += required_sz; 791 } 792 793 return 0; 794 } 795 796 static int 797 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages, 798 uint32_t *page_count) 799 { 800 struct spdk_blob_md_page *cur_page; 801 int rc; 802 uint8_t *buf; 803 size_t remaining_sz; 804 uint64_t last_cluster; 805 806 assert(pages != NULL); 807 assert(page_count != NULL); 808 assert(blob != NULL); 809 assert(blob->state == SPDK_BLOB_STATE_DIRTY); 810 811 *pages = NULL; 812 *page_count = 0; 813 814 /* A blob always has at least 1 page, even if it has no descriptors */ 815 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page); 816 if (rc < 0) { 817 return rc; 818 } 819 820 buf = (uint8_t *)cur_page->descriptors; 821 remaining_sz = sizeof(cur_page->descriptors); 822 823 /* Serialize flags */ 824 _spdk_blob_serialize_flags(blob, buf, &remaining_sz); 825 buf += sizeof(struct spdk_blob_md_descriptor_flags); 826 827 /* Serialize xattrs */ 828 rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false, 829 pages, cur_page, page_count, &buf, &remaining_sz); 830 if (rc < 0) { 831 return rc; 832 } 833 834 /* Serialize internal xattrs */ 835 rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true, 836 pages, cur_page, page_count, &buf, &remaining_sz); 837 if (rc < 0) { 838 return rc; 839 } 840 841 /* Serialize extents */ 842 last_cluster = 0; 843 while (last_cluster < blob->active.num_clusters) { 844 _spdk_blob_serialize_extent(blob, last_cluster, &last_cluster, 845 buf, remaining_sz); 846 847 if (last_cluster == blob->active.num_clusters) { 848 break; 849 } 850 851 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, 852 &cur_page); 853 if (rc < 0) { 854 return rc; 855 } 856 857 buf = (uint8_t *)cur_page->descriptors; 858 remaining_sz = sizeof(cur_page->descriptors); 859 } 860 861 return 0; 862 } 863 864 struct spdk_blob_load_ctx { 865 struct spdk_blob *blob; 866 867 struct spdk_blob_md_page *pages; 868 uint32_t num_pages; 869 spdk_bs_sequence_t *seq; 870 871 spdk_bs_sequence_cpl cb_fn; 872 void *cb_arg; 873 }; 874 875 static uint32_t 876 _spdk_blob_md_page_calc_crc(void *page) 877 { 878 uint32_t crc; 879 880 crc = BLOB_CRC32C_INITIAL; 881 crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc); 882 crc ^= BLOB_CRC32C_INITIAL; 883 884 return crc; 885 886 } 887 888 static void 889 _spdk_blob_load_final(void *cb_arg, int bserrno) 890 { 891 struct spdk_blob_load_ctx *ctx = cb_arg; 892 struct spdk_blob *blob = ctx->blob; 893 894 _spdk_blob_mark_clean(blob); 895 896 ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno); 897 898 /* Free the memory */ 899 spdk_dma_free(ctx->pages); 900 free(ctx); 901 } 902 903 static void 904 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno) 905 { 906 struct spdk_blob_load_ctx *ctx = cb_arg; 907 struct spdk_blob *blob = ctx->blob; 908 909 if (bserrno != 0) { 910 goto error; 911 } 912 913 blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot); 914 915 if (blob->back_bs_dev == NULL) { 916 bserrno = -ENOMEM; 917 goto error; 918 } 919 920 _spdk_blob_load_final(ctx, bserrno); 921 return; 922 923 error: 924 SPDK_ERRLOG("Snapshot fail\n"); 925 _spdk_blob_free(blob); 926 ctx->cb_fn(ctx->seq, NULL, bserrno); 927 spdk_dma_free(ctx->pages); 928 free(ctx); 929 } 930 931 static void 932 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 933 { 934 struct spdk_blob_load_ctx *ctx = cb_arg; 935 struct spdk_blob *blob = ctx->blob; 936 struct spdk_blob_md_page *page; 937 const void *value; 938 size_t len; 939 int rc; 940 uint32_t crc; 941 942 page = &ctx->pages[ctx->num_pages - 1]; 943 crc = _spdk_blob_md_page_calc_crc(page); 944 if (crc != page->crc) { 945 SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages); 946 _spdk_blob_free(blob); 947 ctx->cb_fn(seq, NULL, -EINVAL); 948 spdk_dma_free(ctx->pages); 949 free(ctx); 950 return; 951 } 952 953 if (page->next != SPDK_INVALID_MD_PAGE) { 954 uint32_t next_page = page->next; 955 uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page); 956 957 958 assert(next_lba < (blob->bs->md_start + blob->bs->md_len)); 959 960 /* Read the next page */ 961 ctx->num_pages++; 962 ctx->pages = spdk_dma_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages), 963 sizeof(*page), NULL); 964 if (ctx->pages == NULL) { 965 ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM); 966 free(ctx); 967 return; 968 } 969 970 spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1], 971 next_lba, 972 _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)), 973 _spdk_blob_load_cpl, ctx); 974 return; 975 } 976 977 /* Parse the pages */ 978 rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob); 979 if (rc) { 980 _spdk_blob_free(blob); 981 ctx->cb_fn(seq, NULL, rc); 982 spdk_dma_free(ctx->pages); 983 free(ctx); 984 return; 985 } 986 ctx->seq = seq; 987 988 989 if (spdk_blob_is_thin_provisioned(blob)) { 990 rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true); 991 if (rc == 0) { 992 if (len != sizeof(spdk_blob_id)) { 993 _spdk_blob_free(blob); 994 ctx->cb_fn(seq, NULL, -EINVAL); 995 spdk_dma_free(ctx->pages); 996 free(ctx); 997 return; 998 } 999 /* open snapshot blob and continue in the callback function */ 1000 blob->parent_id = *(spdk_blob_id *)value; 1001 spdk_bs_open_blob(blob->bs, blob->parent_id, 1002 _spdk_blob_load_snapshot_cpl, ctx); 1003 return; 1004 } else { 1005 /* add zeroes_dev for thin provisioned blob */ 1006 blob->back_bs_dev = spdk_bs_create_zeroes_dev(); 1007 } 1008 } else { 1009 /* standard blob */ 1010 blob->back_bs_dev = NULL; 1011 } 1012 _spdk_blob_load_final(ctx, bserrno); 1013 } 1014 1015 /* Load a blob from disk given a blobid */ 1016 static void 1017 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 1018 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 1019 { 1020 struct spdk_blob_load_ctx *ctx; 1021 struct spdk_blob_store *bs; 1022 uint32_t page_num; 1023 uint64_t lba; 1024 1025 _spdk_blob_verify_md_op(blob); 1026 1027 bs = blob->bs; 1028 1029 ctx = calloc(1, sizeof(*ctx)); 1030 if (!ctx) { 1031 cb_fn(seq, cb_arg, -ENOMEM); 1032 return; 1033 } 1034 1035 ctx->blob = blob; 1036 ctx->pages = spdk_dma_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, 1037 SPDK_BS_PAGE_SIZE, NULL); 1038 if (!ctx->pages) { 1039 free(ctx); 1040 cb_fn(seq, cb_arg, -ENOMEM); 1041 return; 1042 } 1043 ctx->num_pages = 1; 1044 ctx->cb_fn = cb_fn; 1045 ctx->cb_arg = cb_arg; 1046 1047 page_num = _spdk_bs_blobid_to_page(blob->id); 1048 lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num); 1049 1050 blob->state = SPDK_BLOB_STATE_LOADING; 1051 1052 spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba, 1053 _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE), 1054 _spdk_blob_load_cpl, ctx); 1055 } 1056 1057 struct spdk_blob_persist_ctx { 1058 struct spdk_blob *blob; 1059 1060 struct spdk_bs_super_block *super; 1061 1062 struct spdk_blob_md_page *pages; 1063 1064 uint64_t idx; 1065 1066 spdk_bs_sequence_t *seq; 1067 spdk_bs_sequence_cpl cb_fn; 1068 void *cb_arg; 1069 }; 1070 1071 static void 1072 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1073 { 1074 struct spdk_blob_persist_ctx *ctx = cb_arg; 1075 struct spdk_blob *blob = ctx->blob; 1076 1077 if (bserrno == 0) { 1078 _spdk_blob_mark_clean(blob); 1079 } 1080 1081 /* Call user callback */ 1082 ctx->cb_fn(seq, ctx->cb_arg, bserrno); 1083 1084 /* Free the memory */ 1085 spdk_dma_free(ctx->pages); 1086 free(ctx); 1087 } 1088 1089 static void 1090 _spdk_blob_persist_unmap_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1091 { 1092 struct spdk_blob_persist_ctx *ctx = cb_arg; 1093 struct spdk_blob *blob = ctx->blob; 1094 struct spdk_blob_store *bs = blob->bs; 1095 void *tmp; 1096 size_t i; 1097 1098 /* Release all clusters that were truncated */ 1099 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1100 uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]); 1101 1102 /* Nothing to release if it was not allocated */ 1103 if (blob->active.clusters[i] != 0) { 1104 _spdk_bs_release_cluster(bs, cluster_num); 1105 } 1106 } 1107 1108 if (blob->active.num_clusters == 0) { 1109 free(blob->active.clusters); 1110 blob->active.clusters = NULL; 1111 blob->active.cluster_array_size = 0; 1112 } else { 1113 tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters); 1114 assert(tmp != NULL); 1115 blob->active.clusters = tmp; 1116 blob->active.cluster_array_size = blob->active.num_clusters; 1117 } 1118 1119 _spdk_blob_persist_complete(seq, ctx, bserrno); 1120 } 1121 1122 static void 1123 _spdk_blob_persist_unmap_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1124 { 1125 struct spdk_blob_persist_ctx *ctx = cb_arg; 1126 struct spdk_blob *blob = ctx->blob; 1127 struct spdk_blob_store *bs = blob->bs; 1128 spdk_bs_batch_t *batch; 1129 size_t i; 1130 uint64_t lba; 1131 uint32_t lba_count; 1132 1133 /* Clusters don't move around in blobs. The list shrinks or grows 1134 * at the end, but no changes ever occur in the middle of the list. 1135 */ 1136 1137 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_unmap_clusters_cpl, ctx); 1138 1139 /* Unmap all clusters that were truncated */ 1140 lba = 0; 1141 lba_count = 0; 1142 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1143 uint64_t next_lba = blob->active.clusters[i]; 1144 uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1); 1145 1146 if (next_lba > 0 && (lba + lba_count) == next_lba) { 1147 /* This cluster is contiguous with the previous one. */ 1148 lba_count += next_lba_count; 1149 continue; 1150 } 1151 1152 /* This cluster is not contiguous with the previous one. */ 1153 1154 /* If a run of LBAs previously existing, send them 1155 * as an unmap. 1156 */ 1157 if (lba_count > 0) { 1158 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 1159 } 1160 1161 /* Start building the next batch */ 1162 lba = next_lba; 1163 if (next_lba > 0) { 1164 lba_count = next_lba_count; 1165 } else { 1166 lba_count = 0; 1167 } 1168 } 1169 1170 /* If we ended with a contiguous set of LBAs, send the unmap now */ 1171 if (lba_count > 0) { 1172 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 1173 } 1174 1175 spdk_bs_batch_close(batch); 1176 } 1177 1178 static void 1179 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1180 { 1181 struct spdk_blob_persist_ctx *ctx = cb_arg; 1182 struct spdk_blob *blob = ctx->blob; 1183 struct spdk_blob_store *bs = blob->bs; 1184 size_t i; 1185 1186 /* This loop starts at 1 because the first page is special and handled 1187 * below. The pages (except the first) are never written in place, 1188 * so any pages in the clean list must be zeroed. 1189 */ 1190 for (i = 1; i < blob->clean.num_pages; i++) { 1191 spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]); 1192 } 1193 1194 if (blob->active.num_pages == 0) { 1195 uint32_t page_num; 1196 1197 page_num = _spdk_bs_blobid_to_page(blob->id); 1198 spdk_bit_array_clear(bs->used_md_pages, page_num); 1199 } 1200 1201 /* Move on to unmapping clusters */ 1202 _spdk_blob_persist_unmap_clusters(seq, ctx, 0); 1203 } 1204 1205 static void 1206 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1207 { 1208 struct spdk_blob_persist_ctx *ctx = cb_arg; 1209 struct spdk_blob *blob = ctx->blob; 1210 struct spdk_blob_store *bs = blob->bs; 1211 uint64_t lba; 1212 uint32_t lba_count; 1213 spdk_bs_batch_t *batch; 1214 size_t i; 1215 1216 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx); 1217 1218 lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE); 1219 1220 /* This loop starts at 1 because the first page is special and handled 1221 * below. The pages (except the first) are never written in place, 1222 * so any pages in the clean list must be zeroed. 1223 */ 1224 for (i = 1; i < blob->clean.num_pages; i++) { 1225 lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]); 1226 1227 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1228 } 1229 1230 /* The first page will only be zeroed if this is a delete. */ 1231 if (blob->active.num_pages == 0) { 1232 uint32_t page_num; 1233 1234 /* The first page in the metadata goes where the blobid indicates */ 1235 page_num = _spdk_bs_blobid_to_page(blob->id); 1236 lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num); 1237 1238 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1239 } 1240 1241 spdk_bs_batch_close(batch); 1242 } 1243 1244 static void 1245 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1246 { 1247 struct spdk_blob_persist_ctx *ctx = cb_arg; 1248 struct spdk_blob *blob = ctx->blob; 1249 struct spdk_blob_store *bs = blob->bs; 1250 uint64_t lba; 1251 uint32_t lba_count; 1252 struct spdk_blob_md_page *page; 1253 1254 if (blob->active.num_pages == 0) { 1255 /* Move on to the next step */ 1256 _spdk_blob_persist_zero_pages(seq, ctx, 0); 1257 return; 1258 } 1259 1260 lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page)); 1261 1262 page = &ctx->pages[0]; 1263 /* The first page in the metadata goes where the blobid indicates */ 1264 lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id)); 1265 1266 spdk_bs_sequence_write_dev(seq, page, lba, lba_count, 1267 _spdk_blob_persist_zero_pages, ctx); 1268 } 1269 1270 static void 1271 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1272 { 1273 struct spdk_blob_persist_ctx *ctx = cb_arg; 1274 struct spdk_blob *blob = ctx->blob; 1275 struct spdk_blob_store *bs = blob->bs; 1276 uint64_t lba; 1277 uint32_t lba_count; 1278 struct spdk_blob_md_page *page; 1279 spdk_bs_batch_t *batch; 1280 size_t i; 1281 1282 /* Clusters don't move around in blobs. The list shrinks or grows 1283 * at the end, but no changes ever occur in the middle of the list. 1284 */ 1285 1286 lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page)); 1287 1288 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx); 1289 1290 /* This starts at 1. The root page is not written until 1291 * all of the others are finished 1292 */ 1293 for (i = 1; i < blob->active.num_pages; i++) { 1294 page = &ctx->pages[i]; 1295 assert(page->sequence_num == i); 1296 1297 lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]); 1298 1299 spdk_bs_batch_write_dev(batch, page, lba, lba_count); 1300 } 1301 1302 spdk_bs_batch_close(batch); 1303 } 1304 1305 static int 1306 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz) 1307 { 1308 uint64_t i; 1309 uint64_t *tmp; 1310 uint64_t lfc; /* lowest free cluster */ 1311 uint64_t num_clusters; 1312 struct spdk_blob_store *bs; 1313 1314 bs = blob->bs; 1315 1316 _spdk_blob_verify_md_op(blob); 1317 1318 if (blob->active.num_clusters == sz) { 1319 return 0; 1320 } 1321 1322 if (blob->active.num_clusters < blob->active.cluster_array_size) { 1323 /* If this blob was resized to be larger, then smaller, then 1324 * larger without syncing, then the cluster array already 1325 * contains spare assigned clusters we can use. 1326 */ 1327 num_clusters = spdk_min(blob->active.cluster_array_size, 1328 sz); 1329 } else { 1330 num_clusters = blob->active.num_clusters; 1331 } 1332 1333 /* Do two passes - one to verify that we can obtain enough clusters 1334 * and another to actually claim them. 1335 */ 1336 1337 if (spdk_blob_is_thin_provisioned(blob) == false) { 1338 lfc = 0; 1339 for (i = num_clusters; i < sz; i++) { 1340 lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc); 1341 if (lfc >= bs->total_clusters) { 1342 /* No more free clusters. Cannot satisfy the request */ 1343 return -ENOSPC; 1344 } 1345 lfc++; 1346 } 1347 } 1348 1349 if (sz > num_clusters) { 1350 /* Expand the cluster array if necessary. 1351 * We only shrink the array when persisting. 1352 */ 1353 tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz); 1354 if (sz > 0 && tmp == NULL) { 1355 return -ENOMEM; 1356 } 1357 memset(tmp + blob->active.cluster_array_size, 0, 1358 sizeof(uint64_t) * (sz - blob->active.cluster_array_size)); 1359 blob->active.clusters = tmp; 1360 blob->active.cluster_array_size = sz; 1361 } 1362 1363 blob->state = SPDK_BLOB_STATE_DIRTY; 1364 1365 if (spdk_blob_is_thin_provisioned(blob) == false) { 1366 lfc = 0; 1367 for (i = num_clusters; i < sz; i++) { 1368 _spdk_bs_allocate_cluster(blob, i, &lfc, true); 1369 lfc++; 1370 } 1371 } 1372 1373 blob->active.num_clusters = sz; 1374 1375 return 0; 1376 } 1377 1378 static void 1379 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx) 1380 { 1381 spdk_bs_sequence_t *seq = ctx->seq; 1382 struct spdk_blob *blob = ctx->blob; 1383 struct spdk_blob_store *bs = blob->bs; 1384 uint64_t i; 1385 uint32_t page_num; 1386 void *tmp; 1387 int rc; 1388 1389 if (blob->active.num_pages == 0) { 1390 /* This is the signal that the blob should be deleted. 1391 * Immediately jump to the clean up routine. */ 1392 assert(blob->clean.num_pages > 0); 1393 ctx->idx = blob->clean.num_pages - 1; 1394 blob->state = SPDK_BLOB_STATE_CLEAN; 1395 _spdk_blob_persist_zero_pages(seq, ctx, 0); 1396 return; 1397 1398 } 1399 1400 /* Generate the new metadata */ 1401 rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages); 1402 if (rc < 0) { 1403 _spdk_blob_persist_complete(seq, ctx, rc); 1404 return; 1405 } 1406 1407 assert(blob->active.num_pages >= 1); 1408 1409 /* Resize the cache of page indices */ 1410 tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages)); 1411 if (!tmp) { 1412 _spdk_blob_persist_complete(seq, ctx, -ENOMEM); 1413 return; 1414 } 1415 blob->active.pages = tmp; 1416 1417 /* Assign this metadata to pages. This requires two passes - 1418 * one to verify that there are enough pages and a second 1419 * to actually claim them. */ 1420 page_num = 0; 1421 /* Note that this loop starts at one. The first page location is fixed by the blobid. */ 1422 for (i = 1; i < blob->active.num_pages; i++) { 1423 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 1424 if (page_num >= spdk_bit_array_capacity(bs->used_md_pages)) { 1425 _spdk_blob_persist_complete(seq, ctx, -ENOMEM); 1426 return; 1427 } 1428 page_num++; 1429 } 1430 1431 page_num = 0; 1432 blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id); 1433 for (i = 1; i < blob->active.num_pages; i++) { 1434 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 1435 ctx->pages[i - 1].next = page_num; 1436 /* Now that previous metadata page is complete, calculate the crc for it. */ 1437 ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]); 1438 blob->active.pages[i] = page_num; 1439 spdk_bit_array_set(bs->used_md_pages, page_num); 1440 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id); 1441 page_num++; 1442 } 1443 ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]); 1444 /* Start writing the metadata from last page to first */ 1445 ctx->idx = blob->active.num_pages - 1; 1446 blob->state = SPDK_BLOB_STATE_CLEAN; 1447 _spdk_blob_persist_write_page_chain(seq, ctx, 0); 1448 } 1449 1450 static void 1451 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1452 { 1453 struct spdk_blob_persist_ctx *ctx = cb_arg; 1454 1455 ctx->blob->bs->clean = 0; 1456 1457 spdk_dma_free(ctx->super); 1458 1459 _spdk_blob_persist_start(ctx); 1460 } 1461 1462 static void 1463 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 1464 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg); 1465 1466 1467 static void 1468 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1469 { 1470 struct spdk_blob_persist_ctx *ctx = cb_arg; 1471 1472 ctx->super->clean = 0; 1473 if (ctx->super->size == 0) { 1474 ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen; 1475 } 1476 1477 _spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx); 1478 } 1479 1480 1481 /* Write a blob to disk */ 1482 static void 1483 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 1484 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 1485 { 1486 struct spdk_blob_persist_ctx *ctx; 1487 1488 _spdk_blob_verify_md_op(blob); 1489 1490 if (blob->state == SPDK_BLOB_STATE_CLEAN) { 1491 cb_fn(seq, cb_arg, 0); 1492 return; 1493 } 1494 1495 ctx = calloc(1, sizeof(*ctx)); 1496 if (!ctx) { 1497 cb_fn(seq, cb_arg, -ENOMEM); 1498 return; 1499 } 1500 ctx->blob = blob; 1501 ctx->seq = seq; 1502 ctx->cb_fn = cb_fn; 1503 ctx->cb_arg = cb_arg; 1504 1505 if (blob->bs->clean) { 1506 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 1507 if (!ctx->super) { 1508 cb_fn(seq, cb_arg, -ENOMEM); 1509 free(ctx); 1510 return; 1511 } 1512 1513 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0), 1514 _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)), 1515 _spdk_blob_persist_dirty, ctx); 1516 } else { 1517 _spdk_blob_persist_start(ctx); 1518 } 1519 } 1520 1521 struct spdk_blob_copy_cluster_ctx { 1522 struct spdk_blob *blob; 1523 uint8_t *buf; 1524 uint64_t page; 1525 uint64_t new_cluster; 1526 spdk_bs_sequence_t *seq; 1527 }; 1528 1529 static void 1530 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno) 1531 { 1532 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1533 struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq; 1534 TAILQ_HEAD(, spdk_bs_request_set) requests; 1535 spdk_bs_user_op_t *op; 1536 1537 TAILQ_INIT(&requests); 1538 TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link); 1539 1540 while (!TAILQ_EMPTY(&requests)) { 1541 op = TAILQ_FIRST(&requests); 1542 TAILQ_REMOVE(&requests, op, link); 1543 if (bserrno == 0) { 1544 spdk_bs_user_op_execute(op); 1545 } else { 1546 spdk_bs_user_op_abort(op); 1547 } 1548 } 1549 1550 spdk_dma_free(ctx->buf); 1551 free(ctx); 1552 } 1553 1554 static void 1555 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno) 1556 { 1557 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1558 1559 if (bserrno) { 1560 uint32_t cluster_number; 1561 1562 if (bserrno == -EEXIST) { 1563 /* The metadata insert failed because another thread 1564 * allocated the cluster first. Free our cluster 1565 * but continue without error. */ 1566 bserrno = 0; 1567 } 1568 1569 cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page); 1570 _spdk_bs_release_cluster(ctx->blob->bs, cluster_number); 1571 } 1572 1573 spdk_bs_sequence_finish(ctx->seq, bserrno); 1574 } 1575 1576 static void 1577 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1578 { 1579 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1580 uint32_t cluster_number; 1581 1582 if (bserrno) { 1583 /* The write failed, so jump to the final completion handler */ 1584 spdk_bs_sequence_finish(seq, bserrno); 1585 return; 1586 } 1587 1588 cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page); 1589 1590 _spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 1591 _spdk_blob_insert_cluster_cpl, ctx); 1592 } 1593 1594 static void 1595 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1596 { 1597 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1598 1599 if (bserrno != 0) { 1600 /* The read failed, so jump to the final completion handler */ 1601 spdk_bs_sequence_finish(seq, bserrno); 1602 return; 1603 } 1604 1605 /* Write whole cluster */ 1606 spdk_bs_sequence_write_dev(seq, ctx->buf, 1607 _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster), 1608 _spdk_bs_cluster_to_lba(ctx->blob->bs, 1), 1609 _spdk_blob_write_copy_cpl, ctx); 1610 } 1611 1612 static void 1613 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob, 1614 struct spdk_io_channel *_ch, 1615 uint64_t offset, spdk_bs_user_op_t *op) 1616 { 1617 struct spdk_bs_cpl cpl; 1618 struct spdk_bs_channel *ch; 1619 struct spdk_blob_copy_cluster_ctx *ctx; 1620 uint32_t cluster_start_page; 1621 uint32_t cluster_number; 1622 int rc; 1623 1624 ch = spdk_io_channel_get_ctx(_ch); 1625 1626 if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) { 1627 /* There are already operations pending. Queue this user op 1628 * and return because it will be re-executed when the outstanding 1629 * cluster allocation completes. */ 1630 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 1631 return; 1632 } 1633 1634 /* Round the page offset down to the first page in the cluster */ 1635 cluster_start_page = _spdk_bs_page_to_cluster_start(blob, offset); 1636 1637 /* Calculate which index in the metadata cluster array the corresponding 1638 * cluster is supposed to be at. */ 1639 cluster_number = _spdk_bs_page_to_cluster(blob->bs, cluster_start_page); 1640 1641 ctx = calloc(1, sizeof(*ctx)); 1642 if (!ctx) { 1643 spdk_bs_user_op_abort(op); 1644 return; 1645 } 1646 1647 assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0); 1648 1649 ctx->blob = blob; 1650 ctx->page = cluster_start_page; 1651 1652 if (blob->parent_id != SPDK_BLOBID_INVALID) { 1653 ctx->buf = spdk_dma_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, NULL); 1654 if (!ctx->buf) { 1655 SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n", 1656 blob->bs->cluster_sz); 1657 free(ctx); 1658 spdk_bs_user_op_abort(op); 1659 return; 1660 } 1661 } 1662 1663 rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false); 1664 if (rc != 0) { 1665 spdk_dma_free(ctx->buf); 1666 free(ctx); 1667 spdk_bs_user_op_abort(op); 1668 return; 1669 } 1670 1671 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1672 cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl; 1673 cpl.u.blob_basic.cb_arg = ctx; 1674 1675 ctx->seq = spdk_bs_sequence_start(_ch, &cpl); 1676 if (!ctx->seq) { 1677 _spdk_bs_release_cluster(blob->bs, ctx->new_cluster); 1678 spdk_dma_free(ctx->buf); 1679 free(ctx); 1680 spdk_bs_user_op_abort(op); 1681 return; 1682 } 1683 1684 /* Queue the user op to block other incoming operations */ 1685 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 1686 1687 if (blob->parent_id != SPDK_BLOBID_INVALID) { 1688 /* Read cluster from backing device */ 1689 spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf, 1690 _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page), 1691 _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz), 1692 _spdk_blob_write_copy, ctx); 1693 } else { 1694 _spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 1695 _spdk_blob_insert_cluster_cpl, ctx); 1696 } 1697 } 1698 1699 static void 1700 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t page, uint64_t length, 1701 uint64_t *lba, uint32_t *lba_count) 1702 { 1703 *lba_count = _spdk_bs_page_to_lba(blob->bs, length); 1704 1705 if (!_spdk_bs_page_is_allocated(blob, page)) { 1706 assert(blob->back_bs_dev != NULL); 1707 *lba = _spdk_bs_dev_page_to_lba(blob->back_bs_dev, page); 1708 *lba_count = _spdk_bs_blob_lba_to_back_dev_lba(blob, *lba_count); 1709 } else { 1710 *lba = _spdk_bs_blob_page_to_lba(blob, page); 1711 } 1712 } 1713 1714 struct op_split_ctx { 1715 struct spdk_blob *blob; 1716 struct spdk_io_channel *channel; 1717 uint64_t page_offset; 1718 uint64_t pages_remaining; 1719 void *curr_payload; 1720 enum spdk_blob_op_type op_type; 1721 spdk_bs_sequence_t *seq; 1722 }; 1723 1724 static void 1725 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno) 1726 { 1727 struct op_split_ctx *ctx = cb_arg; 1728 struct spdk_blob *blob = ctx->blob; 1729 struct spdk_io_channel *ch = ctx->channel; 1730 enum spdk_blob_op_type op_type = ctx->op_type; 1731 uint8_t *buf = ctx->curr_payload; 1732 uint64_t offset = ctx->page_offset; 1733 uint64_t length = ctx->pages_remaining; 1734 uint64_t op_length; 1735 1736 if (bserrno != 0 || ctx->pages_remaining == 0) { 1737 spdk_bs_sequence_finish(ctx->seq, bserrno); 1738 free(ctx); 1739 return; 1740 } 1741 1742 op_length = spdk_min(length, _spdk_bs_num_pages_to_cluster_boundary(blob, offset)); 1743 1744 /* Update length and payload for next operation */ 1745 ctx->pages_remaining -= op_length; 1746 ctx->page_offset += op_length; 1747 if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) { 1748 ctx->curr_payload += (op_length * SPDK_BS_PAGE_SIZE); 1749 } 1750 1751 switch (op_type) { 1752 case SPDK_BLOB_READ: 1753 spdk_blob_io_read(blob, ch, buf, offset, op_length, 1754 _spdk_blob_request_submit_op_split_next, ctx); 1755 break; 1756 case SPDK_BLOB_WRITE: 1757 spdk_blob_io_write(blob, ch, buf, offset, op_length, 1758 _spdk_blob_request_submit_op_split_next, ctx); 1759 break; 1760 case SPDK_BLOB_UNMAP: 1761 spdk_blob_io_unmap(blob, ch, offset, op_length, 1762 _spdk_blob_request_submit_op_split_next, ctx); 1763 break; 1764 case SPDK_BLOB_WRITE_ZEROES: 1765 spdk_blob_io_write_zeroes(blob, ch, offset, op_length, 1766 _spdk_blob_request_submit_op_split_next, ctx); 1767 break; 1768 case SPDK_BLOB_READV: 1769 case SPDK_BLOB_WRITEV: 1770 SPDK_ERRLOG("readv/write not valid for %s\n", __func__); 1771 spdk_bs_sequence_finish(ctx->seq, -EINVAL); 1772 free(ctx); 1773 break; 1774 } 1775 } 1776 1777 static void 1778 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob, 1779 void *payload, uint64_t offset, uint64_t length, 1780 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1781 { 1782 struct op_split_ctx *ctx; 1783 spdk_bs_sequence_t *seq; 1784 struct spdk_bs_cpl cpl; 1785 1786 assert(blob != NULL); 1787 1788 ctx = calloc(1, sizeof(struct op_split_ctx)); 1789 if (ctx == NULL) { 1790 cb_fn(cb_arg, -ENOMEM); 1791 return; 1792 } 1793 1794 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1795 cpl.u.blob_basic.cb_fn = cb_fn; 1796 cpl.u.blob_basic.cb_arg = cb_arg; 1797 1798 seq = spdk_bs_sequence_start(ch, &cpl); 1799 if (!seq) { 1800 free(ctx); 1801 cb_fn(cb_arg, -ENOMEM); 1802 return; 1803 } 1804 1805 ctx->blob = blob; 1806 ctx->channel = ch; 1807 ctx->curr_payload = payload; 1808 ctx->page_offset = offset; 1809 ctx->pages_remaining = length; 1810 ctx->op_type = op_type; 1811 ctx->seq = seq; 1812 1813 _spdk_blob_request_submit_op_split_next(ctx, 0); 1814 } 1815 1816 static void 1817 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob, 1818 void *payload, uint64_t offset, uint64_t length, 1819 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1820 { 1821 struct spdk_bs_cpl cpl; 1822 uint64_t lba; 1823 uint32_t lba_count; 1824 1825 assert(blob != NULL); 1826 1827 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1828 cpl.u.blob_basic.cb_fn = cb_fn; 1829 cpl.u.blob_basic.cb_arg = cb_arg; 1830 1831 _spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 1832 1833 if (blob->frozen_refcnt) { 1834 /* This blob I/O is frozen */ 1835 spdk_bs_user_op_t *op; 1836 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch); 1837 1838 op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 1839 if (!op) { 1840 cb_fn(cb_arg, -ENOMEM); 1841 return; 1842 } 1843 1844 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 1845 1846 return; 1847 } 1848 1849 switch (op_type) { 1850 case SPDK_BLOB_READ: { 1851 spdk_bs_batch_t *batch; 1852 1853 batch = spdk_bs_batch_open(_ch, &cpl); 1854 if (!batch) { 1855 cb_fn(cb_arg, -ENOMEM); 1856 return; 1857 } 1858 1859 if (_spdk_bs_page_is_allocated(blob, offset)) { 1860 /* Read from the blob */ 1861 spdk_bs_batch_read_dev(batch, payload, lba, lba_count); 1862 } else { 1863 /* Read from the backing block device */ 1864 spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count); 1865 } 1866 1867 spdk_bs_batch_close(batch); 1868 break; 1869 } 1870 case SPDK_BLOB_WRITE: 1871 case SPDK_BLOB_WRITE_ZEROES: { 1872 if (_spdk_bs_page_is_allocated(blob, offset)) { 1873 /* Write to the blob */ 1874 spdk_bs_batch_t *batch; 1875 1876 if (lba_count == 0) { 1877 cb_fn(cb_arg, 0); 1878 return; 1879 } 1880 1881 batch = spdk_bs_batch_open(_ch, &cpl); 1882 if (!batch) { 1883 cb_fn(cb_arg, -ENOMEM); 1884 return; 1885 } 1886 1887 if (op_type == SPDK_BLOB_WRITE) { 1888 spdk_bs_batch_write_dev(batch, payload, lba, lba_count); 1889 } else { 1890 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1891 } 1892 1893 spdk_bs_batch_close(batch); 1894 } else { 1895 /* Queue this operation and allocate the cluster */ 1896 spdk_bs_user_op_t *op; 1897 1898 op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 1899 if (!op) { 1900 cb_fn(cb_arg, -ENOMEM); 1901 return; 1902 } 1903 1904 _spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op); 1905 } 1906 break; 1907 } 1908 case SPDK_BLOB_UNMAP: { 1909 spdk_bs_batch_t *batch; 1910 1911 batch = spdk_bs_batch_open(_ch, &cpl); 1912 if (!batch) { 1913 cb_fn(cb_arg, -ENOMEM); 1914 return; 1915 } 1916 1917 if (_spdk_bs_page_is_allocated(blob, offset)) { 1918 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 1919 } 1920 1921 spdk_bs_batch_close(batch); 1922 break; 1923 } 1924 case SPDK_BLOB_READV: 1925 case SPDK_BLOB_WRITEV: 1926 SPDK_ERRLOG("readv/write not valid\n"); 1927 cb_fn(cb_arg, -EINVAL); 1928 break; 1929 } 1930 } 1931 1932 static void 1933 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel, 1934 void *payload, uint64_t offset, uint64_t length, 1935 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1936 { 1937 assert(blob != NULL); 1938 1939 if (blob->data_ro && op_type != SPDK_BLOB_READ) { 1940 cb_fn(cb_arg, -EPERM); 1941 return; 1942 } 1943 1944 if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) { 1945 cb_fn(cb_arg, -EINVAL); 1946 return; 1947 } 1948 1949 if (length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset)) { 1950 _spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length, 1951 cb_fn, cb_arg, op_type); 1952 } else { 1953 _spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length, 1954 cb_fn, cb_arg, op_type); 1955 } 1956 } 1957 1958 struct rw_iov_ctx { 1959 struct spdk_blob *blob; 1960 struct spdk_io_channel *channel; 1961 spdk_blob_op_complete cb_fn; 1962 void *cb_arg; 1963 bool read; 1964 int iovcnt; 1965 struct iovec *orig_iov; 1966 uint64_t page_offset; 1967 uint64_t pages_remaining; 1968 uint64_t pages_done; 1969 struct iovec iov[0]; 1970 }; 1971 1972 static void 1973 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1974 { 1975 assert(cb_arg == NULL); 1976 spdk_bs_sequence_finish(seq, bserrno); 1977 } 1978 1979 static void 1980 _spdk_rw_iov_split_next(void *cb_arg, int bserrno) 1981 { 1982 struct rw_iov_ctx *ctx = cb_arg; 1983 struct spdk_blob *blob = ctx->blob; 1984 struct iovec *iov, *orig_iov; 1985 int iovcnt; 1986 size_t orig_iovoff; 1987 uint64_t page_count, pages_to_boundary, page_offset; 1988 uint64_t byte_count; 1989 1990 if (bserrno != 0 || ctx->pages_remaining == 0) { 1991 ctx->cb_fn(ctx->cb_arg, bserrno); 1992 free(ctx); 1993 return; 1994 } 1995 1996 page_offset = ctx->page_offset; 1997 pages_to_boundary = _spdk_bs_num_pages_to_cluster_boundary(blob, page_offset); 1998 page_count = spdk_min(ctx->pages_remaining, pages_to_boundary); 1999 2000 /* 2001 * Get index and offset into the original iov array for our current position in the I/O sequence. 2002 * byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will 2003 * point to the current position in the I/O sequence. 2004 */ 2005 byte_count = ctx->pages_done * sizeof(struct spdk_blob_md_page); 2006 orig_iov = &ctx->orig_iov[0]; 2007 orig_iovoff = 0; 2008 while (byte_count > 0) { 2009 if (byte_count >= orig_iov->iov_len) { 2010 byte_count -= orig_iov->iov_len; 2011 orig_iov++; 2012 } else { 2013 orig_iovoff = byte_count; 2014 byte_count = 0; 2015 } 2016 } 2017 2018 /* 2019 * Build an iov array for the next I/O in the sequence. byte_count will keep track of how many 2020 * bytes of this next I/O remain to be accounted for in the new iov array. 2021 */ 2022 byte_count = page_count * sizeof(struct spdk_blob_md_page); 2023 iov = &ctx->iov[0]; 2024 iovcnt = 0; 2025 while (byte_count > 0) { 2026 iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff); 2027 iov->iov_base = orig_iov->iov_base + orig_iovoff; 2028 byte_count -= iov->iov_len; 2029 orig_iovoff = 0; 2030 orig_iov++; 2031 iov++; 2032 iovcnt++; 2033 } 2034 2035 ctx->page_offset += page_count; 2036 ctx->pages_done += page_count; 2037 ctx->pages_remaining -= page_count; 2038 iov = &ctx->iov[0]; 2039 2040 if (ctx->read) { 2041 spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, page_offset, 2042 page_count, _spdk_rw_iov_split_next, ctx); 2043 } else { 2044 spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, page_offset, 2045 page_count, _spdk_rw_iov_split_next, ctx); 2046 } 2047 } 2048 2049 static void 2050 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel, 2051 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 2052 spdk_blob_op_complete cb_fn, void *cb_arg, bool read) 2053 { 2054 struct spdk_bs_cpl cpl; 2055 2056 assert(blob != NULL); 2057 2058 if (!read && blob->data_ro) { 2059 cb_fn(cb_arg, -EPERM); 2060 return; 2061 } 2062 2063 if (length == 0) { 2064 cb_fn(cb_arg, 0); 2065 return; 2066 } 2067 2068 if (offset + length > blob->active.num_clusters * blob->bs->pages_per_cluster) { 2069 cb_fn(cb_arg, -EINVAL); 2070 return; 2071 } 2072 2073 /* 2074 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having 2075 * to split a request that spans a cluster boundary. For I/O that do not span a cluster boundary, 2076 * there will be no noticeable difference compared to using a batch. For I/O that do span a cluster 2077 * boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need 2078 * to allocate a separate iov array and split the I/O such that none of the resulting 2079 * smaller I/O cross a cluster boundary. These smaller I/O will be issued in sequence (not in parallel) 2080 * but since this case happens very infrequently, any performance impact will be negligible. 2081 * 2082 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs 2083 * for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them 2084 * in a batch. That would also require creating an intermediate spdk_bs_cpl that would get called 2085 * when the batch was completed, to allow for freeing the memory for the iov arrays. 2086 */ 2087 if (spdk_likely(length <= _spdk_bs_num_pages_to_cluster_boundary(blob, offset))) { 2088 uint32_t lba_count; 2089 uint64_t lba; 2090 2091 _spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 2092 2093 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2094 cpl.u.blob_basic.cb_fn = cb_fn; 2095 cpl.u.blob_basic.cb_arg = cb_arg; 2096 if (blob->frozen_refcnt) { 2097 /* This blob I/O is frozen */ 2098 spdk_bs_user_op_t *op; 2099 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel); 2100 2101 op = spdk_bs_user_op_alloc(_channel, &cpl, read, blob, iov, iovcnt, offset, length); 2102 if (!op) { 2103 cb_fn(cb_arg, -ENOMEM); 2104 return; 2105 } 2106 2107 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 2108 2109 return; 2110 } 2111 2112 if (read) { 2113 spdk_bs_sequence_t *seq; 2114 2115 seq = spdk_bs_sequence_start(_channel, &cpl); 2116 if (!seq) { 2117 cb_fn(cb_arg, -ENOMEM); 2118 return; 2119 } 2120 2121 if (_spdk_bs_page_is_allocated(blob, offset)) { 2122 spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL); 2123 } else { 2124 spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count, 2125 _spdk_rw_iov_done, NULL); 2126 } 2127 } else { 2128 if (_spdk_bs_page_is_allocated(blob, offset)) { 2129 spdk_bs_sequence_t *seq; 2130 2131 seq = spdk_bs_sequence_start(_channel, &cpl); 2132 if (!seq) { 2133 cb_fn(cb_arg, -ENOMEM); 2134 return; 2135 } 2136 2137 spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL); 2138 } else { 2139 /* Queue this operation and allocate the cluster */ 2140 spdk_bs_user_op_t *op; 2141 2142 op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, length); 2143 if (!op) { 2144 cb_fn(cb_arg, -ENOMEM); 2145 return; 2146 } 2147 2148 _spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op); 2149 } 2150 } 2151 } else { 2152 struct rw_iov_ctx *ctx; 2153 2154 ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec)); 2155 if (ctx == NULL) { 2156 cb_fn(cb_arg, -ENOMEM); 2157 return; 2158 } 2159 2160 ctx->blob = blob; 2161 ctx->channel = _channel; 2162 ctx->cb_fn = cb_fn; 2163 ctx->cb_arg = cb_arg; 2164 ctx->read = read; 2165 ctx->orig_iov = iov; 2166 ctx->iovcnt = iovcnt; 2167 ctx->page_offset = offset; 2168 ctx->pages_remaining = length; 2169 ctx->pages_done = 0; 2170 2171 _spdk_rw_iov_split_next(ctx, 0); 2172 } 2173 } 2174 2175 static struct spdk_blob * 2176 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid) 2177 { 2178 struct spdk_blob *blob; 2179 2180 TAILQ_FOREACH(blob, &bs->blobs, link) { 2181 if (blob->id == blobid) { 2182 return blob; 2183 } 2184 } 2185 2186 return NULL; 2187 } 2188 2189 static int 2190 _spdk_bs_channel_create(void *io_device, void *ctx_buf) 2191 { 2192 struct spdk_blob_store *bs = io_device; 2193 struct spdk_bs_channel *channel = ctx_buf; 2194 struct spdk_bs_dev *dev; 2195 uint32_t max_ops = bs->max_channel_ops; 2196 uint32_t i; 2197 2198 dev = bs->dev; 2199 2200 channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set)); 2201 if (!channel->req_mem) { 2202 return -1; 2203 } 2204 2205 TAILQ_INIT(&channel->reqs); 2206 2207 for (i = 0; i < max_ops; i++) { 2208 TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link); 2209 } 2210 2211 channel->bs = bs; 2212 channel->dev = dev; 2213 channel->dev_channel = dev->create_channel(dev); 2214 2215 if (!channel->dev_channel) { 2216 SPDK_ERRLOG("Failed to create device channel.\n"); 2217 free(channel->req_mem); 2218 return -1; 2219 } 2220 2221 TAILQ_INIT(&channel->need_cluster_alloc); 2222 TAILQ_INIT(&channel->queued_io); 2223 2224 return 0; 2225 } 2226 2227 static void 2228 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf) 2229 { 2230 struct spdk_bs_channel *channel = ctx_buf; 2231 spdk_bs_user_op_t *op; 2232 2233 while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) { 2234 op = TAILQ_FIRST(&channel->need_cluster_alloc); 2235 TAILQ_REMOVE(&channel->need_cluster_alloc, op, link); 2236 spdk_bs_user_op_abort(op); 2237 } 2238 2239 while (!TAILQ_EMPTY(&channel->queued_io)) { 2240 op = TAILQ_FIRST(&channel->queued_io); 2241 TAILQ_REMOVE(&channel->queued_io, op, link); 2242 spdk_bs_user_op_abort(op); 2243 } 2244 2245 free(channel->req_mem); 2246 channel->dev->destroy_channel(channel->dev, channel->dev_channel); 2247 } 2248 2249 static void 2250 _spdk_bs_dev_destroy(void *io_device) 2251 { 2252 struct spdk_blob_store *bs = io_device; 2253 struct spdk_blob *blob, *blob_tmp; 2254 2255 bs->dev->destroy(bs->dev); 2256 2257 TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) { 2258 TAILQ_REMOVE(&bs->blobs, blob, link); 2259 _spdk_blob_free(blob); 2260 } 2261 2262 pthread_mutex_destroy(&bs->used_clusters_mutex); 2263 2264 spdk_bit_array_free(&bs->used_blobids); 2265 spdk_bit_array_free(&bs->used_md_pages); 2266 spdk_bit_array_free(&bs->used_clusters); 2267 /* 2268 * If this function is called for any reason except a successful unload, 2269 * the unload_cpl type will be NONE and this will be a nop. 2270 */ 2271 spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err); 2272 2273 free(bs); 2274 } 2275 2276 static int 2277 _spdk_bs_blob_list_add(struct spdk_blob *blob) 2278 { 2279 spdk_blob_id snapshot_id; 2280 struct spdk_blob_list *snapshot_entry = NULL; 2281 struct spdk_blob_list *clone_entry = NULL; 2282 2283 assert(blob != NULL); 2284 2285 snapshot_id = blob->parent_id; 2286 if (snapshot_id == SPDK_BLOBID_INVALID) { 2287 return 0; 2288 } 2289 2290 TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) { 2291 if (snapshot_entry->id == snapshot_id) { 2292 break; 2293 } 2294 } 2295 2296 if (snapshot_entry == NULL) { 2297 /* Snapshot not found */ 2298 snapshot_entry = calloc(1, sizeof(struct spdk_blob_list)); 2299 if (snapshot_entry == NULL) { 2300 return -ENOMEM; 2301 } 2302 snapshot_entry->id = snapshot_id; 2303 TAILQ_INIT(&snapshot_entry->clones); 2304 TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link); 2305 } else { 2306 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 2307 if (clone_entry->id == blob->id) { 2308 break; 2309 } 2310 } 2311 } 2312 2313 if (clone_entry == NULL) { 2314 /* Clone not found */ 2315 clone_entry = calloc(1, sizeof(struct spdk_blob_list)); 2316 if (clone_entry == NULL) { 2317 return -ENOMEM; 2318 } 2319 clone_entry->id = blob->id; 2320 TAILQ_INIT(&clone_entry->clones); 2321 TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link); 2322 snapshot_entry->clone_count++; 2323 } 2324 2325 return 0; 2326 } 2327 2328 static int 2329 _spdk_bs_blob_list_remove(struct spdk_blob *blob) 2330 { 2331 struct spdk_blob_list *snapshot_entry = NULL; 2332 struct spdk_blob_list *clone_entry = NULL; 2333 spdk_blob_id snapshot_id; 2334 2335 assert(blob != NULL); 2336 2337 snapshot_id = blob->parent_id; 2338 if (snapshot_id == SPDK_BLOBID_INVALID) { 2339 return 0; 2340 } 2341 2342 TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) { 2343 if (snapshot_entry->id == snapshot_id) { 2344 break; 2345 } 2346 } 2347 2348 assert(snapshot_entry != NULL); 2349 2350 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 2351 if (clone_entry->id == blob->id) { 2352 break; 2353 } 2354 } 2355 2356 assert(clone_entry != NULL); 2357 2358 blob->parent_id = SPDK_BLOBID_INVALID; 2359 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 2360 free(clone_entry); 2361 2362 snapshot_entry->clone_count--; 2363 if (snapshot_entry->clone_count == 0) { 2364 /* Snapshot have no more clones */ 2365 TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link); 2366 free(snapshot_entry); 2367 } 2368 2369 return 0; 2370 } 2371 2372 static int 2373 _spdk_bs_blob_list_free(struct spdk_blob_store *bs) 2374 { 2375 struct spdk_blob_list *snapshot_entry; 2376 struct spdk_blob_list *snapshot_entry_tmp; 2377 struct spdk_blob_list *clone_entry; 2378 struct spdk_blob_list *clone_entry_tmp; 2379 2380 TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) { 2381 TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) { 2382 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 2383 free(clone_entry); 2384 } 2385 TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link); 2386 free(snapshot_entry); 2387 } 2388 2389 return 0; 2390 } 2391 2392 static void 2393 _spdk_bs_free(struct spdk_blob_store *bs) 2394 { 2395 _spdk_bs_blob_list_free(bs); 2396 2397 spdk_bs_unregister_md_thread(bs); 2398 spdk_io_device_unregister(bs, _spdk_bs_dev_destroy); 2399 } 2400 2401 void 2402 spdk_bs_opts_init(struct spdk_bs_opts *opts) 2403 { 2404 opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ; 2405 opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES; 2406 opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS; 2407 opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS; 2408 memset(&opts->bstype, 0, sizeof(opts->bstype)); 2409 opts->iter_cb_fn = NULL; 2410 opts->iter_cb_arg = NULL; 2411 } 2412 2413 static int 2414 _spdk_bs_opts_verify(struct spdk_bs_opts *opts) 2415 { 2416 if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 || 2417 opts->max_channel_ops == 0) { 2418 SPDK_ERRLOG("Blobstore options cannot be set to 0\n"); 2419 return -1; 2420 } 2421 2422 return 0; 2423 } 2424 2425 static int 2426 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs) 2427 { 2428 struct spdk_blob_store *bs; 2429 uint64_t dev_size; 2430 int rc; 2431 2432 dev_size = dev->blocklen * dev->blockcnt; 2433 if (dev_size < opts->cluster_sz) { 2434 /* Device size cannot be smaller than cluster size of blobstore */ 2435 SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n", 2436 dev_size, opts->cluster_sz); 2437 return -ENOSPC; 2438 } 2439 if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) { 2440 /* Cluster size cannot be smaller than page size */ 2441 SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n", 2442 opts->cluster_sz, SPDK_BS_PAGE_SIZE); 2443 return -EINVAL; 2444 } 2445 bs = calloc(1, sizeof(struct spdk_blob_store)); 2446 if (!bs) { 2447 return -ENOMEM; 2448 } 2449 2450 TAILQ_INIT(&bs->blobs); 2451 TAILQ_INIT(&bs->snapshots); 2452 bs->dev = dev; 2453 bs->md_thread = spdk_get_thread(); 2454 assert(bs->md_thread != NULL); 2455 2456 /* 2457 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an 2458 * even multiple of the cluster size. 2459 */ 2460 bs->cluster_sz = opts->cluster_sz; 2461 bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen); 2462 bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE; 2463 bs->num_free_clusters = bs->total_clusters; 2464 bs->used_clusters = spdk_bit_array_create(bs->total_clusters); 2465 if (bs->used_clusters == NULL) { 2466 free(bs); 2467 return -ENOMEM; 2468 } 2469 2470 bs->max_channel_ops = opts->max_channel_ops; 2471 bs->super_blob = SPDK_BLOBID_INVALID; 2472 memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype)); 2473 2474 /* The metadata is assumed to be at least 1 page */ 2475 bs->used_md_pages = spdk_bit_array_create(1); 2476 bs->used_blobids = spdk_bit_array_create(0); 2477 2478 pthread_mutex_init(&bs->used_clusters_mutex, NULL); 2479 2480 spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy, 2481 sizeof(struct spdk_bs_channel)); 2482 rc = spdk_bs_register_md_thread(bs); 2483 if (rc == -1) { 2484 spdk_io_device_unregister(bs, NULL); 2485 pthread_mutex_destroy(&bs->used_clusters_mutex); 2486 spdk_bit_array_free(&bs->used_blobids); 2487 spdk_bit_array_free(&bs->used_md_pages); 2488 spdk_bit_array_free(&bs->used_clusters); 2489 free(bs); 2490 /* FIXME: this is a lie but don't know how to get a proper error code here */ 2491 return -ENOMEM; 2492 } 2493 2494 *_bs = bs; 2495 return 0; 2496 } 2497 2498 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */ 2499 2500 struct spdk_bs_load_ctx { 2501 struct spdk_blob_store *bs; 2502 struct spdk_bs_super_block *super; 2503 2504 struct spdk_bs_md_mask *mask; 2505 bool in_page_chain; 2506 uint32_t page_index; 2507 uint32_t cur_page; 2508 struct spdk_blob_md_page *page; 2509 bool is_load; 2510 2511 spdk_bs_sequence_t *seq; 2512 spdk_blob_op_with_handle_complete iter_cb_fn; 2513 void *iter_cb_arg; 2514 }; 2515 2516 static void 2517 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 2518 { 2519 assert(bserrno != 0); 2520 2521 spdk_dma_free(ctx->super); 2522 spdk_bs_sequence_finish(seq, bserrno); 2523 /* 2524 * Only free the blobstore when a load fails. If an unload fails (for some reason) 2525 * we want to keep the blobstore in case the caller wants to try again. 2526 */ 2527 if (ctx->is_load) { 2528 _spdk_bs_free(ctx->bs); 2529 } 2530 free(ctx); 2531 } 2532 2533 static void 2534 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask) 2535 { 2536 uint32_t i = 0; 2537 2538 while (true) { 2539 i = spdk_bit_array_find_first_set(array, i); 2540 if (i >= mask->length) { 2541 break; 2542 } 2543 mask->mask[i / 8] |= 1U << (i % 8); 2544 i++; 2545 } 2546 } 2547 2548 static int 2549 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask) 2550 { 2551 struct spdk_bit_array *array; 2552 uint32_t i; 2553 2554 if (spdk_bit_array_resize(array_ptr, mask->length) < 0) { 2555 return -ENOMEM; 2556 } 2557 2558 array = *array_ptr; 2559 for (i = 0; i < mask->length; i++) { 2560 if (mask->mask[i / 8] & (1U << (i % 8))) { 2561 spdk_bit_array_set(array, i); 2562 } 2563 } 2564 2565 return 0; 2566 } 2567 2568 static void 2569 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 2570 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg) 2571 { 2572 /* Update the values in the super block */ 2573 super->super_blob = bs->super_blob; 2574 memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype)); 2575 super->crc = _spdk_blob_md_page_calc_crc(super); 2576 spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0), 2577 _spdk_bs_byte_to_lba(bs, sizeof(*super)), 2578 cb_fn, cb_arg); 2579 } 2580 2581 static void 2582 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2583 { 2584 struct spdk_bs_load_ctx *ctx = arg; 2585 uint64_t mask_size, lba, lba_count; 2586 2587 /* Write out the used clusters mask */ 2588 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 2589 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2590 if (!ctx->mask) { 2591 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2592 return; 2593 } 2594 2595 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS; 2596 ctx->mask->length = ctx->bs->total_clusters; 2597 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters)); 2598 2599 _spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask); 2600 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 2601 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 2602 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2603 } 2604 2605 static void 2606 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2607 { 2608 struct spdk_bs_load_ctx *ctx = arg; 2609 uint64_t mask_size, lba, lba_count; 2610 2611 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 2612 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2613 if (!ctx->mask) { 2614 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2615 return; 2616 } 2617 2618 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES; 2619 ctx->mask->length = ctx->super->md_len; 2620 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages)); 2621 2622 _spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask); 2623 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 2624 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 2625 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2626 } 2627 2628 static void 2629 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2630 { 2631 struct spdk_bs_load_ctx *ctx = arg; 2632 uint64_t mask_size, lba, lba_count; 2633 2634 if (ctx->super->used_blobid_mask_len == 0) { 2635 /* 2636 * This is a pre-v3 on-disk format where the blobid mask does not get 2637 * written to disk. 2638 */ 2639 cb_fn(seq, arg, 0); 2640 return; 2641 } 2642 2643 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 2644 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2645 if (!ctx->mask) { 2646 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2647 return; 2648 } 2649 2650 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS; 2651 ctx->mask->length = ctx->super->md_len; 2652 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids)); 2653 2654 _spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask); 2655 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 2656 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 2657 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2658 } 2659 2660 static void 2661 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno) 2662 { 2663 struct spdk_bs_load_ctx *ctx = arg; 2664 2665 if (bserrno == 0) { 2666 if (ctx->iter_cb_fn) { 2667 ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0); 2668 } 2669 _spdk_bs_blob_list_add(blob); 2670 spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx); 2671 return; 2672 } 2673 2674 if (bserrno == -ENOENT) { 2675 bserrno = 0; 2676 } else { 2677 /* 2678 * This case needs to be looked at further. Same problem 2679 * exists with applications that rely on explicit blob 2680 * iteration. We should just skip the blob that failed 2681 * to load and continue on to the next one. 2682 */ 2683 SPDK_ERRLOG("Error in iterating blobs\n"); 2684 } 2685 2686 ctx->iter_cb_fn = NULL; 2687 2688 spdk_dma_free(ctx->super); 2689 spdk_dma_free(ctx->mask); 2690 spdk_bs_sequence_finish(ctx->seq, bserrno); 2691 free(ctx); 2692 } 2693 2694 static void 2695 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 2696 { 2697 ctx->seq = seq; 2698 spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx); 2699 } 2700 2701 static void 2702 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2703 { 2704 struct spdk_bs_load_ctx *ctx = cb_arg; 2705 int rc; 2706 2707 /* The type must be correct */ 2708 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS); 2709 2710 /* The length of the mask (in bits) must not be greater than 2711 * the length of the buffer (converted to bits) */ 2712 assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8)); 2713 2714 /* The length of the mask must be exactly equal to the size 2715 * (in pages) of the metadata region */ 2716 assert(ctx->mask->length == ctx->super->md_len); 2717 2718 rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask); 2719 if (rc < 0) { 2720 spdk_dma_free(ctx->mask); 2721 _spdk_bs_load_ctx_fail(seq, ctx, rc); 2722 return; 2723 } 2724 2725 _spdk_bs_load_complete(seq, ctx, bserrno); 2726 } 2727 2728 static void 2729 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2730 { 2731 struct spdk_bs_load_ctx *ctx = cb_arg; 2732 uint64_t lba, lba_count, mask_size; 2733 int rc; 2734 2735 /* The type must be correct */ 2736 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS); 2737 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 2738 assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof( 2739 struct spdk_blob_md_page) * 8)); 2740 /* The length of the mask must be exactly equal to the total number of clusters */ 2741 assert(ctx->mask->length == ctx->bs->total_clusters); 2742 2743 rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask); 2744 if (rc < 0) { 2745 spdk_dma_free(ctx->mask); 2746 _spdk_bs_load_ctx_fail(seq, ctx, rc); 2747 return; 2748 } 2749 2750 ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters); 2751 assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters); 2752 2753 spdk_dma_free(ctx->mask); 2754 2755 /* Read the used blobids mask */ 2756 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 2757 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2758 if (!ctx->mask) { 2759 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2760 return; 2761 } 2762 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 2763 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 2764 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2765 _spdk_bs_load_used_blobids_cpl, ctx); 2766 } 2767 2768 static void 2769 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2770 { 2771 struct spdk_bs_load_ctx *ctx = cb_arg; 2772 uint64_t lba, lba_count, mask_size; 2773 int rc; 2774 2775 /* The type must be correct */ 2776 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES); 2777 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 2778 assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE * 2779 8)); 2780 /* The length of the mask must be exactly equal to the size (in pages) of the metadata region */ 2781 assert(ctx->mask->length == ctx->super->md_len); 2782 2783 rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask); 2784 if (rc < 0) { 2785 spdk_dma_free(ctx->mask); 2786 _spdk_bs_load_ctx_fail(seq, ctx, rc); 2787 return; 2788 } 2789 2790 spdk_dma_free(ctx->mask); 2791 2792 /* Read the used clusters mask */ 2793 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 2794 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2795 if (!ctx->mask) { 2796 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2797 return; 2798 } 2799 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 2800 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 2801 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2802 _spdk_bs_load_used_clusters_cpl, ctx); 2803 } 2804 2805 static void 2806 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg) 2807 { 2808 struct spdk_bs_load_ctx *ctx = cb_arg; 2809 uint64_t lba, lba_count, mask_size; 2810 2811 /* Read the used pages mask */ 2812 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 2813 ctx->mask = spdk_dma_zmalloc(mask_size, 0x1000, NULL); 2814 if (!ctx->mask) { 2815 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2816 return; 2817 } 2818 2819 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 2820 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 2821 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2822 _spdk_bs_load_used_pages_cpl, ctx); 2823 } 2824 2825 static int 2826 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs) 2827 { 2828 struct spdk_blob_md_descriptor *desc; 2829 size_t cur_desc = 0; 2830 2831 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 2832 while (cur_desc < sizeof(page->descriptors)) { 2833 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 2834 if (desc->length == 0) { 2835 /* If padding and length are 0, this terminates the page */ 2836 break; 2837 } 2838 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 2839 struct spdk_blob_md_descriptor_extent *desc_extent; 2840 unsigned int i, j; 2841 unsigned int cluster_count = 0; 2842 uint32_t cluster_idx; 2843 2844 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 2845 2846 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 2847 for (j = 0; j < desc_extent->extents[i].length; j++) { 2848 cluster_idx = desc_extent->extents[i].cluster_idx; 2849 /* 2850 * cluster_idx = 0 means an unallocated cluster - don't mark that 2851 * in the used cluster map. 2852 */ 2853 if (cluster_idx != 0) { 2854 spdk_bit_array_set(bs->used_clusters, cluster_idx + j); 2855 if (bs->num_free_clusters == 0) { 2856 return -ENOSPC; 2857 } 2858 bs->num_free_clusters--; 2859 } 2860 cluster_count++; 2861 } 2862 } 2863 if (cluster_count == 0) { 2864 return -EINVAL; 2865 } 2866 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 2867 /* Skip this item */ 2868 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 2869 /* Skip this item */ 2870 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 2871 /* Skip this item */ 2872 } else { 2873 /* Error */ 2874 return -EINVAL; 2875 } 2876 /* Advance to the next descriptor */ 2877 cur_desc += sizeof(*desc) + desc->length; 2878 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 2879 break; 2880 } 2881 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 2882 } 2883 return 0; 2884 } 2885 2886 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx) 2887 { 2888 uint32_t crc; 2889 2890 crc = _spdk_blob_md_page_calc_crc(ctx->page); 2891 if (crc != ctx->page->crc) { 2892 return false; 2893 } 2894 2895 if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) { 2896 return false; 2897 } 2898 return true; 2899 } 2900 2901 static void 2902 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg); 2903 2904 static void 2905 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2906 { 2907 struct spdk_bs_load_ctx *ctx = cb_arg; 2908 2909 _spdk_bs_load_complete(seq, ctx, bserrno); 2910 } 2911 2912 static void 2913 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2914 { 2915 struct spdk_bs_load_ctx *ctx = cb_arg; 2916 2917 spdk_dma_free(ctx->mask); 2918 ctx->mask = NULL; 2919 2920 _spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl); 2921 } 2922 2923 static void 2924 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2925 { 2926 struct spdk_bs_load_ctx *ctx = cb_arg; 2927 2928 spdk_dma_free(ctx->mask); 2929 ctx->mask = NULL; 2930 2931 _spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl); 2932 } 2933 2934 static void 2935 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2936 { 2937 _spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl); 2938 } 2939 2940 static void 2941 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2942 { 2943 struct spdk_bs_load_ctx *ctx = cb_arg; 2944 uint64_t num_md_clusters; 2945 uint64_t i; 2946 uint32_t page_num; 2947 2948 if (bserrno != 0) { 2949 _spdk_bs_load_ctx_fail(seq, ctx, bserrno); 2950 return; 2951 } 2952 2953 page_num = ctx->cur_page; 2954 if (_spdk_bs_load_cur_md_page_valid(ctx) == true) { 2955 if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) { 2956 spdk_bit_array_set(ctx->bs->used_md_pages, page_num); 2957 if (ctx->page->sequence_num == 0) { 2958 spdk_bit_array_set(ctx->bs->used_blobids, page_num); 2959 } 2960 if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) { 2961 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 2962 return; 2963 } 2964 if (ctx->page->next != SPDK_INVALID_MD_PAGE) { 2965 ctx->in_page_chain = true; 2966 ctx->cur_page = ctx->page->next; 2967 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 2968 return; 2969 } 2970 } 2971 } 2972 2973 ctx->in_page_chain = false; 2974 2975 do { 2976 ctx->page_index++; 2977 } while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true); 2978 2979 if (ctx->page_index < ctx->super->md_len) { 2980 ctx->cur_page = ctx->page_index; 2981 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 2982 } else { 2983 /* Claim all of the clusters used by the metadata */ 2984 num_md_clusters = divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster); 2985 for (i = 0; i < num_md_clusters; i++) { 2986 _spdk_bs_claim_cluster(ctx->bs, i); 2987 } 2988 spdk_dma_free(ctx->page); 2989 _spdk_bs_load_write_used_md(seq, ctx, bserrno); 2990 } 2991 } 2992 2993 static void 2994 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg) 2995 { 2996 struct spdk_bs_load_ctx *ctx = cb_arg; 2997 uint64_t lba; 2998 2999 assert(ctx->cur_page < ctx->super->md_len); 3000 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page); 3001 spdk_bs_sequence_read_dev(seq, ctx->page, lba, 3002 _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 3003 _spdk_bs_load_replay_md_cpl, ctx); 3004 } 3005 3006 static void 3007 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg) 3008 { 3009 struct spdk_bs_load_ctx *ctx = cb_arg; 3010 3011 ctx->page_index = 0; 3012 ctx->cur_page = 0; 3013 ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE, 3014 SPDK_BS_PAGE_SIZE, 3015 NULL); 3016 if (!ctx->page) { 3017 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3018 return; 3019 } 3020 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 3021 } 3022 3023 static void 3024 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg) 3025 { 3026 struct spdk_bs_load_ctx *ctx = cb_arg; 3027 int rc; 3028 3029 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len); 3030 if (rc < 0) { 3031 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3032 return; 3033 } 3034 3035 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len); 3036 if (rc < 0) { 3037 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3038 return; 3039 } 3040 3041 rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters); 3042 if (rc < 0) { 3043 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3044 return; 3045 } 3046 3047 ctx->bs->num_free_clusters = ctx->bs->total_clusters; 3048 _spdk_bs_load_replay_md(seq, cb_arg); 3049 } 3050 3051 static void 3052 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3053 { 3054 struct spdk_bs_load_ctx *ctx = cb_arg; 3055 uint32_t crc; 3056 int rc; 3057 static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH]; 3058 3059 if (ctx->super->version > SPDK_BS_VERSION || 3060 ctx->super->version < SPDK_BS_INITIAL_VERSION) { 3061 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3062 return; 3063 } 3064 3065 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 3066 sizeof(ctx->super->signature)) != 0) { 3067 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3068 return; 3069 } 3070 3071 crc = _spdk_blob_md_page_calc_crc(ctx->super); 3072 if (crc != ctx->super->crc) { 3073 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3074 return; 3075 } 3076 3077 if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 3078 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n"); 3079 } else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 3080 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n"); 3081 } else { 3082 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n"); 3083 SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 3084 SPDK_TRACEDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 3085 _spdk_bs_load_ctx_fail(seq, ctx, -ENXIO); 3086 return; 3087 } 3088 3089 if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) { 3090 SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n", 3091 ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size); 3092 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3093 return; 3094 } 3095 3096 if (ctx->super->size == 0) { 3097 ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen; 3098 } 3099 3100 /* Parse the super block */ 3101 ctx->bs->clean = 1; 3102 ctx->bs->cluster_sz = ctx->super->cluster_size; 3103 ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size; 3104 ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE; 3105 rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters); 3106 if (rc < 0) { 3107 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3108 return; 3109 } 3110 ctx->bs->md_start = ctx->super->md_start; 3111 ctx->bs->md_len = ctx->super->md_len; 3112 ctx->bs->total_data_clusters = ctx->bs->total_clusters - divide_round_up( 3113 ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster); 3114 ctx->bs->super_blob = ctx->super->super_blob; 3115 memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype)); 3116 3117 if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) { 3118 _spdk_bs_recover(seq, ctx); 3119 } else { 3120 _spdk_bs_load_read_used_pages(seq, ctx); 3121 } 3122 } 3123 3124 void 3125 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 3126 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 3127 { 3128 struct spdk_blob_store *bs; 3129 struct spdk_bs_cpl cpl; 3130 spdk_bs_sequence_t *seq; 3131 struct spdk_bs_load_ctx *ctx; 3132 struct spdk_bs_opts opts = {}; 3133 int err; 3134 3135 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev); 3136 3137 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 3138 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen); 3139 dev->destroy(dev); 3140 cb_fn(cb_arg, NULL, -EINVAL); 3141 return; 3142 } 3143 3144 if (o) { 3145 opts = *o; 3146 } else { 3147 spdk_bs_opts_init(&opts); 3148 } 3149 3150 if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) { 3151 dev->destroy(dev); 3152 cb_fn(cb_arg, NULL, -EINVAL); 3153 return; 3154 } 3155 3156 err = _spdk_bs_alloc(dev, &opts, &bs); 3157 if (err) { 3158 dev->destroy(dev); 3159 cb_fn(cb_arg, NULL, err); 3160 return; 3161 } 3162 3163 ctx = calloc(1, sizeof(*ctx)); 3164 if (!ctx) { 3165 _spdk_bs_free(bs); 3166 cb_fn(cb_arg, NULL, -ENOMEM); 3167 return; 3168 } 3169 3170 ctx->bs = bs; 3171 ctx->is_load = true; 3172 ctx->iter_cb_fn = opts.iter_cb_fn; 3173 ctx->iter_cb_arg = opts.iter_cb_arg; 3174 3175 /* Allocate memory for the super block */ 3176 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 3177 if (!ctx->super) { 3178 free(ctx); 3179 _spdk_bs_free(bs); 3180 cb_fn(cb_arg, NULL, -ENOMEM); 3181 return; 3182 } 3183 3184 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 3185 cpl.u.bs_handle.cb_fn = cb_fn; 3186 cpl.u.bs_handle.cb_arg = cb_arg; 3187 cpl.u.bs_handle.bs = bs; 3188 3189 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3190 if (!seq) { 3191 spdk_dma_free(ctx->super); 3192 free(ctx); 3193 _spdk_bs_free(bs); 3194 cb_fn(cb_arg, NULL, -ENOMEM); 3195 return; 3196 } 3197 3198 /* Read the super block */ 3199 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3200 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3201 _spdk_bs_load_super_cpl, ctx); 3202 } 3203 3204 /* END spdk_bs_load */ 3205 3206 /* START spdk_bs_dump */ 3207 3208 struct spdk_bs_dump_ctx { 3209 struct spdk_blob_store *bs; 3210 struct spdk_bs_super_block *super; 3211 uint32_t cur_page; 3212 struct spdk_blob_md_page *page; 3213 spdk_bs_sequence_t *seq; 3214 FILE *fp; 3215 spdk_bs_dump_print_xattr print_xattr_fn; 3216 char xattr_name[4096]; 3217 }; 3218 3219 static void 3220 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno) 3221 { 3222 spdk_dma_free(ctx->super); 3223 3224 /* 3225 * We need to defer calling spdk_bs_call_cpl() until after 3226 * dev destruction, so tuck these away for later use. 3227 */ 3228 ctx->bs->unload_err = bserrno; 3229 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 3230 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 3231 3232 spdk_bs_sequence_finish(seq, 0); 3233 _spdk_bs_free(ctx->bs); 3234 free(ctx); 3235 } 3236 3237 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg); 3238 3239 static void 3240 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx) 3241 { 3242 uint32_t page_idx = ctx->cur_page; 3243 struct spdk_blob_md_page *page = ctx->page; 3244 struct spdk_blob_md_descriptor *desc; 3245 size_t cur_desc = 0; 3246 uint32_t crc; 3247 3248 fprintf(ctx->fp, "=========\n"); 3249 fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx); 3250 fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id); 3251 3252 crc = _spdk_blob_md_page_calc_crc(page); 3253 fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch"); 3254 3255 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 3256 while (cur_desc < sizeof(page->descriptors)) { 3257 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 3258 if (desc->length == 0) { 3259 /* If padding and length are 0, this terminates the page */ 3260 break; 3261 } 3262 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 3263 struct spdk_blob_md_descriptor_extent *desc_extent; 3264 unsigned int i; 3265 3266 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 3267 3268 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 3269 if (desc_extent->extents[i].cluster_idx != 0) { 3270 fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32, 3271 desc_extent->extents[i].cluster_idx); 3272 } else { 3273 fprintf(ctx->fp, "Unallocated Extent - "); 3274 } 3275 fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length); 3276 fprintf(ctx->fp, "\n"); 3277 } 3278 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 3279 struct spdk_blob_md_descriptor_xattr *desc_xattr; 3280 uint32_t i; 3281 3282 desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc; 3283 3284 if (desc_xattr->length != 3285 sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) + 3286 desc_xattr->name_length + desc_xattr->value_length) { 3287 } 3288 3289 memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length); 3290 ctx->xattr_name[desc_xattr->name_length] = '\0'; 3291 fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name); 3292 fprintf(ctx->fp, " value = \""); 3293 ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name, 3294 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 3295 desc_xattr->value_length); 3296 fprintf(ctx->fp, "\"\n"); 3297 for (i = 0; i < desc_xattr->value_length; i++) { 3298 if (i % 16 == 0) { 3299 fprintf(ctx->fp, " "); 3300 } 3301 fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i)); 3302 if ((i + 1) % 16 == 0) { 3303 fprintf(ctx->fp, "\n"); 3304 } 3305 } 3306 if (i % 16 != 0) { 3307 fprintf(ctx->fp, "\n"); 3308 } 3309 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 3310 /* TODO */ 3311 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 3312 /* TODO */ 3313 } else { 3314 /* Error */ 3315 } 3316 /* Advance to the next descriptor */ 3317 cur_desc += sizeof(*desc) + desc->length; 3318 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 3319 break; 3320 } 3321 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 3322 } 3323 } 3324 3325 static void 3326 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3327 { 3328 struct spdk_bs_dump_ctx *ctx = cb_arg; 3329 3330 if (bserrno != 0) { 3331 _spdk_bs_dump_finish(seq, ctx, bserrno); 3332 return; 3333 } 3334 3335 if (ctx->page->id != 0) { 3336 _spdk_bs_dump_print_md_page(ctx); 3337 } 3338 3339 ctx->cur_page++; 3340 3341 if (ctx->cur_page < ctx->super->md_len) { 3342 _spdk_bs_dump_read_md_page(seq, cb_arg); 3343 } else { 3344 spdk_dma_free(ctx->page); 3345 _spdk_bs_dump_finish(seq, ctx, 0); 3346 } 3347 } 3348 3349 static void 3350 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg) 3351 { 3352 struct spdk_bs_dump_ctx *ctx = cb_arg; 3353 uint64_t lba; 3354 3355 assert(ctx->cur_page < ctx->super->md_len); 3356 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page); 3357 spdk_bs_sequence_read_dev(seq, ctx->page, lba, 3358 _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 3359 _spdk_bs_dump_read_md_page_cpl, ctx); 3360 } 3361 3362 static void 3363 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3364 { 3365 struct spdk_bs_dump_ctx *ctx = cb_arg; 3366 3367 fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature); 3368 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 3369 sizeof(ctx->super->signature)) != 0) { 3370 fprintf(ctx->fp, "(Mismatch)\n"); 3371 _spdk_bs_dump_finish(seq, ctx, bserrno); 3372 return; 3373 } else { 3374 fprintf(ctx->fp, "(OK)\n"); 3375 } 3376 fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version); 3377 fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc, 3378 (ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch"); 3379 fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype); 3380 fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size); 3381 fprintf(ctx->fp, "Super Blob ID: "); 3382 if (ctx->super->super_blob == SPDK_BLOBID_INVALID) { 3383 fprintf(ctx->fp, "(None)\n"); 3384 } else { 3385 fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob); 3386 } 3387 fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean); 3388 fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start); 3389 fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len); 3390 fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start); 3391 fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len); 3392 fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start); 3393 fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len); 3394 fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start); 3395 fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len); 3396 3397 ctx->cur_page = 0; 3398 ctx->page = spdk_dma_zmalloc(SPDK_BS_PAGE_SIZE, 3399 SPDK_BS_PAGE_SIZE, 3400 NULL); 3401 if (!ctx->page) { 3402 _spdk_bs_dump_finish(seq, ctx, -ENOMEM); 3403 return; 3404 } 3405 _spdk_bs_dump_read_md_page(seq, cb_arg); 3406 } 3407 3408 void 3409 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn, 3410 spdk_bs_op_complete cb_fn, void *cb_arg) 3411 { 3412 struct spdk_blob_store *bs; 3413 struct spdk_bs_cpl cpl; 3414 spdk_bs_sequence_t *seq; 3415 struct spdk_bs_dump_ctx *ctx; 3416 struct spdk_bs_opts opts = {}; 3417 int err; 3418 3419 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev); 3420 3421 spdk_bs_opts_init(&opts); 3422 3423 err = _spdk_bs_alloc(dev, &opts, &bs); 3424 if (err) { 3425 dev->destroy(dev); 3426 cb_fn(cb_arg, err); 3427 return; 3428 } 3429 3430 ctx = calloc(1, sizeof(*ctx)); 3431 if (!ctx) { 3432 _spdk_bs_free(bs); 3433 cb_fn(cb_arg, -ENOMEM); 3434 return; 3435 } 3436 3437 ctx->bs = bs; 3438 ctx->fp = fp; 3439 ctx->print_xattr_fn = print_xattr_fn; 3440 3441 /* Allocate memory for the super block */ 3442 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 3443 if (!ctx->super) { 3444 free(ctx); 3445 _spdk_bs_free(bs); 3446 cb_fn(cb_arg, -ENOMEM); 3447 return; 3448 } 3449 3450 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3451 cpl.u.bs_basic.cb_fn = cb_fn; 3452 cpl.u.bs_basic.cb_arg = cb_arg; 3453 3454 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3455 if (!seq) { 3456 spdk_dma_free(ctx->super); 3457 free(ctx); 3458 _spdk_bs_free(bs); 3459 cb_fn(cb_arg, -ENOMEM); 3460 return; 3461 } 3462 3463 /* Read the super block */ 3464 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3465 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3466 _spdk_bs_dump_super_cpl, ctx); 3467 } 3468 3469 /* END spdk_bs_dump */ 3470 3471 /* START spdk_bs_init */ 3472 3473 struct spdk_bs_init_ctx { 3474 struct spdk_blob_store *bs; 3475 struct spdk_bs_super_block *super; 3476 }; 3477 3478 static void 3479 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3480 { 3481 struct spdk_bs_init_ctx *ctx = cb_arg; 3482 3483 spdk_dma_free(ctx->super); 3484 free(ctx); 3485 3486 spdk_bs_sequence_finish(seq, bserrno); 3487 } 3488 3489 static void 3490 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3491 { 3492 struct spdk_bs_init_ctx *ctx = cb_arg; 3493 3494 /* Write super block */ 3495 spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0), 3496 _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)), 3497 _spdk_bs_init_persist_super_cpl, ctx); 3498 } 3499 3500 void 3501 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 3502 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 3503 { 3504 struct spdk_bs_init_ctx *ctx; 3505 struct spdk_blob_store *bs; 3506 struct spdk_bs_cpl cpl; 3507 spdk_bs_sequence_t *seq; 3508 spdk_bs_batch_t *batch; 3509 uint64_t num_md_lba; 3510 uint64_t num_md_pages; 3511 uint64_t num_md_clusters; 3512 uint32_t i; 3513 struct spdk_bs_opts opts = {}; 3514 int rc; 3515 3516 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev); 3517 3518 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 3519 SPDK_ERRLOG("unsupported dev block length of %d\n", 3520 dev->blocklen); 3521 dev->destroy(dev); 3522 cb_fn(cb_arg, NULL, -EINVAL); 3523 return; 3524 } 3525 3526 if (o) { 3527 opts = *o; 3528 } else { 3529 spdk_bs_opts_init(&opts); 3530 } 3531 3532 if (_spdk_bs_opts_verify(&opts) != 0) { 3533 dev->destroy(dev); 3534 cb_fn(cb_arg, NULL, -EINVAL); 3535 return; 3536 } 3537 3538 rc = _spdk_bs_alloc(dev, &opts, &bs); 3539 if (rc) { 3540 dev->destroy(dev); 3541 cb_fn(cb_arg, NULL, rc); 3542 return; 3543 } 3544 3545 if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) { 3546 /* By default, allocate 1 page per cluster. 3547 * Technically, this over-allocates metadata 3548 * because more metadata will reduce the number 3549 * of usable clusters. This can be addressed with 3550 * more complex math in the future. 3551 */ 3552 bs->md_len = bs->total_clusters; 3553 } else { 3554 bs->md_len = opts.num_md_pages; 3555 } 3556 3557 rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len); 3558 if (rc < 0) { 3559 _spdk_bs_free(bs); 3560 cb_fn(cb_arg, NULL, -ENOMEM); 3561 return; 3562 } 3563 3564 rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len); 3565 if (rc < 0) { 3566 _spdk_bs_free(bs); 3567 cb_fn(cb_arg, NULL, -ENOMEM); 3568 return; 3569 } 3570 3571 ctx = calloc(1, sizeof(*ctx)); 3572 if (!ctx) { 3573 _spdk_bs_free(bs); 3574 cb_fn(cb_arg, NULL, -ENOMEM); 3575 return; 3576 } 3577 3578 ctx->bs = bs; 3579 3580 /* Allocate memory for the super block */ 3581 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 3582 if (!ctx->super) { 3583 free(ctx); 3584 _spdk_bs_free(bs); 3585 cb_fn(cb_arg, NULL, -ENOMEM); 3586 return; 3587 } 3588 memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 3589 sizeof(ctx->super->signature)); 3590 ctx->super->version = SPDK_BS_VERSION; 3591 ctx->super->length = sizeof(*ctx->super); 3592 ctx->super->super_blob = bs->super_blob; 3593 ctx->super->clean = 0; 3594 ctx->super->cluster_size = bs->cluster_sz; 3595 memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype)); 3596 3597 /* Calculate how many pages the metadata consumes at the front 3598 * of the disk. 3599 */ 3600 3601 /* The super block uses 1 page */ 3602 num_md_pages = 1; 3603 3604 /* The used_md_pages mask requires 1 bit per metadata page, rounded 3605 * up to the nearest page, plus a header. 3606 */ 3607 ctx->super->used_page_mask_start = num_md_pages; 3608 ctx->super->used_page_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) + 3609 divide_round_up(bs->md_len, 8), 3610 SPDK_BS_PAGE_SIZE); 3611 num_md_pages += ctx->super->used_page_mask_len; 3612 3613 /* The used_clusters mask requires 1 bit per cluster, rounded 3614 * up to the nearest page, plus a header. 3615 */ 3616 ctx->super->used_cluster_mask_start = num_md_pages; 3617 ctx->super->used_cluster_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) + 3618 divide_round_up(bs->total_clusters, 8), 3619 SPDK_BS_PAGE_SIZE); 3620 num_md_pages += ctx->super->used_cluster_mask_len; 3621 3622 /* The used_blobids mask requires 1 bit per metadata page, rounded 3623 * up to the nearest page, plus a header. 3624 */ 3625 ctx->super->used_blobid_mask_start = num_md_pages; 3626 ctx->super->used_blobid_mask_len = divide_round_up(sizeof(struct spdk_bs_md_mask) + 3627 divide_round_up(bs->md_len, 8), 3628 SPDK_BS_PAGE_SIZE); 3629 num_md_pages += ctx->super->used_blobid_mask_len; 3630 3631 /* The metadata region size was chosen above */ 3632 ctx->super->md_start = bs->md_start = num_md_pages; 3633 ctx->super->md_len = bs->md_len; 3634 num_md_pages += bs->md_len; 3635 3636 num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages); 3637 3638 ctx->super->size = dev->blockcnt * dev->blocklen; 3639 3640 ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super); 3641 3642 num_md_clusters = divide_round_up(num_md_pages, bs->pages_per_cluster); 3643 if (num_md_clusters > bs->total_clusters) { 3644 SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, " 3645 "please decrease number of pages reserved for metadata " 3646 "or increase cluster size.\n"); 3647 spdk_dma_free(ctx->super); 3648 free(ctx); 3649 _spdk_bs_free(bs); 3650 cb_fn(cb_arg, NULL, -ENOMEM); 3651 return; 3652 } 3653 /* Claim all of the clusters used by the metadata */ 3654 for (i = 0; i < num_md_clusters; i++) { 3655 _spdk_bs_claim_cluster(bs, i); 3656 } 3657 3658 bs->total_data_clusters = bs->num_free_clusters; 3659 3660 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 3661 cpl.u.bs_handle.cb_fn = cb_fn; 3662 cpl.u.bs_handle.cb_arg = cb_arg; 3663 cpl.u.bs_handle.bs = bs; 3664 3665 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3666 if (!seq) { 3667 spdk_dma_free(ctx->super); 3668 free(ctx); 3669 _spdk_bs_free(bs); 3670 cb_fn(cb_arg, NULL, -ENOMEM); 3671 return; 3672 } 3673 3674 batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx); 3675 3676 /* Clear metadata space */ 3677 spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba); 3678 /* Trim data clusters */ 3679 spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba); 3680 3681 spdk_bs_batch_close(batch); 3682 } 3683 3684 /* END spdk_bs_init */ 3685 3686 /* START spdk_bs_destroy */ 3687 3688 static void 3689 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3690 { 3691 struct spdk_bs_init_ctx *ctx = cb_arg; 3692 struct spdk_blob_store *bs = ctx->bs; 3693 3694 /* 3695 * We need to defer calling spdk_bs_call_cpl() until after 3696 * dev destruction, so tuck these away for later use. 3697 */ 3698 bs->unload_err = bserrno; 3699 memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 3700 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 3701 3702 spdk_bs_sequence_finish(seq, bserrno); 3703 3704 _spdk_bs_free(bs); 3705 free(ctx); 3706 } 3707 3708 void 3709 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, 3710 void *cb_arg) 3711 { 3712 struct spdk_bs_cpl cpl; 3713 spdk_bs_sequence_t *seq; 3714 struct spdk_bs_init_ctx *ctx; 3715 3716 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n"); 3717 3718 if (!TAILQ_EMPTY(&bs->blobs)) { 3719 SPDK_ERRLOG("Blobstore still has open blobs\n"); 3720 cb_fn(cb_arg, -EBUSY); 3721 return; 3722 } 3723 3724 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3725 cpl.u.bs_basic.cb_fn = cb_fn; 3726 cpl.u.bs_basic.cb_arg = cb_arg; 3727 3728 ctx = calloc(1, sizeof(*ctx)); 3729 if (!ctx) { 3730 cb_fn(cb_arg, -ENOMEM); 3731 return; 3732 } 3733 3734 ctx->bs = bs; 3735 3736 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3737 if (!seq) { 3738 free(ctx); 3739 cb_fn(cb_arg, -ENOMEM); 3740 return; 3741 } 3742 3743 /* Write zeroes to the super block */ 3744 spdk_bs_sequence_write_zeroes_dev(seq, 3745 _spdk_bs_page_to_lba(bs, 0), 3746 _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)), 3747 _spdk_bs_destroy_trim_cpl, ctx); 3748 } 3749 3750 /* END spdk_bs_destroy */ 3751 3752 /* START spdk_bs_unload */ 3753 3754 static void 3755 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3756 { 3757 struct spdk_bs_load_ctx *ctx = cb_arg; 3758 3759 spdk_dma_free(ctx->super); 3760 3761 /* 3762 * We need to defer calling spdk_bs_call_cpl() until after 3763 * dev destruction, so tuck these away for later use. 3764 */ 3765 ctx->bs->unload_err = bserrno; 3766 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 3767 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 3768 3769 spdk_bs_sequence_finish(seq, bserrno); 3770 3771 _spdk_bs_free(ctx->bs); 3772 free(ctx); 3773 } 3774 3775 static void 3776 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3777 { 3778 struct spdk_bs_load_ctx *ctx = cb_arg; 3779 3780 spdk_dma_free(ctx->mask); 3781 ctx->super->clean = 1; 3782 3783 _spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx); 3784 } 3785 3786 static void 3787 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3788 { 3789 struct spdk_bs_load_ctx *ctx = cb_arg; 3790 3791 spdk_dma_free(ctx->mask); 3792 ctx->mask = NULL; 3793 3794 _spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl); 3795 } 3796 3797 static void 3798 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3799 { 3800 struct spdk_bs_load_ctx *ctx = cb_arg; 3801 3802 spdk_dma_free(ctx->mask); 3803 ctx->mask = NULL; 3804 3805 _spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl); 3806 } 3807 3808 static void 3809 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3810 { 3811 _spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl); 3812 } 3813 3814 void 3815 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg) 3816 { 3817 struct spdk_bs_cpl cpl; 3818 spdk_bs_sequence_t *seq; 3819 struct spdk_bs_load_ctx *ctx; 3820 3821 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n"); 3822 3823 if (!TAILQ_EMPTY(&bs->blobs)) { 3824 SPDK_ERRLOG("Blobstore still has open blobs\n"); 3825 cb_fn(cb_arg, -EBUSY); 3826 return; 3827 } 3828 3829 ctx = calloc(1, sizeof(*ctx)); 3830 if (!ctx) { 3831 cb_fn(cb_arg, -ENOMEM); 3832 return; 3833 } 3834 3835 ctx->bs = bs; 3836 ctx->is_load = false; 3837 3838 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 3839 if (!ctx->super) { 3840 free(ctx); 3841 cb_fn(cb_arg, -ENOMEM); 3842 return; 3843 } 3844 3845 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3846 cpl.u.bs_basic.cb_fn = cb_fn; 3847 cpl.u.bs_basic.cb_arg = cb_arg; 3848 3849 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3850 if (!seq) { 3851 spdk_dma_free(ctx->super); 3852 free(ctx); 3853 cb_fn(cb_arg, -ENOMEM); 3854 return; 3855 } 3856 3857 /* Read super block */ 3858 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3859 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3860 _spdk_bs_unload_read_super_cpl, ctx); 3861 } 3862 3863 /* END spdk_bs_unload */ 3864 3865 /* START spdk_bs_set_super */ 3866 3867 struct spdk_bs_set_super_ctx { 3868 struct spdk_blob_store *bs; 3869 struct spdk_bs_super_block *super; 3870 }; 3871 3872 static void 3873 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3874 { 3875 struct spdk_bs_set_super_ctx *ctx = cb_arg; 3876 3877 if (bserrno != 0) { 3878 SPDK_ERRLOG("Unable to write to super block of blobstore\n"); 3879 } 3880 3881 spdk_dma_free(ctx->super); 3882 3883 spdk_bs_sequence_finish(seq, bserrno); 3884 3885 free(ctx); 3886 } 3887 3888 static void 3889 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3890 { 3891 struct spdk_bs_set_super_ctx *ctx = cb_arg; 3892 3893 if (bserrno != 0) { 3894 SPDK_ERRLOG("Unable to read super block of blobstore\n"); 3895 spdk_dma_free(ctx->super); 3896 spdk_bs_sequence_finish(seq, bserrno); 3897 free(ctx); 3898 return; 3899 } 3900 3901 _spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx); 3902 } 3903 3904 void 3905 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid, 3906 spdk_bs_op_complete cb_fn, void *cb_arg) 3907 { 3908 struct spdk_bs_cpl cpl; 3909 spdk_bs_sequence_t *seq; 3910 struct spdk_bs_set_super_ctx *ctx; 3911 3912 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n"); 3913 3914 ctx = calloc(1, sizeof(*ctx)); 3915 if (!ctx) { 3916 cb_fn(cb_arg, -ENOMEM); 3917 return; 3918 } 3919 3920 ctx->bs = bs; 3921 3922 ctx->super = spdk_dma_zmalloc(sizeof(*ctx->super), 0x1000, NULL); 3923 if (!ctx->super) { 3924 free(ctx); 3925 cb_fn(cb_arg, -ENOMEM); 3926 return; 3927 } 3928 3929 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3930 cpl.u.bs_basic.cb_fn = cb_fn; 3931 cpl.u.bs_basic.cb_arg = cb_arg; 3932 3933 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3934 if (!seq) { 3935 spdk_dma_free(ctx->super); 3936 free(ctx); 3937 cb_fn(cb_arg, -ENOMEM); 3938 return; 3939 } 3940 3941 bs->super_blob = blobid; 3942 3943 /* Read super block */ 3944 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3945 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3946 _spdk_bs_set_super_read_cpl, ctx); 3947 } 3948 3949 /* END spdk_bs_set_super */ 3950 3951 void 3952 spdk_bs_get_super(struct spdk_blob_store *bs, 3953 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 3954 { 3955 if (bs->super_blob == SPDK_BLOBID_INVALID) { 3956 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT); 3957 } else { 3958 cb_fn(cb_arg, bs->super_blob, 0); 3959 } 3960 } 3961 3962 uint64_t 3963 spdk_bs_get_cluster_size(struct spdk_blob_store *bs) 3964 { 3965 return bs->cluster_sz; 3966 } 3967 3968 uint64_t 3969 spdk_bs_get_page_size(struct spdk_blob_store *bs) 3970 { 3971 return SPDK_BS_PAGE_SIZE; 3972 } 3973 3974 uint64_t 3975 spdk_bs_free_cluster_count(struct spdk_blob_store *bs) 3976 { 3977 return bs->num_free_clusters; 3978 } 3979 3980 uint64_t 3981 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs) 3982 { 3983 return bs->total_data_clusters; 3984 } 3985 3986 static int 3987 spdk_bs_register_md_thread(struct spdk_blob_store *bs) 3988 { 3989 bs->md_channel = spdk_get_io_channel(bs); 3990 if (!bs->md_channel) { 3991 SPDK_ERRLOG("Failed to get IO channel.\n"); 3992 return -1; 3993 } 3994 3995 return 0; 3996 } 3997 3998 static int 3999 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs) 4000 { 4001 spdk_put_io_channel(bs->md_channel); 4002 4003 return 0; 4004 } 4005 4006 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob) 4007 { 4008 assert(blob != NULL); 4009 4010 return blob->id; 4011 } 4012 4013 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob) 4014 { 4015 assert(blob != NULL); 4016 4017 return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters); 4018 } 4019 4020 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob) 4021 { 4022 assert(blob != NULL); 4023 4024 return blob->active.num_clusters; 4025 } 4026 4027 /* START spdk_bs_create_blob */ 4028 4029 static void 4030 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4031 { 4032 struct spdk_blob *blob = cb_arg; 4033 4034 _spdk_blob_free(blob); 4035 4036 spdk_bs_sequence_finish(seq, bserrno); 4037 } 4038 4039 static int 4040 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs, 4041 bool internal) 4042 { 4043 uint64_t i; 4044 size_t value_len = 0; 4045 int rc; 4046 const void *value = NULL; 4047 if (xattrs->count > 0 && xattrs->get_value == NULL) { 4048 return -EINVAL; 4049 } 4050 for (i = 0; i < xattrs->count; i++) { 4051 xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len); 4052 if (value == NULL || value_len == 0) { 4053 return -EINVAL; 4054 } 4055 rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal); 4056 if (rc < 0) { 4057 return rc; 4058 } 4059 } 4060 return 0; 4061 } 4062 4063 static void 4064 _spdk_blob_set_thin_provision(struct spdk_blob *blob) 4065 { 4066 _spdk_blob_verify_md_op(blob); 4067 blob->invalid_flags |= SPDK_BLOB_THIN_PROV; 4068 blob->state = SPDK_BLOB_STATE_DIRTY; 4069 } 4070 4071 static void 4072 _spdk_bs_create_blob(struct spdk_blob_store *bs, 4073 const struct spdk_blob_opts *opts, 4074 const struct spdk_blob_xattr_opts *internal_xattrs, 4075 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4076 { 4077 struct spdk_blob *blob; 4078 uint32_t page_idx; 4079 struct spdk_bs_cpl cpl; 4080 struct spdk_blob_opts opts_default; 4081 struct spdk_blob_xattr_opts internal_xattrs_default; 4082 spdk_bs_sequence_t *seq; 4083 spdk_blob_id id; 4084 int rc; 4085 4086 assert(spdk_get_thread() == bs->md_thread); 4087 4088 page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0); 4089 if (page_idx >= spdk_bit_array_capacity(bs->used_md_pages)) { 4090 cb_fn(cb_arg, 0, -ENOMEM); 4091 return; 4092 } 4093 spdk_bit_array_set(bs->used_blobids, page_idx); 4094 spdk_bit_array_set(bs->used_md_pages, page_idx); 4095 4096 id = _spdk_bs_page_to_blobid(page_idx); 4097 4098 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx); 4099 4100 blob = _spdk_blob_alloc(bs, id); 4101 if (!blob) { 4102 cb_fn(cb_arg, 0, -ENOMEM); 4103 return; 4104 } 4105 4106 if (!opts) { 4107 spdk_blob_opts_init(&opts_default); 4108 opts = &opts_default; 4109 } 4110 if (!internal_xattrs) { 4111 _spdk_blob_xattrs_init(&internal_xattrs_default); 4112 internal_xattrs = &internal_xattrs_default; 4113 } 4114 4115 rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false); 4116 if (rc < 0) { 4117 _spdk_blob_free(blob); 4118 cb_fn(cb_arg, 0, rc); 4119 return; 4120 } 4121 4122 rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true); 4123 if (rc < 0) { 4124 _spdk_blob_free(blob); 4125 cb_fn(cb_arg, 0, rc); 4126 return; 4127 } 4128 4129 if (opts->thin_provision) { 4130 _spdk_blob_set_thin_provision(blob); 4131 } 4132 4133 rc = _spdk_blob_resize(blob, opts->num_clusters); 4134 if (rc < 0) { 4135 _spdk_blob_free(blob); 4136 cb_fn(cb_arg, 0, rc); 4137 return; 4138 } 4139 cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 4140 cpl.u.blobid.cb_fn = cb_fn; 4141 cpl.u.blobid.cb_arg = cb_arg; 4142 cpl.u.blobid.blobid = blob->id; 4143 4144 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 4145 if (!seq) { 4146 _spdk_blob_free(blob); 4147 cb_fn(cb_arg, 0, -ENOMEM); 4148 return; 4149 } 4150 4151 _spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob); 4152 } 4153 4154 void spdk_bs_create_blob(struct spdk_blob_store *bs, 4155 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4156 { 4157 _spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg); 4158 } 4159 4160 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts, 4161 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4162 { 4163 _spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg); 4164 } 4165 4166 /* END spdk_bs_create_blob */ 4167 4168 /* START blob_cleanup */ 4169 4170 struct spdk_clone_snapshot_ctx { 4171 struct spdk_bs_cpl cpl; 4172 int bserrno; 4173 bool frozen; 4174 4175 struct spdk_io_channel *channel; 4176 4177 /* Current cluster for inflate operation */ 4178 uint64_t cluster; 4179 4180 /* For inflation force allocation of all unallocated clusters and remove 4181 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */ 4182 bool allocate_all; 4183 4184 struct { 4185 spdk_blob_id id; 4186 struct spdk_blob *blob; 4187 } original; 4188 struct { 4189 spdk_blob_id id; 4190 struct spdk_blob *blob; 4191 } new; 4192 4193 /* xattrs specified for snapshot/clones only. They have no impact on 4194 * the original blobs xattrs. */ 4195 const struct spdk_blob_xattr_opts *xattrs; 4196 }; 4197 4198 static void 4199 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno) 4200 { 4201 struct spdk_clone_snapshot_ctx *ctx = cb_arg; 4202 struct spdk_bs_cpl *cpl = &ctx->cpl; 4203 4204 if (bserrno != 0) { 4205 if (ctx->bserrno != 0) { 4206 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 4207 } else { 4208 ctx->bserrno = bserrno; 4209 } 4210 } 4211 4212 switch (cpl->type) { 4213 case SPDK_BS_CPL_TYPE_BLOBID: 4214 cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno); 4215 break; 4216 case SPDK_BS_CPL_TYPE_BLOB_BASIC: 4217 cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno); 4218 break; 4219 default: 4220 SPDK_UNREACHABLE(); 4221 break; 4222 } 4223 4224 free(ctx); 4225 } 4226 4227 static void 4228 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno) 4229 { 4230 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4231 struct spdk_blob *origblob = ctx->original.blob; 4232 4233 if (bserrno != 0) { 4234 if (ctx->bserrno != 0) { 4235 SPDK_ERRLOG("Unfreeze error %d\n", bserrno); 4236 } else { 4237 ctx->bserrno = bserrno; 4238 } 4239 } 4240 4241 ctx->original.id = origblob->id; 4242 spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx); 4243 } 4244 4245 static void 4246 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno) 4247 { 4248 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4249 struct spdk_blob *origblob = ctx->original.blob; 4250 4251 if (bserrno != 0) { 4252 if (ctx->bserrno != 0) { 4253 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 4254 } else { 4255 ctx->bserrno = bserrno; 4256 } 4257 } 4258 4259 if (ctx->frozen) { 4260 /* Unfreeze any outstanding I/O */ 4261 _spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx); 4262 } else { 4263 _spdk_bs_snapshot_unfreeze_cpl(ctx, 0); 4264 } 4265 4266 } 4267 4268 static void 4269 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno) 4270 { 4271 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4272 struct spdk_blob *newblob = ctx->new.blob; 4273 4274 if (bserrno != 0) { 4275 if (ctx->bserrno != 0) { 4276 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 4277 } else { 4278 ctx->bserrno = bserrno; 4279 } 4280 } 4281 4282 ctx->new.id = newblob->id; 4283 spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4284 } 4285 4286 /* END blob_cleanup */ 4287 4288 /* START spdk_bs_create_snapshot */ 4289 4290 static void 4291 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno) 4292 { 4293 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4294 struct spdk_blob *newblob = ctx->new.blob; 4295 4296 if (bserrno != 0) { 4297 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4298 return; 4299 } 4300 4301 /* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */ 4302 bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true); 4303 if (bserrno != 0) { 4304 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4305 return; 4306 } 4307 4308 _spdk_bs_blob_list_add(ctx->original.blob); 4309 4310 spdk_blob_set_read_only(newblob); 4311 4312 /* sync snapshot metadata */ 4313 spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg); 4314 } 4315 4316 static void 4317 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno) 4318 { 4319 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4320 struct spdk_blob *origblob = ctx->original.blob; 4321 struct spdk_blob *newblob = ctx->new.blob; 4322 4323 if (bserrno != 0) { 4324 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4325 return; 4326 } 4327 4328 /* Set internal xattr for snapshot id */ 4329 bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true); 4330 if (bserrno != 0) { 4331 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4332 return; 4333 } 4334 4335 _spdk_bs_blob_list_remove(origblob); 4336 origblob->parent_id = newblob->id; 4337 4338 /* Create new back_bs_dev for snapshot */ 4339 origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob); 4340 if (origblob->back_bs_dev == NULL) { 4341 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL); 4342 return; 4343 } 4344 4345 /* set clone blob as thin provisioned */ 4346 _spdk_blob_set_thin_provision(origblob); 4347 4348 _spdk_bs_blob_list_add(newblob); 4349 4350 /* Zero out origblob cluster map */ 4351 memset(origblob->active.clusters, 0, 4352 origblob->active.num_clusters * sizeof(origblob->active.clusters)); 4353 4354 /* sync clone metadata */ 4355 spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx); 4356 } 4357 4358 static void 4359 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc) 4360 { 4361 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4362 struct spdk_blob *origblob = ctx->original.blob; 4363 struct spdk_blob *newblob = ctx->new.blob; 4364 int bserrno; 4365 4366 if (rc != 0) { 4367 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc); 4368 return; 4369 } 4370 4371 ctx->frozen = true; 4372 4373 /* set new back_bs_dev for snapshot */ 4374 newblob->back_bs_dev = origblob->back_bs_dev; 4375 /* Set invalid flags from origblob */ 4376 newblob->invalid_flags = origblob->invalid_flags; 4377 4378 /* inherit parent from original blob if set */ 4379 newblob->parent_id = origblob->parent_id; 4380 if (origblob->parent_id != SPDK_BLOBID_INVALID) { 4381 /* Set internal xattr for snapshot id */ 4382 bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT, 4383 &origblob->parent_id, sizeof(spdk_blob_id), true); 4384 if (bserrno != 0) { 4385 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4386 return; 4387 } 4388 } 4389 4390 /* Copy cluster map to snapshot */ 4391 memcpy(newblob->active.clusters, origblob->active.clusters, 4392 origblob->active.num_clusters * sizeof(origblob->active.clusters)); 4393 4394 /* sync snapshot metadata */ 4395 spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx); 4396 } 4397 4398 static void 4399 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4400 { 4401 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4402 struct spdk_blob *origblob = ctx->original.blob; 4403 struct spdk_blob *newblob = _blob; 4404 4405 if (bserrno != 0) { 4406 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4407 return; 4408 } 4409 4410 ctx->new.blob = newblob; 4411 4412 _spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx); 4413 } 4414 4415 static void 4416 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 4417 { 4418 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4419 struct spdk_blob *origblob = ctx->original.blob; 4420 4421 if (bserrno != 0) { 4422 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4423 return; 4424 } 4425 4426 ctx->new.id = blobid; 4427 ctx->cpl.u.blobid.blobid = blobid; 4428 4429 spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx); 4430 } 4431 4432 4433 static void 4434 _spdk_bs_xattr_snapshot(void *arg, const char *name, 4435 const void **value, size_t *value_len) 4436 { 4437 assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0); 4438 4439 struct spdk_blob *blob = (struct spdk_blob *)arg; 4440 *value = &blob->id; 4441 *value_len = sizeof(blob->id); 4442 } 4443 4444 static void 4445 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4446 { 4447 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4448 struct spdk_blob_opts opts; 4449 struct spdk_blob_xattr_opts internal_xattrs; 4450 char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS }; 4451 4452 if (bserrno != 0) { 4453 _spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno); 4454 return; 4455 } 4456 4457 ctx->original.blob = _blob; 4458 4459 if (_blob->data_ro || _blob->md_ro) { 4460 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n", 4461 _blob->id); 4462 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL); 4463 return; 4464 } 4465 4466 spdk_blob_opts_init(&opts); 4467 _spdk_blob_xattrs_init(&internal_xattrs); 4468 4469 /* Change the size of new blob to the same as in original blob, 4470 * but do not allocate clusters */ 4471 opts.thin_provision = true; 4472 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 4473 4474 /* If there are any xattrs specified for snapshot, set them now */ 4475 if (ctx->xattrs) { 4476 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 4477 } 4478 /* Set internal xattr SNAPSHOT_IN_PROGRESS */ 4479 internal_xattrs.count = 1; 4480 internal_xattrs.ctx = _blob; 4481 internal_xattrs.names = xattrs_names; 4482 internal_xattrs.get_value = _spdk_bs_xattr_snapshot; 4483 4484 _spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs, 4485 _spdk_bs_snapshot_newblob_create_cpl, ctx); 4486 } 4487 4488 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid, 4489 const struct spdk_blob_xattr_opts *snapshot_xattrs, 4490 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4491 { 4492 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 4493 4494 if (!ctx) { 4495 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 4496 return; 4497 } 4498 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 4499 ctx->cpl.u.blobid.cb_fn = cb_fn; 4500 ctx->cpl.u.blobid.cb_arg = cb_arg; 4501 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 4502 ctx->bserrno = 0; 4503 ctx->frozen = false; 4504 ctx->original.id = blobid; 4505 ctx->xattrs = snapshot_xattrs; 4506 4507 spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx); 4508 } 4509 /* END spdk_bs_create_snapshot */ 4510 4511 /* START spdk_bs_create_clone */ 4512 4513 static void 4514 _spdk_bs_xattr_clone(void *arg, const char *name, 4515 const void **value, size_t *value_len) 4516 { 4517 assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0); 4518 4519 struct spdk_blob *blob = (struct spdk_blob *)arg; 4520 *value = &blob->id; 4521 *value_len = sizeof(blob->id); 4522 } 4523 4524 static void 4525 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4526 { 4527 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4528 struct spdk_blob *clone = _blob; 4529 4530 ctx->new.blob = clone; 4531 _spdk_bs_blob_list_add(clone); 4532 4533 spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4534 } 4535 4536 static void 4537 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 4538 { 4539 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4540 4541 ctx->cpl.u.blobid.blobid = blobid; 4542 spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx); 4543 } 4544 4545 static void 4546 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4547 { 4548 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4549 struct spdk_blob_opts opts; 4550 struct spdk_blob_xattr_opts internal_xattrs; 4551 char *xattr_names[] = { BLOB_SNAPSHOT }; 4552 4553 if (bserrno != 0) { 4554 _spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno); 4555 return; 4556 } 4557 4558 ctx->original.blob = _blob; 4559 4560 if (!_blob->data_ro || !_blob->md_ro) { 4561 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n"); 4562 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL); 4563 return; 4564 } 4565 4566 spdk_blob_opts_init(&opts); 4567 _spdk_blob_xattrs_init(&internal_xattrs); 4568 4569 opts.thin_provision = true; 4570 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 4571 if (ctx->xattrs) { 4572 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 4573 } 4574 4575 /* Set internal xattr BLOB_SNAPSHOT */ 4576 internal_xattrs.count = 1; 4577 internal_xattrs.ctx = _blob; 4578 internal_xattrs.names = xattr_names; 4579 internal_xattrs.get_value = _spdk_bs_xattr_clone; 4580 4581 _spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs, 4582 _spdk_bs_clone_newblob_create_cpl, ctx); 4583 } 4584 4585 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid, 4586 const struct spdk_blob_xattr_opts *clone_xattrs, 4587 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4588 { 4589 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 4590 4591 if (!ctx) { 4592 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 4593 return; 4594 } 4595 4596 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 4597 ctx->cpl.u.blobid.cb_fn = cb_fn; 4598 ctx->cpl.u.blobid.cb_arg = cb_arg; 4599 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 4600 ctx->bserrno = 0; 4601 ctx->xattrs = clone_xattrs; 4602 ctx->original.id = blobid; 4603 4604 spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx); 4605 } 4606 4607 /* END spdk_bs_create_clone */ 4608 4609 /* START spdk_bs_inflate_blob */ 4610 4611 static void 4612 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno) 4613 { 4614 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4615 struct spdk_blob *_blob = ctx->original.blob; 4616 4617 if (bserrno != 0) { 4618 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4619 return; 4620 } 4621 4622 assert(_parent != NULL); 4623 4624 _spdk_bs_blob_list_remove(_blob); 4625 _blob->parent_id = _parent->id; 4626 _spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id, 4627 sizeof(spdk_blob_id), true); 4628 4629 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 4630 _blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent); 4631 _spdk_bs_blob_list_add(_blob); 4632 4633 spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4634 } 4635 4636 static void 4637 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno) 4638 { 4639 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4640 struct spdk_blob *_blob = ctx->original.blob; 4641 struct spdk_blob *_parent; 4642 4643 if (bserrno != 0) { 4644 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4645 return; 4646 } 4647 4648 if (ctx->allocate_all) { 4649 /* remove thin provisioning */ 4650 _spdk_bs_blob_list_remove(_blob); 4651 _spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 4652 _blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV; 4653 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 4654 _blob->back_bs_dev = NULL; 4655 _blob->parent_id = SPDK_BLOBID_INVALID; 4656 } else { 4657 _parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob; 4658 if (_parent->parent_id != SPDK_BLOBID_INVALID) { 4659 /* We must change the parent of the inflated blob */ 4660 spdk_bs_open_blob(_blob->bs, _parent->parent_id, 4661 _spdk_bs_inflate_blob_set_parent_cpl, ctx); 4662 return; 4663 } 4664 4665 _spdk_bs_blob_list_remove(_blob); 4666 _spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 4667 _blob->parent_id = SPDK_BLOBID_INVALID; 4668 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 4669 _blob->back_bs_dev = spdk_bs_create_zeroes_dev(); 4670 } 4671 4672 _blob->state = SPDK_BLOB_STATE_DIRTY; 4673 spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4674 } 4675 4676 /* Check if cluster needs allocation */ 4677 static inline bool 4678 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all) 4679 { 4680 struct spdk_blob_bs_dev *b; 4681 4682 assert(blob != NULL); 4683 4684 if (blob->active.clusters[cluster] != 0) { 4685 /* Cluster is already allocated */ 4686 return false; 4687 } 4688 4689 if (blob->parent_id == SPDK_BLOBID_INVALID) { 4690 /* Blob have no parent blob */ 4691 return allocate_all; 4692 } 4693 4694 b = (struct spdk_blob_bs_dev *)blob->back_bs_dev; 4695 return (allocate_all || b->blob->active.clusters[cluster] != 0); 4696 } 4697 4698 static void 4699 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno) 4700 { 4701 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4702 struct spdk_blob *_blob = ctx->original.blob; 4703 uint64_t offset; 4704 4705 if (bserrno != 0) { 4706 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4707 return; 4708 } 4709 4710 for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) { 4711 if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) { 4712 break; 4713 } 4714 } 4715 4716 if (ctx->cluster < _blob->active.num_clusters) { 4717 offset = _spdk_bs_cluster_to_page(_blob->bs, ctx->cluster); 4718 4719 /* We may safely increment a cluster before write */ 4720 ctx->cluster++; 4721 4722 /* Use zero length write to touch a cluster */ 4723 spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0, 4724 _spdk_bs_inflate_blob_touch_next, ctx); 4725 } else { 4726 _spdk_bs_inflate_blob_done(cb_arg, bserrno); 4727 } 4728 } 4729 4730 static void 4731 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4732 { 4733 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4734 uint64_t lfc; /* lowest free cluster */ 4735 uint64_t i; 4736 4737 if (bserrno != 0) { 4738 _spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno); 4739 return; 4740 } 4741 ctx->original.blob = _blob; 4742 4743 if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) { 4744 /* This blob have no parent, so we cannot decouple it. */ 4745 SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n"); 4746 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL); 4747 return; 4748 } 4749 4750 if (spdk_blob_is_thin_provisioned(_blob) == false) { 4751 /* This is not thin provisioned blob. No need to inflate. */ 4752 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0); 4753 return; 4754 } 4755 4756 /* Do two passes - one to verify that we can obtain enough clusters 4757 * and another to actually claim them. 4758 */ 4759 lfc = 0; 4760 for (i = 0; i < _blob->active.num_clusters; i++) { 4761 if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) { 4762 lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc); 4763 if (lfc >= _blob->bs->total_clusters) { 4764 /* No more free clusters. Cannot satisfy the request */ 4765 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC); 4766 return; 4767 } 4768 lfc++; 4769 } 4770 } 4771 4772 ctx->cluster = 0; 4773 _spdk_bs_inflate_blob_touch_next(ctx, 0); 4774 } 4775 4776 static void 4777 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 4778 spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg) 4779 { 4780 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 4781 4782 if (!ctx) { 4783 cb_fn(cb_arg, -ENOMEM); 4784 return; 4785 } 4786 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 4787 ctx->cpl.u.bs_basic.cb_fn = cb_fn; 4788 ctx->cpl.u.bs_basic.cb_arg = cb_arg; 4789 ctx->bserrno = 0; 4790 ctx->original.id = blobid; 4791 ctx->channel = channel; 4792 ctx->allocate_all = allocate_all; 4793 4794 spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx); 4795 } 4796 4797 void 4798 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 4799 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 4800 { 4801 _spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg); 4802 } 4803 4804 void 4805 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 4806 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 4807 { 4808 _spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg); 4809 } 4810 /* END spdk_bs_inflate_blob */ 4811 4812 /* START spdk_blob_resize */ 4813 struct spdk_bs_resize_ctx { 4814 spdk_blob_op_complete cb_fn; 4815 void *cb_arg; 4816 struct spdk_blob *blob; 4817 uint64_t sz; 4818 int rc; 4819 }; 4820 4821 static void 4822 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc) 4823 { 4824 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 4825 4826 if (rc != 0) { 4827 SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc); 4828 } 4829 4830 if (ctx->rc != 0) { 4831 SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc); 4832 rc = ctx->rc; 4833 } 4834 4835 ctx->blob->resize_in_progress = false; 4836 4837 ctx->cb_fn(ctx->cb_arg, rc); 4838 free(ctx); 4839 } 4840 4841 static void 4842 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc) 4843 { 4844 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 4845 4846 if (rc != 0) { 4847 ctx->blob->resize_in_progress = false; 4848 ctx->cb_fn(ctx->cb_arg, rc); 4849 free(ctx); 4850 return; 4851 } 4852 4853 ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz); 4854 4855 _spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx); 4856 } 4857 4858 void 4859 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg) 4860 { 4861 struct spdk_bs_resize_ctx *ctx; 4862 4863 _spdk_blob_verify_md_op(blob); 4864 4865 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz); 4866 4867 if (blob->md_ro) { 4868 cb_fn(cb_arg, -EPERM); 4869 return; 4870 } 4871 4872 if (sz == blob->active.num_clusters) { 4873 cb_fn(cb_arg, 0); 4874 return; 4875 } 4876 4877 if (blob->resize_in_progress) { 4878 cb_fn(cb_arg, -EBUSY); 4879 return; 4880 } 4881 4882 ctx = calloc(1, sizeof(*ctx)); 4883 if (!ctx) { 4884 cb_fn(cb_arg, -ENOMEM); 4885 return; 4886 } 4887 4888 blob->resize_in_progress = true; 4889 ctx->cb_fn = cb_fn; 4890 ctx->cb_arg = cb_arg; 4891 ctx->blob = blob; 4892 ctx->sz = sz; 4893 _spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx); 4894 } 4895 4896 /* END spdk_blob_resize */ 4897 4898 4899 /* START spdk_bs_delete_blob */ 4900 4901 static void 4902 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno) 4903 { 4904 spdk_bs_sequence_t *seq = cb_arg; 4905 4906 spdk_bs_sequence_finish(seq, bserrno); 4907 } 4908 4909 static void 4910 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4911 { 4912 struct spdk_blob *blob = cb_arg; 4913 4914 if (bserrno != 0) { 4915 /* 4916 * We already removed this blob from the blobstore tailq, so 4917 * we need to free it here since this is the last reference 4918 * to it. 4919 */ 4920 _spdk_blob_free(blob); 4921 _spdk_bs_delete_close_cpl(seq, bserrno); 4922 return; 4923 } 4924 4925 /* 4926 * This will immediately decrement the ref_count and call 4927 * the completion routine since the metadata state is clean. 4928 * By calling spdk_blob_close, we reduce the number of call 4929 * points into code that touches the blob->open_ref count 4930 * and the blobstore's blob list. 4931 */ 4932 spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq); 4933 } 4934 4935 static void 4936 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno) 4937 { 4938 spdk_bs_sequence_t *seq = cb_arg; 4939 uint32_t page_num; 4940 4941 if (bserrno != 0) { 4942 spdk_bs_sequence_finish(seq, bserrno); 4943 return; 4944 } 4945 4946 _spdk_blob_verify_md_op(blob); 4947 4948 if (blob->open_ref > 1) { 4949 /* 4950 * Someone has this blob open (besides this delete context). 4951 * Decrement the ref count directly and return -EBUSY. 4952 */ 4953 blob->open_ref--; 4954 spdk_bs_sequence_finish(seq, -EBUSY); 4955 return; 4956 } 4957 4958 bserrno = _spdk_bs_blob_list_remove(blob); 4959 if (bserrno != 0) { 4960 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Remove blob #%" PRIu64 " from a list\n", blob->id); 4961 spdk_bs_sequence_finish(seq, bserrno); 4962 return; 4963 } 4964 4965 /* 4966 * Remove the blob from the blob_store list now, to ensure it does not 4967 * get returned after this point by _spdk_blob_lookup(). 4968 */ 4969 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 4970 page_num = _spdk_bs_blobid_to_page(blob->id); 4971 spdk_bit_array_clear(blob->bs->used_blobids, page_num); 4972 blob->state = SPDK_BLOB_STATE_DIRTY; 4973 blob->active.num_pages = 0; 4974 _spdk_blob_resize(blob, 0); 4975 4976 _spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob); 4977 } 4978 4979 void 4980 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 4981 spdk_blob_op_complete cb_fn, void *cb_arg) 4982 { 4983 struct spdk_bs_cpl cpl; 4984 spdk_bs_sequence_t *seq; 4985 struct spdk_blob_list *snapshot_entry = NULL; 4986 4987 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid); 4988 4989 assert(spdk_get_thread() == bs->md_thread); 4990 4991 /* Check if this is a snapshot with clones */ 4992 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 4993 if (snapshot_entry->id == blobid) { 4994 break; 4995 } 4996 } 4997 if (snapshot_entry != NULL) { 4998 /* If snapshot have clones, we cannot remove it */ 4999 if (!TAILQ_EMPTY(&snapshot_entry->clones)) { 5000 SPDK_ERRLOG("Cannot remove snapshot with clones\n"); 5001 cb_fn(cb_arg, -EBUSY); 5002 return; 5003 } 5004 } 5005 5006 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 5007 cpl.u.blob_basic.cb_fn = cb_fn; 5008 cpl.u.blob_basic.cb_arg = cb_arg; 5009 5010 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 5011 if (!seq) { 5012 cb_fn(cb_arg, -ENOMEM); 5013 return; 5014 } 5015 5016 spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq); 5017 } 5018 5019 /* END spdk_bs_delete_blob */ 5020 5021 /* START spdk_bs_open_blob */ 5022 5023 static void 5024 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5025 { 5026 struct spdk_blob *blob = cb_arg; 5027 5028 /* If the blob have crc error, we just return NULL. */ 5029 if (blob == NULL) { 5030 seq->cpl.u.blob_handle.blob = NULL; 5031 spdk_bs_sequence_finish(seq, bserrno); 5032 return; 5033 } 5034 5035 blob->open_ref++; 5036 5037 TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link); 5038 5039 spdk_bs_sequence_finish(seq, bserrno); 5040 } 5041 5042 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 5043 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5044 { 5045 struct spdk_blob *blob; 5046 struct spdk_bs_cpl cpl; 5047 spdk_bs_sequence_t *seq; 5048 uint32_t page_num; 5049 5050 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid); 5051 assert(spdk_get_thread() == bs->md_thread); 5052 5053 page_num = _spdk_bs_blobid_to_page(blobid); 5054 if (spdk_bit_array_get(bs->used_blobids, page_num) == false) { 5055 /* Invalid blobid */ 5056 cb_fn(cb_arg, NULL, -ENOENT); 5057 return; 5058 } 5059 5060 blob = _spdk_blob_lookup(bs, blobid); 5061 if (blob) { 5062 blob->open_ref++; 5063 cb_fn(cb_arg, blob, 0); 5064 return; 5065 } 5066 5067 blob = _spdk_blob_alloc(bs, blobid); 5068 if (!blob) { 5069 cb_fn(cb_arg, NULL, -ENOMEM); 5070 return; 5071 } 5072 5073 cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE; 5074 cpl.u.blob_handle.cb_fn = cb_fn; 5075 cpl.u.blob_handle.cb_arg = cb_arg; 5076 cpl.u.blob_handle.blob = blob; 5077 5078 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 5079 if (!seq) { 5080 _spdk_blob_free(blob); 5081 cb_fn(cb_arg, NULL, -ENOMEM); 5082 return; 5083 } 5084 5085 _spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob); 5086 } 5087 /* END spdk_bs_open_blob */ 5088 5089 /* START spdk_blob_set_read_only */ 5090 int spdk_blob_set_read_only(struct spdk_blob *blob) 5091 { 5092 _spdk_blob_verify_md_op(blob); 5093 5094 blob->data_ro_flags |= SPDK_BLOB_READ_ONLY; 5095 5096 blob->state = SPDK_BLOB_STATE_DIRTY; 5097 return 0; 5098 } 5099 /* END spdk_blob_set_read_only */ 5100 5101 /* START spdk_blob_sync_md */ 5102 5103 static void 5104 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5105 { 5106 struct spdk_blob *blob = cb_arg; 5107 5108 if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 5109 blob->data_ro = true; 5110 blob->md_ro = true; 5111 } 5112 5113 spdk_bs_sequence_finish(seq, bserrno); 5114 } 5115 5116 static void 5117 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 5118 { 5119 struct spdk_bs_cpl cpl; 5120 spdk_bs_sequence_t *seq; 5121 5122 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 5123 cpl.u.blob_basic.cb_fn = cb_fn; 5124 cpl.u.blob_basic.cb_arg = cb_arg; 5125 5126 seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl); 5127 if (!seq) { 5128 cb_fn(cb_arg, -ENOMEM); 5129 return; 5130 } 5131 5132 _spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob); 5133 } 5134 5135 void 5136 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 5137 { 5138 _spdk_blob_verify_md_op(blob); 5139 5140 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id); 5141 5142 if (blob->md_ro) { 5143 assert(blob->state == SPDK_BLOB_STATE_CLEAN); 5144 cb_fn(cb_arg, 0); 5145 return; 5146 } 5147 5148 _spdk_blob_sync_md(blob, cb_fn, cb_arg); 5149 } 5150 5151 /* END spdk_blob_sync_md */ 5152 5153 struct spdk_blob_insert_cluster_ctx { 5154 struct spdk_thread *thread; 5155 struct spdk_blob *blob; 5156 uint32_t cluster_num; /* cluster index in blob */ 5157 uint32_t cluster; /* cluster on disk */ 5158 int rc; 5159 spdk_blob_op_complete cb_fn; 5160 void *cb_arg; 5161 }; 5162 5163 static void 5164 _spdk_blob_insert_cluster_msg_cpl(void *arg) 5165 { 5166 struct spdk_blob_insert_cluster_ctx *ctx = arg; 5167 5168 ctx->cb_fn(ctx->cb_arg, ctx->rc); 5169 free(ctx); 5170 } 5171 5172 static void 5173 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno) 5174 { 5175 struct spdk_blob_insert_cluster_ctx *ctx = arg; 5176 5177 ctx->rc = bserrno; 5178 spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx); 5179 } 5180 5181 static void 5182 _spdk_blob_insert_cluster_msg(void *arg) 5183 { 5184 struct spdk_blob_insert_cluster_ctx *ctx = arg; 5185 5186 ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster); 5187 if (ctx->rc != 0) { 5188 spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx); 5189 return; 5190 } 5191 5192 ctx->blob->state = SPDK_BLOB_STATE_DIRTY; 5193 _spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx); 5194 } 5195 5196 static void 5197 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 5198 uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg) 5199 { 5200 struct spdk_blob_insert_cluster_ctx *ctx; 5201 5202 ctx = calloc(1, sizeof(*ctx)); 5203 if (ctx == NULL) { 5204 cb_fn(cb_arg, -ENOMEM); 5205 return; 5206 } 5207 5208 ctx->thread = spdk_get_thread(); 5209 ctx->blob = blob; 5210 ctx->cluster_num = cluster_num; 5211 ctx->cluster = cluster; 5212 ctx->cb_fn = cb_fn; 5213 ctx->cb_arg = cb_arg; 5214 5215 spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx); 5216 } 5217 5218 /* START spdk_blob_close */ 5219 5220 static void 5221 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5222 { 5223 struct spdk_blob *blob = cb_arg; 5224 5225 if (bserrno == 0) { 5226 blob->open_ref--; 5227 if (blob->open_ref == 0) { 5228 /* 5229 * Blobs with active.num_pages == 0 are deleted blobs. 5230 * these blobs are removed from the blob_store list 5231 * when the deletion process starts - so don't try to 5232 * remove them again. 5233 */ 5234 if (blob->active.num_pages > 0) { 5235 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 5236 } 5237 _spdk_blob_free(blob); 5238 } 5239 } 5240 5241 spdk_bs_sequence_finish(seq, bserrno); 5242 } 5243 5244 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 5245 { 5246 struct spdk_bs_cpl cpl; 5247 spdk_bs_sequence_t *seq; 5248 5249 _spdk_blob_verify_md_op(blob); 5250 5251 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id); 5252 5253 if (blob->open_ref == 0) { 5254 cb_fn(cb_arg, -EBADF); 5255 return; 5256 } 5257 5258 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 5259 cpl.u.blob_basic.cb_fn = cb_fn; 5260 cpl.u.blob_basic.cb_arg = cb_arg; 5261 5262 seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl); 5263 if (!seq) { 5264 cb_fn(cb_arg, -ENOMEM); 5265 return; 5266 } 5267 5268 /* Sync metadata */ 5269 _spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob); 5270 } 5271 5272 /* END spdk_blob_close */ 5273 5274 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs) 5275 { 5276 return spdk_get_io_channel(bs); 5277 } 5278 5279 void spdk_bs_free_io_channel(struct spdk_io_channel *channel) 5280 { 5281 spdk_put_io_channel(channel); 5282 } 5283 5284 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel, 5285 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 5286 { 5287 _spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 5288 SPDK_BLOB_UNMAP); 5289 } 5290 5291 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel, 5292 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 5293 { 5294 _spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 5295 SPDK_BLOB_WRITE_ZEROES); 5296 } 5297 5298 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel, 5299 void *payload, uint64_t offset, uint64_t length, 5300 spdk_blob_op_complete cb_fn, void *cb_arg) 5301 { 5302 _spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 5303 SPDK_BLOB_WRITE); 5304 } 5305 5306 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel, 5307 void *payload, uint64_t offset, uint64_t length, 5308 spdk_blob_op_complete cb_fn, void *cb_arg) 5309 { 5310 _spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 5311 SPDK_BLOB_READ); 5312 } 5313 5314 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel, 5315 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 5316 spdk_blob_op_complete cb_fn, void *cb_arg) 5317 { 5318 _spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false); 5319 } 5320 5321 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel, 5322 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 5323 spdk_blob_op_complete cb_fn, void *cb_arg) 5324 { 5325 _spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true); 5326 } 5327 5328 struct spdk_bs_iter_ctx { 5329 int64_t page_num; 5330 struct spdk_blob_store *bs; 5331 5332 spdk_blob_op_with_handle_complete cb_fn; 5333 void *cb_arg; 5334 }; 5335 5336 static void 5337 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5338 { 5339 struct spdk_bs_iter_ctx *ctx = cb_arg; 5340 struct spdk_blob_store *bs = ctx->bs; 5341 spdk_blob_id id; 5342 5343 if (bserrno == 0) { 5344 ctx->cb_fn(ctx->cb_arg, _blob, bserrno); 5345 free(ctx); 5346 return; 5347 } 5348 5349 ctx->page_num++; 5350 ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num); 5351 if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) { 5352 ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT); 5353 free(ctx); 5354 return; 5355 } 5356 5357 id = _spdk_bs_page_to_blobid(ctx->page_num); 5358 5359 spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx); 5360 } 5361 5362 void 5363 spdk_bs_iter_first(struct spdk_blob_store *bs, 5364 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5365 { 5366 struct spdk_bs_iter_ctx *ctx; 5367 5368 ctx = calloc(1, sizeof(*ctx)); 5369 if (!ctx) { 5370 cb_fn(cb_arg, NULL, -ENOMEM); 5371 return; 5372 } 5373 5374 ctx->page_num = -1; 5375 ctx->bs = bs; 5376 ctx->cb_fn = cb_fn; 5377 ctx->cb_arg = cb_arg; 5378 5379 _spdk_bs_iter_cpl(ctx, NULL, -1); 5380 } 5381 5382 static void 5383 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno) 5384 { 5385 struct spdk_bs_iter_ctx *ctx = cb_arg; 5386 5387 _spdk_bs_iter_cpl(ctx, NULL, -1); 5388 } 5389 5390 void 5391 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob, 5392 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5393 { 5394 struct spdk_bs_iter_ctx *ctx; 5395 5396 assert(blob != NULL); 5397 5398 ctx = calloc(1, sizeof(*ctx)); 5399 if (!ctx) { 5400 cb_fn(cb_arg, NULL, -ENOMEM); 5401 return; 5402 } 5403 5404 ctx->page_num = _spdk_bs_blobid_to_page(blob->id); 5405 ctx->bs = bs; 5406 ctx->cb_fn = cb_fn; 5407 ctx->cb_arg = cb_arg; 5408 5409 /* Close the existing blob */ 5410 spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx); 5411 } 5412 5413 static int 5414 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 5415 uint16_t value_len, bool internal) 5416 { 5417 struct spdk_xattr_tailq *xattrs; 5418 struct spdk_xattr *xattr; 5419 5420 _spdk_blob_verify_md_op(blob); 5421 5422 if (blob->md_ro) { 5423 return -EPERM; 5424 } 5425 5426 if (internal) { 5427 xattrs = &blob->xattrs_internal; 5428 blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR; 5429 } else { 5430 xattrs = &blob->xattrs; 5431 } 5432 5433 TAILQ_FOREACH(xattr, xattrs, link) { 5434 if (!strcmp(name, xattr->name)) { 5435 free(xattr->value); 5436 xattr->value_len = value_len; 5437 xattr->value = malloc(value_len); 5438 memcpy(xattr->value, value, value_len); 5439 5440 blob->state = SPDK_BLOB_STATE_DIRTY; 5441 5442 return 0; 5443 } 5444 } 5445 5446 xattr = calloc(1, sizeof(*xattr)); 5447 if (!xattr) { 5448 return -ENOMEM; 5449 } 5450 xattr->name = strdup(name); 5451 xattr->value_len = value_len; 5452 xattr->value = malloc(value_len); 5453 memcpy(xattr->value, value, value_len); 5454 TAILQ_INSERT_TAIL(xattrs, xattr, link); 5455 5456 blob->state = SPDK_BLOB_STATE_DIRTY; 5457 5458 return 0; 5459 } 5460 5461 int 5462 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 5463 uint16_t value_len) 5464 { 5465 return _spdk_blob_set_xattr(blob, name, value, value_len, false); 5466 } 5467 5468 static int 5469 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal) 5470 { 5471 struct spdk_xattr_tailq *xattrs; 5472 struct spdk_xattr *xattr; 5473 5474 _spdk_blob_verify_md_op(blob); 5475 5476 if (blob->md_ro) { 5477 return -EPERM; 5478 } 5479 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 5480 5481 TAILQ_FOREACH(xattr, xattrs, link) { 5482 if (!strcmp(name, xattr->name)) { 5483 TAILQ_REMOVE(xattrs, xattr, link); 5484 free(xattr->value); 5485 free(xattr->name); 5486 free(xattr); 5487 5488 if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) { 5489 blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR; 5490 } 5491 blob->state = SPDK_BLOB_STATE_DIRTY; 5492 5493 return 0; 5494 } 5495 } 5496 5497 return -ENOENT; 5498 } 5499 5500 int 5501 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name) 5502 { 5503 return _spdk_blob_remove_xattr(blob, name, false); 5504 } 5505 5506 static int 5507 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 5508 const void **value, size_t *value_len, bool internal) 5509 { 5510 struct spdk_xattr *xattr; 5511 struct spdk_xattr_tailq *xattrs; 5512 5513 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 5514 5515 TAILQ_FOREACH(xattr, xattrs, link) { 5516 if (!strcmp(name, xattr->name)) { 5517 *value = xattr->value; 5518 *value_len = xattr->value_len; 5519 return 0; 5520 } 5521 } 5522 return -ENOENT; 5523 } 5524 5525 int 5526 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 5527 const void **value, size_t *value_len) 5528 { 5529 _spdk_blob_verify_md_op(blob); 5530 5531 return _spdk_blob_get_xattr_value(blob, name, value, value_len, false); 5532 } 5533 5534 struct spdk_xattr_names { 5535 uint32_t count; 5536 const char *names[0]; 5537 }; 5538 5539 static int 5540 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names) 5541 { 5542 struct spdk_xattr *xattr; 5543 int count = 0; 5544 5545 TAILQ_FOREACH(xattr, xattrs, link) { 5546 count++; 5547 } 5548 5549 *names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *)); 5550 if (*names == NULL) { 5551 return -ENOMEM; 5552 } 5553 5554 TAILQ_FOREACH(xattr, xattrs, link) { 5555 (*names)->names[(*names)->count++] = xattr->name; 5556 } 5557 5558 return 0; 5559 } 5560 5561 int 5562 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names) 5563 { 5564 _spdk_blob_verify_md_op(blob); 5565 5566 return _spdk_blob_get_xattr_names(&blob->xattrs, names); 5567 } 5568 5569 uint32_t 5570 spdk_xattr_names_get_count(struct spdk_xattr_names *names) 5571 { 5572 assert(names != NULL); 5573 5574 return names->count; 5575 } 5576 5577 const char * 5578 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index) 5579 { 5580 if (index >= names->count) { 5581 return NULL; 5582 } 5583 5584 return names->names[index]; 5585 } 5586 5587 void 5588 spdk_xattr_names_free(struct spdk_xattr_names *names) 5589 { 5590 free(names); 5591 } 5592 5593 struct spdk_bs_type 5594 spdk_bs_get_bstype(struct spdk_blob_store *bs) 5595 { 5596 return bs->bstype; 5597 } 5598 5599 void 5600 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype) 5601 { 5602 memcpy(&bs->bstype, &bstype, sizeof(bstype)); 5603 } 5604 5605 bool 5606 spdk_blob_is_read_only(struct spdk_blob *blob) 5607 { 5608 assert(blob != NULL); 5609 return (blob->data_ro || blob->md_ro); 5610 } 5611 5612 bool 5613 spdk_blob_is_snapshot(struct spdk_blob *blob) 5614 { 5615 struct spdk_blob_list *snapshot_entry; 5616 5617 assert(blob != NULL); 5618 5619 TAILQ_FOREACH(snapshot_entry, &blob->bs->snapshots, link) { 5620 if (snapshot_entry->id == blob->id) { 5621 break; 5622 } 5623 } 5624 5625 if (snapshot_entry == NULL) { 5626 return false; 5627 } 5628 5629 return true; 5630 } 5631 5632 bool 5633 spdk_blob_is_clone(struct spdk_blob *blob) 5634 { 5635 assert(blob != NULL); 5636 5637 if (blob->parent_id != SPDK_BLOBID_INVALID) { 5638 assert(spdk_blob_is_thin_provisioned(blob)); 5639 return true; 5640 } 5641 5642 return false; 5643 } 5644 5645 bool 5646 spdk_blob_is_thin_provisioned(struct spdk_blob *blob) 5647 { 5648 assert(blob != NULL); 5649 return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV); 5650 } 5651 5652 spdk_blob_id 5653 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id) 5654 { 5655 struct spdk_blob_list *snapshot_entry = NULL; 5656 struct spdk_blob_list *clone_entry = NULL; 5657 5658 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 5659 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 5660 if (clone_entry->id == blob_id) { 5661 return snapshot_entry->id; 5662 } 5663 } 5664 } 5665 5666 return SPDK_BLOBID_INVALID; 5667 } 5668 5669 int 5670 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids, 5671 size_t *count) 5672 { 5673 struct spdk_blob_list *snapshot_entry, *clone_entry; 5674 size_t n; 5675 5676 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 5677 if (snapshot_entry->id == blobid) { 5678 break; 5679 } 5680 } 5681 if (snapshot_entry == NULL) { 5682 *count = 0; 5683 return 0; 5684 } 5685 5686 if (ids == NULL || *count < snapshot_entry->clone_count) { 5687 *count = snapshot_entry->clone_count; 5688 return -ENOMEM; 5689 } 5690 *count = snapshot_entry->clone_count; 5691 5692 n = 0; 5693 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 5694 ids[n++] = clone_entry->id; 5695 } 5696 5697 return 0; 5698 } 5699 5700 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB) 5701