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 * 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 #include <sys/types.h> 36 #include <sys/diskslice.h> 37 #include <sys/diskmbr.h> 38 #include <sys/stat.h> 39 #include <sys/time.h> 40 #include <sys/sysctl.h> 41 #include <vfs/hammer2/hammer2_disk.h> 42 43 #include <stdio.h> 44 #include <stdlib.h> 45 #include <stdarg.h> 46 #include <stddef.h> 47 #include <unistd.h> 48 #include <string.h> 49 #include <errno.h> 50 #include <fcntl.h> 51 #include <assert.h> 52 #include <err.h> 53 #include <uuid.h> 54 55 #define MAXLABELS 4 56 57 #define hammer2_icrc32(buf, size) iscsi_crc32((buf), (size)) 58 #define hammer2_icrc32c(buf, size, crc) iscsi_crc32_ext((buf), (size), (crc)) 59 uint32_t iscsi_crc32(const void *buf, size_t size); 60 uint32_t iscsi_crc32_ext(const void *buf, size_t size, uint32_t ocrc); 61 62 static hammer2_off_t check_volume(const char *path, int *fdp); 63 static int64_t getsize(const char *str, int64_t minval, int64_t maxval, int pw); 64 static const char *sizetostr(hammer2_off_t size); 65 static uint64_t nowtime(void); 66 static int blkrefary_cmp(const void *b1, const void *b2); 67 static void usage(void); 68 69 static void format_hammer2(int fd, hammer2_off_t total_space, 70 hammer2_off_t free_space); 71 static void alloc_direct(hammer2_off_t *basep, hammer2_blockref_t *bref, 72 size_t bytes); 73 static hammer2_key_t dirhash(const unsigned char *name, size_t len); 74 75 static int Hammer2Version = -1; 76 static int ForceOpt = 0; 77 static uuid_t Hammer2_FSType; /* static filesystem type id for HAMMER2 */ 78 static uuid_t Hammer2_VolFSID; /* unique filesystem id in volu header */ 79 static uuid_t Hammer2_SupCLID; /* PFS cluster id in super-root inode */ 80 static uuid_t Hammer2_SupFSID; /* PFS unique id in super-root inode */ 81 static uuid_t Hammer2_PfsCLID[MAXLABELS]; 82 static uuid_t Hammer2_PfsFSID[MAXLABELS]; 83 static const char *Label[MAXLABELS]; 84 static hammer2_off_t BootAreaSize; 85 static hammer2_off_t AuxAreaSize; 86 static int NLabels; 87 88 #define GIG ((hammer2_off_t)1024*1024*1024) 89 90 int 91 main(int ac, char **av) 92 { 93 uint32_t status; 94 hammer2_off_t total_space; 95 hammer2_off_t free_space; 96 hammer2_off_t reserved_space; 97 int ch; 98 int fd = -1; 99 int i; 100 int nolabels = 0; 101 char *vol_fsid; 102 char *sup_clid_name; 103 char *sup_fsid_name; 104 char *pfs_clid_name; 105 char *pfs_fsid_name; 106 107 Label[NLabels++] = "LOCAL"; 108 109 /* 110 * Sanity check basic filesystem structures. No cookies for us 111 * if it gets broken! 112 */ 113 assert(sizeof(hammer2_volume_data_t) == HAMMER2_VOLUME_BYTES); 114 assert(sizeof(hammer2_inode_data_t) == HAMMER2_INODE_BYTES); 115 assert(sizeof(hammer2_blockref_t) == HAMMER2_BLOCKREF_BYTES); 116 117 /* 118 * Generate a filesystem id and lookup the filesystem type 119 */ 120 srandomdev(); 121 uuidgen(&Hammer2_VolFSID, 1); 122 uuidgen(&Hammer2_SupCLID, 1); 123 uuidgen(&Hammer2_SupFSID, 1); 124 uuid_from_string(HAMMER2_UUID_STRING, &Hammer2_FSType, &status); 125 /*uuid_name_lookup(&Hammer2_FSType, "DragonFly HAMMER2", &status);*/ 126 if (status != uuid_s_ok) { 127 errx(1, "uuids file does not have the DragonFly " 128 "HAMMER2 filesystem type"); 129 } 130 131 /* 132 * Parse arguments 133 */ 134 while ((ch = getopt(ac, av, "fL:b:m:r:V:")) != -1) { 135 switch(ch) { 136 case 'f': 137 ForceOpt = 1; 138 break; 139 case 'L': 140 if (strcasecmp(optarg, "none") == 0) { 141 nolabels = 1; 142 break; 143 } 144 if (NLabels >= MAXLABELS) { 145 errx(1, "Limit of 3 local labels"); 146 } 147 Label[NLabels++] = optarg; 148 if (strlen(Label[NLabels-1]) > HAMMER2_INODE_MAXNAME) { 149 errx(1, "Root directory label too long " 150 "(64 chars max)\n"); 151 } 152 break; 153 case 'b': 154 BootAreaSize = getsize(optarg, 155 HAMMER2_NEWFS_ALIGN, 156 HAMMER2_BOOT_MAX_BYTES, 2); 157 break; 158 case 'r': 159 AuxAreaSize = getsize(optarg, 160 HAMMER2_NEWFS_ALIGN, 161 HAMMER2_REDO_MAX_BYTES, 2); 162 break; 163 case 'V': 164 Hammer2Version = strtol(optarg, NULL, 0); 165 if (Hammer2Version < HAMMER2_VOL_VERSION_MIN || 166 Hammer2Version >= HAMMER2_VOL_VERSION_WIP) { 167 errx(1, 168 "I don't understand how to format " 169 "HAMMER2 version %d\n", 170 Hammer2Version); 171 } 172 break; 173 default: 174 usage(); 175 break; 176 } 177 } 178 179 /* 180 * Adjust Label[] and NLabels 181 */ 182 if (nolabels) { 183 NLabels = 1; 184 } else if (NLabels == 1) { 185 Label[NLabels++] = "BOOT"; 186 Label[NLabels++] = "ROOT"; 187 } 188 189 /* 190 * Check Hammer2 version 191 */ 192 if (Hammer2Version < 0) { 193 size_t olen = sizeof(Hammer2Version); 194 Hammer2Version = HAMMER2_VOL_VERSION_DEFAULT; 195 if (sysctlbyname("vfs.hammer2.supported_version", 196 &Hammer2Version, &olen, NULL, 0) == 0) { 197 if (Hammer2Version >= HAMMER2_VOL_VERSION_WIP) { 198 Hammer2Version = HAMMER2_VOL_VERSION_WIP - 1; 199 fprintf(stderr, 200 "newfs_hammer: WARNING: HAMMER2 VFS " 201 "supports higher version than I " 202 "understand,\n" 203 "using version %d\n", 204 Hammer2Version); 205 } 206 } else { 207 fprintf(stderr, 208 "newfs_hammer: WARNING: HAMMER2 VFS not " 209 "loaded, cannot get version info.\n" 210 "Using version %d\n", 211 HAMMER2_VOL_VERSION_DEFAULT); 212 } 213 } 214 215 /* 216 * Collect volume information. 217 */ 218 ac -= optind; 219 av += optind; 220 221 if (ac != 1) { 222 fprintf(stderr, "Exactly one disk device must be specified\n"); 223 exit(1); 224 } 225 total_space = check_volume(av[0], &fd); 226 227 /* 228 * ~typically 8MB alignment to avoid edge cases for reserved blocks 229 * and so raid stripes (if any) operate efficiently. 230 */ 231 total_space &= ~HAMMER2_VOLUME_ALIGNMASK64; 232 233 /* 234 * Calculate defaults for the boot area size and round to the 235 * volume alignment boundary. 236 */ 237 if (BootAreaSize == 0) { 238 BootAreaSize = HAMMER2_BOOT_NOM_BYTES; 239 while (BootAreaSize > total_space / 20) 240 BootAreaSize >>= 1; 241 if (BootAreaSize < HAMMER2_BOOT_MIN_BYTES) 242 BootAreaSize = HAMMER2_BOOT_MIN_BYTES; 243 } else if (BootAreaSize < HAMMER2_BOOT_MIN_BYTES) { 244 BootAreaSize = HAMMER2_BOOT_MIN_BYTES; 245 } 246 BootAreaSize = (BootAreaSize + HAMMER2_VOLUME_ALIGNMASK64) & 247 ~HAMMER2_VOLUME_ALIGNMASK64; 248 249 /* 250 * Calculate defaults for the redo area size and round to the 251 * volume alignment boundary. 252 */ 253 if (AuxAreaSize == 0) { 254 AuxAreaSize = HAMMER2_REDO_NOM_BYTES; 255 while (AuxAreaSize > total_space / 20) 256 AuxAreaSize >>= 1; 257 if (AuxAreaSize < HAMMER2_REDO_MIN_BYTES) 258 AuxAreaSize = HAMMER2_REDO_MIN_BYTES; 259 } else if (AuxAreaSize < HAMMER2_REDO_MIN_BYTES) { 260 AuxAreaSize = HAMMER2_REDO_MIN_BYTES; 261 } 262 AuxAreaSize = (AuxAreaSize + HAMMER2_VOLUME_ALIGNMASK64) & 263 ~HAMMER2_VOLUME_ALIGNMASK64; 264 265 /* 266 * We'll need to stuff this in the volume header soon. 267 */ 268 uuid_to_string(&Hammer2_VolFSID, &vol_fsid, &status); 269 uuid_to_string(&Hammer2_SupCLID, &sup_clid_name, &status); 270 uuid_to_string(&Hammer2_SupFSID, &sup_fsid_name, &status); 271 272 /* 273 * Calculate the amount of reserved space. HAMMER2_ZONE_SEG (4MB) 274 * is reserved at the beginning of every 2GB of storage, rounded up. 275 * Thus a 200MB filesystem will still have a 4MB reserve area. 276 * 277 * We also include the boot and redo areas in the reserve. The 278 * reserve is used to help 'df' calculate the amount of available 279 * space. 280 */ 281 reserved_space = ((total_space + HAMMER2_ZONE_MASK64) / 282 HAMMER2_ZONE_BYTES64) * HAMMER2_ZONE_SEG64; 283 284 free_space = total_space - reserved_space - 285 BootAreaSize - AuxAreaSize; 286 287 format_hammer2(fd, total_space, free_space); 288 fsync(fd); 289 close(fd); 290 291 printf("---------------------------------------------\n"); 292 printf("version: %d\n", Hammer2Version); 293 printf("total-size: %s (%jd bytes)\n", 294 sizetostr(total_space), 295 (intmax_t)total_space); 296 printf("boot-area-size: %s\n", sizetostr(BootAreaSize)); 297 printf("aux-area-size: %s\n", sizetostr(AuxAreaSize)); 298 printf("topo-reserved: %s\n", sizetostr(reserved_space)); 299 printf("free-space: %s\n", sizetostr(free_space)); 300 printf("vol-fsid: %s\n", vol_fsid); 301 printf("sup-clid: %s\n", sup_clid_name); 302 printf("sup-fsid: %s\n", sup_fsid_name); 303 for (i = 0; i < NLabels; ++i) { 304 printf("PFS \"%s\"\n", Label[i]); 305 uuid_to_string(&Hammer2_PfsCLID[i], &pfs_clid_name, &status); 306 uuid_to_string(&Hammer2_PfsFSID[i], &pfs_fsid_name, &status); 307 printf(" clid %s\n", pfs_clid_name); 308 printf(" fsid %s\n", pfs_fsid_name); 309 } 310 printf("\n"); 311 312 return(0); 313 } 314 315 static 316 void 317 usage(void) 318 { 319 fprintf(stderr, 320 "usage: newfs_hammer -L label [-f] [-b bootsize] " 321 "[-r redosize] [-V version] special ...\n" 322 ); 323 exit(1); 324 } 325 326 /* 327 * Convert the size in bytes to a human readable string. 328 */ 329 static 330 const char * 331 sizetostr(hammer2_off_t size) 332 { 333 static char buf[32]; 334 335 if (size < 1024 / 2) { 336 snprintf(buf, sizeof(buf), "%6.2f", (double)size); 337 } else if (size < 1024 * 1024 / 2) { 338 snprintf(buf, sizeof(buf), "%6.2fKB", 339 (double)size / 1024); 340 } else if (size < 1024 * 1024 * 1024LL / 2) { 341 snprintf(buf, sizeof(buf), "%6.2fMB", 342 (double)size / (1024 * 1024)); 343 } else if (size < 1024 * 1024 * 1024LL * 1024LL / 2) { 344 snprintf(buf, sizeof(buf), "%6.2fGB", 345 (double)size / (1024 * 1024 * 1024LL)); 346 } else { 347 snprintf(buf, sizeof(buf), "%6.2fTB", 348 (double)size / (1024 * 1024 * 1024LL * 1024LL)); 349 } 350 return(buf); 351 } 352 353 /* 354 * Convert a string to a 64 bit signed integer with various requirements. 355 */ 356 static int64_t 357 getsize(const char *str, int64_t minval, int64_t maxval, int powerof2) 358 { 359 int64_t val; 360 char *ptr; 361 362 val = strtoll(str, &ptr, 0); 363 switch(*ptr) { 364 case 't': 365 case 'T': 366 val *= 1024; 367 /* fall through */ 368 case 'g': 369 case 'G': 370 val *= 1024; 371 /* fall through */ 372 case 'm': 373 case 'M': 374 val *= 1024; 375 /* fall through */ 376 case 'k': 377 case 'K': 378 val *= 1024; 379 break; 380 default: 381 errx(1, "Unknown suffix in number '%s'\n", str); 382 /* not reached */ 383 } 384 if (ptr[1]) { 385 errx(1, "Unknown suffix in number '%s'\n", str); 386 /* not reached */ 387 } 388 if (val < minval) { 389 errx(1, "Value too small: %s, min is %s\n", 390 str, sizetostr(minval)); 391 /* not reached */ 392 } 393 if (val > maxval) { 394 errx(1, "Value too large: %s, max is %s\n", 395 str, sizetostr(maxval)); 396 /* not reached */ 397 } 398 if ((powerof2 & 1) && (val ^ (val - 1)) != ((val << 1) - 1)) { 399 errx(1, "Value not power of 2: %s\n", str); 400 /* not reached */ 401 } 402 if ((powerof2 & 2) && (val & HAMMER2_NEWFS_ALIGNMASK)) { 403 errx(1, "Value not an integral multiple of %dK: %s", 404 HAMMER2_NEWFS_ALIGN / 1024, str); 405 /* not reached */ 406 } 407 return(val); 408 } 409 410 static uint64_t 411 nowtime(void) 412 { 413 struct timeval tv; 414 uint64_t xtime; 415 416 gettimeofday(&tv, NULL); 417 xtime = tv.tv_sec * 1000000LL + tv.tv_usec; 418 return(xtime); 419 } 420 421 /* 422 * Figure out how big the volume is. 423 */ 424 static 425 hammer2_off_t 426 check_volume(const char *path, int *fdp) 427 { 428 struct partinfo pinfo; 429 struct stat st; 430 hammer2_off_t size; 431 432 /* 433 * Get basic information about the volume 434 */ 435 *fdp = open(path, O_RDWR); 436 if (*fdp < 0) 437 err(1, "Unable to open %s R+W", path); 438 if (ioctl(*fdp, DIOCGPART, &pinfo) < 0) { 439 /* 440 * Allow the formatting of regular files as HAMMER2 volumes 441 */ 442 if (fstat(*fdp, &st) < 0) 443 err(1, "Unable to stat %s", path); 444 size = st.st_size; 445 } else { 446 /* 447 * When formatting a block device as a HAMMER2 volume the 448 * sector size must be compatible. HAMMER2 uses 64K 449 * filesystem buffers but logical buffers for direct I/O 450 * can be as small as HAMMER2_LOGSIZE (16KB). 451 */ 452 if (pinfo.reserved_blocks) { 453 errx(1, "HAMMER cannot be placed in a partition " 454 "which overlaps the disklabel or MBR"); 455 } 456 if (pinfo.media_blksize > HAMMER2_PBUFSIZE || 457 HAMMER2_PBUFSIZE % pinfo.media_blksize) { 458 errx(1, "A media sector size of %d is not supported", 459 pinfo.media_blksize); 460 } 461 size = pinfo.media_size; 462 } 463 printf("Volume %-15s size %s\n", path, sizetostr(size)); 464 return (size); 465 } 466 467 /* 468 * Create the volume header, the super-root directory inode, and 469 * the writable snapshot subdirectory (named via the label) which 470 * is to be the initial mount point, or at least the first mount point. 471 * 472 * [----reserved_area----][boot_area][aux_area] 473 * [[vol_hdr]... ] [sroot][root] 474 * 475 * The sroot and root inodes eat 512 bytes each. newfs labels can only be 476 * 64 bytes so the root (snapshot) inode does not need to extend past 512 477 * bytes. We use the correct hash slot correct but note that because 478 * directory hashes are chained 16x, any slot in the inode will work. 479 * 480 * Also format the allocation map. 481 * 482 * NOTE: The passed total_space is 8MB-aligned to avoid edge cases. 483 */ 484 static 485 void 486 format_hammer2(int fd, hammer2_off_t total_space, hammer2_off_t free_space) 487 { 488 char *buf = malloc(HAMMER2_PBUFSIZE); 489 hammer2_volume_data_t *vol; 490 hammer2_inode_data_t *rawip; 491 hammer2_blockref_t sroot_blockref; 492 hammer2_blockref_t root_blockref[MAXLABELS]; /* Max 4 labels */ 493 uint64_t now; 494 hammer2_off_t volu_base = 0; 495 hammer2_off_t boot_base = HAMMER2_ZONE_SEG; 496 hammer2_off_t aux_base = boot_base + BootAreaSize; 497 hammer2_off_t alloc_base = aux_base + AuxAreaSize; 498 hammer2_off_t tmp_base; 499 size_t n; 500 int i; 501 502 /* 503 * Clear the entire reserve for the first 2G segment and 504 * make sure we can write to the last block. 505 */ 506 bzero(buf, HAMMER2_PBUFSIZE); 507 tmp_base = volu_base; 508 for (i = 0; i < HAMMER2_ZONE_BLOCKS_SEG; ++i) { 509 n = pwrite(fd, buf, HAMMER2_PBUFSIZE, tmp_base); 510 if (n != HAMMER2_PBUFSIZE) { 511 perror("write"); 512 exit(1); 513 } 514 tmp_base += HAMMER2_PBUFSIZE; 515 } 516 517 n = pwrite(fd, buf, HAMMER2_PBUFSIZE, 518 volu_base + total_space - HAMMER2_PBUFSIZE); 519 if (n != HAMMER2_PBUFSIZE) { 520 perror("write (at-end-of-volume)"); 521 exit(1); 522 } 523 524 /* 525 * Make sure alloc_base won't cross the reserved area at the 526 * beginning of each 2GB zone. 527 * 528 * Reserve space for the super-root inode and the root inode. 529 * Make sure they are in the same 64K block to simplify our code. 530 */ 531 assert((alloc_base & HAMMER2_PBUFMASK) == 0); 532 assert(alloc_base < HAMMER2_ZONE_BYTES64 - HAMMER2_ZONE_SEG); 533 now = nowtime(); 534 bzero(buf, HAMMER2_PBUFSIZE); 535 536 alloc_base &= ~HAMMER2_PBUFMASK64; 537 alloc_direct(&alloc_base, &sroot_blockref, HAMMER2_INODE_BYTES); 538 539 for (i = 0; i < NLabels; ++i) { 540 uuidgen(&Hammer2_PfsCLID[i], 1); 541 uuidgen(&Hammer2_PfsFSID[i], 1); 542 543 alloc_direct(&alloc_base, &root_blockref[i], 544 HAMMER2_INODE_BYTES); 545 assert(((sroot_blockref.data_off ^ root_blockref[i].data_off) & 546 HAMMER2_OFF_MASK_HI) == 0); 547 548 /* 549 * Format the root directory inode, which is left empty. 550 */ 551 rawip = (void *)(buf + (HAMMER2_OFF_MASK_LO & 552 root_blockref[i].data_off)); 553 rawip->version = HAMMER2_INODE_VERSION_ONE; 554 rawip->ctime = now; 555 rawip->mtime = now; 556 /* rawip->atime = now; NOT IMPL MUST BE ZERO */ 557 rawip->btime = now; 558 rawip->type = HAMMER2_OBJTYPE_DIRECTORY; 559 rawip->mode = 0755; 560 rawip->inum = 1; /* root inode, inumber 1 */ 561 rawip->nlinks = 1; /* directory link count compat */ 562 563 rawip->name_len = strlen(Label[i]); 564 bcopy(Label[i], rawip->filename, rawip->name_len); 565 rawip->name_key = dirhash(rawip->filename, rawip->name_len); 566 567 /* 568 * Compression mode and supported copyids. 569 * 570 * Do not allow compression when creating a "BOOT" label 571 * (pfs-create also does the same if the pfs is named "BOOT") 572 */ 573 if (strcasecmp(Label[i], "BOOT") == 0) { 574 rawip->comp_algo = HAMMER2_COMP_AUTOZERO; 575 } else { 576 rawip->comp_algo = HAMMER2_COMP_NEWFS_DEFAULT; 577 } 578 579 rawip->pfs_clid = Hammer2_PfsCLID[i]; 580 rawip->pfs_fsid = Hammer2_PfsFSID[i]; 581 rawip->pfs_type = HAMMER2_PFSTYPE_MASTER; 582 rawip->op_flags |= HAMMER2_OPFLAG_PFSROOT; 583 rawip->pfs_inum = 16; /* first allocatable inode number */ 584 585 /* rawip->u.blockset is left empty */ 586 587 /* 588 * The root blockref will be stored in the super-root inode as 589 * the only directory entry. The copyid here is the actual 590 * copyid of the storage ref. 591 * 592 * The key field for a directory entry's blockref is 593 * essentially the name key for the entry. 594 */ 595 root_blockref[i].key = rawip->name_key; 596 root_blockref[i].copyid = HAMMER2_COPYID_LOCAL; 597 root_blockref[i].keybits = 0; 598 root_blockref[i].check.iscsi32.value = 599 hammer2_icrc32(rawip, sizeof(*rawip)); 600 root_blockref[i].type = HAMMER2_BREF_TYPE_INODE; 601 root_blockref[i].methods = 602 HAMMER2_ENC_CHECK(HAMMER2_CHECK_ISCSI32) | 603 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE); 604 root_blockref[i].mirror_tid = 16; 605 root_blockref[i].flags = HAMMER2_BREF_FLAG_PFSROOT; 606 } 607 608 /* 609 * Format the super-root directory inode, giving it one directory 610 * entry (root_blockref) and fixup the icrc method. 611 * 612 * The superroot contains one directory entry pointing at the root 613 * inode (named via the label). Inodes contain one blockset which 614 * is fully associative so we can put the entry anywhere without 615 * having to worry about the hash. Use index 0. 616 */ 617 rawip = (void *)(buf + (HAMMER2_OFF_MASK_LO & sroot_blockref.data_off)); 618 rawip->version = HAMMER2_INODE_VERSION_ONE; 619 rawip->ctime = now; 620 rawip->mtime = now; 621 /* rawip->atime = now; NOT IMPL MUST BE ZERO */ 622 rawip->btime = now; 623 rawip->type = HAMMER2_OBJTYPE_DIRECTORY; 624 rawip->mode = 0700; /* super-root - root only */ 625 rawip->inum = 0; /* super root inode, inumber 0 */ 626 rawip->nlinks = 2; /* directory link count compat */ 627 628 rawip->name_len = 0; /* super-root is unnamed */ 629 rawip->name_key = 0; 630 631 rawip->comp_algo = HAMMER2_COMP_AUTOZERO; 632 633 /* 634 * The super-root is flagged as a PFS and typically given its own 635 * random FSID, making it possible to mirror an entire HAMMER2 disk 636 * snapshots and all if desired. PFS ids are used to match up 637 * mirror sources and targets and cluster copy sources and targets. 638 * 639 * (XXX whole-disk logical mirroring is not really supported in 640 * the first attempt because each PFS is in its own modify/mirror 641 * transaction id domain, so normal mechanics cannot cross a PFS 642 * boundary). 643 */ 644 rawip->pfs_clid = Hammer2_SupCLID; 645 rawip->pfs_fsid = Hammer2_SupFSID; 646 rawip->pfs_type = HAMMER2_PFSTYPE_SUPROOT; 647 rawip->pfs_inum = 16; /* first allocatable inode number */ 648 649 /* 650 * The super-root has a directory entry pointing to each local 651 * PFS. To avoid having to deal with indirect blocks we can't load 652 * up more than 8 entries, but NLabels is restricted to 4 entries 653 * to leave room for possible future mandatory PFSs. 654 */ 655 qsort(root_blockref, NLabels, sizeof(root_blockref[0]), blkrefary_cmp); 656 for (i = 0; i < NLabels; ++i) 657 rawip->u.blockset.blockref[i] = root_blockref[i]; 658 659 /* 660 * The sroot blockref will be stored in the volume header. 661 */ 662 sroot_blockref.copyid = HAMMER2_COPYID_LOCAL; 663 sroot_blockref.keybits = 0; 664 sroot_blockref.check.iscsi32.value = 665 hammer2_icrc32(rawip, sizeof(*rawip)); 666 sroot_blockref.type = HAMMER2_BREF_TYPE_INODE; 667 sroot_blockref.methods = HAMMER2_ENC_CHECK(HAMMER2_CHECK_ISCSI32) | 668 HAMMER2_ENC_COMP(HAMMER2_COMP_AUTOZERO); 669 sroot_blockref.mirror_tid = 16; 670 rawip = NULL; 671 672 /* 673 * Write out the 64K HAMMER2 block containing the root and sroot. 674 */ 675 n = pwrite(fd, buf, HAMMER2_PBUFSIZE, 676 sroot_blockref.data_off & HAMMER2_OFF_MASK_HI); 677 if (n != HAMMER2_PBUFSIZE) { 678 perror("write"); 679 exit(1); 680 } 681 682 /* 683 * Format the volume header. 684 * 685 * The volume header points to sroot_blockref. Also be absolutely 686 * sure that allocator_beg is set. 687 */ 688 bzero(buf, HAMMER2_PBUFSIZE); 689 vol = (void *)buf; 690 691 vol->magic = HAMMER2_VOLUME_ID_HBO; 692 vol->boot_beg = boot_base; 693 vol->boot_end = boot_base + BootAreaSize; 694 vol->aux_beg = aux_base; 695 vol->aux_end = aux_base + AuxAreaSize; 696 vol->volu_size = total_space; 697 vol->version = Hammer2Version; 698 vol->flags = 0; 699 700 vol->fsid = Hammer2_VolFSID; 701 vol->fstype = Hammer2_FSType; 702 703 vol->peer_type = DMSG_PEER_HAMMER2; /* LNK_CONN identification */ 704 705 vol->allocator_size = free_space; 706 vol->allocator_free = free_space; 707 vol->allocator_beg = alloc_base; 708 709 vol->sroot_blockset.blockref[0] = sroot_blockref; 710 vol->mirror_tid = 16; /* all blockref mirror TIDs set to 16 */ 711 vol->freemap_tid = 16; /* all blockref mirror TIDs set to 16 */ 712 vol->icrc_sects[HAMMER2_VOL_ICRC_SECT1] = 713 hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRC1_OFF, 714 HAMMER2_VOLUME_ICRC1_SIZE); 715 716 /* 717 * Set ICRC_SECT0 after all remaining elements of sect0 have been 718 * populated in the volume header. Note hat ICRC_SECT* (except for 719 * SECT0) are part of sect0. 720 */ 721 vol->icrc_sects[HAMMER2_VOL_ICRC_SECT0] = 722 hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRC0_OFF, 723 HAMMER2_VOLUME_ICRC0_SIZE); 724 vol->icrc_volheader = 725 hammer2_icrc32((char *)vol + HAMMER2_VOLUME_ICRCVH_OFF, 726 HAMMER2_VOLUME_ICRCVH_SIZE); 727 728 /* 729 * Write the volume header and all alternates. 730 */ 731 for (i = 0; i < HAMMER2_NUM_VOLHDRS; ++i) { 732 if (i * HAMMER2_ZONE_BYTES64 >= total_space) 733 break; 734 n = pwrite(fd, buf, HAMMER2_PBUFSIZE, 735 volu_base + i * HAMMER2_ZONE_BYTES64); 736 if (n != HAMMER2_PBUFSIZE) { 737 perror("write"); 738 exit(1); 739 } 740 } 741 742 /* 743 * Cleanup 744 */ 745 free(buf); 746 } 747 748 static void 749 alloc_direct(hammer2_off_t *basep, hammer2_blockref_t *bref, size_t bytes) 750 { 751 int radix; 752 753 radix = 0; 754 assert(bytes); 755 while ((bytes & 1) == 0) { 756 bytes >>= 1; 757 ++radix; 758 } 759 assert(bytes == 1); 760 if (radix < HAMMER2_RADIX_MIN) 761 radix = HAMMER2_RADIX_MIN; 762 763 bzero(bref, sizeof(*bref)); 764 bref->data_off = *basep | radix; 765 bref->vradix = radix; 766 767 *basep += 1U << radix; 768 } 769 770 /* 771 * Borrow HAMMER1's directory hash algorithm #1 with a few modifications. 772 * The filename is split into fields which are hashed separately and then 773 * added together. 774 * 775 * Differences include: bit 63 must be set to 1 for HAMMER2 (HAMMER1 sets 776 * it to 0), this is because bit63=0 is used for hidden hardlinked inodes. 777 * (This means we do not need to do a 0-check/or-with-0x100000000 either). 778 * 779 * Also, the iscsi crc code is used instead of the old crc32 code. 780 */ 781 static hammer2_key_t 782 dirhash(const unsigned char *name, size_t len) 783 { 784 const unsigned char *aname = name; 785 uint32_t crcx; 786 uint64_t key; 787 size_t i; 788 size_t j; 789 790 /* 791 * Filesystem version 6 or better will create directories 792 * using the ALG1 dirhash. This hash breaks the filename 793 * up into domains separated by special characters and 794 * hashes each domain independently. 795 * 796 * We also do a simple sub-sort using the first character 797 * of the filename in the top 5-bits. 798 */ 799 key = 0; 800 801 /* 802 * m32 803 */ 804 crcx = 0; 805 for (i = j = 0; i < len; ++i) { 806 if (aname[i] == '.' || 807 aname[i] == '-' || 808 aname[i] == '_' || 809 aname[i] == '~') { 810 if (i != j) 811 crcx += hammer2_icrc32(aname + j, i - j); 812 j = i + 1; 813 } 814 } 815 if (i != j) 816 crcx += hammer2_icrc32(aname + j, i - j); 817 818 /* 819 * The directory hash utilizes the top 32 bits of the 64-bit key. 820 * Bit 63 must be set to 1. 821 */ 822 crcx |= 0x80000000U; 823 key |= (uint64_t)crcx << 32; 824 825 /* 826 * l16 - crc of entire filename 827 * 828 * This crc reduces degenerate hash collision conditions 829 */ 830 crcx = hammer2_icrc32(aname, len); 831 crcx = crcx ^ (crcx << 16); 832 key |= crcx & 0xFFFF0000U; 833 834 /* 835 * Set bit 15. This allows readdir to strip bit 63 so a positive 836 * 64-bit cookie/offset can always be returned, and still guarantee 837 * that the values 0x0000-0x7FFF are available for artificial entries. 838 * ('.' and '..'). 839 */ 840 key |= 0x8000U; 841 842 return (key); 843 } 844 845 static int 846 blkrefary_cmp(const void *b1, const void *b2) 847 { 848 const hammer2_blockref_t *bref1 = b1; 849 const hammer2_blockref_t *bref2 = b2; 850 if (bref1->key < bref2->key) 851 return(-1); 852 if (bref1->key > bref2->key) 853 return(1); 854 return 0; 855 } 856