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/bit_pool.h" 43 #include "spdk/likely.h" 44 #include "spdk/util.h" 45 #include "spdk/string.h" 46 47 #include "spdk_internal/assert.h" 48 #include "spdk/log.h" 49 50 #include "blobstore.h" 51 52 #define BLOB_CRC32C_INITIAL 0xffffffffUL 53 54 static int bs_register_md_thread(struct spdk_blob_store *bs); 55 static int bs_unregister_md_thread(struct spdk_blob_store *bs); 56 static void blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno); 57 static void blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 58 uint64_t cluster, uint32_t extent, spdk_blob_op_complete cb_fn, void *cb_arg); 59 60 static int blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 61 uint16_t value_len, bool internal); 62 static int blob_get_xattr_value(struct spdk_blob *blob, const char *name, 63 const void **value, size_t *value_len, bool internal); 64 static int blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal); 65 66 static void blob_write_extent_page(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num, 67 spdk_blob_op_complete cb_fn, void *cb_arg); 68 69 static void 70 blob_verify_md_op(struct spdk_blob *blob) 71 { 72 assert(blob != NULL); 73 assert(spdk_get_thread() == blob->bs->md_thread); 74 assert(blob->state != SPDK_BLOB_STATE_LOADING); 75 } 76 77 static struct spdk_blob_list * 78 bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid) 79 { 80 struct spdk_blob_list *snapshot_entry = NULL; 81 82 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 83 if (snapshot_entry->id == blobid) { 84 break; 85 } 86 } 87 88 return snapshot_entry; 89 } 90 91 static void 92 bs_claim_md_page(struct spdk_blob_store *bs, uint32_t page) 93 { 94 assert(page < spdk_bit_array_capacity(bs->used_md_pages)); 95 assert(spdk_bit_array_get(bs->used_md_pages, page) == false); 96 97 spdk_bit_array_set(bs->used_md_pages, page); 98 } 99 100 static void 101 bs_release_md_page(struct spdk_blob_store *bs, uint32_t page) 102 { 103 assert(page < spdk_bit_array_capacity(bs->used_md_pages)); 104 assert(spdk_bit_array_get(bs->used_md_pages, page) == true); 105 106 spdk_bit_array_clear(bs->used_md_pages, page); 107 } 108 109 static uint32_t 110 bs_claim_cluster(struct spdk_blob_store *bs) 111 { 112 uint32_t cluster_num; 113 114 cluster_num = spdk_bit_pool_allocate_bit(bs->used_clusters); 115 if (cluster_num == UINT32_MAX) { 116 return UINT32_MAX; 117 } 118 119 SPDK_DEBUGLOG(blob, "Claiming cluster %u\n", cluster_num); 120 bs->num_free_clusters--; 121 122 return cluster_num; 123 } 124 125 static void 126 bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 127 { 128 assert(cluster_num < spdk_bit_pool_capacity(bs->used_clusters)); 129 assert(spdk_bit_pool_is_allocated(bs->used_clusters, cluster_num) == true); 130 assert(bs->num_free_clusters < bs->total_clusters); 131 132 SPDK_DEBUGLOG(blob, "Releasing cluster %u\n", cluster_num); 133 134 spdk_bit_pool_free_bit(bs->used_clusters, cluster_num); 135 bs->num_free_clusters++; 136 } 137 138 static int 139 blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster) 140 { 141 uint64_t *cluster_lba = &blob->active.clusters[cluster_num]; 142 143 blob_verify_md_op(blob); 144 145 if (*cluster_lba != 0) { 146 return -EEXIST; 147 } 148 149 *cluster_lba = bs_cluster_to_lba(blob->bs, cluster); 150 return 0; 151 } 152 153 static int 154 bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num, 155 uint64_t *cluster, uint32_t *lowest_free_md_page, bool update_map) 156 { 157 uint32_t *extent_page = 0; 158 159 *cluster = bs_claim_cluster(blob->bs); 160 if (*cluster == UINT32_MAX) { 161 /* No more free clusters. Cannot satisfy the request */ 162 return -ENOSPC; 163 } 164 165 if (blob->use_extent_table) { 166 extent_page = bs_cluster_to_extent_page(blob, cluster_num); 167 if (*extent_page == 0) { 168 /* Extent page shall never occupy md_page so start the search from 1 */ 169 if (*lowest_free_md_page == 0) { 170 *lowest_free_md_page = 1; 171 } 172 /* No extent_page is allocated for the cluster */ 173 *lowest_free_md_page = spdk_bit_array_find_first_clear(blob->bs->used_md_pages, 174 *lowest_free_md_page); 175 if (*lowest_free_md_page == UINT32_MAX) { 176 /* No more free md pages. Cannot satisfy the request */ 177 bs_release_cluster(blob->bs, *cluster); 178 return -ENOSPC; 179 } 180 bs_claim_md_page(blob->bs, *lowest_free_md_page); 181 } 182 } 183 184 SPDK_DEBUGLOG(blob, "Claiming cluster %" PRIu64 " for blob %" PRIu64 "\n", *cluster, blob->id); 185 186 if (update_map) { 187 blob_insert_cluster(blob, cluster_num, *cluster); 188 if (blob->use_extent_table && *extent_page == 0) { 189 *extent_page = *lowest_free_md_page; 190 } 191 } 192 193 return 0; 194 } 195 196 static void 197 blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs) 198 { 199 xattrs->count = 0; 200 xattrs->names = NULL; 201 xattrs->ctx = NULL; 202 xattrs->get_value = NULL; 203 } 204 205 void 206 spdk_blob_opts_init(struct spdk_blob_opts *opts, size_t opts_size) 207 { 208 if (!opts) { 209 SPDK_ERRLOG("opts should not be NULL\n"); 210 return; 211 } 212 213 if (!opts_size) { 214 SPDK_ERRLOG("opts_size should not be zero value\n"); 215 return; 216 } 217 218 memset(opts, 0, opts_size); 219 opts->opts_size = opts_size; 220 221 #define FIELD_OK(field) \ 222 offsetof(struct spdk_blob_opts, field) + sizeof(opts->field) <= opts_size 223 224 #define SET_FIELD(field, value) \ 225 if (FIELD_OK(field)) { \ 226 opts->field = value; \ 227 } \ 228 229 SET_FIELD(num_clusters, 0); 230 SET_FIELD(thin_provision, false); 231 SET_FIELD(clear_method, BLOB_CLEAR_WITH_DEFAULT); 232 233 if (FIELD_OK(xattrs)) { 234 blob_xattrs_init(&opts->xattrs); 235 } 236 237 SET_FIELD(use_extent_table, true); 238 239 #undef FIELD_OK 240 #undef SET_FIELD 241 } 242 243 void 244 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts, size_t opts_size) 245 { 246 if (!opts) { 247 SPDK_ERRLOG("opts should not be NULL\n"); 248 return; 249 } 250 251 if (!opts_size) { 252 SPDK_ERRLOG("opts_size should not be zero value\n"); 253 return; 254 } 255 256 memset(opts, 0, opts_size); 257 opts->opts_size = opts_size; 258 259 #define FIELD_OK(field) \ 260 offsetof(struct spdk_blob_open_opts, field) + sizeof(opts->field) <= opts_size 261 262 #define SET_FIELD(field, value) \ 263 if (FIELD_OK(field)) { \ 264 opts->field = value; \ 265 } \ 266 267 SET_FIELD(clear_method, BLOB_CLEAR_WITH_DEFAULT); 268 269 #undef FIELD_OK 270 #undef SET_FILED 271 } 272 273 static struct spdk_blob * 274 blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id) 275 { 276 struct spdk_blob *blob; 277 278 blob = calloc(1, sizeof(*blob)); 279 if (!blob) { 280 return NULL; 281 } 282 283 blob->id = id; 284 blob->bs = bs; 285 286 blob->parent_id = SPDK_BLOBID_INVALID; 287 288 blob->state = SPDK_BLOB_STATE_DIRTY; 289 blob->extent_rle_found = false; 290 blob->extent_table_found = false; 291 blob->active.num_pages = 1; 292 blob->active.pages = calloc(1, sizeof(*blob->active.pages)); 293 if (!blob->active.pages) { 294 free(blob); 295 return NULL; 296 } 297 298 blob->active.pages[0] = bs_blobid_to_page(id); 299 300 TAILQ_INIT(&blob->xattrs); 301 TAILQ_INIT(&blob->xattrs_internal); 302 TAILQ_INIT(&blob->pending_persists); 303 TAILQ_INIT(&blob->persists_to_complete); 304 305 return blob; 306 } 307 308 static void 309 xattrs_free(struct spdk_xattr_tailq *xattrs) 310 { 311 struct spdk_xattr *xattr, *xattr_tmp; 312 313 TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) { 314 TAILQ_REMOVE(xattrs, xattr, link); 315 free(xattr->name); 316 free(xattr->value); 317 free(xattr); 318 } 319 } 320 321 static void 322 blob_free(struct spdk_blob *blob) 323 { 324 assert(blob != NULL); 325 assert(TAILQ_EMPTY(&blob->pending_persists)); 326 assert(TAILQ_EMPTY(&blob->persists_to_complete)); 327 328 free(blob->active.extent_pages); 329 free(blob->clean.extent_pages); 330 free(blob->active.clusters); 331 free(blob->clean.clusters); 332 free(blob->active.pages); 333 free(blob->clean.pages); 334 335 xattrs_free(&blob->xattrs); 336 xattrs_free(&blob->xattrs_internal); 337 338 if (blob->back_bs_dev) { 339 blob->back_bs_dev->destroy(blob->back_bs_dev); 340 } 341 342 free(blob); 343 } 344 345 struct freeze_io_ctx { 346 struct spdk_bs_cpl cpl; 347 struct spdk_blob *blob; 348 }; 349 350 static void 351 blob_io_sync(struct spdk_io_channel_iter *i) 352 { 353 spdk_for_each_channel_continue(i, 0); 354 } 355 356 static void 357 blob_execute_queued_io(struct spdk_io_channel_iter *i) 358 { 359 struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i); 360 struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch); 361 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 362 struct spdk_bs_request_set *set; 363 struct spdk_bs_user_op_args *args; 364 spdk_bs_user_op_t *op, *tmp; 365 366 TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) { 367 set = (struct spdk_bs_request_set *)op; 368 args = &set->u.user_op; 369 370 if (args->blob == ctx->blob) { 371 TAILQ_REMOVE(&ch->queued_io, op, link); 372 bs_user_op_execute(op); 373 } 374 } 375 376 spdk_for_each_channel_continue(i, 0); 377 } 378 379 static void 380 blob_io_cpl(struct spdk_io_channel_iter *i, int status) 381 { 382 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 383 384 ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0); 385 386 free(ctx); 387 } 388 389 static void 390 blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 391 { 392 struct freeze_io_ctx *ctx; 393 394 ctx = calloc(1, sizeof(*ctx)); 395 if (!ctx) { 396 cb_fn(cb_arg, -ENOMEM); 397 return; 398 } 399 400 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 401 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 402 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 403 ctx->blob = blob; 404 405 /* Freeze I/O on blob */ 406 blob->frozen_refcnt++; 407 408 if (blob->frozen_refcnt == 1) { 409 spdk_for_each_channel(blob->bs, blob_io_sync, ctx, blob_io_cpl); 410 } else { 411 cb_fn(cb_arg, 0); 412 free(ctx); 413 } 414 } 415 416 static void 417 blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 418 { 419 struct freeze_io_ctx *ctx; 420 421 ctx = calloc(1, sizeof(*ctx)); 422 if (!ctx) { 423 cb_fn(cb_arg, -ENOMEM); 424 return; 425 } 426 427 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 428 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 429 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 430 ctx->blob = blob; 431 432 assert(blob->frozen_refcnt > 0); 433 434 blob->frozen_refcnt--; 435 436 if (blob->frozen_refcnt == 0) { 437 spdk_for_each_channel(blob->bs, blob_execute_queued_io, ctx, blob_io_cpl); 438 } else { 439 cb_fn(cb_arg, 0); 440 free(ctx); 441 } 442 } 443 444 static int 445 blob_mark_clean(struct spdk_blob *blob) 446 { 447 uint32_t *extent_pages = NULL; 448 uint64_t *clusters = NULL; 449 uint32_t *pages = NULL; 450 451 assert(blob != NULL); 452 453 if (blob->active.num_extent_pages) { 454 assert(blob->active.extent_pages); 455 extent_pages = calloc(blob->active.num_extent_pages, sizeof(*blob->active.extent_pages)); 456 if (!extent_pages) { 457 return -ENOMEM; 458 } 459 memcpy(extent_pages, blob->active.extent_pages, 460 blob->active.num_extent_pages * sizeof(*extent_pages)); 461 } 462 463 if (blob->active.num_clusters) { 464 assert(blob->active.clusters); 465 clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters)); 466 if (!clusters) { 467 free(extent_pages); 468 return -ENOMEM; 469 } 470 memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters)); 471 } 472 473 if (blob->active.num_pages) { 474 assert(blob->active.pages); 475 pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages)); 476 if (!pages) { 477 free(extent_pages); 478 free(clusters); 479 return -ENOMEM; 480 } 481 memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages)); 482 } 483 484 free(blob->clean.extent_pages); 485 free(blob->clean.clusters); 486 free(blob->clean.pages); 487 488 blob->clean.num_extent_pages = blob->active.num_extent_pages; 489 blob->clean.extent_pages = blob->active.extent_pages; 490 blob->clean.num_clusters = blob->active.num_clusters; 491 blob->clean.clusters = blob->active.clusters; 492 blob->clean.num_pages = blob->active.num_pages; 493 blob->clean.pages = blob->active.pages; 494 495 blob->active.extent_pages = extent_pages; 496 blob->active.clusters = clusters; 497 blob->active.pages = pages; 498 499 /* If the metadata was dirtied again while the metadata was being written to disk, 500 * we do not want to revert the DIRTY state back to CLEAN here. 501 */ 502 if (blob->state == SPDK_BLOB_STATE_LOADING) { 503 blob->state = SPDK_BLOB_STATE_CLEAN; 504 } 505 506 return 0; 507 } 508 509 static int 510 blob_deserialize_xattr(struct spdk_blob *blob, 511 struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal) 512 { 513 struct spdk_xattr *xattr; 514 515 if (desc_xattr->length != sizeof(desc_xattr->name_length) + 516 sizeof(desc_xattr->value_length) + 517 desc_xattr->name_length + desc_xattr->value_length) { 518 return -EINVAL; 519 } 520 521 xattr = calloc(1, sizeof(*xattr)); 522 if (xattr == NULL) { 523 return -ENOMEM; 524 } 525 526 xattr->name = malloc(desc_xattr->name_length + 1); 527 if (xattr->name == NULL) { 528 free(xattr); 529 return -ENOMEM; 530 } 531 memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length); 532 xattr->name[desc_xattr->name_length] = '\0'; 533 534 xattr->value = malloc(desc_xattr->value_length); 535 if (xattr->value == NULL) { 536 free(xattr->name); 537 free(xattr); 538 return -ENOMEM; 539 } 540 xattr->value_len = desc_xattr->value_length; 541 memcpy(xattr->value, 542 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 543 desc_xattr->value_length); 544 545 TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link); 546 547 return 0; 548 } 549 550 551 static int 552 blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob) 553 { 554 struct spdk_blob_md_descriptor *desc; 555 size_t cur_desc = 0; 556 void *tmp; 557 558 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 559 while (cur_desc < sizeof(page->descriptors)) { 560 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 561 if (desc->length == 0) { 562 /* If padding and length are 0, this terminates the page */ 563 break; 564 } 565 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 566 struct spdk_blob_md_descriptor_flags *desc_flags; 567 568 desc_flags = (struct spdk_blob_md_descriptor_flags *)desc; 569 570 if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) { 571 return -EINVAL; 572 } 573 574 if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) != 575 SPDK_BLOB_INVALID_FLAGS_MASK) { 576 return -EINVAL; 577 } 578 579 if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) != 580 SPDK_BLOB_DATA_RO_FLAGS_MASK) { 581 blob->data_ro = true; 582 blob->md_ro = true; 583 } 584 585 if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) != 586 SPDK_BLOB_MD_RO_FLAGS_MASK) { 587 blob->md_ro = true; 588 } 589 590 if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 591 blob->data_ro = true; 592 blob->md_ro = true; 593 } 594 595 blob->invalid_flags = desc_flags->invalid_flags; 596 blob->data_ro_flags = desc_flags->data_ro_flags; 597 blob->md_ro_flags = desc_flags->md_ro_flags; 598 599 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) { 600 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 601 unsigned int i, j; 602 unsigned int cluster_count = blob->active.num_clusters; 603 604 if (blob->extent_table_found) { 605 /* Extent Table already present in the md, 606 * both descriptors should never be at the same time. */ 607 return -EINVAL; 608 } 609 blob->extent_rle_found = true; 610 611 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc; 612 613 if (desc_extent_rle->length == 0 || 614 (desc_extent_rle->length % sizeof(desc_extent_rle->extents[0]) != 0)) { 615 return -EINVAL; 616 } 617 618 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 619 for (j = 0; j < desc_extent_rle->extents[i].length; j++) { 620 if (desc_extent_rle->extents[i].cluster_idx != 0) { 621 if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters, 622 desc_extent_rle->extents[i].cluster_idx + j)) { 623 return -EINVAL; 624 } 625 } 626 cluster_count++; 627 } 628 } 629 630 if (cluster_count == 0) { 631 return -EINVAL; 632 } 633 tmp = realloc(blob->active.clusters, cluster_count * sizeof(*blob->active.clusters)); 634 if (tmp == NULL) { 635 return -ENOMEM; 636 } 637 blob->active.clusters = tmp; 638 blob->active.cluster_array_size = cluster_count; 639 640 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 641 for (j = 0; j < desc_extent_rle->extents[i].length; j++) { 642 if (desc_extent_rle->extents[i].cluster_idx != 0) { 643 blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs, 644 desc_extent_rle->extents[i].cluster_idx + j); 645 } else if (spdk_blob_is_thin_provisioned(blob)) { 646 blob->active.clusters[blob->active.num_clusters++] = 0; 647 } else { 648 return -EINVAL; 649 } 650 } 651 } 652 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) { 653 struct spdk_blob_md_descriptor_extent_table *desc_extent_table; 654 uint32_t num_extent_pages = blob->active.num_extent_pages; 655 uint32_t i, j; 656 size_t extent_pages_length; 657 658 desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc; 659 extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters); 660 661 if (blob->extent_rle_found) { 662 /* This means that Extent RLE is present in MD, 663 * both should never be at the same time. */ 664 return -EINVAL; 665 } else if (blob->extent_table_found && 666 desc_extent_table->num_clusters != blob->remaining_clusters_in_et) { 667 /* Number of clusters in this ET does not match number 668 * from previously read EXTENT_TABLE. */ 669 return -EINVAL; 670 } 671 672 blob->extent_table_found = true; 673 674 if (desc_extent_table->length == 0 || 675 (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) { 676 return -EINVAL; 677 } 678 679 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 680 num_extent_pages += desc_extent_table->extent_page[i].num_pages; 681 } 682 683 tmp = realloc(blob->active.extent_pages, num_extent_pages * sizeof(uint32_t)); 684 if (tmp == NULL) { 685 return -ENOMEM; 686 } 687 blob->active.extent_pages = tmp; 688 blob->active.extent_pages_array_size = num_extent_pages; 689 690 blob->remaining_clusters_in_et = desc_extent_table->num_clusters; 691 692 /* Extent table entries contain md page numbers for extent pages. 693 * Zeroes represent unallocated extent pages, those are run-length-encoded. 694 */ 695 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 696 if (desc_extent_table->extent_page[i].page_idx != 0) { 697 assert(desc_extent_table->extent_page[i].num_pages == 1); 698 blob->active.extent_pages[blob->active.num_extent_pages++] = 699 desc_extent_table->extent_page[i].page_idx; 700 } else if (spdk_blob_is_thin_provisioned(blob)) { 701 for (j = 0; j < desc_extent_table->extent_page[i].num_pages; j++) { 702 blob->active.extent_pages[blob->active.num_extent_pages++] = 0; 703 } 704 } else { 705 return -EINVAL; 706 } 707 } 708 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 709 struct spdk_blob_md_descriptor_extent_page *desc_extent; 710 unsigned int i; 711 unsigned int cluster_count = 0; 712 size_t cluster_idx_length; 713 714 if (blob->extent_rle_found) { 715 /* This means that Extent RLE is present in MD, 716 * both should never be at the same time. */ 717 return -EINVAL; 718 } 719 720 desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc; 721 cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx); 722 723 if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) || 724 (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) { 725 return -EINVAL; 726 } 727 728 for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) { 729 if (desc_extent->cluster_idx[i] != 0) { 730 if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters, desc_extent->cluster_idx[i])) { 731 return -EINVAL; 732 } 733 } 734 cluster_count++; 735 } 736 737 if (cluster_count == 0) { 738 return -EINVAL; 739 } 740 741 /* When reading extent pages sequentially starting cluster idx should match 742 * current size of a blob. 743 * If changed to batch reading, this check shall be removed. */ 744 if (desc_extent->start_cluster_idx != blob->active.num_clusters) { 745 return -EINVAL; 746 } 747 748 tmp = realloc(blob->active.clusters, 749 (cluster_count + blob->active.num_clusters) * sizeof(*blob->active.clusters)); 750 if (tmp == NULL) { 751 return -ENOMEM; 752 } 753 blob->active.clusters = tmp; 754 blob->active.cluster_array_size = (cluster_count + blob->active.num_clusters); 755 756 for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) { 757 if (desc_extent->cluster_idx[i] != 0) { 758 blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs, 759 desc_extent->cluster_idx[i]); 760 } else if (spdk_blob_is_thin_provisioned(blob)) { 761 blob->active.clusters[blob->active.num_clusters++] = 0; 762 } else { 763 return -EINVAL; 764 } 765 } 766 assert(desc_extent->start_cluster_idx + cluster_count == blob->active.num_clusters); 767 assert(blob->remaining_clusters_in_et >= cluster_count); 768 blob->remaining_clusters_in_et -= cluster_count; 769 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 770 int rc; 771 772 rc = blob_deserialize_xattr(blob, 773 (struct spdk_blob_md_descriptor_xattr *) desc, false); 774 if (rc != 0) { 775 return rc; 776 } 777 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 778 int rc; 779 780 rc = blob_deserialize_xattr(blob, 781 (struct spdk_blob_md_descriptor_xattr *) desc, true); 782 if (rc != 0) { 783 return rc; 784 } 785 } else { 786 /* Unrecognized descriptor type. Do not fail - just continue to the 787 * next descriptor. If this descriptor is associated with some feature 788 * defined in a newer version of blobstore, that version of blobstore 789 * should create and set an associated feature flag to specify if this 790 * blob can be loaded or not. 791 */ 792 } 793 794 /* Advance to the next descriptor */ 795 cur_desc += sizeof(*desc) + desc->length; 796 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 797 break; 798 } 799 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 800 } 801 802 return 0; 803 } 804 805 static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page); 806 807 static int 808 blob_parse_extent_page(struct spdk_blob_md_page *extent_page, struct spdk_blob *blob) 809 { 810 assert(blob != NULL); 811 assert(blob->state == SPDK_BLOB_STATE_LOADING); 812 813 if (bs_load_cur_extent_page_valid(extent_page) == false) { 814 return -ENOENT; 815 } 816 817 return blob_parse_page(extent_page, blob); 818 } 819 820 static int 821 blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count, 822 struct spdk_blob *blob) 823 { 824 const struct spdk_blob_md_page *page; 825 uint32_t i; 826 int rc; 827 void *tmp; 828 829 assert(page_count > 0); 830 assert(pages[0].sequence_num == 0); 831 assert(blob != NULL); 832 assert(blob->state == SPDK_BLOB_STATE_LOADING); 833 assert(blob->active.clusters == NULL); 834 835 /* The blobid provided doesn't match what's in the MD, this can 836 * happen for example if a bogus blobid is passed in through open. 837 */ 838 if (blob->id != pages[0].id) { 839 SPDK_ERRLOG("Blobid (%" PRIu64 ") doesn't match what's in metadata (%" PRIu64 ")\n", 840 blob->id, pages[0].id); 841 return -ENOENT; 842 } 843 844 tmp = realloc(blob->active.pages, page_count * sizeof(*blob->active.pages)); 845 if (!tmp) { 846 return -ENOMEM; 847 } 848 blob->active.pages = tmp; 849 850 blob->active.pages[0] = pages[0].id; 851 852 for (i = 1; i < page_count; i++) { 853 assert(spdk_bit_array_get(blob->bs->used_md_pages, pages[i - 1].next)); 854 blob->active.pages[i] = pages[i - 1].next; 855 } 856 blob->active.num_pages = page_count; 857 858 for (i = 0; i < page_count; i++) { 859 page = &pages[i]; 860 861 assert(page->id == blob->id); 862 assert(page->sequence_num == i); 863 864 rc = blob_parse_page(page, blob); 865 if (rc != 0) { 866 return rc; 867 } 868 } 869 870 return 0; 871 } 872 873 static int 874 blob_serialize_add_page(const struct spdk_blob *blob, 875 struct spdk_blob_md_page **pages, 876 uint32_t *page_count, 877 struct spdk_blob_md_page **last_page) 878 { 879 struct spdk_blob_md_page *page, *tmp_pages; 880 881 assert(pages != NULL); 882 assert(page_count != NULL); 883 884 *last_page = NULL; 885 if (*page_count == 0) { 886 assert(*pages == NULL); 887 *pages = spdk_malloc(SPDK_BS_PAGE_SIZE, 0, 888 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 889 if (*pages == NULL) { 890 return -ENOMEM; 891 } 892 *page_count = 1; 893 } else { 894 assert(*pages != NULL); 895 tmp_pages = spdk_realloc(*pages, SPDK_BS_PAGE_SIZE * (*page_count + 1), 0); 896 if (tmp_pages == NULL) { 897 return -ENOMEM; 898 } 899 (*page_count)++; 900 *pages = tmp_pages; 901 } 902 903 page = &(*pages)[*page_count - 1]; 904 memset(page, 0, sizeof(*page)); 905 page->id = blob->id; 906 page->sequence_num = *page_count - 1; 907 page->next = SPDK_INVALID_MD_PAGE; 908 *last_page = page; 909 910 return 0; 911 } 912 913 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor. 914 * Update required_sz on both success and failure. 915 * 916 */ 917 static int 918 blob_serialize_xattr(const struct spdk_xattr *xattr, 919 uint8_t *buf, size_t buf_sz, 920 size_t *required_sz, bool internal) 921 { 922 struct spdk_blob_md_descriptor_xattr *desc; 923 924 *required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) + 925 strlen(xattr->name) + 926 xattr->value_len; 927 928 if (buf_sz < *required_sz) { 929 return -1; 930 } 931 932 desc = (struct spdk_blob_md_descriptor_xattr *)buf; 933 934 desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR; 935 desc->length = sizeof(desc->name_length) + 936 sizeof(desc->value_length) + 937 strlen(xattr->name) + 938 xattr->value_len; 939 desc->name_length = strlen(xattr->name); 940 desc->value_length = xattr->value_len; 941 942 memcpy(desc->name, xattr->name, desc->name_length); 943 memcpy((void *)((uintptr_t)desc->name + desc->name_length), 944 xattr->value, 945 desc->value_length); 946 947 return 0; 948 } 949 950 static void 951 blob_serialize_extent_table_entry(const struct spdk_blob *blob, 952 uint64_t start_ep, uint64_t *next_ep, 953 uint8_t **buf, size_t *remaining_sz) 954 { 955 struct spdk_blob_md_descriptor_extent_table *desc; 956 size_t cur_sz; 957 uint64_t i, et_idx; 958 uint32_t extent_page, ep_len; 959 960 /* The buffer must have room for at least num_clusters entry */ 961 cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->num_clusters); 962 if (*remaining_sz < cur_sz) { 963 *next_ep = start_ep; 964 return; 965 } 966 967 desc = (struct spdk_blob_md_descriptor_extent_table *)*buf; 968 desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE; 969 970 desc->num_clusters = blob->active.num_clusters; 971 972 ep_len = 1; 973 et_idx = 0; 974 for (i = start_ep; i < blob->active.num_extent_pages; i++) { 975 if (*remaining_sz < cur_sz + sizeof(desc->extent_page[0])) { 976 /* If we ran out of buffer space, return */ 977 break; 978 } 979 980 extent_page = blob->active.extent_pages[i]; 981 /* Verify that next extent_page is unallocated */ 982 if (extent_page == 0 && 983 (i + 1 < blob->active.num_extent_pages && blob->active.extent_pages[i + 1] == 0)) { 984 ep_len++; 985 continue; 986 } 987 desc->extent_page[et_idx].page_idx = extent_page; 988 desc->extent_page[et_idx].num_pages = ep_len; 989 et_idx++; 990 991 ep_len = 1; 992 cur_sz += sizeof(desc->extent_page[et_idx]); 993 } 994 *next_ep = i; 995 996 desc->length = sizeof(desc->num_clusters) + sizeof(desc->extent_page[0]) * et_idx; 997 *remaining_sz -= sizeof(struct spdk_blob_md_descriptor) + desc->length; 998 *buf += sizeof(struct spdk_blob_md_descriptor) + desc->length; 999 } 1000 1001 static int 1002 blob_serialize_extent_table(const struct spdk_blob *blob, 1003 struct spdk_blob_md_page **pages, 1004 struct spdk_blob_md_page *cur_page, 1005 uint32_t *page_count, uint8_t **buf, 1006 size_t *remaining_sz) 1007 { 1008 uint64_t last_extent_page; 1009 int rc; 1010 1011 last_extent_page = 0; 1012 /* At least single extent table entry has to be always persisted. 1013 * Such case occurs with num_extent_pages == 0. */ 1014 while (last_extent_page <= blob->active.num_extent_pages) { 1015 blob_serialize_extent_table_entry(blob, last_extent_page, &last_extent_page, buf, 1016 remaining_sz); 1017 1018 if (last_extent_page == blob->active.num_extent_pages) { 1019 break; 1020 } 1021 1022 rc = blob_serialize_add_page(blob, pages, page_count, &cur_page); 1023 if (rc < 0) { 1024 return rc; 1025 } 1026 1027 *buf = (uint8_t *)cur_page->descriptors; 1028 *remaining_sz = sizeof(cur_page->descriptors); 1029 } 1030 1031 return 0; 1032 } 1033 1034 static void 1035 blob_serialize_extent_rle(const struct spdk_blob *blob, 1036 uint64_t start_cluster, uint64_t *next_cluster, 1037 uint8_t **buf, size_t *buf_sz) 1038 { 1039 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 1040 size_t cur_sz; 1041 uint64_t i, extent_idx; 1042 uint64_t lba, lba_per_cluster, lba_count; 1043 1044 /* The buffer must have room for at least one extent */ 1045 cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc_extent_rle->extents[0]); 1046 if (*buf_sz < cur_sz) { 1047 *next_cluster = start_cluster; 1048 return; 1049 } 1050 1051 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)*buf; 1052 desc_extent_rle->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE; 1053 1054 lba_per_cluster = bs_cluster_to_lba(blob->bs, 1); 1055 1056 lba = blob->active.clusters[start_cluster]; 1057 lba_count = lba_per_cluster; 1058 extent_idx = 0; 1059 for (i = start_cluster + 1; i < blob->active.num_clusters; i++) { 1060 if ((lba + lba_count) == blob->active.clusters[i] && lba != 0) { 1061 /* Run-length encode sequential non-zero LBA */ 1062 lba_count += lba_per_cluster; 1063 continue; 1064 } else if (lba == 0 && blob->active.clusters[i] == 0) { 1065 /* Run-length encode unallocated clusters */ 1066 lba_count += lba_per_cluster; 1067 continue; 1068 } 1069 desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 1070 desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster; 1071 extent_idx++; 1072 1073 cur_sz += sizeof(desc_extent_rle->extents[extent_idx]); 1074 1075 if (*buf_sz < cur_sz) { 1076 /* If we ran out of buffer space, return */ 1077 *next_cluster = i; 1078 break; 1079 } 1080 1081 lba = blob->active.clusters[i]; 1082 lba_count = lba_per_cluster; 1083 } 1084 1085 if (*buf_sz >= cur_sz) { 1086 desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 1087 desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster; 1088 extent_idx++; 1089 1090 *next_cluster = blob->active.num_clusters; 1091 } 1092 1093 desc_extent_rle->length = sizeof(desc_extent_rle->extents[0]) * extent_idx; 1094 *buf_sz -= sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length; 1095 *buf += sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length; 1096 } 1097 1098 static int 1099 blob_serialize_extents_rle(const struct spdk_blob *blob, 1100 struct spdk_blob_md_page **pages, 1101 struct spdk_blob_md_page *cur_page, 1102 uint32_t *page_count, uint8_t **buf, 1103 size_t *remaining_sz) 1104 { 1105 uint64_t last_cluster; 1106 int rc; 1107 1108 last_cluster = 0; 1109 while (last_cluster < blob->active.num_clusters) { 1110 blob_serialize_extent_rle(blob, last_cluster, &last_cluster, buf, remaining_sz); 1111 1112 if (last_cluster == blob->active.num_clusters) { 1113 break; 1114 } 1115 1116 rc = blob_serialize_add_page(blob, pages, page_count, &cur_page); 1117 if (rc < 0) { 1118 return rc; 1119 } 1120 1121 *buf = (uint8_t *)cur_page->descriptors; 1122 *remaining_sz = sizeof(cur_page->descriptors); 1123 } 1124 1125 return 0; 1126 } 1127 1128 static void 1129 blob_serialize_extent_page(const struct spdk_blob *blob, 1130 uint64_t cluster, struct spdk_blob_md_page *page) 1131 { 1132 struct spdk_blob_md_descriptor_extent_page *desc_extent; 1133 uint64_t i, extent_idx; 1134 uint64_t lba, lba_per_cluster; 1135 uint64_t start_cluster_idx = (cluster / SPDK_EXTENTS_PER_EP) * SPDK_EXTENTS_PER_EP; 1136 1137 desc_extent = (struct spdk_blob_md_descriptor_extent_page *) page->descriptors; 1138 desc_extent->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE; 1139 1140 lba_per_cluster = bs_cluster_to_lba(blob->bs, 1); 1141 1142 desc_extent->start_cluster_idx = start_cluster_idx; 1143 extent_idx = 0; 1144 for (i = start_cluster_idx; i < blob->active.num_clusters; i++) { 1145 lba = blob->active.clusters[i]; 1146 desc_extent->cluster_idx[extent_idx++] = lba / lba_per_cluster; 1147 if (extent_idx >= SPDK_EXTENTS_PER_EP) { 1148 break; 1149 } 1150 } 1151 desc_extent->length = sizeof(desc_extent->start_cluster_idx) + 1152 sizeof(desc_extent->cluster_idx[0]) * extent_idx; 1153 } 1154 1155 static void 1156 blob_serialize_flags(const struct spdk_blob *blob, 1157 uint8_t *buf, size_t *buf_sz) 1158 { 1159 struct spdk_blob_md_descriptor_flags *desc; 1160 1161 /* 1162 * Flags get serialized first, so we should always have room for the flags 1163 * descriptor. 1164 */ 1165 assert(*buf_sz >= sizeof(*desc)); 1166 1167 desc = (struct spdk_blob_md_descriptor_flags *)buf; 1168 desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS; 1169 desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor); 1170 desc->invalid_flags = blob->invalid_flags; 1171 desc->data_ro_flags = blob->data_ro_flags; 1172 desc->md_ro_flags = blob->md_ro_flags; 1173 1174 *buf_sz -= sizeof(*desc); 1175 } 1176 1177 static int 1178 blob_serialize_xattrs(const struct spdk_blob *blob, 1179 const struct spdk_xattr_tailq *xattrs, bool internal, 1180 struct spdk_blob_md_page **pages, 1181 struct spdk_blob_md_page *cur_page, 1182 uint32_t *page_count, uint8_t **buf, 1183 size_t *remaining_sz) 1184 { 1185 const struct spdk_xattr *xattr; 1186 int rc; 1187 1188 TAILQ_FOREACH(xattr, xattrs, link) { 1189 size_t required_sz = 0; 1190 1191 rc = blob_serialize_xattr(xattr, 1192 *buf, *remaining_sz, 1193 &required_sz, internal); 1194 if (rc < 0) { 1195 /* Need to add a new page to the chain */ 1196 rc = blob_serialize_add_page(blob, pages, page_count, 1197 &cur_page); 1198 if (rc < 0) { 1199 spdk_free(*pages); 1200 *pages = NULL; 1201 *page_count = 0; 1202 return rc; 1203 } 1204 1205 *buf = (uint8_t *)cur_page->descriptors; 1206 *remaining_sz = sizeof(cur_page->descriptors); 1207 1208 /* Try again */ 1209 required_sz = 0; 1210 rc = blob_serialize_xattr(xattr, 1211 *buf, *remaining_sz, 1212 &required_sz, internal); 1213 1214 if (rc < 0) { 1215 spdk_free(*pages); 1216 *pages = NULL; 1217 *page_count = 0; 1218 return rc; 1219 } 1220 } 1221 1222 *remaining_sz -= required_sz; 1223 *buf += required_sz; 1224 } 1225 1226 return 0; 1227 } 1228 1229 static int 1230 blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages, 1231 uint32_t *page_count) 1232 { 1233 struct spdk_blob_md_page *cur_page; 1234 int rc; 1235 uint8_t *buf; 1236 size_t remaining_sz; 1237 1238 assert(pages != NULL); 1239 assert(page_count != NULL); 1240 assert(blob != NULL); 1241 assert(blob->state == SPDK_BLOB_STATE_DIRTY); 1242 1243 *pages = NULL; 1244 *page_count = 0; 1245 1246 /* A blob always has at least 1 page, even if it has no descriptors */ 1247 rc = blob_serialize_add_page(blob, pages, page_count, &cur_page); 1248 if (rc < 0) { 1249 return rc; 1250 } 1251 1252 buf = (uint8_t *)cur_page->descriptors; 1253 remaining_sz = sizeof(cur_page->descriptors); 1254 1255 /* Serialize flags */ 1256 blob_serialize_flags(blob, buf, &remaining_sz); 1257 buf += sizeof(struct spdk_blob_md_descriptor_flags); 1258 1259 /* Serialize xattrs */ 1260 rc = blob_serialize_xattrs(blob, &blob->xattrs, false, 1261 pages, cur_page, page_count, &buf, &remaining_sz); 1262 if (rc < 0) { 1263 return rc; 1264 } 1265 1266 /* Serialize internal xattrs */ 1267 rc = blob_serialize_xattrs(blob, &blob->xattrs_internal, true, 1268 pages, cur_page, page_count, &buf, &remaining_sz); 1269 if (rc < 0) { 1270 return rc; 1271 } 1272 1273 if (blob->use_extent_table) { 1274 /* Serialize extent table */ 1275 rc = blob_serialize_extent_table(blob, pages, cur_page, page_count, &buf, &remaining_sz); 1276 } else { 1277 /* Serialize extents */ 1278 rc = blob_serialize_extents_rle(blob, pages, cur_page, page_count, &buf, &remaining_sz); 1279 } 1280 1281 return rc; 1282 } 1283 1284 struct spdk_blob_load_ctx { 1285 struct spdk_blob *blob; 1286 1287 struct spdk_blob_md_page *pages; 1288 uint32_t num_pages; 1289 uint32_t next_extent_page; 1290 spdk_bs_sequence_t *seq; 1291 1292 spdk_bs_sequence_cpl cb_fn; 1293 void *cb_arg; 1294 }; 1295 1296 static uint32_t 1297 blob_md_page_calc_crc(void *page) 1298 { 1299 uint32_t crc; 1300 1301 crc = BLOB_CRC32C_INITIAL; 1302 crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc); 1303 crc ^= BLOB_CRC32C_INITIAL; 1304 1305 return crc; 1306 1307 } 1308 1309 static void 1310 blob_load_final(struct spdk_blob_load_ctx *ctx, int bserrno) 1311 { 1312 struct spdk_blob *blob = ctx->blob; 1313 1314 if (bserrno == 0) { 1315 blob_mark_clean(blob); 1316 } 1317 1318 ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno); 1319 1320 /* Free the memory */ 1321 spdk_free(ctx->pages); 1322 free(ctx); 1323 } 1324 1325 static void 1326 blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno) 1327 { 1328 struct spdk_blob_load_ctx *ctx = cb_arg; 1329 struct spdk_blob *blob = ctx->blob; 1330 1331 if (bserrno == 0) { 1332 blob->back_bs_dev = bs_create_blob_bs_dev(snapshot); 1333 if (blob->back_bs_dev == NULL) { 1334 bserrno = -ENOMEM; 1335 } 1336 } 1337 if (bserrno != 0) { 1338 SPDK_ERRLOG("Snapshot fail\n"); 1339 } 1340 1341 blob_load_final(ctx, bserrno); 1342 } 1343 1344 static void blob_update_clear_method(struct spdk_blob *blob); 1345 1346 static void 1347 blob_load_backing_dev(void *cb_arg) 1348 { 1349 struct spdk_blob_load_ctx *ctx = cb_arg; 1350 struct spdk_blob *blob = ctx->blob; 1351 const void *value; 1352 size_t len; 1353 int rc; 1354 1355 if (spdk_blob_is_thin_provisioned(blob)) { 1356 rc = blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true); 1357 if (rc == 0) { 1358 if (len != sizeof(spdk_blob_id)) { 1359 blob_load_final(ctx, -EINVAL); 1360 return; 1361 } 1362 /* open snapshot blob and continue in the callback function */ 1363 blob->parent_id = *(spdk_blob_id *)value; 1364 spdk_bs_open_blob(blob->bs, blob->parent_id, 1365 blob_load_snapshot_cpl, ctx); 1366 return; 1367 } else { 1368 /* add zeroes_dev for thin provisioned blob */ 1369 blob->back_bs_dev = bs_create_zeroes_dev(); 1370 } 1371 } else { 1372 /* standard blob */ 1373 blob->back_bs_dev = NULL; 1374 } 1375 blob_load_final(ctx, 0); 1376 } 1377 1378 static void 1379 blob_load_cpl_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1380 { 1381 struct spdk_blob_load_ctx *ctx = cb_arg; 1382 struct spdk_blob *blob = ctx->blob; 1383 struct spdk_blob_md_page *page; 1384 uint64_t i; 1385 uint32_t crc; 1386 uint64_t lba; 1387 void *tmp; 1388 uint64_t sz; 1389 1390 if (bserrno) { 1391 SPDK_ERRLOG("Extent page read failed: %d\n", bserrno); 1392 blob_load_final(ctx, bserrno); 1393 return; 1394 } 1395 1396 if (ctx->pages == NULL) { 1397 /* First iteration of this function, allocate buffer for single EXTENT_PAGE */ 1398 ctx->pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0, 1399 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1400 if (!ctx->pages) { 1401 blob_load_final(ctx, -ENOMEM); 1402 return; 1403 } 1404 ctx->num_pages = 1; 1405 ctx->next_extent_page = 0; 1406 } else { 1407 page = &ctx->pages[0]; 1408 crc = blob_md_page_calc_crc(page); 1409 if (crc != page->crc) { 1410 blob_load_final(ctx, -EINVAL); 1411 return; 1412 } 1413 1414 if (page->next != SPDK_INVALID_MD_PAGE) { 1415 blob_load_final(ctx, -EINVAL); 1416 return; 1417 } 1418 1419 bserrno = blob_parse_extent_page(page, blob); 1420 if (bserrno) { 1421 blob_load_final(ctx, bserrno); 1422 return; 1423 } 1424 } 1425 1426 for (i = ctx->next_extent_page; i < blob->active.num_extent_pages; i++) { 1427 if (blob->active.extent_pages[i] != 0) { 1428 /* Extent page was allocated, read and parse it. */ 1429 lba = bs_md_page_to_lba(blob->bs, blob->active.extent_pages[i]); 1430 ctx->next_extent_page = i + 1; 1431 1432 bs_sequence_read_dev(seq, &ctx->pages[0], lba, 1433 bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE), 1434 blob_load_cpl_extents_cpl, ctx); 1435 return; 1436 } else { 1437 /* Thin provisioned blobs can point to unallocated extent pages. 1438 * In this case blob size should be increased by up to the amount left in remaining_clusters_in_et. */ 1439 1440 sz = spdk_min(blob->remaining_clusters_in_et, SPDK_EXTENTS_PER_EP); 1441 blob->active.num_clusters += sz; 1442 blob->remaining_clusters_in_et -= sz; 1443 1444 assert(spdk_blob_is_thin_provisioned(blob)); 1445 assert(i + 1 < blob->active.num_extent_pages || blob->remaining_clusters_in_et == 0); 1446 1447 tmp = realloc(blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters)); 1448 if (tmp == NULL) { 1449 blob_load_final(ctx, -ENOMEM); 1450 return; 1451 } 1452 memset(tmp + sizeof(*blob->active.clusters) * blob->active.cluster_array_size, 0, 1453 sizeof(*blob->active.clusters) * (blob->active.num_clusters - blob->active.cluster_array_size)); 1454 blob->active.clusters = tmp; 1455 blob->active.cluster_array_size = blob->active.num_clusters; 1456 } 1457 } 1458 1459 blob_load_backing_dev(ctx); 1460 } 1461 1462 static void 1463 blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1464 { 1465 struct spdk_blob_load_ctx *ctx = cb_arg; 1466 struct spdk_blob *blob = ctx->blob; 1467 struct spdk_blob_md_page *page; 1468 int rc; 1469 uint32_t crc; 1470 uint32_t current_page; 1471 1472 if (ctx->num_pages == 1) { 1473 current_page = bs_blobid_to_page(blob->id); 1474 } else { 1475 assert(ctx->num_pages != 0); 1476 page = &ctx->pages[ctx->num_pages - 2]; 1477 current_page = page->next; 1478 } 1479 1480 if (bserrno) { 1481 SPDK_ERRLOG("Metadata page %d read failed for blobid %" PRIu64 ": %d\n", 1482 current_page, blob->id, bserrno); 1483 blob_load_final(ctx, bserrno); 1484 return; 1485 } 1486 1487 page = &ctx->pages[ctx->num_pages - 1]; 1488 crc = blob_md_page_calc_crc(page); 1489 if (crc != page->crc) { 1490 SPDK_ERRLOG("Metadata page %d crc mismatch for blobid %" PRIu64 "\n", 1491 current_page, blob->id); 1492 blob_load_final(ctx, -EINVAL); 1493 return; 1494 } 1495 1496 if (page->next != SPDK_INVALID_MD_PAGE) { 1497 struct spdk_blob_md_page *tmp_pages; 1498 uint32_t next_page = page->next; 1499 uint64_t next_lba = bs_md_page_to_lba(blob->bs, next_page); 1500 1501 /* Read the next page */ 1502 tmp_pages = spdk_realloc(ctx->pages, (sizeof(*page) * (ctx->num_pages + 1)), 0); 1503 if (tmp_pages == NULL) { 1504 blob_load_final(ctx, -ENOMEM); 1505 return; 1506 } 1507 ctx->num_pages++; 1508 ctx->pages = tmp_pages; 1509 1510 bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1], 1511 next_lba, 1512 bs_byte_to_lba(blob->bs, sizeof(*page)), 1513 blob_load_cpl, ctx); 1514 return; 1515 } 1516 1517 /* Parse the pages */ 1518 rc = blob_parse(ctx->pages, ctx->num_pages, blob); 1519 if (rc) { 1520 blob_load_final(ctx, rc); 1521 return; 1522 } 1523 1524 if (blob->extent_table_found == true) { 1525 /* If EXTENT_TABLE was found, that means support for it should be enabled. */ 1526 assert(blob->extent_rle_found == false); 1527 blob->use_extent_table = true; 1528 } else { 1529 /* If EXTENT_RLE or no extent_* descriptor was found disable support 1530 * for extent table. No extent_* descriptors means that blob has length of 0 1531 * and no extent_rle descriptors were persisted for it. 1532 * EXTENT_TABLE if used, is always present in metadata regardless of length. */ 1533 blob->use_extent_table = false; 1534 } 1535 1536 /* Check the clear_method stored in metadata vs what may have been passed 1537 * via spdk_bs_open_blob_ext() and update accordingly. 1538 */ 1539 blob_update_clear_method(blob); 1540 1541 spdk_free(ctx->pages); 1542 ctx->pages = NULL; 1543 1544 if (blob->extent_table_found) { 1545 blob_load_cpl_extents_cpl(seq, ctx, 0); 1546 } else { 1547 blob_load_backing_dev(ctx); 1548 } 1549 } 1550 1551 /* Load a blob from disk given a blobid */ 1552 static void 1553 blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 1554 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 1555 { 1556 struct spdk_blob_load_ctx *ctx; 1557 struct spdk_blob_store *bs; 1558 uint32_t page_num; 1559 uint64_t lba; 1560 1561 blob_verify_md_op(blob); 1562 1563 bs = blob->bs; 1564 1565 ctx = calloc(1, sizeof(*ctx)); 1566 if (!ctx) { 1567 cb_fn(seq, cb_arg, -ENOMEM); 1568 return; 1569 } 1570 1571 ctx->blob = blob; 1572 ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, 0); 1573 if (!ctx->pages) { 1574 free(ctx); 1575 cb_fn(seq, cb_arg, -ENOMEM); 1576 return; 1577 } 1578 ctx->num_pages = 1; 1579 ctx->cb_fn = cb_fn; 1580 ctx->cb_arg = cb_arg; 1581 ctx->seq = seq; 1582 1583 page_num = bs_blobid_to_page(blob->id); 1584 lba = bs_md_page_to_lba(blob->bs, page_num); 1585 1586 blob->state = SPDK_BLOB_STATE_LOADING; 1587 1588 bs_sequence_read_dev(seq, &ctx->pages[0], lba, 1589 bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE), 1590 blob_load_cpl, ctx); 1591 } 1592 1593 struct spdk_blob_persist_ctx { 1594 struct spdk_blob *blob; 1595 1596 struct spdk_bs_super_block *super; 1597 1598 struct spdk_blob_md_page *pages; 1599 uint32_t next_extent_page; 1600 struct spdk_blob_md_page *extent_page; 1601 1602 spdk_bs_sequence_t *seq; 1603 spdk_bs_sequence_cpl cb_fn; 1604 void *cb_arg; 1605 TAILQ_ENTRY(spdk_blob_persist_ctx) link; 1606 }; 1607 1608 static void 1609 bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba, 1610 uint32_t lba_count) 1611 { 1612 switch (ctx->blob->clear_method) { 1613 case BLOB_CLEAR_WITH_DEFAULT: 1614 case BLOB_CLEAR_WITH_UNMAP: 1615 bs_batch_unmap_dev(batch, lba, lba_count); 1616 break; 1617 case BLOB_CLEAR_WITH_WRITE_ZEROES: 1618 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1619 break; 1620 case BLOB_CLEAR_WITH_NONE: 1621 default: 1622 break; 1623 } 1624 } 1625 1626 static void blob_persist_check_dirty(struct spdk_blob_persist_ctx *ctx); 1627 1628 static void 1629 blob_persist_complete_cb(void *arg) 1630 { 1631 struct spdk_blob_persist_ctx *ctx = arg; 1632 1633 /* Call user callback */ 1634 ctx->cb_fn(ctx->seq, ctx->cb_arg, 0); 1635 1636 /* Free the memory */ 1637 spdk_free(ctx->pages); 1638 free(ctx); 1639 } 1640 1641 static void 1642 blob_persist_complete(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx, int bserrno) 1643 { 1644 struct spdk_blob_persist_ctx *next_persist, *tmp; 1645 struct spdk_blob *blob = ctx->blob; 1646 1647 if (bserrno == 0) { 1648 blob_mark_clean(blob); 1649 } 1650 1651 assert(ctx == TAILQ_FIRST(&blob->persists_to_complete)); 1652 1653 /* Complete all persists that were pending when the current persist started */ 1654 TAILQ_FOREACH_SAFE(next_persist, &blob->persists_to_complete, link, tmp) { 1655 TAILQ_REMOVE(&blob->persists_to_complete, next_persist, link); 1656 spdk_thread_send_msg(spdk_get_thread(), blob_persist_complete_cb, next_persist); 1657 } 1658 1659 if (TAILQ_EMPTY(&blob->pending_persists)) { 1660 return; 1661 } 1662 1663 /* Queue up all pending persists for completion and start blob persist with first one */ 1664 TAILQ_SWAP(&blob->persists_to_complete, &blob->pending_persists, spdk_blob_persist_ctx, link); 1665 next_persist = TAILQ_FIRST(&blob->persists_to_complete); 1666 1667 blob->state = SPDK_BLOB_STATE_DIRTY; 1668 blob_persist_check_dirty(next_persist); 1669 } 1670 1671 static void 1672 blob_persist_clear_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1673 { 1674 struct spdk_blob_persist_ctx *ctx = cb_arg; 1675 struct spdk_blob *blob = ctx->blob; 1676 struct spdk_blob_store *bs = blob->bs; 1677 size_t i; 1678 1679 if (bserrno != 0) { 1680 blob_persist_complete(seq, ctx, bserrno); 1681 return; 1682 } 1683 1684 /* Release all extent_pages that were truncated */ 1685 for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) { 1686 /* Nothing to release if it was not allocated */ 1687 if (blob->active.extent_pages[i] != 0) { 1688 bs_release_md_page(bs, blob->active.extent_pages[i]); 1689 } 1690 } 1691 1692 if (blob->active.num_extent_pages == 0) { 1693 free(blob->active.extent_pages); 1694 blob->active.extent_pages = NULL; 1695 blob->active.extent_pages_array_size = 0; 1696 } else if (blob->active.num_extent_pages != blob->active.extent_pages_array_size) { 1697 #ifndef __clang_analyzer__ 1698 void *tmp; 1699 1700 /* scan-build really can't figure reallocs, workaround it */ 1701 tmp = realloc(blob->active.extent_pages, sizeof(uint32_t) * blob->active.num_extent_pages); 1702 assert(tmp != NULL); 1703 blob->active.extent_pages = tmp; 1704 #endif 1705 blob->active.extent_pages_array_size = blob->active.num_extent_pages; 1706 } 1707 1708 blob_persist_complete(seq, ctx, bserrno); 1709 } 1710 1711 static void 1712 blob_persist_clear_extents(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx) 1713 { 1714 struct spdk_blob *blob = ctx->blob; 1715 struct spdk_blob_store *bs = blob->bs; 1716 size_t i; 1717 uint64_t lba; 1718 uint32_t lba_count; 1719 spdk_bs_batch_t *batch; 1720 1721 batch = bs_sequence_to_batch(seq, blob_persist_clear_extents_cpl, ctx); 1722 lba_count = bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE); 1723 1724 /* Clear all extent_pages that were truncated */ 1725 for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) { 1726 /* Nothing to clear if it was not allocated */ 1727 if (blob->active.extent_pages[i] != 0) { 1728 lba = bs_md_page_to_lba(bs, blob->active.extent_pages[i]); 1729 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1730 } 1731 } 1732 1733 bs_batch_close(batch); 1734 } 1735 1736 static void 1737 blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1738 { 1739 struct spdk_blob_persist_ctx *ctx = cb_arg; 1740 struct spdk_blob *blob = ctx->blob; 1741 struct spdk_blob_store *bs = blob->bs; 1742 size_t i; 1743 1744 if (bserrno != 0) { 1745 blob_persist_complete(seq, ctx, bserrno); 1746 return; 1747 } 1748 1749 pthread_mutex_lock(&bs->used_clusters_mutex); 1750 /* Release all clusters that were truncated */ 1751 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1752 uint32_t cluster_num = bs_lba_to_cluster(bs, blob->active.clusters[i]); 1753 1754 /* Nothing to release if it was not allocated */ 1755 if (blob->active.clusters[i] != 0) { 1756 bs_release_cluster(bs, cluster_num); 1757 } 1758 } 1759 pthread_mutex_unlock(&bs->used_clusters_mutex); 1760 1761 if (blob->active.num_clusters == 0) { 1762 free(blob->active.clusters); 1763 blob->active.clusters = NULL; 1764 blob->active.cluster_array_size = 0; 1765 } else if (blob->active.num_clusters != blob->active.cluster_array_size) { 1766 #ifndef __clang_analyzer__ 1767 void *tmp; 1768 1769 /* scan-build really can't figure reallocs, workaround it */ 1770 tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * blob->active.num_clusters); 1771 assert(tmp != NULL); 1772 blob->active.clusters = tmp; 1773 1774 #endif 1775 blob->active.cluster_array_size = blob->active.num_clusters; 1776 } 1777 1778 /* Move on to clearing extent pages */ 1779 blob_persist_clear_extents(seq, ctx); 1780 } 1781 1782 static void 1783 blob_persist_clear_clusters(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx) 1784 { 1785 struct spdk_blob *blob = ctx->blob; 1786 struct spdk_blob_store *bs = blob->bs; 1787 spdk_bs_batch_t *batch; 1788 size_t i; 1789 uint64_t lba; 1790 uint32_t lba_count; 1791 1792 /* Clusters don't move around in blobs. The list shrinks or grows 1793 * at the end, but no changes ever occur in the middle of the list. 1794 */ 1795 1796 batch = bs_sequence_to_batch(seq, blob_persist_clear_clusters_cpl, ctx); 1797 1798 /* Clear all clusters that were truncated */ 1799 lba = 0; 1800 lba_count = 0; 1801 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1802 uint64_t next_lba = blob->active.clusters[i]; 1803 uint32_t next_lba_count = bs_cluster_to_lba(bs, 1); 1804 1805 if (next_lba > 0 && (lba + lba_count) == next_lba) { 1806 /* This cluster is contiguous with the previous one. */ 1807 lba_count += next_lba_count; 1808 continue; 1809 } else if (next_lba == 0) { 1810 continue; 1811 } 1812 1813 /* This cluster is not contiguous with the previous one. */ 1814 1815 /* If a run of LBAs previously existing, clear them now */ 1816 if (lba_count > 0) { 1817 bs_batch_clear_dev(ctx, batch, lba, lba_count); 1818 } 1819 1820 /* Start building the next batch */ 1821 lba = next_lba; 1822 if (next_lba > 0) { 1823 lba_count = next_lba_count; 1824 } else { 1825 lba_count = 0; 1826 } 1827 } 1828 1829 /* If we ended with a contiguous set of LBAs, clear them now */ 1830 if (lba_count > 0) { 1831 bs_batch_clear_dev(ctx, batch, lba, lba_count); 1832 } 1833 1834 bs_batch_close(batch); 1835 } 1836 1837 static void 1838 blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1839 { 1840 struct spdk_blob_persist_ctx *ctx = cb_arg; 1841 struct spdk_blob *blob = ctx->blob; 1842 struct spdk_blob_store *bs = blob->bs; 1843 size_t i; 1844 1845 if (bserrno != 0) { 1846 blob_persist_complete(seq, ctx, bserrno); 1847 return; 1848 } 1849 1850 /* This loop starts at 1 because the first page is special and handled 1851 * below. The pages (except the first) are never written in place, 1852 * so any pages in the clean list must be zeroed. 1853 */ 1854 for (i = 1; i < blob->clean.num_pages; i++) { 1855 bs_release_md_page(bs, blob->clean.pages[i]); 1856 } 1857 1858 if (blob->active.num_pages == 0) { 1859 uint32_t page_num; 1860 1861 page_num = bs_blobid_to_page(blob->id); 1862 bs_release_md_page(bs, page_num); 1863 } 1864 1865 /* Move on to clearing clusters */ 1866 blob_persist_clear_clusters(seq, ctx); 1867 } 1868 1869 static void 1870 blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1871 { 1872 struct spdk_blob_persist_ctx *ctx = cb_arg; 1873 struct spdk_blob *blob = ctx->blob; 1874 struct spdk_blob_store *bs = blob->bs; 1875 uint64_t lba; 1876 uint32_t lba_count; 1877 spdk_bs_batch_t *batch; 1878 size_t i; 1879 1880 if (bserrno != 0) { 1881 blob_persist_complete(seq, ctx, bserrno); 1882 return; 1883 } 1884 1885 batch = bs_sequence_to_batch(seq, blob_persist_zero_pages_cpl, ctx); 1886 1887 lba_count = bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE); 1888 1889 /* This loop starts at 1 because the first page is special and handled 1890 * below. The pages (except the first) are never written in place, 1891 * so any pages in the clean list must be zeroed. 1892 */ 1893 for (i = 1; i < blob->clean.num_pages; i++) { 1894 lba = bs_md_page_to_lba(bs, blob->clean.pages[i]); 1895 1896 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1897 } 1898 1899 /* The first page will only be zeroed if this is a delete. */ 1900 if (blob->active.num_pages == 0) { 1901 uint32_t page_num; 1902 1903 /* The first page in the metadata goes where the blobid indicates */ 1904 page_num = bs_blobid_to_page(blob->id); 1905 lba = bs_md_page_to_lba(bs, page_num); 1906 1907 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1908 } 1909 1910 bs_batch_close(batch); 1911 } 1912 1913 static void 1914 blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1915 { 1916 struct spdk_blob_persist_ctx *ctx = cb_arg; 1917 struct spdk_blob *blob = ctx->blob; 1918 struct spdk_blob_store *bs = blob->bs; 1919 uint64_t lba; 1920 uint32_t lba_count; 1921 struct spdk_blob_md_page *page; 1922 1923 if (bserrno != 0) { 1924 blob_persist_complete(seq, ctx, bserrno); 1925 return; 1926 } 1927 1928 if (blob->active.num_pages == 0) { 1929 /* Move on to the next step */ 1930 blob_persist_zero_pages(seq, ctx, 0); 1931 return; 1932 } 1933 1934 lba_count = bs_byte_to_lba(bs, sizeof(*page)); 1935 1936 page = &ctx->pages[0]; 1937 /* The first page in the metadata goes where the blobid indicates */ 1938 lba = bs_md_page_to_lba(bs, bs_blobid_to_page(blob->id)); 1939 1940 bs_sequence_write_dev(seq, page, lba, lba_count, 1941 blob_persist_zero_pages, ctx); 1942 } 1943 1944 static void 1945 blob_persist_write_page_chain(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx) 1946 { 1947 struct spdk_blob *blob = ctx->blob; 1948 struct spdk_blob_store *bs = blob->bs; 1949 uint64_t lba; 1950 uint32_t lba_count; 1951 struct spdk_blob_md_page *page; 1952 spdk_bs_batch_t *batch; 1953 size_t i; 1954 1955 /* Clusters don't move around in blobs. The list shrinks or grows 1956 * at the end, but no changes ever occur in the middle of the list. 1957 */ 1958 1959 lba_count = bs_byte_to_lba(bs, sizeof(*page)); 1960 1961 batch = bs_sequence_to_batch(seq, blob_persist_write_page_root, ctx); 1962 1963 /* This starts at 1. The root page is not written until 1964 * all of the others are finished 1965 */ 1966 for (i = 1; i < blob->active.num_pages; i++) { 1967 page = &ctx->pages[i]; 1968 assert(page->sequence_num == i); 1969 1970 lba = bs_md_page_to_lba(bs, blob->active.pages[i]); 1971 1972 bs_batch_write_dev(batch, page, lba, lba_count); 1973 } 1974 1975 bs_batch_close(batch); 1976 } 1977 1978 static int 1979 blob_resize(struct spdk_blob *blob, uint64_t sz) 1980 { 1981 uint64_t i; 1982 uint64_t *tmp; 1983 uint64_t cluster; 1984 uint32_t lfmd; /* lowest free md page */ 1985 uint64_t num_clusters; 1986 uint32_t *ep_tmp; 1987 uint64_t new_num_ep = 0, current_num_ep = 0; 1988 struct spdk_blob_store *bs; 1989 1990 bs = blob->bs; 1991 1992 blob_verify_md_op(blob); 1993 1994 if (blob->active.num_clusters == sz) { 1995 return 0; 1996 } 1997 1998 if (blob->active.num_clusters < blob->active.cluster_array_size) { 1999 /* If this blob was resized to be larger, then smaller, then 2000 * larger without syncing, then the cluster array already 2001 * contains spare assigned clusters we can use. 2002 */ 2003 num_clusters = spdk_min(blob->active.cluster_array_size, 2004 sz); 2005 } else { 2006 num_clusters = blob->active.num_clusters; 2007 } 2008 2009 if (blob->use_extent_table) { 2010 /* Round up since every cluster beyond current Extent Table size, 2011 * requires new extent page. */ 2012 new_num_ep = spdk_divide_round_up(sz, SPDK_EXTENTS_PER_EP); 2013 current_num_ep = spdk_divide_round_up(num_clusters, SPDK_EXTENTS_PER_EP); 2014 } 2015 2016 /* Check first that we have enough clusters and md pages before we start claiming them. */ 2017 if (sz > num_clusters && spdk_blob_is_thin_provisioned(blob) == false) { 2018 if ((sz - num_clusters) > bs->num_free_clusters) { 2019 return -ENOSPC; 2020 } 2021 lfmd = 0; 2022 for (i = current_num_ep; i < new_num_ep ; i++) { 2023 lfmd = spdk_bit_array_find_first_clear(blob->bs->used_md_pages, lfmd); 2024 if (lfmd == UINT32_MAX) { 2025 /* No more free md pages. Cannot satisfy the request */ 2026 return -ENOSPC; 2027 } 2028 } 2029 } 2030 2031 if (sz > num_clusters) { 2032 /* Expand the cluster array if necessary. 2033 * We only shrink the array when persisting. 2034 */ 2035 tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * sz); 2036 if (sz > 0 && tmp == NULL) { 2037 return -ENOMEM; 2038 } 2039 memset(tmp + blob->active.cluster_array_size, 0, 2040 sizeof(*blob->active.clusters) * (sz - blob->active.cluster_array_size)); 2041 blob->active.clusters = tmp; 2042 blob->active.cluster_array_size = sz; 2043 2044 /* Expand the extents table, only if enough clusters were added */ 2045 if (new_num_ep > current_num_ep && blob->use_extent_table) { 2046 ep_tmp = realloc(blob->active.extent_pages, sizeof(*blob->active.extent_pages) * new_num_ep); 2047 if (new_num_ep > 0 && ep_tmp == NULL) { 2048 return -ENOMEM; 2049 } 2050 memset(ep_tmp + blob->active.extent_pages_array_size, 0, 2051 sizeof(*blob->active.extent_pages) * (new_num_ep - blob->active.extent_pages_array_size)); 2052 blob->active.extent_pages = ep_tmp; 2053 blob->active.extent_pages_array_size = new_num_ep; 2054 } 2055 } 2056 2057 blob->state = SPDK_BLOB_STATE_DIRTY; 2058 2059 if (spdk_blob_is_thin_provisioned(blob) == false) { 2060 cluster = 0; 2061 lfmd = 0; 2062 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 2063 for (i = num_clusters; i < sz; i++) { 2064 bs_allocate_cluster(blob, i, &cluster, &lfmd, true); 2065 lfmd++; 2066 } 2067 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 2068 } 2069 2070 blob->active.num_clusters = sz; 2071 blob->active.num_extent_pages = new_num_ep; 2072 2073 return 0; 2074 } 2075 2076 static void 2077 blob_persist_generate_new_md(struct spdk_blob_persist_ctx *ctx) 2078 { 2079 spdk_bs_sequence_t *seq = ctx->seq; 2080 struct spdk_blob *blob = ctx->blob; 2081 struct spdk_blob_store *bs = blob->bs; 2082 uint64_t i; 2083 uint32_t page_num; 2084 void *tmp; 2085 int rc; 2086 2087 /* Generate the new metadata */ 2088 rc = blob_serialize(blob, &ctx->pages, &blob->active.num_pages); 2089 if (rc < 0) { 2090 blob_persist_complete(seq, ctx, rc); 2091 return; 2092 } 2093 2094 assert(blob->active.num_pages >= 1); 2095 2096 /* Resize the cache of page indices */ 2097 tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages)); 2098 if (!tmp) { 2099 blob_persist_complete(seq, ctx, -ENOMEM); 2100 return; 2101 } 2102 blob->active.pages = tmp; 2103 2104 /* Assign this metadata to pages. This requires two passes - 2105 * one to verify that there are enough pages and a second 2106 * to actually claim them. */ 2107 page_num = 0; 2108 /* Note that this loop starts at one. The first page location is fixed by the blobid. */ 2109 for (i = 1; i < blob->active.num_pages; i++) { 2110 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 2111 if (page_num == UINT32_MAX) { 2112 blob_persist_complete(seq, ctx, -ENOMEM); 2113 return; 2114 } 2115 page_num++; 2116 } 2117 2118 page_num = 0; 2119 blob->active.pages[0] = bs_blobid_to_page(blob->id); 2120 for (i = 1; i < blob->active.num_pages; i++) { 2121 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 2122 ctx->pages[i - 1].next = page_num; 2123 /* Now that previous metadata page is complete, calculate the crc for it. */ 2124 ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]); 2125 blob->active.pages[i] = page_num; 2126 bs_claim_md_page(bs, page_num); 2127 SPDK_DEBUGLOG(blob, "Claiming page %u for blob %" PRIu64 "\n", page_num, blob->id); 2128 page_num++; 2129 } 2130 ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]); 2131 /* Start writing the metadata from last page to first */ 2132 blob->state = SPDK_BLOB_STATE_CLEAN; 2133 blob_persist_write_page_chain(seq, ctx); 2134 } 2135 2136 static void 2137 blob_persist_write_extent_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2138 { 2139 struct spdk_blob_persist_ctx *ctx = cb_arg; 2140 struct spdk_blob *blob = ctx->blob; 2141 size_t i; 2142 uint32_t extent_page_id; 2143 uint32_t page_count = 0; 2144 int rc; 2145 2146 if (ctx->extent_page != NULL) { 2147 spdk_free(ctx->extent_page); 2148 ctx->extent_page = NULL; 2149 } 2150 2151 if (bserrno != 0) { 2152 blob_persist_complete(seq, ctx, bserrno); 2153 return; 2154 } 2155 2156 /* Only write out Extent Pages when blob was resized. */ 2157 for (i = ctx->next_extent_page; i < blob->active.extent_pages_array_size; i++) { 2158 extent_page_id = blob->active.extent_pages[i]; 2159 if (extent_page_id == 0) { 2160 /* No Extent Page to persist */ 2161 assert(spdk_blob_is_thin_provisioned(blob)); 2162 continue; 2163 } 2164 assert(spdk_bit_array_get(blob->bs->used_md_pages, extent_page_id)); 2165 ctx->next_extent_page = i + 1; 2166 rc = blob_serialize_add_page(ctx->blob, &ctx->extent_page, &page_count, &ctx->extent_page); 2167 if (rc < 0) { 2168 blob_persist_complete(seq, ctx, rc); 2169 return; 2170 } 2171 2172 blob->state = SPDK_BLOB_STATE_DIRTY; 2173 blob_serialize_extent_page(blob, i * SPDK_EXTENTS_PER_EP, ctx->extent_page); 2174 2175 ctx->extent_page->crc = blob_md_page_calc_crc(ctx->extent_page); 2176 2177 bs_sequence_write_dev(seq, ctx->extent_page, bs_md_page_to_lba(blob->bs, extent_page_id), 2178 bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE), 2179 blob_persist_write_extent_pages, ctx); 2180 return; 2181 } 2182 2183 blob_persist_generate_new_md(ctx); 2184 } 2185 2186 static void 2187 blob_persist_start(struct spdk_blob_persist_ctx *ctx) 2188 { 2189 spdk_bs_sequence_t *seq = ctx->seq; 2190 struct spdk_blob *blob = ctx->blob; 2191 2192 if (blob->active.num_pages == 0) { 2193 /* This is the signal that the blob should be deleted. 2194 * Immediately jump to the clean up routine. */ 2195 assert(blob->clean.num_pages > 0); 2196 blob->state = SPDK_BLOB_STATE_CLEAN; 2197 blob_persist_zero_pages(seq, ctx, 0); 2198 return; 2199 2200 } 2201 2202 if (blob->clean.num_clusters < blob->active.num_clusters) { 2203 /* Blob was resized up */ 2204 assert(blob->clean.num_extent_pages <= blob->active.num_extent_pages); 2205 ctx->next_extent_page = spdk_max(1, blob->clean.num_extent_pages) - 1; 2206 } else if (blob->active.num_clusters < blob->active.cluster_array_size) { 2207 /* Blob was resized down */ 2208 assert(blob->clean.num_extent_pages >= blob->active.num_extent_pages); 2209 ctx->next_extent_page = spdk_max(1, blob->active.num_extent_pages) - 1; 2210 } else { 2211 /* No change in size occured */ 2212 blob_persist_generate_new_md(ctx); 2213 return; 2214 } 2215 2216 blob_persist_write_extent_pages(seq, ctx, 0); 2217 } 2218 2219 static void 2220 blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2221 { 2222 struct spdk_blob_persist_ctx *ctx = cb_arg; 2223 2224 spdk_free(ctx->super); 2225 2226 if (bserrno != 0) { 2227 blob_persist_complete(seq, ctx, bserrno); 2228 return; 2229 } 2230 2231 ctx->blob->bs->clean = 0; 2232 2233 blob_persist_start(ctx); 2234 } 2235 2236 static void 2237 bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 2238 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg); 2239 2240 2241 static void 2242 blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2243 { 2244 struct spdk_blob_persist_ctx *ctx = cb_arg; 2245 2246 if (bserrno != 0) { 2247 spdk_free(ctx->super); 2248 blob_persist_complete(seq, ctx, bserrno); 2249 return; 2250 } 2251 2252 ctx->super->clean = 0; 2253 if (ctx->super->size == 0) { 2254 ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen; 2255 } 2256 2257 bs_write_super(seq, ctx->blob->bs, ctx->super, blob_persist_dirty_cpl, ctx); 2258 } 2259 2260 static void 2261 blob_persist_check_dirty(struct spdk_blob_persist_ctx *ctx) 2262 { 2263 if (ctx->blob->bs->clean) { 2264 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 2265 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 2266 if (!ctx->super) { 2267 blob_persist_complete(ctx->seq, ctx, -ENOMEM); 2268 return; 2269 } 2270 2271 bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(ctx->blob->bs, 0), 2272 bs_byte_to_lba(ctx->blob->bs, sizeof(*ctx->super)), 2273 blob_persist_dirty, ctx); 2274 } else { 2275 blob_persist_start(ctx); 2276 } 2277 } 2278 2279 /* Write a blob to disk */ 2280 static void 2281 blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 2282 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 2283 { 2284 struct spdk_blob_persist_ctx *ctx; 2285 2286 blob_verify_md_op(blob); 2287 2288 if (blob->state == SPDK_BLOB_STATE_CLEAN && TAILQ_EMPTY(&blob->persists_to_complete)) { 2289 cb_fn(seq, cb_arg, 0); 2290 return; 2291 } 2292 2293 ctx = calloc(1, sizeof(*ctx)); 2294 if (!ctx) { 2295 cb_fn(seq, cb_arg, -ENOMEM); 2296 return; 2297 } 2298 ctx->blob = blob; 2299 ctx->seq = seq; 2300 ctx->cb_fn = cb_fn; 2301 ctx->cb_arg = cb_arg; 2302 2303 /* Multiple blob persists can affect one another, via blob->state or 2304 * blob mutable data changes. To prevent it, queue up the persists. */ 2305 if (!TAILQ_EMPTY(&blob->persists_to_complete)) { 2306 TAILQ_INSERT_TAIL(&blob->pending_persists, ctx, link); 2307 return; 2308 } 2309 TAILQ_INSERT_HEAD(&blob->persists_to_complete, ctx, link); 2310 2311 blob_persist_check_dirty(ctx); 2312 } 2313 2314 struct spdk_blob_copy_cluster_ctx { 2315 struct spdk_blob *blob; 2316 uint8_t *buf; 2317 uint64_t page; 2318 uint64_t new_cluster; 2319 uint32_t new_extent_page; 2320 spdk_bs_sequence_t *seq; 2321 }; 2322 2323 static void 2324 blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno) 2325 { 2326 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2327 struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq; 2328 TAILQ_HEAD(, spdk_bs_request_set) requests; 2329 spdk_bs_user_op_t *op; 2330 2331 TAILQ_INIT(&requests); 2332 TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link); 2333 2334 while (!TAILQ_EMPTY(&requests)) { 2335 op = TAILQ_FIRST(&requests); 2336 TAILQ_REMOVE(&requests, op, link); 2337 if (bserrno == 0) { 2338 bs_user_op_execute(op); 2339 } else { 2340 bs_user_op_abort(op); 2341 } 2342 } 2343 2344 spdk_free(ctx->buf); 2345 free(ctx); 2346 } 2347 2348 static void 2349 blob_insert_cluster_cpl(void *cb_arg, int bserrno) 2350 { 2351 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2352 2353 if (bserrno) { 2354 if (bserrno == -EEXIST) { 2355 /* The metadata insert failed because another thread 2356 * allocated the cluster first. Free our cluster 2357 * but continue without error. */ 2358 bserrno = 0; 2359 } 2360 pthread_mutex_lock(&ctx->blob->bs->used_clusters_mutex); 2361 bs_release_cluster(ctx->blob->bs, ctx->new_cluster); 2362 pthread_mutex_unlock(&ctx->blob->bs->used_clusters_mutex); 2363 if (ctx->new_extent_page != 0) { 2364 bs_release_md_page(ctx->blob->bs, ctx->new_extent_page); 2365 } 2366 } 2367 2368 bs_sequence_finish(ctx->seq, bserrno); 2369 } 2370 2371 static void 2372 blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2373 { 2374 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2375 uint32_t cluster_number; 2376 2377 if (bserrno) { 2378 /* The write failed, so jump to the final completion handler */ 2379 bs_sequence_finish(seq, bserrno); 2380 return; 2381 } 2382 2383 cluster_number = bs_page_to_cluster(ctx->blob->bs, ctx->page); 2384 2385 blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 2386 ctx->new_extent_page, blob_insert_cluster_cpl, ctx); 2387 } 2388 2389 static void 2390 blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2391 { 2392 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2393 2394 if (bserrno != 0) { 2395 /* The read failed, so jump to the final completion handler */ 2396 bs_sequence_finish(seq, bserrno); 2397 return; 2398 } 2399 2400 /* Write whole cluster */ 2401 bs_sequence_write_dev(seq, ctx->buf, 2402 bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster), 2403 bs_cluster_to_lba(ctx->blob->bs, 1), 2404 blob_write_copy_cpl, ctx); 2405 } 2406 2407 static void 2408 bs_allocate_and_copy_cluster(struct spdk_blob *blob, 2409 struct spdk_io_channel *_ch, 2410 uint64_t io_unit, spdk_bs_user_op_t *op) 2411 { 2412 struct spdk_bs_cpl cpl; 2413 struct spdk_bs_channel *ch; 2414 struct spdk_blob_copy_cluster_ctx *ctx; 2415 uint32_t cluster_start_page; 2416 uint32_t cluster_number; 2417 int rc; 2418 2419 ch = spdk_io_channel_get_ctx(_ch); 2420 2421 if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) { 2422 /* There are already operations pending. Queue this user op 2423 * and return because it will be re-executed when the outstanding 2424 * cluster allocation completes. */ 2425 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 2426 return; 2427 } 2428 2429 /* Round the io_unit offset down to the first page in the cluster */ 2430 cluster_start_page = bs_io_unit_to_cluster_start(blob, io_unit); 2431 2432 /* Calculate which index in the metadata cluster array the corresponding 2433 * cluster is supposed to be at. */ 2434 cluster_number = bs_io_unit_to_cluster_number(blob, io_unit); 2435 2436 ctx = calloc(1, sizeof(*ctx)); 2437 if (!ctx) { 2438 bs_user_op_abort(op); 2439 return; 2440 } 2441 2442 assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0); 2443 2444 ctx->blob = blob; 2445 ctx->page = cluster_start_page; 2446 2447 if (blob->parent_id != SPDK_BLOBID_INVALID) { 2448 ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, 2449 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 2450 if (!ctx->buf) { 2451 SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n", 2452 blob->bs->cluster_sz); 2453 free(ctx); 2454 bs_user_op_abort(op); 2455 return; 2456 } 2457 } 2458 2459 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 2460 rc = bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, &ctx->new_extent_page, 2461 false); 2462 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 2463 if (rc != 0) { 2464 spdk_free(ctx->buf); 2465 free(ctx); 2466 bs_user_op_abort(op); 2467 return; 2468 } 2469 2470 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2471 cpl.u.blob_basic.cb_fn = blob_allocate_and_copy_cluster_cpl; 2472 cpl.u.blob_basic.cb_arg = ctx; 2473 2474 ctx->seq = bs_sequence_start(_ch, &cpl); 2475 if (!ctx->seq) { 2476 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 2477 bs_release_cluster(blob->bs, ctx->new_cluster); 2478 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 2479 spdk_free(ctx->buf); 2480 free(ctx); 2481 bs_user_op_abort(op); 2482 return; 2483 } 2484 2485 /* Queue the user op to block other incoming operations */ 2486 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 2487 2488 if (blob->parent_id != SPDK_BLOBID_INVALID) { 2489 /* Read cluster from backing device */ 2490 bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf, 2491 bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page), 2492 bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz), 2493 blob_write_copy, ctx); 2494 } else { 2495 blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 2496 ctx->new_extent_page, blob_insert_cluster_cpl, ctx); 2497 } 2498 } 2499 2500 static inline bool 2501 blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length, 2502 uint64_t *lba, uint32_t *lba_count) 2503 { 2504 *lba_count = length; 2505 2506 if (!bs_io_unit_is_allocated(blob, io_unit)) { 2507 assert(blob->back_bs_dev != NULL); 2508 *lba = bs_io_unit_to_back_dev_lba(blob, io_unit); 2509 *lba_count = bs_io_unit_to_back_dev_lba(blob, *lba_count); 2510 return false; 2511 } else { 2512 *lba = bs_blob_io_unit_to_lba(blob, io_unit); 2513 return true; 2514 } 2515 } 2516 2517 struct op_split_ctx { 2518 struct spdk_blob *blob; 2519 struct spdk_io_channel *channel; 2520 uint64_t io_unit_offset; 2521 uint64_t io_units_remaining; 2522 void *curr_payload; 2523 enum spdk_blob_op_type op_type; 2524 spdk_bs_sequence_t *seq; 2525 }; 2526 2527 static void 2528 blob_request_submit_op_split_next(void *cb_arg, int bserrno) 2529 { 2530 struct op_split_ctx *ctx = cb_arg; 2531 struct spdk_blob *blob = ctx->blob; 2532 struct spdk_io_channel *ch = ctx->channel; 2533 enum spdk_blob_op_type op_type = ctx->op_type; 2534 uint8_t *buf = ctx->curr_payload; 2535 uint64_t offset = ctx->io_unit_offset; 2536 uint64_t length = ctx->io_units_remaining; 2537 uint64_t op_length; 2538 2539 if (bserrno != 0 || ctx->io_units_remaining == 0) { 2540 bs_sequence_finish(ctx->seq, bserrno); 2541 free(ctx); 2542 return; 2543 } 2544 2545 op_length = spdk_min(length, bs_num_io_units_to_cluster_boundary(blob, 2546 offset)); 2547 2548 /* Update length and payload for next operation */ 2549 ctx->io_units_remaining -= op_length; 2550 ctx->io_unit_offset += op_length; 2551 if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) { 2552 ctx->curr_payload += op_length * blob->bs->io_unit_size; 2553 } 2554 2555 switch (op_type) { 2556 case SPDK_BLOB_READ: 2557 spdk_blob_io_read(blob, ch, buf, offset, op_length, 2558 blob_request_submit_op_split_next, ctx); 2559 break; 2560 case SPDK_BLOB_WRITE: 2561 spdk_blob_io_write(blob, ch, buf, offset, op_length, 2562 blob_request_submit_op_split_next, ctx); 2563 break; 2564 case SPDK_BLOB_UNMAP: 2565 spdk_blob_io_unmap(blob, ch, offset, op_length, 2566 blob_request_submit_op_split_next, ctx); 2567 break; 2568 case SPDK_BLOB_WRITE_ZEROES: 2569 spdk_blob_io_write_zeroes(blob, ch, offset, op_length, 2570 blob_request_submit_op_split_next, ctx); 2571 break; 2572 case SPDK_BLOB_READV: 2573 case SPDK_BLOB_WRITEV: 2574 SPDK_ERRLOG("readv/write not valid\n"); 2575 bs_sequence_finish(ctx->seq, -EINVAL); 2576 free(ctx); 2577 break; 2578 } 2579 } 2580 2581 static void 2582 blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob, 2583 void *payload, uint64_t offset, uint64_t length, 2584 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 2585 { 2586 struct op_split_ctx *ctx; 2587 spdk_bs_sequence_t *seq; 2588 struct spdk_bs_cpl cpl; 2589 2590 assert(blob != NULL); 2591 2592 ctx = calloc(1, sizeof(struct op_split_ctx)); 2593 if (ctx == NULL) { 2594 cb_fn(cb_arg, -ENOMEM); 2595 return; 2596 } 2597 2598 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2599 cpl.u.blob_basic.cb_fn = cb_fn; 2600 cpl.u.blob_basic.cb_arg = cb_arg; 2601 2602 seq = bs_sequence_start(ch, &cpl); 2603 if (!seq) { 2604 free(ctx); 2605 cb_fn(cb_arg, -ENOMEM); 2606 return; 2607 } 2608 2609 ctx->blob = blob; 2610 ctx->channel = ch; 2611 ctx->curr_payload = payload; 2612 ctx->io_unit_offset = offset; 2613 ctx->io_units_remaining = length; 2614 ctx->op_type = op_type; 2615 ctx->seq = seq; 2616 2617 blob_request_submit_op_split_next(ctx, 0); 2618 } 2619 2620 static void 2621 blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob, 2622 void *payload, uint64_t offset, uint64_t length, 2623 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 2624 { 2625 struct spdk_bs_cpl cpl; 2626 uint64_t lba; 2627 uint32_t lba_count; 2628 bool is_allocated; 2629 2630 assert(blob != NULL); 2631 2632 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2633 cpl.u.blob_basic.cb_fn = cb_fn; 2634 cpl.u.blob_basic.cb_arg = cb_arg; 2635 2636 if (blob->frozen_refcnt) { 2637 /* This blob I/O is frozen */ 2638 spdk_bs_user_op_t *op; 2639 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch); 2640 2641 op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 2642 if (!op) { 2643 cb_fn(cb_arg, -ENOMEM); 2644 return; 2645 } 2646 2647 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 2648 2649 return; 2650 } 2651 2652 is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 2653 2654 switch (op_type) { 2655 case SPDK_BLOB_READ: { 2656 spdk_bs_batch_t *batch; 2657 2658 batch = bs_batch_open(_ch, &cpl); 2659 if (!batch) { 2660 cb_fn(cb_arg, -ENOMEM); 2661 return; 2662 } 2663 2664 if (is_allocated) { 2665 /* Read from the blob */ 2666 bs_batch_read_dev(batch, payload, lba, lba_count); 2667 } else { 2668 /* Read from the backing block device */ 2669 bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count); 2670 } 2671 2672 bs_batch_close(batch); 2673 break; 2674 } 2675 case SPDK_BLOB_WRITE: 2676 case SPDK_BLOB_WRITE_ZEROES: { 2677 if (is_allocated) { 2678 /* Write to the blob */ 2679 spdk_bs_batch_t *batch; 2680 2681 if (lba_count == 0) { 2682 cb_fn(cb_arg, 0); 2683 return; 2684 } 2685 2686 batch = bs_batch_open(_ch, &cpl); 2687 if (!batch) { 2688 cb_fn(cb_arg, -ENOMEM); 2689 return; 2690 } 2691 2692 if (op_type == SPDK_BLOB_WRITE) { 2693 bs_batch_write_dev(batch, payload, lba, lba_count); 2694 } else { 2695 bs_batch_write_zeroes_dev(batch, lba, lba_count); 2696 } 2697 2698 bs_batch_close(batch); 2699 } else { 2700 /* Queue this operation and allocate the cluster */ 2701 spdk_bs_user_op_t *op; 2702 2703 op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 2704 if (!op) { 2705 cb_fn(cb_arg, -ENOMEM); 2706 return; 2707 } 2708 2709 bs_allocate_and_copy_cluster(blob, _ch, offset, op); 2710 } 2711 break; 2712 } 2713 case SPDK_BLOB_UNMAP: { 2714 spdk_bs_batch_t *batch; 2715 2716 batch = bs_batch_open(_ch, &cpl); 2717 if (!batch) { 2718 cb_fn(cb_arg, -ENOMEM); 2719 return; 2720 } 2721 2722 if (is_allocated) { 2723 bs_batch_unmap_dev(batch, lba, lba_count); 2724 } 2725 2726 bs_batch_close(batch); 2727 break; 2728 } 2729 case SPDK_BLOB_READV: 2730 case SPDK_BLOB_WRITEV: 2731 SPDK_ERRLOG("readv/write not valid\n"); 2732 cb_fn(cb_arg, -EINVAL); 2733 break; 2734 } 2735 } 2736 2737 static void 2738 blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel, 2739 void *payload, uint64_t offset, uint64_t length, 2740 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 2741 { 2742 assert(blob != NULL); 2743 2744 if (blob->data_ro && op_type != SPDK_BLOB_READ) { 2745 cb_fn(cb_arg, -EPERM); 2746 return; 2747 } 2748 2749 if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) { 2750 cb_fn(cb_arg, -EINVAL); 2751 return; 2752 } 2753 if (length <= bs_num_io_units_to_cluster_boundary(blob, offset)) { 2754 blob_request_submit_op_single(_channel, blob, payload, offset, length, 2755 cb_fn, cb_arg, op_type); 2756 } else { 2757 blob_request_submit_op_split(_channel, blob, payload, offset, length, 2758 cb_fn, cb_arg, op_type); 2759 } 2760 } 2761 2762 struct rw_iov_ctx { 2763 struct spdk_blob *blob; 2764 struct spdk_io_channel *channel; 2765 spdk_blob_op_complete cb_fn; 2766 void *cb_arg; 2767 bool read; 2768 int iovcnt; 2769 struct iovec *orig_iov; 2770 uint64_t io_unit_offset; 2771 uint64_t io_units_remaining; 2772 uint64_t io_units_done; 2773 struct iovec iov[0]; 2774 }; 2775 2776 static void 2777 rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2778 { 2779 assert(cb_arg == NULL); 2780 bs_sequence_finish(seq, bserrno); 2781 } 2782 2783 static void 2784 rw_iov_split_next(void *cb_arg, int bserrno) 2785 { 2786 struct rw_iov_ctx *ctx = cb_arg; 2787 struct spdk_blob *blob = ctx->blob; 2788 struct iovec *iov, *orig_iov; 2789 int iovcnt; 2790 size_t orig_iovoff; 2791 uint64_t io_units_count, io_units_to_boundary, io_unit_offset; 2792 uint64_t byte_count; 2793 2794 if (bserrno != 0 || ctx->io_units_remaining == 0) { 2795 ctx->cb_fn(ctx->cb_arg, bserrno); 2796 free(ctx); 2797 return; 2798 } 2799 2800 io_unit_offset = ctx->io_unit_offset; 2801 io_units_to_boundary = bs_num_io_units_to_cluster_boundary(blob, io_unit_offset); 2802 io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary); 2803 /* 2804 * Get index and offset into the original iov array for our current position in the I/O sequence. 2805 * byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will 2806 * point to the current position in the I/O sequence. 2807 */ 2808 byte_count = ctx->io_units_done * blob->bs->io_unit_size; 2809 orig_iov = &ctx->orig_iov[0]; 2810 orig_iovoff = 0; 2811 while (byte_count > 0) { 2812 if (byte_count >= orig_iov->iov_len) { 2813 byte_count -= orig_iov->iov_len; 2814 orig_iov++; 2815 } else { 2816 orig_iovoff = byte_count; 2817 byte_count = 0; 2818 } 2819 } 2820 2821 /* 2822 * Build an iov array for the next I/O in the sequence. byte_count will keep track of how many 2823 * bytes of this next I/O remain to be accounted for in the new iov array. 2824 */ 2825 byte_count = io_units_count * blob->bs->io_unit_size; 2826 iov = &ctx->iov[0]; 2827 iovcnt = 0; 2828 while (byte_count > 0) { 2829 assert(iovcnt < ctx->iovcnt); 2830 iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff); 2831 iov->iov_base = orig_iov->iov_base + orig_iovoff; 2832 byte_count -= iov->iov_len; 2833 orig_iovoff = 0; 2834 orig_iov++; 2835 iov++; 2836 iovcnt++; 2837 } 2838 2839 ctx->io_unit_offset += io_units_count; 2840 ctx->io_units_remaining -= io_units_count; 2841 ctx->io_units_done += io_units_count; 2842 iov = &ctx->iov[0]; 2843 2844 if (ctx->read) { 2845 spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset, 2846 io_units_count, rw_iov_split_next, ctx); 2847 } else { 2848 spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset, 2849 io_units_count, rw_iov_split_next, ctx); 2850 } 2851 } 2852 2853 static void 2854 blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel, 2855 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 2856 spdk_blob_op_complete cb_fn, void *cb_arg, bool read) 2857 { 2858 struct spdk_bs_cpl cpl; 2859 2860 assert(blob != NULL); 2861 2862 if (!read && blob->data_ro) { 2863 cb_fn(cb_arg, -EPERM); 2864 return; 2865 } 2866 2867 if (length == 0) { 2868 cb_fn(cb_arg, 0); 2869 return; 2870 } 2871 2872 if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) { 2873 cb_fn(cb_arg, -EINVAL); 2874 return; 2875 } 2876 2877 /* 2878 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having 2879 * to split a request that spans a cluster boundary. For I/O that do not span a cluster boundary, 2880 * there will be no noticeable difference compared to using a batch. For I/O that do span a cluster 2881 * boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need 2882 * to allocate a separate iov array and split the I/O such that none of the resulting 2883 * smaller I/O cross a cluster boundary. These smaller I/O will be issued in sequence (not in parallel) 2884 * but since this case happens very infrequently, any performance impact will be negligible. 2885 * 2886 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs 2887 * for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them 2888 * in a batch. That would also require creating an intermediate spdk_bs_cpl that would get called 2889 * when the batch was completed, to allow for freeing the memory for the iov arrays. 2890 */ 2891 if (spdk_likely(length <= bs_num_io_units_to_cluster_boundary(blob, offset))) { 2892 uint32_t lba_count; 2893 uint64_t lba; 2894 bool is_allocated; 2895 2896 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2897 cpl.u.blob_basic.cb_fn = cb_fn; 2898 cpl.u.blob_basic.cb_arg = cb_arg; 2899 2900 if (blob->frozen_refcnt) { 2901 /* This blob I/O is frozen */ 2902 enum spdk_blob_op_type op_type; 2903 spdk_bs_user_op_t *op; 2904 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel); 2905 2906 op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV; 2907 op = bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length); 2908 if (!op) { 2909 cb_fn(cb_arg, -ENOMEM); 2910 return; 2911 } 2912 2913 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 2914 2915 return; 2916 } 2917 2918 is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 2919 2920 if (read) { 2921 spdk_bs_sequence_t *seq; 2922 2923 seq = bs_sequence_start(_channel, &cpl); 2924 if (!seq) { 2925 cb_fn(cb_arg, -ENOMEM); 2926 return; 2927 } 2928 2929 if (is_allocated) { 2930 bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL); 2931 } else { 2932 bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count, 2933 rw_iov_done, NULL); 2934 } 2935 } else { 2936 if (is_allocated) { 2937 spdk_bs_sequence_t *seq; 2938 2939 seq = bs_sequence_start(_channel, &cpl); 2940 if (!seq) { 2941 cb_fn(cb_arg, -ENOMEM); 2942 return; 2943 } 2944 2945 bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL); 2946 } else { 2947 /* Queue this operation and allocate the cluster */ 2948 spdk_bs_user_op_t *op; 2949 2950 op = bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, 2951 length); 2952 if (!op) { 2953 cb_fn(cb_arg, -ENOMEM); 2954 return; 2955 } 2956 2957 bs_allocate_and_copy_cluster(blob, _channel, offset, op); 2958 } 2959 } 2960 } else { 2961 struct rw_iov_ctx *ctx; 2962 2963 ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec)); 2964 if (ctx == NULL) { 2965 cb_fn(cb_arg, -ENOMEM); 2966 return; 2967 } 2968 2969 ctx->blob = blob; 2970 ctx->channel = _channel; 2971 ctx->cb_fn = cb_fn; 2972 ctx->cb_arg = cb_arg; 2973 ctx->read = read; 2974 ctx->orig_iov = iov; 2975 ctx->iovcnt = iovcnt; 2976 ctx->io_unit_offset = offset; 2977 ctx->io_units_remaining = length; 2978 ctx->io_units_done = 0; 2979 2980 rw_iov_split_next(ctx, 0); 2981 } 2982 } 2983 2984 static struct spdk_blob * 2985 blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid) 2986 { 2987 struct spdk_blob *blob; 2988 2989 if (spdk_bit_array_get(bs->open_blobids, blobid) == 0) { 2990 return NULL; 2991 } 2992 2993 TAILQ_FOREACH(blob, &bs->blobs, link) { 2994 if (blob->id == blobid) { 2995 return blob; 2996 } 2997 } 2998 2999 return NULL; 3000 } 3001 3002 static void 3003 blob_get_snapshot_and_clone_entries(struct spdk_blob *blob, 3004 struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry) 3005 { 3006 assert(blob != NULL); 3007 *snapshot_entry = NULL; 3008 *clone_entry = NULL; 3009 3010 if (blob->parent_id == SPDK_BLOBID_INVALID) { 3011 return; 3012 } 3013 3014 TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) { 3015 if ((*snapshot_entry)->id == blob->parent_id) { 3016 break; 3017 } 3018 } 3019 3020 if (*snapshot_entry != NULL) { 3021 TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) { 3022 if ((*clone_entry)->id == blob->id) { 3023 break; 3024 } 3025 } 3026 3027 assert(clone_entry != NULL); 3028 } 3029 } 3030 3031 static int 3032 bs_channel_create(void *io_device, void *ctx_buf) 3033 { 3034 struct spdk_blob_store *bs = io_device; 3035 struct spdk_bs_channel *channel = ctx_buf; 3036 struct spdk_bs_dev *dev; 3037 uint32_t max_ops = bs->max_channel_ops; 3038 uint32_t i; 3039 3040 dev = bs->dev; 3041 3042 channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set)); 3043 if (!channel->req_mem) { 3044 return -1; 3045 } 3046 3047 TAILQ_INIT(&channel->reqs); 3048 3049 for (i = 0; i < max_ops; i++) { 3050 TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link); 3051 } 3052 3053 channel->bs = bs; 3054 channel->dev = dev; 3055 channel->dev_channel = dev->create_channel(dev); 3056 3057 if (!channel->dev_channel) { 3058 SPDK_ERRLOG("Failed to create device channel.\n"); 3059 free(channel->req_mem); 3060 return -1; 3061 } 3062 3063 TAILQ_INIT(&channel->need_cluster_alloc); 3064 TAILQ_INIT(&channel->queued_io); 3065 3066 return 0; 3067 } 3068 3069 static void 3070 bs_channel_destroy(void *io_device, void *ctx_buf) 3071 { 3072 struct spdk_bs_channel *channel = ctx_buf; 3073 spdk_bs_user_op_t *op; 3074 3075 while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) { 3076 op = TAILQ_FIRST(&channel->need_cluster_alloc); 3077 TAILQ_REMOVE(&channel->need_cluster_alloc, op, link); 3078 bs_user_op_abort(op); 3079 } 3080 3081 while (!TAILQ_EMPTY(&channel->queued_io)) { 3082 op = TAILQ_FIRST(&channel->queued_io); 3083 TAILQ_REMOVE(&channel->queued_io, op, link); 3084 bs_user_op_abort(op); 3085 } 3086 3087 free(channel->req_mem); 3088 channel->dev->destroy_channel(channel->dev, channel->dev_channel); 3089 } 3090 3091 static void 3092 bs_dev_destroy(void *io_device) 3093 { 3094 struct spdk_blob_store *bs = io_device; 3095 struct spdk_blob *blob, *blob_tmp; 3096 3097 bs->dev->destroy(bs->dev); 3098 3099 TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) { 3100 TAILQ_REMOVE(&bs->blobs, blob, link); 3101 spdk_bit_array_clear(bs->open_blobids, blob->id); 3102 blob_free(blob); 3103 } 3104 3105 pthread_mutex_destroy(&bs->used_clusters_mutex); 3106 3107 spdk_bit_array_free(&bs->open_blobids); 3108 spdk_bit_array_free(&bs->used_blobids); 3109 spdk_bit_array_free(&bs->used_md_pages); 3110 spdk_bit_pool_free(&bs->used_clusters); 3111 /* 3112 * If this function is called for any reason except a successful unload, 3113 * the unload_cpl type will be NONE and this will be a nop. 3114 */ 3115 bs_call_cpl(&bs->unload_cpl, bs->unload_err); 3116 3117 free(bs); 3118 } 3119 3120 static int 3121 bs_blob_list_add(struct spdk_blob *blob) 3122 { 3123 spdk_blob_id snapshot_id; 3124 struct spdk_blob_list *snapshot_entry = NULL; 3125 struct spdk_blob_list *clone_entry = NULL; 3126 3127 assert(blob != NULL); 3128 3129 snapshot_id = blob->parent_id; 3130 if (snapshot_id == SPDK_BLOBID_INVALID) { 3131 return 0; 3132 } 3133 3134 snapshot_entry = bs_get_snapshot_entry(blob->bs, snapshot_id); 3135 if (snapshot_entry == NULL) { 3136 /* Snapshot not found */ 3137 snapshot_entry = calloc(1, sizeof(struct spdk_blob_list)); 3138 if (snapshot_entry == NULL) { 3139 return -ENOMEM; 3140 } 3141 snapshot_entry->id = snapshot_id; 3142 TAILQ_INIT(&snapshot_entry->clones); 3143 TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link); 3144 } else { 3145 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 3146 if (clone_entry->id == blob->id) { 3147 break; 3148 } 3149 } 3150 } 3151 3152 if (clone_entry == NULL) { 3153 /* Clone not found */ 3154 clone_entry = calloc(1, sizeof(struct spdk_blob_list)); 3155 if (clone_entry == NULL) { 3156 return -ENOMEM; 3157 } 3158 clone_entry->id = blob->id; 3159 TAILQ_INIT(&clone_entry->clones); 3160 TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link); 3161 snapshot_entry->clone_count++; 3162 } 3163 3164 return 0; 3165 } 3166 3167 static void 3168 bs_blob_list_remove(struct spdk_blob *blob) 3169 { 3170 struct spdk_blob_list *snapshot_entry = NULL; 3171 struct spdk_blob_list *clone_entry = NULL; 3172 3173 blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry); 3174 3175 if (snapshot_entry == NULL) { 3176 return; 3177 } 3178 3179 blob->parent_id = SPDK_BLOBID_INVALID; 3180 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 3181 free(clone_entry); 3182 3183 snapshot_entry->clone_count--; 3184 } 3185 3186 static int 3187 bs_blob_list_free(struct spdk_blob_store *bs) 3188 { 3189 struct spdk_blob_list *snapshot_entry; 3190 struct spdk_blob_list *snapshot_entry_tmp; 3191 struct spdk_blob_list *clone_entry; 3192 struct spdk_blob_list *clone_entry_tmp; 3193 3194 TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) { 3195 TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) { 3196 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 3197 free(clone_entry); 3198 } 3199 TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link); 3200 free(snapshot_entry); 3201 } 3202 3203 return 0; 3204 } 3205 3206 static void 3207 bs_free(struct spdk_blob_store *bs) 3208 { 3209 bs_blob_list_free(bs); 3210 3211 bs_unregister_md_thread(bs); 3212 spdk_io_device_unregister(bs, bs_dev_destroy); 3213 } 3214 3215 void 3216 spdk_bs_opts_init(struct spdk_bs_opts *opts, size_t opts_size) 3217 { 3218 3219 if (!opts) { 3220 SPDK_ERRLOG("opts should not be NULL\n"); 3221 return; 3222 } 3223 3224 if (!opts_size) { 3225 SPDK_ERRLOG("opts_size should not be zero value\n"); 3226 return; 3227 } 3228 3229 memset(opts, 0, opts_size); 3230 opts->opts_size = opts_size; 3231 3232 #define FIELD_OK(field) \ 3233 offsetof(struct spdk_bs_opts, field) + sizeof(opts->field) <= opts_size 3234 3235 #define SET_FIELD(field, value) \ 3236 if (FIELD_OK(field)) { \ 3237 opts->field = value; \ 3238 } \ 3239 3240 SET_FIELD(cluster_sz, SPDK_BLOB_OPTS_CLUSTER_SZ); 3241 SET_FIELD(num_md_pages, SPDK_BLOB_OPTS_NUM_MD_PAGES); 3242 SET_FIELD(max_md_ops, SPDK_BLOB_OPTS_NUM_MD_PAGES); 3243 SET_FIELD(max_channel_ops, SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS); 3244 SET_FIELD(clear_method, BS_CLEAR_WITH_UNMAP); 3245 3246 if (FIELD_OK(bstype)) { 3247 memset(&opts->bstype, 0, sizeof(opts->bstype)); 3248 } 3249 3250 SET_FIELD(iter_cb_fn, NULL); 3251 SET_FIELD(iter_cb_arg, NULL); 3252 3253 #undef FIELD_OK 3254 #undef SET_FIELD 3255 } 3256 3257 static int 3258 bs_opts_verify(struct spdk_bs_opts *opts) 3259 { 3260 if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 || 3261 opts->max_channel_ops == 0) { 3262 SPDK_ERRLOG("Blobstore options cannot be set to 0\n"); 3263 return -1; 3264 } 3265 3266 return 0; 3267 } 3268 3269 /* START spdk_bs_load */ 3270 3271 /* spdk_bs_load_ctx is used for init, load, unload and dump code paths. */ 3272 3273 struct spdk_bs_load_ctx { 3274 struct spdk_blob_store *bs; 3275 struct spdk_bs_super_block *super; 3276 3277 struct spdk_bs_md_mask *mask; 3278 bool in_page_chain; 3279 uint32_t page_index; 3280 uint32_t cur_page; 3281 struct spdk_blob_md_page *page; 3282 3283 uint64_t num_extent_pages; 3284 uint32_t *extent_page_num; 3285 struct spdk_blob_md_page *extent_pages; 3286 struct spdk_bit_array *used_clusters; 3287 3288 spdk_bs_sequence_t *seq; 3289 spdk_blob_op_with_handle_complete iter_cb_fn; 3290 void *iter_cb_arg; 3291 struct spdk_blob *blob; 3292 spdk_blob_id blobid; 3293 3294 /* These fields are used in the spdk_bs_dump path. */ 3295 FILE *fp; 3296 spdk_bs_dump_print_xattr print_xattr_fn; 3297 char xattr_name[4096]; 3298 }; 3299 3300 static int 3301 bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs, 3302 struct spdk_bs_load_ctx **_ctx) 3303 { 3304 struct spdk_blob_store *bs; 3305 struct spdk_bs_load_ctx *ctx; 3306 uint64_t dev_size; 3307 int rc; 3308 3309 dev_size = dev->blocklen * dev->blockcnt; 3310 if (dev_size < opts->cluster_sz) { 3311 /* Device size cannot be smaller than cluster size of blobstore */ 3312 SPDK_INFOLOG(blob, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n", 3313 dev_size, opts->cluster_sz); 3314 return -ENOSPC; 3315 } 3316 if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) { 3317 /* Cluster size cannot be smaller than page size */ 3318 SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n", 3319 opts->cluster_sz, SPDK_BS_PAGE_SIZE); 3320 return -EINVAL; 3321 } 3322 bs = calloc(1, sizeof(struct spdk_blob_store)); 3323 if (!bs) { 3324 return -ENOMEM; 3325 } 3326 3327 ctx = calloc(1, sizeof(struct spdk_bs_load_ctx)); 3328 if (!ctx) { 3329 free(bs); 3330 return -ENOMEM; 3331 } 3332 3333 ctx->bs = bs; 3334 ctx->iter_cb_fn = opts->iter_cb_fn; 3335 ctx->iter_cb_arg = opts->iter_cb_arg; 3336 3337 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 3338 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3339 if (!ctx->super) { 3340 free(ctx); 3341 free(bs); 3342 return -ENOMEM; 3343 } 3344 3345 TAILQ_INIT(&bs->blobs); 3346 TAILQ_INIT(&bs->snapshots); 3347 bs->dev = dev; 3348 bs->md_thread = spdk_get_thread(); 3349 assert(bs->md_thread != NULL); 3350 3351 /* 3352 * Do not use bs_lba_to_cluster() here since blockcnt may not be an 3353 * even multiple of the cluster size. 3354 */ 3355 bs->cluster_sz = opts->cluster_sz; 3356 bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen); 3357 ctx->used_clusters = spdk_bit_array_create(bs->total_clusters); 3358 if (!ctx->used_clusters) { 3359 spdk_free(ctx->super); 3360 free(ctx); 3361 free(bs); 3362 return -ENOMEM; 3363 } 3364 3365 bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE; 3366 if (spdk_u32_is_pow2(bs->pages_per_cluster)) { 3367 bs->pages_per_cluster_shift = spdk_u32log2(bs->pages_per_cluster); 3368 } 3369 bs->num_free_clusters = bs->total_clusters; 3370 bs->io_unit_size = dev->blocklen; 3371 3372 bs->max_channel_ops = opts->max_channel_ops; 3373 bs->super_blob = SPDK_BLOBID_INVALID; 3374 memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype)); 3375 3376 /* The metadata is assumed to be at least 1 page */ 3377 bs->used_md_pages = spdk_bit_array_create(1); 3378 bs->used_blobids = spdk_bit_array_create(0); 3379 bs->open_blobids = spdk_bit_array_create(0); 3380 3381 pthread_mutex_init(&bs->used_clusters_mutex, NULL); 3382 3383 spdk_io_device_register(bs, bs_channel_create, bs_channel_destroy, 3384 sizeof(struct spdk_bs_channel), "blobstore"); 3385 rc = bs_register_md_thread(bs); 3386 if (rc == -1) { 3387 spdk_io_device_unregister(bs, NULL); 3388 pthread_mutex_destroy(&bs->used_clusters_mutex); 3389 spdk_bit_array_free(&bs->open_blobids); 3390 spdk_bit_array_free(&bs->used_blobids); 3391 spdk_bit_array_free(&bs->used_md_pages); 3392 spdk_bit_array_free(&ctx->used_clusters); 3393 spdk_free(ctx->super); 3394 free(ctx); 3395 free(bs); 3396 /* FIXME: this is a lie but don't know how to get a proper error code here */ 3397 return -ENOMEM; 3398 } 3399 3400 *_ctx = ctx; 3401 *_bs = bs; 3402 return 0; 3403 } 3404 3405 static void 3406 bs_load_ctx_fail(struct spdk_bs_load_ctx *ctx, int bserrno) 3407 { 3408 assert(bserrno != 0); 3409 3410 spdk_free(ctx->super); 3411 bs_sequence_finish(ctx->seq, bserrno); 3412 bs_free(ctx->bs); 3413 spdk_bit_array_free(&ctx->used_clusters); 3414 free(ctx); 3415 } 3416 3417 static void 3418 bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 3419 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg) 3420 { 3421 /* Update the values in the super block */ 3422 super->super_blob = bs->super_blob; 3423 memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype)); 3424 super->crc = blob_md_page_calc_crc(super); 3425 bs_sequence_write_dev(seq, super, bs_page_to_lba(bs, 0), 3426 bs_byte_to_lba(bs, sizeof(*super)), 3427 cb_fn, cb_arg); 3428 } 3429 3430 static void 3431 bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 3432 { 3433 struct spdk_bs_load_ctx *ctx = arg; 3434 uint64_t mask_size, lba, lba_count; 3435 3436 /* Write out the used clusters mask */ 3437 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 3438 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 3439 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3440 if (!ctx->mask) { 3441 bs_load_ctx_fail(ctx, -ENOMEM); 3442 return; 3443 } 3444 3445 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS; 3446 ctx->mask->length = ctx->bs->total_clusters; 3447 /* We could get here through the normal unload path, or through dirty 3448 * shutdown recovery. For the normal unload path, we use the mask from 3449 * the bit pool. For dirty shutdown recovery, we don't have a bit pool yet - 3450 * only the bit array from the load ctx. 3451 */ 3452 if (ctx->bs->used_clusters) { 3453 assert(ctx->mask->length == spdk_bit_pool_capacity(ctx->bs->used_clusters)); 3454 spdk_bit_pool_store_mask(ctx->bs->used_clusters, ctx->mask->mask); 3455 } else { 3456 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->used_clusters)); 3457 spdk_bit_array_store_mask(ctx->used_clusters, ctx->mask->mask); 3458 } 3459 lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 3460 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 3461 bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 3462 } 3463 3464 static void 3465 bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 3466 { 3467 struct spdk_bs_load_ctx *ctx = arg; 3468 uint64_t mask_size, lba, lba_count; 3469 3470 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 3471 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 3472 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3473 if (!ctx->mask) { 3474 bs_load_ctx_fail(ctx, -ENOMEM); 3475 return; 3476 } 3477 3478 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES; 3479 ctx->mask->length = ctx->super->md_len; 3480 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages)); 3481 3482 spdk_bit_array_store_mask(ctx->bs->used_md_pages, ctx->mask->mask); 3483 lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 3484 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 3485 bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 3486 } 3487 3488 static void 3489 bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 3490 { 3491 struct spdk_bs_load_ctx *ctx = arg; 3492 uint64_t mask_size, lba, lba_count; 3493 3494 if (ctx->super->used_blobid_mask_len == 0) { 3495 /* 3496 * This is a pre-v3 on-disk format where the blobid mask does not get 3497 * written to disk. 3498 */ 3499 cb_fn(seq, arg, 0); 3500 return; 3501 } 3502 3503 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 3504 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 3505 SPDK_MALLOC_DMA); 3506 if (!ctx->mask) { 3507 bs_load_ctx_fail(ctx, -ENOMEM); 3508 return; 3509 } 3510 3511 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS; 3512 ctx->mask->length = ctx->super->md_len; 3513 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids)); 3514 3515 spdk_bit_array_store_mask(ctx->bs->used_blobids, ctx->mask->mask); 3516 lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 3517 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 3518 bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 3519 } 3520 3521 static void 3522 blob_set_thin_provision(struct spdk_blob *blob) 3523 { 3524 blob_verify_md_op(blob); 3525 blob->invalid_flags |= SPDK_BLOB_THIN_PROV; 3526 blob->state = SPDK_BLOB_STATE_DIRTY; 3527 } 3528 3529 static void 3530 blob_set_clear_method(struct spdk_blob *blob, enum blob_clear_method clear_method) 3531 { 3532 blob_verify_md_op(blob); 3533 blob->clear_method = clear_method; 3534 blob->md_ro_flags |= (clear_method << SPDK_BLOB_CLEAR_METHOD_SHIFT); 3535 blob->state = SPDK_BLOB_STATE_DIRTY; 3536 } 3537 3538 static void bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno); 3539 3540 static void 3541 bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno) 3542 { 3543 struct spdk_bs_load_ctx *ctx = cb_arg; 3544 spdk_blob_id id; 3545 int64_t page_num; 3546 3547 /* Iterate to next blob (we can't use spdk_bs_iter_next function as our 3548 * last blob has been removed */ 3549 page_num = bs_blobid_to_page(ctx->blobid); 3550 page_num++; 3551 page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num); 3552 if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) { 3553 bs_load_iter(ctx, NULL, -ENOENT); 3554 return; 3555 } 3556 3557 id = bs_page_to_blobid(page_num); 3558 3559 spdk_bs_open_blob(ctx->bs, id, bs_load_iter, ctx); 3560 } 3561 3562 static void 3563 bs_delete_corrupted_close_cb(void *cb_arg, int bserrno) 3564 { 3565 struct spdk_bs_load_ctx *ctx = cb_arg; 3566 3567 if (bserrno != 0) { 3568 SPDK_ERRLOG("Failed to close corrupted blob\n"); 3569 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3570 return; 3571 } 3572 3573 spdk_bs_delete_blob(ctx->bs, ctx->blobid, bs_delete_corrupted_blob_cpl, ctx); 3574 } 3575 3576 static void 3577 bs_delete_corrupted_blob(void *cb_arg, int bserrno) 3578 { 3579 struct spdk_bs_load_ctx *ctx = cb_arg; 3580 uint64_t i; 3581 3582 if (bserrno != 0) { 3583 SPDK_ERRLOG("Failed to close clone of a corrupted blob\n"); 3584 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3585 return; 3586 } 3587 3588 /* Snapshot and clone have the same copy of cluster map and extent pages 3589 * at this point. Let's clear both for snpashot now, 3590 * so that it won't be cleared for clone later when we remove snapshot. 3591 * Also set thin provision to pass data corruption check */ 3592 for (i = 0; i < ctx->blob->active.num_clusters; i++) { 3593 ctx->blob->active.clusters[i] = 0; 3594 } 3595 for (i = 0; i < ctx->blob->active.num_extent_pages; i++) { 3596 ctx->blob->active.extent_pages[i] = 0; 3597 } 3598 3599 ctx->blob->md_ro = false; 3600 3601 blob_set_thin_provision(ctx->blob); 3602 3603 ctx->blobid = ctx->blob->id; 3604 3605 spdk_blob_close(ctx->blob, bs_delete_corrupted_close_cb, ctx); 3606 } 3607 3608 static void 3609 bs_update_corrupted_blob(void *cb_arg, int bserrno) 3610 { 3611 struct spdk_bs_load_ctx *ctx = cb_arg; 3612 3613 if (bserrno != 0) { 3614 SPDK_ERRLOG("Failed to close clone of a corrupted blob\n"); 3615 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3616 return; 3617 } 3618 3619 ctx->blob->md_ro = false; 3620 blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true); 3621 blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true); 3622 spdk_blob_set_read_only(ctx->blob); 3623 3624 if (ctx->iter_cb_fn) { 3625 ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0); 3626 } 3627 bs_blob_list_add(ctx->blob); 3628 3629 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3630 } 3631 3632 static void 3633 bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno) 3634 { 3635 struct spdk_bs_load_ctx *ctx = cb_arg; 3636 3637 if (bserrno != 0) { 3638 SPDK_ERRLOG("Failed to open clone of a corrupted blob\n"); 3639 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3640 return; 3641 } 3642 3643 if (blob->parent_id == ctx->blob->id) { 3644 /* Power failure occured before updating clone (snapshot delete case) 3645 * or after updating clone (creating snapshot case) - keep snapshot */ 3646 spdk_blob_close(blob, bs_update_corrupted_blob, ctx); 3647 } else { 3648 /* Power failure occured after updating clone (snapshot delete case) 3649 * or before updating clone (creating snapshot case) - remove snapshot */ 3650 spdk_blob_close(blob, bs_delete_corrupted_blob, ctx); 3651 } 3652 } 3653 3654 static void 3655 bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno) 3656 { 3657 struct spdk_bs_load_ctx *ctx = arg; 3658 const void *value; 3659 size_t len; 3660 int rc = 0; 3661 3662 if (bserrno == 0) { 3663 /* Examine blob if it is corrupted after power failure. Fix 3664 * the ones that can be fixed and remove any other corrupted 3665 * ones. If it is not corrupted just process it */ 3666 rc = blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true); 3667 if (rc != 0) { 3668 rc = blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true); 3669 if (rc != 0) { 3670 /* Not corrupted - process it and continue with iterating through blobs */ 3671 if (ctx->iter_cb_fn) { 3672 ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0); 3673 } 3674 bs_blob_list_add(blob); 3675 spdk_bs_iter_next(ctx->bs, blob, bs_load_iter, ctx); 3676 return; 3677 } 3678 3679 } 3680 3681 assert(len == sizeof(spdk_blob_id)); 3682 3683 ctx->blob = blob; 3684 3685 /* Open clone to check if we are able to fix this blob or should we remove it */ 3686 spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, bs_examine_clone, ctx); 3687 return; 3688 } else if (bserrno == -ENOENT) { 3689 bserrno = 0; 3690 } else { 3691 /* 3692 * This case needs to be looked at further. Same problem 3693 * exists with applications that rely on explicit blob 3694 * iteration. We should just skip the blob that failed 3695 * to load and continue on to the next one. 3696 */ 3697 SPDK_ERRLOG("Error in iterating blobs\n"); 3698 } 3699 3700 ctx->iter_cb_fn = NULL; 3701 3702 spdk_free(ctx->super); 3703 spdk_free(ctx->mask); 3704 bs_sequence_finish(ctx->seq, bserrno); 3705 free(ctx); 3706 } 3707 3708 static void 3709 bs_load_complete(struct spdk_bs_load_ctx *ctx) 3710 { 3711 ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters); 3712 spdk_bs_iter_first(ctx->bs, bs_load_iter, ctx); 3713 } 3714 3715 static void 3716 bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3717 { 3718 struct spdk_bs_load_ctx *ctx = cb_arg; 3719 int rc; 3720 3721 /* The type must be correct */ 3722 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS); 3723 3724 /* The length of the mask (in bits) must not be greater than 3725 * the length of the buffer (converted to bits) */ 3726 assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8)); 3727 3728 /* The length of the mask must be exactly equal to the size 3729 * (in pages) of the metadata region */ 3730 assert(ctx->mask->length == ctx->super->md_len); 3731 3732 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->mask->length); 3733 if (rc < 0) { 3734 spdk_free(ctx->mask); 3735 bs_load_ctx_fail(ctx, rc); 3736 return; 3737 } 3738 3739 spdk_bit_array_load_mask(ctx->bs->used_blobids, ctx->mask->mask); 3740 bs_load_complete(ctx); 3741 } 3742 3743 static void 3744 bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3745 { 3746 struct spdk_bs_load_ctx *ctx = cb_arg; 3747 uint64_t lba, lba_count, mask_size; 3748 int rc; 3749 3750 if (bserrno != 0) { 3751 bs_load_ctx_fail(ctx, bserrno); 3752 return; 3753 } 3754 3755 /* The type must be correct */ 3756 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS); 3757 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 3758 assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof( 3759 struct spdk_blob_md_page) * 8)); 3760 /* The length of the mask must be exactly equal to the total number of clusters */ 3761 assert(ctx->mask->length == ctx->bs->total_clusters); 3762 3763 rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->mask->length); 3764 if (rc < 0) { 3765 spdk_free(ctx->mask); 3766 bs_load_ctx_fail(ctx, rc); 3767 return; 3768 } 3769 3770 spdk_bit_array_load_mask(ctx->used_clusters, ctx->mask->mask); 3771 ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->used_clusters); 3772 assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters); 3773 3774 spdk_free(ctx->mask); 3775 3776 /* Read the used blobids mask */ 3777 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 3778 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 3779 SPDK_MALLOC_DMA); 3780 if (!ctx->mask) { 3781 bs_load_ctx_fail(ctx, -ENOMEM); 3782 return; 3783 } 3784 lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 3785 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 3786 bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 3787 bs_load_used_blobids_cpl, ctx); 3788 } 3789 3790 static void 3791 bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3792 { 3793 struct spdk_bs_load_ctx *ctx = cb_arg; 3794 uint64_t lba, lba_count, mask_size; 3795 int rc; 3796 3797 if (bserrno != 0) { 3798 bs_load_ctx_fail(ctx, bserrno); 3799 return; 3800 } 3801 3802 /* The type must be correct */ 3803 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES); 3804 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 3805 assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE * 3806 8)); 3807 /* The length of the mask must be exactly equal to the size (in pages) of the metadata region */ 3808 if (ctx->mask->length != ctx->super->md_len) { 3809 SPDK_ERRLOG("mismatched md_len in used_pages mask: " 3810 "mask->length=%" PRIu32 " super->md_len=%" PRIu32 "\n", 3811 ctx->mask->length, ctx->super->md_len); 3812 assert(false); 3813 } 3814 3815 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->mask->length); 3816 if (rc < 0) { 3817 spdk_free(ctx->mask); 3818 bs_load_ctx_fail(ctx, rc); 3819 return; 3820 } 3821 3822 spdk_bit_array_load_mask(ctx->bs->used_md_pages, ctx->mask->mask); 3823 spdk_free(ctx->mask); 3824 3825 /* Read the used clusters mask */ 3826 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 3827 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 3828 SPDK_MALLOC_DMA); 3829 if (!ctx->mask) { 3830 bs_load_ctx_fail(ctx, -ENOMEM); 3831 return; 3832 } 3833 lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 3834 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 3835 bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 3836 bs_load_used_clusters_cpl, ctx); 3837 } 3838 3839 static void 3840 bs_load_read_used_pages(struct spdk_bs_load_ctx *ctx) 3841 { 3842 uint64_t lba, lba_count, mask_size; 3843 3844 /* Read the used pages mask */ 3845 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 3846 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 3847 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3848 if (!ctx->mask) { 3849 bs_load_ctx_fail(ctx, -ENOMEM); 3850 return; 3851 } 3852 3853 lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 3854 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 3855 bs_sequence_read_dev(ctx->seq, ctx->mask, lba, lba_count, 3856 bs_load_used_pages_cpl, ctx); 3857 } 3858 3859 static int 3860 bs_load_replay_md_parse_page(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_page *page) 3861 { 3862 struct spdk_blob_store *bs = ctx->bs; 3863 struct spdk_blob_md_descriptor *desc; 3864 size_t cur_desc = 0; 3865 3866 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 3867 while (cur_desc < sizeof(page->descriptors)) { 3868 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 3869 if (desc->length == 0) { 3870 /* If padding and length are 0, this terminates the page */ 3871 break; 3872 } 3873 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) { 3874 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 3875 unsigned int i, j; 3876 unsigned int cluster_count = 0; 3877 uint32_t cluster_idx; 3878 3879 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc; 3880 3881 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 3882 for (j = 0; j < desc_extent_rle->extents[i].length; j++) { 3883 cluster_idx = desc_extent_rle->extents[i].cluster_idx; 3884 /* 3885 * cluster_idx = 0 means an unallocated cluster - don't mark that 3886 * in the used cluster map. 3887 */ 3888 if (cluster_idx != 0) { 3889 spdk_bit_array_set(ctx->used_clusters, cluster_idx + j); 3890 if (bs->num_free_clusters == 0) { 3891 return -ENOSPC; 3892 } 3893 bs->num_free_clusters--; 3894 } 3895 cluster_count++; 3896 } 3897 } 3898 if (cluster_count == 0) { 3899 return -EINVAL; 3900 } 3901 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 3902 struct spdk_blob_md_descriptor_extent_page *desc_extent; 3903 uint32_t i; 3904 uint32_t cluster_count = 0; 3905 uint32_t cluster_idx; 3906 size_t cluster_idx_length; 3907 3908 desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc; 3909 cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx); 3910 3911 if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) || 3912 (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) { 3913 return -EINVAL; 3914 } 3915 3916 for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) { 3917 cluster_idx = desc_extent->cluster_idx[i]; 3918 /* 3919 * cluster_idx = 0 means an unallocated cluster - don't mark that 3920 * in the used cluster map. 3921 */ 3922 if (cluster_idx != 0) { 3923 if (cluster_idx < desc_extent->start_cluster_idx && 3924 cluster_idx >= desc_extent->start_cluster_idx + cluster_count) { 3925 return -EINVAL; 3926 } 3927 spdk_bit_array_set(ctx->used_clusters, cluster_idx); 3928 if (bs->num_free_clusters == 0) { 3929 return -ENOSPC; 3930 } 3931 bs->num_free_clusters--; 3932 } 3933 cluster_count++; 3934 } 3935 3936 if (cluster_count == 0) { 3937 return -EINVAL; 3938 } 3939 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 3940 /* Skip this item */ 3941 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 3942 /* Skip this item */ 3943 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 3944 /* Skip this item */ 3945 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) { 3946 struct spdk_blob_md_descriptor_extent_table *desc_extent_table; 3947 uint32_t num_extent_pages = ctx->num_extent_pages; 3948 uint32_t i; 3949 size_t extent_pages_length; 3950 void *tmp; 3951 3952 desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc; 3953 extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters); 3954 3955 if (desc_extent_table->length == 0 || 3956 (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) { 3957 return -EINVAL; 3958 } 3959 3960 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 3961 if (desc_extent_table->extent_page[i].page_idx != 0) { 3962 if (desc_extent_table->extent_page[i].num_pages != 1) { 3963 return -EINVAL; 3964 } 3965 num_extent_pages += 1; 3966 } 3967 } 3968 3969 if (num_extent_pages > 0) { 3970 tmp = realloc(ctx->extent_page_num, num_extent_pages * sizeof(uint32_t)); 3971 if (tmp == NULL) { 3972 return -ENOMEM; 3973 } 3974 ctx->extent_page_num = tmp; 3975 3976 /* Extent table entries contain md page numbers for extent pages. 3977 * Zeroes represent unallocated extent pages, those are run-length-encoded. 3978 */ 3979 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 3980 if (desc_extent_table->extent_page[i].page_idx != 0) { 3981 ctx->extent_page_num[ctx->num_extent_pages] = desc_extent_table->extent_page[i].page_idx; 3982 ctx->num_extent_pages += 1; 3983 } 3984 } 3985 } 3986 } else { 3987 /* Error */ 3988 return -EINVAL; 3989 } 3990 /* Advance to the next descriptor */ 3991 cur_desc += sizeof(*desc) + desc->length; 3992 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 3993 break; 3994 } 3995 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 3996 } 3997 return 0; 3998 } 3999 4000 static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page) 4001 { 4002 uint32_t crc; 4003 struct spdk_blob_md_descriptor *desc = (struct spdk_blob_md_descriptor *)page->descriptors; 4004 size_t desc_len; 4005 4006 crc = blob_md_page_calc_crc(page); 4007 if (crc != page->crc) { 4008 return false; 4009 } 4010 4011 /* Extent page should always be of sequence num 0. */ 4012 if (page->sequence_num != 0) { 4013 return false; 4014 } 4015 4016 /* Descriptor type must be EXTENT_PAGE. */ 4017 if (desc->type != SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 4018 return false; 4019 } 4020 4021 /* Descriptor length cannot exceed the page. */ 4022 desc_len = sizeof(*desc) + desc->length; 4023 if (desc_len > sizeof(page->descriptors)) { 4024 return false; 4025 } 4026 4027 /* It has to be the only descriptor in the page. */ 4028 if (desc_len + sizeof(*desc) <= sizeof(page->descriptors)) { 4029 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + desc_len); 4030 if (desc->length != 0) { 4031 return false; 4032 } 4033 } 4034 4035 return true; 4036 } 4037 4038 static bool bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx) 4039 { 4040 uint32_t crc; 4041 struct spdk_blob_md_page *page = ctx->page; 4042 4043 crc = blob_md_page_calc_crc(page); 4044 if (crc != page->crc) { 4045 return false; 4046 } 4047 4048 /* First page of a sequence should match the blobid. */ 4049 if (page->sequence_num == 0 && 4050 bs_page_to_blobid(ctx->cur_page) != page->id) { 4051 return false; 4052 } 4053 assert(bs_load_cur_extent_page_valid(page) == false); 4054 4055 return true; 4056 } 4057 4058 static void 4059 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx); 4060 4061 static void 4062 bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4063 { 4064 struct spdk_bs_load_ctx *ctx = cb_arg; 4065 4066 if (bserrno != 0) { 4067 bs_load_ctx_fail(ctx, bserrno); 4068 return; 4069 } 4070 4071 bs_load_complete(ctx); 4072 } 4073 4074 static void 4075 bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4076 { 4077 struct spdk_bs_load_ctx *ctx = cb_arg; 4078 4079 spdk_free(ctx->mask); 4080 ctx->mask = NULL; 4081 4082 if (bserrno != 0) { 4083 bs_load_ctx_fail(ctx, bserrno); 4084 return; 4085 } 4086 4087 bs_write_used_clusters(seq, ctx, bs_load_write_used_clusters_cpl); 4088 } 4089 4090 static void 4091 bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4092 { 4093 struct spdk_bs_load_ctx *ctx = cb_arg; 4094 4095 spdk_free(ctx->mask); 4096 ctx->mask = NULL; 4097 4098 if (bserrno != 0) { 4099 bs_load_ctx_fail(ctx, bserrno); 4100 return; 4101 } 4102 4103 bs_write_used_blobids(seq, ctx, bs_load_write_used_blobids_cpl); 4104 } 4105 4106 static void 4107 bs_load_write_used_md(struct spdk_bs_load_ctx *ctx) 4108 { 4109 bs_write_used_md(ctx->seq, ctx, bs_load_write_used_pages_cpl); 4110 } 4111 4112 static void 4113 bs_load_replay_md_chain_cpl(struct spdk_bs_load_ctx *ctx) 4114 { 4115 uint64_t num_md_clusters; 4116 uint64_t i; 4117 4118 ctx->in_page_chain = false; 4119 4120 do { 4121 ctx->page_index++; 4122 } while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true); 4123 4124 if (ctx->page_index < ctx->super->md_len) { 4125 ctx->cur_page = ctx->page_index; 4126 bs_load_replay_cur_md_page(ctx); 4127 } else { 4128 /* Claim all of the clusters used by the metadata */ 4129 num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster); 4130 for (i = 0; i < num_md_clusters; i++) { 4131 spdk_bit_array_set(ctx->used_clusters, i); 4132 } 4133 ctx->bs->num_free_clusters -= num_md_clusters; 4134 spdk_free(ctx->page); 4135 bs_load_write_used_md(ctx); 4136 } 4137 } 4138 4139 static void 4140 bs_load_replay_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4141 { 4142 struct spdk_bs_load_ctx *ctx = cb_arg; 4143 uint32_t page_num; 4144 uint64_t i; 4145 4146 if (bserrno != 0) { 4147 spdk_free(ctx->extent_pages); 4148 bs_load_ctx_fail(ctx, bserrno); 4149 return; 4150 } 4151 4152 for (i = 0; i < ctx->num_extent_pages; i++) { 4153 /* Extent pages are only read when present within in chain md. 4154 * Integrity of md is not right if that page was not a valid extent page. */ 4155 if (bs_load_cur_extent_page_valid(&ctx->extent_pages[i]) != true) { 4156 spdk_free(ctx->extent_pages); 4157 bs_load_ctx_fail(ctx, -EILSEQ); 4158 return; 4159 } 4160 4161 page_num = ctx->extent_page_num[i]; 4162 spdk_bit_array_set(ctx->bs->used_md_pages, page_num); 4163 if (bs_load_replay_md_parse_page(ctx, &ctx->extent_pages[i])) { 4164 spdk_free(ctx->extent_pages); 4165 bs_load_ctx_fail(ctx, -EILSEQ); 4166 return; 4167 } 4168 } 4169 4170 spdk_free(ctx->extent_pages); 4171 free(ctx->extent_page_num); 4172 ctx->extent_page_num = NULL; 4173 ctx->num_extent_pages = 0; 4174 4175 bs_load_replay_md_chain_cpl(ctx); 4176 } 4177 4178 static void 4179 bs_load_replay_extent_pages(struct spdk_bs_load_ctx *ctx) 4180 { 4181 spdk_bs_batch_t *batch; 4182 uint32_t page; 4183 uint64_t lba; 4184 uint64_t i; 4185 4186 ctx->extent_pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE * ctx->num_extent_pages, 0, 4187 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 4188 if (!ctx->extent_pages) { 4189 bs_load_ctx_fail(ctx, -ENOMEM); 4190 return; 4191 } 4192 4193 batch = bs_sequence_to_batch(ctx->seq, bs_load_replay_extent_page_cpl, ctx); 4194 4195 for (i = 0; i < ctx->num_extent_pages; i++) { 4196 page = ctx->extent_page_num[i]; 4197 assert(page < ctx->super->md_len); 4198 lba = bs_md_page_to_lba(ctx->bs, page); 4199 bs_batch_read_dev(batch, &ctx->extent_pages[i], lba, 4200 bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE)); 4201 } 4202 4203 bs_batch_close(batch); 4204 } 4205 4206 static void 4207 bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4208 { 4209 struct spdk_bs_load_ctx *ctx = cb_arg; 4210 uint32_t page_num; 4211 struct spdk_blob_md_page *page; 4212 4213 if (bserrno != 0) { 4214 bs_load_ctx_fail(ctx, bserrno); 4215 return; 4216 } 4217 4218 page_num = ctx->cur_page; 4219 page = ctx->page; 4220 if (bs_load_cur_md_page_valid(ctx) == true) { 4221 if (page->sequence_num == 0 || ctx->in_page_chain == true) { 4222 bs_claim_md_page(ctx->bs, page_num); 4223 if (page->sequence_num == 0) { 4224 spdk_bit_array_set(ctx->bs->used_blobids, page_num); 4225 } 4226 if (bs_load_replay_md_parse_page(ctx, page)) { 4227 bs_load_ctx_fail(ctx, -EILSEQ); 4228 return; 4229 } 4230 if (page->next != SPDK_INVALID_MD_PAGE) { 4231 ctx->in_page_chain = true; 4232 ctx->cur_page = page->next; 4233 bs_load_replay_cur_md_page(ctx); 4234 return; 4235 } 4236 if (ctx->num_extent_pages != 0) { 4237 bs_load_replay_extent_pages(ctx); 4238 return; 4239 } 4240 } 4241 } 4242 bs_load_replay_md_chain_cpl(ctx); 4243 } 4244 4245 static void 4246 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx) 4247 { 4248 uint64_t lba; 4249 4250 assert(ctx->cur_page < ctx->super->md_len); 4251 lba = bs_md_page_to_lba(ctx->bs, ctx->cur_page); 4252 bs_sequence_read_dev(ctx->seq, ctx->page, lba, 4253 bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 4254 bs_load_replay_md_cpl, ctx); 4255 } 4256 4257 static void 4258 bs_load_replay_md(struct spdk_bs_load_ctx *ctx) 4259 { 4260 ctx->page_index = 0; 4261 ctx->cur_page = 0; 4262 ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0, 4263 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 4264 if (!ctx->page) { 4265 bs_load_ctx_fail(ctx, -ENOMEM); 4266 return; 4267 } 4268 bs_load_replay_cur_md_page(ctx); 4269 } 4270 4271 static void 4272 bs_recover(struct spdk_bs_load_ctx *ctx) 4273 { 4274 int rc; 4275 4276 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len); 4277 if (rc < 0) { 4278 bs_load_ctx_fail(ctx, -ENOMEM); 4279 return; 4280 } 4281 4282 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len); 4283 if (rc < 0) { 4284 bs_load_ctx_fail(ctx, -ENOMEM); 4285 return; 4286 } 4287 4288 rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters); 4289 if (rc < 0) { 4290 bs_load_ctx_fail(ctx, -ENOMEM); 4291 return; 4292 } 4293 4294 rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->super->md_len); 4295 if (rc < 0) { 4296 bs_load_ctx_fail(ctx, -ENOMEM); 4297 return; 4298 } 4299 4300 ctx->bs->num_free_clusters = ctx->bs->total_clusters; 4301 bs_load_replay_md(ctx); 4302 } 4303 4304 static void 4305 bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4306 { 4307 struct spdk_bs_load_ctx *ctx = cb_arg; 4308 uint32_t crc; 4309 int rc; 4310 static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH]; 4311 4312 if (ctx->super->version > SPDK_BS_VERSION || 4313 ctx->super->version < SPDK_BS_INITIAL_VERSION) { 4314 bs_load_ctx_fail(ctx, -EILSEQ); 4315 return; 4316 } 4317 4318 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 4319 sizeof(ctx->super->signature)) != 0) { 4320 bs_load_ctx_fail(ctx, -EILSEQ); 4321 return; 4322 } 4323 4324 crc = blob_md_page_calc_crc(ctx->super); 4325 if (crc != ctx->super->crc) { 4326 bs_load_ctx_fail(ctx, -EILSEQ); 4327 return; 4328 } 4329 4330 if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 4331 SPDK_DEBUGLOG(blob, "Bstype matched - loading blobstore\n"); 4332 } else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 4333 SPDK_DEBUGLOG(blob, "Bstype wildcard used - loading blobstore regardless bstype\n"); 4334 } else { 4335 SPDK_DEBUGLOG(blob, "Unexpected bstype\n"); 4336 SPDK_LOGDUMP(blob, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 4337 SPDK_LOGDUMP(blob, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 4338 bs_load_ctx_fail(ctx, -ENXIO); 4339 return; 4340 } 4341 4342 if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) { 4343 SPDK_NOTICELOG("Size mismatch, dev size: %" PRIu64 ", blobstore size: %" PRIu64 "\n", 4344 ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size); 4345 bs_load_ctx_fail(ctx, -EILSEQ); 4346 return; 4347 } 4348 4349 if (ctx->super->size == 0) { 4350 ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen; 4351 } 4352 4353 if (ctx->super->io_unit_size == 0) { 4354 ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE; 4355 } 4356 4357 /* Parse the super block */ 4358 ctx->bs->clean = 1; 4359 ctx->bs->cluster_sz = ctx->super->cluster_size; 4360 ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size; 4361 ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE; 4362 if (spdk_u32_is_pow2(ctx->bs->pages_per_cluster)) { 4363 ctx->bs->pages_per_cluster_shift = spdk_u32log2(ctx->bs->pages_per_cluster); 4364 } 4365 ctx->bs->io_unit_size = ctx->super->io_unit_size; 4366 rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters); 4367 if (rc < 0) { 4368 bs_load_ctx_fail(ctx, -ENOMEM); 4369 return; 4370 } 4371 ctx->bs->md_start = ctx->super->md_start; 4372 ctx->bs->md_len = ctx->super->md_len; 4373 rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->bs->md_len); 4374 if (rc < 0) { 4375 bs_load_ctx_fail(ctx, -ENOMEM); 4376 return; 4377 } 4378 4379 ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up( 4380 ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster); 4381 ctx->bs->super_blob = ctx->super->super_blob; 4382 memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype)); 4383 4384 if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) { 4385 bs_recover(ctx); 4386 } else { 4387 bs_load_read_used_pages(ctx); 4388 } 4389 } 4390 4391 static inline int 4392 bs_opts_copy(struct spdk_bs_opts *src, struct spdk_bs_opts *dst) 4393 { 4394 4395 if (!src->opts_size) { 4396 SPDK_ERRLOG("opts_size should not be zero value\n"); 4397 return -1; 4398 } 4399 4400 #define FIELD_OK(field) \ 4401 offsetof(struct spdk_bs_opts, field) + sizeof(src->field) <= src->opts_size 4402 4403 #define SET_FIELD(field) \ 4404 if (FIELD_OK(field)) { \ 4405 dst->field = src->field; \ 4406 } \ 4407 4408 SET_FIELD(cluster_sz); 4409 SET_FIELD(num_md_pages); 4410 SET_FIELD(max_md_ops); 4411 SET_FIELD(max_channel_ops); 4412 SET_FIELD(clear_method); 4413 4414 if (FIELD_OK(bstype)) { 4415 memcpy(&dst->bstype, &src->bstype, sizeof(dst->bstype)); 4416 } 4417 SET_FIELD(iter_cb_fn); 4418 SET_FIELD(iter_cb_arg); 4419 4420 dst->opts_size = src->opts_size; 4421 4422 /* You should not remove this statement, but need to update the assert statement 4423 * if you add a new field, and also add a corresponding SET_FIELD statement */ 4424 SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_opts) == 64, "Incorrect size"); 4425 4426 #undef FIELD_OK 4427 #undef SET_FIELD 4428 4429 return 0; 4430 } 4431 4432 void 4433 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 4434 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 4435 { 4436 struct spdk_blob_store *bs; 4437 struct spdk_bs_cpl cpl; 4438 struct spdk_bs_load_ctx *ctx; 4439 struct spdk_bs_opts opts = {}; 4440 int err; 4441 4442 SPDK_DEBUGLOG(blob, "Loading blobstore from dev %p\n", dev); 4443 4444 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 4445 SPDK_DEBUGLOG(blob, "unsupported dev block length of %d\n", dev->blocklen); 4446 dev->destroy(dev); 4447 cb_fn(cb_arg, NULL, -EINVAL); 4448 return; 4449 } 4450 4451 spdk_bs_opts_init(&opts, sizeof(opts)); 4452 if (o) { 4453 if (bs_opts_copy(o, &opts)) { 4454 return; 4455 } 4456 } 4457 4458 if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) { 4459 dev->destroy(dev); 4460 cb_fn(cb_arg, NULL, -EINVAL); 4461 return; 4462 } 4463 4464 err = bs_alloc(dev, &opts, &bs, &ctx); 4465 if (err) { 4466 dev->destroy(dev); 4467 cb_fn(cb_arg, NULL, err); 4468 return; 4469 } 4470 4471 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 4472 cpl.u.bs_handle.cb_fn = cb_fn; 4473 cpl.u.bs_handle.cb_arg = cb_arg; 4474 cpl.u.bs_handle.bs = bs; 4475 4476 ctx->seq = bs_sequence_start(bs->md_channel, &cpl); 4477 if (!ctx->seq) { 4478 spdk_free(ctx->super); 4479 free(ctx); 4480 bs_free(bs); 4481 cb_fn(cb_arg, NULL, -ENOMEM); 4482 return; 4483 } 4484 4485 /* Read the super block */ 4486 bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0), 4487 bs_byte_to_lba(bs, sizeof(*ctx->super)), 4488 bs_load_super_cpl, ctx); 4489 } 4490 4491 /* END spdk_bs_load */ 4492 4493 /* START spdk_bs_dump */ 4494 4495 static void 4496 bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 4497 { 4498 spdk_free(ctx->super); 4499 4500 /* 4501 * We need to defer calling bs_call_cpl() until after 4502 * dev destruction, so tuck these away for later use. 4503 */ 4504 ctx->bs->unload_err = bserrno; 4505 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 4506 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 4507 4508 bs_sequence_finish(seq, 0); 4509 bs_free(ctx->bs); 4510 free(ctx); 4511 } 4512 4513 static void bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg); 4514 4515 static void 4516 bs_dump_print_md_page(struct spdk_bs_load_ctx *ctx) 4517 { 4518 uint32_t page_idx = ctx->cur_page; 4519 struct spdk_blob_md_page *page = ctx->page; 4520 struct spdk_blob_md_descriptor *desc; 4521 size_t cur_desc = 0; 4522 uint32_t crc; 4523 4524 fprintf(ctx->fp, "=========\n"); 4525 fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx); 4526 fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id); 4527 4528 crc = blob_md_page_calc_crc(page); 4529 fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch"); 4530 4531 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 4532 while (cur_desc < sizeof(page->descriptors)) { 4533 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 4534 if (desc->length == 0) { 4535 /* If padding and length are 0, this terminates the page */ 4536 break; 4537 } 4538 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) { 4539 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 4540 unsigned int i; 4541 4542 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc; 4543 4544 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 4545 if (desc_extent_rle->extents[i].cluster_idx != 0) { 4546 fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32, 4547 desc_extent_rle->extents[i].cluster_idx); 4548 } else { 4549 fprintf(ctx->fp, "Unallocated Extent - "); 4550 } 4551 fprintf(ctx->fp, " Length: %" PRIu32, desc_extent_rle->extents[i].length); 4552 fprintf(ctx->fp, "\n"); 4553 } 4554 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 4555 struct spdk_blob_md_descriptor_extent_page *desc_extent; 4556 unsigned int i; 4557 4558 desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc; 4559 4560 for (i = 0; i < desc_extent->length / sizeof(desc_extent->cluster_idx[0]); i++) { 4561 if (desc_extent->cluster_idx[i] != 0) { 4562 fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32, 4563 desc_extent->cluster_idx[i]); 4564 } else { 4565 fprintf(ctx->fp, "Unallocated Extent"); 4566 } 4567 fprintf(ctx->fp, "\n"); 4568 } 4569 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 4570 struct spdk_blob_md_descriptor_xattr *desc_xattr; 4571 uint32_t i; 4572 4573 desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc; 4574 4575 if (desc_xattr->length != 4576 sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) + 4577 desc_xattr->name_length + desc_xattr->value_length) { 4578 } 4579 4580 memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length); 4581 ctx->xattr_name[desc_xattr->name_length] = '\0'; 4582 fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name); 4583 fprintf(ctx->fp, " value = \""); 4584 ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name, 4585 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 4586 desc_xattr->value_length); 4587 fprintf(ctx->fp, "\"\n"); 4588 for (i = 0; i < desc_xattr->value_length; i++) { 4589 if (i % 16 == 0) { 4590 fprintf(ctx->fp, " "); 4591 } 4592 fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i)); 4593 if ((i + 1) % 16 == 0) { 4594 fprintf(ctx->fp, "\n"); 4595 } 4596 } 4597 if (i % 16 != 0) { 4598 fprintf(ctx->fp, "\n"); 4599 } 4600 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 4601 /* TODO */ 4602 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 4603 /* TODO */ 4604 } else { 4605 /* Error */ 4606 } 4607 /* Advance to the next descriptor */ 4608 cur_desc += sizeof(*desc) + desc->length; 4609 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 4610 break; 4611 } 4612 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 4613 } 4614 } 4615 4616 static void 4617 bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4618 { 4619 struct spdk_bs_load_ctx *ctx = cb_arg; 4620 4621 if (bserrno != 0) { 4622 bs_dump_finish(seq, ctx, bserrno); 4623 return; 4624 } 4625 4626 if (ctx->page->id != 0) { 4627 bs_dump_print_md_page(ctx); 4628 } 4629 4630 ctx->cur_page++; 4631 4632 if (ctx->cur_page < ctx->super->md_len) { 4633 bs_dump_read_md_page(seq, ctx); 4634 } else { 4635 spdk_free(ctx->page); 4636 bs_dump_finish(seq, ctx, 0); 4637 } 4638 } 4639 4640 static void 4641 bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg) 4642 { 4643 struct spdk_bs_load_ctx *ctx = cb_arg; 4644 uint64_t lba; 4645 4646 assert(ctx->cur_page < ctx->super->md_len); 4647 lba = bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page); 4648 bs_sequence_read_dev(seq, ctx->page, lba, 4649 bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 4650 bs_dump_read_md_page_cpl, ctx); 4651 } 4652 4653 static void 4654 bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4655 { 4656 struct spdk_bs_load_ctx *ctx = cb_arg; 4657 4658 fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature); 4659 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 4660 sizeof(ctx->super->signature)) != 0) { 4661 fprintf(ctx->fp, "(Mismatch)\n"); 4662 bs_dump_finish(seq, ctx, bserrno); 4663 return; 4664 } else { 4665 fprintf(ctx->fp, "(OK)\n"); 4666 } 4667 fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version); 4668 fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc, 4669 (ctx->super->crc == blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch"); 4670 fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype); 4671 fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size); 4672 fprintf(ctx->fp, "Super Blob ID: "); 4673 if (ctx->super->super_blob == SPDK_BLOBID_INVALID) { 4674 fprintf(ctx->fp, "(None)\n"); 4675 } else { 4676 fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob); 4677 } 4678 fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean); 4679 fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start); 4680 fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len); 4681 fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start); 4682 fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len); 4683 fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start); 4684 fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len); 4685 fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start); 4686 fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len); 4687 4688 ctx->cur_page = 0; 4689 ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0, 4690 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 4691 if (!ctx->page) { 4692 bs_dump_finish(seq, ctx, -ENOMEM); 4693 return; 4694 } 4695 bs_dump_read_md_page(seq, ctx); 4696 } 4697 4698 void 4699 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn, 4700 spdk_bs_op_complete cb_fn, void *cb_arg) 4701 { 4702 struct spdk_blob_store *bs; 4703 struct spdk_bs_cpl cpl; 4704 spdk_bs_sequence_t *seq; 4705 struct spdk_bs_load_ctx *ctx; 4706 struct spdk_bs_opts opts = {}; 4707 int err; 4708 4709 SPDK_DEBUGLOG(blob, "Dumping blobstore from dev %p\n", dev); 4710 4711 spdk_bs_opts_init(&opts, sizeof(opts)); 4712 4713 err = bs_alloc(dev, &opts, &bs, &ctx); 4714 if (err) { 4715 dev->destroy(dev); 4716 cb_fn(cb_arg, err); 4717 return; 4718 } 4719 4720 ctx->fp = fp; 4721 ctx->print_xattr_fn = print_xattr_fn; 4722 4723 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 4724 cpl.u.bs_basic.cb_fn = cb_fn; 4725 cpl.u.bs_basic.cb_arg = cb_arg; 4726 4727 seq = bs_sequence_start(bs->md_channel, &cpl); 4728 if (!seq) { 4729 spdk_free(ctx->super); 4730 free(ctx); 4731 bs_free(bs); 4732 cb_fn(cb_arg, -ENOMEM); 4733 return; 4734 } 4735 4736 /* Read the super block */ 4737 bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0), 4738 bs_byte_to_lba(bs, sizeof(*ctx->super)), 4739 bs_dump_super_cpl, ctx); 4740 } 4741 4742 /* END spdk_bs_dump */ 4743 4744 /* START spdk_bs_init */ 4745 4746 static void 4747 bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4748 { 4749 struct spdk_bs_load_ctx *ctx = cb_arg; 4750 4751 ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters); 4752 spdk_free(ctx->super); 4753 free(ctx); 4754 4755 bs_sequence_finish(seq, bserrno); 4756 } 4757 4758 static void 4759 bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4760 { 4761 struct spdk_bs_load_ctx *ctx = cb_arg; 4762 4763 /* Write super block */ 4764 bs_sequence_write_dev(seq, ctx->super, bs_page_to_lba(ctx->bs, 0), 4765 bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)), 4766 bs_init_persist_super_cpl, ctx); 4767 } 4768 4769 void 4770 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 4771 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 4772 { 4773 struct spdk_bs_load_ctx *ctx; 4774 struct spdk_blob_store *bs; 4775 struct spdk_bs_cpl cpl; 4776 spdk_bs_sequence_t *seq; 4777 spdk_bs_batch_t *batch; 4778 uint64_t num_md_lba; 4779 uint64_t num_md_pages; 4780 uint64_t num_md_clusters; 4781 uint32_t i; 4782 struct spdk_bs_opts opts = {}; 4783 int rc; 4784 uint64_t lba, lba_count; 4785 4786 SPDK_DEBUGLOG(blob, "Initializing blobstore on dev %p\n", dev); 4787 4788 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 4789 SPDK_ERRLOG("unsupported dev block length of %d\n", 4790 dev->blocklen); 4791 dev->destroy(dev); 4792 cb_fn(cb_arg, NULL, -EINVAL); 4793 return; 4794 } 4795 4796 spdk_bs_opts_init(&opts, sizeof(opts)); 4797 if (o) { 4798 if (bs_opts_copy(o, &opts)) { 4799 return; 4800 } 4801 } 4802 4803 if (bs_opts_verify(&opts) != 0) { 4804 dev->destroy(dev); 4805 cb_fn(cb_arg, NULL, -EINVAL); 4806 return; 4807 } 4808 4809 rc = bs_alloc(dev, &opts, &bs, &ctx); 4810 if (rc) { 4811 dev->destroy(dev); 4812 cb_fn(cb_arg, NULL, rc); 4813 return; 4814 } 4815 4816 if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) { 4817 /* By default, allocate 1 page per cluster. 4818 * Technically, this over-allocates metadata 4819 * because more metadata will reduce the number 4820 * of usable clusters. This can be addressed with 4821 * more complex math in the future. 4822 */ 4823 bs->md_len = bs->total_clusters; 4824 } else { 4825 bs->md_len = opts.num_md_pages; 4826 } 4827 rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len); 4828 if (rc < 0) { 4829 spdk_free(ctx->super); 4830 free(ctx); 4831 bs_free(bs); 4832 cb_fn(cb_arg, NULL, -ENOMEM); 4833 return; 4834 } 4835 4836 rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len); 4837 if (rc < 0) { 4838 spdk_free(ctx->super); 4839 free(ctx); 4840 bs_free(bs); 4841 cb_fn(cb_arg, NULL, -ENOMEM); 4842 return; 4843 } 4844 4845 rc = spdk_bit_array_resize(&bs->open_blobids, bs->md_len); 4846 if (rc < 0) { 4847 spdk_free(ctx->super); 4848 free(ctx); 4849 bs_free(bs); 4850 cb_fn(cb_arg, NULL, -ENOMEM); 4851 return; 4852 } 4853 4854 memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 4855 sizeof(ctx->super->signature)); 4856 ctx->super->version = SPDK_BS_VERSION; 4857 ctx->super->length = sizeof(*ctx->super); 4858 ctx->super->super_blob = bs->super_blob; 4859 ctx->super->clean = 0; 4860 ctx->super->cluster_size = bs->cluster_sz; 4861 ctx->super->io_unit_size = bs->io_unit_size; 4862 memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype)); 4863 4864 /* Calculate how many pages the metadata consumes at the front 4865 * of the disk. 4866 */ 4867 4868 /* The super block uses 1 page */ 4869 num_md_pages = 1; 4870 4871 /* The used_md_pages mask requires 1 bit per metadata page, rounded 4872 * up to the nearest page, plus a header. 4873 */ 4874 ctx->super->used_page_mask_start = num_md_pages; 4875 ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 4876 spdk_divide_round_up(bs->md_len, 8), 4877 SPDK_BS_PAGE_SIZE); 4878 num_md_pages += ctx->super->used_page_mask_len; 4879 4880 /* The used_clusters mask requires 1 bit per cluster, rounded 4881 * up to the nearest page, plus a header. 4882 */ 4883 ctx->super->used_cluster_mask_start = num_md_pages; 4884 ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 4885 spdk_divide_round_up(bs->total_clusters, 8), 4886 SPDK_BS_PAGE_SIZE); 4887 num_md_pages += ctx->super->used_cluster_mask_len; 4888 4889 /* The used_blobids mask requires 1 bit per metadata page, rounded 4890 * up to the nearest page, plus a header. 4891 */ 4892 ctx->super->used_blobid_mask_start = num_md_pages; 4893 ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 4894 spdk_divide_round_up(bs->md_len, 8), 4895 SPDK_BS_PAGE_SIZE); 4896 num_md_pages += ctx->super->used_blobid_mask_len; 4897 4898 /* The metadata region size was chosen above */ 4899 ctx->super->md_start = bs->md_start = num_md_pages; 4900 ctx->super->md_len = bs->md_len; 4901 num_md_pages += bs->md_len; 4902 4903 num_md_lba = bs_page_to_lba(bs, num_md_pages); 4904 4905 ctx->super->size = dev->blockcnt * dev->blocklen; 4906 4907 ctx->super->crc = blob_md_page_calc_crc(ctx->super); 4908 4909 num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster); 4910 if (num_md_clusters > bs->total_clusters) { 4911 SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, " 4912 "please decrease number of pages reserved for metadata " 4913 "or increase cluster size.\n"); 4914 spdk_free(ctx->super); 4915 spdk_bit_array_free(&ctx->used_clusters); 4916 free(ctx); 4917 bs_free(bs); 4918 cb_fn(cb_arg, NULL, -ENOMEM); 4919 return; 4920 } 4921 /* Claim all of the clusters used by the metadata */ 4922 for (i = 0; i < num_md_clusters; i++) { 4923 spdk_bit_array_set(ctx->used_clusters, i); 4924 } 4925 4926 bs->num_free_clusters -= num_md_clusters; 4927 bs->total_data_clusters = bs->num_free_clusters; 4928 4929 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 4930 cpl.u.bs_handle.cb_fn = cb_fn; 4931 cpl.u.bs_handle.cb_arg = cb_arg; 4932 cpl.u.bs_handle.bs = bs; 4933 4934 seq = bs_sequence_start(bs->md_channel, &cpl); 4935 if (!seq) { 4936 spdk_free(ctx->super); 4937 free(ctx); 4938 bs_free(bs); 4939 cb_fn(cb_arg, NULL, -ENOMEM); 4940 return; 4941 } 4942 4943 batch = bs_sequence_to_batch(seq, bs_init_trim_cpl, ctx); 4944 4945 /* Clear metadata space */ 4946 bs_batch_write_zeroes_dev(batch, 0, num_md_lba); 4947 4948 lba = num_md_lba; 4949 while (lba < ctx->bs->dev->blockcnt) { 4950 lba_count = spdk_min(UINT32_MAX, ctx->bs->dev->blockcnt - lba); 4951 switch (opts.clear_method) { 4952 case BS_CLEAR_WITH_UNMAP: 4953 /* Trim data clusters */ 4954 bs_batch_unmap_dev(batch, lba, lba_count); 4955 break; 4956 case BS_CLEAR_WITH_WRITE_ZEROES: 4957 /* Write_zeroes to data clusters */ 4958 bs_batch_write_zeroes_dev(batch, lba, lba_count); 4959 break; 4960 case BS_CLEAR_WITH_NONE: 4961 default: 4962 break; 4963 } 4964 lba += lba_count; 4965 } 4966 4967 bs_batch_close(batch); 4968 } 4969 4970 /* END spdk_bs_init */ 4971 4972 /* START spdk_bs_destroy */ 4973 4974 static void 4975 bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4976 { 4977 struct spdk_bs_load_ctx *ctx = cb_arg; 4978 struct spdk_blob_store *bs = ctx->bs; 4979 4980 /* 4981 * We need to defer calling bs_call_cpl() until after 4982 * dev destruction, so tuck these away for later use. 4983 */ 4984 bs->unload_err = bserrno; 4985 memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 4986 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 4987 4988 bs_sequence_finish(seq, bserrno); 4989 4990 bs_free(bs); 4991 free(ctx); 4992 } 4993 4994 void 4995 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, 4996 void *cb_arg) 4997 { 4998 struct spdk_bs_cpl cpl; 4999 spdk_bs_sequence_t *seq; 5000 struct spdk_bs_load_ctx *ctx; 5001 5002 SPDK_DEBUGLOG(blob, "Destroying blobstore\n"); 5003 5004 if (!TAILQ_EMPTY(&bs->blobs)) { 5005 SPDK_ERRLOG("Blobstore still has open blobs\n"); 5006 cb_fn(cb_arg, -EBUSY); 5007 return; 5008 } 5009 5010 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 5011 cpl.u.bs_basic.cb_fn = cb_fn; 5012 cpl.u.bs_basic.cb_arg = cb_arg; 5013 5014 ctx = calloc(1, sizeof(*ctx)); 5015 if (!ctx) { 5016 cb_fn(cb_arg, -ENOMEM); 5017 return; 5018 } 5019 5020 ctx->bs = bs; 5021 5022 seq = bs_sequence_start(bs->md_channel, &cpl); 5023 if (!seq) { 5024 free(ctx); 5025 cb_fn(cb_arg, -ENOMEM); 5026 return; 5027 } 5028 5029 /* Write zeroes to the super block */ 5030 bs_sequence_write_zeroes_dev(seq, 5031 bs_page_to_lba(bs, 0), 5032 bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)), 5033 bs_destroy_trim_cpl, ctx); 5034 } 5035 5036 /* END spdk_bs_destroy */ 5037 5038 /* START spdk_bs_unload */ 5039 5040 static void 5041 bs_unload_finish(struct spdk_bs_load_ctx *ctx, int bserrno) 5042 { 5043 spdk_bs_sequence_t *seq = ctx->seq; 5044 5045 spdk_free(ctx->super); 5046 5047 /* 5048 * We need to defer calling bs_call_cpl() until after 5049 * dev destruction, so tuck these away for later use. 5050 */ 5051 ctx->bs->unload_err = bserrno; 5052 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 5053 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 5054 5055 bs_sequence_finish(seq, bserrno); 5056 5057 bs_free(ctx->bs); 5058 free(ctx); 5059 } 5060 5061 static void 5062 bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5063 { 5064 struct spdk_bs_load_ctx *ctx = cb_arg; 5065 5066 bs_unload_finish(ctx, bserrno); 5067 } 5068 5069 static void 5070 bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5071 { 5072 struct spdk_bs_load_ctx *ctx = cb_arg; 5073 5074 spdk_free(ctx->mask); 5075 5076 if (bserrno != 0) { 5077 bs_unload_finish(ctx, bserrno); 5078 return; 5079 } 5080 5081 ctx->super->clean = 1; 5082 5083 bs_write_super(seq, ctx->bs, ctx->super, bs_unload_write_super_cpl, ctx); 5084 } 5085 5086 static void 5087 bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5088 { 5089 struct spdk_bs_load_ctx *ctx = cb_arg; 5090 5091 spdk_free(ctx->mask); 5092 ctx->mask = NULL; 5093 5094 if (bserrno != 0) { 5095 bs_unload_finish(ctx, bserrno); 5096 return; 5097 } 5098 5099 bs_write_used_clusters(seq, ctx, bs_unload_write_used_clusters_cpl); 5100 } 5101 5102 static void 5103 bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5104 { 5105 struct spdk_bs_load_ctx *ctx = cb_arg; 5106 5107 spdk_free(ctx->mask); 5108 ctx->mask = NULL; 5109 5110 if (bserrno != 0) { 5111 bs_unload_finish(ctx, bserrno); 5112 return; 5113 } 5114 5115 bs_write_used_blobids(seq, ctx, bs_unload_write_used_blobids_cpl); 5116 } 5117 5118 static void 5119 bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5120 { 5121 struct spdk_bs_load_ctx *ctx = cb_arg; 5122 5123 if (bserrno != 0) { 5124 bs_unload_finish(ctx, bserrno); 5125 return; 5126 } 5127 5128 bs_write_used_md(seq, cb_arg, bs_unload_write_used_pages_cpl); 5129 } 5130 5131 void 5132 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg) 5133 { 5134 struct spdk_bs_cpl cpl; 5135 struct spdk_bs_load_ctx *ctx; 5136 5137 SPDK_DEBUGLOG(blob, "Syncing blobstore\n"); 5138 5139 if (!TAILQ_EMPTY(&bs->blobs)) { 5140 SPDK_ERRLOG("Blobstore still has open blobs\n"); 5141 cb_fn(cb_arg, -EBUSY); 5142 return; 5143 } 5144 5145 ctx = calloc(1, sizeof(*ctx)); 5146 if (!ctx) { 5147 cb_fn(cb_arg, -ENOMEM); 5148 return; 5149 } 5150 5151 ctx->bs = bs; 5152 5153 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 5154 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 5155 if (!ctx->super) { 5156 free(ctx); 5157 cb_fn(cb_arg, -ENOMEM); 5158 return; 5159 } 5160 5161 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 5162 cpl.u.bs_basic.cb_fn = cb_fn; 5163 cpl.u.bs_basic.cb_arg = cb_arg; 5164 5165 ctx->seq = bs_sequence_start(bs->md_channel, &cpl); 5166 if (!ctx->seq) { 5167 spdk_free(ctx->super); 5168 free(ctx); 5169 cb_fn(cb_arg, -ENOMEM); 5170 return; 5171 } 5172 5173 /* Read super block */ 5174 bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0), 5175 bs_byte_to_lba(bs, sizeof(*ctx->super)), 5176 bs_unload_read_super_cpl, ctx); 5177 } 5178 5179 /* END spdk_bs_unload */ 5180 5181 /* START spdk_bs_set_super */ 5182 5183 struct spdk_bs_set_super_ctx { 5184 struct spdk_blob_store *bs; 5185 struct spdk_bs_super_block *super; 5186 }; 5187 5188 static void 5189 bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5190 { 5191 struct spdk_bs_set_super_ctx *ctx = cb_arg; 5192 5193 if (bserrno != 0) { 5194 SPDK_ERRLOG("Unable to write to super block of blobstore\n"); 5195 } 5196 5197 spdk_free(ctx->super); 5198 5199 bs_sequence_finish(seq, bserrno); 5200 5201 free(ctx); 5202 } 5203 5204 static void 5205 bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5206 { 5207 struct spdk_bs_set_super_ctx *ctx = cb_arg; 5208 5209 if (bserrno != 0) { 5210 SPDK_ERRLOG("Unable to read super block of blobstore\n"); 5211 spdk_free(ctx->super); 5212 bs_sequence_finish(seq, bserrno); 5213 free(ctx); 5214 return; 5215 } 5216 5217 bs_write_super(seq, ctx->bs, ctx->super, bs_set_super_write_cpl, ctx); 5218 } 5219 5220 void 5221 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid, 5222 spdk_bs_op_complete cb_fn, void *cb_arg) 5223 { 5224 struct spdk_bs_cpl cpl; 5225 spdk_bs_sequence_t *seq; 5226 struct spdk_bs_set_super_ctx *ctx; 5227 5228 SPDK_DEBUGLOG(blob, "Setting super blob id on blobstore\n"); 5229 5230 ctx = calloc(1, sizeof(*ctx)); 5231 if (!ctx) { 5232 cb_fn(cb_arg, -ENOMEM); 5233 return; 5234 } 5235 5236 ctx->bs = bs; 5237 5238 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 5239 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 5240 if (!ctx->super) { 5241 free(ctx); 5242 cb_fn(cb_arg, -ENOMEM); 5243 return; 5244 } 5245 5246 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 5247 cpl.u.bs_basic.cb_fn = cb_fn; 5248 cpl.u.bs_basic.cb_arg = cb_arg; 5249 5250 seq = bs_sequence_start(bs->md_channel, &cpl); 5251 if (!seq) { 5252 spdk_free(ctx->super); 5253 free(ctx); 5254 cb_fn(cb_arg, -ENOMEM); 5255 return; 5256 } 5257 5258 bs->super_blob = blobid; 5259 5260 /* Read super block */ 5261 bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0), 5262 bs_byte_to_lba(bs, sizeof(*ctx->super)), 5263 bs_set_super_read_cpl, ctx); 5264 } 5265 5266 /* END spdk_bs_set_super */ 5267 5268 void 5269 spdk_bs_get_super(struct spdk_blob_store *bs, 5270 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5271 { 5272 if (bs->super_blob == SPDK_BLOBID_INVALID) { 5273 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT); 5274 } else { 5275 cb_fn(cb_arg, bs->super_blob, 0); 5276 } 5277 } 5278 5279 uint64_t 5280 spdk_bs_get_cluster_size(struct spdk_blob_store *bs) 5281 { 5282 return bs->cluster_sz; 5283 } 5284 5285 uint64_t 5286 spdk_bs_get_page_size(struct spdk_blob_store *bs) 5287 { 5288 return SPDK_BS_PAGE_SIZE; 5289 } 5290 5291 uint64_t 5292 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs) 5293 { 5294 return bs->io_unit_size; 5295 } 5296 5297 uint64_t 5298 spdk_bs_free_cluster_count(struct spdk_blob_store *bs) 5299 { 5300 return bs->num_free_clusters; 5301 } 5302 5303 uint64_t 5304 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs) 5305 { 5306 return bs->total_data_clusters; 5307 } 5308 5309 static int 5310 bs_register_md_thread(struct spdk_blob_store *bs) 5311 { 5312 bs->md_channel = spdk_get_io_channel(bs); 5313 if (!bs->md_channel) { 5314 SPDK_ERRLOG("Failed to get IO channel.\n"); 5315 return -1; 5316 } 5317 5318 return 0; 5319 } 5320 5321 static int 5322 bs_unregister_md_thread(struct spdk_blob_store *bs) 5323 { 5324 spdk_put_io_channel(bs->md_channel); 5325 5326 return 0; 5327 } 5328 5329 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob) 5330 { 5331 assert(blob != NULL); 5332 5333 return blob->id; 5334 } 5335 5336 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob) 5337 { 5338 assert(blob != NULL); 5339 5340 return bs_cluster_to_page(blob->bs, blob->active.num_clusters); 5341 } 5342 5343 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob) 5344 { 5345 assert(blob != NULL); 5346 5347 return spdk_blob_get_num_pages(blob) * bs_io_unit_per_page(blob->bs); 5348 } 5349 5350 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob) 5351 { 5352 assert(blob != NULL); 5353 5354 return blob->active.num_clusters; 5355 } 5356 5357 /* START spdk_bs_create_blob */ 5358 5359 static void 5360 bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5361 { 5362 struct spdk_blob *blob = cb_arg; 5363 uint32_t page_idx = bs_blobid_to_page(blob->id); 5364 5365 if (bserrno != 0) { 5366 spdk_bit_array_clear(blob->bs->used_blobids, page_idx); 5367 bs_release_md_page(blob->bs, page_idx); 5368 } 5369 5370 blob_free(blob); 5371 5372 bs_sequence_finish(seq, bserrno); 5373 } 5374 5375 static int 5376 blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs, 5377 bool internal) 5378 { 5379 uint64_t i; 5380 size_t value_len = 0; 5381 int rc; 5382 const void *value = NULL; 5383 if (xattrs->count > 0 && xattrs->get_value == NULL) { 5384 return -EINVAL; 5385 } 5386 for (i = 0; i < xattrs->count; i++) { 5387 xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len); 5388 if (value == NULL || value_len == 0) { 5389 return -EINVAL; 5390 } 5391 rc = blob_set_xattr(blob, xattrs->names[i], value, value_len, internal); 5392 if (rc < 0) { 5393 return rc; 5394 } 5395 } 5396 return 0; 5397 } 5398 5399 static void 5400 blob_opts_copy(const struct spdk_blob_opts *src, struct spdk_blob_opts *dst) 5401 { 5402 #define FIELD_OK(field) \ 5403 offsetof(struct spdk_blob_opts, field) + sizeof(src->field) <= src->opts_size 5404 5405 #define SET_FIELD(field) \ 5406 if (FIELD_OK(field)) { \ 5407 dst->field = src->field; \ 5408 } \ 5409 5410 SET_FIELD(num_clusters); 5411 SET_FIELD(thin_provision); 5412 SET_FIELD(clear_method); 5413 5414 if (FIELD_OK(xattrs)) { 5415 memcpy(&dst->xattrs, &src->xattrs, sizeof(src->xattrs)); 5416 } 5417 5418 SET_FIELD(use_extent_table); 5419 5420 dst->opts_size = src->opts_size; 5421 5422 /* You should not remove this statement, but need to update the assert statement 5423 * if you add a new field, and also add a corresponding SET_FIELD statement */ 5424 SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_opts) == 64, "Incorrect size"); 5425 5426 #undef FIELD_OK 5427 #undef SET_FIELD 5428 } 5429 5430 static void 5431 bs_create_blob(struct spdk_blob_store *bs, 5432 const struct spdk_blob_opts *opts, 5433 const struct spdk_blob_xattr_opts *internal_xattrs, 5434 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5435 { 5436 struct spdk_blob *blob; 5437 uint32_t page_idx; 5438 struct spdk_bs_cpl cpl; 5439 struct spdk_blob_opts opts_local; 5440 struct spdk_blob_xattr_opts internal_xattrs_default; 5441 spdk_bs_sequence_t *seq; 5442 spdk_blob_id id; 5443 int rc; 5444 5445 assert(spdk_get_thread() == bs->md_thread); 5446 5447 page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0); 5448 if (page_idx == UINT32_MAX) { 5449 cb_fn(cb_arg, 0, -ENOMEM); 5450 return; 5451 } 5452 spdk_bit_array_set(bs->used_blobids, page_idx); 5453 bs_claim_md_page(bs, page_idx); 5454 5455 id = bs_page_to_blobid(page_idx); 5456 5457 SPDK_DEBUGLOG(blob, "Creating blob with id %" PRIu64 " at page %u\n", id, page_idx); 5458 5459 blob = blob_alloc(bs, id); 5460 if (!blob) { 5461 spdk_bit_array_clear(bs->used_blobids, page_idx); 5462 bs_release_md_page(bs, page_idx); 5463 cb_fn(cb_arg, 0, -ENOMEM); 5464 return; 5465 } 5466 5467 spdk_blob_opts_init(&opts_local, sizeof(opts_local)); 5468 if (opts) { 5469 blob_opts_copy(opts, &opts_local); 5470 } 5471 5472 blob->use_extent_table = opts_local.use_extent_table; 5473 if (blob->use_extent_table) { 5474 blob->invalid_flags |= SPDK_BLOB_EXTENT_TABLE; 5475 } 5476 5477 if (!internal_xattrs) { 5478 blob_xattrs_init(&internal_xattrs_default); 5479 internal_xattrs = &internal_xattrs_default; 5480 } 5481 5482 rc = blob_set_xattrs(blob, &opts_local.xattrs, false); 5483 if (rc < 0) { 5484 blob_free(blob); 5485 spdk_bit_array_clear(bs->used_blobids, page_idx); 5486 bs_release_md_page(bs, page_idx); 5487 cb_fn(cb_arg, 0, rc); 5488 return; 5489 } 5490 5491 rc = blob_set_xattrs(blob, internal_xattrs, true); 5492 if (rc < 0) { 5493 blob_free(blob); 5494 spdk_bit_array_clear(bs->used_blobids, page_idx); 5495 bs_release_md_page(bs, page_idx); 5496 cb_fn(cb_arg, 0, rc); 5497 return; 5498 } 5499 5500 if (opts_local.thin_provision) { 5501 blob_set_thin_provision(blob); 5502 } 5503 5504 blob_set_clear_method(blob, opts_local.clear_method); 5505 5506 rc = blob_resize(blob, opts_local.num_clusters); 5507 if (rc < 0) { 5508 blob_free(blob); 5509 spdk_bit_array_clear(bs->used_blobids, page_idx); 5510 bs_release_md_page(bs, page_idx); 5511 cb_fn(cb_arg, 0, rc); 5512 return; 5513 } 5514 cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 5515 cpl.u.blobid.cb_fn = cb_fn; 5516 cpl.u.blobid.cb_arg = cb_arg; 5517 cpl.u.blobid.blobid = blob->id; 5518 5519 seq = bs_sequence_start(bs->md_channel, &cpl); 5520 if (!seq) { 5521 blob_free(blob); 5522 spdk_bit_array_clear(bs->used_blobids, page_idx); 5523 bs_release_md_page(bs, page_idx); 5524 cb_fn(cb_arg, 0, -ENOMEM); 5525 return; 5526 } 5527 5528 blob_persist(seq, blob, bs_create_blob_cpl, blob); 5529 } 5530 5531 void spdk_bs_create_blob(struct spdk_blob_store *bs, 5532 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5533 { 5534 bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg); 5535 } 5536 5537 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts, 5538 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5539 { 5540 bs_create_blob(bs, opts, NULL, cb_fn, cb_arg); 5541 } 5542 5543 /* END spdk_bs_create_blob */ 5544 5545 /* START blob_cleanup */ 5546 5547 struct spdk_clone_snapshot_ctx { 5548 struct spdk_bs_cpl cpl; 5549 int bserrno; 5550 bool frozen; 5551 5552 struct spdk_io_channel *channel; 5553 5554 /* Current cluster for inflate operation */ 5555 uint64_t cluster; 5556 5557 /* For inflation force allocation of all unallocated clusters and remove 5558 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */ 5559 bool allocate_all; 5560 5561 struct { 5562 spdk_blob_id id; 5563 struct spdk_blob *blob; 5564 } original; 5565 struct { 5566 spdk_blob_id id; 5567 struct spdk_blob *blob; 5568 } new; 5569 5570 /* xattrs specified for snapshot/clones only. They have no impact on 5571 * the original blobs xattrs. */ 5572 const struct spdk_blob_xattr_opts *xattrs; 5573 }; 5574 5575 static void 5576 bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno) 5577 { 5578 struct spdk_clone_snapshot_ctx *ctx = cb_arg; 5579 struct spdk_bs_cpl *cpl = &ctx->cpl; 5580 5581 if (bserrno != 0) { 5582 if (ctx->bserrno != 0) { 5583 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 5584 } else { 5585 ctx->bserrno = bserrno; 5586 } 5587 } 5588 5589 switch (cpl->type) { 5590 case SPDK_BS_CPL_TYPE_BLOBID: 5591 cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno); 5592 break; 5593 case SPDK_BS_CPL_TYPE_BLOB_BASIC: 5594 cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno); 5595 break; 5596 default: 5597 SPDK_UNREACHABLE(); 5598 break; 5599 } 5600 5601 free(ctx); 5602 } 5603 5604 static void 5605 bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno) 5606 { 5607 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5608 struct spdk_blob *origblob = ctx->original.blob; 5609 5610 if (bserrno != 0) { 5611 if (ctx->bserrno != 0) { 5612 SPDK_ERRLOG("Unfreeze error %d\n", bserrno); 5613 } else { 5614 ctx->bserrno = bserrno; 5615 } 5616 } 5617 5618 ctx->original.id = origblob->id; 5619 origblob->locked_operation_in_progress = false; 5620 5621 spdk_blob_close(origblob, bs_clone_snapshot_cleanup_finish, ctx); 5622 } 5623 5624 static void 5625 bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno) 5626 { 5627 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5628 struct spdk_blob *origblob = ctx->original.blob; 5629 5630 if (bserrno != 0) { 5631 if (ctx->bserrno != 0) { 5632 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 5633 } else { 5634 ctx->bserrno = bserrno; 5635 } 5636 } 5637 5638 if (ctx->frozen) { 5639 /* Unfreeze any outstanding I/O */ 5640 blob_unfreeze_io(origblob, bs_snapshot_unfreeze_cpl, ctx); 5641 } else { 5642 bs_snapshot_unfreeze_cpl(ctx, 0); 5643 } 5644 5645 } 5646 5647 static void 5648 bs_clone_snapshot_newblob_cleanup(struct spdk_clone_snapshot_ctx *ctx, int bserrno) 5649 { 5650 struct spdk_blob *newblob = ctx->new.blob; 5651 5652 if (bserrno != 0) { 5653 if (ctx->bserrno != 0) { 5654 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 5655 } else { 5656 ctx->bserrno = bserrno; 5657 } 5658 } 5659 5660 ctx->new.id = newblob->id; 5661 spdk_blob_close(newblob, bs_clone_snapshot_origblob_cleanup, ctx); 5662 } 5663 5664 /* END blob_cleanup */ 5665 5666 /* START spdk_bs_create_snapshot */ 5667 5668 static void 5669 bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2) 5670 { 5671 uint64_t *cluster_temp; 5672 uint32_t *extent_page_temp; 5673 5674 cluster_temp = blob1->active.clusters; 5675 blob1->active.clusters = blob2->active.clusters; 5676 blob2->active.clusters = cluster_temp; 5677 5678 extent_page_temp = blob1->active.extent_pages; 5679 blob1->active.extent_pages = blob2->active.extent_pages; 5680 blob2->active.extent_pages = extent_page_temp; 5681 } 5682 5683 static void 5684 bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno) 5685 { 5686 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5687 struct spdk_blob *origblob = ctx->original.blob; 5688 struct spdk_blob *newblob = ctx->new.blob; 5689 5690 if (bserrno != 0) { 5691 bs_snapshot_swap_cluster_maps(newblob, origblob); 5692 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5693 return; 5694 } 5695 5696 /* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */ 5697 bserrno = blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true); 5698 if (bserrno != 0) { 5699 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5700 return; 5701 } 5702 5703 bs_blob_list_add(ctx->original.blob); 5704 5705 spdk_blob_set_read_only(newblob); 5706 5707 /* sync snapshot metadata */ 5708 spdk_blob_sync_md(newblob, bs_clone_snapshot_origblob_cleanup, ctx); 5709 } 5710 5711 static void 5712 bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno) 5713 { 5714 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5715 struct spdk_blob *origblob = ctx->original.blob; 5716 struct spdk_blob *newblob = ctx->new.blob; 5717 5718 if (bserrno != 0) { 5719 /* return cluster map back to original */ 5720 bs_snapshot_swap_cluster_maps(newblob, origblob); 5721 5722 /* Newblob md sync failed. Valid clusters are only present in origblob. 5723 * Since I/O is frozen on origblob, not changes to zeroed out cluster map should have occured. 5724 * Newblob needs to be reverted to thin_provisioned state at creation to properly close. */ 5725 blob_set_thin_provision(newblob); 5726 assert(spdk_mem_all_zero(newblob->active.clusters, 5727 newblob->active.num_clusters * sizeof(*newblob->active.clusters))); 5728 assert(spdk_mem_all_zero(newblob->active.extent_pages, 5729 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages))); 5730 5731 bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 5732 return; 5733 } 5734 5735 /* Set internal xattr for snapshot id */ 5736 bserrno = blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true); 5737 if (bserrno != 0) { 5738 /* return cluster map back to original */ 5739 bs_snapshot_swap_cluster_maps(newblob, origblob); 5740 blob_set_thin_provision(newblob); 5741 bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 5742 return; 5743 } 5744 5745 /* Create new back_bs_dev for snapshot */ 5746 origblob->back_bs_dev = bs_create_blob_bs_dev(newblob); 5747 if (origblob->back_bs_dev == NULL) { 5748 /* return cluster map back to original */ 5749 bs_snapshot_swap_cluster_maps(newblob, origblob); 5750 blob_set_thin_provision(newblob); 5751 bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL); 5752 return; 5753 } 5754 5755 bs_blob_list_remove(origblob); 5756 origblob->parent_id = newblob->id; 5757 /* set clone blob as thin provisioned */ 5758 blob_set_thin_provision(origblob); 5759 5760 bs_blob_list_add(newblob); 5761 5762 /* sync clone metadata */ 5763 spdk_blob_sync_md(origblob, bs_snapshot_origblob_sync_cpl, ctx); 5764 } 5765 5766 static void 5767 bs_snapshot_freeze_cpl(void *cb_arg, int rc) 5768 { 5769 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5770 struct spdk_blob *origblob = ctx->original.blob; 5771 struct spdk_blob *newblob = ctx->new.blob; 5772 int bserrno; 5773 5774 if (rc != 0) { 5775 bs_clone_snapshot_newblob_cleanup(ctx, rc); 5776 return; 5777 } 5778 5779 ctx->frozen = true; 5780 5781 /* set new back_bs_dev for snapshot */ 5782 newblob->back_bs_dev = origblob->back_bs_dev; 5783 /* Set invalid flags from origblob */ 5784 newblob->invalid_flags = origblob->invalid_flags; 5785 5786 /* inherit parent from original blob if set */ 5787 newblob->parent_id = origblob->parent_id; 5788 if (origblob->parent_id != SPDK_BLOBID_INVALID) { 5789 /* Set internal xattr for snapshot id */ 5790 bserrno = blob_set_xattr(newblob, BLOB_SNAPSHOT, 5791 &origblob->parent_id, sizeof(spdk_blob_id), true); 5792 if (bserrno != 0) { 5793 bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 5794 return; 5795 } 5796 } 5797 5798 /* swap cluster maps */ 5799 bs_snapshot_swap_cluster_maps(newblob, origblob); 5800 5801 /* Set the clear method on the new blob to match the original. */ 5802 blob_set_clear_method(newblob, origblob->clear_method); 5803 5804 /* sync snapshot metadata */ 5805 spdk_blob_sync_md(newblob, bs_snapshot_newblob_sync_cpl, ctx); 5806 } 5807 5808 static void 5809 bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5810 { 5811 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5812 struct spdk_blob *origblob = ctx->original.blob; 5813 struct spdk_blob *newblob = _blob; 5814 5815 if (bserrno != 0) { 5816 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5817 return; 5818 } 5819 5820 ctx->new.blob = newblob; 5821 assert(spdk_blob_is_thin_provisioned(newblob)); 5822 assert(spdk_mem_all_zero(newblob->active.clusters, 5823 newblob->active.num_clusters * sizeof(*newblob->active.clusters))); 5824 assert(spdk_mem_all_zero(newblob->active.extent_pages, 5825 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages))); 5826 5827 blob_freeze_io(origblob, bs_snapshot_freeze_cpl, ctx); 5828 } 5829 5830 static void 5831 bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 5832 { 5833 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5834 struct spdk_blob *origblob = ctx->original.blob; 5835 5836 if (bserrno != 0) { 5837 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5838 return; 5839 } 5840 5841 ctx->new.id = blobid; 5842 ctx->cpl.u.blobid.blobid = blobid; 5843 5844 spdk_bs_open_blob(origblob->bs, ctx->new.id, bs_snapshot_newblob_open_cpl, ctx); 5845 } 5846 5847 5848 static void 5849 bs_xattr_snapshot(void *arg, const char *name, 5850 const void **value, size_t *value_len) 5851 { 5852 assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0); 5853 5854 struct spdk_blob *blob = (struct spdk_blob *)arg; 5855 *value = &blob->id; 5856 *value_len = sizeof(blob->id); 5857 } 5858 5859 static void 5860 bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5861 { 5862 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5863 struct spdk_blob_opts opts; 5864 struct spdk_blob_xattr_opts internal_xattrs; 5865 char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS }; 5866 5867 if (bserrno != 0) { 5868 bs_clone_snapshot_cleanup_finish(ctx, bserrno); 5869 return; 5870 } 5871 5872 ctx->original.blob = _blob; 5873 5874 if (_blob->data_ro || _blob->md_ro) { 5875 SPDK_DEBUGLOG(blob, "Cannot create snapshot from read only blob with id %" PRIu64 "\n", 5876 _blob->id); 5877 ctx->bserrno = -EINVAL; 5878 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5879 return; 5880 } 5881 5882 if (_blob->locked_operation_in_progress) { 5883 SPDK_DEBUGLOG(blob, "Cannot create snapshot - another operation in progress\n"); 5884 ctx->bserrno = -EBUSY; 5885 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5886 return; 5887 } 5888 5889 _blob->locked_operation_in_progress = true; 5890 5891 spdk_blob_opts_init(&opts, sizeof(opts)); 5892 blob_xattrs_init(&internal_xattrs); 5893 5894 /* Change the size of new blob to the same as in original blob, 5895 * but do not allocate clusters */ 5896 opts.thin_provision = true; 5897 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 5898 opts.use_extent_table = _blob->use_extent_table; 5899 5900 /* If there are any xattrs specified for snapshot, set them now */ 5901 if (ctx->xattrs) { 5902 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 5903 } 5904 /* Set internal xattr SNAPSHOT_IN_PROGRESS */ 5905 internal_xattrs.count = 1; 5906 internal_xattrs.ctx = _blob; 5907 internal_xattrs.names = xattrs_names; 5908 internal_xattrs.get_value = bs_xattr_snapshot; 5909 5910 bs_create_blob(_blob->bs, &opts, &internal_xattrs, 5911 bs_snapshot_newblob_create_cpl, ctx); 5912 } 5913 5914 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid, 5915 const struct spdk_blob_xattr_opts *snapshot_xattrs, 5916 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5917 { 5918 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 5919 5920 if (!ctx) { 5921 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 5922 return; 5923 } 5924 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 5925 ctx->cpl.u.blobid.cb_fn = cb_fn; 5926 ctx->cpl.u.blobid.cb_arg = cb_arg; 5927 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 5928 ctx->bserrno = 0; 5929 ctx->frozen = false; 5930 ctx->original.id = blobid; 5931 ctx->xattrs = snapshot_xattrs; 5932 5933 spdk_bs_open_blob(bs, ctx->original.id, bs_snapshot_origblob_open_cpl, ctx); 5934 } 5935 /* END spdk_bs_create_snapshot */ 5936 5937 /* START spdk_bs_create_clone */ 5938 5939 static void 5940 bs_xattr_clone(void *arg, const char *name, 5941 const void **value, size_t *value_len) 5942 { 5943 assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0); 5944 5945 struct spdk_blob *blob = (struct spdk_blob *)arg; 5946 *value = &blob->id; 5947 *value_len = sizeof(blob->id); 5948 } 5949 5950 static void 5951 bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5952 { 5953 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5954 struct spdk_blob *clone = _blob; 5955 5956 ctx->new.blob = clone; 5957 bs_blob_list_add(clone); 5958 5959 spdk_blob_close(clone, bs_clone_snapshot_origblob_cleanup, ctx); 5960 } 5961 5962 static void 5963 bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 5964 { 5965 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5966 5967 ctx->cpl.u.blobid.blobid = blobid; 5968 spdk_bs_open_blob(ctx->original.blob->bs, blobid, bs_clone_newblob_open_cpl, ctx); 5969 } 5970 5971 static void 5972 bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5973 { 5974 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5975 struct spdk_blob_opts opts; 5976 struct spdk_blob_xattr_opts internal_xattrs; 5977 char *xattr_names[] = { BLOB_SNAPSHOT }; 5978 5979 if (bserrno != 0) { 5980 bs_clone_snapshot_cleanup_finish(ctx, bserrno); 5981 return; 5982 } 5983 5984 ctx->original.blob = _blob; 5985 5986 if (!_blob->data_ro || !_blob->md_ro) { 5987 SPDK_DEBUGLOG(blob, "Clone not from read-only blob\n"); 5988 ctx->bserrno = -EINVAL; 5989 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5990 return; 5991 } 5992 5993 if (_blob->locked_operation_in_progress) { 5994 SPDK_DEBUGLOG(blob, "Cannot create clone - another operation in progress\n"); 5995 ctx->bserrno = -EBUSY; 5996 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5997 return; 5998 } 5999 6000 _blob->locked_operation_in_progress = true; 6001 6002 spdk_blob_opts_init(&opts, sizeof(opts)); 6003 blob_xattrs_init(&internal_xattrs); 6004 6005 opts.thin_provision = true; 6006 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 6007 opts.use_extent_table = _blob->use_extent_table; 6008 if (ctx->xattrs) { 6009 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 6010 } 6011 6012 /* Set internal xattr BLOB_SNAPSHOT */ 6013 internal_xattrs.count = 1; 6014 internal_xattrs.ctx = _blob; 6015 internal_xattrs.names = xattr_names; 6016 internal_xattrs.get_value = bs_xattr_clone; 6017 6018 bs_create_blob(_blob->bs, &opts, &internal_xattrs, 6019 bs_clone_newblob_create_cpl, ctx); 6020 } 6021 6022 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid, 6023 const struct spdk_blob_xattr_opts *clone_xattrs, 6024 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 6025 { 6026 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 6027 6028 if (!ctx) { 6029 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 6030 return; 6031 } 6032 6033 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 6034 ctx->cpl.u.blobid.cb_fn = cb_fn; 6035 ctx->cpl.u.blobid.cb_arg = cb_arg; 6036 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 6037 ctx->bserrno = 0; 6038 ctx->xattrs = clone_xattrs; 6039 ctx->original.id = blobid; 6040 6041 spdk_bs_open_blob(bs, ctx->original.id, bs_clone_origblob_open_cpl, ctx); 6042 } 6043 6044 /* END spdk_bs_create_clone */ 6045 6046 /* START spdk_bs_inflate_blob */ 6047 6048 static void 6049 bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno) 6050 { 6051 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 6052 struct spdk_blob *_blob = ctx->original.blob; 6053 6054 if (bserrno != 0) { 6055 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 6056 return; 6057 } 6058 6059 assert(_parent != NULL); 6060 6061 bs_blob_list_remove(_blob); 6062 _blob->parent_id = _parent->id; 6063 blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id, 6064 sizeof(spdk_blob_id), true); 6065 6066 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 6067 _blob->back_bs_dev = bs_create_blob_bs_dev(_parent); 6068 bs_blob_list_add(_blob); 6069 6070 spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx); 6071 } 6072 6073 static void 6074 bs_inflate_blob_done(struct spdk_clone_snapshot_ctx *ctx) 6075 { 6076 struct spdk_blob *_blob = ctx->original.blob; 6077 struct spdk_blob *_parent; 6078 6079 if (ctx->allocate_all) { 6080 /* remove thin provisioning */ 6081 bs_blob_list_remove(_blob); 6082 blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 6083 _blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV; 6084 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 6085 _blob->back_bs_dev = NULL; 6086 _blob->parent_id = SPDK_BLOBID_INVALID; 6087 } else { 6088 _parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob; 6089 if (_parent->parent_id != SPDK_BLOBID_INVALID) { 6090 /* We must change the parent of the inflated blob */ 6091 spdk_bs_open_blob(_blob->bs, _parent->parent_id, 6092 bs_inflate_blob_set_parent_cpl, ctx); 6093 return; 6094 } 6095 6096 bs_blob_list_remove(_blob); 6097 blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 6098 _blob->parent_id = SPDK_BLOBID_INVALID; 6099 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 6100 _blob->back_bs_dev = bs_create_zeroes_dev(); 6101 } 6102 6103 _blob->state = SPDK_BLOB_STATE_DIRTY; 6104 spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx); 6105 } 6106 6107 /* Check if cluster needs allocation */ 6108 static inline bool 6109 bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all) 6110 { 6111 struct spdk_blob_bs_dev *b; 6112 6113 assert(blob != NULL); 6114 6115 if (blob->active.clusters[cluster] != 0) { 6116 /* Cluster is already allocated */ 6117 return false; 6118 } 6119 6120 if (blob->parent_id == SPDK_BLOBID_INVALID) { 6121 /* Blob have no parent blob */ 6122 return allocate_all; 6123 } 6124 6125 b = (struct spdk_blob_bs_dev *)blob->back_bs_dev; 6126 return (allocate_all || b->blob->active.clusters[cluster] != 0); 6127 } 6128 6129 static void 6130 bs_inflate_blob_touch_next(void *cb_arg, int bserrno) 6131 { 6132 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 6133 struct spdk_blob *_blob = ctx->original.blob; 6134 uint64_t offset; 6135 6136 if (bserrno != 0) { 6137 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 6138 return; 6139 } 6140 6141 for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) { 6142 if (bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) { 6143 break; 6144 } 6145 } 6146 6147 if (ctx->cluster < _blob->active.num_clusters) { 6148 offset = bs_cluster_to_lba(_blob->bs, ctx->cluster); 6149 6150 /* We may safely increment a cluster before write */ 6151 ctx->cluster++; 6152 6153 /* Use zero length write to touch a cluster */ 6154 spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0, 6155 bs_inflate_blob_touch_next, ctx); 6156 } else { 6157 bs_inflate_blob_done(ctx); 6158 } 6159 } 6160 6161 static void 6162 bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 6163 { 6164 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 6165 uint64_t clusters_needed; 6166 uint64_t i; 6167 6168 if (bserrno != 0) { 6169 bs_clone_snapshot_cleanup_finish(ctx, bserrno); 6170 return; 6171 } 6172 6173 ctx->original.blob = _blob; 6174 6175 if (_blob->locked_operation_in_progress) { 6176 SPDK_DEBUGLOG(blob, "Cannot inflate blob - another operation in progress\n"); 6177 ctx->bserrno = -EBUSY; 6178 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 6179 return; 6180 } 6181 6182 _blob->locked_operation_in_progress = true; 6183 6184 if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) { 6185 /* This blob have no parent, so we cannot decouple it. */ 6186 SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n"); 6187 bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL); 6188 return; 6189 } 6190 6191 if (spdk_blob_is_thin_provisioned(_blob) == false) { 6192 /* This is not thin provisioned blob. No need to inflate. */ 6193 bs_clone_snapshot_origblob_cleanup(ctx, 0); 6194 return; 6195 } 6196 6197 /* Do two passes - one to verify that we can obtain enough clusters 6198 * and another to actually claim them. 6199 */ 6200 clusters_needed = 0; 6201 for (i = 0; i < _blob->active.num_clusters; i++) { 6202 if (bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) { 6203 clusters_needed++; 6204 } 6205 } 6206 6207 if (clusters_needed > _blob->bs->num_free_clusters) { 6208 /* Not enough free clusters. Cannot satisfy the request. */ 6209 bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC); 6210 return; 6211 } 6212 6213 ctx->cluster = 0; 6214 bs_inflate_blob_touch_next(ctx, 0); 6215 } 6216 6217 static void 6218 bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 6219 spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg) 6220 { 6221 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 6222 6223 if (!ctx) { 6224 cb_fn(cb_arg, -ENOMEM); 6225 return; 6226 } 6227 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 6228 ctx->cpl.u.bs_basic.cb_fn = cb_fn; 6229 ctx->cpl.u.bs_basic.cb_arg = cb_arg; 6230 ctx->bserrno = 0; 6231 ctx->original.id = blobid; 6232 ctx->channel = channel; 6233 ctx->allocate_all = allocate_all; 6234 6235 spdk_bs_open_blob(bs, ctx->original.id, bs_inflate_blob_open_cpl, ctx); 6236 } 6237 6238 void 6239 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 6240 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 6241 { 6242 bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg); 6243 } 6244 6245 void 6246 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 6247 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 6248 { 6249 bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg); 6250 } 6251 /* END spdk_bs_inflate_blob */ 6252 6253 /* START spdk_blob_resize */ 6254 struct spdk_bs_resize_ctx { 6255 spdk_blob_op_complete cb_fn; 6256 void *cb_arg; 6257 struct spdk_blob *blob; 6258 uint64_t sz; 6259 int rc; 6260 }; 6261 6262 static void 6263 bs_resize_unfreeze_cpl(void *cb_arg, int rc) 6264 { 6265 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 6266 6267 if (rc != 0) { 6268 SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc); 6269 } 6270 6271 if (ctx->rc != 0) { 6272 SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc); 6273 rc = ctx->rc; 6274 } 6275 6276 ctx->blob->locked_operation_in_progress = false; 6277 6278 ctx->cb_fn(ctx->cb_arg, rc); 6279 free(ctx); 6280 } 6281 6282 static void 6283 bs_resize_freeze_cpl(void *cb_arg, int rc) 6284 { 6285 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 6286 6287 if (rc != 0) { 6288 ctx->blob->locked_operation_in_progress = false; 6289 ctx->cb_fn(ctx->cb_arg, rc); 6290 free(ctx); 6291 return; 6292 } 6293 6294 ctx->rc = blob_resize(ctx->blob, ctx->sz); 6295 6296 blob_unfreeze_io(ctx->blob, bs_resize_unfreeze_cpl, ctx); 6297 } 6298 6299 void 6300 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg) 6301 { 6302 struct spdk_bs_resize_ctx *ctx; 6303 6304 blob_verify_md_op(blob); 6305 6306 SPDK_DEBUGLOG(blob, "Resizing blob %" PRIu64 " to %" PRIu64 " clusters\n", blob->id, sz); 6307 6308 if (blob->md_ro) { 6309 cb_fn(cb_arg, -EPERM); 6310 return; 6311 } 6312 6313 if (sz == blob->active.num_clusters) { 6314 cb_fn(cb_arg, 0); 6315 return; 6316 } 6317 6318 if (blob->locked_operation_in_progress) { 6319 cb_fn(cb_arg, -EBUSY); 6320 return; 6321 } 6322 6323 ctx = calloc(1, sizeof(*ctx)); 6324 if (!ctx) { 6325 cb_fn(cb_arg, -ENOMEM); 6326 return; 6327 } 6328 6329 blob->locked_operation_in_progress = true; 6330 ctx->cb_fn = cb_fn; 6331 ctx->cb_arg = cb_arg; 6332 ctx->blob = blob; 6333 ctx->sz = sz; 6334 blob_freeze_io(blob, bs_resize_freeze_cpl, ctx); 6335 } 6336 6337 /* END spdk_blob_resize */ 6338 6339 6340 /* START spdk_bs_delete_blob */ 6341 6342 static void 6343 bs_delete_close_cpl(void *cb_arg, int bserrno) 6344 { 6345 spdk_bs_sequence_t *seq = cb_arg; 6346 6347 bs_sequence_finish(seq, bserrno); 6348 } 6349 6350 static void 6351 bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 6352 { 6353 struct spdk_blob *blob = cb_arg; 6354 6355 if (bserrno != 0) { 6356 /* 6357 * We already removed this blob from the blobstore tailq, so 6358 * we need to free it here since this is the last reference 6359 * to it. 6360 */ 6361 blob_free(blob); 6362 bs_delete_close_cpl(seq, bserrno); 6363 return; 6364 } 6365 6366 /* 6367 * This will immediately decrement the ref_count and call 6368 * the completion routine since the metadata state is clean. 6369 * By calling spdk_blob_close, we reduce the number of call 6370 * points into code that touches the blob->open_ref count 6371 * and the blobstore's blob list. 6372 */ 6373 spdk_blob_close(blob, bs_delete_close_cpl, seq); 6374 } 6375 6376 struct delete_snapshot_ctx { 6377 struct spdk_blob_list *parent_snapshot_entry; 6378 struct spdk_blob *snapshot; 6379 bool snapshot_md_ro; 6380 struct spdk_blob *clone; 6381 bool clone_md_ro; 6382 spdk_blob_op_with_handle_complete cb_fn; 6383 void *cb_arg; 6384 int bserrno; 6385 uint32_t next_extent_page; 6386 }; 6387 6388 static void 6389 delete_blob_cleanup_finish(void *cb_arg, int bserrno) 6390 { 6391 struct delete_snapshot_ctx *ctx = cb_arg; 6392 6393 if (bserrno != 0) { 6394 SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno); 6395 } 6396 6397 assert(ctx != NULL); 6398 6399 if (bserrno != 0 && ctx->bserrno == 0) { 6400 ctx->bserrno = bserrno; 6401 } 6402 6403 ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno); 6404 free(ctx); 6405 } 6406 6407 static void 6408 delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno) 6409 { 6410 struct delete_snapshot_ctx *ctx = cb_arg; 6411 6412 if (bserrno != 0) { 6413 ctx->bserrno = bserrno; 6414 SPDK_ERRLOG("Clone cleanup error %d\n", bserrno); 6415 } 6416 6417 if (ctx->bserrno != 0) { 6418 assert(blob_lookup(ctx->snapshot->bs, ctx->snapshot->id) == NULL); 6419 TAILQ_INSERT_HEAD(&ctx->snapshot->bs->blobs, ctx->snapshot, link); 6420 spdk_bit_array_set(ctx->snapshot->bs->open_blobids, ctx->snapshot->id); 6421 } 6422 6423 ctx->snapshot->locked_operation_in_progress = false; 6424 ctx->snapshot->md_ro = ctx->snapshot_md_ro; 6425 6426 spdk_blob_close(ctx->snapshot, delete_blob_cleanup_finish, ctx); 6427 } 6428 6429 static void 6430 delete_snapshot_cleanup_clone(void *cb_arg, int bserrno) 6431 { 6432 struct delete_snapshot_ctx *ctx = cb_arg; 6433 6434 ctx->clone->locked_operation_in_progress = false; 6435 ctx->clone->md_ro = ctx->clone_md_ro; 6436 6437 spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx); 6438 } 6439 6440 static void 6441 delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno) 6442 { 6443 struct delete_snapshot_ctx *ctx = cb_arg; 6444 6445 if (bserrno) { 6446 ctx->bserrno = bserrno; 6447 delete_snapshot_cleanup_clone(ctx, 0); 6448 return; 6449 } 6450 6451 ctx->clone->locked_operation_in_progress = false; 6452 spdk_blob_close(ctx->clone, delete_blob_cleanup_finish, ctx); 6453 } 6454 6455 static void 6456 delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno) 6457 { 6458 struct delete_snapshot_ctx *ctx = cb_arg; 6459 struct spdk_blob_list *parent_snapshot_entry = NULL; 6460 struct spdk_blob_list *snapshot_entry = NULL; 6461 struct spdk_blob_list *clone_entry = NULL; 6462 struct spdk_blob_list *snapshot_clone_entry = NULL; 6463 6464 if (bserrno) { 6465 SPDK_ERRLOG("Failed to sync MD on blob\n"); 6466 ctx->bserrno = bserrno; 6467 delete_snapshot_cleanup_clone(ctx, 0); 6468 return; 6469 } 6470 6471 /* Get snapshot entry for the snapshot we want to remove */ 6472 snapshot_entry = bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id); 6473 6474 assert(snapshot_entry != NULL); 6475 6476 /* Remove clone entry in this snapshot (at this point there can be only one clone) */ 6477 clone_entry = TAILQ_FIRST(&snapshot_entry->clones); 6478 assert(clone_entry != NULL); 6479 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 6480 snapshot_entry->clone_count--; 6481 assert(TAILQ_EMPTY(&snapshot_entry->clones)); 6482 6483 if (ctx->snapshot->parent_id != SPDK_BLOBID_INVALID) { 6484 /* This snapshot is at the same time a clone of another snapshot - we need to 6485 * update parent snapshot (remove current clone, add new one inherited from 6486 * the snapshot that is being removed) */ 6487 6488 /* Get snapshot entry for parent snapshot and clone entry within that snapshot for 6489 * snapshot that we are removing */ 6490 blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry, 6491 &snapshot_clone_entry); 6492 6493 /* Switch clone entry in parent snapshot */ 6494 TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link); 6495 TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link); 6496 free(snapshot_clone_entry); 6497 } else { 6498 /* No parent snapshot - just remove clone entry */ 6499 free(clone_entry); 6500 } 6501 6502 /* Restore md_ro flags */ 6503 ctx->clone->md_ro = ctx->clone_md_ro; 6504 ctx->snapshot->md_ro = ctx->snapshot_md_ro; 6505 6506 blob_unfreeze_io(ctx->clone, delete_snapshot_unfreeze_cpl, ctx); 6507 } 6508 6509 static void 6510 delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno) 6511 { 6512 struct delete_snapshot_ctx *ctx = cb_arg; 6513 uint64_t i; 6514 6515 ctx->snapshot->md_ro = false; 6516 6517 if (bserrno) { 6518 SPDK_ERRLOG("Failed to sync MD on clone\n"); 6519 ctx->bserrno = bserrno; 6520 6521 /* Restore snapshot to previous state */ 6522 bserrno = blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true); 6523 if (bserrno != 0) { 6524 delete_snapshot_cleanup_clone(ctx, bserrno); 6525 return; 6526 } 6527 6528 spdk_blob_sync_md(ctx->snapshot, delete_snapshot_cleanup_clone, ctx); 6529 return; 6530 } 6531 6532 /* Clear cluster map entries for snapshot */ 6533 for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) { 6534 if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) { 6535 ctx->snapshot->active.clusters[i] = 0; 6536 } 6537 } 6538 for (i = 0; i < ctx->snapshot->active.num_extent_pages && 6539 i < ctx->clone->active.num_extent_pages; i++) { 6540 if (ctx->clone->active.extent_pages[i] == ctx->snapshot->active.extent_pages[i]) { 6541 ctx->snapshot->active.extent_pages[i] = 0; 6542 } 6543 } 6544 6545 blob_set_thin_provision(ctx->snapshot); 6546 ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY; 6547 6548 if (ctx->parent_snapshot_entry != NULL) { 6549 ctx->snapshot->back_bs_dev = NULL; 6550 } 6551 6552 spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_cpl, ctx); 6553 } 6554 6555 static void 6556 delete_snapshot_update_extent_pages_cpl(struct delete_snapshot_ctx *ctx) 6557 { 6558 /* Delete old backing bs_dev from clone (related to snapshot that will be removed) */ 6559 ctx->clone->back_bs_dev->destroy(ctx->clone->back_bs_dev); 6560 6561 /* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */ 6562 if (ctx->parent_snapshot_entry != NULL) { 6563 /* ...to parent snapshot */ 6564 ctx->clone->parent_id = ctx->parent_snapshot_entry->id; 6565 ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev; 6566 blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id, 6567 sizeof(spdk_blob_id), 6568 true); 6569 } else { 6570 /* ...to blobid invalid and zeroes dev */ 6571 ctx->clone->parent_id = SPDK_BLOBID_INVALID; 6572 ctx->clone->back_bs_dev = bs_create_zeroes_dev(); 6573 blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true); 6574 } 6575 6576 spdk_blob_sync_md(ctx->clone, delete_snapshot_sync_clone_cpl, ctx); 6577 } 6578 6579 static void 6580 delete_snapshot_update_extent_pages(void *cb_arg, int bserrno) 6581 { 6582 struct delete_snapshot_ctx *ctx = cb_arg; 6583 uint32_t *extent_page; 6584 uint64_t i; 6585 6586 for (i = ctx->next_extent_page; i < ctx->snapshot->active.num_extent_pages && 6587 i < ctx->clone->active.num_extent_pages; i++) { 6588 if (ctx->snapshot->active.extent_pages[i] == 0) { 6589 /* No extent page to use from snapshot */ 6590 continue; 6591 } 6592 6593 extent_page = &ctx->clone->active.extent_pages[i]; 6594 if (*extent_page == 0) { 6595 /* Copy extent page from snapshot when clone did not have a matching one */ 6596 *extent_page = ctx->snapshot->active.extent_pages[i]; 6597 continue; 6598 } 6599 6600 /* Clone and snapshot both contain partialy filled matching extent pages. 6601 * Update the clone extent page in place with cluster map containing the mix of both. */ 6602 ctx->next_extent_page = i + 1; 6603 6604 blob_write_extent_page(ctx->clone, *extent_page, i * SPDK_EXTENTS_PER_EP, 6605 delete_snapshot_update_extent_pages, ctx); 6606 return; 6607 } 6608 delete_snapshot_update_extent_pages_cpl(ctx); 6609 } 6610 6611 static void 6612 delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno) 6613 { 6614 struct delete_snapshot_ctx *ctx = cb_arg; 6615 uint64_t i; 6616 6617 /* Temporarily override md_ro flag for clone for MD modification */ 6618 ctx->clone_md_ro = ctx->clone->md_ro; 6619 ctx->clone->md_ro = false; 6620 6621 if (bserrno) { 6622 SPDK_ERRLOG("Failed to sync MD with xattr on blob\n"); 6623 ctx->bserrno = bserrno; 6624 delete_snapshot_cleanup_clone(ctx, 0); 6625 return; 6626 } 6627 6628 /* Copy snapshot map to clone map (only unallocated clusters in clone) */ 6629 for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) { 6630 if (ctx->clone->active.clusters[i] == 0) { 6631 ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i]; 6632 } 6633 } 6634 ctx->next_extent_page = 0; 6635 delete_snapshot_update_extent_pages(ctx, 0); 6636 } 6637 6638 static void 6639 delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno) 6640 { 6641 struct delete_snapshot_ctx *ctx = cb_arg; 6642 6643 if (bserrno) { 6644 SPDK_ERRLOG("Failed to freeze I/O on clone\n"); 6645 ctx->bserrno = bserrno; 6646 delete_snapshot_cleanup_clone(ctx, 0); 6647 return; 6648 } 6649 6650 /* Temporarily override md_ro flag for snapshot for MD modification */ 6651 ctx->snapshot_md_ro = ctx->snapshot->md_ro; 6652 ctx->snapshot->md_ro = false; 6653 6654 /* Mark blob as pending for removal for power failure safety, use clone id for recovery */ 6655 ctx->bserrno = blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id, 6656 sizeof(spdk_blob_id), true); 6657 if (ctx->bserrno != 0) { 6658 delete_snapshot_cleanup_clone(ctx, 0); 6659 return; 6660 } 6661 6662 spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_xattr_cpl, ctx); 6663 } 6664 6665 static void 6666 delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno) 6667 { 6668 struct delete_snapshot_ctx *ctx = cb_arg; 6669 6670 if (bserrno) { 6671 SPDK_ERRLOG("Failed to open clone\n"); 6672 ctx->bserrno = bserrno; 6673 delete_snapshot_cleanup_snapshot(ctx, 0); 6674 return; 6675 } 6676 6677 ctx->clone = clone; 6678 6679 if (clone->locked_operation_in_progress) { 6680 SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress on its clone\n"); 6681 ctx->bserrno = -EBUSY; 6682 spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx); 6683 return; 6684 } 6685 6686 clone->locked_operation_in_progress = true; 6687 6688 blob_freeze_io(clone, delete_snapshot_freeze_io_cb, ctx); 6689 } 6690 6691 static void 6692 update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx) 6693 { 6694 struct spdk_blob_list *snapshot_entry = NULL; 6695 struct spdk_blob_list *clone_entry = NULL; 6696 struct spdk_blob_list *snapshot_clone_entry = NULL; 6697 6698 /* Get snapshot entry for the snapshot we want to remove */ 6699 snapshot_entry = bs_get_snapshot_entry(snapshot->bs, snapshot->id); 6700 6701 assert(snapshot_entry != NULL); 6702 6703 /* Get clone of the snapshot (at this point there can be only one clone) */ 6704 clone_entry = TAILQ_FIRST(&snapshot_entry->clones); 6705 assert(snapshot_entry->clone_count == 1); 6706 assert(clone_entry != NULL); 6707 6708 /* Get snapshot entry for parent snapshot and clone entry within that snapshot for 6709 * snapshot that we are removing */ 6710 blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry, 6711 &snapshot_clone_entry); 6712 6713 spdk_bs_open_blob(snapshot->bs, clone_entry->id, delete_snapshot_open_clone_cb, ctx); 6714 } 6715 6716 static void 6717 bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno) 6718 { 6719 spdk_bs_sequence_t *seq = cb_arg; 6720 struct spdk_blob_list *snapshot_entry = NULL; 6721 uint32_t page_num; 6722 6723 if (bserrno) { 6724 SPDK_ERRLOG("Failed to remove blob\n"); 6725 bs_sequence_finish(seq, bserrno); 6726 return; 6727 } 6728 6729 /* Remove snapshot from the list */ 6730 snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id); 6731 if (snapshot_entry != NULL) { 6732 TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link); 6733 free(snapshot_entry); 6734 } 6735 6736 page_num = bs_blobid_to_page(blob->id); 6737 spdk_bit_array_clear(blob->bs->used_blobids, page_num); 6738 blob->state = SPDK_BLOB_STATE_DIRTY; 6739 blob->active.num_pages = 0; 6740 blob_resize(blob, 0); 6741 6742 blob_persist(seq, blob, bs_delete_persist_cpl, blob); 6743 } 6744 6745 static int 6746 bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone) 6747 { 6748 struct spdk_blob_list *snapshot_entry = NULL; 6749 struct spdk_blob_list *clone_entry = NULL; 6750 struct spdk_blob *clone = NULL; 6751 bool has_one_clone = false; 6752 6753 /* Check if this is a snapshot with clones */ 6754 snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id); 6755 if (snapshot_entry != NULL) { 6756 if (snapshot_entry->clone_count > 1) { 6757 SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n"); 6758 return -EBUSY; 6759 } else if (snapshot_entry->clone_count == 1) { 6760 has_one_clone = true; 6761 } 6762 } 6763 6764 /* Check if someone has this blob open (besides this delete context): 6765 * - open_ref = 1 - only this context opened blob, so it is ok to remove it 6766 * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot 6767 * and that is ok, because we will update it accordingly */ 6768 if (blob->open_ref <= 2 && has_one_clone) { 6769 clone_entry = TAILQ_FIRST(&snapshot_entry->clones); 6770 assert(clone_entry != NULL); 6771 clone = blob_lookup(blob->bs, clone_entry->id); 6772 6773 if (blob->open_ref == 2 && clone == NULL) { 6774 /* Clone is closed and someone else opened this blob */ 6775 SPDK_ERRLOG("Cannot remove snapshot because it is open\n"); 6776 return -EBUSY; 6777 } 6778 6779 *update_clone = true; 6780 return 0; 6781 } 6782 6783 if (blob->open_ref > 1) { 6784 SPDK_ERRLOG("Cannot remove snapshot because it is open\n"); 6785 return -EBUSY; 6786 } 6787 6788 assert(has_one_clone == false); 6789 *update_clone = false; 6790 return 0; 6791 } 6792 6793 static void 6794 bs_delete_enomem_close_cpl(void *cb_arg, int bserrno) 6795 { 6796 spdk_bs_sequence_t *seq = cb_arg; 6797 6798 bs_sequence_finish(seq, -ENOMEM); 6799 } 6800 6801 static void 6802 bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno) 6803 { 6804 spdk_bs_sequence_t *seq = cb_arg; 6805 struct delete_snapshot_ctx *ctx; 6806 bool update_clone = false; 6807 6808 if (bserrno != 0) { 6809 bs_sequence_finish(seq, bserrno); 6810 return; 6811 } 6812 6813 blob_verify_md_op(blob); 6814 6815 ctx = calloc(1, sizeof(*ctx)); 6816 if (ctx == NULL) { 6817 spdk_blob_close(blob, bs_delete_enomem_close_cpl, seq); 6818 return; 6819 } 6820 6821 ctx->snapshot = blob; 6822 ctx->cb_fn = bs_delete_blob_finish; 6823 ctx->cb_arg = seq; 6824 6825 /* Check if blob can be removed and if it is a snapshot with clone on top of it */ 6826 ctx->bserrno = bs_is_blob_deletable(blob, &update_clone); 6827 if (ctx->bserrno) { 6828 spdk_blob_close(blob, delete_blob_cleanup_finish, ctx); 6829 return; 6830 } 6831 6832 if (blob->locked_operation_in_progress) { 6833 SPDK_DEBUGLOG(blob, "Cannot remove blob - another operation in progress\n"); 6834 ctx->bserrno = -EBUSY; 6835 spdk_blob_close(blob, delete_blob_cleanup_finish, ctx); 6836 return; 6837 } 6838 6839 blob->locked_operation_in_progress = true; 6840 6841 /* 6842 * Remove the blob from the blob_store list now, to ensure it does not 6843 * get returned after this point by blob_lookup(). 6844 */ 6845 spdk_bit_array_clear(blob->bs->open_blobids, blob->id); 6846 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 6847 6848 if (update_clone) { 6849 /* This blob is a snapshot with active clone - update clone first */ 6850 update_clone_on_snapshot_deletion(blob, ctx); 6851 } else { 6852 /* This blob does not have any clones - just remove it */ 6853 bs_blob_list_remove(blob); 6854 bs_delete_blob_finish(seq, blob, 0); 6855 free(ctx); 6856 } 6857 } 6858 6859 void 6860 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 6861 spdk_blob_op_complete cb_fn, void *cb_arg) 6862 { 6863 struct spdk_bs_cpl cpl; 6864 spdk_bs_sequence_t *seq; 6865 6866 SPDK_DEBUGLOG(blob, "Deleting blob %" PRIu64 "\n", blobid); 6867 6868 assert(spdk_get_thread() == bs->md_thread); 6869 6870 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 6871 cpl.u.blob_basic.cb_fn = cb_fn; 6872 cpl.u.blob_basic.cb_arg = cb_arg; 6873 6874 seq = bs_sequence_start(bs->md_channel, &cpl); 6875 if (!seq) { 6876 cb_fn(cb_arg, -ENOMEM); 6877 return; 6878 } 6879 6880 spdk_bs_open_blob(bs, blobid, bs_delete_open_cpl, seq); 6881 } 6882 6883 /* END spdk_bs_delete_blob */ 6884 6885 /* START spdk_bs_open_blob */ 6886 6887 static void 6888 bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 6889 { 6890 struct spdk_blob *blob = cb_arg; 6891 struct spdk_blob *existing; 6892 6893 if (bserrno != 0) { 6894 blob_free(blob); 6895 seq->cpl.u.blob_handle.blob = NULL; 6896 bs_sequence_finish(seq, bserrno); 6897 return; 6898 } 6899 6900 existing = blob_lookup(blob->bs, blob->id); 6901 if (existing) { 6902 blob_free(blob); 6903 existing->open_ref++; 6904 seq->cpl.u.blob_handle.blob = existing; 6905 bs_sequence_finish(seq, 0); 6906 return; 6907 } 6908 6909 blob->open_ref++; 6910 6911 spdk_bit_array_set(blob->bs->open_blobids, blob->id); 6912 TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link); 6913 6914 bs_sequence_finish(seq, bserrno); 6915 } 6916 6917 static inline void 6918 blob_open_opts_copy(const struct spdk_blob_open_opts *src, struct spdk_blob_open_opts *dst) 6919 { 6920 #define FIELD_OK(field) \ 6921 offsetof(struct spdk_blob_opts, field) + sizeof(src->field) <= src->opts_size 6922 6923 #define SET_FIELD(field) \ 6924 if (FIELD_OK(field)) { \ 6925 dst->field = src->field; \ 6926 } \ 6927 6928 SET_FIELD(clear_method); 6929 6930 dst->opts_size = src->opts_size; 6931 6932 /* You should not remove this statement, but need to update the assert statement 6933 * if you add a new field, and also add a corresponding SET_FIELD statement */ 6934 SPDK_STATIC_ASSERT(sizeof(struct spdk_blob_open_opts) == 16, "Incorrect size"); 6935 6936 #undef FIELD_OK 6937 #undef SET_FIELD 6938 } 6939 6940 static void 6941 bs_open_blob(struct spdk_blob_store *bs, 6942 spdk_blob_id blobid, 6943 struct spdk_blob_open_opts *opts, 6944 spdk_blob_op_with_handle_complete cb_fn, 6945 void *cb_arg) 6946 { 6947 struct spdk_blob *blob; 6948 struct spdk_bs_cpl cpl; 6949 struct spdk_blob_open_opts opts_local; 6950 spdk_bs_sequence_t *seq; 6951 uint32_t page_num; 6952 6953 SPDK_DEBUGLOG(blob, "Opening blob %" PRIu64 "\n", blobid); 6954 assert(spdk_get_thread() == bs->md_thread); 6955 6956 page_num = bs_blobid_to_page(blobid); 6957 if (spdk_bit_array_get(bs->used_blobids, page_num) == false) { 6958 /* Invalid blobid */ 6959 cb_fn(cb_arg, NULL, -ENOENT); 6960 return; 6961 } 6962 6963 blob = blob_lookup(bs, blobid); 6964 if (blob) { 6965 blob->open_ref++; 6966 cb_fn(cb_arg, blob, 0); 6967 return; 6968 } 6969 6970 blob = blob_alloc(bs, blobid); 6971 if (!blob) { 6972 cb_fn(cb_arg, NULL, -ENOMEM); 6973 return; 6974 } 6975 6976 spdk_blob_open_opts_init(&opts_local, sizeof(opts_local)); 6977 if (opts) { 6978 blob_open_opts_copy(opts, &opts_local); 6979 } 6980 6981 blob->clear_method = opts_local.clear_method; 6982 6983 cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE; 6984 cpl.u.blob_handle.cb_fn = cb_fn; 6985 cpl.u.blob_handle.cb_arg = cb_arg; 6986 cpl.u.blob_handle.blob = blob; 6987 6988 seq = bs_sequence_start(bs->md_channel, &cpl); 6989 if (!seq) { 6990 blob_free(blob); 6991 cb_fn(cb_arg, NULL, -ENOMEM); 6992 return; 6993 } 6994 6995 blob_load(seq, blob, bs_open_blob_cpl, blob); 6996 } 6997 6998 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 6999 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 7000 { 7001 bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg); 7002 } 7003 7004 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid, 7005 struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 7006 { 7007 bs_open_blob(bs, blobid, opts, cb_fn, cb_arg); 7008 } 7009 7010 /* END spdk_bs_open_blob */ 7011 7012 /* START spdk_blob_set_read_only */ 7013 int spdk_blob_set_read_only(struct spdk_blob *blob) 7014 { 7015 blob_verify_md_op(blob); 7016 7017 blob->data_ro_flags |= SPDK_BLOB_READ_ONLY; 7018 7019 blob->state = SPDK_BLOB_STATE_DIRTY; 7020 return 0; 7021 } 7022 /* END spdk_blob_set_read_only */ 7023 7024 /* START spdk_blob_sync_md */ 7025 7026 static void 7027 blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 7028 { 7029 struct spdk_blob *blob = cb_arg; 7030 7031 if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 7032 blob->data_ro = true; 7033 blob->md_ro = true; 7034 } 7035 7036 bs_sequence_finish(seq, bserrno); 7037 } 7038 7039 static void 7040 blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 7041 { 7042 struct spdk_bs_cpl cpl; 7043 spdk_bs_sequence_t *seq; 7044 7045 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 7046 cpl.u.blob_basic.cb_fn = cb_fn; 7047 cpl.u.blob_basic.cb_arg = cb_arg; 7048 7049 seq = bs_sequence_start(blob->bs->md_channel, &cpl); 7050 if (!seq) { 7051 cb_fn(cb_arg, -ENOMEM); 7052 return; 7053 } 7054 7055 blob_persist(seq, blob, blob_sync_md_cpl, blob); 7056 } 7057 7058 void 7059 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 7060 { 7061 blob_verify_md_op(blob); 7062 7063 SPDK_DEBUGLOG(blob, "Syncing blob %" PRIu64 "\n", blob->id); 7064 7065 if (blob->md_ro) { 7066 assert(blob->state == SPDK_BLOB_STATE_CLEAN); 7067 cb_fn(cb_arg, 0); 7068 return; 7069 } 7070 7071 blob_sync_md(blob, cb_fn, cb_arg); 7072 } 7073 7074 /* END spdk_blob_sync_md */ 7075 7076 struct spdk_blob_insert_cluster_ctx { 7077 struct spdk_thread *thread; 7078 struct spdk_blob *blob; 7079 uint32_t cluster_num; /* cluster index in blob */ 7080 uint32_t cluster; /* cluster on disk */ 7081 uint32_t extent_page; /* extent page on disk */ 7082 int rc; 7083 spdk_blob_op_complete cb_fn; 7084 void *cb_arg; 7085 }; 7086 7087 static void 7088 blob_insert_cluster_msg_cpl(void *arg) 7089 { 7090 struct spdk_blob_insert_cluster_ctx *ctx = arg; 7091 7092 ctx->cb_fn(ctx->cb_arg, ctx->rc); 7093 free(ctx); 7094 } 7095 7096 static void 7097 blob_insert_cluster_msg_cb(void *arg, int bserrno) 7098 { 7099 struct spdk_blob_insert_cluster_ctx *ctx = arg; 7100 7101 ctx->rc = bserrno; 7102 spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx); 7103 } 7104 7105 static void 7106 blob_insert_new_ep_cb(void *arg, int bserrno) 7107 { 7108 struct spdk_blob_insert_cluster_ctx *ctx = arg; 7109 uint32_t *extent_page; 7110 7111 extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num); 7112 *extent_page = ctx->extent_page; 7113 ctx->blob->state = SPDK_BLOB_STATE_DIRTY; 7114 blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx); 7115 } 7116 7117 static void 7118 blob_persist_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 7119 { 7120 struct spdk_blob_md_page *page = cb_arg; 7121 7122 bs_sequence_finish(seq, bserrno); 7123 spdk_free(page); 7124 } 7125 7126 static void 7127 blob_write_extent_page(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num, 7128 spdk_blob_op_complete cb_fn, void *cb_arg) 7129 { 7130 spdk_bs_sequence_t *seq; 7131 struct spdk_bs_cpl cpl; 7132 struct spdk_blob_md_page *page = NULL; 7133 uint32_t page_count = 0; 7134 int rc; 7135 7136 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 7137 cpl.u.blob_basic.cb_fn = cb_fn; 7138 cpl.u.blob_basic.cb_arg = cb_arg; 7139 7140 seq = bs_sequence_start(blob->bs->md_channel, &cpl); 7141 if (!seq) { 7142 cb_fn(cb_arg, -ENOMEM); 7143 return; 7144 } 7145 rc = blob_serialize_add_page(blob, &page, &page_count, &page); 7146 if (rc < 0) { 7147 bs_sequence_finish(seq, rc); 7148 return; 7149 } 7150 7151 blob_serialize_extent_page(blob, cluster_num, page); 7152 7153 page->crc = blob_md_page_calc_crc(page); 7154 7155 assert(spdk_bit_array_get(blob->bs->used_md_pages, extent) == true); 7156 7157 bs_sequence_write_dev(seq, page, bs_md_page_to_lba(blob->bs, extent), 7158 bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE), 7159 blob_persist_extent_page_cpl, page); 7160 } 7161 7162 static void 7163 blob_insert_cluster_msg(void *arg) 7164 { 7165 struct spdk_blob_insert_cluster_ctx *ctx = arg; 7166 uint32_t *extent_page; 7167 7168 ctx->rc = blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster); 7169 if (ctx->rc != 0) { 7170 spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx); 7171 return; 7172 } 7173 7174 if (ctx->blob->use_extent_table == false) { 7175 /* Extent table is not used, proceed with sync of md that will only use extents_rle. */ 7176 ctx->blob->state = SPDK_BLOB_STATE_DIRTY; 7177 blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx); 7178 return; 7179 } 7180 7181 extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num); 7182 if (*extent_page == 0) { 7183 /* Extent page requires allocation. 7184 * It was already claimed in the used_md_pages map and placed in ctx. */ 7185 assert(ctx->extent_page != 0); 7186 assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true); 7187 blob_write_extent_page(ctx->blob, ctx->extent_page, ctx->cluster_num, 7188 blob_insert_new_ep_cb, ctx); 7189 } else { 7190 /* It is possible for original thread to allocate extent page for 7191 * different cluster in the same extent page. In such case proceed with 7192 * updating the existing extent page, but release the additional one. */ 7193 if (ctx->extent_page != 0) { 7194 assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true); 7195 bs_release_md_page(ctx->blob->bs, ctx->extent_page); 7196 ctx->extent_page = 0; 7197 } 7198 /* Extent page already allocated. 7199 * Every cluster allocation, requires just an update of single extent page. */ 7200 blob_write_extent_page(ctx->blob, *extent_page, ctx->cluster_num, 7201 blob_insert_cluster_msg_cb, ctx); 7202 } 7203 } 7204 7205 static void 7206 blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 7207 uint64_t cluster, uint32_t extent_page, spdk_blob_op_complete cb_fn, void *cb_arg) 7208 { 7209 struct spdk_blob_insert_cluster_ctx *ctx; 7210 7211 ctx = calloc(1, sizeof(*ctx)); 7212 if (ctx == NULL) { 7213 cb_fn(cb_arg, -ENOMEM); 7214 return; 7215 } 7216 7217 ctx->thread = spdk_get_thread(); 7218 ctx->blob = blob; 7219 ctx->cluster_num = cluster_num; 7220 ctx->cluster = cluster; 7221 ctx->extent_page = extent_page; 7222 ctx->cb_fn = cb_fn; 7223 ctx->cb_arg = cb_arg; 7224 7225 spdk_thread_send_msg(blob->bs->md_thread, blob_insert_cluster_msg, ctx); 7226 } 7227 7228 /* START spdk_blob_close */ 7229 7230 static void 7231 blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 7232 { 7233 struct spdk_blob *blob = cb_arg; 7234 7235 if (bserrno == 0) { 7236 blob->open_ref--; 7237 if (blob->open_ref == 0) { 7238 /* 7239 * Blobs with active.num_pages == 0 are deleted blobs. 7240 * these blobs are removed from the blob_store list 7241 * when the deletion process starts - so don't try to 7242 * remove them again. 7243 */ 7244 if (blob->active.num_pages > 0) { 7245 spdk_bit_array_clear(blob->bs->open_blobids, blob->id); 7246 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 7247 } 7248 blob_free(blob); 7249 } 7250 } 7251 7252 bs_sequence_finish(seq, bserrno); 7253 } 7254 7255 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 7256 { 7257 struct spdk_bs_cpl cpl; 7258 spdk_bs_sequence_t *seq; 7259 7260 blob_verify_md_op(blob); 7261 7262 SPDK_DEBUGLOG(blob, "Closing blob %" PRIu64 "\n", blob->id); 7263 7264 if (blob->open_ref == 0) { 7265 cb_fn(cb_arg, -EBADF); 7266 return; 7267 } 7268 7269 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 7270 cpl.u.blob_basic.cb_fn = cb_fn; 7271 cpl.u.blob_basic.cb_arg = cb_arg; 7272 7273 seq = bs_sequence_start(blob->bs->md_channel, &cpl); 7274 if (!seq) { 7275 cb_fn(cb_arg, -ENOMEM); 7276 return; 7277 } 7278 7279 /* Sync metadata */ 7280 blob_persist(seq, blob, blob_close_cpl, blob); 7281 } 7282 7283 /* END spdk_blob_close */ 7284 7285 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs) 7286 { 7287 return spdk_get_io_channel(bs); 7288 } 7289 7290 void spdk_bs_free_io_channel(struct spdk_io_channel *channel) 7291 { 7292 spdk_put_io_channel(channel); 7293 } 7294 7295 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel, 7296 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 7297 { 7298 blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 7299 SPDK_BLOB_UNMAP); 7300 } 7301 7302 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel, 7303 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 7304 { 7305 blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 7306 SPDK_BLOB_WRITE_ZEROES); 7307 } 7308 7309 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel, 7310 void *payload, uint64_t offset, uint64_t length, 7311 spdk_blob_op_complete cb_fn, void *cb_arg) 7312 { 7313 blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 7314 SPDK_BLOB_WRITE); 7315 } 7316 7317 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel, 7318 void *payload, uint64_t offset, uint64_t length, 7319 spdk_blob_op_complete cb_fn, void *cb_arg) 7320 { 7321 blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 7322 SPDK_BLOB_READ); 7323 } 7324 7325 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel, 7326 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 7327 spdk_blob_op_complete cb_fn, void *cb_arg) 7328 { 7329 blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false); 7330 } 7331 7332 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel, 7333 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 7334 spdk_blob_op_complete cb_fn, void *cb_arg) 7335 { 7336 blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true); 7337 } 7338 7339 struct spdk_bs_iter_ctx { 7340 int64_t page_num; 7341 struct spdk_blob_store *bs; 7342 7343 spdk_blob_op_with_handle_complete cb_fn; 7344 void *cb_arg; 7345 }; 7346 7347 static void 7348 bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 7349 { 7350 struct spdk_bs_iter_ctx *ctx = cb_arg; 7351 struct spdk_blob_store *bs = ctx->bs; 7352 spdk_blob_id id; 7353 7354 if (bserrno == 0) { 7355 ctx->cb_fn(ctx->cb_arg, _blob, bserrno); 7356 free(ctx); 7357 return; 7358 } 7359 7360 ctx->page_num++; 7361 ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num); 7362 if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) { 7363 ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT); 7364 free(ctx); 7365 return; 7366 } 7367 7368 id = bs_page_to_blobid(ctx->page_num); 7369 7370 spdk_bs_open_blob(bs, id, bs_iter_cpl, ctx); 7371 } 7372 7373 void 7374 spdk_bs_iter_first(struct spdk_blob_store *bs, 7375 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 7376 { 7377 struct spdk_bs_iter_ctx *ctx; 7378 7379 ctx = calloc(1, sizeof(*ctx)); 7380 if (!ctx) { 7381 cb_fn(cb_arg, NULL, -ENOMEM); 7382 return; 7383 } 7384 7385 ctx->page_num = -1; 7386 ctx->bs = bs; 7387 ctx->cb_fn = cb_fn; 7388 ctx->cb_arg = cb_arg; 7389 7390 bs_iter_cpl(ctx, NULL, -1); 7391 } 7392 7393 static void 7394 bs_iter_close_cpl(void *cb_arg, int bserrno) 7395 { 7396 struct spdk_bs_iter_ctx *ctx = cb_arg; 7397 7398 bs_iter_cpl(ctx, NULL, -1); 7399 } 7400 7401 void 7402 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob, 7403 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 7404 { 7405 struct spdk_bs_iter_ctx *ctx; 7406 7407 assert(blob != NULL); 7408 7409 ctx = calloc(1, sizeof(*ctx)); 7410 if (!ctx) { 7411 cb_fn(cb_arg, NULL, -ENOMEM); 7412 return; 7413 } 7414 7415 ctx->page_num = bs_blobid_to_page(blob->id); 7416 ctx->bs = bs; 7417 ctx->cb_fn = cb_fn; 7418 ctx->cb_arg = cb_arg; 7419 7420 /* Close the existing blob */ 7421 spdk_blob_close(blob, bs_iter_close_cpl, ctx); 7422 } 7423 7424 static int 7425 blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 7426 uint16_t value_len, bool internal) 7427 { 7428 struct spdk_xattr_tailq *xattrs; 7429 struct spdk_xattr *xattr; 7430 size_t desc_size; 7431 void *tmp; 7432 7433 blob_verify_md_op(blob); 7434 7435 if (blob->md_ro) { 7436 return -EPERM; 7437 } 7438 7439 desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len; 7440 if (desc_size > SPDK_BS_MAX_DESC_SIZE) { 7441 SPDK_DEBUGLOG(blob, "Xattr '%s' of size %zu does not fix into single page %zu\n", name, 7442 desc_size, SPDK_BS_MAX_DESC_SIZE); 7443 return -ENOMEM; 7444 } 7445 7446 if (internal) { 7447 xattrs = &blob->xattrs_internal; 7448 blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR; 7449 } else { 7450 xattrs = &blob->xattrs; 7451 } 7452 7453 TAILQ_FOREACH(xattr, xattrs, link) { 7454 if (!strcmp(name, xattr->name)) { 7455 tmp = malloc(value_len); 7456 if (!tmp) { 7457 return -ENOMEM; 7458 } 7459 7460 free(xattr->value); 7461 xattr->value_len = value_len; 7462 xattr->value = tmp; 7463 memcpy(xattr->value, value, value_len); 7464 7465 blob->state = SPDK_BLOB_STATE_DIRTY; 7466 7467 return 0; 7468 } 7469 } 7470 7471 xattr = calloc(1, sizeof(*xattr)); 7472 if (!xattr) { 7473 return -ENOMEM; 7474 } 7475 7476 xattr->name = strdup(name); 7477 if (!xattr->name) { 7478 free(xattr); 7479 return -ENOMEM; 7480 } 7481 7482 xattr->value_len = value_len; 7483 xattr->value = malloc(value_len); 7484 if (!xattr->value) { 7485 free(xattr->name); 7486 free(xattr); 7487 return -ENOMEM; 7488 } 7489 memcpy(xattr->value, value, value_len); 7490 TAILQ_INSERT_TAIL(xattrs, xattr, link); 7491 7492 blob->state = SPDK_BLOB_STATE_DIRTY; 7493 7494 return 0; 7495 } 7496 7497 int 7498 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 7499 uint16_t value_len) 7500 { 7501 return blob_set_xattr(blob, name, value, value_len, false); 7502 } 7503 7504 static int 7505 blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal) 7506 { 7507 struct spdk_xattr_tailq *xattrs; 7508 struct spdk_xattr *xattr; 7509 7510 blob_verify_md_op(blob); 7511 7512 if (blob->md_ro) { 7513 return -EPERM; 7514 } 7515 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 7516 7517 TAILQ_FOREACH(xattr, xattrs, link) { 7518 if (!strcmp(name, xattr->name)) { 7519 TAILQ_REMOVE(xattrs, xattr, link); 7520 free(xattr->value); 7521 free(xattr->name); 7522 free(xattr); 7523 7524 if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) { 7525 blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR; 7526 } 7527 blob->state = SPDK_BLOB_STATE_DIRTY; 7528 7529 return 0; 7530 } 7531 } 7532 7533 return -ENOENT; 7534 } 7535 7536 int 7537 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name) 7538 { 7539 return blob_remove_xattr(blob, name, false); 7540 } 7541 7542 static int 7543 blob_get_xattr_value(struct spdk_blob *blob, const char *name, 7544 const void **value, size_t *value_len, bool internal) 7545 { 7546 struct spdk_xattr *xattr; 7547 struct spdk_xattr_tailq *xattrs; 7548 7549 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 7550 7551 TAILQ_FOREACH(xattr, xattrs, link) { 7552 if (!strcmp(name, xattr->name)) { 7553 *value = xattr->value; 7554 *value_len = xattr->value_len; 7555 return 0; 7556 } 7557 } 7558 return -ENOENT; 7559 } 7560 7561 int 7562 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 7563 const void **value, size_t *value_len) 7564 { 7565 blob_verify_md_op(blob); 7566 7567 return blob_get_xattr_value(blob, name, value, value_len, false); 7568 } 7569 7570 struct spdk_xattr_names { 7571 uint32_t count; 7572 const char *names[0]; 7573 }; 7574 7575 static int 7576 blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names) 7577 { 7578 struct spdk_xattr *xattr; 7579 int count = 0; 7580 7581 TAILQ_FOREACH(xattr, xattrs, link) { 7582 count++; 7583 } 7584 7585 *names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *)); 7586 if (*names == NULL) { 7587 return -ENOMEM; 7588 } 7589 7590 TAILQ_FOREACH(xattr, xattrs, link) { 7591 (*names)->names[(*names)->count++] = xattr->name; 7592 } 7593 7594 return 0; 7595 } 7596 7597 int 7598 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names) 7599 { 7600 blob_verify_md_op(blob); 7601 7602 return blob_get_xattr_names(&blob->xattrs, names); 7603 } 7604 7605 uint32_t 7606 spdk_xattr_names_get_count(struct spdk_xattr_names *names) 7607 { 7608 assert(names != NULL); 7609 7610 return names->count; 7611 } 7612 7613 const char * 7614 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index) 7615 { 7616 if (index >= names->count) { 7617 return NULL; 7618 } 7619 7620 return names->names[index]; 7621 } 7622 7623 void 7624 spdk_xattr_names_free(struct spdk_xattr_names *names) 7625 { 7626 free(names); 7627 } 7628 7629 struct spdk_bs_type 7630 spdk_bs_get_bstype(struct spdk_blob_store *bs) 7631 { 7632 return bs->bstype; 7633 } 7634 7635 void 7636 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype) 7637 { 7638 memcpy(&bs->bstype, &bstype, sizeof(bstype)); 7639 } 7640 7641 bool 7642 spdk_blob_is_read_only(struct spdk_blob *blob) 7643 { 7644 assert(blob != NULL); 7645 return (blob->data_ro || blob->md_ro); 7646 } 7647 7648 bool 7649 spdk_blob_is_snapshot(struct spdk_blob *blob) 7650 { 7651 struct spdk_blob_list *snapshot_entry; 7652 7653 assert(blob != NULL); 7654 7655 snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id); 7656 if (snapshot_entry == NULL) { 7657 return false; 7658 } 7659 7660 return true; 7661 } 7662 7663 bool 7664 spdk_blob_is_clone(struct spdk_blob *blob) 7665 { 7666 assert(blob != NULL); 7667 7668 if (blob->parent_id != SPDK_BLOBID_INVALID) { 7669 assert(spdk_blob_is_thin_provisioned(blob)); 7670 return true; 7671 } 7672 7673 return false; 7674 } 7675 7676 bool 7677 spdk_blob_is_thin_provisioned(struct spdk_blob *blob) 7678 { 7679 assert(blob != NULL); 7680 return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV); 7681 } 7682 7683 static void 7684 blob_update_clear_method(struct spdk_blob *blob) 7685 { 7686 enum blob_clear_method stored_cm; 7687 7688 assert(blob != NULL); 7689 7690 /* If BLOB_CLEAR_WITH_DEFAULT was passed in, use the setting stored 7691 * in metadata previously. If something other than the default was 7692 * specified, ignore stored value and used what was passed in. 7693 */ 7694 stored_cm = ((blob->md_ro_flags & SPDK_BLOB_CLEAR_METHOD) >> SPDK_BLOB_CLEAR_METHOD_SHIFT); 7695 7696 if (blob->clear_method == BLOB_CLEAR_WITH_DEFAULT) { 7697 blob->clear_method = stored_cm; 7698 } else if (blob->clear_method != stored_cm) { 7699 SPDK_WARNLOG("Using passed in clear method 0x%x instead of stored value of 0x%x\n", 7700 blob->clear_method, stored_cm); 7701 } 7702 } 7703 7704 spdk_blob_id 7705 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id) 7706 { 7707 struct spdk_blob_list *snapshot_entry = NULL; 7708 struct spdk_blob_list *clone_entry = NULL; 7709 7710 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 7711 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 7712 if (clone_entry->id == blob_id) { 7713 return snapshot_entry->id; 7714 } 7715 } 7716 } 7717 7718 return SPDK_BLOBID_INVALID; 7719 } 7720 7721 int 7722 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids, 7723 size_t *count) 7724 { 7725 struct spdk_blob_list *snapshot_entry, *clone_entry; 7726 size_t n; 7727 7728 snapshot_entry = bs_get_snapshot_entry(bs, blobid); 7729 if (snapshot_entry == NULL) { 7730 *count = 0; 7731 return 0; 7732 } 7733 7734 if (ids == NULL || *count < snapshot_entry->clone_count) { 7735 *count = snapshot_entry->clone_count; 7736 return -ENOMEM; 7737 } 7738 *count = snapshot_entry->clone_count; 7739 7740 n = 0; 7741 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 7742 ids[n++] = clone_entry->id; 7743 } 7744 7745 return 0; 7746 } 7747 7748 SPDK_LOG_REGISTER_COMPONENT(blob) 7749