1 /* 2 * Copyright (c) 2011-2014 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@dragonflybsd.org> 6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the 17 * distribution. 18 * 3. Neither the name of The DragonFly Project nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific, prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 36 #ifndef _VFS_HAMMER2_DISK_H_ 37 #define _VFS_HAMMER2_DISK_H_ 38 39 #ifndef _SYS_UUID_H_ 40 #include <sys/uuid.h> 41 #endif 42 #ifndef _SYS_DMSG_H_ 43 #include <sys/dmsg.h> 44 #endif 45 46 /* 47 * The structures below represent the on-disk media structures for the HAMMER2 48 * filesystem. Note that all fields for on-disk structures are naturally 49 * aligned. The host endian format is typically used - compatibility is 50 * possible if the implementation detects reversed endian and adjusts accesses 51 * accordingly. 52 * 53 * HAMMER2 primarily revolves around the directory topology: inodes, 54 * directory entries, and block tables. Block device buffer cache buffers 55 * are always 64KB. Logical file buffers are typically 16KB. All data 56 * references utilize 64-bit byte offsets. 57 * 58 * Free block management is handled independently using blocks reserved by 59 * the media topology. 60 */ 61 62 /* 63 * The data at the end of a file or directory may be a fragment in order 64 * to optimize storage efficiency. The minimum fragment size is 1KB. 65 * Since allocations are in powers of 2 fragments must also be sized in 66 * powers of 2 (1024, 2048, ... 65536). 67 * 68 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K), 69 * which is 2^16. Larger extents may be supported in the future. Smaller 70 * fragments might be supported in the future (down to 64 bytes is possible), 71 * but probably will not be. 72 * 73 * A full indirect block use supports 512 x 128-byte blockrefs in a 64KB 74 * buffer. Indirect blocks down to 1KB are supported to keep small 75 * directories small. 76 * 77 * A maximally sized file (2^64-1 bytes) requires ~6 indirect block levels 78 * using 64KB indirect blocks (128 byte refs, 512 or radix 9 per indblk). 79 * 80 * 16(datablk) + 9 + 9 + 9 + 9 + 9 + 9 = ~70. 81 * 16(datablk) + 7 + 9 + 9 + 9 + 9 + 9 = ~68. (smaller top level indblk) 82 * 83 * The actual depth depends on copies redundancy and whether the filesystem 84 * has chosen to use a smaller indirect block size at the top level or not. 85 */ 86 #define HAMMER2_ALLOC_MIN 1024 /* minimum allocation size */ 87 #define HAMMER2_RADIX_MIN 10 /* minimum allocation size 2^N */ 88 #define HAMMER2_ALLOC_MAX 65536 /* maximum allocation size */ 89 #define HAMMER2_RADIX_MAX 16 /* maximum allocation size 2^N */ 90 #define HAMMER2_RADIX_KEY 64 /* number of bits in key */ 91 92 /* 93 * MINALLOCSIZE - The minimum allocation size. This can be smaller 94 * or larger than the minimum physical IO size. 95 * 96 * NOTE: Should not be larger than 1K since inodes 97 * are 1K. 98 * 99 * MINIOSIZE - The minimum IO size. This must be less than 100 * or equal to HAMMER2_LBUFSIZE. 101 * 102 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups. 103 * 104 * HAMMER2_PBUFSIZE - Topological block size used by files for all 105 * blocks except the block straddling EOF. 106 * 107 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB 108 * (space represented by a level0 bitmap). 109 */ 110 111 #define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX) 112 #define HAMMER2_SEGRADIX HAMMER2_FREEMAP_LEVEL0_RADIX 113 114 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 115 #define HAMMER2_PBUFSIZE 65536 116 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 117 #define HAMMER2_LBUFSIZE 16384 118 119 /* 120 * Generally speaking we want to use 16K and 64K I/Os 121 */ 122 #define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX 123 #define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE 124 125 #define HAMMER2_IND_BYTES_MIN 4096 126 #define HAMMER2_IND_BYTES_NOM HAMMER2_LBUFSIZE 127 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 128 #define HAMMER2_IND_RADIX_MIN 12 129 #define HAMMER2_IND_RADIX_NOM HAMMER2_LBUFRADIX 130 #define HAMMER2_IND_RADIX_MAX HAMMER2_PBUFRADIX 131 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 132 sizeof(hammer2_blockref_t)) 133 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 134 sizeof(hammer2_blockref_t)) 135 136 /* 137 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 138 * any element can occur at any index and holes can be anywhere. As a 139 * future optimization we will be able to flag that such arrays are sorted 140 * and thus optimize lookups, but for now we don't. 141 * 142 * Inodes embed either 512 bytes of direct data or an array of 8 blockrefs, 143 * resulting in highly efficient storage for files <= 512 bytes and for files 144 * <= 512KB. Up to 8 directory entries can be referenced from a directory 145 * without requiring an indirect block. 146 * 147 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented), 148 * or 64KB (1024 blockrefs / ~64MB represented). 149 */ 150 #define HAMMER2_SET_RADIX 2 /* radix 2 = 4 entries */ 151 #define HAMMER2_SET_COUNT (1 << HAMMER2_SET_RADIX) 152 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 153 #define HAMMER2_EMBEDDED_RADIX 9 154 155 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 156 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 157 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 158 159 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 160 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 161 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 162 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 163 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 164 165 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 166 167 /* 168 * A HAMMER2 filesystem is always sized in multiples of 8MB. 169 * 170 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment 171 * contains the volume header (or backup volume header), the free block 172 * table, and possibly other information in the future. 173 * 174 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 175 * 176 * +-----------------------+ 177 * | Volume Hdr | block 0 volume header & alternates 178 * +-----------------------+ (first four zones only) 179 * | FreeBlk Section A | block 1-4 180 * +-----------------------+ 181 * | FreeBlk Section B | block 5-8 182 * +-----------------------+ 183 * | FreeBlk Section C | block 9-12 184 * +-----------------------+ 185 * | FreeBlk Section D | block 13-16 186 * +-----------------------+ 187 * | | block 17...63 188 * | reserved | 189 * | | 190 * +-----------------------+ 191 * 192 * The first few 2GB zones contain volume headers and volume header backups. 193 * After that the volume header block# is reserved for future use. Similarly, 194 * there are many blocks related to various Freemap levels which are not 195 * used in every segment and those are also reserved for future use. 196 * 197 * Freemap (see the FREEMAP document) 198 * 199 * The freemap utilizes blocks #1-16 in 8 sets of 4 blocks. Each block in 200 * a set represents a level of depth in the freemap topology. Eight sets 201 * exist to prevent live updates from disturbing the state of the freemap 202 * were a crash/reboot to occur. That is, a live update is not committed 203 * until the update's flush reaches the volume root. There are FOUR volume 204 * roots representing the last four synchronization points, so the freemap 205 * must be consistent no matter which volume root is chosen by the mount 206 * code. 207 * 208 * Each freemap set is 4 x 64K blocks and represents the 2GB, 2TB, 2PB, 209 * and 2EB indirect map. The volume header itself has a set of 8 freemap 210 * blockrefs representing another 3 bits, giving us a total 64 bits of 211 * representable address space. 212 * 213 * The Level 0 64KB block represents 2GB of storage represented by 214 * (64 x struct hammer2_bmap_data). Each structure represents 2MB of storage 215 * and has a 256 bit bitmap, using 2 bits to represent a 16KB chunk of 216 * storage. These 2 bits represent the following states: 217 * 218 * 00 Free 219 * 01 (reserved) (Possibly partially allocated) 220 * 10 Possibly free 221 * 11 Allocated 222 * 223 * One important thing to note here is that the freemap resolution is 16KB, 224 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps 225 * track of sub-allocations in memory, which means that on a unmount or reboot 226 * the entire 16KB of a partially allocated block will be considered fully 227 * allocated. It is possible for fragmentation to build up over time, but 228 * defragmentation is fairly easy to accomplish since all modifications 229 * allocate a new block. 230 * 231 * The Second thing to note is that due to the way snapshots and inode 232 * replication works, deleting a file cannot immediately free the related 233 * space. Furthermore, deletions often do not bother to traverse the 234 * block subhierarchy being deleted. And to go even further, whole 235 * sub-directory trees can be deleted simply by deleting the directory inode 236 * at the top. So even though we have a symbol to represent a 'possibly free' 237 * block (binary 10), only the bulk free scanning code can actually use it. 238 * Normal 'rm's or other deletions do not. 239 * 240 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 241 * (i.e. a multiple of 2MB). VOLUME_ALIGN must be >= ZONE_SEG. 242 * 243 * In Summary: 244 * 245 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block 246 * from the next set). The new copy is reused until a flush occurs at 247 * which point the next modification will then rotate to the next set. 248 * 249 * (2) A total of 10 freemap sets is required. 250 * 251 * - 8 sets - 2 sets per volume header backup x 4 volume header backups 252 * - 2 sets used as backing store for the bulk freemap scan. 253 * - The freemap recovery scan which runs on-mount just uses the inactive 254 * set for whichever volume header was selected by the mount code. 255 * 256 */ 257 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 258 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 259 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 260 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 261 262 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 263 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 264 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 265 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 266 267 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 268 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 269 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 270 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 271 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 272 273 #define HAMMER2_ZONE_FREEMAP_INC 5 /* 5 deep */ 274 275 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 276 #define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */ 277 #define HAMMER2_ZONE_FREEMAP_01 6 /* normal freemap rotation */ 278 #define HAMMER2_ZONE_FREEMAP_02 11 /* normal freemap rotation */ 279 #define HAMMER2_ZONE_FREEMAP_03 16 /* normal freemap rotation */ 280 #define HAMMER2_ZONE_FREEMAP_04 21 /* normal freemap rotation */ 281 #define HAMMER2_ZONE_FREEMAP_05 26 /* normal freemap rotation */ 282 #define HAMMER2_ZONE_FREEMAP_06 31 /* normal freemap rotation */ 283 #define HAMMER2_ZONE_FREEMAP_07 36 /* normal freemap rotation */ 284 #define HAMMER2_ZONE_FREEMAP_END 41 /* (non-inclusive) */ 285 286 #define HAMMER2_ZONE_UNUSED41 41 287 #define HAMMER2_ZONE_UNUSED42 42 288 #define HAMMER2_ZONE_UNUSED43 43 289 #define HAMMER2_ZONE_UNUSED44 44 290 #define HAMMER2_ZONE_UNUSED45 45 291 #define HAMMER2_ZONE_UNUSED46 46 292 #define HAMMER2_ZONE_UNUSED47 47 293 #define HAMMER2_ZONE_UNUSED48 48 294 #define HAMMER2_ZONE_UNUSED49 49 295 #define HAMMER2_ZONE_UNUSED50 50 296 #define HAMMER2_ZONE_UNUSED51 51 297 #define HAMMER2_ZONE_UNUSED52 52 298 #define HAMMER2_ZONE_UNUSED53 53 299 #define HAMMER2_ZONE_UNUSED54 54 300 #define HAMMER2_ZONE_UNUSED55 55 301 #define HAMMER2_ZONE_UNUSED56 56 302 #define HAMMER2_ZONE_UNUSED57 57 303 #define HAMMER2_ZONE_UNUSED58 58 304 #define HAMMER2_ZONE_UNUSED59 59 305 #define HAMMER2_ZONE_UNUSED60 60 306 #define HAMMER2_ZONE_UNUSED61 61 307 #define HAMMER2_ZONE_UNUSED62 62 308 #define HAMMER2_ZONE_UNUSED63 63 309 #define HAMMER2_ZONE_END 64 /* non-inclusive */ 310 311 #define HAMMER2_NFREEMAPS 8 /* FREEMAP_00 - FREEMAP_07 */ 312 313 /* relative to FREEMAP_x */ 314 #define HAMMER2_ZONEFM_LEVEL1 0 /* 1GB leafmap */ 315 #define HAMMER2_ZONEFM_LEVEL2 1 /* 256GB indmap */ 316 #define HAMMER2_ZONEFM_LEVEL3 2 /* 64TB indmap */ 317 #define HAMMER2_ZONEFM_LEVEL4 3 /* 16PB indmap */ 318 #define HAMMER2_ZONEFM_LEVEL5 4 /* 4EB indmap */ 319 /* LEVEL6 is a set of 4 blockrefs in the volume header 16EB */ 320 321 /* 322 * Freemap radix. Assumes a set-count of 4, 128-byte blockrefs, 323 * 32KB indirect block for freemap (LEVELN_PSIZE below). 324 * 325 * Leaf entry represents 4MB of storage broken down into a 512-bit 326 * bitmap, 2-bits per entry. So course bitmap item represents 16KB. 327 */ 328 #if HAMMER2_SET_COUNT != 4 329 #error "hammer2_disk.h - freemap assumes SET_COUNT is 4" 330 #endif 331 #define HAMMER2_FREEMAP_LEVEL6_RADIX 64 /* 16EB (end) */ 332 #define HAMMER2_FREEMAP_LEVEL5_RADIX 62 /* 4EB */ 333 #define HAMMER2_FREEMAP_LEVEL4_RADIX 54 /* 16PB */ 334 #define HAMMER2_FREEMAP_LEVEL3_RADIX 46 /* 64TB */ 335 #define HAMMER2_FREEMAP_LEVEL2_RADIX 38 /* 256GB */ 336 #define HAMMER2_FREEMAP_LEVEL1_RADIX 30 /* 1GB */ 337 #define HAMMER2_FREEMAP_LEVEL0_RADIX 22 /* 4MB (128by in l-1 leaf) */ 338 339 #define HAMMER2_FREEMAP_LEVELN_PSIZE 32768 /* physical bytes */ 340 341 #define HAMMER2_FREEMAP_LEVEL5_SIZE ((hammer2_off_t)1 << \ 342 HAMMER2_FREEMAP_LEVEL5_RADIX) 343 #define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \ 344 HAMMER2_FREEMAP_LEVEL4_RADIX) 345 #define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \ 346 HAMMER2_FREEMAP_LEVEL3_RADIX) 347 #define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \ 348 HAMMER2_FREEMAP_LEVEL2_RADIX) 349 #define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \ 350 HAMMER2_FREEMAP_LEVEL1_RADIX) 351 #define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \ 352 HAMMER2_FREEMAP_LEVEL0_RADIX) 353 354 #define HAMMER2_FREEMAP_LEVEL5_MASK (HAMMER2_FREEMAP_LEVEL5_SIZE - 1) 355 #define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1) 356 #define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1) 357 #define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1) 358 #define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1) 359 #define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1) 360 361 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 362 sizeof(hammer2_bmap_data_t)) 363 364 /* 365 * 16KB bitmap granularity (x2 bits per entry). 366 */ 367 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 368 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 369 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 370 371 /* 372 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE. 373 * 374 * 8 x 64-bit elements, 2 bits per block. 375 * 32 blocks (radix 5) per element. 376 * representing INDEX_SIZE bytes worth of storage per element. 377 */ 378 379 typedef uint64_t hammer2_bitmap_t; 380 381 #define HAMMER2_BMAP_ALLONES ((hammer2_bitmap_t)-1) 382 #define HAMMER2_BMAP_ELEMENTS 8 383 #define HAMMER2_BMAP_BITS_PER_ELEMENT 64 384 #define HAMMER2_BMAP_INDEX_RADIX 5 /* 32 blocks per element */ 385 #define HAMMER2_BMAP_BLOCKS_PER_ELEMENT (1 << HAMMER2_BMAP_INDEX_RADIX) 386 387 #define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * \ 388 HAMMER2_BMAP_BLOCKS_PER_ELEMENT) 389 #define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1) 390 391 #define HAMMER2_BMAP_SIZE (HAMMER2_BMAP_INDEX_SIZE * \ 392 HAMMER2_BMAP_ELEMENTS) 393 #define HAMMER2_BMAP_MASK (HAMMER2_BMAP_SIZE - 1) 394 395 /* 396 * Two linear areas can be reserved after the initial 2MB segment in the base 397 * zone (the one starting at offset 0). These areas are NOT managed by the 398 * block allocator and do not fall under HAMMER2 crc checking rules based 399 * at the volume header (but can be self-CRCd internally, depending). 400 */ 401 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 402 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 403 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 404 405 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 406 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 407 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 408 409 /* 410 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 411 * Transaction ids are monotonic. 412 * 413 * We utilize 32-bit iSCSI CRCs. 414 */ 415 typedef uint64_t hammer2_tid_t; 416 typedef uint64_t hammer2_off_t; 417 typedef uint64_t hammer2_key_t; 418 typedef uint32_t hammer2_crc32_t; 419 420 /* 421 * Miscellanious ranges (all are unsigned). 422 */ 423 #define HAMMER2_TID_MIN 1ULL 424 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 425 #define HAMMER2_KEY_MIN 0ULL 426 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 427 #define HAMMER2_OFFSET_MIN 0ULL 428 #define HAMMER2_OFFSET_MAX 0xFFFFFFFFFFFFFFFFULL 429 430 /* 431 * HAMMER2 data offset special cases and masking. 432 * 433 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 434 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 435 * 436 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 437 * of the data offset field specifies how large the data chunk being pointed 438 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 439 * needed in the low bits of the data offset field). However, the practical 440 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 441 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but 442 * we fully intend to support larger extents in the future. 443 * 444 * WARNING! A radix of 0 (such as when data_off is all 0's) is a special 445 * case which means no data associated with the blockref, and 446 * not the '1 byte' it would otherwise calculate to. 447 */ 448 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 449 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 450 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 451 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 452 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 453 #define HAMMER2_MAX_COPIES 6 454 455 /* 456 * HAMMER2 directory support and pre-defined keys 457 */ 458 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 459 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 460 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 461 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 462 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 463 464 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 465 #define HAMMER2_BOOT_KEY 0xd9b36ce135528000ULL /* sroot to BOOT PFS */ 466 467 /************************************************************************ 468 * DMSG SUPPORT * 469 ************************************************************************ 470 * LNK_VOLCONF 471 * 472 * All HAMMER2 directories directly under the super-root on your local 473 * media can be mounted separately, even if they share the same physical 474 * device. 475 * 476 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 477 * cluster via local media. The local media does not have to participate 478 * in the cluster, other than to provide the hammer2_volconf[] array and 479 * root inode for the mount. 480 * 481 * This is important: The mount device path you specify serves to bootstrap 482 * your entry into the cluster, but your mount will make active connections 483 * to ALL copy elements in the hammer2_volconf[] array which match the 484 * PFSID of the directory in the super-root that you specified. The local 485 * media path does not have to be mentioned in this array but becomes part 486 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 487 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 488 * 489 * The actual cluster may be far larger than the elements you list in the 490 * hammer2_volconf[] array. You list only the elements you wish to 491 * directly connect to and you are able to access the rest of the cluster 492 * indirectly through those connections. 493 * 494 * WARNING! This structure must be exactly 128 bytes long for its config 495 * array to fit in the volume header. 496 */ 497 struct hammer2_volconf { 498 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 499 uint8_t inprog; /* 01 operation in progress, or 0 */ 500 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 501 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 502 uint16_t flags; /* 04-05 flags field */ 503 uint8_t error; /* 06 last operational error */ 504 uint8_t priority; /* 07 priority and round-robin flag */ 505 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 506 uint8_t reserved08[23]; /* 09-1F */ 507 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 508 uint8_t label[16]; /* 30-3F import/export label */ 509 uint8_t path[64]; /* 40-7F target specification string or key */ 510 } __packed; 511 512 typedef struct hammer2_volconf hammer2_volconf_t; 513 514 #define DMSG_VOLF_ENABLED 0x0001 515 #define DMSG_VOLF_INPROG 0x0002 516 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 517 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 518 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 519 520 struct dmsg_lnk_hammer2_volconf { 521 dmsg_hdr_t head; 522 hammer2_volconf_t copy; /* copy spec */ 523 int32_t index; 524 int32_t unused01; 525 uuid_t mediaid; 526 int64_t reserved02[32]; 527 } __packed; 528 529 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 530 531 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 532 dmsg_lnk_hammer2_volconf) 533 534 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 535 536 /* 537 * HAMMER2 directory entry header (embedded in blockref) exactly 16 bytes 538 */ 539 struct hammer2_dirent_head { 540 hammer2_tid_t inum; /* inode number */ 541 uint16_t namlen; /* name length */ 542 uint8_t type; /* OBJTYPE_* */ 543 uint8_t unused0B; 544 uint8_t unused0C[4]; 545 } __packed; 546 547 typedef struct hammer2_dirent_head hammer2_dirent_head_t; 548 549 /* 550 * The media block reference structure. This forms the core of the HAMMER2 551 * media topology recursion. This 128-byte data structure is embedded in the 552 * volume header, in inodes (which are also directory entries), and in 553 * indirect blocks. 554 * 555 * A blockref references a single media item, which typically can be a 556 * directory entry (aka inode), indirect block, or data block. 557 * 558 * The primary feature a blockref represents is the ability to validate 559 * the entire tree underneath it via its check code. Any modification to 560 * anything propagates up the blockref tree all the way to the root, replacing 561 * the related blocks and compounding the generated check code. 562 * 563 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, or as 564 * complex as a 512 bit cryptographic hash. I originally used a 64-byte 565 * blockref but later expanded it to 128 bytes to be able to support the 566 * larger check code as well as to embed statistics for quota operation. 567 * 568 * Simple check codes are not sufficient for unverified dedup. Even with 569 * a maximally-sized check code unverified dedup should only be used in 570 * in subdirectory trees where you do not need 100% data integrity. 571 * 572 * Unverified dedup is deduping based on meta-data only without verifying 573 * that the data blocks are actually identical. Verified dedup guarantees 574 * integrity but is a far more I/O-expensive operation. 575 * 576 * -- 577 * 578 * mirror_tid - per cluster node modified (propagated upward by flush) 579 * modify_tid - clc record modified (not propagated). 580 * update_tid - clc record updated (propagated upward on verification) 581 * 582 * CLC - Stands for 'Cluster Level Change', identifiers which are identical 583 * within the topology across all cluster nodes (when fully 584 * synchronized). 585 * 586 * NOTE: The range of keys represented by the blockref is (key) to 587 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 588 * blocks bottom-up, inserting a new root when radix expansion 589 * is required. 590 * 591 * leaf_count - Helps manage leaf collapse calculations when indirect 592 * blocks become mostly empty. This value caps out at 593 * HAMMER2_BLOCKREF_LEAF_MAX (65535). 594 * 595 * Used by the chain code to determine when to pull leafs up 596 * from nearly empty indirect blocks. For the purposes of this 597 * calculation, BREF_TYPE_INODE is considered a leaf, along 598 * with DIRENT and DATA. 599 * 600 * RESERVED FIELDS 601 * 602 * A number of blockref fields are reserved and should generally be set to 603 * 0 for future compatibility. 604 * 605 * FUTURE BLOCKREF EXPANSION 606 * 607 * CONTENT ADDRESSABLE INDEXING (future) - Using a 256 or 512-bit check code. 608 */ 609 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 610 uint8_t type; /* type of underlying item */ 611 uint8_t methods; /* check method & compression method */ 612 uint8_t copyid; /* specify which copy this is */ 613 uint8_t keybits; /* #of keybits masked off 0=leaf */ 614 uint8_t vradix; /* virtual data/meta-data size */ 615 uint8_t flags; /* blockref flags */ 616 uint16_t leaf_count; /* leaf aggregation count */ 617 hammer2_key_t key; /* key specification */ 618 hammer2_tid_t mirror_tid; /* media flush topology & freemap */ 619 hammer2_tid_t modify_tid; /* clc modify (not propagated) */ 620 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 621 hammer2_tid_t update_tid; /* clc modify (propagated upward) */ 622 union { 623 char buf[16]; 624 625 /* 626 * Directory entry header (BREF_TYPE_DIRENT) 627 * 628 * NOTE: check.buf contains filename if <= 64 bytes. Longer 629 * filenames are stored in a data reference of size 630 * HAMMER2_ALLOC_MIN (at least 256, typically 1024). 631 * 632 * NOTE: inode structure may contain a copy of a recently 633 * associated filename, for recovery purposes. 634 * 635 * NOTE: Superroot entries are INODEs, not DIRENTs. Code 636 * allows both cases. 637 */ 638 hammer2_dirent_head_t dirent; 639 640 /* 641 * Statistics aggregation (BREF_TYPE_INODE, BREF_TYPE_INDIRECT) 642 */ 643 struct { 644 hammer2_key_t data_count; 645 hammer2_key_t inode_count; 646 } stats; 647 } embed; 648 union { /* check info */ 649 char buf[64]; 650 struct { 651 uint32_t value; 652 uint32_t reserved[15]; 653 } iscsi32; 654 struct { 655 uint64_t value; 656 uint64_t reserved[7]; 657 } xxhash64; 658 struct { 659 char data[24]; 660 char reserved[40]; 661 } sha192; 662 struct { 663 char data[32]; 664 char reserved[32]; 665 } sha256; 666 struct { 667 char data[64]; 668 } sha512; 669 670 /* 671 * Freemap hints are embedded in addition to the icrc32. 672 * 673 * bigmask - Radixes available for allocation (0-31). 674 * Heuristical (may be permissive but not 675 * restrictive). Typically only radix values 676 * 10-16 are used (i.e. (1<<10) through (1<<16)). 677 * 678 * avail - Total available space remaining, in bytes 679 */ 680 struct { 681 uint32_t icrc32; 682 uint32_t bigmask; /* available radixes */ 683 uint64_t avail; /* total available bytes */ 684 char reserved[48]; 685 } freemap; 686 } check; 687 } __packed; 688 689 typedef struct hammer2_blockref hammer2_blockref_t; 690 691 #define HAMMER2_BLOCKREF_BYTES 128 /* blockref struct in bytes */ 692 #define HAMMER2_BLOCKREF_RADIX 7 693 694 #define HAMMER2_BLOCKREF_LEAF_MAX 65535 695 696 /* 697 * On-media and off-media blockref types. 698 * 699 * types >= 128 are pseudo values that should never be present on-media. 700 */ 701 #define HAMMER2_BREF_TYPE_EMPTY 0 702 #define HAMMER2_BREF_TYPE_INODE 1 703 #define HAMMER2_BREF_TYPE_INDIRECT 2 704 #define HAMMER2_BREF_TYPE_DATA 3 705 #define HAMMER2_BREF_TYPE_DIRENT 4 706 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 707 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 708 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 709 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 710 711 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 712 #define HAMMER2_BREF_FLAG_ZERO 0x02 713 714 /* 715 * Encode/decode check mode and compression mode for 716 * bref.methods. The compression level is not encoded in 717 * bref.methods. 718 */ 719 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 720 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 721 #define HAMMER2_ENC_COMP(n) ((n) & 15) 722 #define HAMMER2_DEC_COMP(n) ((n) & 15) 723 724 #define HAMMER2_CHECK_NONE 0 725 #define HAMMER2_CHECK_DISABLED 1 726 #define HAMMER2_CHECK_ISCSI32 2 727 #define HAMMER2_CHECK_XXHASH64 3 728 #define HAMMER2_CHECK_SHA192 4 729 #define HAMMER2_CHECK_FREEMAP 5 730 731 #define HAMMER2_CHECK_DEFAULT HAMMER2_CHECK_XXHASH64 732 733 /* user-specifiable check modes only */ 734 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 735 "xxhash64", "sha192" } 736 #define HAMMER2_CHECK_STRINGS_COUNT 5 737 738 /* 739 * Encode/decode check or compression algorithm request in 740 * ipdata->meta.check_algo and ipdata->meta.comp_algo. 741 */ 742 #define HAMMER2_ENC_ALGO(n) (n) 743 #define HAMMER2_DEC_ALGO(n) ((n) & 15) 744 #define HAMMER2_ENC_LEVEL(n) ((n) << 4) 745 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 746 747 #define HAMMER2_COMP_NONE 0 748 #define HAMMER2_COMP_AUTOZERO 1 749 #define HAMMER2_COMP_LZ4 2 750 #define HAMMER2_COMP_ZLIB 3 751 752 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 753 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 754 #define HAMMER2_COMP_STRINGS_COUNT 4 755 756 /* 757 * Passed to hammer2_chain_create(), causes methods to be inherited from 758 * parent. 759 */ 760 #define HAMMER2_METH_DEFAULT -1 761 762 /* 763 * HAMMER2 block references are collected into sets of 4 blockrefs. These 764 * sets are fully associative, meaning the elements making up a set are 765 * not sorted in any way and may contain duplicate entries, holes, or 766 * entries which shortcut multiple levels of indirection. Sets are used 767 * in various ways: 768 * 769 * (1) When redundancy is desired a set may contain several duplicate 770 * entries pointing to different copies of the same data. Up to 4 copies 771 * are supported. 772 * 773 * (2) The blockrefs in a set can shortcut multiple levels of indirections 774 * within the bounds imposed by the parent of set. 775 * 776 * When a set fills up another level of indirection is inserted, moving 777 * some or all of the set's contents into indirect blocks placed under the 778 * set. This is a top-down approach in that indirect blocks are not created 779 * until the set actually becomes full (that is, the entries in the set can 780 * shortcut the indirect blocks when the set is not full). Depending on how 781 * things are filled multiple indirect blocks will eventually be created. 782 * 783 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 784 * are also treated as fully set-associative. 785 */ 786 struct hammer2_blockset { 787 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 788 }; 789 790 typedef struct hammer2_blockset hammer2_blockset_t; 791 792 /* 793 * Catch programmer snafus 794 */ 795 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 796 #error "hammer2 direct radix is incorrect" 797 #endif 798 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 799 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 800 #endif 801 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 802 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 803 #endif 804 805 /* 806 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 807 * 808 * Each 128-byte entry contains the bitmap and meta-data required to manage 809 * a LEVEL0 (128KB) block of storage. The storage is managed in 128 x 1KB 810 * chunks. 811 * 812 * A smaller allocation granularity is supported via a linear iterator and/or 813 * must otherwise be tracked in ram. 814 * 815 * (data structure must be 128 bytes exactly) 816 * 817 * linear - A BYTE linear allocation offset used for sub-16KB allocations 818 * only. May contain values between 0 and 2MB. Must be ignored 819 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 820 * used to sub-allocate within the 16KB block (which is already 821 * marked as allocated in the bitmap). 822 * 823 * Sub-allocations need only be 1KB-aligned and do not have to be 824 * size-aligned, and 16KB or larger allocations do not update this 825 * field, resulting in pretty good packing. 826 * 827 * Please note that file data granularity may be limited by 828 * other issues such as buffer cache direct-mapping and the 829 * desire to support sector sizes up to 16KB (so H2 only issues 830 * I/O's in multiples of 16KB anyway). 831 * 832 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 833 * free. Used to cluster device buffers so all elements must have 834 * the same device block size, but may mix logical sizes. 835 * 836 * Typically integrated with the blockref type in the upper 8 bits 837 * to localize inodes and indrect blocks, improving bulk free scans 838 * and directory scans. 839 * 840 * bitmap - Two bits per 16KB allocation block arranged in arrays of 841 * 32-bit elements, 256x2 bits representing ~4MB worth of media 842 * storage. Bit patterns are as follows: 843 * 844 * 00 Unallocated 845 * 01 (reserved) 846 * 10 Possibly free 847 * 11 Allocated 848 */ 849 struct hammer2_bmap_data { 850 int32_t linear; /* 00 linear sub-granular allocation offset */ 851 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 852 uint8_t reserved06; /* 06 */ 853 uint8_t reserved07; /* 07 */ 854 uint32_t reserved08; /* 08 */ 855 uint32_t reserved0C; /* 0C */ 856 uint32_t reserved10; /* 10 */ 857 uint32_t reserved14; /* 14 */ 858 uint32_t reserved18; /* 18 */ 859 uint32_t avail; /* 1C */ 860 uint32_t reserved20[8]; /* 20-3F 256 bits manages 128K/1KB/2-bits */ 861 /* 40-7F 512 bits manages 4MB of storage */ 862 hammer2_bitmap_t bitmapq[HAMMER2_BMAP_ELEMENTS]; 863 } __packed; 864 865 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 866 867 /* 868 * In HAMMER2 inodes ARE directory entries, with a special exception for 869 * hardlinks. The inode number is stored in the inode rather than being 870 * based on the location of the inode (since the location moves every time 871 * the inode or anything underneath the inode is modified). 872 * 873 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 874 * for the filename, and 512 bytes worth of direct file data OR an embedded 875 * blockset. The in-memory hammer2_inode structure contains only the mostly- 876 * node-independent meta-data portion (some flags are node-specific and will 877 * not be synchronized). The rest of the inode is node-specific and chain I/O 878 * is required to obtain it. 879 * 880 * Directories represent one inode per blockref. Inodes are not laid out 881 * as a file but instead are represented by the related blockrefs. The 882 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 883 * that blocksets are fully associative, so a certain degree efficiency is 884 * achieved just from that. 885 * 886 * Up to 512 bytes of direct data can be embedded in an inode, and since 887 * inodes are essentially directory entries this also means that small data 888 * files end up simply being laid out linearly in the directory, resulting 889 * in fewer seeks and highly optimal access. 890 * 891 * The compression mode can be changed at any time in the inode and is 892 * recorded on a blockref-by-blockref basis. 893 * 894 * Hardlinks are supported via the inode map. Essentially the way a hardlink 895 * works is that all individual directory entries representing the same file 896 * are special cased and specify the same inode number. The actual file 897 * is placed in the nearest parent directory that is parent to all instances 898 * of the hardlink. If all hardlinks to a file are in the same directory 899 * the actual file will also be placed in that directory. This file uses 900 * the inode number as the directory entry key and is invisible to normal 901 * directory scans. Real directory entry keys are differentiated from the 902 * inode number key via bit 63. Access to the hardlink silently looks up 903 * the real file and forwards all operations to that file. Removal of the 904 * last hardlink also removes the real file. 905 * 906 * (attr_tid) is only updated when the inode's specific attributes or regular 907 * file size has changed, and affects path lookups and stat. (attr_tid) 908 * represents a special cache coherency lock under the inode. The inode 909 * blockref's modify_tid will always cover it. 910 * 911 * (dirent_tid) is only updated when an entry under a directory inode has 912 * been created, deleted, renamed, or had its attributes change, and affects 913 * directory lookups and scans. (dirent_tid) represents another special cache 914 * coherency lock under the inode. The inode blockref's modify_tid will 915 * always cover it. 916 */ 917 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 918 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 919 #define HAMMER2_INODE_VERSION_ONE 1 920 921 #define HAMMER2_INODE_START 1024 /* dynamically allocated */ 922 923 struct hammer2_inode_meta { 924 uint16_t version; /* 0000 inode data version */ 925 uint8_t reserved02; /* 0002 */ 926 uint8_t pfs_subtype; /* 0003 pfs sub-type */ 927 928 /* 929 * core inode attributes, inode type, misc flags 930 */ 931 uint32_t uflags; /* 0004 chflags */ 932 uint32_t rmajor; /* 0008 available for device nodes */ 933 uint32_t rminor; /* 000C available for device nodes */ 934 uint64_t ctime; /* 0010 inode change time */ 935 uint64_t mtime; /* 0018 modified time */ 936 uint64_t atime; /* 0020 access time (unsupported) */ 937 uint64_t btime; /* 0028 birth time */ 938 uuid_t uid; /* 0030 uid / degenerate unix uid */ 939 uuid_t gid; /* 0040 gid / degenerate unix gid */ 940 941 uint8_t type; /* 0050 object type */ 942 uint8_t op_flags; /* 0051 operational flags */ 943 uint16_t cap_flags; /* 0052 capability flags */ 944 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 945 946 /* 947 * inode size, identification, localized recursive configuration 948 * for compression and backup copies. 949 * 950 * NOTE: Nominal parent inode number (iparent) is only applicable 951 * for directories but can also help for files during 952 * catastrophic recovery. 953 */ 954 hammer2_tid_t inum; /* 0058 inode number */ 955 hammer2_off_t size; /* 0060 size of file */ 956 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 957 hammer2_tid_t iparent; /* 0070 nominal parent inum */ 958 hammer2_key_t name_key; /* 0078 full filename key */ 959 uint16_t name_len; /* 0080 filename length */ 960 uint8_t ncopies; /* 0082 ncopies to local media */ 961 uint8_t comp_algo; /* 0083 compression request & algo */ 962 963 /* 964 * These fields are currently only applicable to PFSROOTs. 965 * 966 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 967 * identify an instance of a PFS in the cluster because 968 * a mount may contain more than one copy of the PFS as 969 * a separate node. {pfs_clid, pfs_fsid} must be used for 970 * registration in the cluster. 971 */ 972 uint8_t target_type; /* 0084 hardlink target type */ 973 uint8_t check_algo; /* 0085 check code request & algo */ 974 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 975 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 976 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 977 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 978 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 979 980 /* 981 * Quotas and aggregate sub-tree inode and data counters. Note that 982 * quotas are not replicated downward, they are explicitly set by 983 * the sysop and in-memory structures keep track of inheritence. 984 */ 985 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 986 hammer2_key_t unusedB8; /* 00B8 subtree byte count */ 987 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 988 hammer2_key_t unusedC8; /* 00C8 subtree inode count */ 989 990 /* 991 * The last snapshot tid is tested against modify_tid to determine 992 * when a copy must be made of a data block whos check mode has been 993 * disabled (a disabled check mode allows data blocks to be updated 994 * in place instead of copy-on-write). 995 */ 996 hammer2_tid_t pfs_lsnap_tid; /* 00D0 last snapshot tid */ 997 hammer2_tid_t reservedD8; /* 00D8 (avail) */ 998 999 /* 1000 * Tracks (possibly degenerate) free areas covering all sub-tree 1001 * allocations under inode, not counting the inode itself. 1002 * 0/0 indicates empty entry. fully set-associative. 1003 * 1004 * (not yet implemented) 1005 */ 1006 uint64_t decrypt_check; /* 00E0 decryption validator */ 1007 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */ 1008 } __packed; 1009 1010 typedef struct hammer2_inode_meta hammer2_inode_meta_t; 1011 1012 struct hammer2_inode_data { 1013 hammer2_inode_meta_t meta; /* 0000-00FF */ 1014 unsigned char filename[HAMMER2_INODE_MAXNAME]; 1015 /* 0100-01FF (256 char, unterminated) */ 1016 union { /* 0200-03FF (64x8 = 512 bytes) */ 1017 struct hammer2_blockset blockset; 1018 char data[HAMMER2_EMBEDDED_BYTES]; 1019 } u; 1020 } __packed; 1021 1022 typedef struct hammer2_inode_data hammer2_inode_data_t; 1023 1024 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 1025 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 1026 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 1027 1028 #define HAMMER2_OBJTYPE_UNKNOWN 0 1029 #define HAMMER2_OBJTYPE_DIRECTORY 1 1030 #define HAMMER2_OBJTYPE_REGFILE 2 1031 #define HAMMER2_OBJTYPE_FIFO 4 1032 #define HAMMER2_OBJTYPE_CDEV 5 1033 #define HAMMER2_OBJTYPE_BDEV 6 1034 #define HAMMER2_OBJTYPE_SOFTLINK 7 1035 #define HAMMER2_OBJTYPE_UNUSED08 8 1036 #define HAMMER2_OBJTYPE_SOCKET 9 1037 #define HAMMER2_OBJTYPE_WHITEOUT 10 1038 1039 #define HAMMER2_COPYID_NONE 0 1040 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 1041 1042 #define HAMMER2_COPYID_COUNT 256 1043 1044 /* 1045 * PFS types identify the role of a PFS within a cluster. The PFS types 1046 * is stored on media and in LNK_SPAN messages and used in other places. 1047 * 1048 * The low 4 bits specify the current active type while the high 4 bits 1049 * specify the transition target if the PFS is being upgraded or downgraded, 1050 * If the upper 4 bits are not zero it may effect how a PFS is used during 1051 * the transition. 1052 * 1053 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until 1054 * at least all MASTERs have updated their pfs_nmasters field. And upgrading 1055 * a SLAVE to a MASTER cannot complete until the new prospective master has 1056 * been fully synchronized (though theoretically full synchronization is 1057 * not required if a (new) quorum of other masters are fully synchronized). 1058 * 1059 * It generally does not matter which PFS element you actually mount, you 1060 * are mounting 'the cluster'. So, for example, a network mount will mount 1061 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions. 1062 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs 1063 * must be directly mounted. 1064 */ 1065 #define HAMMER2_PFSTYPE_NONE 0x00 1066 #define HAMMER2_PFSTYPE_CACHE 0x01 1067 #define HAMMER2_PFSTYPE_UNUSED02 0x02 1068 #define HAMMER2_PFSTYPE_SLAVE 0x03 1069 #define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04 1070 #define HAMMER2_PFSTYPE_SOFT_MASTER 0x05 1071 #define HAMMER2_PFSTYPE_MASTER 0x06 1072 #define HAMMER2_PFSTYPE_UNUSED07 0x07 1073 #define HAMMER2_PFSTYPE_SUPROOT 0x08 1074 #define HAMMER2_PFSTYPE_DUMMY 0x09 1075 #define HAMMER2_PFSTYPE_MAX 16 1076 1077 #define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */ 1078 #define HAMMER2_PFSTRAN_CACHE 0x10 1079 #define HAMMER2_PFSTRAN_UNMUSED20 0x20 1080 #define HAMMER2_PFSTRAN_SLAVE 0x30 1081 #define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40 1082 #define HAMMER2_PFSTRAN_SOFT_MASTER 0x50 1083 #define HAMMER2_PFSTRAN_MASTER 0x60 1084 #define HAMMER2_PFSTRAN_UNUSED70 0x70 1085 #define HAMMER2_PFSTRAN_SUPROOT 0x80 1086 #define HAMMER2_PFSTRAN_DUMMY 0x90 1087 1088 #define HAMMER2_PFS_DEC(n) ((n) & 0x0F) 1089 #define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F) 1090 #define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4) 1091 1092 #define HAMMER2_PFSSUBTYPE_NONE 0 1093 #define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */ 1094 #define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */ 1095 1096 /* 1097 * PFS mode of operation is a bitmask. This is typically not stored 1098 * on-media, but defined here because the field may be used in dmsgs. 1099 */ 1100 #define HAMMER2_PFSMODE_QUORUM 0x01 1101 #define HAMMER2_PFSMODE_RW 0x02 1102 1103 /* 1104 * Allocation Table 1105 * 1106 */ 1107 1108 1109 /* 1110 * Flags (8 bits) - blockref, for freemap only 1111 * 1112 * Note that the minimum chunk size is 1KB so we could theoretically have 1113 * 10 bits here, but we might have some future extension that allows a 1114 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 1115 */ 1116 #define HAMMER2_AVF_SELMASK 0x03 /* select group */ 1117 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 1118 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 1119 #define HAMMER2_AVF_RESERVED10 0x10 1120 #define HAMMER2_AVF_RESERVED20 0x20 1121 #define HAMMER2_AVF_RESERVED40 0x40 1122 #define HAMMER2_AVF_RESERVED80 0x80 1123 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 1124 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 1125 1126 #define HAMMER2_AV_SELECT_A 0x00 1127 #define HAMMER2_AV_SELECT_B 0x01 1128 #define HAMMER2_AV_SELECT_C 0x02 1129 #define HAMMER2_AV_SELECT_D 0x03 1130 1131 /* 1132 * The volume header eats a 64K block. There is currently an issue where 1133 * we want to try to fit all nominal filesystem updates in a 512-byte section 1134 * but it may be a lost cause due to the need for a blockset. 1135 * 1136 * All information is stored in host byte order. The volume header's magic 1137 * number may be checked to determine the byte order. If you wish to mount 1138 * between machines w/ different endian modes you'll need filesystem code 1139 * which acts on the media data consistently (either all one way or all the 1140 * other). Our code currently does not do that. 1141 * 1142 * A read-write mount may have to recover missing allocations by doing an 1143 * incremental mirror scan looking for modifications made after alloc_tid. 1144 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 1145 * operations are usually very, very fast. 1146 * 1147 * Read-only mounts do not need to do any recovery, access to the filesystem 1148 * topology is always consistent after a crash (is always consistent, period). 1149 * However, there may be shortcutted blockref updates present from deep in 1150 * the tree which are stored in the volumeh eader and must be tracked on 1151 * the fly. 1152 * 1153 * NOTE: The copyinfo[] array contains the configuration for both the 1154 * cluster connections and any local media copies. The volume 1155 * header will be replicated for each local media copy. 1156 * 1157 * The mount command may specify multiple medias or just one and 1158 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 1159 * array on mount. 1160 * 1161 * NOTE: root_blockref points to the super-root directory, not the root 1162 * directory. The root directory will be a subdirectory under the 1163 * super-root. 1164 * 1165 * The super-root directory contains all root directories and all 1166 * snapshots (readonly or writable). It is possible to do a 1167 * null-mount of the super-root using special path constructions 1168 * relative to your mounted root. 1169 * 1170 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 1171 * a PFS, including mirroring and storage quota operations, and this is 1172 * prefered over creating discrete PFSs in the super-root. Instead 1173 * the super-root is most typically used to create writable snapshots, 1174 * alternative roots, and so forth. The super-root is also used by 1175 * the automatic snapshotting mechanism. 1176 */ 1177 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 1178 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 1179 1180 struct hammer2_volume_data { 1181 /* 1182 * sector #0 - 512 bytes 1183 */ 1184 uint64_t magic; /* 0000 Signature */ 1185 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 1186 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 1187 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 1188 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 1189 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 1190 1191 uint32_t version; /* 0030 */ 1192 uint32_t flags; /* 0034 */ 1193 uint8_t copyid; /* 0038 copyid of phys vol */ 1194 uint8_t freemap_version; /* 0039 freemap algorithm */ 1195 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 1196 uint8_t reserved003B; /* 003B */ 1197 uint32_t reserved003C; /* 003C */ 1198 1199 uuid_t fsid; /* 0040 */ 1200 uuid_t fstype; /* 0050 */ 1201 1202 /* 1203 * allocator_size is precalculated at newfs time and does not include 1204 * reserved blocks, boot, or redo areas. 1205 * 1206 * Initial non-reserved-area allocations do not use the freemap 1207 * but instead adjust alloc_iterator. Dynamic allocations take 1208 * over starting at (allocator_beg). This makes newfs_hammer2's 1209 * job a lot easier and can also serve as a testing jig. 1210 */ 1211 hammer2_off_t allocator_size; /* 0060 Total data space */ 1212 hammer2_off_t allocator_free; /* 0068 Free space */ 1213 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 1214 1215 /* 1216 * mirror_tid reflects the highest committed change for this 1217 * block device regardless of whether it is to the super-root 1218 * or to a PFS or whatever. 1219 * 1220 * freemap_tid reflects the highest committed freemap change for 1221 * this block device. 1222 */ 1223 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 1224 hammer2_tid_t reserved0080; /* 0080 */ 1225 hammer2_tid_t reserved0088; /* 0088 */ 1226 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 1227 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 1228 hammer2_tid_t reserved00A0[5]; /* 00A0-00C7 */ 1229 1230 /* 1231 * Copyids are allocated dynamically from the copyexists bitmap. 1232 * An id from the active copies set (up to 8, see copyinfo later on) 1233 * may still exist after the copy set has been removed from the 1234 * volume header and its bit will remain active in the bitmap and 1235 * cannot be reused until it is 100% removed from the hierarchy. 1236 */ 1237 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1238 char reserved0140[248]; /* 00E8-01DF */ 1239 1240 /* 1241 * 32 bit CRC array at the end of the first 512 byte sector. 1242 * 1243 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1244 * the other icrc's except this one). 1245 * 1246 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1247 * the blockset for the root. 1248 * 1249 * icrc_sects[5] - Sector 2 1250 * icrc_sects[4] - Sector 3 1251 * icrc_sects[3] - Sector 4 (the freemap blockset) 1252 */ 1253 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1254 1255 /* 1256 * sector #1 - 512 bytes 1257 * 1258 * The entire sector is used by a blockset. 1259 */ 1260 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1261 1262 /* 1263 * sector #2-7 1264 */ 1265 char sector2[512]; /* 0400-05FF reserved */ 1266 char sector3[512]; /* 0600-07FF reserved */ 1267 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1268 char sector5[512]; /* 0A00-0BFF reserved */ 1269 char sector6[512]; /* 0C00-0DFF reserved */ 1270 char sector7[512]; /* 0E00-0FFF reserved */ 1271 1272 /* 1273 * sector #8-71 - 32768 bytes 1274 * 1275 * Contains the configuration for up to 256 copyinfo targets. These 1276 * specify local and remote copies operating as masters or slaves. 1277 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1278 * indicates the local media). 1279 * 1280 * Each inode contains a set of up to 8 copyids, either inherited 1281 * from its parent or explicitly specified in the inode, which 1282 * indexes into this array. 1283 */ 1284 /* 1000-8FFF copyinfo config */ 1285 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1286 1287 /* 1288 * Remaining sections are reserved for future use. 1289 */ 1290 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1291 1292 /* 1293 * icrc on entire volume header 1294 */ 1295 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1296 } __packed; 1297 1298 typedef struct hammer2_volume_data hammer2_volume_data_t; 1299 1300 /* 1301 * Various parts of the volume header have their own iCRCs. 1302 * 1303 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1304 * and not included the icrc calculation. 1305 * 1306 * The second 512 bytes also has its own iCRC but it is stored in the first 1307 * 512 bytes so it covers the entire second 512 bytes. 1308 * 1309 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1310 * which is where the iCRC for the whole volume is stored. This is currently 1311 * a catch-all for anything not individually iCRCd. 1312 */ 1313 #define HAMMER2_VOL_ICRC_SECT0 7 1314 #define HAMMER2_VOL_ICRC_SECT1 6 1315 1316 #define HAMMER2_VOLUME_BYTES 65536 1317 1318 #define HAMMER2_VOLUME_ICRC0_OFF 0 1319 #define HAMMER2_VOLUME_ICRC1_OFF 512 1320 #define HAMMER2_VOLUME_ICRCVH_OFF 0 1321 1322 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1323 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 1324 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1325 1326 #define HAMMER2_VOL_VERSION_MIN 1 1327 #define HAMMER2_VOL_VERSION_DEFAULT 1 1328 #define HAMMER2_VOL_VERSION_WIP 2 1329 1330 #define HAMMER2_NUM_VOLHDRS 4 1331 1332 union hammer2_media_data { 1333 hammer2_volume_data_t voldata; 1334 hammer2_inode_data_t ipdata; 1335 hammer2_blockset_t blkset; 1336 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1337 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1338 char buf[HAMMER2_PBUFSIZE]; 1339 } __packed; 1340 1341 typedef union hammer2_media_data hammer2_media_data_t; 1342 1343 #endif /* !_VFS_HAMMER2_DISK_H_ */ 1344