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