1/*- 2 * Copyright (c) 2011-2018 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 * $FreeBSD$ 36 */ 37 38#ifndef _HAMMER2_DISK_H_ 39#define _HAMMER2_DISK_H_ 40 41#ifndef _SYS_UUID_H_ 42#include <sys/uuid.h> 43#endif 44#ifndef _SYS_DMSG_H_ 45/* 46 * dmsg_hdr must be 64 bytes 47 */ 48struct dmsg_hdr { 49 uint16_t magic; /* 00 sanity, synchro, endian */ 50 uint16_t reserved02; /* 02 */ 51 uint32_t salt; /* 04 random salt helps w/crypto */ 52 53 uint64_t msgid; /* 08 message transaction id */ 54 uint64_t circuit; /* 10 circuit id or 0 */ 55 uint64_t reserved18; /* 18 */ 56 57 uint32_t cmd; /* 20 flags | cmd | hdr_size / ALIGN */ 58 uint32_t aux_crc; /* 24 auxillary data crc */ 59 uint32_t aux_bytes; /* 28 auxillary data length (bytes) */ 60 uint32_t error; /* 2C error code or 0 */ 61 uint64_t aux_descr; /* 30 negotiated OOB data descr */ 62 uint32_t reserved38; /* 38 */ 63 uint32_t hdr_crc; /* 3C (aligned) extended header crc */ 64}; 65 66typedef struct dmsg_hdr dmsg_hdr_t; 67#endif 68 69/* 70 * The structures below represent the on-disk media structures for the HAMMER2 71 * filesystem. Note that all fields for on-disk structures are naturally 72 * aligned. The host endian format is typically used - compatibility is 73 * possible if the implementation detects reversed endian and adjusts accesses 74 * accordingly. 75 * 76 * HAMMER2 primarily revolves around the directory topology: inodes, 77 * directory entries, and block tables. Block device buffer cache buffers 78 * are always 64KB. Logical file buffers are typically 16KB. All data 79 * references utilize 64-bit byte offsets. 80 * 81 * Free block management is handled independently using blocks reserved by 82 * the media topology. 83 */ 84 85/* 86 * The data at the end of a file or directory may be a fragment in order 87 * to optimize storage efficiency. The minimum fragment size is 1KB. 88 * Since allocations are in powers of 2 fragments must also be sized in 89 * powers of 2 (1024, 2048, ... 65536). 90 * 91 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K), 92 * which is 2^16. Larger extents may be supported in the future. Smaller 93 * fragments might be supported in the future (down to 64 bytes is possible), 94 * but probably will not be. 95 * 96 * A full indirect block use supports 512 x 128-byte blockrefs in a 64KB 97 * buffer. Indirect blocks down to 1KB are supported to keep small 98 * directories small. 99 * 100 * A maximally sized file (2^64-1 bytes) requires ~6 indirect block levels 101 * using 64KB indirect blocks (128 byte refs, 512 or radix 9 per indblk). 102 * 103 * 16(datablk) + 9 + 9 + 9 + 9 + 9 + 9 = ~70. 104 * 16(datablk) + 7 + 9 + 9 + 9 + 9 + 9 = ~68. (smaller top level indblk) 105 * 106 * The actual depth depends on copies redundancy and whether the filesystem 107 * has chosen to use a smaller indirect block size at the top level or not. 108 */ 109#define HAMMER2_ALLOC_MIN 1024 /* minimum allocation size */ 110#define HAMMER2_RADIX_MIN 10 /* minimum allocation size 2^N */ 111#define HAMMER2_ALLOC_MAX 65536 /* maximum allocation size */ 112#define HAMMER2_RADIX_MAX 16 /* maximum allocation size 2^N */ 113#define HAMMER2_RADIX_KEY 64 /* number of bits in key */ 114 115/* 116 * MINALLOCSIZE - The minimum allocation size. This can be smaller 117 * or larger than the minimum physical IO size. 118 * 119 * NOTE: Should not be larger than 1K since inodes 120 * are 1K. 121 * 122 * MINIOSIZE - The minimum IO size. This must be less than 123 * or equal to HAMMER2_LBUFSIZE. 124 * 125 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups. 126 * 127 * HAMMER2_PBUFSIZE - Topological block size used by files for all 128 * blocks except the block straddling EOF. 129 * 130 * HAMMER2_SEGSIZE - Allocation map segment size, typically 4MB 131 * (space represented by a level0 bitmap). 132 */ 133 134#define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX) 135#define HAMMER2_SEGRADIX HAMMER2_FREEMAP_LEVEL0_RADIX 136 137#define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 138#define HAMMER2_PBUFSIZE 65536 139#define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 140#define HAMMER2_LBUFSIZE 16384 141 142/* 143 * Generally speaking we want to use 16K and 64K I/Os 144 */ 145#define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX 146#define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE 147 148#define HAMMER2_IND_BYTES_MIN 4096 149#define HAMMER2_IND_BYTES_NOM HAMMER2_LBUFSIZE 150#define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 151#define HAMMER2_IND_RADIX_MIN 12 152#define HAMMER2_IND_RADIX_NOM HAMMER2_LBUFRADIX 153#define HAMMER2_IND_RADIX_MAX HAMMER2_PBUFRADIX 154#define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 155 sizeof(hammer2_blockref_t)) 156#define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 157 sizeof(hammer2_blockref_t)) 158 159/* 160 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 161 * any element can occur at any index and holes can be anywhere. As a 162 * future optimization we will be able to flag that such arrays are sorted 163 * and thus optimize lookups, but for now we don't. 164 * 165 * Inodes embed either 512 bytes of direct data or an array of 4 blockrefs, 166 * resulting in highly efficient storage for files <= 512 bytes and for files 167 * <= 512KB. Up to 4 directory entries can be referenced from a directory 168 * without requiring an indirect block. 169 * 170 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented), 171 * or 64KB (1024 blockrefs / ~64MB represented). 172 */ 173#define HAMMER2_SET_RADIX 2 /* radix 2 = 4 entries */ 174#define HAMMER2_SET_COUNT (1 << HAMMER2_SET_RADIX) 175#define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 176#define HAMMER2_EMBEDDED_RADIX 9 177 178#define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 179#define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 180#define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 181 182#define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 183#define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 184#define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 185#define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 186#define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 187 188#define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 189 190/* 191 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment 192 * contains the volume header (or backup volume header), the free block 193 * table, and possibly other information in the future. A 4MB segment for 194 * freemap is reserved at the beginning of every 1GB. 195 * 196 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 197 * 198 * ========== 199 * 0 volume header (for the first four 2GB zones) 200 * 1 freemap00 level1 FREEMAP_LEAF (256 x 128B bitmap data per 1GB) 201 * 2 level2 FREEMAP_NODE (256 x 128B indirect block per 256GB) 202 * 3 level3 FREEMAP_NODE (256 x 128B indirect block per 64TB) 203 * 4 level4 FREEMAP_NODE (256 x 128B indirect block per 16PB) 204 * 5 level5 FREEMAP_NODE (256 x 128B indirect block per 4EB) 205 * 6 freemap01 level1 (rotation) 206 * 7 level2 207 * 8 level3 208 * 9 level4 209 * 10 level5 210 * 11 freemap02 level1 (rotation) 211 * 12 level2 212 * 13 level3 213 * 14 level4 214 * 15 level5 215 * 16 freemap03 level1 (rotation) 216 * 17 level2 217 * 18 level3 218 * 19 level4 219 * 20 level5 220 * 21 freemap04 level1 (rotation) 221 * 22 level2 222 * 23 level3 223 * 24 level4 224 * 25 level5 225 * 26 freemap05 level1 (rotation) 226 * 27 level2 227 * 28 level3 228 * 29 level4 229 * 30 level5 230 * 31 freemap06 level1 (rotation) 231 * 32 level2 232 * 33 level3 233 * 34 level4 234 * 35 level5 235 * 36 freemap07 level1 (rotation) 236 * 37 level2 237 * 38 level3 238 * 39 level4 239 * 40 level5 240 * 41 unused 241 * .. unused 242 * 63 unused 243 * ========== 244 * 245 * The first four 2GB zones contain volume headers and volume header backups. 246 * After that the volume header block# is reserved for future use. Similarly, 247 * there are many blocks related to various Freemap levels which are not 248 * used in every segment and those are also reserved for future use. 249 * Note that each FREEMAP_LEAF or FREEMAP_NODE uses 32KB out of 64KB slot. 250 * 251 * Freemap (see the FREEMAP document) 252 * 253 * The freemap utilizes blocks #1-40 in 8 sets of 5 blocks. Each block in 254 * a set represents a level of depth in the freemap topology. Eight sets 255 * exist to prevent live updates from disturbing the state of the freemap 256 * were a crash/reboot to occur. That is, a live update is not committed 257 * until the update's flush reaches the volume root. There are FOUR volume 258 * roots representing the last four synchronization points, so the freemap 259 * must be consistent no matter which volume root is chosen by the mount 260 * code. 261 * 262 * Each freemap set is 5 x 64K blocks and represents the 1GB, 256GB, 64TB, 263 * 16PB and 4EB indirect map. The volume header itself has a set of 4 freemap 264 * blockrefs representing another 2 bits, giving us a total 64 bits of 265 * representable address space. 266 * 267 * The Level 0 64KB block represents 1GB of storage represented by 32KB 268 * (256 x struct hammer2_bmap_data). Each structure represents 4MB of storage 269 * and has a 512 bit bitmap, using 2 bits to represent a 16KB chunk of 270 * storage. These 2 bits represent the following states: 271 * 272 * 00 Free 273 * 01 (reserved) (Possibly partially allocated) 274 * 10 Possibly free 275 * 11 Allocated 276 * 277 * One important thing to note here is that the freemap resolution is 16KB, 278 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps 279 * track of sub-allocations in memory, which means that on a unmount or reboot 280 * the entire 16KB of a partially allocated block will be considered fully 281 * allocated. It is possible for fragmentation to build up over time, but 282 * defragmentation is fairly easy to accomplish since all modifications 283 * allocate a new block. 284 * 285 * The Second thing to note is that due to the way snapshots and inode 286 * replication works, deleting a file cannot immediately free the related 287 * space. Furthermore, deletions often do not bother to traverse the 288 * block subhierarchy being deleted. And to go even further, whole 289 * sub-directory trees can be deleted simply by deleting the directory inode 290 * at the top. So even though we have a symbol to represent a 'possibly free' 291 * block (binary 10), only the bulk free scanning code can actually use it. 292 * Normal 'rm's or other deletions do not. 293 * 294 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 295 * (i.e. a multiple of 4MB). VOLUME_ALIGN must be >= ZONE_SEG. 296 * 297 * In Summary: 298 * 299 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block 300 * from the next set). The new copy is reused until a flush occurs at 301 * which point the next modification will then rotate to the next set. 302 */ 303#define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 304#define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 305#define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 306#define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 307 308#define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 309#define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 310#define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 311#define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 312 313#define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 314#define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 315#define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 316#define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 317#define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 318 319#define HAMMER2_ZONE_FREEMAP_INC 5 /* 5 deep */ 320 321#define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 322#define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */ 323#define HAMMER2_ZONE_FREEMAP_01 6 /* normal freemap rotation */ 324#define HAMMER2_ZONE_FREEMAP_02 11 /* normal freemap rotation */ 325#define HAMMER2_ZONE_FREEMAP_03 16 /* normal freemap rotation */ 326#define HAMMER2_ZONE_FREEMAP_04 21 /* normal freemap rotation */ 327#define HAMMER2_ZONE_FREEMAP_05 26 /* normal freemap rotation */ 328#define HAMMER2_ZONE_FREEMAP_06 31 /* normal freemap rotation */ 329#define HAMMER2_ZONE_FREEMAP_07 36 /* normal freemap rotation */ 330#define HAMMER2_ZONE_FREEMAP_END 41 /* (non-inclusive) */ 331 332#define HAMMER2_ZONE_UNUSED41 41 333#define HAMMER2_ZONE_UNUSED42 42 334#define HAMMER2_ZONE_UNUSED43 43 335#define HAMMER2_ZONE_UNUSED44 44 336#define HAMMER2_ZONE_UNUSED45 45 337#define HAMMER2_ZONE_UNUSED46 46 338#define HAMMER2_ZONE_UNUSED47 47 339#define HAMMER2_ZONE_UNUSED48 48 340#define HAMMER2_ZONE_UNUSED49 49 341#define HAMMER2_ZONE_UNUSED50 50 342#define HAMMER2_ZONE_UNUSED51 51 343#define HAMMER2_ZONE_UNUSED52 52 344#define HAMMER2_ZONE_UNUSED53 53 345#define HAMMER2_ZONE_UNUSED54 54 346#define HAMMER2_ZONE_UNUSED55 55 347#define HAMMER2_ZONE_UNUSED56 56 348#define HAMMER2_ZONE_UNUSED57 57 349#define HAMMER2_ZONE_UNUSED58 58 350#define HAMMER2_ZONE_UNUSED59 59 351#define HAMMER2_ZONE_UNUSED60 60 352#define HAMMER2_ZONE_UNUSED61 61 353#define HAMMER2_ZONE_UNUSED62 62 354#define HAMMER2_ZONE_UNUSED63 63 355#define HAMMER2_ZONE_END 64 /* non-inclusive */ 356 357#define HAMMER2_NFREEMAPS 8 /* FREEMAP_00 - FREEMAP_07 */ 358 359 /* relative to FREEMAP_x */ 360#define HAMMER2_ZONEFM_LEVEL1 0 /* 1GB leafmap */ 361#define HAMMER2_ZONEFM_LEVEL2 1 /* 256GB indmap */ 362#define HAMMER2_ZONEFM_LEVEL3 2 /* 64TB indmap */ 363#define HAMMER2_ZONEFM_LEVEL4 3 /* 16PB indmap */ 364#define HAMMER2_ZONEFM_LEVEL5 4 /* 4EB indmap */ 365/* LEVEL6 is a set of 4 blockrefs in the volume header 16EB */ 366 367/* 368 * Freemap radix. Assumes a set-count of 4, 128-byte blockrefs, 369 * 32KB indirect block for freemap (LEVELN_PSIZE below). 370 * 371 * Leaf entry represents 4MB of storage broken down into a 512-bit 372 * bitmap, 2-bits per entry. So course bitmap item represents 16KB. 373 */ 374#if HAMMER2_SET_COUNT != 4 375#error "hammer2_disk.h - freemap assumes SET_COUNT is 4" 376#endif 377#define HAMMER2_FREEMAP_LEVEL6_RADIX 64 /* 16EB (end) */ 378#define HAMMER2_FREEMAP_LEVEL5_RADIX 62 /* 4EB */ 379#define HAMMER2_FREEMAP_LEVEL4_RADIX 54 /* 16PB */ 380#define HAMMER2_FREEMAP_LEVEL3_RADIX 46 /* 64TB */ 381#define HAMMER2_FREEMAP_LEVEL2_RADIX 38 /* 256GB */ 382#define HAMMER2_FREEMAP_LEVEL1_RADIX 30 /* 1GB */ 383#define HAMMER2_FREEMAP_LEVEL0_RADIX 22 /* 4MB (128by in l-1 leaf) */ 384 385#define HAMMER2_FREEMAP_LEVELN_PSIZE 32768 /* physical bytes */ 386 387#define HAMMER2_FREEMAP_LEVEL5_SIZE ((hammer2_off_t)1 << \ 388 HAMMER2_FREEMAP_LEVEL5_RADIX) 389#define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \ 390 HAMMER2_FREEMAP_LEVEL4_RADIX) 391#define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \ 392 HAMMER2_FREEMAP_LEVEL3_RADIX) 393#define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \ 394 HAMMER2_FREEMAP_LEVEL2_RADIX) 395#define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \ 396 HAMMER2_FREEMAP_LEVEL1_RADIX) 397#define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \ 398 HAMMER2_FREEMAP_LEVEL0_RADIX) 399 400#define HAMMER2_FREEMAP_LEVEL5_MASK (HAMMER2_FREEMAP_LEVEL5_SIZE - 1) 401#define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1) 402#define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1) 403#define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1) 404#define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1) 405#define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1) 406 407#define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 408 sizeof(hammer2_bmap_data_t)) 409 410/* 411 * XXX I made a mistake and made the reserved area begin at each LEVEL1 zone, 412 * which is on a 1GB demark. This will eat a little more space but for 413 * now we retain compatibility and make FMZONEBASE every 1GB 414 */ 415#define H2FMZONEBASE(key) ((key) & ~HAMMER2_FREEMAP_LEVEL1_MASK) 416#define H2FMBASE(key, radix) ((key) & ~(((hammer2_off_t)1 << (radix)) - 1)) 417 418/* 419 * 16KB bitmap granularity (x2 bits per entry). 420 */ 421#define HAMMER2_FREEMAP_BLOCK_RADIX 14 422#define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 423#define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 424 425/* 426 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE. 427 * 428 * 8 x 64-bit elements, 2 bits per block. 429 * 32 blocks (radix 5) per element. 430 * representing INDEX_SIZE bytes worth of storage per element. 431 */ 432 433typedef uint64_t hammer2_bitmap_t; 434 435#define HAMMER2_BMAP_ALLONES ((hammer2_bitmap_t)-1) 436#define HAMMER2_BMAP_ELEMENTS 8 437#define HAMMER2_BMAP_BITS_PER_ELEMENT 64 438#define HAMMER2_BMAP_INDEX_RADIX 5 /* 32 blocks per element */ 439#define HAMMER2_BMAP_BLOCKS_PER_ELEMENT (1 << HAMMER2_BMAP_INDEX_RADIX) 440 441#define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * \ 442 HAMMER2_BMAP_BLOCKS_PER_ELEMENT) 443#define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1) 444 445#define HAMMER2_BMAP_SIZE (HAMMER2_BMAP_INDEX_SIZE * \ 446 HAMMER2_BMAP_ELEMENTS) 447#define HAMMER2_BMAP_MASK (HAMMER2_BMAP_SIZE - 1) 448 449/* 450 * Two linear areas can be reserved after the initial 4MB segment in the base 451 * zone (the one starting at offset 0). These areas are NOT managed by the 452 * block allocator and do not fall under HAMMER2 crc checking rules based 453 * at the volume header (but can be self-CRCd internally, depending). 454 */ 455#define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 456#define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 457#define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 458 459#define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 460#define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 461#define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 462 463/* 464 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 465 * Transaction ids are monotonic. 466 * 467 * We utilize 32-bit iSCSI CRCs. 468 */ 469typedef uint64_t hammer2_tid_t; 470typedef uint64_t hammer2_off_t; 471typedef uint64_t hammer2_key_t; 472typedef uint32_t hammer2_crc32_t; 473 474/* 475 * Miscellanious ranges (all are unsigned). 476 */ 477#define HAMMER2_TID_MIN 1ULL 478#define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 479#define HAMMER2_KEY_MIN 0ULL 480#define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 481#define HAMMER2_OFFSET_MIN 0ULL 482#define HAMMER2_OFFSET_MAX 0xFFFFFFFFFFFFFFFFULL 483 484/* 485 * HAMMER2 data offset special cases and masking. 486 * 487 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 488 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 489 * 490 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 491 * of the data offset field specifies how large the data chunk being pointed 492 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 493 * needed in the low bits of the data offset field). However, the practical 494 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 495 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but 496 * we fully intend to support larger extents in the future. 497 * 498 * WARNING! A radix of 0 (such as when data_off is all 0's) is a special 499 * case which means no data associated with the blockref, and 500 * not the '1 byte' it would otherwise calculate to. 501 */ 502#define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 503#define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 504#define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 505#define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 506#define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 507#define HAMMER2_MAX_COPIES 6 508 509/* 510 * HAMMER2 directory support and pre-defined keys 511 */ 512#define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 513#define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 514#define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 515#define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 516#define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 517 518#define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 519#define HAMMER2_BOOT_KEY 0xd9b36ce135528000ULL /* sroot to BOOT PFS */ 520 521/************************************************************************ 522 * DMSG SUPPORT * 523 ************************************************************************ 524 * LNK_VOLCONF 525 * 526 * All HAMMER2 directories directly under the super-root on your local 527 * media can be mounted separately, even if they share the same physical 528 * device. 529 * 530 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 531 * cluster via local media. The local media does not have to participate 532 * in the cluster, other than to provide the hammer2_volconf[] array and 533 * root inode for the mount. 534 * 535 * This is important: The mount device path you specify serves to bootstrap 536 * your entry into the cluster, but your mount will make active connections 537 * to ALL copy elements in the hammer2_volconf[] array which match the 538 * PFSID of the directory in the super-root that you specified. The local 539 * media path does not have to be mentioned in this array but becomes part 540 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 541 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 542 * 543 * The actual cluster may be far larger than the elements you list in the 544 * hammer2_volconf[] array. You list only the elements you wish to 545 * directly connect to and you are able to access the rest of the cluster 546 * indirectly through those connections. 547 * 548 * WARNING! This structure must be exactly 128 bytes long for its config 549 * array to fit in the volume header. 550 */ 551struct hammer2_volconf { 552 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 553 uint8_t inprog; /* 01 operation in progress, or 0 */ 554 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 555 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 556 uint16_t flags; /* 04-05 flags field */ 557 uint8_t error; /* 06 last operational error */ 558 uint8_t priority; /* 07 priority and round-robin flag */ 559 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 560 uint8_t reserved08[23]; /* 09-1F */ 561 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 562 uint8_t label[16]; /* 30-3F import/export label */ 563 uint8_t path[64]; /* 40-7F target specification string or key */ 564} __packed; 565 566typedef struct hammer2_volconf hammer2_volconf_t; 567 568#define DMSG_VOLF_ENABLED 0x0001 569#define DMSG_VOLF_INPROG 0x0002 570#define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 571#define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 572#define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 573 574struct dmsg_lnk_hammer2_volconf { 575 dmsg_hdr_t head; 576 hammer2_volconf_t copy; /* copy spec */ 577 int32_t index; 578 int32_t unused01; 579 uuid_t mediaid; 580 int64_t reserved02[32]; 581} __packed; 582 583typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 584 585#define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 586 dmsg_lnk_hammer2_volconf) 587 588#define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 589 590/* 591 * HAMMER2 directory entry header (embedded in blockref) exactly 16 bytes 592 */ 593struct hammer2_dirent_head { 594 hammer2_tid_t inum; /* inode number */ 595 uint16_t namlen; /* name length */ 596 uint8_t type; /* OBJTYPE_* */ 597 uint8_t unused0B; 598 uint8_t unused0C[4]; 599} __packed; 600 601typedef struct hammer2_dirent_head hammer2_dirent_head_t; 602 603/* 604 * The media block reference structure. This forms the core of the HAMMER2 605 * media topology recursion. This 128-byte data structure is embedded in the 606 * volume header, in inodes (which are also directory entries), and in 607 * indirect blocks. 608 * 609 * A blockref references a single media item, which typically can be a 610 * directory entry (aka inode), indirect block, or data block. 611 * 612 * The primary feature a blockref represents is the ability to validate 613 * the entire tree underneath it via its check code. Any modification to 614 * anything propagates up the blockref tree all the way to the root, replacing 615 * the related blocks and compounding the generated check code. 616 * 617 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, or as 618 * complex as a 512 bit cryptographic hash. I originally used a 64-byte 619 * blockref but later expanded it to 128 bytes to be able to support the 620 * larger check code as well as to embed statistics for quota operation. 621 * 622 * Simple check codes are not sufficient for unverified dedup. Even with 623 * a maximally-sized check code unverified dedup should only be used in 624 * in subdirectory trees where you do not need 100% data integrity. 625 * 626 * Unverified dedup is deduping based on meta-data only without verifying 627 * that the data blocks are actually identical. Verified dedup guarantees 628 * integrity but is a far more I/O-expensive operation. 629 * 630 * -- 631 * 632 * mirror_tid - per cluster node modified (propagated upward by flush) 633 * modify_tid - clc record modified (not propagated). 634 * update_tid - clc record updated (propagated upward on verification) 635 * 636 * CLC - Stands for 'Cluster Level Change', identifiers which are identical 637 * within the topology across all cluster nodes (when fully 638 * synchronized). 639 * 640 * NOTE: The range of keys represented by the blockref is (key) to 641 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 642 * blocks bottom-up, inserting a new root when radix expansion 643 * is required. 644 * 645 * leaf_count - Helps manage leaf collapse calculations when indirect 646 * blocks become mostly empty. This value caps out at 647 * HAMMER2_BLOCKREF_LEAF_MAX (65535). 648 * 649 * Used by the chain code to determine when to pull leafs up 650 * from nearly empty indirect blocks. For the purposes of this 651 * calculation, BREF_TYPE_INODE is considered a leaf, along 652 * with DIRENT and DATA. 653 * 654 * RESERVED FIELDS 655 * 656 * A number of blockref fields are reserved and should generally be set to 657 * 0 for future compatibility. 658 * 659 * FUTURE BLOCKREF EXPANSION 660 * 661 * CONTENT ADDRESSABLE INDEXING (future) - Using a 256 or 512-bit check code. 662 */ 663struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 664 uint8_t type; /* type of underlying item */ 665 uint8_t methods; /* check method & compression method */ 666 uint8_t copyid; /* specify which copy this is */ 667 uint8_t keybits; /* #of keybits masked off 0=leaf */ 668 uint8_t vradix; /* virtual data/meta-data size */ 669 uint8_t flags; /* blockref flags */ 670 uint16_t leaf_count; /* leaf aggregation count */ 671 hammer2_key_t key; /* key specification */ 672 hammer2_tid_t mirror_tid; /* media flush topology & freemap */ 673 hammer2_tid_t modify_tid; /* clc modify (not propagated) */ 674 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 675 hammer2_tid_t update_tid; /* clc modify (propagated upward) */ 676 union { 677 char buf[16]; 678 679 /* 680 * Directory entry header (BREF_TYPE_DIRENT) 681 * 682 * NOTE: check.buf contains filename if <= 64 bytes. Longer 683 * filenames are stored in a data reference of size 684 * HAMMER2_ALLOC_MIN (at least 256, typically 1024). 685 * 686 * NOTE: inode structure may contain a copy of a recently 687 * associated filename, for recovery purposes. 688 * 689 * NOTE: Superroot entries are INODEs, not DIRENTs. Code 690 * allows both cases. 691 */ 692 hammer2_dirent_head_t dirent; 693 694 /* 695 * Statistics aggregation (BREF_TYPE_INODE, BREF_TYPE_INDIRECT) 696 */ 697 struct { 698 hammer2_key_t data_count; 699 hammer2_key_t inode_count; 700 } stats; 701 } embed; 702 union { /* check info */ 703 char buf[64]; 704 struct { 705 uint32_t value; 706 uint32_t reserved[15]; 707 } iscsi32; 708 struct { 709 uint64_t value; 710 uint64_t reserved[7]; 711 } xxhash64; 712 struct { 713 char data[24]; 714 char reserved[40]; 715 } sha192; 716 struct { 717 char data[32]; 718 char reserved[32]; 719 } sha256; 720 struct { 721 char data[64]; 722 } sha512; 723 724 /* 725 * Freemap hints are embedded in addition to the icrc32. 726 * 727 * bigmask - Radixes available for allocation (0-31). 728 * Heuristical (may be permissive but not 729 * restrictive). Typically only radix values 730 * 10-16 are used (i.e. (1<<10) through (1<<16)). 731 * 732 * avail - Total available space remaining, in bytes 733 */ 734 struct { 735 uint32_t icrc32; 736 uint32_t bigmask; /* available radixes */ 737 uint64_t avail; /* total available bytes */ 738 char reserved[48]; 739 } freemap; 740 } check; 741} __packed; 742 743typedef struct hammer2_blockref hammer2_blockref_t; 744 745#define HAMMER2_BLOCKREF_BYTES 128 /* blockref struct in bytes */ 746#define HAMMER2_BLOCKREF_RADIX 7 747 748#define HAMMER2_BLOCKREF_LEAF_MAX 65535 749 750/* 751 * On-media and off-media blockref types. 752 * 753 * types >= 128 are pseudo values that should never be present on-media. 754 */ 755#define HAMMER2_BREF_TYPE_EMPTY 0 756#define HAMMER2_BREF_TYPE_INODE 1 757#define HAMMER2_BREF_TYPE_INDIRECT 2 758#define HAMMER2_BREF_TYPE_DATA 3 759#define HAMMER2_BREF_TYPE_DIRENT 4 760#define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 761#define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 762#define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 763#define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 764 765#define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 766#define HAMMER2_BREF_FLAG_ZERO 0x02 767 768/* 769 * Encode/decode check mode and compression mode for 770 * bref.methods. The compression level is not encoded in 771 * bref.methods. 772 */ 773#define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 774#define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 775#define HAMMER2_ENC_COMP(n) ((n) & 15) 776#define HAMMER2_DEC_COMP(n) ((n) & 15) 777 778#define HAMMER2_CHECK_NONE 0 779#define HAMMER2_CHECK_DISABLED 1 780#define HAMMER2_CHECK_ISCSI32 2 781#define HAMMER2_CHECK_XXHASH64 3 782#define HAMMER2_CHECK_SHA192 4 783#define HAMMER2_CHECK_FREEMAP 5 784 785#define HAMMER2_CHECK_DEFAULT HAMMER2_CHECK_XXHASH64 786 787/* user-specifiable check modes only */ 788#define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 789 "xxhash64", "sha192" } 790#define HAMMER2_CHECK_STRINGS_COUNT 5 791 792/* 793 * Encode/decode check or compression algorithm request in 794 * ipdata->meta.check_algo and ipdata->meta.comp_algo. 795 */ 796#define HAMMER2_ENC_ALGO(n) (n) 797#define HAMMER2_DEC_ALGO(n) ((n) & 15) 798#define HAMMER2_ENC_LEVEL(n) ((n) << 4) 799#define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 800 801#define HAMMER2_COMP_NONE 0 802#define HAMMER2_COMP_AUTOZERO 1 803#define HAMMER2_COMP_LZ4 2 804#define HAMMER2_COMP_ZLIB 3 805 806#define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 807#define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 808#define HAMMER2_COMP_STRINGS_COUNT 4 809 810/* 811 * Passed to hammer2_chain_create(), causes methods to be inherited from 812 * parent. 813 */ 814#define HAMMER2_METH_DEFAULT -1 815 816/* 817 * HAMMER2 block references are collected into sets of 4 blockrefs. These 818 * sets are fully associative, meaning the elements making up a set are 819 * not sorted in any way and may contain duplicate entries, holes, or 820 * entries which shortcut multiple levels of indirection. Sets are used 821 * in various ways: 822 * 823 * (1) When redundancy is desired a set may contain several duplicate 824 * entries pointing to different copies of the same data. Up to 4 copies 825 * are supported. 826 * 827 * (2) The blockrefs in a set can shortcut multiple levels of indirections 828 * within the bounds imposed by the parent of set. 829 * 830 * When a set fills up another level of indirection is inserted, moving 831 * some or all of the set's contents into indirect blocks placed under the 832 * set. This is a top-down approach in that indirect blocks are not created 833 * until the set actually becomes full (that is, the entries in the set can 834 * shortcut the indirect blocks when the set is not full). Depending on how 835 * things are filled multiple indirect blocks will eventually be created. 836 * 837 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 838 * are also treated as fully set-associative. 839 */ 840struct hammer2_blockset { 841 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 842}; 843 844typedef struct hammer2_blockset hammer2_blockset_t; 845 846/* 847 * Catch programmer snafus 848 */ 849#if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 850#error "hammer2 direct radix is incorrect" 851#endif 852#if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 853#error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 854#endif 855#if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 856#error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 857#endif 858 859/* 860 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 861 * 862 * Each 128-byte entry contains the bitmap and meta-data required to manage 863 * a LEVEL0 (4MB) block of storage. The storage is managed in 256 x 16KB 864 * chunks. 865 * 866 * A smaller allocation granularity is supported via a linear iterator and/or 867 * must otherwise be tracked in ram. 868 * 869 * (data structure must be 128 bytes exactly) 870 * 871 * linear - A BYTE linear allocation offset used for sub-16KB allocations 872 * only. May contain values between 0 and 4MB. Must be ignored 873 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 874 * used to sub-allocate within the 16KB block (which is already 875 * marked as allocated in the bitmap). 876 * 877 * Sub-allocations need only be 1KB-aligned and do not have to be 878 * size-aligned, and 16KB or larger allocations do not update this 879 * field, resulting in pretty good packing. 880 * 881 * Please note that file data granularity may be limited by 882 * other issues such as buffer cache direct-mapping and the 883 * desire to support sector sizes up to 16KB (so H2 only issues 884 * I/O's in multiples of 16KB anyway). 885 * 886 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 887 * free. Used to cluster device buffers so all elements must have 888 * the same device block size, but may mix logical sizes. 889 * 890 * Typically integrated with the blockref type in the upper 8 bits 891 * to localize inodes and indrect blocks, improving bulk free scans 892 * and directory scans. 893 * 894 * bitmap - Two bits per 16KB allocation block arranged in arrays of 895 * 64-bit elements, 256x2 bits representing ~4MB worth of media 896 * storage. Bit patterns are as follows: 897 * 898 * 00 Unallocated 899 * 01 (reserved) 900 * 10 Possibly free 901 * 11 Allocated 902 */ 903struct hammer2_bmap_data { 904 int32_t linear; /* 00 linear sub-granular allocation offset */ 905 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 906 uint8_t reserved06; /* 06 */ 907 uint8_t reserved07; /* 07 */ 908 uint32_t reserved08; /* 08 */ 909 uint32_t reserved0C; /* 0C */ 910 uint32_t reserved10; /* 10 */ 911 uint32_t reserved14; /* 14 */ 912 uint32_t reserved18; /* 18 */ 913 uint32_t avail; /* 1C */ 914 uint32_t reserved20[8]; /* 20-3F 256 bits manages 128K/1KB/2-bits */ 915 /* 40-7F 512 bits manages 4MB of storage */ 916 hammer2_bitmap_t bitmapq[HAMMER2_BMAP_ELEMENTS]; 917} __packed; 918 919typedef struct hammer2_bmap_data hammer2_bmap_data_t; 920 921/* 922 * XXX "Inodes ARE directory entries" is no longer the case. Hardlinks are 923 * dirents which refer to the same inode#, which is how filesystems usually 924 * implement hardlink. The following comments need to be updated. 925 * 926 * In HAMMER2 inodes ARE directory entries, with a special exception for 927 * hardlinks. The inode number is stored in the inode rather than being 928 * based on the location of the inode (since the location moves every time 929 * the inode or anything underneath the inode is modified). 930 * 931 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 932 * for the filename, and 512 bytes worth of direct file data OR an embedded 933 * blockset. The in-memory hammer2_inode structure contains only the mostly- 934 * node-independent meta-data portion (some flags are node-specific and will 935 * not be synchronized). The rest of the inode is node-specific and chain I/O 936 * is required to obtain it. 937 * 938 * Directories represent one inode per blockref. Inodes are not laid out 939 * as a file but instead are represented by the related blockrefs. The 940 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 941 * that blocksets are fully associative, so a certain degree efficiency is 942 * achieved just from that. 943 * 944 * Up to 512 bytes of direct data can be embedded in an inode, and since 945 * inodes are essentially directory entries this also means that small data 946 * files end up simply being laid out linearly in the directory, resulting 947 * in fewer seeks and highly optimal access. 948 * 949 * The compression mode can be changed at any time in the inode and is 950 * recorded on a blockref-by-blockref basis. 951 * 952 * Hardlinks are supported via the inode map. Essentially the way a hardlink 953 * works is that all individual directory entries representing the same file 954 * are special cased and specify the same inode number. The actual file 955 * is placed in the nearest parent directory that is parent to all instances 956 * of the hardlink. If all hardlinks to a file are in the same directory 957 * the actual file will also be placed in that directory. This file uses 958 * the inode number as the directory entry key and is invisible to normal 959 * directory scans. Real directory entry keys are differentiated from the 960 * inode number key via bit 63. Access to the hardlink silently looks up 961 * the real file and forwards all operations to that file. Removal of the 962 * last hardlink also removes the real file. 963 */ 964#define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 965#define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 966#define HAMMER2_INODE_VERSION_ONE 1 967 968#define HAMMER2_INODE_START 1024 /* dynamically allocated */ 969 970struct hammer2_inode_meta { 971 uint16_t version; /* 0000 inode data version */ 972 uint8_t reserved02; /* 0002 */ 973 uint8_t pfs_subtype; /* 0003 pfs sub-type */ 974 975 /* 976 * core inode attributes, inode type, misc flags 977 */ 978 uint32_t uflags; /* 0004 chflags */ 979 uint32_t rmajor; /* 0008 available for device nodes */ 980 uint32_t rminor; /* 000C available for device nodes */ 981 uint64_t ctime; /* 0010 inode change time */ 982 uint64_t mtime; /* 0018 modified time */ 983 uint64_t atime; /* 0020 access time (unsupported) */ 984 uint64_t btime; /* 0028 birth time */ 985 uuid_t uid; /* 0030 uid / degenerate unix uid */ 986 uuid_t gid; /* 0040 gid / degenerate unix gid */ 987 988 uint8_t type; /* 0050 object type */ 989 uint8_t op_flags; /* 0051 operational flags */ 990 uint16_t cap_flags; /* 0052 capability flags */ 991 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 992 993 /* 994 * inode size, identification, localized recursive configuration 995 * for compression and backup copies. 996 * 997 * NOTE: Nominal parent inode number (iparent) is only applicable 998 * for directories but can also help for files during 999 * catastrophic recovery. 1000 */ 1001 hammer2_tid_t inum; /* 0058 inode number */ 1002 hammer2_off_t size; /* 0060 size of file */ 1003 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 1004 hammer2_tid_t iparent; /* 0070 nominal parent inum */ 1005 hammer2_key_t name_key; /* 0078 full filename key */ 1006 uint16_t name_len; /* 0080 filename length */ 1007 uint8_t ncopies; /* 0082 ncopies to local media */ 1008 uint8_t comp_algo; /* 0083 compression request & algo */ 1009 1010 /* 1011 * These fields are currently only applicable to PFSROOTs. 1012 * 1013 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 1014 * identify an instance of a PFS in the cluster because 1015 * a mount may contain more than one copy of the PFS as 1016 * a separate node. {pfs_clid, pfs_fsid} must be used for 1017 * registration in the cluster. 1018 */ 1019 uint8_t target_type; /* 0084 hardlink target type */ 1020 uint8_t check_algo; /* 0085 check code request & algo */ 1021 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 1022 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 1023 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 1024 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 1025 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 1026 1027 /* 1028 * Quotas and aggregate sub-tree inode and data counters. Note that 1029 * quotas are not replicated downward, they are explicitly set by 1030 * the sysop and in-memory structures keep track of inheritance. 1031 */ 1032 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 1033 hammer2_key_t unusedB8; /* 00B8 subtree byte count */ 1034 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 1035 hammer2_key_t unusedC8; /* 00C8 subtree inode count */ 1036 1037 /* 1038 * The last snapshot tid is tested against modify_tid to determine 1039 * when a copy must be made of a data block whos check mode has been 1040 * disabled (a disabled check mode allows data blocks to be updated 1041 * in place instead of copy-on-write). 1042 */ 1043 hammer2_tid_t pfs_lsnap_tid; /* 00D0 last snapshot tid */ 1044 hammer2_tid_t reservedD8; /* 00D8 (avail) */ 1045 1046 /* 1047 * Tracks (possibly degenerate) free areas covering all sub-tree 1048 * allocations under inode, not counting the inode itself. 1049 * 0/0 indicates empty entry. fully set-associative. 1050 * 1051 * (not yet implemented) 1052 */ 1053 uint64_t decrypt_check; /* 00E0 decryption validator */ 1054 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */ 1055} __packed; 1056 1057typedef struct hammer2_inode_meta hammer2_inode_meta_t; 1058 1059struct hammer2_inode_data { 1060 hammer2_inode_meta_t meta; /* 0000-00FF */ 1061 unsigned char filename[HAMMER2_INODE_MAXNAME]; 1062 /* 0100-01FF (256 char, unterminated) */ 1063 union { /* 0200-03FF (64x8 = 512 bytes) */ 1064 hammer2_blockset_t blockset; 1065 char data[HAMMER2_EMBEDDED_BYTES]; 1066 } u; 1067} __packed; 1068 1069typedef struct hammer2_inode_data hammer2_inode_data_t; 1070 1071#define HAMMER2_OPFLAG_DIRECTDATA 0x01 1072#define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 1073#define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 1074 1075#define HAMMER2_OBJTYPE_UNKNOWN 0 1076#define HAMMER2_OBJTYPE_DIRECTORY 1 1077#define HAMMER2_OBJTYPE_REGFILE 2 1078#define HAMMER2_OBJTYPE_FIFO 4 1079#define HAMMER2_OBJTYPE_CDEV 5 1080#define HAMMER2_OBJTYPE_BDEV 6 1081#define HAMMER2_OBJTYPE_SOFTLINK 7 1082#define HAMMER2_OBJTYPE_UNUSED08 8 1083#define HAMMER2_OBJTYPE_SOCKET 9 1084#define HAMMER2_OBJTYPE_WHITEOUT 10 1085 1086#define HAMMER2_COPYID_NONE 0 1087#define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 1088 1089#define HAMMER2_COPYID_COUNT 256 1090 1091/* 1092 * PFS types identify the role of a PFS within a cluster. The PFS types 1093 * is stored on media and in LNK_SPAN messages and used in other places. 1094 * 1095 * The low 4 bits specify the current active type while the high 4 bits 1096 * specify the transition target if the PFS is being upgraded or downgraded, 1097 * If the upper 4 bits are not zero it may effect how a PFS is used during 1098 * the transition. 1099 * 1100 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until 1101 * at least all MASTERs have updated their pfs_nmasters field. And upgrading 1102 * a SLAVE to a MASTER cannot complete until the new prospective master has 1103 * been fully synchronized (though theoretically full synchronization is 1104 * not required if a (new) quorum of other masters are fully synchronized). 1105 * 1106 * It generally does not matter which PFS element you actually mount, you 1107 * are mounting 'the cluster'. So, for example, a network mount will mount 1108 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions. 1109 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs 1110 * must be directly mounted. 1111 */ 1112#define HAMMER2_PFSTYPE_NONE 0x00 1113#define HAMMER2_PFSTYPE_CACHE 0x01 1114#define HAMMER2_PFSTYPE_UNUSED02 0x02 1115#define HAMMER2_PFSTYPE_SLAVE 0x03 1116#define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04 1117#define HAMMER2_PFSTYPE_SOFT_MASTER 0x05 1118#define HAMMER2_PFSTYPE_MASTER 0x06 1119#define HAMMER2_PFSTYPE_UNUSED07 0x07 1120#define HAMMER2_PFSTYPE_SUPROOT 0x08 1121#define HAMMER2_PFSTYPE_DUMMY 0x09 1122#define HAMMER2_PFSTYPE_MAX 16 1123 1124#define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */ 1125#define HAMMER2_PFSTRAN_CACHE 0x10 1126#define HAMMER2_PFSTRAN_UNMUSED20 0x20 1127#define HAMMER2_PFSTRAN_SLAVE 0x30 1128#define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40 1129#define HAMMER2_PFSTRAN_SOFT_MASTER 0x50 1130#define HAMMER2_PFSTRAN_MASTER 0x60 1131#define HAMMER2_PFSTRAN_UNUSED70 0x70 1132#define HAMMER2_PFSTRAN_SUPROOT 0x80 1133#define HAMMER2_PFSTRAN_DUMMY 0x90 1134 1135#define HAMMER2_PFS_DEC(n) ((n) & 0x0F) 1136#define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F) 1137#define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4) 1138 1139#define HAMMER2_PFSSUBTYPE_NONE 0 1140#define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */ 1141#define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */ 1142 1143/* 1144 * PFS mode of operation is a bitmask. This is typically not stored 1145 * on-media, but defined here because the field may be used in dmsgs. 1146 */ 1147#define HAMMER2_PFSMODE_QUORUM 0x01 1148#define HAMMER2_PFSMODE_RW 0x02 1149 1150/* 1151 * Allocation Table 1152 * 1153 */ 1154 1155 1156/* 1157 * Flags (8 bits) - blockref, for freemap only 1158 * 1159 * Note that the minimum chunk size is 1KB so we could theoretically have 1160 * 10 bits here, but we might have some future extension that allows a 1161 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 1162 */ 1163#define HAMMER2_AVF_SELMASK 0x03 /* select group */ 1164#define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 1165#define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 1166#define HAMMER2_AVF_RESERVED10 0x10 1167#define HAMMER2_AVF_RESERVED20 0x20 1168#define HAMMER2_AVF_RESERVED40 0x40 1169#define HAMMER2_AVF_RESERVED80 0x80 1170#define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 1171#define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 1172 1173#define HAMMER2_AV_SELECT_A 0x00 1174#define HAMMER2_AV_SELECT_B 0x01 1175#define HAMMER2_AV_SELECT_C 0x02 1176#define HAMMER2_AV_SELECT_D 0x03 1177 1178/* 1179 * The volume header eats a 64K block. There is currently an issue where 1180 * we want to try to fit all nominal filesystem updates in a 512-byte section 1181 * but it may be a lost cause due to the need for a blockset. 1182 * 1183 * All information is stored in host byte order. The volume header's magic 1184 * number may be checked to determine the byte order. If you wish to mount 1185 * between machines w/ different endian modes you'll need filesystem code 1186 * which acts on the media data consistently (either all one way or all the 1187 * other). Our code currently does not do that. 1188 * 1189 * A read-write mount may have to recover missing allocations by doing an 1190 * incremental mirror scan looking for modifications made after alloc_tid. 1191 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 1192 * operations are usually very, very fast. 1193 * 1194 * Read-only mounts do not need to do any recovery, access to the filesystem 1195 * topology is always consistent after a crash (is always consistent, period). 1196 * However, there may be shortcutted blockref updates present from deep in 1197 * the tree which are stored in the volumeh eader and must be tracked on 1198 * the fly. 1199 * 1200 * NOTE: The copyinfo[] array contains the configuration for both the 1201 * cluster connections and any local media copies. The volume 1202 * header will be replicated for each local media copy. 1203 * 1204 * The mount command may specify multiple medias or just one and 1205 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 1206 * array on mount. 1207 * 1208 * NOTE: root_blockref points to the super-root directory, not the root 1209 * directory. The root directory will be a subdirectory under the 1210 * super-root. 1211 * 1212 * The super-root directory contains all root directories and all 1213 * snapshots (readonly or writable). It is possible to do a 1214 * null-mount of the super-root using special path constructions 1215 * relative to your mounted root. 1216 * 1217 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 1218 * a PFS, including mirroring and storage quota operations, and this is 1219 * preferred over creating discrete PFSs in the super-root. Instead 1220 * the super-root is most typically used to create writable snapshots, 1221 * alternative roots, and so forth. The super-root is also used by 1222 * the automatic snapshotting mechanism. 1223 */ 1224#define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 1225#define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 1226 1227struct hammer2_volume_data { 1228 /* 1229 * sector #0 - 512 bytes 1230 */ 1231 uint64_t magic; /* 0000 Signature */ 1232 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 1233 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 1234 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 1235 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 1236 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 1237 1238 uint32_t version; /* 0030 */ 1239 uint32_t flags; /* 0034 */ 1240 uint8_t copyid; /* 0038 copyid of phys vol */ 1241 uint8_t freemap_version; /* 0039 freemap algorithm */ 1242 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 1243 uint8_t reserved003B; /* 003B */ 1244 uint32_t reserved003C; /* 003C */ 1245 1246 uuid_t fsid; /* 0040 */ 1247 uuid_t fstype; /* 0050 */ 1248 1249 /* 1250 * allocator_size is precalculated at newfs time and does not include 1251 * reserved blocks, boot, or redo areas. 1252 * 1253 * Initial non-reserved-area allocations do not use the freemap 1254 * but instead adjust alloc_iterator. Dynamic allocations take 1255 * over starting at (allocator_beg). This makes newfs_hammer2's 1256 * job a lot easier and can also serve as a testing jig. 1257 */ 1258 hammer2_off_t allocator_size; /* 0060 Total data space */ 1259 hammer2_off_t allocator_free; /* 0068 Free space */ 1260 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 1261 1262 /* 1263 * mirror_tid reflects the highest committed change for this 1264 * block device regardless of whether it is to the super-root 1265 * or to a PFS or whatever. 1266 * 1267 * freemap_tid reflects the highest committed freemap change for 1268 * this block device. 1269 */ 1270 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 1271 hammer2_tid_t reserved0080; /* 0080 */ 1272 hammer2_tid_t reserved0088; /* 0088 */ 1273 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 1274 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 1275 hammer2_tid_t reserved00A0[5]; /* 00A0-00C7 */ 1276 1277 /* 1278 * Copyids are allocated dynamically from the copyexists bitmap. 1279 * An id from the active copies set (up to 8, see copyinfo later on) 1280 * may still exist after the copy set has been removed from the 1281 * volume header and its bit will remain active in the bitmap and 1282 * cannot be reused until it is 100% removed from the hierarchy. 1283 */ 1284 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1285 char reserved0140[248]; /* 00E8-01DF */ 1286 1287 /* 1288 * 32 bit CRC array at the end of the first 512 byte sector. 1289 * 1290 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1291 * the other icrc's except this one). 1292 * 1293 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1294 * the blockset for the root. 1295 * 1296 * icrc_sects[5] - Sector 2 1297 * icrc_sects[4] - Sector 3 1298 * icrc_sects[3] - Sector 4 (the freemap blockset) 1299 */ 1300 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1301 1302 /* 1303 * sector #1 - 512 bytes 1304 * 1305 * The entire sector is used by a blockset. 1306 */ 1307 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1308 1309 /* 1310 * sector #2-7 1311 */ 1312 char sector2[512]; /* 0400-05FF reserved */ 1313 char sector3[512]; /* 0600-07FF reserved */ 1314 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1315 char sector5[512]; /* 0A00-0BFF reserved */ 1316 char sector6[512]; /* 0C00-0DFF reserved */ 1317 char sector7[512]; /* 0E00-0FFF reserved */ 1318 1319 /* 1320 * sector #8-71 - 32768 bytes 1321 * 1322 * Contains the configuration for up to 256 copyinfo targets. These 1323 * specify local and remote copies operating as masters or slaves. 1324 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1325 * indicates the local media). 1326 * 1327 * Each inode contains a set of up to 8 copyids, either inherited 1328 * from its parent or explicitly specified in the inode, which 1329 * indexes into this array. 1330 */ 1331 /* 1000-8FFF copyinfo config */ 1332 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1333 1334 /* 1335 * Remaining sections are reserved for future use. 1336 */ 1337 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1338 1339 /* 1340 * icrc on entire volume header 1341 */ 1342 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1343} __packed; 1344 1345typedef struct hammer2_volume_data hammer2_volume_data_t; 1346 1347/* 1348 * Various parts of the volume header have their own iCRCs. 1349 * 1350 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1351 * and not included the icrc calculation. 1352 * 1353 * The second 512 bytes also has its own iCRC but it is stored in the first 1354 * 512 bytes so it covers the entire second 512 bytes. 1355 * 1356 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1357 * which is where the iCRC for the whole volume is stored. This is currently 1358 * a catch-all for anything not individually iCRCd. 1359 */ 1360#define HAMMER2_VOL_ICRC_SECT0 7 1361#define HAMMER2_VOL_ICRC_SECT1 6 1362 1363#define HAMMER2_VOLUME_BYTES 65536 1364 1365#define HAMMER2_VOLUME_ICRC0_OFF 0 1366#define HAMMER2_VOLUME_ICRC1_OFF 512 1367#define HAMMER2_VOLUME_ICRCVH_OFF 0 1368 1369#define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1370#define HAMMER2_VOLUME_ICRC1_SIZE (512) 1371#define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1372 1373#define HAMMER2_VOL_VERSION_MIN 1 1374#define HAMMER2_VOL_VERSION_DEFAULT 1 1375#define HAMMER2_VOL_VERSION_WIP 2 1376 1377#define HAMMER2_NUM_VOLHDRS 4 1378 1379union hammer2_media_data { 1380 hammer2_volume_data_t voldata; 1381 hammer2_inode_data_t ipdata; 1382 hammer2_blockset_t blkset; 1383 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1384 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1385 char buf[HAMMER2_PBUFSIZE]; 1386} __packed; 1387 1388typedef union hammer2_media_data hammer2_media_data_t; 1389 1390#endif /* !_HAMMER2_DISK_H_ */ 1391