1/* 2 * Copyright (c) 2007 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <[email protected]> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35#ifndef VFS_HAMMER_DISK_H_ 36#define VFS_HAMMER_DISK_H_ 37 38#include <sys/endian.h> 39#include <sys/types.h> 40 41#ifndef _SYS_UUID_H_ 42#include <sys/uuid.h> 43#endif 44 45/* 46 * The structures below represent the on-disk format for a HAMMER 47 * filesystem. Note that all fields for on-disk structures are naturally 48 * aligned. HAMMER uses little endian for fields in on-disk structures. 49 * HAMMER doesn't support big endian arch, but is planned. 50 * 51 * Most of HAMMER revolves around the concept of an object identifier. An 52 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object 53 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups 54 * and mirrors to retain varying amounts of filesystem history by removing 55 * any possibility of conflict through identifier reuse. 56 * 57 * A HAMMER filesystem may span multiple volumes. 58 * 59 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem 60 * I/O is done in multiples of 16K. 61 * 62 * 64K X-bufs are used for blocks >= a file's 1MB mark. 63 * 64 * Per-volume storage limit: 52 bits 4096 TB 65 * Per-Zone storage limit: 60 bits 1 MTB 66 * Per-filesystem storage limit: 60 bits 1 MTB 67 */ 68#define HAMMER_BUFSIZE 16384 69#define HAMMER_XBUFSIZE 65536 70#define HAMMER_HBUFSIZE (HAMMER_BUFSIZE / 2) 71#define HAMMER_XDEMARC (1024 * 1024) 72#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1) 73#define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1) 74 75#define HAMMER_BUFSIZE64 ((uint64_t)HAMMER_BUFSIZE) 76#define HAMMER_BUFMASK64 ((uint64_t)HAMMER_BUFMASK) 77 78#define HAMMER_XBUFSIZE64 ((uint64_t)HAMMER_XBUFSIZE) 79#define HAMMER_XBUFMASK64 ((uint64_t)HAMMER_XBUFMASK) 80 81#define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */ 82#define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */ 83#define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */ 84#define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */ 85 86#define HAMMER_OFF_BAD ((hammer_off_t)-1) 87 88#define HAMMER_BUFSIZE_DOALIGN(offset) \ 89 (((offset) + HAMMER_BUFMASK) & ~HAMMER_BUFMASK) 90#define HAMMER_BUFSIZE64_DOALIGN(offset) \ 91 (((offset) + HAMMER_BUFMASK64) & ~HAMMER_BUFMASK64) 92 93#define HAMMER_XBUFSIZE_DOALIGN(offset) \ 94 (((offset) + HAMMER_XBUFMASK) & ~HAMMER_XBUFMASK) 95#define HAMMER_XBUFSIZE64_DOALIGN(offset) \ 96 (((offset) + HAMMER_XBUFMASK64) & ~HAMMER_XBUFMASK64) 97 98/* 99 * The current limit of volumes that can make up a HAMMER FS 100 */ 101#define HAMMER_MAX_VOLUMES 256 102 103/* 104 * Reserved space for (future) header junk after the volume header. 105 */ 106#define HAMMER_MIN_VOL_JUNK (HAMMER_BUFSIZE * 16) /* 256 KB */ 107#define HAMMER_MAX_VOL_JUNK HAMMER_MIN_VOL_JUNK 108#define HAMMER_VOL_JUNK_SIZE HAMMER_MIN_VOL_JUNK 109 110/* 111 * Hammer transaction ids are 64 bit unsigned integers and are usually 112 * synchronized with the time of day in nanoseconds. 113 * 114 * Hammer offsets are used for FIFO indexing and embed a cycle counter 115 * and volume number in addition to the offset. Most offsets are required 116 * to be 16 KB aligned. 117 */ 118typedef uint64_t hammer_tid_t; 119typedef uint64_t hammer_off_t; 120typedef uint32_t hammer_crc_t; 121typedef uuid_t hammer_uuid_t; 122 123#define HAMMER_MIN_TID 0ULL /* unsigned */ 124#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 125#define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */ 126#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */ 127#define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */ 128#define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */ 129#define HAMMER_MIN_RECTYPE 0x0U /* unsigned */ 130#define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */ 131#define HAMMER_MIN_OFFSET 0ULL /* unsigned */ 132#define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 133 134/* 135 * hammer_off_t has several different encodings. Note that not all zones 136 * encode a vol_no. Zone bits are not a part of filesystem capacity as 137 * the zone bits aren't directly or indirectly mapped to physical volumes. 138 * 139 * In other words, HAMMER's logical filesystem offset consists of 64 bits, 140 * but the filesystem is considered 60 bits filesystem, not 64 bits. 141 * The maximum filesystem capacity is 1EB, not 16EB. 142 * 143 * zone 0: available, a big-block that contains the offset is unused 144 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header) 145 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer) 146 * zone 3 (z,o): undo/redo fifo - fixed zone-2 offset array in volume header 147 * zone 4 (z,v,o): freemap - only real blockmap 148 * zone 8 (z,v,o): B-Tree - actually zone-2 address 149 * zone 9 (z,v,o): meta - actually zone-2 address 150 * zone 10 (z,v,o): large-data - actually zone-2 address 151 * zone 11 (z,v,o): small-data - actually zone-2 address 152 * zone 15: unavailable, usually the offset is beyond volume size 153 * 154 * layer1/layer2 direct map: 155 * Maximum HAMMER filesystem capacity from volume aspect 156 * 2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB (long offset) 157 * <-------------------------------------------------------------> 158 * 8bits 52bits (short offset) 159 * <------><-----------------------------------------------------> 160 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo 161 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo 162 * <-----------------><------------------><----------------------> 163 * 18bits 19bits 23bits 164 * <-------------------------------------------------------------> 165 * 2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB 166 * Maximum HAMMER filesystem capacity from blockmap aspect 167 * 168 * volume#0 layout 169 * +-------------------------> offset 0 of a device/partition 170 * | volume header (1928 bytes) 171 * | the rest of header junk space (HAMMER_BUFSIZE aligned) 172 * +-------------------------> vol_bot_beg 173 * | boot area (HAMMER_BUFSIZE aligned) 174 * +-------------------------> vol_mem_beg 175 * | memory log (HAMMER_BUFSIZE aligned) 176 * +-------------------------> vol_buf_beg (physical offset of zone-2) 177 * | zone-4 big-block for layer1 178 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE 179 * | zone-4 big-blocks for layer2 180 * | ... (1 big-block per 4TB space) 181 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ... 182 * | zone-3 big-blocks for UNDO/REDO FIFO 183 * | ... (max 128 big-blocks) 184 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ... 185 * | zone-8 big-block for root B-Tree node/etc 186 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ... 187 * | zone-9 big-block for root inode/PFS/etc 188 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ... 189 * | zone-X big-blocks 190 * | ... (big-blocks for new zones after newfs_hammer) 191 * | ... 192 * | ... 193 * | ... 194 * | ... 195 * +-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned) 196 * +-------------------------> end of a device/partition 197 * 198 * volume#N layout (0<N<256) 199 * +-------------------------> offset 0 of a device/partition 200 * | volume header (1928 bytes) 201 * | the rest of header junk space (HAMMER_BUFSIZE aligned) 202 * +-------------------------> vol_bot_beg 203 * | boot area (HAMMER_BUFSIZE aligned) 204 * +-------------------------> vol_mem_beg 205 * | memory log (HAMMER_BUFSIZE aligned) 206 * +-------------------------> vol_buf_beg (physical offset of zone-2) 207 * | zone-4 big-blocks for layer2 208 * | ... (1 big-block per 4TB space) 209 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ... 210 * | zone-X big-blocks 211 * | ... (unused until volume#(N-1) runs out of space) 212 * | ... 213 * | ... 214 * | ... 215 * | ... 216 * +-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned) 217 * +-------------------------> end of a device/partition 218 */ 219 220#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL 221#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL 222#define HAMMER_ZONE_UNDO 0x3000000000000000ULL 223#define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL 224#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL /* not used */ 225#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL /* not used */ 226#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL /* not used */ 227#define HAMMER_ZONE_BTREE 0x8000000000000000ULL 228#define HAMMER_ZONE_META 0x9000000000000000ULL 229#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL 230#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL 231#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL /* not used */ 232#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL /* not used */ 233#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL /* not used */ 234#define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL 235 236#define HAMMER_ZONE_RAW_VOLUME_INDEX 1 237#define HAMMER_ZONE_RAW_BUFFER_INDEX 2 238#define HAMMER_ZONE_UNDO_INDEX 3 239#define HAMMER_ZONE_FREEMAP_INDEX 4 240#define HAMMER_ZONE_BTREE_INDEX 8 241#define HAMMER_ZONE_META_INDEX 9 242#define HAMMER_ZONE_LARGE_DATA_INDEX 10 243#define HAMMER_ZONE_SMALL_DATA_INDEX 11 244#define HAMMER_ZONE_UNAVAIL_INDEX 15 245 246#define HAMMER_MAX_ZONES 16 247 248#define HAMMER_ZONE(offset) ((offset) & HAMMER_OFF_ZONE_MASK) 249 250#define hammer_is_zone_raw_volume(offset) \ 251 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME) 252#define hammer_is_zone_raw_buffer(offset) \ 253 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER) 254#define hammer_is_zone_undo(offset) \ 255 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO) 256#define hammer_is_zone_freemap(offset) \ 257 (HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP) 258#define hammer_is_zone_btree(offset) \ 259 (HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE) 260#define hammer_is_zone_meta(offset) \ 261 (HAMMER_ZONE(offset) == HAMMER_ZONE_META) 262#define hammer_is_zone_large_data(offset) \ 263 (HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA) 264#define hammer_is_zone_small_data(offset) \ 265 (HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA) 266#define hammer_is_zone_unavail(offset) \ 267 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL) 268#define hammer_is_zone_data(offset) \ 269 (hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset)) 270 271#define hammer_is_index_record(zone) \ 272 ((zone) >= HAMMER_ZONE_BTREE_INDEX && \ 273 (zone) < HAMMER_MAX_ZONES) 274 275#define hammer_is_zone_record(offset) \ 276 hammer_is_index_record(HAMMER_ZONE_DECODE(offset)) 277 278#define hammer_is_index_direct_xlated(zone) \ 279 (((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) || \ 280 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) || \ 281 hammer_is_index_record(zone)) 282 283#define hammer_is_zone_direct_xlated(offset) \ 284 hammer_is_index_direct_xlated(HAMMER_ZONE_DECODE(offset)) 285 286#define HAMMER_ZONE_ENCODE(zone, ham_off) \ 287 (((hammer_off_t)(zone) << 60) | (ham_off)) 288#define HAMMER_ZONE_DECODE(ham_off) \ 289 ((int)(((hammer_off_t)(ham_off) >> 60))) 290 291#define HAMMER_VOL_ENCODE(vol_no) \ 292 ((hammer_off_t)((vol_no) & 255) << 52) 293#define HAMMER_VOL_DECODE(ham_off) \ 294 ((int)(((hammer_off_t)(ham_off) >> 52) & 255)) 295 296#define HAMMER_OFF_SHORT_ENCODE(offset) \ 297 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK) 298#define HAMMER_OFF_LONG_ENCODE(offset) \ 299 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK) 300 301#define HAMMER_ENCODE(zone, vol_no, offset) \ 302 (((hammer_off_t)(zone) << 60) | \ 303 HAMMER_VOL_ENCODE(vol_no) | \ 304 HAMMER_OFF_SHORT_ENCODE(offset)) 305#define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \ 306 HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset) 307#define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \ 308 HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset) 309#define HAMMER_ENCODE_UNDO(offset) \ 310 HAMMER_ENCODE(HAMMER_ZONE_UNDO_INDEX, HAMMER_ROOT_VOLNO, offset) 311#define HAMMER_ENCODE_FREEMAP(vol_no, offset) \ 312 HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset) 313 314/* 315 * Translate a zone address to zone-X address. 316 */ 317#define hammer_xlate_to_zoneX(zone, offset) \ 318 HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK) 319#define hammer_xlate_to_zone2(offset) \ 320 hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset)) 321 322#define hammer_data_zone(data_len) \ 323 (((data_len) >= HAMMER_BUFSIZE) ? \ 324 HAMMER_ZONE_LARGE_DATA : \ 325 HAMMER_ZONE_SMALL_DATA) 326#define hammer_data_zone_index(data_len) \ 327 (((data_len) >= HAMMER_BUFSIZE) ? \ 328 HAMMER_ZONE_LARGE_DATA_INDEX : \ 329 HAMMER_ZONE_SMALL_DATA_INDEX) 330 331/* 332 * Big-Block backing store 333 * 334 * A blockmap is a two-level map which translates a blockmap-backed zone 335 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the 336 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two 337 * layers gives us 18+19+23 = 60 bits of address space. 338 * 339 * When using hinting for a blockmap lookup, the hint is lost when the 340 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's. 341 * HINTBLOCK is a heuristic. 342 */ 343#define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4) 344#define HAMMER_HINTBLOCK_MASK64 ((uint64_t)HAMMER_HINTBLOCK_SIZE - 1) 345#define HAMMER_BIGBLOCK_SIZE (8192 * 1024) 346#define HAMMER_BIGBLOCK_SIZE64 ((uint64_t)HAMMER_BIGBLOCK_SIZE) 347#define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1) 348#define HAMMER_BIGBLOCK_MASK64 ((uint64_t)HAMMER_BIGBLOCK_SIZE - 1) 349#define HAMMER_BIGBLOCK_BITS 23 350#if 0 351#define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024) 352#endif 353#if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE 354#error "HAMMER_BIGBLOCK_BITS BROKEN" 355#endif 356 357#define HAMMER_BUFFERS_PER_BIGBLOCK \ 358 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE) 359#define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \ 360 (HAMMER_BUFFERS_PER_BIGBLOCK - 1) 361#define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \ 362 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK) 363 364#define HAMMER_BIGBLOCK_DOALIGN(offset) \ 365 (((offset) + HAMMER_BIGBLOCK_MASK64) & ~HAMMER_BIGBLOCK_MASK64) 366 367/* 368 * Maximum number of mirrors operating in master mode (multi-master 369 * clustering and mirroring). Note that HAMMER1 does not support 370 * multi-master clustering as of 2015. 371 */ 372#define HAMMER_MAX_MASTERS 16 373 374/* 375 * The blockmap is somewhat of a degenerate structure. HAMMER only actually 376 * uses it in its original incarnation to implement the freemap. 377 * 378 * zone:1 raw volume (no blockmap) 379 * zone:2 raw buffer (no blockmap) 380 * zone:3 undomap (direct layer2 array in volume header) 381 * zone:4 freemap (the only real blockmap) 382 * zone:8-15 zone id used to classify big-block only, address is actually 383 * a zone-2 address. 384 */ 385typedef struct hammer_blockmap { 386 hammer_off_t phys_offset; /* zone-2 offset only used by zone-4 */ 387 hammer_off_t first_offset; /* zone-X offset only used by zone-3 */ 388 hammer_off_t next_offset; /* zone-X offset for allocation */ 389 hammer_off_t alloc_offset; /* zone-X offset only used by zone-3 */ 390 uint32_t reserved01; 391 hammer_crc_t entry_crc; 392} *hammer_blockmap_t; 393 394#define HAMMER_BLOCKMAP_CRCSIZE \ 395 offsetof(struct hammer_blockmap, entry_crc) 396 397/* 398 * The blockmap is a 2-layer entity made up of big-blocks. The first layer 399 * contains 262144 32-byte entries (18 bits), the second layer contains 400 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps. 401 * 18+19+23 = 60 bits. The top four bits are the zone id. 402 * 403 * Currently only the freemap utilizes both layers in all their glory. 404 * All primary data/meta-data zones actually encode a zone-2 address 405 * requiring no real blockmap translation. 406 * 407 * The freemap uses the upper 8 bits of layer-1 to identify the volume, 408 * thus any space allocated via the freemap can be directly translated 409 * to a zone:2 (or zone:8-15) address. 410 * 411 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23] 412 */ 413 414/* 415 * 32 bytes layer1 entry for 8MB big-block. 416 * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries, 417 * which equals bits assigned for layer1 in zone-2 address. 418 */ 419typedef struct hammer_blockmap_layer1 { 420 hammer_off_t blocks_free; /* big-blocks free */ 421 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */ 422 hammer_off_t reserved01; 423 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */ 424 /* (not yet used) */ 425 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/ 426} *hammer_blockmap_layer1_t; 427 428#define HAMMER_LAYER1_CRCSIZE \ 429 offsetof(struct hammer_blockmap_layer1, layer1_crc) 430 431/* 432 * 16 bytes layer2 entry for 8MB big-blocks. 433 * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries, 434 * which equals bits assigned for layer2 in zone-2 address. 435 * 436 * NOTE: bytes_free is signed and can legally go negative if/when data 437 * de-dup occurs. This field will never go higher than 438 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE 439 * the big-block is completely free. 440 */ 441typedef struct hammer_blockmap_layer2 { 442 uint8_t zone; /* typed allocation zone */ 443 uint8_t reserved01; 444 uint16_t reserved02; 445 uint32_t append_off; /* allocatable space index */ 446 int32_t bytes_free; /* bytes free within this big-block */ 447 hammer_crc_t entry_crc; 448} *hammer_blockmap_layer2_t; 449 450#define HAMMER_LAYER2_CRCSIZE \ 451 offsetof(struct hammer_blockmap_layer2, entry_crc) 452 453#define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL) 454 455#define HAMMER_BLOCKMAP_RADIX1 /* 2^18 = 262144 */ \ 456 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))) 457#define HAMMER_BLOCKMAP_RADIX2 /* 2^19 = 524288 */ \ 458 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))) 459 460#define HAMMER_BLOCKMAP_LAYER1 /* 2^(18+19+23) = 1EB */ \ 461 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2) 462#define HAMMER_BLOCKMAP_LAYER2 /* 2^(19+23) = 4TB */ \ 463 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64) 464 465#define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1) 466#define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1) 467 468#define HAMMER_BLOCKMAP_LAYER2_DOALIGN(offset) \ 469 (((offset) + HAMMER_BLOCKMAP_LAYER2_MASK) & \ 470 ~HAMMER_BLOCKMAP_LAYER2_MASK) 471 472/* 473 * Index within layer1 or layer2 big-block for the entry representing 474 * a zone-2 physical offset. 475 */ 476#define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) \ 477 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \ 478 HAMMER_BLOCKMAP_LAYER2)) 479 480#define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) \ 481 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \ 482 HAMMER_BIGBLOCK_SIZE64)) 483 484/* 485 * Byte offset within layer1 or layer2 big-block for the entry representing 486 * a zone-2 physical offset. Multiply the index by sizeof(blockmap_layer). 487 */ 488#define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \ 489 (HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) * \ 490 sizeof(struct hammer_blockmap_layer1)) 491 492#define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \ 493 (HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) * \ 494 sizeof(struct hammer_blockmap_layer2)) 495 496/* 497 * Move on to offset 0 of the next layer1 or layer2. 498 */ 499#define HAMMER_ZONE_LAYER1_NEXT_OFFSET(offset) \ 500 (((offset) + HAMMER_BLOCKMAP_LAYER2) & ~HAMMER_BLOCKMAP_LAYER2_MASK) 501 502#define HAMMER_ZONE_LAYER2_NEXT_OFFSET(offset) \ 503 (((offset) + HAMMER_BIGBLOCK_SIZE) & ~HAMMER_BIGBLOCK_MASK64) 504 505/* 506 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume 507 * header with an array of zone-2 offsets. A maximum of (128x8MB) = 1GB, 508 * and minimum of (64x8MB) = 512MB may be reserved. The size of the undo 509 * fifo is usually set a newfs time. 510 */ 511#define HAMMER_MIN_UNDO_BIGBLOCKS 64 512#define HAMMER_MAX_UNDO_BIGBLOCKS 128 513 514/* 515 * All on-disk HAMMER structures which make up elements of the UNDO FIFO 516 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure 517 * contains all the information required to validate the fifo element 518 * and to scan the fifo in either direction. The head is typically embedded 519 * in higher level hammer on-disk structures while the tail is typically 520 * out-of-band. hdr_size is the size of the whole mess, including the tail. 521 * 522 * All undo structures are guaranteed to not cross a 16K filesystem 523 * buffer boundary. Most undo structures are fairly small. Data spaces 524 * are not immediately reused by HAMMER so file data is not usually recorded 525 * as part of an UNDO. 526 * 527 * PAD elements are allowed to take up only 8 bytes of space as a special 528 * case, containing only hdr_signature, hdr_type, and hdr_size fields, 529 * and with the tail overloaded onto the head structure for 8 bytes total. 530 * 531 * Every undo record has a sequence number. This number is unrelated to 532 * transaction ids and instead collects the undo transactions associated 533 * with a single atomic operation. A larger transactional operation, such 534 * as a remove(), may consist of several smaller atomic operations 535 * representing raw meta-data operations. 536 * 537 * HAMMER VERSION 4 CHANGES 538 * 539 * In HAMMER version 4 the undo structure alignment is reduced from 16384 540 * to 512 bytes in order to ensure that each 512 byte sector begins with 541 * a header. The hdr_seq field in the header is a 32 bit sequence number 542 * which allows the recovery code to detect missing sectors 543 * without relying on the 32-bit crc and to definitively identify the current 544 * undo sequence space without having to rely on information from the volume 545 * header. In addition, new REDO entries in the undo space are used to 546 * record write, write/extend, and transaction id updates. 547 * 548 * The grand result is: 549 * 550 * (1) The volume header no longer needs to be synchronized for most 551 * flush and fsync operations. 552 * 553 * (2) Most fsync operations need only lay down REDO records 554 * 555 * (3) Data overwrite for nohistory operations covered by REDO records 556 * can be supported (instead of rolling a new block allocation), 557 * by rolling UNDO for the prior contents of the data. 558 * 559 * HAMMER VERSION 5 CHANGES 560 * 561 * Hammer version 5 contains a minor adjustment making layer2's bytes_free 562 * field signed, allowing dedup to push it into the negative domain. 563 */ 564#define HAMMER_HEAD_ALIGN 8 565#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1) 566#define HAMMER_HEAD_DOALIGN(bytes) \ 567 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK) 568 569#define HAMMER_UNDO_ALIGN 512 570#define HAMMER_UNDO_ALIGN64 ((uint64_t)512) 571#define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1) 572#define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1) 573#define HAMMER_UNDO_DOALIGN(offset) \ 574 (((offset) + HAMMER_UNDO_MASK) & ~HAMMER_UNDO_MASK64) 575 576typedef struct hammer_fifo_head { 577 uint16_t hdr_signature; 578 uint16_t hdr_type; 579 uint32_t hdr_size; /* Aligned size of the whole mess */ 580 uint32_t hdr_seq; /* Sequence number */ 581 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */ 582} *hammer_fifo_head_t; 583 584#define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc) 585 586typedef struct hammer_fifo_tail { 587 uint16_t tail_signature; 588 uint16_t tail_type; 589 uint32_t tail_size; /* aligned size of the whole mess */ 590} *hammer_fifo_tail_t; 591 592/* 593 * Fifo header types. 594 * 595 * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER 596 * version ever gets bumped again. It exists only to keep compatibility with 597 * older versions. 598 */ 599#define HAMMER_HEAD_TYPE_PAD (0x0040U | 0x8000U) 600#define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */ 601#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */ 602#define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */ 603 604#define HAMMER_HEAD_SIGNATURE 0xC84EU 605#define HAMMER_TAIL_SIGNATURE 0xC74FU 606 607/* 608 * Misc FIFO structures. 609 * 610 * UNDO - Raw meta-data media updates. 611 */ 612typedef struct hammer_fifo_undo { 613 struct hammer_fifo_head head; 614 hammer_off_t undo_offset; /* zone-1,2 offset */ 615 int32_t undo_data_bytes; 616 int32_t undo_reserved01; 617 /* followed by data */ 618} *hammer_fifo_undo_t; 619 620/* 621 * REDO (HAMMER version 4+) - Logical file writes/truncates. 622 * 623 * REDOs contain information which will be duplicated in a later meta-data 624 * update, allowing fast write()+fsync() operations. REDOs can be ignored 625 * without harming filesystem integrity but must be processed if fsync() 626 * semantics are desired. 627 * 628 * Unlike UNDOs which are processed backwards within the recovery span, 629 * REDOs must be processed forwards starting further back (starting outside 630 * the recovery span). 631 * 632 * WRITE - Write logical file (with payload). Executed both 633 * out-of-span and in-span. Out-of-span WRITEs may be 634 * filtered out by TERMs. 635 * 636 * TRUNC - Truncate logical file (no payload). Executed both 637 * out-of-span and in-span. Out-of-span WRITEs may be 638 * filtered out by TERMs. 639 * 640 * TERM_* - Indicates meta-data was committed (if out-of-span) or 641 * will be rolled-back (in-span). Any out-of-span TERMs 642 * matching earlier WRITEs remove those WRITEs from 643 * consideration as they might conflict with a later data 644 * commit (which is not being rolled-back). 645 * 646 * SYNC - The earliest in-span SYNC (the last one when scanning 647 * backwards) tells the recovery code how far out-of-span 648 * it must go to run REDOs. 649 * 650 * NOTE: WRITEs do not always have matching TERMs even under 651 * perfect conditions because truncations might remove the 652 * buffers from consideration. I/O problems can also remove 653 * buffers from consideration. 654 * 655 * TRUNCSs do not always have matching TERMs because several 656 * truncations may be aggregated together into a single TERM. 657 */ 658typedef struct hammer_fifo_redo { 659 struct hammer_fifo_head head; 660 int64_t redo_objid; /* file being written */ 661 hammer_off_t redo_offset; /* logical offset in file */ 662 int32_t redo_data_bytes; 663 uint32_t redo_flags; 664 uint32_t redo_localization; 665 uint32_t redo_reserved01; 666 uint64_t redo_reserved02; 667 /* followed by data */ 668} *hammer_fifo_redo_t; 669 670#define HAMMER_REDO_WRITE 0x00000001 671#define HAMMER_REDO_TRUNC 0x00000002 672#define HAMMER_REDO_TERM_WRITE 0x00000004 673#define HAMMER_REDO_TERM_TRUNC 0x00000008 674#define HAMMER_REDO_SYNC 0x00000010 675 676typedef union hammer_fifo_any { 677 struct hammer_fifo_head head; 678 struct hammer_fifo_undo undo; 679 struct hammer_fifo_redo redo; 680} *hammer_fifo_any_t; 681 682/* 683 * Volume header types 684 */ 685#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */ 686#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */ 687 688/* 689 * HAMMER Volume header 690 * 691 * A HAMMER filesystem can be built from 1-256 block devices, each block 692 * device contains a volume header followed by however many buffers fit 693 * into the volume. 694 * 695 * One of the volumes making up a HAMMER filesystem is the root volume. 696 * The root volume is always volume #0 which is the first block device path 697 * specified by newfs_hammer(8). All HAMMER volumes have a volume header, 698 * however the root volume may be the only volume that has valid values for 699 * some fields in the header. 700 * 701 * Special field notes: 702 * 703 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes) 704 * vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes) 705 * vol_buf_beg - offset of the first buffer in volume 706 * vol_buf_end - offset of volume EOF (on buffer boundary) 707 * 708 * The memory log area allows a kernel to cache new records and data 709 * in memory without allocating space in the actual filesystem to hold 710 * the records and data. In the event that a filesystem becomes full, 711 * any records remaining in memory can be flushed to the memory log 712 * area. This allows the kernel to immediately return success. 713 * 714 * The buffer offset is a physical offset of zone-2 offset. The lower 715 * 52 bits of the zone-2 offset is added to the buffer offset of each 716 * volume to generate an actual I/O offset within the block device. 717 * 718 * NOTE: boot area and memory log are currently not used. 719 */ 720 721/* 722 * Filesystem type string 723 */ 724#define HAMMER_FSTYPE_STRING "DragonFly HAMMER" 725 726/* 727 * These macros are only used by userspace when userspace commands either 728 * initialize or add a new HAMMER volume. 729 */ 730#define HAMMER_BOOT_MINBYTES (32*1024) 731#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024) 732#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024) 733 734#define HAMMER_MEM_MINBYTES (256*1024) 735#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024) 736#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024) 737 738typedef struct hammer_volume_ondisk { 739 uint64_t vol_signature; /* HAMMER_FSBUF_VOLUME for a valid header */ 740 741 /* 742 * These are relative to block device offset, not zone offsets. 743 */ 744 int64_t vol_bot_beg; /* offset of boot area */ 745 int64_t vol_mem_beg; /* offset of memory log */ 746 int64_t vol_buf_beg; /* offset of the first buffer in volume */ 747 int64_t vol_buf_end; /* offset of volume EOF (on buffer boundary) */ 748 int64_t vol_reserved01; 749 750 hammer_uuid_t vol_fsid; /* identify filesystem */ 751 hammer_uuid_t vol_fstype; /* identify filesystem type */ 752 char vol_label[64]; /* filesystem label */ 753 754 int32_t vol_no; /* volume number within filesystem */ 755 int32_t vol_count; /* number of volumes making up filesystem */ 756 757 uint32_t vol_version; /* version control information */ 758 hammer_crc_t vol_crc; /* header crc */ 759 uint32_t vol_flags; /* volume flags */ 760 uint32_t vol_rootvol; /* the root volume number (must be 0) */ 761 762 uint32_t vol_reserved[8]; 763 764 /* 765 * These fields are initialized and space is reserved in every 766 * volume making up a HAMMER filesytem, but only the root volume 767 * contains valid data. Note that vol0_stat_bigblocks does not 768 * include big-blocks for freemap and undomap initially allocated 769 * by newfs_hammer(8). 770 */ 771 int64_t vol0_stat_bigblocks; /* total big-blocks when fs is empty */ 772 int64_t vol0_stat_freebigblocks;/* number of free big-blocks */ 773 int64_t vol0_reserved01; 774 int64_t vol0_stat_inodes; /* for statfs only */ 775 int64_t vol0_reserved02; 776 hammer_off_t vol0_btree_root; /* B-Tree root offset in zone-8 */ 777 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */ 778 hammer_off_t vol0_reserved03; 779 780 /* 781 * Blockmaps for zones. Not all zones use a blockmap. Note that 782 * the entire root blockmap is cached in the hammer_mount structure. 783 */ 784 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES]; 785 786 /* 787 * Array of zone-2 addresses for undo FIFO. 788 */ 789 hammer_off_t vol0_undo_array[HAMMER_MAX_UNDO_BIGBLOCKS]; 790} *hammer_volume_ondisk_t; 791 792#define HAMMER_ROOT_VOLNO 0 793 794#define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */ 795 796#define HAMMER_VOL_CRCSIZE1 \ 797 offsetof(struct hammer_volume_ondisk, vol_crc) 798#define HAMMER_VOL_CRCSIZE2 \ 799 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \ 800 sizeof(hammer_crc_t)) 801 802#define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */ 803#define HAMMER_VOL_VERSION_DEFAULT 7 /* newfs default version */ 804#define HAMMER_VOL_VERSION_WIP 8 /* version >= this is WIP */ 805#define HAMMER_VOL_VERSION_MAX 7 /* maximum supported version */ 806 807#define HAMMER_VOL_VERSION_ONE 1 808#define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */ 809#define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */ 810#define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */ 811#define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */ 812#define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */ 813#define HAMMER_VOL_VERSION_SEVEN 7 /* use the faster iscsi_crc */ 814 815/* 816 * Translate a zone-2 address to physical address 817 */ 818#define hammer_xlate_to_phys(volume, zone2_offset) \ 819 ((volume)->vol_buf_beg + HAMMER_OFF_SHORT_ENCODE(zone2_offset)) 820 821/* 822 * Translate a zone-3 address to zone-2 address 823 */ 824#define HAMMER_UNDO_INDEX(zone3_offset) \ 825 (HAMMER_OFF_SHORT_ENCODE(zone3_offset) / HAMMER_BIGBLOCK_SIZE) 826 827#define hammer_xlate_to_undo(volume, zone3_offset) \ 828 ((volume)->vol0_undo_array[HAMMER_UNDO_INDEX(zone3_offset)] + \ 829 (zone3_offset & HAMMER_BIGBLOCK_MASK64)) 830 831/* 832 * Effective per-volume filesystem capacity including big-blocks for layer1/2 833 */ 834#define HAMMER_VOL_BUF_SIZE(volume) \ 835 ((volume)->vol_buf_end - (volume)->vol_buf_beg) 836 837/* 838 * Record types are fairly straightforward. The B-Tree includes the record 839 * type in its index sort. 840 */ 841#define HAMMER_RECTYPE_UNKNOWN 0x0000 842#define HAMMER_RECTYPE_INODE 0x0001 /* inode in obj_id space */ 843#define HAMMER_RECTYPE_DATA 0x0010 844#define HAMMER_RECTYPE_DIRENTRY 0x0011 845#define HAMMER_RECTYPE_DB 0x0012 846#define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */ 847#define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */ 848#define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */ 849#define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */ 850#define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */ 851#define HAMMER_RECTYPE_MAX 0xFFFF 852 853#define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1) 854#define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT 855 856#define HAMMER_FIXKEY_SYMLINK 1 857 858#define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */ 859#define HAMMER_OBJTYPE_DIRECTORY 1 860#define HAMMER_OBJTYPE_REGFILE 2 861#define HAMMER_OBJTYPE_DBFILE 3 862#define HAMMER_OBJTYPE_FIFO 4 863#define HAMMER_OBJTYPE_CDEV 5 864#define HAMMER_OBJTYPE_BDEV 6 865#define HAMMER_OBJTYPE_SOFTLINK 7 866#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */ 867#define HAMMER_OBJTYPE_SOCKET 9 868 869/* 870 * HAMMER inode attribute data 871 * 872 * The data reference for a HAMMER inode points to this structure. Any 873 * modifications to the contents of this structure will result in a 874 * replacement operation. 875 * 876 * parent_obj_id is only valid for directories (which cannot be hard-linked), 877 * and specifies the parent directory obj_id. This field will also be set 878 * for non-directory inodes as a recovery aid, but can wind up holding 879 * stale information. However, since object id's are not reused, the worse 880 * that happens is that the recovery code is unable to use it. 881 * A parent_obj_id of 0 means it's a root inode of root or non-root PFS. 882 * 883 * NOTE: Future note on directory hardlinks. We can implement a record type 884 * which allows us to point to multiple parent directories. 885 */ 886typedef struct hammer_inode_data { 887 uint16_t version; /* inode data version */ 888 uint16_t mode; /* basic unix permissions */ 889 uint32_t uflags; /* chflags */ 890 uint32_t rmajor; /* used by device nodes */ 891 uint32_t rminor; /* used by device nodes */ 892 uint64_t ctime; 893 int64_t parent_obj_id; /* parent directory obj_id */ 894 hammer_uuid_t uid; 895 hammer_uuid_t gid; 896 897 uint8_t obj_type; 898 uint8_t cap_flags; /* capability support flags (extension) */ 899 uint16_t reserved01; 900 uint32_t reserved02; 901 uint64_t nlinks; /* hard links */ 902 uint64_t size; /* filesystem object size */ 903 union { 904 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */ 905 } ext; 906 uint64_t mtime; /* mtime must be second-to-last */ 907 uint64_t atime; /* atime must be last */ 908} *hammer_inode_data_t; 909 910/* 911 * Neither mtime nor atime upates are CRCd by the B-Tree element. 912 * mtime updates have UNDO, atime updates do not. 913 */ 914#define HAMMER_INODE_CRCSIZE \ 915 offsetof(struct hammer_inode_data, mtime) 916 917#define HAMMER_INODE_DATA_VERSION 1 918#define HAMMER_OBJID_ROOT 1 /* root inodes # */ 919#define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */ 920 921/* 922 * Capability & implementation flags. 923 * 924 * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization 925 * for directory entries. Also see HAMMER_DIR_INODE_LOCALIZATION(). 926 */ 927#define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */ 928#define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00 929#define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01 930#define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02 931#define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03 932#define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */ 933 934#define HAMMER_DATA_DOALIGN(offset) \ 935 (((offset) + 15) & ~15) 936#define HAMMER_DATA_DOALIGN_WITH(type, offset) \ 937 (((type)(offset) + 15) & (~(type)15)) 938 939/* 940 * A HAMMER directory entry associates a HAMMER filesystem object with a 941 * namespace. It is hooked into a pseudo-filesystem (with its own inode 942 * numbering space) in the filesystem by setting the high 16 bits of the 943 * localization field. The low 16 bits must be 0 and are reserved for 944 * future use. 945 * 946 * Directory entries are indexed with a 128 bit namekey rather then an 947 * offset. A portion of the namekey is an iterator/randomizer to deal 948 * with collisions. 949 * 950 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds 951 * the filesystem object type of obj_id, e.g. a den_type equivalent. 952 * It is not stored in hammer_direntry_data. 953 * 954 * NOTE: name field / the filename data reference is NOT terminated with \0. 955 */ 956typedef struct hammer_direntry_data { 957 int64_t obj_id; /* object being referenced */ 958 uint32_t localization; /* identify pseudo-filesystem */ 959 uint32_t reserved01; 960 char name[16]; /* name (extended) */ 961} *hammer_direntry_data_t; 962 963#define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_direntry_data, name[0]) 964#define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_direntry_data, name[nlen]) 965 966/* 967 * Symlink data which does not fit in the inode is stored in a separate 968 * FIX type record. 969 */ 970typedef struct hammer_symlink_data { 971 char name[16]; /* name (extended) */ 972} *hammer_symlink_data_t; 973 974#define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0]) 975 976/* 977 * The root inode for the primary filesystem and root inode for any 978 * pseudo-fs may be tagged with an optional data structure using 979 * HAMMER_RECTYPE_PFS and localization id. This structure allows 980 * the node to be used as a mirroring master or slave. 981 * 982 * When operating as a slave CD's into the node automatically become read-only 983 * and as-of sync_end_tid. 984 * 985 * When operating as a master the read PFSD info sets sync_end_tid to 986 * the most recently flushed TID. 987 * 988 * sync_low_tid is not yet used but will represent the highest pruning 989 * end-point, after which full history is available. 990 * 991 * We need to pack this structure making it equally sized on both 32-bit and 992 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is 993 * send over the wire in hammer mirror operations. Only on 64-bit machines 994 * the size of this struct differ when packed or not. This leads us to the 995 * situation where old 64-bit systems (using the non-packed structure), 996 * which were never able to mirror to/from 32-bit systems, are now no longer 997 * able to mirror to/from newer 64-bit systems (using the packed structure). 998 */ 999struct hammer_pseudofs_data { 1000 hammer_tid_t sync_low_tid; /* full history beyond this point */ 1001 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */ 1002 hammer_tid_t sync_end_tid; /* current synchronizatoin point */ 1003 uint64_t sync_beg_ts; /* real-time of last completed sync */ 1004 uint64_t sync_end_ts; /* initiation of current sync cycle */ 1005 hammer_uuid_t shared_uuid; /* shared uuid (match required) */ 1006 hammer_uuid_t unique_uuid; /* unique uuid of this master/slave */ 1007 int32_t reserved01; /* reserved for future master_id */ 1008 int32_t mirror_flags; /* misc flags */ 1009 char label[64]; /* filesystem space label */ 1010 char snapshots[64]; /* softlink dir for pruning */ 1011 int32_t reserved02; /* was prune_{time,freq} */ 1012 int32_t reserved03; /* was reblock_{time,freq} */ 1013 int32_t reserved04; /* was snapshot_freq */ 1014 int32_t prune_min; /* do not prune recent history */ 1015 int32_t prune_max; /* do not retain history beyond here */ 1016 int32_t reserved[16]; 1017} __packed; 1018 1019typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t; 1020 1021#define HAMMER_PFSD_SLAVE 0x00000001 1022#define HAMMER_PFSD_DELETED 0x80000000 1023 1024#define hammer_is_pfs_slave(pfsd) \ 1025 (((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0) 1026#define hammer_is_pfs_master(pfsd) \ 1027 (!hammer_is_pfs_slave(pfsd)) 1028#define hammer_is_pfs_deleted(pfsd) \ 1029 (((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0) 1030 1031#define HAMMER_MAX_PFS 65536 1032#define HAMMER_MAX_PFSID (HAMMER_MAX_PFS - 1) 1033#define HAMMER_ROOT_PFSID 0 1034 1035/* 1036 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }. 1037 * 1038 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot 1039 * records are mirrored but may be independantly managed once they are laid down on 1040 * a slave. 1041 * 1042 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the 1043 * results. 1044 * 1045 * NOTE: Reserved fields must be zero (as usual) 1046 */ 1047typedef struct hammer_snapshot_data { 1048 hammer_tid_t tid; /* the snapshot TID itself (== key) */ 1049 uint64_t ts; /* real-time when snapshot was made */ 1050 uint64_t reserved01; 1051 uint64_t reserved02; 1052 char label[64]; /* user-supplied description */ 1053 uint64_t reserved03[4]; 1054} *hammer_snapshot_data_t; 1055 1056/* 1057 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }. 1058 * 1059 * Used to store the hammer cleanup config. This data is not mirrored. 1060 */ 1061typedef struct hammer_config_data { 1062 char text[1024]; 1063} *hammer_config_data_t; 1064 1065/* 1066 * Rollup various structures embedded as record data 1067 */ 1068typedef union hammer_data_ondisk { 1069 struct hammer_direntry_data entry; 1070 struct hammer_inode_data inode; 1071 struct hammer_symlink_data symlink; 1072 struct hammer_pseudofs_data pfsd; 1073 struct hammer_snapshot_data snap; 1074 struct hammer_config_data config; 1075} *hammer_data_ondisk_t; 1076 1077/* 1078 * Ondisk layout of B-Tree related structures 1079 */ 1080#include "hammer_btree.h" 1081 1082#define HAMMER_DIR_INODE_LOCALIZATION(ino_data) \ 1083 (((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ? \ 1084 HAMMER_LOCALIZE_INODE : \ 1085 HAMMER_LOCALIZE_MISC) 1086 1087#endif /* !VFS_HAMMER_DISK_H_ */ 1088