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 #ifndef SPDK_BLOBSTORE_H 35 #define SPDK_BLOBSTORE_H 36 37 #include "spdk/assert.h" 38 #include "spdk/blob.h" 39 #include "spdk/queue.h" 40 #include "spdk/util.h" 41 42 #include "request.h" 43 44 /* In Memory Data Structures 45 * 46 * The following data structures exist only in memory. 47 */ 48 49 #define SPDK_BLOB_OPTS_CLUSTER_SZ (1024 * 1024) 50 #define SPDK_BLOB_OPTS_NUM_MD_PAGES UINT32_MAX 51 #define SPDK_BLOB_OPTS_MAX_MD_OPS 32 52 #define SPDK_BLOB_OPTS_DEFAULT_CHANNEL_OPS 512 53 #define SPDK_BLOB_BLOBID_HIGH_BIT (1ULL << 32) 54 55 struct spdk_xattr { 56 /* TODO: reorder for best packing */ 57 uint32_t index; 58 char *name; 59 void *value; 60 uint16_t value_len; 61 TAILQ_ENTRY(spdk_xattr) link; 62 }; 63 64 /* The mutable part of the blob data that is sync'd to 65 * disk. The data in here is both mutable and persistent. 66 */ 67 struct spdk_blob_mut_data { 68 /* Number of data clusters in the blob */ 69 uint64_t num_clusters; 70 71 /* Array LBAs that are the beginning of a cluster, in 72 * the order they appear in the blob. 73 */ 74 uint64_t *clusters; 75 76 /* The size of the clusters array. This is greater than or 77 * equal to 'num_clusters'. 78 */ 79 size_t cluster_array_size; 80 81 /* Number of metadata pages */ 82 uint32_t num_pages; 83 84 /* Array of page offsets into the metadata region, in 85 * the order of the metadata page sequence. 86 */ 87 uint32_t *pages; 88 }; 89 90 enum spdk_blob_state { 91 /* The blob in-memory version does not match the on-disk 92 * version. 93 */ 94 SPDK_BLOB_STATE_DIRTY, 95 96 /* The blob in memory version of the blob matches the on disk 97 * version. 98 */ 99 SPDK_BLOB_STATE_CLEAN, 100 101 /* The in-memory state being synchronized with the on-disk 102 * blob state. */ 103 SPDK_BLOB_STATE_LOADING, 104 105 /* The disk state is being synchronized with the current 106 * blob state. 107 */ 108 SPDK_BLOB_STATE_SYNCING, 109 }; 110 111 struct spdk_blob_data { 112 struct spdk_blob_store *bs; 113 114 uint32_t open_ref; 115 116 spdk_blob_id id; 117 118 enum spdk_blob_state state; 119 120 /* Two copies of the mutable data. One is a version 121 * that matches the last known data on disk (clean). 122 * The other (active) is the current data. Syncing 123 * a blob makes the clean match the active. 124 */ 125 struct spdk_blob_mut_data clean; 126 struct spdk_blob_mut_data active; 127 128 bool invalid; 129 bool data_ro; 130 bool md_ro; 131 132 uint64_t invalid_flags; 133 uint64_t data_ro_flags; 134 uint64_t md_ro_flags; 135 136 struct spdk_bs_dev *back_bs_dev; 137 138 /* TODO: The xattrs are mutable, but we don't want to be 139 * copying them unecessarily. Figure this out. 140 */ 141 TAILQ_HEAD(spdk_xattr_tailq, spdk_xattr) xattrs; 142 143 TAILQ_ENTRY(spdk_blob_data) link; 144 }; 145 146 #define __blob_to_data(x) ((struct spdk_blob_data *)(x)) 147 #define __data_to_blob(x) ((struct spdk_blob *)(x)) 148 149 struct spdk_blob_store { 150 uint64_t md_start; /* Offset from beginning of disk, in pages */ 151 uint32_t md_len; /* Count, in pages */ 152 153 struct spdk_io_channel *md_channel; 154 uint32_t max_channel_ops; 155 156 struct spdk_thread *md_thread; 157 158 struct spdk_bs_dev *dev; 159 160 struct spdk_bit_array *used_md_pages; 161 struct spdk_bit_array *used_clusters; 162 struct spdk_bit_array *used_blobids; 163 164 pthread_mutex_t used_clusters_mutex; 165 166 uint32_t cluster_sz; 167 uint64_t total_clusters; 168 uint64_t total_data_clusters; 169 uint64_t num_free_clusters; 170 uint32_t pages_per_cluster; 171 172 spdk_blob_id super_blob; 173 struct spdk_bs_type bstype; 174 175 struct spdk_bs_cpl unload_cpl; 176 int unload_err; 177 178 TAILQ_HEAD(, spdk_blob_data) blobs; 179 }; 180 181 struct spdk_bs_channel { 182 struct spdk_bs_request_set *req_mem; 183 TAILQ_HEAD(, spdk_bs_request_set) reqs; 184 185 struct spdk_blob_store *bs; 186 187 struct spdk_bs_dev *dev; 188 struct spdk_io_channel *dev_channel; 189 190 TAILQ_HEAD(, spdk_bs_request_set) need_cluster_alloc; 191 }; 192 193 /** operation type */ 194 enum spdk_blob_op_type { 195 SPDK_BLOB_WRITE, 196 SPDK_BLOB_READ, 197 SPDK_BLOB_UNMAP, 198 SPDK_BLOB_WRITE_ZEROES, 199 SPDK_BLOB_WRITEV, 200 SPDK_BLOB_READV, 201 }; 202 203 /* On-Disk Data Structures 204 * 205 * The following data structures exist on disk. 206 */ 207 #define SPDK_BS_INITIAL_VERSION 1 208 #define SPDK_BS_VERSION 3 /* current version */ 209 210 #pragma pack(push, 1) 211 212 #define SPDK_MD_MASK_TYPE_USED_PAGES 0 213 #define SPDK_MD_MASK_TYPE_USED_CLUSTERS 1 214 #define SPDK_MD_MASK_TYPE_USED_BLOBIDS 2 215 216 struct spdk_bs_md_mask { 217 uint8_t type; 218 uint32_t length; /* In bits */ 219 uint8_t mask[0]; 220 }; 221 222 #define SPDK_MD_DESCRIPTOR_TYPE_PADDING 0 223 #define SPDK_MD_DESCRIPTOR_TYPE_EXTENT 1 224 #define SPDK_MD_DESCRIPTOR_TYPE_XATTR 2 225 #define SPDK_MD_DESCRIPTOR_TYPE_FLAGS 3 226 227 struct spdk_blob_md_descriptor_xattr { 228 uint8_t type; 229 uint32_t length; 230 231 uint16_t name_length; 232 uint16_t value_length; 233 234 char name[0]; 235 /* String name immediately followed by string value. */ 236 }; 237 238 struct spdk_blob_md_descriptor_extent { 239 uint8_t type; 240 uint32_t length; 241 242 struct { 243 uint32_t cluster_idx; 244 uint32_t length; /* In units of clusters */ 245 } extents[0]; 246 }; 247 248 #define SPDK_BLOB_THIN_PROV (1ULL << 0) 249 #define SPDK_BLOB_INVALID_FLAGS_MASK SPDK_BLOB_THIN_PROV 250 251 #define SPDK_BLOB_READ_ONLY (1ULL << 0) 252 #define SPDK_BLOB_DATA_RO_FLAGS_MASK SPDK_BLOB_READ_ONLY 253 #define SPDK_BLOB_MD_RO_FLAGS_MASK 0 254 255 #define spdk_blob_is_thin_provisioned(blob) (blob->invalid_flags & SPDK_BLOB_THIN_PROV) 256 257 struct spdk_blob_md_descriptor_flags { 258 uint8_t type; 259 uint32_t length; 260 261 /* 262 * If a flag in invalid_flags is set that the application is not aware of, 263 * it will not allow the blob to be opened. 264 */ 265 uint64_t invalid_flags; 266 267 /* 268 * If a flag in data_ro_flags is set that the application is not aware of, 269 * allow the blob to be opened in data_read_only and md_read_only mode. 270 */ 271 uint64_t data_ro_flags; 272 273 /* 274 * If a flag in md_ro_flags is set the the application is not aware of, 275 * allow the blob to be opened in md_read_only mode. 276 */ 277 uint64_t md_ro_flags; 278 }; 279 280 struct spdk_blob_md_descriptor { 281 uint8_t type; 282 uint32_t length; 283 }; 284 285 #define SPDK_INVALID_MD_PAGE UINT32_MAX 286 287 struct spdk_blob_md_page { 288 spdk_blob_id id; 289 290 uint32_t sequence_num; 291 uint32_t reserved0; 292 293 /* Descriptors here */ 294 uint8_t descriptors[4072]; 295 296 uint32_t next; 297 uint32_t crc; 298 }; 299 #define SPDK_BS_PAGE_SIZE 0x1000 300 SPDK_STATIC_ASSERT(SPDK_BS_PAGE_SIZE == sizeof(struct spdk_blob_md_page), "Invalid md page size"); 301 302 #define SPDK_BS_SUPER_BLOCK_SIG "SPDKBLOB" 303 304 struct spdk_bs_super_block { 305 uint8_t signature[8]; 306 uint32_t version; 307 uint32_t length; 308 uint32_t clean; /* If there was a clean shutdown, this is 1. */ 309 spdk_blob_id super_blob; 310 311 uint32_t cluster_size; /* In bytes */ 312 313 uint32_t used_page_mask_start; /* Offset from beginning of disk, in pages */ 314 uint32_t used_page_mask_len; /* Count, in pages */ 315 316 uint32_t used_cluster_mask_start; /* Offset from beginning of disk, in pages */ 317 uint32_t used_cluster_mask_len; /* Count, in pages */ 318 319 uint32_t md_start; /* Offset from beginning of disk, in pages */ 320 uint32_t md_len; /* Count, in pages */ 321 322 struct spdk_bs_type bstype; /* blobstore type */ 323 324 uint32_t used_blobid_mask_start; /* Offset from beginning of disk, in pages */ 325 uint32_t used_blobid_mask_len; /* Count, in pages */ 326 327 uint8_t reserved[4012]; 328 uint32_t crc; 329 }; 330 SPDK_STATIC_ASSERT(sizeof(struct spdk_bs_super_block) == 0x1000, "Invalid super block size"); 331 332 #pragma pack(pop) 333 334 struct spdk_bs_dev *spdk_bs_create_zeroes_dev(void); 335 336 /* Unit Conversions 337 * 338 * The blobstore works with several different units: 339 * - Byte: Self explanatory 340 * - LBA: The logical blocks on the backing storage device. 341 * - Page: The read/write units of blobs and metadata. This is 342 * an offset into a blob in units of 4KiB. 343 * - Cluster Index: The disk is broken into a sequential list of 344 * clusters. This is the offset from the beginning. 345 * 346 * NOTE: These conversions all act on simple magnitudes, not with any sort 347 * of knowledge about the blobs themselves. For instance, converting 348 * a page to an lba with the conversion function below simply converts 349 * a number of pages to an equivalent number of lbas, but that 350 * lba certainly isn't the right lba that corresponds to a page offset 351 * for a particular blob. 352 */ 353 static inline uint64_t 354 _spdk_bs_byte_to_lba(struct spdk_blob_store *bs, uint64_t length) 355 { 356 assert(length % bs->dev->blocklen == 0); 357 358 return length / bs->dev->blocklen; 359 } 360 361 static inline uint64_t 362 _spdk_bs_dev_byte_to_lba(struct spdk_bs_dev *bs_dev, uint64_t length) 363 { 364 assert(length % bs_dev->blocklen == 0); 365 366 return length / bs_dev->blocklen; 367 } 368 369 static inline uint64_t 370 _spdk_bs_lba_to_byte(struct spdk_blob_store *bs, uint64_t lba) 371 { 372 return lba * bs->dev->blocklen; 373 } 374 375 static inline uint64_t 376 _spdk_bs_page_to_lba(struct spdk_blob_store *bs, uint64_t page) 377 { 378 return page * SPDK_BS_PAGE_SIZE / bs->dev->blocklen; 379 } 380 381 static inline uint64_t 382 _spdk_bs_dev_page_to_lba(struct spdk_bs_dev *bs_dev, uint64_t page) 383 { 384 return page * SPDK_BS_PAGE_SIZE / bs_dev->blocklen; 385 } 386 387 static inline uint32_t 388 _spdk_bs_lba_to_page(struct spdk_blob_store *bs, uint64_t lba) 389 { 390 uint64_t lbas_per_page; 391 392 lbas_per_page = SPDK_BS_PAGE_SIZE / bs->dev->blocklen; 393 394 assert(lba % lbas_per_page == 0); 395 396 return lba / lbas_per_page; 397 } 398 399 static inline uint64_t 400 _spdk_bs_dev_lba_to_page(struct spdk_bs_dev *bs_dev, uint64_t lba) 401 { 402 uint64_t lbas_per_page; 403 404 lbas_per_page = SPDK_BS_PAGE_SIZE / bs_dev->blocklen; 405 406 assert(lba % lbas_per_page == 0); 407 408 return lba / lbas_per_page; 409 } 410 411 static inline uint64_t 412 _spdk_bs_cluster_to_page(struct spdk_blob_store *bs, uint32_t cluster) 413 { 414 return cluster * bs->pages_per_cluster; 415 } 416 417 static inline uint32_t 418 _spdk_bs_page_to_cluster(struct spdk_blob_store *bs, uint64_t page) 419 { 420 assert(page % bs->pages_per_cluster == 0); 421 422 return page / bs->pages_per_cluster; 423 } 424 425 static inline uint64_t 426 _spdk_bs_cluster_to_lba(struct spdk_blob_store *bs, uint32_t cluster) 427 { 428 return cluster * (bs->cluster_sz / bs->dev->blocklen); 429 } 430 431 static inline uint32_t 432 _spdk_bs_lba_to_cluster(struct spdk_blob_store *bs, uint64_t lba) 433 { 434 assert(lba % (bs->cluster_sz / bs->dev->blocklen) == 0); 435 436 return lba / (bs->cluster_sz / bs->dev->blocklen); 437 } 438 439 static inline uint64_t 440 _spdk_bs_blob_lba_to_back_dev_lba(struct spdk_blob_data *blob, uint64_t lba) 441 { 442 return lba * blob->bs->dev->blocklen / blob->back_bs_dev->blocklen; 443 } 444 445 static inline uint64_t 446 _spdk_bs_blob_lba_from_back_dev_lba(struct spdk_blob_data *blob, uint64_t lba) 447 { 448 return lba * blob->back_bs_dev->blocklen / blob->bs->dev->blocklen; 449 } 450 451 /* End basic conversions */ 452 453 static inline uint32_t 454 _spdk_bs_blobid_to_page(spdk_blob_id id) 455 { 456 return id & 0xFFFFFFFF; 457 } 458 459 /* The blob id is a 64 bit number. The lower 32 bits are the page_idx. The upper 460 * 32 bits are not currently used. Stick a 1 there just to catch bugs where the 461 * code assumes blob id == page_idx. 462 */ 463 static inline spdk_blob_id 464 _spdk_bs_page_to_blobid(uint32_t page_idx) 465 { 466 return SPDK_BLOB_BLOBID_HIGH_BIT | page_idx; 467 } 468 469 /* Given a page offset into a blob, look up the LBA for the 470 * start of that page. 471 */ 472 static inline uint64_t 473 _spdk_bs_blob_page_to_lba(struct spdk_blob_data *blob, uint32_t page) 474 { 475 uint64_t lba; 476 uint32_t pages_per_cluster; 477 478 pages_per_cluster = blob->bs->pages_per_cluster; 479 480 assert(page < blob->active.num_clusters * pages_per_cluster); 481 482 lba = blob->active.clusters[page / pages_per_cluster]; 483 lba += _spdk_bs_page_to_lba(blob->bs, page % pages_per_cluster); 484 485 return lba; 486 } 487 488 /* Given a page offset into a blob, look up the number of pages until the 489 * next cluster boundary. 490 */ 491 static inline uint32_t 492 _spdk_bs_num_pages_to_cluster_boundary(struct spdk_blob_data *blob, uint32_t page) 493 { 494 uint32_t pages_per_cluster; 495 496 pages_per_cluster = blob->bs->pages_per_cluster; 497 498 return pages_per_cluster - (page % pages_per_cluster); 499 } 500 501 /* Given a page offset into a blob, look up the number of pages into blob to beginning of current cluster */ 502 static inline uint32_t 503 _spdk_bs_page_to_cluster_start(struct spdk_blob_data *blob, uint32_t page) 504 { 505 uint32_t pages_per_cluster; 506 507 pages_per_cluster = blob->bs->pages_per_cluster; 508 509 return page - (page % pages_per_cluster); 510 } 511 512 /* Given a page offset into a blob, look up if it is from allocated cluster. */ 513 static inline bool 514 _spdk_bs_page_is_allocated(struct spdk_blob_data *blob, uint32_t page) 515 { 516 uint64_t lba; 517 uint32_t pages_per_cluster; 518 519 pages_per_cluster = blob->bs->pages_per_cluster; 520 521 assert(page < blob->active.num_clusters * pages_per_cluster); 522 523 lba = blob->active.clusters[page / pages_per_cluster]; 524 525 if (lba == 0) { 526 assert(spdk_blob_is_thin_provisioned(blob)); 527 return false; 528 } else { 529 return true; 530 } 531 } 532 533 #endif 534