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