1 /* 2 * Copyright (c) 2011-2019 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 _VFS_HAMMER2_DISK_H_ 37 #define _VFS_HAMMER2_DISK_H_ 38 39 #ifndef _SYS_UUID_H_ 40 #include <sys/uuid.h> 41 #endif 42 #ifndef _SYS_DMSG_H_ 43 #include <sys/dmsg.h> 44 #endif 45 46 /* 47 * The structures below represent the on-disk media structures for the HAMMER2 48 * filesystem. Note that all fields for on-disk structures are naturally 49 * aligned. The host endian format is typically used - compatibility is 50 * possible if the implementation detects reversed endian and adjusts accesses 51 * accordingly. 52 * 53 * HAMMER2 primarily revolves around the directory topology: inodes, 54 * directory entries, and block tables. Block device buffer cache buffers 55 * are always 64KB. Logical file buffers are typically 16KB. All data 56 * references utilize 64-bit byte offsets. 57 * 58 * Free block management is handled independently using blocks reserved by 59 * the media topology. 60 */ 61 62 /* 63 * The data at the end of a file or directory may be a fragment in order 64 * to optimize storage efficiency. The minimum fragment size is 1KB. 65 * Since allocations are in powers of 2 fragments must also be sized in 66 * powers of 2 (1024, 2048, ... 65536). 67 * 68 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K), 69 * which is 2^16. Larger extents may be supported in the future. Smaller 70 * fragments might be supported in the future (down to 64 bytes is possible), 71 * but probably will not be. 72 * 73 * A full indirect block use supports 512 x 128-byte blockrefs in a 64KB 74 * buffer. Indirect blocks down to 1KB are supported to keep small 75 * directories small. 76 * 77 * A maximally sized file (2^64-1 bytes) requires ~6 indirect block levels 78 * using 64KB indirect blocks (128 byte refs, 512 or radix 9 per indblk). 79 * 80 * 16(datablk) + 9 + 9 + 9 + 9 + 9 + 9 = ~70. 81 * 16(datablk) + 7 + 9 + 9 + 9 + 9 + 9 = ~68. (smaller top level indblk) 82 * 83 * The actual depth depends on copies redundancy and whether the filesystem 84 * has chosen to use a smaller indirect block size at the top level or not. 85 */ 86 #define HAMMER2_ALLOC_MIN 1024 /* minimum allocation size */ 87 #define HAMMER2_RADIX_MIN 10 /* minimum allocation size 2^N */ 88 #define HAMMER2_ALLOC_MAX 65536 /* maximum allocation size */ 89 #define HAMMER2_RADIX_MAX 16 /* maximum allocation size 2^N */ 90 #define HAMMER2_RADIX_KEY 64 /* number of bits in key */ 91 92 /* 93 * MINALLOCSIZE - The minimum allocation size. This can be smaller 94 * or larger than the minimum physical IO size. 95 * 96 * NOTE: Should not be larger than 1K since inodes 97 * are 1K. 98 * 99 * MINIOSIZE - The minimum IO size. This must be less than 100 * or equal to HAMMER2_LBUFSIZE. 101 * 102 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups. 103 * 104 * HAMMER2_PBUFSIZE - Topological block size used by files for all 105 * blocks except the block straddling EOF. 106 * 107 * HAMMER2_SEGSIZE - Allocation map segment size, typically 4MB 108 * (space represented by a level0 bitmap). 109 */ 110 111 #define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX) 112 #define HAMMER2_SEGRADIX HAMMER2_FREEMAP_LEVEL0_RADIX 113 114 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 115 #define HAMMER2_PBUFSIZE 65536 116 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 117 #define HAMMER2_LBUFSIZE 16384 118 119 /* 120 * Generally speaking we want to use 16K and 64K I/Os 121 */ 122 #define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX 123 #define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE 124 125 #define HAMMER2_IND_BYTES_MIN 4096 126 #define HAMMER2_IND_BYTES_NOM HAMMER2_LBUFSIZE 127 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 128 #define HAMMER2_IND_RADIX_MIN 12 129 #define HAMMER2_IND_RADIX_NOM HAMMER2_LBUFRADIX 130 #define HAMMER2_IND_RADIX_MAX HAMMER2_PBUFRADIX 131 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 132 sizeof(hammer2_blockref_t)) 133 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 134 sizeof(hammer2_blockref_t)) 135 136 /* 137 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 138 * any element can occur at any index and holes can be anywhere. As a 139 * future optimization we will be able to flag that such arrays are sorted 140 * and thus optimize lookups, but for now we don't. 141 * 142 * Inodes embed either 512 bytes of direct data or an array of 4 blockrefs, 143 * resulting in highly efficient storage for files <= 512 bytes and for files 144 * <= 512KB. Up to 4 directory entries can be referenced from a directory 145 * without requiring an indirect block. 146 * 147 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented), 148 * or 64KB (1024 blockrefs / ~64MB represented). 149 */ 150 #define HAMMER2_SET_RADIX 2 /* radix 2 = 4 entries */ 151 #define HAMMER2_SET_COUNT (1 << HAMMER2_SET_RADIX) 152 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 153 #define HAMMER2_EMBEDDED_RADIX 9 154 155 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 156 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 157 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 158 159 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 160 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 161 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 162 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 163 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 164 165 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 166 167 /* 168 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment 169 * contains the volume header (or backup volume header), the free block 170 * table, and possibly other information in the future. A 4MB segment for 171 * freemap is reserved at the beginning of every 1GB. 172 * 173 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 174 * 175 * ========== 176 * 0 volume header (for the first four 2GB zones) 177 * 1 freemap00 level1 FREEMAP_LEAF (256 x 128B bitmap data per 1GB) 178 * 2 level2 FREEMAP_NODE (256 x 128B indirect block per 256GB) 179 * 3 level3 FREEMAP_NODE (256 x 128B indirect block per 64TB) 180 * 4 level4 FREEMAP_NODE (256 x 128B indirect block per 16PB) 181 * 5 level5 FREEMAP_NODE (256 x 128B indirect block per 4EB) 182 * 6 freemap01 level1 (rotation) 183 * 7 level2 184 * 8 level3 185 * 9 level4 186 * 10 level5 187 * 11 freemap02 level1 (rotation) 188 * 12 level2 189 * 13 level3 190 * 14 level4 191 * 15 level5 192 * 16 freemap03 level1 (rotation) 193 * 17 level2 194 * 18 level3 195 * 19 level4 196 * 20 level5 197 * 21 freemap04 level1 (rotation) 198 * 22 level2 199 * 23 level3 200 * 24 level4 201 * 25 level5 202 * 26 freemap05 level1 (rotation) 203 * 27 level2 204 * 28 level3 205 * 29 level4 206 * 30 level5 207 * 31 freemap06 level1 (rotation) 208 * 32 level2 209 * 33 level3 210 * 34 level4 211 * 35 level5 212 * 36 freemap07 level1 (rotation) 213 * 37 level2 214 * 38 level3 215 * 39 level4 216 * 40 level5 217 * 41 unused 218 * .. unused 219 * 63 unused 220 * ========== 221 * 222 * The first four 2GB zones contain volume headers and volume header backups. 223 * After that the volume header block# is reserved for future use. Similarly, 224 * there are many blocks related to various Freemap levels which are not 225 * used in every segment and those are also reserved for future use. 226 * Note that each FREEMAP_LEAF or FREEMAP_NODE uses 32KB out of 64KB slot. 227 * 228 * Freemap (see the FREEMAP document) 229 * 230 * The freemap utilizes blocks #1-40 in 8 sets of 5 blocks. Each block in 231 * a set represents a level of depth in the freemap topology. Eight sets 232 * exist to prevent live updates from disturbing the state of the freemap 233 * were a crash/reboot to occur. That is, a live update is not committed 234 * until the update's flush reaches the volume root. There are FOUR volume 235 * roots representing the last four synchronization points, so the freemap 236 * must be consistent no matter which volume root is chosen by the mount 237 * code. 238 * 239 * Each freemap set is 5 x 64K blocks and represents the 1GB, 256GB, 64TB, 240 * 16PB and 4EB indirect map. The volume header itself has a set of 4 freemap 241 * blockrefs representing another 2 bits, giving us a total 64 bits of 242 * representable address space. 243 * 244 * The Level 0 64KB block represents 1GB of storage represented by 32KB 245 * (256 x struct hammer2_bmap_data). Each structure represents 4MB of storage 246 * and has a 512 bit bitmap, using 2 bits to represent a 16KB chunk of 247 * storage. These 2 bits represent the following states: 248 * 249 * 00 Free 250 * 01 (reserved) (Possibly partially allocated) 251 * 10 Possibly free 252 * 11 Allocated 253 * 254 * One important thing to note here is that the freemap resolution is 16KB, 255 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps 256 * track of sub-allocations in memory, which means that on a unmount or reboot 257 * the entire 16KB of a partially allocated block will be considered fully 258 * allocated. It is possible for fragmentation to build up over time, but 259 * defragmentation is fairly easy to accomplish since all modifications 260 * allocate a new block. 261 * 262 * The Second thing to note is that due to the way snapshots and inode 263 * replication works, deleting a file cannot immediately free the related 264 * space. Furthermore, deletions often do not bother to traverse the 265 * block subhierarchy being deleted. And to go even further, whole 266 * sub-directory trees can be deleted simply by deleting the directory inode 267 * at the top. So even though we have a symbol to represent a 'possibly free' 268 * block (binary 10), only the bulk free scanning code can actually use it. 269 * Normal 'rm's or other deletions do not. 270 * 271 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 272 * (i.e. a multiple of 4MB). VOLUME_ALIGN must be >= ZONE_SEG. 273 * 274 * In Summary: 275 * 276 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block 277 * from the next set). The new copy is reused until a flush occurs at 278 * which point the next modification will then rotate to the next set. 279 */ 280 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 281 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 282 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 283 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 284 285 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 286 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 287 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 288 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 289 290 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 291 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 292 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 293 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 294 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 295 296 #define HAMMER2_ZONE_FREEMAP_INC 5 /* 5 deep */ 297 298 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 299 #define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */ 300 #define HAMMER2_ZONE_FREEMAP_01 6 /* normal freemap rotation */ 301 #define HAMMER2_ZONE_FREEMAP_02 11 /* normal freemap rotation */ 302 #define HAMMER2_ZONE_FREEMAP_03 16 /* normal freemap rotation */ 303 #define HAMMER2_ZONE_FREEMAP_04 21 /* normal freemap rotation */ 304 #define HAMMER2_ZONE_FREEMAP_05 26 /* normal freemap rotation */ 305 #define HAMMER2_ZONE_FREEMAP_06 31 /* normal freemap rotation */ 306 #define HAMMER2_ZONE_FREEMAP_07 36 /* normal freemap rotation */ 307 #define HAMMER2_ZONE_FREEMAP_END 41 /* (non-inclusive) */ 308 309 #define HAMMER2_ZONE_UNUSED41 41 310 #define HAMMER2_ZONE_UNUSED42 42 311 #define HAMMER2_ZONE_UNUSED43 43 312 #define HAMMER2_ZONE_UNUSED44 44 313 #define HAMMER2_ZONE_UNUSED45 45 314 #define HAMMER2_ZONE_UNUSED46 46 315 #define HAMMER2_ZONE_UNUSED47 47 316 #define HAMMER2_ZONE_UNUSED48 48 317 #define HAMMER2_ZONE_UNUSED49 49 318 #define HAMMER2_ZONE_UNUSED50 50 319 #define HAMMER2_ZONE_UNUSED51 51 320 #define HAMMER2_ZONE_UNUSED52 52 321 #define HAMMER2_ZONE_UNUSED53 53 322 #define HAMMER2_ZONE_UNUSED54 54 323 #define HAMMER2_ZONE_UNUSED55 55 324 #define HAMMER2_ZONE_UNUSED56 56 325 #define HAMMER2_ZONE_UNUSED57 57 326 #define HAMMER2_ZONE_UNUSED58 58 327 #define HAMMER2_ZONE_UNUSED59 59 328 #define HAMMER2_ZONE_UNUSED60 60 329 #define HAMMER2_ZONE_UNUSED61 61 330 #define HAMMER2_ZONE_UNUSED62 62 331 #define HAMMER2_ZONE_UNUSED63 63 332 #define HAMMER2_ZONE_END 64 /* non-inclusive */ 333 334 #define HAMMER2_NFREEMAPS 8 /* FREEMAP_00 - FREEMAP_07 */ 335 336 /* relative to FREEMAP_x */ 337 #define HAMMER2_ZONEFM_LEVEL1 0 /* 1GB leafmap */ 338 #define HAMMER2_ZONEFM_LEVEL2 1 /* 256GB indmap */ 339 #define HAMMER2_ZONEFM_LEVEL3 2 /* 64TB indmap */ 340 #define HAMMER2_ZONEFM_LEVEL4 3 /* 16PB indmap */ 341 #define HAMMER2_ZONEFM_LEVEL5 4 /* 4EB indmap */ 342 /* LEVEL6 is a set of 4 blockrefs in the volume header 16EB */ 343 344 /* 345 * Freemap radix. Assumes a set-count of 4, 128-byte blockrefs, 346 * 32KB indirect block for freemap (LEVELN_PSIZE below). 347 * 348 * Leaf entry represents 4MB of storage broken down into a 512-bit 349 * bitmap, 2-bits per entry. So course bitmap item represents 16KB. 350 */ 351 #if HAMMER2_SET_COUNT != 4 352 #error "hammer2_disk.h - freemap assumes SET_COUNT is 4" 353 #endif 354 #define HAMMER2_FREEMAP_LEVEL6_RADIX 64 /* 16EB (end) */ 355 #define HAMMER2_FREEMAP_LEVEL5_RADIX 62 /* 4EB */ 356 #define HAMMER2_FREEMAP_LEVEL4_RADIX 54 /* 16PB */ 357 #define HAMMER2_FREEMAP_LEVEL3_RADIX 46 /* 64TB */ 358 #define HAMMER2_FREEMAP_LEVEL2_RADIX 38 /* 256GB */ 359 #define HAMMER2_FREEMAP_LEVEL1_RADIX 30 /* 1GB */ 360 #define HAMMER2_FREEMAP_LEVEL0_RADIX 22 /* 4MB (128by in l-1 leaf) */ 361 362 #define HAMMER2_FREEMAP_LEVELN_PSIZE 32768 /* physical bytes */ 363 364 #define HAMMER2_FREEMAP_LEVEL5_SIZE ((hammer2_off_t)1 << \ 365 HAMMER2_FREEMAP_LEVEL5_RADIX) 366 #define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \ 367 HAMMER2_FREEMAP_LEVEL4_RADIX) 368 #define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \ 369 HAMMER2_FREEMAP_LEVEL3_RADIX) 370 #define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \ 371 HAMMER2_FREEMAP_LEVEL2_RADIX) 372 #define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \ 373 HAMMER2_FREEMAP_LEVEL1_RADIX) 374 #define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \ 375 HAMMER2_FREEMAP_LEVEL0_RADIX) 376 377 #define HAMMER2_FREEMAP_LEVEL5_MASK (HAMMER2_FREEMAP_LEVEL5_SIZE - 1) 378 #define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1) 379 #define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1) 380 #define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1) 381 #define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1) 382 #define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1) 383 384 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 385 sizeof(hammer2_bmap_data_t)) 386 387 /* 388 * XXX I made a mistake and made the reserved area begin at each LEVEL1 zone, 389 * which is on a 1GB demark. This will eat a little more space but for 390 * now we retain compatibility and make FMZONEBASE every 1GB 391 */ 392 #define H2FMZONEBASE(key) ((key) & ~HAMMER2_FREEMAP_LEVEL1_MASK) 393 #define H2FMBASE(key, radix) rounddown2(key, (hammer2_off_t)1 << (radix)) 394 395 /* 396 * 16KB bitmap granularity (x2 bits per entry). 397 */ 398 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 399 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 400 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 401 402 /* 403 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE. 404 * 405 * 8 x 64-bit elements, 2 bits per block. 406 * 32 blocks (radix 5) per element. 407 * representing INDEX_SIZE bytes worth of storage per element. 408 */ 409 410 typedef uint64_t hammer2_bitmap_t; 411 412 #define HAMMER2_BMAP_ALLONES ((hammer2_bitmap_t)-1) 413 #define HAMMER2_BMAP_ELEMENTS 8 414 #define HAMMER2_BMAP_BITS_PER_ELEMENT 64 415 #define HAMMER2_BMAP_INDEX_RADIX 5 /* 32 blocks per element */ 416 #define HAMMER2_BMAP_BLOCKS_PER_ELEMENT (1 << HAMMER2_BMAP_INDEX_RADIX) 417 418 #define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * \ 419 HAMMER2_BMAP_BLOCKS_PER_ELEMENT) 420 #define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1) 421 422 #define HAMMER2_BMAP_SIZE (HAMMER2_BMAP_INDEX_SIZE * \ 423 HAMMER2_BMAP_ELEMENTS) 424 #define HAMMER2_BMAP_MASK (HAMMER2_BMAP_SIZE - 1) 425 426 /* 427 * Two linear areas can be reserved after the initial 4MB segment in the base 428 * zone (the one starting at offset 0). These areas are NOT managed by the 429 * block allocator and do not fall under HAMMER2 crc checking rules based 430 * at the volume header (but can be self-CRCd internally, depending). 431 */ 432 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 433 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 434 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 435 436 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 437 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 438 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 439 440 /* 441 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 442 * Transaction ids are monotonic. 443 * 444 * We utilize 32-bit iSCSI CRCs. 445 */ 446 typedef uint64_t hammer2_tid_t; 447 typedef uint64_t hammer2_off_t; 448 typedef uint64_t hammer2_key_t; 449 typedef uint32_t hammer2_crc32_t; 450 451 /* 452 * Miscellanious ranges (all are unsigned). 453 */ 454 #define HAMMER2_TID_MIN 1ULL 455 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 456 #define HAMMER2_KEY_MIN 0ULL 457 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 458 #define HAMMER2_OFFSET_MIN 0ULL 459 #define HAMMER2_OFFSET_MAX 0xFFFFFFFFFFFFFFFFULL 460 461 /* 462 * HAMMER2 data offset special cases and masking. 463 * 464 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 465 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 466 * 467 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 468 * of the data offset field specifies how large the data chunk being pointed 469 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 470 * needed in the low bits of the data offset field). However, the practical 471 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 472 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but 473 * we fully intend to support larger extents in the future. 474 * 475 * WARNING! A radix of 0 (such as when data_off is all 0's) is a special 476 * case which means no data associated with the blockref, and 477 * not the '1 byte' it would otherwise calculate to. 478 */ 479 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 480 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 481 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 482 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 483 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 484 #define HAMMER2_MAX_COPIES 6 485 486 /* 487 * HAMMER2 directory support and pre-defined keys 488 */ 489 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 490 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 491 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 492 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 493 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 494 495 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 496 #define HAMMER2_BOOT_KEY 0xd9b36ce135528000ULL /* sroot to BOOT PFS */ 497 498 /************************************************************************ 499 * DMSG SUPPORT * 500 ************************************************************************ 501 * LNK_VOLCONF 502 * 503 * All HAMMER2 directories directly under the super-root on your local 504 * media can be mounted separately, even if they share the same physical 505 * device. 506 * 507 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 508 * cluster via local media. The local media does not have to participate 509 * in the cluster, other than to provide the hammer2_volconf[] array and 510 * root inode for the mount. 511 * 512 * This is important: The mount device path you specify serves to bootstrap 513 * your entry into the cluster, but your mount will make active connections 514 * to ALL copy elements in the hammer2_volconf[] array which match the 515 * PFSID of the directory in the super-root that you specified. The local 516 * media path does not have to be mentioned in this array but becomes part 517 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 518 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 519 * 520 * The actual cluster may be far larger than the elements you list in the 521 * hammer2_volconf[] array. You list only the elements you wish to 522 * directly connect to and you are able to access the rest of the cluster 523 * indirectly through those connections. 524 * 525 * WARNING! This structure must be exactly 128 bytes long for its config 526 * array to fit in the volume header. 527 */ 528 struct hammer2_volconf { 529 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 530 uint8_t inprog; /* 01 operation in progress, or 0 */ 531 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 532 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 533 uint16_t flags; /* 04-05 flags field */ 534 uint8_t error; /* 06 last operational error */ 535 uint8_t priority; /* 07 priority and round-robin flag */ 536 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 537 uint8_t reserved08[23]; /* 09-1F */ 538 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 539 uint8_t label[16]; /* 30-3F import/export label */ 540 uint8_t path[64]; /* 40-7F target specification string or key */ 541 } __packed; 542 543 typedef struct hammer2_volconf hammer2_volconf_t; 544 545 #define DMSG_VOLF_ENABLED 0x0001 546 #define DMSG_VOLF_INPROG 0x0002 547 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 548 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 549 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 550 551 struct dmsg_lnk_hammer2_volconf { 552 dmsg_hdr_t head; 553 hammer2_volconf_t copy; /* copy spec */ 554 int32_t index; 555 int32_t unused01; 556 uuid_t mediaid; 557 int64_t reserved02[32]; 558 } __packed; 559 560 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 561 562 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 563 dmsg_lnk_hammer2_volconf) 564 565 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 566 567 /* 568 * HAMMER2 directory entry header (embedded in blockref) exactly 16 bytes 569 */ 570 struct hammer2_dirent_head { 571 hammer2_tid_t inum; /* inode number */ 572 uint16_t namlen; /* name length */ 573 uint8_t type; /* OBJTYPE_* */ 574 uint8_t unused0B; 575 uint8_t unused0C[4]; 576 } __packed; 577 578 typedef struct hammer2_dirent_head hammer2_dirent_head_t; 579 580 /* 581 * The media block reference structure. This forms the core of the HAMMER2 582 * media topology recursion. This 128-byte data structure is embedded in the 583 * volume header, in inodes (which are also directory entries), and in 584 * indirect blocks. 585 * 586 * A blockref references a single media item, which typically can be a 587 * directory entry (aka inode), indirect block, or data block. 588 * 589 * The primary feature a blockref represents is the ability to validate 590 * the entire tree underneath it via its check code. Any modification to 591 * anything propagates up the blockref tree all the way to the root, replacing 592 * the related blocks and compounding the generated check code. 593 * 594 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, or as 595 * complex as a 512 bit cryptographic hash. I originally used a 64-byte 596 * blockref but later expanded it to 128 bytes to be able to support the 597 * larger check code as well as to embed statistics for quota operation. 598 * 599 * Simple check codes are not sufficient for unverified dedup. Even with 600 * a maximally-sized check code unverified dedup should only be used in 601 * in subdirectory trees where you do not need 100% data integrity. 602 * 603 * Unverified dedup is deduping based on meta-data only without verifying 604 * that the data blocks are actually identical. Verified dedup guarantees 605 * integrity but is a far more I/O-expensive operation. 606 * 607 * -- 608 * 609 * mirror_tid - per cluster node modified (propagated upward by flush) 610 * modify_tid - clc record modified (not propagated). 611 * update_tid - clc record updated (propagated upward on verification) 612 * 613 * CLC - Stands for 'Cluster Level Change', identifiers which are identical 614 * within the topology across all cluster nodes (when fully 615 * synchronized). 616 * 617 * NOTE: The range of keys represented by the blockref is (key) to 618 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 619 * blocks bottom-up, inserting a new root when radix expansion 620 * is required. 621 * 622 * leaf_count - Helps manage leaf collapse calculations when indirect 623 * blocks become mostly empty. This value caps out at 624 * HAMMER2_BLOCKREF_LEAF_MAX (65535). 625 * 626 * Used by the chain code to determine when to pull leafs up 627 * from nearly empty indirect blocks. For the purposes of this 628 * calculation, BREF_TYPE_INODE is considered a leaf, along 629 * with DIRENT and DATA. 630 * 631 * RESERVED FIELDS 632 * 633 * A number of blockref fields are reserved and should generally be set to 634 * 0 for future compatibility. 635 * 636 * FUTURE BLOCKREF EXPANSION 637 * 638 * CONTENT ADDRESSABLE INDEXING (future) - Using a 256 or 512-bit check code. 639 */ 640 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 641 uint8_t type; /* type of underlying item */ 642 uint8_t methods; /* check method & compression method */ 643 uint8_t copyid; /* specify which copy this is */ 644 uint8_t keybits; /* #of keybits masked off 0=leaf */ 645 uint8_t vradix; /* virtual data/meta-data size */ 646 uint8_t flags; /* blockref flags */ 647 uint16_t leaf_count; /* leaf aggregation count */ 648 hammer2_key_t key; /* key specification */ 649 hammer2_tid_t mirror_tid; /* media flush topology & freemap */ 650 hammer2_tid_t modify_tid; /* clc modify (not propagated) */ 651 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 652 hammer2_tid_t update_tid; /* clc modify (propagated upward) */ 653 union { 654 char buf[16]; 655 656 /* 657 * Directory entry header (BREF_TYPE_DIRENT) 658 * 659 * NOTE: check.buf contains filename if <= 64 bytes. Longer 660 * filenames are stored in a data reference of size 661 * HAMMER2_ALLOC_MIN (at least 256, typically 1024). 662 * 663 * NOTE: inode structure may contain a copy of a recently 664 * associated filename, for recovery purposes. 665 * 666 * NOTE: Superroot entries are INODEs, not DIRENTs. Code 667 * allows both cases. 668 */ 669 hammer2_dirent_head_t dirent; 670 671 /* 672 * Statistics aggregation (BREF_TYPE_INODE, BREF_TYPE_INDIRECT) 673 */ 674 struct { 675 hammer2_key_t data_count; 676 hammer2_key_t inode_count; 677 } stats; 678 } embed; 679 union { /* check info */ 680 char buf[64]; 681 struct { 682 uint32_t value; 683 uint32_t reserved[15]; 684 } iscsi32; 685 struct { 686 uint64_t value; 687 uint64_t reserved[7]; 688 } xxhash64; 689 struct { 690 char data[24]; 691 char reserved[40]; 692 } sha192; 693 struct { 694 char data[32]; 695 char reserved[32]; 696 } sha256; 697 struct { 698 char data[64]; 699 } sha512; 700 701 /* 702 * Freemap hints are embedded in addition to the icrc32. 703 * 704 * bigmask - Radixes available for allocation (0-31). 705 * Heuristical (may be permissive but not 706 * restrictive). Typically only radix values 707 * 10-16 are used (i.e. (1<<10) through (1<<16)). 708 * 709 * avail - Total available space remaining, in bytes 710 */ 711 struct { 712 uint32_t icrc32; 713 uint32_t bigmask; /* available radixes */ 714 uint64_t avail; /* total available bytes */ 715 char reserved[48]; 716 } freemap; 717 } check; 718 } __packed; 719 720 typedef struct hammer2_blockref hammer2_blockref_t; 721 722 #define HAMMER2_BLOCKREF_BYTES 128 /* blockref struct in bytes */ 723 #define HAMMER2_BLOCKREF_RADIX 7 724 725 #define HAMMER2_BLOCKREF_LEAF_MAX 65535 726 727 /* 728 * On-media and off-media blockref types. 729 * 730 * types >= 128 are pseudo values that should never be present on-media. 731 */ 732 #define HAMMER2_BREF_TYPE_EMPTY 0 733 #define HAMMER2_BREF_TYPE_INODE 1 734 #define HAMMER2_BREF_TYPE_INDIRECT 2 735 #define HAMMER2_BREF_TYPE_DATA 3 736 #define HAMMER2_BREF_TYPE_DIRENT 4 737 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 738 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 739 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 740 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 741 742 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 743 #define HAMMER2_BREF_FLAG_ZERO 0x02 /* NO LONGER USED */ 744 #define HAMMER2_BREF_FLAG_EMERG_MIP 0x04 /* emerg modified-in-place */ 745 746 /* 747 * Encode/decode check mode and compression mode for 748 * bref.methods. The compression level is not encoded in 749 * bref.methods. 750 */ 751 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 752 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 753 #define HAMMER2_ENC_COMP(n) ((n) & 15) 754 #define HAMMER2_DEC_COMP(n) ((n) & 15) 755 756 #define HAMMER2_CHECK_NONE 0 757 #define HAMMER2_CHECK_DISABLED 1 758 #define HAMMER2_CHECK_ISCSI32 2 759 #define HAMMER2_CHECK_XXHASH64 3 760 #define HAMMER2_CHECK_SHA192 4 761 #define HAMMER2_CHECK_FREEMAP 5 762 763 #define HAMMER2_CHECK_DEFAULT HAMMER2_CHECK_XXHASH64 764 765 /* user-specifiable check modes only */ 766 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 767 "xxhash64", "sha192" } 768 #define HAMMER2_CHECK_STRINGS_COUNT 5 769 770 /* 771 * Encode/decode check or compression algorithm request in 772 * ipdata->meta.check_algo and ipdata->meta.comp_algo. 773 */ 774 #define HAMMER2_ENC_ALGO(n) (n) 775 #define HAMMER2_DEC_ALGO(n) ((n) & 15) 776 #define HAMMER2_ENC_LEVEL(n) ((n) << 4) 777 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 778 779 #define HAMMER2_COMP_NONE 0 780 #define HAMMER2_COMP_AUTOZERO 1 781 #define HAMMER2_COMP_LZ4 2 782 #define HAMMER2_COMP_ZLIB 3 783 784 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 785 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 786 #define HAMMER2_COMP_STRINGS_COUNT 4 787 788 /* 789 * Passed to hammer2_chain_create(), causes methods to be inherited from 790 * parent. 791 */ 792 #define HAMMER2_METH_DEFAULT -1 793 794 /* 795 * HAMMER2 block references are collected into sets of 4 blockrefs. These 796 * sets are fully associative, meaning the elements making up a set are 797 * not sorted in any way and may contain duplicate entries, holes, or 798 * entries which shortcut multiple levels of indirection. Sets are used 799 * in various ways: 800 * 801 * (1) When redundancy is desired a set may contain several duplicate 802 * entries pointing to different copies of the same data. Up to 4 copies 803 * are supported. 804 * 805 * (2) The blockrefs in a set can shortcut multiple levels of indirections 806 * within the bounds imposed by the parent of set. 807 * 808 * When a set fills up another level of indirection is inserted, moving 809 * some or all of the set's contents into indirect blocks placed under the 810 * set. This is a top-down approach in that indirect blocks are not created 811 * until the set actually becomes full (that is, the entries in the set can 812 * shortcut the indirect blocks when the set is not full). Depending on how 813 * things are filled multiple indirect blocks will eventually be created. 814 * 815 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 816 * are also treated as fully set-associative. 817 */ 818 struct hammer2_blockset { 819 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 820 }; 821 822 typedef struct hammer2_blockset hammer2_blockset_t; 823 824 /* 825 * Catch programmer snafus 826 */ 827 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 828 #error "hammer2 direct radix is incorrect" 829 #endif 830 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 831 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 832 #endif 833 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 834 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 835 #endif 836 837 /* 838 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 839 * 840 * Each 128-byte entry contains the bitmap and meta-data required to manage 841 * a LEVEL0 (4MB) block of storage. The storage is managed in 256 x 16KB 842 * chunks. 843 * 844 * A smaller allocation granularity is supported via a linear iterator and/or 845 * must otherwise be tracked in ram. 846 * 847 * (data structure must be 128 bytes exactly) 848 * 849 * linear - A BYTE linear allocation offset used for sub-16KB allocations 850 * only. May contain values between 0 and 4MB. Must be ignored 851 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 852 * used to sub-allocate within the 16KB block (which is already 853 * marked as allocated in the bitmap). 854 * 855 * Sub-allocations need only be 1KB-aligned and do not have to be 856 * size-aligned, and 16KB or larger allocations do not update this 857 * field, resulting in pretty good packing. 858 * 859 * Please note that file data granularity may be limited by 860 * other issues such as buffer cache direct-mapping and the 861 * desire to support sector sizes up to 16KB (so H2 only issues 862 * I/O's in multiples of 16KB anyway). 863 * 864 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 865 * free. Used to cluster device buffers so all elements must have 866 * the same device block size, but may mix logical sizes. 867 * 868 * Typically integrated with the blockref type in the upper 8 bits 869 * to localize inodes and indrect blocks, improving bulk free scans 870 * and directory scans. 871 * 872 * bitmap - Two bits per 16KB allocation block arranged in arrays of 873 * 64-bit elements, 256x2 bits representing ~4MB worth of media 874 * storage. Bit patterns are as follows: 875 * 876 * 00 Unallocated 877 * 01 (reserved) 878 * 10 Possibly free 879 * 11 Allocated 880 */ 881 struct hammer2_bmap_data { 882 int32_t linear; /* 00 linear sub-granular allocation offset */ 883 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 884 uint8_t reserved06; /* 06 */ 885 uint8_t reserved07; /* 07 */ 886 uint32_t reserved08; /* 08 */ 887 uint32_t reserved0C; /* 0C */ 888 uint32_t reserved10; /* 10 */ 889 uint32_t reserved14; /* 14 */ 890 uint32_t reserved18; /* 18 */ 891 uint32_t avail; /* 1C */ 892 uint32_t reserved20[8]; /* 20-3F 256 bits manages 128K/1KB/2-bits */ 893 /* 40-7F 512 bits manages 4MB of storage */ 894 hammer2_bitmap_t bitmapq[HAMMER2_BMAP_ELEMENTS]; 895 } __packed; 896 897 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 898 899 /* 900 * XXX "Inodes ARE directory entries" is no longer the case. Hardlinks are 901 * dirents which refer to the same inode#, which is how filesystems usually 902 * implement hardlink. The following comments need to be updated. 903 * 904 * In HAMMER2 inodes ARE directory entries, with a special exception for 905 * hardlinks. The inode number is stored in the inode rather than being 906 * based on the location of the inode (since the location moves every time 907 * the inode or anything underneath the inode is modified). 908 * 909 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 910 * for the filename, and 512 bytes worth of direct file data OR an embedded 911 * blockset. The in-memory hammer2_inode structure contains only the mostly- 912 * node-independent meta-data portion (some flags are node-specific and will 913 * not be synchronized). The rest of the inode is node-specific and chain I/O 914 * is required to obtain it. 915 * 916 * Directories represent one inode per blockref. Inodes are not laid out 917 * as a file but instead are represented by the related blockrefs. The 918 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 919 * that blocksets are fully associative, so a certain degree efficiency is 920 * achieved just from that. 921 * 922 * Up to 512 bytes of direct data can be embedded in an inode, and since 923 * inodes are essentially directory entries this also means that small data 924 * files end up simply being laid out linearly in the directory, resulting 925 * in fewer seeks and highly optimal access. 926 * 927 * The compression mode can be changed at any time in the inode and is 928 * recorded on a blockref-by-blockref basis. 929 * 930 * Hardlinks are supported via the inode map. Essentially the way a hardlink 931 * works is that all individual directory entries representing the same file 932 * are special cased and specify the same inode number. The actual file 933 * is placed in the nearest parent directory that is parent to all instances 934 * of the hardlink. If all hardlinks to a file are in the same directory 935 * the actual file will also be placed in that directory. This file uses 936 * the inode number as the directory entry key and is invisible to normal 937 * directory scans. Real directory entry keys are differentiated from the 938 * inode number key via bit 63. Access to the hardlink silently looks up 939 * the real file and forwards all operations to that file. Removal of the 940 * last hardlink also removes the real file. 941 */ 942 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 943 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 944 #define HAMMER2_INODE_VERSION_ONE 1 945 946 #define HAMMER2_INODE_START 1024 /* dynamically allocated */ 947 948 struct hammer2_inode_meta { 949 uint16_t version; /* 0000 inode data version */ 950 uint8_t reserved02; /* 0002 */ 951 uint8_t pfs_subtype; /* 0003 pfs sub-type */ 952 953 /* 954 * core inode attributes, inode type, misc flags 955 */ 956 uint32_t uflags; /* 0004 chflags */ 957 uint32_t rmajor; /* 0008 available for device nodes */ 958 uint32_t rminor; /* 000C available for device nodes */ 959 uint64_t ctime; /* 0010 inode change time */ 960 uint64_t mtime; /* 0018 modified time */ 961 uint64_t atime; /* 0020 access time (unsupported) */ 962 uint64_t btime; /* 0028 birth time */ 963 uuid_t uid; /* 0030 uid / degenerate unix uid */ 964 uuid_t gid; /* 0040 gid / degenerate unix gid */ 965 966 uint8_t type; /* 0050 object type */ 967 uint8_t op_flags; /* 0051 operational flags */ 968 uint16_t cap_flags; /* 0052 capability flags */ 969 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 970 971 /* 972 * inode size, identification, localized recursive configuration 973 * for compression and backup copies. 974 * 975 * NOTE: Nominal parent inode number (iparent) is only applicable 976 * for directories but can also help for files during 977 * catastrophic recovery. 978 */ 979 hammer2_tid_t inum; /* 0058 inode number */ 980 hammer2_off_t size; /* 0060 size of file */ 981 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 982 hammer2_tid_t iparent; /* 0070 nominal parent inum */ 983 hammer2_key_t name_key; /* 0078 full filename key */ 984 uint16_t name_len; /* 0080 filename length */ 985 uint8_t ncopies; /* 0082 ncopies to local media */ 986 uint8_t comp_algo; /* 0083 compression request & algo */ 987 988 /* 989 * These fields are currently only applicable to PFSROOTs. 990 * 991 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 992 * identify an instance of a PFS in the cluster because 993 * a mount may contain more than one copy of the PFS as 994 * a separate node. {pfs_clid, pfs_fsid} must be used for 995 * registration in the cluster. 996 */ 997 uint8_t target_type; /* 0084 hardlink target type */ 998 uint8_t check_algo; /* 0085 check code request & algo */ 999 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 1000 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 1001 hammer2_tid_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 1002 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 1003 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 1004 1005 /* 1006 * Quotas and aggregate sub-tree inode and data counters. Note that 1007 * quotas are not replicated downward, they are explicitly set by 1008 * the sysop and in-memory structures keep track of inheritance. 1009 */ 1010 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 1011 hammer2_key_t unusedB8; /* 00B8 subtree byte count */ 1012 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 1013 hammer2_key_t unusedC8; /* 00C8 subtree inode count */ 1014 1015 /* 1016 * The last snapshot tid is tested against modify_tid to determine 1017 * when a copy must be made of a data block whos check mode has been 1018 * disabled (a disabled check mode allows data blocks to be updated 1019 * in place instead of copy-on-write). 1020 */ 1021 hammer2_tid_t pfs_lsnap_tid; /* 00D0 last snapshot tid */ 1022 hammer2_tid_t reservedD8; /* 00D8 (avail) */ 1023 1024 /* 1025 * Tracks (possibly degenerate) free areas covering all sub-tree 1026 * allocations under inode, not counting the inode itself. 1027 * 0/0 indicates empty entry. fully set-associative. 1028 * 1029 * (not yet implemented) 1030 */ 1031 uint64_t decrypt_check; /* 00E0 decryption validator */ 1032 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */ 1033 } __packed; 1034 1035 typedef struct hammer2_inode_meta hammer2_inode_meta_t; 1036 1037 struct hammer2_inode_data { 1038 hammer2_inode_meta_t meta; /* 0000-00FF */ 1039 unsigned char filename[HAMMER2_INODE_MAXNAME]; 1040 /* 0100-01FF (256 char, unterminated) */ 1041 union { /* 0200-03FF (64x8 = 512 bytes) */ 1042 hammer2_blockset_t blockset; 1043 char data[HAMMER2_EMBEDDED_BYTES]; 1044 } u; 1045 } __packed; 1046 1047 typedef struct hammer2_inode_data hammer2_inode_data_t; 1048 1049 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 1050 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 1051 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 1052 1053 #define HAMMER2_OBJTYPE_UNKNOWN 0 1054 #define HAMMER2_OBJTYPE_DIRECTORY 1 1055 #define HAMMER2_OBJTYPE_REGFILE 2 1056 #define HAMMER2_OBJTYPE_FIFO 4 1057 #define HAMMER2_OBJTYPE_CDEV 5 1058 #define HAMMER2_OBJTYPE_BDEV 6 1059 #define HAMMER2_OBJTYPE_SOFTLINK 7 1060 #define HAMMER2_OBJTYPE_UNUSED08 8 1061 #define HAMMER2_OBJTYPE_SOCKET 9 1062 #define HAMMER2_OBJTYPE_WHITEOUT 10 1063 1064 #define HAMMER2_COPYID_NONE 0 1065 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 1066 1067 #define HAMMER2_COPYID_COUNT 256 1068 1069 /* 1070 * PFS types identify the role of a PFS within a cluster. The PFS types 1071 * is stored on media and in LNK_SPAN messages and used in other places. 1072 * 1073 * The low 4 bits specify the current active type while the high 4 bits 1074 * specify the transition target if the PFS is being upgraded or downgraded, 1075 * If the upper 4 bits are not zero it may effect how a PFS is used during 1076 * the transition. 1077 * 1078 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until 1079 * at least all MASTERs have updated their pfs_nmasters field. And upgrading 1080 * a SLAVE to a MASTER cannot complete until the new prospective master has 1081 * been fully synchronized (though theoretically full synchronization is 1082 * not required if a (new) quorum of other masters are fully synchronized). 1083 * 1084 * It generally does not matter which PFS element you actually mount, you 1085 * are mounting 'the cluster'. So, for example, a network mount will mount 1086 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions. 1087 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs 1088 * must be directly mounted. 1089 */ 1090 #define HAMMER2_PFSTYPE_NONE 0x00 1091 #define HAMMER2_PFSTYPE_CACHE 0x01 1092 #define HAMMER2_PFSTYPE_UNUSED02 0x02 1093 #define HAMMER2_PFSTYPE_SLAVE 0x03 1094 #define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04 1095 #define HAMMER2_PFSTYPE_SOFT_MASTER 0x05 1096 #define HAMMER2_PFSTYPE_MASTER 0x06 1097 #define HAMMER2_PFSTYPE_UNUSED07 0x07 1098 #define HAMMER2_PFSTYPE_SUPROOT 0x08 1099 #define HAMMER2_PFSTYPE_DUMMY 0x09 1100 #define HAMMER2_PFSTYPE_MAX 16 1101 1102 #define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */ 1103 #define HAMMER2_PFSTRAN_CACHE 0x10 1104 #define HAMMER2_PFSTRAN_UNMUSED20 0x20 1105 #define HAMMER2_PFSTRAN_SLAVE 0x30 1106 #define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40 1107 #define HAMMER2_PFSTRAN_SOFT_MASTER 0x50 1108 #define HAMMER2_PFSTRAN_MASTER 0x60 1109 #define HAMMER2_PFSTRAN_UNUSED70 0x70 1110 #define HAMMER2_PFSTRAN_SUPROOT 0x80 1111 #define HAMMER2_PFSTRAN_DUMMY 0x90 1112 1113 #define HAMMER2_PFS_DEC(n) ((n) & 0x0F) 1114 #define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F) 1115 #define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4) 1116 1117 #define HAMMER2_PFSSUBTYPE_NONE 0 1118 #define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */ 1119 #define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */ 1120 1121 /* 1122 * PFS mode of operation is a bitmask. This is typically not stored 1123 * on-media, but defined here because the field may be used in dmsgs. 1124 */ 1125 #define HAMMER2_PFSMODE_QUORUM 0x01 1126 #define HAMMER2_PFSMODE_RW 0x02 1127 1128 /* 1129 * Allocation Table 1130 * 1131 */ 1132 1133 1134 /* 1135 * Flags (8 bits) - blockref, for freemap only 1136 * 1137 * Note that the minimum chunk size is 1KB so we could theoretically have 1138 * 10 bits here, but we might have some future extension that allows a 1139 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 1140 */ 1141 #define HAMMER2_AVF_SELMASK 0x03 /* select group */ 1142 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 1143 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 1144 #define HAMMER2_AVF_RESERVED10 0x10 1145 #define HAMMER2_AVF_RESERVED20 0x20 1146 #define HAMMER2_AVF_RESERVED40 0x40 1147 #define HAMMER2_AVF_RESERVED80 0x80 1148 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 1149 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 1150 1151 #define HAMMER2_AV_SELECT_A 0x00 1152 #define HAMMER2_AV_SELECT_B 0x01 1153 #define HAMMER2_AV_SELECT_C 0x02 1154 #define HAMMER2_AV_SELECT_D 0x03 1155 1156 /* 1157 * The volume header eats a 64K block. There is currently an issue where 1158 * we want to try to fit all nominal filesystem updates in a 512-byte section 1159 * but it may be a lost cause due to the need for a blockset. 1160 * 1161 * All information is stored in host byte order. The volume header's magic 1162 * number may be checked to determine the byte order. If you wish to mount 1163 * between machines w/ different endian modes you'll need filesystem code 1164 * which acts on the media data consistently (either all one way or all the 1165 * other). Our code currently does not do that. 1166 * 1167 * A read-write mount may have to recover missing allocations by doing an 1168 * incremental mirror scan looking for modifications made after alloc_tid. 1169 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 1170 * operations are usually very, very fast. 1171 * 1172 * Read-only mounts do not need to do any recovery, access to the filesystem 1173 * topology is always consistent after a crash (is always consistent, period). 1174 * However, there may be shortcutted blockref updates present from deep in 1175 * the tree which are stored in the volumeh eader and must be tracked on 1176 * the fly. 1177 * 1178 * NOTE: The copyinfo[] array contains the configuration for both the 1179 * cluster connections and any local media copies. The volume 1180 * header will be replicated for each local media copy. 1181 * 1182 * The mount command may specify multiple medias or just one and 1183 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 1184 * array on mount. 1185 * 1186 * NOTE: root_blockref points to the super-root directory, not the root 1187 * directory. The root directory will be a subdirectory under the 1188 * super-root. 1189 * 1190 * The super-root directory contains all root directories and all 1191 * snapshots (readonly or writable). It is possible to do a 1192 * null-mount of the super-root using special path constructions 1193 * relative to your mounted root. 1194 * 1195 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 1196 * a PFS, including mirroring and storage quota operations, and this is 1197 * prefered over creating discrete PFSs in the super-root. Instead 1198 * the super-root is most typically used to create writable snapshots, 1199 * alternative roots, and so forth. The super-root is also used by 1200 * the automatic snapshotting mechanism. 1201 */ 1202 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 1203 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 1204 1205 struct hammer2_volume_data { 1206 /* 1207 * sector #0 - 512 bytes 1208 */ 1209 uint64_t magic; /* 0000 Signature */ 1210 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 1211 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 1212 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 1213 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 1214 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 1215 1216 uint32_t version; /* 0030 */ 1217 uint32_t flags; /* 0034 */ 1218 uint8_t copyid; /* 0038 copyid of phys vol */ 1219 uint8_t freemap_version; /* 0039 freemap algorithm */ 1220 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 1221 uint8_t reserved003B; /* 003B */ 1222 uint32_t reserved003C; /* 003C */ 1223 1224 uuid_t fsid; /* 0040 */ 1225 uuid_t fstype; /* 0050 */ 1226 1227 /* 1228 * allocator_size is precalculated at newfs time and does not include 1229 * reserved blocks, boot, or redo areas. 1230 * 1231 * Initial non-reserved-area allocations do not use the freemap 1232 * but instead adjust alloc_iterator. Dynamic allocations take 1233 * over starting at (allocator_beg). This makes newfs_hammer2's 1234 * job a lot easier and can also serve as a testing jig. 1235 */ 1236 hammer2_off_t allocator_size; /* 0060 Total data space */ 1237 hammer2_off_t allocator_free; /* 0068 Free space */ 1238 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 1239 1240 /* 1241 * mirror_tid reflects the highest committed change for this 1242 * block device regardless of whether it is to the super-root 1243 * or to a PFS or whatever. 1244 * 1245 * freemap_tid reflects the highest committed freemap change for 1246 * this block device. 1247 */ 1248 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 1249 hammer2_tid_t reserved0080; /* 0080 */ 1250 hammer2_tid_t reserved0088; /* 0088 */ 1251 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 1252 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 1253 hammer2_tid_t reserved00A0[5]; /* 00A0-00C7 */ 1254 1255 /* 1256 * Copyids are allocated dynamically from the copyexists bitmap. 1257 * An id from the active copies set (up to 8, see copyinfo later on) 1258 * may still exist after the copy set has been removed from the 1259 * volume header and its bit will remain active in the bitmap and 1260 * cannot be reused until it is 100% removed from the hierarchy. 1261 */ 1262 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1263 char reserved0140[248]; /* 00E8-01DF */ 1264 1265 /* 1266 * 32 bit CRC array at the end of the first 512 byte sector. 1267 * 1268 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1269 * the other icrc's except this one). 1270 * 1271 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1272 * the blockset for the root. 1273 * 1274 * icrc_sects[5] - Sector 2 1275 * icrc_sects[4] - Sector 3 1276 * icrc_sects[3] - Sector 4 (the freemap blockset) 1277 */ 1278 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1279 1280 /* 1281 * sector #1 - 512 bytes 1282 * 1283 * The entire sector is used by a blockset. 1284 */ 1285 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1286 1287 /* 1288 * sector #2-7 1289 */ 1290 char sector2[512]; /* 0400-05FF reserved */ 1291 char sector3[512]; /* 0600-07FF reserved */ 1292 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1293 char sector5[512]; /* 0A00-0BFF reserved */ 1294 char sector6[512]; /* 0C00-0DFF reserved */ 1295 char sector7[512]; /* 0E00-0FFF reserved */ 1296 1297 /* 1298 * sector #8-71 - 32768 bytes 1299 * 1300 * Contains the configuration for up to 256 copyinfo targets. These 1301 * specify local and remote copies operating as masters or slaves. 1302 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1303 * indicates the local media). 1304 * 1305 * Each inode contains a set of up to 8 copyids, either inherited 1306 * from its parent or explicitly specified in the inode, which 1307 * indexes into this array. 1308 */ 1309 /* 1000-8FFF copyinfo config */ 1310 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1311 1312 /* 1313 * Remaining sections are reserved for future use. 1314 */ 1315 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1316 1317 /* 1318 * icrc on entire volume header 1319 */ 1320 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1321 } __packed; 1322 1323 typedef struct hammer2_volume_data hammer2_volume_data_t; 1324 1325 /* 1326 * Various parts of the volume header have their own iCRCs. 1327 * 1328 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1329 * and not included the icrc calculation. 1330 * 1331 * The second 512 bytes also has its own iCRC but it is stored in the first 1332 * 512 bytes so it covers the entire second 512 bytes. 1333 * 1334 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1335 * which is where the iCRC for the whole volume is stored. This is currently 1336 * a catch-all for anything not individually iCRCd. 1337 */ 1338 #define HAMMER2_VOL_ICRC_SECT0 7 1339 #define HAMMER2_VOL_ICRC_SECT1 6 1340 1341 #define HAMMER2_VOLUME_BYTES 65536 1342 1343 #define HAMMER2_VOLUME_ICRC0_OFF 0 1344 #define HAMMER2_VOLUME_ICRC1_OFF 512 1345 #define HAMMER2_VOLUME_ICRCVH_OFF 0 1346 1347 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1348 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 1349 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1350 1351 #define HAMMER2_VOL_VERSION_MIN 1 1352 #define HAMMER2_VOL_VERSION_DEFAULT 1 1353 #define HAMMER2_VOL_VERSION_WIP 2 1354 1355 #define HAMMER2_NUM_VOLHDRS 4 1356 1357 union hammer2_media_data { 1358 hammer2_volume_data_t voldata; 1359 hammer2_inode_data_t ipdata; 1360 hammer2_blockset_t blkset; 1361 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1362 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1363 char buf[HAMMER2_PBUFSIZE]; 1364 } __packed; 1365 1366 typedef union hammer2_media_data hammer2_media_data_t; 1367 1368 #endif /* !_VFS_HAMMER2_DISK_H_ */ 1369