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 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups. 94 * 95 * HAMMER2_PBUFSIZE - Topological block size used by files for all 96 * blocks except the block straddling EOF. 97 * 98 * HAMMER2_SEGSIZE - Allocation map segment size, typically 4MB 99 * (space represented by a level0 bitmap). 100 */ 101 102 #define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX) 103 #define HAMMER2_SEGRADIX HAMMER2_FREEMAP_LEVEL0_RADIX 104 105 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 106 #define HAMMER2_PBUFSIZE 65536 107 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 108 #define HAMMER2_LBUFSIZE 16384 109 110 #define HAMMER2_IND_BYTES_MIN 4096 111 #define HAMMER2_IND_BYTES_NOM HAMMER2_LBUFSIZE 112 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 113 #define HAMMER2_IND_RADIX_MIN 12 114 #define HAMMER2_IND_RADIX_NOM HAMMER2_LBUFRADIX 115 #define HAMMER2_IND_RADIX_MAX HAMMER2_PBUFRADIX 116 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 117 sizeof(hammer2_blockref_t)) 118 #define HAMMER2_IND_COUNT_NOM (HAMMER2_IND_BYTES_NOM / \ 119 sizeof(hammer2_blockref_t)) 120 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 121 sizeof(hammer2_blockref_t)) 122 123 /* 124 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 125 * any element can occur at any index and holes can be anywhere. 126 * 127 * Inodes embed either 512 bytes of direct data or an array of 4 blockrefs, 128 * resulting in highly efficient storage for files <= 512 bytes and for files 129 * <= 512KB. Up to 4 directory entries can be referenced from a directory 130 * without requiring an indirect block. 131 */ 132 #define HAMMER2_SET_RADIX 2 /* radix 2 = 4 entries */ 133 #define HAMMER2_SET_COUNT (1 << HAMMER2_SET_RADIX) 134 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 135 #define HAMMER2_EMBEDDED_RADIX 9 136 137 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 138 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 139 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 140 141 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 142 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 143 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 144 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 145 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 146 147 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 148 149 /* 150 * A 4MB segment is reserved at the beginning of each 1GB. This segment 151 * contains the volume header (or backup volume header), the free block 152 * table, and possibly other information in the future. 153 * 154 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 155 * 156 * ========== 157 * 0 volume header (for the first four 2GB zones) 158 * 1 freemap00 level1 FREEMAP_LEAF (256 x 128B bitmap data per 1GB) 159 * 2 level2 FREEMAP_NODE (256 x 128B indirect block per 256GB) 160 * 3 level3 FREEMAP_NODE (256 x 128B indirect block per 64TB) 161 * 4 level4 FREEMAP_NODE (256 x 128B indirect block per 16PB) 162 * 5 level5 FREEMAP_NODE (256 x 128B indirect block per 4EB) 163 * 6 freemap01 level1 (rotation) 164 * 7 level2 165 * 8 level3 166 * 9 level4 167 * 10 level5 168 * 11 freemap02 level1 (rotation) 169 * 12 level2 170 * 13 level3 171 * 14 level4 172 * 15 level5 173 * 16 freemap03 level1 (rotation) 174 * 17 level2 175 * 18 level3 176 * 19 level4 177 * 20 level5 178 * 21 freemap04 level1 (rotation) 179 * 22 level2 180 * 23 level3 181 * 24 level4 182 * 25 level5 183 * 26 freemap05 level1 (rotation) 184 * 27 level2 185 * 28 level3 186 * 29 level4 187 * 30 level5 188 * 31 freemap06 level1 (rotation) 189 * 32 level2 190 * 33 level3 191 * 34 level4 192 * 35 level5 193 * 36 freemap07 level1 (rotation) 194 * 37 level2 195 * 38 level3 196 * 39 level4 197 * 40 level5 198 * 41 unused 199 * .. unused 200 * 63 unused 201 * ========== 202 * 203 * The first four 2GB zones contain volume headers and volume header backups. 204 * After that the volume header block# is reserved for future use. Similarly, 205 * there are many blocks related to various Freemap levels which are not 206 * used in every segment and those are also reserved for future use. 207 * Note that each FREEMAP_LEAF or FREEMAP_NODE uses 32KB out of 64KB slot. 208 * 209 * Freemap (see the FREEMAP document) 210 * 211 * The freemap utilizes blocks #1-40 in 8 sets of 5 blocks. Each block in 212 * a set represents a level of depth in the freemap topology. Eight sets 213 * exist to prevent live updates from disturbing the state of the freemap 214 * were a crash/reboot to occur. That is, a live update is not committed 215 * until the update's flush reaches the volume root. There are FOUR volume 216 * roots representing the last four synchronization points, so the freemap 217 * must be consistent no matter which volume root is chosen by the mount 218 * code. 219 * 220 * Each freemap set is 5 x 64K blocks and represents the 1GB, 256GB, 64TB, 221 * 16PB and 4EB indirect map. The volume header itself has a set of 4 freemap 222 * blockrefs representing another 2 bits, giving us a total 64 bits of 223 * representable address space. 224 * 225 * The Level 0 64KB block represents 1GB of storage represented by 32KB 226 * (256 x struct hammer2_bmap_data). Each structure represents 4MB of storage 227 * and has a 512 bit bitmap, using 2 bits to represent a 16KB chunk of 228 * storage. These 2 bits represent the following states: 229 * 230 * 00 Free 231 * 01 (reserved) (Possibly partially allocated) 232 * 10 Possibly free 233 * 11 Allocated 234 * 235 * One important thing to note here is that the freemap resolution is 16KB, 236 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps 237 * track of sub-allocations in memory, which means that on a unmount or reboot 238 * the entire 16KB of a partially allocated block will be considered fully 239 * allocated. It is possible for fragmentation to build up over time, but 240 * defragmentation is fairly easy to accomplish since all modifications 241 * allocate a new block. 242 * 243 * The Second thing to note is that due to the way snapshots and inode 244 * replication works, deleting a file cannot immediately free the related 245 * space. Furthermore, deletions often do not bother to traverse the 246 * block subhierarchy being deleted. And to go even further, whole 247 * sub-directory trees can be deleted simply by deleting the directory inode 248 * at the top. So even though we have a symbol to represent a 'possibly free' 249 * block (binary 10), only the bulk free scanning code can actually use it. 250 * Normal 'rm's or other deletions do not. 251 * 252 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 253 * (i.e. a multiple of 4MB). VOLUME_ALIGN must be >= ZONE_SEG. 254 * 255 * In Summary: 256 * 257 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block 258 * from the next set). The new copy is reused until a flush occurs at 259 * which point the next modification will then rotate to the next set. 260 */ 261 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 262 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 263 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 264 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 265 266 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 267 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 268 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 269 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 270 271 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 272 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 273 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 274 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 275 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 276 277 #define HAMMER2_ZONE_FREEMAP_INC 5 /* 5 deep */ 278 279 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 280 #define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */ 281 #define HAMMER2_ZONE_FREEMAP_01 6 /* normal freemap rotation */ 282 #define HAMMER2_ZONE_FREEMAP_02 11 /* normal freemap rotation */ 283 #define HAMMER2_ZONE_FREEMAP_03 16 /* normal freemap rotation */ 284 #define HAMMER2_ZONE_FREEMAP_04 21 /* normal freemap rotation */ 285 #define HAMMER2_ZONE_FREEMAP_05 26 /* normal freemap rotation */ 286 #define HAMMER2_ZONE_FREEMAP_06 31 /* normal freemap rotation */ 287 #define HAMMER2_ZONE_FREEMAP_07 36 /* normal freemap rotation */ 288 #define HAMMER2_ZONE_FREEMAP_END 41 /* (non-inclusive) */ 289 290 #define HAMMER2_ZONE_UNUSED41 41 291 #define HAMMER2_ZONE_UNUSED42 42 292 #define HAMMER2_ZONE_UNUSED43 43 293 #define HAMMER2_ZONE_UNUSED44 44 294 #define HAMMER2_ZONE_UNUSED45 45 295 #define HAMMER2_ZONE_UNUSED46 46 296 #define HAMMER2_ZONE_UNUSED47 47 297 #define HAMMER2_ZONE_UNUSED48 48 298 #define HAMMER2_ZONE_UNUSED49 49 299 #define HAMMER2_ZONE_UNUSED50 50 300 #define HAMMER2_ZONE_UNUSED51 51 301 #define HAMMER2_ZONE_UNUSED52 52 302 #define HAMMER2_ZONE_UNUSED53 53 303 #define HAMMER2_ZONE_UNUSED54 54 304 #define HAMMER2_ZONE_UNUSED55 55 305 #define HAMMER2_ZONE_UNUSED56 56 306 #define HAMMER2_ZONE_UNUSED57 57 307 #define HAMMER2_ZONE_UNUSED58 58 308 #define HAMMER2_ZONE_UNUSED59 59 309 #define HAMMER2_ZONE_UNUSED60 60 310 #define HAMMER2_ZONE_UNUSED61 61 311 #define HAMMER2_ZONE_UNUSED62 62 312 #define HAMMER2_ZONE_UNUSED63 63 313 #define HAMMER2_ZONE_END 64 /* non-inclusive */ 314 315 #define HAMMER2_NFREEMAPS 8 /* FREEMAP_00 - FREEMAP_07 */ 316 317 /* relative to FREEMAP_x */ 318 #define HAMMER2_ZONEFM_LEVEL1 0 /* 1GB leafmap */ 319 #define HAMMER2_ZONEFM_LEVEL2 1 /* 256GB indmap */ 320 #define HAMMER2_ZONEFM_LEVEL3 2 /* 64TB indmap */ 321 #define HAMMER2_ZONEFM_LEVEL4 3 /* 16PB indmap */ 322 #define HAMMER2_ZONEFM_LEVEL5 4 /* 4EB indmap */ 323 /* LEVEL6 is a set of 4 blockrefs in the volume header 16EB */ 324 325 /* 326 * Freemap radix. Assumes a set-count of 4, 128-byte blockrefs, 327 * 32KB indirect block for freemap (LEVELN_PSIZE below). 328 * 329 * Leaf entry represents 4MB of storage broken down into a 512-bit 330 * bitmap, 2-bits per entry. So course bitmap item represents 16KB. 331 */ 332 #if HAMMER2_SET_COUNT != 4 333 #error "hammer2_disk.h - freemap assumes SET_COUNT is 4" 334 #endif 335 #define HAMMER2_FREEMAP_LEVEL6_RADIX 64 /* 16EB (end) */ 336 #define HAMMER2_FREEMAP_LEVEL5_RADIX 62 /* 4EB */ 337 #define HAMMER2_FREEMAP_LEVEL4_RADIX 54 /* 16PB */ 338 #define HAMMER2_FREEMAP_LEVEL3_RADIX 46 /* 64TB */ 339 #define HAMMER2_FREEMAP_LEVEL2_RADIX 38 /* 256GB */ 340 #define HAMMER2_FREEMAP_LEVEL1_RADIX 30 /* 1GB */ 341 #define HAMMER2_FREEMAP_LEVEL0_RADIX 22 /* 4MB (x 256 in l-1 leaf) */ 342 343 #define HAMMER2_FREEMAP_LEVELN_PSIZE 32768 /* physical bytes */ 344 345 #define HAMMER2_FREEMAP_LEVEL5_SIZE ((hammer2_off_t)1 << \ 346 HAMMER2_FREEMAP_LEVEL5_RADIX) 347 #define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \ 348 HAMMER2_FREEMAP_LEVEL4_RADIX) 349 #define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \ 350 HAMMER2_FREEMAP_LEVEL3_RADIX) 351 #define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \ 352 HAMMER2_FREEMAP_LEVEL2_RADIX) 353 #define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \ 354 HAMMER2_FREEMAP_LEVEL1_RADIX) 355 #define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \ 356 HAMMER2_FREEMAP_LEVEL0_RADIX) 357 358 #define HAMMER2_FREEMAP_LEVEL5_MASK (HAMMER2_FREEMAP_LEVEL5_SIZE - 1) 359 #define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1) 360 #define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1) 361 #define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1) 362 #define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1) 363 #define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1) 364 365 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 366 sizeof(hammer2_bmap_data_t)) 367 368 /* 369 * XXX I made a mistake and made the reserved area begin at each LEVEL1 zone, 370 * which is on a 1GB demark. This will eat a little more space but for 371 * now we retain compatibility and make FMZONEBASE every 1GB 372 */ 373 #define H2FMZONEBASE(key) ((key) & ~HAMMER2_FREEMAP_LEVEL1_MASK) 374 #define H2FMBASE(key, radix) rounddown2(key, (hammer2_off_t)1 << (radix)) 375 376 /* 377 * 16KB bitmap granularity (x2 bits per entry). 378 */ 379 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 380 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 381 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 382 383 /* 384 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE. 385 * 386 * 8 x 64-bit elements, 2 bits per block. 387 * 32 blocks (radix 5) per element. 388 * representing INDEX_SIZE bytes worth of storage per element. 389 */ 390 391 typedef uint64_t hammer2_bitmap_t; 392 393 #define HAMMER2_BMAP_ALLONES ((hammer2_bitmap_t)-1) 394 #define HAMMER2_BMAP_ELEMENTS 8 395 #define HAMMER2_BMAP_BITS_PER_ELEMENT 64 396 #define HAMMER2_BMAP_INDEX_RADIX 5 /* 32 blocks per element */ 397 #define HAMMER2_BMAP_BLOCKS_PER_ELEMENT (1 << HAMMER2_BMAP_INDEX_RADIX) 398 399 #define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * \ 400 HAMMER2_BMAP_BLOCKS_PER_ELEMENT) 401 #define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1) 402 403 #define HAMMER2_BMAP_SIZE (HAMMER2_BMAP_INDEX_SIZE * \ 404 HAMMER2_BMAP_ELEMENTS) 405 #define HAMMER2_BMAP_MASK (HAMMER2_BMAP_SIZE - 1) 406 407 /* 408 * Two linear areas can be reserved after the initial 4MB segment in the base 409 * zone (the one starting at offset 0). These areas are NOT managed by the 410 * block allocator and do not fall under HAMMER2 crc checking rules based 411 * at the volume header (but can be self-CRCd internally, depending). 412 */ 413 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 414 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 415 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 416 417 #define HAMMER2_AUX_MIN_BYTES HAMMER2_VOLUME_ALIGN 418 #define HAMMER2_AUX_NOM_BYTES (256*1024*1024) 419 #define HAMMER2_AUX_MAX_BYTES (1024*1024*1024) 420 421 /* 422 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 423 * Transaction ids are monotonic. 424 * 425 * We utilize 32-bit iSCSI CRCs. 426 */ 427 typedef uint64_t hammer2_tid_t; 428 typedef uint64_t hammer2_off_t; 429 typedef uint64_t hammer2_key_t; 430 typedef uint32_t hammer2_crc32_t; 431 432 /* 433 * Miscellanious ranges (all are unsigned). 434 */ 435 #define HAMMER2_TID_MIN 1ULL 436 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 437 #define HAMMER2_KEY_MIN 0ULL 438 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 439 440 /* 441 * HAMMER2 data offset special cases and masking. 442 * 443 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 444 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 445 * 446 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 447 * of the data offset field specifies how large the data chunk being pointed 448 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 449 * needed in the low bits of the data offset field). However, the practical 450 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 451 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but 452 * we fully intend to support larger extents in the future. 453 * 454 * WARNING! A radix of 0 (such as when data_off is all 0's) is a special 455 * case which means no data associated with the blockref, and 456 * not the '1 byte' it would otherwise calculate to. 457 */ 458 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 459 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 460 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 461 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 462 463 /* 464 * HAMMER2 directory support and pre-defined keys 465 */ 466 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 467 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 468 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 469 #if 0 470 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 471 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 472 #endif 473 474 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 475 #define HAMMER2_BOOT_KEY 0xd9b36ce135528000ULL /* sroot to BOOT PFS */ 476 477 /************************************************************************ 478 * DMSG SUPPORT * 479 ************************************************************************ 480 * LNK_VOLCONF 481 * 482 * All HAMMER2 directories directly under the super-root on your local 483 * media can be mounted separately, even if they share the same physical 484 * device. 485 * 486 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 487 * cluster via local media. The local media does not have to participate 488 * in the cluster, other than to provide the hammer2_volconf[] array and 489 * root inode for the mount. 490 * 491 * This is important: The mount device path you specify serves to bootstrap 492 * your entry into the cluster, but your mount will make active connections 493 * to ALL copy elements in the hammer2_volconf[] array which match the 494 * PFSID of the directory in the super-root that you specified. The local 495 * media path does not have to be mentioned in this array but becomes part 496 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 497 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 498 * 499 * The actual cluster may be far larger than the elements you list in the 500 * hammer2_volconf[] array. You list only the elements you wish to 501 * directly connect to and you are able to access the rest of the cluster 502 * indirectly through those connections. 503 * 504 * WARNING! This structure must be exactly 128 bytes long for its config 505 * array to fit in the volume header. 506 */ 507 struct hammer2_volconf { 508 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 509 uint8_t inprog; /* 01 operation in progress, or 0 */ 510 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 511 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 512 uint16_t flags; /* 04-05 flags field */ 513 uint8_t error; /* 06 last operational error */ 514 uint8_t priority; /* 07 priority and round-robin flag */ 515 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 516 uint8_t reserved08[23]; /* 09-1F */ 517 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 518 uint8_t label[16]; /* 30-3F import/export label */ 519 uint8_t path[64]; /* 40-7F target specification string or key */ 520 } __packed; 521 522 typedef struct hammer2_volconf hammer2_volconf_t; 523 524 #define DMSG_VOLF_ENABLED 0x0001 525 #define DMSG_VOLF_INPROG 0x0002 526 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 527 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 528 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 529 530 struct dmsg_lnk_hammer2_volconf { 531 dmsg_hdr_t head; 532 hammer2_volconf_t copy; /* copy spec */ 533 int32_t index; 534 int32_t unused01; 535 uuid_t mediaid; 536 int64_t reserved02[32]; 537 } __packed; 538 539 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 540 541 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 542 dmsg_lnk_hammer2_volconf) 543 544 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 545 546 /* 547 * HAMMER2 directory entry header (embedded in blockref) exactly 16 bytes 548 */ 549 struct hammer2_dirent_head { 550 hammer2_tid_t inum; /* inode number */ 551 uint16_t namlen; /* name length */ 552 uint8_t type; /* OBJTYPE_* */ 553 uint8_t unused0B; 554 uint8_t unused0C[4]; 555 } __packed; 556 557 typedef struct hammer2_dirent_head hammer2_dirent_head_t; 558 559 /* 560 * The media block reference structure. This forms the core of the HAMMER2 561 * media topology recursion. This 128-byte data structure is embedded in the 562 * volume header, in inodes (which are also directory entries), and in 563 * indirect blocks. 564 * 565 * A blockref references a single media item, which typically can be a 566 * directory entry (aka inode), indirect block, or data block. 567 * 568 * The primary feature a blockref represents is the ability to validate 569 * the entire tree underneath it via its check code. Any modification to 570 * anything propagates up the blockref tree all the way to the root, replacing 571 * the related blocks and compounding the generated check code. 572 * 573 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, or as 574 * complex as a 512 bit cryptographic hash. I originally used a 64-byte 575 * blockref but later expanded it to 128 bytes to be able to support the 576 * larger check code as well as to embed statistics for quota operation. 577 * 578 * Simple check codes are not sufficient for unverified dedup. Even with 579 * a maximally-sized check code unverified dedup should only be used in 580 * in subdirectory trees where you do not need 100% data integrity. 581 * 582 * Unverified dedup is deduping based on meta-data only without verifying 583 * that the data blocks are actually identical. Verified dedup guarantees 584 * integrity but is a far more I/O-expensive operation. 585 * 586 * -- 587 * 588 * mirror_tid - per cluster node modified (propagated upward by flush) 589 * modify_tid - clc record modified (not propagated). 590 * update_tid - clc record updated (propagated upward on verification) 591 * 592 * CLC - Stands for 'Cluster Level Change', identifiers which are identical 593 * within the topology across all cluster nodes (when fully 594 * synchronized). 595 * 596 * NOTE: The range of keys represented by the blockref is (key) to 597 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 598 * blocks bottom-up, inserting a new root when radix expansion 599 * is required. 600 * 601 * leaf_count - Helps manage leaf collapse calculations when indirect 602 * blocks become mostly empty. This value caps out at 603 * HAMMER2_BLOCKREF_LEAF_MAX (65535). 604 * 605 * Used by the chain code to determine when to pull leafs up 606 * from nearly empty indirect blocks. For the purposes of this 607 * calculation, BREF_TYPE_INODE is considered a leaf, along 608 * with DIRENT and DATA. 609 * 610 * RESERVED FIELDS 611 * 612 * A number of blockref fields are reserved and should generally be set to 613 * 0 for future compatibility. 614 * 615 * FUTURE BLOCKREF EXPANSION 616 * 617 * CONTENT ADDRESSABLE INDEXING (future) - Using a 256 or 512-bit check code. 618 */ 619 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 620 uint8_t type; /* type of underlying item */ 621 uint8_t methods; /* check method & compression method */ 622 uint8_t copyid; /* specify which copy this is */ 623 uint8_t keybits; /* #of keybits masked off 0=leaf */ 624 uint8_t vradix; /* virtual data/meta-data size */ 625 uint8_t flags; /* blockref flags */ 626 uint16_t leaf_count; /* leaf aggregation count */ 627 hammer2_key_t key; /* key specification */ 628 hammer2_tid_t mirror_tid; /* media flush topology & freemap */ 629 hammer2_tid_t modify_tid; /* clc modify (not propagated) */ 630 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 631 hammer2_tid_t update_tid; /* clc modify (propagated upward) */ 632 union { 633 char buf[16]; 634 635 /* 636 * Directory entry header (BREF_TYPE_DIRENT) 637 * 638 * NOTE: check.buf contains filename if <= 64 bytes. Longer 639 * filenames are stored in a data reference of size 640 * HAMMER2_ALLOC_MIN (at least 256, typically 1024). 641 * 642 * NOTE: inode structure may contain a copy of a recently 643 * associated filename, for recovery purposes. 644 * 645 * NOTE: Superroot entries are INODEs, not DIRENTs. Code 646 * allows both cases. 647 */ 648 hammer2_dirent_head_t dirent; 649 650 /* 651 * Statistics aggregation (BREF_TYPE_INODE, BREF_TYPE_INDIRECT) 652 */ 653 struct { 654 hammer2_key_t data_count; 655 hammer2_key_t inode_count; 656 } stats; 657 } embed; 658 union { /* check info */ 659 char buf[64]; 660 struct { 661 uint32_t value; 662 uint32_t reserved[15]; 663 } iscsi32; 664 struct { 665 uint64_t value; 666 uint64_t reserved[7]; 667 } xxhash64; 668 struct { 669 char data[24]; 670 char reserved[40]; 671 } sha192; 672 struct { 673 char data[32]; 674 char reserved[32]; 675 } sha256; 676 struct { 677 char data[64]; 678 } sha512; 679 680 /* 681 * Freemap hints are embedded in addition to the icrc32. 682 * 683 * bigmask - Radixes available for allocation (0-31). 684 * Heuristical (may be permissive but not 685 * restrictive). Typically only radix values 686 * 10-16 are used (i.e. (1<<10) through (1<<16)). 687 * 688 * avail - Total available space remaining, in bytes 689 */ 690 struct { 691 uint32_t icrc32; 692 uint32_t bigmask; /* available radixes */ 693 uint64_t avail; /* total available bytes */ 694 char reserved[48]; 695 } freemap; 696 } check; 697 } __packed; 698 699 typedef struct hammer2_blockref hammer2_blockref_t; 700 701 #define HAMMER2_BLOCKREF_BYTES 128 /* blockref struct in bytes */ 702 #define HAMMER2_BLOCKREF_RADIX 7 703 704 #define HAMMER2_BLOCKREF_LEAF_MAX 65535 705 706 /* 707 * On-media and off-media blockref types. 708 * 709 * types >= 128 are pseudo values that should never be present on-media. 710 */ 711 #define HAMMER2_BREF_TYPE_EMPTY 0 712 #define HAMMER2_BREF_TYPE_INODE 1 713 #define HAMMER2_BREF_TYPE_INDIRECT 2 714 #define HAMMER2_BREF_TYPE_DATA 3 715 #define HAMMER2_BREF_TYPE_DIRENT 4 716 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 717 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 718 #define HAMMER2_BREF_TYPE_INVALID 7 719 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 720 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 721 722 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 723 #define HAMMER2_BREF_FLAG_ZERO 0x02 /* NO LONGER USED */ 724 #define HAMMER2_BREF_FLAG_EMERG_MIP 0x04 /* emerg modified-in-place */ 725 726 /* 727 * Encode/decode check mode and compression mode for 728 * bref.methods. The compression level is not encoded in 729 * bref.methods. 730 */ 731 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 732 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 733 #define HAMMER2_ENC_COMP(n) ((n) & 15) 734 #define HAMMER2_DEC_COMP(n) ((n) & 15) 735 736 #define HAMMER2_CHECK_NONE 0 737 #define HAMMER2_CHECK_DISABLED 1 738 #define HAMMER2_CHECK_ISCSI32 2 739 #define HAMMER2_CHECK_XXHASH64 3 740 #define HAMMER2_CHECK_SHA192 4 741 #define HAMMER2_CHECK_FREEMAP 5 742 743 #define HAMMER2_CHECK_DEFAULT HAMMER2_CHECK_XXHASH64 744 745 /* user-specifiable check modes only */ 746 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 747 "xxhash64", "sha192" } 748 #define HAMMER2_CHECK_STRINGS_COUNT 5 749 750 /* 751 * Encode/decode check or compression algorithm request in 752 * ipdata->meta.check_algo and ipdata->meta.comp_algo. 753 */ 754 #define HAMMER2_ENC_ALGO(n) (n) 755 #define HAMMER2_DEC_ALGO(n) ((n) & 15) 756 #define HAMMER2_ENC_LEVEL(n) ((n) << 4) 757 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 758 759 #define HAMMER2_COMP_NONE 0 760 #define HAMMER2_COMP_AUTOZERO 1 761 #define HAMMER2_COMP_LZ4 2 762 #define HAMMER2_COMP_ZLIB 3 763 764 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 765 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 766 #define HAMMER2_COMP_STRINGS_COUNT 4 767 768 /* 769 * Passed to hammer2_chain_create(), causes methods to be inherited from 770 * parent. 771 */ 772 #define HAMMER2_METH_DEFAULT -1 773 774 /* 775 * HAMMER2 block references are collected into sets of 4 blockrefs. These 776 * sets are fully associative, meaning the elements making up a set may 777 * contain duplicate entries, holes, but valid elements are always sorted. 778 * 779 * When redundancy is desired a set may contain several duplicate 780 * entries pointing to different copies of the same data. Up to 4 copies 781 * are supported. Not implemented. 782 * 783 * When a set fills up another level of indirection is inserted, moving 784 * some or all of the set's contents into indirect blocks placed under the 785 * set. This is a top-down approach in that indirect blocks are not created 786 * until the set actually becomes full (that is, the entries in the set can 787 * shortcut the indirect blocks when the set is not full). Depending on how 788 * things are filled multiple indirect blocks will eventually be created. 789 */ 790 struct hammer2_blockset { 791 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 792 }; 793 794 typedef struct hammer2_blockset hammer2_blockset_t; 795 796 /* 797 * Catch programmer snafus 798 */ 799 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 800 #error "hammer2 direct radix is incorrect" 801 #endif 802 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 803 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 804 #endif 805 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 806 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 807 #endif 808 809 /* 810 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 811 * 812 * Each 128-byte entry contains the bitmap and meta-data required to manage 813 * a LEVEL0 (4MB) block of storage. The storage is managed in 256 x 16KB 814 * chunks. 815 * 816 * A smaller allocation granularity is supported via a linear iterator and/or 817 * must otherwise be tracked in ram. 818 * 819 * (data structure must be 128 bytes exactly) 820 * 821 * linear - A BYTE linear allocation offset used for sub-16KB allocations 822 * only. May contain values between 0 and 4MB. Must be ignored 823 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 824 * used to sub-allocate within the 16KB block (which is already 825 * marked as allocated in the bitmap). 826 * 827 * Sub-allocations need only be 1KB-aligned and do not have to be 828 * size-aligned, and 16KB or larger allocations do not update this 829 * field, resulting in pretty good packing. 830 * 831 * Please note that file data granularity may be limited by 832 * other issues such as buffer cache direct-mapping and the 833 * desire to support sector sizes up to 16KB (so H2 only issues 834 * I/O's in multiples of 16KB anyway). 835 * 836 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 837 * free. Used to cluster device buffers so all elements must have 838 * the same device block size, but may mix logical sizes. 839 * 840 * Typically integrated with the blockref type in the upper 8 bits 841 * to localize inodes and indrect blocks, improving bulk free scans 842 * and directory scans. 843 * 844 * bitmap - Two bits per 16KB allocation block arranged in arrays of 845 * 64-bit elements, 256x2 bits representing ~4MB worth of media 846 * storage. Bit patterns are as follows: 847 * 848 * 00 Unallocated 849 * 01 (reserved) 850 * 10 Possibly free 851 * 11 Allocated 852 * 853 * ========== 854 * level6 freemap 855 * blockref[0] : 4EB 856 * blockref[1] : 4EB 857 * blockref[2] : 4EB 858 * blockref[3] : 4EB 859 * ----------------------------------------------------------------------- 860 * 4 x 128B = 512B : 4 x 4EB = 16EB 861 * 862 * level2-5 FREEMAP_NODE 863 * blockref[0] : 1GB,256GB,64TB,16PB 864 * blockref[1] : 1GB,256GB,64TB,16PB 865 * ... 866 * blockref[255] : 1GB,256GB,64TB,16PB 867 * ----------------------------------------------------------------------- 868 * 256 x 128B = 32KB : 256 x 1GB,256GB,64TB,16PB = 256GB,64TB,16PB,4EB 869 * 870 * level1 FREEMAP_LEAF 871 * bmap_data[0] : 8 x 8B = 512bits = 256 x 2bits -> 256 x 16KB = 4MB 872 * bmap_data[1] : 8 x 8B = 512bits = 256 x 2bits -> 256 x 16KB = 4MB 873 * ... 874 * bmap_data[255] : 8 x 8B = 512bits = 256 x 2bits -> 256 x 16KB = 4MB 875 * ----------------------------------------------------------------------- 876 * 256 x 128B = 32KB : 256 x 4MB = 1GB 877 * ========== 878 */ 879 struct hammer2_bmap_data { 880 int32_t linear; /* 00 linear sub-granular allocation offset */ 881 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 882 uint8_t reserved06; /* 06 */ 883 uint8_t reserved07; /* 07 */ 884 uint32_t reserved08; /* 08 */ 885 uint32_t reserved0C; /* 0C */ 886 uint32_t reserved10; /* 10 */ 887 uint32_t reserved14; /* 14 */ 888 uint32_t reserved18; /* 18 */ 889 uint32_t avail; /* 1C */ 890 uint32_t reserved20[8]; /* 20-3F 256 bits manages 128K/1KB/2-bits */ 891 /* 40-7F 512 bits manages 4MB of storage */ 892 hammer2_bitmap_t bitmapq[HAMMER2_BMAP_ELEMENTS]; 893 } __packed; 894 895 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 896 897 /* 898 * The inode number is stored in the inode rather than being 899 * based on the location of the inode (since the location moves every time 900 * the inode or anything underneath the inode is modified). 901 * 902 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 903 * for the filename, and 512 bytes worth of direct file data OR an embedded 904 * blockset. The in-memory hammer2_inode structure contains only the mostly- 905 * node-independent meta-data portion (some flags are node-specific and will 906 * not be synchronized). The rest of the inode is node-specific and chain I/O 907 * is required to obtain it. 908 * 909 * Directories represent one inode per blockref. Inodes are not laid out 910 * as a file but instead are represented by the related blockrefs. The 911 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 912 * that blocksets are fully associative, so a certain degree efficiency is 913 * achieved just from that. 914 * 915 * Up to 512 bytes of direct data can be embedded in an inode, and since 916 * inodes are essentially directory entries this also means that small data 917 * files end up simply being laid out linearly in the directory, resulting 918 * in fewer seeks and highly optimal access. 919 * 920 * The compression mode can be changed at any time in the inode and is 921 * recorded on a blockref-by-blockref basis. 922 */ 923 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 924 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 925 #define HAMMER2_INODE_VERSION_ONE 1 926 927 #define HAMMER2_INODE_START 1024 /* dynamically allocated */ 928 929 struct hammer2_inode_meta { 930 uint16_t version; /* 0000 inode data version */ 931 uint8_t reserved02; /* 0002 */ 932 uint8_t pfs_subtype; /* 0003 pfs sub-type */ 933 934 /* 935 * core inode attributes, inode type, misc flags 936 */ 937 uint32_t uflags; /* 0004 chflags */ 938 uint32_t rmajor; /* 0008 available for device nodes */ 939 uint32_t rminor; /* 000C available for device nodes */ 940 uint64_t ctime; /* 0010 inode change time */ 941 uint64_t mtime; /* 0018 modified time */ 942 uint64_t atime; /* 0020 access time (unsupported) */ 943 uint64_t btime; /* 0028 birth time */ 944 uuid_t uid; /* 0030 uid / degenerate unix uid */ 945 uuid_t gid; /* 0040 gid / degenerate unix gid */ 946 947 uint8_t type; /* 0050 object type */ 948 uint8_t op_flags; /* 0051 operational flags */ 949 uint16_t cap_flags; /* 0052 capability flags */ 950 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 951 952 /* 953 * inode size, identification, localized recursive configuration 954 * for compression and backup copies. 955 * 956 * NOTE: Nominal parent inode number (iparent) is only applicable 957 * for directories but can also help for files during 958 * catastrophic recovery. 959 */ 960 hammer2_tid_t inum; /* 0058 inode number */ 961 hammer2_off_t size; /* 0060 size of file */ 962 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 963 hammer2_tid_t iparent; /* 0070 nominal parent inum */ 964 hammer2_key_t name_key; /* 0078 full filename key */ 965 uint16_t name_len; /* 0080 filename length */ 966 uint8_t ncopies; /* 0082 ncopies to local media */ 967 uint8_t comp_algo; /* 0083 compression request & algo */ 968 969 /* 970 * These fields are currently only applicable to PFSROOTs. 971 * 972 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 973 * identify an instance of a PFS in the cluster because 974 * a mount may contain more than one copy of the PFS as 975 * a separate node. {pfs_clid, pfs_fsid} must be used for 976 * registration in the cluster. 977 */ 978 uint8_t target_type; /* 0084 hardlink target type */ 979 uint8_t check_algo; /* 0085 check code request & algo */ 980 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 981 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 982 hammer2_tid_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 983 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 984 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 985 986 /* 987 * Quotas and aggregate sub-tree inode and data counters. Note that 988 * quotas are not replicated downward, they are explicitly set by 989 * the sysop and in-memory structures keep track of inheritance. 990 */ 991 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 992 hammer2_key_t unusedB8; /* 00B8 subtree byte count */ 993 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 994 hammer2_key_t unusedC8; /* 00C8 subtree inode count */ 995 996 /* 997 * The last snapshot tid is tested against modify_tid to determine 998 * when a copy must be made of a data block whos check mode has been 999 * disabled (a disabled check mode allows data blocks to be updated 1000 * in place instead of copy-on-write). 1001 */ 1002 hammer2_tid_t pfs_lsnap_tid; /* 00D0 last snapshot tid */ 1003 hammer2_tid_t reservedD8; /* 00D8 (avail) */ 1004 1005 /* 1006 * Tracks (possibly degenerate) free areas covering all sub-tree 1007 * allocations under inode, not counting the inode itself. 1008 * 0/0 indicates empty entry. fully set-associative. 1009 * 1010 * (not yet implemented) 1011 */ 1012 uint64_t decrypt_check; /* 00E0 decryption validator */ 1013 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */ 1014 } __packed; 1015 1016 typedef struct hammer2_inode_meta hammer2_inode_meta_t; 1017 1018 struct hammer2_inode_data { 1019 hammer2_inode_meta_t meta; /* 0000-00FF */ 1020 unsigned char filename[HAMMER2_INODE_MAXNAME]; 1021 /* 0100-01FF (256 char, unterminated) */ 1022 union { /* 0200-03FF (64x8 = 512 bytes) */ 1023 hammer2_blockset_t blockset; 1024 char data[HAMMER2_EMBEDDED_BYTES]; 1025 } u; 1026 } __packed; 1027 1028 typedef struct hammer2_inode_data hammer2_inode_data_t; 1029 1030 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 1031 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 1032 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 1033 1034 #define HAMMER2_OBJTYPE_UNKNOWN 0 1035 #define HAMMER2_OBJTYPE_DIRECTORY 1 1036 #define HAMMER2_OBJTYPE_REGFILE 2 1037 #define HAMMER2_OBJTYPE_FIFO 4 1038 #define HAMMER2_OBJTYPE_CDEV 5 1039 #define HAMMER2_OBJTYPE_BDEV 6 1040 #define HAMMER2_OBJTYPE_SOFTLINK 7 1041 #define HAMMER2_OBJTYPE_UNUSED08 8 1042 #define HAMMER2_OBJTYPE_SOCKET 9 1043 #define HAMMER2_OBJTYPE_WHITEOUT 10 1044 1045 #define HAMMER2_COPYID_NONE 0 1046 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 1047 1048 #define HAMMER2_COPYID_COUNT 256 1049 1050 /* 1051 * PFS types identify the role of a PFS within a cluster. The PFS types 1052 * is stored on media and in LNK_SPAN messages and used in other places. 1053 * 1054 * The low 4 bits specify the current active type while the high 4 bits 1055 * specify the transition target if the PFS is being upgraded or downgraded, 1056 * If the upper 4 bits are not zero it may effect how a PFS is used during 1057 * the transition. 1058 * 1059 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until 1060 * at least all MASTERs have updated their pfs_nmasters field. And upgrading 1061 * a SLAVE to a MASTER cannot complete until the new prospective master has 1062 * been fully synchronized (though theoretically full synchronization is 1063 * not required if a (new) quorum of other masters are fully synchronized). 1064 * 1065 * It generally does not matter which PFS element you actually mount, you 1066 * are mounting 'the cluster'. So, for example, a network mount will mount 1067 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions. 1068 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs 1069 * must be directly mounted. 1070 */ 1071 #define HAMMER2_PFSTYPE_NONE 0x00 1072 #define HAMMER2_PFSTYPE_CACHE 0x01 1073 #define HAMMER2_PFSTYPE_UNUSED02 0x02 1074 #define HAMMER2_PFSTYPE_SLAVE 0x03 1075 #define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04 1076 #define HAMMER2_PFSTYPE_SOFT_MASTER 0x05 1077 #define HAMMER2_PFSTYPE_MASTER 0x06 1078 #define HAMMER2_PFSTYPE_UNUSED07 0x07 1079 #define HAMMER2_PFSTYPE_SUPROOT 0x08 1080 #define HAMMER2_PFSTYPE_DUMMY 0x09 1081 #define HAMMER2_PFSTYPE_MAX 16 1082 1083 #define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */ 1084 #define HAMMER2_PFSTRAN_CACHE 0x10 1085 #define HAMMER2_PFSTRAN_UNMUSED20 0x20 1086 #define HAMMER2_PFSTRAN_SLAVE 0x30 1087 #define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40 1088 #define HAMMER2_PFSTRAN_SOFT_MASTER 0x50 1089 #define HAMMER2_PFSTRAN_MASTER 0x60 1090 #define HAMMER2_PFSTRAN_UNUSED70 0x70 1091 #define HAMMER2_PFSTRAN_SUPROOT 0x80 1092 #define HAMMER2_PFSTRAN_DUMMY 0x90 1093 1094 #define HAMMER2_PFS_DEC(n) ((n) & 0x0F) 1095 #define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F) 1096 #define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4) 1097 1098 #define HAMMER2_PFSSUBTYPE_NONE 0 1099 #define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */ 1100 #define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */ 1101 1102 /* 1103 * PFS mode of operation is a bitmask. This is typically not stored 1104 * on-media, but defined here because the field may be used in dmsgs. 1105 */ 1106 #define HAMMER2_PFSMODE_QUORUM 0x01 1107 #define HAMMER2_PFSMODE_RW 0x02 1108 1109 /* 1110 * The volume header eats a 64K block at the beginning of each 2GB zone 1111 * up to four copies. 1112 * 1113 * All information is stored in host byte order. The volume header's magic 1114 * number may be checked to determine the byte order. If you wish to mount 1115 * between machines w/ different endian modes you'll need filesystem code 1116 * which acts on the media data consistently (either all one way or all the 1117 * other). Our code currently does not do that. 1118 * 1119 * A read-write mount may have to recover missing allocations by doing an 1120 * incremental mirror scan looking for modifications made after alloc_tid. 1121 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 1122 * operations are usually very, very fast. 1123 * 1124 * Read-only mounts do not need to do any recovery, access to the filesystem 1125 * topology is always consistent after a crash (is always consistent, period). 1126 * However, there may be shortcutted blockref updates present from deep in 1127 * the tree which are stored in the volumeh eader and must be tracked on 1128 * the fly. 1129 * 1130 * NOTE: The copyinfo[] array contains the configuration for both the 1131 * cluster connections and any local media copies. The volume 1132 * header will be replicated for each local media copy. 1133 * 1134 * The mount command may specify multiple medias or just one and 1135 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 1136 * array on mount. 1137 * 1138 * NOTE: sroot_blockset points to the super-root directory, not the root 1139 * directory. The root directory will be a subdirectory under the 1140 * super-root. 1141 * 1142 * The super-root directory contains all root directories and all 1143 * snapshots (readonly or writable). It is possible to do a 1144 * null-mount of the super-root using special path constructions 1145 * relative to your mounted root. 1146 */ 1147 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 1148 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 1149 1150 /* 1151 * If volume version is HAMMER2_VOL_VERSION_MULTI_VOLUMES or above, max 1152 * HAMMER2_MAX_VOLUMES volumes are supported. There must be 1 (and only 1) 1153 * volume with volume id HAMMER2_ROOT_VOLUME. 1154 * Otherwise filesystem only supports 1 volume, and that volume must have 1155 * volume id HAMMER2_ROOT_VOLUME(0) which was a reserved field then. 1156 */ 1157 #define HAMMER2_MAX_VOLUMES 64 1158 #define HAMMER2_ROOT_VOLUME 0 1159 1160 struct hammer2_volume_data { 1161 /* 1162 * sector #0 - 512 bytes 1163 */ 1164 uint64_t magic; /* 0000 Signature */ 1165 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 1166 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 1167 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 1168 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 1169 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 1170 1171 uint32_t version; /* 0030 */ 1172 uint32_t flags; /* 0034 */ 1173 uint8_t copyid; /* 0038 copyid of phys vol */ 1174 uint8_t freemap_version; /* 0039 freemap algorithm */ 1175 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 1176 uint8_t volu_id; /* 003B */ 1177 uint8_t nvolumes; /* 003C */ 1178 uint8_t reserved003D; /* 003D */ 1179 uint16_t reserved003E; /* 003E */ 1180 1181 uuid_t fsid; /* 0040 */ 1182 uuid_t fstype; /* 0050 */ 1183 1184 /* 1185 * allocator_size is precalculated at newfs time and does not include 1186 * reserved blocks, boot, or aux areas. 1187 * 1188 * Initial non-reserved-area allocations do not use the freemap 1189 * but instead adjust alloc_iterator. Dynamic allocations take 1190 * over starting at (allocator_beg). This makes newfs_hammer2's 1191 * job a lot easier and can also serve as a testing jig. 1192 */ 1193 hammer2_off_t allocator_size; /* 0060 Total data space */ 1194 hammer2_off_t allocator_free; /* 0068 Free space */ 1195 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 1196 1197 /* 1198 * mirror_tid reflects the highest committed change for this 1199 * block device regardless of whether it is to the super-root 1200 * or to a PFS or whatever. 1201 * 1202 * freemap_tid reflects the highest committed freemap change for 1203 * this block device. 1204 */ 1205 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 1206 hammer2_tid_t reserved0080; /* 0080 */ 1207 hammer2_tid_t reserved0088; /* 0088 */ 1208 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 1209 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 1210 hammer2_tid_t reserved00A0[4]; /* 00A0-00BF */ 1211 1212 hammer2_off_t total_size; /* 00C0 Total volume size, bytes */ 1213 1214 /* 1215 * Copyids are allocated dynamically from the copyexists bitmap. 1216 * An id from the active copies set (up to 8, see copyinfo later on) 1217 * may still exist after the copy set has been removed from the 1218 * volume header and its bit will remain active in the bitmap and 1219 * cannot be reused until it is 100% removed from the hierarchy. 1220 */ 1221 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1222 char reserved0140[248]; /* 00E8-01DF */ 1223 1224 /* 1225 * 32 bit CRC array at the end of the first 512 byte sector. 1226 * 1227 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1228 * the other icrc's except this one). 1229 * 1230 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1231 * the blockset for the root. 1232 * 1233 * icrc_sects[5] - Sector 2 1234 * icrc_sects[4] - Sector 3 1235 * icrc_sects[3] - Sector 4 (the freemap blockset) 1236 */ 1237 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1238 1239 /* 1240 * sector #1 - 512 bytes 1241 * 1242 * The entire sector is used by a blockset, but currently only first 1243 * blockref is used. 1244 */ 1245 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1246 1247 /* 1248 * sector #2-6 1249 */ 1250 char sector2[512]; /* 0400-05FF reserved */ 1251 char sector3[512]; /* 0600-07FF reserved */ 1252 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1253 char sector5[512]; /* 0A00-0BFF reserved */ 1254 char sector6[512]; /* 0C00-0DFF reserved */ 1255 1256 /* 1257 * sector #7 - 512 bytes 1258 * Maximum 64 volume offsets within logical offset. 1259 */ 1260 hammer2_off_t volu_loff[HAMMER2_MAX_VOLUMES]; 1261 1262 /* 1263 * sector #8-71 - 32768 bytes 1264 * 1265 * Contains the configuration for up to 256 copyinfo targets. These 1266 * specify local and remote copies operating as masters or slaves. 1267 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1268 * indicates the local media). 1269 */ 1270 /* 1000-8FFF copyinfo config */ 1271 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1272 1273 /* 1274 * Remaining sections are reserved for future use. 1275 */ 1276 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1277 1278 /* 1279 * icrc on entire volume header 1280 */ 1281 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1282 } __packed; 1283 1284 typedef struct hammer2_volume_data hammer2_volume_data_t; 1285 1286 /* 1287 * Various parts of the volume header have their own iCRCs. 1288 * 1289 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1290 * and not included the icrc calculation. 1291 * 1292 * The second 512 bytes also has its own iCRC but it is stored in the first 1293 * 512 bytes so it covers the entire second 512 bytes. 1294 * 1295 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1296 * which is where the iCRC for the whole volume is stored. This is currently 1297 * a catch-all for anything not individually iCRCd. 1298 */ 1299 #define HAMMER2_VOL_ICRC_SECT0 7 1300 #define HAMMER2_VOL_ICRC_SECT1 6 1301 1302 #define HAMMER2_VOLUME_BYTES 65536 1303 1304 #define HAMMER2_VOLUME_ICRC0_OFF 0 1305 #define HAMMER2_VOLUME_ICRC1_OFF 512 1306 #define HAMMER2_VOLUME_ICRCVH_OFF 0 1307 1308 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1309 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 1310 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1311 1312 #define HAMMER2_VOL_VERSION_MULTI_VOLUMES 2 1313 1314 #define HAMMER2_VOL_VERSION_MIN 1 1315 #define HAMMER2_VOL_VERSION_DEFAULT HAMMER2_VOL_VERSION_MULTI_VOLUMES 1316 #define HAMMER2_VOL_VERSION_WIP (HAMMER2_VOL_VERSION_MULTI_VOLUMES + 1) 1317 1318 #define HAMMER2_NUM_VOLHDRS 4 1319 1320 union hammer2_media_data { 1321 hammer2_volume_data_t voldata; 1322 hammer2_inode_data_t ipdata; 1323 hammer2_blockset_t blkset; 1324 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1325 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1326 char buf[HAMMER2_PBUFSIZE]; 1327 } __packed; 1328 1329 typedef union hammer2_media_data hammer2_media_data_t; 1330 1331 #endif /* !_VFS_HAMMER2_DISK_H_ */ 1332