1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/stdinc.h" 35 36 #include "spdk/blob.h" 37 #include "spdk/crc32.h" 38 #include "spdk/env.h" 39 #include "spdk/queue.h" 40 #include "spdk/thread.h" 41 #include "spdk/bit_array.h" 42 #include "spdk/likely.h" 43 #include "spdk/util.h" 44 45 #include "spdk_internal/assert.h" 46 #include "spdk_internal/log.h" 47 48 #include "blobstore.h" 49 50 #define BLOB_CRC32C_INITIAL 0xffffffffUL 51 52 static int spdk_bs_register_md_thread(struct spdk_blob_store *bs); 53 static int spdk_bs_unregister_md_thread(struct spdk_blob_store *bs); 54 static void _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno); 55 static void _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 56 uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg); 57 58 static int _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 59 uint16_t value_len, bool internal); 60 static int _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 61 const void **value, size_t *value_len, bool internal); 62 static int _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal); 63 64 static void 65 _spdk_blob_verify_md_op(struct spdk_blob *blob) 66 { 67 assert(blob != NULL); 68 assert(spdk_get_thread() == blob->bs->md_thread); 69 assert(blob->state != SPDK_BLOB_STATE_LOADING); 70 } 71 72 static struct spdk_blob_list * 73 _spdk_bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid) 74 { 75 struct spdk_blob_list *snapshot_entry = NULL; 76 77 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 78 if (snapshot_entry->id == blobid) { 79 break; 80 } 81 } 82 83 return snapshot_entry; 84 } 85 86 static void 87 _spdk_bs_claim_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 88 { 89 assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters)); 90 assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == false); 91 assert(bs->num_free_clusters > 0); 92 93 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num); 94 95 spdk_bit_array_set(bs->used_clusters, cluster_num); 96 bs->num_free_clusters--; 97 } 98 99 static int 100 _spdk_blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster) 101 { 102 uint64_t *cluster_lba = &blob->active.clusters[cluster_num]; 103 104 _spdk_blob_verify_md_op(blob); 105 106 if (*cluster_lba != 0) { 107 return -EEXIST; 108 } 109 110 *cluster_lba = _spdk_bs_cluster_to_lba(blob->bs, cluster); 111 return 0; 112 } 113 114 static int 115 _spdk_bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num, 116 uint64_t *lowest_free_cluster, bool update_map) 117 { 118 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 119 *lowest_free_cluster = spdk_bit_array_find_first_clear(blob->bs->used_clusters, 120 *lowest_free_cluster); 121 if (*lowest_free_cluster == UINT32_MAX) { 122 /* No more free clusters. Cannot satisfy the request */ 123 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 124 return -ENOSPC; 125 } 126 127 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *lowest_free_cluster, blob->id); 128 _spdk_bs_claim_cluster(blob->bs, *lowest_free_cluster); 129 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 130 131 if (update_map) { 132 _spdk_blob_insert_cluster(blob, cluster_num, *lowest_free_cluster); 133 } 134 135 return 0; 136 } 137 138 static void 139 _spdk_bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 140 { 141 assert(cluster_num < spdk_bit_array_capacity(bs->used_clusters)); 142 assert(spdk_bit_array_get(bs->used_clusters, cluster_num) == true); 143 assert(bs->num_free_clusters < bs->total_clusters); 144 145 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num); 146 147 pthread_mutex_lock(&bs->used_clusters_mutex); 148 spdk_bit_array_clear(bs->used_clusters, cluster_num); 149 bs->num_free_clusters++; 150 pthread_mutex_unlock(&bs->used_clusters_mutex); 151 } 152 153 static void 154 _spdk_blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs) 155 { 156 xattrs->count = 0; 157 xattrs->names = NULL; 158 xattrs->ctx = NULL; 159 xattrs->get_value = NULL; 160 } 161 162 void 163 spdk_blob_opts_init(struct spdk_blob_opts *opts) 164 { 165 opts->num_clusters = 0; 166 opts->thin_provision = false; 167 _spdk_blob_xattrs_init(&opts->xattrs); 168 } 169 170 void 171 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts) 172 { 173 opts->clear_method = BLOB_CLEAR_WITH_UNMAP; 174 } 175 176 static struct spdk_blob * 177 _spdk_blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id) 178 { 179 struct spdk_blob *blob; 180 181 blob = calloc(1, sizeof(*blob)); 182 if (!blob) { 183 return NULL; 184 } 185 186 blob->id = id; 187 blob->bs = bs; 188 189 blob->parent_id = SPDK_BLOBID_INVALID; 190 191 blob->state = SPDK_BLOB_STATE_DIRTY; 192 blob->active.num_pages = 1; 193 blob->active.pages = calloc(1, sizeof(*blob->active.pages)); 194 if (!blob->active.pages) { 195 free(blob); 196 return NULL; 197 } 198 199 blob->active.pages[0] = _spdk_bs_blobid_to_page(id); 200 201 TAILQ_INIT(&blob->xattrs); 202 TAILQ_INIT(&blob->xattrs_internal); 203 204 return blob; 205 } 206 207 static void 208 _spdk_xattrs_free(struct spdk_xattr_tailq *xattrs) 209 { 210 struct spdk_xattr *xattr, *xattr_tmp; 211 212 TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) { 213 TAILQ_REMOVE(xattrs, xattr, link); 214 free(xattr->name); 215 free(xattr->value); 216 free(xattr); 217 } 218 } 219 220 static void 221 _spdk_blob_free(struct spdk_blob *blob) 222 { 223 assert(blob != NULL); 224 225 free(blob->active.clusters); 226 free(blob->clean.clusters); 227 free(blob->active.pages); 228 free(blob->clean.pages); 229 230 _spdk_xattrs_free(&blob->xattrs); 231 _spdk_xattrs_free(&blob->xattrs_internal); 232 233 if (blob->back_bs_dev) { 234 blob->back_bs_dev->destroy(blob->back_bs_dev); 235 } 236 237 free(blob); 238 } 239 240 struct freeze_io_ctx { 241 struct spdk_bs_cpl cpl; 242 struct spdk_blob *blob; 243 }; 244 245 static void 246 _spdk_blob_io_sync(struct spdk_io_channel_iter *i) 247 { 248 spdk_for_each_channel_continue(i, 0); 249 } 250 251 static void 252 _spdk_blob_execute_queued_io(struct spdk_io_channel_iter *i) 253 { 254 struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i); 255 struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch); 256 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 257 struct spdk_bs_request_set *set; 258 struct spdk_bs_user_op_args *args; 259 spdk_bs_user_op_t *op, *tmp; 260 261 TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) { 262 set = (struct spdk_bs_request_set *)op; 263 args = &set->u.user_op; 264 265 if (args->blob == ctx->blob) { 266 TAILQ_REMOVE(&ch->queued_io, op, link); 267 spdk_bs_user_op_execute(op); 268 } 269 } 270 271 spdk_for_each_channel_continue(i, 0); 272 } 273 274 static void 275 _spdk_blob_io_cpl(struct spdk_io_channel_iter *i, int status) 276 { 277 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 278 279 ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0); 280 281 free(ctx); 282 } 283 284 static void 285 _spdk_blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 286 { 287 struct freeze_io_ctx *ctx; 288 289 ctx = calloc(1, sizeof(*ctx)); 290 if (!ctx) { 291 cb_fn(cb_arg, -ENOMEM); 292 return; 293 } 294 295 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 296 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 297 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 298 ctx->blob = blob; 299 300 /* Freeze I/O on blob */ 301 blob->frozen_refcnt++; 302 303 if (blob->frozen_refcnt == 1) { 304 spdk_for_each_channel(blob->bs, _spdk_blob_io_sync, ctx, _spdk_blob_io_cpl); 305 } else { 306 cb_fn(cb_arg, 0); 307 free(ctx); 308 } 309 } 310 311 static void 312 _spdk_blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 313 { 314 struct freeze_io_ctx *ctx; 315 316 ctx = calloc(1, sizeof(*ctx)); 317 if (!ctx) { 318 cb_fn(cb_arg, -ENOMEM); 319 return; 320 } 321 322 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 323 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 324 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 325 ctx->blob = blob; 326 327 assert(blob->frozen_refcnt > 0); 328 329 blob->frozen_refcnt--; 330 331 if (blob->frozen_refcnt == 0) { 332 spdk_for_each_channel(blob->bs, _spdk_blob_execute_queued_io, ctx, _spdk_blob_io_cpl); 333 } else { 334 cb_fn(cb_arg, 0); 335 free(ctx); 336 } 337 } 338 339 static int 340 _spdk_blob_mark_clean(struct spdk_blob *blob) 341 { 342 uint64_t *clusters = NULL; 343 uint32_t *pages = NULL; 344 345 assert(blob != NULL); 346 347 if (blob->active.num_clusters) { 348 assert(blob->active.clusters); 349 clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters)); 350 if (!clusters) { 351 return -ENOMEM; 352 } 353 memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*clusters)); 354 } 355 356 if (blob->active.num_pages) { 357 assert(blob->active.pages); 358 pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages)); 359 if (!pages) { 360 free(clusters); 361 return -ENOMEM; 362 } 363 memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*pages)); 364 } 365 366 free(blob->clean.clusters); 367 free(blob->clean.pages); 368 369 blob->clean.num_clusters = blob->active.num_clusters; 370 blob->clean.clusters = blob->active.clusters; 371 blob->clean.num_pages = blob->active.num_pages; 372 blob->clean.pages = blob->active.pages; 373 374 blob->active.clusters = clusters; 375 blob->active.pages = pages; 376 377 /* If the metadata was dirtied again while the metadata was being written to disk, 378 * we do not want to revert the DIRTY state back to CLEAN here. 379 */ 380 if (blob->state == SPDK_BLOB_STATE_LOADING) { 381 blob->state = SPDK_BLOB_STATE_CLEAN; 382 } 383 384 return 0; 385 } 386 387 static int 388 _spdk_blob_deserialize_xattr(struct spdk_blob *blob, 389 struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal) 390 { 391 struct spdk_xattr *xattr; 392 393 if (desc_xattr->length != sizeof(desc_xattr->name_length) + 394 sizeof(desc_xattr->value_length) + 395 desc_xattr->name_length + desc_xattr->value_length) { 396 return -EINVAL; 397 } 398 399 xattr = calloc(1, sizeof(*xattr)); 400 if (xattr == NULL) { 401 return -ENOMEM; 402 } 403 404 xattr->name = malloc(desc_xattr->name_length + 1); 405 if (xattr->name == NULL) { 406 free(xattr); 407 return -ENOMEM; 408 } 409 memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length); 410 xattr->name[desc_xattr->name_length] = '\0'; 411 412 xattr->value = malloc(desc_xattr->value_length); 413 if (xattr->value == NULL) { 414 free(xattr->name); 415 free(xattr); 416 return -ENOMEM; 417 } 418 xattr->value_len = desc_xattr->value_length; 419 memcpy(xattr->value, 420 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 421 desc_xattr->value_length); 422 423 TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link); 424 425 return 0; 426 } 427 428 429 static int 430 _spdk_blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob) 431 { 432 struct spdk_blob_md_descriptor *desc; 433 size_t cur_desc = 0; 434 void *tmp; 435 436 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 437 while (cur_desc < sizeof(page->descriptors)) { 438 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 439 if (desc->length == 0) { 440 /* If padding and length are 0, this terminates the page */ 441 break; 442 } 443 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 444 struct spdk_blob_md_descriptor_flags *desc_flags; 445 446 desc_flags = (struct spdk_blob_md_descriptor_flags *)desc; 447 448 if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) { 449 return -EINVAL; 450 } 451 452 if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) != 453 SPDK_BLOB_INVALID_FLAGS_MASK) { 454 return -EINVAL; 455 } 456 457 if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) != 458 SPDK_BLOB_DATA_RO_FLAGS_MASK) { 459 blob->data_ro = true; 460 blob->md_ro = true; 461 } 462 463 if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) != 464 SPDK_BLOB_MD_RO_FLAGS_MASK) { 465 blob->md_ro = true; 466 } 467 468 if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 469 blob->data_ro = true; 470 blob->md_ro = true; 471 } 472 473 blob->invalid_flags = desc_flags->invalid_flags; 474 blob->data_ro_flags = desc_flags->data_ro_flags; 475 blob->md_ro_flags = desc_flags->md_ro_flags; 476 477 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 478 struct spdk_blob_md_descriptor_extent *desc_extent; 479 unsigned int i, j; 480 unsigned int cluster_count = blob->active.num_clusters; 481 482 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 483 484 if (desc_extent->length == 0 || 485 (desc_extent->length % sizeof(desc_extent->extents[0]) != 0)) { 486 return -EINVAL; 487 } 488 489 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 490 for (j = 0; j < desc_extent->extents[i].length; j++) { 491 if (desc_extent->extents[i].cluster_idx != 0) { 492 if (!spdk_bit_array_get(blob->bs->used_clusters, 493 desc_extent->extents[i].cluster_idx + j)) { 494 return -EINVAL; 495 } 496 } 497 cluster_count++; 498 } 499 } 500 501 if (cluster_count == 0) { 502 return -EINVAL; 503 } 504 tmp = realloc(blob->active.clusters, cluster_count * sizeof(uint64_t)); 505 if (tmp == NULL) { 506 return -ENOMEM; 507 } 508 blob->active.clusters = tmp; 509 blob->active.cluster_array_size = cluster_count; 510 511 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 512 for (j = 0; j < desc_extent->extents[i].length; j++) { 513 if (desc_extent->extents[i].cluster_idx != 0) { 514 blob->active.clusters[blob->active.num_clusters++] = _spdk_bs_cluster_to_lba(blob->bs, 515 desc_extent->extents[i].cluster_idx + j); 516 } else if (spdk_blob_is_thin_provisioned(blob)) { 517 blob->active.clusters[blob->active.num_clusters++] = 0; 518 } else { 519 return -EINVAL; 520 } 521 } 522 } 523 524 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 525 int rc; 526 527 rc = _spdk_blob_deserialize_xattr(blob, 528 (struct spdk_blob_md_descriptor_xattr *) desc, false); 529 if (rc != 0) { 530 return rc; 531 } 532 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 533 int rc; 534 535 rc = _spdk_blob_deserialize_xattr(blob, 536 (struct spdk_blob_md_descriptor_xattr *) desc, true); 537 if (rc != 0) { 538 return rc; 539 } 540 } else { 541 /* Unrecognized descriptor type. Do not fail - just continue to the 542 * next descriptor. If this descriptor is associated with some feature 543 * defined in a newer version of blobstore, that version of blobstore 544 * should create and set an associated feature flag to specify if this 545 * blob can be loaded or not. 546 */ 547 } 548 549 /* Advance to the next descriptor */ 550 cur_desc += sizeof(*desc) + desc->length; 551 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 552 break; 553 } 554 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 555 } 556 557 return 0; 558 } 559 560 static int 561 _spdk_blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count, 562 struct spdk_blob *blob) 563 { 564 const struct spdk_blob_md_page *page; 565 uint32_t i; 566 int rc; 567 568 assert(page_count > 0); 569 assert(pages[0].sequence_num == 0); 570 assert(blob != NULL); 571 assert(blob->state == SPDK_BLOB_STATE_LOADING); 572 assert(blob->active.clusters == NULL); 573 574 /* The blobid provided doesn't match what's in the MD, this can 575 * happen for example if a bogus blobid is passed in through open. 576 */ 577 if (blob->id != pages[0].id) { 578 SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n", 579 blob->id, pages[0].id); 580 return -ENOENT; 581 } 582 583 for (i = 0; i < page_count; i++) { 584 page = &pages[i]; 585 586 assert(page->id == blob->id); 587 assert(page->sequence_num == i); 588 589 rc = _spdk_blob_parse_page(page, blob); 590 if (rc != 0) { 591 return rc; 592 } 593 } 594 595 return 0; 596 } 597 598 static int 599 _spdk_blob_serialize_add_page(const struct spdk_blob *blob, 600 struct spdk_blob_md_page **pages, 601 uint32_t *page_count, 602 struct spdk_blob_md_page **last_page) 603 { 604 struct spdk_blob_md_page *page; 605 606 assert(pages != NULL); 607 assert(page_count != NULL); 608 609 if (*page_count == 0) { 610 assert(*pages == NULL); 611 *page_count = 1; 612 *pages = spdk_malloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE, 613 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 614 } else { 615 assert(*pages != NULL); 616 (*page_count)++; 617 *pages = spdk_realloc(*pages, 618 SPDK_BS_PAGE_SIZE * (*page_count), 619 SPDK_BS_PAGE_SIZE); 620 } 621 622 if (*pages == NULL) { 623 *page_count = 0; 624 *last_page = NULL; 625 return -ENOMEM; 626 } 627 628 page = &(*pages)[*page_count - 1]; 629 memset(page, 0, sizeof(*page)); 630 page->id = blob->id; 631 page->sequence_num = *page_count - 1; 632 page->next = SPDK_INVALID_MD_PAGE; 633 *last_page = page; 634 635 return 0; 636 } 637 638 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor. 639 * Update required_sz on both success and failure. 640 * 641 */ 642 static int 643 _spdk_blob_serialize_xattr(const struct spdk_xattr *xattr, 644 uint8_t *buf, size_t buf_sz, 645 size_t *required_sz, bool internal) 646 { 647 struct spdk_blob_md_descriptor_xattr *desc; 648 649 *required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) + 650 strlen(xattr->name) + 651 xattr->value_len; 652 653 if (buf_sz < *required_sz) { 654 return -1; 655 } 656 657 desc = (struct spdk_blob_md_descriptor_xattr *)buf; 658 659 desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR; 660 desc->length = sizeof(desc->name_length) + 661 sizeof(desc->value_length) + 662 strlen(xattr->name) + 663 xattr->value_len; 664 desc->name_length = strlen(xattr->name); 665 desc->value_length = xattr->value_len; 666 667 memcpy(desc->name, xattr->name, desc->name_length); 668 memcpy((void *)((uintptr_t)desc->name + desc->name_length), 669 xattr->value, 670 desc->value_length); 671 672 return 0; 673 } 674 675 static void 676 _spdk_blob_serialize_extent(const struct spdk_blob *blob, 677 uint64_t start_cluster, uint64_t *next_cluster, 678 uint8_t *buf, size_t buf_sz) 679 { 680 struct spdk_blob_md_descriptor_extent *desc; 681 size_t cur_sz; 682 uint64_t i, extent_idx; 683 uint64_t lba, lba_per_cluster, lba_count; 684 685 /* The buffer must have room for at least one extent */ 686 cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->extents[0]); 687 if (buf_sz < cur_sz) { 688 *next_cluster = start_cluster; 689 return; 690 } 691 692 desc = (struct spdk_blob_md_descriptor_extent *)buf; 693 desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT; 694 695 lba_per_cluster = _spdk_bs_cluster_to_lba(blob->bs, 1); 696 697 lba = blob->active.clusters[start_cluster]; 698 lba_count = lba_per_cluster; 699 extent_idx = 0; 700 for (i = start_cluster + 1; i < blob->active.num_clusters; i++) { 701 if ((lba + lba_count) == blob->active.clusters[i]) { 702 lba_count += lba_per_cluster; 703 continue; 704 } else if (lba == 0 && blob->active.clusters[i] == 0) { 705 lba_count += lba_per_cluster; 706 continue; 707 } 708 desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 709 desc->extents[extent_idx].length = lba_count / lba_per_cluster; 710 extent_idx++; 711 712 cur_sz += sizeof(desc->extents[extent_idx]); 713 714 if (buf_sz < cur_sz) { 715 /* If we ran out of buffer space, return */ 716 desc->length = sizeof(desc->extents[0]) * extent_idx; 717 *next_cluster = i; 718 return; 719 } 720 721 lba = blob->active.clusters[i]; 722 lba_count = lba_per_cluster; 723 } 724 725 desc->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 726 desc->extents[extent_idx].length = lba_count / lba_per_cluster; 727 extent_idx++; 728 729 desc->length = sizeof(desc->extents[0]) * extent_idx; 730 *next_cluster = blob->active.num_clusters; 731 732 return; 733 } 734 735 static void 736 _spdk_blob_serialize_flags(const struct spdk_blob *blob, 737 uint8_t *buf, size_t *buf_sz) 738 { 739 struct spdk_blob_md_descriptor_flags *desc; 740 741 /* 742 * Flags get serialized first, so we should always have room for the flags 743 * descriptor. 744 */ 745 assert(*buf_sz >= sizeof(*desc)); 746 747 desc = (struct spdk_blob_md_descriptor_flags *)buf; 748 desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS; 749 desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor); 750 desc->invalid_flags = blob->invalid_flags; 751 desc->data_ro_flags = blob->data_ro_flags; 752 desc->md_ro_flags = blob->md_ro_flags; 753 754 *buf_sz -= sizeof(*desc); 755 } 756 757 static int 758 _spdk_blob_serialize_xattrs(const struct spdk_blob *blob, 759 const struct spdk_xattr_tailq *xattrs, bool internal, 760 struct spdk_blob_md_page **pages, 761 struct spdk_blob_md_page *cur_page, 762 uint32_t *page_count, uint8_t **buf, 763 size_t *remaining_sz) 764 { 765 const struct spdk_xattr *xattr; 766 int rc; 767 768 TAILQ_FOREACH(xattr, xattrs, link) { 769 size_t required_sz = 0; 770 771 rc = _spdk_blob_serialize_xattr(xattr, 772 *buf, *remaining_sz, 773 &required_sz, internal); 774 if (rc < 0) { 775 /* Need to add a new page to the chain */ 776 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, 777 &cur_page); 778 if (rc < 0) { 779 spdk_free(*pages); 780 *pages = NULL; 781 *page_count = 0; 782 return rc; 783 } 784 785 *buf = (uint8_t *)cur_page->descriptors; 786 *remaining_sz = sizeof(cur_page->descriptors); 787 788 /* Try again */ 789 required_sz = 0; 790 rc = _spdk_blob_serialize_xattr(xattr, 791 *buf, *remaining_sz, 792 &required_sz, internal); 793 794 if (rc < 0) { 795 spdk_free(*pages); 796 *pages = NULL; 797 *page_count = 0; 798 return rc; 799 } 800 } 801 802 *remaining_sz -= required_sz; 803 *buf += required_sz; 804 } 805 806 return 0; 807 } 808 809 static int 810 _spdk_blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages, 811 uint32_t *page_count) 812 { 813 struct spdk_blob_md_page *cur_page; 814 int rc; 815 uint8_t *buf; 816 size_t remaining_sz; 817 uint64_t last_cluster; 818 819 assert(pages != NULL); 820 assert(page_count != NULL); 821 assert(blob != NULL); 822 assert(blob->state == SPDK_BLOB_STATE_DIRTY); 823 824 *pages = NULL; 825 *page_count = 0; 826 827 /* A blob always has at least 1 page, even if it has no descriptors */ 828 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, &cur_page); 829 if (rc < 0) { 830 return rc; 831 } 832 833 buf = (uint8_t *)cur_page->descriptors; 834 remaining_sz = sizeof(cur_page->descriptors); 835 836 /* Serialize flags */ 837 _spdk_blob_serialize_flags(blob, buf, &remaining_sz); 838 buf += sizeof(struct spdk_blob_md_descriptor_flags); 839 840 /* Serialize xattrs */ 841 rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs, false, 842 pages, cur_page, page_count, &buf, &remaining_sz); 843 if (rc < 0) { 844 return rc; 845 } 846 847 /* Serialize internal xattrs */ 848 rc = _spdk_blob_serialize_xattrs(blob, &blob->xattrs_internal, true, 849 pages, cur_page, page_count, &buf, &remaining_sz); 850 if (rc < 0) { 851 return rc; 852 } 853 854 /* Serialize extents */ 855 last_cluster = 0; 856 while (last_cluster < blob->active.num_clusters) { 857 _spdk_blob_serialize_extent(blob, last_cluster, &last_cluster, 858 buf, remaining_sz); 859 860 if (last_cluster == blob->active.num_clusters) { 861 break; 862 } 863 864 rc = _spdk_blob_serialize_add_page(blob, pages, page_count, 865 &cur_page); 866 if (rc < 0) { 867 return rc; 868 } 869 870 buf = (uint8_t *)cur_page->descriptors; 871 remaining_sz = sizeof(cur_page->descriptors); 872 } 873 874 return 0; 875 } 876 877 struct spdk_blob_load_ctx { 878 struct spdk_blob *blob; 879 880 struct spdk_blob_md_page *pages; 881 uint32_t num_pages; 882 spdk_bs_sequence_t *seq; 883 884 spdk_bs_sequence_cpl cb_fn; 885 void *cb_arg; 886 }; 887 888 static uint32_t 889 _spdk_blob_md_page_calc_crc(void *page) 890 { 891 uint32_t crc; 892 893 crc = BLOB_CRC32C_INITIAL; 894 crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc); 895 crc ^= BLOB_CRC32C_INITIAL; 896 897 return crc; 898 899 } 900 901 static void 902 _spdk_blob_load_final(void *cb_arg, int bserrno) 903 { 904 struct spdk_blob_load_ctx *ctx = cb_arg; 905 struct spdk_blob *blob = ctx->blob; 906 907 _spdk_blob_mark_clean(blob); 908 909 ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno); 910 911 /* Free the memory */ 912 spdk_free(ctx->pages); 913 free(ctx); 914 } 915 916 static void 917 _spdk_blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno) 918 { 919 struct spdk_blob_load_ctx *ctx = cb_arg; 920 struct spdk_blob *blob = ctx->blob; 921 922 if (bserrno != 0) { 923 goto error; 924 } 925 926 blob->back_bs_dev = spdk_bs_create_blob_bs_dev(snapshot); 927 928 if (blob->back_bs_dev == NULL) { 929 bserrno = -ENOMEM; 930 goto error; 931 } 932 933 _spdk_blob_load_final(ctx, bserrno); 934 return; 935 936 error: 937 SPDK_ERRLOG("Snapshot fail\n"); 938 _spdk_blob_free(blob); 939 ctx->cb_fn(ctx->seq, NULL, bserrno); 940 spdk_free(ctx->pages); 941 free(ctx); 942 } 943 944 static void 945 _spdk_blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 946 { 947 struct spdk_blob_load_ctx *ctx = cb_arg; 948 struct spdk_blob *blob = ctx->blob; 949 struct spdk_blob_md_page *page; 950 const void *value; 951 size_t len; 952 int rc; 953 uint32_t crc; 954 955 if (bserrno) { 956 SPDK_ERRLOG("Metadata page read failed: %d\n", bserrno); 957 _spdk_blob_free(blob); 958 ctx->cb_fn(seq, NULL, bserrno); 959 spdk_free(ctx->pages); 960 free(ctx); 961 return; 962 } 963 964 page = &ctx->pages[ctx->num_pages - 1]; 965 crc = _spdk_blob_md_page_calc_crc(page); 966 if (crc != page->crc) { 967 SPDK_ERRLOG("Metadata page %d crc mismatch\n", ctx->num_pages); 968 _spdk_blob_free(blob); 969 ctx->cb_fn(seq, NULL, -EINVAL); 970 spdk_free(ctx->pages); 971 free(ctx); 972 return; 973 } 974 975 if (page->next != SPDK_INVALID_MD_PAGE) { 976 uint32_t next_page = page->next; 977 uint64_t next_lba = _spdk_bs_page_to_lba(blob->bs, blob->bs->md_start + next_page); 978 979 980 assert(next_lba < (blob->bs->md_start + blob->bs->md_len)); 981 982 /* Read the next page */ 983 ctx->num_pages++; 984 ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages), 985 sizeof(*page)); 986 if (ctx->pages == NULL) { 987 ctx->cb_fn(seq, ctx->cb_arg, -ENOMEM); 988 free(ctx); 989 return; 990 } 991 992 spdk_bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1], 993 next_lba, 994 _spdk_bs_byte_to_lba(blob->bs, sizeof(*page)), 995 _spdk_blob_load_cpl, ctx); 996 return; 997 } 998 999 /* Parse the pages */ 1000 rc = _spdk_blob_parse(ctx->pages, ctx->num_pages, blob); 1001 if (rc) { 1002 _spdk_blob_free(blob); 1003 ctx->cb_fn(seq, NULL, rc); 1004 spdk_free(ctx->pages); 1005 free(ctx); 1006 return; 1007 } 1008 ctx->seq = seq; 1009 1010 1011 if (spdk_blob_is_thin_provisioned(blob)) { 1012 rc = _spdk_blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true); 1013 if (rc == 0) { 1014 if (len != sizeof(spdk_blob_id)) { 1015 _spdk_blob_free(blob); 1016 ctx->cb_fn(seq, NULL, -EINVAL); 1017 spdk_free(ctx->pages); 1018 free(ctx); 1019 return; 1020 } 1021 /* open snapshot blob and continue in the callback function */ 1022 blob->parent_id = *(spdk_blob_id *)value; 1023 spdk_bs_open_blob(blob->bs, blob->parent_id, 1024 _spdk_blob_load_snapshot_cpl, ctx); 1025 return; 1026 } else { 1027 /* add zeroes_dev for thin provisioned blob */ 1028 blob->back_bs_dev = spdk_bs_create_zeroes_dev(); 1029 } 1030 } else { 1031 /* standard blob */ 1032 blob->back_bs_dev = NULL; 1033 } 1034 _spdk_blob_load_final(ctx, bserrno); 1035 } 1036 1037 /* Load a blob from disk given a blobid */ 1038 static void 1039 _spdk_blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 1040 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 1041 { 1042 struct spdk_blob_load_ctx *ctx; 1043 struct spdk_blob_store *bs; 1044 uint32_t page_num; 1045 uint64_t lba; 1046 1047 _spdk_blob_verify_md_op(blob); 1048 1049 bs = blob->bs; 1050 1051 ctx = calloc(1, sizeof(*ctx)); 1052 if (!ctx) { 1053 cb_fn(seq, cb_arg, -ENOMEM); 1054 return; 1055 } 1056 1057 ctx->blob = blob; 1058 ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE); 1059 if (!ctx->pages) { 1060 free(ctx); 1061 cb_fn(seq, cb_arg, -ENOMEM); 1062 return; 1063 } 1064 ctx->num_pages = 1; 1065 ctx->cb_fn = cb_fn; 1066 ctx->cb_arg = cb_arg; 1067 1068 page_num = _spdk_bs_blobid_to_page(blob->id); 1069 lba = _spdk_bs_page_to_lba(blob->bs, bs->md_start + page_num); 1070 1071 blob->state = SPDK_BLOB_STATE_LOADING; 1072 1073 spdk_bs_sequence_read_dev(seq, &ctx->pages[0], lba, 1074 _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE), 1075 _spdk_blob_load_cpl, ctx); 1076 } 1077 1078 struct spdk_blob_persist_ctx { 1079 struct spdk_blob *blob; 1080 1081 struct spdk_bs_super_block *super; 1082 1083 struct spdk_blob_md_page *pages; 1084 1085 uint64_t idx; 1086 1087 spdk_bs_sequence_t *seq; 1088 spdk_bs_sequence_cpl cb_fn; 1089 void *cb_arg; 1090 }; 1091 1092 static void 1093 spdk_bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba, 1094 uint32_t lba_count) 1095 { 1096 if (ctx->blob->clear_method == BLOB_CLEAR_WITH_DEFAULT || 1097 ctx->blob->clear_method == BLOB_CLEAR_WITH_UNMAP) { 1098 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 1099 } else if (ctx->blob->clear_method == BLOB_CLEAR_WITH_WRITE_ZEROES) { 1100 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1101 } 1102 } 1103 1104 static void 1105 _spdk_blob_persist_complete(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1106 { 1107 struct spdk_blob_persist_ctx *ctx = cb_arg; 1108 struct spdk_blob *blob = ctx->blob; 1109 1110 if (bserrno == 0) { 1111 _spdk_blob_mark_clean(blob); 1112 } 1113 1114 /* Call user callback */ 1115 ctx->cb_fn(seq, ctx->cb_arg, bserrno); 1116 1117 /* Free the memory */ 1118 spdk_free(ctx->pages); 1119 free(ctx); 1120 } 1121 1122 static void 1123 _spdk_blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1124 { 1125 struct spdk_blob_persist_ctx *ctx = cb_arg; 1126 struct spdk_blob *blob = ctx->blob; 1127 struct spdk_blob_store *bs = blob->bs; 1128 void *tmp; 1129 size_t i; 1130 1131 /* Release all clusters that were truncated */ 1132 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1133 uint32_t cluster_num = _spdk_bs_lba_to_cluster(bs, blob->active.clusters[i]); 1134 1135 /* Nothing to release if it was not allocated */ 1136 if (blob->active.clusters[i] != 0) { 1137 _spdk_bs_release_cluster(bs, cluster_num); 1138 } 1139 } 1140 1141 if (blob->active.num_clusters == 0) { 1142 free(blob->active.clusters); 1143 blob->active.clusters = NULL; 1144 blob->active.cluster_array_size = 0; 1145 } else if (blob->active.num_clusters != blob->active.cluster_array_size) { 1146 tmp = realloc(blob->active.clusters, sizeof(uint64_t) * blob->active.num_clusters); 1147 assert(tmp != NULL); 1148 blob->active.clusters = tmp; 1149 blob->active.cluster_array_size = blob->active.num_clusters; 1150 } 1151 1152 _spdk_blob_persist_complete(seq, ctx, bserrno); 1153 } 1154 1155 static void 1156 _spdk_blob_persist_clear_clusters(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1157 { 1158 struct spdk_blob_persist_ctx *ctx = cb_arg; 1159 struct spdk_blob *blob = ctx->blob; 1160 struct spdk_blob_store *bs = blob->bs; 1161 spdk_bs_batch_t *batch; 1162 size_t i; 1163 uint64_t lba; 1164 uint32_t lba_count; 1165 1166 /* Clusters don't move around in blobs. The list shrinks or grows 1167 * at the end, but no changes ever occur in the middle of the list. 1168 */ 1169 1170 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_clear_clusters_cpl, ctx); 1171 1172 /* Clear all clusters that were truncated */ 1173 lba = 0; 1174 lba_count = 0; 1175 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1176 uint64_t next_lba = blob->active.clusters[i]; 1177 uint32_t next_lba_count = _spdk_bs_cluster_to_lba(bs, 1); 1178 1179 if (next_lba > 0 && (lba + lba_count) == next_lba) { 1180 /* This cluster is contiguous with the previous one. */ 1181 lba_count += next_lba_count; 1182 continue; 1183 } 1184 1185 /* This cluster is not contiguous with the previous one. */ 1186 1187 /* If a run of LBAs previously existing, clear them now */ 1188 if (lba_count > 0) { 1189 spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count); 1190 } 1191 1192 /* Start building the next batch */ 1193 lba = next_lba; 1194 if (next_lba > 0) { 1195 lba_count = next_lba_count; 1196 } else { 1197 lba_count = 0; 1198 } 1199 } 1200 1201 /* If we ended with a contiguous set of LBAs, clear them now */ 1202 if (lba_count > 0) { 1203 spdk_bs_batch_clear_dev(ctx, batch, lba, lba_count); 1204 } 1205 1206 spdk_bs_batch_close(batch); 1207 } 1208 1209 static void 1210 _spdk_blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1211 { 1212 struct spdk_blob_persist_ctx *ctx = cb_arg; 1213 struct spdk_blob *blob = ctx->blob; 1214 struct spdk_blob_store *bs = blob->bs; 1215 size_t i; 1216 1217 /* This loop starts at 1 because the first page is special and handled 1218 * below. The pages (except the first) are never written in place, 1219 * so any pages in the clean list must be zeroed. 1220 */ 1221 for (i = 1; i < blob->clean.num_pages; i++) { 1222 spdk_bit_array_clear(bs->used_md_pages, blob->clean.pages[i]); 1223 } 1224 1225 if (blob->active.num_pages == 0) { 1226 uint32_t page_num; 1227 1228 page_num = _spdk_bs_blobid_to_page(blob->id); 1229 spdk_bit_array_clear(bs->used_md_pages, page_num); 1230 } 1231 1232 /* Move on to clearing clusters */ 1233 _spdk_blob_persist_clear_clusters(seq, ctx, 0); 1234 } 1235 1236 static void 1237 _spdk_blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1238 { 1239 struct spdk_blob_persist_ctx *ctx = cb_arg; 1240 struct spdk_blob *blob = ctx->blob; 1241 struct spdk_blob_store *bs = blob->bs; 1242 uint64_t lba; 1243 uint32_t lba_count; 1244 spdk_bs_batch_t *batch; 1245 size_t i; 1246 1247 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_zero_pages_cpl, ctx); 1248 1249 lba_count = _spdk_bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE); 1250 1251 /* This loop starts at 1 because the first page is special and handled 1252 * below. The pages (except the first) are never written in place, 1253 * so any pages in the clean list must be zeroed. 1254 */ 1255 for (i = 1; i < blob->clean.num_pages; i++) { 1256 lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->clean.pages[i]); 1257 1258 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1259 } 1260 1261 /* The first page will only be zeroed if this is a delete. */ 1262 if (blob->active.num_pages == 0) { 1263 uint32_t page_num; 1264 1265 /* The first page in the metadata goes where the blobid indicates */ 1266 page_num = _spdk_bs_blobid_to_page(blob->id); 1267 lba = _spdk_bs_page_to_lba(bs, bs->md_start + page_num); 1268 1269 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1270 } 1271 1272 spdk_bs_batch_close(batch); 1273 } 1274 1275 static void 1276 _spdk_blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1277 { 1278 struct spdk_blob_persist_ctx *ctx = cb_arg; 1279 struct spdk_blob *blob = ctx->blob; 1280 struct spdk_blob_store *bs = blob->bs; 1281 uint64_t lba; 1282 uint32_t lba_count; 1283 struct spdk_blob_md_page *page; 1284 1285 if (blob->active.num_pages == 0) { 1286 /* Move on to the next step */ 1287 _spdk_blob_persist_zero_pages(seq, ctx, 0); 1288 return; 1289 } 1290 1291 lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page)); 1292 1293 page = &ctx->pages[0]; 1294 /* The first page in the metadata goes where the blobid indicates */ 1295 lba = _spdk_bs_page_to_lba(bs, bs->md_start + _spdk_bs_blobid_to_page(blob->id)); 1296 1297 spdk_bs_sequence_write_dev(seq, page, lba, lba_count, 1298 _spdk_blob_persist_zero_pages, ctx); 1299 } 1300 1301 static void 1302 _spdk_blob_persist_write_page_chain(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1303 { 1304 struct spdk_blob_persist_ctx *ctx = cb_arg; 1305 struct spdk_blob *blob = ctx->blob; 1306 struct spdk_blob_store *bs = blob->bs; 1307 uint64_t lba; 1308 uint32_t lba_count; 1309 struct spdk_blob_md_page *page; 1310 spdk_bs_batch_t *batch; 1311 size_t i; 1312 1313 /* Clusters don't move around in blobs. The list shrinks or grows 1314 * at the end, but no changes ever occur in the middle of the list. 1315 */ 1316 1317 lba_count = _spdk_bs_byte_to_lba(bs, sizeof(*page)); 1318 1319 batch = spdk_bs_sequence_to_batch(seq, _spdk_blob_persist_write_page_root, ctx); 1320 1321 /* This starts at 1. The root page is not written until 1322 * all of the others are finished 1323 */ 1324 for (i = 1; i < blob->active.num_pages; i++) { 1325 page = &ctx->pages[i]; 1326 assert(page->sequence_num == i); 1327 1328 lba = _spdk_bs_page_to_lba(bs, bs->md_start + blob->active.pages[i]); 1329 1330 spdk_bs_batch_write_dev(batch, page, lba, lba_count); 1331 } 1332 1333 spdk_bs_batch_close(batch); 1334 } 1335 1336 static int 1337 _spdk_blob_resize(struct spdk_blob *blob, uint64_t sz) 1338 { 1339 uint64_t i; 1340 uint64_t *tmp; 1341 uint64_t lfc; /* lowest free cluster */ 1342 uint64_t num_clusters; 1343 struct spdk_blob_store *bs; 1344 1345 bs = blob->bs; 1346 1347 _spdk_blob_verify_md_op(blob); 1348 1349 if (blob->active.num_clusters == sz) { 1350 return 0; 1351 } 1352 1353 if (blob->active.num_clusters < blob->active.cluster_array_size) { 1354 /* If this blob was resized to be larger, then smaller, then 1355 * larger without syncing, then the cluster array already 1356 * contains spare assigned clusters we can use. 1357 */ 1358 num_clusters = spdk_min(blob->active.cluster_array_size, 1359 sz); 1360 } else { 1361 num_clusters = blob->active.num_clusters; 1362 } 1363 1364 /* Do two passes - one to verify that we can obtain enough clusters 1365 * and another to actually claim them. 1366 */ 1367 1368 if (spdk_blob_is_thin_provisioned(blob) == false) { 1369 lfc = 0; 1370 for (i = num_clusters; i < sz; i++) { 1371 lfc = spdk_bit_array_find_first_clear(bs->used_clusters, lfc); 1372 if (lfc == UINT32_MAX) { 1373 /* No more free clusters. Cannot satisfy the request */ 1374 return -ENOSPC; 1375 } 1376 lfc++; 1377 } 1378 } 1379 1380 if (sz > num_clusters) { 1381 /* Expand the cluster array if necessary. 1382 * We only shrink the array when persisting. 1383 */ 1384 tmp = realloc(blob->active.clusters, sizeof(uint64_t) * sz); 1385 if (sz > 0 && tmp == NULL) { 1386 return -ENOMEM; 1387 } 1388 memset(tmp + blob->active.cluster_array_size, 0, 1389 sizeof(uint64_t) * (sz - blob->active.cluster_array_size)); 1390 blob->active.clusters = tmp; 1391 blob->active.cluster_array_size = sz; 1392 } 1393 1394 blob->state = SPDK_BLOB_STATE_DIRTY; 1395 1396 if (spdk_blob_is_thin_provisioned(blob) == false) { 1397 lfc = 0; 1398 for (i = num_clusters; i < sz; i++) { 1399 _spdk_bs_allocate_cluster(blob, i, &lfc, true); 1400 lfc++; 1401 } 1402 } 1403 1404 blob->active.num_clusters = sz; 1405 1406 return 0; 1407 } 1408 1409 static void 1410 _spdk_blob_persist_start(struct spdk_blob_persist_ctx *ctx) 1411 { 1412 spdk_bs_sequence_t *seq = ctx->seq; 1413 struct spdk_blob *blob = ctx->blob; 1414 struct spdk_blob_store *bs = blob->bs; 1415 uint64_t i; 1416 uint32_t page_num; 1417 void *tmp; 1418 int rc; 1419 1420 if (blob->active.num_pages == 0) { 1421 /* This is the signal that the blob should be deleted. 1422 * Immediately jump to the clean up routine. */ 1423 assert(blob->clean.num_pages > 0); 1424 ctx->idx = blob->clean.num_pages - 1; 1425 blob->state = SPDK_BLOB_STATE_CLEAN; 1426 _spdk_blob_persist_zero_pages(seq, ctx, 0); 1427 return; 1428 1429 } 1430 1431 /* Generate the new metadata */ 1432 rc = _spdk_blob_serialize(blob, &ctx->pages, &blob->active.num_pages); 1433 if (rc < 0) { 1434 _spdk_blob_persist_complete(seq, ctx, rc); 1435 return; 1436 } 1437 1438 assert(blob->active.num_pages >= 1); 1439 1440 /* Resize the cache of page indices */ 1441 tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages)); 1442 if (!tmp) { 1443 _spdk_blob_persist_complete(seq, ctx, -ENOMEM); 1444 return; 1445 } 1446 blob->active.pages = tmp; 1447 1448 /* Assign this metadata to pages. This requires two passes - 1449 * one to verify that there are enough pages and a second 1450 * to actually claim them. */ 1451 page_num = 0; 1452 /* Note that this loop starts at one. The first page location is fixed by the blobid. */ 1453 for (i = 1; i < blob->active.num_pages; i++) { 1454 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 1455 if (page_num == UINT32_MAX) { 1456 _spdk_blob_persist_complete(seq, ctx, -ENOMEM); 1457 return; 1458 } 1459 page_num++; 1460 } 1461 1462 page_num = 0; 1463 blob->active.pages[0] = _spdk_bs_blobid_to_page(blob->id); 1464 for (i = 1; i < blob->active.num_pages; i++) { 1465 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 1466 ctx->pages[i - 1].next = page_num; 1467 /* Now that previous metadata page is complete, calculate the crc for it. */ 1468 ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]); 1469 blob->active.pages[i] = page_num; 1470 spdk_bit_array_set(bs->used_md_pages, page_num); 1471 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id); 1472 page_num++; 1473 } 1474 ctx->pages[i - 1].crc = _spdk_blob_md_page_calc_crc(&ctx->pages[i - 1]); 1475 /* Start writing the metadata from last page to first */ 1476 ctx->idx = blob->active.num_pages - 1; 1477 blob->state = SPDK_BLOB_STATE_CLEAN; 1478 _spdk_blob_persist_write_page_chain(seq, ctx, 0); 1479 } 1480 1481 static void 1482 _spdk_blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1483 { 1484 struct spdk_blob_persist_ctx *ctx = cb_arg; 1485 1486 ctx->blob->bs->clean = 0; 1487 1488 spdk_free(ctx->super); 1489 1490 _spdk_blob_persist_start(ctx); 1491 } 1492 1493 static void 1494 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 1495 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg); 1496 1497 1498 static void 1499 _spdk_blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1500 { 1501 struct spdk_blob_persist_ctx *ctx = cb_arg; 1502 1503 ctx->super->clean = 0; 1504 if (ctx->super->size == 0) { 1505 ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen; 1506 } 1507 1508 _spdk_bs_write_super(seq, ctx->blob->bs, ctx->super, _spdk_blob_persist_dirty_cpl, ctx); 1509 } 1510 1511 1512 /* Write a blob to disk */ 1513 static void 1514 _spdk_blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 1515 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 1516 { 1517 struct spdk_blob_persist_ctx *ctx; 1518 1519 _spdk_blob_verify_md_op(blob); 1520 1521 if (blob->state == SPDK_BLOB_STATE_CLEAN) { 1522 cb_fn(seq, cb_arg, 0); 1523 return; 1524 } 1525 1526 ctx = calloc(1, sizeof(*ctx)); 1527 if (!ctx) { 1528 cb_fn(seq, cb_arg, -ENOMEM); 1529 return; 1530 } 1531 ctx->blob = blob; 1532 ctx->seq = seq; 1533 ctx->cb_fn = cb_fn; 1534 ctx->cb_arg = cb_arg; 1535 1536 if (blob->bs->clean) { 1537 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 1538 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1539 if (!ctx->super) { 1540 cb_fn(seq, cb_arg, -ENOMEM); 1541 free(ctx); 1542 return; 1543 } 1544 1545 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(blob->bs, 0), 1546 _spdk_bs_byte_to_lba(blob->bs, sizeof(*ctx->super)), 1547 _spdk_blob_persist_dirty, ctx); 1548 } else { 1549 _spdk_blob_persist_start(ctx); 1550 } 1551 } 1552 1553 struct spdk_blob_copy_cluster_ctx { 1554 struct spdk_blob *blob; 1555 uint8_t *buf; 1556 uint64_t page; 1557 uint64_t new_cluster; 1558 spdk_bs_sequence_t *seq; 1559 }; 1560 1561 static void 1562 _spdk_blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno) 1563 { 1564 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1565 struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq; 1566 TAILQ_HEAD(, spdk_bs_request_set) requests; 1567 spdk_bs_user_op_t *op; 1568 1569 TAILQ_INIT(&requests); 1570 TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link); 1571 1572 while (!TAILQ_EMPTY(&requests)) { 1573 op = TAILQ_FIRST(&requests); 1574 TAILQ_REMOVE(&requests, op, link); 1575 if (bserrno == 0) { 1576 spdk_bs_user_op_execute(op); 1577 } else { 1578 spdk_bs_user_op_abort(op); 1579 } 1580 } 1581 1582 spdk_free(ctx->buf); 1583 free(ctx); 1584 } 1585 1586 static void 1587 _spdk_blob_insert_cluster_cpl(void *cb_arg, int bserrno) 1588 { 1589 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1590 1591 if (bserrno) { 1592 uint32_t cluster_number; 1593 1594 if (bserrno == -EEXIST) { 1595 /* The metadata insert failed because another thread 1596 * allocated the cluster first. Free our cluster 1597 * but continue without error. */ 1598 bserrno = 0; 1599 } 1600 1601 cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page); 1602 _spdk_bs_release_cluster(ctx->blob->bs, cluster_number); 1603 } 1604 1605 spdk_bs_sequence_finish(ctx->seq, bserrno); 1606 } 1607 1608 static void 1609 _spdk_blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1610 { 1611 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1612 uint32_t cluster_number; 1613 1614 if (bserrno) { 1615 /* The write failed, so jump to the final completion handler */ 1616 spdk_bs_sequence_finish(seq, bserrno); 1617 return; 1618 } 1619 1620 cluster_number = _spdk_bs_page_to_cluster(ctx->blob->bs, ctx->page); 1621 1622 _spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 1623 _spdk_blob_insert_cluster_cpl, ctx); 1624 } 1625 1626 static void 1627 _spdk_blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1628 { 1629 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 1630 1631 if (bserrno != 0) { 1632 /* The read failed, so jump to the final completion handler */ 1633 spdk_bs_sequence_finish(seq, bserrno); 1634 return; 1635 } 1636 1637 /* Write whole cluster */ 1638 spdk_bs_sequence_write_dev(seq, ctx->buf, 1639 _spdk_bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster), 1640 _spdk_bs_cluster_to_lba(ctx->blob->bs, 1), 1641 _spdk_blob_write_copy_cpl, ctx); 1642 } 1643 1644 static void 1645 _spdk_bs_allocate_and_copy_cluster(struct spdk_blob *blob, 1646 struct spdk_io_channel *_ch, 1647 uint64_t io_unit, spdk_bs_user_op_t *op) 1648 { 1649 struct spdk_bs_cpl cpl; 1650 struct spdk_bs_channel *ch; 1651 struct spdk_blob_copy_cluster_ctx *ctx; 1652 uint32_t cluster_start_page; 1653 uint32_t cluster_number; 1654 int rc; 1655 1656 ch = spdk_io_channel_get_ctx(_ch); 1657 1658 if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) { 1659 /* There are already operations pending. Queue this user op 1660 * and return because it will be re-executed when the outstanding 1661 * cluster allocation completes. */ 1662 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 1663 return; 1664 } 1665 1666 /* Round the io_unit offset down to the first page in the cluster */ 1667 cluster_start_page = _spdk_bs_io_unit_to_cluster_start(blob, io_unit); 1668 1669 /* Calculate which index in the metadata cluster array the corresponding 1670 * cluster is supposed to be at. */ 1671 cluster_number = _spdk_bs_io_unit_to_cluster_number(blob, io_unit); 1672 1673 ctx = calloc(1, sizeof(*ctx)); 1674 if (!ctx) { 1675 spdk_bs_user_op_abort(op); 1676 return; 1677 } 1678 1679 assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0); 1680 1681 ctx->blob = blob; 1682 ctx->page = cluster_start_page; 1683 1684 if (blob->parent_id != SPDK_BLOBID_INVALID) { 1685 ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, 1686 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1687 if (!ctx->buf) { 1688 SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n", 1689 blob->bs->cluster_sz); 1690 free(ctx); 1691 spdk_bs_user_op_abort(op); 1692 return; 1693 } 1694 } 1695 1696 rc = _spdk_bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, false); 1697 if (rc != 0) { 1698 spdk_free(ctx->buf); 1699 free(ctx); 1700 spdk_bs_user_op_abort(op); 1701 return; 1702 } 1703 1704 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1705 cpl.u.blob_basic.cb_fn = _spdk_blob_allocate_and_copy_cluster_cpl; 1706 cpl.u.blob_basic.cb_arg = ctx; 1707 1708 ctx->seq = spdk_bs_sequence_start(_ch, &cpl); 1709 if (!ctx->seq) { 1710 _spdk_bs_release_cluster(blob->bs, ctx->new_cluster); 1711 spdk_free(ctx->buf); 1712 free(ctx); 1713 spdk_bs_user_op_abort(op); 1714 return; 1715 } 1716 1717 /* Queue the user op to block other incoming operations */ 1718 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 1719 1720 if (blob->parent_id != SPDK_BLOBID_INVALID) { 1721 /* Read cluster from backing device */ 1722 spdk_bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf, 1723 _spdk_bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page), 1724 _spdk_bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz), 1725 _spdk_blob_write_copy, ctx); 1726 } else { 1727 _spdk_blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 1728 _spdk_blob_insert_cluster_cpl, ctx); 1729 } 1730 } 1731 1732 static void 1733 _spdk_blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length, 1734 uint64_t *lba, uint32_t *lba_count) 1735 { 1736 *lba_count = length; 1737 1738 if (!_spdk_bs_io_unit_is_allocated(blob, io_unit)) { 1739 assert(blob->back_bs_dev != NULL); 1740 *lba = _spdk_bs_io_unit_to_back_dev_lba(blob, io_unit); 1741 *lba_count = _spdk_bs_io_unit_to_back_dev_lba(blob, *lba_count); 1742 } else { 1743 *lba = _spdk_bs_blob_io_unit_to_lba(blob, io_unit); 1744 } 1745 } 1746 1747 struct op_split_ctx { 1748 struct spdk_blob *blob; 1749 struct spdk_io_channel *channel; 1750 uint64_t io_unit_offset; 1751 uint64_t io_units_remaining; 1752 void *curr_payload; 1753 enum spdk_blob_op_type op_type; 1754 spdk_bs_sequence_t *seq; 1755 }; 1756 1757 static void 1758 _spdk_blob_request_submit_op_split_next(void *cb_arg, int bserrno) 1759 { 1760 struct op_split_ctx *ctx = cb_arg; 1761 struct spdk_blob *blob = ctx->blob; 1762 struct spdk_io_channel *ch = ctx->channel; 1763 enum spdk_blob_op_type op_type = ctx->op_type; 1764 uint8_t *buf = ctx->curr_payload; 1765 uint64_t offset = ctx->io_unit_offset; 1766 uint64_t length = ctx->io_units_remaining; 1767 uint64_t op_length; 1768 1769 if (bserrno != 0 || ctx->io_units_remaining == 0) { 1770 spdk_bs_sequence_finish(ctx->seq, bserrno); 1771 free(ctx); 1772 return; 1773 } 1774 1775 op_length = spdk_min(length, _spdk_bs_num_io_units_to_cluster_boundary(blob, 1776 offset)); 1777 1778 /* Update length and payload for next operation */ 1779 ctx->io_units_remaining -= op_length; 1780 ctx->io_unit_offset += op_length; 1781 if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) { 1782 ctx->curr_payload += op_length * blob->bs->io_unit_size; 1783 } 1784 1785 switch (op_type) { 1786 case SPDK_BLOB_READ: 1787 spdk_blob_io_read(blob, ch, buf, offset, op_length, 1788 _spdk_blob_request_submit_op_split_next, ctx); 1789 break; 1790 case SPDK_BLOB_WRITE: 1791 spdk_blob_io_write(blob, ch, buf, offset, op_length, 1792 _spdk_blob_request_submit_op_split_next, ctx); 1793 break; 1794 case SPDK_BLOB_UNMAP: 1795 spdk_blob_io_unmap(blob, ch, offset, op_length, 1796 _spdk_blob_request_submit_op_split_next, ctx); 1797 break; 1798 case SPDK_BLOB_WRITE_ZEROES: 1799 spdk_blob_io_write_zeroes(blob, ch, offset, op_length, 1800 _spdk_blob_request_submit_op_split_next, ctx); 1801 break; 1802 case SPDK_BLOB_READV: 1803 case SPDK_BLOB_WRITEV: 1804 SPDK_ERRLOG("readv/write not valid for %s\n", __func__); 1805 spdk_bs_sequence_finish(ctx->seq, -EINVAL); 1806 free(ctx); 1807 break; 1808 } 1809 } 1810 1811 static void 1812 _spdk_blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob, 1813 void *payload, uint64_t offset, uint64_t length, 1814 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1815 { 1816 struct op_split_ctx *ctx; 1817 spdk_bs_sequence_t *seq; 1818 struct spdk_bs_cpl cpl; 1819 1820 assert(blob != NULL); 1821 1822 ctx = calloc(1, sizeof(struct op_split_ctx)); 1823 if (ctx == NULL) { 1824 cb_fn(cb_arg, -ENOMEM); 1825 return; 1826 } 1827 1828 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1829 cpl.u.blob_basic.cb_fn = cb_fn; 1830 cpl.u.blob_basic.cb_arg = cb_arg; 1831 1832 seq = spdk_bs_sequence_start(ch, &cpl); 1833 if (!seq) { 1834 free(ctx); 1835 cb_fn(cb_arg, -ENOMEM); 1836 return; 1837 } 1838 1839 ctx->blob = blob; 1840 ctx->channel = ch; 1841 ctx->curr_payload = payload; 1842 ctx->io_unit_offset = offset; 1843 ctx->io_units_remaining = length; 1844 ctx->op_type = op_type; 1845 ctx->seq = seq; 1846 1847 _spdk_blob_request_submit_op_split_next(ctx, 0); 1848 } 1849 1850 static void 1851 _spdk_blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob, 1852 void *payload, uint64_t offset, uint64_t length, 1853 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1854 { 1855 struct spdk_bs_cpl cpl; 1856 uint64_t lba; 1857 uint32_t lba_count; 1858 1859 assert(blob != NULL); 1860 1861 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 1862 cpl.u.blob_basic.cb_fn = cb_fn; 1863 cpl.u.blob_basic.cb_arg = cb_arg; 1864 1865 _spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 1866 1867 if (blob->frozen_refcnt) { 1868 /* This blob I/O is frozen */ 1869 spdk_bs_user_op_t *op; 1870 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch); 1871 1872 op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 1873 if (!op) { 1874 cb_fn(cb_arg, -ENOMEM); 1875 return; 1876 } 1877 1878 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 1879 1880 return; 1881 } 1882 1883 switch (op_type) { 1884 case SPDK_BLOB_READ: { 1885 spdk_bs_batch_t *batch; 1886 1887 batch = spdk_bs_batch_open(_ch, &cpl); 1888 if (!batch) { 1889 cb_fn(cb_arg, -ENOMEM); 1890 return; 1891 } 1892 1893 if (_spdk_bs_io_unit_is_allocated(blob, offset)) { 1894 /* Read from the blob */ 1895 spdk_bs_batch_read_dev(batch, payload, lba, lba_count); 1896 } else { 1897 /* Read from the backing block device */ 1898 spdk_bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count); 1899 } 1900 1901 spdk_bs_batch_close(batch); 1902 break; 1903 } 1904 case SPDK_BLOB_WRITE: 1905 case SPDK_BLOB_WRITE_ZEROES: { 1906 if (_spdk_bs_io_unit_is_allocated(blob, offset)) { 1907 /* Write to the blob */ 1908 spdk_bs_batch_t *batch; 1909 1910 if (lba_count == 0) { 1911 cb_fn(cb_arg, 0); 1912 return; 1913 } 1914 1915 batch = spdk_bs_batch_open(_ch, &cpl); 1916 if (!batch) { 1917 cb_fn(cb_arg, -ENOMEM); 1918 return; 1919 } 1920 1921 if (op_type == SPDK_BLOB_WRITE) { 1922 spdk_bs_batch_write_dev(batch, payload, lba, lba_count); 1923 } else { 1924 spdk_bs_batch_write_zeroes_dev(batch, lba, lba_count); 1925 } 1926 1927 spdk_bs_batch_close(batch); 1928 } else { 1929 /* Queue this operation and allocate the cluster */ 1930 spdk_bs_user_op_t *op; 1931 1932 op = spdk_bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 1933 if (!op) { 1934 cb_fn(cb_arg, -ENOMEM); 1935 return; 1936 } 1937 1938 _spdk_bs_allocate_and_copy_cluster(blob, _ch, offset, op); 1939 } 1940 break; 1941 } 1942 case SPDK_BLOB_UNMAP: { 1943 spdk_bs_batch_t *batch; 1944 1945 batch = spdk_bs_batch_open(_ch, &cpl); 1946 if (!batch) { 1947 cb_fn(cb_arg, -ENOMEM); 1948 return; 1949 } 1950 1951 if (_spdk_bs_io_unit_is_allocated(blob, offset)) { 1952 spdk_bs_batch_unmap_dev(batch, lba, lba_count); 1953 } 1954 1955 spdk_bs_batch_close(batch); 1956 break; 1957 } 1958 case SPDK_BLOB_READV: 1959 case SPDK_BLOB_WRITEV: 1960 SPDK_ERRLOG("readv/write not valid\n"); 1961 cb_fn(cb_arg, -EINVAL); 1962 break; 1963 } 1964 } 1965 1966 static void 1967 _spdk_blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel, 1968 void *payload, uint64_t offset, uint64_t length, 1969 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 1970 { 1971 assert(blob != NULL); 1972 1973 if (blob->data_ro && op_type != SPDK_BLOB_READ) { 1974 cb_fn(cb_arg, -EPERM); 1975 return; 1976 } 1977 1978 if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) { 1979 cb_fn(cb_arg, -EINVAL); 1980 return; 1981 } 1982 if (length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset)) { 1983 _spdk_blob_request_submit_op_single(_channel, blob, payload, offset, length, 1984 cb_fn, cb_arg, op_type); 1985 } else { 1986 _spdk_blob_request_submit_op_split(_channel, blob, payload, offset, length, 1987 cb_fn, cb_arg, op_type); 1988 } 1989 } 1990 1991 struct rw_iov_ctx { 1992 struct spdk_blob *blob; 1993 struct spdk_io_channel *channel; 1994 spdk_blob_op_complete cb_fn; 1995 void *cb_arg; 1996 bool read; 1997 int iovcnt; 1998 struct iovec *orig_iov; 1999 uint64_t io_unit_offset; 2000 uint64_t io_units_remaining; 2001 uint64_t io_units_done; 2002 struct iovec iov[0]; 2003 }; 2004 2005 static void 2006 _spdk_rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2007 { 2008 assert(cb_arg == NULL); 2009 spdk_bs_sequence_finish(seq, bserrno); 2010 } 2011 2012 static void 2013 _spdk_rw_iov_split_next(void *cb_arg, int bserrno) 2014 { 2015 struct rw_iov_ctx *ctx = cb_arg; 2016 struct spdk_blob *blob = ctx->blob; 2017 struct iovec *iov, *orig_iov; 2018 int iovcnt; 2019 size_t orig_iovoff; 2020 uint64_t io_units_count, io_units_to_boundary, io_unit_offset; 2021 uint64_t byte_count; 2022 2023 if (bserrno != 0 || ctx->io_units_remaining == 0) { 2024 ctx->cb_fn(ctx->cb_arg, bserrno); 2025 free(ctx); 2026 return; 2027 } 2028 2029 io_unit_offset = ctx->io_unit_offset; 2030 io_units_to_boundary = _spdk_bs_num_io_units_to_cluster_boundary(blob, io_unit_offset); 2031 io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary); 2032 /* 2033 * Get index and offset into the original iov array for our current position in the I/O sequence. 2034 * byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will 2035 * point to the current position in the I/O sequence. 2036 */ 2037 byte_count = ctx->io_units_done * blob->bs->io_unit_size; 2038 orig_iov = &ctx->orig_iov[0]; 2039 orig_iovoff = 0; 2040 while (byte_count > 0) { 2041 if (byte_count >= orig_iov->iov_len) { 2042 byte_count -= orig_iov->iov_len; 2043 orig_iov++; 2044 } else { 2045 orig_iovoff = byte_count; 2046 byte_count = 0; 2047 } 2048 } 2049 2050 /* 2051 * Build an iov array for the next I/O in the sequence. byte_count will keep track of how many 2052 * bytes of this next I/O remain to be accounted for in the new iov array. 2053 */ 2054 byte_count = io_units_count * blob->bs->io_unit_size; 2055 iov = &ctx->iov[0]; 2056 iovcnt = 0; 2057 while (byte_count > 0) { 2058 assert(iovcnt < ctx->iovcnt); 2059 iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff); 2060 iov->iov_base = orig_iov->iov_base + orig_iovoff; 2061 byte_count -= iov->iov_len; 2062 orig_iovoff = 0; 2063 orig_iov++; 2064 iov++; 2065 iovcnt++; 2066 } 2067 2068 ctx->io_unit_offset += io_units_count; 2069 ctx->io_units_remaining -= io_units_count; 2070 ctx->io_units_done += io_units_count; 2071 iov = &ctx->iov[0]; 2072 2073 if (ctx->read) { 2074 spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset, 2075 io_units_count, _spdk_rw_iov_split_next, ctx); 2076 } else { 2077 spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset, 2078 io_units_count, _spdk_rw_iov_split_next, ctx); 2079 } 2080 } 2081 2082 static void 2083 _spdk_blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel, 2084 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 2085 spdk_blob_op_complete cb_fn, void *cb_arg, bool read) 2086 { 2087 struct spdk_bs_cpl cpl; 2088 2089 assert(blob != NULL); 2090 2091 if (!read && blob->data_ro) { 2092 cb_fn(cb_arg, -EPERM); 2093 return; 2094 } 2095 2096 if (length == 0) { 2097 cb_fn(cb_arg, 0); 2098 return; 2099 } 2100 2101 if (offset + length > _spdk_bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) { 2102 cb_fn(cb_arg, -EINVAL); 2103 return; 2104 } 2105 2106 /* 2107 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having 2108 * to split a request that spans a cluster boundary. For I/O that do not span a cluster boundary, 2109 * there will be no noticeable difference compared to using a batch. For I/O that do span a cluster 2110 * boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need 2111 * to allocate a separate iov array and split the I/O such that none of the resulting 2112 * smaller I/O cross a cluster boundary. These smaller I/O will be issued in sequence (not in parallel) 2113 * but since this case happens very infrequently, any performance impact will be negligible. 2114 * 2115 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs 2116 * for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them 2117 * in a batch. That would also require creating an intermediate spdk_bs_cpl that would get called 2118 * when the batch was completed, to allow for freeing the memory for the iov arrays. 2119 */ 2120 if (spdk_likely(length <= _spdk_bs_num_io_units_to_cluster_boundary(blob, offset))) { 2121 uint32_t lba_count; 2122 uint64_t lba; 2123 2124 _spdk_blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 2125 2126 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2127 cpl.u.blob_basic.cb_fn = cb_fn; 2128 cpl.u.blob_basic.cb_arg = cb_arg; 2129 if (blob->frozen_refcnt) { 2130 /* This blob I/O is frozen */ 2131 enum spdk_blob_op_type op_type; 2132 spdk_bs_user_op_t *op; 2133 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel); 2134 2135 op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV; 2136 op = spdk_bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length); 2137 if (!op) { 2138 cb_fn(cb_arg, -ENOMEM); 2139 return; 2140 } 2141 2142 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 2143 2144 return; 2145 } 2146 2147 if (read) { 2148 spdk_bs_sequence_t *seq; 2149 2150 seq = spdk_bs_sequence_start(_channel, &cpl); 2151 if (!seq) { 2152 cb_fn(cb_arg, -ENOMEM); 2153 return; 2154 } 2155 2156 if (_spdk_bs_io_unit_is_allocated(blob, offset)) { 2157 spdk_bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL); 2158 } else { 2159 spdk_bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count, 2160 _spdk_rw_iov_done, NULL); 2161 } 2162 } else { 2163 if (_spdk_bs_io_unit_is_allocated(blob, offset)) { 2164 spdk_bs_sequence_t *seq; 2165 2166 seq = spdk_bs_sequence_start(_channel, &cpl); 2167 if (!seq) { 2168 cb_fn(cb_arg, -ENOMEM); 2169 return; 2170 } 2171 2172 spdk_bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, _spdk_rw_iov_done, NULL); 2173 } else { 2174 /* Queue this operation and allocate the cluster */ 2175 spdk_bs_user_op_t *op; 2176 2177 op = spdk_bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, 2178 length); 2179 if (!op) { 2180 cb_fn(cb_arg, -ENOMEM); 2181 return; 2182 } 2183 2184 _spdk_bs_allocate_and_copy_cluster(blob, _channel, offset, op); 2185 } 2186 } 2187 } else { 2188 struct rw_iov_ctx *ctx; 2189 2190 ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec)); 2191 if (ctx == NULL) { 2192 cb_fn(cb_arg, -ENOMEM); 2193 return; 2194 } 2195 2196 ctx->blob = blob; 2197 ctx->channel = _channel; 2198 ctx->cb_fn = cb_fn; 2199 ctx->cb_arg = cb_arg; 2200 ctx->read = read; 2201 ctx->orig_iov = iov; 2202 ctx->iovcnt = iovcnt; 2203 ctx->io_unit_offset = offset; 2204 ctx->io_units_remaining = length; 2205 ctx->io_units_done = 0; 2206 2207 _spdk_rw_iov_split_next(ctx, 0); 2208 } 2209 } 2210 2211 static struct spdk_blob * 2212 _spdk_blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid) 2213 { 2214 struct spdk_blob *blob; 2215 2216 TAILQ_FOREACH(blob, &bs->blobs, link) { 2217 if (blob->id == blobid) { 2218 return blob; 2219 } 2220 } 2221 2222 return NULL; 2223 } 2224 2225 static void 2226 _spdk_blob_get_snapshot_and_clone_entries(struct spdk_blob *blob, 2227 struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry) 2228 { 2229 assert(blob != NULL); 2230 *snapshot_entry = NULL; 2231 *clone_entry = NULL; 2232 2233 if (blob->parent_id == SPDK_BLOBID_INVALID) { 2234 return; 2235 } 2236 2237 TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) { 2238 if ((*snapshot_entry)->id == blob->parent_id) { 2239 break; 2240 } 2241 } 2242 2243 if (*snapshot_entry != NULL) { 2244 TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) { 2245 if ((*clone_entry)->id == blob->id) { 2246 break; 2247 } 2248 } 2249 2250 assert(clone_entry != NULL); 2251 } 2252 } 2253 2254 static int 2255 _spdk_bs_channel_create(void *io_device, void *ctx_buf) 2256 { 2257 struct spdk_blob_store *bs = io_device; 2258 struct spdk_bs_channel *channel = ctx_buf; 2259 struct spdk_bs_dev *dev; 2260 uint32_t max_ops = bs->max_channel_ops; 2261 uint32_t i; 2262 2263 dev = bs->dev; 2264 2265 channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set)); 2266 if (!channel->req_mem) { 2267 return -1; 2268 } 2269 2270 TAILQ_INIT(&channel->reqs); 2271 2272 for (i = 0; i < max_ops; i++) { 2273 TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link); 2274 } 2275 2276 channel->bs = bs; 2277 channel->dev = dev; 2278 channel->dev_channel = dev->create_channel(dev); 2279 2280 if (!channel->dev_channel) { 2281 SPDK_ERRLOG("Failed to create device channel.\n"); 2282 free(channel->req_mem); 2283 return -1; 2284 } 2285 2286 TAILQ_INIT(&channel->need_cluster_alloc); 2287 TAILQ_INIT(&channel->queued_io); 2288 2289 return 0; 2290 } 2291 2292 static void 2293 _spdk_bs_channel_destroy(void *io_device, void *ctx_buf) 2294 { 2295 struct spdk_bs_channel *channel = ctx_buf; 2296 spdk_bs_user_op_t *op; 2297 2298 while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) { 2299 op = TAILQ_FIRST(&channel->need_cluster_alloc); 2300 TAILQ_REMOVE(&channel->need_cluster_alloc, op, link); 2301 spdk_bs_user_op_abort(op); 2302 } 2303 2304 while (!TAILQ_EMPTY(&channel->queued_io)) { 2305 op = TAILQ_FIRST(&channel->queued_io); 2306 TAILQ_REMOVE(&channel->queued_io, op, link); 2307 spdk_bs_user_op_abort(op); 2308 } 2309 2310 free(channel->req_mem); 2311 channel->dev->destroy_channel(channel->dev, channel->dev_channel); 2312 } 2313 2314 static void 2315 _spdk_bs_dev_destroy(void *io_device) 2316 { 2317 struct spdk_blob_store *bs = io_device; 2318 struct spdk_blob *blob, *blob_tmp; 2319 2320 bs->dev->destroy(bs->dev); 2321 2322 TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) { 2323 TAILQ_REMOVE(&bs->blobs, blob, link); 2324 _spdk_blob_free(blob); 2325 } 2326 2327 pthread_mutex_destroy(&bs->used_clusters_mutex); 2328 2329 spdk_bit_array_free(&bs->used_blobids); 2330 spdk_bit_array_free(&bs->used_md_pages); 2331 spdk_bit_array_free(&bs->used_clusters); 2332 /* 2333 * If this function is called for any reason except a successful unload, 2334 * the unload_cpl type will be NONE and this will be a nop. 2335 */ 2336 spdk_bs_call_cpl(&bs->unload_cpl, bs->unload_err); 2337 2338 free(bs); 2339 } 2340 2341 static int 2342 _spdk_bs_blob_list_add(struct spdk_blob *blob) 2343 { 2344 spdk_blob_id snapshot_id; 2345 struct spdk_blob_list *snapshot_entry = NULL; 2346 struct spdk_blob_list *clone_entry = NULL; 2347 2348 assert(blob != NULL); 2349 2350 snapshot_id = blob->parent_id; 2351 if (snapshot_id == SPDK_BLOBID_INVALID) { 2352 return 0; 2353 } 2354 2355 snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, snapshot_id); 2356 if (snapshot_entry == NULL) { 2357 /* Snapshot not found */ 2358 snapshot_entry = calloc(1, sizeof(struct spdk_blob_list)); 2359 if (snapshot_entry == NULL) { 2360 return -ENOMEM; 2361 } 2362 snapshot_entry->id = snapshot_id; 2363 TAILQ_INIT(&snapshot_entry->clones); 2364 TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link); 2365 } else { 2366 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 2367 if (clone_entry->id == blob->id) { 2368 break; 2369 } 2370 } 2371 } 2372 2373 if (clone_entry == NULL) { 2374 /* Clone not found */ 2375 clone_entry = calloc(1, sizeof(struct spdk_blob_list)); 2376 if (clone_entry == NULL) { 2377 return -ENOMEM; 2378 } 2379 clone_entry->id = blob->id; 2380 TAILQ_INIT(&clone_entry->clones); 2381 TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link); 2382 snapshot_entry->clone_count++; 2383 } 2384 2385 return 0; 2386 } 2387 2388 static void 2389 _spdk_bs_blob_list_remove(struct spdk_blob *blob) 2390 { 2391 struct spdk_blob_list *snapshot_entry = NULL; 2392 struct spdk_blob_list *clone_entry = NULL; 2393 2394 _spdk_blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry); 2395 2396 if (snapshot_entry == NULL) { 2397 return; 2398 } 2399 2400 blob->parent_id = SPDK_BLOBID_INVALID; 2401 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 2402 free(clone_entry); 2403 2404 snapshot_entry->clone_count--; 2405 } 2406 2407 static int 2408 _spdk_bs_blob_list_free(struct spdk_blob_store *bs) 2409 { 2410 struct spdk_blob_list *snapshot_entry; 2411 struct spdk_blob_list *snapshot_entry_tmp; 2412 struct spdk_blob_list *clone_entry; 2413 struct spdk_blob_list *clone_entry_tmp; 2414 2415 TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) { 2416 TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) { 2417 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 2418 free(clone_entry); 2419 } 2420 TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link); 2421 free(snapshot_entry); 2422 } 2423 2424 return 0; 2425 } 2426 2427 static void 2428 _spdk_bs_free(struct spdk_blob_store *bs) 2429 { 2430 _spdk_bs_blob_list_free(bs); 2431 2432 spdk_bs_unregister_md_thread(bs); 2433 spdk_io_device_unregister(bs, _spdk_bs_dev_destroy); 2434 } 2435 2436 void 2437 spdk_bs_opts_init(struct spdk_bs_opts *opts) 2438 { 2439 opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ; 2440 opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES; 2441 opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS; 2442 opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS; 2443 opts->clear_method = BS_CLEAR_WITH_UNMAP; 2444 memset(&opts->bstype, 0, sizeof(opts->bstype)); 2445 opts->iter_cb_fn = NULL; 2446 opts->iter_cb_arg = NULL; 2447 } 2448 2449 static int 2450 _spdk_bs_opts_verify(struct spdk_bs_opts *opts) 2451 { 2452 if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 || 2453 opts->max_channel_ops == 0) { 2454 SPDK_ERRLOG("Blobstore options cannot be set to 0\n"); 2455 return -1; 2456 } 2457 2458 return 0; 2459 } 2460 2461 static int 2462 _spdk_bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs) 2463 { 2464 struct spdk_blob_store *bs; 2465 uint64_t dev_size; 2466 int rc; 2467 2468 dev_size = dev->blocklen * dev->blockcnt; 2469 if (dev_size < opts->cluster_sz) { 2470 /* Device size cannot be smaller than cluster size of blobstore */ 2471 SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n", 2472 dev_size, opts->cluster_sz); 2473 return -ENOSPC; 2474 } 2475 if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) { 2476 /* Cluster size cannot be smaller than page size */ 2477 SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n", 2478 opts->cluster_sz, SPDK_BS_PAGE_SIZE); 2479 return -EINVAL; 2480 } 2481 bs = calloc(1, sizeof(struct spdk_blob_store)); 2482 if (!bs) { 2483 return -ENOMEM; 2484 } 2485 2486 TAILQ_INIT(&bs->blobs); 2487 TAILQ_INIT(&bs->snapshots); 2488 bs->dev = dev; 2489 bs->md_thread = spdk_get_thread(); 2490 assert(bs->md_thread != NULL); 2491 2492 /* 2493 * Do not use _spdk_bs_lba_to_cluster() here since blockcnt may not be an 2494 * even multiple of the cluster size. 2495 */ 2496 bs->cluster_sz = opts->cluster_sz; 2497 bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen); 2498 bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE; 2499 bs->num_free_clusters = bs->total_clusters; 2500 bs->used_clusters = spdk_bit_array_create(bs->total_clusters); 2501 bs->io_unit_size = dev->blocklen; 2502 if (bs->used_clusters == NULL) { 2503 free(bs); 2504 return -ENOMEM; 2505 } 2506 2507 bs->max_channel_ops = opts->max_channel_ops; 2508 bs->super_blob = SPDK_BLOBID_INVALID; 2509 memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype)); 2510 2511 /* The metadata is assumed to be at least 1 page */ 2512 bs->used_md_pages = spdk_bit_array_create(1); 2513 bs->used_blobids = spdk_bit_array_create(0); 2514 2515 pthread_mutex_init(&bs->used_clusters_mutex, NULL); 2516 2517 spdk_io_device_register(bs, _spdk_bs_channel_create, _spdk_bs_channel_destroy, 2518 sizeof(struct spdk_bs_channel), "blobstore"); 2519 rc = spdk_bs_register_md_thread(bs); 2520 if (rc == -1) { 2521 spdk_io_device_unregister(bs, NULL); 2522 pthread_mutex_destroy(&bs->used_clusters_mutex); 2523 spdk_bit_array_free(&bs->used_blobids); 2524 spdk_bit_array_free(&bs->used_md_pages); 2525 spdk_bit_array_free(&bs->used_clusters); 2526 free(bs); 2527 /* FIXME: this is a lie but don't know how to get a proper error code here */ 2528 return -ENOMEM; 2529 } 2530 2531 *_bs = bs; 2532 return 0; 2533 } 2534 2535 /* START spdk_bs_load, spdk_bs_load_ctx will used for both load and unload. */ 2536 2537 struct spdk_bs_load_ctx { 2538 struct spdk_blob_store *bs; 2539 struct spdk_bs_super_block *super; 2540 2541 struct spdk_bs_md_mask *mask; 2542 bool in_page_chain; 2543 uint32_t page_index; 2544 uint32_t cur_page; 2545 struct spdk_blob_md_page *page; 2546 2547 spdk_bs_sequence_t *seq; 2548 spdk_blob_op_with_handle_complete iter_cb_fn; 2549 void *iter_cb_arg; 2550 }; 2551 2552 static void 2553 _spdk_bs_load_ctx_fail(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 2554 { 2555 assert(bserrno != 0); 2556 2557 spdk_free(ctx->super); 2558 spdk_bs_sequence_finish(seq, bserrno); 2559 _spdk_bs_free(ctx->bs); 2560 free(ctx); 2561 } 2562 2563 static void 2564 _spdk_bs_set_mask(struct spdk_bit_array *array, struct spdk_bs_md_mask *mask) 2565 { 2566 uint32_t i = 0; 2567 2568 while (true) { 2569 i = spdk_bit_array_find_first_set(array, i); 2570 if (i >= mask->length) { 2571 break; 2572 } 2573 mask->mask[i / 8] |= 1U << (i % 8); 2574 i++; 2575 } 2576 } 2577 2578 static int 2579 _spdk_bs_load_mask(struct spdk_bit_array **array_ptr, struct spdk_bs_md_mask *mask) 2580 { 2581 struct spdk_bit_array *array; 2582 uint32_t i; 2583 2584 if (spdk_bit_array_resize(array_ptr, mask->length) < 0) { 2585 return -ENOMEM; 2586 } 2587 2588 array = *array_ptr; 2589 for (i = 0; i < mask->length; i++) { 2590 if (mask->mask[i / 8] & (1U << (i % 8))) { 2591 spdk_bit_array_set(array, i); 2592 } 2593 } 2594 2595 return 0; 2596 } 2597 2598 static void 2599 _spdk_bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 2600 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg) 2601 { 2602 /* Update the values in the super block */ 2603 super->super_blob = bs->super_blob; 2604 memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype)); 2605 super->crc = _spdk_blob_md_page_calc_crc(super); 2606 spdk_bs_sequence_write_dev(seq, super, _spdk_bs_page_to_lba(bs, 0), 2607 _spdk_bs_byte_to_lba(bs, sizeof(*super)), 2608 cb_fn, cb_arg); 2609 } 2610 2611 static void 2612 _spdk_bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2613 { 2614 struct spdk_bs_load_ctx *ctx = arg; 2615 uint64_t mask_size, lba, lba_count; 2616 2617 /* Write out the used clusters mask */ 2618 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 2619 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 2620 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 2621 if (!ctx->mask) { 2622 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2623 return; 2624 } 2625 2626 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS; 2627 ctx->mask->length = ctx->bs->total_clusters; 2628 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_clusters)); 2629 2630 _spdk_bs_set_mask(ctx->bs->used_clusters, ctx->mask); 2631 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 2632 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 2633 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2634 } 2635 2636 static void 2637 _spdk_bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2638 { 2639 struct spdk_bs_load_ctx *ctx = arg; 2640 uint64_t mask_size, lba, lba_count; 2641 2642 if (seq->bserrno) { 2643 _spdk_bs_load_ctx_fail(seq, ctx, seq->bserrno); 2644 return; 2645 } 2646 2647 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 2648 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 2649 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 2650 if (!ctx->mask) { 2651 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2652 return; 2653 } 2654 2655 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES; 2656 ctx->mask->length = ctx->super->md_len; 2657 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages)); 2658 2659 _spdk_bs_set_mask(ctx->bs->used_md_pages, ctx->mask); 2660 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 2661 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 2662 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2663 } 2664 2665 static void 2666 _spdk_bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 2667 { 2668 struct spdk_bs_load_ctx *ctx = arg; 2669 uint64_t mask_size, lba, lba_count; 2670 2671 if (ctx->super->used_blobid_mask_len == 0) { 2672 /* 2673 * This is a pre-v3 on-disk format where the blobid mask does not get 2674 * written to disk. 2675 */ 2676 cb_fn(seq, arg, 0); 2677 return; 2678 } 2679 2680 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 2681 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 2682 SPDK_MALLOC_DMA); 2683 if (!ctx->mask) { 2684 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2685 return; 2686 } 2687 2688 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS; 2689 ctx->mask->length = ctx->super->md_len; 2690 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids)); 2691 2692 _spdk_bs_set_mask(ctx->bs->used_blobids, ctx->mask); 2693 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 2694 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 2695 spdk_bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 2696 } 2697 2698 static void 2699 _spdk_bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno) 2700 { 2701 struct spdk_bs_load_ctx *ctx = arg; 2702 2703 if (bserrno == 0) { 2704 if (ctx->iter_cb_fn) { 2705 ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0); 2706 } 2707 _spdk_bs_blob_list_add(blob); 2708 spdk_bs_iter_next(ctx->bs, blob, _spdk_bs_load_iter, ctx); 2709 return; 2710 } 2711 2712 if (bserrno == -ENOENT) { 2713 bserrno = 0; 2714 } else { 2715 /* 2716 * This case needs to be looked at further. Same problem 2717 * exists with applications that rely on explicit blob 2718 * iteration. We should just skip the blob that failed 2719 * to load and continue on to the next one. 2720 */ 2721 SPDK_ERRLOG("Error in iterating blobs\n"); 2722 } 2723 2724 ctx->iter_cb_fn = NULL; 2725 2726 spdk_free(ctx->super); 2727 spdk_free(ctx->mask); 2728 spdk_bs_sequence_finish(ctx->seq, bserrno); 2729 free(ctx); 2730 } 2731 2732 static void 2733 _spdk_bs_load_complete(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 2734 { 2735 ctx->seq = seq; 2736 spdk_bs_iter_first(ctx->bs, _spdk_bs_load_iter, ctx); 2737 } 2738 2739 static void 2740 _spdk_bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2741 { 2742 struct spdk_bs_load_ctx *ctx = cb_arg; 2743 int rc; 2744 2745 /* The type must be correct */ 2746 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS); 2747 2748 /* The length of the mask (in bits) must not be greater than 2749 * the length of the buffer (converted to bits) */ 2750 assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8)); 2751 2752 /* The length of the mask must be exactly equal to the size 2753 * (in pages) of the metadata region */ 2754 assert(ctx->mask->length == ctx->super->md_len); 2755 2756 rc = _spdk_bs_load_mask(&ctx->bs->used_blobids, ctx->mask); 2757 if (rc < 0) { 2758 spdk_free(ctx->mask); 2759 _spdk_bs_load_ctx_fail(seq, ctx, rc); 2760 return; 2761 } 2762 2763 _spdk_bs_load_complete(seq, ctx, bserrno); 2764 } 2765 2766 static void 2767 _spdk_bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2768 { 2769 struct spdk_bs_load_ctx *ctx = cb_arg; 2770 uint64_t lba, lba_count, mask_size; 2771 int rc; 2772 2773 /* The type must be correct */ 2774 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS); 2775 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 2776 assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof( 2777 struct spdk_blob_md_page) * 8)); 2778 /* The length of the mask must be exactly equal to the total number of clusters */ 2779 assert(ctx->mask->length == ctx->bs->total_clusters); 2780 2781 rc = _spdk_bs_load_mask(&ctx->bs->used_clusters, ctx->mask); 2782 if (rc < 0) { 2783 spdk_free(ctx->mask); 2784 _spdk_bs_load_ctx_fail(seq, ctx, rc); 2785 return; 2786 } 2787 2788 ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->bs->used_clusters); 2789 assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters); 2790 2791 spdk_free(ctx->mask); 2792 2793 /* Read the used blobids mask */ 2794 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 2795 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 2796 SPDK_MALLOC_DMA); 2797 if (!ctx->mask) { 2798 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2799 return; 2800 } 2801 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 2802 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 2803 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2804 _spdk_bs_load_used_blobids_cpl, ctx); 2805 } 2806 2807 static void 2808 _spdk_bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2809 { 2810 struct spdk_bs_load_ctx *ctx = cb_arg; 2811 uint64_t lba, lba_count, mask_size; 2812 int rc; 2813 2814 /* The type must be correct */ 2815 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES); 2816 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 2817 assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE * 2818 8)); 2819 /* The length of the mask must be exactly equal to the size (in pages) of the metadata region */ 2820 assert(ctx->mask->length == ctx->super->md_len); 2821 2822 rc = _spdk_bs_load_mask(&ctx->bs->used_md_pages, ctx->mask); 2823 if (rc < 0) { 2824 spdk_free(ctx->mask); 2825 _spdk_bs_load_ctx_fail(seq, ctx, rc); 2826 return; 2827 } 2828 2829 spdk_free(ctx->mask); 2830 2831 /* Read the used clusters mask */ 2832 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 2833 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 2834 SPDK_MALLOC_DMA); 2835 if (!ctx->mask) { 2836 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2837 return; 2838 } 2839 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 2840 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 2841 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2842 _spdk_bs_load_used_clusters_cpl, ctx); 2843 } 2844 2845 static void 2846 _spdk_bs_load_read_used_pages(spdk_bs_sequence_t *seq, void *cb_arg) 2847 { 2848 struct spdk_bs_load_ctx *ctx = cb_arg; 2849 uint64_t lba, lba_count, mask_size; 2850 2851 /* Read the used pages mask */ 2852 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 2853 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 2854 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 2855 if (!ctx->mask) { 2856 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 2857 return; 2858 } 2859 2860 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 2861 lba_count = _spdk_bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 2862 spdk_bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 2863 _spdk_bs_load_used_pages_cpl, ctx); 2864 } 2865 2866 static int 2867 _spdk_bs_load_replay_md_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob_store *bs) 2868 { 2869 struct spdk_blob_md_descriptor *desc; 2870 size_t cur_desc = 0; 2871 2872 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 2873 while (cur_desc < sizeof(page->descriptors)) { 2874 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 2875 if (desc->length == 0) { 2876 /* If padding and length are 0, this terminates the page */ 2877 break; 2878 } 2879 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 2880 struct spdk_blob_md_descriptor_extent *desc_extent; 2881 unsigned int i, j; 2882 unsigned int cluster_count = 0; 2883 uint32_t cluster_idx; 2884 2885 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 2886 2887 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 2888 for (j = 0; j < desc_extent->extents[i].length; j++) { 2889 cluster_idx = desc_extent->extents[i].cluster_idx; 2890 /* 2891 * cluster_idx = 0 means an unallocated cluster - don't mark that 2892 * in the used cluster map. 2893 */ 2894 if (cluster_idx != 0) { 2895 spdk_bit_array_set(bs->used_clusters, cluster_idx + j); 2896 if (bs->num_free_clusters == 0) { 2897 return -ENOSPC; 2898 } 2899 bs->num_free_clusters--; 2900 } 2901 cluster_count++; 2902 } 2903 } 2904 if (cluster_count == 0) { 2905 return -EINVAL; 2906 } 2907 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 2908 /* Skip this item */ 2909 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 2910 /* Skip this item */ 2911 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 2912 /* Skip this item */ 2913 } else { 2914 /* Error */ 2915 return -EINVAL; 2916 } 2917 /* Advance to the next descriptor */ 2918 cur_desc += sizeof(*desc) + desc->length; 2919 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 2920 break; 2921 } 2922 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 2923 } 2924 return 0; 2925 } 2926 2927 static bool _spdk_bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx) 2928 { 2929 uint32_t crc; 2930 2931 crc = _spdk_blob_md_page_calc_crc(ctx->page); 2932 if (crc != ctx->page->crc) { 2933 return false; 2934 } 2935 2936 if (_spdk_bs_page_to_blobid(ctx->cur_page) != ctx->page->id) { 2937 return false; 2938 } 2939 return true; 2940 } 2941 2942 static void 2943 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg); 2944 2945 static void 2946 _spdk_bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2947 { 2948 struct spdk_bs_load_ctx *ctx = cb_arg; 2949 2950 _spdk_bs_load_complete(seq, ctx, bserrno); 2951 } 2952 2953 static void 2954 _spdk_bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2955 { 2956 struct spdk_bs_load_ctx *ctx = cb_arg; 2957 2958 spdk_free(ctx->mask); 2959 ctx->mask = NULL; 2960 2961 _spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_load_write_used_clusters_cpl); 2962 } 2963 2964 static void 2965 _spdk_bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2966 { 2967 struct spdk_bs_load_ctx *ctx = cb_arg; 2968 2969 spdk_free(ctx->mask); 2970 ctx->mask = NULL; 2971 2972 _spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_load_write_used_blobids_cpl); 2973 } 2974 2975 static void 2976 _spdk_bs_load_write_used_md(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2977 { 2978 _spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_load_write_used_pages_cpl); 2979 } 2980 2981 static void 2982 _spdk_bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2983 { 2984 struct spdk_bs_load_ctx *ctx = cb_arg; 2985 uint64_t num_md_clusters; 2986 uint64_t i; 2987 uint32_t page_num; 2988 2989 if (bserrno != 0) { 2990 _spdk_bs_load_ctx_fail(seq, ctx, bserrno); 2991 return; 2992 } 2993 2994 page_num = ctx->cur_page; 2995 if (_spdk_bs_load_cur_md_page_valid(ctx) == true) { 2996 if (ctx->page->sequence_num == 0 || ctx->in_page_chain == true) { 2997 spdk_bit_array_set(ctx->bs->used_md_pages, page_num); 2998 if (ctx->page->sequence_num == 0) { 2999 spdk_bit_array_set(ctx->bs->used_blobids, page_num); 3000 } 3001 if (_spdk_bs_load_replay_md_parse_page(ctx->page, ctx->bs)) { 3002 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3003 return; 3004 } 3005 if (ctx->page->next != SPDK_INVALID_MD_PAGE) { 3006 ctx->in_page_chain = true; 3007 ctx->cur_page = ctx->page->next; 3008 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 3009 return; 3010 } 3011 } 3012 } 3013 3014 ctx->in_page_chain = false; 3015 3016 do { 3017 ctx->page_index++; 3018 } while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true); 3019 3020 if (ctx->page_index < ctx->super->md_len) { 3021 ctx->cur_page = ctx->page_index; 3022 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 3023 } else { 3024 /* Claim all of the clusters used by the metadata */ 3025 num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster); 3026 for (i = 0; i < num_md_clusters; i++) { 3027 _spdk_bs_claim_cluster(ctx->bs, i); 3028 } 3029 spdk_free(ctx->page); 3030 _spdk_bs_load_write_used_md(seq, ctx, bserrno); 3031 } 3032 } 3033 3034 static void 3035 _spdk_bs_load_replay_cur_md_page(spdk_bs_sequence_t *seq, void *cb_arg) 3036 { 3037 struct spdk_bs_load_ctx *ctx = cb_arg; 3038 uint64_t lba; 3039 3040 assert(ctx->cur_page < ctx->super->md_len); 3041 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page); 3042 spdk_bs_sequence_read_dev(seq, ctx->page, lba, 3043 _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 3044 _spdk_bs_load_replay_md_cpl, ctx); 3045 } 3046 3047 static void 3048 _spdk_bs_load_replay_md(spdk_bs_sequence_t *seq, void *cb_arg) 3049 { 3050 struct spdk_bs_load_ctx *ctx = cb_arg; 3051 3052 ctx->page_index = 0; 3053 ctx->cur_page = 0; 3054 ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE, 3055 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3056 if (!ctx->page) { 3057 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3058 return; 3059 } 3060 _spdk_bs_load_replay_cur_md_page(seq, cb_arg); 3061 } 3062 3063 static void 3064 _spdk_bs_recover(spdk_bs_sequence_t *seq, void *cb_arg) 3065 { 3066 struct spdk_bs_load_ctx *ctx = cb_arg; 3067 int rc; 3068 3069 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len); 3070 if (rc < 0) { 3071 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3072 return; 3073 } 3074 3075 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len); 3076 if (rc < 0) { 3077 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3078 return; 3079 } 3080 3081 rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters); 3082 if (rc < 0) { 3083 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3084 return; 3085 } 3086 3087 ctx->bs->num_free_clusters = ctx->bs->total_clusters; 3088 _spdk_bs_load_replay_md(seq, cb_arg); 3089 } 3090 3091 static void 3092 _spdk_bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3093 { 3094 struct spdk_bs_load_ctx *ctx = cb_arg; 3095 uint32_t crc; 3096 int rc; 3097 static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH]; 3098 3099 if (ctx->super->version > SPDK_BS_VERSION || 3100 ctx->super->version < SPDK_BS_INITIAL_VERSION) { 3101 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3102 return; 3103 } 3104 3105 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 3106 sizeof(ctx->super->signature)) != 0) { 3107 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3108 return; 3109 } 3110 3111 crc = _spdk_blob_md_page_calc_crc(ctx->super); 3112 if (crc != ctx->super->crc) { 3113 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3114 return; 3115 } 3116 3117 if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 3118 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n"); 3119 } else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 3120 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n"); 3121 } else { 3122 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n"); 3123 SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 3124 SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 3125 _spdk_bs_load_ctx_fail(seq, ctx, -ENXIO); 3126 return; 3127 } 3128 3129 if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) { 3130 SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n", 3131 ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size); 3132 _spdk_bs_load_ctx_fail(seq, ctx, -EILSEQ); 3133 return; 3134 } 3135 3136 if (ctx->super->size == 0) { 3137 ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen; 3138 } 3139 3140 if (ctx->super->io_unit_size == 0) { 3141 ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE; 3142 } 3143 3144 /* Parse the super block */ 3145 ctx->bs->clean = 1; 3146 ctx->bs->cluster_sz = ctx->super->cluster_size; 3147 ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size; 3148 ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE; 3149 ctx->bs->io_unit_size = ctx->super->io_unit_size; 3150 rc = spdk_bit_array_resize(&ctx->bs->used_clusters, ctx->bs->total_clusters); 3151 if (rc < 0) { 3152 _spdk_bs_load_ctx_fail(seq, ctx, -ENOMEM); 3153 return; 3154 } 3155 ctx->bs->md_start = ctx->super->md_start; 3156 ctx->bs->md_len = ctx->super->md_len; 3157 ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up( 3158 ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster); 3159 ctx->bs->super_blob = ctx->super->super_blob; 3160 memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype)); 3161 3162 if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) { 3163 _spdk_bs_recover(seq, ctx); 3164 } else { 3165 _spdk_bs_load_read_used_pages(seq, ctx); 3166 } 3167 } 3168 3169 void 3170 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 3171 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 3172 { 3173 struct spdk_blob_store *bs; 3174 struct spdk_bs_cpl cpl; 3175 spdk_bs_sequence_t *seq; 3176 struct spdk_bs_load_ctx *ctx; 3177 struct spdk_bs_opts opts = {}; 3178 int err; 3179 3180 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev); 3181 3182 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 3183 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen); 3184 dev->destroy(dev); 3185 cb_fn(cb_arg, NULL, -EINVAL); 3186 return; 3187 } 3188 3189 if (o) { 3190 opts = *o; 3191 } else { 3192 spdk_bs_opts_init(&opts); 3193 } 3194 3195 if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) { 3196 dev->destroy(dev); 3197 cb_fn(cb_arg, NULL, -EINVAL); 3198 return; 3199 } 3200 3201 err = _spdk_bs_alloc(dev, &opts, &bs); 3202 if (err) { 3203 dev->destroy(dev); 3204 cb_fn(cb_arg, NULL, err); 3205 return; 3206 } 3207 3208 ctx = calloc(1, sizeof(*ctx)); 3209 if (!ctx) { 3210 _spdk_bs_free(bs); 3211 cb_fn(cb_arg, NULL, -ENOMEM); 3212 return; 3213 } 3214 3215 ctx->bs = bs; 3216 ctx->iter_cb_fn = opts.iter_cb_fn; 3217 ctx->iter_cb_arg = opts.iter_cb_arg; 3218 3219 /* Allocate memory for the super block */ 3220 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 3221 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3222 if (!ctx->super) { 3223 free(ctx); 3224 _spdk_bs_free(bs); 3225 cb_fn(cb_arg, NULL, -ENOMEM); 3226 return; 3227 } 3228 3229 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 3230 cpl.u.bs_handle.cb_fn = cb_fn; 3231 cpl.u.bs_handle.cb_arg = cb_arg; 3232 cpl.u.bs_handle.bs = bs; 3233 3234 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3235 if (!seq) { 3236 spdk_free(ctx->super); 3237 free(ctx); 3238 _spdk_bs_free(bs); 3239 cb_fn(cb_arg, NULL, -ENOMEM); 3240 return; 3241 } 3242 3243 /* Read the super block */ 3244 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3245 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3246 _spdk_bs_load_super_cpl, ctx); 3247 } 3248 3249 /* END spdk_bs_load */ 3250 3251 /* START spdk_bs_dump */ 3252 3253 struct spdk_bs_dump_ctx { 3254 struct spdk_blob_store *bs; 3255 struct spdk_bs_super_block *super; 3256 uint32_t cur_page; 3257 struct spdk_blob_md_page *page; 3258 spdk_bs_sequence_t *seq; 3259 FILE *fp; 3260 spdk_bs_dump_print_xattr print_xattr_fn; 3261 char xattr_name[4096]; 3262 }; 3263 3264 static void 3265 _spdk_bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_dump_ctx *ctx, int bserrno) 3266 { 3267 spdk_free(ctx->super); 3268 3269 /* 3270 * We need to defer calling spdk_bs_call_cpl() until after 3271 * dev destruction, so tuck these away for later use. 3272 */ 3273 ctx->bs->unload_err = bserrno; 3274 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 3275 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 3276 3277 spdk_bs_sequence_finish(seq, 0); 3278 _spdk_bs_free(ctx->bs); 3279 free(ctx); 3280 } 3281 3282 static void _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg); 3283 3284 static void 3285 _spdk_bs_dump_print_md_page(struct spdk_bs_dump_ctx *ctx) 3286 { 3287 uint32_t page_idx = ctx->cur_page; 3288 struct spdk_blob_md_page *page = ctx->page; 3289 struct spdk_blob_md_descriptor *desc; 3290 size_t cur_desc = 0; 3291 uint32_t crc; 3292 3293 fprintf(ctx->fp, "=========\n"); 3294 fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx); 3295 fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id); 3296 3297 crc = _spdk_blob_md_page_calc_crc(page); 3298 fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch"); 3299 3300 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 3301 while (cur_desc < sizeof(page->descriptors)) { 3302 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 3303 if (desc->length == 0) { 3304 /* If padding and length are 0, this terminates the page */ 3305 break; 3306 } 3307 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT) { 3308 struct spdk_blob_md_descriptor_extent *desc_extent; 3309 unsigned int i; 3310 3311 desc_extent = (struct spdk_blob_md_descriptor_extent *)desc; 3312 3313 for (i = 0; i < desc_extent->length / sizeof(desc_extent->extents[0]); i++) { 3314 if (desc_extent->extents[i].cluster_idx != 0) { 3315 fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32, 3316 desc_extent->extents[i].cluster_idx); 3317 } else { 3318 fprintf(ctx->fp, "Unallocated Extent - "); 3319 } 3320 fprintf(ctx->fp, " Length: %" PRIu32, desc_extent->extents[i].length); 3321 fprintf(ctx->fp, "\n"); 3322 } 3323 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 3324 struct spdk_blob_md_descriptor_xattr *desc_xattr; 3325 uint32_t i; 3326 3327 desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc; 3328 3329 if (desc_xattr->length != 3330 sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) + 3331 desc_xattr->name_length + desc_xattr->value_length) { 3332 } 3333 3334 memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length); 3335 ctx->xattr_name[desc_xattr->name_length] = '\0'; 3336 fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name); 3337 fprintf(ctx->fp, " value = \""); 3338 ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name, 3339 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 3340 desc_xattr->value_length); 3341 fprintf(ctx->fp, "\"\n"); 3342 for (i = 0; i < desc_xattr->value_length; i++) { 3343 if (i % 16 == 0) { 3344 fprintf(ctx->fp, " "); 3345 } 3346 fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i)); 3347 if ((i + 1) % 16 == 0) { 3348 fprintf(ctx->fp, "\n"); 3349 } 3350 } 3351 if (i % 16 != 0) { 3352 fprintf(ctx->fp, "\n"); 3353 } 3354 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 3355 /* TODO */ 3356 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 3357 /* TODO */ 3358 } else { 3359 /* Error */ 3360 } 3361 /* Advance to the next descriptor */ 3362 cur_desc += sizeof(*desc) + desc->length; 3363 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 3364 break; 3365 } 3366 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 3367 } 3368 } 3369 3370 static void 3371 _spdk_bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3372 { 3373 struct spdk_bs_dump_ctx *ctx = cb_arg; 3374 3375 if (bserrno != 0) { 3376 _spdk_bs_dump_finish(seq, ctx, bserrno); 3377 return; 3378 } 3379 3380 if (ctx->page->id != 0) { 3381 _spdk_bs_dump_print_md_page(ctx); 3382 } 3383 3384 ctx->cur_page++; 3385 3386 if (ctx->cur_page < ctx->super->md_len) { 3387 _spdk_bs_dump_read_md_page(seq, cb_arg); 3388 } else { 3389 spdk_free(ctx->page); 3390 _spdk_bs_dump_finish(seq, ctx, 0); 3391 } 3392 } 3393 3394 static void 3395 _spdk_bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg) 3396 { 3397 struct spdk_bs_dump_ctx *ctx = cb_arg; 3398 uint64_t lba; 3399 3400 assert(ctx->cur_page < ctx->super->md_len); 3401 lba = _spdk_bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page); 3402 spdk_bs_sequence_read_dev(seq, ctx->page, lba, 3403 _spdk_bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 3404 _spdk_bs_dump_read_md_page_cpl, ctx); 3405 } 3406 3407 static void 3408 _spdk_bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3409 { 3410 struct spdk_bs_dump_ctx *ctx = cb_arg; 3411 3412 fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature); 3413 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 3414 sizeof(ctx->super->signature)) != 0) { 3415 fprintf(ctx->fp, "(Mismatch)\n"); 3416 _spdk_bs_dump_finish(seq, ctx, bserrno); 3417 return; 3418 } else { 3419 fprintf(ctx->fp, "(OK)\n"); 3420 } 3421 fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version); 3422 fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc, 3423 (ctx->super->crc == _spdk_blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch"); 3424 fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype); 3425 fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size); 3426 fprintf(ctx->fp, "Super Blob ID: "); 3427 if (ctx->super->super_blob == SPDK_BLOBID_INVALID) { 3428 fprintf(ctx->fp, "(None)\n"); 3429 } else { 3430 fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob); 3431 } 3432 fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean); 3433 fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start); 3434 fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len); 3435 fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start); 3436 fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len); 3437 fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start); 3438 fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len); 3439 fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start); 3440 fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len); 3441 3442 ctx->cur_page = 0; 3443 ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, SPDK_BS_PAGE_SIZE, 3444 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3445 if (!ctx->page) { 3446 _spdk_bs_dump_finish(seq, ctx, -ENOMEM); 3447 return; 3448 } 3449 _spdk_bs_dump_read_md_page(seq, cb_arg); 3450 } 3451 3452 void 3453 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn, 3454 spdk_bs_op_complete cb_fn, void *cb_arg) 3455 { 3456 struct spdk_blob_store *bs; 3457 struct spdk_bs_cpl cpl; 3458 spdk_bs_sequence_t *seq; 3459 struct spdk_bs_dump_ctx *ctx; 3460 struct spdk_bs_opts opts = {}; 3461 int err; 3462 3463 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev); 3464 3465 spdk_bs_opts_init(&opts); 3466 3467 err = _spdk_bs_alloc(dev, &opts, &bs); 3468 if (err) { 3469 dev->destroy(dev); 3470 cb_fn(cb_arg, err); 3471 return; 3472 } 3473 3474 ctx = calloc(1, sizeof(*ctx)); 3475 if (!ctx) { 3476 _spdk_bs_free(bs); 3477 cb_fn(cb_arg, -ENOMEM); 3478 return; 3479 } 3480 3481 ctx->bs = bs; 3482 ctx->fp = fp; 3483 ctx->print_xattr_fn = print_xattr_fn; 3484 3485 /* Allocate memory for the super block */ 3486 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 3487 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3488 if (!ctx->super) { 3489 free(ctx); 3490 _spdk_bs_free(bs); 3491 cb_fn(cb_arg, -ENOMEM); 3492 return; 3493 } 3494 3495 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3496 cpl.u.bs_basic.cb_fn = cb_fn; 3497 cpl.u.bs_basic.cb_arg = cb_arg; 3498 3499 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3500 if (!seq) { 3501 spdk_free(ctx->super); 3502 free(ctx); 3503 _spdk_bs_free(bs); 3504 cb_fn(cb_arg, -ENOMEM); 3505 return; 3506 } 3507 3508 /* Read the super block */ 3509 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3510 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3511 _spdk_bs_dump_super_cpl, ctx); 3512 } 3513 3514 /* END spdk_bs_dump */ 3515 3516 /* START spdk_bs_init */ 3517 3518 struct spdk_bs_init_ctx { 3519 struct spdk_blob_store *bs; 3520 struct spdk_bs_super_block *super; 3521 }; 3522 3523 static void 3524 _spdk_bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3525 { 3526 struct spdk_bs_init_ctx *ctx = cb_arg; 3527 3528 spdk_free(ctx->super); 3529 free(ctx); 3530 3531 spdk_bs_sequence_finish(seq, bserrno); 3532 } 3533 3534 static void 3535 _spdk_bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3536 { 3537 struct spdk_bs_init_ctx *ctx = cb_arg; 3538 3539 /* Write super block */ 3540 spdk_bs_sequence_write_dev(seq, ctx->super, _spdk_bs_page_to_lba(ctx->bs, 0), 3541 _spdk_bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)), 3542 _spdk_bs_init_persist_super_cpl, ctx); 3543 } 3544 3545 void 3546 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 3547 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 3548 { 3549 struct spdk_bs_init_ctx *ctx; 3550 struct spdk_blob_store *bs; 3551 struct spdk_bs_cpl cpl; 3552 spdk_bs_sequence_t *seq; 3553 spdk_bs_batch_t *batch; 3554 uint64_t num_md_lba; 3555 uint64_t num_md_pages; 3556 uint64_t num_md_clusters; 3557 uint32_t i; 3558 struct spdk_bs_opts opts = {}; 3559 int rc; 3560 3561 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev); 3562 3563 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 3564 SPDK_ERRLOG("unsupported dev block length of %d\n", 3565 dev->blocklen); 3566 dev->destroy(dev); 3567 cb_fn(cb_arg, NULL, -EINVAL); 3568 return; 3569 } 3570 3571 if (o) { 3572 opts = *o; 3573 } else { 3574 spdk_bs_opts_init(&opts); 3575 } 3576 3577 if (_spdk_bs_opts_verify(&opts) != 0) { 3578 dev->destroy(dev); 3579 cb_fn(cb_arg, NULL, -EINVAL); 3580 return; 3581 } 3582 3583 rc = _spdk_bs_alloc(dev, &opts, &bs); 3584 if (rc) { 3585 dev->destroy(dev); 3586 cb_fn(cb_arg, NULL, rc); 3587 return; 3588 } 3589 3590 if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) { 3591 /* By default, allocate 1 page per cluster. 3592 * Technically, this over-allocates metadata 3593 * because more metadata will reduce the number 3594 * of usable clusters. This can be addressed with 3595 * more complex math in the future. 3596 */ 3597 bs->md_len = bs->total_clusters; 3598 } else { 3599 bs->md_len = opts.num_md_pages; 3600 } 3601 rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len); 3602 if (rc < 0) { 3603 _spdk_bs_free(bs); 3604 cb_fn(cb_arg, NULL, -ENOMEM); 3605 return; 3606 } 3607 3608 rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len); 3609 if (rc < 0) { 3610 _spdk_bs_free(bs); 3611 cb_fn(cb_arg, NULL, -ENOMEM); 3612 return; 3613 } 3614 3615 ctx = calloc(1, sizeof(*ctx)); 3616 if (!ctx) { 3617 _spdk_bs_free(bs); 3618 cb_fn(cb_arg, NULL, -ENOMEM); 3619 return; 3620 } 3621 3622 ctx->bs = bs; 3623 3624 /* Allocate memory for the super block */ 3625 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 3626 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3627 if (!ctx->super) { 3628 free(ctx); 3629 _spdk_bs_free(bs); 3630 cb_fn(cb_arg, NULL, -ENOMEM); 3631 return; 3632 } 3633 memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 3634 sizeof(ctx->super->signature)); 3635 ctx->super->version = SPDK_BS_VERSION; 3636 ctx->super->length = sizeof(*ctx->super); 3637 ctx->super->super_blob = bs->super_blob; 3638 ctx->super->clean = 0; 3639 ctx->super->cluster_size = bs->cluster_sz; 3640 ctx->super->io_unit_size = bs->io_unit_size; 3641 memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype)); 3642 3643 /* Calculate how many pages the metadata consumes at the front 3644 * of the disk. 3645 */ 3646 3647 /* The super block uses 1 page */ 3648 num_md_pages = 1; 3649 3650 /* The used_md_pages mask requires 1 bit per metadata page, rounded 3651 * up to the nearest page, plus a header. 3652 */ 3653 ctx->super->used_page_mask_start = num_md_pages; 3654 ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 3655 spdk_divide_round_up(bs->md_len, 8), 3656 SPDK_BS_PAGE_SIZE); 3657 num_md_pages += ctx->super->used_page_mask_len; 3658 3659 /* The used_clusters mask requires 1 bit per cluster, rounded 3660 * up to the nearest page, plus a header. 3661 */ 3662 ctx->super->used_cluster_mask_start = num_md_pages; 3663 ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 3664 spdk_divide_round_up(bs->total_clusters, 8), 3665 SPDK_BS_PAGE_SIZE); 3666 num_md_pages += ctx->super->used_cluster_mask_len; 3667 3668 /* The used_blobids mask requires 1 bit per metadata page, rounded 3669 * up to the nearest page, plus a header. 3670 */ 3671 ctx->super->used_blobid_mask_start = num_md_pages; 3672 ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 3673 spdk_divide_round_up(bs->md_len, 8), 3674 SPDK_BS_PAGE_SIZE); 3675 num_md_pages += ctx->super->used_blobid_mask_len; 3676 3677 /* The metadata region size was chosen above */ 3678 ctx->super->md_start = bs->md_start = num_md_pages; 3679 ctx->super->md_len = bs->md_len; 3680 num_md_pages += bs->md_len; 3681 3682 num_md_lba = _spdk_bs_page_to_lba(bs, num_md_pages); 3683 3684 ctx->super->size = dev->blockcnt * dev->blocklen; 3685 3686 ctx->super->crc = _spdk_blob_md_page_calc_crc(ctx->super); 3687 3688 num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster); 3689 if (num_md_clusters > bs->total_clusters) { 3690 SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, " 3691 "please decrease number of pages reserved for metadata " 3692 "or increase cluster size.\n"); 3693 spdk_free(ctx->super); 3694 free(ctx); 3695 _spdk_bs_free(bs); 3696 cb_fn(cb_arg, NULL, -ENOMEM); 3697 return; 3698 } 3699 /* Claim all of the clusters used by the metadata */ 3700 for (i = 0; i < num_md_clusters; i++) { 3701 _spdk_bs_claim_cluster(bs, i); 3702 } 3703 3704 bs->total_data_clusters = bs->num_free_clusters; 3705 3706 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 3707 cpl.u.bs_handle.cb_fn = cb_fn; 3708 cpl.u.bs_handle.cb_arg = cb_arg; 3709 cpl.u.bs_handle.bs = bs; 3710 3711 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3712 if (!seq) { 3713 spdk_free(ctx->super); 3714 free(ctx); 3715 _spdk_bs_free(bs); 3716 cb_fn(cb_arg, NULL, -ENOMEM); 3717 return; 3718 } 3719 3720 batch = spdk_bs_sequence_to_batch(seq, _spdk_bs_init_trim_cpl, ctx); 3721 3722 /* Clear metadata space */ 3723 spdk_bs_batch_write_zeroes_dev(batch, 0, num_md_lba); 3724 3725 if (opts.clear_method == BS_CLEAR_WITH_UNMAP) { 3726 /* Trim data clusters */ 3727 spdk_bs_batch_unmap_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba); 3728 } else if (opts.clear_method == BS_CLEAR_WITH_WRITE_ZEROES) { 3729 /* Write_zeroes to data clusters */ 3730 spdk_bs_batch_write_zeroes_dev(batch, num_md_lba, ctx->bs->dev->blockcnt - num_md_lba); 3731 } 3732 3733 spdk_bs_batch_close(batch); 3734 } 3735 3736 /* END spdk_bs_init */ 3737 3738 /* START spdk_bs_destroy */ 3739 3740 static void 3741 _spdk_bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3742 { 3743 struct spdk_bs_init_ctx *ctx = cb_arg; 3744 struct spdk_blob_store *bs = ctx->bs; 3745 3746 /* 3747 * We need to defer calling spdk_bs_call_cpl() until after 3748 * dev destruction, so tuck these away for later use. 3749 */ 3750 bs->unload_err = bserrno; 3751 memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 3752 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 3753 3754 spdk_bs_sequence_finish(seq, bserrno); 3755 3756 _spdk_bs_free(bs); 3757 free(ctx); 3758 } 3759 3760 void 3761 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, 3762 void *cb_arg) 3763 { 3764 struct spdk_bs_cpl cpl; 3765 spdk_bs_sequence_t *seq; 3766 struct spdk_bs_init_ctx *ctx; 3767 3768 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n"); 3769 3770 if (!TAILQ_EMPTY(&bs->blobs)) { 3771 SPDK_ERRLOG("Blobstore still has open blobs\n"); 3772 cb_fn(cb_arg, -EBUSY); 3773 return; 3774 } 3775 3776 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3777 cpl.u.bs_basic.cb_fn = cb_fn; 3778 cpl.u.bs_basic.cb_arg = cb_arg; 3779 3780 ctx = calloc(1, sizeof(*ctx)); 3781 if (!ctx) { 3782 cb_fn(cb_arg, -ENOMEM); 3783 return; 3784 } 3785 3786 ctx->bs = bs; 3787 3788 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3789 if (!seq) { 3790 free(ctx); 3791 cb_fn(cb_arg, -ENOMEM); 3792 return; 3793 } 3794 3795 /* Write zeroes to the super block */ 3796 spdk_bs_sequence_write_zeroes_dev(seq, 3797 _spdk_bs_page_to_lba(bs, 0), 3798 _spdk_bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)), 3799 _spdk_bs_destroy_trim_cpl, ctx); 3800 } 3801 3802 /* END spdk_bs_destroy */ 3803 3804 /* START spdk_bs_unload */ 3805 3806 static void 3807 _spdk_bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3808 { 3809 struct spdk_bs_load_ctx *ctx = cb_arg; 3810 3811 spdk_free(ctx->super); 3812 3813 /* 3814 * We need to defer calling spdk_bs_call_cpl() until after 3815 * dev destruction, so tuck these away for later use. 3816 */ 3817 ctx->bs->unload_err = bserrno; 3818 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 3819 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 3820 3821 spdk_bs_sequence_finish(seq, bserrno); 3822 3823 _spdk_bs_free(ctx->bs); 3824 free(ctx); 3825 } 3826 3827 static void 3828 _spdk_bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3829 { 3830 struct spdk_bs_load_ctx *ctx = cb_arg; 3831 3832 spdk_free(ctx->mask); 3833 ctx->super->clean = 1; 3834 3835 _spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_unload_write_super_cpl, ctx); 3836 } 3837 3838 static void 3839 _spdk_bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3840 { 3841 struct spdk_bs_load_ctx *ctx = cb_arg; 3842 3843 spdk_free(ctx->mask); 3844 ctx->mask = NULL; 3845 3846 _spdk_bs_write_used_clusters(seq, cb_arg, _spdk_bs_unload_write_used_clusters_cpl); 3847 } 3848 3849 static void 3850 _spdk_bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3851 { 3852 struct spdk_bs_load_ctx *ctx = cb_arg; 3853 3854 spdk_free(ctx->mask); 3855 ctx->mask = NULL; 3856 3857 _spdk_bs_write_used_blobids(seq, cb_arg, _spdk_bs_unload_write_used_blobids_cpl); 3858 } 3859 3860 static void 3861 _spdk_bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3862 { 3863 _spdk_bs_write_used_md(seq, cb_arg, _spdk_bs_unload_write_used_pages_cpl); 3864 } 3865 3866 void 3867 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg) 3868 { 3869 struct spdk_bs_cpl cpl; 3870 spdk_bs_sequence_t *seq; 3871 struct spdk_bs_load_ctx *ctx; 3872 3873 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n"); 3874 3875 if (!TAILQ_EMPTY(&bs->blobs)) { 3876 SPDK_ERRLOG("Blobstore still has open blobs\n"); 3877 cb_fn(cb_arg, -EBUSY); 3878 return; 3879 } 3880 3881 ctx = calloc(1, sizeof(*ctx)); 3882 if (!ctx) { 3883 cb_fn(cb_arg, -ENOMEM); 3884 return; 3885 } 3886 3887 ctx->bs = bs; 3888 3889 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 3890 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3891 if (!ctx->super) { 3892 free(ctx); 3893 cb_fn(cb_arg, -ENOMEM); 3894 return; 3895 } 3896 3897 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3898 cpl.u.bs_basic.cb_fn = cb_fn; 3899 cpl.u.bs_basic.cb_arg = cb_arg; 3900 3901 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3902 if (!seq) { 3903 spdk_free(ctx->super); 3904 free(ctx); 3905 cb_fn(cb_arg, -ENOMEM); 3906 return; 3907 } 3908 3909 /* Read super block */ 3910 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3911 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3912 _spdk_bs_unload_read_super_cpl, ctx); 3913 } 3914 3915 /* END spdk_bs_unload */ 3916 3917 /* START spdk_bs_set_super */ 3918 3919 struct spdk_bs_set_super_ctx { 3920 struct spdk_blob_store *bs; 3921 struct spdk_bs_super_block *super; 3922 }; 3923 3924 static void 3925 _spdk_bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3926 { 3927 struct spdk_bs_set_super_ctx *ctx = cb_arg; 3928 3929 if (bserrno != 0) { 3930 SPDK_ERRLOG("Unable to write to super block of blobstore\n"); 3931 } 3932 3933 spdk_free(ctx->super); 3934 3935 spdk_bs_sequence_finish(seq, bserrno); 3936 3937 free(ctx); 3938 } 3939 3940 static void 3941 _spdk_bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3942 { 3943 struct spdk_bs_set_super_ctx *ctx = cb_arg; 3944 3945 if (bserrno != 0) { 3946 SPDK_ERRLOG("Unable to read super block of blobstore\n"); 3947 spdk_free(ctx->super); 3948 spdk_bs_sequence_finish(seq, bserrno); 3949 free(ctx); 3950 return; 3951 } 3952 3953 _spdk_bs_write_super(seq, ctx->bs, ctx->super, _spdk_bs_set_super_write_cpl, ctx); 3954 } 3955 3956 void 3957 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid, 3958 spdk_bs_op_complete cb_fn, void *cb_arg) 3959 { 3960 struct spdk_bs_cpl cpl; 3961 spdk_bs_sequence_t *seq; 3962 struct spdk_bs_set_super_ctx *ctx; 3963 3964 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n"); 3965 3966 ctx = calloc(1, sizeof(*ctx)); 3967 if (!ctx) { 3968 cb_fn(cb_arg, -ENOMEM); 3969 return; 3970 } 3971 3972 ctx->bs = bs; 3973 3974 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 3975 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3976 if (!ctx->super) { 3977 free(ctx); 3978 cb_fn(cb_arg, -ENOMEM); 3979 return; 3980 } 3981 3982 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 3983 cpl.u.bs_basic.cb_fn = cb_fn; 3984 cpl.u.bs_basic.cb_arg = cb_arg; 3985 3986 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 3987 if (!seq) { 3988 spdk_free(ctx->super); 3989 free(ctx); 3990 cb_fn(cb_arg, -ENOMEM); 3991 return; 3992 } 3993 3994 bs->super_blob = blobid; 3995 3996 /* Read super block */ 3997 spdk_bs_sequence_read_dev(seq, ctx->super, _spdk_bs_page_to_lba(bs, 0), 3998 _spdk_bs_byte_to_lba(bs, sizeof(*ctx->super)), 3999 _spdk_bs_set_super_read_cpl, ctx); 4000 } 4001 4002 /* END spdk_bs_set_super */ 4003 4004 void 4005 spdk_bs_get_super(struct spdk_blob_store *bs, 4006 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4007 { 4008 if (bs->super_blob == SPDK_BLOBID_INVALID) { 4009 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT); 4010 } else { 4011 cb_fn(cb_arg, bs->super_blob, 0); 4012 } 4013 } 4014 4015 uint64_t 4016 spdk_bs_get_cluster_size(struct spdk_blob_store *bs) 4017 { 4018 return bs->cluster_sz; 4019 } 4020 4021 uint64_t 4022 spdk_bs_get_page_size(struct spdk_blob_store *bs) 4023 { 4024 return SPDK_BS_PAGE_SIZE; 4025 } 4026 4027 uint64_t 4028 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs) 4029 { 4030 return bs->io_unit_size; 4031 } 4032 4033 uint64_t 4034 spdk_bs_free_cluster_count(struct spdk_blob_store *bs) 4035 { 4036 return bs->num_free_clusters; 4037 } 4038 4039 uint64_t 4040 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs) 4041 { 4042 return bs->total_data_clusters; 4043 } 4044 4045 static int 4046 spdk_bs_register_md_thread(struct spdk_blob_store *bs) 4047 { 4048 bs->md_channel = spdk_get_io_channel(bs); 4049 if (!bs->md_channel) { 4050 SPDK_ERRLOG("Failed to get IO channel.\n"); 4051 return -1; 4052 } 4053 4054 return 0; 4055 } 4056 4057 static int 4058 spdk_bs_unregister_md_thread(struct spdk_blob_store *bs) 4059 { 4060 spdk_put_io_channel(bs->md_channel); 4061 4062 return 0; 4063 } 4064 4065 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob) 4066 { 4067 assert(blob != NULL); 4068 4069 return blob->id; 4070 } 4071 4072 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob) 4073 { 4074 assert(blob != NULL); 4075 4076 return _spdk_bs_cluster_to_page(blob->bs, blob->active.num_clusters); 4077 } 4078 4079 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob) 4080 { 4081 assert(blob != NULL); 4082 4083 return spdk_blob_get_num_pages(blob) * _spdk_bs_io_unit_per_page(blob->bs); 4084 } 4085 4086 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob) 4087 { 4088 assert(blob != NULL); 4089 4090 return blob->active.num_clusters; 4091 } 4092 4093 /* START spdk_bs_create_blob */ 4094 4095 static void 4096 _spdk_bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4097 { 4098 struct spdk_blob *blob = cb_arg; 4099 4100 _spdk_blob_free(blob); 4101 4102 spdk_bs_sequence_finish(seq, bserrno); 4103 } 4104 4105 static int 4106 _spdk_blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs, 4107 bool internal) 4108 { 4109 uint64_t i; 4110 size_t value_len = 0; 4111 int rc; 4112 const void *value = NULL; 4113 if (xattrs->count > 0 && xattrs->get_value == NULL) { 4114 return -EINVAL; 4115 } 4116 for (i = 0; i < xattrs->count; i++) { 4117 xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len); 4118 if (value == NULL || value_len == 0) { 4119 return -EINVAL; 4120 } 4121 rc = _spdk_blob_set_xattr(blob, xattrs->names[i], value, value_len, internal); 4122 if (rc < 0) { 4123 return rc; 4124 } 4125 } 4126 return 0; 4127 } 4128 4129 static void 4130 _spdk_blob_set_thin_provision(struct spdk_blob *blob) 4131 { 4132 _spdk_blob_verify_md_op(blob); 4133 blob->invalid_flags |= SPDK_BLOB_THIN_PROV; 4134 blob->state = SPDK_BLOB_STATE_DIRTY; 4135 } 4136 4137 static void 4138 _spdk_bs_create_blob(struct spdk_blob_store *bs, 4139 const struct spdk_blob_opts *opts, 4140 const struct spdk_blob_xattr_opts *internal_xattrs, 4141 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4142 { 4143 struct spdk_blob *blob; 4144 uint32_t page_idx; 4145 struct spdk_bs_cpl cpl; 4146 struct spdk_blob_opts opts_default; 4147 struct spdk_blob_xattr_opts internal_xattrs_default; 4148 spdk_bs_sequence_t *seq; 4149 spdk_blob_id id; 4150 int rc; 4151 4152 assert(spdk_get_thread() == bs->md_thread); 4153 4154 page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0); 4155 if (page_idx == UINT32_MAX) { 4156 cb_fn(cb_arg, 0, -ENOMEM); 4157 return; 4158 } 4159 spdk_bit_array_set(bs->used_blobids, page_idx); 4160 spdk_bit_array_set(bs->used_md_pages, page_idx); 4161 4162 id = _spdk_bs_page_to_blobid(page_idx); 4163 4164 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx); 4165 4166 blob = _spdk_blob_alloc(bs, id); 4167 if (!blob) { 4168 cb_fn(cb_arg, 0, -ENOMEM); 4169 return; 4170 } 4171 4172 if (!opts) { 4173 spdk_blob_opts_init(&opts_default); 4174 opts = &opts_default; 4175 } 4176 if (!internal_xattrs) { 4177 _spdk_blob_xattrs_init(&internal_xattrs_default); 4178 internal_xattrs = &internal_xattrs_default; 4179 } 4180 4181 rc = _spdk_blob_set_xattrs(blob, &opts->xattrs, false); 4182 if (rc < 0) { 4183 _spdk_blob_free(blob); 4184 cb_fn(cb_arg, 0, rc); 4185 return; 4186 } 4187 4188 rc = _spdk_blob_set_xattrs(blob, internal_xattrs, true); 4189 if (rc < 0) { 4190 _spdk_blob_free(blob); 4191 cb_fn(cb_arg, 0, rc); 4192 return; 4193 } 4194 4195 if (opts->thin_provision) { 4196 _spdk_blob_set_thin_provision(blob); 4197 } 4198 4199 rc = _spdk_blob_resize(blob, opts->num_clusters); 4200 if (rc < 0) { 4201 _spdk_blob_free(blob); 4202 cb_fn(cb_arg, 0, rc); 4203 return; 4204 } 4205 cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 4206 cpl.u.blobid.cb_fn = cb_fn; 4207 cpl.u.blobid.cb_arg = cb_arg; 4208 cpl.u.blobid.blobid = blob->id; 4209 4210 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 4211 if (!seq) { 4212 _spdk_blob_free(blob); 4213 cb_fn(cb_arg, 0, -ENOMEM); 4214 return; 4215 } 4216 4217 _spdk_blob_persist(seq, blob, _spdk_bs_create_blob_cpl, blob); 4218 } 4219 4220 void spdk_bs_create_blob(struct spdk_blob_store *bs, 4221 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4222 { 4223 _spdk_bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg); 4224 } 4225 4226 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts, 4227 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4228 { 4229 _spdk_bs_create_blob(bs, opts, NULL, cb_fn, cb_arg); 4230 } 4231 4232 /* END spdk_bs_create_blob */ 4233 4234 /* START blob_cleanup */ 4235 4236 struct spdk_clone_snapshot_ctx { 4237 struct spdk_bs_cpl cpl; 4238 int bserrno; 4239 bool frozen; 4240 4241 struct spdk_io_channel *channel; 4242 4243 /* Current cluster for inflate operation */ 4244 uint64_t cluster; 4245 4246 /* For inflation force allocation of all unallocated clusters and remove 4247 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */ 4248 bool allocate_all; 4249 4250 struct { 4251 spdk_blob_id id; 4252 struct spdk_blob *blob; 4253 } original; 4254 struct { 4255 spdk_blob_id id; 4256 struct spdk_blob *blob; 4257 } new; 4258 4259 /* xattrs specified for snapshot/clones only. They have no impact on 4260 * the original blobs xattrs. */ 4261 const struct spdk_blob_xattr_opts *xattrs; 4262 }; 4263 4264 static void 4265 _spdk_bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno) 4266 { 4267 struct spdk_clone_snapshot_ctx *ctx = cb_arg; 4268 struct spdk_bs_cpl *cpl = &ctx->cpl; 4269 4270 if (bserrno != 0) { 4271 if (ctx->bserrno != 0) { 4272 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 4273 } else { 4274 ctx->bserrno = bserrno; 4275 } 4276 } 4277 4278 switch (cpl->type) { 4279 case SPDK_BS_CPL_TYPE_BLOBID: 4280 cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno); 4281 break; 4282 case SPDK_BS_CPL_TYPE_BLOB_BASIC: 4283 cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno); 4284 break; 4285 default: 4286 SPDK_UNREACHABLE(); 4287 break; 4288 } 4289 4290 free(ctx); 4291 } 4292 4293 static void 4294 _spdk_bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno) 4295 { 4296 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4297 struct spdk_blob *origblob = ctx->original.blob; 4298 4299 if (bserrno != 0) { 4300 if (ctx->bserrno != 0) { 4301 SPDK_ERRLOG("Unfreeze error %d\n", bserrno); 4302 } else { 4303 ctx->bserrno = bserrno; 4304 } 4305 } 4306 4307 ctx->original.id = origblob->id; 4308 origblob->locked_operation_in_progress = false; 4309 4310 spdk_blob_close(origblob, _spdk_bs_clone_snapshot_cleanup_finish, ctx); 4311 } 4312 4313 static void 4314 _spdk_bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno) 4315 { 4316 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4317 struct spdk_blob *origblob = ctx->original.blob; 4318 4319 if (bserrno != 0) { 4320 if (ctx->bserrno != 0) { 4321 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 4322 } else { 4323 ctx->bserrno = bserrno; 4324 } 4325 } 4326 4327 if (ctx->frozen) { 4328 /* Unfreeze any outstanding I/O */ 4329 _spdk_blob_unfreeze_io(origblob, _spdk_bs_snapshot_unfreeze_cpl, ctx); 4330 } else { 4331 _spdk_bs_snapshot_unfreeze_cpl(ctx, 0); 4332 } 4333 4334 } 4335 4336 static void 4337 _spdk_bs_clone_snapshot_newblob_cleanup(void *cb_arg, int bserrno) 4338 { 4339 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4340 struct spdk_blob *newblob = ctx->new.blob; 4341 4342 if (bserrno != 0) { 4343 if (ctx->bserrno != 0) { 4344 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 4345 } else { 4346 ctx->bserrno = bserrno; 4347 } 4348 } 4349 4350 ctx->new.id = newblob->id; 4351 spdk_blob_close(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4352 } 4353 4354 /* END blob_cleanup */ 4355 4356 /* START spdk_bs_create_snapshot */ 4357 4358 static void 4359 _spdk_bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2) 4360 { 4361 uint64_t *cluster_temp; 4362 4363 cluster_temp = blob1->active.clusters; 4364 blob1->active.clusters = blob2->active.clusters; 4365 blob2->active.clusters = cluster_temp; 4366 } 4367 4368 static void 4369 _spdk_bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno) 4370 { 4371 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4372 struct spdk_blob *origblob = ctx->original.blob; 4373 struct spdk_blob *newblob = ctx->new.blob; 4374 4375 if (bserrno != 0) { 4376 _spdk_bs_snapshot_swap_cluster_maps(newblob, origblob); 4377 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4378 return; 4379 } 4380 4381 /* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */ 4382 bserrno = _spdk_blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true); 4383 if (bserrno != 0) { 4384 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4385 return; 4386 } 4387 4388 _spdk_bs_blob_list_add(ctx->original.blob); 4389 4390 spdk_blob_set_read_only(newblob); 4391 4392 /* sync snapshot metadata */ 4393 spdk_blob_sync_md(newblob, _spdk_bs_clone_snapshot_origblob_cleanup, cb_arg); 4394 } 4395 4396 static void 4397 _spdk_bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno) 4398 { 4399 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4400 struct spdk_blob *origblob = ctx->original.blob; 4401 struct spdk_blob *newblob = ctx->new.blob; 4402 4403 if (bserrno != 0) { 4404 /* return cluster map back to original */ 4405 _spdk_bs_snapshot_swap_cluster_maps(newblob, origblob); 4406 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4407 return; 4408 } 4409 4410 /* Set internal xattr for snapshot id */ 4411 bserrno = _spdk_blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true); 4412 if (bserrno != 0) { 4413 /* return cluster map back to original */ 4414 _spdk_bs_snapshot_swap_cluster_maps(newblob, origblob); 4415 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4416 return; 4417 } 4418 4419 _spdk_bs_blob_list_remove(origblob); 4420 origblob->parent_id = newblob->id; 4421 4422 /* Create new back_bs_dev for snapshot */ 4423 origblob->back_bs_dev = spdk_bs_create_blob_bs_dev(newblob); 4424 if (origblob->back_bs_dev == NULL) { 4425 /* return cluster map back to original */ 4426 _spdk_bs_snapshot_swap_cluster_maps(newblob, origblob); 4427 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL); 4428 return; 4429 } 4430 4431 /* set clone blob as thin provisioned */ 4432 _spdk_blob_set_thin_provision(origblob); 4433 4434 _spdk_bs_blob_list_add(newblob); 4435 4436 /* sync clone metadata */ 4437 spdk_blob_sync_md(origblob, _spdk_bs_snapshot_origblob_sync_cpl, ctx); 4438 } 4439 4440 static void 4441 _spdk_bs_snapshot_freeze_cpl(void *cb_arg, int rc) 4442 { 4443 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4444 struct spdk_blob *origblob = ctx->original.blob; 4445 struct spdk_blob *newblob = ctx->new.blob; 4446 int bserrno; 4447 4448 if (rc != 0) { 4449 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, rc); 4450 return; 4451 } 4452 4453 ctx->frozen = true; 4454 4455 /* set new back_bs_dev for snapshot */ 4456 newblob->back_bs_dev = origblob->back_bs_dev; 4457 /* Set invalid flags from origblob */ 4458 newblob->invalid_flags = origblob->invalid_flags; 4459 4460 /* inherit parent from original blob if set */ 4461 newblob->parent_id = origblob->parent_id; 4462 if (origblob->parent_id != SPDK_BLOBID_INVALID) { 4463 /* Set internal xattr for snapshot id */ 4464 bserrno = _spdk_blob_set_xattr(newblob, BLOB_SNAPSHOT, 4465 &origblob->parent_id, sizeof(spdk_blob_id), true); 4466 if (bserrno != 0) { 4467 _spdk_bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 4468 return; 4469 } 4470 } 4471 4472 /* swap cluster maps */ 4473 _spdk_bs_snapshot_swap_cluster_maps(newblob, origblob); 4474 4475 /* sync snapshot metadata */ 4476 spdk_blob_sync_md(newblob, _spdk_bs_snapshot_newblob_sync_cpl, ctx); 4477 } 4478 4479 static void 4480 _spdk_bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4481 { 4482 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4483 struct spdk_blob *origblob = ctx->original.blob; 4484 struct spdk_blob *newblob = _blob; 4485 4486 if (bserrno != 0) { 4487 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4488 return; 4489 } 4490 4491 ctx->new.blob = newblob; 4492 4493 /* Zero out newblob cluster map */ 4494 memset(newblob->active.clusters, 0, 4495 newblob->active.num_clusters * sizeof(newblob->active.clusters)); 4496 4497 _spdk_blob_freeze_io(origblob, _spdk_bs_snapshot_freeze_cpl, ctx); 4498 } 4499 4500 static void 4501 _spdk_bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 4502 { 4503 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4504 struct spdk_blob *origblob = ctx->original.blob; 4505 4506 if (bserrno != 0) { 4507 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4508 return; 4509 } 4510 4511 ctx->new.id = blobid; 4512 ctx->cpl.u.blobid.blobid = blobid; 4513 4514 spdk_bs_open_blob(origblob->bs, ctx->new.id, _spdk_bs_snapshot_newblob_open_cpl, ctx); 4515 } 4516 4517 4518 static void 4519 _spdk_bs_xattr_snapshot(void *arg, const char *name, 4520 const void **value, size_t *value_len) 4521 { 4522 assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0); 4523 4524 struct spdk_blob *blob = (struct spdk_blob *)arg; 4525 *value = &blob->id; 4526 *value_len = sizeof(blob->id); 4527 } 4528 4529 static void 4530 _spdk_bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4531 { 4532 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4533 struct spdk_blob_opts opts; 4534 struct spdk_blob_xattr_opts internal_xattrs; 4535 char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS }; 4536 4537 if (bserrno != 0) { 4538 _spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno); 4539 return; 4540 } 4541 4542 ctx->original.blob = _blob; 4543 4544 if (_blob->data_ro || _blob->md_ro) { 4545 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n", 4546 _blob->id); 4547 ctx->bserrno = -EINVAL; 4548 spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx); 4549 return; 4550 } 4551 4552 if (_blob->locked_operation_in_progress) { 4553 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot - another operation in progress\n"); 4554 ctx->bserrno = -EBUSY; 4555 spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx); 4556 return; 4557 } 4558 4559 _blob->locked_operation_in_progress = true; 4560 4561 spdk_blob_opts_init(&opts); 4562 _spdk_blob_xattrs_init(&internal_xattrs); 4563 4564 /* Change the size of new blob to the same as in original blob, 4565 * but do not allocate clusters */ 4566 opts.thin_provision = true; 4567 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 4568 4569 /* If there are any xattrs specified for snapshot, set them now */ 4570 if (ctx->xattrs) { 4571 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 4572 } 4573 /* Set internal xattr SNAPSHOT_IN_PROGRESS */ 4574 internal_xattrs.count = 1; 4575 internal_xattrs.ctx = _blob; 4576 internal_xattrs.names = xattrs_names; 4577 internal_xattrs.get_value = _spdk_bs_xattr_snapshot; 4578 4579 _spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs, 4580 _spdk_bs_snapshot_newblob_create_cpl, ctx); 4581 } 4582 4583 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid, 4584 const struct spdk_blob_xattr_opts *snapshot_xattrs, 4585 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4586 { 4587 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 4588 4589 if (!ctx) { 4590 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 4591 return; 4592 } 4593 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 4594 ctx->cpl.u.blobid.cb_fn = cb_fn; 4595 ctx->cpl.u.blobid.cb_arg = cb_arg; 4596 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 4597 ctx->bserrno = 0; 4598 ctx->frozen = false; 4599 ctx->original.id = blobid; 4600 ctx->xattrs = snapshot_xattrs; 4601 4602 spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_snapshot_origblob_open_cpl, ctx); 4603 } 4604 /* END spdk_bs_create_snapshot */ 4605 4606 /* START spdk_bs_create_clone */ 4607 4608 static void 4609 _spdk_bs_xattr_clone(void *arg, const char *name, 4610 const void **value, size_t *value_len) 4611 { 4612 assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0); 4613 4614 struct spdk_blob *blob = (struct spdk_blob *)arg; 4615 *value = &blob->id; 4616 *value_len = sizeof(blob->id); 4617 } 4618 4619 static void 4620 _spdk_bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4621 { 4622 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4623 struct spdk_blob *clone = _blob; 4624 4625 ctx->new.blob = clone; 4626 _spdk_bs_blob_list_add(clone); 4627 4628 spdk_blob_close(clone, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4629 } 4630 4631 static void 4632 _spdk_bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 4633 { 4634 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4635 4636 ctx->cpl.u.blobid.blobid = blobid; 4637 spdk_bs_open_blob(ctx->original.blob->bs, blobid, _spdk_bs_clone_newblob_open_cpl, ctx); 4638 } 4639 4640 static void 4641 _spdk_bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4642 { 4643 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4644 struct spdk_blob_opts opts; 4645 struct spdk_blob_xattr_opts internal_xattrs; 4646 char *xattr_names[] = { BLOB_SNAPSHOT }; 4647 4648 if (bserrno != 0) { 4649 _spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno); 4650 return; 4651 } 4652 4653 ctx->original.blob = _blob; 4654 4655 if (!_blob->data_ro || !_blob->md_ro) { 4656 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n"); 4657 ctx->bserrno = -EINVAL; 4658 spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx); 4659 return; 4660 } 4661 4662 if (_blob->locked_operation_in_progress) { 4663 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create clone - another operation in progress\n"); 4664 ctx->bserrno = -EBUSY; 4665 spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx); 4666 return; 4667 } 4668 4669 _blob->locked_operation_in_progress = true; 4670 4671 spdk_blob_opts_init(&opts); 4672 _spdk_blob_xattrs_init(&internal_xattrs); 4673 4674 opts.thin_provision = true; 4675 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 4676 if (ctx->xattrs) { 4677 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 4678 } 4679 4680 /* Set internal xattr BLOB_SNAPSHOT */ 4681 internal_xattrs.count = 1; 4682 internal_xattrs.ctx = _blob; 4683 internal_xattrs.names = xattr_names; 4684 internal_xattrs.get_value = _spdk_bs_xattr_clone; 4685 4686 _spdk_bs_create_blob(_blob->bs, &opts, &internal_xattrs, 4687 _spdk_bs_clone_newblob_create_cpl, ctx); 4688 } 4689 4690 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid, 4691 const struct spdk_blob_xattr_opts *clone_xattrs, 4692 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 4693 { 4694 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 4695 4696 if (!ctx) { 4697 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 4698 return; 4699 } 4700 4701 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 4702 ctx->cpl.u.blobid.cb_fn = cb_fn; 4703 ctx->cpl.u.blobid.cb_arg = cb_arg; 4704 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 4705 ctx->bserrno = 0; 4706 ctx->xattrs = clone_xattrs; 4707 ctx->original.id = blobid; 4708 4709 spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_clone_origblob_open_cpl, ctx); 4710 } 4711 4712 /* END spdk_bs_create_clone */ 4713 4714 /* START spdk_bs_inflate_blob */ 4715 4716 static void 4717 _spdk_bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno) 4718 { 4719 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4720 struct spdk_blob *_blob = ctx->original.blob; 4721 4722 if (bserrno != 0) { 4723 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4724 return; 4725 } 4726 4727 assert(_parent != NULL); 4728 4729 _spdk_bs_blob_list_remove(_blob); 4730 _blob->parent_id = _parent->id; 4731 _spdk_blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id, 4732 sizeof(spdk_blob_id), true); 4733 4734 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 4735 _blob->back_bs_dev = spdk_bs_create_blob_bs_dev(_parent); 4736 _spdk_bs_blob_list_add(_blob); 4737 4738 spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4739 } 4740 4741 static void 4742 _spdk_bs_inflate_blob_done(void *cb_arg, int bserrno) 4743 { 4744 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4745 struct spdk_blob *_blob = ctx->original.blob; 4746 struct spdk_blob *_parent; 4747 4748 if (bserrno != 0) { 4749 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4750 return; 4751 } 4752 4753 if (ctx->allocate_all) { 4754 /* remove thin provisioning */ 4755 _spdk_bs_blob_list_remove(_blob); 4756 _spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 4757 _blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV; 4758 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 4759 _blob->back_bs_dev = NULL; 4760 _blob->parent_id = SPDK_BLOBID_INVALID; 4761 } else { 4762 _parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob; 4763 if (_parent->parent_id != SPDK_BLOBID_INVALID) { 4764 /* We must change the parent of the inflated blob */ 4765 spdk_bs_open_blob(_blob->bs, _parent->parent_id, 4766 _spdk_bs_inflate_blob_set_parent_cpl, ctx); 4767 return; 4768 } 4769 4770 _spdk_bs_blob_list_remove(_blob); 4771 _spdk_blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 4772 _blob->parent_id = SPDK_BLOBID_INVALID; 4773 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 4774 _blob->back_bs_dev = spdk_bs_create_zeroes_dev(); 4775 } 4776 4777 _blob->state = SPDK_BLOB_STATE_DIRTY; 4778 spdk_blob_sync_md(_blob, _spdk_bs_clone_snapshot_origblob_cleanup, ctx); 4779 } 4780 4781 /* Check if cluster needs allocation */ 4782 static inline bool 4783 _spdk_bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all) 4784 { 4785 struct spdk_blob_bs_dev *b; 4786 4787 assert(blob != NULL); 4788 4789 if (blob->active.clusters[cluster] != 0) { 4790 /* Cluster is already allocated */ 4791 return false; 4792 } 4793 4794 if (blob->parent_id == SPDK_BLOBID_INVALID) { 4795 /* Blob have no parent blob */ 4796 return allocate_all; 4797 } 4798 4799 b = (struct spdk_blob_bs_dev *)blob->back_bs_dev; 4800 return (allocate_all || b->blob->active.clusters[cluster] != 0); 4801 } 4802 4803 static void 4804 _spdk_bs_inflate_blob_touch_next(void *cb_arg, int bserrno) 4805 { 4806 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4807 struct spdk_blob *_blob = ctx->original.blob; 4808 uint64_t offset; 4809 4810 if (bserrno != 0) { 4811 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 4812 return; 4813 } 4814 4815 for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) { 4816 if (_spdk_bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) { 4817 break; 4818 } 4819 } 4820 4821 if (ctx->cluster < _blob->active.num_clusters) { 4822 offset = _spdk_bs_cluster_to_lba(_blob->bs, ctx->cluster); 4823 4824 /* We may safely increment a cluster before write */ 4825 ctx->cluster++; 4826 4827 /* Use zero length write to touch a cluster */ 4828 spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0, 4829 _spdk_bs_inflate_blob_touch_next, ctx); 4830 } else { 4831 _spdk_bs_inflate_blob_done(cb_arg, bserrno); 4832 } 4833 } 4834 4835 static void 4836 _spdk_bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 4837 { 4838 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 4839 uint64_t lfc; /* lowest free cluster */ 4840 uint64_t i; 4841 4842 if (bserrno != 0) { 4843 _spdk_bs_clone_snapshot_cleanup_finish(ctx, bserrno); 4844 return; 4845 } 4846 4847 ctx->original.blob = _blob; 4848 4849 if (_blob->locked_operation_in_progress) { 4850 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot inflate blob - another operation in progress\n"); 4851 ctx->bserrno = -EBUSY; 4852 spdk_blob_close(_blob, _spdk_bs_clone_snapshot_cleanup_finish, ctx); 4853 return; 4854 } 4855 4856 _blob->locked_operation_in_progress = true; 4857 4858 if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) { 4859 /* This blob have no parent, so we cannot decouple it. */ 4860 SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n"); 4861 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL); 4862 return; 4863 } 4864 4865 if (spdk_blob_is_thin_provisioned(_blob) == false) { 4866 /* This is not thin provisioned blob. No need to inflate. */ 4867 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, 0); 4868 return; 4869 } 4870 4871 /* Do two passes - one to verify that we can obtain enough clusters 4872 * and another to actually claim them. 4873 */ 4874 lfc = 0; 4875 for (i = 0; i < _blob->active.num_clusters; i++) { 4876 if (_spdk_bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) { 4877 lfc = spdk_bit_array_find_first_clear(_blob->bs->used_clusters, lfc); 4878 if (lfc == UINT32_MAX) { 4879 /* No more free clusters. Cannot satisfy the request */ 4880 _spdk_bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC); 4881 return; 4882 } 4883 lfc++; 4884 } 4885 } 4886 4887 ctx->cluster = 0; 4888 _spdk_bs_inflate_blob_touch_next(ctx, 0); 4889 } 4890 4891 static void 4892 _spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 4893 spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg) 4894 { 4895 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 4896 4897 if (!ctx) { 4898 cb_fn(cb_arg, -ENOMEM); 4899 return; 4900 } 4901 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 4902 ctx->cpl.u.bs_basic.cb_fn = cb_fn; 4903 ctx->cpl.u.bs_basic.cb_arg = cb_arg; 4904 ctx->bserrno = 0; 4905 ctx->original.id = blobid; 4906 ctx->channel = channel; 4907 ctx->allocate_all = allocate_all; 4908 4909 spdk_bs_open_blob(bs, ctx->original.id, _spdk_bs_inflate_blob_open_cpl, ctx); 4910 } 4911 4912 void 4913 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 4914 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 4915 { 4916 _spdk_bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg); 4917 } 4918 4919 void 4920 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 4921 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 4922 { 4923 _spdk_bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg); 4924 } 4925 /* END spdk_bs_inflate_blob */ 4926 4927 /* START spdk_blob_resize */ 4928 struct spdk_bs_resize_ctx { 4929 spdk_blob_op_complete cb_fn; 4930 void *cb_arg; 4931 struct spdk_blob *blob; 4932 uint64_t sz; 4933 int rc; 4934 }; 4935 4936 static void 4937 _spdk_bs_resize_unfreeze_cpl(void *cb_arg, int rc) 4938 { 4939 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 4940 4941 if (rc != 0) { 4942 SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc); 4943 } 4944 4945 if (ctx->rc != 0) { 4946 SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc); 4947 rc = ctx->rc; 4948 } 4949 4950 ctx->blob->locked_operation_in_progress = false; 4951 4952 ctx->cb_fn(ctx->cb_arg, rc); 4953 free(ctx); 4954 } 4955 4956 static void 4957 _spdk_bs_resize_freeze_cpl(void *cb_arg, int rc) 4958 { 4959 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 4960 4961 if (rc != 0) { 4962 ctx->blob->locked_operation_in_progress = false; 4963 ctx->cb_fn(ctx->cb_arg, rc); 4964 free(ctx); 4965 return; 4966 } 4967 4968 ctx->rc = _spdk_blob_resize(ctx->blob, ctx->sz); 4969 4970 _spdk_blob_unfreeze_io(ctx->blob, _spdk_bs_resize_unfreeze_cpl, ctx); 4971 } 4972 4973 void 4974 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg) 4975 { 4976 struct spdk_bs_resize_ctx *ctx; 4977 4978 _spdk_blob_verify_md_op(blob); 4979 4980 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz); 4981 4982 if (blob->md_ro) { 4983 cb_fn(cb_arg, -EPERM); 4984 return; 4985 } 4986 4987 if (sz == blob->active.num_clusters) { 4988 cb_fn(cb_arg, 0); 4989 return; 4990 } 4991 4992 if (blob->locked_operation_in_progress) { 4993 cb_fn(cb_arg, -EBUSY); 4994 return; 4995 } 4996 4997 ctx = calloc(1, sizeof(*ctx)); 4998 if (!ctx) { 4999 cb_fn(cb_arg, -ENOMEM); 5000 return; 5001 } 5002 5003 blob->locked_operation_in_progress = true; 5004 ctx->cb_fn = cb_fn; 5005 ctx->cb_arg = cb_arg; 5006 ctx->blob = blob; 5007 ctx->sz = sz; 5008 _spdk_blob_freeze_io(blob, _spdk_bs_resize_freeze_cpl, ctx); 5009 } 5010 5011 /* END spdk_blob_resize */ 5012 5013 5014 /* START spdk_bs_delete_blob */ 5015 5016 static void 5017 _spdk_bs_delete_ebusy_close_cpl(void *cb_arg, int bserrno) 5018 { 5019 spdk_bs_sequence_t *seq = cb_arg; 5020 5021 spdk_bs_sequence_finish(seq, -EBUSY); 5022 } 5023 5024 static void 5025 _spdk_bs_delete_close_cpl(void *cb_arg, int bserrno) 5026 { 5027 spdk_bs_sequence_t *seq = cb_arg; 5028 5029 spdk_bs_sequence_finish(seq, bserrno); 5030 } 5031 5032 static void 5033 _spdk_bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5034 { 5035 struct spdk_blob *blob = cb_arg; 5036 5037 if (bserrno != 0) { 5038 /* 5039 * We already removed this blob from the blobstore tailq, so 5040 * we need to free it here since this is the last reference 5041 * to it. 5042 */ 5043 _spdk_blob_free(blob); 5044 _spdk_bs_delete_close_cpl(seq, bserrno); 5045 return; 5046 } 5047 5048 /* 5049 * This will immediately decrement the ref_count and call 5050 * the completion routine since the metadata state is clean. 5051 * By calling spdk_blob_close, we reduce the number of call 5052 * points into code that touches the blob->open_ref count 5053 * and the blobstore's blob list. 5054 */ 5055 spdk_blob_close(blob, _spdk_bs_delete_close_cpl, seq); 5056 } 5057 5058 static int 5059 _spdk_bs_is_blob_deletable(struct spdk_blob *blob) 5060 { 5061 struct spdk_blob_list *snapshot_entry = NULL; 5062 5063 if (blob->open_ref > 1) { 5064 /* Someone has this blob open (besides this delete context). */ 5065 return -EBUSY; 5066 } 5067 5068 /* Check if this is a snapshot with clones */ 5069 snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id); 5070 if (snapshot_entry != NULL) { 5071 /* If snapshot have clones, we cannot remove it */ 5072 if (!TAILQ_EMPTY(&snapshot_entry->clones)) { 5073 SPDK_ERRLOG("Cannot remove snapshot with clones\n"); 5074 return -EBUSY; 5075 } 5076 } 5077 5078 return 0; 5079 } 5080 5081 static void 5082 _spdk_bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno) 5083 { 5084 spdk_bs_sequence_t *seq = cb_arg; 5085 struct spdk_blob_list *snapshot = NULL; 5086 uint32_t page_num; 5087 5088 if (bserrno != 0) { 5089 spdk_bs_sequence_finish(seq, bserrno); 5090 return; 5091 } 5092 5093 _spdk_blob_verify_md_op(blob); 5094 5095 bserrno = _spdk_bs_is_blob_deletable(blob); 5096 if (bserrno) { 5097 spdk_blob_close(blob, _spdk_bs_delete_ebusy_close_cpl, seq); 5098 return; 5099 } 5100 5101 _spdk_bs_blob_list_remove(blob); 5102 5103 if (blob->locked_operation_in_progress) { 5104 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress\n"); 5105 spdk_blob_close(blob, _spdk_bs_delete_ebusy_close_cpl, seq); 5106 return; 5107 } 5108 5109 blob->locked_operation_in_progress = true; 5110 5111 /* 5112 * Remove the blob from the blob_store list now, to ensure it does not 5113 * get returned after this point by _spdk_blob_lookup(). 5114 */ 5115 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 5116 5117 /* If blob is a snapshot then remove it from the list */ 5118 snapshot = _spdk_bs_get_snapshot_entry(blob->bs, blob->id); 5119 if (snapshot != NULL) { 5120 TAILQ_REMOVE(&blob->bs->snapshots, snapshot, link); 5121 free(snapshot); 5122 } 5123 5124 page_num = _spdk_bs_blobid_to_page(blob->id); 5125 spdk_bit_array_clear(blob->bs->used_blobids, page_num); 5126 blob->state = SPDK_BLOB_STATE_DIRTY; 5127 blob->active.num_pages = 0; 5128 _spdk_blob_resize(blob, 0); 5129 5130 _spdk_blob_persist(seq, blob, _spdk_bs_delete_persist_cpl, blob); 5131 } 5132 5133 void 5134 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 5135 spdk_blob_op_complete cb_fn, void *cb_arg) 5136 { 5137 struct spdk_bs_cpl cpl; 5138 spdk_bs_sequence_t *seq; 5139 5140 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid); 5141 5142 assert(spdk_get_thread() == bs->md_thread); 5143 5144 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 5145 cpl.u.blob_basic.cb_fn = cb_fn; 5146 cpl.u.blob_basic.cb_arg = cb_arg; 5147 5148 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 5149 if (!seq) { 5150 cb_fn(cb_arg, -ENOMEM); 5151 return; 5152 } 5153 5154 spdk_bs_open_blob(bs, blobid, _spdk_bs_delete_open_cpl, seq); 5155 } 5156 5157 /* END spdk_bs_delete_blob */ 5158 5159 /* START spdk_bs_open_blob */ 5160 5161 static void 5162 _spdk_bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5163 { 5164 struct spdk_blob *blob = cb_arg; 5165 5166 /* If the blob have crc error, we just return NULL. */ 5167 if (blob == NULL) { 5168 seq->cpl.u.blob_handle.blob = NULL; 5169 spdk_bs_sequence_finish(seq, bserrno); 5170 return; 5171 } 5172 5173 blob->open_ref++; 5174 5175 TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link); 5176 5177 spdk_bs_sequence_finish(seq, bserrno); 5178 } 5179 5180 static void _spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 5181 struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5182 { 5183 struct spdk_blob *blob; 5184 struct spdk_bs_cpl cpl; 5185 struct spdk_blob_open_opts opts_default; 5186 spdk_bs_sequence_t *seq; 5187 uint32_t page_num; 5188 5189 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid); 5190 assert(spdk_get_thread() == bs->md_thread); 5191 5192 page_num = _spdk_bs_blobid_to_page(blobid); 5193 if (spdk_bit_array_get(bs->used_blobids, page_num) == false) { 5194 /* Invalid blobid */ 5195 cb_fn(cb_arg, NULL, -ENOENT); 5196 return; 5197 } 5198 5199 blob = _spdk_blob_lookup(bs, blobid); 5200 if (blob) { 5201 blob->open_ref++; 5202 cb_fn(cb_arg, blob, 0); 5203 return; 5204 } 5205 5206 blob = _spdk_blob_alloc(bs, blobid); 5207 if (!blob) { 5208 cb_fn(cb_arg, NULL, -ENOMEM); 5209 return; 5210 } 5211 5212 if (!opts) { 5213 spdk_blob_open_opts_init(&opts_default); 5214 opts = &opts_default; 5215 } 5216 5217 blob->clear_method = opts->clear_method; 5218 5219 cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE; 5220 cpl.u.blob_handle.cb_fn = cb_fn; 5221 cpl.u.blob_handle.cb_arg = cb_arg; 5222 cpl.u.blob_handle.blob = blob; 5223 5224 seq = spdk_bs_sequence_start(bs->md_channel, &cpl); 5225 if (!seq) { 5226 _spdk_blob_free(blob); 5227 cb_fn(cb_arg, NULL, -ENOMEM); 5228 return; 5229 } 5230 5231 _spdk_blob_load(seq, blob, _spdk_bs_open_blob_cpl, blob); 5232 } 5233 5234 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 5235 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5236 { 5237 _spdk_bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg); 5238 } 5239 5240 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid, 5241 struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5242 { 5243 _spdk_bs_open_blob(bs, blobid, opts, cb_fn, cb_arg); 5244 } 5245 5246 /* END spdk_bs_open_blob */ 5247 5248 /* START spdk_blob_set_read_only */ 5249 int spdk_blob_set_read_only(struct spdk_blob *blob) 5250 { 5251 _spdk_blob_verify_md_op(blob); 5252 5253 blob->data_ro_flags |= SPDK_BLOB_READ_ONLY; 5254 5255 blob->state = SPDK_BLOB_STATE_DIRTY; 5256 return 0; 5257 } 5258 /* END spdk_blob_set_read_only */ 5259 5260 /* START spdk_blob_sync_md */ 5261 5262 static void 5263 _spdk_blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5264 { 5265 struct spdk_blob *blob = cb_arg; 5266 5267 if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 5268 blob->data_ro = true; 5269 blob->md_ro = true; 5270 } 5271 5272 spdk_bs_sequence_finish(seq, bserrno); 5273 } 5274 5275 static void 5276 _spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 5277 { 5278 struct spdk_bs_cpl cpl; 5279 spdk_bs_sequence_t *seq; 5280 5281 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 5282 cpl.u.blob_basic.cb_fn = cb_fn; 5283 cpl.u.blob_basic.cb_arg = cb_arg; 5284 5285 seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl); 5286 if (!seq) { 5287 cb_fn(cb_arg, -ENOMEM); 5288 return; 5289 } 5290 5291 _spdk_blob_persist(seq, blob, _spdk_blob_sync_md_cpl, blob); 5292 } 5293 5294 void 5295 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 5296 { 5297 _spdk_blob_verify_md_op(blob); 5298 5299 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id); 5300 5301 if (blob->md_ro) { 5302 assert(blob->state == SPDK_BLOB_STATE_CLEAN); 5303 cb_fn(cb_arg, 0); 5304 return; 5305 } 5306 5307 _spdk_blob_sync_md(blob, cb_fn, cb_arg); 5308 } 5309 5310 /* END spdk_blob_sync_md */ 5311 5312 struct spdk_blob_insert_cluster_ctx { 5313 struct spdk_thread *thread; 5314 struct spdk_blob *blob; 5315 uint32_t cluster_num; /* cluster index in blob */ 5316 uint32_t cluster; /* cluster on disk */ 5317 int rc; 5318 spdk_blob_op_complete cb_fn; 5319 void *cb_arg; 5320 }; 5321 5322 static void 5323 _spdk_blob_insert_cluster_msg_cpl(void *arg) 5324 { 5325 struct spdk_blob_insert_cluster_ctx *ctx = arg; 5326 5327 ctx->cb_fn(ctx->cb_arg, ctx->rc); 5328 free(ctx); 5329 } 5330 5331 static void 5332 _spdk_blob_insert_cluster_msg_cb(void *arg, int bserrno) 5333 { 5334 struct spdk_blob_insert_cluster_ctx *ctx = arg; 5335 5336 ctx->rc = bserrno; 5337 spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx); 5338 } 5339 5340 static void 5341 _spdk_blob_insert_cluster_msg(void *arg) 5342 { 5343 struct spdk_blob_insert_cluster_ctx *ctx = arg; 5344 5345 ctx->rc = _spdk_blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster); 5346 if (ctx->rc != 0) { 5347 spdk_thread_send_msg(ctx->thread, _spdk_blob_insert_cluster_msg_cpl, ctx); 5348 return; 5349 } 5350 5351 ctx->blob->state = SPDK_BLOB_STATE_DIRTY; 5352 _spdk_blob_sync_md(ctx->blob, _spdk_blob_insert_cluster_msg_cb, ctx); 5353 } 5354 5355 static void 5356 _spdk_blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 5357 uint64_t cluster, spdk_blob_op_complete cb_fn, void *cb_arg) 5358 { 5359 struct spdk_blob_insert_cluster_ctx *ctx; 5360 5361 ctx = calloc(1, sizeof(*ctx)); 5362 if (ctx == NULL) { 5363 cb_fn(cb_arg, -ENOMEM); 5364 return; 5365 } 5366 5367 ctx->thread = spdk_get_thread(); 5368 ctx->blob = blob; 5369 ctx->cluster_num = cluster_num; 5370 ctx->cluster = cluster; 5371 ctx->cb_fn = cb_fn; 5372 ctx->cb_arg = cb_arg; 5373 5374 spdk_thread_send_msg(blob->bs->md_thread, _spdk_blob_insert_cluster_msg, ctx); 5375 } 5376 5377 /* START spdk_blob_close */ 5378 5379 static void 5380 _spdk_blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5381 { 5382 struct spdk_blob *blob = cb_arg; 5383 5384 if (bserrno == 0) { 5385 blob->open_ref--; 5386 if (blob->open_ref == 0) { 5387 /* 5388 * Blobs with active.num_pages == 0 are deleted blobs. 5389 * these blobs are removed from the blob_store list 5390 * when the deletion process starts - so don't try to 5391 * remove them again. 5392 */ 5393 if (blob->active.num_pages > 0) { 5394 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 5395 } 5396 _spdk_blob_free(blob); 5397 } 5398 } 5399 5400 spdk_bs_sequence_finish(seq, bserrno); 5401 } 5402 5403 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 5404 { 5405 struct spdk_bs_cpl cpl; 5406 spdk_bs_sequence_t *seq; 5407 5408 _spdk_blob_verify_md_op(blob); 5409 5410 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id); 5411 5412 if (blob->open_ref == 0) { 5413 cb_fn(cb_arg, -EBADF); 5414 return; 5415 } 5416 5417 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 5418 cpl.u.blob_basic.cb_fn = cb_fn; 5419 cpl.u.blob_basic.cb_arg = cb_arg; 5420 5421 seq = spdk_bs_sequence_start(blob->bs->md_channel, &cpl); 5422 if (!seq) { 5423 cb_fn(cb_arg, -ENOMEM); 5424 return; 5425 } 5426 5427 /* Sync metadata */ 5428 _spdk_blob_persist(seq, blob, _spdk_blob_close_cpl, blob); 5429 } 5430 5431 /* END spdk_blob_close */ 5432 5433 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs) 5434 { 5435 return spdk_get_io_channel(bs); 5436 } 5437 5438 void spdk_bs_free_io_channel(struct spdk_io_channel *channel) 5439 { 5440 spdk_put_io_channel(channel); 5441 } 5442 5443 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel, 5444 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 5445 { 5446 _spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 5447 SPDK_BLOB_UNMAP); 5448 } 5449 5450 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel, 5451 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 5452 { 5453 _spdk_blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 5454 SPDK_BLOB_WRITE_ZEROES); 5455 } 5456 5457 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel, 5458 void *payload, uint64_t offset, uint64_t length, 5459 spdk_blob_op_complete cb_fn, void *cb_arg) 5460 { 5461 _spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 5462 SPDK_BLOB_WRITE); 5463 } 5464 5465 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel, 5466 void *payload, uint64_t offset, uint64_t length, 5467 spdk_blob_op_complete cb_fn, void *cb_arg) 5468 { 5469 _spdk_blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 5470 SPDK_BLOB_READ); 5471 } 5472 5473 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel, 5474 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 5475 spdk_blob_op_complete cb_fn, void *cb_arg) 5476 { 5477 _spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false); 5478 } 5479 5480 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel, 5481 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 5482 spdk_blob_op_complete cb_fn, void *cb_arg) 5483 { 5484 _spdk_blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true); 5485 } 5486 5487 struct spdk_bs_iter_ctx { 5488 int64_t page_num; 5489 struct spdk_blob_store *bs; 5490 5491 spdk_blob_op_with_handle_complete cb_fn; 5492 void *cb_arg; 5493 }; 5494 5495 static void 5496 _spdk_bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5497 { 5498 struct spdk_bs_iter_ctx *ctx = cb_arg; 5499 struct spdk_blob_store *bs = ctx->bs; 5500 spdk_blob_id id; 5501 5502 if (bserrno == 0) { 5503 ctx->cb_fn(ctx->cb_arg, _blob, bserrno); 5504 free(ctx); 5505 return; 5506 } 5507 5508 ctx->page_num++; 5509 ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num); 5510 if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) { 5511 ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT); 5512 free(ctx); 5513 return; 5514 } 5515 5516 id = _spdk_bs_page_to_blobid(ctx->page_num); 5517 5518 spdk_bs_open_blob(bs, id, _spdk_bs_iter_cpl, ctx); 5519 } 5520 5521 void 5522 spdk_bs_iter_first(struct spdk_blob_store *bs, 5523 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5524 { 5525 struct spdk_bs_iter_ctx *ctx; 5526 5527 ctx = calloc(1, sizeof(*ctx)); 5528 if (!ctx) { 5529 cb_fn(cb_arg, NULL, -ENOMEM); 5530 return; 5531 } 5532 5533 ctx->page_num = -1; 5534 ctx->bs = bs; 5535 ctx->cb_fn = cb_fn; 5536 ctx->cb_arg = cb_arg; 5537 5538 _spdk_bs_iter_cpl(ctx, NULL, -1); 5539 } 5540 5541 static void 5542 _spdk_bs_iter_close_cpl(void *cb_arg, int bserrno) 5543 { 5544 struct spdk_bs_iter_ctx *ctx = cb_arg; 5545 5546 _spdk_bs_iter_cpl(ctx, NULL, -1); 5547 } 5548 5549 void 5550 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob, 5551 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 5552 { 5553 struct spdk_bs_iter_ctx *ctx; 5554 5555 assert(blob != NULL); 5556 5557 ctx = calloc(1, sizeof(*ctx)); 5558 if (!ctx) { 5559 cb_fn(cb_arg, NULL, -ENOMEM); 5560 return; 5561 } 5562 5563 ctx->page_num = _spdk_bs_blobid_to_page(blob->id); 5564 ctx->bs = bs; 5565 ctx->cb_fn = cb_fn; 5566 ctx->cb_arg = cb_arg; 5567 5568 /* Close the existing blob */ 5569 spdk_blob_close(blob, _spdk_bs_iter_close_cpl, ctx); 5570 } 5571 5572 static int 5573 _spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 5574 uint16_t value_len, bool internal) 5575 { 5576 struct spdk_xattr_tailq *xattrs; 5577 struct spdk_xattr *xattr; 5578 5579 _spdk_blob_verify_md_op(blob); 5580 5581 if (blob->md_ro) { 5582 return -EPERM; 5583 } 5584 5585 if (internal) { 5586 xattrs = &blob->xattrs_internal; 5587 blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR; 5588 } else { 5589 xattrs = &blob->xattrs; 5590 } 5591 5592 TAILQ_FOREACH(xattr, xattrs, link) { 5593 if (!strcmp(name, xattr->name)) { 5594 free(xattr->value); 5595 xattr->value_len = value_len; 5596 xattr->value = malloc(value_len); 5597 memcpy(xattr->value, value, value_len); 5598 5599 blob->state = SPDK_BLOB_STATE_DIRTY; 5600 5601 return 0; 5602 } 5603 } 5604 5605 xattr = calloc(1, sizeof(*xattr)); 5606 if (!xattr) { 5607 return -ENOMEM; 5608 } 5609 xattr->name = strdup(name); 5610 xattr->value_len = value_len; 5611 xattr->value = malloc(value_len); 5612 memcpy(xattr->value, value, value_len); 5613 TAILQ_INSERT_TAIL(xattrs, xattr, link); 5614 5615 blob->state = SPDK_BLOB_STATE_DIRTY; 5616 5617 return 0; 5618 } 5619 5620 int 5621 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 5622 uint16_t value_len) 5623 { 5624 return _spdk_blob_set_xattr(blob, name, value, value_len, false); 5625 } 5626 5627 static int 5628 _spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal) 5629 { 5630 struct spdk_xattr_tailq *xattrs; 5631 struct spdk_xattr *xattr; 5632 5633 _spdk_blob_verify_md_op(blob); 5634 5635 if (blob->md_ro) { 5636 return -EPERM; 5637 } 5638 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 5639 5640 TAILQ_FOREACH(xattr, xattrs, link) { 5641 if (!strcmp(name, xattr->name)) { 5642 TAILQ_REMOVE(xattrs, xattr, link); 5643 free(xattr->value); 5644 free(xattr->name); 5645 free(xattr); 5646 5647 if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) { 5648 blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR; 5649 } 5650 blob->state = SPDK_BLOB_STATE_DIRTY; 5651 5652 return 0; 5653 } 5654 } 5655 5656 return -ENOENT; 5657 } 5658 5659 int 5660 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name) 5661 { 5662 return _spdk_blob_remove_xattr(blob, name, false); 5663 } 5664 5665 static int 5666 _spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 5667 const void **value, size_t *value_len, bool internal) 5668 { 5669 struct spdk_xattr *xattr; 5670 struct spdk_xattr_tailq *xattrs; 5671 5672 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 5673 5674 TAILQ_FOREACH(xattr, xattrs, link) { 5675 if (!strcmp(name, xattr->name)) { 5676 *value = xattr->value; 5677 *value_len = xattr->value_len; 5678 return 0; 5679 } 5680 } 5681 return -ENOENT; 5682 } 5683 5684 int 5685 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 5686 const void **value, size_t *value_len) 5687 { 5688 _spdk_blob_verify_md_op(blob); 5689 5690 return _spdk_blob_get_xattr_value(blob, name, value, value_len, false); 5691 } 5692 5693 struct spdk_xattr_names { 5694 uint32_t count; 5695 const char *names[0]; 5696 }; 5697 5698 static int 5699 _spdk_blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names) 5700 { 5701 struct spdk_xattr *xattr; 5702 int count = 0; 5703 5704 TAILQ_FOREACH(xattr, xattrs, link) { 5705 count++; 5706 } 5707 5708 *names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *)); 5709 if (*names == NULL) { 5710 return -ENOMEM; 5711 } 5712 5713 TAILQ_FOREACH(xattr, xattrs, link) { 5714 (*names)->names[(*names)->count++] = xattr->name; 5715 } 5716 5717 return 0; 5718 } 5719 5720 int 5721 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names) 5722 { 5723 _spdk_blob_verify_md_op(blob); 5724 5725 return _spdk_blob_get_xattr_names(&blob->xattrs, names); 5726 } 5727 5728 uint32_t 5729 spdk_xattr_names_get_count(struct spdk_xattr_names *names) 5730 { 5731 assert(names != NULL); 5732 5733 return names->count; 5734 } 5735 5736 const char * 5737 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index) 5738 { 5739 if (index >= names->count) { 5740 return NULL; 5741 } 5742 5743 return names->names[index]; 5744 } 5745 5746 void 5747 spdk_xattr_names_free(struct spdk_xattr_names *names) 5748 { 5749 free(names); 5750 } 5751 5752 struct spdk_bs_type 5753 spdk_bs_get_bstype(struct spdk_blob_store *bs) 5754 { 5755 return bs->bstype; 5756 } 5757 5758 void 5759 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype) 5760 { 5761 memcpy(&bs->bstype, &bstype, sizeof(bstype)); 5762 } 5763 5764 bool 5765 spdk_blob_is_read_only(struct spdk_blob *blob) 5766 { 5767 assert(blob != NULL); 5768 return (blob->data_ro || blob->md_ro); 5769 } 5770 5771 bool 5772 spdk_blob_is_snapshot(struct spdk_blob *blob) 5773 { 5774 struct spdk_blob_list *snapshot_entry; 5775 5776 assert(blob != NULL); 5777 5778 snapshot_entry = _spdk_bs_get_snapshot_entry(blob->bs, blob->id); 5779 if (snapshot_entry == NULL) { 5780 return false; 5781 } 5782 5783 return true; 5784 } 5785 5786 bool 5787 spdk_blob_is_clone(struct spdk_blob *blob) 5788 { 5789 assert(blob != NULL); 5790 5791 if (blob->parent_id != SPDK_BLOBID_INVALID) { 5792 assert(spdk_blob_is_thin_provisioned(blob)); 5793 return true; 5794 } 5795 5796 return false; 5797 } 5798 5799 bool 5800 spdk_blob_is_thin_provisioned(struct spdk_blob *blob) 5801 { 5802 assert(blob != NULL); 5803 return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV); 5804 } 5805 5806 spdk_blob_id 5807 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id) 5808 { 5809 struct spdk_blob_list *snapshot_entry = NULL; 5810 struct spdk_blob_list *clone_entry = NULL; 5811 5812 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 5813 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 5814 if (clone_entry->id == blob_id) { 5815 return snapshot_entry->id; 5816 } 5817 } 5818 } 5819 5820 return SPDK_BLOBID_INVALID; 5821 } 5822 5823 int 5824 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids, 5825 size_t *count) 5826 { 5827 struct spdk_blob_list *snapshot_entry, *clone_entry; 5828 size_t n; 5829 5830 snapshot_entry = _spdk_bs_get_snapshot_entry(bs, blobid); 5831 if (snapshot_entry == NULL) { 5832 *count = 0; 5833 return 0; 5834 } 5835 5836 if (ids == NULL || *count < snapshot_entry->clone_count) { 5837 *count = snapshot_entry->clone_count; 5838 return -ENOMEM; 5839 } 5840 *count = snapshot_entry->clone_count; 5841 5842 n = 0; 5843 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 5844 ids[n++] = clone_entry->id; 5845 } 5846 5847 return 0; 5848 } 5849 5850 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB) 5851