1 /* 2 * Copyright (c) 2011-2012 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@dragonflybsd.org> 6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the 17 * distribution. 18 * 3. Neither the name of The DragonFly Project nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific, prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 #ifndef VFS_HAMMER2_DISK_H_ 36 #define VFS_HAMMER2_DISK_H_ 37 38 #ifndef _SYS_UUID_H_ 39 #include <sys/uuid.h> 40 #endif 41 #ifndef _SYS_DMSG_H_ 42 #include <sys/dmsg.h> 43 #endif 44 45 /* 46 * The structures below represent the on-disk media structures for the HAMMER2 47 * filesystem. Note that all fields for on-disk structures are naturally 48 * aligned. The host endian format is typically used - compatibility is 49 * possible if the implementation detects reversed endian and adjusts accesses 50 * accordingly. 51 * 52 * HAMMER2 primarily revolves around the directory topology: inodes, 53 * directory entries, and block tables. Block device buffer cache buffers 54 * are always 64KB. Logical file buffers are typically 16KB. All data 55 * references utilize 64-bit byte offsets. 56 * 57 * Free block management is handled independently using blocks reserved by 58 * the media topology. 59 */ 60 61 /* 62 * The data at the end of a file or directory may be a fragment in order 63 * to optimize storage efficiency. The minimum fragment size is 1KB. 64 * Since allocations are in powers of 2 fragments must also be sized in 65 * powers of 2 (1024, 2048, ... 65536). 66 * 67 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K), 68 * which is 2^16. Larger extents may be supported in the future. Smaller 69 * fragments might be supported in the future (down to 64 bytes is possible), 70 * but probably will not be. 71 * 72 * A full indirect block use supports 1024 x 64-byte blockrefs in a 64KB 73 * buffer. Indirect blocks down to 1KB are supported to keep small 74 * directories small. 75 * 76 * A maximally sized file (2^64-1 bytes) requires 5 indirect block levels. 77 * The hammer2_blockset in the volume header or file inode has another 8 78 * entries, giving us 66+3 = 69 bits of address space. However, some bits 79 * are taken up by (potentially) requests for redundant copies. HAMMER2 80 * currently supports up to 8 copies, which brings the address space down 81 * to 66 bits and gives us 2 bits of leeway. 82 */ 83 #define HAMMER2_MIN_ALLOC 1024 /* minimum allocation size */ 84 #define HAMMER2_MIN_RADIX 10 /* minimum allocation size 2^N */ 85 #define HAMMER2_MAX_ALLOC 65536 /* maximum allocation size */ 86 #define HAMMER2_MAX_RADIX 16 /* maximum allocation size 2^N */ 87 #define HAMMER2_KEY_RADIX 64 /* number of bits in key */ 88 89 /* 90 * MINALLOCSIZE - The minimum allocation size. This can be smaller 91 * or larger than the minimum physical IO size. 92 * 93 * NOTE: Should not be larger than 1K since inodes 94 * are 1K. 95 * 96 * MINIOSIZE - The minimum IO size. This must be less than 97 * or equal to HAMMER2_LBUFSIZE. 98 * 99 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups. 100 * 101 * HAMMER2_PBUFSIZE - Topological block size used by files for all 102 * blocks except the block straddling EOF. 103 * 104 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB 105 * (space represented by a level0 bitmap). 106 */ 107 108 #define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX) 109 110 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 111 #define HAMMER2_PBUFSIZE 65536 112 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 113 #define HAMMER2_LBUFSIZE 16384 114 115 /* 116 * Generally speaking we want to use 16K and 64K I/Os 117 */ 118 #define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX 119 #define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE 120 121 #define HAMMER2_IND_BYTES_MIN HAMMER2_LBUFSIZE 122 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 123 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 124 sizeof(hammer2_blockref_t)) 125 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 126 sizeof(hammer2_blockref_t)) 127 128 /* 129 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 130 * any element can occur at any index and holes can be anywhere. As a 131 * future optimization we will be able to flag that such arrays are sorted 132 * and thus optimize lookups, but for now we don't. 133 * 134 * Inodes embed either 512 bytes of direct data or an array of 8 blockrefs, 135 * resulting in highly efficient storage for files <= 512 bytes and for files 136 * <= 512KB. Up to 8 directory entries can be referenced from a directory 137 * without requiring an indirect block. 138 * 139 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented), 140 * or 64KB (1024 blockrefs / ~64MB represented). 141 */ 142 #define HAMMER2_SET_COUNT 8 /* direct entries */ 143 #define HAMMER2_SET_RADIX 3 144 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 145 #define HAMMER2_EMBEDDED_RADIX 9 146 147 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 148 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 149 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 150 151 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 152 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 153 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 154 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 155 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 156 157 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 158 159 /* 160 * A HAMMER2 filesystem is always sized in multiples of 8MB. 161 * 162 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment 163 * contains the volume header (or backup volume header), the free block 164 * table, and possibly other information in the future. 165 * 166 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 167 * 168 * +-----------------------+ 169 * | Volume Hdr | block 0 volume header & alternates 170 * +-----------------------+ (first four zones only) 171 * | FreeBlk Section A | block 1-8 172 * +-----------------------+ 173 * | FreeBlk Section B | block 9-16 174 * +-----------------------+ 175 * | FreeBlk Section C | block 17-24 176 * +-----------------------+ 177 * | FreeBlk Section D | block 25-32 178 * +-----------------------+ 179 * | | block 33...63 180 * | reserved | 181 * | | 182 * +-----------------------+ 183 * 184 * The first few 2GB zones contain volume headers and volume header backups. 185 * After that the volume header block# is reserved. 186 * 187 * The freemap utilizes blocks #1-32 for now, see the FREEMAP document. 188 * The Free block table has a resolution of 1KB 189 * 190 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 191 * (i.e. a multiple of 2MB). VOLUME_ALIGN must be >= ZONE_SEG. 192 */ 193 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 194 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 195 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 196 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 197 198 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 199 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 200 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 201 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 202 203 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 204 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 205 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 206 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 207 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 208 209 /* 210 * 64 x 64KB blocks are reserved at the base of each 2GB zone. These blocks 211 * are used to store the volume header or volume header backups, allocation 212 * tree, and other information in the future. 213 * 214 * All specified blocks are not necessarily used in all 2GB zones. However, 215 * dead areas are reserved for future use and MUST NOT BE USED for other 216 * purposes. 217 * 218 * The freemap is arranged into four groups. Modifications rotate through 219 * the groups on a block by block basis (so all the blocks are not necessarily 220 * synchronized to the same group). Because the freemap is flushed 221 * independent of the main filesystem, the freemap only really needs two 222 * groups to operate efficiently. 223 * 224 * 225 * 226 */ 227 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 228 #define HAMMER2_ZONE_FREEMAP_A 1 /* freemap layer group A */ 229 #define HAMMER2_ZONE_FREEMAP_B 9 /* freemap layer group B */ 230 #define HAMMER2_ZONE_FREEMAP_C 17 /* freemap layer group C */ 231 #define HAMMER2_ZONE_FREEMAP_D 25 /* freemap layer group D */ 232 233 /* relative to FREEMAP_x */ 234 #define HAMMER2_ZONEFM_LEVEL0 0 /* 256KB bitmap (4 blks) */ 235 #define HAMMER2_ZONEFM_LEVEL1 4 /* 2GB indmap */ 236 #define HAMMER2_ZONEFM_LEVEL2 5 /* 2TB indmap */ 237 #define HAMMER2_ZONEFM_LEVEL3 6 /* 2PB indmap */ 238 #define HAMMER2_ZONEFM_LEVEL4 7 /* 2EB indmap */ 239 /* LEVEL5 is a set of 8 blockrefs in the volume header 16EB */ 240 241 #define HAMMER2_ZONE_BLOCK49 49 /* future */ 242 #define HAMMER2_ZONE_BLOCK50 50 /* future */ 243 #define HAMMER2_ZONE_BLOCK51 51 /* future */ 244 #define HAMMER2_ZONE_BLOCK52 52 /* future */ 245 #define HAMMER2_ZONE_BLOCK53 53 /* future */ 246 #define HAMMER2_ZONE_BLOCK54 54 /* future */ 247 #define HAMMER2_ZONE_BLOCK55 55 /* future */ 248 #define HAMMER2_ZONE_BLOCK56 56 /* future */ 249 #define HAMMER2_ZONE_BLOCK57 57 /* future */ 250 #define HAMMER2_ZONE_BLOCK58 58 /* future */ 251 #define HAMMER2_ZONE_BLOCK59 59 /* future */ 252 253 #define HAMMER2_ZONE_BLOCK60 60 /* future */ 254 #define HAMMER2_ZONE_BLOCK61 61 /* future */ 255 #define HAMMER2_ZONE_BLOCK62 62 /* future */ 256 #define HAMMER2_ZONE_BLOCK63 63 /* future */ 257 258 /* 259 * Freemap radii. Please note that LEVEL 1 blockref array entries 260 * point to 256-byte sections of the bitmap representing 2MB of storage. 261 * Even though the chain structures represent only 256 bytes, they are 262 * mapped using larger 16K or 64K buffer cache buffers. 263 */ 264 #define HAMMER2_FREEMAP_LEVEL5_RADIX 64 /* 16EB */ 265 #define HAMMER2_FREEMAP_LEVEL4_RADIX 61 /* 2EB */ 266 #define HAMMER2_FREEMAP_LEVEL3_RADIX 51 /* 2PB */ 267 #define HAMMER2_FREEMAP_LEVEL2_RADIX 41 /* 2TB */ 268 #define HAMMER2_FREEMAP_LEVEL1_RADIX 31 /* 2GB (256KB of bitmap) */ 269 #define HAMMER2_FREEMAP_LEVEL0_RADIX 21 /* 2MB (256 bytes of bitmap) */ 270 271 #define HAMMER2_FREEMAP_LEVELN_PSIZE 65536 /* physical bytes */ 272 #define HAMMER2_FREEMAP_LEVEL0_PSIZE 256 /* physical bytes */ 273 274 275 /* 276 * Two linear areas can be reserved after the initial 2MB segment in the base 277 * zone (the one starting at offset 0). These areas are NOT managed by the 278 * block allocator and do not fall under HAMMER2 crc checking rules based 279 * at the volume header (but can be self-CRCd internally, depending). 280 */ 281 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 282 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 283 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 284 285 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 286 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 287 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 288 289 /* 290 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 291 * Transaction ids are monotonic. 292 * 293 * We utilize 32-bit iSCSI CRCs. 294 */ 295 typedef uint64_t hammer2_tid_t; 296 typedef uint64_t hammer2_off_t; 297 typedef uint64_t hammer2_key_t; 298 typedef uint32_t hammer2_crc32_t; 299 300 /* 301 * Miscellanious ranges (all are unsigned). 302 */ 303 #define HAMMER2_MIN_TID 1ULL 304 #define HAMMER2_MAX_TID 0xFFFFFFFFFFFFFFFFULL 305 #define HAMMER2_MIN_KEY 0ULL 306 #define HAMMER2_MAX_KEY 0xFFFFFFFFFFFFFFFFULL 307 #define HAMMER2_MIN_OFFSET 0ULL 308 #define HAMMER2_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL 309 310 /* 311 * HAMMER2 data offset special cases and masking. 312 * 313 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 314 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 315 * 316 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 317 * of the data offset field specifies how large the data chunk being pointed 318 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 319 * needed in the low bits of the data offset field). However, the practical 320 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 321 * HAMMER2_MIN_RADIX to 10. The maximum radix is currently 16 (64KB), but 322 * we fully intend to support larger extents in the future. 323 */ 324 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 325 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 326 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 327 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 328 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 329 #define HAMMER2_MAX_COPIES 6 330 331 /* 332 * HAMMER2 directory support and pre-defined keys 333 */ 334 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 335 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 336 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 337 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 338 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 339 340 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 341 342 /* 343 * The media block reference structure. This forms the core of the HAMMER2 344 * media topology recursion. This 64-byte data structure is embedded in the 345 * volume header, in inodes (which are also directory entries), and in 346 * indirect blocks. 347 * 348 * A blockref references a single media item, which typically can be a 349 * directory entry (aka inode), indirect block, or data block. 350 * 351 * The primary feature a blockref represents is the ability to validate 352 * the entire tree underneath it via its check code. Any modification to 353 * anything propagates up the blockref tree all the way to the root, replacing 354 * the related blocks. Propagations can shortcut to the volume root to 355 * implement the 'fast syncing' feature but this only delays the eventual 356 * propagation. 357 * 358 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, 359 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum 360 * supported check code size, which is not sufficient for unverified dedup 361 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when 362 * farming web data). HAMMER2 has an unverified dedup feature for just this 363 * purpose. 364 * 365 * -- 366 * 367 * NOTE: The range of keys represented by the blockref is (key) to 368 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 369 * blocks bottom-up, inserting a new root when radix expansion 370 * is required. 371 */ 372 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 373 uint8_t type; /* type of underlying item */ 374 uint8_t methods; /* check method & compression method */ 375 uint8_t copyid; /* specify which copy this is */ 376 uint8_t keybits; /* #of keybits masked off 0=leaf */ 377 uint8_t vradix; /* virtual data/meta-data size */ 378 uint8_t flags; /* blockref flags */ 379 uint8_t reserved06; 380 uint8_t reserved07; 381 hammer2_key_t key; /* key specification */ 382 hammer2_tid_t mirror_tid; /* propagate for mirror scan */ 383 hammer2_tid_t modify_tid; /* modifications sans propagation */ 384 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 385 union { /* check info */ 386 char buf[24]; 387 struct { 388 uint32_t value; 389 uint32_t unused[5]; 390 } iscsi32; 391 struct { 392 uint64_t value; 393 uint64_t unused[2]; 394 } crc64; 395 struct { 396 char data[24]; 397 } sha192; 398 399 /* 400 * Freemap hints are embedded in addition to the icrc32. 401 * 402 * biggest - Largest possible allocation 2^N within sub-tree. 403 * typically initialized to 64 in freemap_blockref 404 * and reduced as-needed when a request fails. 405 * 406 * An allocation > 2^N is guaranteed to fail. An 407 * allocation <= 2^N MAY fail, and if it does the 408 * biggest hint will be adjusted downward. 409 * 410 * Used when allocating space. 411 * 412 * radix - (Leaf only) once assigned, radix for clustering. 413 * All device I/O can cluster within the 2MB 414 * segment. 415 */ 416 struct { 417 uint32_t icrc32; 418 uint8_t biggest; 419 uint8_t radix; /* 0, LBUFRADIX, PBUFRADIX */ 420 uint8_t reserved06; 421 uint8_t reserved07; 422 uint64_t avail; /* total available bytes */ 423 uint64_t unused; /* unused must be 0 */ 424 } freemap; 425 } check; 426 }; 427 428 typedef struct hammer2_blockref hammer2_blockref_t; 429 430 #if 0 431 #define HAMMER2_BREF_SYNC1 0x01 /* modification synchronized */ 432 #define HAMMER2_BREF_SYNC2 0x02 /* modification committed */ 433 #define HAMMER2_BREF_DESYNCCHLD 0x04 /* desynchronize children */ 434 #define HAMMER2_BREF_DELETED 0x80 /* indicates a deletion */ 435 #endif 436 437 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */ 438 439 #define HAMMER2_BREF_TYPE_EMPTY 0 440 #define HAMMER2_BREF_TYPE_INODE 1 441 #define HAMMER2_BREF_TYPE_INDIRECT 2 442 #define HAMMER2_BREF_TYPE_DATA 3 443 #define HAMMER2_BREF_TYPE_UNUSED04 4 444 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 445 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 446 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 447 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 448 449 #define HAMMER2_ENC_CHECK(n) ((n) << 4) 450 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 451 452 #define HAMMER2_CHECK_NONE 0 453 #define HAMMER2_CHECK_ISCSI32 1 454 #define HAMMER2_CHECK_CRC64 2 455 #define HAMMER2_CHECK_SHA192 3 456 #define HAMMER2_CHECK_FREEMAP 4 457 458 #define HAMMER2_ENC_COMP(n) (n) 459 #define HAMMER2_DEC_COMP(n) ((n) & 15) 460 461 #define HAMMER2_COMP_NONE 0 462 #define HAMMER2_COMP_AUTOZERO 1 463 464 465 /* 466 * HAMMER2 block references are collected into sets of 8 blockrefs. These 467 * sets are fully associative, meaning the elements making up a set are 468 * not sorted in any way and may contain duplicate entries, holes, or 469 * entries which shortcut multiple levels of indirection. Sets are used 470 * in various ways: 471 * 472 * (1) When redundancy is desired a set may contain several duplicate 473 * entries pointing to different copies of the same data. Up to 8 copies 474 * are supported but the set structure becomes a bit inefficient once 475 * you go over 4. 476 * 477 * (2) The blockrefs in a set can shortcut multiple levels of indirections 478 * within the bounds imposed by the parent of set. 479 * 480 * When a set fills up another level of indirection is inserted, moving 481 * some or all of the set's contents into indirect blocks placed under the 482 * set. This is a top-down approach in that indirect blocks are not created 483 * until the set actually becomes full (that is, the entries in the set can 484 * shortcut the indirect blocks when the set is not full). Depending on how 485 * things are filled multiple indirect blocks will eventually be created. 486 * 487 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 488 * are also treated as fully set-associative. 489 */ 490 struct hammer2_blockset { 491 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 492 }; 493 494 typedef struct hammer2_blockset hammer2_blockset_t; 495 496 /* 497 * Catch programmer snafus 498 */ 499 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 500 #error "hammer2 direct radix is incorrect" 501 #endif 502 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 503 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 504 #endif 505 #if (1 << HAMMER2_MIN_RADIX) != HAMMER2_MIN_ALLOC 506 #error "HAMMER2_MIN_RADIX and HAMMER2_MIN_ALLOC are inconsistent" 507 #endif 508 509 /* 510 * The media indirect block structure. 511 */ 512 struct hammer2_indblock_data { 513 hammer2_blockref_t blockref[HAMMER2_IND_COUNT_MAX]; 514 }; 515 516 typedef struct hammer2_indblock_data hammer2_indblock_data_t; 517 518 struct hammer2_bmap_data { 519 uint64_t array[HAMMER2_FREEMAP_LEVEL0_PSIZE / sizeof(uint64_t)]; 520 }; 521 522 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 523 524 /* 525 * In HAMMER2 inodes ARE directory entries, with a special exception for 526 * hardlinks. The inode number is stored in the inode rather than being 527 * based on the location of the inode (since the location moves every time 528 * the inode or anything underneath the inode is modified). 529 * 530 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 531 * for the filename, and 512 bytes worth of direct file data OR an embedded 532 * blockset. 533 * 534 * Directories represent one inode per blockref. Inodes are not laid out 535 * as a file but instead are represented by the related blockrefs. The 536 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 537 * that blocksets are fully associative, so a certain degree efficiency is 538 * achieved just from that. 539 * 540 * Up to 512 bytes of direct data can be embedded in an inode, and since 541 * inodes are essentially directory entries this also means that small data 542 * files end up simply being laid out linearly in the directory, resulting 543 * in fewer seeks and highly optimal access. 544 * 545 * The compression mode can be changed at any time in the inode and is 546 * recorded on a blockref-by-blockref basis. 547 * 548 * Hardlinks are supported via the inode map. Essentially the way a hardlink 549 * works is that all individual directory entries representing the same file 550 * are special cased and specify the same inode number. The actual file 551 * is placed in the nearest parent directory that is parent to all instances 552 * of the hardlink. If all hardlinks to a file are in the same directory 553 * the actual file will also be placed in that directory. This file uses 554 * the inode number as the directory entry key and is invisible to normal 555 * directory scans. Real directory entry keys are differentiated from the 556 * inode number key via bit 63. Access to the hardlink silently looks up 557 * the real file and forwards all operations to that file. Removal of the 558 * last hardlink also removes the real file. 559 * 560 * (attr_tid) is only updated when the inode's specific attributes or regular 561 * file size has changed, and affects path lookups and stat. (attr_tid) 562 * represents a special cache coherency lock under the inode. The inode 563 * blockref's modify_tid will always cover it. 564 * 565 * (dirent_tid) is only updated when an entry under a directory inode has 566 * been created, deleted, renamed, or had its attributes change, and affects 567 * directory lookups and scans. (dirent_tid) represents another special cache 568 * coherency lock under the inode. The inode blockref's modify_tid will 569 * always cover it. 570 */ 571 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 572 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 573 #define HAMMER2_INODE_VERSION_ONE 1 574 575 struct hammer2_inode_data { 576 uint16_t version; /* 0000 inode data version */ 577 uint16_t reserved02; /* 0002 */ 578 579 /* 580 * core inode attributes, inode type, misc flags 581 */ 582 uint32_t uflags; /* 0004 chflags */ 583 uint32_t rmajor; /* 0008 available for device nodes */ 584 uint32_t rminor; /* 000C available for device nodes */ 585 uint64_t ctime; /* 0010 inode change time */ 586 uint64_t mtime; /* 0018 modified time */ 587 uint64_t atime; /* 0020 access time (unsupported) */ 588 uint64_t btime; /* 0028 birth time */ 589 uuid_t uid; /* 0030 uid / degenerate unix uid */ 590 uuid_t gid; /* 0040 gid / degenerate unix gid */ 591 592 uint8_t type; /* 0050 object type */ 593 uint8_t op_flags; /* 0051 operational flags */ 594 uint16_t cap_flags; /* 0052 capability flags */ 595 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 596 597 /* 598 * inode size, identification, localized recursive configuration 599 * for compression and backup copies. 600 */ 601 hammer2_tid_t inum; /* 0058 inode number */ 602 hammer2_off_t size; /* 0060 size of file */ 603 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 604 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */ 605 hammer2_key_t name_key; /* 0078 full filename key */ 606 uint16_t name_len; /* 0080 filename length */ 607 uint8_t ncopies; /* 0082 ncopies to local media */ 608 uint8_t comp_algo; /* 0083 compression request & algo */ 609 610 /* 611 * These fields are currently only applicable to PFSROOTs. 612 * 613 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 614 * identify an instance of a PFS in the cluster because 615 * a mount may contain more than one copy of the PFS as 616 * a separate node. {pfs_clid, pfs_fsid} must be used for 617 * registration in the cluster. 618 */ 619 uint8_t target_type; /* 0084 hardlink target type */ 620 uint8_t reserved85; /* 0085 */ 621 uint8_t reserved86; /* 0086 */ 622 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 623 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 624 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 625 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 626 627 /* 628 * Quotas and cumulative sub-tree counters. 629 */ 630 hammer2_off_t data_quota; /* 00B0 subtree quota in bytes */ 631 hammer2_off_t data_count; /* 00B8 subtree byte count */ 632 hammer2_off_t inode_quota; /* 00C0 subtree quota inode count */ 633 hammer2_off_t inode_count; /* 00C8 subtree inode count */ 634 hammer2_tid_t attr_tid; /* 00D0 attributes changed */ 635 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */ 636 uint64_t reservedE0; /* 00E0 */ 637 uint64_t reservedE8; /* 00E8 */ 638 uint64_t reservedF0; /* 00F0 */ 639 uint64_t reservedF8; /* 00F8 */ 640 641 unsigned char filename[HAMMER2_INODE_MAXNAME]; 642 /* 0100-01FF (256 char, unterminated) */ 643 union { /* 0200-03FF (64x8 = 512 bytes) */ 644 struct hammer2_blockset blockset; 645 char data[HAMMER2_EMBEDDED_BYTES]; 646 } u; 647 }; 648 649 typedef struct hammer2_inode_data hammer2_inode_data_t; 650 651 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 652 #define HAMMER2_OPFLAG_PFSROOT 0x02 653 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 654 655 #define HAMMER2_OBJTYPE_UNKNOWN 0 656 #define HAMMER2_OBJTYPE_DIRECTORY 1 657 #define HAMMER2_OBJTYPE_REGFILE 2 658 #define HAMMER2_OBJTYPE_FIFO 4 659 #define HAMMER2_OBJTYPE_CDEV 5 660 #define HAMMER2_OBJTYPE_BDEV 6 661 #define HAMMER2_OBJTYPE_SOFTLINK 7 662 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */ 663 #define HAMMER2_OBJTYPE_SOCKET 9 664 #define HAMMER2_OBJTYPE_WHITEOUT 10 665 666 #define HAMMER2_COPYID_NONE 0 667 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 668 669 /* 670 * PEER types identify connections and help cluster controller filter 671 * out unwanted SPANs. 672 */ 673 #define HAMMER2_PEER_NONE DMSG_PEER_NONE 674 #define HAMMER2_PEER_CLUSTER DMSG_PEER_CLUSTER 675 #define HAMMER2_PEER_BLOCK DMSG_PEER_BLOCK 676 #define HAMMER2_PEER_HAMMER2 DMSG_PEER_HAMMER2 677 678 #define HAMMER2_COPYID_COUNT DMSG_COPYID_COUNT 679 680 /* 681 * PFS types identify a PFS on media and in LNK_SPAN messages. 682 */ 683 #define HAMMER2_PFSTYPE_NONE DMSG_PFSTYPE_NONE 684 #define HAMMER2_PFSTYPE_ADMIN DMSG_PFSTYPE_ADMIN 685 #define HAMMER2_PFSTYPE_CLIENT DMSG_PFSTYPE_CLIENT 686 #define HAMMER2_PFSTYPE_CACHE DMSG_PFSTYPE_CACHE 687 #define HAMMER2_PFSTYPE_COPY DMSG_PFSTYPE_COPY 688 #define HAMMER2_PFSTYPE_SLAVE DMSG_PFSTYPE_SLAVE 689 #define HAMMER2_PFSTYPE_SOFT_SLAVE DMSG_PFSTYPE_SOFT_SLAVE 690 #define HAMMER2_PFSTYPE_SOFT_MASTER DMSG_PFSTYPE_SOFT_MASTER 691 #define HAMMER2_PFSTYPE_MASTER DMSG_PFSTYPE_MASTER 692 #define HAMMER2_PFSTYPE_SNAPSHOT DMSG_PFSTYPE_SNAPSHOT 693 #define HAMMER2_PFSTYPE_MAX DMSG_PFSTYPE_MAX 694 695 /* 696 * Allocation Table 697 * 698 */ 699 700 701 /* 702 * Flags (8 bits) - blockref, for freemap only 703 * 704 * Note that the minimum chunk size is 1KB so we could theoretically have 705 * 10 bits here, but we might have some future extension that allows a 706 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 707 */ 708 #define HAMMER2_AVF_SELMASK 0x03 /* select group */ 709 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 710 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 711 #define HAMMER2_AVF_RESERVED10 0x10 712 #define HAMMER2_AVF_RESERVED20 0x20 713 #define HAMMER2_AVF_RESERVED40 0x40 714 #define HAMMER2_AVF_RESERVED80 0x80 715 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 716 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 717 718 #define HAMMER2_AV_SELECT_A 0x00 719 #define HAMMER2_AV_SELECT_B 0x01 720 #define HAMMER2_AV_SELECT_C 0x02 721 #define HAMMER2_AV_SELECT_D 0x03 722 723 /* 724 * The volume header eats a 64K block. There is currently an issue where 725 * we want to try to fit all nominal filesystem updates in a 512-byte section 726 * but it may be a lost cause due to the need for a blockset. 727 * 728 * All information is stored in host byte order. The volume header's magic 729 * number may be checked to determine the byte order. If you wish to mount 730 * between machines w/ different endian modes you'll need filesystem code 731 * which acts on the media data consistently (either all one way or all the 732 * other). Our code currently does not do that. 733 * 734 * A read-write mount may have to recover missing allocations by doing an 735 * incremental mirror scan looking for modifications made after alloc_tid. 736 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 737 * operations are usually very, very fast. 738 * 739 * Read-only mounts do not need to do any recovery, access to the filesystem 740 * topology is always consistent after a crash (is always consistent, period). 741 * However, there may be shortcutted blockref updates present from deep in 742 * the tree which are stored in the volumeh eader and must be tracked on 743 * the fly. 744 * 745 * NOTE: The copyinfo[] array contains the configuration for both the 746 * cluster connections and any local media copies. The volume 747 * header will be replicated for each local media copy. 748 * 749 * The mount command may specify multiple medias or just one and 750 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 751 * array on mount. 752 * 753 * NOTE: root_blockref points to the super-root directory, not the root 754 * directory. The root directory will be a subdirectory under the 755 * super-root. 756 * 757 * The super-root directory contains all root directories and all 758 * snapshots (readonly or writable). It is possible to do a 759 * null-mount of the super-root using special path constructions 760 * relative to your mounted root. 761 * 762 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 763 * a PFS, including mirroring and storage quota operations, and this is 764 * prefered over creating discrete PFSs in the super-root. Instead 765 * the super-root is most typically used to create writable snapshots, 766 * alternative roots, and so forth. The super-root is also used by 767 * the automatic snapshotting mechanism. 768 */ 769 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 770 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 771 772 struct hammer2_volume_data { 773 /* 774 * sector #0 - 512 bytes 775 */ 776 uint64_t magic; /* 0000 Signature */ 777 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 778 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 779 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 780 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 781 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 782 783 uint32_t version; /* 0030 */ 784 uint32_t flags; /* 0034 */ 785 uint8_t copyid; /* 0038 copyid of phys vol */ 786 uint8_t freemap_version; /* 0039 freemap algorithm */ 787 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 788 uint8_t reserved003B; /* 003B */ 789 uint32_t reserved003C; /* 003C */ 790 791 uuid_t fsid; /* 0040 */ 792 uuid_t fstype; /* 0050 */ 793 794 /* 795 * allocator_size is precalculated at newfs time and does not include 796 * reserved blocks, boot, or redo areas. 797 * 798 * Initial non-reserved-area allocations do not use the freemap 799 * but instead adjust alloc_iterator. Dynamic allocations take 800 * over starting at (allocator_beg). This makes newfs_hammer2's 801 * job a lot easier and can also serve as a testing jig. 802 */ 803 hammer2_off_t allocator_size; /* 0060 Total data space */ 804 hammer2_off_t allocator_free; /* 0068 Free space */ 805 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 806 hammer2_tid_t mirror_tid; /* 0078 best committed tid */ 807 hammer2_tid_t alloc_tid; /* 0080 Alloctable modify tid */ 808 hammer2_blockref_t reserved0088; /* 0088-00C7 */ 809 810 /* 811 * Copyids are allocated dynamically from the copyexists bitmap. 812 * An id from the active copies set (up to 8, see copyinfo later on) 813 * may still exist after the copy set has been removed from the 814 * volume header and its bit will remain active in the bitmap and 815 * cannot be reused until it is 100% removed from the hierarchy. 816 */ 817 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 818 char reserved0140[248]; /* 00E8-01DF */ 819 820 /* 821 * 32 bit CRC array at the end of the first 512 byte sector. 822 * 823 * icrc_sects[7] - First 512-4 bytes of volume header (including all 824 * the other icrc's except this one). 825 * 826 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 827 * the blockset for the root. 828 * 829 * icrc_sects[5] - Sector 2 830 * icrc_sects[4] - Sector 3 831 * icrc_sects[3] - Sector 4 (the freemap blockset) 832 */ 833 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 834 835 /* 836 * sector #1 - 512 bytes 837 * 838 * The entire sector is used by a blockset. 839 */ 840 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 841 842 /* 843 * sector #2-7 844 */ 845 char sector2[512]; /* 0400-05FF reserved */ 846 char sector3[512]; /* 0600-07FF reserved */ 847 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 848 char sector5[512]; /* 0A00-0BFF reserved */ 849 char sector6[512]; /* 0C00-0DFF reserved */ 850 char sector7[512]; /* 0E00-0FFF reserved */ 851 852 /* 853 * sector #8-71 - 32768 bytes 854 * 855 * Contains the configuration for up to 256 copyinfo targets. These 856 * specify local and remote copies operating as masters or slaves. 857 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 858 * indicates the local media). 859 * 860 * Each inode contains a set of up to 8 copyids, either inherited 861 * from its parent or explicitly specified in the inode, which 862 * indexes into this array. 863 */ 864 /* 1000-8FFF copyinfo config */ 865 dmsg_vol_data_t copyinfo[HAMMER2_COPYID_COUNT]; 866 867 /* 868 * Remaining sections are reserved for future use. 869 */ 870 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 871 872 /* 873 * icrc on entire volume header 874 */ 875 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 876 }; 877 878 typedef struct hammer2_volume_data hammer2_volume_data_t; 879 880 /* 881 * Various parts of the volume header have their own iCRCs. 882 * 883 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 884 * and not included the icrc calculation. 885 * 886 * The second 512 bytes also has its own iCRC but it is stored in the first 887 * 512 bytes so it covers the entire second 512 bytes. 888 * 889 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 890 * which is where the iCRC for the whole volume is stored. This is currently 891 * a catch-all for anything not individually iCRCd. 892 */ 893 #define HAMMER2_VOL_ICRC_SECT0 7 894 #define HAMMER2_VOL_ICRC_SECT1 6 895 896 #define HAMMER2_VOLUME_BYTES 65536 897 898 #define HAMMER2_VOLUME_ICRC0_OFF 0 899 #define HAMMER2_VOLUME_ICRC1_OFF 512 900 #define HAMMER2_VOLUME_ICRCVH_OFF 0 901 902 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 903 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 904 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 905 906 #define HAMMER2_VOL_VERSION_MIN 1 907 #define HAMMER2_VOL_VERSION_DEFAULT 1 908 #define HAMMER2_VOL_VERSION_WIP 2 909 910 #define HAMMER2_NUM_VOLHDRS 4 911 912 union hammer2_media_data { 913 hammer2_volume_data_t voldata; 914 hammer2_inode_data_t ipdata; 915 hammer2_indblock_data_t npdata; 916 hammer2_bmap_data_t bmdata; 917 char buf[HAMMER2_PBUFSIZE]; 918 }; 919 920 typedef union hammer2_media_data hammer2_media_data_t; 921 922 #endif 923