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_MAX (HAMMER2_IND_BYTES_MAX / \ 119 sizeof(hammer2_blockref_t)) 120 121 /* 122 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 123 * any element can occur at any index and holes can be anywhere. 124 * 125 * Inodes embed either 512 bytes of direct data or an array of 4 blockrefs, 126 * resulting in highly efficient storage for files <= 512 bytes and for files 127 * <= 512KB. Up to 4 directory entries can be referenced from a directory 128 * without requiring an indirect block. 129 */ 130 #define HAMMER2_SET_RADIX 2 /* radix 2 = 4 entries */ 131 #define HAMMER2_SET_COUNT (1 << HAMMER2_SET_RADIX) 132 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 133 #define HAMMER2_EMBEDDED_RADIX 9 134 135 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 136 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 137 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 138 139 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 140 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 141 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 142 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 143 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 144 145 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 146 147 /* 148 * A 4MB segment is reserved at the beginning of each 1GB. This segment 149 * contains the volume header (or backup volume header), the free block 150 * table, and possibly other information in the future. 151 * 152 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 153 * 154 * ========== 155 * 0 volume header (for the first four 2GB zones) 156 * 1 freemap00 level1 FREEMAP_LEAF (256 x 128B bitmap data per 1GB) 157 * 2 level2 FREEMAP_NODE (256 x 128B indirect block per 256GB) 158 * 3 level3 FREEMAP_NODE (256 x 128B indirect block per 64TB) 159 * 4 level4 FREEMAP_NODE (256 x 128B indirect block per 16PB) 160 * 5 level5 FREEMAP_NODE (256 x 128B indirect block per 4EB) 161 * 6 freemap01 level1 (rotation) 162 * 7 level2 163 * 8 level3 164 * 9 level4 165 * 10 level5 166 * 11 freemap02 level1 (rotation) 167 * 12 level2 168 * 13 level3 169 * 14 level4 170 * 15 level5 171 * 16 freemap03 level1 (rotation) 172 * 17 level2 173 * 18 level3 174 * 19 level4 175 * 20 level5 176 * 21 freemap04 level1 (rotation) 177 * 22 level2 178 * 23 level3 179 * 24 level4 180 * 25 level5 181 * 26 freemap05 level1 (rotation) 182 * 27 level2 183 * 28 level3 184 * 29 level4 185 * 30 level5 186 * 31 freemap06 level1 (rotation) 187 * 32 level2 188 * 33 level3 189 * 34 level4 190 * 35 level5 191 * 36 freemap07 level1 (rotation) 192 * 37 level2 193 * 38 level3 194 * 39 level4 195 * 40 level5 196 * 41 unused 197 * .. unused 198 * 63 unused 199 * ========== 200 * 201 * The first four 2GB zones contain volume headers and volume header backups. 202 * After that the volume header block# is reserved for future use. Similarly, 203 * there are many blocks related to various Freemap levels which are not 204 * used in every segment and those are also reserved for future use. 205 * Note that each FREEMAP_LEAF or FREEMAP_NODE uses 32KB out of 64KB slot. 206 * 207 * Freemap (see the FREEMAP document) 208 * 209 * The freemap utilizes blocks #1-40 in 8 sets of 5 blocks. Each block in 210 * a set represents a level of depth in the freemap topology. Eight sets 211 * exist to prevent live updates from disturbing the state of the freemap 212 * were a crash/reboot to occur. That is, a live update is not committed 213 * until the update's flush reaches the volume root. There are FOUR volume 214 * roots representing the last four synchronization points, so the freemap 215 * must be consistent no matter which volume root is chosen by the mount 216 * code. 217 * 218 * Each freemap set is 5 x 64K blocks and represents the 1GB, 256GB, 64TB, 219 * 16PB and 4EB indirect map. The volume header itself has a set of 4 freemap 220 * blockrefs representing another 2 bits, giving us a total 64 bits of 221 * representable address space. 222 * 223 * The Level 0 64KB block represents 1GB of storage represented by 32KB 224 * (256 x struct hammer2_bmap_data). Each structure represents 4MB of storage 225 * and has a 512 bit bitmap, using 2 bits to represent a 16KB chunk of 226 * storage. These 2 bits represent the following states: 227 * 228 * 00 Free 229 * 01 (reserved) (Possibly partially allocated) 230 * 10 Possibly free 231 * 11 Allocated 232 * 233 * One important thing to note here is that the freemap resolution is 16KB, 234 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps 235 * track of sub-allocations in memory, which means that on a unmount or reboot 236 * the entire 16KB of a partially allocated block will be considered fully 237 * allocated. It is possible for fragmentation to build up over time, but 238 * defragmentation is fairly easy to accomplish since all modifications 239 * allocate a new block. 240 * 241 * The Second thing to note is that due to the way snapshots and inode 242 * replication works, deleting a file cannot immediately free the related 243 * space. Furthermore, deletions often do not bother to traverse the 244 * block subhierarchy being deleted. And to go even further, whole 245 * sub-directory trees can be deleted simply by deleting the directory inode 246 * at the top. So even though we have a symbol to represent a 'possibly free' 247 * block (binary 10), only the bulk free scanning code can actually use it. 248 * Normal 'rm's or other deletions do not. 249 * 250 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 251 * (i.e. a multiple of 4MB). VOLUME_ALIGN must be >= ZONE_SEG. 252 * 253 * In Summary: 254 * 255 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block 256 * from the next set). The new copy is reused until a flush occurs at 257 * which point the next modification will then rotate to the next set. 258 */ 259 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 260 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 261 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 262 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 263 264 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 265 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 266 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 267 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 268 269 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 270 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 271 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 272 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 273 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 274 275 #define HAMMER2_ZONE_FREEMAP_INC 5 /* 5 deep */ 276 277 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 278 #define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */ 279 #define HAMMER2_ZONE_FREEMAP_01 6 /* normal freemap rotation */ 280 #define HAMMER2_ZONE_FREEMAP_02 11 /* normal freemap rotation */ 281 #define HAMMER2_ZONE_FREEMAP_03 16 /* normal freemap rotation */ 282 #define HAMMER2_ZONE_FREEMAP_04 21 /* normal freemap rotation */ 283 #define HAMMER2_ZONE_FREEMAP_05 26 /* normal freemap rotation */ 284 #define HAMMER2_ZONE_FREEMAP_06 31 /* normal freemap rotation */ 285 #define HAMMER2_ZONE_FREEMAP_07 36 /* normal freemap rotation */ 286 #define HAMMER2_ZONE_FREEMAP_END 41 /* (non-inclusive) */ 287 288 #define HAMMER2_ZONE_UNUSED41 41 289 #define HAMMER2_ZONE_UNUSED42 42 290 #define HAMMER2_ZONE_UNUSED43 43 291 #define HAMMER2_ZONE_UNUSED44 44 292 #define HAMMER2_ZONE_UNUSED45 45 293 #define HAMMER2_ZONE_UNUSED46 46 294 #define HAMMER2_ZONE_UNUSED47 47 295 #define HAMMER2_ZONE_UNUSED48 48 296 #define HAMMER2_ZONE_UNUSED49 49 297 #define HAMMER2_ZONE_UNUSED50 50 298 #define HAMMER2_ZONE_UNUSED51 51 299 #define HAMMER2_ZONE_UNUSED52 52 300 #define HAMMER2_ZONE_UNUSED53 53 301 #define HAMMER2_ZONE_UNUSED54 54 302 #define HAMMER2_ZONE_UNUSED55 55 303 #define HAMMER2_ZONE_UNUSED56 56 304 #define HAMMER2_ZONE_UNUSED57 57 305 #define HAMMER2_ZONE_UNUSED58 58 306 #define HAMMER2_ZONE_UNUSED59 59 307 #define HAMMER2_ZONE_UNUSED60 60 308 #define HAMMER2_ZONE_UNUSED61 61 309 #define HAMMER2_ZONE_UNUSED62 62 310 #define HAMMER2_ZONE_UNUSED63 63 311 #define HAMMER2_ZONE_END 64 /* non-inclusive */ 312 313 #define HAMMER2_NFREEMAPS 8 /* FREEMAP_00 - FREEMAP_07 */ 314 315 /* relative to FREEMAP_x */ 316 #define HAMMER2_ZONEFM_LEVEL1 0 /* 1GB leafmap */ 317 #define HAMMER2_ZONEFM_LEVEL2 1 /* 256GB indmap */ 318 #define HAMMER2_ZONEFM_LEVEL3 2 /* 64TB indmap */ 319 #define HAMMER2_ZONEFM_LEVEL4 3 /* 16PB indmap */ 320 #define HAMMER2_ZONEFM_LEVEL5 4 /* 4EB indmap */ 321 /* LEVEL6 is a set of 4 blockrefs in the volume header 16EB */ 322 323 /* 324 * Freemap radix. Assumes a set-count of 4, 128-byte blockrefs, 325 * 32KB indirect block for freemap (LEVELN_PSIZE below). 326 * 327 * Leaf entry represents 4MB of storage broken down into a 512-bit 328 * bitmap, 2-bits per entry. So course bitmap item represents 16KB. 329 */ 330 #if HAMMER2_SET_COUNT != 4 331 #error "hammer2_disk.h - freemap assumes SET_COUNT is 4" 332 #endif 333 #define HAMMER2_FREEMAP_LEVEL6_RADIX 64 /* 16EB (end) */ 334 #define HAMMER2_FREEMAP_LEVEL5_RADIX 62 /* 4EB */ 335 #define HAMMER2_FREEMAP_LEVEL4_RADIX 54 /* 16PB */ 336 #define HAMMER2_FREEMAP_LEVEL3_RADIX 46 /* 64TB */ 337 #define HAMMER2_FREEMAP_LEVEL2_RADIX 38 /* 256GB */ 338 #define HAMMER2_FREEMAP_LEVEL1_RADIX 30 /* 1GB */ 339 #define HAMMER2_FREEMAP_LEVEL0_RADIX 22 /* 4MB (x 256 in l-1 leaf) */ 340 341 #define HAMMER2_FREEMAP_LEVELN_PSIZE 32768 /* physical bytes */ 342 343 #define HAMMER2_FREEMAP_LEVEL5_SIZE ((hammer2_off_t)1 << \ 344 HAMMER2_FREEMAP_LEVEL5_RADIX) 345 #define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \ 346 HAMMER2_FREEMAP_LEVEL4_RADIX) 347 #define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \ 348 HAMMER2_FREEMAP_LEVEL3_RADIX) 349 #define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \ 350 HAMMER2_FREEMAP_LEVEL2_RADIX) 351 #define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \ 352 HAMMER2_FREEMAP_LEVEL1_RADIX) 353 #define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \ 354 HAMMER2_FREEMAP_LEVEL0_RADIX) 355 356 #define HAMMER2_FREEMAP_LEVEL5_MASK (HAMMER2_FREEMAP_LEVEL5_SIZE - 1) 357 #define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1) 358 #define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1) 359 #define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1) 360 #define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1) 361 #define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1) 362 363 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 364 sizeof(hammer2_bmap_data_t)) 365 366 /* 367 * XXX I made a mistake and made the reserved area begin at each LEVEL1 zone, 368 * which is on a 1GB demark. This will eat a little more space but for 369 * now we retain compatibility and make FMZONEBASE every 1GB 370 */ 371 #define H2FMZONEBASE(key) ((key) & ~HAMMER2_FREEMAP_LEVEL1_MASK) 372 #define H2FMBASE(key, radix) rounddown2(key, (hammer2_off_t)1 << (radix)) 373 374 /* 375 * 16KB bitmap granularity (x2 bits per entry). 376 */ 377 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 378 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 379 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 380 381 /* 382 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE. 383 * 384 * 8 x 64-bit elements, 2 bits per block. 385 * 32 blocks (radix 5) per element. 386 * representing INDEX_SIZE bytes worth of storage per element. 387 */ 388 389 typedef uint64_t hammer2_bitmap_t; 390 391 #define HAMMER2_BMAP_ALLONES ((hammer2_bitmap_t)-1) 392 #define HAMMER2_BMAP_ELEMENTS 8 393 #define HAMMER2_BMAP_BITS_PER_ELEMENT 64 394 #define HAMMER2_BMAP_INDEX_RADIX 5 /* 32 blocks per element */ 395 #define HAMMER2_BMAP_BLOCKS_PER_ELEMENT (1 << HAMMER2_BMAP_INDEX_RADIX) 396 397 #define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * \ 398 HAMMER2_BMAP_BLOCKS_PER_ELEMENT) 399 #define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1) 400 401 #define HAMMER2_BMAP_SIZE (HAMMER2_BMAP_INDEX_SIZE * \ 402 HAMMER2_BMAP_ELEMENTS) 403 #define HAMMER2_BMAP_MASK (HAMMER2_BMAP_SIZE - 1) 404 405 /* 406 * Two linear areas can be reserved after the initial 4MB segment in the base 407 * zone (the one starting at offset 0). These areas are NOT managed by the 408 * block allocator and do not fall under HAMMER2 crc checking rules based 409 * at the volume header (but can be self-CRCd internally, depending). 410 */ 411 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 412 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 413 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 414 415 #define HAMMER2_AUX_MIN_BYTES HAMMER2_VOLUME_ALIGN 416 #define HAMMER2_AUX_NOM_BYTES (256*1024*1024) 417 #define HAMMER2_AUX_MAX_BYTES (1024*1024*1024) 418 419 /* 420 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 421 * Transaction ids are monotonic. 422 * 423 * We utilize 32-bit iSCSI CRCs. 424 */ 425 typedef uint64_t hammer2_tid_t; 426 typedef uint64_t hammer2_off_t; 427 typedef uint64_t hammer2_key_t; 428 typedef uint32_t hammer2_crc32_t; 429 430 /* 431 * Miscellanious ranges (all are unsigned). 432 */ 433 #define HAMMER2_TID_MIN 1ULL 434 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 435 #define HAMMER2_KEY_MIN 0ULL 436 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 437 #define HAMMER2_OFFSET_MIN 0ULL 438 #define HAMMER2_OFFSET_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 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 470 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 471 472 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 473 #define HAMMER2_BOOT_KEY 0xd9b36ce135528000ULL /* sroot to BOOT PFS */ 474 475 /************************************************************************ 476 * DMSG SUPPORT * 477 ************************************************************************ 478 * LNK_VOLCONF 479 * 480 * All HAMMER2 directories directly under the super-root on your local 481 * media can be mounted separately, even if they share the same physical 482 * device. 483 * 484 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 485 * cluster via local media. The local media does not have to participate 486 * in the cluster, other than to provide the hammer2_volconf[] array and 487 * root inode for the mount. 488 * 489 * This is important: The mount device path you specify serves to bootstrap 490 * your entry into the cluster, but your mount will make active connections 491 * to ALL copy elements in the hammer2_volconf[] array which match the 492 * PFSID of the directory in the super-root that you specified. The local 493 * media path does not have to be mentioned in this array but becomes part 494 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 495 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 496 * 497 * The actual cluster may be far larger than the elements you list in the 498 * hammer2_volconf[] array. You list only the elements you wish to 499 * directly connect to and you are able to access the rest of the cluster 500 * indirectly through those connections. 501 * 502 * WARNING! This structure must be exactly 128 bytes long for its config 503 * array to fit in the volume header. 504 */ 505 struct hammer2_volconf { 506 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 507 uint8_t inprog; /* 01 operation in progress, or 0 */ 508 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 509 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 510 uint16_t flags; /* 04-05 flags field */ 511 uint8_t error; /* 06 last operational error */ 512 uint8_t priority; /* 07 priority and round-robin flag */ 513 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 514 uint8_t reserved08[23]; /* 09-1F */ 515 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 516 uint8_t label[16]; /* 30-3F import/export label */ 517 uint8_t path[64]; /* 40-7F target specification string or key */ 518 } __packed; 519 520 typedef struct hammer2_volconf hammer2_volconf_t; 521 522 #define DMSG_VOLF_ENABLED 0x0001 523 #define DMSG_VOLF_INPROG 0x0002 524 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 525 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 526 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 527 528 struct dmsg_lnk_hammer2_volconf { 529 dmsg_hdr_t head; 530 hammer2_volconf_t copy; /* copy spec */ 531 int32_t index; 532 int32_t unused01; 533 uuid_t mediaid; 534 int64_t reserved02[32]; 535 } __packed; 536 537 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 538 539 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 540 dmsg_lnk_hammer2_volconf) 541 542 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 543 544 /* 545 * HAMMER2 directory entry header (embedded in blockref) exactly 16 bytes 546 */ 547 struct hammer2_dirent_head { 548 hammer2_tid_t inum; /* inode number */ 549 uint16_t namlen; /* name length */ 550 uint8_t type; /* OBJTYPE_* */ 551 uint8_t unused0B; 552 uint8_t unused0C[4]; 553 } __packed; 554 555 typedef struct hammer2_dirent_head hammer2_dirent_head_t; 556 557 /* 558 * The media block reference structure. This forms the core of the HAMMER2 559 * media topology recursion. This 128-byte data structure is embedded in the 560 * volume header, in inodes (which are also directory entries), and in 561 * indirect blocks. 562 * 563 * A blockref references a single media item, which typically can be a 564 * directory entry (aka inode), indirect block, or data block. 565 * 566 * The primary feature a blockref represents is the ability to validate 567 * the entire tree underneath it via its check code. Any modification to 568 * anything propagates up the blockref tree all the way to the root, replacing 569 * the related blocks and compounding the generated check code. 570 * 571 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, or as 572 * complex as a 512 bit cryptographic hash. I originally used a 64-byte 573 * blockref but later expanded it to 128 bytes to be able to support the 574 * larger check code as well as to embed statistics for quota operation. 575 * 576 * Simple check codes are not sufficient for unverified dedup. Even with 577 * a maximally-sized check code unverified dedup should only be used in 578 * in subdirectory trees where you do not need 100% data integrity. 579 * 580 * Unverified dedup is deduping based on meta-data only without verifying 581 * that the data blocks are actually identical. Verified dedup guarantees 582 * integrity but is a far more I/O-expensive operation. 583 * 584 * -- 585 * 586 * mirror_tid - per cluster node modified (propagated upward by flush) 587 * modify_tid - clc record modified (not propagated). 588 * update_tid - clc record updated (propagated upward on verification) 589 * 590 * CLC - Stands for 'Cluster Level Change', identifiers which are identical 591 * within the topology across all cluster nodes (when fully 592 * synchronized). 593 * 594 * NOTE: The range of keys represented by the blockref is (key) to 595 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 596 * blocks bottom-up, inserting a new root when radix expansion 597 * is required. 598 * 599 * leaf_count - Helps manage leaf collapse calculations when indirect 600 * blocks become mostly empty. This value caps out at 601 * HAMMER2_BLOCKREF_LEAF_MAX (65535). 602 * 603 * Used by the chain code to determine when to pull leafs up 604 * from nearly empty indirect blocks. For the purposes of this 605 * calculation, BREF_TYPE_INODE is considered a leaf, along 606 * with DIRENT and DATA. 607 * 608 * RESERVED FIELDS 609 * 610 * A number of blockref fields are reserved and should generally be set to 611 * 0 for future compatibility. 612 * 613 * FUTURE BLOCKREF EXPANSION 614 * 615 * CONTENT ADDRESSABLE INDEXING (future) - Using a 256 or 512-bit check code. 616 */ 617 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 618 uint8_t type; /* type of underlying item */ 619 uint8_t methods; /* check method & compression method */ 620 uint8_t copyid; /* specify which copy this is */ 621 uint8_t keybits; /* #of keybits masked off 0=leaf */ 622 uint8_t vradix; /* virtual data/meta-data size */ 623 uint8_t flags; /* blockref flags */ 624 uint16_t leaf_count; /* leaf aggregation count */ 625 hammer2_key_t key; /* key specification */ 626 hammer2_tid_t mirror_tid; /* media flush topology & freemap */ 627 hammer2_tid_t modify_tid; /* clc modify (not propagated) */ 628 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 629 hammer2_tid_t update_tid; /* clc modify (propagated upward) */ 630 union { 631 char buf[16]; 632 633 /* 634 * Directory entry header (BREF_TYPE_DIRENT) 635 * 636 * NOTE: check.buf contains filename if <= 64 bytes. Longer 637 * filenames are stored in a data reference of size 638 * HAMMER2_ALLOC_MIN (at least 256, typically 1024). 639 * 640 * NOTE: inode structure may contain a copy of a recently 641 * associated filename, for recovery purposes. 642 * 643 * NOTE: Superroot entries are INODEs, not DIRENTs. Code 644 * allows both cases. 645 */ 646 hammer2_dirent_head_t dirent; 647 648 /* 649 * Statistics aggregation (BREF_TYPE_INODE, BREF_TYPE_INDIRECT) 650 */ 651 struct { 652 hammer2_key_t data_count; 653 hammer2_key_t inode_count; 654 } stats; 655 } embed; 656 union { /* check info */ 657 char buf[64]; 658 struct { 659 uint32_t value; 660 uint32_t reserved[15]; 661 } iscsi32; 662 struct { 663 uint64_t value; 664 uint64_t reserved[7]; 665 } xxhash64; 666 struct { 667 char data[24]; 668 char reserved[40]; 669 } sha192; 670 struct { 671 char data[32]; 672 char reserved[32]; 673 } sha256; 674 struct { 675 char data[64]; 676 } sha512; 677 678 /* 679 * Freemap hints are embedded in addition to the icrc32. 680 * 681 * bigmask - Radixes available for allocation (0-31). 682 * Heuristical (may be permissive but not 683 * restrictive). Typically only radix values 684 * 10-16 are used (i.e. (1<<10) through (1<<16)). 685 * 686 * avail - Total available space remaining, in bytes 687 */ 688 struct { 689 uint32_t icrc32; 690 uint32_t bigmask; /* available radixes */ 691 uint64_t avail; /* total available bytes */ 692 char reserved[48]; 693 } freemap; 694 } check; 695 } __packed; 696 697 typedef struct hammer2_blockref hammer2_blockref_t; 698 699 #define HAMMER2_BLOCKREF_BYTES 128 /* blockref struct in bytes */ 700 #define HAMMER2_BLOCKREF_RADIX 7 701 702 #define HAMMER2_BLOCKREF_LEAF_MAX 65535 703 704 /* 705 * On-media and off-media blockref types. 706 * 707 * types >= 128 are pseudo values that should never be present on-media. 708 */ 709 #define HAMMER2_BREF_TYPE_EMPTY 0 710 #define HAMMER2_BREF_TYPE_INODE 1 711 #define HAMMER2_BREF_TYPE_INDIRECT 2 712 #define HAMMER2_BREF_TYPE_DATA 3 713 #define HAMMER2_BREF_TYPE_DIRENT 4 714 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 715 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 716 #define HAMMER2_BREF_TYPE_INVALID 7 717 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 718 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 719 720 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 721 #define HAMMER2_BREF_FLAG_ZERO 0x02 /* NO LONGER USED */ 722 #define HAMMER2_BREF_FLAG_EMERG_MIP 0x04 /* emerg modified-in-place */ 723 724 /* 725 * Encode/decode check mode and compression mode for 726 * bref.methods. The compression level is not encoded in 727 * bref.methods. 728 */ 729 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 730 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 731 #define HAMMER2_ENC_COMP(n) ((n) & 15) 732 #define HAMMER2_DEC_COMP(n) ((n) & 15) 733 734 #define HAMMER2_CHECK_NONE 0 735 #define HAMMER2_CHECK_DISABLED 1 736 #define HAMMER2_CHECK_ISCSI32 2 737 #define HAMMER2_CHECK_XXHASH64 3 738 #define HAMMER2_CHECK_SHA192 4 739 #define HAMMER2_CHECK_FREEMAP 5 740 741 #define HAMMER2_CHECK_DEFAULT HAMMER2_CHECK_XXHASH64 742 743 /* user-specifiable check modes only */ 744 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 745 "xxhash64", "sha192" } 746 #define HAMMER2_CHECK_STRINGS_COUNT 5 747 748 /* 749 * Encode/decode check or compression algorithm request in 750 * ipdata->meta.check_algo and ipdata->meta.comp_algo. 751 */ 752 #define HAMMER2_ENC_ALGO(n) (n) 753 #define HAMMER2_DEC_ALGO(n) ((n) & 15) 754 #define HAMMER2_ENC_LEVEL(n) ((n) << 4) 755 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 756 757 #define HAMMER2_COMP_NONE 0 758 #define HAMMER2_COMP_AUTOZERO 1 759 #define HAMMER2_COMP_LZ4 2 760 #define HAMMER2_COMP_ZLIB 3 761 762 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 763 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 764 #define HAMMER2_COMP_STRINGS_COUNT 4 765 766 /* 767 * Passed to hammer2_chain_create(), causes methods to be inherited from 768 * parent. 769 */ 770 #define HAMMER2_METH_DEFAULT -1 771 772 /* 773 * HAMMER2 block references are collected into sets of 4 blockrefs. These 774 * sets are fully associative, meaning the elements making up a set may 775 * contain duplicate entries, holes, but valid elements are always sorted. 776 * 777 * When redundancy is desired a set may contain several duplicate 778 * entries pointing to different copies of the same data. Up to 4 copies 779 * are supported. Not implemented. 780 * 781 * When a set fills up another level of indirection is inserted, moving 782 * some or all of the set's contents into indirect blocks placed under the 783 * set. This is a top-down approach in that indirect blocks are not created 784 * until the set actually becomes full (that is, the entries in the set can 785 * shortcut the indirect blocks when the set is not full). Depending on how 786 * things are filled multiple indirect blocks will eventually be created. 787 */ 788 struct hammer2_blockset { 789 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 790 }; 791 792 typedef struct hammer2_blockset hammer2_blockset_t; 793 794 /* 795 * Catch programmer snafus 796 */ 797 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 798 #error "hammer2 direct radix is incorrect" 799 #endif 800 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 801 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 802 #endif 803 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 804 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 805 #endif 806 807 /* 808 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 809 * 810 * Each 128-byte entry contains the bitmap and meta-data required to manage 811 * a LEVEL0 (4MB) block of storage. The storage is managed in 256 x 16KB 812 * chunks. 813 * 814 * A smaller allocation granularity is supported via a linear iterator and/or 815 * must otherwise be tracked in ram. 816 * 817 * (data structure must be 128 bytes exactly) 818 * 819 * linear - A BYTE linear allocation offset used for sub-16KB allocations 820 * only. May contain values between 0 and 4MB. Must be ignored 821 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 822 * used to sub-allocate within the 16KB block (which is already 823 * marked as allocated in the bitmap). 824 * 825 * Sub-allocations need only be 1KB-aligned and do not have to be 826 * size-aligned, and 16KB or larger allocations do not update this 827 * field, resulting in pretty good packing. 828 * 829 * Please note that file data granularity may be limited by 830 * other issues such as buffer cache direct-mapping and the 831 * desire to support sector sizes up to 16KB (so H2 only issues 832 * I/O's in multiples of 16KB anyway). 833 * 834 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 835 * free. Used to cluster device buffers so all elements must have 836 * the same device block size, but may mix logical sizes. 837 * 838 * Typically integrated with the blockref type in the upper 8 bits 839 * to localize inodes and indrect blocks, improving bulk free scans 840 * and directory scans. 841 * 842 * bitmap - Two bits per 16KB allocation block arranged in arrays of 843 * 64-bit elements, 256x2 bits representing ~4MB worth of media 844 * storage. Bit patterns are as follows: 845 * 846 * 00 Unallocated 847 * 01 (reserved) 848 * 10 Possibly free 849 * 11 Allocated 850 * 851 * ========== 852 * level6 freemap 853 * blockref[0] : 4EB 854 * blockref[1] : 4EB 855 * blockref[2] : 4EB 856 * blockref[3] : 4EB 857 * ----------------------------------------------------------------------- 858 * 4 x 128B = 512B : 4 x 4EB = 16EB 859 * 860 * level2-5 FREEMAP_NODE 861 * blockref[0] : 1GB,256GB,64TB,16PB 862 * blockref[1] : 1GB,256GB,64TB,16PB 863 * ... 864 * blockref[255] : 1GB,256GB,64TB,16PB 865 * ----------------------------------------------------------------------- 866 * 256 x 128B = 32KB : 256 x 1GB,256GB,64TB,16PB = 256GB,64TB,16PB,4EB 867 * 868 * level1 FREEMAP_LEAF 869 * bmap_data[0] : 8 x 8B = 512bits = 256 x 2bits -> 256 x 16KB = 4MB 870 * bmap_data[1] : 8 x 8B = 512bits = 256 x 2bits -> 256 x 16KB = 4MB 871 * ... 872 * bmap_data[255] : 8 x 8B = 512bits = 256 x 2bits -> 256 x 16KB = 4MB 873 * ----------------------------------------------------------------------- 874 * 256 x 128B = 32KB : 256 x 4MB = 1GB 875 * ========== 876 */ 877 struct hammer2_bmap_data { 878 int32_t linear; /* 00 linear sub-granular allocation offset */ 879 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 880 uint8_t reserved06; /* 06 */ 881 uint8_t reserved07; /* 07 */ 882 uint32_t reserved08; /* 08 */ 883 uint32_t reserved0C; /* 0C */ 884 uint32_t reserved10; /* 10 */ 885 uint32_t reserved14; /* 14 */ 886 uint32_t reserved18; /* 18 */ 887 uint32_t avail; /* 1C */ 888 uint32_t reserved20[8]; /* 20-3F 256 bits manages 128K/1KB/2-bits */ 889 /* 40-7F 512 bits manages 4MB of storage */ 890 hammer2_bitmap_t bitmapq[HAMMER2_BMAP_ELEMENTS]; 891 } __packed; 892 893 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 894 895 /* 896 * The inode number is stored in the inode rather than being 897 * based on the location of the inode (since the location moves every time 898 * the inode or anything underneath the inode is modified). 899 * 900 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 901 * for the filename, and 512 bytes worth of direct file data OR an embedded 902 * blockset. The in-memory hammer2_inode structure contains only the mostly- 903 * node-independent meta-data portion (some flags are node-specific and will 904 * not be synchronized). The rest of the inode is node-specific and chain I/O 905 * is required to obtain it. 906 * 907 * Directories represent one inode per blockref. Inodes are not laid out 908 * as a file but instead are represented by the related blockrefs. The 909 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 910 * that blocksets are fully associative, so a certain degree efficiency is 911 * achieved just from that. 912 * 913 * Up to 512 bytes of direct data can be embedded in an inode, and since 914 * inodes are essentially directory entries this also means that small data 915 * files end up simply being laid out linearly in the directory, resulting 916 * in fewer seeks and highly optimal access. 917 * 918 * The compression mode can be changed at any time in the inode and is 919 * recorded on a blockref-by-blockref basis. 920 */ 921 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 922 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 923 #define HAMMER2_INODE_VERSION_ONE 1 924 925 #define HAMMER2_INODE_START 1024 /* dynamically allocated */ 926 927 struct hammer2_inode_meta { 928 uint16_t version; /* 0000 inode data version */ 929 uint8_t reserved02; /* 0002 */ 930 uint8_t pfs_subtype; /* 0003 pfs sub-type */ 931 932 /* 933 * core inode attributes, inode type, misc flags 934 */ 935 uint32_t uflags; /* 0004 chflags */ 936 uint32_t rmajor; /* 0008 available for device nodes */ 937 uint32_t rminor; /* 000C available for device nodes */ 938 uint64_t ctime; /* 0010 inode change time */ 939 uint64_t mtime; /* 0018 modified time */ 940 uint64_t atime; /* 0020 access time (unsupported) */ 941 uint64_t btime; /* 0028 birth time */ 942 uuid_t uid; /* 0030 uid / degenerate unix uid */ 943 uuid_t gid; /* 0040 gid / degenerate unix gid */ 944 945 uint8_t type; /* 0050 object type */ 946 uint8_t op_flags; /* 0051 operational flags */ 947 uint16_t cap_flags; /* 0052 capability flags */ 948 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 949 950 /* 951 * inode size, identification, localized recursive configuration 952 * for compression and backup copies. 953 * 954 * NOTE: Nominal parent inode number (iparent) is only applicable 955 * for directories but can also help for files during 956 * catastrophic recovery. 957 */ 958 hammer2_tid_t inum; /* 0058 inode number */ 959 hammer2_off_t size; /* 0060 size of file */ 960 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 961 hammer2_tid_t iparent; /* 0070 nominal parent inum */ 962 hammer2_key_t name_key; /* 0078 full filename key */ 963 uint16_t name_len; /* 0080 filename length */ 964 uint8_t ncopies; /* 0082 ncopies to local media */ 965 uint8_t comp_algo; /* 0083 compression request & algo */ 966 967 /* 968 * These fields are currently only applicable to PFSROOTs. 969 * 970 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 971 * identify an instance of a PFS in the cluster because 972 * a mount may contain more than one copy of the PFS as 973 * a separate node. {pfs_clid, pfs_fsid} must be used for 974 * registration in the cluster. 975 */ 976 uint8_t target_type; /* 0084 hardlink target type */ 977 uint8_t check_algo; /* 0085 check code request & algo */ 978 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 979 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 980 hammer2_tid_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 981 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 982 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 983 984 /* 985 * Quotas and aggregate sub-tree inode and data counters. Note that 986 * quotas are not replicated downward, they are explicitly set by 987 * the sysop and in-memory structures keep track of inheritance. 988 */ 989 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 990 hammer2_key_t unusedB8; /* 00B8 subtree byte count */ 991 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 992 hammer2_key_t unusedC8; /* 00C8 subtree inode count */ 993 994 /* 995 * The last snapshot tid is tested against modify_tid to determine 996 * when a copy must be made of a data block whos check mode has been 997 * disabled (a disabled check mode allows data blocks to be updated 998 * in place instead of copy-on-write). 999 */ 1000 hammer2_tid_t pfs_lsnap_tid; /* 00D0 last snapshot tid */ 1001 hammer2_tid_t reservedD8; /* 00D8 (avail) */ 1002 1003 /* 1004 * Tracks (possibly degenerate) free areas covering all sub-tree 1005 * allocations under inode, not counting the inode itself. 1006 * 0/0 indicates empty entry. fully set-associative. 1007 * 1008 * (not yet implemented) 1009 */ 1010 uint64_t decrypt_check; /* 00E0 decryption validator */ 1011 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */ 1012 } __packed; 1013 1014 typedef struct hammer2_inode_meta hammer2_inode_meta_t; 1015 1016 struct hammer2_inode_data { 1017 hammer2_inode_meta_t meta; /* 0000-00FF */ 1018 unsigned char filename[HAMMER2_INODE_MAXNAME]; 1019 /* 0100-01FF (256 char, unterminated) */ 1020 union { /* 0200-03FF (64x8 = 512 bytes) */ 1021 hammer2_blockset_t blockset; 1022 char data[HAMMER2_EMBEDDED_BYTES]; 1023 } u; 1024 } __packed; 1025 1026 typedef struct hammer2_inode_data hammer2_inode_data_t; 1027 1028 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 1029 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 1030 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 1031 1032 #define HAMMER2_OBJTYPE_UNKNOWN 0 1033 #define HAMMER2_OBJTYPE_DIRECTORY 1 1034 #define HAMMER2_OBJTYPE_REGFILE 2 1035 #define HAMMER2_OBJTYPE_FIFO 4 1036 #define HAMMER2_OBJTYPE_CDEV 5 1037 #define HAMMER2_OBJTYPE_BDEV 6 1038 #define HAMMER2_OBJTYPE_SOFTLINK 7 1039 #define HAMMER2_OBJTYPE_UNUSED08 8 1040 #define HAMMER2_OBJTYPE_SOCKET 9 1041 #define HAMMER2_OBJTYPE_WHITEOUT 10 1042 1043 #define HAMMER2_COPYID_NONE 0 1044 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 1045 1046 #define HAMMER2_COPYID_COUNT 256 1047 1048 /* 1049 * PFS types identify the role of a PFS within a cluster. The PFS types 1050 * is stored on media and in LNK_SPAN messages and used in other places. 1051 * 1052 * The low 4 bits specify the current active type while the high 4 bits 1053 * specify the transition target if the PFS is being upgraded or downgraded, 1054 * If the upper 4 bits are not zero it may effect how a PFS is used during 1055 * the transition. 1056 * 1057 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until 1058 * at least all MASTERs have updated their pfs_nmasters field. And upgrading 1059 * a SLAVE to a MASTER cannot complete until the new prospective master has 1060 * been fully synchronized (though theoretically full synchronization is 1061 * not required if a (new) quorum of other masters are fully synchronized). 1062 * 1063 * It generally does not matter which PFS element you actually mount, you 1064 * are mounting 'the cluster'. So, for example, a network mount will mount 1065 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions. 1066 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs 1067 * must be directly mounted. 1068 */ 1069 #define HAMMER2_PFSTYPE_NONE 0x00 1070 #define HAMMER2_PFSTYPE_CACHE 0x01 1071 #define HAMMER2_PFSTYPE_UNUSED02 0x02 1072 #define HAMMER2_PFSTYPE_SLAVE 0x03 1073 #define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04 1074 #define HAMMER2_PFSTYPE_SOFT_MASTER 0x05 1075 #define HAMMER2_PFSTYPE_MASTER 0x06 1076 #define HAMMER2_PFSTYPE_UNUSED07 0x07 1077 #define HAMMER2_PFSTYPE_SUPROOT 0x08 1078 #define HAMMER2_PFSTYPE_DUMMY 0x09 1079 #define HAMMER2_PFSTYPE_MAX 16 1080 1081 #define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */ 1082 #define HAMMER2_PFSTRAN_CACHE 0x10 1083 #define HAMMER2_PFSTRAN_UNMUSED20 0x20 1084 #define HAMMER2_PFSTRAN_SLAVE 0x30 1085 #define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40 1086 #define HAMMER2_PFSTRAN_SOFT_MASTER 0x50 1087 #define HAMMER2_PFSTRAN_MASTER 0x60 1088 #define HAMMER2_PFSTRAN_UNUSED70 0x70 1089 #define HAMMER2_PFSTRAN_SUPROOT 0x80 1090 #define HAMMER2_PFSTRAN_DUMMY 0x90 1091 1092 #define HAMMER2_PFS_DEC(n) ((n) & 0x0F) 1093 #define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F) 1094 #define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4) 1095 1096 #define HAMMER2_PFSSUBTYPE_NONE 0 1097 #define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */ 1098 #define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */ 1099 1100 /* 1101 * PFS mode of operation is a bitmask. This is typically not stored 1102 * on-media, but defined here because the field may be used in dmsgs. 1103 */ 1104 #define HAMMER2_PFSMODE_QUORUM 0x01 1105 #define HAMMER2_PFSMODE_RW 0x02 1106 1107 /* 1108 * The volume header eats a 64K block at the beginning of each 2GB zone 1109 * up to four copies. 1110 * 1111 * All information is stored in host byte order. The volume header's magic 1112 * number may be checked to determine the byte order. If you wish to mount 1113 * between machines w/ different endian modes you'll need filesystem code 1114 * which acts on the media data consistently (either all one way or all the 1115 * other). Our code currently does not do that. 1116 * 1117 * A read-write mount may have to recover missing allocations by doing an 1118 * incremental mirror scan looking for modifications made after alloc_tid. 1119 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 1120 * operations are usually very, very fast. 1121 * 1122 * Read-only mounts do not need to do any recovery, access to the filesystem 1123 * topology is always consistent after a crash (is always consistent, period). 1124 * However, there may be shortcutted blockref updates present from deep in 1125 * the tree which are stored in the volumeh eader and must be tracked on 1126 * the fly. 1127 * 1128 * NOTE: The copyinfo[] array contains the configuration for both the 1129 * cluster connections and any local media copies. The volume 1130 * header will be replicated for each local media copy. 1131 * 1132 * The mount command may specify multiple medias or just one and 1133 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 1134 * array on mount. 1135 * 1136 * NOTE: sroot_blockset points to the super-root directory, not the root 1137 * directory. The root directory will be a subdirectory under the 1138 * super-root. 1139 * 1140 * The super-root directory contains all root directories and all 1141 * snapshots (readonly or writable). It is possible to do a 1142 * null-mount of the super-root using special path constructions 1143 * relative to your mounted root. 1144 */ 1145 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 1146 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 1147 1148 /* 1149 * If volume version is HAMMER2_VOL_VERSION_MULTI_VOLUMES or above, max 1150 * HAMMER2_MAX_VOLUMES volumes are supported. There must be 1 (and only 1) 1151 * volume with volume id HAMMER2_ROOT_VOLUME. 1152 * Otherwise filesystem only supports 1 volume, and that volume must have 1153 * volume id HAMMER2_ROOT_VOLUME(0) which was a reserved field then. 1154 */ 1155 #define HAMMER2_MAX_VOLUMES 64 1156 #define HAMMER2_ROOT_VOLUME 0 1157 1158 struct hammer2_volume_data { 1159 /* 1160 * sector #0 - 512 bytes 1161 */ 1162 uint64_t magic; /* 0000 Signature */ 1163 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 1164 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 1165 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 1166 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 1167 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 1168 1169 uint32_t version; /* 0030 */ 1170 uint32_t flags; /* 0034 */ 1171 uint8_t copyid; /* 0038 copyid of phys vol */ 1172 uint8_t freemap_version; /* 0039 freemap algorithm */ 1173 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 1174 uint8_t volu_id; /* 003B */ 1175 uint8_t nvolumes; /* 003C */ 1176 uint8_t reserved003D; /* 003D */ 1177 uint16_t reserved003E; /* 003E */ 1178 1179 uuid_t fsid; /* 0040 */ 1180 uuid_t fstype; /* 0050 */ 1181 1182 /* 1183 * allocator_size is precalculated at newfs time and does not include 1184 * reserved blocks, boot, or aux areas. 1185 * 1186 * Initial non-reserved-area allocations do not use the freemap 1187 * but instead adjust alloc_iterator. Dynamic allocations take 1188 * over starting at (allocator_beg). This makes newfs_hammer2's 1189 * job a lot easier and can also serve as a testing jig. 1190 */ 1191 hammer2_off_t allocator_size; /* 0060 Total data space */ 1192 hammer2_off_t allocator_free; /* 0068 Free space */ 1193 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 1194 1195 /* 1196 * mirror_tid reflects the highest committed change for this 1197 * block device regardless of whether it is to the super-root 1198 * or to a PFS or whatever. 1199 * 1200 * freemap_tid reflects the highest committed freemap change for 1201 * this block device. 1202 */ 1203 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 1204 hammer2_tid_t reserved0080; /* 0080 */ 1205 hammer2_tid_t reserved0088; /* 0088 */ 1206 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 1207 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 1208 hammer2_tid_t reserved00A0[4]; /* 00A0-00BF */ 1209 1210 hammer2_off_t total_size; /* 00C0 Total volume size, bytes */ 1211 1212 /* 1213 * Copyids are allocated dynamically from the copyexists bitmap. 1214 * An id from the active copies set (up to 8, see copyinfo later on) 1215 * may still exist after the copy set has been removed from the 1216 * volume header and its bit will remain active in the bitmap and 1217 * cannot be reused until it is 100% removed from the hierarchy. 1218 */ 1219 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1220 char reserved0140[248]; /* 00E8-01DF */ 1221 1222 /* 1223 * 32 bit CRC array at the end of the first 512 byte sector. 1224 * 1225 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1226 * the other icrc's except this one). 1227 * 1228 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1229 * the blockset for the root. 1230 * 1231 * icrc_sects[5] - Sector 2 1232 * icrc_sects[4] - Sector 3 1233 * icrc_sects[3] - Sector 4 (the freemap blockset) 1234 */ 1235 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1236 1237 /* 1238 * sector #1 - 512 bytes 1239 * 1240 * The entire sector is used by a blockset, but currently only first 1241 * blockref is used. 1242 */ 1243 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1244 1245 /* 1246 * sector #2-6 1247 */ 1248 char sector2[512]; /* 0400-05FF reserved */ 1249 char sector3[512]; /* 0600-07FF reserved */ 1250 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1251 char sector5[512]; /* 0A00-0BFF reserved */ 1252 char sector6[512]; /* 0C00-0DFF reserved */ 1253 1254 /* 1255 * sector #7 - 512 bytes 1256 * Maximum 64 volume offsets within logical offset. 1257 */ 1258 hammer2_off_t volu_loff[HAMMER2_MAX_VOLUMES]; 1259 1260 /* 1261 * sector #8-71 - 32768 bytes 1262 * 1263 * Contains the configuration for up to 256 copyinfo targets. These 1264 * specify local and remote copies operating as masters or slaves. 1265 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1266 * indicates the local media). 1267 */ 1268 /* 1000-8FFF copyinfo config */ 1269 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1270 1271 /* 1272 * Remaining sections are reserved for future use. 1273 */ 1274 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1275 1276 /* 1277 * icrc on entire volume header 1278 */ 1279 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1280 } __packed; 1281 1282 typedef struct hammer2_volume_data hammer2_volume_data_t; 1283 1284 /* 1285 * Various parts of the volume header have their own iCRCs. 1286 * 1287 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1288 * and not included the icrc calculation. 1289 * 1290 * The second 512 bytes also has its own iCRC but it is stored in the first 1291 * 512 bytes so it covers the entire second 512 bytes. 1292 * 1293 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1294 * which is where the iCRC for the whole volume is stored. This is currently 1295 * a catch-all for anything not individually iCRCd. 1296 */ 1297 #define HAMMER2_VOL_ICRC_SECT0 7 1298 #define HAMMER2_VOL_ICRC_SECT1 6 1299 1300 #define HAMMER2_VOLUME_BYTES 65536 1301 1302 #define HAMMER2_VOLUME_ICRC0_OFF 0 1303 #define HAMMER2_VOLUME_ICRC1_OFF 512 1304 #define HAMMER2_VOLUME_ICRCVH_OFF 0 1305 1306 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1307 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 1308 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1309 1310 #define HAMMER2_VOL_VERSION_MULTI_VOLUMES 2 1311 1312 #define HAMMER2_VOL_VERSION_MIN 1 1313 #define HAMMER2_VOL_VERSION_DEFAULT HAMMER2_VOL_VERSION_MULTI_VOLUMES 1314 #define HAMMER2_VOL_VERSION_WIP (HAMMER2_VOL_VERSION_MULTI_VOLUMES + 1) 1315 1316 #define HAMMER2_NUM_VOLHDRS 4 1317 1318 union hammer2_media_data { 1319 hammer2_volume_data_t voldata; 1320 hammer2_inode_data_t ipdata; 1321 hammer2_blockset_t blkset; 1322 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1323 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1324 char buf[HAMMER2_PBUFSIZE]; 1325 } __packed; 1326 1327 typedef union hammer2_media_data hammer2_media_data_t; 1328 1329 #endif /* !_VFS_HAMMER2_DISK_H_ */ 1330