1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/stdinc.h" 35 36 #include "spdk/blob.h" 37 #include "spdk/crc32.h" 38 #include "spdk/env.h" 39 #include "spdk/queue.h" 40 #include "spdk/thread.h" 41 #include "spdk/bit_array.h" 42 #include "spdk/bit_pool.h" 43 #include "spdk/likely.h" 44 #include "spdk/util.h" 45 #include "spdk/string.h" 46 47 #include "spdk_internal/assert.h" 48 #include "spdk_internal/log.h" 49 50 #include "blobstore.h" 51 52 #define BLOB_CRC32C_INITIAL 0xffffffffUL 53 54 static int bs_register_md_thread(struct spdk_blob_store *bs); 55 static int bs_unregister_md_thread(struct spdk_blob_store *bs); 56 static void blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno); 57 static void blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 58 uint64_t cluster, uint32_t extent, spdk_blob_op_complete cb_fn, void *cb_arg); 59 60 static int blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 61 uint16_t value_len, bool internal); 62 static int blob_get_xattr_value(struct spdk_blob *blob, const char *name, 63 const void **value, size_t *value_len, bool internal); 64 static int blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal); 65 66 static void blob_insert_extent(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num, 67 spdk_blob_op_complete cb_fn, void *cb_arg); 68 69 static void 70 blob_verify_md_op(struct spdk_blob *blob) 71 { 72 assert(blob != NULL); 73 assert(spdk_get_thread() == blob->bs->md_thread); 74 assert(blob->state != SPDK_BLOB_STATE_LOADING); 75 } 76 77 static struct spdk_blob_list * 78 bs_get_snapshot_entry(struct spdk_blob_store *bs, spdk_blob_id blobid) 79 { 80 struct spdk_blob_list *snapshot_entry = NULL; 81 82 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 83 if (snapshot_entry->id == blobid) { 84 break; 85 } 86 } 87 88 return snapshot_entry; 89 } 90 91 static void 92 bs_claim_md_page(struct spdk_blob_store *bs, uint32_t page) 93 { 94 assert(page < spdk_bit_array_capacity(bs->used_md_pages)); 95 assert(spdk_bit_array_get(bs->used_md_pages, page) == false); 96 97 spdk_bit_array_set(bs->used_md_pages, page); 98 } 99 100 static void 101 bs_release_md_page(struct spdk_blob_store *bs, uint32_t page) 102 { 103 assert(page < spdk_bit_array_capacity(bs->used_md_pages)); 104 assert(spdk_bit_array_get(bs->used_md_pages, page) == true); 105 106 spdk_bit_array_clear(bs->used_md_pages, page); 107 } 108 109 static uint32_t 110 bs_claim_cluster(struct spdk_blob_store *bs) 111 { 112 uint32_t cluster_num; 113 114 cluster_num = spdk_bit_pool_allocate_bit(bs->used_clusters); 115 if (cluster_num == UINT32_MAX) { 116 return UINT32_MAX; 117 } 118 119 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %u\n", cluster_num); 120 bs->num_free_clusters--; 121 122 return cluster_num; 123 } 124 125 static void 126 bs_release_cluster(struct spdk_blob_store *bs, uint32_t cluster_num) 127 { 128 assert(cluster_num < spdk_bit_pool_capacity(bs->used_clusters)); 129 assert(spdk_bit_pool_is_allocated(bs->used_clusters, cluster_num) == true); 130 assert(bs->num_free_clusters < bs->total_clusters); 131 132 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Releasing cluster %u\n", cluster_num); 133 134 spdk_bit_pool_free_bit(bs->used_clusters, cluster_num); 135 bs->num_free_clusters++; 136 } 137 138 static int 139 blob_insert_cluster(struct spdk_blob *blob, uint32_t cluster_num, uint64_t cluster) 140 { 141 uint64_t *cluster_lba = &blob->active.clusters[cluster_num]; 142 143 blob_verify_md_op(blob); 144 145 if (*cluster_lba != 0) { 146 return -EEXIST; 147 } 148 149 *cluster_lba = bs_cluster_to_lba(blob->bs, cluster); 150 return 0; 151 } 152 153 static int 154 bs_allocate_cluster(struct spdk_blob *blob, uint32_t cluster_num, 155 uint64_t *cluster, uint32_t *lowest_free_md_page, bool update_map) 156 { 157 uint32_t *extent_page = 0; 158 159 *cluster = bs_claim_cluster(blob->bs); 160 if (*cluster == UINT32_MAX) { 161 /* No more free clusters. Cannot satisfy the request */ 162 return -ENOSPC; 163 } 164 165 if (blob->use_extent_table) { 166 extent_page = bs_cluster_to_extent_page(blob, cluster_num); 167 if (*extent_page == 0) { 168 /* No extent_page is allocated for the cluster */ 169 *lowest_free_md_page = spdk_bit_array_find_first_clear(blob->bs->used_md_pages, 170 *lowest_free_md_page); 171 if (*lowest_free_md_page == UINT32_MAX) { 172 /* No more free md pages. Cannot satisfy the request */ 173 bs_release_cluster(blob->bs, *cluster); 174 return -ENOSPC; 175 } 176 bs_claim_md_page(blob->bs, *lowest_free_md_page); 177 } 178 } 179 180 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming cluster %lu for blob %lu\n", *cluster, blob->id); 181 182 if (update_map) { 183 blob_insert_cluster(blob, cluster_num, *cluster); 184 if (blob->use_extent_table && *extent_page == 0) { 185 *extent_page = *lowest_free_md_page; 186 } 187 } 188 189 return 0; 190 } 191 192 static void 193 blob_xattrs_init(struct spdk_blob_xattr_opts *xattrs) 194 { 195 xattrs->count = 0; 196 xattrs->names = NULL; 197 xattrs->ctx = NULL; 198 xattrs->get_value = NULL; 199 } 200 201 void 202 spdk_blob_opts_init(struct spdk_blob_opts *opts) 203 { 204 opts->num_clusters = 0; 205 opts->thin_provision = false; 206 opts->clear_method = BLOB_CLEAR_WITH_DEFAULT; 207 blob_xattrs_init(&opts->xattrs); 208 opts->use_extent_table = true; 209 } 210 211 void 212 spdk_blob_open_opts_init(struct spdk_blob_open_opts *opts) 213 { 214 opts->clear_method = BLOB_CLEAR_WITH_DEFAULT; 215 } 216 217 static struct spdk_blob * 218 blob_alloc(struct spdk_blob_store *bs, spdk_blob_id id) 219 { 220 struct spdk_blob *blob; 221 222 blob = calloc(1, sizeof(*blob)); 223 if (!blob) { 224 return NULL; 225 } 226 227 blob->id = id; 228 blob->bs = bs; 229 230 blob->parent_id = SPDK_BLOBID_INVALID; 231 232 blob->state = SPDK_BLOB_STATE_DIRTY; 233 blob->extent_rle_found = false; 234 blob->extent_table_found = false; 235 blob->active.num_pages = 1; 236 blob->active.pages = calloc(1, sizeof(*blob->active.pages)); 237 if (!blob->active.pages) { 238 free(blob); 239 return NULL; 240 } 241 242 blob->active.pages[0] = bs_blobid_to_page(id); 243 244 TAILQ_INIT(&blob->xattrs); 245 TAILQ_INIT(&blob->xattrs_internal); 246 TAILQ_INIT(&blob->pending_persists); 247 248 return blob; 249 } 250 251 static void 252 xattrs_free(struct spdk_xattr_tailq *xattrs) 253 { 254 struct spdk_xattr *xattr, *xattr_tmp; 255 256 TAILQ_FOREACH_SAFE(xattr, xattrs, link, xattr_tmp) { 257 TAILQ_REMOVE(xattrs, xattr, link); 258 free(xattr->name); 259 free(xattr->value); 260 free(xattr); 261 } 262 } 263 264 static void 265 blob_free(struct spdk_blob *blob) 266 { 267 assert(blob != NULL); 268 assert(TAILQ_EMPTY(&blob->pending_persists)); 269 270 free(blob->active.extent_pages); 271 free(blob->clean.extent_pages); 272 free(blob->active.clusters); 273 free(blob->clean.clusters); 274 free(blob->active.pages); 275 free(blob->clean.pages); 276 277 xattrs_free(&blob->xattrs); 278 xattrs_free(&blob->xattrs_internal); 279 280 if (blob->back_bs_dev) { 281 blob->back_bs_dev->destroy(blob->back_bs_dev); 282 } 283 284 free(blob); 285 } 286 287 struct freeze_io_ctx { 288 struct spdk_bs_cpl cpl; 289 struct spdk_blob *blob; 290 }; 291 292 static void 293 blob_io_sync(struct spdk_io_channel_iter *i) 294 { 295 spdk_for_each_channel_continue(i, 0); 296 } 297 298 static void 299 blob_execute_queued_io(struct spdk_io_channel_iter *i) 300 { 301 struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i); 302 struct spdk_bs_channel *ch = spdk_io_channel_get_ctx(_ch); 303 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 304 struct spdk_bs_request_set *set; 305 struct spdk_bs_user_op_args *args; 306 spdk_bs_user_op_t *op, *tmp; 307 308 TAILQ_FOREACH_SAFE(op, &ch->queued_io, link, tmp) { 309 set = (struct spdk_bs_request_set *)op; 310 args = &set->u.user_op; 311 312 if (args->blob == ctx->blob) { 313 TAILQ_REMOVE(&ch->queued_io, op, link); 314 bs_user_op_execute(op); 315 } 316 } 317 318 spdk_for_each_channel_continue(i, 0); 319 } 320 321 static void 322 blob_io_cpl(struct spdk_io_channel_iter *i, int status) 323 { 324 struct freeze_io_ctx *ctx = spdk_io_channel_iter_get_ctx(i); 325 326 ctx->cpl.u.blob_basic.cb_fn(ctx->cpl.u.blob_basic.cb_arg, 0); 327 328 free(ctx); 329 } 330 331 static void 332 blob_freeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 333 { 334 struct freeze_io_ctx *ctx; 335 336 ctx = calloc(1, sizeof(*ctx)); 337 if (!ctx) { 338 cb_fn(cb_arg, -ENOMEM); 339 return; 340 } 341 342 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 343 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 344 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 345 ctx->blob = blob; 346 347 /* Freeze I/O on blob */ 348 blob->frozen_refcnt++; 349 350 if (blob->frozen_refcnt == 1) { 351 spdk_for_each_channel(blob->bs, blob_io_sync, ctx, blob_io_cpl); 352 } else { 353 cb_fn(cb_arg, 0); 354 free(ctx); 355 } 356 } 357 358 static void 359 blob_unfreeze_io(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 360 { 361 struct freeze_io_ctx *ctx; 362 363 ctx = calloc(1, sizeof(*ctx)); 364 if (!ctx) { 365 cb_fn(cb_arg, -ENOMEM); 366 return; 367 } 368 369 ctx->cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 370 ctx->cpl.u.blob_basic.cb_fn = cb_fn; 371 ctx->cpl.u.blob_basic.cb_arg = cb_arg; 372 ctx->blob = blob; 373 374 assert(blob->frozen_refcnt > 0); 375 376 blob->frozen_refcnt--; 377 378 if (blob->frozen_refcnt == 0) { 379 spdk_for_each_channel(blob->bs, blob_execute_queued_io, ctx, blob_io_cpl); 380 } else { 381 cb_fn(cb_arg, 0); 382 free(ctx); 383 } 384 } 385 386 static int 387 blob_mark_clean(struct spdk_blob *blob) 388 { 389 uint32_t *extent_pages = NULL; 390 uint64_t *clusters = NULL; 391 uint32_t *pages = NULL; 392 393 assert(blob != NULL); 394 395 if (blob->active.num_extent_pages) { 396 assert(blob->active.extent_pages); 397 extent_pages = calloc(blob->active.num_extent_pages, sizeof(*blob->active.extent_pages)); 398 if (!extent_pages) { 399 return -ENOMEM; 400 } 401 memcpy(extent_pages, blob->active.extent_pages, 402 blob->active.num_extent_pages * sizeof(*extent_pages)); 403 } 404 405 if (blob->active.num_clusters) { 406 assert(blob->active.clusters); 407 clusters = calloc(blob->active.num_clusters, sizeof(*blob->active.clusters)); 408 if (!clusters) { 409 free(extent_pages); 410 return -ENOMEM; 411 } 412 memcpy(clusters, blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters)); 413 } 414 415 if (blob->active.num_pages) { 416 assert(blob->active.pages); 417 pages = calloc(blob->active.num_pages, sizeof(*blob->active.pages)); 418 if (!pages) { 419 free(extent_pages); 420 free(clusters); 421 return -ENOMEM; 422 } 423 memcpy(pages, blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages)); 424 } 425 426 free(blob->clean.extent_pages); 427 free(blob->clean.clusters); 428 free(blob->clean.pages); 429 430 blob->clean.num_extent_pages = blob->active.num_extent_pages; 431 blob->clean.extent_pages = blob->active.extent_pages; 432 blob->clean.num_clusters = blob->active.num_clusters; 433 blob->clean.clusters = blob->active.clusters; 434 blob->clean.num_pages = blob->active.num_pages; 435 blob->clean.pages = blob->active.pages; 436 437 blob->active.extent_pages = extent_pages; 438 blob->active.clusters = clusters; 439 blob->active.pages = pages; 440 441 /* If the metadata was dirtied again while the metadata was being written to disk, 442 * we do not want to revert the DIRTY state back to CLEAN here. 443 */ 444 if (blob->state == SPDK_BLOB_STATE_LOADING) { 445 blob->state = SPDK_BLOB_STATE_CLEAN; 446 } 447 448 return 0; 449 } 450 451 static int 452 blob_deserialize_xattr(struct spdk_blob *blob, 453 struct spdk_blob_md_descriptor_xattr *desc_xattr, bool internal) 454 { 455 struct spdk_xattr *xattr; 456 457 if (desc_xattr->length != sizeof(desc_xattr->name_length) + 458 sizeof(desc_xattr->value_length) + 459 desc_xattr->name_length + desc_xattr->value_length) { 460 return -EINVAL; 461 } 462 463 xattr = calloc(1, sizeof(*xattr)); 464 if (xattr == NULL) { 465 return -ENOMEM; 466 } 467 468 xattr->name = malloc(desc_xattr->name_length + 1); 469 if (xattr->name == NULL) { 470 free(xattr); 471 return -ENOMEM; 472 } 473 memcpy(xattr->name, desc_xattr->name, desc_xattr->name_length); 474 xattr->name[desc_xattr->name_length] = '\0'; 475 476 xattr->value = malloc(desc_xattr->value_length); 477 if (xattr->value == NULL) { 478 free(xattr->name); 479 free(xattr); 480 return -ENOMEM; 481 } 482 xattr->value_len = desc_xattr->value_length; 483 memcpy(xattr->value, 484 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 485 desc_xattr->value_length); 486 487 TAILQ_INSERT_TAIL(internal ? &blob->xattrs_internal : &blob->xattrs, xattr, link); 488 489 return 0; 490 } 491 492 493 static int 494 blob_parse_page(const struct spdk_blob_md_page *page, struct spdk_blob *blob) 495 { 496 struct spdk_blob_md_descriptor *desc; 497 size_t cur_desc = 0; 498 void *tmp; 499 500 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 501 while (cur_desc < sizeof(page->descriptors)) { 502 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 503 if (desc->length == 0) { 504 /* If padding and length are 0, this terminates the page */ 505 break; 506 } 507 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 508 struct spdk_blob_md_descriptor_flags *desc_flags; 509 510 desc_flags = (struct spdk_blob_md_descriptor_flags *)desc; 511 512 if (desc_flags->length != sizeof(*desc_flags) - sizeof(*desc)) { 513 return -EINVAL; 514 } 515 516 if ((desc_flags->invalid_flags | SPDK_BLOB_INVALID_FLAGS_MASK) != 517 SPDK_BLOB_INVALID_FLAGS_MASK) { 518 return -EINVAL; 519 } 520 521 if ((desc_flags->data_ro_flags | SPDK_BLOB_DATA_RO_FLAGS_MASK) != 522 SPDK_BLOB_DATA_RO_FLAGS_MASK) { 523 blob->data_ro = true; 524 blob->md_ro = true; 525 } 526 527 if ((desc_flags->md_ro_flags | SPDK_BLOB_MD_RO_FLAGS_MASK) != 528 SPDK_BLOB_MD_RO_FLAGS_MASK) { 529 blob->md_ro = true; 530 } 531 532 if ((desc_flags->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 533 blob->data_ro = true; 534 blob->md_ro = true; 535 } 536 537 blob->invalid_flags = desc_flags->invalid_flags; 538 blob->data_ro_flags = desc_flags->data_ro_flags; 539 blob->md_ro_flags = desc_flags->md_ro_flags; 540 541 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) { 542 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 543 unsigned int i, j; 544 unsigned int cluster_count = blob->active.num_clusters; 545 546 if (blob->extent_table_found) { 547 /* Extent Table already present in the md, 548 * both descriptors should never be at the same time. */ 549 return -EINVAL; 550 } 551 blob->extent_rle_found = true; 552 553 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc; 554 555 if (desc_extent_rle->length == 0 || 556 (desc_extent_rle->length % sizeof(desc_extent_rle->extents[0]) != 0)) { 557 return -EINVAL; 558 } 559 560 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 561 for (j = 0; j < desc_extent_rle->extents[i].length; j++) { 562 if (desc_extent_rle->extents[i].cluster_idx != 0) { 563 if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters, 564 desc_extent_rle->extents[i].cluster_idx + j)) { 565 return -EINVAL; 566 } 567 } 568 cluster_count++; 569 } 570 } 571 572 if (cluster_count == 0) { 573 return -EINVAL; 574 } 575 tmp = realloc(blob->active.clusters, cluster_count * sizeof(*blob->active.clusters)); 576 if (tmp == NULL) { 577 return -ENOMEM; 578 } 579 blob->active.clusters = tmp; 580 blob->active.cluster_array_size = cluster_count; 581 582 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 583 for (j = 0; j < desc_extent_rle->extents[i].length; j++) { 584 if (desc_extent_rle->extents[i].cluster_idx != 0) { 585 blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs, 586 desc_extent_rle->extents[i].cluster_idx + j); 587 } else if (spdk_blob_is_thin_provisioned(blob)) { 588 blob->active.clusters[blob->active.num_clusters++] = 0; 589 } else { 590 return -EINVAL; 591 } 592 } 593 } 594 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) { 595 struct spdk_blob_md_descriptor_extent_table *desc_extent_table; 596 uint32_t num_extent_pages = blob->active.num_extent_pages; 597 uint32_t i, j; 598 size_t extent_pages_length; 599 600 desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc; 601 extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters); 602 603 if (blob->extent_rle_found) { 604 /* This means that Extent RLE is present in MD, 605 * both should never be at the same time. */ 606 return -EINVAL; 607 } else if (blob->extent_table_found && 608 desc_extent_table->num_clusters != blob->remaining_clusters_in_et) { 609 /* Number of clusters in this ET does not match number 610 * from previously read EXTENT_TABLE. */ 611 return -EINVAL; 612 } 613 614 blob->extent_table_found = true; 615 616 if (desc_extent_table->length == 0 || 617 (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) { 618 return -EINVAL; 619 } 620 621 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 622 num_extent_pages += desc_extent_table->extent_page[i].num_pages; 623 } 624 625 tmp = realloc(blob->active.extent_pages, num_extent_pages * sizeof(uint32_t)); 626 if (tmp == NULL) { 627 return -ENOMEM; 628 } 629 blob->active.extent_pages = tmp; 630 blob->active.extent_pages_array_size = num_extent_pages; 631 632 blob->remaining_clusters_in_et = desc_extent_table->num_clusters; 633 634 /* Extent table entries contain md page numbers for extent pages. 635 * Zeroes represent unallocated extent pages, those are run-length-encoded. 636 */ 637 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 638 if (desc_extent_table->extent_page[i].page_idx != 0) { 639 assert(desc_extent_table->extent_page[i].num_pages == 1); 640 blob->active.extent_pages[blob->active.num_extent_pages++] = 641 desc_extent_table->extent_page[i].page_idx; 642 } else if (spdk_blob_is_thin_provisioned(blob)) { 643 for (j = 0; j < desc_extent_table->extent_page[i].num_pages; j++) { 644 blob->active.extent_pages[blob->active.num_extent_pages++] = 0; 645 } 646 } else { 647 return -EINVAL; 648 } 649 } 650 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 651 struct spdk_blob_md_descriptor_extent_page *desc_extent; 652 unsigned int i; 653 unsigned int cluster_count = 0; 654 size_t cluster_idx_length; 655 656 if (blob->extent_rle_found) { 657 /* This means that Extent RLE is present in MD, 658 * both should never be at the same time. */ 659 return -EINVAL; 660 } 661 662 desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc; 663 cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx); 664 665 if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) || 666 (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) { 667 return -EINVAL; 668 } 669 670 for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) { 671 if (desc_extent->cluster_idx[i] != 0) { 672 if (!spdk_bit_pool_is_allocated(blob->bs->used_clusters, desc_extent->cluster_idx[i])) { 673 return -EINVAL; 674 } 675 } 676 cluster_count++; 677 } 678 679 if (cluster_count == 0) { 680 return -EINVAL; 681 } 682 683 /* When reading extent pages sequentially starting cluster idx should match 684 * current size of a blob. 685 * If changed to batch reading, this check shall be removed. */ 686 if (desc_extent->start_cluster_idx != blob->active.num_clusters) { 687 return -EINVAL; 688 } 689 690 tmp = realloc(blob->active.clusters, 691 (cluster_count + blob->active.num_clusters) * sizeof(*blob->active.clusters)); 692 if (tmp == NULL) { 693 return -ENOMEM; 694 } 695 blob->active.clusters = tmp; 696 blob->active.cluster_array_size = (cluster_count + blob->active.num_clusters); 697 698 for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) { 699 if (desc_extent->cluster_idx[i] != 0) { 700 blob->active.clusters[blob->active.num_clusters++] = bs_cluster_to_lba(blob->bs, 701 desc_extent->cluster_idx[i]); 702 } else if (spdk_blob_is_thin_provisioned(blob)) { 703 blob->active.clusters[blob->active.num_clusters++] = 0; 704 } else { 705 return -EINVAL; 706 } 707 } 708 assert(desc_extent->start_cluster_idx + cluster_count == blob->active.num_clusters); 709 assert(blob->remaining_clusters_in_et >= cluster_count); 710 blob->remaining_clusters_in_et -= cluster_count; 711 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 712 int rc; 713 714 rc = blob_deserialize_xattr(blob, 715 (struct spdk_blob_md_descriptor_xattr *) desc, false); 716 if (rc != 0) { 717 return rc; 718 } 719 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 720 int rc; 721 722 rc = blob_deserialize_xattr(blob, 723 (struct spdk_blob_md_descriptor_xattr *) desc, true); 724 if (rc != 0) { 725 return rc; 726 } 727 } else { 728 /* Unrecognized descriptor type. Do not fail - just continue to the 729 * next descriptor. If this descriptor is associated with some feature 730 * defined in a newer version of blobstore, that version of blobstore 731 * should create and set an associated feature flag to specify if this 732 * blob can be loaded or not. 733 */ 734 } 735 736 /* Advance to the next descriptor */ 737 cur_desc += sizeof(*desc) + desc->length; 738 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 739 break; 740 } 741 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 742 } 743 744 return 0; 745 } 746 747 static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page); 748 749 static int 750 blob_parse_extent_page(struct spdk_blob_md_page *extent_page, struct spdk_blob *blob) 751 { 752 assert(blob != NULL); 753 assert(blob->state == SPDK_BLOB_STATE_LOADING); 754 755 if (bs_load_cur_extent_page_valid(extent_page) == false) { 756 return -ENOENT; 757 } 758 759 return blob_parse_page(extent_page, blob); 760 } 761 762 static int 763 blob_parse(const struct spdk_blob_md_page *pages, uint32_t page_count, 764 struct spdk_blob *blob) 765 { 766 const struct spdk_blob_md_page *page; 767 uint32_t i; 768 int rc; 769 770 assert(page_count > 0); 771 assert(pages[0].sequence_num == 0); 772 assert(blob != NULL); 773 assert(blob->state == SPDK_BLOB_STATE_LOADING); 774 assert(blob->active.clusters == NULL); 775 776 /* The blobid provided doesn't match what's in the MD, this can 777 * happen for example if a bogus blobid is passed in through open. 778 */ 779 if (blob->id != pages[0].id) { 780 SPDK_ERRLOG("Blobid (%lu) doesn't match what's in metadata (%lu)\n", 781 blob->id, pages[0].id); 782 return -ENOENT; 783 } 784 785 for (i = 0; i < page_count; i++) { 786 page = &pages[i]; 787 788 assert(page->id == blob->id); 789 assert(page->sequence_num == i); 790 791 rc = blob_parse_page(page, blob); 792 if (rc != 0) { 793 return rc; 794 } 795 } 796 797 return 0; 798 } 799 800 static int 801 blob_serialize_add_page(const struct spdk_blob *blob, 802 struct spdk_blob_md_page **pages, 803 uint32_t *page_count, 804 struct spdk_blob_md_page **last_page) 805 { 806 struct spdk_blob_md_page *page; 807 808 assert(pages != NULL); 809 assert(page_count != NULL); 810 811 if (*page_count == 0) { 812 assert(*pages == NULL); 813 *page_count = 1; 814 *pages = spdk_malloc(SPDK_BS_PAGE_SIZE, 0, 815 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 816 } else { 817 assert(*pages != NULL); 818 (*page_count)++; 819 *pages = spdk_realloc(*pages, SPDK_BS_PAGE_SIZE * (*page_count), 0); 820 } 821 822 if (*pages == NULL) { 823 *page_count = 0; 824 *last_page = NULL; 825 return -ENOMEM; 826 } 827 828 page = &(*pages)[*page_count - 1]; 829 memset(page, 0, sizeof(*page)); 830 page->id = blob->id; 831 page->sequence_num = *page_count - 1; 832 page->next = SPDK_INVALID_MD_PAGE; 833 *last_page = page; 834 835 return 0; 836 } 837 838 /* Transform the in-memory representation 'xattr' into an on-disk xattr descriptor. 839 * Update required_sz on both success and failure. 840 * 841 */ 842 static int 843 blob_serialize_xattr(const struct spdk_xattr *xattr, 844 uint8_t *buf, size_t buf_sz, 845 size_t *required_sz, bool internal) 846 { 847 struct spdk_blob_md_descriptor_xattr *desc; 848 849 *required_sz = sizeof(struct spdk_blob_md_descriptor_xattr) + 850 strlen(xattr->name) + 851 xattr->value_len; 852 853 if (buf_sz < *required_sz) { 854 return -1; 855 } 856 857 desc = (struct spdk_blob_md_descriptor_xattr *)buf; 858 859 desc->type = internal ? SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL : SPDK_MD_DESCRIPTOR_TYPE_XATTR; 860 desc->length = sizeof(desc->name_length) + 861 sizeof(desc->value_length) + 862 strlen(xattr->name) + 863 xattr->value_len; 864 desc->name_length = strlen(xattr->name); 865 desc->value_length = xattr->value_len; 866 867 memcpy(desc->name, xattr->name, desc->name_length); 868 memcpy((void *)((uintptr_t)desc->name + desc->name_length), 869 xattr->value, 870 desc->value_length); 871 872 return 0; 873 } 874 875 static void 876 blob_serialize_extent_table_entry(const struct spdk_blob *blob, 877 uint64_t start_ep, uint64_t *next_ep, 878 uint8_t **buf, size_t *remaining_sz) 879 { 880 struct spdk_blob_md_descriptor_extent_table *desc; 881 size_t cur_sz; 882 uint64_t i, et_idx; 883 uint32_t extent_page, ep_len; 884 885 /* The buffer must have room for at least num_clusters entry */ 886 cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc->num_clusters); 887 if (*remaining_sz < cur_sz) { 888 *next_ep = start_ep; 889 return; 890 } 891 892 desc = (struct spdk_blob_md_descriptor_extent_table *)*buf; 893 desc->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE; 894 895 desc->num_clusters = blob->active.num_clusters; 896 897 ep_len = 1; 898 et_idx = 0; 899 for (i = start_ep; i < blob->active.num_extent_pages; i++) { 900 if (*remaining_sz < cur_sz + sizeof(desc->extent_page[0])) { 901 /* If we ran out of buffer space, return */ 902 break; 903 } 904 905 extent_page = blob->active.extent_pages[i]; 906 /* Verify that next extent_page is unallocated */ 907 if (extent_page == 0 && 908 (i + 1 < blob->active.num_extent_pages && blob->active.extent_pages[i + 1] == 0)) { 909 ep_len++; 910 continue; 911 } 912 desc->extent_page[et_idx].page_idx = extent_page; 913 desc->extent_page[et_idx].num_pages = ep_len; 914 et_idx++; 915 916 ep_len = 1; 917 cur_sz += sizeof(desc->extent_page[et_idx]); 918 } 919 *next_ep = i; 920 921 desc->length = sizeof(desc->num_clusters) + sizeof(desc->extent_page[0]) * et_idx; 922 *remaining_sz -= sizeof(struct spdk_blob_md_descriptor) + desc->length; 923 *buf += sizeof(struct spdk_blob_md_descriptor) + desc->length; 924 } 925 926 static int 927 blob_serialize_extent_table(const struct spdk_blob *blob, 928 struct spdk_blob_md_page **pages, 929 struct spdk_blob_md_page *cur_page, 930 uint32_t *page_count, uint8_t **buf, 931 size_t *remaining_sz) 932 { 933 uint64_t last_extent_page; 934 int rc; 935 936 last_extent_page = 0; 937 /* At least single extent table entry has to be always persisted. 938 * Such case occurs with num_extent_pages == 0. */ 939 while (last_extent_page <= blob->active.num_extent_pages) { 940 blob_serialize_extent_table_entry(blob, last_extent_page, &last_extent_page, buf, 941 remaining_sz); 942 943 if (last_extent_page == blob->active.num_extent_pages) { 944 break; 945 } 946 947 rc = blob_serialize_add_page(blob, pages, page_count, &cur_page); 948 if (rc < 0) { 949 return rc; 950 } 951 952 *buf = (uint8_t *)cur_page->descriptors; 953 *remaining_sz = sizeof(cur_page->descriptors); 954 } 955 956 return 0; 957 } 958 959 static void 960 blob_serialize_extent_rle(const struct spdk_blob *blob, 961 uint64_t start_cluster, uint64_t *next_cluster, 962 uint8_t **buf, size_t *buf_sz) 963 { 964 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 965 size_t cur_sz; 966 uint64_t i, extent_idx; 967 uint64_t lba, lba_per_cluster, lba_count; 968 969 /* The buffer must have room for at least one extent */ 970 cur_sz = sizeof(struct spdk_blob_md_descriptor) + sizeof(desc_extent_rle->extents[0]); 971 if (*buf_sz < cur_sz) { 972 *next_cluster = start_cluster; 973 return; 974 } 975 976 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)*buf; 977 desc_extent_rle->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE; 978 979 lba_per_cluster = bs_cluster_to_lba(blob->bs, 1); 980 981 lba = blob->active.clusters[start_cluster]; 982 lba_count = lba_per_cluster; 983 extent_idx = 0; 984 for (i = start_cluster + 1; i < blob->active.num_clusters; i++) { 985 if ((lba + lba_count) == blob->active.clusters[i] && lba != 0) { 986 /* Run-length encode sequential non-zero LBA */ 987 lba_count += lba_per_cluster; 988 continue; 989 } else if (lba == 0 && blob->active.clusters[i] == 0) { 990 /* Run-length encode unallocated clusters */ 991 lba_count += lba_per_cluster; 992 continue; 993 } 994 desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 995 desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster; 996 extent_idx++; 997 998 cur_sz += sizeof(desc_extent_rle->extents[extent_idx]); 999 1000 if (*buf_sz < cur_sz) { 1001 /* If we ran out of buffer space, return */ 1002 *next_cluster = i; 1003 break; 1004 } 1005 1006 lba = blob->active.clusters[i]; 1007 lba_count = lba_per_cluster; 1008 } 1009 1010 if (*buf_sz >= cur_sz) { 1011 desc_extent_rle->extents[extent_idx].cluster_idx = lba / lba_per_cluster; 1012 desc_extent_rle->extents[extent_idx].length = lba_count / lba_per_cluster; 1013 extent_idx++; 1014 1015 *next_cluster = blob->active.num_clusters; 1016 } 1017 1018 desc_extent_rle->length = sizeof(desc_extent_rle->extents[0]) * extent_idx; 1019 *buf_sz -= sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length; 1020 *buf += sizeof(struct spdk_blob_md_descriptor) + desc_extent_rle->length; 1021 } 1022 1023 static int 1024 blob_serialize_extents_rle(const struct spdk_blob *blob, 1025 struct spdk_blob_md_page **pages, 1026 struct spdk_blob_md_page *cur_page, 1027 uint32_t *page_count, uint8_t **buf, 1028 size_t *remaining_sz) 1029 { 1030 uint64_t last_cluster; 1031 int rc; 1032 1033 last_cluster = 0; 1034 while (last_cluster < blob->active.num_clusters) { 1035 blob_serialize_extent_rle(blob, last_cluster, &last_cluster, buf, remaining_sz); 1036 1037 if (last_cluster == blob->active.num_clusters) { 1038 break; 1039 } 1040 1041 rc = blob_serialize_add_page(blob, pages, page_count, &cur_page); 1042 if (rc < 0) { 1043 return rc; 1044 } 1045 1046 *buf = (uint8_t *)cur_page->descriptors; 1047 *remaining_sz = sizeof(cur_page->descriptors); 1048 } 1049 1050 return 0; 1051 } 1052 1053 static void 1054 blob_serialize_extent_page(const struct spdk_blob *blob, 1055 uint64_t cluster, struct spdk_blob_md_page *page) 1056 { 1057 struct spdk_blob_md_descriptor_extent_page *desc_extent; 1058 uint64_t i, extent_idx; 1059 uint64_t lba, lba_per_cluster; 1060 uint64_t start_cluster_idx = (cluster / SPDK_EXTENTS_PER_EP) * SPDK_EXTENTS_PER_EP; 1061 1062 desc_extent = (struct spdk_blob_md_descriptor_extent_page *) page->descriptors; 1063 desc_extent->type = SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE; 1064 1065 lba_per_cluster = bs_cluster_to_lba(blob->bs, 1); 1066 1067 desc_extent->start_cluster_idx = start_cluster_idx; 1068 extent_idx = 0; 1069 for (i = start_cluster_idx; i < blob->active.num_clusters; i++) { 1070 lba = blob->active.clusters[i]; 1071 desc_extent->cluster_idx[extent_idx++] = lba / lba_per_cluster; 1072 if (extent_idx >= SPDK_EXTENTS_PER_EP) { 1073 break; 1074 } 1075 } 1076 desc_extent->length = sizeof(desc_extent->start_cluster_idx) + 1077 sizeof(desc_extent->cluster_idx[0]) * extent_idx; 1078 } 1079 1080 static void 1081 blob_serialize_flags(const struct spdk_blob *blob, 1082 uint8_t *buf, size_t *buf_sz) 1083 { 1084 struct spdk_blob_md_descriptor_flags *desc; 1085 1086 /* 1087 * Flags get serialized first, so we should always have room for the flags 1088 * descriptor. 1089 */ 1090 assert(*buf_sz >= sizeof(*desc)); 1091 1092 desc = (struct spdk_blob_md_descriptor_flags *)buf; 1093 desc->type = SPDK_MD_DESCRIPTOR_TYPE_FLAGS; 1094 desc->length = sizeof(*desc) - sizeof(struct spdk_blob_md_descriptor); 1095 desc->invalid_flags = blob->invalid_flags; 1096 desc->data_ro_flags = blob->data_ro_flags; 1097 desc->md_ro_flags = blob->md_ro_flags; 1098 1099 *buf_sz -= sizeof(*desc); 1100 } 1101 1102 static int 1103 blob_serialize_xattrs(const struct spdk_blob *blob, 1104 const struct spdk_xattr_tailq *xattrs, bool internal, 1105 struct spdk_blob_md_page **pages, 1106 struct spdk_blob_md_page *cur_page, 1107 uint32_t *page_count, uint8_t **buf, 1108 size_t *remaining_sz) 1109 { 1110 const struct spdk_xattr *xattr; 1111 int rc; 1112 1113 TAILQ_FOREACH(xattr, xattrs, link) { 1114 size_t required_sz = 0; 1115 1116 rc = blob_serialize_xattr(xattr, 1117 *buf, *remaining_sz, 1118 &required_sz, internal); 1119 if (rc < 0) { 1120 /* Need to add a new page to the chain */ 1121 rc = blob_serialize_add_page(blob, pages, page_count, 1122 &cur_page); 1123 if (rc < 0) { 1124 spdk_free(*pages); 1125 *pages = NULL; 1126 *page_count = 0; 1127 return rc; 1128 } 1129 1130 *buf = (uint8_t *)cur_page->descriptors; 1131 *remaining_sz = sizeof(cur_page->descriptors); 1132 1133 /* Try again */ 1134 required_sz = 0; 1135 rc = blob_serialize_xattr(xattr, 1136 *buf, *remaining_sz, 1137 &required_sz, internal); 1138 1139 if (rc < 0) { 1140 spdk_free(*pages); 1141 *pages = NULL; 1142 *page_count = 0; 1143 return rc; 1144 } 1145 } 1146 1147 *remaining_sz -= required_sz; 1148 *buf += required_sz; 1149 } 1150 1151 return 0; 1152 } 1153 1154 static int 1155 blob_serialize(const struct spdk_blob *blob, struct spdk_blob_md_page **pages, 1156 uint32_t *page_count) 1157 { 1158 struct spdk_blob_md_page *cur_page; 1159 int rc; 1160 uint8_t *buf; 1161 size_t remaining_sz; 1162 1163 assert(pages != NULL); 1164 assert(page_count != NULL); 1165 assert(blob != NULL); 1166 assert(blob->state == SPDK_BLOB_STATE_DIRTY); 1167 1168 *pages = NULL; 1169 *page_count = 0; 1170 1171 /* A blob always has at least 1 page, even if it has no descriptors */ 1172 rc = blob_serialize_add_page(blob, pages, page_count, &cur_page); 1173 if (rc < 0) { 1174 return rc; 1175 } 1176 1177 buf = (uint8_t *)cur_page->descriptors; 1178 remaining_sz = sizeof(cur_page->descriptors); 1179 1180 /* Serialize flags */ 1181 blob_serialize_flags(blob, buf, &remaining_sz); 1182 buf += sizeof(struct spdk_blob_md_descriptor_flags); 1183 1184 /* Serialize xattrs */ 1185 rc = blob_serialize_xattrs(blob, &blob->xattrs, false, 1186 pages, cur_page, page_count, &buf, &remaining_sz); 1187 if (rc < 0) { 1188 return rc; 1189 } 1190 1191 /* Serialize internal xattrs */ 1192 rc = blob_serialize_xattrs(blob, &blob->xattrs_internal, true, 1193 pages, cur_page, page_count, &buf, &remaining_sz); 1194 if (rc < 0) { 1195 return rc; 1196 } 1197 1198 if (blob->use_extent_table) { 1199 /* Serialize extent table */ 1200 rc = blob_serialize_extent_table(blob, pages, cur_page, page_count, &buf, &remaining_sz); 1201 } else { 1202 /* Serialize extents */ 1203 rc = blob_serialize_extents_rle(blob, pages, cur_page, page_count, &buf, &remaining_sz); 1204 } 1205 1206 return rc; 1207 } 1208 1209 struct spdk_blob_load_ctx { 1210 struct spdk_blob *blob; 1211 1212 struct spdk_blob_md_page *pages; 1213 uint32_t num_pages; 1214 uint32_t next_extent_page; 1215 spdk_bs_sequence_t *seq; 1216 1217 spdk_bs_sequence_cpl cb_fn; 1218 void *cb_arg; 1219 }; 1220 1221 static uint32_t 1222 blob_md_page_calc_crc(void *page) 1223 { 1224 uint32_t crc; 1225 1226 crc = BLOB_CRC32C_INITIAL; 1227 crc = spdk_crc32c_update(page, SPDK_BS_PAGE_SIZE - 4, crc); 1228 crc ^= BLOB_CRC32C_INITIAL; 1229 1230 return crc; 1231 1232 } 1233 1234 static void 1235 blob_load_final(struct spdk_blob_load_ctx *ctx, int bserrno) 1236 { 1237 struct spdk_blob *blob = ctx->blob; 1238 1239 if (bserrno == 0) { 1240 blob_mark_clean(blob); 1241 } 1242 1243 ctx->cb_fn(ctx->seq, ctx->cb_arg, bserrno); 1244 1245 /* Free the memory */ 1246 spdk_free(ctx->pages); 1247 free(ctx); 1248 } 1249 1250 static void 1251 blob_load_snapshot_cpl(void *cb_arg, struct spdk_blob *snapshot, int bserrno) 1252 { 1253 struct spdk_blob_load_ctx *ctx = cb_arg; 1254 struct spdk_blob *blob = ctx->blob; 1255 1256 if (bserrno == 0) { 1257 blob->back_bs_dev = bs_create_blob_bs_dev(snapshot); 1258 if (blob->back_bs_dev == NULL) { 1259 bserrno = -ENOMEM; 1260 } 1261 } 1262 if (bserrno != 0) { 1263 SPDK_ERRLOG("Snapshot fail\n"); 1264 } 1265 1266 blob_load_final(ctx, bserrno); 1267 } 1268 1269 static void blob_update_clear_method(struct spdk_blob *blob); 1270 1271 static void 1272 blob_load_backing_dev(void *cb_arg) 1273 { 1274 struct spdk_blob_load_ctx *ctx = cb_arg; 1275 struct spdk_blob *blob = ctx->blob; 1276 const void *value; 1277 size_t len; 1278 int rc; 1279 1280 if (spdk_blob_is_thin_provisioned(blob)) { 1281 rc = blob_get_xattr_value(blob, BLOB_SNAPSHOT, &value, &len, true); 1282 if (rc == 0) { 1283 if (len != sizeof(spdk_blob_id)) { 1284 blob_load_final(ctx, -EINVAL); 1285 return; 1286 } 1287 /* open snapshot blob and continue in the callback function */ 1288 blob->parent_id = *(spdk_blob_id *)value; 1289 spdk_bs_open_blob(blob->bs, blob->parent_id, 1290 blob_load_snapshot_cpl, ctx); 1291 return; 1292 } else { 1293 /* add zeroes_dev for thin provisioned blob */ 1294 blob->back_bs_dev = bs_create_zeroes_dev(); 1295 } 1296 } else { 1297 /* standard blob */ 1298 blob->back_bs_dev = NULL; 1299 } 1300 blob_load_final(ctx, 0); 1301 } 1302 1303 static void 1304 blob_load_cpl_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1305 { 1306 struct spdk_blob_load_ctx *ctx = cb_arg; 1307 struct spdk_blob *blob = ctx->blob; 1308 struct spdk_blob_md_page *page; 1309 uint64_t i; 1310 uint32_t crc; 1311 uint64_t lba; 1312 void *tmp; 1313 uint64_t sz; 1314 1315 if (bserrno) { 1316 SPDK_ERRLOG("Extent page read failed: %d\n", bserrno); 1317 blob_load_final(ctx, bserrno); 1318 return; 1319 } 1320 1321 if (ctx->pages == NULL) { 1322 /* First iteration of this function, allocate buffer for single EXTENT_PAGE */ 1323 ctx->pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0, 1324 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 1325 if (!ctx->pages) { 1326 blob_load_final(ctx, -ENOMEM); 1327 return; 1328 } 1329 ctx->num_pages = 1; 1330 ctx->next_extent_page = 0; 1331 } else { 1332 page = &ctx->pages[0]; 1333 crc = blob_md_page_calc_crc(page); 1334 if (crc != page->crc) { 1335 blob_load_final(ctx, -EINVAL); 1336 return; 1337 } 1338 1339 if (page->next != SPDK_INVALID_MD_PAGE) { 1340 blob_load_final(ctx, -EINVAL); 1341 return; 1342 } 1343 1344 bserrno = blob_parse_extent_page(page, blob); 1345 if (bserrno) { 1346 blob_load_final(ctx, bserrno); 1347 return; 1348 } 1349 } 1350 1351 for (i = ctx->next_extent_page; i < blob->active.num_extent_pages; i++) { 1352 if (blob->active.extent_pages[i] != 0) { 1353 /* Extent page was allocated, read and parse it. */ 1354 lba = bs_md_page_to_lba(blob->bs, blob->active.extent_pages[i]); 1355 ctx->next_extent_page = i + 1; 1356 1357 bs_sequence_read_dev(seq, &ctx->pages[0], lba, 1358 bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE), 1359 blob_load_cpl_extents_cpl, ctx); 1360 return; 1361 } else { 1362 /* Thin provisioned blobs can point to unallocated extent pages. 1363 * In this case blob size should be increased by up to the amount left in remaining_clusters_in_et. */ 1364 1365 sz = spdk_min(blob->remaining_clusters_in_et, SPDK_EXTENTS_PER_EP); 1366 blob->active.num_clusters += sz; 1367 blob->remaining_clusters_in_et -= sz; 1368 1369 assert(spdk_blob_is_thin_provisioned(blob)); 1370 assert(i + 1 < blob->active.num_extent_pages || blob->remaining_clusters_in_et == 0); 1371 1372 tmp = realloc(blob->active.clusters, blob->active.num_clusters * sizeof(*blob->active.clusters)); 1373 if (tmp == NULL) { 1374 blob_load_final(ctx, -ENOMEM); 1375 return; 1376 } 1377 memset(tmp + sizeof(*blob->active.clusters) * blob->active.cluster_array_size, 0, 1378 sizeof(*blob->active.clusters) * (blob->active.num_clusters - blob->active.cluster_array_size)); 1379 blob->active.clusters = tmp; 1380 blob->active.cluster_array_size = blob->active.num_clusters; 1381 } 1382 } 1383 1384 blob_load_backing_dev(ctx); 1385 } 1386 1387 static void 1388 blob_load_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1389 { 1390 struct spdk_blob_load_ctx *ctx = cb_arg; 1391 struct spdk_blob *blob = ctx->blob; 1392 struct spdk_blob_md_page *page; 1393 int rc; 1394 uint32_t crc; 1395 uint32_t current_page; 1396 1397 if (ctx->num_pages == 1) { 1398 current_page = bs_blobid_to_page(blob->id); 1399 } else { 1400 assert(ctx->num_pages != 0); 1401 page = &ctx->pages[ctx->num_pages - 2]; 1402 current_page = page->next; 1403 } 1404 1405 if (bserrno) { 1406 SPDK_ERRLOG("Metadata page %d read failed for blobid %lu: %d\n", 1407 current_page, blob->id, bserrno); 1408 blob_load_final(ctx, bserrno); 1409 return; 1410 } 1411 1412 page = &ctx->pages[ctx->num_pages - 1]; 1413 crc = blob_md_page_calc_crc(page); 1414 if (crc != page->crc) { 1415 SPDK_ERRLOG("Metadata page %d crc mismatch for blobid %lu\n", 1416 current_page, blob->id); 1417 blob_load_final(ctx, -EINVAL); 1418 return; 1419 } 1420 1421 if (page->next != SPDK_INVALID_MD_PAGE) { 1422 uint32_t next_page = page->next; 1423 uint64_t next_lba = bs_md_page_to_lba(blob->bs, next_page); 1424 1425 /* Read the next page */ 1426 ctx->num_pages++; 1427 ctx->pages = spdk_realloc(ctx->pages, (sizeof(*page) * ctx->num_pages), 0); 1428 if (ctx->pages == NULL) { 1429 blob_load_final(ctx, -ENOMEM); 1430 return; 1431 } 1432 1433 bs_sequence_read_dev(seq, &ctx->pages[ctx->num_pages - 1], 1434 next_lba, 1435 bs_byte_to_lba(blob->bs, sizeof(*page)), 1436 blob_load_cpl, ctx); 1437 return; 1438 } 1439 1440 /* Parse the pages */ 1441 rc = blob_parse(ctx->pages, ctx->num_pages, blob); 1442 if (rc) { 1443 blob_load_final(ctx, rc); 1444 return; 1445 } 1446 1447 if (blob->extent_table_found == true) { 1448 /* If EXTENT_TABLE was found, that means support for it should be enabled. */ 1449 assert(blob->extent_rle_found == false); 1450 blob->use_extent_table = true; 1451 } else { 1452 /* If EXTENT_RLE or no extent_* descriptor was found disable support 1453 * for extent table. No extent_* descriptors means that blob has length of 0 1454 * and no extent_rle descriptors were persisted for it. 1455 * EXTENT_TABLE if used, is always present in metadata regardless of length. */ 1456 blob->use_extent_table = false; 1457 } 1458 1459 /* Check the clear_method stored in metadata vs what may have been passed 1460 * via spdk_bs_open_blob_ext() and update accordingly. 1461 */ 1462 blob_update_clear_method(blob); 1463 1464 spdk_free(ctx->pages); 1465 ctx->pages = NULL; 1466 1467 if (blob->extent_table_found) { 1468 blob_load_cpl_extents_cpl(seq, ctx, 0); 1469 } else { 1470 blob_load_backing_dev(ctx); 1471 } 1472 } 1473 1474 /* Load a blob from disk given a blobid */ 1475 static void 1476 blob_load(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 1477 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 1478 { 1479 struct spdk_blob_load_ctx *ctx; 1480 struct spdk_blob_store *bs; 1481 uint32_t page_num; 1482 uint64_t lba; 1483 1484 blob_verify_md_op(blob); 1485 1486 bs = blob->bs; 1487 1488 ctx = calloc(1, sizeof(*ctx)); 1489 if (!ctx) { 1490 cb_fn(seq, cb_arg, -ENOMEM); 1491 return; 1492 } 1493 1494 ctx->blob = blob; 1495 ctx->pages = spdk_realloc(ctx->pages, SPDK_BS_PAGE_SIZE, 0); 1496 if (!ctx->pages) { 1497 free(ctx); 1498 cb_fn(seq, cb_arg, -ENOMEM); 1499 return; 1500 } 1501 ctx->num_pages = 1; 1502 ctx->cb_fn = cb_fn; 1503 ctx->cb_arg = cb_arg; 1504 ctx->seq = seq; 1505 1506 page_num = bs_blobid_to_page(blob->id); 1507 lba = bs_md_page_to_lba(blob->bs, page_num); 1508 1509 blob->state = SPDK_BLOB_STATE_LOADING; 1510 1511 bs_sequence_read_dev(seq, &ctx->pages[0], lba, 1512 bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE), 1513 blob_load_cpl, ctx); 1514 } 1515 1516 struct spdk_blob_persist_ctx { 1517 struct spdk_blob *blob; 1518 1519 struct spdk_bs_super_block *super; 1520 1521 struct spdk_blob_md_page *pages; 1522 uint32_t next_extent_page; 1523 struct spdk_blob_md_page *extent_page; 1524 1525 spdk_bs_sequence_t *seq; 1526 spdk_bs_sequence_cpl cb_fn; 1527 void *cb_arg; 1528 TAILQ_ENTRY(spdk_blob_persist_ctx) link; 1529 }; 1530 1531 static void 1532 bs_batch_clear_dev(struct spdk_blob_persist_ctx *ctx, spdk_bs_batch_t *batch, uint64_t lba, 1533 uint32_t lba_count) 1534 { 1535 switch (ctx->blob->clear_method) { 1536 case BLOB_CLEAR_WITH_DEFAULT: 1537 case BLOB_CLEAR_WITH_UNMAP: 1538 bs_batch_unmap_dev(batch, lba, lba_count); 1539 break; 1540 case BLOB_CLEAR_WITH_WRITE_ZEROES: 1541 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1542 break; 1543 case BLOB_CLEAR_WITH_NONE: 1544 default: 1545 break; 1546 } 1547 } 1548 1549 static void blob_persist_check_dirty(struct spdk_blob_persist_ctx *ctx); 1550 1551 static void 1552 blob_persist_complete(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx, int bserrno) 1553 { 1554 struct spdk_blob_persist_ctx *next_persist; 1555 struct spdk_blob *blob = ctx->blob; 1556 1557 if (bserrno == 0) { 1558 blob_mark_clean(blob); 1559 } 1560 1561 assert(ctx == TAILQ_FIRST(&blob->pending_persists)); 1562 TAILQ_REMOVE(&blob->pending_persists, ctx, link); 1563 1564 next_persist = TAILQ_FIRST(&blob->pending_persists); 1565 1566 /* Call user callback */ 1567 ctx->cb_fn(seq, ctx->cb_arg, bserrno); 1568 1569 /* Free the memory */ 1570 spdk_free(ctx->pages); 1571 free(ctx); 1572 1573 if (next_persist != NULL) { 1574 blob_persist_check_dirty(next_persist); 1575 } 1576 } 1577 1578 static void 1579 blob_persist_clear_extents_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1580 { 1581 struct spdk_blob_persist_ctx *ctx = cb_arg; 1582 struct spdk_blob *blob = ctx->blob; 1583 struct spdk_blob_store *bs = blob->bs; 1584 size_t i; 1585 1586 if (bserrno != 0) { 1587 blob_persist_complete(seq, ctx, bserrno); 1588 return; 1589 } 1590 1591 /* Release all extent_pages that were truncated */ 1592 for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) { 1593 /* Nothing to release if it was not allocated */ 1594 if (blob->active.extent_pages[i] != 0) { 1595 bs_release_md_page(bs, blob->active.extent_pages[i]); 1596 } 1597 } 1598 1599 if (blob->active.num_extent_pages == 0) { 1600 free(blob->active.extent_pages); 1601 blob->active.extent_pages = NULL; 1602 blob->active.extent_pages_array_size = 0; 1603 } else if (blob->active.num_extent_pages != blob->active.extent_pages_array_size) { 1604 #ifndef __clang_analyzer__ 1605 void *tmp; 1606 1607 /* scan-build really can't figure reallocs, workaround it */ 1608 tmp = realloc(blob->active.extent_pages, sizeof(uint32_t) * blob->active.num_extent_pages); 1609 assert(tmp != NULL); 1610 blob->active.extent_pages = tmp; 1611 #endif 1612 blob->active.extent_pages_array_size = blob->active.num_extent_pages; 1613 } 1614 1615 blob_persist_complete(seq, ctx, bserrno); 1616 } 1617 1618 static void 1619 blob_persist_clear_extents(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx) 1620 { 1621 struct spdk_blob *blob = ctx->blob; 1622 struct spdk_blob_store *bs = blob->bs; 1623 size_t i; 1624 uint64_t lba; 1625 uint32_t lba_count; 1626 spdk_bs_batch_t *batch; 1627 1628 batch = bs_sequence_to_batch(seq, blob_persist_clear_extents_cpl, ctx); 1629 lba_count = bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE); 1630 1631 /* Clear all extent_pages that were truncated */ 1632 for (i = blob->active.num_extent_pages; i < blob->active.extent_pages_array_size; i++) { 1633 /* Nothing to clear if it was not allocated */ 1634 if (blob->active.extent_pages[i] != 0) { 1635 lba = bs_md_page_to_lba(bs, blob->clean.extent_pages[i]); 1636 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1637 } 1638 } 1639 1640 bs_batch_close(batch); 1641 } 1642 1643 static void 1644 blob_persist_clear_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1645 { 1646 struct spdk_blob_persist_ctx *ctx = cb_arg; 1647 struct spdk_blob *blob = ctx->blob; 1648 struct spdk_blob_store *bs = blob->bs; 1649 size_t i; 1650 1651 if (bserrno != 0) { 1652 blob_persist_complete(seq, ctx, bserrno); 1653 return; 1654 } 1655 1656 pthread_mutex_lock(&bs->used_clusters_mutex); 1657 /* Release all clusters that were truncated */ 1658 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1659 uint32_t cluster_num = bs_lba_to_cluster(bs, blob->active.clusters[i]); 1660 1661 /* Nothing to release if it was not allocated */ 1662 if (blob->active.clusters[i] != 0) { 1663 bs_release_cluster(bs, cluster_num); 1664 } 1665 } 1666 pthread_mutex_unlock(&bs->used_clusters_mutex); 1667 1668 if (blob->active.num_clusters == 0) { 1669 free(blob->active.clusters); 1670 blob->active.clusters = NULL; 1671 blob->active.cluster_array_size = 0; 1672 } else if (blob->active.num_clusters != blob->active.cluster_array_size) { 1673 #ifndef __clang_analyzer__ 1674 void *tmp; 1675 1676 /* scan-build really can't figure reallocs, workaround it */ 1677 tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * blob->active.num_clusters); 1678 assert(tmp != NULL); 1679 blob->active.clusters = tmp; 1680 1681 #endif 1682 blob->active.cluster_array_size = blob->active.num_clusters; 1683 } 1684 1685 /* Move on to clearing extent pages */ 1686 blob_persist_clear_extents(seq, ctx); 1687 } 1688 1689 static void 1690 blob_persist_clear_clusters(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx) 1691 { 1692 struct spdk_blob *blob = ctx->blob; 1693 struct spdk_blob_store *bs = blob->bs; 1694 spdk_bs_batch_t *batch; 1695 size_t i; 1696 uint64_t lba; 1697 uint32_t lba_count; 1698 1699 /* Clusters don't move around in blobs. The list shrinks or grows 1700 * at the end, but no changes ever occur in the middle of the list. 1701 */ 1702 1703 batch = bs_sequence_to_batch(seq, blob_persist_clear_clusters_cpl, ctx); 1704 1705 /* Clear all clusters that were truncated */ 1706 lba = 0; 1707 lba_count = 0; 1708 for (i = blob->active.num_clusters; i < blob->active.cluster_array_size; i++) { 1709 uint64_t next_lba = blob->active.clusters[i]; 1710 uint32_t next_lba_count = bs_cluster_to_lba(bs, 1); 1711 1712 if (next_lba > 0 && (lba + lba_count) == next_lba) { 1713 /* This cluster is contiguous with the previous one. */ 1714 lba_count += next_lba_count; 1715 continue; 1716 } 1717 1718 /* This cluster is not contiguous with the previous one. */ 1719 1720 /* If a run of LBAs previously existing, clear them now */ 1721 if (lba_count > 0) { 1722 bs_batch_clear_dev(ctx, batch, lba, lba_count); 1723 } 1724 1725 /* Start building the next batch */ 1726 lba = next_lba; 1727 if (next_lba > 0) { 1728 lba_count = next_lba_count; 1729 } else { 1730 lba_count = 0; 1731 } 1732 } 1733 1734 /* If we ended with a contiguous set of LBAs, clear them now */ 1735 if (lba_count > 0) { 1736 bs_batch_clear_dev(ctx, batch, lba, lba_count); 1737 } 1738 1739 bs_batch_close(batch); 1740 } 1741 1742 static void 1743 blob_persist_zero_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1744 { 1745 struct spdk_blob_persist_ctx *ctx = cb_arg; 1746 struct spdk_blob *blob = ctx->blob; 1747 struct spdk_blob_store *bs = blob->bs; 1748 size_t i; 1749 1750 if (bserrno != 0) { 1751 blob_persist_complete(seq, ctx, bserrno); 1752 return; 1753 } 1754 1755 /* This loop starts at 1 because the first page is special and handled 1756 * below. The pages (except the first) are never written in place, 1757 * so any pages in the clean list must be zeroed. 1758 */ 1759 for (i = 1; i < blob->clean.num_pages; i++) { 1760 bs_release_md_page(bs, blob->clean.pages[i]); 1761 } 1762 1763 if (blob->active.num_pages == 0) { 1764 uint32_t page_num; 1765 1766 page_num = bs_blobid_to_page(blob->id); 1767 bs_release_md_page(bs, page_num); 1768 } 1769 1770 /* Move on to clearing clusters */ 1771 blob_persist_clear_clusters(seq, ctx); 1772 } 1773 1774 static void 1775 blob_persist_zero_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1776 { 1777 struct spdk_blob_persist_ctx *ctx = cb_arg; 1778 struct spdk_blob *blob = ctx->blob; 1779 struct spdk_blob_store *bs = blob->bs; 1780 uint64_t lba; 1781 uint32_t lba_count; 1782 spdk_bs_batch_t *batch; 1783 size_t i; 1784 1785 if (bserrno != 0) { 1786 blob_persist_complete(seq, ctx, bserrno); 1787 return; 1788 } 1789 1790 batch = bs_sequence_to_batch(seq, blob_persist_zero_pages_cpl, ctx); 1791 1792 lba_count = bs_byte_to_lba(bs, SPDK_BS_PAGE_SIZE); 1793 1794 /* This loop starts at 1 because the first page is special and handled 1795 * below. The pages (except the first) are never written in place, 1796 * so any pages in the clean list must be zeroed. 1797 */ 1798 for (i = 1; i < blob->clean.num_pages; i++) { 1799 lba = bs_md_page_to_lba(bs, blob->clean.pages[i]); 1800 1801 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1802 } 1803 1804 /* The first page will only be zeroed if this is a delete. */ 1805 if (blob->active.num_pages == 0) { 1806 uint32_t page_num; 1807 1808 /* The first page in the metadata goes where the blobid indicates */ 1809 page_num = bs_blobid_to_page(blob->id); 1810 lba = bs_md_page_to_lba(bs, page_num); 1811 1812 bs_batch_write_zeroes_dev(batch, lba, lba_count); 1813 } 1814 1815 bs_batch_close(batch); 1816 } 1817 1818 static void 1819 blob_persist_write_page_root(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 1820 { 1821 struct spdk_blob_persist_ctx *ctx = cb_arg; 1822 struct spdk_blob *blob = ctx->blob; 1823 struct spdk_blob_store *bs = blob->bs; 1824 uint64_t lba; 1825 uint32_t lba_count; 1826 struct spdk_blob_md_page *page; 1827 1828 if (bserrno != 0) { 1829 blob_persist_complete(seq, ctx, bserrno); 1830 return; 1831 } 1832 1833 if (blob->active.num_pages == 0) { 1834 /* Move on to the next step */ 1835 blob_persist_zero_pages(seq, ctx, 0); 1836 return; 1837 } 1838 1839 lba_count = bs_byte_to_lba(bs, sizeof(*page)); 1840 1841 page = &ctx->pages[0]; 1842 /* The first page in the metadata goes where the blobid indicates */ 1843 lba = bs_md_page_to_lba(bs, bs_blobid_to_page(blob->id)); 1844 1845 bs_sequence_write_dev(seq, page, lba, lba_count, 1846 blob_persist_zero_pages, ctx); 1847 } 1848 1849 static void 1850 blob_persist_write_page_chain(spdk_bs_sequence_t *seq, struct spdk_blob_persist_ctx *ctx) 1851 { 1852 struct spdk_blob *blob = ctx->blob; 1853 struct spdk_blob_store *bs = blob->bs; 1854 uint64_t lba; 1855 uint32_t lba_count; 1856 struct spdk_blob_md_page *page; 1857 spdk_bs_batch_t *batch; 1858 size_t i; 1859 1860 /* Clusters don't move around in blobs. The list shrinks or grows 1861 * at the end, but no changes ever occur in the middle of the list. 1862 */ 1863 1864 lba_count = bs_byte_to_lba(bs, sizeof(*page)); 1865 1866 batch = bs_sequence_to_batch(seq, blob_persist_write_page_root, ctx); 1867 1868 /* This starts at 1. The root page is not written until 1869 * all of the others are finished 1870 */ 1871 for (i = 1; i < blob->active.num_pages; i++) { 1872 page = &ctx->pages[i]; 1873 assert(page->sequence_num == i); 1874 1875 lba = bs_md_page_to_lba(bs, blob->active.pages[i]); 1876 1877 bs_batch_write_dev(batch, page, lba, lba_count); 1878 } 1879 1880 bs_batch_close(batch); 1881 } 1882 1883 static int 1884 blob_resize(struct spdk_blob *blob, uint64_t sz) 1885 { 1886 uint64_t i; 1887 uint64_t *tmp; 1888 uint64_t cluster; 1889 uint32_t lfmd; /* lowest free md page */ 1890 uint64_t num_clusters; 1891 uint32_t *ep_tmp; 1892 uint64_t new_num_ep = 0, current_num_ep = 0; 1893 struct spdk_blob_store *bs; 1894 1895 bs = blob->bs; 1896 1897 blob_verify_md_op(blob); 1898 1899 if (blob->active.num_clusters == sz) { 1900 return 0; 1901 } 1902 1903 if (blob->active.num_clusters < blob->active.cluster_array_size) { 1904 /* If this blob was resized to be larger, then smaller, then 1905 * larger without syncing, then the cluster array already 1906 * contains spare assigned clusters we can use. 1907 */ 1908 num_clusters = spdk_min(blob->active.cluster_array_size, 1909 sz); 1910 } else { 1911 num_clusters = blob->active.num_clusters; 1912 } 1913 1914 if (blob->use_extent_table) { 1915 /* Round up since every cluster beyond current Extent Table size, 1916 * requires new extent page. */ 1917 new_num_ep = spdk_divide_round_up(sz, SPDK_EXTENTS_PER_EP); 1918 current_num_ep = spdk_divide_round_up(num_clusters, SPDK_EXTENTS_PER_EP); 1919 } 1920 1921 /* Check first that we have enough clusters and md pages before we start claiming them. */ 1922 if (sz > num_clusters && spdk_blob_is_thin_provisioned(blob) == false) { 1923 if ((sz - num_clusters) > bs->num_free_clusters) { 1924 return -ENOSPC; 1925 } 1926 lfmd = 0; 1927 for (i = current_num_ep; i < new_num_ep ; i++) { 1928 lfmd = spdk_bit_array_find_first_clear(blob->bs->used_md_pages, lfmd); 1929 if (lfmd == UINT32_MAX) { 1930 /* No more free md pages. Cannot satisfy the request */ 1931 return -ENOSPC; 1932 } 1933 } 1934 } 1935 1936 if (sz > num_clusters) { 1937 /* Expand the cluster array if necessary. 1938 * We only shrink the array when persisting. 1939 */ 1940 tmp = realloc(blob->active.clusters, sizeof(*blob->active.clusters) * sz); 1941 if (sz > 0 && tmp == NULL) { 1942 return -ENOMEM; 1943 } 1944 memset(tmp + blob->active.cluster_array_size, 0, 1945 sizeof(*blob->active.clusters) * (sz - blob->active.cluster_array_size)); 1946 blob->active.clusters = tmp; 1947 blob->active.cluster_array_size = sz; 1948 1949 /* Expand the extents table, only if enough clusters were added */ 1950 if (new_num_ep > current_num_ep && blob->use_extent_table) { 1951 ep_tmp = realloc(blob->active.extent_pages, sizeof(*blob->active.extent_pages) * new_num_ep); 1952 if (new_num_ep > 0 && ep_tmp == NULL) { 1953 return -ENOMEM; 1954 } 1955 memset(ep_tmp + blob->active.extent_pages_array_size, 0, 1956 sizeof(*blob->active.extent_pages) * (new_num_ep - blob->active.extent_pages_array_size)); 1957 blob->active.extent_pages = ep_tmp; 1958 blob->active.extent_pages_array_size = new_num_ep; 1959 } 1960 } 1961 1962 blob->state = SPDK_BLOB_STATE_DIRTY; 1963 1964 if (spdk_blob_is_thin_provisioned(blob) == false) { 1965 cluster = 0; 1966 lfmd = 0; 1967 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 1968 for (i = num_clusters; i < sz; i++) { 1969 bs_allocate_cluster(blob, i, &cluster, &lfmd, true); 1970 lfmd++; 1971 } 1972 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 1973 } 1974 1975 blob->active.num_clusters = sz; 1976 blob->active.num_extent_pages = new_num_ep; 1977 1978 return 0; 1979 } 1980 1981 static void 1982 blob_persist_generate_new_md(struct spdk_blob_persist_ctx *ctx) 1983 { 1984 spdk_bs_sequence_t *seq = ctx->seq; 1985 struct spdk_blob *blob = ctx->blob; 1986 struct spdk_blob_store *bs = blob->bs; 1987 uint64_t i; 1988 uint32_t page_num; 1989 void *tmp; 1990 int rc; 1991 1992 /* Generate the new metadata */ 1993 rc = blob_serialize(blob, &ctx->pages, &blob->active.num_pages); 1994 if (rc < 0) { 1995 blob_persist_complete(seq, ctx, rc); 1996 return; 1997 } 1998 1999 assert(blob->active.num_pages >= 1); 2000 2001 /* Resize the cache of page indices */ 2002 tmp = realloc(blob->active.pages, blob->active.num_pages * sizeof(*blob->active.pages)); 2003 if (!tmp) { 2004 blob_persist_complete(seq, ctx, -ENOMEM); 2005 return; 2006 } 2007 blob->active.pages = tmp; 2008 2009 /* Assign this metadata to pages. This requires two passes - 2010 * one to verify that there are enough pages and a second 2011 * to actually claim them. */ 2012 page_num = 0; 2013 /* Note that this loop starts at one. The first page location is fixed by the blobid. */ 2014 for (i = 1; i < blob->active.num_pages; i++) { 2015 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 2016 if (page_num == UINT32_MAX) { 2017 blob_persist_complete(seq, ctx, -ENOMEM); 2018 return; 2019 } 2020 page_num++; 2021 } 2022 2023 page_num = 0; 2024 blob->active.pages[0] = bs_blobid_to_page(blob->id); 2025 for (i = 1; i < blob->active.num_pages; i++) { 2026 page_num = spdk_bit_array_find_first_clear(bs->used_md_pages, page_num); 2027 ctx->pages[i - 1].next = page_num; 2028 /* Now that previous metadata page is complete, calculate the crc for it. */ 2029 ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]); 2030 blob->active.pages[i] = page_num; 2031 bs_claim_md_page(bs, page_num); 2032 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Claiming page %u for blob %lu\n", page_num, blob->id); 2033 page_num++; 2034 } 2035 ctx->pages[i - 1].crc = blob_md_page_calc_crc(&ctx->pages[i - 1]); 2036 /* Start writing the metadata from last page to first */ 2037 blob->state = SPDK_BLOB_STATE_CLEAN; 2038 blob_persist_write_page_chain(seq, ctx); 2039 } 2040 2041 static void 2042 blob_persist_write_extent_pages(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2043 { 2044 struct spdk_blob_persist_ctx *ctx = cb_arg; 2045 struct spdk_blob *blob = ctx->blob; 2046 size_t i; 2047 uint32_t extent_page_id; 2048 uint32_t page_count = 0; 2049 int rc; 2050 2051 if (ctx->extent_page != NULL) { 2052 spdk_free(ctx->extent_page); 2053 ctx->extent_page = NULL; 2054 } 2055 2056 if (bserrno != 0) { 2057 blob_persist_complete(seq, ctx, bserrno); 2058 return; 2059 } 2060 2061 /* Only write out changed extent pages */ 2062 for (i = ctx->next_extent_page; i < blob->active.num_extent_pages; i++) { 2063 extent_page_id = blob->active.extent_pages[i]; 2064 if (extent_page_id == 0) { 2065 /* No Extent Page to persist */ 2066 assert(spdk_blob_is_thin_provisioned(blob)); 2067 continue; 2068 } 2069 /* Writing out new extent page for the first time. Either active extent pages is larger 2070 * than clean extent pages or there was no extent page assigned due to thin provisioning. */ 2071 if (i >= blob->clean.extent_pages_array_size || blob->clean.extent_pages[i] == 0) { 2072 blob->state = SPDK_BLOB_STATE_DIRTY; 2073 assert(spdk_bit_array_get(blob->bs->used_md_pages, extent_page_id)); 2074 ctx->next_extent_page = i + 1; 2075 rc = blob_serialize_add_page(ctx->blob, &ctx->extent_page, &page_count, &ctx->extent_page); 2076 if (rc < 0) { 2077 blob_persist_complete(seq, ctx, rc); 2078 return; 2079 } 2080 2081 blob_serialize_extent_page(blob, i * SPDK_EXTENTS_PER_EP, ctx->extent_page); 2082 2083 ctx->extent_page->crc = blob_md_page_calc_crc(ctx->extent_page); 2084 2085 bs_sequence_write_dev(seq, ctx->extent_page, bs_md_page_to_lba(blob->bs, extent_page_id), 2086 bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE), 2087 blob_persist_write_extent_pages, ctx); 2088 return; 2089 } 2090 assert(blob->clean.extent_pages[i] != 0); 2091 } 2092 2093 blob_persist_generate_new_md(ctx); 2094 } 2095 2096 static void 2097 blob_persist_start(struct spdk_blob_persist_ctx *ctx) 2098 { 2099 spdk_bs_sequence_t *seq = ctx->seq; 2100 struct spdk_blob *blob = ctx->blob; 2101 2102 if (blob->active.num_pages == 0) { 2103 /* This is the signal that the blob should be deleted. 2104 * Immediately jump to the clean up routine. */ 2105 assert(blob->clean.num_pages > 0); 2106 blob->state = SPDK_BLOB_STATE_CLEAN; 2107 blob_persist_zero_pages(seq, ctx, 0); 2108 return; 2109 2110 } 2111 2112 blob_persist_write_extent_pages(seq, ctx, 0); 2113 } 2114 2115 static void 2116 blob_persist_dirty_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2117 { 2118 struct spdk_blob_persist_ctx *ctx = cb_arg; 2119 2120 spdk_free(ctx->super); 2121 2122 if (bserrno != 0) { 2123 blob_persist_complete(seq, ctx, bserrno); 2124 return; 2125 } 2126 2127 ctx->blob->bs->clean = 0; 2128 2129 blob_persist_start(ctx); 2130 } 2131 2132 static void 2133 bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 2134 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg); 2135 2136 2137 static void 2138 blob_persist_dirty(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2139 { 2140 struct spdk_blob_persist_ctx *ctx = cb_arg; 2141 2142 if (bserrno != 0) { 2143 spdk_free(ctx->super); 2144 blob_persist_complete(seq, ctx, bserrno); 2145 return; 2146 } 2147 2148 ctx->super->clean = 0; 2149 if (ctx->super->size == 0) { 2150 ctx->super->size = ctx->blob->bs->dev->blockcnt * ctx->blob->bs->dev->blocklen; 2151 } 2152 2153 bs_write_super(seq, ctx->blob->bs, ctx->super, blob_persist_dirty_cpl, ctx); 2154 } 2155 2156 static void 2157 blob_persist_check_dirty(struct spdk_blob_persist_ctx *ctx) 2158 { 2159 if (ctx->blob->bs->clean) { 2160 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 2161 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 2162 if (!ctx->super) { 2163 blob_persist_complete(ctx->seq, ctx, -ENOMEM); 2164 return; 2165 } 2166 2167 bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(ctx->blob->bs, 0), 2168 bs_byte_to_lba(ctx->blob->bs, sizeof(*ctx->super)), 2169 blob_persist_dirty, ctx); 2170 } else { 2171 blob_persist_start(ctx); 2172 } 2173 } 2174 2175 /* Write a blob to disk */ 2176 static void 2177 blob_persist(spdk_bs_sequence_t *seq, struct spdk_blob *blob, 2178 spdk_bs_sequence_cpl cb_fn, void *cb_arg) 2179 { 2180 struct spdk_blob_persist_ctx *ctx; 2181 2182 blob_verify_md_op(blob); 2183 2184 if (blob->state == SPDK_BLOB_STATE_CLEAN && TAILQ_EMPTY(&blob->pending_persists)) { 2185 cb_fn(seq, cb_arg, 0); 2186 return; 2187 } 2188 2189 ctx = calloc(1, sizeof(*ctx)); 2190 if (!ctx) { 2191 cb_fn(seq, cb_arg, -ENOMEM); 2192 return; 2193 } 2194 ctx->blob = blob; 2195 ctx->seq = seq; 2196 ctx->cb_fn = cb_fn; 2197 ctx->cb_arg = cb_arg; 2198 ctx->next_extent_page = 0; 2199 2200 /* Multiple blob persists can affect one another, via blob->state or 2201 * blob mutable data changes. To prevent it, queue up the persists. */ 2202 if (!TAILQ_EMPTY(&blob->pending_persists)) { 2203 TAILQ_INSERT_TAIL(&blob->pending_persists, ctx, link); 2204 return; 2205 } 2206 TAILQ_INSERT_HEAD(&blob->pending_persists, ctx, link); 2207 2208 blob_persist_check_dirty(ctx); 2209 } 2210 2211 struct spdk_blob_copy_cluster_ctx { 2212 struct spdk_blob *blob; 2213 uint8_t *buf; 2214 uint64_t page; 2215 uint64_t new_cluster; 2216 uint32_t new_extent_page; 2217 spdk_bs_sequence_t *seq; 2218 }; 2219 2220 static void 2221 blob_allocate_and_copy_cluster_cpl(void *cb_arg, int bserrno) 2222 { 2223 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2224 struct spdk_bs_request_set *set = (struct spdk_bs_request_set *)ctx->seq; 2225 TAILQ_HEAD(, spdk_bs_request_set) requests; 2226 spdk_bs_user_op_t *op; 2227 2228 TAILQ_INIT(&requests); 2229 TAILQ_SWAP(&set->channel->need_cluster_alloc, &requests, spdk_bs_request_set, link); 2230 2231 while (!TAILQ_EMPTY(&requests)) { 2232 op = TAILQ_FIRST(&requests); 2233 TAILQ_REMOVE(&requests, op, link); 2234 if (bserrno == 0) { 2235 bs_user_op_execute(op); 2236 } else { 2237 bs_user_op_abort(op); 2238 } 2239 } 2240 2241 spdk_free(ctx->buf); 2242 free(ctx); 2243 } 2244 2245 static void 2246 blob_insert_cluster_cpl(void *cb_arg, int bserrno) 2247 { 2248 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2249 2250 if (bserrno) { 2251 if (bserrno == -EEXIST) { 2252 /* The metadata insert failed because another thread 2253 * allocated the cluster first. Free our cluster 2254 * but continue without error. */ 2255 bserrno = 0; 2256 } 2257 pthread_mutex_lock(&ctx->blob->bs->used_clusters_mutex); 2258 bs_release_cluster(ctx->blob->bs, ctx->new_cluster); 2259 pthread_mutex_unlock(&ctx->blob->bs->used_clusters_mutex); 2260 if (ctx->new_extent_page != 0) { 2261 bs_release_md_page(ctx->blob->bs, ctx->new_extent_page); 2262 } 2263 } 2264 2265 bs_sequence_finish(ctx->seq, bserrno); 2266 } 2267 2268 static void 2269 blob_write_copy_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2270 { 2271 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2272 uint32_t cluster_number; 2273 2274 if (bserrno) { 2275 /* The write failed, so jump to the final completion handler */ 2276 bs_sequence_finish(seq, bserrno); 2277 return; 2278 } 2279 2280 cluster_number = bs_page_to_cluster(ctx->blob->bs, ctx->page); 2281 2282 blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 2283 ctx->new_extent_page, blob_insert_cluster_cpl, ctx); 2284 } 2285 2286 static void 2287 blob_write_copy(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2288 { 2289 struct spdk_blob_copy_cluster_ctx *ctx = cb_arg; 2290 2291 if (bserrno != 0) { 2292 /* The read failed, so jump to the final completion handler */ 2293 bs_sequence_finish(seq, bserrno); 2294 return; 2295 } 2296 2297 /* Write whole cluster */ 2298 bs_sequence_write_dev(seq, ctx->buf, 2299 bs_cluster_to_lba(ctx->blob->bs, ctx->new_cluster), 2300 bs_cluster_to_lba(ctx->blob->bs, 1), 2301 blob_write_copy_cpl, ctx); 2302 } 2303 2304 static void 2305 bs_allocate_and_copy_cluster(struct spdk_blob *blob, 2306 struct spdk_io_channel *_ch, 2307 uint64_t io_unit, spdk_bs_user_op_t *op) 2308 { 2309 struct spdk_bs_cpl cpl; 2310 struct spdk_bs_channel *ch; 2311 struct spdk_blob_copy_cluster_ctx *ctx; 2312 uint32_t cluster_start_page; 2313 uint32_t cluster_number; 2314 int rc; 2315 2316 ch = spdk_io_channel_get_ctx(_ch); 2317 2318 if (!TAILQ_EMPTY(&ch->need_cluster_alloc)) { 2319 /* There are already operations pending. Queue this user op 2320 * and return because it will be re-executed when the outstanding 2321 * cluster allocation completes. */ 2322 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 2323 return; 2324 } 2325 2326 /* Round the io_unit offset down to the first page in the cluster */ 2327 cluster_start_page = bs_io_unit_to_cluster_start(blob, io_unit); 2328 2329 /* Calculate which index in the metadata cluster array the corresponding 2330 * cluster is supposed to be at. */ 2331 cluster_number = bs_io_unit_to_cluster_number(blob, io_unit); 2332 2333 ctx = calloc(1, sizeof(*ctx)); 2334 if (!ctx) { 2335 bs_user_op_abort(op); 2336 return; 2337 } 2338 2339 assert(blob->bs->cluster_sz % blob->back_bs_dev->blocklen == 0); 2340 2341 ctx->blob = blob; 2342 ctx->page = cluster_start_page; 2343 2344 if (blob->parent_id != SPDK_BLOBID_INVALID) { 2345 ctx->buf = spdk_malloc(blob->bs->cluster_sz, blob->back_bs_dev->blocklen, 2346 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 2347 if (!ctx->buf) { 2348 SPDK_ERRLOG("DMA allocation for cluster of size = %" PRIu32 " failed.\n", 2349 blob->bs->cluster_sz); 2350 free(ctx); 2351 bs_user_op_abort(op); 2352 return; 2353 } 2354 } 2355 2356 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 2357 rc = bs_allocate_cluster(blob, cluster_number, &ctx->new_cluster, &ctx->new_extent_page, 2358 false); 2359 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 2360 if (rc != 0) { 2361 spdk_free(ctx->buf); 2362 free(ctx); 2363 bs_user_op_abort(op); 2364 return; 2365 } 2366 2367 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2368 cpl.u.blob_basic.cb_fn = blob_allocate_and_copy_cluster_cpl; 2369 cpl.u.blob_basic.cb_arg = ctx; 2370 2371 ctx->seq = bs_sequence_start(_ch, &cpl); 2372 if (!ctx->seq) { 2373 pthread_mutex_lock(&blob->bs->used_clusters_mutex); 2374 bs_release_cluster(blob->bs, ctx->new_cluster); 2375 pthread_mutex_unlock(&blob->bs->used_clusters_mutex); 2376 spdk_free(ctx->buf); 2377 free(ctx); 2378 bs_user_op_abort(op); 2379 return; 2380 } 2381 2382 /* Queue the user op to block other incoming operations */ 2383 TAILQ_INSERT_TAIL(&ch->need_cluster_alloc, op, link); 2384 2385 if (blob->parent_id != SPDK_BLOBID_INVALID) { 2386 /* Read cluster from backing device */ 2387 bs_sequence_read_bs_dev(ctx->seq, blob->back_bs_dev, ctx->buf, 2388 bs_dev_page_to_lba(blob->back_bs_dev, cluster_start_page), 2389 bs_dev_byte_to_lba(blob->back_bs_dev, blob->bs->cluster_sz), 2390 blob_write_copy, ctx); 2391 } else { 2392 blob_insert_cluster_on_md_thread(ctx->blob, cluster_number, ctx->new_cluster, 2393 ctx->new_extent_page, blob_insert_cluster_cpl, ctx); 2394 } 2395 } 2396 2397 static inline bool 2398 blob_calculate_lba_and_lba_count(struct spdk_blob *blob, uint64_t io_unit, uint64_t length, 2399 uint64_t *lba, uint32_t *lba_count) 2400 { 2401 *lba_count = length; 2402 2403 if (!bs_io_unit_is_allocated(blob, io_unit)) { 2404 assert(blob->back_bs_dev != NULL); 2405 *lba = bs_io_unit_to_back_dev_lba(blob, io_unit); 2406 *lba_count = bs_io_unit_to_back_dev_lba(blob, *lba_count); 2407 return false; 2408 } else { 2409 *lba = bs_blob_io_unit_to_lba(blob, io_unit); 2410 return true; 2411 } 2412 } 2413 2414 struct op_split_ctx { 2415 struct spdk_blob *blob; 2416 struct spdk_io_channel *channel; 2417 uint64_t io_unit_offset; 2418 uint64_t io_units_remaining; 2419 void *curr_payload; 2420 enum spdk_blob_op_type op_type; 2421 spdk_bs_sequence_t *seq; 2422 }; 2423 2424 static void 2425 blob_request_submit_op_split_next(void *cb_arg, int bserrno) 2426 { 2427 struct op_split_ctx *ctx = cb_arg; 2428 struct spdk_blob *blob = ctx->blob; 2429 struct spdk_io_channel *ch = ctx->channel; 2430 enum spdk_blob_op_type op_type = ctx->op_type; 2431 uint8_t *buf = ctx->curr_payload; 2432 uint64_t offset = ctx->io_unit_offset; 2433 uint64_t length = ctx->io_units_remaining; 2434 uint64_t op_length; 2435 2436 if (bserrno != 0 || ctx->io_units_remaining == 0) { 2437 bs_sequence_finish(ctx->seq, bserrno); 2438 free(ctx); 2439 return; 2440 } 2441 2442 op_length = spdk_min(length, bs_num_io_units_to_cluster_boundary(blob, 2443 offset)); 2444 2445 /* Update length and payload for next operation */ 2446 ctx->io_units_remaining -= op_length; 2447 ctx->io_unit_offset += op_length; 2448 if (op_type == SPDK_BLOB_WRITE || op_type == SPDK_BLOB_READ) { 2449 ctx->curr_payload += op_length * blob->bs->io_unit_size; 2450 } 2451 2452 switch (op_type) { 2453 case SPDK_BLOB_READ: 2454 spdk_blob_io_read(blob, ch, buf, offset, op_length, 2455 blob_request_submit_op_split_next, ctx); 2456 break; 2457 case SPDK_BLOB_WRITE: 2458 spdk_blob_io_write(blob, ch, buf, offset, op_length, 2459 blob_request_submit_op_split_next, ctx); 2460 break; 2461 case SPDK_BLOB_UNMAP: 2462 spdk_blob_io_unmap(blob, ch, offset, op_length, 2463 blob_request_submit_op_split_next, ctx); 2464 break; 2465 case SPDK_BLOB_WRITE_ZEROES: 2466 spdk_blob_io_write_zeroes(blob, ch, offset, op_length, 2467 blob_request_submit_op_split_next, ctx); 2468 break; 2469 case SPDK_BLOB_READV: 2470 case SPDK_BLOB_WRITEV: 2471 SPDK_ERRLOG("readv/write not valid\n"); 2472 bs_sequence_finish(ctx->seq, -EINVAL); 2473 free(ctx); 2474 break; 2475 } 2476 } 2477 2478 static void 2479 blob_request_submit_op_split(struct spdk_io_channel *ch, struct spdk_blob *blob, 2480 void *payload, uint64_t offset, uint64_t length, 2481 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 2482 { 2483 struct op_split_ctx *ctx; 2484 spdk_bs_sequence_t *seq; 2485 struct spdk_bs_cpl cpl; 2486 2487 assert(blob != NULL); 2488 2489 ctx = calloc(1, sizeof(struct op_split_ctx)); 2490 if (ctx == NULL) { 2491 cb_fn(cb_arg, -ENOMEM); 2492 return; 2493 } 2494 2495 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2496 cpl.u.blob_basic.cb_fn = cb_fn; 2497 cpl.u.blob_basic.cb_arg = cb_arg; 2498 2499 seq = bs_sequence_start(ch, &cpl); 2500 if (!seq) { 2501 free(ctx); 2502 cb_fn(cb_arg, -ENOMEM); 2503 return; 2504 } 2505 2506 ctx->blob = blob; 2507 ctx->channel = ch; 2508 ctx->curr_payload = payload; 2509 ctx->io_unit_offset = offset; 2510 ctx->io_units_remaining = length; 2511 ctx->op_type = op_type; 2512 ctx->seq = seq; 2513 2514 blob_request_submit_op_split_next(ctx, 0); 2515 } 2516 2517 static void 2518 blob_request_submit_op_single(struct spdk_io_channel *_ch, struct spdk_blob *blob, 2519 void *payload, uint64_t offset, uint64_t length, 2520 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 2521 { 2522 struct spdk_bs_cpl cpl; 2523 uint64_t lba; 2524 uint32_t lba_count; 2525 bool is_allocated; 2526 2527 assert(blob != NULL); 2528 2529 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2530 cpl.u.blob_basic.cb_fn = cb_fn; 2531 cpl.u.blob_basic.cb_arg = cb_arg; 2532 2533 is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 2534 2535 if (blob->frozen_refcnt) { 2536 /* This blob I/O is frozen */ 2537 spdk_bs_user_op_t *op; 2538 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_ch); 2539 2540 op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 2541 if (!op) { 2542 cb_fn(cb_arg, -ENOMEM); 2543 return; 2544 } 2545 2546 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 2547 2548 return; 2549 } 2550 2551 switch (op_type) { 2552 case SPDK_BLOB_READ: { 2553 spdk_bs_batch_t *batch; 2554 2555 batch = bs_batch_open(_ch, &cpl); 2556 if (!batch) { 2557 cb_fn(cb_arg, -ENOMEM); 2558 return; 2559 } 2560 2561 if (is_allocated) { 2562 /* Read from the blob */ 2563 bs_batch_read_dev(batch, payload, lba, lba_count); 2564 } else { 2565 /* Read from the backing block device */ 2566 bs_batch_read_bs_dev(batch, blob->back_bs_dev, payload, lba, lba_count); 2567 } 2568 2569 bs_batch_close(batch); 2570 break; 2571 } 2572 case SPDK_BLOB_WRITE: 2573 case SPDK_BLOB_WRITE_ZEROES: { 2574 if (is_allocated) { 2575 /* Write to the blob */ 2576 spdk_bs_batch_t *batch; 2577 2578 if (lba_count == 0) { 2579 cb_fn(cb_arg, 0); 2580 return; 2581 } 2582 2583 batch = bs_batch_open(_ch, &cpl); 2584 if (!batch) { 2585 cb_fn(cb_arg, -ENOMEM); 2586 return; 2587 } 2588 2589 if (op_type == SPDK_BLOB_WRITE) { 2590 bs_batch_write_dev(batch, payload, lba, lba_count); 2591 } else { 2592 bs_batch_write_zeroes_dev(batch, lba, lba_count); 2593 } 2594 2595 bs_batch_close(batch); 2596 } else { 2597 /* Queue this operation and allocate the cluster */ 2598 spdk_bs_user_op_t *op; 2599 2600 op = bs_user_op_alloc(_ch, &cpl, op_type, blob, payload, 0, offset, length); 2601 if (!op) { 2602 cb_fn(cb_arg, -ENOMEM); 2603 return; 2604 } 2605 2606 bs_allocate_and_copy_cluster(blob, _ch, offset, op); 2607 } 2608 break; 2609 } 2610 case SPDK_BLOB_UNMAP: { 2611 spdk_bs_batch_t *batch; 2612 2613 batch = bs_batch_open(_ch, &cpl); 2614 if (!batch) { 2615 cb_fn(cb_arg, -ENOMEM); 2616 return; 2617 } 2618 2619 if (is_allocated) { 2620 bs_batch_unmap_dev(batch, lba, lba_count); 2621 } 2622 2623 bs_batch_close(batch); 2624 break; 2625 } 2626 case SPDK_BLOB_READV: 2627 case SPDK_BLOB_WRITEV: 2628 SPDK_ERRLOG("readv/write not valid\n"); 2629 cb_fn(cb_arg, -EINVAL); 2630 break; 2631 } 2632 } 2633 2634 static void 2635 blob_request_submit_op(struct spdk_blob *blob, struct spdk_io_channel *_channel, 2636 void *payload, uint64_t offset, uint64_t length, 2637 spdk_blob_op_complete cb_fn, void *cb_arg, enum spdk_blob_op_type op_type) 2638 { 2639 assert(blob != NULL); 2640 2641 if (blob->data_ro && op_type != SPDK_BLOB_READ) { 2642 cb_fn(cb_arg, -EPERM); 2643 return; 2644 } 2645 2646 if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) { 2647 cb_fn(cb_arg, -EINVAL); 2648 return; 2649 } 2650 if (length <= bs_num_io_units_to_cluster_boundary(blob, offset)) { 2651 blob_request_submit_op_single(_channel, blob, payload, offset, length, 2652 cb_fn, cb_arg, op_type); 2653 } else { 2654 blob_request_submit_op_split(_channel, blob, payload, offset, length, 2655 cb_fn, cb_arg, op_type); 2656 } 2657 } 2658 2659 struct rw_iov_ctx { 2660 struct spdk_blob *blob; 2661 struct spdk_io_channel *channel; 2662 spdk_blob_op_complete cb_fn; 2663 void *cb_arg; 2664 bool read; 2665 int iovcnt; 2666 struct iovec *orig_iov; 2667 uint64_t io_unit_offset; 2668 uint64_t io_units_remaining; 2669 uint64_t io_units_done; 2670 struct iovec iov[0]; 2671 }; 2672 2673 static void 2674 rw_iov_done(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 2675 { 2676 assert(cb_arg == NULL); 2677 bs_sequence_finish(seq, bserrno); 2678 } 2679 2680 static void 2681 rw_iov_split_next(void *cb_arg, int bserrno) 2682 { 2683 struct rw_iov_ctx *ctx = cb_arg; 2684 struct spdk_blob *blob = ctx->blob; 2685 struct iovec *iov, *orig_iov; 2686 int iovcnt; 2687 size_t orig_iovoff; 2688 uint64_t io_units_count, io_units_to_boundary, io_unit_offset; 2689 uint64_t byte_count; 2690 2691 if (bserrno != 0 || ctx->io_units_remaining == 0) { 2692 ctx->cb_fn(ctx->cb_arg, bserrno); 2693 free(ctx); 2694 return; 2695 } 2696 2697 io_unit_offset = ctx->io_unit_offset; 2698 io_units_to_boundary = bs_num_io_units_to_cluster_boundary(blob, io_unit_offset); 2699 io_units_count = spdk_min(ctx->io_units_remaining, io_units_to_boundary); 2700 /* 2701 * Get index and offset into the original iov array for our current position in the I/O sequence. 2702 * byte_count will keep track of how many bytes remaining until orig_iov and orig_iovoff will 2703 * point to the current position in the I/O sequence. 2704 */ 2705 byte_count = ctx->io_units_done * blob->bs->io_unit_size; 2706 orig_iov = &ctx->orig_iov[0]; 2707 orig_iovoff = 0; 2708 while (byte_count > 0) { 2709 if (byte_count >= orig_iov->iov_len) { 2710 byte_count -= orig_iov->iov_len; 2711 orig_iov++; 2712 } else { 2713 orig_iovoff = byte_count; 2714 byte_count = 0; 2715 } 2716 } 2717 2718 /* 2719 * Build an iov array for the next I/O in the sequence. byte_count will keep track of how many 2720 * bytes of this next I/O remain to be accounted for in the new iov array. 2721 */ 2722 byte_count = io_units_count * blob->bs->io_unit_size; 2723 iov = &ctx->iov[0]; 2724 iovcnt = 0; 2725 while (byte_count > 0) { 2726 assert(iovcnt < ctx->iovcnt); 2727 iov->iov_len = spdk_min(byte_count, orig_iov->iov_len - orig_iovoff); 2728 iov->iov_base = orig_iov->iov_base + orig_iovoff; 2729 byte_count -= iov->iov_len; 2730 orig_iovoff = 0; 2731 orig_iov++; 2732 iov++; 2733 iovcnt++; 2734 } 2735 2736 ctx->io_unit_offset += io_units_count; 2737 ctx->io_units_remaining -= io_units_count; 2738 ctx->io_units_done += io_units_count; 2739 iov = &ctx->iov[0]; 2740 2741 if (ctx->read) { 2742 spdk_blob_io_readv(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset, 2743 io_units_count, rw_iov_split_next, ctx); 2744 } else { 2745 spdk_blob_io_writev(ctx->blob, ctx->channel, iov, iovcnt, io_unit_offset, 2746 io_units_count, rw_iov_split_next, ctx); 2747 } 2748 } 2749 2750 static void 2751 blob_request_submit_rw_iov(struct spdk_blob *blob, struct spdk_io_channel *_channel, 2752 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 2753 spdk_blob_op_complete cb_fn, void *cb_arg, bool read) 2754 { 2755 struct spdk_bs_cpl cpl; 2756 2757 assert(blob != NULL); 2758 2759 if (!read && blob->data_ro) { 2760 cb_fn(cb_arg, -EPERM); 2761 return; 2762 } 2763 2764 if (length == 0) { 2765 cb_fn(cb_arg, 0); 2766 return; 2767 } 2768 2769 if (offset + length > bs_cluster_to_lba(blob->bs, blob->active.num_clusters)) { 2770 cb_fn(cb_arg, -EINVAL); 2771 return; 2772 } 2773 2774 /* 2775 * For now, we implement readv/writev using a sequence (instead of a batch) to account for having 2776 * to split a request that spans a cluster boundary. For I/O that do not span a cluster boundary, 2777 * there will be no noticeable difference compared to using a batch. For I/O that do span a cluster 2778 * boundary, the target LBAs (after blob offset to LBA translation) may not be contiguous, so we need 2779 * to allocate a separate iov array and split the I/O such that none of the resulting 2780 * smaller I/O cross a cluster boundary. These smaller I/O will be issued in sequence (not in parallel) 2781 * but since this case happens very infrequently, any performance impact will be negligible. 2782 * 2783 * This could be optimized in the future to allocate a big enough iov array to account for all of the iovs 2784 * for all of the smaller I/Os, pre-build all of the iov arrays for the smaller I/Os, then issue them 2785 * in a batch. That would also require creating an intermediate spdk_bs_cpl that would get called 2786 * when the batch was completed, to allow for freeing the memory for the iov arrays. 2787 */ 2788 if (spdk_likely(length <= bs_num_io_units_to_cluster_boundary(blob, offset))) { 2789 uint32_t lba_count; 2790 uint64_t lba; 2791 bool is_allocated; 2792 2793 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 2794 cpl.u.blob_basic.cb_fn = cb_fn; 2795 cpl.u.blob_basic.cb_arg = cb_arg; 2796 2797 if (blob->frozen_refcnt) { 2798 /* This blob I/O is frozen */ 2799 enum spdk_blob_op_type op_type; 2800 spdk_bs_user_op_t *op; 2801 struct spdk_bs_channel *bs_channel = spdk_io_channel_get_ctx(_channel); 2802 2803 op_type = read ? SPDK_BLOB_READV : SPDK_BLOB_WRITEV; 2804 op = bs_user_op_alloc(_channel, &cpl, op_type, blob, iov, iovcnt, offset, length); 2805 if (!op) { 2806 cb_fn(cb_arg, -ENOMEM); 2807 return; 2808 } 2809 2810 TAILQ_INSERT_TAIL(&bs_channel->queued_io, op, link); 2811 2812 return; 2813 } 2814 2815 is_allocated = blob_calculate_lba_and_lba_count(blob, offset, length, &lba, &lba_count); 2816 2817 if (read) { 2818 spdk_bs_sequence_t *seq; 2819 2820 seq = bs_sequence_start(_channel, &cpl); 2821 if (!seq) { 2822 cb_fn(cb_arg, -ENOMEM); 2823 return; 2824 } 2825 2826 if (is_allocated) { 2827 bs_sequence_readv_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL); 2828 } else { 2829 bs_sequence_readv_bs_dev(seq, blob->back_bs_dev, iov, iovcnt, lba, lba_count, 2830 rw_iov_done, NULL); 2831 } 2832 } else { 2833 if (is_allocated) { 2834 spdk_bs_sequence_t *seq; 2835 2836 seq = bs_sequence_start(_channel, &cpl); 2837 if (!seq) { 2838 cb_fn(cb_arg, -ENOMEM); 2839 return; 2840 } 2841 2842 bs_sequence_writev_dev(seq, iov, iovcnt, lba, lba_count, rw_iov_done, NULL); 2843 } else { 2844 /* Queue this operation and allocate the cluster */ 2845 spdk_bs_user_op_t *op; 2846 2847 op = bs_user_op_alloc(_channel, &cpl, SPDK_BLOB_WRITEV, blob, iov, iovcnt, offset, 2848 length); 2849 if (!op) { 2850 cb_fn(cb_arg, -ENOMEM); 2851 return; 2852 } 2853 2854 bs_allocate_and_copy_cluster(blob, _channel, offset, op); 2855 } 2856 } 2857 } else { 2858 struct rw_iov_ctx *ctx; 2859 2860 ctx = calloc(1, sizeof(struct rw_iov_ctx) + iovcnt * sizeof(struct iovec)); 2861 if (ctx == NULL) { 2862 cb_fn(cb_arg, -ENOMEM); 2863 return; 2864 } 2865 2866 ctx->blob = blob; 2867 ctx->channel = _channel; 2868 ctx->cb_fn = cb_fn; 2869 ctx->cb_arg = cb_arg; 2870 ctx->read = read; 2871 ctx->orig_iov = iov; 2872 ctx->iovcnt = iovcnt; 2873 ctx->io_unit_offset = offset; 2874 ctx->io_units_remaining = length; 2875 ctx->io_units_done = 0; 2876 2877 rw_iov_split_next(ctx, 0); 2878 } 2879 } 2880 2881 static struct spdk_blob * 2882 blob_lookup(struct spdk_blob_store *bs, spdk_blob_id blobid) 2883 { 2884 struct spdk_blob *blob; 2885 2886 if (spdk_bit_array_get(bs->open_blobids, blobid) == 0) { 2887 return NULL; 2888 } 2889 2890 TAILQ_FOREACH(blob, &bs->blobs, link) { 2891 if (blob->id == blobid) { 2892 return blob; 2893 } 2894 } 2895 2896 return NULL; 2897 } 2898 2899 static void 2900 blob_get_snapshot_and_clone_entries(struct spdk_blob *blob, 2901 struct spdk_blob_list **snapshot_entry, struct spdk_blob_list **clone_entry) 2902 { 2903 assert(blob != NULL); 2904 *snapshot_entry = NULL; 2905 *clone_entry = NULL; 2906 2907 if (blob->parent_id == SPDK_BLOBID_INVALID) { 2908 return; 2909 } 2910 2911 TAILQ_FOREACH(*snapshot_entry, &blob->bs->snapshots, link) { 2912 if ((*snapshot_entry)->id == blob->parent_id) { 2913 break; 2914 } 2915 } 2916 2917 if (*snapshot_entry != NULL) { 2918 TAILQ_FOREACH(*clone_entry, &(*snapshot_entry)->clones, link) { 2919 if ((*clone_entry)->id == blob->id) { 2920 break; 2921 } 2922 } 2923 2924 assert(clone_entry != NULL); 2925 } 2926 } 2927 2928 static int 2929 bs_channel_create(void *io_device, void *ctx_buf) 2930 { 2931 struct spdk_blob_store *bs = io_device; 2932 struct spdk_bs_channel *channel = ctx_buf; 2933 struct spdk_bs_dev *dev; 2934 uint32_t max_ops = bs->max_channel_ops; 2935 uint32_t i; 2936 2937 dev = bs->dev; 2938 2939 channel->req_mem = calloc(max_ops, sizeof(struct spdk_bs_request_set)); 2940 if (!channel->req_mem) { 2941 return -1; 2942 } 2943 2944 TAILQ_INIT(&channel->reqs); 2945 2946 for (i = 0; i < max_ops; i++) { 2947 TAILQ_INSERT_TAIL(&channel->reqs, &channel->req_mem[i], link); 2948 } 2949 2950 channel->bs = bs; 2951 channel->dev = dev; 2952 channel->dev_channel = dev->create_channel(dev); 2953 2954 if (!channel->dev_channel) { 2955 SPDK_ERRLOG("Failed to create device channel.\n"); 2956 free(channel->req_mem); 2957 return -1; 2958 } 2959 2960 TAILQ_INIT(&channel->need_cluster_alloc); 2961 TAILQ_INIT(&channel->queued_io); 2962 2963 return 0; 2964 } 2965 2966 static void 2967 bs_channel_destroy(void *io_device, void *ctx_buf) 2968 { 2969 struct spdk_bs_channel *channel = ctx_buf; 2970 spdk_bs_user_op_t *op; 2971 2972 while (!TAILQ_EMPTY(&channel->need_cluster_alloc)) { 2973 op = TAILQ_FIRST(&channel->need_cluster_alloc); 2974 TAILQ_REMOVE(&channel->need_cluster_alloc, op, link); 2975 bs_user_op_abort(op); 2976 } 2977 2978 while (!TAILQ_EMPTY(&channel->queued_io)) { 2979 op = TAILQ_FIRST(&channel->queued_io); 2980 TAILQ_REMOVE(&channel->queued_io, op, link); 2981 bs_user_op_abort(op); 2982 } 2983 2984 free(channel->req_mem); 2985 channel->dev->destroy_channel(channel->dev, channel->dev_channel); 2986 } 2987 2988 static void 2989 bs_dev_destroy(void *io_device) 2990 { 2991 struct spdk_blob_store *bs = io_device; 2992 struct spdk_blob *blob, *blob_tmp; 2993 2994 bs->dev->destroy(bs->dev); 2995 2996 TAILQ_FOREACH_SAFE(blob, &bs->blobs, link, blob_tmp) { 2997 TAILQ_REMOVE(&bs->blobs, blob, link); 2998 spdk_bit_array_clear(bs->open_blobids, blob->id); 2999 blob_free(blob); 3000 } 3001 3002 pthread_mutex_destroy(&bs->used_clusters_mutex); 3003 3004 spdk_bit_array_free(&bs->open_blobids); 3005 spdk_bit_array_free(&bs->used_blobids); 3006 spdk_bit_array_free(&bs->used_md_pages); 3007 spdk_bit_pool_free(&bs->used_clusters); 3008 /* 3009 * If this function is called for any reason except a successful unload, 3010 * the unload_cpl type will be NONE and this will be a nop. 3011 */ 3012 bs_call_cpl(&bs->unload_cpl, bs->unload_err); 3013 3014 free(bs); 3015 } 3016 3017 static int 3018 bs_blob_list_add(struct spdk_blob *blob) 3019 { 3020 spdk_blob_id snapshot_id; 3021 struct spdk_blob_list *snapshot_entry = NULL; 3022 struct spdk_blob_list *clone_entry = NULL; 3023 3024 assert(blob != NULL); 3025 3026 snapshot_id = blob->parent_id; 3027 if (snapshot_id == SPDK_BLOBID_INVALID) { 3028 return 0; 3029 } 3030 3031 snapshot_entry = bs_get_snapshot_entry(blob->bs, snapshot_id); 3032 if (snapshot_entry == NULL) { 3033 /* Snapshot not found */ 3034 snapshot_entry = calloc(1, sizeof(struct spdk_blob_list)); 3035 if (snapshot_entry == NULL) { 3036 return -ENOMEM; 3037 } 3038 snapshot_entry->id = snapshot_id; 3039 TAILQ_INIT(&snapshot_entry->clones); 3040 TAILQ_INSERT_TAIL(&blob->bs->snapshots, snapshot_entry, link); 3041 } else { 3042 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 3043 if (clone_entry->id == blob->id) { 3044 break; 3045 } 3046 } 3047 } 3048 3049 if (clone_entry == NULL) { 3050 /* Clone not found */ 3051 clone_entry = calloc(1, sizeof(struct spdk_blob_list)); 3052 if (clone_entry == NULL) { 3053 return -ENOMEM; 3054 } 3055 clone_entry->id = blob->id; 3056 TAILQ_INIT(&clone_entry->clones); 3057 TAILQ_INSERT_TAIL(&snapshot_entry->clones, clone_entry, link); 3058 snapshot_entry->clone_count++; 3059 } 3060 3061 return 0; 3062 } 3063 3064 static void 3065 bs_blob_list_remove(struct spdk_blob *blob) 3066 { 3067 struct spdk_blob_list *snapshot_entry = NULL; 3068 struct spdk_blob_list *clone_entry = NULL; 3069 3070 blob_get_snapshot_and_clone_entries(blob, &snapshot_entry, &clone_entry); 3071 3072 if (snapshot_entry == NULL) { 3073 return; 3074 } 3075 3076 blob->parent_id = SPDK_BLOBID_INVALID; 3077 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 3078 free(clone_entry); 3079 3080 snapshot_entry->clone_count--; 3081 } 3082 3083 static int 3084 bs_blob_list_free(struct spdk_blob_store *bs) 3085 { 3086 struct spdk_blob_list *snapshot_entry; 3087 struct spdk_blob_list *snapshot_entry_tmp; 3088 struct spdk_blob_list *clone_entry; 3089 struct spdk_blob_list *clone_entry_tmp; 3090 3091 TAILQ_FOREACH_SAFE(snapshot_entry, &bs->snapshots, link, snapshot_entry_tmp) { 3092 TAILQ_FOREACH_SAFE(clone_entry, &snapshot_entry->clones, link, clone_entry_tmp) { 3093 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 3094 free(clone_entry); 3095 } 3096 TAILQ_REMOVE(&bs->snapshots, snapshot_entry, link); 3097 free(snapshot_entry); 3098 } 3099 3100 return 0; 3101 } 3102 3103 static void 3104 bs_free(struct spdk_blob_store *bs) 3105 { 3106 bs_blob_list_free(bs); 3107 3108 bs_unregister_md_thread(bs); 3109 spdk_io_device_unregister(bs, bs_dev_destroy); 3110 } 3111 3112 void 3113 spdk_bs_opts_init(struct spdk_bs_opts *opts) 3114 { 3115 opts->cluster_sz = SPDK_BLOB_OPTS_CLUSTER_SZ; 3116 opts->num_md_pages = SPDK_BLOB_OPTS_NUM_MD_PAGES; 3117 opts->max_md_ops = SPDK_BLOB_OPTS_MAX_MD_OPS; 3118 opts->max_channel_ops = SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS; 3119 opts->clear_method = BS_CLEAR_WITH_UNMAP; 3120 memset(&opts->bstype, 0, sizeof(opts->bstype)); 3121 opts->iter_cb_fn = NULL; 3122 opts->iter_cb_arg = NULL; 3123 } 3124 3125 static int 3126 bs_opts_verify(struct spdk_bs_opts *opts) 3127 { 3128 if (opts->cluster_sz == 0 || opts->num_md_pages == 0 || opts->max_md_ops == 0 || 3129 opts->max_channel_ops == 0) { 3130 SPDK_ERRLOG("Blobstore options cannot be set to 0\n"); 3131 return -1; 3132 } 3133 3134 return 0; 3135 } 3136 3137 /* START spdk_bs_load */ 3138 3139 /* spdk_bs_load_ctx is used for init, load, unload and dump code paths. */ 3140 3141 struct spdk_bs_load_ctx { 3142 struct spdk_blob_store *bs; 3143 struct spdk_bs_super_block *super; 3144 3145 struct spdk_bs_md_mask *mask; 3146 bool in_page_chain; 3147 uint32_t page_index; 3148 uint32_t cur_page; 3149 struct spdk_blob_md_page *page; 3150 3151 uint64_t num_extent_pages; 3152 uint32_t *extent_page_num; 3153 struct spdk_blob_md_page *extent_pages; 3154 struct spdk_bit_array *used_clusters; 3155 3156 spdk_bs_sequence_t *seq; 3157 spdk_blob_op_with_handle_complete iter_cb_fn; 3158 void *iter_cb_arg; 3159 struct spdk_blob *blob; 3160 spdk_blob_id blobid; 3161 3162 /* These fields are used in the spdk_bs_dump path. */ 3163 FILE *fp; 3164 spdk_bs_dump_print_xattr print_xattr_fn; 3165 char xattr_name[4096]; 3166 }; 3167 3168 static int 3169 bs_alloc(struct spdk_bs_dev *dev, struct spdk_bs_opts *opts, struct spdk_blob_store **_bs, 3170 struct spdk_bs_load_ctx **_ctx) 3171 { 3172 struct spdk_blob_store *bs; 3173 struct spdk_bs_load_ctx *ctx; 3174 uint64_t dev_size; 3175 int rc; 3176 3177 dev_size = dev->blocklen * dev->blockcnt; 3178 if (dev_size < opts->cluster_sz) { 3179 /* Device size cannot be smaller than cluster size of blobstore */ 3180 SPDK_INFOLOG(SPDK_LOG_BLOB, "Device size %" PRIu64 " is smaller than cluster size %" PRIu32 "\n", 3181 dev_size, opts->cluster_sz); 3182 return -ENOSPC; 3183 } 3184 if (opts->cluster_sz < SPDK_BS_PAGE_SIZE) { 3185 /* Cluster size cannot be smaller than page size */ 3186 SPDK_ERRLOG("Cluster size %" PRIu32 " is smaller than page size %d\n", 3187 opts->cluster_sz, SPDK_BS_PAGE_SIZE); 3188 return -EINVAL; 3189 } 3190 bs = calloc(1, sizeof(struct spdk_blob_store)); 3191 if (!bs) { 3192 return -ENOMEM; 3193 } 3194 3195 ctx = calloc(1, sizeof(struct spdk_bs_load_ctx)); 3196 if (!ctx) { 3197 free(bs); 3198 return -ENOMEM; 3199 } 3200 3201 ctx->bs = bs; 3202 ctx->iter_cb_fn = opts->iter_cb_fn; 3203 ctx->iter_cb_arg = opts->iter_cb_arg; 3204 3205 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 3206 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3207 if (!ctx->super) { 3208 free(ctx); 3209 free(bs); 3210 return -ENOMEM; 3211 } 3212 3213 TAILQ_INIT(&bs->blobs); 3214 TAILQ_INIT(&bs->snapshots); 3215 bs->dev = dev; 3216 bs->md_thread = spdk_get_thread(); 3217 assert(bs->md_thread != NULL); 3218 3219 /* 3220 * Do not use bs_lba_to_cluster() here since blockcnt may not be an 3221 * even multiple of the cluster size. 3222 */ 3223 bs->cluster_sz = opts->cluster_sz; 3224 bs->total_clusters = dev->blockcnt / (bs->cluster_sz / dev->blocklen); 3225 ctx->used_clusters = spdk_bit_array_create(bs->total_clusters); 3226 if (!ctx->used_clusters) { 3227 spdk_free(ctx->super); 3228 free(ctx); 3229 free(bs); 3230 return -ENOMEM; 3231 } 3232 3233 bs->pages_per_cluster = bs->cluster_sz / SPDK_BS_PAGE_SIZE; 3234 if (spdk_u32_is_pow2(bs->pages_per_cluster)) { 3235 bs->pages_per_cluster_shift = spdk_u32log2(bs->pages_per_cluster); 3236 } 3237 bs->num_free_clusters = bs->total_clusters; 3238 bs->io_unit_size = dev->blocklen; 3239 3240 bs->max_channel_ops = opts->max_channel_ops; 3241 bs->super_blob = SPDK_BLOBID_INVALID; 3242 memcpy(&bs->bstype, &opts->bstype, sizeof(opts->bstype)); 3243 3244 /* The metadata is assumed to be at least 1 page */ 3245 bs->used_md_pages = spdk_bit_array_create(1); 3246 bs->used_blobids = spdk_bit_array_create(0); 3247 bs->open_blobids = spdk_bit_array_create(0); 3248 3249 pthread_mutex_init(&bs->used_clusters_mutex, NULL); 3250 3251 spdk_io_device_register(bs, bs_channel_create, bs_channel_destroy, 3252 sizeof(struct spdk_bs_channel), "blobstore"); 3253 rc = bs_register_md_thread(bs); 3254 if (rc == -1) { 3255 spdk_io_device_unregister(bs, NULL); 3256 pthread_mutex_destroy(&bs->used_clusters_mutex); 3257 spdk_bit_array_free(&bs->open_blobids); 3258 spdk_bit_array_free(&bs->used_blobids); 3259 spdk_bit_array_free(&bs->used_md_pages); 3260 spdk_bit_array_free(&ctx->used_clusters); 3261 spdk_free(ctx->super); 3262 free(ctx); 3263 free(bs); 3264 /* FIXME: this is a lie but don't know how to get a proper error code here */ 3265 return -ENOMEM; 3266 } 3267 3268 *_ctx = ctx; 3269 *_bs = bs; 3270 return 0; 3271 } 3272 3273 static void 3274 bs_load_ctx_fail(struct spdk_bs_load_ctx *ctx, int bserrno) 3275 { 3276 assert(bserrno != 0); 3277 3278 spdk_free(ctx->super); 3279 bs_sequence_finish(ctx->seq, bserrno); 3280 bs_free(ctx->bs); 3281 spdk_bit_array_free(&ctx->used_clusters); 3282 free(ctx); 3283 } 3284 3285 static void 3286 bs_write_super(spdk_bs_sequence_t *seq, struct spdk_blob_store *bs, 3287 struct spdk_bs_super_block *super, spdk_bs_sequence_cpl cb_fn, void *cb_arg) 3288 { 3289 /* Update the values in the super block */ 3290 super->super_blob = bs->super_blob; 3291 memcpy(&super->bstype, &bs->bstype, sizeof(bs->bstype)); 3292 super->crc = blob_md_page_calc_crc(super); 3293 bs_sequence_write_dev(seq, super, bs_page_to_lba(bs, 0), 3294 bs_byte_to_lba(bs, sizeof(*super)), 3295 cb_fn, cb_arg); 3296 } 3297 3298 static void 3299 bs_write_used_clusters(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 3300 { 3301 struct spdk_bs_load_ctx *ctx = arg; 3302 uint64_t mask_size, lba, lba_count; 3303 3304 /* Write out the used clusters mask */ 3305 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 3306 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 3307 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3308 if (!ctx->mask) { 3309 bs_load_ctx_fail(ctx, -ENOMEM); 3310 return; 3311 } 3312 3313 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_CLUSTERS; 3314 ctx->mask->length = ctx->bs->total_clusters; 3315 /* We could get here through the normal unload path, or through dirty 3316 * shutdown recovery. For the normal unload path, we use the mask from 3317 * the bit pool. For dirty shutdown recovery, we don't have a bit pool yet - 3318 * only the bit array from the load ctx. 3319 */ 3320 if (ctx->bs->used_clusters) { 3321 assert(ctx->mask->length == spdk_bit_pool_capacity(ctx->bs->used_clusters)); 3322 spdk_bit_pool_store_mask(ctx->bs->used_clusters, ctx->mask->mask); 3323 } else { 3324 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->used_clusters)); 3325 spdk_bit_array_store_mask(ctx->used_clusters, ctx->mask->mask); 3326 } 3327 lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 3328 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 3329 bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 3330 } 3331 3332 static void 3333 bs_write_used_md(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 3334 { 3335 struct spdk_bs_load_ctx *ctx = arg; 3336 uint64_t mask_size, lba, lba_count; 3337 3338 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 3339 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 3340 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3341 if (!ctx->mask) { 3342 bs_load_ctx_fail(ctx, -ENOMEM); 3343 return; 3344 } 3345 3346 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_PAGES; 3347 ctx->mask->length = ctx->super->md_len; 3348 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_md_pages)); 3349 3350 spdk_bit_array_store_mask(ctx->bs->used_md_pages, ctx->mask->mask); 3351 lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 3352 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 3353 bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 3354 } 3355 3356 static void 3357 bs_write_used_blobids(spdk_bs_sequence_t *seq, void *arg, spdk_bs_sequence_cpl cb_fn) 3358 { 3359 struct spdk_bs_load_ctx *ctx = arg; 3360 uint64_t mask_size, lba, lba_count; 3361 3362 if (ctx->super->used_blobid_mask_len == 0) { 3363 /* 3364 * This is a pre-v3 on-disk format where the blobid mask does not get 3365 * written to disk. 3366 */ 3367 cb_fn(seq, arg, 0); 3368 return; 3369 } 3370 3371 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 3372 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 3373 SPDK_MALLOC_DMA); 3374 if (!ctx->mask) { 3375 bs_load_ctx_fail(ctx, -ENOMEM); 3376 return; 3377 } 3378 3379 ctx->mask->type = SPDK_MD_MASK_TYPE_USED_BLOBIDS; 3380 ctx->mask->length = ctx->super->md_len; 3381 assert(ctx->mask->length == spdk_bit_array_capacity(ctx->bs->used_blobids)); 3382 3383 spdk_bit_array_store_mask(ctx->bs->used_blobids, ctx->mask->mask); 3384 lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 3385 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 3386 bs_sequence_write_dev(seq, ctx->mask, lba, lba_count, cb_fn, arg); 3387 } 3388 3389 static void 3390 blob_set_thin_provision(struct spdk_blob *blob) 3391 { 3392 blob_verify_md_op(blob); 3393 blob->invalid_flags |= SPDK_BLOB_THIN_PROV; 3394 blob->state = SPDK_BLOB_STATE_DIRTY; 3395 } 3396 3397 static void 3398 blob_set_clear_method(struct spdk_blob *blob, enum blob_clear_method clear_method) 3399 { 3400 blob_verify_md_op(blob); 3401 blob->clear_method = clear_method; 3402 blob->md_ro_flags |= (clear_method << SPDK_BLOB_CLEAR_METHOD_SHIFT); 3403 blob->state = SPDK_BLOB_STATE_DIRTY; 3404 } 3405 3406 static void bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno); 3407 3408 static void 3409 bs_delete_corrupted_blob_cpl(void *cb_arg, int bserrno) 3410 { 3411 struct spdk_bs_load_ctx *ctx = cb_arg; 3412 spdk_blob_id id; 3413 int64_t page_num; 3414 3415 /* Iterate to next blob (we can't use spdk_bs_iter_next function as our 3416 * last blob has been removed */ 3417 page_num = bs_blobid_to_page(ctx->blobid); 3418 page_num++; 3419 page_num = spdk_bit_array_find_first_set(ctx->bs->used_blobids, page_num); 3420 if (page_num >= spdk_bit_array_capacity(ctx->bs->used_blobids)) { 3421 bs_load_iter(ctx, NULL, -ENOENT); 3422 return; 3423 } 3424 3425 id = bs_page_to_blobid(page_num); 3426 3427 spdk_bs_open_blob(ctx->bs, id, bs_load_iter, ctx); 3428 } 3429 3430 static void 3431 bs_delete_corrupted_close_cb(void *cb_arg, int bserrno) 3432 { 3433 struct spdk_bs_load_ctx *ctx = cb_arg; 3434 3435 if (bserrno != 0) { 3436 SPDK_ERRLOG("Failed to close corrupted blob\n"); 3437 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3438 return; 3439 } 3440 3441 spdk_bs_delete_blob(ctx->bs, ctx->blobid, bs_delete_corrupted_blob_cpl, ctx); 3442 } 3443 3444 static void 3445 bs_delete_corrupted_blob(void *cb_arg, int bserrno) 3446 { 3447 struct spdk_bs_load_ctx *ctx = cb_arg; 3448 uint64_t i; 3449 3450 if (bserrno != 0) { 3451 SPDK_ERRLOG("Failed to close clone of a corrupted blob\n"); 3452 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3453 return; 3454 } 3455 3456 /* Snapshot and clone have the same copy of cluster map and extent pages 3457 * at this point. Let's clear both for snpashot now, 3458 * so that it won't be cleared for clone later when we remove snapshot. 3459 * Also set thin provision to pass data corruption check */ 3460 for (i = 0; i < ctx->blob->active.num_clusters; i++) { 3461 ctx->blob->active.clusters[i] = 0; 3462 } 3463 for (i = 0; i < ctx->blob->active.num_extent_pages; i++) { 3464 ctx->blob->active.extent_pages[i] = 0; 3465 } 3466 3467 ctx->blob->md_ro = false; 3468 3469 blob_set_thin_provision(ctx->blob); 3470 3471 ctx->blobid = ctx->blob->id; 3472 3473 spdk_blob_close(ctx->blob, bs_delete_corrupted_close_cb, ctx); 3474 } 3475 3476 static void 3477 bs_update_corrupted_blob(void *cb_arg, int bserrno) 3478 { 3479 struct spdk_bs_load_ctx *ctx = cb_arg; 3480 3481 if (bserrno != 0) { 3482 SPDK_ERRLOG("Failed to close clone of a corrupted blob\n"); 3483 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3484 return; 3485 } 3486 3487 ctx->blob->md_ro = false; 3488 blob_remove_xattr(ctx->blob, SNAPSHOT_PENDING_REMOVAL, true); 3489 blob_remove_xattr(ctx->blob, SNAPSHOT_IN_PROGRESS, true); 3490 spdk_blob_set_read_only(ctx->blob); 3491 3492 if (ctx->iter_cb_fn) { 3493 ctx->iter_cb_fn(ctx->iter_cb_arg, ctx->blob, 0); 3494 } 3495 bs_blob_list_add(ctx->blob); 3496 3497 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3498 } 3499 3500 static void 3501 bs_examine_clone(void *cb_arg, struct spdk_blob *blob, int bserrno) 3502 { 3503 struct spdk_bs_load_ctx *ctx = cb_arg; 3504 3505 if (bserrno != 0) { 3506 SPDK_ERRLOG("Failed to open clone of a corrupted blob\n"); 3507 spdk_bs_iter_next(ctx->bs, ctx->blob, bs_load_iter, ctx); 3508 return; 3509 } 3510 3511 if (blob->parent_id == ctx->blob->id) { 3512 /* Power failure occured before updating clone (snapshot delete case) 3513 * or after updating clone (creating snapshot case) - keep snapshot */ 3514 spdk_blob_close(blob, bs_update_corrupted_blob, ctx); 3515 } else { 3516 /* Power failure occured after updating clone (snapshot delete case) 3517 * or before updating clone (creating snapshot case) - remove snapshot */ 3518 spdk_blob_close(blob, bs_delete_corrupted_blob, ctx); 3519 } 3520 } 3521 3522 static void 3523 bs_load_iter(void *arg, struct spdk_blob *blob, int bserrno) 3524 { 3525 struct spdk_bs_load_ctx *ctx = arg; 3526 const void *value; 3527 size_t len; 3528 int rc = 0; 3529 3530 if (bserrno == 0) { 3531 /* Examine blob if it is corrupted after power failure. Fix 3532 * the ones that can be fixed and remove any other corrupted 3533 * ones. If it is not corrupted just process it */ 3534 rc = blob_get_xattr_value(blob, SNAPSHOT_PENDING_REMOVAL, &value, &len, true); 3535 if (rc != 0) { 3536 rc = blob_get_xattr_value(blob, SNAPSHOT_IN_PROGRESS, &value, &len, true); 3537 if (rc != 0) { 3538 /* Not corrupted - process it and continue with iterating through blobs */ 3539 if (ctx->iter_cb_fn) { 3540 ctx->iter_cb_fn(ctx->iter_cb_arg, blob, 0); 3541 } 3542 bs_blob_list_add(blob); 3543 spdk_bs_iter_next(ctx->bs, blob, bs_load_iter, ctx); 3544 return; 3545 } 3546 3547 } 3548 3549 assert(len == sizeof(spdk_blob_id)); 3550 3551 ctx->blob = blob; 3552 3553 /* Open clone to check if we are able to fix this blob or should we remove it */ 3554 spdk_bs_open_blob(ctx->bs, *(spdk_blob_id *)value, bs_examine_clone, ctx); 3555 return; 3556 } else if (bserrno == -ENOENT) { 3557 bserrno = 0; 3558 } else { 3559 /* 3560 * This case needs to be looked at further. Same problem 3561 * exists with applications that rely on explicit blob 3562 * iteration. We should just skip the blob that failed 3563 * to load and continue on to the next one. 3564 */ 3565 SPDK_ERRLOG("Error in iterating blobs\n"); 3566 } 3567 3568 ctx->iter_cb_fn = NULL; 3569 3570 spdk_free(ctx->super); 3571 spdk_free(ctx->mask); 3572 bs_sequence_finish(ctx->seq, bserrno); 3573 free(ctx); 3574 } 3575 3576 static void 3577 bs_load_complete(struct spdk_bs_load_ctx *ctx) 3578 { 3579 ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters); 3580 spdk_bs_iter_first(ctx->bs, bs_load_iter, ctx); 3581 } 3582 3583 static void 3584 bs_load_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3585 { 3586 struct spdk_bs_load_ctx *ctx = cb_arg; 3587 int rc; 3588 3589 /* The type must be correct */ 3590 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_BLOBIDS); 3591 3592 /* The length of the mask (in bits) must not be greater than 3593 * the length of the buffer (converted to bits) */ 3594 assert(ctx->mask->length <= (ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE * 8)); 3595 3596 /* The length of the mask must be exactly equal to the size 3597 * (in pages) of the metadata region */ 3598 assert(ctx->mask->length == ctx->super->md_len); 3599 3600 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->mask->length); 3601 if (rc < 0) { 3602 spdk_free(ctx->mask); 3603 bs_load_ctx_fail(ctx, rc); 3604 return; 3605 } 3606 3607 spdk_bit_array_load_mask(ctx->bs->used_blobids, ctx->mask->mask); 3608 bs_load_complete(ctx); 3609 } 3610 3611 static void 3612 bs_load_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3613 { 3614 struct spdk_bs_load_ctx *ctx = cb_arg; 3615 uint64_t lba, lba_count, mask_size; 3616 int rc; 3617 3618 if (bserrno != 0) { 3619 bs_load_ctx_fail(ctx, bserrno); 3620 return; 3621 } 3622 3623 /* The type must be correct */ 3624 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_CLUSTERS); 3625 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 3626 assert(ctx->mask->length <= (ctx->super->used_cluster_mask_len * sizeof( 3627 struct spdk_blob_md_page) * 8)); 3628 /* The length of the mask must be exactly equal to the total number of clusters */ 3629 assert(ctx->mask->length == ctx->bs->total_clusters); 3630 3631 rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->mask->length); 3632 if (rc < 0) { 3633 spdk_free(ctx->mask); 3634 bs_load_ctx_fail(ctx, rc); 3635 return; 3636 } 3637 3638 spdk_bit_array_load_mask(ctx->used_clusters, ctx->mask->mask); 3639 ctx->bs->num_free_clusters = spdk_bit_array_count_clear(ctx->used_clusters); 3640 assert(ctx->bs->num_free_clusters <= ctx->bs->total_clusters); 3641 3642 spdk_free(ctx->mask); 3643 3644 /* Read the used blobids mask */ 3645 mask_size = ctx->super->used_blobid_mask_len * SPDK_BS_PAGE_SIZE; 3646 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 3647 SPDK_MALLOC_DMA); 3648 if (!ctx->mask) { 3649 bs_load_ctx_fail(ctx, -ENOMEM); 3650 return; 3651 } 3652 lba = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_start); 3653 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_blobid_mask_len); 3654 bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 3655 bs_load_used_blobids_cpl, ctx); 3656 } 3657 3658 static void 3659 bs_load_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3660 { 3661 struct spdk_bs_load_ctx *ctx = cb_arg; 3662 uint64_t lba, lba_count, mask_size; 3663 int rc; 3664 3665 if (bserrno != 0) { 3666 bs_load_ctx_fail(ctx, bserrno); 3667 return; 3668 } 3669 3670 /* The type must be correct */ 3671 assert(ctx->mask->type == SPDK_MD_MASK_TYPE_USED_PAGES); 3672 /* The length of the mask (in bits) must not be greater than the length of the buffer (converted to bits) */ 3673 assert(ctx->mask->length <= (ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE * 3674 8)); 3675 /* The length of the mask must be exactly equal to the size (in pages) of the metadata region */ 3676 assert(ctx->mask->length == ctx->super->md_len); 3677 3678 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->mask->length); 3679 if (rc < 0) { 3680 spdk_free(ctx->mask); 3681 bs_load_ctx_fail(ctx, rc); 3682 return; 3683 } 3684 3685 spdk_bit_array_load_mask(ctx->bs->used_md_pages, ctx->mask->mask); 3686 spdk_free(ctx->mask); 3687 3688 /* Read the used clusters mask */ 3689 mask_size = ctx->super->used_cluster_mask_len * SPDK_BS_PAGE_SIZE; 3690 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, SPDK_ENV_SOCKET_ID_ANY, 3691 SPDK_MALLOC_DMA); 3692 if (!ctx->mask) { 3693 bs_load_ctx_fail(ctx, -ENOMEM); 3694 return; 3695 } 3696 lba = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_start); 3697 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_cluster_mask_len); 3698 bs_sequence_read_dev(seq, ctx->mask, lba, lba_count, 3699 bs_load_used_clusters_cpl, ctx); 3700 } 3701 3702 static void 3703 bs_load_read_used_pages(struct spdk_bs_load_ctx *ctx) 3704 { 3705 uint64_t lba, lba_count, mask_size; 3706 3707 /* Read the used pages mask */ 3708 mask_size = ctx->super->used_page_mask_len * SPDK_BS_PAGE_SIZE; 3709 ctx->mask = spdk_zmalloc(mask_size, 0x1000, NULL, 3710 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 3711 if (!ctx->mask) { 3712 bs_load_ctx_fail(ctx, -ENOMEM); 3713 return; 3714 } 3715 3716 lba = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_start); 3717 lba_count = bs_page_to_lba(ctx->bs, ctx->super->used_page_mask_len); 3718 bs_sequence_read_dev(ctx->seq, ctx->mask, lba, lba_count, 3719 bs_load_used_pages_cpl, ctx); 3720 } 3721 3722 static int 3723 bs_load_replay_md_parse_page(struct spdk_bs_load_ctx *ctx, struct spdk_blob_md_page *page) 3724 { 3725 struct spdk_blob_store *bs = ctx->bs; 3726 struct spdk_blob_md_descriptor *desc; 3727 size_t cur_desc = 0; 3728 3729 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 3730 while (cur_desc < sizeof(page->descriptors)) { 3731 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 3732 if (desc->length == 0) { 3733 /* If padding and length are 0, this terminates the page */ 3734 break; 3735 } 3736 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) { 3737 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 3738 unsigned int i, j; 3739 unsigned int cluster_count = 0; 3740 uint32_t cluster_idx; 3741 3742 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc; 3743 3744 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 3745 for (j = 0; j < desc_extent_rle->extents[i].length; j++) { 3746 cluster_idx = desc_extent_rle->extents[i].cluster_idx; 3747 /* 3748 * cluster_idx = 0 means an unallocated cluster - don't mark that 3749 * in the used cluster map. 3750 */ 3751 if (cluster_idx != 0) { 3752 spdk_bit_array_set(ctx->used_clusters, cluster_idx + j); 3753 if (bs->num_free_clusters == 0) { 3754 return -ENOSPC; 3755 } 3756 bs->num_free_clusters--; 3757 } 3758 cluster_count++; 3759 } 3760 } 3761 if (cluster_count == 0) { 3762 return -EINVAL; 3763 } 3764 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 3765 struct spdk_blob_md_descriptor_extent_page *desc_extent; 3766 uint32_t i; 3767 uint32_t cluster_count = 0; 3768 uint32_t cluster_idx; 3769 size_t cluster_idx_length; 3770 3771 desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc; 3772 cluster_idx_length = desc_extent->length - sizeof(desc_extent->start_cluster_idx); 3773 3774 if (desc_extent->length <= sizeof(desc_extent->start_cluster_idx) || 3775 (cluster_idx_length % sizeof(desc_extent->cluster_idx[0]) != 0)) { 3776 return -EINVAL; 3777 } 3778 3779 for (i = 0; i < cluster_idx_length / sizeof(desc_extent->cluster_idx[0]); i++) { 3780 cluster_idx = desc_extent->cluster_idx[i]; 3781 /* 3782 * cluster_idx = 0 means an unallocated cluster - don't mark that 3783 * in the used cluster map. 3784 */ 3785 if (cluster_idx != 0) { 3786 if (cluster_idx < desc_extent->start_cluster_idx && 3787 cluster_idx >= desc_extent->start_cluster_idx + cluster_count) { 3788 return -EINVAL; 3789 } 3790 spdk_bit_array_set(ctx->used_clusters, cluster_idx); 3791 if (bs->num_free_clusters == 0) { 3792 return -ENOSPC; 3793 } 3794 bs->num_free_clusters--; 3795 } 3796 cluster_count++; 3797 } 3798 3799 if (cluster_count == 0) { 3800 return -EINVAL; 3801 } 3802 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 3803 /* Skip this item */ 3804 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 3805 /* Skip this item */ 3806 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 3807 /* Skip this item */ 3808 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_TABLE) { 3809 struct spdk_blob_md_descriptor_extent_table *desc_extent_table; 3810 uint32_t num_extent_pages = ctx->num_extent_pages; 3811 uint32_t i; 3812 size_t extent_pages_length; 3813 void *tmp; 3814 3815 desc_extent_table = (struct spdk_blob_md_descriptor_extent_table *)desc; 3816 extent_pages_length = desc_extent_table->length - sizeof(desc_extent_table->num_clusters); 3817 3818 if (desc_extent_table->length == 0 || 3819 (extent_pages_length % sizeof(desc_extent_table->extent_page[0]) != 0)) { 3820 return -EINVAL; 3821 } 3822 3823 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 3824 if (desc_extent_table->extent_page[i].page_idx != 0) { 3825 if (desc_extent_table->extent_page[i].num_pages != 1) { 3826 return -EINVAL; 3827 } 3828 num_extent_pages += 1; 3829 } 3830 } 3831 3832 if (num_extent_pages > 0) { 3833 tmp = realloc(ctx->extent_page_num, num_extent_pages * sizeof(uint32_t)); 3834 if (tmp == NULL) { 3835 return -ENOMEM; 3836 } 3837 ctx->extent_page_num = tmp; 3838 3839 /* Extent table entries contain md page numbers for extent pages. 3840 * Zeroes represent unallocated extent pages, those are run-length-encoded. 3841 */ 3842 for (i = 0; i < extent_pages_length / sizeof(desc_extent_table->extent_page[0]); i++) { 3843 if (desc_extent_table->extent_page[i].page_idx != 0) { 3844 ctx->extent_page_num[ctx->num_extent_pages] = desc_extent_table->extent_page[i].page_idx; 3845 ctx->num_extent_pages += 1; 3846 } 3847 } 3848 } 3849 } else { 3850 /* Error */ 3851 return -EINVAL; 3852 } 3853 /* Advance to the next descriptor */ 3854 cur_desc += sizeof(*desc) + desc->length; 3855 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 3856 break; 3857 } 3858 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 3859 } 3860 return 0; 3861 } 3862 3863 static bool bs_load_cur_extent_page_valid(struct spdk_blob_md_page *page) 3864 { 3865 uint32_t crc; 3866 struct spdk_blob_md_descriptor *desc = (struct spdk_blob_md_descriptor *)page->descriptors; 3867 size_t desc_len; 3868 3869 crc = blob_md_page_calc_crc(page); 3870 if (crc != page->crc) { 3871 return false; 3872 } 3873 3874 /* Extent page should always be of sequence num 0. */ 3875 if (page->sequence_num != 0) { 3876 return false; 3877 } 3878 3879 /* Descriptor type must be EXTENT_PAGE. */ 3880 if (desc->type != SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 3881 return false; 3882 } 3883 3884 /* Descriptor length cannot exceed the page. */ 3885 desc_len = sizeof(*desc) + desc->length; 3886 if (desc_len > sizeof(page->descriptors)) { 3887 return false; 3888 } 3889 3890 /* It has to be the only descriptor in the page. */ 3891 if (desc_len + sizeof(*desc) <= sizeof(page->descriptors)) { 3892 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + desc_len); 3893 if (desc->length != 0) { 3894 return false; 3895 } 3896 } 3897 3898 return true; 3899 } 3900 3901 static bool bs_load_cur_md_page_valid(struct spdk_bs_load_ctx *ctx) 3902 { 3903 uint32_t crc; 3904 struct spdk_blob_md_page *page = ctx->page; 3905 3906 crc = blob_md_page_calc_crc(page); 3907 if (crc != page->crc) { 3908 return false; 3909 } 3910 3911 /* First page of a sequence should match the blobid. */ 3912 if (page->sequence_num == 0 && 3913 bs_page_to_blobid(ctx->cur_page) != page->id) { 3914 return false; 3915 } 3916 assert(bs_load_cur_extent_page_valid(page) == false); 3917 3918 return true; 3919 } 3920 3921 static void 3922 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx); 3923 3924 static void 3925 bs_load_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3926 { 3927 struct spdk_bs_load_ctx *ctx = cb_arg; 3928 3929 if (bserrno != 0) { 3930 bs_load_ctx_fail(ctx, bserrno); 3931 return; 3932 } 3933 3934 bs_load_complete(ctx); 3935 } 3936 3937 static void 3938 bs_load_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3939 { 3940 struct spdk_bs_load_ctx *ctx = cb_arg; 3941 3942 spdk_free(ctx->mask); 3943 ctx->mask = NULL; 3944 3945 if (bserrno != 0) { 3946 bs_load_ctx_fail(ctx, bserrno); 3947 return; 3948 } 3949 3950 bs_write_used_clusters(seq, ctx, bs_load_write_used_clusters_cpl); 3951 } 3952 3953 static void 3954 bs_load_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 3955 { 3956 struct spdk_bs_load_ctx *ctx = cb_arg; 3957 3958 spdk_free(ctx->mask); 3959 ctx->mask = NULL; 3960 3961 if (bserrno != 0) { 3962 bs_load_ctx_fail(ctx, bserrno); 3963 return; 3964 } 3965 3966 bs_write_used_blobids(seq, ctx, bs_load_write_used_blobids_cpl); 3967 } 3968 3969 static void 3970 bs_load_write_used_md(struct spdk_bs_load_ctx *ctx) 3971 { 3972 bs_write_used_md(ctx->seq, ctx, bs_load_write_used_pages_cpl); 3973 } 3974 3975 static void 3976 bs_load_replay_md_chain_cpl(struct spdk_bs_load_ctx *ctx) 3977 { 3978 uint64_t num_md_clusters; 3979 uint64_t i; 3980 3981 ctx->in_page_chain = false; 3982 3983 do { 3984 ctx->page_index++; 3985 } while (spdk_bit_array_get(ctx->bs->used_md_pages, ctx->page_index) == true); 3986 3987 if (ctx->page_index < ctx->super->md_len) { 3988 ctx->cur_page = ctx->page_index; 3989 bs_load_replay_cur_md_page(ctx); 3990 } else { 3991 /* Claim all of the clusters used by the metadata */ 3992 num_md_clusters = spdk_divide_round_up(ctx->super->md_len, ctx->bs->pages_per_cluster); 3993 for (i = 0; i < num_md_clusters; i++) { 3994 spdk_bit_array_set(ctx->used_clusters, i); 3995 } 3996 ctx->bs->num_free_clusters -= num_md_clusters; 3997 spdk_free(ctx->page); 3998 bs_load_write_used_md(ctx); 3999 } 4000 } 4001 4002 static void 4003 bs_load_replay_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4004 { 4005 struct spdk_bs_load_ctx *ctx = cb_arg; 4006 uint32_t page_num; 4007 uint64_t i; 4008 4009 if (bserrno != 0) { 4010 spdk_free(ctx->extent_pages); 4011 bs_load_ctx_fail(ctx, bserrno); 4012 return; 4013 } 4014 4015 for (i = 0; i < ctx->num_extent_pages; i++) { 4016 /* Extent pages are only read when present within in chain md. 4017 * Integrity of md is not right if that page was not a valid extent page. */ 4018 if (bs_load_cur_extent_page_valid(&ctx->extent_pages[i]) != true) { 4019 spdk_free(ctx->extent_pages); 4020 bs_load_ctx_fail(ctx, -EILSEQ); 4021 return; 4022 } 4023 4024 page_num = ctx->extent_page_num[i]; 4025 spdk_bit_array_set(ctx->bs->used_md_pages, page_num); 4026 if (bs_load_replay_md_parse_page(ctx, &ctx->extent_pages[i])) { 4027 spdk_free(ctx->extent_pages); 4028 bs_load_ctx_fail(ctx, -EILSEQ); 4029 return; 4030 } 4031 } 4032 4033 spdk_free(ctx->extent_pages); 4034 free(ctx->extent_page_num); 4035 ctx->extent_page_num = NULL; 4036 ctx->num_extent_pages = 0; 4037 4038 bs_load_replay_md_chain_cpl(ctx); 4039 } 4040 4041 static void 4042 bs_load_replay_extent_pages(struct spdk_bs_load_ctx *ctx) 4043 { 4044 spdk_bs_batch_t *batch; 4045 uint32_t page; 4046 uint64_t lba; 4047 uint64_t i; 4048 4049 ctx->extent_pages = spdk_zmalloc(SPDK_BS_PAGE_SIZE * ctx->num_extent_pages, 0, 4050 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 4051 if (!ctx->extent_pages) { 4052 bs_load_ctx_fail(ctx, -ENOMEM); 4053 return; 4054 } 4055 4056 batch = bs_sequence_to_batch(ctx->seq, bs_load_replay_extent_page_cpl, ctx); 4057 4058 for (i = 0; i < ctx->num_extent_pages; i++) { 4059 page = ctx->extent_page_num[i]; 4060 assert(page < ctx->super->md_len); 4061 lba = bs_md_page_to_lba(ctx->bs, page); 4062 bs_batch_read_dev(batch, &ctx->extent_pages[i], lba, 4063 bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE)); 4064 } 4065 4066 bs_batch_close(batch); 4067 } 4068 4069 static void 4070 bs_load_replay_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4071 { 4072 struct spdk_bs_load_ctx *ctx = cb_arg; 4073 uint32_t page_num; 4074 struct spdk_blob_md_page *page; 4075 4076 if (bserrno != 0) { 4077 bs_load_ctx_fail(ctx, bserrno); 4078 return; 4079 } 4080 4081 page_num = ctx->cur_page; 4082 page = ctx->page; 4083 if (bs_load_cur_md_page_valid(ctx) == true) { 4084 if (page->sequence_num == 0 || ctx->in_page_chain == true) { 4085 bs_claim_md_page(ctx->bs, page_num); 4086 if (page->sequence_num == 0) { 4087 spdk_bit_array_set(ctx->bs->used_blobids, page_num); 4088 } 4089 if (bs_load_replay_md_parse_page(ctx, page)) { 4090 bs_load_ctx_fail(ctx, -EILSEQ); 4091 return; 4092 } 4093 if (page->next != SPDK_INVALID_MD_PAGE) { 4094 ctx->in_page_chain = true; 4095 ctx->cur_page = page->next; 4096 bs_load_replay_cur_md_page(ctx); 4097 return; 4098 } 4099 if (ctx->num_extent_pages != 0) { 4100 bs_load_replay_extent_pages(ctx); 4101 return; 4102 } 4103 } 4104 } 4105 bs_load_replay_md_chain_cpl(ctx); 4106 } 4107 4108 static void 4109 bs_load_replay_cur_md_page(struct spdk_bs_load_ctx *ctx) 4110 { 4111 uint64_t lba; 4112 4113 assert(ctx->cur_page < ctx->super->md_len); 4114 lba = bs_md_page_to_lba(ctx->bs, ctx->cur_page); 4115 bs_sequence_read_dev(ctx->seq, ctx->page, lba, 4116 bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 4117 bs_load_replay_md_cpl, ctx); 4118 } 4119 4120 static void 4121 bs_load_replay_md(struct spdk_bs_load_ctx *ctx) 4122 { 4123 ctx->page_index = 0; 4124 ctx->cur_page = 0; 4125 ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0, 4126 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 4127 if (!ctx->page) { 4128 bs_load_ctx_fail(ctx, -ENOMEM); 4129 return; 4130 } 4131 bs_load_replay_cur_md_page(ctx); 4132 } 4133 4134 static void 4135 bs_recover(struct spdk_bs_load_ctx *ctx) 4136 { 4137 int rc; 4138 4139 rc = spdk_bit_array_resize(&ctx->bs->used_md_pages, ctx->super->md_len); 4140 if (rc < 0) { 4141 bs_load_ctx_fail(ctx, -ENOMEM); 4142 return; 4143 } 4144 4145 rc = spdk_bit_array_resize(&ctx->bs->used_blobids, ctx->super->md_len); 4146 if (rc < 0) { 4147 bs_load_ctx_fail(ctx, -ENOMEM); 4148 return; 4149 } 4150 4151 rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters); 4152 if (rc < 0) { 4153 bs_load_ctx_fail(ctx, -ENOMEM); 4154 return; 4155 } 4156 4157 rc = spdk_bit_array_resize(&ctx->bs->open_blobids, ctx->super->md_len); 4158 if (rc < 0) { 4159 bs_load_ctx_fail(ctx, -ENOMEM); 4160 return; 4161 } 4162 4163 ctx->bs->num_free_clusters = ctx->bs->total_clusters; 4164 bs_load_replay_md(ctx); 4165 } 4166 4167 static void 4168 bs_load_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4169 { 4170 struct spdk_bs_load_ctx *ctx = cb_arg; 4171 uint32_t crc; 4172 int rc; 4173 static const char zeros[SPDK_BLOBSTORE_TYPE_LENGTH]; 4174 4175 if (ctx->super->version > SPDK_BS_VERSION || 4176 ctx->super->version < SPDK_BS_INITIAL_VERSION) { 4177 bs_load_ctx_fail(ctx, -EILSEQ); 4178 return; 4179 } 4180 4181 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 4182 sizeof(ctx->super->signature)) != 0) { 4183 bs_load_ctx_fail(ctx, -EILSEQ); 4184 return; 4185 } 4186 4187 crc = blob_md_page_calc_crc(ctx->super); 4188 if (crc != ctx->super->crc) { 4189 bs_load_ctx_fail(ctx, -EILSEQ); 4190 return; 4191 } 4192 4193 if (memcmp(&ctx->bs->bstype, &ctx->super->bstype, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 4194 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype matched - loading blobstore\n"); 4195 } else if (memcmp(&ctx->bs->bstype, zeros, SPDK_BLOBSTORE_TYPE_LENGTH) == 0) { 4196 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Bstype wildcard used - loading blobstore regardless bstype\n"); 4197 } else { 4198 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Unexpected bstype\n"); 4199 SPDK_LOGDUMP(SPDK_LOG_BLOB, "Expected:", ctx->bs->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 4200 SPDK_LOGDUMP(SPDK_LOG_BLOB, "Found:", ctx->super->bstype.bstype, SPDK_BLOBSTORE_TYPE_LENGTH); 4201 bs_load_ctx_fail(ctx, -ENXIO); 4202 return; 4203 } 4204 4205 if (ctx->super->size > ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen) { 4206 SPDK_NOTICELOG("Size mismatch, dev size: %lu, blobstore size: %lu\n", 4207 ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen, ctx->super->size); 4208 bs_load_ctx_fail(ctx, -EILSEQ); 4209 return; 4210 } 4211 4212 if (ctx->super->size == 0) { 4213 ctx->super->size = ctx->bs->dev->blockcnt * ctx->bs->dev->blocklen; 4214 } 4215 4216 if (ctx->super->io_unit_size == 0) { 4217 ctx->super->io_unit_size = SPDK_BS_PAGE_SIZE; 4218 } 4219 4220 /* Parse the super block */ 4221 ctx->bs->clean = 1; 4222 ctx->bs->cluster_sz = ctx->super->cluster_size; 4223 ctx->bs->total_clusters = ctx->super->size / ctx->super->cluster_size; 4224 ctx->bs->pages_per_cluster = ctx->bs->cluster_sz / SPDK_BS_PAGE_SIZE; 4225 if (spdk_u32_is_pow2(ctx->bs->pages_per_cluster)) { 4226 ctx->bs->pages_per_cluster_shift = spdk_u32log2(ctx->bs->pages_per_cluster); 4227 } 4228 ctx->bs->io_unit_size = ctx->super->io_unit_size; 4229 rc = spdk_bit_array_resize(&ctx->used_clusters, ctx->bs->total_clusters); 4230 if (rc < 0) { 4231 bs_load_ctx_fail(ctx, -ENOMEM); 4232 return; 4233 } 4234 ctx->bs->md_start = ctx->super->md_start; 4235 ctx->bs->md_len = ctx->super->md_len; 4236 ctx->bs->total_data_clusters = ctx->bs->total_clusters - spdk_divide_round_up( 4237 ctx->bs->md_start + ctx->bs->md_len, ctx->bs->pages_per_cluster); 4238 ctx->bs->super_blob = ctx->super->super_blob; 4239 memcpy(&ctx->bs->bstype, &ctx->super->bstype, sizeof(ctx->super->bstype)); 4240 4241 if (ctx->super->used_blobid_mask_len == 0 || ctx->super->clean == 0) { 4242 bs_recover(ctx); 4243 } else { 4244 bs_load_read_used_pages(ctx); 4245 } 4246 } 4247 4248 void 4249 spdk_bs_load(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 4250 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 4251 { 4252 struct spdk_blob_store *bs; 4253 struct spdk_bs_cpl cpl; 4254 struct spdk_bs_load_ctx *ctx; 4255 struct spdk_bs_opts opts = {}; 4256 int err; 4257 4258 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Loading blobstore from dev %p\n", dev); 4259 4260 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 4261 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "unsupported dev block length of %d\n", dev->blocklen); 4262 dev->destroy(dev); 4263 cb_fn(cb_arg, NULL, -EINVAL); 4264 return; 4265 } 4266 4267 if (o) { 4268 opts = *o; 4269 } else { 4270 spdk_bs_opts_init(&opts); 4271 } 4272 4273 if (opts.max_md_ops == 0 || opts.max_channel_ops == 0) { 4274 dev->destroy(dev); 4275 cb_fn(cb_arg, NULL, -EINVAL); 4276 return; 4277 } 4278 4279 err = bs_alloc(dev, &opts, &bs, &ctx); 4280 if (err) { 4281 dev->destroy(dev); 4282 cb_fn(cb_arg, NULL, err); 4283 return; 4284 } 4285 4286 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 4287 cpl.u.bs_handle.cb_fn = cb_fn; 4288 cpl.u.bs_handle.cb_arg = cb_arg; 4289 cpl.u.bs_handle.bs = bs; 4290 4291 ctx->seq = bs_sequence_start(bs->md_channel, &cpl); 4292 if (!ctx->seq) { 4293 spdk_free(ctx->super); 4294 free(ctx); 4295 bs_free(bs); 4296 cb_fn(cb_arg, NULL, -ENOMEM); 4297 return; 4298 } 4299 4300 /* Read the super block */ 4301 bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0), 4302 bs_byte_to_lba(bs, sizeof(*ctx->super)), 4303 bs_load_super_cpl, ctx); 4304 } 4305 4306 /* END spdk_bs_load */ 4307 4308 /* START spdk_bs_dump */ 4309 4310 static void 4311 bs_dump_finish(spdk_bs_sequence_t *seq, struct spdk_bs_load_ctx *ctx, int bserrno) 4312 { 4313 spdk_free(ctx->super); 4314 4315 /* 4316 * We need to defer calling bs_call_cpl() until after 4317 * dev destruction, so tuck these away for later use. 4318 */ 4319 ctx->bs->unload_err = bserrno; 4320 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 4321 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 4322 4323 bs_sequence_finish(seq, 0); 4324 bs_free(ctx->bs); 4325 free(ctx); 4326 } 4327 4328 static void bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg); 4329 4330 static void 4331 bs_dump_print_md_page(struct spdk_bs_load_ctx *ctx) 4332 { 4333 uint32_t page_idx = ctx->cur_page; 4334 struct spdk_blob_md_page *page = ctx->page; 4335 struct spdk_blob_md_descriptor *desc; 4336 size_t cur_desc = 0; 4337 uint32_t crc; 4338 4339 fprintf(ctx->fp, "=========\n"); 4340 fprintf(ctx->fp, "Metadata Page Index: %" PRIu32 " (0x%" PRIx32 ")\n", page_idx, page_idx); 4341 fprintf(ctx->fp, "Blob ID: 0x%" PRIx64 "\n", page->id); 4342 4343 crc = blob_md_page_calc_crc(page); 4344 fprintf(ctx->fp, "CRC: 0x%" PRIx32 " (%s)\n", page->crc, crc == page->crc ? "OK" : "Mismatch"); 4345 4346 desc = (struct spdk_blob_md_descriptor *)page->descriptors; 4347 while (cur_desc < sizeof(page->descriptors)) { 4348 if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_PADDING) { 4349 if (desc->length == 0) { 4350 /* If padding and length are 0, this terminates the page */ 4351 break; 4352 } 4353 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_RLE) { 4354 struct spdk_blob_md_descriptor_extent_rle *desc_extent_rle; 4355 unsigned int i; 4356 4357 desc_extent_rle = (struct spdk_blob_md_descriptor_extent_rle *)desc; 4358 4359 for (i = 0; i < desc_extent_rle->length / sizeof(desc_extent_rle->extents[0]); i++) { 4360 if (desc_extent_rle->extents[i].cluster_idx != 0) { 4361 fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32, 4362 desc_extent_rle->extents[i].cluster_idx); 4363 } else { 4364 fprintf(ctx->fp, "Unallocated Extent - "); 4365 } 4366 fprintf(ctx->fp, " Length: %" PRIu32, desc_extent_rle->extents[i].length); 4367 fprintf(ctx->fp, "\n"); 4368 } 4369 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_EXTENT_PAGE) { 4370 struct spdk_blob_md_descriptor_extent_page *desc_extent; 4371 unsigned int i; 4372 4373 desc_extent = (struct spdk_blob_md_descriptor_extent_page *)desc; 4374 4375 for (i = 0; i < desc_extent->length / sizeof(desc_extent->cluster_idx[0]); i++) { 4376 if (desc_extent->cluster_idx[i] != 0) { 4377 fprintf(ctx->fp, "Allocated Extent - Start: %" PRIu32, 4378 desc_extent->cluster_idx[i]); 4379 } else { 4380 fprintf(ctx->fp, "Unallocated Extent"); 4381 } 4382 fprintf(ctx->fp, "\n"); 4383 } 4384 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR) { 4385 struct spdk_blob_md_descriptor_xattr *desc_xattr; 4386 uint32_t i; 4387 4388 desc_xattr = (struct spdk_blob_md_descriptor_xattr *)desc; 4389 4390 if (desc_xattr->length != 4391 sizeof(desc_xattr->name_length) + sizeof(desc_xattr->value_length) + 4392 desc_xattr->name_length + desc_xattr->value_length) { 4393 } 4394 4395 memcpy(ctx->xattr_name, desc_xattr->name, desc_xattr->name_length); 4396 ctx->xattr_name[desc_xattr->name_length] = '\0'; 4397 fprintf(ctx->fp, "XATTR: name = \"%s\"\n", ctx->xattr_name); 4398 fprintf(ctx->fp, " value = \""); 4399 ctx->print_xattr_fn(ctx->fp, ctx->super->bstype.bstype, ctx->xattr_name, 4400 (void *)((uintptr_t)desc_xattr->name + desc_xattr->name_length), 4401 desc_xattr->value_length); 4402 fprintf(ctx->fp, "\"\n"); 4403 for (i = 0; i < desc_xattr->value_length; i++) { 4404 if (i % 16 == 0) { 4405 fprintf(ctx->fp, " "); 4406 } 4407 fprintf(ctx->fp, "%02" PRIx8 " ", *((uint8_t *)desc_xattr->name + desc_xattr->name_length + i)); 4408 if ((i + 1) % 16 == 0) { 4409 fprintf(ctx->fp, "\n"); 4410 } 4411 } 4412 if (i % 16 != 0) { 4413 fprintf(ctx->fp, "\n"); 4414 } 4415 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_XATTR_INTERNAL) { 4416 /* TODO */ 4417 } else if (desc->type == SPDK_MD_DESCRIPTOR_TYPE_FLAGS) { 4418 /* TODO */ 4419 } else { 4420 /* Error */ 4421 } 4422 /* Advance to the next descriptor */ 4423 cur_desc += sizeof(*desc) + desc->length; 4424 if (cur_desc + sizeof(*desc) > sizeof(page->descriptors)) { 4425 break; 4426 } 4427 desc = (struct spdk_blob_md_descriptor *)((uintptr_t)page->descriptors + cur_desc); 4428 } 4429 } 4430 4431 static void 4432 bs_dump_read_md_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4433 { 4434 struct spdk_bs_load_ctx *ctx = cb_arg; 4435 4436 if (bserrno != 0) { 4437 bs_dump_finish(seq, ctx, bserrno); 4438 return; 4439 } 4440 4441 if (ctx->page->id != 0) { 4442 bs_dump_print_md_page(ctx); 4443 } 4444 4445 ctx->cur_page++; 4446 4447 if (ctx->cur_page < ctx->super->md_len) { 4448 bs_dump_read_md_page(seq, ctx); 4449 } else { 4450 spdk_free(ctx->page); 4451 bs_dump_finish(seq, ctx, 0); 4452 } 4453 } 4454 4455 static void 4456 bs_dump_read_md_page(spdk_bs_sequence_t *seq, void *cb_arg) 4457 { 4458 struct spdk_bs_load_ctx *ctx = cb_arg; 4459 uint64_t lba; 4460 4461 assert(ctx->cur_page < ctx->super->md_len); 4462 lba = bs_page_to_lba(ctx->bs, ctx->super->md_start + ctx->cur_page); 4463 bs_sequence_read_dev(seq, ctx->page, lba, 4464 bs_byte_to_lba(ctx->bs, SPDK_BS_PAGE_SIZE), 4465 bs_dump_read_md_page_cpl, ctx); 4466 } 4467 4468 static void 4469 bs_dump_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4470 { 4471 struct spdk_bs_load_ctx *ctx = cb_arg; 4472 4473 fprintf(ctx->fp, "Signature: \"%.8s\" ", ctx->super->signature); 4474 if (memcmp(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 4475 sizeof(ctx->super->signature)) != 0) { 4476 fprintf(ctx->fp, "(Mismatch)\n"); 4477 bs_dump_finish(seq, ctx, bserrno); 4478 return; 4479 } else { 4480 fprintf(ctx->fp, "(OK)\n"); 4481 } 4482 fprintf(ctx->fp, "Version: %" PRIu32 "\n", ctx->super->version); 4483 fprintf(ctx->fp, "CRC: 0x%x (%s)\n", ctx->super->crc, 4484 (ctx->super->crc == blob_md_page_calc_crc(ctx->super)) ? "OK" : "Mismatch"); 4485 fprintf(ctx->fp, "Blobstore Type: %.*s\n", SPDK_BLOBSTORE_TYPE_LENGTH, ctx->super->bstype.bstype); 4486 fprintf(ctx->fp, "Cluster Size: %" PRIu32 "\n", ctx->super->cluster_size); 4487 fprintf(ctx->fp, "Super Blob ID: "); 4488 if (ctx->super->super_blob == SPDK_BLOBID_INVALID) { 4489 fprintf(ctx->fp, "(None)\n"); 4490 } else { 4491 fprintf(ctx->fp, "%" PRIu64 "\n", ctx->super->super_blob); 4492 } 4493 fprintf(ctx->fp, "Clean: %" PRIu32 "\n", ctx->super->clean); 4494 fprintf(ctx->fp, "Used Metadata Page Mask Start: %" PRIu32 "\n", ctx->super->used_page_mask_start); 4495 fprintf(ctx->fp, "Used Metadata Page Mask Length: %" PRIu32 "\n", ctx->super->used_page_mask_len); 4496 fprintf(ctx->fp, "Used Cluster Mask Start: %" PRIu32 "\n", ctx->super->used_cluster_mask_start); 4497 fprintf(ctx->fp, "Used Cluster Mask Length: %" PRIu32 "\n", ctx->super->used_cluster_mask_len); 4498 fprintf(ctx->fp, "Used Blob ID Mask Start: %" PRIu32 "\n", ctx->super->used_blobid_mask_start); 4499 fprintf(ctx->fp, "Used Blob ID Mask Length: %" PRIu32 "\n", ctx->super->used_blobid_mask_len); 4500 fprintf(ctx->fp, "Metadata Start: %" PRIu32 "\n", ctx->super->md_start); 4501 fprintf(ctx->fp, "Metadata Length: %" PRIu32 "\n", ctx->super->md_len); 4502 4503 ctx->cur_page = 0; 4504 ctx->page = spdk_zmalloc(SPDK_BS_PAGE_SIZE, 0, 4505 NULL, SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 4506 if (!ctx->page) { 4507 bs_dump_finish(seq, ctx, -ENOMEM); 4508 return; 4509 } 4510 bs_dump_read_md_page(seq, ctx); 4511 } 4512 4513 void 4514 spdk_bs_dump(struct spdk_bs_dev *dev, FILE *fp, spdk_bs_dump_print_xattr print_xattr_fn, 4515 spdk_bs_op_complete cb_fn, void *cb_arg) 4516 { 4517 struct spdk_blob_store *bs; 4518 struct spdk_bs_cpl cpl; 4519 spdk_bs_sequence_t *seq; 4520 struct spdk_bs_load_ctx *ctx; 4521 struct spdk_bs_opts opts = {}; 4522 int err; 4523 4524 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Dumping blobstore from dev %p\n", dev); 4525 4526 spdk_bs_opts_init(&opts); 4527 4528 err = bs_alloc(dev, &opts, &bs, &ctx); 4529 if (err) { 4530 dev->destroy(dev); 4531 cb_fn(cb_arg, err); 4532 return; 4533 } 4534 4535 ctx->fp = fp; 4536 ctx->print_xattr_fn = print_xattr_fn; 4537 4538 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 4539 cpl.u.bs_basic.cb_fn = cb_fn; 4540 cpl.u.bs_basic.cb_arg = cb_arg; 4541 4542 seq = bs_sequence_start(bs->md_channel, &cpl); 4543 if (!seq) { 4544 spdk_free(ctx->super); 4545 free(ctx); 4546 bs_free(bs); 4547 cb_fn(cb_arg, -ENOMEM); 4548 return; 4549 } 4550 4551 /* Read the super block */ 4552 bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0), 4553 bs_byte_to_lba(bs, sizeof(*ctx->super)), 4554 bs_dump_super_cpl, ctx); 4555 } 4556 4557 /* END spdk_bs_dump */ 4558 4559 /* START spdk_bs_init */ 4560 4561 static void 4562 bs_init_persist_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4563 { 4564 struct spdk_bs_load_ctx *ctx = cb_arg; 4565 4566 ctx->bs->used_clusters = spdk_bit_pool_create_from_array(ctx->used_clusters); 4567 spdk_free(ctx->super); 4568 free(ctx); 4569 4570 bs_sequence_finish(seq, bserrno); 4571 } 4572 4573 static void 4574 bs_init_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4575 { 4576 struct spdk_bs_load_ctx *ctx = cb_arg; 4577 4578 /* Write super block */ 4579 bs_sequence_write_dev(seq, ctx->super, bs_page_to_lba(ctx->bs, 0), 4580 bs_byte_to_lba(ctx->bs, sizeof(*ctx->super)), 4581 bs_init_persist_super_cpl, ctx); 4582 } 4583 4584 void 4585 spdk_bs_init(struct spdk_bs_dev *dev, struct spdk_bs_opts *o, 4586 spdk_bs_op_with_handle_complete cb_fn, void *cb_arg) 4587 { 4588 struct spdk_bs_load_ctx *ctx; 4589 struct spdk_blob_store *bs; 4590 struct spdk_bs_cpl cpl; 4591 spdk_bs_sequence_t *seq; 4592 spdk_bs_batch_t *batch; 4593 uint64_t num_md_lba; 4594 uint64_t num_md_pages; 4595 uint64_t num_md_clusters; 4596 uint32_t i; 4597 struct spdk_bs_opts opts = {}; 4598 int rc; 4599 uint64_t lba, lba_count; 4600 4601 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Initializing blobstore on dev %p\n", dev); 4602 4603 if ((SPDK_BS_PAGE_SIZE % dev->blocklen) != 0) { 4604 SPDK_ERRLOG("unsupported dev block length of %d\n", 4605 dev->blocklen); 4606 dev->destroy(dev); 4607 cb_fn(cb_arg, NULL, -EINVAL); 4608 return; 4609 } 4610 4611 if (o) { 4612 opts = *o; 4613 } else { 4614 spdk_bs_opts_init(&opts); 4615 } 4616 4617 if (bs_opts_verify(&opts) != 0) { 4618 dev->destroy(dev); 4619 cb_fn(cb_arg, NULL, -EINVAL); 4620 return; 4621 } 4622 4623 rc = bs_alloc(dev, &opts, &bs, &ctx); 4624 if (rc) { 4625 dev->destroy(dev); 4626 cb_fn(cb_arg, NULL, rc); 4627 return; 4628 } 4629 4630 if (opts.num_md_pages == SPDK_BLOB_OPTS_NUM_MD_PAGES) { 4631 /* By default, allocate 1 page per cluster. 4632 * Technically, this over-allocates metadata 4633 * because more metadata will reduce the number 4634 * of usable clusters. This can be addressed with 4635 * more complex math in the future. 4636 */ 4637 bs->md_len = bs->total_clusters; 4638 } else { 4639 bs->md_len = opts.num_md_pages; 4640 } 4641 rc = spdk_bit_array_resize(&bs->used_md_pages, bs->md_len); 4642 if (rc < 0) { 4643 spdk_free(ctx->super); 4644 free(ctx); 4645 bs_free(bs); 4646 cb_fn(cb_arg, NULL, -ENOMEM); 4647 return; 4648 } 4649 4650 rc = spdk_bit_array_resize(&bs->used_blobids, bs->md_len); 4651 if (rc < 0) { 4652 spdk_free(ctx->super); 4653 free(ctx); 4654 bs_free(bs); 4655 cb_fn(cb_arg, NULL, -ENOMEM); 4656 return; 4657 } 4658 4659 rc = spdk_bit_array_resize(&bs->open_blobids, bs->md_len); 4660 if (rc < 0) { 4661 spdk_free(ctx->super); 4662 free(ctx); 4663 bs_free(bs); 4664 cb_fn(cb_arg, NULL, -ENOMEM); 4665 return; 4666 } 4667 4668 memcpy(ctx->super->signature, SPDK_BS_SUPER_BLOCK_SIG, 4669 sizeof(ctx->super->signature)); 4670 ctx->super->version = SPDK_BS_VERSION; 4671 ctx->super->length = sizeof(*ctx->super); 4672 ctx->super->super_blob = bs->super_blob; 4673 ctx->super->clean = 0; 4674 ctx->super->cluster_size = bs->cluster_sz; 4675 ctx->super->io_unit_size = bs->io_unit_size; 4676 memcpy(&ctx->super->bstype, &bs->bstype, sizeof(bs->bstype)); 4677 4678 /* Calculate how many pages the metadata consumes at the front 4679 * of the disk. 4680 */ 4681 4682 /* The super block uses 1 page */ 4683 num_md_pages = 1; 4684 4685 /* The used_md_pages mask requires 1 bit per metadata page, rounded 4686 * up to the nearest page, plus a header. 4687 */ 4688 ctx->super->used_page_mask_start = num_md_pages; 4689 ctx->super->used_page_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 4690 spdk_divide_round_up(bs->md_len, 8), 4691 SPDK_BS_PAGE_SIZE); 4692 num_md_pages += ctx->super->used_page_mask_len; 4693 4694 /* The used_clusters mask requires 1 bit per cluster, rounded 4695 * up to the nearest page, plus a header. 4696 */ 4697 ctx->super->used_cluster_mask_start = num_md_pages; 4698 ctx->super->used_cluster_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 4699 spdk_divide_round_up(bs->total_clusters, 8), 4700 SPDK_BS_PAGE_SIZE); 4701 num_md_pages += ctx->super->used_cluster_mask_len; 4702 4703 /* The used_blobids mask requires 1 bit per metadata page, rounded 4704 * up to the nearest page, plus a header. 4705 */ 4706 ctx->super->used_blobid_mask_start = num_md_pages; 4707 ctx->super->used_blobid_mask_len = spdk_divide_round_up(sizeof(struct spdk_bs_md_mask) + 4708 spdk_divide_round_up(bs->md_len, 8), 4709 SPDK_BS_PAGE_SIZE); 4710 num_md_pages += ctx->super->used_blobid_mask_len; 4711 4712 /* The metadata region size was chosen above */ 4713 ctx->super->md_start = bs->md_start = num_md_pages; 4714 ctx->super->md_len = bs->md_len; 4715 num_md_pages += bs->md_len; 4716 4717 num_md_lba = bs_page_to_lba(bs, num_md_pages); 4718 4719 ctx->super->size = dev->blockcnt * dev->blocklen; 4720 4721 ctx->super->crc = blob_md_page_calc_crc(ctx->super); 4722 4723 num_md_clusters = spdk_divide_round_up(num_md_pages, bs->pages_per_cluster); 4724 if (num_md_clusters > bs->total_clusters) { 4725 SPDK_ERRLOG("Blobstore metadata cannot use more clusters than is available, " 4726 "please decrease number of pages reserved for metadata " 4727 "or increase cluster size.\n"); 4728 spdk_free(ctx->super); 4729 spdk_bit_array_free(&ctx->used_clusters); 4730 free(ctx); 4731 bs_free(bs); 4732 cb_fn(cb_arg, NULL, -ENOMEM); 4733 return; 4734 } 4735 /* Claim all of the clusters used by the metadata */ 4736 for (i = 0; i < num_md_clusters; i++) { 4737 spdk_bit_array_set(ctx->used_clusters, i); 4738 } 4739 4740 bs->num_free_clusters -= num_md_clusters; 4741 bs->total_data_clusters = bs->num_free_clusters; 4742 4743 cpl.type = SPDK_BS_CPL_TYPE_BS_HANDLE; 4744 cpl.u.bs_handle.cb_fn = cb_fn; 4745 cpl.u.bs_handle.cb_arg = cb_arg; 4746 cpl.u.bs_handle.bs = bs; 4747 4748 seq = bs_sequence_start(bs->md_channel, &cpl); 4749 if (!seq) { 4750 spdk_free(ctx->super); 4751 free(ctx); 4752 bs_free(bs); 4753 cb_fn(cb_arg, NULL, -ENOMEM); 4754 return; 4755 } 4756 4757 batch = bs_sequence_to_batch(seq, bs_init_trim_cpl, ctx); 4758 4759 /* Clear metadata space */ 4760 bs_batch_write_zeroes_dev(batch, 0, num_md_lba); 4761 4762 lba = num_md_lba; 4763 while (lba < ctx->bs->dev->blockcnt) { 4764 lba_count = spdk_min(UINT32_MAX, ctx->bs->dev->blockcnt - lba); 4765 switch (opts.clear_method) { 4766 case BS_CLEAR_WITH_UNMAP: 4767 /* Trim data clusters */ 4768 bs_batch_unmap_dev(batch, lba, lba_count); 4769 break; 4770 case BS_CLEAR_WITH_WRITE_ZEROES: 4771 /* Write_zeroes to data clusters */ 4772 bs_batch_write_zeroes_dev(batch, lba, lba_count); 4773 break; 4774 case BS_CLEAR_WITH_NONE: 4775 default: 4776 break; 4777 } 4778 lba += lba_count; 4779 } 4780 4781 bs_batch_close(batch); 4782 } 4783 4784 /* END spdk_bs_init */ 4785 4786 /* START spdk_bs_destroy */ 4787 4788 static void 4789 bs_destroy_trim_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4790 { 4791 struct spdk_bs_load_ctx *ctx = cb_arg; 4792 struct spdk_blob_store *bs = ctx->bs; 4793 4794 /* 4795 * We need to defer calling bs_call_cpl() until after 4796 * dev destruction, so tuck these away for later use. 4797 */ 4798 bs->unload_err = bserrno; 4799 memcpy(&bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 4800 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 4801 4802 bs_sequence_finish(seq, bserrno); 4803 4804 bs_free(bs); 4805 free(ctx); 4806 } 4807 4808 void 4809 spdk_bs_destroy(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, 4810 void *cb_arg) 4811 { 4812 struct spdk_bs_cpl cpl; 4813 spdk_bs_sequence_t *seq; 4814 struct spdk_bs_load_ctx *ctx; 4815 4816 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Destroying blobstore\n"); 4817 4818 if (!TAILQ_EMPTY(&bs->blobs)) { 4819 SPDK_ERRLOG("Blobstore still has open blobs\n"); 4820 cb_fn(cb_arg, -EBUSY); 4821 return; 4822 } 4823 4824 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 4825 cpl.u.bs_basic.cb_fn = cb_fn; 4826 cpl.u.bs_basic.cb_arg = cb_arg; 4827 4828 ctx = calloc(1, sizeof(*ctx)); 4829 if (!ctx) { 4830 cb_fn(cb_arg, -ENOMEM); 4831 return; 4832 } 4833 4834 ctx->bs = bs; 4835 4836 seq = bs_sequence_start(bs->md_channel, &cpl); 4837 if (!seq) { 4838 free(ctx); 4839 cb_fn(cb_arg, -ENOMEM); 4840 return; 4841 } 4842 4843 /* Write zeroes to the super block */ 4844 bs_sequence_write_zeroes_dev(seq, 4845 bs_page_to_lba(bs, 0), 4846 bs_byte_to_lba(bs, sizeof(struct spdk_bs_super_block)), 4847 bs_destroy_trim_cpl, ctx); 4848 } 4849 4850 /* END spdk_bs_destroy */ 4851 4852 /* START spdk_bs_unload */ 4853 4854 static void 4855 bs_unload_finish(struct spdk_bs_load_ctx *ctx, int bserrno) 4856 { 4857 spdk_bs_sequence_t *seq = ctx->seq; 4858 4859 spdk_free(ctx->super); 4860 4861 /* 4862 * We need to defer calling bs_call_cpl() until after 4863 * dev destruction, so tuck these away for later use. 4864 */ 4865 ctx->bs->unload_err = bserrno; 4866 memcpy(&ctx->bs->unload_cpl, &seq->cpl, sizeof(struct spdk_bs_cpl)); 4867 seq->cpl.type = SPDK_BS_CPL_TYPE_NONE; 4868 4869 bs_sequence_finish(seq, bserrno); 4870 4871 bs_free(ctx->bs); 4872 free(ctx); 4873 } 4874 4875 static void 4876 bs_unload_write_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4877 { 4878 struct spdk_bs_load_ctx *ctx = cb_arg; 4879 4880 bs_unload_finish(ctx, bserrno); 4881 } 4882 4883 static void 4884 bs_unload_write_used_clusters_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4885 { 4886 struct spdk_bs_load_ctx *ctx = cb_arg; 4887 4888 spdk_free(ctx->mask); 4889 4890 if (bserrno != 0) { 4891 bs_unload_finish(ctx, bserrno); 4892 return; 4893 } 4894 4895 ctx->super->clean = 1; 4896 4897 bs_write_super(seq, ctx->bs, ctx->super, bs_unload_write_super_cpl, ctx); 4898 } 4899 4900 static void 4901 bs_unload_write_used_blobids_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4902 { 4903 struct spdk_bs_load_ctx *ctx = cb_arg; 4904 4905 spdk_free(ctx->mask); 4906 ctx->mask = NULL; 4907 4908 if (bserrno != 0) { 4909 bs_unload_finish(ctx, bserrno); 4910 return; 4911 } 4912 4913 bs_write_used_clusters(seq, ctx, bs_unload_write_used_clusters_cpl); 4914 } 4915 4916 static void 4917 bs_unload_write_used_pages_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4918 { 4919 struct spdk_bs_load_ctx *ctx = cb_arg; 4920 4921 spdk_free(ctx->mask); 4922 ctx->mask = NULL; 4923 4924 if (bserrno != 0) { 4925 bs_unload_finish(ctx, bserrno); 4926 return; 4927 } 4928 4929 bs_write_used_blobids(seq, ctx, bs_unload_write_used_blobids_cpl); 4930 } 4931 4932 static void 4933 bs_unload_read_super_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 4934 { 4935 struct spdk_bs_load_ctx *ctx = cb_arg; 4936 4937 if (bserrno != 0) { 4938 bs_unload_finish(ctx, bserrno); 4939 return; 4940 } 4941 4942 bs_write_used_md(seq, cb_arg, bs_unload_write_used_pages_cpl); 4943 } 4944 4945 void 4946 spdk_bs_unload(struct spdk_blob_store *bs, spdk_bs_op_complete cb_fn, void *cb_arg) 4947 { 4948 struct spdk_bs_cpl cpl; 4949 struct spdk_bs_load_ctx *ctx; 4950 4951 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blobstore\n"); 4952 4953 if (!TAILQ_EMPTY(&bs->blobs)) { 4954 SPDK_ERRLOG("Blobstore still has open blobs\n"); 4955 cb_fn(cb_arg, -EBUSY); 4956 return; 4957 } 4958 4959 ctx = calloc(1, sizeof(*ctx)); 4960 if (!ctx) { 4961 cb_fn(cb_arg, -ENOMEM); 4962 return; 4963 } 4964 4965 ctx->bs = bs; 4966 4967 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 4968 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 4969 if (!ctx->super) { 4970 free(ctx); 4971 cb_fn(cb_arg, -ENOMEM); 4972 return; 4973 } 4974 4975 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 4976 cpl.u.bs_basic.cb_fn = cb_fn; 4977 cpl.u.bs_basic.cb_arg = cb_arg; 4978 4979 ctx->seq = bs_sequence_start(bs->md_channel, &cpl); 4980 if (!ctx->seq) { 4981 spdk_free(ctx->super); 4982 free(ctx); 4983 cb_fn(cb_arg, -ENOMEM); 4984 return; 4985 } 4986 4987 /* Read super block */ 4988 bs_sequence_read_dev(ctx->seq, ctx->super, bs_page_to_lba(bs, 0), 4989 bs_byte_to_lba(bs, sizeof(*ctx->super)), 4990 bs_unload_read_super_cpl, ctx); 4991 } 4992 4993 /* END spdk_bs_unload */ 4994 4995 /* START spdk_bs_set_super */ 4996 4997 struct spdk_bs_set_super_ctx { 4998 struct spdk_blob_store *bs; 4999 struct spdk_bs_super_block *super; 5000 }; 5001 5002 static void 5003 bs_set_super_write_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5004 { 5005 struct spdk_bs_set_super_ctx *ctx = cb_arg; 5006 5007 if (bserrno != 0) { 5008 SPDK_ERRLOG("Unable to write to super block of blobstore\n"); 5009 } 5010 5011 spdk_free(ctx->super); 5012 5013 bs_sequence_finish(seq, bserrno); 5014 5015 free(ctx); 5016 } 5017 5018 static void 5019 bs_set_super_read_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5020 { 5021 struct spdk_bs_set_super_ctx *ctx = cb_arg; 5022 5023 if (bserrno != 0) { 5024 SPDK_ERRLOG("Unable to read super block of blobstore\n"); 5025 spdk_free(ctx->super); 5026 bs_sequence_finish(seq, bserrno); 5027 free(ctx); 5028 return; 5029 } 5030 5031 bs_write_super(seq, ctx->bs, ctx->super, bs_set_super_write_cpl, ctx); 5032 } 5033 5034 void 5035 spdk_bs_set_super(struct spdk_blob_store *bs, spdk_blob_id blobid, 5036 spdk_bs_op_complete cb_fn, void *cb_arg) 5037 { 5038 struct spdk_bs_cpl cpl; 5039 spdk_bs_sequence_t *seq; 5040 struct spdk_bs_set_super_ctx *ctx; 5041 5042 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Setting super blob id on blobstore\n"); 5043 5044 ctx = calloc(1, sizeof(*ctx)); 5045 if (!ctx) { 5046 cb_fn(cb_arg, -ENOMEM); 5047 return; 5048 } 5049 5050 ctx->bs = bs; 5051 5052 ctx->super = spdk_zmalloc(sizeof(*ctx->super), 0x1000, NULL, 5053 SPDK_ENV_SOCKET_ID_ANY, SPDK_MALLOC_DMA); 5054 if (!ctx->super) { 5055 free(ctx); 5056 cb_fn(cb_arg, -ENOMEM); 5057 return; 5058 } 5059 5060 cpl.type = SPDK_BS_CPL_TYPE_BS_BASIC; 5061 cpl.u.bs_basic.cb_fn = cb_fn; 5062 cpl.u.bs_basic.cb_arg = cb_arg; 5063 5064 seq = bs_sequence_start(bs->md_channel, &cpl); 5065 if (!seq) { 5066 spdk_free(ctx->super); 5067 free(ctx); 5068 cb_fn(cb_arg, -ENOMEM); 5069 return; 5070 } 5071 5072 bs->super_blob = blobid; 5073 5074 /* Read super block */ 5075 bs_sequence_read_dev(seq, ctx->super, bs_page_to_lba(bs, 0), 5076 bs_byte_to_lba(bs, sizeof(*ctx->super)), 5077 bs_set_super_read_cpl, ctx); 5078 } 5079 5080 /* END spdk_bs_set_super */ 5081 5082 void 5083 spdk_bs_get_super(struct spdk_blob_store *bs, 5084 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5085 { 5086 if (bs->super_blob == SPDK_BLOBID_INVALID) { 5087 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOENT); 5088 } else { 5089 cb_fn(cb_arg, bs->super_blob, 0); 5090 } 5091 } 5092 5093 uint64_t 5094 spdk_bs_get_cluster_size(struct spdk_blob_store *bs) 5095 { 5096 return bs->cluster_sz; 5097 } 5098 5099 uint64_t 5100 spdk_bs_get_page_size(struct spdk_blob_store *bs) 5101 { 5102 return SPDK_BS_PAGE_SIZE; 5103 } 5104 5105 uint64_t 5106 spdk_bs_get_io_unit_size(struct spdk_blob_store *bs) 5107 { 5108 return bs->io_unit_size; 5109 } 5110 5111 uint64_t 5112 spdk_bs_free_cluster_count(struct spdk_blob_store *bs) 5113 { 5114 return bs->num_free_clusters; 5115 } 5116 5117 uint64_t 5118 spdk_bs_total_data_cluster_count(struct spdk_blob_store *bs) 5119 { 5120 return bs->total_data_clusters; 5121 } 5122 5123 static int 5124 bs_register_md_thread(struct spdk_blob_store *bs) 5125 { 5126 bs->md_channel = spdk_get_io_channel(bs); 5127 if (!bs->md_channel) { 5128 SPDK_ERRLOG("Failed to get IO channel.\n"); 5129 return -1; 5130 } 5131 5132 return 0; 5133 } 5134 5135 static int 5136 bs_unregister_md_thread(struct spdk_blob_store *bs) 5137 { 5138 spdk_put_io_channel(bs->md_channel); 5139 5140 return 0; 5141 } 5142 5143 spdk_blob_id spdk_blob_get_id(struct spdk_blob *blob) 5144 { 5145 assert(blob != NULL); 5146 5147 return blob->id; 5148 } 5149 5150 uint64_t spdk_blob_get_num_pages(struct spdk_blob *blob) 5151 { 5152 assert(blob != NULL); 5153 5154 return bs_cluster_to_page(blob->bs, blob->active.num_clusters); 5155 } 5156 5157 uint64_t spdk_blob_get_num_io_units(struct spdk_blob *blob) 5158 { 5159 assert(blob != NULL); 5160 5161 return spdk_blob_get_num_pages(blob) * bs_io_unit_per_page(blob->bs); 5162 } 5163 5164 uint64_t spdk_blob_get_num_clusters(struct spdk_blob *blob) 5165 { 5166 assert(blob != NULL); 5167 5168 return blob->active.num_clusters; 5169 } 5170 5171 /* START spdk_bs_create_blob */ 5172 5173 static void 5174 bs_create_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 5175 { 5176 struct spdk_blob *blob = cb_arg; 5177 uint32_t page_idx = bs_blobid_to_page(blob->id); 5178 5179 if (bserrno != 0) { 5180 spdk_bit_array_clear(blob->bs->used_blobids, page_idx); 5181 bs_release_md_page(blob->bs, page_idx); 5182 } 5183 5184 blob_free(blob); 5185 5186 bs_sequence_finish(seq, bserrno); 5187 } 5188 5189 static int 5190 blob_set_xattrs(struct spdk_blob *blob, const struct spdk_blob_xattr_opts *xattrs, 5191 bool internal) 5192 { 5193 uint64_t i; 5194 size_t value_len = 0; 5195 int rc; 5196 const void *value = NULL; 5197 if (xattrs->count > 0 && xattrs->get_value == NULL) { 5198 return -EINVAL; 5199 } 5200 for (i = 0; i < xattrs->count; i++) { 5201 xattrs->get_value(xattrs->ctx, xattrs->names[i], &value, &value_len); 5202 if (value == NULL || value_len == 0) { 5203 return -EINVAL; 5204 } 5205 rc = blob_set_xattr(blob, xattrs->names[i], value, value_len, internal); 5206 if (rc < 0) { 5207 return rc; 5208 } 5209 } 5210 return 0; 5211 } 5212 5213 static void 5214 bs_create_blob(struct spdk_blob_store *bs, 5215 const struct spdk_blob_opts *opts, 5216 const struct spdk_blob_xattr_opts *internal_xattrs, 5217 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5218 { 5219 struct spdk_blob *blob; 5220 uint32_t page_idx; 5221 struct spdk_bs_cpl cpl; 5222 struct spdk_blob_opts opts_default; 5223 struct spdk_blob_xattr_opts internal_xattrs_default; 5224 spdk_bs_sequence_t *seq; 5225 spdk_blob_id id; 5226 int rc; 5227 5228 assert(spdk_get_thread() == bs->md_thread); 5229 5230 page_idx = spdk_bit_array_find_first_clear(bs->used_md_pages, 0); 5231 if (page_idx == UINT32_MAX) { 5232 cb_fn(cb_arg, 0, -ENOMEM); 5233 return; 5234 } 5235 spdk_bit_array_set(bs->used_blobids, page_idx); 5236 bs_claim_md_page(bs, page_idx); 5237 5238 id = bs_page_to_blobid(page_idx); 5239 5240 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Creating blob with id %lu at page %u\n", id, page_idx); 5241 5242 blob = blob_alloc(bs, id); 5243 if (!blob) { 5244 spdk_bit_array_clear(bs->used_blobids, page_idx); 5245 bs_release_md_page(bs, page_idx); 5246 cb_fn(cb_arg, 0, -ENOMEM); 5247 return; 5248 } 5249 5250 if (!opts) { 5251 spdk_blob_opts_init(&opts_default); 5252 opts = &opts_default; 5253 } 5254 5255 blob->use_extent_table = opts->use_extent_table; 5256 if (blob->use_extent_table) { 5257 blob->invalid_flags |= SPDK_BLOB_EXTENT_TABLE; 5258 } 5259 5260 if (!internal_xattrs) { 5261 blob_xattrs_init(&internal_xattrs_default); 5262 internal_xattrs = &internal_xattrs_default; 5263 } 5264 5265 rc = blob_set_xattrs(blob, &opts->xattrs, false); 5266 if (rc < 0) { 5267 blob_free(blob); 5268 spdk_bit_array_clear(bs->used_blobids, page_idx); 5269 bs_release_md_page(bs, page_idx); 5270 cb_fn(cb_arg, 0, rc); 5271 return; 5272 } 5273 5274 rc = blob_set_xattrs(blob, internal_xattrs, true); 5275 if (rc < 0) { 5276 blob_free(blob); 5277 spdk_bit_array_clear(bs->used_blobids, page_idx); 5278 bs_release_md_page(bs, page_idx); 5279 cb_fn(cb_arg, 0, rc); 5280 return; 5281 } 5282 5283 if (opts->thin_provision) { 5284 blob_set_thin_provision(blob); 5285 } 5286 5287 blob_set_clear_method(blob, opts->clear_method); 5288 5289 rc = blob_resize(blob, opts->num_clusters); 5290 if (rc < 0) { 5291 blob_free(blob); 5292 spdk_bit_array_clear(bs->used_blobids, page_idx); 5293 bs_release_md_page(bs, page_idx); 5294 cb_fn(cb_arg, 0, rc); 5295 return; 5296 } 5297 cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 5298 cpl.u.blobid.cb_fn = cb_fn; 5299 cpl.u.blobid.cb_arg = cb_arg; 5300 cpl.u.blobid.blobid = blob->id; 5301 5302 seq = bs_sequence_start(bs->md_channel, &cpl); 5303 if (!seq) { 5304 blob_free(blob); 5305 spdk_bit_array_clear(bs->used_blobids, page_idx); 5306 bs_release_md_page(bs, page_idx); 5307 cb_fn(cb_arg, 0, -ENOMEM); 5308 return; 5309 } 5310 5311 blob_persist(seq, blob, bs_create_blob_cpl, blob); 5312 } 5313 5314 void spdk_bs_create_blob(struct spdk_blob_store *bs, 5315 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5316 { 5317 bs_create_blob(bs, NULL, NULL, cb_fn, cb_arg); 5318 } 5319 5320 void spdk_bs_create_blob_ext(struct spdk_blob_store *bs, const struct spdk_blob_opts *opts, 5321 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5322 { 5323 bs_create_blob(bs, opts, NULL, cb_fn, cb_arg); 5324 } 5325 5326 /* END spdk_bs_create_blob */ 5327 5328 /* START blob_cleanup */ 5329 5330 struct spdk_clone_snapshot_ctx { 5331 struct spdk_bs_cpl cpl; 5332 int bserrno; 5333 bool frozen; 5334 5335 struct spdk_io_channel *channel; 5336 5337 /* Current cluster for inflate operation */ 5338 uint64_t cluster; 5339 5340 /* For inflation force allocation of all unallocated clusters and remove 5341 * thin-provisioning. Otherwise only decouple parent and keep clone thin. */ 5342 bool allocate_all; 5343 5344 struct { 5345 spdk_blob_id id; 5346 struct spdk_blob *blob; 5347 } original; 5348 struct { 5349 spdk_blob_id id; 5350 struct spdk_blob *blob; 5351 } new; 5352 5353 /* xattrs specified for snapshot/clones only. They have no impact on 5354 * the original blobs xattrs. */ 5355 const struct spdk_blob_xattr_opts *xattrs; 5356 }; 5357 5358 static void 5359 bs_clone_snapshot_cleanup_finish(void *cb_arg, int bserrno) 5360 { 5361 struct spdk_clone_snapshot_ctx *ctx = cb_arg; 5362 struct spdk_bs_cpl *cpl = &ctx->cpl; 5363 5364 if (bserrno != 0) { 5365 if (ctx->bserrno != 0) { 5366 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 5367 } else { 5368 ctx->bserrno = bserrno; 5369 } 5370 } 5371 5372 switch (cpl->type) { 5373 case SPDK_BS_CPL_TYPE_BLOBID: 5374 cpl->u.blobid.cb_fn(cpl->u.blobid.cb_arg, cpl->u.blobid.blobid, ctx->bserrno); 5375 break; 5376 case SPDK_BS_CPL_TYPE_BLOB_BASIC: 5377 cpl->u.blob_basic.cb_fn(cpl->u.blob_basic.cb_arg, ctx->bserrno); 5378 break; 5379 default: 5380 SPDK_UNREACHABLE(); 5381 break; 5382 } 5383 5384 free(ctx); 5385 } 5386 5387 static void 5388 bs_snapshot_unfreeze_cpl(void *cb_arg, int bserrno) 5389 { 5390 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5391 struct spdk_blob *origblob = ctx->original.blob; 5392 5393 if (bserrno != 0) { 5394 if (ctx->bserrno != 0) { 5395 SPDK_ERRLOG("Unfreeze error %d\n", bserrno); 5396 } else { 5397 ctx->bserrno = bserrno; 5398 } 5399 } 5400 5401 ctx->original.id = origblob->id; 5402 origblob->locked_operation_in_progress = false; 5403 5404 spdk_blob_close(origblob, bs_clone_snapshot_cleanup_finish, ctx); 5405 } 5406 5407 static void 5408 bs_clone_snapshot_origblob_cleanup(void *cb_arg, int bserrno) 5409 { 5410 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5411 struct spdk_blob *origblob = ctx->original.blob; 5412 5413 if (bserrno != 0) { 5414 if (ctx->bserrno != 0) { 5415 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 5416 } else { 5417 ctx->bserrno = bserrno; 5418 } 5419 } 5420 5421 if (ctx->frozen) { 5422 /* Unfreeze any outstanding I/O */ 5423 blob_unfreeze_io(origblob, bs_snapshot_unfreeze_cpl, ctx); 5424 } else { 5425 bs_snapshot_unfreeze_cpl(ctx, 0); 5426 } 5427 5428 } 5429 5430 static void 5431 bs_clone_snapshot_newblob_cleanup(struct spdk_clone_snapshot_ctx *ctx, int bserrno) 5432 { 5433 struct spdk_blob *newblob = ctx->new.blob; 5434 5435 if (bserrno != 0) { 5436 if (ctx->bserrno != 0) { 5437 SPDK_ERRLOG("Cleanup error %d\n", bserrno); 5438 } else { 5439 ctx->bserrno = bserrno; 5440 } 5441 } 5442 5443 ctx->new.id = newblob->id; 5444 spdk_blob_close(newblob, bs_clone_snapshot_origblob_cleanup, ctx); 5445 } 5446 5447 /* END blob_cleanup */ 5448 5449 /* START spdk_bs_create_snapshot */ 5450 5451 static void 5452 bs_snapshot_swap_cluster_maps(struct spdk_blob *blob1, struct spdk_blob *blob2) 5453 { 5454 uint64_t *cluster_temp; 5455 uint32_t *extent_page_temp; 5456 5457 cluster_temp = blob1->active.clusters; 5458 blob1->active.clusters = blob2->active.clusters; 5459 blob2->active.clusters = cluster_temp; 5460 5461 extent_page_temp = blob1->active.extent_pages; 5462 blob1->active.extent_pages = blob2->active.extent_pages; 5463 blob2->active.extent_pages = extent_page_temp; 5464 } 5465 5466 static void 5467 bs_snapshot_origblob_sync_cpl(void *cb_arg, int bserrno) 5468 { 5469 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5470 struct spdk_blob *origblob = ctx->original.blob; 5471 struct spdk_blob *newblob = ctx->new.blob; 5472 5473 if (bserrno != 0) { 5474 bs_snapshot_swap_cluster_maps(newblob, origblob); 5475 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5476 return; 5477 } 5478 5479 /* Remove metadata descriptor SNAPSHOT_IN_PROGRESS */ 5480 bserrno = blob_remove_xattr(newblob, SNAPSHOT_IN_PROGRESS, true); 5481 if (bserrno != 0) { 5482 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5483 return; 5484 } 5485 5486 bs_blob_list_add(ctx->original.blob); 5487 5488 spdk_blob_set_read_only(newblob); 5489 5490 /* sync snapshot metadata */ 5491 spdk_blob_sync_md(newblob, bs_clone_snapshot_origblob_cleanup, ctx); 5492 } 5493 5494 static void 5495 bs_snapshot_newblob_sync_cpl(void *cb_arg, int bserrno) 5496 { 5497 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5498 struct spdk_blob *origblob = ctx->original.blob; 5499 struct spdk_blob *newblob = ctx->new.blob; 5500 5501 if (bserrno != 0) { 5502 /* return cluster map back to original */ 5503 bs_snapshot_swap_cluster_maps(newblob, origblob); 5504 5505 /* Newblob md sync failed. Valid clusters are only present in origblob. 5506 * Since I/O is frozen on origblob, not changes to zeroed out cluster map should have occured. 5507 * Newblob needs to be reverted to thin_provisioned state at creation to properly close. */ 5508 blob_set_thin_provision(newblob); 5509 assert(spdk_mem_all_zero(newblob->active.clusters, 5510 newblob->active.num_clusters * sizeof(*newblob->active.clusters))); 5511 assert(spdk_mem_all_zero(newblob->active.extent_pages, 5512 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages))); 5513 5514 bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 5515 return; 5516 } 5517 5518 /* Set internal xattr for snapshot id */ 5519 bserrno = blob_set_xattr(origblob, BLOB_SNAPSHOT, &newblob->id, sizeof(spdk_blob_id), true); 5520 if (bserrno != 0) { 5521 /* return cluster map back to original */ 5522 bs_snapshot_swap_cluster_maps(newblob, origblob); 5523 bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 5524 return; 5525 } 5526 5527 bs_blob_list_remove(origblob); 5528 origblob->parent_id = newblob->id; 5529 5530 /* Create new back_bs_dev for snapshot */ 5531 origblob->back_bs_dev = bs_create_blob_bs_dev(newblob); 5532 if (origblob->back_bs_dev == NULL) { 5533 /* return cluster map back to original */ 5534 bs_snapshot_swap_cluster_maps(newblob, origblob); 5535 bs_clone_snapshot_newblob_cleanup(ctx, -EINVAL); 5536 return; 5537 } 5538 5539 /* set clone blob as thin provisioned */ 5540 blob_set_thin_provision(origblob); 5541 5542 bs_blob_list_add(newblob); 5543 5544 /* sync clone metadata */ 5545 spdk_blob_sync_md(origblob, bs_snapshot_origblob_sync_cpl, ctx); 5546 } 5547 5548 static void 5549 bs_snapshot_freeze_cpl(void *cb_arg, int rc) 5550 { 5551 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5552 struct spdk_blob *origblob = ctx->original.blob; 5553 struct spdk_blob *newblob = ctx->new.blob; 5554 int bserrno; 5555 5556 if (rc != 0) { 5557 bs_clone_snapshot_newblob_cleanup(ctx, rc); 5558 return; 5559 } 5560 5561 ctx->frozen = true; 5562 5563 /* set new back_bs_dev for snapshot */ 5564 newblob->back_bs_dev = origblob->back_bs_dev; 5565 /* Set invalid flags from origblob */ 5566 newblob->invalid_flags = origblob->invalid_flags; 5567 5568 /* inherit parent from original blob if set */ 5569 newblob->parent_id = origblob->parent_id; 5570 if (origblob->parent_id != SPDK_BLOBID_INVALID) { 5571 /* Set internal xattr for snapshot id */ 5572 bserrno = blob_set_xattr(newblob, BLOB_SNAPSHOT, 5573 &origblob->parent_id, sizeof(spdk_blob_id), true); 5574 if (bserrno != 0) { 5575 bs_clone_snapshot_newblob_cleanup(ctx, bserrno); 5576 return; 5577 } 5578 } 5579 5580 /* swap cluster maps */ 5581 bs_snapshot_swap_cluster_maps(newblob, origblob); 5582 5583 /* Set the clear method on the new blob to match the original. */ 5584 blob_set_clear_method(newblob, origblob->clear_method); 5585 5586 /* sync snapshot metadata */ 5587 spdk_blob_sync_md(newblob, bs_snapshot_newblob_sync_cpl, ctx); 5588 } 5589 5590 static void 5591 bs_snapshot_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5592 { 5593 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5594 struct spdk_blob *origblob = ctx->original.blob; 5595 struct spdk_blob *newblob = _blob; 5596 5597 if (bserrno != 0) { 5598 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5599 return; 5600 } 5601 5602 ctx->new.blob = newblob; 5603 assert(spdk_blob_is_thin_provisioned(newblob)); 5604 assert(spdk_mem_all_zero(newblob->active.clusters, 5605 newblob->active.num_clusters * sizeof(*newblob->active.clusters))); 5606 assert(spdk_mem_all_zero(newblob->active.extent_pages, 5607 newblob->active.num_extent_pages * sizeof(*newblob->active.extent_pages))); 5608 5609 blob_freeze_io(origblob, bs_snapshot_freeze_cpl, ctx); 5610 } 5611 5612 static void 5613 bs_snapshot_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 5614 { 5615 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5616 struct spdk_blob *origblob = ctx->original.blob; 5617 5618 if (bserrno != 0) { 5619 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5620 return; 5621 } 5622 5623 ctx->new.id = blobid; 5624 ctx->cpl.u.blobid.blobid = blobid; 5625 5626 spdk_bs_open_blob(origblob->bs, ctx->new.id, bs_snapshot_newblob_open_cpl, ctx); 5627 } 5628 5629 5630 static void 5631 bs_xattr_snapshot(void *arg, const char *name, 5632 const void **value, size_t *value_len) 5633 { 5634 assert(strncmp(name, SNAPSHOT_IN_PROGRESS, sizeof(SNAPSHOT_IN_PROGRESS)) == 0); 5635 5636 struct spdk_blob *blob = (struct spdk_blob *)arg; 5637 *value = &blob->id; 5638 *value_len = sizeof(blob->id); 5639 } 5640 5641 static void 5642 bs_snapshot_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5643 { 5644 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5645 struct spdk_blob_opts opts; 5646 struct spdk_blob_xattr_opts internal_xattrs; 5647 char *xattrs_names[] = { SNAPSHOT_IN_PROGRESS }; 5648 5649 if (bserrno != 0) { 5650 bs_clone_snapshot_cleanup_finish(ctx, bserrno); 5651 return; 5652 } 5653 5654 ctx->original.blob = _blob; 5655 5656 if (_blob->data_ro || _blob->md_ro) { 5657 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot from read only blob with id %lu\n", 5658 _blob->id); 5659 ctx->bserrno = -EINVAL; 5660 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5661 return; 5662 } 5663 5664 if (_blob->locked_operation_in_progress) { 5665 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create snapshot - another operation in progress\n"); 5666 ctx->bserrno = -EBUSY; 5667 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5668 return; 5669 } 5670 5671 _blob->locked_operation_in_progress = true; 5672 5673 spdk_blob_opts_init(&opts); 5674 blob_xattrs_init(&internal_xattrs); 5675 5676 /* Change the size of new blob to the same as in original blob, 5677 * but do not allocate clusters */ 5678 opts.thin_provision = true; 5679 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 5680 opts.use_extent_table = _blob->use_extent_table; 5681 5682 /* If there are any xattrs specified for snapshot, set them now */ 5683 if (ctx->xattrs) { 5684 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 5685 } 5686 /* Set internal xattr SNAPSHOT_IN_PROGRESS */ 5687 internal_xattrs.count = 1; 5688 internal_xattrs.ctx = _blob; 5689 internal_xattrs.names = xattrs_names; 5690 internal_xattrs.get_value = bs_xattr_snapshot; 5691 5692 bs_create_blob(_blob->bs, &opts, &internal_xattrs, 5693 bs_snapshot_newblob_create_cpl, ctx); 5694 } 5695 5696 void spdk_bs_create_snapshot(struct spdk_blob_store *bs, spdk_blob_id blobid, 5697 const struct spdk_blob_xattr_opts *snapshot_xattrs, 5698 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5699 { 5700 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 5701 5702 if (!ctx) { 5703 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 5704 return; 5705 } 5706 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 5707 ctx->cpl.u.blobid.cb_fn = cb_fn; 5708 ctx->cpl.u.blobid.cb_arg = cb_arg; 5709 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 5710 ctx->bserrno = 0; 5711 ctx->frozen = false; 5712 ctx->original.id = blobid; 5713 ctx->xattrs = snapshot_xattrs; 5714 5715 spdk_bs_open_blob(bs, ctx->original.id, bs_snapshot_origblob_open_cpl, ctx); 5716 } 5717 /* END spdk_bs_create_snapshot */ 5718 5719 /* START spdk_bs_create_clone */ 5720 5721 static void 5722 bs_xattr_clone(void *arg, const char *name, 5723 const void **value, size_t *value_len) 5724 { 5725 assert(strncmp(name, BLOB_SNAPSHOT, sizeof(BLOB_SNAPSHOT)) == 0); 5726 5727 struct spdk_blob *blob = (struct spdk_blob *)arg; 5728 *value = &blob->id; 5729 *value_len = sizeof(blob->id); 5730 } 5731 5732 static void 5733 bs_clone_newblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5734 { 5735 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5736 struct spdk_blob *clone = _blob; 5737 5738 ctx->new.blob = clone; 5739 bs_blob_list_add(clone); 5740 5741 spdk_blob_close(clone, bs_clone_snapshot_origblob_cleanup, ctx); 5742 } 5743 5744 static void 5745 bs_clone_newblob_create_cpl(void *cb_arg, spdk_blob_id blobid, int bserrno) 5746 { 5747 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5748 5749 ctx->cpl.u.blobid.blobid = blobid; 5750 spdk_bs_open_blob(ctx->original.blob->bs, blobid, bs_clone_newblob_open_cpl, ctx); 5751 } 5752 5753 static void 5754 bs_clone_origblob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5755 { 5756 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5757 struct spdk_blob_opts opts; 5758 struct spdk_blob_xattr_opts internal_xattrs; 5759 char *xattr_names[] = { BLOB_SNAPSHOT }; 5760 5761 if (bserrno != 0) { 5762 bs_clone_snapshot_cleanup_finish(ctx, bserrno); 5763 return; 5764 } 5765 5766 ctx->original.blob = _blob; 5767 5768 if (!_blob->data_ro || !_blob->md_ro) { 5769 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Clone not from read-only blob\n"); 5770 ctx->bserrno = -EINVAL; 5771 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5772 return; 5773 } 5774 5775 if (_blob->locked_operation_in_progress) { 5776 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot create clone - another operation in progress\n"); 5777 ctx->bserrno = -EBUSY; 5778 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5779 return; 5780 } 5781 5782 _blob->locked_operation_in_progress = true; 5783 5784 spdk_blob_opts_init(&opts); 5785 blob_xattrs_init(&internal_xattrs); 5786 5787 opts.thin_provision = true; 5788 opts.num_clusters = spdk_blob_get_num_clusters(_blob); 5789 opts.use_extent_table = _blob->use_extent_table; 5790 if (ctx->xattrs) { 5791 memcpy(&opts.xattrs, ctx->xattrs, sizeof(*ctx->xattrs)); 5792 } 5793 5794 /* Set internal xattr BLOB_SNAPSHOT */ 5795 internal_xattrs.count = 1; 5796 internal_xattrs.ctx = _blob; 5797 internal_xattrs.names = xattr_names; 5798 internal_xattrs.get_value = bs_xattr_clone; 5799 5800 bs_create_blob(_blob->bs, &opts, &internal_xattrs, 5801 bs_clone_newblob_create_cpl, ctx); 5802 } 5803 5804 void spdk_bs_create_clone(struct spdk_blob_store *bs, spdk_blob_id blobid, 5805 const struct spdk_blob_xattr_opts *clone_xattrs, 5806 spdk_blob_op_with_id_complete cb_fn, void *cb_arg) 5807 { 5808 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 5809 5810 if (!ctx) { 5811 cb_fn(cb_arg, SPDK_BLOBID_INVALID, -ENOMEM); 5812 return; 5813 } 5814 5815 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOBID; 5816 ctx->cpl.u.blobid.cb_fn = cb_fn; 5817 ctx->cpl.u.blobid.cb_arg = cb_arg; 5818 ctx->cpl.u.blobid.blobid = SPDK_BLOBID_INVALID; 5819 ctx->bserrno = 0; 5820 ctx->xattrs = clone_xattrs; 5821 ctx->original.id = blobid; 5822 5823 spdk_bs_open_blob(bs, ctx->original.id, bs_clone_origblob_open_cpl, ctx); 5824 } 5825 5826 /* END spdk_bs_create_clone */ 5827 5828 /* START spdk_bs_inflate_blob */ 5829 5830 static void 5831 bs_inflate_blob_set_parent_cpl(void *cb_arg, struct spdk_blob *_parent, int bserrno) 5832 { 5833 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5834 struct spdk_blob *_blob = ctx->original.blob; 5835 5836 if (bserrno != 0) { 5837 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5838 return; 5839 } 5840 5841 assert(_parent != NULL); 5842 5843 bs_blob_list_remove(_blob); 5844 _blob->parent_id = _parent->id; 5845 blob_set_xattr(_blob, BLOB_SNAPSHOT, &_blob->parent_id, 5846 sizeof(spdk_blob_id), true); 5847 5848 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 5849 _blob->back_bs_dev = bs_create_blob_bs_dev(_parent); 5850 bs_blob_list_add(_blob); 5851 5852 spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx); 5853 } 5854 5855 static void 5856 bs_inflate_blob_done(struct spdk_clone_snapshot_ctx *ctx) 5857 { 5858 struct spdk_blob *_blob = ctx->original.blob; 5859 struct spdk_blob *_parent; 5860 5861 if (ctx->allocate_all) { 5862 /* remove thin provisioning */ 5863 bs_blob_list_remove(_blob); 5864 blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 5865 _blob->invalid_flags = _blob->invalid_flags & ~SPDK_BLOB_THIN_PROV; 5866 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 5867 _blob->back_bs_dev = NULL; 5868 _blob->parent_id = SPDK_BLOBID_INVALID; 5869 } else { 5870 _parent = ((struct spdk_blob_bs_dev *)(_blob->back_bs_dev))->blob; 5871 if (_parent->parent_id != SPDK_BLOBID_INVALID) { 5872 /* We must change the parent of the inflated blob */ 5873 spdk_bs_open_blob(_blob->bs, _parent->parent_id, 5874 bs_inflate_blob_set_parent_cpl, ctx); 5875 return; 5876 } 5877 5878 bs_blob_list_remove(_blob); 5879 blob_remove_xattr(_blob, BLOB_SNAPSHOT, true); 5880 _blob->parent_id = SPDK_BLOBID_INVALID; 5881 _blob->back_bs_dev->destroy(_blob->back_bs_dev); 5882 _blob->back_bs_dev = bs_create_zeroes_dev(); 5883 } 5884 5885 _blob->state = SPDK_BLOB_STATE_DIRTY; 5886 spdk_blob_sync_md(_blob, bs_clone_snapshot_origblob_cleanup, ctx); 5887 } 5888 5889 /* Check if cluster needs allocation */ 5890 static inline bool 5891 bs_cluster_needs_allocation(struct spdk_blob *blob, uint64_t cluster, bool allocate_all) 5892 { 5893 struct spdk_blob_bs_dev *b; 5894 5895 assert(blob != NULL); 5896 5897 if (blob->active.clusters[cluster] != 0) { 5898 /* Cluster is already allocated */ 5899 return false; 5900 } 5901 5902 if (blob->parent_id == SPDK_BLOBID_INVALID) { 5903 /* Blob have no parent blob */ 5904 return allocate_all; 5905 } 5906 5907 b = (struct spdk_blob_bs_dev *)blob->back_bs_dev; 5908 return (allocate_all || b->blob->active.clusters[cluster] != 0); 5909 } 5910 5911 static void 5912 bs_inflate_blob_touch_next(void *cb_arg, int bserrno) 5913 { 5914 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5915 struct spdk_blob *_blob = ctx->original.blob; 5916 uint64_t offset; 5917 5918 if (bserrno != 0) { 5919 bs_clone_snapshot_origblob_cleanup(ctx, bserrno); 5920 return; 5921 } 5922 5923 for (; ctx->cluster < _blob->active.num_clusters; ctx->cluster++) { 5924 if (bs_cluster_needs_allocation(_blob, ctx->cluster, ctx->allocate_all)) { 5925 break; 5926 } 5927 } 5928 5929 if (ctx->cluster < _blob->active.num_clusters) { 5930 offset = bs_cluster_to_lba(_blob->bs, ctx->cluster); 5931 5932 /* We may safely increment a cluster before write */ 5933 ctx->cluster++; 5934 5935 /* Use zero length write to touch a cluster */ 5936 spdk_blob_io_write(_blob, ctx->channel, NULL, offset, 0, 5937 bs_inflate_blob_touch_next, ctx); 5938 } else { 5939 bs_inflate_blob_done(ctx); 5940 } 5941 } 5942 5943 static void 5944 bs_inflate_blob_open_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 5945 { 5946 struct spdk_clone_snapshot_ctx *ctx = (struct spdk_clone_snapshot_ctx *)cb_arg; 5947 uint64_t clusters_needed; 5948 uint64_t i; 5949 5950 if (bserrno != 0) { 5951 bs_clone_snapshot_cleanup_finish(ctx, bserrno); 5952 return; 5953 } 5954 5955 ctx->original.blob = _blob; 5956 5957 if (_blob->locked_operation_in_progress) { 5958 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot inflate blob - another operation in progress\n"); 5959 ctx->bserrno = -EBUSY; 5960 spdk_blob_close(_blob, bs_clone_snapshot_cleanup_finish, ctx); 5961 return; 5962 } 5963 5964 _blob->locked_operation_in_progress = true; 5965 5966 if (!ctx->allocate_all && _blob->parent_id == SPDK_BLOBID_INVALID) { 5967 /* This blob have no parent, so we cannot decouple it. */ 5968 SPDK_ERRLOG("Cannot decouple parent of blob with no parent.\n"); 5969 bs_clone_snapshot_origblob_cleanup(ctx, -EINVAL); 5970 return; 5971 } 5972 5973 if (spdk_blob_is_thin_provisioned(_blob) == false) { 5974 /* This is not thin provisioned blob. No need to inflate. */ 5975 bs_clone_snapshot_origblob_cleanup(ctx, 0); 5976 return; 5977 } 5978 5979 /* Do two passes - one to verify that we can obtain enough clusters 5980 * and another to actually claim them. 5981 */ 5982 clusters_needed = 0; 5983 for (i = 0; i < _blob->active.num_clusters; i++) { 5984 if (bs_cluster_needs_allocation(_blob, i, ctx->allocate_all)) { 5985 clusters_needed++; 5986 } 5987 } 5988 5989 if (clusters_needed > _blob->bs->num_free_clusters) { 5990 /* Not enough free clusters. Cannot satisfy the request. */ 5991 bs_clone_snapshot_origblob_cleanup(ctx, -ENOSPC); 5992 return; 5993 } 5994 5995 ctx->cluster = 0; 5996 bs_inflate_blob_touch_next(ctx, 0); 5997 } 5998 5999 static void 6000 bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 6001 spdk_blob_id blobid, bool allocate_all, spdk_blob_op_complete cb_fn, void *cb_arg) 6002 { 6003 struct spdk_clone_snapshot_ctx *ctx = calloc(1, sizeof(*ctx)); 6004 6005 if (!ctx) { 6006 cb_fn(cb_arg, -ENOMEM); 6007 return; 6008 } 6009 ctx->cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 6010 ctx->cpl.u.bs_basic.cb_fn = cb_fn; 6011 ctx->cpl.u.bs_basic.cb_arg = cb_arg; 6012 ctx->bserrno = 0; 6013 ctx->original.id = blobid; 6014 ctx->channel = channel; 6015 ctx->allocate_all = allocate_all; 6016 6017 spdk_bs_open_blob(bs, ctx->original.id, bs_inflate_blob_open_cpl, ctx); 6018 } 6019 6020 void 6021 spdk_bs_inflate_blob(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 6022 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 6023 { 6024 bs_inflate_blob(bs, channel, blobid, true, cb_fn, cb_arg); 6025 } 6026 6027 void 6028 spdk_bs_blob_decouple_parent(struct spdk_blob_store *bs, struct spdk_io_channel *channel, 6029 spdk_blob_id blobid, spdk_blob_op_complete cb_fn, void *cb_arg) 6030 { 6031 bs_inflate_blob(bs, channel, blobid, false, cb_fn, cb_arg); 6032 } 6033 /* END spdk_bs_inflate_blob */ 6034 6035 /* START spdk_blob_resize */ 6036 struct spdk_bs_resize_ctx { 6037 spdk_blob_op_complete cb_fn; 6038 void *cb_arg; 6039 struct spdk_blob *blob; 6040 uint64_t sz; 6041 int rc; 6042 }; 6043 6044 static void 6045 bs_resize_unfreeze_cpl(void *cb_arg, int rc) 6046 { 6047 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 6048 6049 if (rc != 0) { 6050 SPDK_ERRLOG("Unfreeze failed, rc=%d\n", rc); 6051 } 6052 6053 if (ctx->rc != 0) { 6054 SPDK_ERRLOG("Unfreeze failed, ctx->rc=%d\n", ctx->rc); 6055 rc = ctx->rc; 6056 } 6057 6058 ctx->blob->locked_operation_in_progress = false; 6059 6060 ctx->cb_fn(ctx->cb_arg, rc); 6061 free(ctx); 6062 } 6063 6064 static void 6065 bs_resize_freeze_cpl(void *cb_arg, int rc) 6066 { 6067 struct spdk_bs_resize_ctx *ctx = (struct spdk_bs_resize_ctx *)cb_arg; 6068 6069 if (rc != 0) { 6070 ctx->blob->locked_operation_in_progress = false; 6071 ctx->cb_fn(ctx->cb_arg, rc); 6072 free(ctx); 6073 return; 6074 } 6075 6076 ctx->rc = blob_resize(ctx->blob, ctx->sz); 6077 6078 blob_unfreeze_io(ctx->blob, bs_resize_unfreeze_cpl, ctx); 6079 } 6080 6081 void 6082 spdk_blob_resize(struct spdk_blob *blob, uint64_t sz, spdk_blob_op_complete cb_fn, void *cb_arg) 6083 { 6084 struct spdk_bs_resize_ctx *ctx; 6085 6086 blob_verify_md_op(blob); 6087 6088 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Resizing blob %lu to %lu clusters\n", blob->id, sz); 6089 6090 if (blob->md_ro) { 6091 cb_fn(cb_arg, -EPERM); 6092 return; 6093 } 6094 6095 if (sz == blob->active.num_clusters) { 6096 cb_fn(cb_arg, 0); 6097 return; 6098 } 6099 6100 if (blob->locked_operation_in_progress) { 6101 cb_fn(cb_arg, -EBUSY); 6102 return; 6103 } 6104 6105 ctx = calloc(1, sizeof(*ctx)); 6106 if (!ctx) { 6107 cb_fn(cb_arg, -ENOMEM); 6108 return; 6109 } 6110 6111 blob->locked_operation_in_progress = true; 6112 ctx->cb_fn = cb_fn; 6113 ctx->cb_arg = cb_arg; 6114 ctx->blob = blob; 6115 ctx->sz = sz; 6116 blob_freeze_io(blob, bs_resize_freeze_cpl, ctx); 6117 } 6118 6119 /* END spdk_blob_resize */ 6120 6121 6122 /* START spdk_bs_delete_blob */ 6123 6124 static void 6125 bs_delete_close_cpl(void *cb_arg, int bserrno) 6126 { 6127 spdk_bs_sequence_t *seq = cb_arg; 6128 6129 bs_sequence_finish(seq, bserrno); 6130 } 6131 6132 static void 6133 bs_delete_persist_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 6134 { 6135 struct spdk_blob *blob = cb_arg; 6136 6137 if (bserrno != 0) { 6138 /* 6139 * We already removed this blob from the blobstore tailq, so 6140 * we need to free it here since this is the last reference 6141 * to it. 6142 */ 6143 blob_free(blob); 6144 bs_delete_close_cpl(seq, bserrno); 6145 return; 6146 } 6147 6148 /* 6149 * This will immediately decrement the ref_count and call 6150 * the completion routine since the metadata state is clean. 6151 * By calling spdk_blob_close, we reduce the number of call 6152 * points into code that touches the blob->open_ref count 6153 * and the blobstore's blob list. 6154 */ 6155 spdk_blob_close(blob, bs_delete_close_cpl, seq); 6156 } 6157 6158 struct delete_snapshot_ctx { 6159 struct spdk_blob_list *parent_snapshot_entry; 6160 struct spdk_blob *snapshot; 6161 bool snapshot_md_ro; 6162 struct spdk_blob *clone; 6163 bool clone_md_ro; 6164 spdk_blob_op_with_handle_complete cb_fn; 6165 void *cb_arg; 6166 int bserrno; 6167 }; 6168 6169 static void 6170 delete_blob_cleanup_finish(void *cb_arg, int bserrno) 6171 { 6172 struct delete_snapshot_ctx *ctx = cb_arg; 6173 6174 if (bserrno != 0) { 6175 SPDK_ERRLOG("Snapshot cleanup error %d\n", bserrno); 6176 } 6177 6178 assert(ctx != NULL); 6179 6180 if (bserrno != 0 && ctx->bserrno == 0) { 6181 ctx->bserrno = bserrno; 6182 } 6183 6184 ctx->cb_fn(ctx->cb_arg, ctx->snapshot, ctx->bserrno); 6185 free(ctx); 6186 } 6187 6188 static void 6189 delete_snapshot_cleanup_snapshot(void *cb_arg, int bserrno) 6190 { 6191 struct delete_snapshot_ctx *ctx = cb_arg; 6192 6193 if (bserrno != 0) { 6194 ctx->bserrno = bserrno; 6195 SPDK_ERRLOG("Clone cleanup error %d\n", bserrno); 6196 } 6197 6198 if (ctx->bserrno != 0) { 6199 assert(blob_lookup(ctx->snapshot->bs, ctx->snapshot->id) == NULL); 6200 TAILQ_INSERT_HEAD(&ctx->snapshot->bs->blobs, ctx->snapshot, link); 6201 spdk_bit_array_set(ctx->snapshot->bs->open_blobids, ctx->snapshot->id); 6202 } 6203 6204 ctx->snapshot->locked_operation_in_progress = false; 6205 ctx->snapshot->md_ro = ctx->snapshot_md_ro; 6206 6207 spdk_blob_close(ctx->snapshot, delete_blob_cleanup_finish, ctx); 6208 } 6209 6210 static void 6211 delete_snapshot_cleanup_clone(void *cb_arg, int bserrno) 6212 { 6213 struct delete_snapshot_ctx *ctx = cb_arg; 6214 6215 ctx->clone->locked_operation_in_progress = false; 6216 ctx->clone->md_ro = ctx->clone_md_ro; 6217 6218 spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx); 6219 } 6220 6221 static void 6222 delete_snapshot_unfreeze_cpl(void *cb_arg, int bserrno) 6223 { 6224 struct delete_snapshot_ctx *ctx = cb_arg; 6225 6226 if (bserrno) { 6227 ctx->bserrno = bserrno; 6228 delete_snapshot_cleanup_clone(ctx, 0); 6229 return; 6230 } 6231 6232 ctx->clone->locked_operation_in_progress = false; 6233 spdk_blob_close(ctx->clone, delete_blob_cleanup_finish, ctx); 6234 } 6235 6236 static void 6237 delete_snapshot_sync_snapshot_cpl(void *cb_arg, int bserrno) 6238 { 6239 struct delete_snapshot_ctx *ctx = cb_arg; 6240 struct spdk_blob_list *parent_snapshot_entry = NULL; 6241 struct spdk_blob_list *snapshot_entry = NULL; 6242 struct spdk_blob_list *clone_entry = NULL; 6243 struct spdk_blob_list *snapshot_clone_entry = NULL; 6244 6245 if (bserrno) { 6246 SPDK_ERRLOG("Failed to sync MD on blob\n"); 6247 ctx->bserrno = bserrno; 6248 delete_snapshot_cleanup_clone(ctx, 0); 6249 return; 6250 } 6251 6252 /* Get snapshot entry for the snapshot we want to remove */ 6253 snapshot_entry = bs_get_snapshot_entry(ctx->snapshot->bs, ctx->snapshot->id); 6254 6255 assert(snapshot_entry != NULL); 6256 6257 /* Remove clone entry in this snapshot (at this point there can be only one clone) */ 6258 clone_entry = TAILQ_FIRST(&snapshot_entry->clones); 6259 assert(clone_entry != NULL); 6260 TAILQ_REMOVE(&snapshot_entry->clones, clone_entry, link); 6261 snapshot_entry->clone_count--; 6262 assert(TAILQ_EMPTY(&snapshot_entry->clones)); 6263 6264 if (ctx->snapshot->parent_id != SPDK_BLOBID_INVALID) { 6265 /* This snapshot is at the same time a clone of another snapshot - we need to 6266 * update parent snapshot (remove current clone, add new one inherited from 6267 * the snapshot that is being removed) */ 6268 6269 /* Get snapshot entry for parent snapshot and clone entry within that snapshot for 6270 * snapshot that we are removing */ 6271 blob_get_snapshot_and_clone_entries(ctx->snapshot, &parent_snapshot_entry, 6272 &snapshot_clone_entry); 6273 6274 /* Switch clone entry in parent snapshot */ 6275 TAILQ_INSERT_TAIL(&parent_snapshot_entry->clones, clone_entry, link); 6276 TAILQ_REMOVE(&parent_snapshot_entry->clones, snapshot_clone_entry, link); 6277 free(snapshot_clone_entry); 6278 } else { 6279 /* No parent snapshot - just remove clone entry */ 6280 free(clone_entry); 6281 } 6282 6283 /* Restore md_ro flags */ 6284 ctx->clone->md_ro = ctx->clone_md_ro; 6285 ctx->snapshot->md_ro = ctx->snapshot_md_ro; 6286 6287 blob_unfreeze_io(ctx->clone, delete_snapshot_unfreeze_cpl, ctx); 6288 } 6289 6290 static void 6291 delete_snapshot_sync_clone_cpl(void *cb_arg, int bserrno) 6292 { 6293 struct delete_snapshot_ctx *ctx = cb_arg; 6294 uint64_t i; 6295 6296 ctx->snapshot->md_ro = false; 6297 6298 if (bserrno) { 6299 SPDK_ERRLOG("Failed to sync MD on clone\n"); 6300 ctx->bserrno = bserrno; 6301 6302 /* Restore snapshot to previous state */ 6303 bserrno = blob_remove_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, true); 6304 if (bserrno != 0) { 6305 delete_snapshot_cleanup_clone(ctx, bserrno); 6306 return; 6307 } 6308 6309 spdk_blob_sync_md(ctx->snapshot, delete_snapshot_cleanup_clone, ctx); 6310 return; 6311 } 6312 6313 /* Clear cluster map entries for snapshot */ 6314 for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) { 6315 if (ctx->clone->active.clusters[i] == ctx->snapshot->active.clusters[i]) { 6316 ctx->snapshot->active.clusters[i] = 0; 6317 } 6318 } 6319 for (i = 0; i < ctx->snapshot->active.num_extent_pages && 6320 i < ctx->clone->active.num_extent_pages; i++) { 6321 if (ctx->clone->active.extent_pages[i] == ctx->snapshot->active.extent_pages[i]) { 6322 ctx->snapshot->active.extent_pages[i] = 0; 6323 } 6324 } 6325 6326 blob_set_thin_provision(ctx->snapshot); 6327 ctx->snapshot->state = SPDK_BLOB_STATE_DIRTY; 6328 6329 if (ctx->parent_snapshot_entry != NULL) { 6330 ctx->snapshot->back_bs_dev = NULL; 6331 } 6332 6333 spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_cpl, ctx); 6334 } 6335 6336 static void 6337 delete_snapshot_sync_snapshot_xattr_cpl(void *cb_arg, int bserrno) 6338 { 6339 struct delete_snapshot_ctx *ctx = cb_arg; 6340 uint64_t i; 6341 6342 /* Temporarily override md_ro flag for clone for MD modification */ 6343 ctx->clone_md_ro = ctx->clone->md_ro; 6344 ctx->clone->md_ro = false; 6345 6346 if (bserrno) { 6347 SPDK_ERRLOG("Failed to sync MD with xattr on blob\n"); 6348 ctx->bserrno = bserrno; 6349 delete_snapshot_cleanup_clone(ctx, 0); 6350 return; 6351 } 6352 6353 /* Copy snapshot map to clone map (only unallocated clusters in clone) */ 6354 for (i = 0; i < ctx->snapshot->active.num_clusters && i < ctx->clone->active.num_clusters; i++) { 6355 if (ctx->clone->active.clusters[i] == 0) { 6356 ctx->clone->active.clusters[i] = ctx->snapshot->active.clusters[i]; 6357 } 6358 } 6359 for (i = 0; i < ctx->snapshot->active.num_extent_pages && 6360 i < ctx->clone->active.num_extent_pages; i++) { 6361 if (ctx->clone->active.extent_pages[i] == 0) { 6362 ctx->clone->active.extent_pages[i] = ctx->snapshot->active.extent_pages[i]; 6363 } 6364 } 6365 6366 /* Delete old backing bs_dev from clone (related to snapshot that will be removed) */ 6367 ctx->clone->back_bs_dev->destroy(ctx->clone->back_bs_dev); 6368 6369 /* Set/remove snapshot xattr and switch parent ID and backing bs_dev on clone... */ 6370 if (ctx->parent_snapshot_entry != NULL) { 6371 /* ...to parent snapshot */ 6372 ctx->clone->parent_id = ctx->parent_snapshot_entry->id; 6373 ctx->clone->back_bs_dev = ctx->snapshot->back_bs_dev; 6374 blob_set_xattr(ctx->clone, BLOB_SNAPSHOT, &ctx->parent_snapshot_entry->id, 6375 sizeof(spdk_blob_id), 6376 true); 6377 } else { 6378 /* ...to blobid invalid and zeroes dev */ 6379 ctx->clone->parent_id = SPDK_BLOBID_INVALID; 6380 ctx->clone->back_bs_dev = bs_create_zeroes_dev(); 6381 blob_remove_xattr(ctx->clone, BLOB_SNAPSHOT, true); 6382 } 6383 6384 spdk_blob_sync_md(ctx->clone, delete_snapshot_sync_clone_cpl, ctx); 6385 } 6386 6387 static void 6388 delete_snapshot_freeze_io_cb(void *cb_arg, int bserrno) 6389 { 6390 struct delete_snapshot_ctx *ctx = cb_arg; 6391 6392 if (bserrno) { 6393 SPDK_ERRLOG("Failed to freeze I/O on clone\n"); 6394 ctx->bserrno = bserrno; 6395 delete_snapshot_cleanup_clone(ctx, 0); 6396 return; 6397 } 6398 6399 /* Temporarily override md_ro flag for snapshot for MD modification */ 6400 ctx->snapshot_md_ro = ctx->snapshot->md_ro; 6401 ctx->snapshot->md_ro = false; 6402 6403 /* Mark blob as pending for removal for power failure safety, use clone id for recovery */ 6404 ctx->bserrno = blob_set_xattr(ctx->snapshot, SNAPSHOT_PENDING_REMOVAL, &ctx->clone->id, 6405 sizeof(spdk_blob_id), true); 6406 if (ctx->bserrno != 0) { 6407 delete_snapshot_cleanup_clone(ctx, 0); 6408 return; 6409 } 6410 6411 spdk_blob_sync_md(ctx->snapshot, delete_snapshot_sync_snapshot_xattr_cpl, ctx); 6412 } 6413 6414 static void 6415 delete_snapshot_open_clone_cb(void *cb_arg, struct spdk_blob *clone, int bserrno) 6416 { 6417 struct delete_snapshot_ctx *ctx = cb_arg; 6418 6419 if (bserrno) { 6420 SPDK_ERRLOG("Failed to open clone\n"); 6421 ctx->bserrno = bserrno; 6422 delete_snapshot_cleanup_snapshot(ctx, 0); 6423 return; 6424 } 6425 6426 ctx->clone = clone; 6427 6428 if (clone->locked_operation_in_progress) { 6429 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress on its clone\n"); 6430 ctx->bserrno = -EBUSY; 6431 spdk_blob_close(ctx->clone, delete_snapshot_cleanup_snapshot, ctx); 6432 return; 6433 } 6434 6435 clone->locked_operation_in_progress = true; 6436 6437 blob_freeze_io(clone, delete_snapshot_freeze_io_cb, ctx); 6438 } 6439 6440 static void 6441 update_clone_on_snapshot_deletion(struct spdk_blob *snapshot, struct delete_snapshot_ctx *ctx) 6442 { 6443 struct spdk_blob_list *snapshot_entry = NULL; 6444 struct spdk_blob_list *clone_entry = NULL; 6445 struct spdk_blob_list *snapshot_clone_entry = NULL; 6446 6447 /* Get snapshot entry for the snapshot we want to remove */ 6448 snapshot_entry = bs_get_snapshot_entry(snapshot->bs, snapshot->id); 6449 6450 assert(snapshot_entry != NULL); 6451 6452 /* Get clone of the snapshot (at this point there can be only one clone) */ 6453 clone_entry = TAILQ_FIRST(&snapshot_entry->clones); 6454 assert(snapshot_entry->clone_count == 1); 6455 assert(clone_entry != NULL); 6456 6457 /* Get snapshot entry for parent snapshot and clone entry within that snapshot for 6458 * snapshot that we are removing */ 6459 blob_get_snapshot_and_clone_entries(snapshot, &ctx->parent_snapshot_entry, 6460 &snapshot_clone_entry); 6461 6462 spdk_bs_open_blob(snapshot->bs, clone_entry->id, delete_snapshot_open_clone_cb, ctx); 6463 } 6464 6465 static void 6466 bs_delete_blob_finish(void *cb_arg, struct spdk_blob *blob, int bserrno) 6467 { 6468 spdk_bs_sequence_t *seq = cb_arg; 6469 struct spdk_blob_list *snapshot_entry = NULL; 6470 uint32_t page_num; 6471 6472 if (bserrno) { 6473 SPDK_ERRLOG("Failed to remove blob\n"); 6474 bs_sequence_finish(seq, bserrno); 6475 return; 6476 } 6477 6478 /* Remove snapshot from the list */ 6479 snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id); 6480 if (snapshot_entry != NULL) { 6481 TAILQ_REMOVE(&blob->bs->snapshots, snapshot_entry, link); 6482 free(snapshot_entry); 6483 } 6484 6485 page_num = bs_blobid_to_page(blob->id); 6486 spdk_bit_array_clear(blob->bs->used_blobids, page_num); 6487 blob->state = SPDK_BLOB_STATE_DIRTY; 6488 blob->active.num_pages = 0; 6489 blob_resize(blob, 0); 6490 6491 blob_persist(seq, blob, bs_delete_persist_cpl, blob); 6492 } 6493 6494 static int 6495 bs_is_blob_deletable(struct spdk_blob *blob, bool *update_clone) 6496 { 6497 struct spdk_blob_list *snapshot_entry = NULL; 6498 struct spdk_blob_list *clone_entry = NULL; 6499 struct spdk_blob *clone = NULL; 6500 bool has_one_clone = false; 6501 6502 /* Check if this is a snapshot with clones */ 6503 snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id); 6504 if (snapshot_entry != NULL) { 6505 if (snapshot_entry->clone_count > 1) { 6506 SPDK_ERRLOG("Cannot remove snapshot with more than one clone\n"); 6507 return -EBUSY; 6508 } else if (snapshot_entry->clone_count == 1) { 6509 has_one_clone = true; 6510 } 6511 } 6512 6513 /* Check if someone has this blob open (besides this delete context): 6514 * - open_ref = 1 - only this context opened blob, so it is ok to remove it 6515 * - open_ref <= 2 && has_one_clone = true - clone is holding snapshot 6516 * and that is ok, because we will update it accordingly */ 6517 if (blob->open_ref <= 2 && has_one_clone) { 6518 clone_entry = TAILQ_FIRST(&snapshot_entry->clones); 6519 assert(clone_entry != NULL); 6520 clone = blob_lookup(blob->bs, clone_entry->id); 6521 6522 if (blob->open_ref == 2 && clone == NULL) { 6523 /* Clone is closed and someone else opened this blob */ 6524 SPDK_ERRLOG("Cannot remove snapshot because it is open\n"); 6525 return -EBUSY; 6526 } 6527 6528 *update_clone = true; 6529 return 0; 6530 } 6531 6532 if (blob->open_ref > 1) { 6533 SPDK_ERRLOG("Cannot remove snapshot because it is open\n"); 6534 return -EBUSY; 6535 } 6536 6537 assert(has_one_clone == false); 6538 *update_clone = false; 6539 return 0; 6540 } 6541 6542 static void 6543 bs_delete_enomem_close_cpl(void *cb_arg, int bserrno) 6544 { 6545 spdk_bs_sequence_t *seq = cb_arg; 6546 6547 bs_sequence_finish(seq, -ENOMEM); 6548 } 6549 6550 static void 6551 bs_delete_open_cpl(void *cb_arg, struct spdk_blob *blob, int bserrno) 6552 { 6553 spdk_bs_sequence_t *seq = cb_arg; 6554 struct delete_snapshot_ctx *ctx; 6555 bool update_clone = false; 6556 6557 if (bserrno != 0) { 6558 bs_sequence_finish(seq, bserrno); 6559 return; 6560 } 6561 6562 blob_verify_md_op(blob); 6563 6564 ctx = calloc(1, sizeof(*ctx)); 6565 if (ctx == NULL) { 6566 spdk_blob_close(blob, bs_delete_enomem_close_cpl, seq); 6567 return; 6568 } 6569 6570 ctx->snapshot = blob; 6571 ctx->cb_fn = bs_delete_blob_finish; 6572 ctx->cb_arg = seq; 6573 6574 /* Check if blob can be removed and if it is a snapshot with clone on top of it */ 6575 ctx->bserrno = bs_is_blob_deletable(blob, &update_clone); 6576 if (ctx->bserrno) { 6577 spdk_blob_close(blob, delete_blob_cleanup_finish, ctx); 6578 return; 6579 } 6580 6581 if (blob->locked_operation_in_progress) { 6582 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Cannot remove blob - another operation in progress\n"); 6583 ctx->bserrno = -EBUSY; 6584 spdk_blob_close(blob, delete_blob_cleanup_finish, ctx); 6585 return; 6586 } 6587 6588 blob->locked_operation_in_progress = true; 6589 6590 /* 6591 * Remove the blob from the blob_store list now, to ensure it does not 6592 * get returned after this point by blob_lookup(). 6593 */ 6594 spdk_bit_array_clear(blob->bs->open_blobids, blob->id); 6595 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 6596 6597 if (update_clone) { 6598 /* This blob is a snapshot with active clone - update clone first */ 6599 update_clone_on_snapshot_deletion(blob, ctx); 6600 } else { 6601 /* This blob does not have any clones - just remove it */ 6602 bs_blob_list_remove(blob); 6603 bs_delete_blob_finish(seq, blob, 0); 6604 free(ctx); 6605 } 6606 } 6607 6608 void 6609 spdk_bs_delete_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 6610 spdk_blob_op_complete cb_fn, void *cb_arg) 6611 { 6612 struct spdk_bs_cpl cpl; 6613 spdk_bs_sequence_t *seq; 6614 6615 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Deleting blob %lu\n", blobid); 6616 6617 assert(spdk_get_thread() == bs->md_thread); 6618 6619 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 6620 cpl.u.blob_basic.cb_fn = cb_fn; 6621 cpl.u.blob_basic.cb_arg = cb_arg; 6622 6623 seq = bs_sequence_start(bs->md_channel, &cpl); 6624 if (!seq) { 6625 cb_fn(cb_arg, -ENOMEM); 6626 return; 6627 } 6628 6629 spdk_bs_open_blob(bs, blobid, bs_delete_open_cpl, seq); 6630 } 6631 6632 /* END spdk_bs_delete_blob */ 6633 6634 /* START spdk_bs_open_blob */ 6635 6636 static void 6637 bs_open_blob_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 6638 { 6639 struct spdk_blob *blob = cb_arg; 6640 struct spdk_blob *existing; 6641 6642 if (bserrno != 0) { 6643 blob_free(blob); 6644 seq->cpl.u.blob_handle.blob = NULL; 6645 bs_sequence_finish(seq, bserrno); 6646 return; 6647 } 6648 6649 existing = blob_lookup(blob->bs, blob->id); 6650 if (existing) { 6651 blob_free(blob); 6652 existing->open_ref++; 6653 seq->cpl.u.blob_handle.blob = existing; 6654 bs_sequence_finish(seq, 0); 6655 return; 6656 } 6657 6658 blob->open_ref++; 6659 6660 spdk_bit_array_set(blob->bs->open_blobids, blob->id); 6661 TAILQ_INSERT_HEAD(&blob->bs->blobs, blob, link); 6662 6663 bs_sequence_finish(seq, bserrno); 6664 } 6665 6666 static void 6667 bs_open_blob(struct spdk_blob_store *bs, 6668 spdk_blob_id blobid, 6669 struct spdk_blob_open_opts *opts, 6670 spdk_blob_op_with_handle_complete cb_fn, 6671 void *cb_arg) 6672 { 6673 struct spdk_blob *blob; 6674 struct spdk_bs_cpl cpl; 6675 struct spdk_blob_open_opts opts_default; 6676 spdk_bs_sequence_t *seq; 6677 uint32_t page_num; 6678 6679 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Opening blob %lu\n", blobid); 6680 assert(spdk_get_thread() == bs->md_thread); 6681 6682 page_num = bs_blobid_to_page(blobid); 6683 if (spdk_bit_array_get(bs->used_blobids, page_num) == false) { 6684 /* Invalid blobid */ 6685 cb_fn(cb_arg, NULL, -ENOENT); 6686 return; 6687 } 6688 6689 blob = blob_lookup(bs, blobid); 6690 if (blob) { 6691 blob->open_ref++; 6692 cb_fn(cb_arg, blob, 0); 6693 return; 6694 } 6695 6696 blob = blob_alloc(bs, blobid); 6697 if (!blob) { 6698 cb_fn(cb_arg, NULL, -ENOMEM); 6699 return; 6700 } 6701 6702 if (!opts) { 6703 spdk_blob_open_opts_init(&opts_default); 6704 opts = &opts_default; 6705 } 6706 6707 blob->clear_method = opts->clear_method; 6708 6709 cpl.type = SPDK_BS_CPL_TYPE_BLOB_HANDLE; 6710 cpl.u.blob_handle.cb_fn = cb_fn; 6711 cpl.u.blob_handle.cb_arg = cb_arg; 6712 cpl.u.blob_handle.blob = blob; 6713 6714 seq = bs_sequence_start(bs->md_channel, &cpl); 6715 if (!seq) { 6716 blob_free(blob); 6717 cb_fn(cb_arg, NULL, -ENOMEM); 6718 return; 6719 } 6720 6721 blob_load(seq, blob, bs_open_blob_cpl, blob); 6722 } 6723 6724 void spdk_bs_open_blob(struct spdk_blob_store *bs, spdk_blob_id blobid, 6725 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 6726 { 6727 bs_open_blob(bs, blobid, NULL, cb_fn, cb_arg); 6728 } 6729 6730 void spdk_bs_open_blob_ext(struct spdk_blob_store *bs, spdk_blob_id blobid, 6731 struct spdk_blob_open_opts *opts, spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 6732 { 6733 bs_open_blob(bs, blobid, opts, cb_fn, cb_arg); 6734 } 6735 6736 /* END spdk_bs_open_blob */ 6737 6738 /* START spdk_blob_set_read_only */ 6739 int spdk_blob_set_read_only(struct spdk_blob *blob) 6740 { 6741 blob_verify_md_op(blob); 6742 6743 blob->data_ro_flags |= SPDK_BLOB_READ_ONLY; 6744 6745 blob->state = SPDK_BLOB_STATE_DIRTY; 6746 return 0; 6747 } 6748 /* END spdk_blob_set_read_only */ 6749 6750 /* START spdk_blob_sync_md */ 6751 6752 static void 6753 blob_sync_md_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 6754 { 6755 struct spdk_blob *blob = cb_arg; 6756 6757 if (bserrno == 0 && (blob->data_ro_flags & SPDK_BLOB_READ_ONLY)) { 6758 blob->data_ro = true; 6759 blob->md_ro = true; 6760 } 6761 6762 bs_sequence_finish(seq, bserrno); 6763 } 6764 6765 static void 6766 blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 6767 { 6768 struct spdk_bs_cpl cpl; 6769 spdk_bs_sequence_t *seq; 6770 6771 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 6772 cpl.u.blob_basic.cb_fn = cb_fn; 6773 cpl.u.blob_basic.cb_arg = cb_arg; 6774 6775 seq = bs_sequence_start(blob->bs->md_channel, &cpl); 6776 if (!seq) { 6777 cb_fn(cb_arg, -ENOMEM); 6778 return; 6779 } 6780 6781 blob_persist(seq, blob, blob_sync_md_cpl, blob); 6782 } 6783 6784 void 6785 spdk_blob_sync_md(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 6786 { 6787 blob_verify_md_op(blob); 6788 6789 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Syncing blob %lu\n", blob->id); 6790 6791 if (blob->md_ro) { 6792 assert(blob->state == SPDK_BLOB_STATE_CLEAN); 6793 cb_fn(cb_arg, 0); 6794 return; 6795 } 6796 6797 blob_sync_md(blob, cb_fn, cb_arg); 6798 } 6799 6800 /* END spdk_blob_sync_md */ 6801 6802 struct spdk_blob_insert_cluster_ctx { 6803 struct spdk_thread *thread; 6804 struct spdk_blob *blob; 6805 uint32_t cluster_num; /* cluster index in blob */ 6806 uint32_t cluster; /* cluster on disk */ 6807 uint32_t extent_page; /* extent page on disk */ 6808 int rc; 6809 spdk_blob_op_complete cb_fn; 6810 void *cb_arg; 6811 }; 6812 6813 static void 6814 blob_insert_cluster_msg_cpl(void *arg) 6815 { 6816 struct spdk_blob_insert_cluster_ctx *ctx = arg; 6817 6818 ctx->cb_fn(ctx->cb_arg, ctx->rc); 6819 free(ctx); 6820 } 6821 6822 static void 6823 blob_insert_cluster_msg_cb(void *arg, int bserrno) 6824 { 6825 struct spdk_blob_insert_cluster_ctx *ctx = arg; 6826 6827 ctx->rc = bserrno; 6828 spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx); 6829 } 6830 6831 static void 6832 blob_persist_extent_page_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 6833 { 6834 struct spdk_blob_md_page *page = cb_arg; 6835 6836 bs_sequence_finish(seq, bserrno); 6837 spdk_free(page); 6838 } 6839 6840 static void 6841 blob_insert_extent(struct spdk_blob *blob, uint32_t extent, uint64_t cluster_num, 6842 spdk_blob_op_complete cb_fn, void *cb_arg) 6843 { 6844 spdk_bs_sequence_t *seq; 6845 struct spdk_bs_cpl cpl; 6846 struct spdk_blob_md_page *page = NULL; 6847 uint32_t page_count = 0; 6848 int rc; 6849 6850 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 6851 cpl.u.blob_basic.cb_fn = cb_fn; 6852 cpl.u.blob_basic.cb_arg = cb_arg; 6853 6854 seq = bs_sequence_start(blob->bs->md_channel, &cpl); 6855 if (!seq) { 6856 cb_fn(cb_arg, -ENOMEM); 6857 return; 6858 } 6859 rc = blob_serialize_add_page(blob, &page, &page_count, &page); 6860 if (rc < 0) { 6861 bs_sequence_finish(seq, rc); 6862 return; 6863 } 6864 6865 blob_serialize_extent_page(blob, cluster_num, page); 6866 6867 page->crc = blob_md_page_calc_crc(page); 6868 6869 assert(spdk_bit_array_get(blob->bs->used_md_pages, extent) == true); 6870 6871 bs_sequence_write_dev(seq, page, bs_md_page_to_lba(blob->bs, extent), 6872 bs_byte_to_lba(blob->bs, SPDK_BS_PAGE_SIZE), 6873 blob_persist_extent_page_cpl, page); 6874 } 6875 6876 static void 6877 blob_insert_cluster_msg(void *arg) 6878 { 6879 struct spdk_blob_insert_cluster_ctx *ctx = arg; 6880 uint32_t *extent_page; 6881 6882 ctx->rc = blob_insert_cluster(ctx->blob, ctx->cluster_num, ctx->cluster); 6883 if (ctx->rc != 0) { 6884 spdk_thread_send_msg(ctx->thread, blob_insert_cluster_msg_cpl, ctx); 6885 return; 6886 } 6887 6888 if (ctx->blob->use_extent_table == false) { 6889 /* Extent table is not used, proceed with sync of md that will only use extents_rle. */ 6890 ctx->blob->state = SPDK_BLOB_STATE_DIRTY; 6891 blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx); 6892 return; 6893 } 6894 6895 extent_page = bs_cluster_to_extent_page(ctx->blob, ctx->cluster_num); 6896 if (*extent_page == 0) { 6897 /* Extent page requires allocation. 6898 * It was already claimed in the used_md_pages map and placed in ctx. 6899 * Blob persist will take care of writing out new extent page on disk. */ 6900 assert(ctx->extent_page != 0); 6901 assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true); 6902 *extent_page = ctx->extent_page; 6903 ctx->blob->state = SPDK_BLOB_STATE_DIRTY; 6904 blob_sync_md(ctx->blob, blob_insert_cluster_msg_cb, ctx); 6905 } else { 6906 /* It is possible for original thread to allocate extent page for 6907 * different cluster in the same extent page. In such case proceed with 6908 * updating the existing extent page, but release the additional one. */ 6909 if (ctx->extent_page != 0) { 6910 assert(spdk_bit_array_get(ctx->blob->bs->used_md_pages, ctx->extent_page) == true); 6911 bs_release_md_page(ctx->blob->bs, ctx->extent_page); 6912 ctx->extent_page = 0; 6913 } 6914 /* Extent page already allocated. 6915 * Every cluster allocation, requires just an update of single extent page. */ 6916 blob_insert_extent(ctx->blob, *extent_page, ctx->cluster_num, 6917 blob_insert_cluster_msg_cb, ctx); 6918 } 6919 } 6920 6921 static void 6922 blob_insert_cluster_on_md_thread(struct spdk_blob *blob, uint32_t cluster_num, 6923 uint64_t cluster, uint32_t extent_page, spdk_blob_op_complete cb_fn, void *cb_arg) 6924 { 6925 struct spdk_blob_insert_cluster_ctx *ctx; 6926 6927 ctx = calloc(1, sizeof(*ctx)); 6928 if (ctx == NULL) { 6929 cb_fn(cb_arg, -ENOMEM); 6930 return; 6931 } 6932 6933 ctx->thread = spdk_get_thread(); 6934 ctx->blob = blob; 6935 ctx->cluster_num = cluster_num; 6936 ctx->cluster = cluster; 6937 ctx->extent_page = extent_page; 6938 ctx->cb_fn = cb_fn; 6939 ctx->cb_arg = cb_arg; 6940 6941 spdk_thread_send_msg(blob->bs->md_thread, blob_insert_cluster_msg, ctx); 6942 } 6943 6944 /* START spdk_blob_close */ 6945 6946 static void 6947 blob_close_cpl(spdk_bs_sequence_t *seq, void *cb_arg, int bserrno) 6948 { 6949 struct spdk_blob *blob = cb_arg; 6950 6951 if (bserrno == 0) { 6952 blob->open_ref--; 6953 if (blob->open_ref == 0) { 6954 /* 6955 * Blobs with active.num_pages == 0 are deleted blobs. 6956 * these blobs are removed from the blob_store list 6957 * when the deletion process starts - so don't try to 6958 * remove them again. 6959 */ 6960 if (blob->active.num_pages > 0) { 6961 spdk_bit_array_clear(blob->bs->open_blobids, blob->id); 6962 TAILQ_REMOVE(&blob->bs->blobs, blob, link); 6963 } 6964 blob_free(blob); 6965 } 6966 } 6967 6968 bs_sequence_finish(seq, bserrno); 6969 } 6970 6971 void spdk_blob_close(struct spdk_blob *blob, spdk_blob_op_complete cb_fn, void *cb_arg) 6972 { 6973 struct spdk_bs_cpl cpl; 6974 spdk_bs_sequence_t *seq; 6975 6976 blob_verify_md_op(blob); 6977 6978 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Closing blob %lu\n", blob->id); 6979 6980 if (blob->open_ref == 0) { 6981 cb_fn(cb_arg, -EBADF); 6982 return; 6983 } 6984 6985 cpl.type = SPDK_BS_CPL_TYPE_BLOB_BASIC; 6986 cpl.u.blob_basic.cb_fn = cb_fn; 6987 cpl.u.blob_basic.cb_arg = cb_arg; 6988 6989 seq = bs_sequence_start(blob->bs->md_channel, &cpl); 6990 if (!seq) { 6991 cb_fn(cb_arg, -ENOMEM); 6992 return; 6993 } 6994 6995 /* Sync metadata */ 6996 blob_persist(seq, blob, blob_close_cpl, blob); 6997 } 6998 6999 /* END spdk_blob_close */ 7000 7001 struct spdk_io_channel *spdk_bs_alloc_io_channel(struct spdk_blob_store *bs) 7002 { 7003 return spdk_get_io_channel(bs); 7004 } 7005 7006 void spdk_bs_free_io_channel(struct spdk_io_channel *channel) 7007 { 7008 spdk_put_io_channel(channel); 7009 } 7010 7011 void spdk_blob_io_unmap(struct spdk_blob *blob, struct spdk_io_channel *channel, 7012 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 7013 { 7014 blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 7015 SPDK_BLOB_UNMAP); 7016 } 7017 7018 void spdk_blob_io_write_zeroes(struct spdk_blob *blob, struct spdk_io_channel *channel, 7019 uint64_t offset, uint64_t length, spdk_blob_op_complete cb_fn, void *cb_arg) 7020 { 7021 blob_request_submit_op(blob, channel, NULL, offset, length, cb_fn, cb_arg, 7022 SPDK_BLOB_WRITE_ZEROES); 7023 } 7024 7025 void spdk_blob_io_write(struct spdk_blob *blob, struct spdk_io_channel *channel, 7026 void *payload, uint64_t offset, uint64_t length, 7027 spdk_blob_op_complete cb_fn, void *cb_arg) 7028 { 7029 blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 7030 SPDK_BLOB_WRITE); 7031 } 7032 7033 void spdk_blob_io_read(struct spdk_blob *blob, struct spdk_io_channel *channel, 7034 void *payload, uint64_t offset, uint64_t length, 7035 spdk_blob_op_complete cb_fn, void *cb_arg) 7036 { 7037 blob_request_submit_op(blob, channel, payload, offset, length, cb_fn, cb_arg, 7038 SPDK_BLOB_READ); 7039 } 7040 7041 void spdk_blob_io_writev(struct spdk_blob *blob, struct spdk_io_channel *channel, 7042 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 7043 spdk_blob_op_complete cb_fn, void *cb_arg) 7044 { 7045 blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, false); 7046 } 7047 7048 void spdk_blob_io_readv(struct spdk_blob *blob, struct spdk_io_channel *channel, 7049 struct iovec *iov, int iovcnt, uint64_t offset, uint64_t length, 7050 spdk_blob_op_complete cb_fn, void *cb_arg) 7051 { 7052 blob_request_submit_rw_iov(blob, channel, iov, iovcnt, offset, length, cb_fn, cb_arg, true); 7053 } 7054 7055 struct spdk_bs_iter_ctx { 7056 int64_t page_num; 7057 struct spdk_blob_store *bs; 7058 7059 spdk_blob_op_with_handle_complete cb_fn; 7060 void *cb_arg; 7061 }; 7062 7063 static void 7064 bs_iter_cpl(void *cb_arg, struct spdk_blob *_blob, int bserrno) 7065 { 7066 struct spdk_bs_iter_ctx *ctx = cb_arg; 7067 struct spdk_blob_store *bs = ctx->bs; 7068 spdk_blob_id id; 7069 7070 if (bserrno == 0) { 7071 ctx->cb_fn(ctx->cb_arg, _blob, bserrno); 7072 free(ctx); 7073 return; 7074 } 7075 7076 ctx->page_num++; 7077 ctx->page_num = spdk_bit_array_find_first_set(bs->used_blobids, ctx->page_num); 7078 if (ctx->page_num >= spdk_bit_array_capacity(bs->used_blobids)) { 7079 ctx->cb_fn(ctx->cb_arg, NULL, -ENOENT); 7080 free(ctx); 7081 return; 7082 } 7083 7084 id = bs_page_to_blobid(ctx->page_num); 7085 7086 spdk_bs_open_blob(bs, id, bs_iter_cpl, ctx); 7087 } 7088 7089 void 7090 spdk_bs_iter_first(struct spdk_blob_store *bs, 7091 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 7092 { 7093 struct spdk_bs_iter_ctx *ctx; 7094 7095 ctx = calloc(1, sizeof(*ctx)); 7096 if (!ctx) { 7097 cb_fn(cb_arg, NULL, -ENOMEM); 7098 return; 7099 } 7100 7101 ctx->page_num = -1; 7102 ctx->bs = bs; 7103 ctx->cb_fn = cb_fn; 7104 ctx->cb_arg = cb_arg; 7105 7106 bs_iter_cpl(ctx, NULL, -1); 7107 } 7108 7109 static void 7110 bs_iter_close_cpl(void *cb_arg, int bserrno) 7111 { 7112 struct spdk_bs_iter_ctx *ctx = cb_arg; 7113 7114 bs_iter_cpl(ctx, NULL, -1); 7115 } 7116 7117 void 7118 spdk_bs_iter_next(struct spdk_blob_store *bs, struct spdk_blob *blob, 7119 spdk_blob_op_with_handle_complete cb_fn, void *cb_arg) 7120 { 7121 struct spdk_bs_iter_ctx *ctx; 7122 7123 assert(blob != NULL); 7124 7125 ctx = calloc(1, sizeof(*ctx)); 7126 if (!ctx) { 7127 cb_fn(cb_arg, NULL, -ENOMEM); 7128 return; 7129 } 7130 7131 ctx->page_num = bs_blobid_to_page(blob->id); 7132 ctx->bs = bs; 7133 ctx->cb_fn = cb_fn; 7134 ctx->cb_arg = cb_arg; 7135 7136 /* Close the existing blob */ 7137 spdk_blob_close(blob, bs_iter_close_cpl, ctx); 7138 } 7139 7140 static int 7141 blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 7142 uint16_t value_len, bool internal) 7143 { 7144 struct spdk_xattr_tailq *xattrs; 7145 struct spdk_xattr *xattr; 7146 size_t desc_size; 7147 void *tmp; 7148 7149 blob_verify_md_op(blob); 7150 7151 if (blob->md_ro) { 7152 return -EPERM; 7153 } 7154 7155 desc_size = sizeof(struct spdk_blob_md_descriptor_xattr) + strlen(name) + value_len; 7156 if (desc_size > SPDK_BS_MAX_DESC_SIZE) { 7157 SPDK_DEBUGLOG(SPDK_LOG_BLOB, "Xattr '%s' of size %ld does not fix into single page %ld\n", name, 7158 desc_size, SPDK_BS_MAX_DESC_SIZE); 7159 return -ENOMEM; 7160 } 7161 7162 if (internal) { 7163 xattrs = &blob->xattrs_internal; 7164 blob->invalid_flags |= SPDK_BLOB_INTERNAL_XATTR; 7165 } else { 7166 xattrs = &blob->xattrs; 7167 } 7168 7169 TAILQ_FOREACH(xattr, xattrs, link) { 7170 if (!strcmp(name, xattr->name)) { 7171 tmp = malloc(value_len); 7172 if (!tmp) { 7173 return -ENOMEM; 7174 } 7175 7176 free(xattr->value); 7177 xattr->value_len = value_len; 7178 xattr->value = tmp; 7179 memcpy(xattr->value, value, value_len); 7180 7181 blob->state = SPDK_BLOB_STATE_DIRTY; 7182 7183 return 0; 7184 } 7185 } 7186 7187 xattr = calloc(1, sizeof(*xattr)); 7188 if (!xattr) { 7189 return -ENOMEM; 7190 } 7191 7192 xattr->name = strdup(name); 7193 if (!xattr->name) { 7194 free(xattr); 7195 return -ENOMEM; 7196 } 7197 7198 xattr->value_len = value_len; 7199 xattr->value = malloc(value_len); 7200 if (!xattr->value) { 7201 free(xattr->name); 7202 free(xattr); 7203 return -ENOMEM; 7204 } 7205 memcpy(xattr->value, value, value_len); 7206 TAILQ_INSERT_TAIL(xattrs, xattr, link); 7207 7208 blob->state = SPDK_BLOB_STATE_DIRTY; 7209 7210 return 0; 7211 } 7212 7213 int 7214 spdk_blob_set_xattr(struct spdk_blob *blob, const char *name, const void *value, 7215 uint16_t value_len) 7216 { 7217 return blob_set_xattr(blob, name, value, value_len, false); 7218 } 7219 7220 static int 7221 blob_remove_xattr(struct spdk_blob *blob, const char *name, bool internal) 7222 { 7223 struct spdk_xattr_tailq *xattrs; 7224 struct spdk_xattr *xattr; 7225 7226 blob_verify_md_op(blob); 7227 7228 if (blob->md_ro) { 7229 return -EPERM; 7230 } 7231 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 7232 7233 TAILQ_FOREACH(xattr, xattrs, link) { 7234 if (!strcmp(name, xattr->name)) { 7235 TAILQ_REMOVE(xattrs, xattr, link); 7236 free(xattr->value); 7237 free(xattr->name); 7238 free(xattr); 7239 7240 if (internal && TAILQ_EMPTY(&blob->xattrs_internal)) { 7241 blob->invalid_flags &= ~SPDK_BLOB_INTERNAL_XATTR; 7242 } 7243 blob->state = SPDK_BLOB_STATE_DIRTY; 7244 7245 return 0; 7246 } 7247 } 7248 7249 return -ENOENT; 7250 } 7251 7252 int 7253 spdk_blob_remove_xattr(struct spdk_blob *blob, const char *name) 7254 { 7255 return blob_remove_xattr(blob, name, false); 7256 } 7257 7258 static int 7259 blob_get_xattr_value(struct spdk_blob *blob, const char *name, 7260 const void **value, size_t *value_len, bool internal) 7261 { 7262 struct spdk_xattr *xattr; 7263 struct spdk_xattr_tailq *xattrs; 7264 7265 xattrs = internal ? &blob->xattrs_internal : &blob->xattrs; 7266 7267 TAILQ_FOREACH(xattr, xattrs, link) { 7268 if (!strcmp(name, xattr->name)) { 7269 *value = xattr->value; 7270 *value_len = xattr->value_len; 7271 return 0; 7272 } 7273 } 7274 return -ENOENT; 7275 } 7276 7277 int 7278 spdk_blob_get_xattr_value(struct spdk_blob *blob, const char *name, 7279 const void **value, size_t *value_len) 7280 { 7281 blob_verify_md_op(blob); 7282 7283 return blob_get_xattr_value(blob, name, value, value_len, false); 7284 } 7285 7286 struct spdk_xattr_names { 7287 uint32_t count; 7288 const char *names[0]; 7289 }; 7290 7291 static int 7292 blob_get_xattr_names(struct spdk_xattr_tailq *xattrs, struct spdk_xattr_names **names) 7293 { 7294 struct spdk_xattr *xattr; 7295 int count = 0; 7296 7297 TAILQ_FOREACH(xattr, xattrs, link) { 7298 count++; 7299 } 7300 7301 *names = calloc(1, sizeof(struct spdk_xattr_names) + count * sizeof(char *)); 7302 if (*names == NULL) { 7303 return -ENOMEM; 7304 } 7305 7306 TAILQ_FOREACH(xattr, xattrs, link) { 7307 (*names)->names[(*names)->count++] = xattr->name; 7308 } 7309 7310 return 0; 7311 } 7312 7313 int 7314 spdk_blob_get_xattr_names(struct spdk_blob *blob, struct spdk_xattr_names **names) 7315 { 7316 blob_verify_md_op(blob); 7317 7318 return blob_get_xattr_names(&blob->xattrs, names); 7319 } 7320 7321 uint32_t 7322 spdk_xattr_names_get_count(struct spdk_xattr_names *names) 7323 { 7324 assert(names != NULL); 7325 7326 return names->count; 7327 } 7328 7329 const char * 7330 spdk_xattr_names_get_name(struct spdk_xattr_names *names, uint32_t index) 7331 { 7332 if (index >= names->count) { 7333 return NULL; 7334 } 7335 7336 return names->names[index]; 7337 } 7338 7339 void 7340 spdk_xattr_names_free(struct spdk_xattr_names *names) 7341 { 7342 free(names); 7343 } 7344 7345 struct spdk_bs_type 7346 spdk_bs_get_bstype(struct spdk_blob_store *bs) 7347 { 7348 return bs->bstype; 7349 } 7350 7351 void 7352 spdk_bs_set_bstype(struct spdk_blob_store *bs, struct spdk_bs_type bstype) 7353 { 7354 memcpy(&bs->bstype, &bstype, sizeof(bstype)); 7355 } 7356 7357 bool 7358 spdk_blob_is_read_only(struct spdk_blob *blob) 7359 { 7360 assert(blob != NULL); 7361 return (blob->data_ro || blob->md_ro); 7362 } 7363 7364 bool 7365 spdk_blob_is_snapshot(struct spdk_blob *blob) 7366 { 7367 struct spdk_blob_list *snapshot_entry; 7368 7369 assert(blob != NULL); 7370 7371 snapshot_entry = bs_get_snapshot_entry(blob->bs, blob->id); 7372 if (snapshot_entry == NULL) { 7373 return false; 7374 } 7375 7376 return true; 7377 } 7378 7379 bool 7380 spdk_blob_is_clone(struct spdk_blob *blob) 7381 { 7382 assert(blob != NULL); 7383 7384 if (blob->parent_id != SPDK_BLOBID_INVALID) { 7385 assert(spdk_blob_is_thin_provisioned(blob)); 7386 return true; 7387 } 7388 7389 return false; 7390 } 7391 7392 bool 7393 spdk_blob_is_thin_provisioned(struct spdk_blob *blob) 7394 { 7395 assert(blob != NULL); 7396 return !!(blob->invalid_flags & SPDK_BLOB_THIN_PROV); 7397 } 7398 7399 static void 7400 blob_update_clear_method(struct spdk_blob *blob) 7401 { 7402 enum blob_clear_method stored_cm; 7403 7404 assert(blob != NULL); 7405 7406 /* If BLOB_CLEAR_WITH_DEFAULT was passed in, use the setting stored 7407 * in metadata previously. If something other than the default was 7408 * specified, ignore stored value and used what was passed in. 7409 */ 7410 stored_cm = ((blob->md_ro_flags & SPDK_BLOB_CLEAR_METHOD) >> SPDK_BLOB_CLEAR_METHOD_SHIFT); 7411 7412 if (blob->clear_method == BLOB_CLEAR_WITH_DEFAULT) { 7413 blob->clear_method = stored_cm; 7414 } else if (blob->clear_method != stored_cm) { 7415 SPDK_WARNLOG("Using passed in clear method 0x%x instead of stored value of 0x%x\n", 7416 blob->clear_method, stored_cm); 7417 } 7418 } 7419 7420 spdk_blob_id 7421 spdk_blob_get_parent_snapshot(struct spdk_blob_store *bs, spdk_blob_id blob_id) 7422 { 7423 struct spdk_blob_list *snapshot_entry = NULL; 7424 struct spdk_blob_list *clone_entry = NULL; 7425 7426 TAILQ_FOREACH(snapshot_entry, &bs->snapshots, link) { 7427 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 7428 if (clone_entry->id == blob_id) { 7429 return snapshot_entry->id; 7430 } 7431 } 7432 } 7433 7434 return SPDK_BLOBID_INVALID; 7435 } 7436 7437 int 7438 spdk_blob_get_clones(struct spdk_blob_store *bs, spdk_blob_id blobid, spdk_blob_id *ids, 7439 size_t *count) 7440 { 7441 struct spdk_blob_list *snapshot_entry, *clone_entry; 7442 size_t n; 7443 7444 snapshot_entry = bs_get_snapshot_entry(bs, blobid); 7445 if (snapshot_entry == NULL) { 7446 *count = 0; 7447 return 0; 7448 } 7449 7450 if (ids == NULL || *count < snapshot_entry->clone_count) { 7451 *count = snapshot_entry->clone_count; 7452 return -ENOMEM; 7453 } 7454 *count = snapshot_entry->clone_count; 7455 7456 n = 0; 7457 TAILQ_FOREACH(clone_entry, &snapshot_entry->clones, link) { 7458 ids[n++] = clone_entry->id; 7459 } 7460 7461 return 0; 7462 } 7463 7464 SPDK_LOG_REGISTER_COMPONENT("blob", SPDK_LOG_BLOB) 7465