1/* $NetBSD: mke2fs.c,v 1.27 2024/02/22 14:11:52 tsutsui Exp $ */ 2 3/*- 4 * Copyright (c) 2007 Izumi Tsutsui. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27/* 28 * Copyright (c) 1980, 1989, 1993 29 * The Regents of the University of California. All rights reserved. 30 * 31 * Redistribution and use in source and binary forms, with or without 32 * modification, are permitted provided that the following conditions 33 * are met: 34 * 1. Redistributions of source code must retain the above copyright 35 * notice, this list of conditions and the following disclaimer. 36 * 2. Redistributions in binary form must reproduce the above copyright 37 * notice, this list of conditions and the following disclaimer in the 38 * documentation and/or other materials provided with the distribution. 39 * 3. Neither the name of the University nor the names of its contributors 40 * may be used to endorse or promote products derived from this software 41 * without specific prior written permission. 42 * 43 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 46 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 53 * SUCH DAMAGE. 54 */ 55 56/* 57 * Copyright (c) 1997 Manuel Bouyer. 58 * 59 * Redistribution and use in source and binary forms, with or without 60 * modification, are permitted provided that the following conditions 61 * are met: 62 * 1. Redistributions of source code must retain the above copyright 63 * notice, this list of conditions and the following disclaimer. 64 * 2. Redistributions in binary form must reproduce the above copyright 65 * notice, this list of conditions and the following disclaimer in the 66 * documentation and/or other materials provided with the distribution. 67 * 68 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 69 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 70 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 71 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 72 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 73 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 74 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 75 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 76 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 77 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 78 */ 79 80/* 81 * mke2fs.c: "re-invent (dumb but non-GPLed) wheel as a fun project" 82 * 83 * In spite of this name, there is no piece of code 84 * derived from GPLed e2fsprogs written for Linux. 85 * I referred them only to see how each structure 86 * member should be initialized. 87 * 88 * Reference: 89 * - All NetBSD sources under src/sys/ufs/ext2fs and src/sbin/fsck_ext2fs 90 * - Ext2fs Home Page 91 * https://e2fsprogs.sourceforge.net/ext2.html 92 * - Design and Implementation of the Second Extended Filesystem 93 * https://e2fsprogs.sourceforge.net/ext2intro.html 94 * - Linux Documentation "The Second Extended Filesystem" 95 * https://www.kernel.org/doc/Documentation/filesystems/ext2.txt 96 */ 97 98#include <sys/cdefs.h> 99#ifndef lint 100#if 0 101static char sccsid[] = "@(#)mkfs.c 8.11 (Berkeley) 5/3/95"; 102#else 103__RCSID("$NetBSD: mke2fs.c,v 1.27 2024/02/22 14:11:52 tsutsui Exp $"); 104#endif 105#endif /* not lint */ 106 107#include <sys/param.h> 108#include <sys/mman.h> 109#include <sys/time.h> 110#include <ufs/ext2fs/ext2fs_dinode.h> 111#include <ufs/ext2fs/ext2fs_dir.h> 112#include <ufs/ext2fs/ext2fs.h> 113#include <sys/ioctl.h> 114 115#include <err.h> 116#include <errno.h> 117#include <string.h> 118#include <unistd.h> 119#include <stdlib.h> 120#include <stddef.h> 121#include <stdio.h> 122#include <uuid.h> 123 124#include "extern.h" 125 126static void initcg(uint); 127static void zap_old_sblock(int); 128static uint cgoverhead(uint); 129static int fsinit(const struct timeval *); 130static int makedir(struct ext2fs_direct *, int); 131static void copy_dir(struct ext2fs_direct *, struct ext2fs_direct *); 132static void init_resizeino(const struct timeval *); 133static uint32_t alloc(uint32_t, uint16_t); 134static void iput(struct ext2fs_dinode *, ino_t); 135static void rdfs(daddr_t, int, void *); 136static void wtfs(daddr_t, int, void *); 137static int ilog2(uint); 138static int skpc(int, size_t, uint8_t *); 139 140/* XXX: some of these macro should be into <ufs/ext2fs/ext2fs.h>? */ 141#define EXT2_DEF_MAX_MNT_COUNT 20 142#define EXT2_DEF_FSCKINTV (180 * 24 * 60 * 60) /* 180 days */ 143#define EXT2_RESERVED_INODES (EXT2_FIRSTINO - 1) 144#define EXT2_UMASK 0755 145 146#define EXT2_INO_INDEX(ino) ((ino) - 1) /* no inode zero */ 147 148#define EXT2_LOSTFOUNDSIZE 16384 149#define EXT2_LOSTFOUNDINO EXT2_FIRSTINO /* XXX: not quite */ 150#define EXT2_LOSTFOUNDUMASK 0700 151 152#define EXT2_RESIZEINOUMASK 0600 153 154#define NBLOCK_SUPERBLOCK 1 155#define NBLOCK_BLOCK_BITMAP 1 156#define NBLOCK_INODE_BITMAP 1 157 158#define cgbase(fs, c) \ 159 ((fs)->e2fs.e2fs_first_dblock + (fs)->e2fs.e2fs_bpg * (c)) 160 161 162/* 163 * ext2fs super block and group descriptor structures 164 * 165 * We don't have to use or setup whole in-memory m_ext2fs structure, 166 * but prepare it to use several macro defined in kernel headers. 167 */ 168union { 169 struct m_ext2fs m_ext2fs; 170 char pad[SBSIZE]; 171} ext2fsun; 172#define sblock ext2fsun.m_ext2fs 173#define gd ext2fsun.m_ext2fs.e2fs_gd 174 175static uint8_t *iobuf; /* for superblock and group descriptors */ 176static int iobufsize; 177 178static uint8_t buf[MAXBSIZE]; /* for initcg() and makedir() ops */ 179 180static int fsi, fso; 181 182void 183mke2fs(const char *fsys, int fi, int fo) 184{ 185 struct timeval tv; 186 int64_t minfssize; 187 uint bcount, fbcount, ficount; 188 uint blocks_gd, blocks_per_cg, inodes_per_cg, iblocks_per_cg; 189 uint minblocks_per_cg, blocks_lastcg; 190 uint ncg, cylno, sboff; 191 uuid_t uuid; 192 uint32_t uustat; 193 int i, len, col, delta, fld_width, max_cols; 194 struct winsize winsize; 195 196 gettimeofday(&tv, NULL); 197 fsi = fi; 198 fso = fo; 199 200 /* 201 * collect and verify the block and fragment sizes 202 */ 203 if (!powerof2(bsize)) { 204 errx(EXIT_FAILURE, 205 "block size must be a power of 2, not %u", 206 bsize); 207 } 208 if (!powerof2(fsize)) { 209 errx(EXIT_FAILURE, 210 "fragment size must be a power of 2, not %u", 211 fsize); 212 } 213 if (fsize < sectorsize) { 214 errx(EXIT_FAILURE, 215 "fragment size %u is too small, minimum is %u", 216 fsize, sectorsize); 217 } 218 if (bsize < MINBSIZE) { 219 errx(EXIT_FAILURE, 220 "block size %u is too small, minimum is %u", 221 bsize, MINBSIZE); 222 } 223 if (bsize > EXT2_MAXBSIZE) { 224 errx(EXIT_FAILURE, 225 "block size %u is too large, maximum is %u", 226 bsize, MAXBSIZE); 227 } 228 if (bsize != fsize) { 229 /* 230 * There is no fragment support on current ext2fs (yet?), 231 * but some kernel code refers fsize or fpg as bsize or bpg 232 * and Linux seems to set the same values to them. 233 */ 234 errx(EXIT_FAILURE, 235 "block size (%u) can't be different from " 236 "fragment size (%u)", 237 bsize, fsize); 238 } 239 240 /* variable inodesize is REV1 feature */ 241 if (Oflag == 0 && inodesize != EXT2_REV0_DINODE_SIZE) { 242 errx(EXIT_FAILURE, "GOOD_OLD_REV file system format" 243 " doesn't support %d byte inode", inodesize); 244 } 245 246 sblock.e2fs.e2fs_log_bsize = ilog2(bsize) - LOG_MINBSIZE; 247 /* Umm, why not e2fs_log_fsize? */ 248 sblock.e2fs.e2fs_fsize = ilog2(fsize) - LOG_MINBSIZE; 249 250 sblock.e2fs_bsize = bsize; 251 sblock.e2fs_bshift = sblock.e2fs.e2fs_log_bsize + LOG_MINBSIZE; 252 sblock.e2fs_qbmask = sblock.e2fs_bsize - 1; 253 sblock.e2fs_bmask = ~sblock.e2fs_qbmask; 254 sblock.e2fs_fsbtodb = ilog2(sblock.e2fs_bsize) - ilog2(sectorsize); 255 sblock.e2fs_ipb = sblock.e2fs_bsize / inodesize; 256 257 /* 258 * Ext2fs preserves BBSIZE (1024 bytes) space at the top for 259 * bootloader (though it is not enough at all for our bootloader). 260 * If bsize == BBSIZE we have to preserve one block. 261 * If bsize > BBSIZE, the first block already contains BBSIZE space 262 * before superblock because superblock is allocated at SBOFF and 263 * bsize is a power of two (i.e. 2048 bytes or more). 264 */ 265 sblock.e2fs.e2fs_first_dblock = (sblock.e2fs_bsize > BBSIZE) ? 0 : 1; 266 minfssize = EXT2_FSBTODB(&sblock, 267 sblock.e2fs.e2fs_first_dblock + 268 NBLOCK_SUPERBLOCK + 269 1 /* at least one group descriptor */ + 270 NBLOCK_BLOCK_BITMAP + 271 NBLOCK_INODE_BITMAP + 272 1 /* at least one inode table block */ + 273 1 /* at least one data block for rootdir */ + 274 1 /* at least one data block for data */ 275 ); /* XXX and more? */ 276 277 if (fssize < minfssize) 278 errx(EXIT_FAILURE, "Filesystem size %" PRId64 279 " < minimum size of %" PRId64, fssize, minfssize); 280 281 bcount = EXT2_DBTOFSB(&sblock, fssize); 282 283 /* 284 * While many people claim that ext2fs is a (bad) clone of ufs/ffs, 285 * it isn't actual ffs so maybe we should call it "block group" 286 * as their native name rather than ffs derived "cylinder group." 287 * But we'll use the latter here since other kernel sources use it. 288 * (I also agree "cylinder" based allocation is obsolete though) 289 */ 290 291 /* maybe "simple is the best" */ 292 blocks_per_cg = sblock.e2fs_bsize * NBBY; 293 294 ncg = howmany(bcount - sblock.e2fs.e2fs_first_dblock, blocks_per_cg); 295 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 296 297 /* check range of inode number */ 298 if (num_inodes < EXT2_FIRSTINO) 299 num_inodes = EXT2_FIRSTINO; /* needs reserved inodes + 1 */ 300 if (num_inodes > UINT16_MAX * ncg) 301 num_inodes = UINT16_MAX * ncg; /* ext2bgd_nifree is uint16_t */ 302 303 inodes_per_cg = num_inodes / ncg; 304 iblocks_per_cg = howmany(inodesize * inodes_per_cg, bsize); 305 306 /* Check that the last cylinder group has enough space for inodes */ 307 minblocks_per_cg = 308 NBLOCK_BLOCK_BITMAP + 309 NBLOCK_INODE_BITMAP + 310 iblocks_per_cg + 311 1; /* at least one data block */ 312 if (Oflag == 0 || cg_has_sb(ncg - 1) != 0) 313 minblocks_per_cg += NBLOCK_SUPERBLOCK + blocks_gd; 314 315 blocks_lastcg = bcount - sblock.e2fs.e2fs_first_dblock - 316 blocks_per_cg * (ncg - 1); 317 if (blocks_lastcg < minblocks_per_cg) { 318 /* 319 * Since we make all the cylinder groups the same size, the 320 * last will only be small if there are more than one 321 * cylinder groups. If the last one is too small to store 322 * filesystem data, just kill it. 323 * 324 * XXX: Does fsck_ext2fs(8) properly handle this case? 325 */ 326 bcount -= blocks_lastcg; 327 ncg--; 328 blocks_lastcg = blocks_per_cg; 329 blocks_gd = howmany(sizeof(struct ext2_gd) * ncg, bsize); 330 inodes_per_cg = num_inodes / ncg; 331 } 332 /* roundup inodes_per_cg to make it use whole inode table blocks */ 333 inodes_per_cg = roundup(inodes_per_cg, sblock.e2fs_ipb); 334 num_inodes = inodes_per_cg * ncg; 335 iblocks_per_cg = inodes_per_cg / sblock.e2fs_ipb; 336 337 /* XXX: probably we should check these adjusted values again */ 338 339 sblock.e2fs.e2fs_bcount = bcount; 340 sblock.e2fs.e2fs_icount = num_inodes; 341 342 sblock.e2fs_ncg = ncg; 343 sblock.e2fs_ngdb = blocks_gd; 344 sblock.e2fs_itpg = iblocks_per_cg; 345 346 sblock.e2fs.e2fs_rbcount = sblock.e2fs.e2fs_bcount * minfree / 100; 347 /* e2fs_fbcount will be accounted later */ 348 /* e2fs_ficount will be accounted later */ 349 350 sblock.e2fs.e2fs_bpg = blocks_per_cg; 351 sblock.e2fs.e2fs_fpg = blocks_per_cg; 352 353 sblock.e2fs.e2fs_ipg = inodes_per_cg; 354 355 sblock.e2fs.e2fs_mtime = 0; 356 sblock.e2fs.e2fs_wtime = tv.tv_sec; 357 sblock.e2fs.e2fs_mnt_count = 0; 358 /* XXX: should add some entropy to avoid checking all fs at once? */ 359 sblock.e2fs.e2fs_max_mnt_count = EXT2_DEF_MAX_MNT_COUNT; 360 361 sblock.e2fs.e2fs_magic = E2FS_MAGIC; 362 sblock.e2fs.e2fs_state = E2FS_ISCLEAN; 363 sblock.e2fs.e2fs_beh = E2FS_BEH_DEFAULT; 364 sblock.e2fs.e2fs_minrev = 0; 365 sblock.e2fs.e2fs_lastfsck = tv.tv_sec; 366 sblock.e2fs.e2fs_fsckintv = EXT2_DEF_FSCKINTV; 367 368 /* 369 * Maybe we can use E2FS_OS_FREEBSD here and it would be more proper, 370 * but the purpose of this newfs_ext2fs(8) command is to provide 371 * a filesystem which can be recognized by firmware on some 372 * Linux based appliances that can load bootstrap files only from 373 * (their native) ext2fs, and anyway we will (and should) try to 374 * act like them as much as possible. 375 * 376 * Anyway, I hope that all newer such boxes will keep their support 377 * for the "GOOD_OLD_REV" ext2fs. 378 */ 379 sblock.e2fs.e2fs_creator = E2FS_OS_LINUX; 380 381 if (Oflag == 0) { 382 sblock.e2fs.e2fs_rev = E2FS_REV0; 383 sblock.e2fs.e2fs_features_compat = 0; 384 sblock.e2fs.e2fs_features_incompat = 0; 385 sblock.e2fs.e2fs_features_rocompat = 0; 386 } else { 387 sblock.e2fs.e2fs_rev = E2FS_REV1; 388 /* 389 * e2fsprogs say "REV1" is "dynamic" so 390 * it isn't quite a version and maybe it means 391 * "extended from REV0 so check compat features." 392 * 393 * XXX: We don't have any native tool to activate 394 * the EXT2F_COMPAT_RESIZE feature and 395 * fsck_ext2fs(8) might not fix structures for it. 396 */ 397 sblock.e2fs.e2fs_features_compat = EXT2F_COMPAT_RESIZE; 398 sblock.e2fs.e2fs_features_incompat = EXT2F_INCOMPAT_FTYPE; 399 sblock.e2fs.e2fs_features_rocompat = 400 EXT2F_ROCOMPAT_SPARSESUPER | EXT2F_ROCOMPAT_LARGEFILE; 401 } 402 403 sblock.e2fs.e2fs_ruid = geteuid(); 404 sblock.e2fs.e2fs_rgid = getegid(); 405 406 sblock.e2fs.e2fs_first_ino = EXT2_FIRSTINO; 407 sblock.e2fs.e2fs_inode_size = inodesize; 408 409 /* e2fs_block_group_nr is set on writing superblock to each group */ 410 411 uuid_create(&uuid, &uustat); 412 if (uustat != uuid_s_ok) 413 errx(EXIT_FAILURE, "Failed to generate uuid"); 414 uuid_enc_le(sblock.e2fs.e2fs_uuid, &uuid); 415 if (volname != NULL) { 416 if (strlen(volname) > sizeof(sblock.e2fs.e2fs_vname)) 417 errx(EXIT_FAILURE, "Volume name is too long"); 418 strlcpy(sblock.e2fs.e2fs_vname, volname, 419 sizeof(sblock.e2fs.e2fs_vname)); 420 } 421 422 sblock.e2fs.e2fs_fsmnt[0] = '\0'; 423 sblock.e2fs_fsmnt[0] = '\0'; 424 425 sblock.e2fs.e2fs_algo = 0; /* XXX unsupported? */ 426 sblock.e2fs.e2fs_prealloc = 0; /* XXX unsupported? */ 427 sblock.e2fs.e2fs_dir_prealloc = 0; /* XXX unsupported? */ 428 429 /* calculate blocks for reserved group descriptors for resize */ 430 sblock.e2fs.e2fs_reserved_ngdb = 0; 431 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 432 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) { 433 uint64_t target_blocks; 434 uint target_ncg, target_ngdb, reserved_ngdb; 435 436 /* reserve descriptors for size as 1024 times as current */ 437 target_blocks = 438 (sblock.e2fs.e2fs_bcount - sblock.e2fs.e2fs_first_dblock) 439 * 1024ULL; 440 /* number of blocks must be in uint32_t */ 441 if (target_blocks > UINT32_MAX) 442 target_blocks = UINT32_MAX; 443 target_ncg = howmany(target_blocks, sblock.e2fs.e2fs_bpg); 444 target_ngdb = howmany(sizeof(struct ext2_gd) * target_ncg, 445 sblock.e2fs_bsize); 446 /* 447 * Reserved group descriptor blocks are preserved as 448 * the second level double indirect reference blocks in 449 * the EXT2_RESIZEINO inode, so the maximum number of 450 * the blocks is EXT2_NINDIR(fs). 451 * (see also descriptions in init_resizeino() function) 452 * 453 * We check a number including current e2fs_ngdb here 454 * because they will be moved into reserved gdb on 455 * possible future size shrink, though e2fsprogs don't 456 * seem to care about it. 457 */ 458 if (target_ngdb > EXT2_NINDIR(&sblock)) 459 target_ngdb = EXT2_NINDIR(&sblock); 460 461 reserved_ngdb = target_ngdb - sblock.e2fs_ngdb; 462 463 /* make sure reserved_ngdb fits in the last cg */ 464 if (reserved_ngdb >= blocks_lastcg - cgoverhead(ncg - 1)) 465 reserved_ngdb = blocks_lastcg - cgoverhead(ncg - 1); 466 if (reserved_ngdb == 0) { 467 /* if no space for reserved gdb, disable the feature */ 468 sblock.e2fs.e2fs_features_compat &= 469 ~EXT2F_COMPAT_RESIZE; 470 } 471 sblock.e2fs.e2fs_reserved_ngdb = reserved_ngdb; 472 } 473 474 /* 475 * Initialize group descriptors 476 */ 477 gd = malloc(sblock.e2fs_ngdb * bsize); 478 if (gd == NULL) 479 errx(EXIT_FAILURE, "Can't allocate descriptors buffer"); 480 memset(gd, 0, sblock.e2fs_ngdb * bsize); 481 482 fbcount = 0; 483 ficount = 0; 484 for (cylno = 0; cylno < ncg; cylno++) { 485 uint boffset; 486 487 boffset = cgbase(&sblock, cylno); 488 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 489 (sblock.e2fs.e2fs_features_rocompat & 490 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 491 cg_has_sb(cylno)) { 492 boffset += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 493 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 494 (sblock.e2fs.e2fs_features_compat & 495 EXT2F_COMPAT_RESIZE) != 0) 496 boffset += sblock.e2fs.e2fs_reserved_ngdb; 497 } 498 gd[cylno].ext2bgd_b_bitmap = h2fs32(boffset); 499 boffset += NBLOCK_BLOCK_BITMAP; 500 gd[cylno].ext2bgd_i_bitmap = h2fs32(boffset); 501 boffset += NBLOCK_INODE_BITMAP; 502 gd[cylno].ext2bgd_i_tables = h2fs32(boffset); 503 if (cylno == (ncg - 1)) 504 gd[cylno].ext2bgd_nbfree = h2fs16( 505 blocks_lastcg - cgoverhead(cylno)); 506 else 507 gd[cylno].ext2bgd_nbfree = h2fs16( 508 sblock.e2fs.e2fs_bpg - cgoverhead(cylno)); 509 fbcount += fs2h16(gd[cylno].ext2bgd_nbfree); 510 if (cylno == 0) { 511 /* take reserved inodes off nifree */ 512 gd[cylno].ext2bgd_nifree = h2fs16( 513 sblock.e2fs.e2fs_ipg-EXT2_RESERVED_INODES); 514 } else { 515 gd[cylno].ext2bgd_nifree = 516 h2fs16(sblock.e2fs.e2fs_ipg); 517 } 518 ficount += fs2h16(gd[cylno].ext2bgd_nifree); 519 gd[cylno].ext2bgd_ndirs = 0; 520 } 521 sblock.e2fs.e2fs_fbcount = fbcount; 522 sblock.e2fs.e2fs_ficount = ficount; 523 524 /* 525 * Dump out summary information about file system. 526 */ 527 if (verbosity > 0) { 528 printf("%s: %u.%1uMB (%" PRId64 " sectors) " 529 "block size %u, fragment size %u\n", 530 fsys, 531 (uint)(((uint64_t)bcount * bsize) / (1024 * 1024)), 532 (uint)((uint64_t)bcount * bsize - 533 rounddown((uint64_t)bcount * bsize, 1024 * 1024)) 534 / 1024 / 100, 535 fssize, bsize, fsize); 536 printf("\tusing %u block groups of %u.0MB, %u blks, " 537 "%u inodes.\n", 538 ncg, bsize * sblock.e2fs.e2fs_bpg / (1024 * 1024), 539 sblock.e2fs.e2fs_bpg, sblock.e2fs.e2fs_ipg); 540 } 541 542 /* 543 * allocate space for superblock and group descriptors 544 */ 545 iobufsize = (NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb) * sblock.e2fs_bsize; 546 iobuf = mmap(0, iobufsize, PROT_READ|PROT_WRITE, 547 MAP_ANON|MAP_PRIVATE, -1, 0); 548 if (iobuf == NULL) 549 errx(EXIT_FAILURE, "Cannot allocate I/O buffer"); 550 memset(iobuf, 0, iobufsize); 551 552 /* 553 * We now start writing to the filesystem 554 */ 555 556 if (!Nflag) { 557 static const uint pbsize[] = { 1024, 2048, 4096, 0 }; 558 uint pblock; 559 /* 560 * Validate the given file system size. 561 * Verify that its last block can actually be accessed. 562 * Convert to file system fragment sized units. 563 */ 564 if (fssize <= 0) 565 errx(EXIT_FAILURE, "Preposterous size %" PRId64, 566 fssize); 567 wtfs(fssize - 1, sectorsize, iobuf); 568 569 /* 570 * Ensure there is nothing that looks like a filesystem 571 * superblock anywhere other than where ours will be. 572 * 573 * Ext2fs superblock is always placed at the same SBOFF, 574 * so we just zap possible first backups. 575 */ 576 for (i = 0; pbsize[i] != 0; i++) { 577 pblock = (pbsize[i] > BBSIZE) ? 0 : 1; /* 1st dblk */ 578 pblock += pbsize[i] * NBBY; /* next bg */ 579 /* zap first backup */ 580 zap_old_sblock(pblock * pbsize[i]); 581 } 582 /* 583 * Also zap possbile FFS magic leftover to prevent 584 * kernel vfs_mountroot() and bootloadres from mis-recognizing 585 * this file system as FFS. 586 */ 587 zap_old_sblock(8192); /* SBLOCK_UFS1 */ 588 zap_old_sblock(65536); /* SBLOCK_UFS2 */ 589 } 590 591 if (verbosity >= 3) 592 printf("super-block backups (for fsck_ext2fs -b #) at:\n"); 593 /* If we are printing more than one line of numbers, line up columns */ 594 fld_width = verbosity < 4 ? 1 : snprintf(NULL, 0, "%" PRIu64, 595 (uint64_t)cgbase(&sblock, ncg - 1)); 596 /* Get terminal width */ 597 if (ioctl(fileno(stdout), TIOCGWINSZ, &winsize) == 0 && 598 winsize.ws_col != 0) 599 max_cols = winsize.ws_col; 600 else 601 max_cols = 80; 602 if (Nflag && verbosity == 3) 603 /* Leave space to add " ..." after one row of numbers */ 604 max_cols -= 4; 605#define BASE 0x10000 /* For some fixed-point maths */ 606 col = 0; 607 delta = verbosity > 2 ? 0 : max_cols * BASE / ncg; 608 for (cylno = 0; cylno < ncg; cylno++) { 609 fflush(stdout); 610 initcg(cylno); 611 if (verbosity < 2) 612 continue; 613 /* the first one is a master, not backup */ 614 if (cylno == 0) 615 continue; 616 /* skip if this cylinder doesn't have a backup */ 617 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 618 (sblock.e2fs.e2fs_features_rocompat & 619 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 620 cg_has_sb(cylno) == 0) 621 continue; 622 623 if (delta > 0) { 624 if (Nflag) 625 /* No point doing dots for -N */ 626 break; 627 /* Print dots scaled to end near RH margin */ 628 for (col += delta; col > BASE; col -= BASE) 629 printf("."); 630 continue; 631 } 632 /* Print superblock numbers */ 633 len = printf("%s%*" PRIu64 ",", (col ? " " : ""), fld_width, 634 (uint64_t)cgbase(&sblock, cylno)); 635 col += len; 636 if (col + len < max_cols) 637 /* Next number fits */ 638 continue; 639 /* Next number won't fit, need a newline */ 640 if (verbosity <= 3) { 641 /* Print dots for subsequent cylinder groups */ 642 delta = sblock.e2fs_ncg - cylno - 1; 643 if (delta != 0) { 644 if (Nflag) { 645 printf(" ..."); 646 break; 647 } 648 delta = max_cols * BASE / delta; 649 } 650 } 651 col = 0; 652 printf("\n"); 653 } 654#undef BASE 655 if (col > 0) 656 printf("\n"); 657 if (Nflag) 658 return; 659 660 /* 661 * Now construct the initial file system, 662 */ 663 if (fsinit(&tv) == 0) 664 errx(EXIT_FAILURE, "Error making filesystem"); 665 /* 666 * Write out the superblock and group descriptors 667 */ 668 sblock.e2fs.e2fs_block_group_nr = 0; 669 sboff = 0; 670 if (cgbase(&sblock, 0) == 0) { 671 /* 672 * If the first block contains the boot block sectors, 673 * (i.e. in case of sblock.e2fs.e2fs_bsize > BBSIZE) 674 * we have to preserve data in it. 675 */ 676 sboff = SBOFF; 677 } 678 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 679 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + sblock.e2fs_bsize), 680 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 681 wtfs(EXT2_FSBTODB(&sblock, cgbase(&sblock, 0)) + sboff / sectorsize, 682 iobufsize - sboff, iobuf + sboff); 683 684 munmap(iobuf, iobufsize); 685} 686 687/* 688 * Initialize a cylinder (block) group. 689 */ 690void 691initcg(uint cylno) 692{ 693 uint nblcg, i, j, sboff; 694 struct ext2fs_dinode *dp; 695 696 /* 697 * Make a copy of the superblock and group descriptors. 698 */ 699 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 700 (sblock.e2fs.e2fs_features_rocompat & 701 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 702 cg_has_sb(cylno)) { 703 sblock.e2fs.e2fs_block_group_nr = cylno; 704 sboff = 0; 705 if (cgbase(&sblock, cylno) == 0) { 706 /* preserve data in bootblock in cg0 */ 707 sboff = SBOFF; 708 } 709 e2fs_sbsave(&sblock.e2fs, (struct ext2fs *)(iobuf + sboff)); 710 e2fs_cgsave(gd, (struct ext2_gd *)(iobuf + 711 sblock.e2fs_bsize * NBLOCK_SUPERBLOCK), 712 sizeof(struct ext2_gd) * sblock.e2fs_ncg); 713 /* write superblock and group descriptor backups */ 714 wtfs(EXT2_FSBTODB(&sblock, cgbase(&sblock, cylno)) + 715 sboff / sectorsize, iobufsize - sboff, iobuf + sboff); 716 } 717 718 /* 719 * Initialize block bitmap. 720 */ 721 memset(buf, 0, sblock.e2fs_bsize); 722 if (cylno == (sblock.e2fs_ncg - 1)) { 723 /* The last group could have less blocks than e2fs_bpg. */ 724 nblcg = sblock.e2fs.e2fs_bcount - 725 cgbase(&sblock, sblock.e2fs_ncg - 1); 726 for (i = nblcg; i < roundup(nblcg, NBBY); i++) 727 setbit(buf, i); 728 memset(&buf[i / NBBY], ~0U, sblock.e2fs.e2fs_bpg - i); 729 } 730 /* set overhead (superblock, group descriptor etc.) blocks used */ 731 for (i = 0; i < cgoverhead(cylno) / NBBY; i++) 732 buf[i] = ~0; 733 i = i * NBBY; 734 for (; i < cgoverhead(cylno); i++) 735 setbit(buf, i); 736 wtfs(EXT2_FSBTODB(&sblock, fs2h32(gd[cylno].ext2bgd_b_bitmap)), 737 sblock.e2fs_bsize, buf); 738 739 /* 740 * Initialize inode bitmap. 741 * 742 * Assume e2fs_ipg is a multiple of NBBY since 743 * it's a multiple of e2fs_ipb (as we did above). 744 * Note even (possibly smaller) the last group has the same e2fs_ipg. 745 */ 746 i = sblock.e2fs.e2fs_ipg / NBBY; 747 memset(buf, 0, i); 748 memset(buf + i, ~0U, sblock.e2fs_bsize - i); 749 if (cylno == 0) { 750 /* mark reserved inodes */ 751 for (i = 1; i < EXT2_FIRSTINO; i++) 752 setbit(buf, EXT2_INO_INDEX(i)); 753 } 754 wtfs(EXT2_FSBTODB(&sblock, fs2h32(gd[cylno].ext2bgd_i_bitmap)), 755 sblock.e2fs_bsize, buf); 756 757 /* 758 * Initialize inode tables. 759 * 760 * Just initialize generation numbers for NFS security. 761 * XXX: sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_valloc() seems 762 * to override these generated numbers. 763 */ 764 memset(buf, 0, sblock.e2fs_bsize); 765 for (i = 0; i < sblock.e2fs_itpg; i++) { 766 for (j = 0; j < sblock.e2fs_ipb; j++) { 767 dp = (struct ext2fs_dinode *)(buf + inodesize * j); 768 /* h2fs32() just for consistency */ 769 dp->e2di_gen = h2fs32(arc4random()); 770 } 771 wtfs(EXT2_FSBTODB(&sblock, fs2h32(gd[cylno].ext2bgd_i_tables) + i), 772 sblock.e2fs_bsize, buf); 773 } 774} 775 776/* 777 * Zap possible lingering old superblock data 778 */ 779static void 780zap_old_sblock(int sblkoff) 781{ 782 static int cg0_data; 783 uint32_t oldfs[SBSIZE / sizeof(uint32_t)]; 784 static const struct fsm { 785 uint32_t offset; 786 uint32_t magic; 787 uint32_t mask; 788 } fs_magics[] = { 789 {offsetof(struct ext2fs, e2fs_magic) / 4, E2FS_MAGIC, 0xffff}, 790 {offsetof(struct ext2fs, e2fs_magic) / 4, 791 E2FS_MAGIC << 16, 0xffff0000}, 792 {14, 0xef530000, 0xffff0000}, /* EXT2FS (big) */ 793 {0x55c / 4, 0x00011954, ~0U}, /* FS_UFS1_MAGIC */ 794 {0x55c / 4, 0x19540119, ~0U}, /* FS_UFS2_MAGIC */ 795 {0, 0x70162, ~0U}, /* LFS_MAGIC */ 796 {.offset = ~0U}, 797 }; 798 const struct fsm *fsm; 799 800 if (Nflag) 801 return; 802 803 /* don't override data before superblock */ 804 if (sblkoff < SBOFF) 805 return; 806 807 if (cg0_data == 0) { 808 cg0_data = 809 ((daddr_t)sblock.e2fs.e2fs_first_dblock + cgoverhead(0)) * 810 sblock.e2fs_bsize; 811 } 812 813 /* Ignore anything that is beyond our filesystem */ 814 if (sblkoff / sectorsize >= fssize) 815 return; 816 /* Zero anything inside our filesystem... */ 817 if (sblkoff >= sblock.e2fs.e2fs_first_dblock * bsize) { 818 /* ...unless we will write that area anyway */ 819 if (sblkoff >= cg0_data) 820 /* assume iobuf is zero'ed here */ 821 wtfs(sblkoff / sectorsize, 822 roundup(SBSIZE, sectorsize), iobuf); 823 return; 824 } 825 826 /* 827 * The sector might contain boot code, so we must validate it 828 * 829 * XXX: ext2fs won't preserve data after SBOFF, 830 * but first_dblock could have a different value. 831 */ 832 rdfs(sblkoff / sectorsize, sizeof(oldfs), &oldfs); 833 for (fsm = fs_magics;; fsm++) { 834 uint32_t v; 835 if (fsm->mask == 0) 836 return; 837 v = oldfs[fsm->offset]; 838 if ((v & fsm->mask) == fsm->magic || 839 (bswap32(v) & fsm->mask) == fsm->magic) 840 break; 841 } 842 843 /* Just zap the magic number */ 844 oldfs[fsm->offset] = 0; 845 wtfs(sblkoff / sectorsize, sizeof(oldfs), &oldfs); 846} 847 848/* 849 * uint cgoverhead(uint c) 850 * 851 * Return a number of reserved blocks on the specified group. 852 * XXX: should be shared with src/sbin/fsck_ext2fs/setup.c 853 */ 854uint 855cgoverhead(uint c) 856{ 857 uint overh; 858 859 overh = NBLOCK_BLOCK_BITMAP + NBLOCK_INODE_BITMAP + sblock.e2fs_itpg; 860 861 if (sblock.e2fs.e2fs_rev == E2FS_REV0 || 862 (sblock.e2fs.e2fs_features_rocompat & 863 EXT2F_ROCOMPAT_SPARSESUPER) == 0 || 864 cg_has_sb(c) != 0) { 865 overh += NBLOCK_SUPERBLOCK + sblock.e2fs_ngdb; 866 867 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 868 (sblock.e2fs.e2fs_features_compat & 869 EXT2F_COMPAT_RESIZE) != 0) 870 overh += sblock.e2fs.e2fs_reserved_ngdb; 871 } 872 873 return overh; 874} 875 876/* 877 * Initialize the file system 878 */ 879 880#define LOSTDIR /* e2fsck complains if there is no lost+found */ 881 882#define PREDEFDIR 2 883 884#ifdef LOSTDIR 885#define PREDEFROOTDIR (PREDEFDIR + 1) 886#else 887#define PREDEFROOTDIR PREDEFDIR 888#endif 889 890struct ext2fs_direct root_dir[] = { 891 { EXT2_ROOTINO, 0, 1, 0, "." }, 892 { EXT2_ROOTINO, 0, 2, 0, ".." }, 893#ifdef LOSTDIR 894 { EXT2_LOSTFOUNDINO, 0, 10, 0, "lost+found" }, 895#endif 896}; 897 898#ifdef LOSTDIR 899struct ext2fs_direct lost_found_dir[] = { 900 { EXT2_LOSTFOUNDINO, 0, 1, 0, "." }, 901 { EXT2_ROOTINO, 0, 2, 0, ".." }, 902}; 903struct ext2fs_direct pad_dir = { 0, sizeof(struct ext2fs_direct), 0, 0, "" }; 904#endif 905 906int 907fsinit(const struct timeval *tv) 908{ 909 struct ext2fs_dinode node; 910#ifdef LOSTDIR 911 uint i, nblks_lostfound, blk; 912#endif 913 914 /* 915 * Initialize the inode for the resizefs feature 916 */ 917 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 918 (sblock.e2fs.e2fs_features_compat & EXT2F_COMPAT_RESIZE) != 0) 919 init_resizeino(tv); 920 921 /* 922 * Initialize the node 923 */ 924 925#ifdef LOSTDIR 926 /* 927 * Create the lost+found directory 928 */ 929 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 930 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 931 lost_found_dir[0].e2d_type = EXT2_FT_DIR; 932 lost_found_dir[1].e2d_type = EXT2_FT_DIR; 933 } 934 (void)makedir(lost_found_dir, __arraycount(lost_found_dir)); 935 936 /* prepare a bit large directory for preserved files */ 937 nblks_lostfound = EXT2_LOSTFOUNDSIZE / sblock.e2fs_bsize; 938 /* ...but only with direct blocks */ 939 if (nblks_lostfound > EXT2FS_NDADDR) 940 nblks_lostfound = EXT2FS_NDADDR; 941 942 memset(&node, 0, sizeof(node)); 943 node.e2di_mode = EXT2_IFDIR | EXT2_LOSTFOUNDUMASK; 944 node.e2di_uid = geteuid(); 945 node.e2di_size = sblock.e2fs_bsize * nblks_lostfound; 946 node.e2di_atime = tv->tv_sec; 947 node.e2di_ctime = tv->tv_sec; 948 node.e2di_mtime = tv->tv_sec; 949 node.e2di_gid = getegid(); 950 node.e2di_nlink = PREDEFDIR; 951 /* e2di_nblock is a number of disk blocks, not ext2fs blocks */ 952 node.e2di_nblock = EXT2_FSBTODB(&sblock, nblks_lostfound); 953 node.e2di_blocks[0] = alloc(sblock.e2fs_bsize, node.e2di_mode); 954 if (node.e2di_blocks[0] == 0) { 955 printf("%s: can't allocate block for lost+found\n", __func__); 956 return 0; 957 } 958 for (i = 1; i < nblks_lostfound; i++) { 959 blk = alloc(sblock.e2fs_bsize, 0); 960 if (blk == 0) { 961 printf("%s: can't allocate blocks for lost+found\n", 962 __func__); 963 return 0; 964 } 965 node.e2di_blocks[i] = blk; 966 } 967 wtfs(EXT2_FSBTODB(&sblock, node.e2di_blocks[0]), 968 sblock.e2fs_bsize, buf); 969 pad_dir.e2d_reclen = sblock.e2fs_bsize; 970 for (i = 1; i < nblks_lostfound; i++) { 971 memset(buf, 0, sblock.e2fs_bsize); 972 copy_dir(&pad_dir, (struct ext2fs_direct *)buf); 973 wtfs(EXT2_FSBTODB(&sblock, node.e2di_blocks[i]), 974 sblock.e2fs_bsize, buf); 975 } 976 iput(&node, EXT2_LOSTFOUNDINO); 977#endif 978 /* 979 * create the root directory 980 */ 981 memset(&node, 0, sizeof(node)); 982 if (sblock.e2fs.e2fs_rev > E2FS_REV0 && 983 sblock.e2fs.e2fs_features_incompat & EXT2F_INCOMPAT_FTYPE) { 984 root_dir[0].e2d_type = EXT2_FT_DIR; 985 root_dir[1].e2d_type = EXT2_FT_DIR; 986#ifdef LOSTDIR 987 root_dir[2].e2d_type = EXT2_FT_DIR; 988#endif 989 } 990 node.e2di_mode = EXT2_IFDIR | EXT2_UMASK; 991 node.e2di_uid = geteuid(); 992 node.e2di_size = makedir(root_dir, __arraycount(root_dir)); 993 node.e2di_atime = tv->tv_sec; 994 node.e2di_ctime = tv->tv_sec; 995 node.e2di_mtime = tv->tv_sec; 996 node.e2di_gid = getegid(); 997 node.e2di_nlink = PREDEFROOTDIR; 998 /* e2di_nblock is a number of disk block, not ext2fs block */ 999 node.e2di_nblock = EXT2_FSBTODB(&sblock, 1); 1000 node.e2di_blocks[0] = alloc(node.e2di_size, node.e2di_mode); 1001 if (node.e2di_blocks[0] == 0) { 1002 printf("%s: can't allocate block for root dir\n", __func__); 1003 return 0; 1004 } 1005 wtfs(EXT2_FSBTODB(&sblock, node.e2di_blocks[0]), 1006 sblock.e2fs_bsize, buf); 1007 iput(&node, EXT2_ROOTINO); 1008 return 1; 1009} 1010 1011/* 1012 * Construct a set of directory entries in "buf". 1013 * return size of directory. 1014 */ 1015int 1016makedir(struct ext2fs_direct *protodir, int entries) 1017{ 1018 uint8_t *cp; 1019 uint i, spcleft; 1020 uint dirblksiz; 1021 1022 dirblksiz = sblock.e2fs_bsize; 1023 memset(buf, 0, dirblksiz); 1024 spcleft = dirblksiz; 1025 for (cp = buf, i = 0; i < entries - 1; i++) { 1026 protodir[i].e2d_reclen = EXT2FS_DIRSIZ(protodir[i].e2d_namlen); 1027 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1028 cp += protodir[i].e2d_reclen; 1029 spcleft -= protodir[i].e2d_reclen; 1030 } 1031 protodir[i].e2d_reclen = spcleft; 1032 copy_dir(&protodir[i], (struct ext2fs_direct *)cp); 1033 return dirblksiz; 1034} 1035 1036/* 1037 * Copy a direntry to a buffer, in fs byte order 1038 */ 1039static void 1040copy_dir(struct ext2fs_direct *dir, struct ext2fs_direct *dbuf) 1041{ 1042 1043 memcpy(dbuf, dir, EXT2FS_DIRSIZ(dir->e2d_namlen)); 1044 dbuf->e2d_ino = h2fs32(dir->e2d_ino); 1045 dbuf->e2d_reclen = h2fs16(dir->e2d_reclen); 1046} 1047 1048/* 1049 * void init_resizeino(const struct timeval *tv); 1050 * 1051 * Initialize the EXT2_RESEIZE_INO inode to preserve 1052 * reserved group descriptor blocks for future growth of this ext2fs. 1053 */ 1054void 1055init_resizeino(const struct timeval *tv) 1056{ 1057 struct ext2fs_dinode node; 1058 uint64_t isize; 1059 uint32_t *dindir_block, *reserved_gdb; 1060 uint nblock, i, cylno, n; 1061 1062 memset(&node, 0, sizeof(node)); 1063 1064 /* 1065 * Note this function only prepares required structures for 1066 * future resize. It's a quite different work to implement 1067 * a utility like resize_ext2fs(8) which handles actual 1068 * resize ops even on offline. 1069 * 1070 * Anyway, I'm not sure if there is any documentation about 1071 * this resize ext2fs feature and related data structures, 1072 * and I've written this function based on things what I see 1073 * on some existing implementation and real file system data 1074 * created by existing tools. To be honest, they are not 1075 * so easy to read, so I will try to implement it here without 1076 * any dumb optimization for people who would eventually 1077 * work on "yet another wheel" like resize_ext2fs(8). 1078 */ 1079 1080 /* 1081 * I'm not sure what type is appropriate for this inode. 1082 * The release notes of e2fsprogs says they changed e2fsck to allow 1083 * IFREG for RESIZEINO since a certain resize tool used it. Hmm. 1084 */ 1085 node.e2di_mode = EXT2_IFREG | EXT2_RESIZEINOUMASK; 1086 node.e2di_uid = geteuid(); 1087 node.e2di_atime = tv->tv_sec; 1088 node.e2di_ctime = tv->tv_sec; 1089 node.e2di_mtime = tv->tv_sec; 1090 node.e2di_gid = getegid(); 1091 node.e2di_nlink = 1; 1092 1093 /* 1094 * To preserve the reserved group descriptor blocks, 1095 * EXT2_RESIZEINO uses only double indirect reference 1096 * blocks in its inode entries. 1097 * 1098 * All entries for direct, single indirect and triple 1099 * indirect references are left zero'ed. Maybe it's safe 1100 * because no write operation will happen with this inode. 1101 * 1102 * We have to allocate a block for the first level double 1103 * indirect reference block. Indexes of inode entries in 1104 * this first level dindirect block are corresponding to 1105 * indexes of group descriptors including both used (e2fs_ngdb) 1106 * and reserved (e2fs_reserved_ngdb) group descriptor blocks. 1107 * 1108 * Inode entries of indexes for used (e2fs_ngdb) descriptors are 1109 * left zero'ed. Entries for reserved (e2fs_reserved_ngdb) ones 1110 * have block numbers of actual reserved group descriptors 1111 * allocated at block group zero. This means e2fs_reserved_ngdb 1112 * blocks are reserved as the second level dindirect reference 1113 * blocks, and they actually contain block numbers of indirect 1114 * references. It may be safe since they don't have to keep any 1115 * data yet. 1116 * 1117 * Each these second dindirect blocks (i.e. reserved group 1118 * descriptor blocks in the first block group) should have 1119 * block numbers of its backups in all other block groups. 1120 * I.e. reserved_ngdb[0] block in block group 0 contains block 1121 * numbers of resreved_ngdb[0] from group 1 through (e2fs_ncg - 1). 1122 * The number of backups can be determined by the 1123 * EXT2_ROCOMPAT_SPARSESUPER feature and cg_has_sb() macro 1124 * as done in the above initcg() function. 1125 */ 1126 1127 /* set e2di_size which occupies whole blocks through DINDIR blocks */ 1128 isize = (uint64_t)sblock.e2fs_bsize * EXT2FS_NDADDR + 1129 (uint64_t)sblock.e2fs_bsize * EXT2_NINDIR(&sblock) + 1130 (uint64_t)sblock.e2fs_bsize * EXT2_NINDIR(&sblock) * 1131 EXT2_NINDIR(&sblock); 1132 if (isize > UINT32_MAX && 1133 (sblock.e2fs.e2fs_features_rocompat & 1134 EXT2F_ROCOMPAT_LARGEFILE) == 0) { 1135 /* XXX should enable it here and update all backups? */ 1136 errx(EXIT_FAILURE, "%s: large_file rocompat feature is " 1137 "required to enable resize feature for this filesystem", 1138 __func__); 1139 } 1140 /* upper 32bit is stored into e2di_size_high on REV1 feature */ 1141 node.e2di_size = isize & UINT32_MAX; 1142 node.e2di_size_high = isize >> 32; 1143 1144#define SINGLE 0 /* index of single indirect block */ 1145#define DOUBLE 1 /* index of double indirect block */ 1146#define TRIPLE 2 /* index of triple indirect block */ 1147 1148 /* zero out entries for direct references */ 1149 for (i = 0; i < EXT2FS_NDADDR; i++) 1150 node.e2di_blocks[i] = 0; 1151 /* also zero out entries for single and triple indirect references */ 1152 node.e2di_blocks[EXT2FS_NDADDR + SINGLE] = 0; 1153 node.e2di_blocks[EXT2FS_NDADDR + TRIPLE] = 0; 1154 1155 /* allocate a block for the first level double indirect reference */ 1156 node.e2di_blocks[EXT2FS_NDADDR + DOUBLE] = 1157 alloc(sblock.e2fs_bsize, node.e2di_mode); 1158 if (node.e2di_blocks[EXT2FS_NDADDR + DOUBLE] == 0) 1159 errx(EXIT_FAILURE, "%s: Can't allocate a dindirect block", 1160 __func__); 1161 1162 /* account this first block */ 1163 nblock = EXT2_FSBTODB(&sblock, 1); 1164 1165 /* allocate buffer to set data in the dindirect block */ 1166 dindir_block = malloc(sblock.e2fs_bsize); 1167 if (dindir_block == NULL) 1168 errx(EXIT_FAILURE, 1169 "%s: Can't allocate buffer for a dindirect block", 1170 __func__); 1171 1172 /* allocate buffer to set data in the group descriptor blocks */ 1173 reserved_gdb = malloc(sblock.e2fs_bsize); 1174 if (reserved_gdb == NULL) 1175 errx(EXIT_FAILURE, 1176 "%s: Can't allocate buffer for group descriptor blocks", 1177 __func__); 1178 1179 /* 1180 * Setup block entries in the first level dindirect blocks 1181 */ 1182 for (i = 0; i < sblock.e2fs_ngdb; i++) { 1183 /* no need to handle used group descriptor blocks */ 1184 dindir_block[i] = 0; 1185 } 1186 for (; i < sblock.e2fs_ngdb + sblock.e2fs.e2fs_reserved_ngdb; i++) { 1187 /* 1188 * point reserved group descriptor block in the first 1189 * (i.e. master) block group 1190 * 1191 * XXX: e2fsprogs seem to use "(i % EXT2_NINDIR(&sblock))" here 1192 * to store maximum EXT2_NINDIR(&sblock) reserved gdbs. 1193 * I'm not sure what will be done on future filesystem 1194 * shrink in that case on their way. 1195 */ 1196 if (i >= EXT2_NINDIR(&sblock)) 1197 errx(EXIT_FAILURE, "%s: too many reserved " 1198 "group descriptors (%u) for resize inode", 1199 __func__, sblock.e2fs.e2fs_reserved_ngdb); 1200 dindir_block[i] = 1201 h2fs32(cgbase(&sblock, 0) + NBLOCK_SUPERBLOCK + i); 1202 1203 /* 1204 * Setup block entries in the second dindirect blocks 1205 * (which are primary reserved group descriptor blocks) 1206 * to point their backups. 1207 */ 1208 for (n = 0, cylno = 1; cylno < sblock.e2fs_ncg; cylno++) { 1209 /* skip block groups without backup */ 1210 if ((sblock.e2fs.e2fs_features_rocompat & 1211 EXT2F_ROCOMPAT_SPARSESUPER) != 0 && 1212 cg_has_sb(cylno) == 0) 1213 continue; 1214 1215 if (n >= EXT2_NINDIR(&sblock)) 1216 errx(EXIT_FAILURE, "%s: too many block groups " 1217 "for the resize feature", __func__); 1218 /* 1219 * These blocks are already reserved in 1220 * initcg() so no need to use alloc() here. 1221 */ 1222 reserved_gdb[n++] = h2fs32(cgbase(&sblock, cylno) + 1223 NBLOCK_SUPERBLOCK + i); 1224 nblock += EXT2_FSBTODB(&sblock, 1); 1225 } 1226 for (; n < EXT2_NINDIR(&sblock); n++) 1227 reserved_gdb[n] = 0; 1228 1229 /* write group descriptor block as the second dindirect refs */ 1230 wtfs(EXT2_FSBTODB(&sblock, fs2h32(dindir_block[i])), 1231 sblock.e2fs_bsize, reserved_gdb); 1232 nblock += EXT2_FSBTODB(&sblock, 1); 1233 } 1234 for (; i < EXT2_NINDIR(&sblock); i++) { 1235 /* leave trailing entries unallocated */ 1236 dindir_block[i] = 0; 1237 } 1238 free(reserved_gdb); 1239 1240 /* finally write the first level dindirect block */ 1241 wtfs(EXT2_FSBTODB(&sblock, node.e2di_blocks[EXT2FS_NDADDR + DOUBLE]), 1242 sblock.e2fs_bsize, dindir_block); 1243 free(dindir_block); 1244 1245 node.e2di_nblock = nblock; 1246 iput(&node, EXT2_RESIZEINO); 1247} 1248 1249/* 1250 * uint32_t alloc(uint32_t size, uint16_t mode) 1251 * 1252 * Allocate a block (from cylinder group 0) 1253 * Reference: src/sys/ufs/ext2fs/ext2fs_alloc.c:ext2fs_alloccg() 1254 */ 1255uint32_t 1256alloc(uint32_t size, uint16_t mode) 1257{ 1258 uint32_t loc, bno; 1259 uint8_t *bbp; 1260 uint len, map, i; 1261 1262 if (gd[0].ext2bgd_nbfree == 0) 1263 return 0; 1264 1265 if (size > sblock.e2fs_bsize) 1266 return 0; 1267 1268 bbp = malloc(sblock.e2fs_bsize); 1269 if (bbp == NULL) 1270 return 0; 1271 rdfs(EXT2_FSBTODB(&sblock, fs2h32(gd[0].ext2bgd_b_bitmap)), 1272 sblock.e2fs_bsize, bbp); 1273 1274 /* XXX: kernel uses e2fs_fpg here */ 1275 len = sblock.e2fs.e2fs_bpg / NBBY; 1276 1277#if 0 /* no need block allocation for root or lost+found dir */ 1278 for (loc = 0; loc < len; loc++) { 1279 if (bbp[loc] == 0) { 1280 bno = loc * NBBY; 1281 goto gotit; 1282 } 1283 } 1284#endif 1285 1286 loc = skpc(~0U, len, bbp); 1287 if (loc == 0) { 1288 free(bbp); 1289 return 0; 1290 } 1291 loc = len - loc; 1292 map = bbp[loc]; 1293 bno = loc * NBBY; 1294 for (i = 0; i < NBBY; i++, bno++) { 1295 if ((map & (1 << i)) == 0) 1296 goto gotit; 1297 } 1298 free(bbp); 1299 return 0; 1300 1301 gotit: 1302 if (isset(bbp, bno)) 1303 errx(EXIT_FAILURE, "%s: inconsistent bitmap", __func__); 1304 1305 setbit(bbp, bno); 1306 wtfs(EXT2_FSBTODB(&sblock, fs2h32(gd[0].ext2bgd_b_bitmap)), 1307 sblock.e2fs_bsize, bbp); 1308 free(bbp); 1309 /* XXX: modified group descriptors won't be written into backups */ 1310 gd[0].ext2bgd_nbfree = h2fs16(fs2h16(gd[0].ext2bgd_nbfree)-1); 1311 if ((mode & EXT2_IFDIR) != 0) 1312 gd[0].ext2bgd_ndirs = h2fs16(fs2h16(gd[0].ext2bgd_ndirs)+1); 1313 sblock.e2fs.e2fs_fbcount--; 1314 1315 return sblock.e2fs.e2fs_first_dblock + bno; 1316} 1317 1318/* 1319 * void iput(struct ext2fs_dinode *ip, ino_t ino) 1320 * 1321 * Put an inode entry into the corresponding table. 1322 */ 1323static void 1324iput(struct ext2fs_dinode *ip, ino_t ino) 1325{ 1326 daddr_t d; 1327 uint c, i; 1328 struct ext2fs_dinode *dp; 1329 uint8_t *bp; 1330 1331 bp = malloc(sblock.e2fs_bsize); 1332 if (bp == NULL) 1333 errx(EXIT_FAILURE, "%s: can't allocate buffer for inode", 1334 __func__); 1335 1336 /* 1337 * Reserved inodes are allocated and accounted in initcg() 1338 * so skip checks of the bitmap and allocation for them. 1339 */ 1340 if (ino >= EXT2_FIRSTINO) { 1341 c = ino_to_cg(&sblock, ino); 1342 1343 /* sanity check */ 1344 if (gd[c].ext2bgd_nifree == 0) 1345 errx(EXIT_FAILURE, 1346 "%s: no free inode %" PRIu64 " in block group %u", 1347 __func__, (uint64_t)ino, c); 1348 1349 /* update inode bitmap */ 1350 rdfs(EXT2_FSBTODB(&sblock, fs2h32(gd[0].ext2bgd_i_bitmap)), 1351 sblock.e2fs_bsize, bp); 1352 1353 /* more sanity */ 1354 if (isset(bp, EXT2_INO_INDEX(ino))) 1355 errx(EXIT_FAILURE, "%s: inode %" PRIu64 1356 " already in use", __func__, (uint64_t)ino); 1357 setbit(bp, EXT2_INO_INDEX(ino)); 1358 wtfs(EXT2_FSBTODB(&sblock, fs2h32(gd[0].ext2bgd_i_bitmap)), 1359 sblock.e2fs_bsize, bp); 1360 gd[c].ext2bgd_nifree = h2fs16(fs2h16(gd[c].ext2bgd_nifree)-1); 1361 sblock.e2fs.e2fs_ficount--; 1362 } 1363 1364 if (ino >= sblock.e2fs.e2fs_ipg * sblock.e2fs_ncg) 1365 errx(EXIT_FAILURE, "%s: inode value out of range (%" PRIu64 1366 ")", __func__, (uint64_t)ino); 1367 1368 /* update an inode entry in the table */ 1369 d = EXT2_FSBTODB(&sblock, ino_to_fsba(&sblock, ino)); 1370 rdfs(d, sblock.e2fs_bsize, bp); 1371 1372 dp = (struct ext2fs_dinode *)(bp + 1373 inodesize * ino_to_fsbo(&sblock, ino)); 1374 e2fs_isave(ip, dp, EXT2_DINODE_SIZE(&sblock)); 1375 /* e2fs_i_bswap() doesn't swap e2di_blocks addrs */ 1376 if ((ip->e2di_mode & EXT2_IFMT) != EXT2_IFLNK) { 1377 for (i = 0; i < EXT2FS_NDADDR + EXT2FS_NIADDR; i++) 1378 dp->e2di_blocks[i] = h2fs32(ip->e2di_blocks[i]); 1379 } 1380 /* h2fs32() just for consistency */ 1381 dp->e2di_gen = h2fs32(arc4random()); 1382 1383 wtfs(d, sblock.e2fs_bsize, bp); 1384 free(bp); 1385} 1386 1387/* 1388 * Read a block from the file system 1389 */ 1390void 1391rdfs(daddr_t bno, int size, void *bf) 1392{ 1393 int n; 1394 off_t offset; 1395 1396 offset = bno; 1397 n = pread(fsi, bf, size, offset * sectorsize); 1398 if (n != size) 1399 err(EXIT_FAILURE, "%s: read error for sector %" PRId64, 1400 __func__, (int64_t)bno); 1401} 1402 1403/* 1404 * Write a block to the file system 1405 */ 1406void 1407wtfs(daddr_t bno, int size, void *bf) 1408{ 1409 int n; 1410 off_t offset; 1411 1412 if (Nflag) 1413 return; 1414 offset = bno; 1415 n = pwrite(fso, bf, size, offset * sectorsize); 1416 if (n != size) 1417 err(EXIT_FAILURE, "%s: write error for sector %" PRId64, 1418 __func__, (int64_t)bno); 1419} 1420 1421int 1422ilog2(uint val) 1423{ 1424 1425 if (val == 0 || !powerof2(val)) 1426 errx(EXIT_FAILURE, "%s: %u is not a power of 2", 1427 __func__, val); 1428 1429 return ffs(val) - 1; 1430} 1431 1432/* 1433 * int skpc(int mask, size_t size, uint8_t *cp) 1434 * 1435 * Locate an unsigned character of value mask inside cp[]. 1436 * (from src/sys/lib/libkern/skpc.c) 1437 */ 1438int 1439skpc(int mask, size_t size, uint8_t *cp) 1440{ 1441 uint8_t *end; 1442 1443 end = &cp[size]; 1444 while (cp < end && *cp == (uint8_t)mask) 1445 cp++; 1446 1447 return end - cp; 1448} 1449