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 5 /* freemap layer group B */ 230 #define HAMMER2_ZONE_FREEMAP_C 9 /* freemap layer group C */ 231 #define HAMMER2_ZONE_FREEMAP_D 13 /* freemap layer group D */ 232 233 /* relative to FREEMAP_x */ 234 #define HAMMER2_ZONEFM_LEVEL1 0 /* 2GB leafmap */ 235 #define HAMMER2_ZONEFM_LEVEL2 1 /* 2TB indmap */ 236 #define HAMMER2_ZONEFM_LEVEL3 2 /* 2PB indmap */ 237 #define HAMMER2_ZONEFM_LEVEL4 3 /* 2EB indmap */ 238 /* LEVEL5 is a set of 8 blockrefs in the volume header 16EB */ 239 240 241 /* 242 * Freemap radii. Please note that LEVEL 1 blockref array entries 243 * point to 256-byte sections of the bitmap representing 2MB of storage. 244 * Even though the chain structures represent only 256 bytes, they are 245 * mapped using larger 16K or 64K buffer cache buffers. 246 */ 247 #define HAMMER2_FREEMAP_LEVEL5_RADIX 64 /* 16EB */ 248 #define HAMMER2_FREEMAP_LEVEL4_RADIX 61 /* 2EB */ 249 #define HAMMER2_FREEMAP_LEVEL3_RADIX 51 /* 2PB */ 250 #define HAMMER2_FREEMAP_LEVEL2_RADIX 41 /* 2TB */ 251 #define HAMMER2_FREEMAP_LEVEL1_RADIX 31 /* 2GB */ 252 #define HAMMER2_FREEMAP_LEVEL0_RADIX 21 /* 2MB (entry in l-1 leaf) */ 253 254 #define HAMMER2_FREEMAP_LEVELN_PSIZE 65536 /* physical bytes */ 255 256 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 257 sizeof(hammer2_bmap_data_t)) 258 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 259 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 260 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 261 262 /* 263 * Two linear areas can be reserved after the initial 2MB segment in the base 264 * zone (the one starting at offset 0). These areas are NOT managed by the 265 * block allocator and do not fall under HAMMER2 crc checking rules based 266 * at the volume header (but can be self-CRCd internally, depending). 267 */ 268 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 269 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 270 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 271 272 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 273 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 274 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 275 276 /* 277 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 278 * Transaction ids are monotonic. 279 * 280 * We utilize 32-bit iSCSI CRCs. 281 */ 282 typedef uint64_t hammer2_tid_t; 283 typedef uint64_t hammer2_off_t; 284 typedef uint64_t hammer2_key_t; 285 typedef uint32_t hammer2_crc32_t; 286 287 /* 288 * Miscellanious ranges (all are unsigned). 289 */ 290 #define HAMMER2_MIN_TID 1ULL 291 #define HAMMER2_MAX_TID 0xFFFFFFFFFFFFFFFFULL 292 #define HAMMER2_MIN_KEY 0ULL 293 #define HAMMER2_MAX_KEY 0xFFFFFFFFFFFFFFFFULL 294 #define HAMMER2_MIN_OFFSET 0ULL 295 #define HAMMER2_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL 296 297 /* 298 * HAMMER2 data offset special cases and masking. 299 * 300 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 301 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 302 * 303 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 304 * of the data offset field specifies how large the data chunk being pointed 305 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 306 * needed in the low bits of the data offset field). However, the practical 307 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 308 * HAMMER2_MIN_RADIX to 10. The maximum radix is currently 16 (64KB), but 309 * we fully intend to support larger extents in the future. 310 */ 311 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 312 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 313 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 314 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 315 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 316 #define HAMMER2_MAX_COPIES 6 317 318 /* 319 * HAMMER2 directory support and pre-defined keys 320 */ 321 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 322 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 323 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 324 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 325 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 326 327 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 328 329 /* 330 * The media block reference structure. This forms the core of the HAMMER2 331 * media topology recursion. This 64-byte data structure is embedded in the 332 * volume header, in inodes (which are also directory entries), and in 333 * indirect blocks. 334 * 335 * A blockref references a single media item, which typically can be a 336 * directory entry (aka inode), indirect block, or data block. 337 * 338 * The primary feature a blockref represents is the ability to validate 339 * the entire tree underneath it via its check code. Any modification to 340 * anything propagates up the blockref tree all the way to the root, replacing 341 * the related blocks. Propagations can shortcut to the volume root to 342 * implement the 'fast syncing' feature but this only delays the eventual 343 * propagation. 344 * 345 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, 346 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum 347 * supported check code size, which is not sufficient for unverified dedup 348 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when 349 * farming web data). HAMMER2 has an unverified dedup feature for just this 350 * purpose. 351 * 352 * -- 353 * 354 * NOTE: The range of keys represented by the blockref is (key) to 355 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 356 * blocks bottom-up, inserting a new root when radix expansion 357 * is required. 358 */ 359 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 360 uint8_t type; /* type of underlying item */ 361 uint8_t methods; /* check method & compression method */ 362 uint8_t copyid; /* specify which copy this is */ 363 uint8_t keybits; /* #of keybits masked off 0=leaf */ 364 uint8_t vradix; /* virtual data/meta-data size */ 365 uint8_t flags; /* blockref flags */ 366 uint8_t reserved06; 367 uint8_t reserved07; 368 hammer2_key_t key; /* key specification */ 369 hammer2_tid_t mirror_tid; /* propagate for mirror scan */ 370 hammer2_tid_t modify_tid; /* modifications sans propagation */ 371 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 372 union { /* check info */ 373 char buf[24]; 374 struct { 375 uint32_t value; 376 uint32_t unused[5]; 377 } iscsi32; 378 struct { 379 uint64_t value; 380 uint64_t unused[2]; 381 } crc64; 382 struct { 383 char data[24]; 384 } sha192; 385 386 /* 387 * Freemap hints are embedded in addition to the icrc32. 388 * 389 * bigmask - Radixes available for allocation (0-31). 390 * Heuristical (may be permissive but not 391 * restrictive). Typically only radix values 392 * 10-16 are used (i.e. (1<<10) through (1<<16)). 393 * 394 * avail - Total available space remaining, in bytes 395 */ 396 struct { 397 uint32_t icrc32; 398 uint32_t bigmask; /* available radixes */ 399 uint64_t avail; /* total available bytes */ 400 uint64_t unused; /* unused must be 0 */ 401 } freemap; 402 } check; 403 }; 404 405 typedef struct hammer2_blockref hammer2_blockref_t; 406 407 #if 0 408 #define HAMMER2_BREF_SYNC1 0x01 /* modification synchronized */ 409 #define HAMMER2_BREF_SYNC2 0x02 /* modification committed */ 410 #define HAMMER2_BREF_DESYNCCHLD 0x04 /* desynchronize children */ 411 #define HAMMER2_BREF_DELETED 0x80 /* indicates a deletion */ 412 #endif 413 414 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */ 415 416 #define HAMMER2_BREF_TYPE_EMPTY 0 417 #define HAMMER2_BREF_TYPE_INODE 1 418 #define HAMMER2_BREF_TYPE_INDIRECT 2 419 #define HAMMER2_BREF_TYPE_DATA 3 420 #define HAMMER2_BREF_TYPE_UNUSED04 4 421 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 422 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 423 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 424 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 425 426 #define HAMMER2_ENC_CHECK(n) ((n) << 4) 427 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 428 429 #define HAMMER2_CHECK_NONE 0 430 #define HAMMER2_CHECK_ISCSI32 1 431 #define HAMMER2_CHECK_CRC64 2 432 #define HAMMER2_CHECK_SHA192 3 433 #define HAMMER2_CHECK_FREEMAP 4 434 435 #define HAMMER2_ENC_COMP(n) (n) 436 #define HAMMER2_DEC_COMP(n) ((n) & 15) 437 438 #define HAMMER2_COMP_NONE 0 439 #define HAMMER2_COMP_AUTOZERO 1 440 441 442 /* 443 * HAMMER2 block references are collected into sets of 8 blockrefs. These 444 * sets are fully associative, meaning the elements making up a set are 445 * not sorted in any way and may contain duplicate entries, holes, or 446 * entries which shortcut multiple levels of indirection. Sets are used 447 * in various ways: 448 * 449 * (1) When redundancy is desired a set may contain several duplicate 450 * entries pointing to different copies of the same data. Up to 8 copies 451 * are supported but the set structure becomes a bit inefficient once 452 * you go over 4. 453 * 454 * (2) The blockrefs in a set can shortcut multiple levels of indirections 455 * within the bounds imposed by the parent of set. 456 * 457 * When a set fills up another level of indirection is inserted, moving 458 * some or all of the set's contents into indirect blocks placed under the 459 * set. This is a top-down approach in that indirect blocks are not created 460 * until the set actually becomes full (that is, the entries in the set can 461 * shortcut the indirect blocks when the set is not full). Depending on how 462 * things are filled multiple indirect blocks will eventually be created. 463 * 464 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 465 * are also treated as fully set-associative. 466 */ 467 struct hammer2_blockset { 468 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 469 }; 470 471 typedef struct hammer2_blockset hammer2_blockset_t; 472 473 /* 474 * Catch programmer snafus 475 */ 476 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 477 #error "hammer2 direct radix is incorrect" 478 #endif 479 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 480 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 481 #endif 482 #if (1 << HAMMER2_MIN_RADIX) != HAMMER2_MIN_ALLOC 483 #error "HAMMER2_MIN_RADIX and HAMMER2_MIN_ALLOC are inconsistent" 484 #endif 485 486 /* 487 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 488 * 489 * Each 64-byte entry contains the bitmap and meta-data required to manage 490 * a LEVEL0 (2MB) block of storage. The storage is managed in 128 x 16KB 491 * chunks. Smaller allocation granularity is supported via a linear iterator 492 * and/or must otherwise be tracked in ram. 493 * 494 * (data structure must be 64 bytes exactly) 495 * 496 * linear - A BYTE linear allocation offset. May contain values between 497 * 0 and 2MB. Any value which is 16KB-aligned is effective neutral 498 * (forces the bitmap to be checked), whereas intermediate values 499 * allow iterative allocations from a bitmap that is already marked 500 * allocated. 501 * 502 * Please note that file data granularity may be limited by 503 * other issues such as buffer cache direct-mapping and the 504 * desire to support sector sizes up to 16KB (so H2 only issues 505 * I/O's in multiples of 16KB anyway). 506 * 507 * radix - Once assigned, radix for clustering. Cleared to 0 only if 508 * the entire leaf becomes free (related device buffer cache buffers 509 * must also be destroyed to avoid later overlap assertions). All 510 * chain's within this 2MB segment must match the clustering radix. 511 * 512 * Device I/O size may be further adjusted to the minimum (16KB), 513 * even if radix is smaller. This forms the I/O clustering radix. 514 * 515 * Value will typically be 10-16 (1KB to 64KB). Smaller values may 516 * be allowed in the future (probably unnecessary to add that 517 * complication though). 518 * 519 * bitmap - Two bits per 16KB allocation block arranged in arrays of 520 * 32-bit elements, 128x2 bits representing ~2MB worth of media 521 * storage. Bit patterns are as follows: 522 * 523 * 00 Unallocated 524 * 01 Armed for bulk free scan 525 * 10 Possibly free 526 * 11 Allocated 527 */ 528 struct hammer2_bmap_data { 529 int32_t linear; /* 00 linear sub-granular allocation offset */ 530 uint8_t reserved04; /* 04 */ 531 uint8_t radix; /* 05 cluster I/O 0, LBUFRADIX, PBUFRADIX */ 532 uint8_t reserved06; /* 06 */ 533 uint8_t reserved07; /* 07 */ 534 uint32_t reserved08; /* 08 */ 535 uint32_t reserved0C; /* 0C */ 536 uint32_t reserved10; /* 10 */ 537 uint32_t reserved14; /* 14 */ 538 uint32_t reserved18; /* 18 */ 539 uint32_t avail; /* 1C */ 540 uint32_t bitmap[8]; /* 20-3F 256 bits manages 2MB/16KB/2-bits */ 541 }; 542 543 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 544 545 /* 546 * In HAMMER2 inodes ARE directory entries, with a special exception for 547 * hardlinks. The inode number is stored in the inode rather than being 548 * based on the location of the inode (since the location moves every time 549 * the inode or anything underneath the inode is modified). 550 * 551 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 552 * for the filename, and 512 bytes worth of direct file data OR an embedded 553 * blockset. 554 * 555 * Directories represent one inode per blockref. Inodes are not laid out 556 * as a file but instead are represented by the related blockrefs. The 557 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 558 * that blocksets are fully associative, so a certain degree efficiency is 559 * achieved just from that. 560 * 561 * Up to 512 bytes of direct data can be embedded in an inode, and since 562 * inodes are essentially directory entries this also means that small data 563 * files end up simply being laid out linearly in the directory, resulting 564 * in fewer seeks and highly optimal access. 565 * 566 * The compression mode can be changed at any time in the inode and is 567 * recorded on a blockref-by-blockref basis. 568 * 569 * Hardlinks are supported via the inode map. Essentially the way a hardlink 570 * works is that all individual directory entries representing the same file 571 * are special cased and specify the same inode number. The actual file 572 * is placed in the nearest parent directory that is parent to all instances 573 * of the hardlink. If all hardlinks to a file are in the same directory 574 * the actual file will also be placed in that directory. This file uses 575 * the inode number as the directory entry key and is invisible to normal 576 * directory scans. Real directory entry keys are differentiated from the 577 * inode number key via bit 63. Access to the hardlink silently looks up 578 * the real file and forwards all operations to that file. Removal of the 579 * last hardlink also removes the real file. 580 * 581 * (attr_tid) is only updated when the inode's specific attributes or regular 582 * file size has changed, and affects path lookups and stat. (attr_tid) 583 * represents a special cache coherency lock under the inode. The inode 584 * blockref's modify_tid will always cover it. 585 * 586 * (dirent_tid) is only updated when an entry under a directory inode has 587 * been created, deleted, renamed, or had its attributes change, and affects 588 * directory lookups and scans. (dirent_tid) represents another special cache 589 * coherency lock under the inode. The inode blockref's modify_tid will 590 * always cover it. 591 */ 592 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 593 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 594 #define HAMMER2_INODE_VERSION_ONE 1 595 596 struct hammer2_inode_data { 597 uint16_t version; /* 0000 inode data version */ 598 uint16_t reserved02; /* 0002 */ 599 600 /* 601 * core inode attributes, inode type, misc flags 602 */ 603 uint32_t uflags; /* 0004 chflags */ 604 uint32_t rmajor; /* 0008 available for device nodes */ 605 uint32_t rminor; /* 000C available for device nodes */ 606 uint64_t ctime; /* 0010 inode change time */ 607 uint64_t mtime; /* 0018 modified time */ 608 uint64_t atime; /* 0020 access time (unsupported) */ 609 uint64_t btime; /* 0028 birth time */ 610 uuid_t uid; /* 0030 uid / degenerate unix uid */ 611 uuid_t gid; /* 0040 gid / degenerate unix gid */ 612 613 uint8_t type; /* 0050 object type */ 614 uint8_t op_flags; /* 0051 operational flags */ 615 uint16_t cap_flags; /* 0052 capability flags */ 616 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 617 618 /* 619 * inode size, identification, localized recursive configuration 620 * for compression and backup copies. 621 */ 622 hammer2_tid_t inum; /* 0058 inode number */ 623 hammer2_off_t size; /* 0060 size of file */ 624 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 625 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */ 626 hammer2_key_t name_key; /* 0078 full filename key */ 627 uint16_t name_len; /* 0080 filename length */ 628 uint8_t ncopies; /* 0082 ncopies to local media */ 629 uint8_t comp_algo; /* 0083 compression request & algo */ 630 631 /* 632 * These fields are currently only applicable to PFSROOTs. 633 * 634 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 635 * identify an instance of a PFS in the cluster because 636 * a mount may contain more than one copy of the PFS as 637 * a separate node. {pfs_clid, pfs_fsid} must be used for 638 * registration in the cluster. 639 */ 640 uint8_t target_type; /* 0084 hardlink target type */ 641 uint8_t reserved85; /* 0085 */ 642 uint8_t reserved86; /* 0086 */ 643 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 644 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 645 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 646 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 647 648 /* 649 * Quotas and cumulative sub-tree counters. 650 */ 651 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 652 hammer2_key_t data_count; /* 00B8 subtree byte count */ 653 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 654 hammer2_key_t inode_count; /* 00C8 subtree inode count */ 655 hammer2_tid_t attr_tid; /* 00D0 attributes changed */ 656 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */ 657 658 /* 659 * Tracks (possibly degenerate) free areas covering all sub-tree 660 * allocations under inode, not counting the inode itself. 661 * 0/0 indicates empty entry. fully set-associative. 662 */ 663 hammer2_off_t freezones[4]; /* 00E0/E8/F0/F8 base|radix */ 664 665 unsigned char filename[HAMMER2_INODE_MAXNAME]; 666 /* 0100-01FF (256 char, unterminated) */ 667 union { /* 0200-03FF (64x8 = 512 bytes) */ 668 struct hammer2_blockset blockset; 669 char data[HAMMER2_EMBEDDED_BYTES]; 670 } u; 671 }; 672 673 typedef struct hammer2_inode_data hammer2_inode_data_t; 674 675 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 676 #define HAMMER2_OPFLAG_PFSROOT 0x02 677 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 678 679 #define HAMMER2_OBJTYPE_UNKNOWN 0 680 #define HAMMER2_OBJTYPE_DIRECTORY 1 681 #define HAMMER2_OBJTYPE_REGFILE 2 682 #define HAMMER2_OBJTYPE_FIFO 4 683 #define HAMMER2_OBJTYPE_CDEV 5 684 #define HAMMER2_OBJTYPE_BDEV 6 685 #define HAMMER2_OBJTYPE_SOFTLINK 7 686 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */ 687 #define HAMMER2_OBJTYPE_SOCKET 9 688 #define HAMMER2_OBJTYPE_WHITEOUT 10 689 690 #define HAMMER2_COPYID_NONE 0 691 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 692 693 /* 694 * PEER types identify connections and help cluster controller filter 695 * out unwanted SPANs. 696 */ 697 #define HAMMER2_PEER_NONE DMSG_PEER_NONE 698 #define HAMMER2_PEER_CLUSTER DMSG_PEER_CLUSTER 699 #define HAMMER2_PEER_BLOCK DMSG_PEER_BLOCK 700 #define HAMMER2_PEER_HAMMER2 DMSG_PEER_HAMMER2 701 702 #define HAMMER2_COPYID_COUNT DMSG_COPYID_COUNT 703 704 /* 705 * PFS types identify a PFS on media and in LNK_SPAN messages. 706 */ 707 #define HAMMER2_PFSTYPE_NONE DMSG_PFSTYPE_NONE 708 #define HAMMER2_PFSTYPE_ADMIN DMSG_PFSTYPE_ADMIN 709 #define HAMMER2_PFSTYPE_CLIENT DMSG_PFSTYPE_CLIENT 710 #define HAMMER2_PFSTYPE_CACHE DMSG_PFSTYPE_CACHE 711 #define HAMMER2_PFSTYPE_COPY DMSG_PFSTYPE_COPY 712 #define HAMMER2_PFSTYPE_SLAVE DMSG_PFSTYPE_SLAVE 713 #define HAMMER2_PFSTYPE_SOFT_SLAVE DMSG_PFSTYPE_SOFT_SLAVE 714 #define HAMMER2_PFSTYPE_SOFT_MASTER DMSG_PFSTYPE_SOFT_MASTER 715 #define HAMMER2_PFSTYPE_MASTER DMSG_PFSTYPE_MASTER 716 #define HAMMER2_PFSTYPE_SNAPSHOT DMSG_PFSTYPE_SNAPSHOT 717 #define HAMMER2_PFSTYPE_MAX DMSG_PFSTYPE_MAX 718 719 /* 720 * Allocation Table 721 * 722 */ 723 724 725 /* 726 * Flags (8 bits) - blockref, for freemap only 727 * 728 * Note that the minimum chunk size is 1KB so we could theoretically have 729 * 10 bits here, but we might have some future extension that allows a 730 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 731 */ 732 #define HAMMER2_AVF_SELMASK 0x03 /* select group */ 733 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 734 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 735 #define HAMMER2_AVF_RESERVED10 0x10 736 #define HAMMER2_AVF_RESERVED20 0x20 737 #define HAMMER2_AVF_RESERVED40 0x40 738 #define HAMMER2_AVF_RESERVED80 0x80 739 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 740 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 741 742 #define HAMMER2_AV_SELECT_A 0x00 743 #define HAMMER2_AV_SELECT_B 0x01 744 #define HAMMER2_AV_SELECT_C 0x02 745 #define HAMMER2_AV_SELECT_D 0x03 746 747 /* 748 * The volume header eats a 64K block. There is currently an issue where 749 * we want to try to fit all nominal filesystem updates in a 512-byte section 750 * but it may be a lost cause due to the need for a blockset. 751 * 752 * All information is stored in host byte order. The volume header's magic 753 * number may be checked to determine the byte order. If you wish to mount 754 * between machines w/ different endian modes you'll need filesystem code 755 * which acts on the media data consistently (either all one way or all the 756 * other). Our code currently does not do that. 757 * 758 * A read-write mount may have to recover missing allocations by doing an 759 * incremental mirror scan looking for modifications made after alloc_tid. 760 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 761 * operations are usually very, very fast. 762 * 763 * Read-only mounts do not need to do any recovery, access to the filesystem 764 * topology is always consistent after a crash (is always consistent, period). 765 * However, there may be shortcutted blockref updates present from deep in 766 * the tree which are stored in the volumeh eader and must be tracked on 767 * the fly. 768 * 769 * NOTE: The copyinfo[] array contains the configuration for both the 770 * cluster connections and any local media copies. The volume 771 * header will be replicated for each local media copy. 772 * 773 * The mount command may specify multiple medias or just one and 774 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 775 * array on mount. 776 * 777 * NOTE: root_blockref points to the super-root directory, not the root 778 * directory. The root directory will be a subdirectory under the 779 * super-root. 780 * 781 * The super-root directory contains all root directories and all 782 * snapshots (readonly or writable). It is possible to do a 783 * null-mount of the super-root using special path constructions 784 * relative to your mounted root. 785 * 786 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 787 * a PFS, including mirroring and storage quota operations, and this is 788 * prefered over creating discrete PFSs in the super-root. Instead 789 * the super-root is most typically used to create writable snapshots, 790 * alternative roots, and so forth. The super-root is also used by 791 * the automatic snapshotting mechanism. 792 */ 793 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 794 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 795 796 struct hammer2_volume_data { 797 /* 798 * sector #0 - 512 bytes 799 */ 800 uint64_t magic; /* 0000 Signature */ 801 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 802 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 803 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 804 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 805 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 806 807 uint32_t version; /* 0030 */ 808 uint32_t flags; /* 0034 */ 809 uint8_t copyid; /* 0038 copyid of phys vol */ 810 uint8_t freemap_version; /* 0039 freemap algorithm */ 811 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 812 uint8_t reserved003B; /* 003B */ 813 uint32_t reserved003C; /* 003C */ 814 815 uuid_t fsid; /* 0040 */ 816 uuid_t fstype; /* 0050 */ 817 818 /* 819 * allocator_size is precalculated at newfs time and does not include 820 * reserved blocks, boot, or redo areas. 821 * 822 * Initial non-reserved-area allocations do not use the freemap 823 * but instead adjust alloc_iterator. Dynamic allocations take 824 * over starting at (allocator_beg). This makes newfs_hammer2's 825 * job a lot easier and can also serve as a testing jig. 826 */ 827 hammer2_off_t allocator_size; /* 0060 Total data space */ 828 hammer2_off_t allocator_free; /* 0068 Free space */ 829 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 830 hammer2_tid_t mirror_tid; /* 0078 best committed tid */ 831 hammer2_tid_t alloc_tid; /* 0080 Alloctable modify tid */ 832 hammer2_blockref_t reserved0088; /* 0088-00C7 */ 833 834 /* 835 * Copyids are allocated dynamically from the copyexists bitmap. 836 * An id from the active copies set (up to 8, see copyinfo later on) 837 * may still exist after the copy set has been removed from the 838 * volume header and its bit will remain active in the bitmap and 839 * cannot be reused until it is 100% removed from the hierarchy. 840 */ 841 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 842 char reserved0140[248]; /* 00E8-01DF */ 843 844 /* 845 * 32 bit CRC array at the end of the first 512 byte sector. 846 * 847 * icrc_sects[7] - First 512-4 bytes of volume header (including all 848 * the other icrc's except this one). 849 * 850 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 851 * the blockset for the root. 852 * 853 * icrc_sects[5] - Sector 2 854 * icrc_sects[4] - Sector 3 855 * icrc_sects[3] - Sector 4 (the freemap blockset) 856 */ 857 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 858 859 /* 860 * sector #1 - 512 bytes 861 * 862 * The entire sector is used by a blockset. 863 */ 864 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 865 866 /* 867 * sector #2-7 868 */ 869 char sector2[512]; /* 0400-05FF reserved */ 870 char sector3[512]; /* 0600-07FF reserved */ 871 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 872 char sector5[512]; /* 0A00-0BFF reserved */ 873 char sector6[512]; /* 0C00-0DFF reserved */ 874 char sector7[512]; /* 0E00-0FFF reserved */ 875 876 /* 877 * sector #8-71 - 32768 bytes 878 * 879 * Contains the configuration for up to 256 copyinfo targets. These 880 * specify local and remote copies operating as masters or slaves. 881 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 882 * indicates the local media). 883 * 884 * Each inode contains a set of up to 8 copyids, either inherited 885 * from its parent or explicitly specified in the inode, which 886 * indexes into this array. 887 */ 888 /* 1000-8FFF copyinfo config */ 889 dmsg_vol_data_t copyinfo[HAMMER2_COPYID_COUNT]; 890 891 /* 892 * Remaining sections are reserved for future use. 893 */ 894 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 895 896 /* 897 * icrc on entire volume header 898 */ 899 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 900 }; 901 902 typedef struct hammer2_volume_data hammer2_volume_data_t; 903 904 /* 905 * Various parts of the volume header have their own iCRCs. 906 * 907 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 908 * and not included the icrc calculation. 909 * 910 * The second 512 bytes also has its own iCRC but it is stored in the first 911 * 512 bytes so it covers the entire second 512 bytes. 912 * 913 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 914 * which is where the iCRC for the whole volume is stored. This is currently 915 * a catch-all for anything not individually iCRCd. 916 */ 917 #define HAMMER2_VOL_ICRC_SECT0 7 918 #define HAMMER2_VOL_ICRC_SECT1 6 919 920 #define HAMMER2_VOLUME_BYTES 65536 921 922 #define HAMMER2_VOLUME_ICRC0_OFF 0 923 #define HAMMER2_VOLUME_ICRC1_OFF 512 924 #define HAMMER2_VOLUME_ICRCVH_OFF 0 925 926 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 927 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 928 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 929 930 #define HAMMER2_VOL_VERSION_MIN 1 931 #define HAMMER2_VOL_VERSION_DEFAULT 1 932 #define HAMMER2_VOL_VERSION_WIP 2 933 934 #define HAMMER2_NUM_VOLHDRS 4 935 936 union hammer2_media_data { 937 hammer2_volume_data_t voldata; 938 hammer2_inode_data_t ipdata; 939 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 940 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 941 char buf[HAMMER2_PBUFSIZE]; 942 }; 943 944 typedef union hammer2_media_data hammer2_media_data_t; 945 946 #endif 947